Copper Drives Remodeling of Metabolic State and Progression of Clear Cell Renal Cell Carcinoma

Clear cell renal cell carcinoma (ccRCC) is the most common renal cancer, typically treated with surgery for localized cases. Yet, 30% to 50% of patients relapse within 5 years postsurgery (1). Although pembrolizumab, an immune checkpoint inhibitor, is used postoperatively, it reduces recurrence by only 10% (2). The absence of biomarkers to predict recurrence or treatment response underscores the need to understand the mechanisms of ccRCC relapse and identify prognostic biomarkers for better adjuvant therapies.

The hallmark of ccRCC is chromosome 3p loss, which affects tumor suppressors VHL, PBRM1, SETD2, and BAP1. 3p loss occurs through a chromothripsis mechanism and is associated with the amplification of chromosome 5q (3–5). Inactivation of VHL and resulting activation of hypoxia-inducible transcription factors (HIF) lead to a pseudohypoxic metabolic phenotype. This includes activation of aerobic glycolysis (Warburg effect), the pentose phosphate pathway (PPP), and serine/glycine biosynthesis, and inhibition of pyruvate entry into the tricarboxylic acid (TCA) cycle with an overall decrease in electron transport chain (ETC) and mitochondrial activity (6, 7). Moreover, ccRCC accumulates lipids and glycogen.

The progression of ccRCC is tightly linked to metabolic reprogramming. This includes a decrease in lipid content, activation of glutathione (GSH) metabolism, a decrease in aspartate levels (8, 9), and activation of the TCA cycle, as well as respiratory complex I activity, maintaining NAD⁺/NADH redox balance (10). Notably, a multiomic analysis of tobacco smoke exposure, a known risk factor for ccRCC, revealed a shift in metabolism toward oxidative phosphorylation (OxPhos). This shift was linked to increased Cu accumulation and its allocation to cytochrome c oxidase (CuCOX), implying a role for Cu in regulating renal cancer cell metabolism (11).

Cu is a metal cofactor of several enzymes critical for cell survival and proliferation. In CuCOX, Cu serves as a crucial cofactor, binding to MT-CO1 and MT-CO2, mitochondrially encoded core subunits of the CuCOX complex, which transfers electrons to molecular oxygen during aerobic respiration (12). High levels of Cu have been reported in sera and tumor tissues in various cancers, whereas Cu chelators have been developed for clinical trials (13–16). Cu-induced OxPhos promotes tumor growth by supporting ATP production (13, 15, 17). Other growth effects of Cu include angiogenesis, proliferation, and metalloallosteric regulation of MEK1/2 and ULK1/2 by labile Cu in BRAF^V600E-driven tumors (18, 19). Acute exposure of cells to supraphysiologic Cu levels has toxic effects, leading to cuproptosis, a process of Cu-triggered mitochondrial cell death (20). However, during prolonged exposures, there is an adaptation and resistance to Cu toxicity, leading to cuproplasia, i.e., Cu-induced growth effects (21).

Here, we investigated the metabolic responses to Cu accumulation during ccRCC growth and progression, finding elevated Cu levels and increased Cu allocation to CuCOX in advanced or relapsing tumors. Low dietary Cu reduced or prevented xenograft tumor growth, whereas chronic exposure to high Cu in culture activated bioenergetic, biosynthetic, and redox homeostasis pathways. Cu enhanced mitochondrial oxygen consumption, respiratory supercomplex formation, PPP activity, and de novo nucleotide biosynthesis. Crucially, Cu coordinated glucose oxidation with glucose-derived GSH production to maintain redox balance, with disruption leading to cell death via H₂O₂ and Cu toxicity. Single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics revealed coordinated induction of respiratory complex subunits and GSH- and Cu-related genes during ccRCC progression. Our study demonstrates that Cu mediates metabolic reprogramming by enhancing mitochondrial CuCOX function, detoxifying Cu-associated oxidative stress, and fueling cancer cell proliferation supporting ccRCC progression.

Cu-induced mechanisms driving ccRCC progression were comprehensively explored through multiomic approaches, metallomics, metabolomics, lipidomics, and transcriptomics, using patient tumors, xenograft models, and cell line analyses (Fig. 1A).

To investigate the role of Cu in primary ccRCC, we employed size-exclusion chromatography coupled to inductively coupled plasma mass spectrometry (SEC-ICP-MS) on specimens obtained from the University of Cincinnati Cancer Center Biospecimen Shared Resource Tumor Bank. Tumors from white male patients, the demographic most commonly affected by ccRCC (22–24), were categorized into localized (stages I + II) and advanced (stages III + IV) tumors. Advanced tumors had significantly higher total Cu and Cu allocated to CuCOX than lower-stage tumors (Fig. 1B and C). In contrast, no differences were found in Cu bound to metallothioneins (MT) or small molecules in the low molecular weight fraction (Fig. 1D and E). The content of Cu in CuCOX was significantly correlated with the total Cu concentration only in advanced tumors (Fig. 1F; Supplementary Fig. S1A), suggesting a shift in Cu handling machinery in advanced tumors that favors its allocation to CuCOX. In contrast, the levels of Cu bound to MTs were significantly correlated with the total Cu concentration independently of the tumor stage, consistent with the role of MTs sequestering a fraction of the total Cu pool (Fig. 1F; Supplementary Fig. S1A). Patients with more advanced tumors also had higher levels of Cu in their sera (Fig. 1G), and there was a significant positive correlation between Cu levels in tumor tissues and sera from patients with advanced tumors but not patients with stage I + II ccRCCs (Fig. 1H; Supplementary Fig. S1B). Note that serum Cu levels in patients with ccRCC were higher than the 10 to 22 μmol/L concentrations observed in unaffected individuals (https://www.ebmconsult.com/articles/lab-test-copper-blood-level), independently of tumor stage.

Given higher Cu levels in advanced ccRCCs, we sought to evaluate whether Cu levels in surgically removed specimens predicted the likelihood of recurrence. We investigated a cohort of stage III tumors from patients with either recurrent disease within 2 years postsurgery (S3RL) or those who remained disease-free (S3DF). Tumors from patients with S3RL showed higher levels of total Cu and CuCOX content compared with those from patients with S3DF (Fig. 1I and J; Supplementary Fig. S1C), even if the levels of Cu in the sera collected before surgery did not differ (Supplementary Fig. S1D). Notably, levels of Cu in CuCOX were significantly correlated with total Cu levels in patients with S3RL but not S3DF (Fig. 1K; Supplementary Fig. S1E). These findings underscore Cu’s critical role in ccRCC progression, particularly through its impact on CuCOX levels and support using tumor Cu content and CuCOX allocation as prognostic markers.

To test the role of Cu in tumor growth, we examined the effects of dietary Cu exposure on the growth of 786-O RCC cell– and patient-derived xenografts in mice. Mice were fed custom-made chow based on an identical formula except for high and low Cu (Cu^Hi and Cu^Lo diets, respectively) or standard chow and custom-made Cu^Lo diet. A Cu^Lo diet significantly inhibited the growth of orthotopic (Fig. 1L and M) and subcutaneous (Supplementary Fig. S1F–S1H) xenografts formed by 786-O cells compared with a Cu^Hi or standard (Supplementary Fig. S1I and S1J) diets. The orthotopic xenografts grown in mice on a Cu^Hi diet showed local invasiveness and kidney infiltration (Supplementary Fig. S1K). Mice fed with a Cu^Hi diet had significantly higher serum Cu concentrations than those on a Cu^Lo diet (Fig. 1N). Moreover, the sera Cu levels in mice on the Cu^Hi diet were comparable with those measured in patients’ sera (Fig. 1G), whereas mice on the Cu^Lo diet were within the range of levels measured in unaffected subjects. Because of the small size, insufficient tumor material could be recovered from orthotopic xenografts grown in mice fed a Cu^Lo diet for metallomic analysis. Therefore, we measured Cu and CuCOX in subcutaneous xenografts. Total Cu and Cu levels allocated to CuCOX were significantly decreased in xenografts from mice fed a Cu^Lo diet (Supplementary Fig. S1L and S1M). The role of Cu in tumor growth was further corroborated in experiments using the XP374d PDX (25). This PDX was selected based on a kidney-localized advanced stage, demographics, and tumor Cu levels measured in early passage consistent with tumors analyzed in Fig. 1B–K. A Cu^Lo diet significantly decreased tumor growth, total Cu levels in tumors and sera, and Cu allocation to CuCOX (Supplementary Fig. 1O–1S). Similar to the cell line xenografts, sera Cu levels in mice with fed Cu^Hi and Cu^Lo diets were comparable with those in patients and unaffected subjects, respectively (Fig. 1S). In all experiments with dietary Cu manipulations, there was no effect on mouse body weight (Supplementary Fig. S1N). These data point to the activation of a Cu-induced molecular program, including Cu allocation to CuCOX, which drives growth in ccRCC.

To investigate the mechanistic role of Cu in ccRCC metabolic reprogramming, we adapted 786-O and RCC4 renal cancer cells to Cu^Hi conditions. Cells were cultured in media with gradually increasing Cu concentrations over 2 weeks until the intracellular Cu content reached a steady state (Supplementary Fig. S1O). Cu was delivered in a biologically relevant manner by binding it to serum proteins through preincubation of CuSO₄ with serum, which was verified by SEC-ICP-MS analysis. Isogenic Cu^Lo cells were treated identically, except for Cu addition, and cells were continuously cultured in Cu^Hi or Cu^Lo media. The total Cu content (Fig. 1T) and Cu in CuCOX measured using SEC-ICP-MS (Fig. 1U) were significantly increased in both Cu^Hi cell lines. 786-O cells accumulated more Cu than RCC4 cells, which may be attributed to the fact that 786-O cells represent more advanced ccRCC, characterized by the expression of only HIF2α, whereas HIF1α is genetically lost (26). In contrast, RCC4 cells express both HIFαs. Loss of HIF1α is an event associated with the advancement of ccRCC (26, 27).

RNA sequencing (RNA-seq) identified 532 differentially expressed genes (DEG) between Cu^Hi and Cu^Lo 786-O cells (Supplementary Table S1A). A signature comprising the top 200 DEGs with the strongest P values stratified the ccRCCs from white males available in the TCGA_KIRC_RNASeqV2_2019 data set [iLINCS (28)] into two main clusters strongly associated with the stage (Fig. 1V). These data further support the role of Cu in tumor progression. Pathway enrichment analysis showed activation of several metabolic pathways, including NRF2, OxPhos, PPP, and GSH metabolism in 786-O Cu^Hi cells (Fig. 1W; Supplementary Fig. S1P; Supplementary Table S1B). Notably, metabolic pathways enriched in 786-O Cu^Hi cells showed significant overlap with pathways enriched in a subpopulation 6 of cancer cells that we identified in scRNA-seq as characterized by concordant activation of glycolysis, ETC, and OxPhos activity (Fig. 1X).

To evaluate the significance of the transcriptomic signature, we conducted a metabolomic analysis of Cu^Hi and Cu^Lo 786-O cells. These studies revealed an increased abundance of several intermediates from the TCA cycle and metabolites related to GSH (Fig. 1Y; Supplementary Table S2), aligning with profiles found in patients with advanced ccRCC samples (8, 9). Overall, these metabolite analyses are consistent with transcriptomic analysis. Both transcriptomic and metabolomic data validate using RCC cell lines, particularly 786-O Cu^Hi cells, as a suitable model for studying Cu’s biochemical and metabolic effects in ccRCC.

The Cu^Hi cells exhibited increased mitochondrial oxygen consumption rate (OCR), including basal and maximal respiration, as measured using Seahorse XFe96, and elevated ATP production (Fig. 2A–D). The induction of OCR was accompanied by significant biochemical remodeling of ETC respiratory complexes at the proteome and lipidome levels. Blue Native PAGE analysis (29), revealed that Cu enhanced the formation of the respiratory super complex (RSC; Fig. 2E) and increased CuCOX activity in this complex as measured using an in-gel activity assay (Fig. 2F and G) without altering the total protein expression of several representative subunits for each complex (Supplementary Fig. S2A). However, Cu increased the expression of Cu-binding CuCOX subunits MT-CO1 and MT-CO2 (Fig. 2H). In addition, Cu induced the expression of COX7A2L (SCAF1), a regulatory subunit that governs the formation of RSCs (30–32), and enhanced the association of COX7A2L with the RSC (Fig. 2H–J). Moreover, Cu stimulated the association of COX17, a mitochondrial chaperone responsible for delivering Cu to CuCOX (33), with lower molecular weight CuCOX complexes before their incorporation into RSC, supporting the Cu-induced allocation of Cu to CuCOX (Fig. 2K).

The assembly and activity of RSCs occur in mitochondrial cristae and require specific lipids such as cardiolipins (CL). CLs are exclusively synthesized in mitochondria, where their remodeling is essential for the proper organization and folding of cristae, the maintenance mitochondrial ultrastructure, and the function of the ETC and OxPhos (34–36). In line with the remodeling of RSCs, LC-MS lipidomics analysis of mitochondrial fractions from Cu^Hi and Cu^Lo 786-O and RCC4 cells revealed a significant increase in CLs in Cu^Hi cells (Fig. 2L; Supplementary Fig. S2B and S2C; Supplementary Tables S3A and S3B). CLs also increased in 786-O xenografts and XP374d tumors grown in mice fed with a Cu^Hi compared with a Cu^Lo diet (Fig. 2M; Supplementary Fig. S2D and S2E; Supplementary Tables S3C and S3D). These data indicate Cu-induced coordinated reprogramming of the mitochondrial proteome and lipidome to stimulate ETC and oxygen consumption. At the same time, chronic Cu exposure did not affect the overall mitochondrial mass measured using MitoTracker Green in 786-O cells, whereas a minor increase was measured in RCC4 cells (Supplementary Fig. S2F).

Next, we analyzed the effects of Cu on the contribution of glucose and glutamine, major carbon and nitrogen donors, to nucleotides using the flux of ¹³C and ¹⁵N from [¹³C₆]-glucose or [¹³C₅,¹⁵N₂]-glutamine (Supplementary Tables S4A and S4B; Supplementary Fig. S2G–2I). We observed a significantly higher uptake of ¹³C glucose in Cu^Hi cells, whereas labeled glutamine uptake did not differ (Supplementary Fig. S2H and S2I). Cu significantly increased flux from [¹³C₆]-glucose into 6-phospho-D-gluconate, indicating activation of PPP (Fig. 2N). Cu also significantly increased the fraction of the ¹³C₅ labeled ATP, CTP, and UTP, although not GTP, in the total pool of each nucleotide, an indication of increased utilization of ribose derived from glucose via the PPP (Fig. 2O; Supplementary Fig. S2J). This is consistent with transcriptomic data demonstrating increased PPP gene expression in Cu^Hi cells (Supplementary Fig. S1P). Cu increased the fractions of labeled ATP, GTP, UTP, and CTP in the total pool of each nucleotide after 24 hours of incubation with [¹³C₅,¹⁵N₂]-glutamine (Fig. 2P; Supplementary Fig. S2K). These data demonstrate that Cu activates mitochondrial glucose oxidation and biosynthesis of nucleotides, coupled with extensive protein and lipid remodeling of ETC complexes, which enhances anabolic pathways and promotes overall oncogenic activity.

To understand the functional role of Cu-induced CuCOX activity, we inhibited CuCOX using potassium cyanide (KCN), a well-established inhibitor (37). The acute inhibition of CuCOX activity by KCN in Cu^Hi but not Cu^Lo cells was confirmed through the CuCOX in-gel activity assay (Supplementary Fig. S2L). KCN treatment was selectively cytotoxic for Cu^Hi, whereas Cu^Lo cells remained unaffected (Supplementary Fig. S2M). This cytotoxicity was rescued by aspartate (Supplementary Fig. S2N), which is consistent with the previously reported role of ETC activity in supplying aspartate for nucleotide biosynthesis (38). These data support the essential survival function of CuCOX in Cu^Hi cells, a function that is not necessary in Cu^Lo cells.

To specifically investigate the role of CuCOX in Cu-driven tumor growth, we stably knocked down COX4I1, a subunit crucial for CuCOX biogenesis (39). This knockdown resulted in a significant decrease in MT-CO1 levels, and the expression of the COX4I2 isoform did not compensate for the loss of COX4I1 (Supplementary Fig. S2O). Cells with COX4I1 knockdown showed diminished basal OCR and ATP production, along with a blunted Cu-dependent induction of OCR (Supplementary Fig. S2P and S2Q). In xenograft models, COX4I1 knockdown did not affect tumor growth in mice on the Cu^Lo diet (Fig. 2Q and R). However, in mice fed the Cu^Hi diet, COX4I1 knockdown abolished the Cu-dependent increase in tumor growth and resulted in significantly smaller tumors than those fed a Cu^Lo diet (Fig. 2Q and R). This supports the complex role of CuCOX activity in providing the biosynthetic substrates and bioenergy necessary for proliferation and survival. The decrease in CuCOX in tumors was confirmed by SEC-ICP-MS (Supplementary Fig. S2R). Importantly, all tumors formed by control cells in mice fed the Cu^Hi diet were locally invasive, infiltrating peritumoral fat tissue, muscles, and nerves (Supplementary Fig. S2S and S2T). In contrast, tumors with COX4I1 knockdown grown in mice fed the Cu^Hi diet, as well as tumors formed in mice fed the Cu^Lo diet, were encapsulated and only invasive into the peritumoral fat (Supplementary Fig. S2S and S2T). Tumor formation did not affect the mice’s body weight (Supplementary Fig. S2U).

To determine if pharmacologic inhibitors of ETC affect tumor growth in mice fed a Cu^Hi diet, mice bearing XP374d tumors were dosed for 3 weeks with 5 mg/kg IACS-10579, an inhibitor of complex I (40), or vehicle administered by oral gavage every other day. The treatment had a partial but significant effect inhibiting tumor growth (Fig. 2S–U); however, it was accompanied by body weight loss, indicating general toxicity (Supplementary Fig. S2V). The impact on tumor growth was consistent with partial significant inhibition of CuCOX as measured using SEC-ICP-MS (Supplementary Fig. S2W).

Importantly, we found that the areas of tumor local invasiveness into the kidney tissue or around lipid droplets or muscle tissue showed enrichment for MT-CO2 staining in cancer cells both in orthotopic (Fig. 2V) or subcutaneous (Supplementary Fig. S2X) xenografts. The anti–MT-CO2 antibody is specific for human MT-CO2 (Supplementary Fig. S2Y). These data underscore the pivotal and causative role of CuCOX and ETC activity in Cu-driven tumor growth and progression.

To further understand the functional role of glucose and glutamine metabolism in Cu^Hi and Cu^Lo cells, we determined the effects of glucose or glutamine starvation on cell survival. Cu^Hi cells were markedly more sensitive to glucose starvation than Cu^Lo cells (Fig. 3A), an effect largely rescued by exogenous pyruvate or dimethylα-ketoglutarate (DMKG; Fig. 3B). In contrast, glutamine starvation induced a similar degree of cell death in both Cu^Hi and Cu^Lo cells (Supplementary Fig. S3A), partially rescued by DMKG but not pyruvate (Supplementary Fig. S3B). In parallel studies, we observed that Cu^Hi cells were also selectively hypersensitive to the inhibition of GSH biosynthesis with buthionine sulphoximine [BSO (41); Fig. 3C]. GSH and GSSG levels were higher in Cu^Hi as compared with Cu^Lo cells (Fig. 3D), complementing and validating metabolomic data showing increased abundance of intermediates of GSH pathway in Cu^Hi cells (Fig. 1Y; Supplementary Table S2). Next, we examined the effects of glucose and glutamine starvation on cellular GSH levels. Glucose starvation decreased only GSH but not GSSG in Cu^Hi, whereas Cu^Lo cells remained unaffected (Fig. 3D). GSH/GSSG ratio was lower in Cu^Hi cells when compared with Cu^Lo cells under normal glucose conditions and was further decreased by glucose starvation only in Cu^Hi cells (Supplementary Fig. S3C). The decrease in GSH/GSSG ratio in low glucose was rescued by pyruvate (Fig. 3E). In contrast, glutamine starvation led to substantial and similar depletion of both GSH and GSSG in Cu^Lo and Cu^Hi cells (Fig. 3D). The critical role of glucose-derived GSH in the survival of Cu^Hi cells was demonstrated by the observation that exogenously supplied ethyl ester of GSH (GSH-EE) significantly prevented the cytotoxicity induced by glucose starvation (Fig. 3F). Additionally, BSO blocked the rescue of the cytotoxic effect of glucose starvation by exogenous pyruvate (Fig. 3G). In contrast, cell death induced by glutamine starvation in Cu^Hi and Cu^Lo cells was not prevented by exogenous GSH-EE (Supplementary Fig. S3B).

We identified three pools of labeled intracellular GSH using metabolic labeling with [¹³C₆]-glucose or [¹³C₅,¹⁵N₂]-glutamine (Supplementary Tables S4A and S4B). [¹³C₂]GSH was the predominant glucose-derived form of labeled GSH in cells treated with labeled glucose, and its abundance was increased in Cu^Hi cells, contributing 14% of the total GSH in Cu^Hi and Cu^Lo cells (Fig. 3H and I; Supplementary Fig. S3D). The abundance and the fraction of glutamine-derived [¹³C₅,¹⁵N₁]GSH were increased in Cu^Hi cells and represented most of the total GSH (Fig. 3I and J; Supplementary Fig. S3E). In contrast, Cu did not change the abundance and fraction of glutamine-derived [¹³C₅]GSH, contributing about 20% of total cellular GSH (Fig. 3I and J; Supplementary Fig. S3E).

The biosynthesis of GSH requires cysteine, which is obtained from the reduction of cysteine. Cysteine is imported in exchange for glutamate through the cysteine/glutamate antiporter, SLC7A11 (42). This antiporter functions through an exchange mechanism, transporting intracellular glutamate out of the cell while simultaneously bringing extracellular cysteine into the cell. SLC7A11 is an NRF2 target that was upregulated in Cu^Hi cells (Supplementary Fig. S3F and S3G). In line with GSH labeling, chronic Cu treatment increased excretion into the culture media of glucose-derived [¹³C₂]-glutamate and glutamine-derived [¹³C₅,¹⁵N₁]-glutamate, but not [¹³C₅]-glutamate (Fig. 3K; Supplementary Fig. S3H). This indicates that utilization of the glutamate derived from different sources for Cu-induced GSH biosynthesis is matched by glutamate utilization for cysteine exchange at the plasma membrane.

We validated the flux of carbon from pyruvate to GSH by measuring the enrichment of intracellular GSH, GSSG, and γ-glutamyl-cysteine when glucose-starved cells were treated with [2,3-¹³C₂]-pyruvate, an effect prevented by co-treatment with BSO (Fig. 3L; Supplementary Fig. S3I; Supplementary Table S4C). Importantly, fractions of these [¹³C₂]-labeled GSH metabolites derived from ¹³C-labeled pyruvate were similar to the fraction of ¹³C₆ glucose-derived [¹³C₂]GSH after 24 hours of labeling, and they were respectively at 11.6%, 12.1%, and 16.1%. We also detected 14% of [¹³C₂]-ketoglutarate and 17% of [¹³C₂]-glutamate from labeled pyruvate, supporting the role of the TCA cycle in the biosynthesis of a GSH pool from glucose. Finally, the abundance of [¹³C₂]GSH and total GSH, as well as of [¹³C₂]-glutamate and total glutamate, substantially decreased in Cu^Hi cells with COX4I1 knockdown (Fig. 3M; Supplementary Fig. S3J; Supplementary Table S4D). COX4I1 knockdown also led to a significant decrease in intracellular cysteine consistent with its correlation with glutamate excretion via the SLC7A11 antiporter, whereas there was no change in the glycine abundance. Neither of these amino acids was labeled with glucose-derived ¹³C (Supplementary Fig. S3K). These data indicate that the biosynthesis of glucose-derived GSH is causatively coupled to the activity of CuCOX and the ETC.

The NRF2 transcription factor regulates GSH biosynthesis (43). Cu^Hi cells showed induction of the NRF2 signature (Fig. 1W). Cu^Hi cells exhibited increased nuclear localization of NRF2 and decreased cytosolic expression of its inhibitory partner, KEAP1 (Supplementary Fig. S3L). They also increased the expression of several NRF2 target genes, including catalytic and modulatory subunits of glutamate-cysteine ligase (GCLC and GCLM), the rate-limiting enzyme in GSH biosynthesis (Supplementary Fig. S3F and S3G).

These results demonstrate that the Cu-driven activity of COX and ETC drives glucose-derived biosynthesis of a fraction of GSH that is essential for maintaining Cu^Hi cells’ viability. Moreover, data point to the biochemical and functional compartmentalization of glucose- and glutamine-derived GSH.

Glucose-derived pyruvate is used in several biochemical reactions. The cytosol converts it to lactate by lactate dehydrogenase (LDH), regenerating the NAD needed for active glycolysis. It enters the mitochondria via the pyruvate carriers MPC1/MPC2 to be converted to acetyl-CoA by the pyruvate dehydrogenase (PDH) complex or to oxaloacetate by pyruvate carboxylase (PC). Pyruvate is also converted to alanine by glutamate pyruvate transaminase (GPT; Fig. 4A). In ccRCC, conversion of pyruvate to acetyl-CoA is inhibited by HIF-induced PDH kinase (PDK1) that phosphorylates the PDHA subunit, which inhibits PDH activity (44, 45).

Stable isotope tracing using [¹³C₆]-glucose in Cu^Hi cells revealed a significantly increased fraction of [¹³C₃]-pyruvate after 24 hours of labeling (Fig. 4B; Supplementary Fig. S4A; Supplementary Table S4A). There was a significant increase in labeled [¹³C₃]-lactate after 5 hours (Supplementary Fig. S4A) and an increased abundance and fraction of [¹³C₃]-alanine after 24 hours of incubation with labeled glucose (Fig. 4C; Supplementary Fig. S4A). Moreover, after 24 hours, there was a significant increase in the abundance and fraction of [¹³C₂]-citrate (Fig. 4D; Supplementary Fig. S4A) and abundance of [¹³C₂]-ketoglutarate (Fig. 4E; Supplementary Fig. S4A), indicating an increased entry and processing of glucose carbon via the TCA cycle. In contrast, labeling of the [¹³C₃]-TCA cycle intermediates was very low. The fraction of [¹³C₃]-citrate decreased in Cu^Hi cells (from 6.6% ± 0.12% to 5.6% ± 0.22%, P = 0.0004), whereas the fraction of [¹³C₃]-ketoglutarate did not change (Supplementary Table S4A), indicating lack of Cu-mediated effects on PC pathway activity. The data point to an increased pyruvate flux to the TCA cycle via PDH, which indicates that the inhibitory effect of PDK1 on PDH activity is decreased in Cu^Hi cells. In line with this, in Cu^Hi cells, we found reduced phosphorylation of PDHA Ser232, Ser293, and Ser300 without significant changes in the expression of PDK1 or phosphatase, PDP1 (Fig. 4F).

Next, we investigated the functional role of pyruvate processing pathways in rescuing glucose starvation-dependent cell death. Inhibition of LDH with the selective inhibitor GNE-140 (46) slightly increased pyruvate’s rescue of glucose starvation-induced cytotoxicity in 786-O cells but had no effect in RCC4 Cu^Hi cells (Fig. 4G). Inhibition of MPC1/2 pyruvate transporters with UK-5099 did not affect pyruvate rescue during glucose starvation in either cell line. In many cancers, including ccRCC, the activity of MPC1/2 pyruvate transporters is low, and low MPC levels are associated with poor prognosis (Supplementary Fig. S4B; refs. 47, 48). This result points to the alternative pathways of pyruvate entry into mitochondria.

Notably, an inhibitor of GPTs (GPT/GPT2), cycloserine (49) reversed pyruvate rescue of glucose starvation-induced cell death (Fig. 4G), indicating that pyruvate may enter mitochondria via alanine transport. The expression of the predominantly cytosolic GPT isoform is very low in renal cancer cells (The Human Protein Atlas) and was not detected in 786-O and RCC4 cells (Supplementary Fig. S4C). However, mitochondrial GPT2 isoform was detected in the cytosol (Supplementary Fig. S4D). Genetic inhibition of GPT2 using siRNA and validated shRNA (50, 51) reversed pyruvate rescue of glucose starvation-induced cell death (Fig. 4H; Supplementary Fig. S4E). The functional effectiveness of the GPT2 knockdown was further validated by a complete loss of glucose-derived [¹³C₃]-alanine (Fig. 4I; Supplementary Fig. S4F; Supplementary Table S4E). The accumulation of the glycolytic metabolites [¹³C₃]-phosphoenolpyruvate (PEP) and [¹³C₃]-phosphoglycerate (PGA) was increased, likely in a feedback mechanism, indicating downstream inhibition of the pathway using glucose carbon for alanine biosynthesis (Fig. 4I; Supplementary Fig. S4F; Supplementary Table S4E). GPT2 knockdown significantly decreased the abundance of [¹³C₂]-glutamate, indicating that GPT2 was required to generate glucose-derived glutamate (Fig. 4I; Supplementary Fig. S4F; Supplementary Table S4E). Expression levels of GPT2 were minimally affected in Cu^Hi as compared with Cu^Lo cells (Supplementary Fig. S4C).

Surprisingly, adding alanine to the culture media did not reverse glucose starvation-induced cell death (Fig. 4G). To determine the utilization of extracellular alanine, we treated cells with [¹³C₃,¹⁵N₁]-alanine under normal and glucose starvation conditions. Consistent with the contribution of glucose to endogenous alanine biosynthesis, glucose starvation reduced cellular levels of endogenous alanine. Exogenous, uniformly labeled alanine uptake was less significantly decreased (Supplementary Fig. S4G; Supplementary Table S4F). Analysis of the flux of ¹³C and ¹⁵N from labeled exogenous alanine into GSH revealed that only ¹⁵N₁ from alanine was detected in GSH, and this labeling decreased in low glucose conditions (Supplementary Fig. S4H; Supplementary Table S4F), supporting that glucose-derived carbon is necessary for GSH biosynthesis. The level of the ¹³C₂ label in GSH was very small and, although increased, did not compensate for the loss of endogenous alanine (Supplementary Fig. S4H; Supplementary Table S4F). These data suggest that endogenous and exogenous pools of alanine are utilized and regulated by compartmentalized routes. The limited use of exogenous alanine for metabolic activity indicates that it is predominantly reserved for protein synthesis. A similar utilization of exogenous alanine for protein synthesis but not for metabolic activity was described during the activation of T lymphocytes (52). We propose that pyruvate undergoes transamination to alanine in the cytosol, and alanine enters mitochondria, where it is converted back to pyruvate. GPT2 activity is then further required for selective transamination of glucose-derived [¹³C₂]-ketoglutarate to [¹³C₂]-glutamate, which is used in GSH biosynthesis and is essential for the survival of Cu^Hi cells.

GSH biosynthesis is localized to the cytosol, thus glucose-derived [¹³C₂]-glutamate needs to exit mitochondria. Two mitochondrial glutamate carriers, SLC25A22 and SLC25A18 (GC1 and GC2) show bidirectional activity symporting glutamate and H⁺ to and from mitochondria (53–55). We prioritized the analysis of SLC25A22 because its high mRNA level is associated with significantly worse survival in ccRCC (Supplementary Fig. S4I). Knockdown of the SLC25A22 transporter reversed rescue of glucose starvation-induced cytotoxicity by exogenously provided pyruvate (Fig. 4J; Supplementary Fig. S4J), indicating that export of glutamate from mitochondria by SLC25A22 is essential for cell survival in Cu^Hi cells. Finally, we determined that knockdowns of GPT2 and SLC25A22 significantly diminished the flux of carbon from [¹³C₆]-glucose into [¹³C₂]GSH as well as total levels of GSH (Fig. 4K; Supplementary Fig. S4K; Supplementary Table S4E).

Overall, these data support the role of GPT2 and SLC25A22 in producing and delivering mitochondrial glutamate derived from glucose to the cytosol for the biosynthesis of a pool of GSH (Fig. 4L). This mechanism couples the biosynthesis of GSH and regulation of redox homeostasis with glucose oxidation and ETC activity, an essential process for the survival of Cu^Hi cells.

Cu^Hi cells show constitutive oxidative stress as demonstrated by a reduced GSH/GSSG ratio (Supplementary Fig. S3C). Glucose starvation further lowered GSH/GSSG ratios in Cu^Hi cells (Supplementary Fig. S3C). Glucose starvation augmented OCR in both Cu^Lo and Cu^Hi cells, and the effect was stronger in Cu^Hi cells (Fig. 5A). It was not reversed by exogenous pyruvate treatment (Fig. 5A). The increase in OCR was accompanied by augmented mitochondrial superoxide accumulation measured using MitoSOX that was significantly higher in Cu^Hi cells and was reversed by pyruvate (Fig. 5B). Using the CM-H2DCFDA probe, which is more specific for H₂O₂, we observed an increase in H₂O₂ levels in response to glucose starvation exclusively in Cu^Hi cells. This effect was completely reversed by the addition of pyruvate (Fig. 5C). Moreover, the ability of pyruvate to counteract H₂O₂ accumulation in glucose-starved cells was blocked by inhibiting GSH biosynthesis with BSO (Fig. 5C) or by knocking down GPT2 (Fig. 5D) and SLC25A22 (Fig. 5E). In line with this, treatment with exogenous GSH-EE inhibited the accumulation of H₂O₂ (Fig. 5F).

Because mitochondrial OCR is the major source of reactive oxygen species, we investigated if the entry of GSH into mitochondria is necessary to rescue glucose starvation-induced cell death by pyruvate. Two GSH importers were recently identified, SLC25A39 and SLC25A40 (56, 57). The expression of SLC25A40 was induced in Cu^Hi cells (Supplementary Fig. S5A). In contrast, the expression of another mitochondrial GSH importer, SLC25A39, was decreased in Cu^Hi 786-O cells or minimally affected in RCC4 cells (Supplementary Fig. S5A). Knockdown of SLC25A40 (Supplementary Fig. S5B) abolished pyruvate rescue of glucose starvation-induced accumulation of H₂O₂ and cytotoxicity (Fig. 5G and H). In contrast, knockdown of SLC25A39 did not affect the pyruvate rescue of glucose starvation-induced cell death (Supplementary Fig. S5C). Note that the expression of SLC25A39 was validated by its induction by BSO/erastin treatment in HEK293 cells (Supplementary Fig. S5D; ref. 56). However, this treatment did not affect SLC25A39 expression in RCC cells; instead, it induced SLC25A40 expression in 786-O cells (Supplementary Fig. S5D), pointing to potential tissue-specific activities of these transporters.

Expression levels of enzymes detoxifying reactive oxygen species, such as SOD1, SOD2, catalase, mitochondrial GSH–independent peroxiredoxin 3 (PRDX3), and cytosolic GSH-dependent peroxiredoxin 6 (PRDX6), were not affected by Cu or glucose concentration, whereas expression of mitochondrial GSH peroxidases (GPX1/2) was decreased in Cu^Hi cells, an effect that may be attributed to the known activity of Cu in reducing expression of cellular selenoproteins (Supplementary Fig. S5E; refs. 58, 59). These results indicate that GSH is involved in alternative mechanisms preventing the accumulation of H₂O₂ by modulating H₂O₂ generation or decomposition or both.

Next, we investigated the mechanism of cell death induced by glucose starvation. Treatment of glucose-starved Cu^Hi cells with PEG-catalase prevented death, supporting a causal role of H₂O₂ in the death of Cu^Hi cells (Fig. 5I). The critical role of ETC-induced H₂O₂ production during glucose starvation in cell death was further validated by COX4I1 knockdown, which reduced H₂O₂ levels and partially prevented the cytotoxicity (Fig. 5J and K), even though GSH production was also decreased (Fig. 3M). Glucose-starvation-induced cell death was entirely prevented by Cu-free media applied for the duration of starvation or by tetrathiomolybdate, an intracellular Cu chelator (Fig. 5L; ref. 60). These data are consistent with Fenton-like chemistry involving the labile Cu and H₂O₂ in cell death during glucose deprivation in Cu^Hi cells. Fenton reactions generate hydroxyl radicals that mediate broad oxidative stress, causing cell death via oxidation of nucleic acids, lipids, and proteins. Importantly, knockdown of COX4I1 or treatment with IACS-10759 increased the fraction of low molecular weight labile Cu in tumors in mice fed with a Cu^Hi diet (Supplementary Fig. S5F and S5G), supporting the toxic role of free Cu and the biogenesis of CuCOX as a physiologic mechanism preventing this toxicity.

Finally, we investigated if the Cu-induced production of glucose-derived GSH is required for tumor growth. Cells with stable knockdowns of GPT2, SLC25A22, and SLC25A40 showed significantly reduced formation of xenograft tumors in mice fed a Cu^Hi diet (Fig. 5M and N) supporting Cu-dependent GSH biosynthesis as an essential survival event promoting tumor growth in response to Cu accumulation. These tumors were encapsulated and noninvasive. Notably, the weight of knockdown tumors was not significantly different from those of the COX4I1 knockdown (as shown in Fig. 2Q and R), supporting that both GSH production and ETC are tumor-promoting activities of Cu. Tumor formation did not affect mouse body weight (Supplementary Fig. S5H).

These data indicate that Cu-dependent activation of CuCOX leads, on the one hand, to increased OCR and H₂O₂ production and, on the other hand, to a coordinated biosynthesis of glucose-derived GSH to counteract the effects of H₂O₂ generated by ETC activity to prevent toxicity of the Fenton-like reaction with Cu (Fig. 5O). The absence of this GSH activity results in cell death.

To determine the activity of ETC, OxPhos, Cu-GSH pathways during the progression of ccRCC, we investigated the expression of genes encoding subunits of the ETC, as well as Cu and GSH-related genes in three independent cohorts of tumors using single-cell, bulk, and spatial transcriptomic data. Meta-analysis of single-cell transcriptomic data of 18 ccRCCs from two established cohorts (61, 62) was used to identify subpopulations of cancer cells. Cancer cells were identified by expression of the HIF target, carbonic anhydrase 9 (CA9; Supplementary Fig. S6A), and confirmed by inferred copy variant number analysis showing loss of chromosome 3p and gain of chromosome 5q (Supplementary Fig. S6B), the two most frequent genomic alterations in ccRCC (5). Clustering of cancer cells by their transcriptional similarity established 15 subpopulations present in stage I, stage III, and metastatic ccRCCs (Fig. 6A; Supplementary Fig. S6C; Supplementary Table S5A). The stability of cell subpopulations was assessed using repeated reclustering with randomly selected 80% of cells. Subpopulations 2, 6, and 10 were identified as the most stable with a median of max Jaccard index >0.6 (Supplementary Fig. S6D; ref. 63). Subpopulation 11, which was also relatively stable, was identified as most similar to proximal tubule epithelial cells and used as the starting point for pseudotime analysis using Monocle3. The pseudotime scores are indicated by a gray-to-purple color scale in which gray and purple correspond to early and late, respectively (Fig. 6B). The subpopulations 4, 0, and 1 were identified as early, whereas three stable populations 2, 6, and 10 were characterized as late. Based on the pseudotime scores, all cells were divided into early (lowest 33%), intermediate (middle 33%), and late (highest 33%) categories. Consistent with this classification, there was a clear association between these categories and tumor progression, with early scoring subpopulations associated with stage I, intermediate with stage III, and late with metastatic tumors, as indicated by χ² statistic residuals (Fig. 6C).

Next, we examined the metabolic and proliferative states of the identified cancer cell subpopulations using curated gene signatures (Supplementary Table S5B). A clear, stage-dependent global transcriptional reprogramming indicated a reduction of a glycolytic phenotype and enhancement of oxidative metabolism during the progression of ccRCC. The expression of glycolytic genes and HIF targets decreased in stage III and metastatic tumors compared with stage I tumors in several subpopulations of cancer cells (Fig. 6D; Supplementary Fig. S6E). In contrast, the expression of a superset of genes encoding all subunits of mitochondrial respiratory complexes (ETC/OxPhos) or individual complexes, genes associated with Cu allocation, and genes related to GSH metabolism were increased in more advanced tumors (Fig. 6D; Supplementary Fig. S6E). Notably, the NRF2 transcriptomic signature was similarly induced (Supplementary Fig. S6E), which is in line with the established role of this transcription factor in the regulation of GSH biosynthesis and ROS detoxification (43, 64) and with our cell line data (Fig. 1W; Supplementary Fig. S3F, S3G, and S3L). The expression levels of gene sets in metabolic pathways for ETC/OxPhos, Cu, and GSH were significantly correlated across all cell subpopulations (Fig. 6E). Interestingly, there was a lack of concordance in the expression of ETC/OxPhos and glycolytic genes in most subpopulations, except for the late subpopulation 6, in which both sets of genes were expressed together (Fig. 6E). Visualization of expression patterns for metabolic and other gene sets further revealed strong concordant expression of glycolytic-, ETC-, GSH-, and NRF2-related genes in subpopulation 6 (Fig. 6F). This implicates activation of full glucose oxidation accompanied by induction of GSH redox scavenging system. The early subpopulation 4 was also characterized by induction of metabolic genes, including ETC, OxPhos, TCA cycle, and fatty acid metabolism, but not glycolysis, implicating non–glucose-derived entry of electrons into the ETC (Fig. 6F). This is consistent with the primary role of fatty acid oxidation in healthy proximal tubule epithelial cells (65). Subpopulation 6 was transcriptionally similar to the cell culture model, 786-O Cu^Hi cells (Fig. 1X). Stable subpopulation 10 was noteworthy for higher cell cycle and DNA repair gene expression, indicating a proliferative phenotype (Fig. 6F). This subpopulation also showed increased expression of OxPhos-related genes (Fig. 6F).

The significance of the identified mitochondrial phenotype in the progression of ccRCC was further corroborated by the bulk transcriptomic analysis of stage III ccRCCs from the TCGA KIRC Firehose Legacy cohort. We focused on patients with known information about progression, i.e., patients who either remained disease-free (S3DF) or relapsed (S3RL) within 2 years after the resection of the primary tumor (Supplementary Fig. S7A). Note that tumors from patients with S3RL showed significantly higher levels of total Cu and CuCOX than those from patients with S3DF (Fig. 1H–J). We identified 1,267 DEGs, of which 340 and 927 were upregulated in S3RL and S3DF, respectively (Supplementary Fig. S7B; Supplementary Table S6). Pathway enrichment analysis identified ETC, mitochondrial respiration, and mitochondrial translation in tumors from S3RL. Immune response genes, including genes for MHC-II, were upregulated in tumors from patients with S3DF (Supplementary Fig. S7C). Notably, the mechanistic connection between altered ETC activity and expression of MHC-I genes was recently reported via epigenetic reprogramming (66), and the expression of MHC-II on renal cancer cells has some prognostic values (67). We constructed a 23-gene signature containing genes encoding a subset of ETC subunits, mitochondrial ribosomal proteins, and MHC-II that classified ccRCC for low, intermediate, and high risk of disease recurrence (Fig. 7A). Using the 23-gene signature and leave-one-out (tumor) cross-validation, we developed supervised machine learning models to predict relapse versus disease-free status from gene expression levels. These models (random forest and LASSO) achieved high classification accuracy, as indicated by the area under ROC curves (Fig. 7B), supporting the predictive solid power of the 23-gene signature. In contrast, the ClearCode34 signature that stratifies ccRCCs into cohorts with different prognoses (68) was not predictive (Supplementary Fig. S7D and S7E).

To reveal spatial organization and local heterogeneity of the cancer cell subpopulations in ccRCCs, we performed 10x Genomics Visium sequencing of five stage III ccRCCs from the University of Cincinnati Cancer Center Biospecimen Shared Resource. We identified nine distinct spots (spatial clusters) based on patterns of differential gene expression (Fig. 7C; Supplementary Table S7). There was an overall significant positive correlation between the expression of ETC/OxPhos-, Cu-, and GSH-related gene sets across all spatial clusters (Fig. 7D), which was particularly strong within spatial cluster 0 (Fig. 7E–G). Moreover, spatial cluster 0 showed significant enrichment for all metabolically dynamic single-cell subpopulations 4, 6, and 11 (Fig. 7H). Thus, metabolically active early and late subpopulations are localized in proximity to one another during tumor progression, suggesting a potential interaction or cooperation between these subpopulations that may contribute to tumor progression.

Mapping the spatial clusters onto the tumor sections showed that spots within the same cluster were often located next to each other, indicating clonal expansion (Fig. 7I; Supplementary Fig. S7F and S7G). Areas of tumors corresponding to spatial cluster 0 showed concordant expression of ETC/OxPhos and Cu-related genes (Fig. 7J), and ETC/OxPhos and GSH-related gene sets (Fig. 7K), corroborating scRNA-seq data. In contrast, spots from cluster 7 were scattered across sections.

To evaluate the adjacency between spatial cluster pairs in tumor sections, we quantified the nonrandom association of spots from each cluster pair by comparing observed versus shuffled spots. Notably, we observed significant adjacency only between spots from spatial clusters 0 and 7 (Fig. 7L and 7M). Interestingly, spatial cluster 7 shares a transcriptional signature with subpopulation 10 identified in scRNA-seq analysis, characterized by a proliferative phenotype (Fig. 7N). These findings suggest functional interactions between metabolically active and proliferating cells, which could impact tumor progression. In line with the transcriptional prediction, IHC staining of sections of orthotopic tumors invading kidney tissues revealed clustering of cancer cells expressing the mitochondrial respiration marker MT-CO2. The proliferating cells, identified by the Ki67 marker, were located near or within these MT-CO2-positive clusters. Interestingly, many Ki67-positive cells also co-expressed MT-CO2. These clusters, characterized by metabolic activity and cell proliferation, were observed in areas where tumor cells infiltrated adjacent kidney tissue in the orthotopic xenograft model (Supplementary Fig. S7H). In contrast, adjacency between spatial clusters 0 and 3 was reduced compared with reshuffled data (Fig. 7L).

Collectively, the data from three independent cohorts of ccRCCs and comprehensive analysis by single-cell, spatial, and bulk transcriptomics demonstrate the activation of mitochondrial OxPhos associated with Cu and GSH metabolism as a key metabolic process activated during ccRCC progression.

Cu-driven metabotranscriptional reprogramming in ccRCC promotes an oncogenic state enabling tumor growth and progression. As ccRCC evolves, blood and tumor tissues show increased Cu levels, facilitating a range of bioenergetic and biosynthetic advantages through allocation to CuCOX. However, to survive and proliferate in this Cu-enriched environment, cancer cells must adapt to counteract the toxic effects of Cu, particularly its potential to form ROS through Fenton-like reactions with H₂O₂. This adaptation is achieved through a tightly coordinated metabolic shift, in which glucose oxidation becomes coupled to GSH biosynthesis, enabling detoxification and maintaining cellular redox balance. The need for enhanced GSH metabolism during the progression of ccRCC has also been observed in a clinical category of ccRCC associated with poor prognosis, marked by an acute inflammatory response, metabolic reprogramming limiting lipid accumulation, and stimulation of proliferation rates, and activation of GSH metabolism (69).

Cu-mediated stimulation of glucose oxidation requires Cu-driven induction of CuCOX biogenesis and underscores the essential role of cytochrome c oxidase as a rheostat in mediating the Cu-regulated metabolic landscape. A possible explanation for this effect is that glucose oxidation is driven by the increased capability of ETC to transfer electrons to oxygen. However, given the very complex and selective mechanism by which glucose carbon is processed for GSH biosynthesis and the accompanying disinhibition of PDH, it is likely that CuCOX and/or Cu orchestrate events upstream of ETC that allow metabolic remodeling of the metabolism. For example, the binding of Cu to MT-CO1 and MT-CO2 may affect the availability of dynamic Cu to act as an allosteric regulator of mitochondrial proteins. The role of CuCOX as a metabolic determinant of cell fate, i.e., survival or death by apoptosis, was previously proposed (70). Unlike glucose metabolism, glutamine metabolism remains similar in Cu^Hi and Cu^Lo cells. Glutamine-derived GSH supports vital cellular processes, such as protein folding in the ER, but it does not have a distinct role under high Cu conditions.

Cu-regulated biochemical and functional compartmentalization of GSH production and activity is intriguing, and its mechanism will require further investigation. Glucose-derived GSH interfered with the accumulation of H₂O₂ in Cu^Hi cells. We propose that this occurs primarily in mitochondria. Interestingly, we did not detect increased expression of GSH-dependent enzymes known to decompose H₂O₂, and levels of GPX1/2 were substantially decreased in Cu^Hi cells. GSH may regulate the activity of a novel mechanism leading to the removal of H₂O₂ or may directly scavenge H₂O₂ (71). In vitro experiments demonstrated that CuCOX binds and can decompose H₂O₂ (72), whereas bacterial cytochrome c peroxidase uses H₂O₂ as the final acceptor of electrons under anoxic conditions (73). Similar activities of CuCOX in Cu^Hi cancer cells may regulate the accumulation of H₂O₂ in a GSH-dependent manner. Mitochondria may also use glucose-derived GSH to maintain the redox state of ETC assembly or glutathionylation of mitochondrial dehydrogenases and subunits of complex I, preventing the formation of superoxide and H₂O₂ while maintaining increased OCR (74–77).

Importantly, Cu leads to the recovery of PDH by inhibiting the phosphorylation of PDHA by PDK. As there was no significant change in PDK or PDH phosphatase protein expression, Cu effects may be caused by allosteric regulation. PDK is inhibited by pyruvate (78). Pyruvate-alanine transfer of cytosolic pyruvate to the mitochondria may result in higher mitochondrial concentration of pyruvate and inhibition of PDK activity. We did not detect any effects of Cu on PC involvement in glucose-derived GSH biosynthesis, although PC was essential for the biosynthesis of glucose-derived GSH in pancreatic islets (79).

The mechanism of alanine entry into the mitochondria must be further investigated. Two transporters are currently known to import small neutral amino acids into mitochondria and may be candidates: SLC25A38 and the members of the sideroflexin (SFXN) family. SLC25A38 is a glycine transporter required for heme synthesis (80). Its gene is located on chromosome 3p and is deleted in 16% of ccRCCs (TCGA KIRC, Firehose Legacy Cohort), consistent with loss of 3p as the most frequent genetic alteration. SFXN1 is a mitochondrial serine importer required for one-carbon metabolism and nucleotide biosynthesis, which has been shown to transport alanine, glycine, and cysteine (81). It is also critical for ETC integrity and activity (82). SFXN1 is located on chromosome 5q and is amplified, or its mRNA levels are increased in 29% of ccRCCs (TCGA KIRC, Firehose Legacy Cohort), consistent with 5q amplification being the second most frequent genetic alteration in ccRCCs.

We propose that cancer cells accumulate Cu during tumor evolution, which drives metabolic reprogramming in specific subpopulations. This reprogramming could allow certain metabolically adaptable cancer cells to shift into a proliferative state by enhancing their bioenergetic, biosynthetic, and redox functions due to genetic and epigenetic modifications driven by their altered metabolism. Alternatively, metabolically active cells may directly interact with subpopulations of scattered, proliferating cancer cells, supporting their rapid growth via released or consumed metabolites. Changes in ketoglutarate and succinate have intrinsic and extrinsic epigenetic effects (83, 84), whereas ROS affect multiple signaling pathways and genomic stability (85, 86). Recently, the role of Cu²⁺ as a direct metal catalyst activating H₂O₂-dependent oxidation of NADH to regenerate NAD was proposed to have metabolic consequences for epigenome and macrophage plasticity (87).

Cu-dependent reprogramming of ccRCC metabolism presents several clinically relevant vulnerabilities. Inhibition of ETC activity with IACS-10579 was partially effective but also toxic. One clinical trial investigated the antitumor effects of a Cu chelator, tetrathiomolybdate, in advanced ccRCC in a small cohort of patients (88). However, the results were inconclusive, and a direct assessment of Cu levels and speciation in both blood and tumor tissues, which we consider essential for appropriate patient stratification in this treatment, was not performed. Moreover, Cu chelation may be a double-edged sword. Cu chelation may paradoxically rescue cancer cells with high Cu content in the regions of tumors that are exposed to low glucose concentration, leading to resistance. The Cu-driven GSH phenotype of advanced ccRCCs may be susceptible to chemodynamic therapy, leveraging Fenton-like chemistry to convert intracellular H₂O₂ into toxic hydroxyl radicals. Cu is particularly suited to drive this reaction because its efficiency in driving Fenton chemistry is much faster than iron and is not affected by pH (89). Metal–organic framework nanoplatforms have been used to deliver Cu and to enhance levels of H₂O₂, whereas their intracellular dissociation utilizes and depletes GSH leading to cell death (90). This treatment induces a redox catastrophe like that caused by glucose starvation of Cu^Hi cells. Complementary strategies targeting subpopulations of cells with metabolic and proliferative states should be considered. Finally, measurements of Cu and CuCOX in primary tumors may serve as a simple prognostic biomarker.

Recently, a novel Cu-induced programmed cell death, cuproptosis, was proposed that involves aggregation of lipoylated proteins of the TCA cycle, followed by loss of iron–sulfur clusters and proteotoxic stress (20). We did not detect any changes in accumulation, lipoylation, or aggregation of lipoylated proteins to indicate cuproptosis in cell death caused by glucose starvation in Cu^Hi cells. This may be related to different biological contexts including chronic rather than acute Cu exposure. However, engaging cuproptosis machinery in ccRCC containing high Cu levels may be an alternative therapeutic approach.

The mechanisms leading to higher Cu accumulation during ccRCC advancement are under investigation. It is possible that exposures of patients to higher levels of environmental Cu, such as tobacco and e-cigarette smoking, result in higher uptake and accumulation in the tumors (11). However, it is also possible that cancer-related mechanisms at organismal and tumor levels participate in mobilizing Cu from its storage in the liver. Positive correlations between Cu levels in tumors and sera, and between total tumor Cu and CuCOX in advanced tumors support increased active uptake of Cu by tumor cells and its selective allocation to CuCOX. The roles of high-affinity Cu transporter, CTR1 (91), endocytosis, macropinocytosis, an actin-dependent process of nutrient uptake (92), and CD44 (87) are under investigation.

Collectively, we show how organometallic metabolism is deregulated in the progression of ccRCCs and how this metabolic plasticity supports the growth and survival advantage of specific clones associated with tumor progression.

Independent cohorts of ccRCCs, primarily from white males, were used for different measurements. Further validation on a demographically diverse single larger cohort in which all multiomic studies can analyze each tumor is needed to strengthen and extend the analysis for potential clinical application. A larger cohort of PDXs, with well-characterized Cu levels and allocation in the primary tumors, will be required for further validation and therapeutic explorations. Mechanistically, an understanding of the compartmentalization of GSH pools and events by which GSH prevents the accumulation of H₂O₂ will need to be further investigated. Finally, a fundamental mechanism by which increased Cu exposure leads to the assembly of CuCOX is unknown.

For Cu and metabolomic measurements and spatial transcriptomics, deidentified ccRCC specimens and associated clinical data from white male patients were obtained from the University of Cincinnati Cancer Center Biospecimens Shared Resource under the University of Cincinnati–approved IRB protocol 2013-4600. All samples were reviewed by pathologists and derived from regions with more than 80% cancer cells. Fresh-frozen tissues were used for metallomic and metabolic assays and FFPE tissues were used for spatial transcriptomics. For scRNA-seq, human data were obtained from two published data sets for 18 primary ccRCCs, labeled as cohort 1 (61) and cohort 2 (62). For the TCGA analysis, the KIRC Firehose Legacy cohort was limited to white males. Clinical fields from the KIRC dataset defined two groups of interest based on patient sex, race, tumor stage, and disease-free survival. “S3DF” consisted of tumors from stage III patients who remained disease-free for at least 24 months after surgery (n = 22). “S3RL” consisted of tumors from stage III patients whose disease recurred within 24 months (n = 20). Of the 20 patients in the latter group, four were removed after oncologist review of pathology reports (available on cBioPortal) because of non–clear-cell morphology.

All animal experiments were conducted in compliance with the University of Cincinnati IACUC-approved protocol 24-03-14-01. Five- to six-week-old Nod-SCID gamma (RRID: IMSR_JAX:005557) or athymic nude (RRID: IMSR_JAX:002019) male mice were used for all experiments. For orthotopic xenografts and PDX implants, mice were administered pre-emptive analgesics. Tumor cells (1 × 10⁶ in 35 μL volume of 50% Matrigel, Corning, #354234) were injected unilaterally under the kidney capsule from a retroperitoneal approach. The incisions were closed in two layers. For subcutaneous xenografts, mice were bilaterally injected with the same number of cells in 200 μL of 50% Matrigel. PDX model XP374d was obtained from the University of Texas Southwestern Medical Center Kidney Cancer Program and was previously described (25). Tumor fragments of approximately 2 mm were implanted subcutaneously into the flank. Mice were divided into two groups, one fed with Cu deficient diet (Envigo Teklad, TD.80388), and the other with either standard chow or matched custom-made high Cu diet (Envigo Teklad, TD.220421). In the case of subcutaneous tumors, the tumor dimensions were measured weekly using calipers, and tumor volume was calculated using the formula 1/2 × W × W × L. For IACS-10759 treatment, XP374d tumors were allowed to grow for approximately 7 weeks to reach an average volume of 200 to 400 mm³, after which mice were treated with 5-mg/kg IACS-10759 (Cayman Chemicals, 25867) suspended in 4% DMSO (Stemsol, PP1350)/0.5% methylcellulose (Sigma, M0262) or vehicle control by oral gavage every other day for 3 weeks.

Human ccRCC cell lines 786-O (RRID: CVCL_1051) and RCC4 (RRID: CVCL_0498) were cultured in DMEM/F12 (Cytiva, SH30023) with 10% fetal bovine serum (Gibco, 6000-044) at 37°C and 5% CO₂. Cell lines were regularly authenticated by STR profiling (LabCorp) and tested for mycoplasma (MycoAlert, Lonza, LT07-218). To establish a chronic Cu^Hi exposure model, cells were gradually adapted to high Cu conditions by incrementally increasing Cu concentration over 2 weeks, transitioning them to media containing 7.5-, 15-, 22.5-, and 30-μmol/L Cu (Supplementary Fig. S1O). Subsequently, cells were carried in DMEM/F12 with 10% FBS medium containing 30-μmol/L Cu. To create high Cu media that simulates Cu bound to blood proteins, 100% FBS was first incubated overnight with CuSO₄ at 4°C and then 10-fold diluted with the media. The binding of Cu to serum proteins and the final total concentration of Cu in the media was routinely validated by ICP-MS. Cells were used up to 15 passages from the switch to Cu^Hi media.

The total Cu levels and Cu speciation in lysed tumors or cell lines were analyzed by ICP-MS and SEC-ICP-MS as described in detail in ref. 11. The total Cu concentration was determined by the external calibration method in an Agilent 8900 ICP-MS/MS system (RRID: SCR_019460). For quality control analysis, a certified reference material (NIST SRM 1577c) was analyzed in parallel, with sulfur used to normalize the Cu content. The size exclusion separations were carried out in a TSKgel QC-PAK GFC 300 column (7.8 × 150 mm, 5 μm) in an Agilent 1260 HPLC equipped with a quaternary pump, an autosampler, and a diode arrange detector. This analysis was performed with a mobile phase composition of 50-mmol/L ammonium acetate in 0.5% methanol. The flow rate of the mobile phase was 0.65 mL/minute and the injection volume was 80 μL. The LC captured the full UV-Vis spectra from 220 to 750 nm, whereas the ICP-MS/MS system was operated in TRA mode, under oxygen reaction mode (at 2 mL/minute), with an integration time of 0.1 s including ⁶³Cu, ⁶⁵Cu, and ³²S → ⁴⁸SO isotopes. The quantification was performed by the external calibration method after integrating the peak areas of the Cu chromatograms of a mix of metalloprotein standards (GFS, Bio-Rad Laboratories, 1511901) and normalized by the total area of the respective chromatogram at 280 nm. Total Cu content represents the Cu concentration normalized to the total protein concentration. Cu levels in ovalbumin were measured against standards using total elemental analysis, which served as quality control to ensure the accuracy of the quantification method.

Labeled metabolites were purchased from Cambridge Isotope Laboratories. For glucose and glutamine tracing experiments, 786-O cells were plated in quadruplicate in 60-mm plates in DMEM/F12 (without glutamine and glucose; Biowest, L0091) supplemented with 10-mmol/L glucose, 2.5-mmol/L glutamine, and 10% dialyzed FBS (Gibco, 26400-044). After 24 hours, the media was changed to the same media supplemented with 5-mmol/L glucose (unlabeled or ¹³C₆ labeled, CLM-1396-1), 2.5-mmol/L glutamine (unlabeled or ¹³C₅,¹⁵N₂ labeled, CNLM-1275-H), and 10% dialyzed FBS and collected at the indicated time points. For the pyruvate tracing experiment, cells were changed to DMEM (Gibco, A1443001) supplemented with 0.05 mmol/L glucose, 4 mmol/L glutamine, 2 mmol/L 2,3-¹³C₂ pyruvate (CLM-3507), and 10% dialyzed FBS, with or without 10-μmol/L BSO for 16 hours. For alanine tracing, cells were changed to DMEM supplemented with 5 or 0.05 mmol/L glucose, 2.5 mmol/L glutamine, 1-mmol/L ¹³C₃,¹⁵N₁L-alanine (CNLM-534-H), and 10% dialyzed FBS for 12 hours. Media from each sample was collected and stored at −80°C for analysis. Cells on an additional plate were harvested and counted for normalization for each time point. The remaining cell plates were washed three times with 2 mL of cold PBS. After the final wash, PBS was aspirated, and plates were placed onto dry ice. Then, 0.9 mL of ice-cold 5:3:2 MeOH:MeCN:water (v/v/v) was added to each plate. The cells were scraped, and lysate was transferred to prechilled 1.5-mL microcentrifuge tubes. Metabolomics extractions and analyses were performed as previously described (93) at the University of Colorado Anschutz Medical Campus Cancer Center Mass Spectrometry Shared Resource Core Facility (RRID: SCR_021988). The extracts were vortexed for 30 minutes at 4°C, clarified via centrifugation (10 minutes, 15,000 x g, 4°C), and 500 μL of supernatant dried under vacuum. The residue was reconstituted in 5:3:2 MeOH:MeCN:water (v/v/v) to a normalized concentration of 2e⁶ cells/mL based on cell counts. Metabolite extracts were analyzed (10 μL per injection) using ultra-high-pressure liquid chromatography coupled to mass spectrometry (UHPLC-MS-Vanquish and Q Exactive, Thermo Fisher Scientific). Metabolites were resolved on a Phenomenex Kinetex C18 column (2.1 × 150 mm, 1.7 μm) at 45°C using a 5-minute gradient method in positive and negative ion modes (separate runs) over the scan range 65 to 975 m/z exactly as previously described (93). Following data acquisition, .raw files were converted to.mzXML using RawConverter, then metabolites were assigned and peaks were integrated using Maven (Princeton University) in conjunction with the KEGG database and an in-house standard library. ¹³C₂ isotopologue peak areas were corrected for the natural abundance of ¹³C₂ in the parent compound. Quality control was assessed using technical replicates run at each sequence’s beginning, end, and middle as previously described (94, 95). Fluxes of labeled carbon and nitrogen into the TCA cycle, nucleotides, and GSH are presented at points where the most significant enrichment was observed.

Lipids were extracted from purified mitochondria or xenograft samples (10 mg/sample) using a modified Bligh and Dyer method. Briefly, mitochondrial pellets or xenograft tissues were homogenized in water:methanol:dichloromethanane (DCM) 1:1:1 (v/v/v) and centrifuged at 2,671 x g for 5 minutes. The organic (upper layer) was collected, and the aqueous (bottom) layer was re-extracted by phase separation. The combined organic layers were dried using a SpeedVac Concentrator and resuspended in DCM:methanol:isopropanol (2:1:1, v/v/v) containing 8 mmol/L ammonium fluoride. SPLASH LipidoMix was used as an internal standard. A bicinchoninic acid protein assay was performed on the aqueous protein layer to normalize the lipid concentrations in each sample.

Reversed-phase chromatographic separation was performed with the Accucore C30 column: 3 μm, 2.1 × 150 mm (Thermo Fisher Scientific). The column was maintained at 35°C and the tray at 20°C. Solvent A was 10-mmol/L ammonium formate (LC-MS grade) in 60:40 acetonitrile (ACN):water (LC-MS grade) with 0.1% formic acid (FA, LC-MS grade). Solvent B was composed of 10-mmol/L ammonium formate with 90:10 IPA:ACN with 0.1% FA. The flow rate was 250 μL/minute, and the injection volume was 10 μL. The gradient was 50% solvent A (3%–50%). The Orbitrap (Thermo Fisher Scientific) mass spectrometer was operated separately for each sample under heated electrospray ionization in positive and negative modes. The spray voltage was 3.5 and 2.4 kV for positive and negative modes; the heated capillary was held at 35°C and the heater at 27°C. The sheath gas flow rate was 45 units, and the auxiliary gas was 8 units. A full scan (m/z 250–2,000) used a resolution of 30,000 at m/z 200 with an automatic gain control target of 2 × 10⁵ ions and a maximum ion injection time of 100 ms. Normalized collision energy settings were 25%, 30%, and 35%.

Lipid identification and relative quantification were performed with LipidSearch 4.1 (Thermo Fisher Scientific, RRID: SCR_023716; ref. 96), followed by Lipidsig (97). The search criteria were as follows: product search; parent m/z tolerance 5 ppm; product m/z tolerance 10 ppm; product ion intensity threshold 1%; filters: top rank, main isomer peak, FA priority; and quantification: m/z tolerance 5 ppm, retention time tolerance 1 minute. The following adducts were allowed in positive mode: +H, +NH_4, +H H₂O, +H 2H₂O, and +2H and in negative mode: −H, +HCOO, +CH₃COO, and −2H. CLs were identified as −H, −2H, and +HCOO adducts.

OCR was measured using a Seahorse XFe96 Analyzer (Agilent, S7800B, RRID: SCR_019545). Cells were changed to DMEM (Agilent, 103575-100) with 10 mmol/L glucose, 4 mmol/L glutamine, 0.5 mmol/L pyruvate or other indicated conditions for 48 hours after plating and then incubated in a 37°C non-CO₂ incubator for 1 hour prior to the assay in fresh media without serum. OCR was measured after each sequential injection of 1 μmol/L oligomycin A, 0.25 (786-O) or 2 μmol/L (RCC4) FCCP, and 0.5 μmol/L rotenone/0.5 μmol/L antimycin A/2.5 μmol/L Hoechst. To calculate basal (predrug injection), ATP coupled (drop post–oligomycin A injection), and maximal (post-FCCP injection) OCR rates, data were exported to a Seahorse XF Cell Mito Stress Test Report Generator using the Seahorse Wave Desktop Software (Agilent, RRID: SCR_024491). Hoechst fluorescent intensity was used to normalize the cell number.

Mitochondria isolation was performed using the human mitochondria isolation kit (Miltenyi, 130-094-532). Cells were pelleted and resuspended in 500-μL lysis buffer supplemented with 1× protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 1861281) and 10 μmol/L phenylmethylsulfonyl fluoride (Sigma, P7626). Cells were lysed by passing 13 times through a 27G needle. Equal amounts of lysates were diluted with 4.5 mL of 1× separation buffer and rocked at 4°C for 1 hour with 50 μL anti-TOM22 microbeads. LS columns were prewashed, loaded with samples, and washed 3 times with 3 mL separation buffer. Beads were eluted by removing columns from the magnet, plunging with 1 mL separation buffer, and centrifuged at 15,000 x g for 15 minutes. For mitochondrial-enriched fraction, cells were lysed as described above. Nuclei were pelleted at 800 × g for 5 minutes. Mitochondria were pelleted at 15,000 × g for 15 minutes, washed once with PBS, and lysed in RIPA buffer.

For nuclear soluble fraction, cell pellets were incubated for 5 minutes on ice in 10 mmol/L HEPES pH 7.5, 10 mmol/L KCl, 1.5 mmol/L MgCl₂, 1 mmol/L DTT, 10% glycerol, 0.1% Triton X-100, and 1× phosphatase inhibitor cocktail. Samples were centrifuged at 1,700 × g for 5 minutes, and the supernatant was taken as the cytoplasmic fraction. Nuclear pellets were extracted in buffer containing 10 mmol/L HEPES pH 7.5, 300 mmol/L NaCl, 1.5 mmol/L MgCl₂, 1 mmol/L DTT, and 10% glycerol at 4°C for 1 hour. Samples were centrifuged, and the supernatant was collected as the soluble nuclear fraction.

Samples were run on Bio-Rad Criterion XT 12% (345008) or 4% to 12% (3450124) Bis-Tris gels with 1× NuPAGE MOPS running buffer (Thermo Fisher Scientific, NP0001). Proteins were transferred to 0.2-μm polyvinylidene difluoride (PVDF) membranes. Blots were developed using Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, 32106), SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, 34095), or SuperSignal West Atto Ultimate Sensitivity Substrate (Thermo Fisher Scientific, A38555) and imaged on a Bio-Rad ChemiDoc Imaging System. Restore PLUS Western Blot Stripping Buffer (Thermo Fisher Scientific, 46430) was used to strip blots for reprobing. Blots or scanned gels were quantified using Image Studio Lite software (LI-COR, RRID: SCR_013715) or ImageJ (RRID: SCR_003070). The following antibodies were used: ATP5A (1:1,000, Abcam, ab14748, RRID: AB_301447); catalase (1:1,000, Cell Signaling Technology, 12980, RRID: AB_2798079); CD98 (1:1,000, Santa Cruz Biotechnology, sc-376815, RRID: AB_2938854); COX4I1 (1:1,000, Cell Signaling Technology, 4844, RRID: AB_2085427); COX4I2 (1:1,000, Proteintech, 11463-1-AP, RRID: AB_2085287); COX17 (1:1,000, My BioSource, mbs2526660, RRID: AB_3076222); COX7A2L (1:1,000, Proteintech, 11416-1-AP, RRID: AB_2245402); GAPDH (1:10,000, Abcam, ab8245, RRID: AB_2107448); GCLC (1:1,000, Novus Biologicals, H00002729-M01, RRID: AB_2294387); GCLM (1:1,000, Abcam, ab126704, RRID: AB_11127439); GPT/ALT1 (1:500, Proteintech, 16897-1-AP, RRID: AB_2230815); GPT2 (1:1,000, Santa Cruz Biotechnology, sc-398383, RRID: AB_2927429); GPX1/2 (1:1,000, Santa Cruz Biotechnology, sc-133160, RRID: sc-133160); KEAP1 (1:1,000, Cell Signaling Technology, 8047, RRID: AB_10860776); MT-CO1 (1:1,000, Abcam, ab203912, RRID: AB_2801537); MT-CO2 (1:1,000, Abbexa, abx125706, RRID: AB_3076223); NDUFS3 (1:1,000, Proteintech, 15066-1-AP, RRID: AB_2151109); NRF1 (1:1,000, Cell Signaling Technology, 46743, RRID: AB_2732888); NRF2 (1:1,000, Cell Signaling Technology, 12721, RRID: AB_2715528); PCB (1:1,000, Santa Cruz Biotechnology, sc-365673, RRID: AB_10842023); PDH-E1α (1:1,000, Santa Cruz Biotechnology, sc-377092, RRID: AB_2716767); PDK1 (1:1,000, Cell Signaling Technology, 3062, RRID: AB_2236832); PDP1/PPM2C (1:1,000, Proteintech, 21176-1-AP, RRID: AB_2878824); PGC-1α (1:1,000, Cell Signaling Technology, 2178, RRID: AB_823600); phospho-PDH E1 Alpha (pSer232; 1:1,000, Proteintech, 29582-1-AP, RRID: AB_2918327); phospho-PDH E1 Alpha (pSer300; 1:1,000, Proteintech, 80572-1-RR, RRID: AB_2918903); phospho-pyruvate dehydrogenase E1-alpha subunit (pSer293; 1:1,000, Novus, NB110-93479, RRID: AB_1237282); PRDX3 (1:1,000, Proteintech, 10664-1-AP, RRID: AB_2284207); PRDX6 (1:1,000, Proteintech, 13585-1-AP, RRID: AB_2168637); SDHA (1:1,000, Cell Signaling Technology, 11998, RRID: AB_2750900); SLC25A22 (1:1,000, Proteintech, 25402-1-AP, RRID: AB_2880060); SLC25A39 (1:1,000, Proteintech, 14963-1-AP, RRID: AB_2878095); SLC25A40 (1:1,000, Cell Signaling Technology, 38244, RRID: AB_3076224); SOD1 (1:1,000, Cell Signaling Technology, 2770, RRID: AB_2302392); SOD2 (1:1,000, Proteintech, 24127-1-AP, RRID: AB_2879437); TOM20 (F-10; 1:1,000, Santa Cruz Biotechnology, sc-17764, RRID: AB_628381); TOM22 (1:1,000, Cell Signaling Technology, 90704, RRID: AB_3076225); UQCRC1 (1:1,000, My BioSource, mbs9415267, RRID: AB_3076226); xCT/SLC7A11 (1:1,000, Cell Signaling Technology, 12691, RRID: AB_2687474); YY1 (H-10; 1:1,000, Santa Cruz Biotechnology, sc-7341, RRID: AB_2257497); goat anti-rabbit IgG HRP-linked antibody (1:5,000, Cell Signaling Technology, 7074, RRID: AB_2099233); and horse anti-mouse IgG HRP-linked antibody (1:5,000, Cell Signaling Technology, 7076, RRID: AB_330924).

Mitochondria were extracted with 1× sample buffer (NativePAGE Sample Prep Kit, Invitrogen, BN2008) containing 1% digitonin for 15 minutes on ice. Lysates were cleared by centrifuging at 20,000 x g for 20 minutes, and then, 0.25% G-250 Sample Additive was added to the supernatant. Samples were run on NativePAGE 3% to 12% Bis-Tris gels (Invitrogen, BN1001BOX) with NativeMark Unstained Protein Standard (Invitrogen, LC0725). Running buffers were prepared using the NativePAGE Running Buffer Kit (Invitrogen, BN2007); 1× running buffer was used as the anode buffer in a Thermo XCell SureLock Mini-Cell (EI0001). Gels were run at 150 V for 40 minutes with dark blue cathode buffer (1× cathode buffer additive in running buffer), followed by an additional 2 hours using light blue cathode buffer (0.1× cathode buffer additive). The gels were then transferred to 0.2-μm PVDF for 1.5 hours at 40 V in a transfer buffer containing 25.6-mmol/L Tris and 190.55-mmol/L glycine. After the transfer, the PVDF membranes were fixed in 8% acetic acid for 15 minutes, destained several times with methanol, and used for western blot analysis.

Mitochondrial lysates were prepared and subjected to BN-PAGE with the following modification: the gel was initially run for the first 40 minutes using a light blue cathode buffer, after which it was switched to a clear cathode buffer (without cathode buffer additive) for the remainder of the running time. Following the run, the gels were rinsed several times with water and incubated for 1 to 16 hours at 37°C in a solution containing 45 mmol/L phosphate buffer (pH 7.4), 0.5 mg/mL diaminobenzidine, and 1 mmol/L bovine cytochrome c (Sigma, C3131). After band development, the gels were rinsed with water and scanned.

Cytotoxicity was assessed through positive staining with propidium iodide and/or annexin V using flow cytometry with the FITC annexin V/Dead Cell Apoptosis Kit (Invitrogen, V13242). Cells were plated, and 24 hours (for RCC4) or 48 hours (for 786-O) later, the media was changed to DMEM supplemented with 10% FBS and the indicated reagents. The analysis was conducted 72 hours (for RCC4) or 24 hours (for 786-O) after the media change. Cells were trypsinized, pelleted, and resuspended in annexin binding buffer containing 1 μg/mL propidium iodide and 1:20 diluted FITC annexin V. After a 15-minute incubation, samples were diluted 1:8 with binding buffer. Fluorescence for 20,000 events per sample was measured using a 488 nm laser with a 505LP/530/30 emission filter (for FITC) and a 561-nm laser with a 600LP/610/20 emission filter (for PI) on a BD LSRFortessa (RRID: SCR_018655).

Drug sensitivity dose curves were generated using the CyQUANT NF Cell Proliferation Assay (Invitrogen, C35006) using 96-well plates. BSO, KCN, or KCN with aspartate were added 1:1 to the cell plate in six technical replicates. After 48 hours, 30 μL of 1× HBSS with 1:1,000 dye and 1:1,000 dye delivery reagent was added to each well. Plates were incubated for 30 minutes in a standard tissue culture incubator. Fluorescence was measured with excitation at 485 nm and emission at 530 nm on a BMG Labtech CLARIOstar. Fluorescence was normalized relative to the control samples.

RCC cell lines were transfected with siRNAs at final concentrations of 50 nmol/L using Lipofectamine 3000 according to the manufacturer’s protocol. Viral transductions included the addition of 2 μg/mL polybrene (Millipore, TR-1003-G) before viruses were administered. Lentiviral shRNA constructs were VSV-G envelope packaged, concentrated, and used at a 1:40 dilution. All control samples were treated with nontarget constructs. The following ON-TARGET plus siRNA (Dharmacon) were used: nontargeting control pool (D-001810-10-20), human GPT2 SMARTpool (L-004173-01-0005), and human SLC25A22 SMARTpool (L-007482-01-0005). The following shRNA constructs (Dharmacon) were used: pLKO nontargeting control, (RHS6848), COX4I1 (RHS3979-200800477), GPT2 shRNA 1 (RHS3979-201764341), SLC25A22 shRNA 1 (RHS3979-200798849), SLC25A22 shRNA 2 (RHS3979-200798847) SLC25A40 shRNA 1 (RHS3979-201778002), SLC25A40 shRNA 2 (RHS3979-201778003), SMARTvector Lentiviral nontargeting shRNA (E20230306C), and SMARTvector lentiviral SLC25A39 shRNA (V3SVHSHC-5016884). For COX4I1 knockdowns and xenograft experiments, cells were selected with and maintained in puromycin for stable expression of shRNAs. SLC25A22 shRNA 1, GPT2 shRNA 1, and SLC25A40 shRNA 1 were used for xenograft experiments.

GSH was measured using the GSH-Glo Assay (Promega, V6911) according to the manufacturer’s protocol. 786-O and RCC4 cells were plated in duplicates on a 96-well white-walled plate. Media was changed to DMEM + 10% FBS with indicated glucose or glutamine concentrations. For GSH measurements, cells were incubated in 50 μL of 1× GSH-Glo Reaction Buffer with 1:100 Luciferin substrate and 1:100 GSH-S-transferase for 30 minutes at room temperature. For total GSH measurements, 1-mmol/L DTT was added to the reaction. Next, 50 μL of reconstituted Luciferin Detection Reagent was added, and plates were incubated for another 15 minutes at room temperature. Luminescence was measured on a BMG Labtech CLARIOstar. Negative control values were subtracted from each measurement. A nonreduced (GSH) sample measurement was subtracted from a reduced sample (GSH + GSSG) measurement to calculate the GSSG levels. A matched plate was treated with the CyQUANT NF Proliferation assay reagent, and fluorescence was used to normalize GSH measurements to cell counts.

Cells were plated and treated as described in the cytotoxicity section. Cells were treated with 500-nmol/L MitoSOX Red (Invitrogen, M36008) and 100-nmol/L MitoTracker Green (Invitrogen, M7514) or 1.5 μmol/L CM-H2DCFDA (Invitrogen, C6827) for 30 minutes at 37°C and then resuspended in PBS + 1% BSA + 0.1 μg/mL DAPI (Invitrogen, D3571). The geometric mean of the fluorescent intensity was measured with a 488 nm laser and 505LP/530/30 emission filter (MitoTracker Green/CM-H2DCFDA), 405-nm laser and 600LP/610/20 emission filter (MitoSOX red), and 405-nm laser and 450/50 emission filter (DAPI) on a BD LSRFortessa for 20,000 events/sample. DAPI was used to gate for live cells. An unstained control was used to subtract background fluorescence.

Tumor sections were stained with hematoxylin and eosin (H&E) or analyzed via IHC performed at the Integrated Pathology Research Core (RRID: SCR_022637) at Cincinnati Children Hospital Medical Center according to standard protocols. The following antibodies were used: Ki-67 (Roche Ventana, 05278384001, RRID: AB_2631262 or Santa Cruz Biotechnology, sc-23900, RRID: AB_627859) and MT-CO2 (1:100, Novus, NBP3-16283, RRID: AB_3599404). This anti–MT-CO2 antibody is specific to human MT-CO2 and does not cross-react with mouse MT-CO2.

Cells plated on glass coverslips were fixed with 100% methanol at −20°C for 5 minutes. Cells were permeabilized with 0.1% saponin, blocked with PBS containing 0.1% saponin and 1% BSA for 30 minutes, and incubated with primary antibody for 1 hour at 37°C. Coverslips were washed and incubated with Alexa Fluor–labeled secondary antibodies for 30 minutes at room temperature. Finally, coverslips were washed and mounted using DAPI Fluoromount-G and analyzed by confocal microscope. The following antibodies were used: GPT2 (1:100, Proteintech, 16757-1-AP, RRID: AB_2112098), TOM20 (1:2,000, Santa Cruz Biotechnology, 17764, RRID: AB_628381), goat anti-mouse IgG Alexa Fluor 555 conjugate (1:1,000, Thermo Fisher Scientific, A-21422, RRID: AB_2535844), and goat anti-rabbit IgG Alexa Fluor 488 (1:1,000, Thermo Fisher Scientific, A-11008, RRID: AB_143165).

RNA was purified from cells using TriReagent (MRC, TR 118) according to manufacturer protocol, resuspended in nuclease-free water, and its concentration was determined with a NanoDrop. cDNA was synthesized using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4368814). The final cDNA was diluted to 8 ng/uL. qPCR was performed using 1× Fast SYBR Green Master mix (Applied Biosystems, 4385612), 400 nmol/L primers, and 20 ng cDNA using the Fast setting on an Applied Biosystems QuantStudio7 (RRID: SCR_020245). Analysis was performed using the ΔΔCt method using PP1A as the housekeeping gene and normalizing relative to Cu^Lo. The following primer sets synthesized by Integrated DNA Technologies were used: NQO1: GGATGAGACACCACTGTATTT and CTCCTCATCCTGTACCTT; HO1: GGGCCAGCAACAAAGTGCAAGATT and TCGCCACCAGAAAGCTGAGTGTAA; GCLC: TGCCCAGAGTTACTTGGATCAGCA and AGAGGCATGGTACTGTAGCCAGTT; GCLM: CTGCTGTGTGATGCCACCAGATTT and GTGCGCTTGAATGTCAGGAATGCT; MT1E: CCCTTTGCTCGAAATGGA and AACAGCAGCTCTTCTTGC; MT1X: TTTCCTCTTGATCGGGAACTC and GGCACAGGAGCCAACAG; SLC7A11: GGCATTTGGACGCTACAT and CACTACAGTTATGCCCACAG.

Data points represent biological replicates, and data are expressed as mean ± SEM. Significance was determined using a two-tailed t test, one-sample t test, paired t test, or one-way or two-way ANOVA with a Holm–Šidák post hoc test, as indicated using SigmaPlot v14/v15 (RRID: SCR_003210) or GraphPad Prism (RRID: SCR_002798). Graphs were created in GraphPad Prism.

RNA was extracted using RNAlater ICE (Ambion, AM7030) and miRNA isolation kit (Ambion, AM1560). The quality of RNA was checked using Bioanalyzer RNA 6000 Nano Kit (Agilent). PolyA RNA was extracted using NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB) and used as input for RNA-seq. RNA-seq was performed in the University of Cincinnati Genomics, Epigenomics and Sequencing Core. RNA-seq libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit (NEB). After library QC analysis using Bioanalyzer High Sensitivity DNA kit (Agilent) and library quantification using NEBNext Library Quant Kit (NEB), the sequencing was performed under the setting of single read 1 × 51 bp to generate ∼30 million reads per sample on HiSeq 1000 sequencer (Illumina).

Differential gene expression analysis was performed using edgeR (RRID: SCR_012802), comparing three Cu^Lo and Cu^Hi samples (Fig. 1V). Gene set enrichment analysis (GSEA, RRID: SCR_003199) was conducted using the clusterProfiler package (RRID: SCR_016884; ref. 98) in R, utilizing a curated library that integrates an in-house developed NRF2 signature with KEGG (RRID: SCR_012773) terms. The input-ranked gene list for GSEA was obtained by the edgeR-generated log₂ fold change in descending order. Subsequently, the dotplot and gseaplot functions from the clusterProfiler package were employed to generate the enrichment and GSEA plots. Next, the top 200 DEGs identified by edgeR were used to stratify the TCGA ccRCC cohort. Specifically, the TCGA_KIRC_RNASeqV2_2019 dataset (n = 309) from the ilincs portal (28) and the k-means (k = 2) algorithm were used to perform clustering analysis for both genes and tumor samples, limited to white males only. Consensus pathways enriched in 786-O Cu^Hi cells and in the subpopulation 6 identified in scRNA-seq were ranked using Fisher consensus of adjusted P values obtained from individual enrichment analyses (Fig. 1X).

scRNA-seq data for 18 ccRCCs were acquired from two published studies: Cohort 1 (61) and Cohort 2 (62). Cells were excluded if they had fewer than 1,000 detected UMIs, expressed fewer than 500 genes, or had mitochondrial gene content greater than 10%. Genes expressed in fewer than 20 cells were also discarded. Harmony (RRID: SCR_022206; ref. 99) integration algorithm was used with Seurat v4 (RRID: SCR_016341) standard preprocessing pipeline to integrate single-cell datasets from different tumors while removing confounding batch effects. Thirty principal components and the Louvain algorithm were used to perform dimensionality reduction and clustering in Seurat, using RunPCA, RunUMAP, FindNeighbors, and FindClusters functions. Cancer cells were identified using expression of CA9 and predicted loss of chromosome 3p. Copy number inference was conducted using the infercnv (RRID: SCR_021140). Proximal tubular cells identified in both cohorts via label transferring (100, 101) were used as the reference cells for infercnv analysis. Seurat generated UMAPs to visualize cancer cells. The expressions of gene sets were computed using the Seurat AddModuleScore function, employing 24 bins and 100 random genes as the background for each bin.

To assess cluster stability, a subsampling was performed by randomly selecting 80% of the cells from the original ccRCC single-cell datasets 100 times. The Jaccard index was computed as a measure of cluster stability by comparing clusters obtained from the original dataset with those obtained using 100 subsamples (63) and plotted using ggplot2 in R.

The residual contributions to the Pearson correlation coefficient (PCC) were used to assess patterns of correlations between gene sets expression levels across subpopulations. To define the PCC residuals, the covariance of the two gene sets was computed using cells from each subpopulation and divided by the product of their standard deviations calculated using all cells.

To characterize the tumor cell evolution landscape within the heterogeneous ccRCC cells, pseudotime trajectory analysis was performed using Monocle3 (v1.3.7; RRID: SCR_018685). The Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) embeddings from the original single-cell ccRCC Seurat analysis were used. The proximal tubule renal epithelial (PT) cells are the origin of ccRCC cells (102). Therefore, inferred cancer cells were merged with PT cells extracted from normal samples of ccRCC cohort 1 (61), and clustered using Seurat as described in the above section. Subpopulation 11 of inferred cancer cells overlapped with PT cells, and its resemblance to PT cells was validated by having the lowest expression of CA9 and a less clear pattern of inferred 3p chromosomal arm deletion among cancer subpopulations (Supplementary Fig. S6A and S6B). The centroid of cluster 11 comprising these PT-like cells retained initially in our analysis was used as the starting point for the Monocle3 trajectory analysis. The pseudotime scores were computed for each cell and mapped into UMAPs to highlight the predicted evolution of cancer subpopulations (Fig. 6B). The association between cancer stage (S1, S3, and Meta) and discretized pseudotime (early, intermediate, and late) was assessed by computing the χ² residuals for each pseudotime state and cancer stage, with the strength of association indicated by the size of circles, whereas positive and negative associations are shown in red and blue, respectively (Fig. 6C).

An in-house developed heatmap function (https://github.com/mjarek66git/ccRCC) was used to generate the heatmap in Fig. 6F. Seurat normalized and batch-corrected gene expression values for the top 3,000 most variable genes were quantile renormalized per cell, and the distributions were shifted by a constant to obtain all positive values. Cells with less than 500 counts and genes with less than 100 counts were excluded. KEGG and Hallmark and curated metabolic gene sets were used. The expression of the gene set was defined as the average of individual gene expression values. For visualization, gene set expression values across all cells were discretized, with the values in the top quartile (high expression) shown as yellow, the bottom quartile (low expression) as blue, and intermediate values as black. Gene sets (columns) were clustered using hierarchical clustering and 1-PCC as the dissimilarity measure, whereas cells (rows) were ordered using scRNA-seq cancer cell subpopulations from the scRNA-seq Seurat analysis. The order of cell clusters was defined using hierarchical clustering of cluster centroids and 1-PCC as the dissimilarity measure. Uninformative gene sets that did not show association with the Seurat cell clusters or did not cluster with at least one other supergene at the level of PCC = 0.5 (note the built-in redundancy between KEGG, Hallmark, and curated supergenes) were excluded from the heatmap.

TCGA Firehose Legacy Kidney Renal Clear-cell carcinoma data were used to analyze differential gene expression between S3DF and S3RL tumors. The “RNA_Seq_v2_expression_median” dataset downloaded from cBioPortal (RRID: SCR_014555), consisting of RSEM values for 20,531 genes across 534 samples, was transformed using log₂(RSEM + 1) and quantile renormalized column-wise (sample-wise). Z-score transformation was applied to gene expression rows to emphasize the relative expression value of a gene across samples. Differential gene expression between S3DF and S3RL tumors was assessed using the t.test package.

ComplexHeatmap (RRID: SCR_017270) generated a gene expression heatmap for 1,267 DEGs with t test P values ≤0.05. Samples were partitioned into two clusters (k = 2) using k-means. Hierarchical clustering was performed on rows and columns (within each k-means cluster) using Euclidean distance and average agglomeration. Fisher’s exact test was performed with fisher.test to determine the significance of sample group separation between k-means clusters, and other classes of samples defined by tumor grade, survival, and mutation type.

GSEA v4.1.0 software was used to assess pathway enrichment between groups, using the MSigDB (RRID: SCR_016863) c5.all.v7.2 collection, consisting of 14,765 ontology gene sets. Signed fold change values, scaled by the corresponding P values from the earlier differential expression analysis, i.e., −log₁₀(P value) × sign(log₂ fold change), were used to rank genes.

Of the 1,267 genes identified to be differentially expressed, ETC, mitochondrial ribosomal protein, and MHC class II (MHC-II) genes were combined to generate a 23-gene signature capable of stratifying the samples into distinct clusters. To assess the discriminatory power of such a defined signature, samples were partitioned into three slices (k = 3) using k-means, whereas hierarchical clustering was performed on columns (within each cluster) using Euclidean distance and average agglomeration. Fisher’s exact test was performed with fisher.test to determine the significance of sample group separation between clusters, and ComplexHeatmap was used to generate the heatmap.

S3DF versus S3RL class membership was predicted for each sample using random forest (RF) machine learning models trained and assessed using a one-sample-leave-one-out cross-validation framework and 23 gene expression values as input features. R function randomForest was called with default parameters for training, and the left-out sample’s class label was predicted using predict.randomForest based on the RF majority vote. For each of the 38 RFs, feature (gene) importance was assessed by a mean decrease in Gini index. Additionally, the proportion of trees voting S3RL was recorded for use in later ROC curve generation. In addition, penalized LASSO regression models were trained and assessed using the same cross-validation framework using cv.glmnet function.

The experiments were carried out in the Advanced Genomics Core at the University of Michigan according to Visium 10× protocol (10× Genomics). FFPE sections of tumor fragments were analyzed by the pathologist, and regions containing predominantly cancer tissues with RNA quality higher than DV200 50% were used as capture areas, H&E stained and examined for spatial transcriptomics. The results were processed by Space Ranger (RRID: SCR_025848) to acquire the expression matrix of each sample. Seurat was used for further processing and data normalization. As a quality control step, spots with fewer than 500 UMIs or more than 30% mitochondrial or ribosomal reads were filtered out.

To analyze the cancer cell regions, spatial spots from each slide were clustered using Seurat, and the pipeline was used to analyze scRNA-seq data. Small spatial clusters with very low total mRNA counts, likely corresponding to connective tissue, were removed. To further discard regions of slides with high stromal or immune cells, we applied the label transfer in Seurat (101) and the distinct cell type expression profiles identified in scRNA-seq analysis from a cohort 1(100). Finally, the selected regions were verified using CA9 expression.

Normalized gene expression profiles of cancer cells were integrated and clustered using Seurat. The integration features were selected from the 2,000 most variable genes using SelectIntegrationFeatures, and the integration was performed by using PrepSCTIntegration, FindIntegrationAnchors, and IntegrateData functions. The Louvain algorithm was applied to 30 principal components using FindNeighbors and FindClusters functions. To evaluate the optimum number of spatial clusters FindClusters function was performed with resolution values ranging from 0.3 to 1.2 at intervals of 0.05, and cluster quality was evaluated at each resolution. For each resolution value, the clustered cells were subsampled 100 times with 500 spots, and the silhouette score was computed for these 500 spots and their cluster labels; Pearson correlation was used as the distance metric in the computation of the silhouette score in the silhouette function. This procedure was repeated for 100 random samples of 500 spots to compute the average silhouette score’s mean and standard deviation at each resolution value. The highest resolution value that maximizes the mean silhouette score before a major reduction in the silhouette score was selected as the optimal resolution. Differential gene expression between clusters was computed by the MAST hurdle model for single-cell gene expression modeling, as implemented in the Seurat FindAllMarkers command, with a log fold change threshold of 0.25 and minimum fractional expression threshold of 0.25, indicating that the resulting gene markers for each cluster are restricted to those with a log fold change greater than 0.25 and nonzero expression in at least 25% of the cells in the cluster.

The module score of metabolic gene modules was calculated for each cancer spot using the Seurat AddModuleScore function with 24 bins and 100 control genes. The association of these module expressions with the spatial clusters was analyzed by correlating discretized values of module scores with spatial clusters. Module scores were discretized using three bins defined as the 1 quartile, between the 1 and 3 quartile, and the 4 quartile, thus defining three states with low, medium, and high expression of ETC/OxPhos, Cu, and GSH gene modules as shown in Fig. 7.

The module score of the top 100 DEGs in each single-cell subpopulation was calculated for the cancer cell spots in the tumor section using the Seurat v4 AddModuleScore function. The module scores were then standardized using the scale function in R. For each spatial cluster, the null hypothesis of the mean of zero for each normalized module score of single-cell clusters’ gene sets was tested using a one-sample Student’s t test using the t.test function in R. The P value and deviation from the mean of zero are reported for each spatial/single-cell cluster pair.

We generated normalized distributions of the number of adjacent spots from the same cluster for each cluster. To assess the nonrandomness of the distribution, spot labels were reshuffled while keeping the same number of spots in each cluster. The adjacency of spots from the same cluster was calculated as the difference in the distribution between the observed and shuffled data using the Kolmogorov–Smirnov test. To quantify adjacency between a reference cluster (e.g., cluster 0) and other clusters, the χ² test was calculated by comparing the number of reference cluster spots with zero versus more than one adjacent spot from other clusters using observed versus reshuffled spot labels for the other clusters.

The raw RNA-seq data from 786-O cell line were deposited into Gene Expression Omnibus with accession number GSE250028 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE250028); the raw sequence data from spatial transcriptomics were deposited into Gene Expression Omnibus with accession number GSE250163 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE250163). Codes were deposited in a public GitHub repository (https://github.com/mjarek66git/ccRCC). Any additional information reasonably needed to reanalyze the data is available from the lead contact.

B. Vemuri, J.A. Landero Figueroa, J.Meller, J.T. Cunningham, M.F. Czyzyk-Krzeska report a patent for US Patent App.17/327,100, 2022 pending. No disclosures were reported by the other authors.

M.E. Bischoff: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. B. Shamsaei: Conceptualization, formal analysis, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Yang: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. D. Secic: Formal analysis, investigation, visualization, methodology, writing–review and editing. B. Vemuri: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. J.A. Reisz: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. D’Alessandro: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. C. Bartolacci: Conceptualization, resources, investigation, visualization, methodology, writing–original draft, writing–review and editing. R. Adamczak: Conceptualization, investigation, visualization, methodology, writing–review and editing. L. Schmidt: Investigation, visualization, methodology, writing–review and editing. J. Wang: Investigation, visualization, methodology, writing–review and editing. A. Martines: Investigation, visualization, methodology. J. Venkat: Investigation, visualization. V.T. Tcheuyap: Investigation. J. Biesiada: Investigation, visualization, methodology, writing–review and editing. C.A. Behrmann: Investigation, methodology. K.E. Vest: Conceptualization, resources, supervision, writing–original draft, writing–review and editing. J. Brugarolas: Conceptualization, resources, investigation, methodology, writing–review and editing. P.P. Scaglioni: Conceptualization, resources, supervision, funding acquisition, writing–review and editing. D.R. Plas: Conceptualization, Supervision, funding acquisition, writing–original draft, writing–review and editing. K.C. Patra: Conceptualization, supervision, writing–original draft, writing–review and editing. S. Gulati: Conceptualization, resources, funding acquisition, writing–original draft, writing–review and editing. J.A. Landero Figueroa: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J. Meller: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. J.T. Cunningham: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. M.F. Czyzyk-Krzeska: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The following grants supported the work: M.F. Czyzyk-Krzeska (R01GM128216, R01CA287260, and 2I01BX001110 BLR&D VA Merit Award); J.T. Cunningham (R01CA230904 and R35 GM133561); D.R. Plas (R01CA239657); P.P. Scaglioni (R01CA259845 and LCS Foundation); K.E. Vest (R35GM146878); K.C. Patra (R37CA272854); S. Gulati (K08CA273542); M.E. Bischoff (T32CA17846); B. Shamsaei (T32CA236764); D. Secic (T32ES007250); and J. Meller (University of Cincinnati Cancer Center Pilot Grant). J.A. Reisz and A. D’Alessandro, on behalf of the University of Colorado SOM Metabolomics Core, acknowledge support from the University of Colorado Cancer Center via NCI P30CA046934 and technical contributions of Rachel Culp-Hill, Shoun Bevers, and Abby Grier. J. Brugarolas, V.T. Tcheuyap, and the PDX platform were supported by P50CA196516. We thank Drew Smith and Betsy DiPasquale in The Integrated Pathology Research Core at Cincinnati Children Hospital Medical Center for histopathology stainings; Kelsey Dillehay McKillip and Farah Sagin in the University of Colorado Cancer Center Biospecimen Shared Resources for the procurement of specimens; Birgit Ehmer for preparing immunofluorescent images; and Rose Bacon, Lindsey Huether, and Addison Cooper (Cooper Graphics) for preparing the figures.