AADAT-Driven Metabolic Control of Malate and CoQ Shapes Immune Evasion in Triple-Negative Breast Cancer.

Introduction
Triple-negative breast cancer (TNBC) is a heterogeneous disease with poor prognosis due to a high propensity for early metastasis and recurrence following standard-of-care chemotherapy1. The presence of tumor-infiltrating lymphocytes (TILs), particularly cytotoxic CD8+ T cells, has been strongly associated with improved response to neoadjuvant chemotherapy and favorable survival outcomes in TNBC patients2–4. These findings have provided a rationale for the use of immune checkpoint blockade (ICB), including therapies targeting CTLA-4 and the PD-1/PD-L1 axis, to enhance anti-tumor immune responses in TNBC5. While the addition of ICB to chemotherapy has shown clinical benefit in a subset of patients—most notably in the KEYNOTE-522 trial, where pembrolizumab improved pathologic complete response (pCR) from 50% to 65% and increased overall survival from 81.7% to 86.6%—many patients remain unresponsive or develop resistance to treatment6–9. Moreover, the high incidence of immune-related adverse events10,11 underscores the need for more targeted, tolerable strategies to sensitize tumors to ICB.
A significant obstacle to optimizing immunotherapy is the immunosuppressive tumor microenvironment (TME), shaped in part by tumor-intrinsic metabolic programs12. In TNBC, the metabolic crosstalk between tumor and immune cells profoundly influences antigen presentation, T cell infiltration, and effector function. For example, hypoxia-driven metabolic changes in the TME can reduce MHC expression13, suppress interferon-gamma (IFN-γ) production14, upregulate PD-L113, and enhance aerobic glycolysis13—all of which facilitate immune escape. Beyond hypoxia, many metabolic intermediates directly modulate immune cell differentiation and function15–18, and reprogramming these axes has emerged as a promising avenue to overcome therapeutic resistance.
Tryptophan catabolism through the kynurenine pathway (KP) illustrates a mechanism of immune evasion where the depletion of tryptophan and the buildup of immunosuppressive metabolites suppress cytotoxic T cells and promote regulatory T cells. The KP’s first enzymatic step, catalyzed by indoleamine 2,3-dioxygenase (IDO1) and tryptophan 2,3-dioxygenase (TDO2), has been associated with poor prognosis and immune suppression in breast tumors19–21. While most studies have focused on IDO1 and TDO2, recent evidence suggests that downstream KP enzymes, including kynureninase and kynurenine-3-monooxygenase, may also contribute to TNBC progression22,23. However, the immunoregulatory potential of other KP enzymes remains poorly defined.
In this context, we identified the mitochondrial enzyme alpha-aminoadipate aminotransferase (AADAT), also known as kynurenine aminotransferase, as a previously uncharacterized immunosuppressive regulator in TNBC. AADAT catalyzes the conversion of kynurenine to kynurenic acid24 and, in our study, was found to be markedly upregulated in TNBC tumors. Beyond its canonical aminotransferase activity, AADAT also functions as a mitochondrial tyrosine transaminase25 that generates 4-hydroxyphenylpyruvate, an obligate precursor of 4-hydroxybenzoate and thus the benzoquinone headgroup of coenzyme Q10 (CoQ10), coupling KP flux to mitochondrial CoQ10 biosynthesis26. In mammalian cancer cells, stress-induced up-regulation of CoQ10 biosynthetic genes enhances oxidative phosphorylation and antioxidant capacity. At the same time, pharmacologic blockade of CoQ10 synthesis increases chemotherapy-induced cytotoxicity27.
Our research reveals a new link between AADAT and anti-tumor immunity in TNBC. Specifically, AADAT suppresses CD8+ T cell activity within the tumor microenvironment by lowering secreted malate levels and preserving CoQ10-dependent redox balance in the tumors. Increasing malate in the tumor microenvironment, either by blocking AADAT or by supplementing the metabolite, reprograms CD8+ T cell metabolism and function. This restores response to immune checkpoint inhibitors and chemotherapy in de novo or acquired immunotherapy-resistant TNBC and increases patient survival without added toxicity. The findings highlight malate as an essential immunoregulatory metabolite and suggest that metabolic reprogramming could be a valuable strategy to overcome immune resistance in aggressive breast cancers.

Results

AADAT Expression Correlates with Reduced CD8+ TILs and Poor Prognosis in TNBC
To investigate the role of the kynurenine pathway (KP, Fig. 1a) in immune escape, we examined publicly available breast cancer patient-derived microarray and IHC data on KP gene expression and CD8+ TILs in breast tumors27. As expected from an earlier study (28), higher levels of KP enzymes such as IDO1, IDO2, and HAAO were linked to increased CD8+ TILs (Figs. S1a–c). Notably, only Alpha-Amino Adipate Amino Transferase (AADAT) expression was uniquely associated with reduced CD8+ TIL infiltration (Fig. 1b, c).
Tissue Microarray (TMA) analysis revealed that AADAT was elevated in malignant breast tumors relative to adjacent benign tissue, without differences between localized and metastatic disease (Fig. 1d). TCGA analysis confirmed that AADAT expression was increased in the more aggressive HER2-like and TNBC subtypes compared to Luminal and Normal tumors (Fig. 1e), findings corroborated in the METABRIC dataset28, where AADAT was enriched in HER2-enriched and Basal-like subtypes (Fig. 1f). Analysis of multiple publicly available gene expression datasets in The University of Alabama at Birmingham Cancer data analysis portal (UALCAN)29 further demonstrated that AADAT expression is significantly higher in ER-negative or TNBC tumors compared to ER-positive tumors or adjacent benign tissues (Figs. S1 d-j). AADAT protein expression was also elevated in TNBC tumors compared to luminal and HER2+ breast cancers in the CPTAC dataset (Fig. S1k). Across datasets, higher AADAT expression predicted worse clinical outcomes in breast cancer and within TNBC (Figs. 1g–i). In TNBC samples, AADAT expression negatively correlated with predicted T cell infiltration using TIMER (Tumor Immune Estimation Resource30) analysis (Fig. 1j), a finding confirmed by IHC on TNBC TMAs showing fewer CD3+ and CD8+ TILs in AADAT-high tumors (Figs. 1k, S2a) and validated using a spatial gene expression dataset31 (Fig. 1l). Gene set enrichment analysis on NanoString data32 revealed enrichment of cytotoxic T cell–associated signatures, including allograft rejection and IFN-γ response, in TNBC samples with low AADAT expression (Figs. S2 b-d), supporting an inverse relationship between AADAT and CD8+ TILs.

AADAT Promotes Tumor Growth and Suppresses Immune Infiltration in TNBC
To determine whether AADAT directly influences immune exclusion, we knocked down (KD) AADAT in E0771 and 4T1 cells using two independent shRNAs (shAADAT-1 and shAADAT-2) delivered by a lentiviral vector, along with reconstituting AADAT expression for rescue (shAADAT-2+AADAT). We confirmed KD and rescue through qPCR, peptide reaction monitoring (PRM) proteomics, and a functional assessment of the KYNA: KYN ratio (Figs. 2a, S3a, b). Remarkably, AADAT KD significantly reduced tumor initiation and delayed tumor growth following orthotopic injection of both E0771 and 4T1 cells into the mammary fat pad of syngeneic wild-type C57BL/6J and BALB/cJ female mice, respectively, which was rescued by re-expression of AADAT (Figs. 2b, c, and S3c, d). In CD8 KO mice, AADAT KD did not delay tumor initiation, indicating that CD8+ T cells are required for the observed delay (Figs. S3e, f). However, once established, AADAT KD tumors grew more slowly in CD8 KO mice (Figs. 2d, e), suggesting the presence of additional CD8+ T cell -independent mechanisms. Flow cytometry analysis of E0771 tumors revealed increased infiltration of immune cells, including monocytes, macrophages, B cells, and CD4+ T cells after AADAT KD (Figs. S3 g-j), indicating that AADAT impacts the immune landscape in TNBC. To study metastasis, we created an inducible AADAT KD in E0771 cell line (Fig. 2f). Spontaneous pulmonary metastasis, assessed one month following resection of size-matched primary tumors derived from the above-mentioned cells, was significantly decreased in mice with induced AADAT KD tumors compared to mice with control tumors that were not given doxycycline (Fig. 2g).

AADAT Inhibition Sensitizes TNBC to Immune Checkpoint Blockade
To determine whether the immunomodulatory role of AADAT extends to clinical outcomes, we evaluated its association with response to immune checkpoint therapy. Analysis of a melanoma dataset (GSE91061) showed that low AADAT expression in patients was significantly associated with better immunotherapy response, indicated by a higher pathological response in patients with low versus high AADAT tumors (58% vs 33%, Fig. 3a). Based on this, we tested whether AADAT depletion could improve immunotherapy effectiveness in orthotopic TNBC mouse models. Inducible AADAT KD, combined with anti–PD–1 and anti–CTLA–4, significantly reduced tumor growth compared to monotherapy in the E0771 and 4T1 models (Figs. 3b–d). Analysis of AADAT KD and control 4T1 tumors, shown in Fig. 3c, using imaging mass cytometry (IMC) with a panel of 14 antibodies (Supplementary Table 1) targeting murine immune and tumor cells (Fig. S4), revealed significantly increased interactions (see Methods for details) between tumor cells and CD8+ T cells in AADAT KD compared to control tumors (Fig. 3e). This suggests that targeting AADAT enhances immune-mediated tumor clearance.

AADAT reduces the buildup and release of immunostimulatory malate
Given that AADAT depletion enhances antitumor immunity, we next explored the metabolic basis for this phenotype by profiling the secretome of AADAT-deficient TNBC cells. Unbiased metabolomics revealed increased malate levels in conditioned media (CM) after AADAT KD in E0771-ova+ cells (Fig. 4a). To determine the clinical relevance of this finding, we examined the relationship between intra-tumoral malate levels, immune infiltration, and clinical outcomes in TNBC patients.
Using high-resolution imaging mass spectrometry (IMS) and IMC on TNBC TMAs (n=31, Supplementary Table 2 for clinical data), we quantified malate and immune cell markers (Supplementary Table 3) in their spatial context. We then aligned IMS malate spots with IMC spots, yielding 5,245 matched pairs. We analyzed the correlation between IMS malate spots and IMC spots for both functional (CD3, CD4, CD8a, CD45, CD45RO) and exhausted T cell markers (CD8a, Ki67, PD-L1, PD-1, CD152, VEGF) across 31 TNBC samples. Figures 4b and 4c show a very similar spatial pattern between functional T cell markers and malate. Notably, this spatial similarity is not seen with fumarate, phenylalanine, leucine, citrulline, and methyl-histidine (Fig. 4d), all of which were also elevated in the conditioned media (CM) following AADAT KD in E0771-ova+ cells (Fig. 4a). Furthermore, analysis revealed a positive correlation between malate levels and the IMC cluster of functional T cells, and a negative correlation with the cluster of exhausted T cells (Figs. 4e, f). However, no significant correlation was observed between the remaining metabolites mentioned above, except methyl histidine, and either the functional T-cell or the exhausted T-cell clusters (Fig. S5). Interestingly, TNBC patients with a higher number of intra-tumoral malate-enriched IMS spots had better 10-year survival after standard chemotherapy compared to those with fewer such spots (Fig. 4g). Additionally, Cox-Proportional Hazard analysis indicated a hazard ratio of 0.835 (P<0.05) for co-localized malate levels with functional T cells, after adjusting for age and cancer stage (Supplementary Table 4). These data suggest that malate serves as a biomarker of a “hot” immune microenvironment and may functionally enhance anti-tumor immunity in TNBC.
We next confirmed elevated levels of secreted malate in murine E0771-ova+ and 4T1 cells by targeted metabolomics, which either had inducible KDs of AADAT (Figs. S6a, b-AADAT expression levels and Figs. 4 h- malate levels) or a constitutive KD of AADAT (Fig. 2a
bottom panel- AADAT expression levels and Fig. 4i- malate levels). Additionally, it was validated in human MDA-MB-231 TNBC cells with separate shRNA-mediated constitutive KDs of AADAT (Fig. S6c- AADAT expression levels and Figs. S6d, e malate levels). Significantly, re-expressing AADAT in 4T1 cells with a constitutive KD reversed malate elevation (Figs. 2a
bottom panel and 4i), indicating AADAT’s direct role in regulating malate secretion in TNBC cells.

AADAT regulates malate secretion by regulating CoQ10 redox homeostasis
Pathway analysis of the CM metabolome in E0771-ova+ cells with inducible AADAT KD, compared with uninduced control cells, revealed changes in amino acid metabolism, the TCA cycle, and CoA biosynthesis (Fig. 5a), suggesting mitochondrial stress. Consistent with this, E0771-ova+ cells with inducible AADAT KD and E0771 cells with constitutive AADAT KD showed significantly lower extracellular acidification rate (ECAR, Figs. S6f-h), oxygen consumption rate (OCR, Figs. 5b,c, S6i), and intracellular malate-to-oxaloacetate ratio (Fig. 5d, a surrogate for mitochondrial redox capacity33 or NADH: NAD+ ratio, not seen in Dox control shown in Fig. S6j) compared to control cells. Moreover, unbiased metabolomic profiles of CM from AADAT-KD cells also revealed elevated malate levels, along with amino acids, including tyrosine, and decreased levels of TCA intermediates such as α-ketoglutarate (α-KG) and succinate, as well as the tyrosine breakdown product, 4-hydroxyphenylpyruvate, a necessary precursor for CoQ10 synthesis (Fig. 4a). The pattern of decreased succinate and α-KG, along with reduced levels of the CoQ10 precursor 4-hydroxyphenylpyruvate, and decreased OCR seen with AADAT KD cells (Figs. 5b, c), suggests impaired mitochondrial oxidative metabolism, indicating TCA cycle stalling, disrupted electron transport, and redox imbalance. It also suggests that AADAT regulates CoQ10 levels, and its loss creates a redox bottleneck, due to reductive stress (Fig. 5d). These findings indicate a synergism between reversal of normal α-KG/malate antiporter activity due to high intracellular malate and elevated NADH/NAD+ levels caused by reduced CoQ10, promoting mitochondrial malate production via malate dehydrogenase 2 (MDH2), which leads to increased malate secretion. High levels of secreted malate create an immunostimulatory metabolic signal in the TME.
To confirm the link between CoQ10 levels and malate secretion in AADAT KD cells, we added CoQ10 to E0771-ova+ cells with inducible AADAT KD. As expected, CoQ10 supplementation reversed the increased intracellular and secreted malate levels observed with AADAT KD (Figs. 5e, f, S6k).
We then investigated the impact of malate on TNBC tumor growth using C57BL/6J female mice orthotopically injected with E0771 cells. Malate supplementation in drinking water significantly reduced tumor growth in treated mice compared with untreated controls (Fig. 5g), demonstrating its tumor-reducing effects in vivo.

Malate Recapitulates AADAT KD Effects on T Cell Activation
Based on our previous findings, we performed co-culture experiments to test the ability of activated CD8+ T cells isolated from OT-1 mice to kill murine TNBC cells (see methods section). The CD8+ T cells were then co-cultured with E0771-ova+ cells with or without inducible KD of AADAT (Fig. 6a). Additionally, we included a co-culture with uninduced E0771-ova+ and CD8+ T cells where the media was supplemented with 2.5 mM malate (Fig. 6a). Interestingly, CD8+ T cells showed significantly increased killing of E0771-ova+ cells with AADAT KD compared to uninduced controls (Fig. 6b). Furthermore, supplementation of malate also resulted in a similar level of CD8+ T cell killing on the E0771-ova+ control cells (Fig. 6b). Malate supplementation did not affect either tumor cells or T cells when cultured separately (Figs. S7a-c). Additionally, an unrelated metabolite like tryptophan did not alter the ability of CD8+ T cells to kill E0771-ova+ cells in a similar co-culture setup (Fig. 7d). Along the same lines, treatment with malate, but not fumarate, significantly enhanced the killing of patient-derived TNBC organoids (with baseline HER-2 levels) by HER-2-directed CAR-T cells, but not by non-targeting CAR-T cells (Fig. S8).
To gain mechanistic insights, we performed bulk RNA-seq analysis of CD8+ T cells obtained from co-cultures described in Fig. 6a. As shown in Fig. 6c, CD8+ T cells exposed to either AADAT KD cells or malate-supplemented control cells displayed significantly altered, overlapping gene expression signatures. Gene Set Enrichment Analysis (GSEA) of this common set of altered genes revealed enrichment for inflammatory pathways, including those describing TNF-α signaling and the IFN-γ response (Figs. 6d). Consistent with this, elevated intracellular levels of both TNF-α and IFN-γ were confirmed by flow cytometry in CD8+ T cells treated with malate (Figs. 6e, f), suggesting a possible mechanism for malate-induced activation of CD8+ T cells.
To further assess whether malate-induced activation of inflammatory pathways in CD8+ T cells results from metabolic rewiring, we examined the metabolome of malate-treated CD8+ T cells. Compared to untreated CD8+ T cells, malate-treated T cells exhibited significantly increased glycolytic and TCA cycle metabolites, a higher NAD+: NADH ratio, and elevated levels of ROS (Figs. 6g–i), all of which indicate improved metabolic fitness that could enhance CD8+ T-cell effector function. Collectively, these data demonstrate that malate supplementation mimics AADAT KD by promoting CD8+ T cell effector function through enhanced metabolic fitness and activation of the inflammatory pathway, thereby driving anti-tumor immunity.

Malate Enhances the Efficacy of Standard-of-Care Immunochemotherapy in TNBC Models
Given the association between lower AADAT expression and improved immunotherapy response in patient data, and our discovery that malate’s immunostimulatory effects enhance immunotherapy efficacy when AADAT is knocked down, we investigated whether malate supplementation could overcome resistance in TNBC tumors unresponsive to standard immunochemotherapy. To examine this, 5% malate was administered daily in drinking water for 12 days to BALB/cJ female mice bearing orthotopic de novo immunotherapy resistant 4T1 TNBC tumors, in combination with anti–PD-1 (200μg, intraperitoneal injection) and PTX (10 mg/kg, intra-orbital injection), given for six cycles every 3 days (Fig. 7a). On day 12 after tumor implantation, malate supplementation was lowered to 1% in both the malate-only group and the group receiving malate combined with chemo-immunotherapy until the study concluded (Fig. 7a). Remarkably, combining malate with anti–PD-1 and PTX significantly decreased tumor volume and increased survival compared to vehicle, or anti–PD-1 plus PTX treatment groups. (Figs. 7b, c).
Additionally, we conducted similar experiments with E0771 and 2208L cells that were made resistant to PTX and anti-PD1 in vitro, referred to as E0771-Res1 and 2208-L Res2, respectively34. These cells were orthotopically injected into syngeneic wild-type BALB/cJ mice, which were treated with a similar regimen (Fig. 7d). However, malate levels after Day 12 were alternated weekly between 1% and 5% until the study concluded, as these cells were inherently resistant to chemo-immunotherapy agents (Fig. 7d). Importantly, like the results in the 4T1 model, the combination of malate with anti-PD1 and PTX resulted in significantly reduced tumor volumes in both these resistant models (Figs. 7e, f), and increased survival (Fig. 7g) compared to vehicle-treated or chemo-immunotherapy-treated controls. Furthermore, histopathological examination revealed that animals with 2208L-Res2 tumors treated with a combination of malate, anti-PD1, and paclitaxel exhibited no signs of toxicity in the liver or lungs. (Fig. S9).
Overall, these results indicate that malate reverses immunotherapy resistance, enhances both local tumor control and systemic anti-tumor immunity, while improving survival.
These findings collectively show that specific TNBC subgroups with high AADAT expression maintain a CoQ10-supported redox imbalance, which prevents the buildup and export of malate. Lower malate levels in the tumor microenvironment (TME) reprogram CD8+ T cell metabolism, leading to an exhausted phenotype (Fig. 8). Conversely, adding malate to TNBC tumors improves the effectiveness of standard immune and chemotherapy by enhancing T cell activation and infiltration, reducing tumor growth, increasing survival, and providing a clinically feasible approach to improve patient outcomes in TNBC (Fig. 8).

Discussion
The aggressive nature, high recurrence rates, and limited actionable targets of TNBC continue to pose significant challenges despite advancements in immunochemotherapy, underscoring an urgent need for novel therapeutic strategies35. While immune checkpoint inhibitors (ICIs) like pembrolizumab, in combination with chemotherapy, have demonstrated survival benefits, a substantial proportion of patients still experience disease progression or recurrence36. This report highlights a novel immune-metabolic axis involving AADAT and malate, which significantly impacts the tumor immune microenvironment (TIME) and therapeutic response in TNBC. Our findings position AADAT as a critical driver of immune exclusion and identify malate as a potent, clinically translatable immunomodulatory adjuvant, offering a promising strategy to overcome resistance and improve outcomes in this challenging disease.
The kynurenine pathway (KP) of tryptophan metabolism is a well-established regulator of immune responses, frequently co-opted by tumors to promote immune evasion37. Canonical KP enzymes such as indoleamine-2,3-dioxygenase 1 (IDO1) and IDO2 are known to induce immune resistance by depleting tryptophan and generating immunosuppressive kynurenines, often correlating with an influx of T cells that are subsequently rendered dysfunctional or exhausted within the immunosuppressive TME38. This study unveils AADAT as a distinct and previously unappreciated player within the KP, exhibiting a unique inverse correlation with CD8+ T cell infiltration and poor prognosis in breast cancer, particularly TNBC. This stands in stark contrast to the positive correlation observed between IDO1, IDO2, and HAAO expression and CD8+ tumor-infiltrating lymphocytes (TILs). The established understanding of IDO1 and IDO2 suggests that they suppress the immune response by depleting tryptophan and producing immunosuppressive kynurenines, often correlating with an increase in TILs, which is interpreted as a compensatory immune response or an active, albeit suppressed, immune microenvironment39.
AADAT’s inverse correlation, which is uniquely linked to decreased CD8+ TIL infiltration, indicates a different mechanism from KP. This mechanism involves reprogramming tumor mitochondria, which subsequently leads to downstream events resulting in T cell exhaustion. This deviation from the canonical KP immunosuppression model represents a significant novelty for understanding immune evasion and designing targeted therapeutic interventions. The clinical relevance of AADAT in TNBC is further underscored by its elevated expression in malignant breast tumors, particularly in aggressive HER2-like and TNBC subtypes, and its consistent prediction of worse clinical outcomes across multiple patient datasets40. Given TNBC’s reliance on metabolic reprogramming and its aggressive nature with limited therapeutic targets, identifying AADAT as a key metabolic enzyme that promotes tumor growth and immune evasion positions it as a significant novel therapeutic target35.
Direct evidence of AADAT’s pro-tumorigenic role was established through genetic knockdown experiments, where depletion of AADAT in TNBC cell lines significantly delayed tumor formation and growth in immunocompetent mice. This tumor-suppressive effect was primarily mediated by CD8+ T cells, as AADAT knockdown did not delay tumor initiation in CD8 knockout mice. However, although established AADAT knockdown tumors grew more slowly, even in CD8 knockout mice, this suggests the presence of additional CD8-independent mechanisms. AADAT’s role in shaping the tumor immune microenvironment extends beyond CD8+ T cells, as evidenced by flow cytometry, which reveals increased infiltration of macrophages and CD4+ T cells following AADAT knockdown. Furthermore, inducible AADAT depletion after tumor establishment significantly reduced spontaneous pulmonary metastases post-resection. While many TME factors contribute to immunosuppression, AADAT’s unique correlation with reduced CD8+ TILs suggests it actively contributes to an “immune desert” phenotype, which presents a more challenging barrier to immunotherapy than an “inflamed but suppressed” TME41. This implies that AADAT inhibition might be particularly effective in converting immunologically “cold” tumors to “hot” ones, thereby enhancing anti-tumor immunity. The translational relevance of AADAT inhibition in sensitizing TNBC to immunotherapy is supported by clinical and preclinical data, including a melanoma dataset (GSE91061) where low AADAT expression correlated with improved immunotherapy response42. Experimentally, AADAT depletion synergized with anti-PD-1 and anti-CTLA-4 combination therapy, leading to significantly reduced tumor growth. Imaging mass cytometry demonstrated increased interactions between tumor cells and CD8+ T cells in AADAT knockdown tumors, highlighting AADAT as a promising target to enhance the efficacy of immune checkpoint blockade (ICB) in TNBC, a critical need given the variable response rates to current ICB strategies43.
While AADAT is a primary enzyme in the kynurenine and lysine catabolism pathways, it also exhibits aminotransferase activity in the tyrosine metabolism pathway, underscoring the importance of considering this broader metabolic impact when interpreting our findings. This positions AADAT as a critical node where amino acid metabolism intersects with immune regulation in TNBC, aligning with the broader concept of metabolic vulnerabilities in cancer cells that hijack normal metabolic pathways to create an immunosuppressive environment43.
Unbiased metabolomics revealed a profound impact of AADAT on mitochondrial function, potentially via CoQ10, a critical cofactor in the electron transport chain and a potent antioxidant44. Specifically, increased malate levels were observed in conditioned media following AADAT knockdown in TNBC cells, accompanied by impaired mitochondrial respiration, increased intracellular malate-to-oxaloacetate ratios (a surrogate for mitochondrial NADH: NAD+33), and decreased ATP production. Crucially, CoQ10 supplementation normalized malate levels in AADAT knockdown cells. The finding that AADAT directly regulates CoQ10 levels, thereby impacting mitochondrial function and immune evasion, represents a novel mechanistic link. While CoQ10’s role in mitochondrial health and immunity is well-established, and its deficiency has been linked to various diseases, including cancer45, identifying AADAT as an upstream regulator of CoQ10 provides a new target for modulating this critical pathway. Interestingly, studies have indicated that high CoQ pathway activity can correlate with worse prognosis and lower immune infiltration in breast cancer46, aligning with the current findings and suggesting that dysregulation of CoQ10 (e.g., AADAT-mediated reduction leading to compensatory malate secretion) might be beneficial for anti-tumor immunity. This suggests a complex interplay in which optimal CoQ10 levels are crucial and perturbing them in the context of AADAT inhibition may be a therapeutic option. The compensatory secretion of malate following AADAT knockdown and CoQ10 impairment suggests a metabolic adaptation by tumor cells, highlighting how disruption of one metabolic pathway leads to the extracellular accumulation of another metabolite (malate).
We believe that the metabolic adaptation leading to malate accumulation involves two separate processes, both regulated by AADAT. AADAT/KATII is known to produce α-KG via the TCA cycle by utilizing several glucogenic amino acids25. Knocking down AADAT could block α-KG from entering the TCA cycle and from reversing the activity of the α-KG/malate antiporter, thereby decreasing the import of cytosolic malate into mitochondria and increasing its export, both of which are limited by α-KG import. Additionally, as mentioned earlier, AADAT reduces CoQ10 levels, leading to reductive stress (high NADH/NAD), which may enhance malate production via mitochondrial Malate Dehydrogenase 2 (MDH2). These downstream effects of AADAT—namely, increased mitochondrial malate synthesis and decreased cytosolic malate import—could lead to a net buildup of intracellular malate, subsequently exported into the tumor microenvironment by tumor cells. Supporting this, our study shows that exogenous CoQ10 supplementation can normalize both intracellular and secreted malate levels in tumor cells. Future experiments measuring tumor CoQ10 levels and assessing the ability of external CoQ10 to alleviate reductive stress will be key in confirming this novel metabolic reprogramming, which results in increased malate secretion into the tumor microenvironment.
Strikingly, malate supplementation mimicked the anti-tumor effects of AADAT knockdown in immunotherapy-resistant preclinical models. Malate treatment reduced tumor growth and enhanced CD8+ T cell-mediated killing of TNBC cells in co-culture assays, establishing malate as a direct effector of the observed anti-tumor immunity. The specific role of malate was confirmed by the absence of CAR-T cell cytotoxicity when fumarate was used, which serves as a substrate for malate in the TCA cycle and was also secreted into the conditioned medium (CM) of AADAT KD containing E0771-Ova+ cells (Fig. 4a). Additionally, among the metabolites that were elevated in the CM of AADAT KD containing E0771-Ova+ cells, only malate showed overlapping spatial distribution with functional T cells in patient tumors. Furthermore, intra-tumoral malate levels were significantly linked to patient outcomes after chemotherapy. These findings prompted further investigation into the immunostimulatory properties of malate.
Bulk RNA-seq and flow cytometry revealed that malate treatment enriched for inflammatory pathways, including TNF-α and IFN-γ, and increased glycolysis, TCA cycle metabolites, the NAD+: NADH ratio, and reactive oxygen species (ROS) levels in CD8+ T cells47. These metabolic shifts are consistent with enhanced effector function and metabolic fitness, which are critical for robust anti-tumor immunity48.
By demonstrating that AADAT knockdown in tumors impairs mitochondrial respiration (via CoQ10 reduction) and that malate can restore T cell metabolic fitness, the study uncovers a new upstream regulatory mechanism for T cell function, potentially preventing T cell exhaustion. This establishes a sophisticated, multi-step connection between the tumor’s metabolism and the immune cell’s function.
In a highly relevant preclinical setting, adding malate to drinking water significantly reversed resistance to standard treatment regimens that combine anti-PD-1 and PTX in TNBC tumors. Malate supplementation led to notable decreases in tumor size and enhanced survival in the orthotopic 4T1 model, which was de novo resistant to immunotherapy, as well as in the E0771-Res1 and 2208-L Res 2 TNBC models that had developed resistance to PTX and anti-PD-134 Notably, our preliminary unpublished findings suggest that E0771-Res1 and 2208-L Res 2 follow distinct pathways to acquire resistance to immunotherapy. This demonstrates malate’s potential as a non-toxic, potent metabolic adjuvant to reverse anti-PD1 resistance by enhancing T cell function. This is evident in patient-derived TNBC tumors, which show a positive correlation between intra-tumoral malate levels and clusters containing functional T cells, and a negative correlation with clusters of exhausted T cells. Most importantly, TNBC patients with high intra-tumoral malate levels had improved 10-year survival following standard-of-care chemotherapy compared with those with low malate levels. Notably, malate has been shown to inhibit cancer growth in other contexts, for instance, by inhibiting 6-phosphogluconate dehydrogenase (6PGD) and DNA synthesis in lung cancer49.
These findings position malate as a novel, clinically translatable adjuvant that may overcome resistance mechanisms in TNBC. It further supports malate’s role as a biomarker for a “hot” immune microenvironment and its functional contribution to anti-tumor immunity in patients. The findings directly address the significant challenges in TNBC treatment, characterized by limited therapeutic targets and a poor prognosis35. By identifying AADAT as a novel target and malate as an effective adjuvant for immunochemotherapy, this research offers a promising new strategy to improve outcomes in this aggressive subtype, particularly for patients who may not respond optimally to current ICB plus chemotherapy regimens. This provides a new, potentially well-tolerated approach to enhance outcomes for a patient population that urgently requires improved therapies. Provocatively, it also establishes a basis for developing dietary strategies to address immunotherapy resistance in TNBC.
In summary, this study identifies AADAT as a novel, critical regulator of the tumor immune microenvironment in TNBC, driving immune exclusion and promoting tumor progression. A unique immune-metabolic axis is unveiled, in which AADAT modulates mitochondrial function via CoQ10, leading to compensatory malate secretion when AADAT is reduced. Crucially, the evidence demonstrates that malate acts as a potent immunomodulator, enhancing CD8+ T cell effector function and sensitizing immunotherapy-resistant TNBC to standard-of-care immunochemotherapy, ultimately improving survival in preclinical models. The robust clinical validation of intra-tumoral malate as a biomarker for immune activation and improved patient outcomes underscores the profound translational potential of these findings. This research provides a compelling rationale for targeting AADAT and for exploring malate supplementation as a novel, accessible, non-toxic, and effective adjuvant strategy to enhance anti-tumor immunity and improve clinical outcomes for patients with TNBC.

Materials and Methods

Patient cohorts
The NCI cohort, which includes matched transcriptomics and immunohistochemistry data from 67 breast cancer patients collected by the National Cancer Institute, as previously described, was analyzed to identify KP enzymes associated with immune cell infiltration [19]. AADAT protein expression in TNBC patients was measured by IHC (Rabbit Polyclonal Antibody, Cat No: TA317921 from Origene, Rockville, MD) and related to immune cell marker expression in the Baylor Scott & White Health (BSWH) dataset, collected as previously described, and in a commercially available tissue microarray (US Biomax Inc, Cat# BR2082a, Rockville, MD) [20]. Publicly available METABRIC and TCGA breast cancer datasets were obtained as described previously and analyzed to associate AADAT mRNA expression with clinical outcomes [20,21]. For all datasets, samples with missing data were excluded.
All human studies were conducted under protocols 300009407 (University of Alabama at Birmingham) and at Roswell Park Comprehensive Cancer Center, with patient consent obtained from Roswell Park. Tissue cores from 31 tumors were analyzed. Alongside self-reported race, we collected clinical data including receptor status, tumor grade, stage, chemotherapy history, and 10-year follow-up data. All tumors were obtained during surgery. Subsequently, in the adjuvant setting, all patients were treated at the same hospital following standard care protocols.
Tissues were deparaffinized in xylene and rehydrated through graded alcohols. For antigen retrieval, the slide was pressure-cooked for 10 minutes. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 5 minutes. Slides were blocked with 3% goat serum, then incubated with AADAT antibody (10 μg/mL, Cat# TA331923, Origene, Rockville, MD) at room temperature for 1 hour in humidity chambers. The HRP-conjugated goat anti-mouse/anti-rabbit secondary antibody (Jackson Immunoresearch Laboratories Inc., West Grove, PA) was applied for 40 minutes. The antigen-antibody complex was visualized using diaminobenzidine (Sigma-Aldrich, St. Louis, MO) for 7 minutes. Slides were counterstained with hematoxylin (Sigma-Aldrich, St. Louis, MO). Positive controls were included in each staining run; negative controls were prepared by omitting the primary antibody. Slides were then dehydrated in alcohols, cleared in three xylene baths, and mounted with Permount (Sigma-Aldrich, St. Louis, MO). Histology scores were assigned by Dr. Chandandeep, a clinical breast pathologist, based on multiplying the intensity and extent of IHC staining.

Cell lines and cell culture
E0771, E0771-Res1 and E0771-ova+ cells (gift of Dr. Xiang Zhang) were cultured in RPMI-1640 medium (Thermo Fisher, Waltham, MA) supplemented with 10 mM HEPES buffer (Thermo Fisher). 4T1 and 2208L-Res2 cells34 (gift of Dr. Xiang Zhang) were grown in DMEM medium (Corning, Corning, NY). All media contained 10% Fetal Bovine Serum (Gibco, Cat # A52094–01) and 1% Penicillin/Streptomycin solution (Corning, Cat # 30–002-Cl). Cell lines were confirmed by Short-Tandem Repeat analysis at MD Anderson Cancer Center and cultured in a humidified incubator at 37°C with 5% CO2. All cells were routinely checked for mycoplasma contamination using the MycoAlert Detection Kit (Lonza, Cat# LT07–418). To generate stable Aadat knockdown, E0771 and 4T1 cells were infected with lentiviral vectors containing two independent shRNAs targeting this gene, obtained from the BCM Cell-based Assay Screening Service (C-BASS, Houston, TX). To re-express Aadat, E0771 and 4T1 cells were infected with lentiviral vectors expressing the Aadat open reading frame (GeneCopoeia, Cat# EX-Mm06255-Lv151). In all cell lines, selection for knockdown was performed with puromycin (0.5 μg/ml for E0771 cells, 1 μg/ml for 4T1 cells), or for re-expression with neomycin (1 μg/ml), for at least one week. Inducible Aadat knockdown in E0771, E0771-ova+ and 4T1 cells was achieved using a lentiviral vector containing doxycycline-inducible shRNA targeting Aadat (C-BASS) Cells were induced with 2 μg/ml doxycycline and 4 μg/ml puromycin. shRNA details are mentioned in Supplementary Table 5. All the cell lines used in this study to generate genetic knockdown of AADAT and its re-expression are summarized in Fig. S10.

RNA extraction and qRT-PCR
Procedures were performed as described in our previous publication50. Primers used in this study are listed in Supplementary Table 6.

Metabolomics to analyze Kynurenine/Kynurenic acid ratio
Three replicates of 0.3 × 10^6 E0771 cells were plated and cultured overnight in standard media. The cells were then incubated with media containing 200 μM L-Tryptophan (13C(U); 15N2, Cambridge Isotope Laboratories, Cat# CNLM-2475-H, Tewksbury, MA) for 24 hours. After this period, cells were washed with PBS and stored at −80°C. Metabolites were extracted using a liquid-liquid method and analyzed by high-throughput LC-MS/MS as previously described (23,24). The tryptophan pathway metabolites were separated on a ZORBAX Eclipse XDB C-18 HPLC column under positive ionization mode. For ESI positive mode, mobile phases were 0.1% formic acid in water (A) and acetonitrile (B). The gradient flow was: 0–2 min 2% B; 2–12 min 10% B; 12–17 min 80% B; then re-equilibrated until 25 min back to 2% B. The flow rate was 0.3 ml/min, and the injection volume was 10 μL. Data acquisition was performed via multiple reaction monitoring (MRM) with a 6495 Triple Quadrupole MS coupled to an HPLC system (Agilent Technologies, Santa Clara, CA) using Agilent Mass Hunter Software (25). Peak integration and data analysis were carried out with Agilent Mass Hunter Quantitative Analysis software. The total metabolite pool was calculated by summing labeled and unlabeled fractions for each metabolite.

Targeted metabolomics for malate and oxaloacetate in tumor cell-conditioned media
TCA cycle metabolites, including malate and oxaloacetate, were extracted from cell culture media, with pooled samples serving as quality controls, following the previously described extraction protocol50–52. Malate was separated using a Luna 3 μm NH2 (100 Å) HPLC column (Phenomenex, Torrance, CA), with mobile phases consisting of 5 mM ammonium acetate in water (A) at pH 9.9 and 100% acetonitrile (B). Metabolite separation was carried out on an Agilent 1290 Infinity HPLC system, with data collected on an Agilent 6495B Triple Quadrupole mass spectrometer (Agilent Technologies, Santa Clara, CA) in MRM mode with negative ionization. Data analysis involved Agilent Mass Hunter Quantitation software, including manual verification of integrations and peak evaluation across samples. Results were normalized using spiked isotopically labeled internal standards and log2-transformed. Differential metabolites were identified using the Benjamini-Hochberg method with an FDR cutoff of 0.25, following Student’s t-test.

Untargeted metabolomics in tumor cell-conditioned media
For untargeted metabolomics, three volumes of methanol/acetonitrile (50/50, v/v) were added to cell culture media. Samples were vortexed for 5 min, kept at −20°C for 10 min, then centrifuged at 4°C and 15,000 rpm for 10 min. Extracts were dried using a GeneVac EZ-2 Plus SpeedVac and reconstituted in 100 μL of methanol/water (50/50, v/v). 10 μL of supernatant from each sample was pooled to create QC samples. Separation was performed using a Thermo Scientific Vanquish Horizon UHPLC system. HPLC analysis employed a Waters ACQUITY HSS T3 reversed-phase column for reverse-phase separation and a Waters ACQUITY BEH amide HILIC column for HILIC separation. Reversed-phase mobile phases were 0.1% formic acid in water (A) and methanol (B). HILIC mobile phases were 0.1% formic acid in ammonium formate, with varying solvent compositions. Gradient flow: 0–3 min at 1% B; 4–11 min at 50% B; 12–21 min at 95% B; re-equilibration for 30 min. Columns operated at 50°C, with a 300 μL/min flow rate and a 2 μL injection volume. Data acquired on a Thermo Orbitrap IQ-X and processed with Compound Discoverer 3.3.3.2 for peak detection and ID, using retention time, MS/MS databases, mzCloud, NIST 2020 HRMS, and HMDB (<5 ppm). QC runs every 10 samples as a reference. Data analysis included log2 transformation and median IQR normalization. Differential metabolites identified via Student’s t-test (p-values) and FDR correction (FDR<0.25).

Statistical analysis for unbiased metabolomics
After data acquisition, the missing values for metabolites were imputed using the K nearest-neighbor (KNN) method. Then the data were log2 transformed and followed by the median IQR normalization. The compound-by-compound t-test was applied to identify the top differentially regulated metabolites that passed the nominal P value of p<0.05, followed by the Benjamini-Hochberg procedure for false discovery rate (FDR<0.25) correction, accounting for multiple comparisons53.

Pathway Enrichment for Differential Metabolites
We performed a pathway enrichment analysis via hypergeometric enrichment for core-enriched metabolites against Hallmark gene sets in the Molecular Signatures Database (MSigDB). A nominal P<0.05 and a false discovery rate (FDR) <0.1% were used as thresholds for significance of Enrichment.

Proteomics
Cell pellets were lysed in 50 mM ammonium bicarbonate with 1 mM CaCl 2 through 3 minutes of sonication. Then, 25 μg of protein was digested with 0. 5 μg of trypsin/Lys- C for 12 hours at 37 ° C. Tryptic peptides were fractionated using a homemade reverse-phase C 18 column in a pipette tip. The peptides were eluted and separated with a stepwise gradient of increasing acetonitrile (2%, 4%, 6%, 8%, 10%, 12%, 14%, 16%, 18%, 20%, 22%, 24%, 26%, 28%, 30%) at pH 10. These fractions were combined into five groups (2 + 12 + 12, 4 + 14 + 24, 6 + 16 + 26, 8 + 18 + 28, 10 + 20 + 30) and vacuum dried. The dried peptides were analyzed on Easy- nLC 1000 nanoflow HPLC coupled with Orbitrap Fusion mass spectrometers (Thermo Fisher Scientific). The setup included a 2 cm × 100 μm trap column and a 5 cm × 150 μm separation column packed with 1. 9 μm Reprosil- Pur Basic C 18 beads. Peptides were separated using a 75- minute discontinuous gradient of 4–26% acetonitrile with 0. 0.1% formic acid at an 800 nl/min flow rate. The mass spectrometer operated under Xcalibur software version 4.1. 1, combining data- dependent acquisition (DDA) and Parallel Reaction Monitoring (PRM). In DDA mode, the top 3 peaks in the full MS scan (300–1400 m/z) at 120, 000 resolution were selected for fragmentation. MS/MS spectra via higher- energy collisional dissociation (HCD) were obtained in rapid scan mode in the ion trap. For PRM, seven specific AADAT peptides were monitored: EIYELAR (+ 2, m/z 447. 447.2463), FEDDLIKR EIYELAR (+ 2, m/z 581. 581.2844; + 3, m/z 345. 859), FLTATSLAR (+ 2, m/z 490. 490.2891), GLAEWHVPK (+ 2, m/z 518. 518.7893), SAAFTVENGSTIR (+ 3, m/z 451. 451.5726), TTADILSK (+ 2, m/z 424. 424.7461), QLIEEK (+ 2, m/z 380. 380.2205). The MS/MS spectra were searched against the Kynurenine/alpha- aminoadipate aminotransferase (UniProtKB/Swiss- Prot: Q 9 WVM 8. 1) protein sequence in Proteome Discoverer 2. 1 (Thermo Fisher), using the Mascot algorithm (Mascot 2. 4, Matrix Science). Allowed parameters included a precursor mass tolerance of 20 ppm, fragment mass tolerance of 0. 5 Da, up to two missed cleavages, and dynamic modifications such as N- terminal acetylation and methionine oxidation. PRM target peptides were manually validated.

Mice
C57Bl/6J (WT, B6) were obtained from the Center for Comparative Medicine at BCM. B6.129S2-Cd8atm1Mak/J (B6.CD8KO) were a gift from Dr. Xiang Zhang and bred in our facilities. BALB/cJ mice were obtained from Jackson Laboratory. CD8KO BALB/cJ mice were a gift from Dr. Xiang Zhang and bred in our facilities.

Breast tumor models
Orthotopic tumors were generated by injecting 50 μl of breast cancer cells (0.5 × 10^6 E0771, 0.2 × 10^6 4T1, 1.5 × 10^5 E0771-Res1, and 1.2 × 10^5 2208L-Res2) in PBS mixed with 50 μl of Matrigel (Corning, Cat# 354248, Corning, NY) into the fourth mammary fat pad of mice. Bilateral tumors were established for tumor take and growth studies. Tumor growth was monitored three times weekly using calipers, and tumor volume was calculated with the formula V = π/6 × length × widtĥ2.

In-vitro and In-vivo treatments
The in-vitro treatments used in this study include Malate (or L-(−)-Malic acid, Sigma, Cat no. M7397), N-Acetyl-L-cysteine (NAC, Sigma, Cat no. A9165) and CoQ10 (Sigma-Aldrich, Cat no. C9538). Throughout various experimental setups, cells were treated with 2.5 mM Malate, 2.5 mM NAC and 10μM CoQ10 for 24 hours unless otherwise specified. For Malate and NAC treatment, a 100mM stock solution of Malate and NAC was prepared by dissolving the commercially available powdered form in molecular biology grade water at room temperature. The final concentration of 2.5mM was achieved by adding the required volume of stock solution in prewarmed media. For CoQ10 treatment, a 1 mM CoQ10 stock solution was prepared by dissolving CoQ10 in 100% ethanol and heating for 5 min to 65°C. 10μl of 1 mM CoQ10 was then added to 90 μl Intralipid (Sigma, Cat no.-I141) and heated to 65°C for 5 min. This solution was then added immediately to the cells to achieve a final concentration of 10 μM. In vivo studies involved treating mice with Doxycycline (Dox), anti-PD-1, anti-CTLA-4, PTX, and Malate. These treatments were administered separately or in combination, depending on the study aims. Dox was used to induce shRNA targeting Aadat in breast tumors by providing mice with water containing 2 mg/mL Dox, prepared twice weekly. For immune checkpoint blockade, each animal received intraperitoneal injections of 100 μg anti-CTLA-4 (clone 9D9) and 200 μg anti-PD-1 (clone RMP1–14), or equivalent amounts of isotype-matched control antibodies (mouse IgG2b (clone MPC-11), rat IgG2a (clone 2A3), BioXCel, Lebanon, NH) at the indicated times post-implantation. PTX was administered at 10 mg/kg body weight via intravenous retro-orbital injection. The injectable PTX was prepared by dissolving it in ethanol to achieve a concentration of 20 mg/mL, then mixing it with an equal volume of Cremophor® EL (Millipore, Cat no. 238470) to reach a final concentration of 10 mg/mL. The stock solution was stored at −20°C. On the day of injection, a 2 mg/mL solution was prepared in 1X PBS, and 100 μL of this working solution was injected per animal on the specified days. Five percent (w/v) Malate (Sigma, Cat no. M7397) was provided in drinking water starting from Day 3 of the study (Day 0 being the day of tumor cell injection into the mammary fat pad). The drinking water was replaced with a fresh 5% Malate solution every 24 hours for each cage. For the 4T1 and E0771 orthotopic mouse models, animals were maintained on 1% (w/v) Malate water until the study endpoint, after 12 days of treatment with 5% Malate water. For the E0771-Res1 and 2208L-Res2 orthotopic mouse model, animals alternated between 1% and 5% Malate water weekly until respective study endpoints, after 12 days of continuous 5% Malate water treatment.

Immunofluorescence analysis
Frozen tumor sections were prepared as previously described (18). Primary antibodies against CD31 (1:100, R&D Systems, Cat# AF3628, Minneapolis, MN), VE-Cadherin (1:100, R&D Systems, Cat# AF1002, Minneapolis, MN), and NG2 (1:200, Sigma-Aldrich, Cat# AB5320, St. Louis, MO) were incubated with the samples overnight at 4°C. Subsequently, donkey anti-goat IgG, Alexa Fluor 488 (1:500, Jackson ImmunoResearch, West Grove, PA) or donkey anti-rabbit IgG, Alexa Fluor 555 (1:500, Invitrogen, Waltham, MA) were applied for 2 hours at room temperature prior to imaging.

Flow cytometry of resected tumors
After resection, tumors were dissociated into single cells using the mouse Tumor Dissociation Kit (Miltenyi Biotec, Cat# 130-096-730, Auburn, CA). Cells were incubated on ice for 10 minutes with CD32 FcR blocker (1:100, Tonbo, clone 2.4G2, San Diego, CA). Subsequently, they were stained with conjugated antibodies for 25 minutes on ice in the dark using FACS buffer (PBS with 1% FBS), then washed twice with the same buffer. Data acquisition was performed on a BD LSR Fortessa, and analysis was done with FlowJo v10.0. To quantify absolute numbers of tumor-infiltrating immune cells, liquid counting beads (BD Biosciences, Franklin Lakes, NJ) were employed. The antibodies used, diluted 1:200, are listed in Supplementary Table 7.

Seahorse XF Mito Stress Assay
Mitochondrial respiration was evaluated using the Seahorse XF Analyzer (Agilent Technologies) according to the manufacturer’s protocol. E0771-ova+ control and AADAT-iKD cells (both iKD1 and iKD2) were seeded in 9–10 wells per experimental run. A total of 3 runs were performed with cells taken from different passages. For each well, 20,000 cells were seeded in 100 μL of complete growth medium per well in an XF Cell Culture Microplate and incubated for 24 hours at 37 °C. The sensor cartridge was hydrated overnight in XF Calibrant at 37 °C in a non-CO2 incubator. On the day of the assay, cells were washed and equilibrated with Seahorse assay medium (XF Base Medium supplemented with 1 mM sodium pyruvate, 2 mM glutamine, and 10 mM glucose) and incubated for 45–60 minutes at 37 °C in a non-CO2 incubator. Oligomycin (1 μM), FCCP (1 μM), and rotenone/antimycin A (0.5 μM) were prepared in assay medium and loaded into Ports A, B, and C, respectively. Following calibration, the assay was performed according to standard procedures. Data were normalized with protein concentration per sample (μg/μl) in Wave 2.6.0. Data were analysed using Wave software (Agilent Technologies).

T-cell kill assay
Mouse spleen was isolated after sacrificing a 6–8-week-old female OT-1 mouse obtained from The Jackson Laboratory (C57BL/6-Tg(TcraTcrb)1100Mjb/J, Strain #: 003831). The tissue was immediately washed with 1X PBS supplemented with 2% FBS and 1% Penicillin-Streptomycin (hereafter referred to as FACS buffer). Mechanical disruption of the tissue was performed to obtain a single-cell suspension with FACS buffer. The cell suspension was then filtered through a cell strainer (pluriSelect USA pluriStrainer Mini 10 μm (Cell Strainer), Cat. No. 43-10010-60) and collected in a 5 ml polystyrene tube. The strained cell suspension was centrifuged at 500g for 5 minutes at room temperature. The pelleted cells were incubated in 1X RBC lysis buffer (TONBO biosciences, Cat No. TNB-4300) for 10 minutes at room temperature. After another centrifugation under the same conditions, FCδ receptors were blocked by adding Purified Anti-Mouse CD16 / CD32 (FC shield) from Cytek Biosciences, Cat. No. 70–0161-U500. Negative selection of CD8+ T cells was performed by incubating the cells with a 1:100 dilution of biotin-labeled anti-mouse CD4 (TONBO Biosciences Cat. # 30–0041-U100), CD11c (TONBO Biosciences Cat. # 30–0114-U100), CD11b, CD45R, Ly-6G, and Ly-6C antibodies (BD PharmigenTM, Biotin Mouse Lineage Panel, Cat. No. 559971) at 4°C for 15 minutes. This was followed by centrifugation and resuspension of the pellet in 2 ml of FACS buffer with a 1:40 dilution of Streptavidin Particles Plus-DM (BD IMagTM, Cat. # 557812). Magnetic separation of these beads finally yields a CD8+ T cell population in the cell suspension. The cells were plated with growth media and T cell activation beads (25 μl per 10^6 T cells) (Thermo Fisher Scientific, DynabeadsTM Mouse T-Activator CD3/CD28, Cat. No. 11456D). The T cell growth media consisted of 9 ml RPMI, 1 ml heat-inactivated FBS, 40 μl 1:1000 β-mercaptoethanol, 4 μl of 10 ng/μl IL-2, and 100 μl Penicillin-Streptomycin solution. After 48 hours of activation, activated CD8+ T cells were co-cultured with tumor cells (1:5 ratio) and/or with treatment conditions (as detailed in the respective result sections and figures). After 24 hours of co-culture and treatment (where applicable), the entire well of co-cultured cells was harvested for flow cytometry analysis to assess tumor cell survival and T cell populations by staining with 1:200 dilutions of CD45 (Biolegend, Cat #103155), CD8 (Cytek, Cat # 35–0081-U500), and CD3e antibodies (Biolegend, Cat # 155703). Liquid counting beads (CountBrightTM Plus Absolute Counting Beads, Invitrogen, Cat. No. C36995) were used to quantify the absolute number of tumor and CD8+ T cells. Data were acquired using a BD LSR Fortessa and analyzed with FlowJo v10.0.

PDXO and CAR T-cell co-culture
Breast Cancer PDXO Establishment: Preparation of patient-derived xenograft organoids (PDXOs) was achieved using 10–15 tumor fragments measuring 2–3 mm3 in 5 mL of warm advanced DMEM/F12 medium supplemented with Miltenyi tumor dissociation kit enzymes (Miltenyi Biotec, Auburn) and 10 mM Y-27632. Dissociation was performed using a GentleMACS Octo Dissociator with heaters (Miltenyi Biotec, Auburn) using the pre-set tough tumor type pre-set TDK program. Organotypic fragments were collected by centrifugation at 300 × g for 3 minutes. Fragments were washed in advanced DMEM/F12 medium and centrifuged to increase the organoid to single cell ratio as described in Scherer et al. 2023 (PMID: 37402170). Organoids were counted and embedded in organoid Matrigel (Corning, Glendale). PDXOs were maintained in 6-well plates (Genesee Scientific, El Cajon, CA, USA) and cultured in 200-μl domes of Matrigel matrix for organoid culture (Corning, Corning, NY, USA). PDXOs were covered with Advanced DMEM/F12 (Thermo Fisher, Waltham, MA, USA) supplemented with 5% FBS, 10 mM HEPES (Thermo Fisher), 1X Glutamax (Thermo Fisher), 1 μg/ml hydrocortisone (Sigma-Aldrich, Burlington, MA, USA), 50 μg/ml gentamicin (Genesee Scientific), 10 ng/ml hEGF (Sigma-Aldrich), and 10 μM Y-27632 (Selleck Chemicals, Houston, TX, USA). Organoids were considered established once they were authenticated though STR analysis, had a stable doubling time, and were devoid of stromal cells.
We obtained healthy donor peripheral blood mononuclear cells (PBMCs) through an institutional review board (IRB)-approved protocol at Baylor College of medicine. PBMCs were isolated using Lymphoprep according to manufacturer’s instructions (Axis-Shield PoC AS, Dundee, Scotland). T cells were activated in 24-well non-tissue culture-treated plates coated with OKT3 (1 mg/mL; Ortho Biotech, Bridgewater, NJ) and NA/LE anti-human CD28 antibodies (1 mg/mL; BD Biosciences, San Jose, CA) at a density of 1 × 106 cells per well. H84T CAR-T cells post transduction were expanded in conventional CTL media (45% RPMI-1640 media (Hyclone Laboratories, Marlborough, MA), 45% Click’s medium (Irvine Scientific, Santa Ana, CA), 10% heat-inactivated FBS (Hyclone Laboratories, Marlborough, MA) and 2 mmol/glutaMAX (Gibco by Life Technologies, Carlsbad, CA)) supplemented with 10ng/ml recombinant human IL-7 and IL-15 cytokines. Retroviral transduction was performed as previously described54. The vector encoding the HER.2 directed CAR (2nd generation HER2.28z; clone FRP5) was a kind gift from Dr. Stephen Gottschalk. Transduction efficiency was measured by flow cytometry using a chimeric Erb2-Fc fusion protein (R&D Systems, Minneapolis, MN) with AF-647 anti-Fc antibody (Southern Biotech, Birmingham, AL).
Organoids were expanded prior to seeding in a microplate by culturing approximately twelve 75 ml domes in a 10 cm dish at a density of 1 × 10^5 – 2 × 10^5 cells per dome. Following 7 days of growth, organoids were collected by dispase (50 units/mL) solution (i.e. Dispase stock supplemented with 20% FBS and 10 mM Y-27632) treatment for 10–15 minutes at 37 °C. Dispase solution was washed off, and organoids were strained through a 100 mm strainer to ensure uniform size prior to plating. 500 organoids were seeded in suspension alone or in co-culture with HER2-directed CAR T or non-transduced T cells (ratio of 1 organoid to 50 T cells) stained with CFSE (ThermoScientific, Cat C34554) into 96-well plates pre-coated with a 15 mL base layer of Matrigel in a volume of 100 mL complete medium supplemented with 5% Matrigel. Complete medium was supplemented with either 2.5 mM Malate or Fumarate and their corresponding vehicle controls (i.e. water or ethanol, respectively). Growth was tracked using green integrated intensity and quantified over time using the IncuCyte software.

T-cell RNA Sequencing and Analysis
The CD8+ T cell pellets were harvested from the 1:5 T cell: tumor cell co-culture as described in the ‘T cell kill activity’ methodology section. The cells were then washed with Phosphate Buffer Saline (PBS). Three million cells were harvested and counted per replicate (n=3). RNA extraction was performed using the Aurum™ Total RNA Mini Kit (Bio-Rad). Libraries were prepared using the ABclonal mRNA-seq Library Prep Kit for Illumina (poly (A) selection). RNA integrity and quantification were confirmed on the AATI Fragment Analyzer and quantitative PCR. Quality assessment and sequencing were performed by Novogene Corporation, Sacramento. Sequencing was performed using Illumina NovaSeq X Plus (PE150).

RNA-seq Data Processing and Differential Expression Analysis
First, FAstQC was used to assess the quality of raw RNA-seq reads (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). It was followed by trimming of low-quality bases and adapter sequences using Trim Galore (https://github.com/FelixKrueger/TrimGalore). Cleaned reads were then aligned to the human reference genome (GRCm38/mm38) using HISAT255 with default parameters. The resulting SAM files were converted to BAM format, sorted, and indexed using SAMtools56. Gene-level read counts were quantified using featureCounts57 from Subread. Only uniquely mapped reads overlapping exons were counted toward gene-level expression estimates. The raw count matrix was subsequently imported into DESeq258 in R for normalization and differential expression analysis. DESeq2’s default pipeline, which employs a negative binomial generalized linear model, was used to estimate size factors and dispersions, followed by Wald tests for pairwise comparisons between experimental conditions. P-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method. Genes with an adjusted p-value (FDR) < 0.10 and |log2 fold change| ≥ 1 were considered significantly differentially expressed.

Reactive Oxygen Species (ROS) analysis
The ROS analysis of T cells, which are suspension cells, was performed using the ROS-ID® Hypoxia/Oxidative Stress Detection Kit from Enzo (Cat no.- ENZ-51042). The T cells were harvested from the spleen of 8–10-week-old female C57Bl/6J (WT, B6) mice and activated as described in the ‘T cell kill assay’ section. The activated T cells were then cultured under four conditions: positive control (Pyocyanine treatment at a later stage), negative control (2.5 mM N-Acetyl Cysteine), untreated, and Malate-treated (2.5 mM Malate). The negative control and Malate-treated cells were grown in 2.5 mM NAC and Malate, respectively, for 24 hours. Simultaneously, the untreated and positive control cells were cultured in T-cell medium. After 24 hours, the cells were centrifuged at 400 × g for 5 minutes and the supernatant was removed. Cells from all culture conditions, except the positive control, were then resuspended in 200 μl of ROS-ID® Hypoxia/Oxidative Stress Detection Mix (6 μl of the oxidative stress reagent provided in the kit in 10 ml of 1X PBS) and incubated under standard cell culture conditions (37°C, 5% CO2) for 3.5 hours. The positive control cells were centrifuged and incubated in 200 μl of detection mix along with 500 μM Pyocyanin under the same conditions for 30 minutes. After the incubation, the cells were centrifuged at 400 × g for 5 minutes to remove the detection mix (and Pyocyanin, for positive control), then washed in 1X PBS. Finally, the cells were resuspended in 500 μl of 1X PBS and analyzed by flow cytometry. Data were acquired using BD LSR Fortessa and analyzed with FlowJo v10.0. ROS detection was performed using a filter set compatible with Fluorescein (Ex/Em: 490/525 nm).

NAD+/NADH assay
The T-cells were harvested from the spleen of C57Bl/6J (WT, B6) 8–10-week-old female mice and activated as described in the ‘T cell kill assay’ section. The activated T-cells were then cultured under two conditions: untreated and malate-treated (2.5 mM malate) for 24 hours. NAD/NADH levels of these cells under the respective treatment groups were measured using the NAD/NADH-Glo™ Assay kit (Promega, Catalog no. G9071). 1 × 10^4 cells were plated per well in 50 μl of the respective culture media in a 96-well white-walled tissue culture plate. Then, 50 μl of NAD/NADH-Glo™ Detection reagent (prepared as instructed in the kit) was added to the wells containing the cells. The plate was gently shaken to mix and lyse the cells. It was then incubated for 60 minutes at room temperature, and luminescence was measured using a luminometer.

T-cell cytokine analysis
The levels of T-cell intracellular cytokines (TNF-α and IFN-δ) were measured using flow cytometry. The T-cells were activated as described in the ‘T cell kill assay’ section. The activated T-cells were then cultured with or without 2.5 mM Malate for 24 hours. The CD8+ activated T cells from the two treatment groups were harvested, followed by blocking FCγ receptors as previously described. The nuclei were then stained with 1 μl/ml of Ghost Dye™ UV 450 (Cytek, Catalog no. 13–0868) in FACS buffer for 30 minutes at 4°C protected from light. After washing the cells 1–2 times with FACS buffer, they were incubated in Fixation/Permeabilization solution (BD Cytofix/Cytoperm™ Fixation/Permeabilization Kit, Catalog no. 554714) for 20 minutes at 4°C. The cells were then washed twice with 1 ml of 1X BD Perm/Wash™ buffer. Subsequently, they were centrifuged at 500 × g for 5 minutes. This was followed by incubating the cells in 50 μl of Perm/Wash™ buffer (BD Cytofix/Cytoperm™ Fixation/Permeabilization Kit, Catalog no. 554714) containing a 1:200 dilution of PE anti-mouse TNF-α and Alexa Fluor® 700 anti-mouse IFN-γ (Biolegend, Catalog no. 506305 and 505823 respectively) for 30 minutes at 4°C protected from light. Wash the cells twice with 1ml of 1X BD Perm/Wash™ buffer and then resuspend the cells in FACS buffer prior to flow cytometric analysis.

Metabolomics of CD8+ T cells and conditioned media
The CD8+ T cells were isolated from spleen of 3 female WT C57Bl/6J mice and activated as described above in the ‘T cell kill assay’ section. Post the 48hours of activation, the CD8+ T cells were cultured for 24 hours under 2 independent conditions, vehicle and 2.5mM Malate. The CD8+ T cells isolated from 3 independent animals were cultured in 2 wells providing 2 technical replicates for each of the 3 biological replicates. Each well was seeded with 5 × 105 cells. After 24h treatment, whole well content (Cells + Media) was centrifuged at 500–600 × g for 10 minutes. 1ml of the supernatant (Conditioned media) was collected and flash froze in Liquid N2, followed by storage at −80°C. The whole cell pellet was rinsed gently with 1ml of 150mM ammonium acetate and the incubated in 700μl of 100% ice-cold methanol for 10–15 mins at 4°C. 300 μl of ice-cold HPLC-grade water was then added and the tubes were flash froze in Liquid N2. Later, the cell samples were stored at −80°C till the time samples were processed. At the time of assay, 30 ul of media sample was pipetted to a 1.5ml Eppendorf tube. After adding 120ul of methanol, the samples were vortexed and centrifuged at 20000 rcf for 10 minutes. The supernatant was pipetted to a glass vial and dried under a gentle nitrogen flow. The cell extract samples were dried under a gentle nitrogen flow. Dried samples (medium and cell pellets) were reconstituted in 50 ul of80% methanol and added 50 μL of 105 mM 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) (methanol solution), 50 μL of 175 mM 3-NPH (75 % methanol solution), and 50 μL of 6% pyridine in methanol. After vertexing, the samples were put in 4 °C for 30 minutes. The samples were dried after derivatization and reconstituted in 50% methanol pending LC/MS analysis. The samples were analyzed with UPLC-TQ absolute system (Waters). Metabolites separation was performed on a BEH C18 column (100mm in length and 1mm in ID, Waters). Metabolites were detected in Multiple Reaction Monitoring (MRM) mode. The data were collected in MassLynx and analyzed in TargetLynx.

Imaging mass cytometry (Mass CyTOF)
31 TNBC FFPE tumors, each with a 1 mm core, were analyzed from the Roswell Park TNBC TMA (clinical data summarized in Supplementary Table 1). Regions of Interest (ROIs) were chosen by Dr. Gutierrez. In this cohort, ROIs measured 600 μm by 600 μm, covering the entire 1 mm core. IMC analysis was performed by the Houston Methodist Research Institute’s Immune Monitoring Core. Sample preparation involved staining tissues with metal-tagged antibodies verified by a pathologist, optimized for CyTOF® imaging. A total of 32 metal-tagged antibodies were used to examine the samples (see Supplementary Table 2), enabling detailed analysis of immune, stromal, and tumor cell heterogeneity, as well as various cell subsets and functional phenotypes within the TME. All antibodies were prepared following the manufacturer’s protocols from Standard BioTools, measured for absorbance, and stored in Candor PBS Antibody Stabilization solution at 4 °C. FFPE tissue sections underwent baking, dewaxing in xylene, rehydration with graded alcohols, and heat-induced epitope retrieval using an EZ-Retriever System at 95 °C with a Tris-Tween20 buffer at pH 9 for 20 min. After blocking with 3% BSA in TBS, sections were incubated overnight with a master antibody mix, washed, and stained for nuclei with Cell-ID Intercalator (Standard BioTools). Finally, sections were washed, air-dried, and stored for ablation. Data acquisition was performed through ablation using the Hyperion system (Standard BioTools). The data from Roswell Park TNBC cohort is detailed in our previous publication59.

Image acquisition by Hyperion and data analysis
Image acquisition was carried out in collaboration with the Flow Cytometry core at Methodist Hospital Research Institute. Tissue analysis involved using a Helios time-of-flight mass cytometer connected to the Hyperion Imaging System (Fluidigm). Before acquisition, the imaging system was auto-tuned with a 3-element tuning slide (Fluidigm) following the manufacturer’s instructions. ROIs for imaging were chosen. After flushing the ablation chamber with Helium, tissue sections were ablated spot-by-spot using a UV laser at 200 Hz and 1μm resolution. The data were saved in Fluidigm’s MCD format and exported as 16-bit OME TIFF files for further analysis.

Processing of the Imaging Mass Cytometry data
The IMC data were stored in MCD files and initially processed with imctools to generate multichannel TIFF images. Cells were then segmented using Steinbock, with DeepCell chosen as the segmentation algorithm, based on Ir191 and Ir193 channels for nuclei and surface markers for cell membranes. Using these cell masks, we created a cell-by-protein matrix that summarized pixel intensities for each cell and marker. The cell position was defined by the centroid of each mask. Data were log-transformed and z-scored, first across all cells and then across all proteins. We performed K-means clustering with 20 clusters, setting nstart=100,000 for robust results. Differential proteins between clusters were identified through one-vs-one comparisons, and clusters were named after their differentially expressed markers.

Spatial cell-cell interaction analysis from IMC data
We initially created a Delauney spatial graph connecting individual cells within each ROI. Subsequently, we conducted a spatial proximity analysis using the Giotto pipeline with the 20 cell-type cluster labels. The Cell Proximity Enrichment feature in Giotto, involving 1000 random simulations, detected cell-type interaction enrichment or depletion relative to randomized labels, where cell types are shuffled. This process produced a z-score for each cell-type pair, reflecting either enrichment (positive z-score) or depletion (negative z-score) of interactions for each ROI. We performed group comparisons between AADAT knockdown and wildtype conditions using a linear mixed model that accounts for variability among animals and the repeated measures from multiple ROIs per animal.

Sample preparation for IMS data
TMA sections were dewaxed with 2 times 3 minutes with fresh xylene and allowed to air dry. Fiducial marks were added at the corners of the slide, and an optical image was acquired at 4800 dpi using an Epson Perfection V600 flatbed scanner. The slide was coated with 10 mg/mL 1,5-diaminonaphthalene in 50% acetonitrile using an HTX M5 Robotic Reagent Sprayer over 5 passes with the following parameters: a flow rate of 0.010 mL/min, a track speed of 1200 mm/min, a track spacing of 3 mm, a nozzle height of 40 mm, a nozzle temperature of 60°C, and a crisscross track pattern.

IMS data collection
Metabolite image data was collected in negative ion mode at 50 μm resolution on a Bruker timsTOF fleX QTOF mass spectrometer using FlexImaging 5.1. Instrument parameters were optimized for metabolites as follows: an m/z range of 50–600, TIMS OFF, a Funnel 1 RF of 75.0 Vpp, a Funnel 2 RF of 100.0 Vpp, a Multipole RF of 150.0 Vpp, a Collision Energy of 10.0 eV, a Collision RF of 500.0 Vpp, a Quadrupole Ion Energy of 5.0 eV, a Transfer Time of 50.0 μs, and a Pre Pulse Storage of 5.0 μs. A total of 300 laser shots were summed per pixel.

IMS data analysis
Image data were visualized using SCiLS Lab 2025b and Root Mean Square normalized before further analysis. Peaks were manually selected from the average spectrum with an integration window of ±15 ppm. Putative metabolite IDs were assigned using the MetaboScape 2025 plugin for SCiLS, allowing a 10-ppm mass tolerance. Pixel data for the entire TMA were exported to a .csv file using the SCiLS Lab API for R to allow for integration with IMC data.

Association between T cell clusters and malate
The relationship between malate levels and various cell-type proportions, focusing on functional T cells and exhausted T cells, was examined using binned data. First, all spatial IMS-IMC spots were sorted by malate concentration. We then divided these spots into percentile bins (every 5th percentile, totaling 20 bins). For each bin, we calculated the average malate level per spot from IMS data and the average cell-type proportions per spot from IMC data at the same position. The mean malate level was plotted against the mean cell-type abundance across all bins. A trend line was added to show the association direction, and Pearson correlation was used to measure its strength.

Spatial distribution of malate and the cell types
Spots were ranked based on their malate levels, and the top and bottom 10% were selected and subsetted for each ROI. For each ROI, the malate concentration at the spot level was plotted against the corresponding cell type proportion. Spots with malate levels above zero are shown in yellow, with the color intensity proportional to the concentration. Similarly, the cell type concentration for spots with malate present is shown in red, with the intensity indicating the abundance.

Survival analysis and Cox proportional hazard analysis
Survival analysis was conducted with the survival R package and visualized using the autoplot function from ggplot2. Patients were grouped based on their average malate levels. Kaplan-Meier curves were then generated using R’s survival library. For Cox Proportional Hazards analysis, we linked spatial malate levels to patients’ 10-year survival status in the Roswell Park TNBC cohort. Each cohort included patient attributes such as age, stage, and grade. We developed multivariate models to examine how spatial malate levels relate to survival, while controlling for covariates such as age and stage. The model was specified as “Surv(time, status) ~ malateHighSpots + age + stage,” where malateHighSpots is the count of high-malate spots per patient derived from spatial IMS data. Hazard ratios and P-values were then calculated.

Statistical analysis
Data analysis was performed using GraphPad Prism (v10.6.1) or R. Sample sizes and the specific statistical tests are detailed in the figures or their legends. A p-value under 0.05 was deemed statistically significant. Unless specified otherwise, data are shown as mean ± SEM. Both in vitro and in vivo datasets were tested for normality using the Shapiro-Wilk test, and parametric tests were only applied to data passing this test. All in vitro experiments were conducted in triplicate. Biological and technical replicates are clearly identified in the figure legends for all experiments.