Vitamin B6 Competition in the Tumor Microenvironment Hampers Antitumor Functions of NK Cells

Tumor-induced immune suppression plays a central role in cancer cells’ evasion of immune surveillance and resistance to immunotherapies (1). Both experimental and epidemiologic studies have shown that nutritional factors, including macronutrients (proteins, carbohydrates, and fatty acids) and micronutrients (vitamins and minerals), can regulate cancer progression and immune homeostasis (2). Activation of oncogenic pathways in the cancer cells rewires their metabolic profiles and increases the consumption of nutrient resources from the tumor microenvironments (TME), resulting in the deprivation of vital metabolites for surrounding noncancerous cells, including immune cells (3). The nutritional competition in the TME has been shown to induce the exhaustion of antitumor immune cells and promote immunosuppression (4, 5). Thus, understanding the mechanism of metabolic cooperation and competition in the TME may provide additional therapeutic opportunities against cancer.

Despite success in other cancers, immunotherapies showed limited efficacy against pancreatic ductal adenocarcinoma (PDAC; 6). PDAC TMEs are immunologically cold, posing a great challenge specifically for T cell–based immunotherapies (7). As central innate immune effectors, natural killer (NK)-cell–mediated elimination of malignant cells is MHC-unrestricted and independent of tumor-associated antigen presentation (8). Loss of MHC-I molecules, which is prevalent in PDAC (9, 10), theoretically sensitizes cancer cells to NK cells. Notably, NK-cell degranulation is a predictive prognostic factor in PDAC (11). Therefore, NK cell–based immunotherapies present promising approaches to improve the clinical outcomes for patients with PDAC.

The bioinformatic analysis of The Cancer Genome Atlas (TCGA) patient cohort with QuanTIseq deconvolution method (12) showed that PDAC patients with high NK infiltration in the tumor displayed improved survival, compared with the ones with low NK infiltrations (Supplementary Fig. S1A). Consistent with the in silico analysis results, enrichment of NK cell–specific gene transcripts, NCAM1 and KLRK1, was associated with a longer overall survival in PDAC (Supplementary Fig. S1A). Transcripts for NK-cell receptor activating ligands were detectable in 11 PDAC cell lines (Supplementary Fig. S1B). Moreover, these ligands showed increased mRNA expression in most PDAC cell lines than an immortalized human pancreatic normal epithelial cell line (HPNE) (Supplementary Fig. S1B).

Given the cancer-protective role of tumor-infiltrating NKs, we treated PDAC tumor-bearing mice with recombinant IL15, a cytokine that induces the proliferation and activation of NK-cells. However, increased intratumoral NK-cell population post IL15 treatment did not improve the survival of mice orthotopically implanted with KPC tumor cells (murine pancreatic cancer cells derived from an LSL-Kras^G12D/+; LSL-Trp53^R172H/+; Pdx-1-Cre genetically engineered mouse; Supplementary Fig. S1C–S1E). These results indicate that perhaps NK cells in the PDAC TME are functionally incompetent. Furthermore, peripheral blood–derived human NK cells were incompetent in reducing the growth of PDAC organoids in coculture assays (Supplementary Fig. S1F and S1G). Two-dimensional adherent as well as three-dimensional spheroid culture studies indicated that PDAC cells are more resistant to NK cell–mediated cytotoxicity, as compared with HPNE cells (Supplementary Fig. S1H–S1J). Moreover, NK cells displayed reduced expression of activation markers (IFNγ and CD107a), upon coculturing with PDAC cells compared with HPNE cells (Supplementary Fig. S1I).

Next, we investigated tumor cell–induced transcriptomic changes in NK cells by performing RNA sequencing (RNA-seq). Principal component analysis (PCA) of RNA-seq data from NK cells pre-cocultured with K562, HPNE, or pancreatic tumor cells (T3M4, CFPAC1, or Capan2) showed distinct transcriptional profiles (Fig. 1A). Gene set enrichment analysis (GSEA) suggested that pancreatic tumor cells reduce the inflammatory response signals in NKs (Fig. 1B). In coculture assays, PDAC cells downregulated the mRNA expression of genes involved in NK cell–mediated cytotoxicity (GZMA, PRF1, and IFNG) with a parallel increase of key inhibitory receptors (TIGIT, LAG3, and KLRC1) on NK cells (Fig. 1C). Correspondingly, mouse KPC tumor–infiltrating NK cells displayed significantly higher levels of NKG2A and TIGIT compared with peripheral NK cells from healthy control mice (Fig. 1D and E). These results suggest that PDAC cells significantly impair NK-cell functions that translate into a limited cytotoxic effect of NK cells during tumor growth progression.

PDAC cell lines showed differential susceptibility to NK cell–mediated killing (Supplementary Fig. S1K). However, the susceptibility to NK cells did not correlate with the expression levels of NK cell–activating or inhibitory ligands in PDAC cells (Supplementary Fig. S1L and S1M), indicating that PDAC cell–mediated inhibition of NK-cell activity is likely through indirect interactions. Accordingly, coculturing of NK cells with PDAC cells or PDAC organoids, prior to the subsequent cytotoxicity assay, significantly reduced NK cell–mediated killing of other sensitive target cells (K562) and attenuated the expression of IFNγ in NK cells upon coculturing with K562 (Supplementary Fig. S1N–S1P). Moreover, NK-cell activating ligands (UBLP-Fc, MICA-Fc) and anti-NKp30 failed to induce IFNγ expression in the NK cells that were previously cocultured with PDAC cells (Supplementary Fig. S1Q). These results suggest that the immunosuppressive effect is potentially caused by the conditioned media (CM) from PDAC cells. Indeed, replacing the PDAC cell CM with fresh media (FM) drastically increased the NK cell–mediated killing of PDAC cells (Fig. 1F and H; Supplementary Fig. S1R).

To identify PDAC CM–mediated immunosuppressive mecha­nisms that control NK-cell function, PDAC CM was fractionated by a 3 kDa cutoff centrifugal filter into >3 kDa (enriched mainly for proteins) and <3 kDa (enriched for metabolites) parts. Metabolites (<3 kDa enriched fraction), rather than proteins, contributed to PDAC CM–mediated suppression of NK-cell function as evidenced by attenuated killing, degranulation (CD107a expression), and IFNγ production (Fig. 1I–K; Supplementary Fig. S1S).

To identify tumor-derived metabolites that contribute to NK-cell dysfunction, we performed LC-MS/MS-based metabolomics of NKs post-coculture with PDAC cells or control HPNE cells. Of note, partial least squares discriminant analysis (PLS-DA) showed distinct intracellular metabolite profiles among NK cells cultured alone or cocultured with PDAC or HPNE cells (Fig. 1L). NK cells cocultured with PDAC cells displayed a significant alteration in Tryptophan (Trp), arginine, alanine, and Vitamin B6 (VB6) metabolic pathways (Fig. 1M; Supplementary Fig. S1T and S1U). Variable importance projection (VIP) analysis, which ranks the metabolites based on their overall importance for separating different samples, showed that the intermediates of Trp–Kyn metabolic pathway [Tryptophan (Trp), Kynurenine (Kyn), and 3-hydroxyanthralinic acid (3-HAA)] were differentially altered in NK cells precocultured with or without PDAC or HPNE cells (Supplementary Fig. S1T). NK cells demonstrated significantly higher levels of Kyn upon coculture with PDAC cells as compared with HPNE or K562 (Supplementary Fig. S1V). Consistent with the in vitro results, the KPC mouse model as well as patients with PDAC displayed significantly higher serum Kyn levels as compared with controls (Supplementary Fig. S1W and S1X). In line with previous reports (13–15), NK cells pretreated with Kyn significantly reduced NK cell–mediated cytotoxicity to K562 cells (Supplementary Fig. S1Y). Besides the Trp/Kyn pathway, we noted that vitamin B6 (pyridoxine) was also important for distinguishing NK cells cocultured with or without PDAC or HPNE cells (Supplementary Fig. S1T and S1U). Vitamins actively participate in the maintenance of immune response, where they serve as cofactors and play vital roles in immune regulation (16). VB6 metabolism showed significant alterations in PDAC cell–cocultured NKs (Fig. 1M; Supplementary Fig. S1T and S1U). PDAC cell–cocultured NKs demonstrated a significant decrease in the intracellular levels of pyridoxine (a common form of VB6 in food) and pyridoxal phosphate (PLP, the active form of VB6; Fig. 1N and O). The level of pyridoxine, the form of VB6 in cell culture media, was significantly lower in PDAC-NK coculture-conditioned media (Supplementary Fig. S1Z). Additionally, attenuated levels of pyridoxine and PLP were recorded in the interstitial fluid from primary pancreatic tumors and the plasma samples from orthotopic PDAC tumor-bearing mice compared with healthy controls (Fig. 1P and Q). Also, patients with PDAC displayed significantly lower PLP levels in serum compared with control gallbladder patients (Fig. 1R). Of note, removal or reduction of pyridoxine levels in cell culture medium impaired NK cell–mediated cytotoxicity against K562 and HPNE cells (Fig. 1S). Together, these results indicate that PDAC cells regulate NK-cell function by altering PLP and Kyn levels in the coculture milieu and TME.

Our study showed that alterations in VB6 and Kyn levels regulate NK-cell activity. Trp/Kyn have an inhibitory impact on NK-cell function (14, 15). Importantly, VB6 is a coenzyme for multiple enzymatic steps in the Trp/Kyn pathway and other metabolic pathways such as one-carbon (1C) metabolism and glycogenolysis (16, 17). Thus, we next examined if targeting Trp/Kyn and supplementing VB6 alone or in combination could lower the deleterious effects of PDAC cells on NK cells. VB6 (PLP, active form of VB6, 100 nmol/L) supplementation and treatment with IDO1 inhibitor (IDO1i, 100 nmol/L, Epacadostat) elevated the intracellular PLP and reduced Kyn to control levels in NK cells cocultured with PDAC cells (CFPAC1 or T3M4; Supplementary Fig. S2A andS2B). Treatment of VB6 and IDO1i enhanced the NK-cell cytotoxicity against CFPAC1 and T3M4 cells (Fig. 2A and B). The treatment also increased the percentage of CD107a and IFNγ-positive NK cells (Fig. 2C). Similarly, adding VB6 and IDO1i to the CM of PDAC cells successfully restored the NK-cell cytotoxicity against K562 cells (Supplementary Fig. S2C). In the presence of VB6 and IDO1i, NK cells significantly restricted the growth of PDAC cell 3D spheroids and patient-derived organoids (Supplementary Fig. S2D–S2F). To further examine the effects of VB6 and IDO1i on NK cell–mediated immunity against PDAC in vivo, we utilized a syngeneic orthotopic mouse model of PDAC (Supplementary Fig. S2G). IDO1i, 100 mg/kg/day, and VB6 (PLP,100 mg/kg) treatment led to a significant peripheral PLP enrichment and Kyn reduction in tumor-bearing mice compared with saline-treated counterparts (Supplementary Fig. S2H). Comparable tumor volume and tumor weight were observed in vehicle (CTL), VB6, IDO1i, and IL15 treatment alone groups. In contrast, a significant tumor growth delay was observed in mice treated with a therapeutic combination of IDO1i and VB6 and a triple combination of IL15, IDO1i, and VB6 (Fig. 2D and E).

We then assessed the impact of the triple combination on the immune microenvironment in treated tumors. Combined treatment with IDO1i and VB6 triggered a significant accumulation of tumor-infiltrating NK cells (Fig. 2F). IL15 treatment further increased the NK-cell population in IDO1i and VB6-treated pancreatic tumors. Furthermore, NK cells derived from double or triple combination-treated tumors displayed an activated phenotype (Fig. 2G). However, all the treatment cohorts displayed a comparable number of total tumor-infiltrating T cells (Supplementary Fig. S2I and S2J). VB6-supplemented tumor-bearing mice showed a modestly increased level of Th17 (IL17⁺,CD4⁺ T cell) ratio, but other subtypes of CD4⁺ T cells were not altered (Supplementary Fig. S2K). Although the proportion of IFNγ⁺ CD8⁺ T cells was increased in IDO1i and VB6-treated tumors, the proportions of CD69⁺CD8⁺ T cells, PD-1⁺CD8⁺ T cells, and TIGIT⁺CD8⁺ T cells were similar in all cohorts (Supplementary Fig. S2L).

The combination of IDO1i and VB6 significantly increased the survival of KPC tumor-bearing mice. IL15 further improved survival in mice treated with IDO1i and VB6 (Fig. 2H). These results suggest that VB6 supplementation and inhibition of Trp/Kyn metabolism decreases PDAC tumor burden in vivo. Increased accumulation and activity of intratumoral NKs in IDO1i and VB6-treated tumors indicate that perhaps NK cells contribute to the treatment efficacy. To interrogate the contribution of NKs in antitumor efficacy of IDO1i and VB6-treated mice, we eliminated NKs by using anti-NK1.1 antibody (clone PK136; Supplementary Fig. S2M). NK-cell depletion significantly attenuated the increased survival in the IDO1i and VB6 combination cohort (Fig. 2I). To determine if T cells play a role in controlling the tumor growth in the combination treatment group, CD4⁺ and CD8⁺ T cells were depleted in mice administered with IL15⁺IDO1i⁺VB6 combination (Supplementary Fig. S2N). Depletion of CD4⁺ and CD8⁺ T cells in the mice with combination treatment slightly increased tumor growth compared with the isotype control; however, IL15⁺IDO1i⁺VB6 treatment significantly reduced tumor weight even in the absence of CD4 and CD8 T cells (Supplementary Fig. S2O), suggesting the limited or inexistent role of CD8/CD4 T in our model.

Given the failure of IDO inhibitors against PDAC in the clinic, we interrogated the other VB6-dependent pathways that might underlie NK-cell activity in our model. To investigate this, we performed LC-MS/MS analysis of human primary NKs after being stimulated with HPNE target cells under different concentrations of VB6 (pyridoxine; Supplementary Fig. S3A). Reduction of pyridoxine to 10% of the standard alpha-MEM medium (1.0 mg/mL) significantly reduced the PLP level in NK cells and weakened the killing activity of NK cells against HPNE and K562 cells (Supplementary Fig. S3B and S3C). PLS-DA models showed distinct intracellular metabolic profiles for NKs cultured at different pyridoxine levels with or without HPNE stimulation (Supplementary Fig. S3D). Notably, VB6 reduction significantly affected glycolysis, gluconeogenesis, and VB6 metabolic pathways in HPNE-stimulated NK cells (Supplementary Fig. S3E). Metabolic intermediates of glycolysis pathways (such as F/G-6P, Lactate, and 3-PG/2-PG) were important for segregating the metabolic profiles of nonstimulated and HPNE-stimulated NK cells with or without VB6 supplementation (Supplementary Fig. S3F).

Subsequently, we assessed the metabolic profile of NKs precultured in PDAC cell CM (CFPAC1 CM and T3M4 CM) for 48 hours at different VB6 doses and then cultured with or without HPNE stimulation. (Supplementary Fig. S3G). Supplementation with extra pyridoxine restored the PLP level in NK cells cultured in PDAC CM (Supplementary Fig. S3H and S3I). The PLS-DA analysis demonstrated that NK cells derived from HPNE CM, PDAC CM, and PDAC CM with or without pyridoxine supplementation displayed different metabolic profiles (Supplementary Fig. S3J). Overall, VB6 supplementation significantly affected glycolysis, gluconeogenesis, and VB6 metabolic pathways in HPNE-stimulated NK cells irrespective of prior culture conditions (Supplementary Fig. S3K).

Although PLP is not a cofactor for glycolysis enzymes, it actively participates in glycogen catabolism as a coenzyme for glycogen phosphorylase (PYG; ref. 18), which releases glucose from glycogen (Supplementary Fig. S3L). In our data, two key intermediates in glycogen catabolism, glucose-1-phosphate (Glc-1-P) and UDP-glucose, were among the top metabolites in PLS-DA-VIP analysis of NK cells with VB6 reduction (Supplementary Fig. S3F). Of note, Glc-1-P and UDP-glucose (suggesting less glycogen breakdown) were drastically downregulated in NKs upon coculturing with PDAC cells but not under conditions of monoculture or coculture with HPNE cells (Fig. 3A and 3B).

Amyloglucosidase (AMG)-based assay showed detectable intracellular glycogen in primary mouse and human NK cells, as well as NK92 cells (Supplementary Fig. S3M). Glycogen phosphorylase inhibitor (GPI)-treated NKs displayed glycogen aggregates under the electron microscope, confirming the existence of glycogen bodies in NK cells (Fig. 3C). Subsequently, we noted a drastic reduction in glycogen levels in human and mouse NK cells upon stimulation with HPNE cells or mouse pancreatic epithelial cells (mPEC; Fig. 3D). Inhibition of glycogen breakdown by using PYG inhibitor GPI impaired NK cell–mediated cytotoxicity and cytokine production (Fig. 3E and F; Supplementary Fig. S3N), suggesting that the PYG-mediated glycogenolysis is required for NK-cell effector functions.

We next determined the effect of PDAC CM on glycogen content in NK cells (Supplementary Fig. S3O). The glycogen content of NK cells remained unaltered upon culture in PDAC CM without stimulation with target cells (Supplementary Fig. S3P). Although both HPNE and PDAC stimulation could reduce glycogen content in NK cells, glycogen levels were significantly higher in PDAC CM-treated and PDAC cell-stimulated NKs as compared with HPNE CM-treated and HPNE-stimulated NKs (Fig. 3G). Supplementation of VB6 (PLP) abolished the PDAC cell–induced blockade of glycogen degradation and increased the levels of glucose-1-phosphate and UDP-glucose in NKs (Fig. 3H–J; Supplementary Fig. S3Q). Inhibition of PYG by GPI abrogated the VB6-enhanced glycogen degradation in CFPAC1-cocultured NKs (Supplementary Fig. S3R). Moreover, GPI treatment impeded VB6-mediated rescue of NK-cell activity in PDAC cells-cocultured conditions (Fig. 3K; Supplementary Fig. S3S). These data suggest PDAC cell–induced VB6 depletion mediates impairment of NK-cell glycogenolysis, which may be needed for NK-cell activation under acute stimulation.

Given the role of glycogen in supporting the effector function of NKs, we interrogated the enzyme involved in glycogen metabolism in NK cells. Glycogen synthase (GYS) and glycogen phosphorylase (PYG) are the rate-limiting enzymes for in vivo glycogen synthesis and catabolism, respectively (19). NK cells expressed GYS1 and PYGB isozymes (Fig. 3L and M). PDAC CM did not modulate GYS1 and PYGB expression in NK cells (Fig. 3N; Supplementary Fig. S4A and S4B). Knockdown of GYS1 and PYGB did not affect NK-cell proliferation (Supplementary Fig. S4C and S4D). Although the knockdown of GYS1 decreased glycogen content in NK cells (Supplementary Fig. S4E), PYGB knockdown induced a slight increase in the accumulation of glycogen in NK cells (Supplementary Fig. S4E). Of note, GYS1 or PYGB knockdown impaired NK-cell cytotoxicity against K562 and HPNE cells (Fig. 3O and P; Supplementary Fig. S4F). Silencing of GYS1 or PYGB also impaired cytokine production and degranulation in NKs upon stimulating with K562 and HPNE cells (Fig. 3Q and R; Supplementary Fig. S4G). Moreover, knockdown of GYS1 or PYGB completely blocked VB6-mediated enhancement of NK-cell cytotoxicity against CFPAC1 cells and PDAC organoids (Supplementary Figs. 3S, S4H, and S4I). Silencing of GYS1 and PYGB also impaired NK cell–mediated cytotoxicity in NK92 cells and impeded VB6 (PLP)-induced restoration of NK92 cell cytotoxicity against PDAC cells (Supplementary Fig. S4J–S4R). These data further demonstrate that the breakdown of VB6-mediated regulation of glycogenolysis modulates NK-cell effector function.

NK cells demonstrated increased glucose-1-phosphate levels upon short-term stimulation by target HPNE cells (Fig. 3A). However, reducing VB6 concentration decreased the levels of glucose-1-phosphate and UDP-glucose in NKs stimulated by HPNE-cocultured (Fig. 4A and B). Glucose-1-phosphate is both a precursor of glycogen synthesis and the product of the overall reaction of the breakdown of glycogen (20). Glucose-1-phosphate can be further converted to glucose-6-phosphate, which fuels cells and often ends up in glycolysis or the tricarboxylic acid (TCA) cycle (21). Thus, we speculated that glycogen degradation might provide the energy source for glycolysis or the TCA cycle in HPNE-stimulated NK cells. Short-term HPNE stimulation resulted in an increased rate of glycolysis, which was supported by increased levels of glycolytic intermediates, such as glucose-6-phosphate/fructose 1-phosphate (F-6-P/G-6-P), Fructose 1,6-bisphosphate (F-1,6-BP), 3-phosphoglycerate/2-phosphoglycerate (3PG/2PG), and lactate, in NK cells (Fig. 4C). But NK cells maintained a relatively stable level of the TCA cycle metabolites post stimulation by HPNE cells (Supplementary Fig. S5A). The HPNE-mediated increase in glycolysis was reduced by VB6 limitation in the cell culture medium (Fig. 4C). Reducing VB6 levels, however, had no significant effect on the TCA cycle metabolites in the control or HPNE-treated NK cells (Supplementary Fig. S5A). These results demonstrated that glycogenolysis supports the increased glycolysis in HPNE-stimulated NK cells.

To further determine if glycogen contributes directly to the glycolytic metabolites pool in NK cells, we devised a glycogen labeling strategy to trace the footprints of glycogen-derived glucose (ref. 22; Supplementary Fig. S5B and S5C). NK cells were first fed with U-¹³C-glucose to label glycogen for 72 hours, followed by culturing in unlabeled media for 2 hours prior to stimulation (Supplementary Fig. S5D). Analysis of ¹³C-labeling patterns showed that U-¹³C-glycogen contributed to the increased levels of glucose-1-P (Supplementary Fig. S5E) and glycolytic metabolites in HPNE-stimulated NKs, including F-6-P/G-6-P (M + 6), 3PG/2PG (M + 3), phosphoenolpyruvate (PEP, M + 3), and Pyruvate (M + 3; Supplementary Fig. S5F). In contrast, there was a mild decrease of ¹³C-glycogen-derived carbon in some TCA cycle intermediates (Supplementary Fig. S5G). We also verified these results using a reversed labeling strategy in which NK cells were fed with ¹²C-glucose, and then the medium was changed to U-¹³C-glucose for two hours before stimulation (Fig. 4D). Consistent with the result of U-¹³C-glycogen, ¹²C-glycogen-derived glucose also contributed to the increased levels of glucose-1-P and glycolytic intermediates in HPNE-stimulated NKs (Fig. 4E–4G). These data suggest that target cells induce NK cells to metabolize glycogen through glycolysis to support the effector functions. Notably, although HPNE-mediated stimulation increased the rate of glycolysis in NK cells, the removal of glucose did not attenuate the cytotoxicity of NK cells in an acute response (Fig. 4H and I). Blocking glycogen breakdown significantly inhibited the effector functions of NK cells, regardless of the presence of glucose (Fig. 4J).

We next determined metabolic alterations of NK cells in vivo. We isolated NK1.1⁺ cells and murine EpCAM⁺ cells (tumor cells) from tumors treated with or without IL15 + IDO1i + VB6 (PLP, 100 mg/kg) treatment. Splenic NK cells from healthy mice served as a control. The IL15 + IDO1i + VB6 treatment decreased Kyn levels and increased PLP levels in the tumor interstitial fluid (TIF; Supplementary Fig. S5H). Tumor cells showed higher levels of PLP and Kyn than NK cells in the control group. Treatment with IL15 + IDO1i + VB6 increased PLP and decreased Kyn in NKs (Supplementary Fig. S5I). Moreover, NKs from the tumor control group showed higher levels of Kyn but lower levels of PLP than the NKs from the spleens of the healthy mice group (Supplementary Fig. S5I). Also, the IL15 + IDO1i + VB6 treatment increased levels of glucose-1-P and several glycolytic pathway intermediates (Supplementary Fig. S5J).

A recent meta-analysis on the association of VB6 with pancreatic cancer risk showed that the risk of pancreatic cancer decreased by 9% for every 10 nmol/L increments in blood PLP levels (23, 24). Our results showed lower serum PLP levels in PDAC patients (Fig. 1R). However, the reason for the reduced level of VB6 in PDAC is not clear. The 48-hour CM of PDAC cells had a significantly lower pyridoxine than the CM from HPNE, K562, and NK cells (Fig. 5A). Contrastingly, the intracellular pyridoxine and PLP levels were much higher in PDAC cells than in HPNE, K562, and NK cells (Fig. 5B). These data suggested a competitive VB6 consumption by PDAC cells in the coculture milieu. To further confirm our hypothesis, we compared the VB6 metabolism in different cell types, by feeding them with ¹³C-labeled pyridoxine from 0 hours to 48 hours (Fig. 5C). The intracellular ¹³C-pyridoxine quickly reached saturation in all cells; there was no difference in the ¹³C-pyridoxine fraction between PDAC cells and other cells even at 1 hour after labeling (Fig. 5D). However, the rate at which ¹³C-PLP replaced ¹²C-PLP was much faster in PDAC cells than in HPNE, K562, and NK cells (Fig. 5E). In cancer cells, the ¹³C-PLP replaced around 80% of the unlabeled PLP within one hour, while in other cells, this process took more than 24 hours.

The disproportionate consumption and high metabolic rate of VB6 pointed out that VB6 may play critical roles in cancer cell proliferation and maintenance in PDAC. Removal or reduction of VB6 from the medium inhibited the proliferation of PDAC cell lines both in 2D and 3D culture conditions (Fig. 5F and G). Depletion or reduction of VB6 also suppressed the growth of PDAC organoids (Fig. 5H and I; Supplementary Fig. S6A and S6B). To test the VB6 dependency of PDAC tumors for in vivo growth, C57BL/6 mice were administrated with a VB6-deficient diet for 5 weeks to reduce systemic PLP levels prior to KPC1245 tumor cell implantation (Fig. 5J and K). Mice fed on a VB6-depleted diet demonstrated significantly decreased tumor growth (Fig. 5L).

To explore the downstream mechanism by which VB6 contributes to PDAC cell proliferation, we compared the metabolomic profiles of four PDAC cell lines cultured in low (pyridoxine, 0.1 mg/L) or standard (pyridoxine, 1.0 mg/L) VB6 media. Metabolic pathway analysis showed that reducing VB6 altered a wide range of metabolites in PDAC cells. The top altered metabolic pathways included nucleotides, glutathione, and glycine/serine/methionine metabolism (Supplementary Fig. S6C). These results suggested that 1C metabolism may underlie the VB6-mediated increase in PDAC cell proliferation. 1C metabolism supports DNA synthesis (purines and thymidine), amino acid homeostasis (glycine, serine, and methionine), and redox defense (glutathione) in cells (25, 26). Of note, VB6 is a cofactor for multiple key enzymes involved in 1C metabolic reactions, including serine hydroxymethyltransferase (SHMT), cystathionine β-synthase (CBS), and cystathionine γ-lyase (CGL; ref. 27; Supplementary Fig. S6D). In line with this speculation, 1C metabolic pathway intermediates, including glycine, cystathionine, glutathione (GSH), and S-adenosyl methionine (SAM), were significantly decreased in PDAC cells cultured under low VB6 conditions (Fig. 6A; Supplementary Fig. S6E). Strikingly, supplying both formate and supraphysiologic glycine to supplement 1C metabolism, substantially rescued the growth of human and mouse PDAC cells under VB6-depleted conditions (Fig. 6B and C).

Glycine and formate are downstream products of the SHMT-catalyzed reaction in the 1C pathway (Supplementary Fig. S6D). Therefore, we focused on investigating the involvement of SHMT in VB6-regulated cancer cell growth. To further explore the functional relevance of SHMT enzymes and VB6 in PDAC, we determined PDAC cell growth upon SHMT1 or SHMT2 knockdown and under control and VB6-depleted conditions. The silencing of SHMT1 and SHMT2 reduced PDAC cell growth to levels comparable to VB6 (pyridoxine) depletion (Fig. 6D–G; Supplementary Fig. S6F and S6G). However, knockdown of SHMT1/2 did not further reduce PDAC cell growth upon VB6 depletion. Moreover, the presence of VB6 was unable to rescue cell proliferation following SHMT1/2 deletion. SHIN1, a recently identified small-molecule SHMT1/2 inhibitor (28), exerted potent antiproliferation activity on PDAC cells (Fig. 6H and Supplementary Fig. S6H) in both 2D and 3D culturing conditions. Similarly, SHMT1/2 inhibition effectively blocked cancer cell growth in human PDAC organoids (Fig. 6I and J). These results demonstrate that SHMTs are essentially required for VB6-mediated PDAC cell growth under culture conditions.

As SHIN1 has a poor half-life in vivo (28), we determined tumor growth for KPC1245 and KPC8069 pancreatic tumor cells upon Shmt1 and Shmt2 knockdown. Stable Shmt1/2 knockdown KPC1245 cell lines were prepared using shRNA. Similar to human PDAC cells, in the presence of VB6 (pyridoxine), murine Shmt1/2-knockdown KPC1245 cells showed decreased cell proliferation compared with controls (shCTL). Removal of VB6 and knockdown of Shmt1/2 exerted a comparable antiproliferative effect on the murine PDAC cells. However, no additive or synergistic effects were noted between conditions of VB6 depletion and silencing of Shmt1/2 (Fig. 6K–6M; and Supplementary Fig. S6I). Furthermore, Shmt1/2 knockdown KPC1245 cells showed reduced tumor burden in C57BL/6J mice compared with the controls (Fig. 6N and O). These results suggest that SHMT1/2 is a bona fide metabolic target against PDAC cells.

Although NK cells require VB6 for effector function, PDAC cells utilize VB6 to promote proliferation. Therefore, we speculated that adding VB6 to the TME could have complex effects on PDAC progression. We tested this notion in PDAC mice models (Supplementary Fig. S7A). Mice fed on a VB6-free diet demonstrated attenuated tumor growth as compared with mice fed on a standard diet (pyridoxine, 7 mg/kg; Supplementary Fig. S7B). However, tumor growth rates were comparable in the mice fed on standard or high VB6 diets (70 mg/kg). Furthermore, tumors from high VB6 diet-fed mice demonstrated an increased abundance of NK cells as compared with the tumors from VB6-lacking and standard diets (Supplementary Fig. S7C). The NK cells from high VB6 tumors also showed enhanced cytotoxic function as evidenced by elevated levels IFNγ and NKG2D (Supplementary Fig. S7D). Depletion of NK cells did not affect tumor growth in the mice fed with VB6-free or standard diets, but it resulted in the accelerated growth of tumors in the mice fed with high VB6 diet (Supplementary Fig. S7E and S7F). These results indicate that increasing VB6 could simultaneously promote tumor growth and rescue NK-cell effector functions in PDAC.

Next, we investigate if targeting 1C metabolism in tumor cells regulates NK-cell tumoricidal response. Silencing SHMT1 and SHMT2 blocked VB6-mediated growth promotion in PDAC cells (shown earlier in Fig. 6D–G; Supplementary Fig. S6F and S6G). However, it did not affect the consumption of VB6 and the production of Kyn in CFAPC1 and T3M4 cells (Supplementary Fig. S7G). Treatment with VB6 (PLP, 100 nmol/L) and IDO1i enhanced NK-cell cytotoxicity against control as well as SHMT1/2-knockdown tumor cells (Supplementary Fig. S7H and S7I). Conditioned medium from control and Shmt1/2-knockdown KPC1245 cells had a comparable immunosuppressive effect on murine NK cells as indicated by the reduced NK killing activity and downregulated expression of IFNγ and CD107a in NK cells (Supplementary Fig. S7J–S7L). VB6 and IDO1i rescued the effector functions of murine NK cells from KPC1245 CM-induced exhaustion (Supplementary Fig. S7J–S7L). These data showed that the silencing of SHMT1/2 did not weaken the inhibitory function of PDAC cells on NK-cell immunity, and the treatment of VB6 and IDO1i could still promote the NK cell–mediated killing of SHMT1/2-knockdown PDAC cells.

As VB6 is utilized by both PDAC and NK cells, we examined the possibility of supporting NK-cell function by supplementing with VB6, along with simultaneous inhibition of VB6 utilization by cancer cells, as a therapeutic strategy against PDAC. VB6 (pyridoxine) level did not affect SHMT1 and SHMT2 expression in NK cells (Fig. 7A). Further, treatment with SHMT inhibitor (SHIN1) moderately affected NK-cell viability even at higher concentrations (Fig. 7B). Also, knockdown of SHMT1 and SHMT2 did not affect proliferation and effector functions of NK cells (Fig. 7C–7F). SHIN1 combined with IDO1i and VB6 (PLP,100 nmol/L) supplementation drastically reduced KPC1245 and CFPAC1 cell growth when cocultured with NK cells (Fig. 7G and H). Altogether, our data suggest that VB6 supplementation along with SHMT inhibition attenuates PDAC cell proliferation without affecting the cytotoxicity of NK cells against PDAC cells. To confirm these findings in vivo, control and Shmt1/2-knockdown KPC1245 were orthotopically implanted into the pancreas of mice fed with VB6-free (0 mg/kg), standard (pyridoxine, 7 mg/Kg), or high VB6 (70 mg/kg) diets and treated with IL15 and IDO1 inhibitor (Supplementary Fig. S7M). In line with in vitro results, mice fed on a VB6-deficient diet demonstrated restricted tumor growth compared with mice fed on the standard diet (Fig. 7I; Supplementary Fig. S7N). Also, Shmt1/2-knockdown KPC1245 cells showed attenuated tumor growth in mice fed on standard and high VB6 diets, comparable to KPC1245 control tumor-bearing mice fed on a VB6-deficient diet. Importantly, IL15 and IDO1i treatment further suppressed the growth of Shmt1/2-knockdown KPC1245 tumors (Fig. 7I; Supplementary Fig. S7N). Notably, tumors from the mice fed with a high VB6 diet showed an increased frequency of tumor-infiltrating NK cells (CD3^-NK1.1⁺) as compared with tumors from the mice fed on VB6-deficient or standard diets (Fig. 7J). The treatment of IL15 combined with IDO1 inhibitor further markedly increased NK cells in high VB6 diet-fed mice (Fig. 7J). Additionally, NK cells from IL15 and IDO1 inhibitor–treated high VB6 diet-fed mice demonstrated increased production of effector molecules, such as GZMB, IFNγ, and NKG2D, as well as reduced expression of inhibitory markers TIGIT and NKG2A, compared with high VB6 diet alone cohort (Fig. 7K and L; Supplementary Fig. S7O). To test the long-term robustness of this combination treatment, we compared survival in independent mice cohorts fed with standard and high VB6 diets and treated with or without IL15 and IDO1 inhibitors. The high VB6 diet alone did not affect the survival of the tumor-bearing mice but the addition of IL15 and IDO inhibitor combination to the high VB6 diet extended the survival by more than 10 days (median survival: 30.5 vs. 42 days; Fig. 7M). Mice implanted with Shmt1/2-knockdown KPC1245 cells exhibited relatively longer survival than the mice implanted with control KPC1245 cells (median survival: 30.5 vs. 48.5 days under standard VB6 diet). Combined treatment of IL15 and IDO1 inhibitor in Shmt1/2-knockdown tumor-bearing mice led to a remarkable improvement in survival upon administration with a high VB6 diet, as compared with WT tumor-bearing mice with IL15 and IDO1i treatments (Fig. 7M). In addition, more than 40% (5 out of 12) of Shmt1/2-knockdown tumor-bearing mice treated with IL15 plus IDO1i combination exhibited tumor-free status for more than 9 months. Because VB6 supports PDAC cell proliferation, long-term VB6 supplementation may promote cancer progression. Therefore, we tested the effect of short-term and long-term VB6 supplementation (short-term: PLP injection for 4 days or 8 days; long-term: PLP injection for 16 days) with IL15 and IDO1i treatment on tumor growth and NK-cell activation in both control and Shmt1/2 knockdown tumor models. IL15 and IDO1i injections with 4-day PLP supplementation significantly reduced tumor growth, but PLP supplementation for 8 days and 16 days showed better antitumor efficacy (Supplementary Fig. S7P). The 8-day and 16-day treatments also showed better effects on the NK-cell population and activation than the 4-day PLP supplementation (Supplementary Fig. S7Q). Together, these data demonstrate that enhancing NK-cell immunity by targeting VB6 and Trp/Kyn pathways combined with inhibiting SHMT1/2 in cancer cells exerts potent anti-PDAC activity.

Unlike T cells, natural killer (NK) cells are not MHC restricted (29). Notably, NK-cell degranulation is a predictive prognostic factor in PDAC, and hence, NK cell–based therapies hold promise against PDAC. However, increasing evidence demonstrated that the PDAC tumor is comprised of a crisscross network of signals among immune cells, stroma, and cancer cells, creating a potent immunosuppressive microenvironment that results in dysfunction of effector immune cells, including NK cells (6). Therefore, increasing NK cells alone has been largely ineffective in controlling PDAC progression. Here, we uncovered a new immunosuppressive mechanism in PDAC (Fig. 7N), wherein PDAC cells maintain fast-paced VB6 metabolism to promote their own proliferation, competitively limiting the amount of VB6 available for antitumor immune cells in the microenvironment. The deficiency of VB6 in NK cells limits the utilization of glycogen, which is necessary for NK-cell effector functions during an acute response.

There is no doubt that a sufficient level of VB6 is necessary to promote and maintain the good health of our body. Although the detailed mechanisms are unclear, limiting VB6 levels through dietary deficiency of VB6 has been shown to impair T-cell responses in mice and humans (30–33), while a high dose of vitamin B6 supplementation increased immune responses in critically ill patients (34). Our study demonstrates a novel metabolic mechanism of VB6 deficiency-caused NK-cell dysfunction in cancer. The acute stimulation by target cells triggers glycogenolysis in NK cells, which supports the NK-cell effector functions. Depletion of VB6 crippled the glycogen degradation by limiting the enzyme activity of PYGB, and thus NK-cell cytotoxicity. VB6 is also the coenzyme for several key enzymatic reactions in the Trp–Kyn pathway (16). One of the interesting outcomes of this work was the observation that IDO1 inhibition and VB6 supplement had an additive effect on restoring NK-cell function in PDAC, indicating that VB6 and Kyn function through independent mechanisms in PDAC caused NK-cell immunosuppression. Additional research is needed to evaluate the cross-talk between VB6 and Trp–Kyn metabolism in PDAC burden-caused immune system impairments.

Glycogen is a crucial carbohydrate energy storage for various cells and can serve as a fuel source for “fight or flight” reactions during acute emergencies (19). Glycogen metabolism has been previously reported in regulating the immune response in both innate and adaptive immune cells. Glycogenolysis can support glycolytic metabolism in dendritic cells during the early stage of lipopolysaccharide (LPS)-induced immune responses (35). In macrophages and CD8⁺ memory T cells, glycogen metabolism facilitates the maintenance of cell survival and functions by promoting the generation of NADPH through producing glucose-6-phosphate and the subsequent pentose phosphate pathway (PPP; refs. 36, 37). In this study, we demonstrated that glycogenolysis contributes to the cytotoxicity of NK cells against tumor cells, and glycogen-derived glucose is a critical energy resource for NK-cell effector functions during acute activation. Our results further showed that the limitation of glucose uptake had moderate effects on NK-cell cytotoxicity against target cells. Identification of this metabolic phenotype provides new insight into how glucose metabolism regulates NK cell functions, but it also raises the question of why NK cells utilize glycogen to support the early glycolytic reprogramming for an acute immune response instead of using glucose directly. Perhaps the generation of glucose-6-phosphate from glycogen is more efficient than the from glucose in activated NK cells. Another explanation is that perhaps glucose transport efficiency is insufficient to support the energy demand in NK cells during acute activation. Indeed, the upregulated expression of glucose transporter (GLUT1, also known as SLC2A1) had been observed in NK cells after overnight stimulation with IL2 (38), suggesting NK cells may need time to upregulate the glucose utilization machinery before employing extracellular glucose to support their activation.

The lack of VB6 limited the proliferation of PDAC cells in vitro, while decreased circulating levels of VB6 showed an inverse association with pancreatic cancer risk (23, 24), supporting our conclusion that vitamin B6 is required for optimal immune surveillance. PDAC and other cancer cells, such as acute myeloid leukemia, require VB6 to maintain tumor growth (17, 39). This requirement provides a new vulnerable anticancer target. Given the potential systemic adverse effects and the immunosuppressive influence (40, 41), directly targeting/limiting VB6 is not a reasonable choice. Here we showed that VB6 accelerated the growth of PDAC cells mainly by promoting 1C cycle. Therefore, drugs targeting the 1C metabolic network can provide a starting point for developing therapeutics against PDAC. In addition, disruption of 1C metabolism by inhibiting SHMT showed a minimal effect on NK-cell proliferation and activation. Therefore, drugs targeting the SHMT combined with NK cell–mediated immunotherapy can further impede PDAC progression. SHIN2, the first in vivo active SHMT1/2 inhibitor was recently developed (42), and it shows potent anticancer activity in T-cell acute lymphoblastic leukemia models. Hence, future studies could evaluate the effect of SHIN2 and the combination treatment in PDAC models.

Although our study revealed a few key facts on the mutual relationship between VB6 and PDAC progression and NK-cell activation, the molecular and cellular circuitries underlying our observations remain obscure. The first question is how the VB6 level is decreased in PDAC patient serum and TIF. Our data showed that PDAC cells consumed more VB6 than normal cells and reduced VB6 in their culture medium, suggesting the decrease of VB6, especially in the PDAC TME, at least in part due to the competitive consumption of VB6 by PDAC cells. However, the VB6 metabolic pathway is more complicated in vivo than in the culture system. It is unknown whether PDAC tumor burden systemically affects the absorption or the catabolism of VB6 in the body. For instance, the richest sources of vitamin B6 are animal foods (41), while digestive difficulty is one of the most common symptoms associated with pancreatic cancer because pancreatic enzymes are responsible for digesting many foods, especially animal foods (43). Another question is the role of VB6 during PDAC initiation. Our data demonstrated that VB6 has progrowth effects on PDAC and helps PDAC cell growth. However, elevated circulating levels of VB6 and intense consumption of VB6-containing food correlate with a reduced incidence of PDAC (23, 24), indicating that VB6 also has antineoplastic effects on PDAC. According to this study, a possible explanation for the antioncogenic effects of VB6 is that VB6 promoted cancer immunosurveillance and antitumor immune responses during cancer initiation. Thus, more studies should be conducted to address the functions of VB6 in the early stage of malignant transformation and PDAC development.

Overall, our study implies that VB6 coordinates multiple activities that are critical for both NK-cell activation and PDAC cell division. Selectively targeting VB6 metabolism in tumor cells while boosting VB6 levels in TME can enhance NK-cell antitumor immunity and impede PDAC progression simultaneously. Hence, our study provides the proof-of-concept for a promising therapeutic approach to combine NK-cell immunotherapy and metabolic therapy against PDAC.

K562 (RRID:CVCL_0004), NK92 (RRID:CVCL_NK92), B16F10 (RRID:CVCL_XH27), YAC-1(RRID:CVCL_2244), HEK 293T(RRID:CVCL_0063), and human PDAC cell lines Capan1(RRID:CVCL_0237), Capan2(RRID:CVCL_0026), AsPC1(RRID:CVCL_0152), PANC-1 (RRID:CVCL_0480), CFPAC1(RRID:CVCL_1119), HPAF-II (RRID:CVCL_0313), SW1990 (RRID:CVCL_1723), and MIAPaCa2 (RRID:CVCL_0428) were obtained from ATCC and cultured as per the instructions. The T3M4 (RRID:CVCL_4056), S2-007 (RRID:CVCL_B279), Patu8902 (RRID:CVCL_1845), and S2013 (RRID:CVCL_B280), HPNE (RRID:CVCL_C466) cell lines were kindly provided by Dr. Michael A. Hollingsworth [Eppley Institute, University of Nebraska Medical Center (UNMC), Omaha, NE]. The murine pancreatic cancer cell lines KPC1245 and KPC8069 derived from pancreatic tumors in Kras^G12D/+; Trp53R172H/+; Pdx1-Cretg/+ (KPC) mice were generously provided by Dr. David Tuveson (CSHL, KPC1245) and Dr. Michael A. Hollingsworth (KPC8069), respectively. All the cell lines are routinely tested for Mycoplasma contamination every six months using a Mycoplasma kit (ABM, Inc., cat. #G238). Primary human cells from healthy donors were isolated from peripheral blood mononuclear cells (PBMC) collected in the Elutriation Core Facility of UNMC and Oklahoma Blood Institute. NK cells were isolated using an unattachment NK-cell isolation kit (Miltenyi Biotec, Inc., cat. #130-092-657). Primary NK and NK92 cells cultured in alpha-MEM with 0.2 mmol/L inositol (Thermo Fisher Scientific, cat. #AC122261000); 0.1 mmol/L 2-mercaptoethanol (Thermo Fisher Scientific, cat. #21985023); 0.02 mmol/L folic acid (Thermo Fisher Scientific, Inc., cat. #BP25195); 100 U/mL recombinant IL2 (PeproTech, Inc., cat. #200-02); adjusted to a final concentration of 5% horse serum (Sigma-Aldrich, cat. #H1270) and 5% FBS (Biowest, Inc., cat. #S1620). The cells were routinely tested for Mycoplasma contamination.

Casein-based OpenStandard Diet with 15 kcal% fat (Research Diets Inc.) were used to prepare the control diet (7 mg/kg pyridoxine, cat. #D11112201i), the VB6-deficient diet (no VB6, cat. #D20102801i), and the VB6 excess diet (70 mg/kg pyridoxine, cat. #D20102803i). To investigate the influence of VB6 on tumor growth and immune system, 12-week-old male C57BL/6J mice (RRID:IMSR_JAX:000664), weighing an average of 25 g, were obtained from The Jackson Laboratory and randomly divided into 3 dietary groups, including the control group (7 mg/kg VB6), the deficient group (no VB6), and the excess group (70 mg/kg VB6). The mice were prefed with the diets for 5 weeks before tumor cell implantation and continued throughout the experiment. Tumor growth rate and survival time were used as main indicators to evaluate the effects of different treatments on KPC tumors.

Mice were treated with recombinant murine IL15 (PeproTech Inc., cat. #210-15; 2 μg per mouse/day, i.p.) on day 7, 9, and 11 postimplantations. For VB6 treatment, the mice were given pyridoxal 5′-phosphate hydrate (PLP, Sigma-Aldrich Inc, cat. #P9255, 100 mg/kg, daily i.p.) from day 7 postimplantations. For IDO1i treatment, the mice were treated with INCB024360 (Epacadostat, AdooQ BioScience Inc., cat. #A15554, 100 mg/kg, daily oral gavage) from 7 day of postimplantations. Primary antibodies used for the study include anti-GYS1 antibody (Abcam Inc., cat. #ab40810, RRID:AB_732660), anti-PYGB antibody (MyBioSource Inc., cat. #MBS611408), anti-SHMT1 antibody (Cell Signaling Technology Inc., cat. #80715, RRID:AB_2799957), and anti-SHMT2 antibody (Cell Signaling Technology Inc., cat. #33443). All animal experiments were conducted under the protocols approved by the institutional animal care and use committee at the UNMC and Oklahoma University Health Science Center.

The RNA-seq results of 176 TGCA human PDAC tumors (upper quartile normalized RSEM data for batch-corrected mRNA gene expression) were obtained from the Broad Institute FireBrowse portal (http://gdac.broadinstitute.org). The clinical information was collected from the Memorial Sloan Kettering Cancer Center cBioPortal (RRID:SCR_014555; http://www.cbioportal.org; ref. 44). The relative frequency of immune cell subtypes in TCGA PDAC tumors were analyzed by QuanTIseq (RRID:SCR_022993), a deconvolution-based method for quantifying the fractions of immune cell types from bulk RNA-seq data (12). Survival analysis of patients with PDAC stratified by the infiltrated immune cell subtypes (top 25% vs. bottom 25%) was based on QuanTIseq analysis results. The Kaplan–Meier survival curves were generated using Prism 9 software with log-rank (Mantel–Cox) test.

Murine and human NK cells were isolated using the autoMACS system using a mouse NK-cell isolation kit (Miltenyi Biotec Inc., cat. #130-115-818) and human NK cell Isolation Kit (Miltenyi Biotec Inc., cat. #130-092-657), respectively. Murine NK cells were obtained from healthy mice spleen whereas human NK cells were collected from the PBMCs. PBMC samples were collected from the elutriation Core Facility at UNMC (Omaha, NE). The purity of NKs was confirmed by flow cytometry. For some studies, human NK cells were cultured for 24 hours to 72 days in complete alpha-MEM media (Thermo Fisher Scientific Inc., cat. #12561056) supplemented with 100 U/mL of recombinant human IL2 (NIH Biological Research branch). The human PDAC organoid strains PA417, PA717, PANC-137, and PANC-193 were generated and cultured previously (45). In pancreatic tumor cells, SHMT1 and SHMT2 were knocked down using siRNAs targeting SHMT1 (Horizon Discovery Inc., cat. #L-004617-00-000) and SHMT2 (Horizon Discovery Inc., cat. #L-004906-00-0005).

The RNA-seq and preliminary bioinformatic analysis were performed by the BGI Genomics Co. Primer sequences used for qRT-PCR were as follows: GYS1: Forward 5′-GCGCTCACGTCTTCACTACTG-3′, reverse 5′-TCCAGATGCCCATAAAAATGGC-3′; GYS2: forward 5′-TGAAGTTGCTTGGGAAGTGAC-3′, reverse 5′-AGGTTCACACTGTTCCACCTG-3′; PYGL: forward 5′-CAGCCTATGGATACGGCATTC-3′, reverse 5′-CGGTGTTGGTGTGTTCTACTTT-3′; PYGB: forward 5′-AGGTGCGGAAGAGCTTCAAC-3′, reverse 5′-TCGCGCTCGTAGTAGTGCT-3′; PYGM: forward 5′-CCATGCCCTACGATACGCC-3′, reverse 5′-TAGCCACCGACATTGAAGTCC-3′. The PCR was performed and analyzed on QuantStudio Real-Time PCR System (Thermo Fisher Scientific, Inc).

Cells were seeded at 5 × 10⁵ cells per well in 6-well plates and cultured using complete RPMI media (United State Biologicals, Inc., cat. #R9010-07; 5% FBS) for 24 hours. Next, the medium was changed to RPMI media with M + 4 13C pyridoxine (Cambridge Isotope Laboratories, Inc. CLM-7563-PK). After 0, 1, 6, 24, or 48 hours of treatments, the intracellular metabolites were harvested using 80% methanol. The relative levels of metabolites were calculated by performing LC/MS-MS–based metabolomic analysis utilizing Acquity UPLC-coupled Waters Xevo TQ-S mass spectrometer. The metabolite levels were calculated using relative peak areas.

NK-cell glycogen content was measured using Glycogen Colorimetric/Fluorometric Assay Kit (BioVision, Inc., cat. #K646). LC/MS-MS–based Metabolomics Analysis of Polar Metabolites was performed as per previous protocols (46). Tumors were enzymatically digested for immune cell profile as per previous protocols (47).

NKs were cultured in alpha-MEM medium supplemented with 100 U/mL of recombinant human IL2 containing 4,500 mg/L glucose with or without GPI (1,4-dideoxy-1,4-imino-D-Arabinitol, Cayman Chemicals, Inc., cat. #20939, 400 μmol/L) treatment for 72 hours. NKs were then collected, washed, and fixed overnight in 0.1 mol/L phosphate buffer with 2% paraformaldehyde and 2% glutaraldehyde. NK-cell samples were then post-fixed in 1% osmium tetroxide (Sigma-Aldrich, Inc., cat. #75632), following dehydration in graded ethanol with propylene oxide, which was used as a transition fluid. After overnight polymerization, samples were embedded in araldite resin, sectioned at 100 nm, and stained with 2% uranyl acetate and Reynold lead citrate. Samples were examined under a Zeiss EM10A electron microscope at the UNMC core facility.

The statistical analysis was performed using GraphPad Prism software (GraphPad Software Inc.). The statistical significance was determined using unpaired Student t test (two-tailed), one-way ANOVA with Tukey multiple comparisons test, two-way ANOVA with Tukey test, or as indicated in figure legends. ***, P < 0.001; **, P < 0.01; *, P < 0.05. P < 0.05 was considered significant. For tumor growth rate and survival analysis, each mouse was considered a separate biological sample. The Kaplan–Meier plots with the Mantel–Cox log-rank test were used to compare survival in animal cohorts.

All original data sets for RNA-seq have been submitted to the Sequence Read Archive of NCBI (BioProject ID: PRJNA1030050). All the cell lines used in this study can be made available from the authors upon request with materials transfer agreements.

P.K. Singh reports grants from NIH/NCI during the conduct of the study. K. Mehla reports grants from NIH/NCI during the conduct of the study. No other disclosures were reported.

C. He: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft. D. Wang: Data curation, formal analysis, investigation, methodology. S.K. Shukla: Data curation, investigation. T. Hu: Data curation, investigation. R. Thakur: Formal analysis, Investigation. X. Fu: Resources. R.J. King: Investigation. S.S. Kollala: Formal analysis, investigation. K.S. Attri: Investigation. D. Murthy: Investigation. N.V. Chaika: Investigation. Y. Fujii: Investigation. D. Gonzalez: Investigation. C.G. Pacheco: Investigation. Y. Qiu: Resources. P.K. Singh: Supervision, funding acquisition, writing–review and editing. J.W. Locasale: Writing–review and editing. K. Mehla: Conceptualization, resources, supervision, funding acquisition, methodology, writing–review and editing.

The authors thank D. Tuveson (Cold Spring Harbor Laboratory) and M.A. Hollingsworth [University of Nebraska Medical Center (UNMC)] for providing the KPC cell lines. The authors thank T. Bargar of Electron Microscopy Core Facility (UNMC) for providing assistance with Electron Microscopy. The authors thank B.T. Hackfort of Echocardiography Imaging Facility (UNMC) for helping longitudinal imaging of tumors with 3D volumetric analysis. The authors thank V.B. Smith and C. Semerad of the Flow Cytometry Research Facility (UNMC) for assisting us in designing flow cytometry panels and data analysis. This work was supported in part by funding from the NIH grant R37CA276924 to K. Mehla and R01CA163649, R01CA270234, and U54CA274329 (NCI) to P.K. Singh. The authors would also like to acknowledge the Fred & Pamela Buffett Cancer Center Support Grant (P30CA036727, NCI) and OU Health Stephenson Cancer Center support grant (P30CA225520, NCI) for supporting shared resources.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.