Targeting DGAT1 reprograms lipid landscape and restores CD8⁺ T cell immunity in pancreatic cancer Authors: Zhongkun Liu, Lang Chen, Tong Zhang, Yuliang Wang, Panpan Ma, Xiangwei Zhang, Ronghui Liu, Ruiying Zhang, Xuanhao Zhang, Qingqing Su, Jinyan Huang, Da Wang, Qi Zhang, Cunqi Ye, Yun-Hua Liu Published: 2026 ABSTRACT AbstractLipid accumulation is a hallmark of the pancreatic ductal adenocarcinoma (PDAC) tumor microenvironment, yet effective strategies to reprogram this lipid-rich niche and restore anti-tumor immunity remain limited. Here, we show that diacylglycerol O-acyltransferase 1 (DGAT1) as a tumor-intrinsic metabolic checkpoint that promotes immune evasion. DGAT1 inhibition rewires tumor lipid metabolism by promoting increased fatty acid uptake and redistribution, thereby depleting extracellular free fatty acids that impair CD8⁺ T cell function. Mechanistically, decreased palmitate availability alleviates endoplasmic reticulum stress, preserves FOXO1 activity, and supports stem-like CD8⁺ T cell differentiation. This competitive lipid remodeling enhances memory potential, restrains terminal exhaustion, and sensitizes PDAC tumors to PD-1 checkpoint blockade in vivo. Together, our findings identify tumor–immune lipid crosstalk as a key barrier to effective immunity in PDAC and establish DGAT1 as a promising therapeutic target to restore T cell function and improve immunotherapy response.Subject terms: Cancer metabolism, Pancreatic cancer, Cancer microenvironment Introduction Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers, with minimal benefit from immunotherapy despite advances in other malignancies1,2. This resistance reflects the profoundly immunosuppressive tumor microenvironment (TME), where nutrient competition and metabolic constraints shape both tumor and immune cell function3,4. Among these constraints, lipid accumulation is a defining metabolic hallmark of PDAC and has emerged as a critical driver of immune evasion5,6. Within this nutrient-deprived but lipid-rich niche, tumor and immune cells adopt altered metabolic states that reinforce suppression and accelerate tumor progression7,8. Lipid metabolism is increasingly recognized as a central determinant of immune cell fate9. Tumor-infiltrating regulatory T cells rely on lipid uptake for persistence10–12, dendritic cells lose antigen-presenting capacity through lipid accumulation13–15, and myeloid cells enhance immunosuppressive activity via arachidonic acid–prostaglandin E₂ pathways16,17. In CD8⁺ T cells, excessive uptake of long-chain fatty acids, oxidized lipoproteins, or cholesterol accelerates exhaustion and ferroptosis5,18,19, whereas certain lipid species, including linoleic acid, trans-vaccenic acid, and microbiota-derived short-chain fatty acids, promote metabolic fitness and effector function20–22. These findings underscore the context-dependent roles of distinct lipid species in shaping immunity. The metabolic interplay between tumor and immune cells within the TME shapes the immune response9. Lipids are not only a source of metabolic energy but also act as structural components of cellular membranes and signaling molecules23,24. Different lipid classes, including free fatty acids (FFAs), triacylglycerols (TAGs), and phospholipids, play distinct roles in modulating immune cell behavior25,26. However, despite growing recognition of the impact of lipid metabolism on tumor progression and immune evasion27,28, the specific lipid species driving immune evasion remain poorly defined. More importantly, effective strategies to reprogram the lipid-rich TME to restore anti-tumor immunity have yet to be fully elucidated. To address this gap, we conducted metabolism-focused CRISPR screens in PDAC, which identified diacylglycerol O-acyltransferase 1 (DGAT1), the rate-limiting enzyme in TAG biosynthesis, as a central regulator of immune evasion29. We found that DGAT1 inhibition rewires tumor lipid metabolism by enhancing fatty acid uptake and redistribution into phospholipids and remodeled TAG species, contributing to reduced extracellular free fatty acid levels in the tumor microenvironment. This metabolic remodeling alleviated ER stress and stabilized FOXO1-dependent transcriptional programs in CD8⁺ T cells, promoting stem-like differentiation, sustaining fitness, and sensitizing tumors to immune checkpoint blockade (ICB). Together, our findings define DGAT1 as a tumor-intrinsic and therapeutically actionable metabolic checkpoint that links FFA metabolism to T cell dysfunction in PDAC. Results CRISPR screening identifies DGAT1 as a metabolic checkpoint in PDAC Cancer cells exhibit distinct metabolic alterations that not only support their rapid growth but also profoundly reshape the surrounding TME, fostering a pro-tumorigenic niche. To investigate the role of metabolic alterations in driving pancreatic cancer progression, we compiled large-scale, publicly available single-cell RNA sequencing (scRNA-seq) datasets from human primary pancreatic lesions, liver metastases, and corresponding adjacent healthy tissues30–33. Following stringent quality control and filtering procedures, we obtained a dataset comprising 439,464 single cells from 75 PDAC patients. Major cell types were annotated through unsupervised clustering combined with established gene markers. To examine how tumor-intrinsic metabolic states influence T cell function and differentiation, we assessed the correlation between the metabolic profiles of tumor cells and distinct CD8⁺ T cell subsets—specifically, exhausted CD8⁺ T cells versus progenitor or memory-like CD8⁺ T cells34. Notably, tumor cell-enriched metabolic pathways that showed a positive correlation with exhausted CD8⁺ T cells included fatty acid metabolism, adipogenesis, folate metabolism, and synthesis of unsaturated FFAs. In contrast, pathways such as bile acid, taurine, and glutathione metabolism were negatively correlated with progenitor or memory-like CD8⁺ T cells (Fig. 1A and Supplementary Dataset 1). CRISPR screening identifies DGAT1 as a metabolic checkpoint in PDAC. A Schematic of the integrated computational workflow used to analyze human PDAC scRNA-seq datasets, linking tumor-intrinsic metabolic pathway activity with CD8⁺ T cell functional and differentiation states. Top enriched metabolic pathways in tumor cells are shown on the right. B Parallel in vitro and in vivo CRISPR screens using a metabolism-focused sgRNA library. MAGeCK analysis and robust rank aggregation (RRA) identified top genes selectively depleted in tumors grown in immunocompetent C57BL/6 mice compared with in vitro cultures. Two-sided tests were used, and P values were adjusted for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) method. C, D Dgat1 knockdown in subcutaneous KPCmut tumors in nude and C57BL/6 mice. Tumor volume (C) and weight (D) are shown. Data represent mean ± SD; n = 8 mice per group (nude) and n = 6 mice per group (C57BL/6). Two-way ANOVA (tumor growth) and one-way ANOVA (tumor weight). E Kaplan–Meier survival analysis of C57BL/6 mice orthotopically implanted KPCmut tumors. Pancreas weight at the endpoint is shown on the right. Data represent mean ± SD; n = 10 mice per group. Log-rank test for survival and one-way ANOVA for pancreas weight. F scRNA-seq was performed on all live leukocytes isolated from KPCflox tumors. UMAP plot shows the distribution of clusters, colored by annotated cell type combined from both groups, the relative fraction of each cluster was shown on the right. Tumors were pooled from 5 mice per group. G UMAP of tumor-infiltrating T cells isolated from KPCflox tumors showing annotated T cell subsets, the relative fraction of each cluster was shown on the right. Tumors were pooled from 5 mice per group. H Flow cytometric analysis of tumor-infiltrating CD3+ T cells isolated from the KPCflox tumors. Data represent mean ± SD; n = 6 biologically independent samples. Two-tailed t-test. I Flow cytometric analysis of IFNγ+ and TNFα+ expression in CD8+ TILs isolated from KPCflox tumors. Data represent mean ± SD; n = 5 biologically independent samples. Two-tailed t-test. Data in B and F–I are results from one experiment. Data in C–E are representative results of two independent experiments. Source Data are provided as a Source Data file. To systematically identify novel metabolic determinants whose loss may enhance innate or adaptive anti-tumor immunity, we performed a metabolism-focused CRISPR/Cas9 screening in the KPCmut cell line (derived from KrasG12D;Trp53R172H, KPCmut mouse model)35. This sublibrary contains around 10,000 sgRNAs (~1000 genes) targeting metabolic enzymes, transporters, and other regulators relevant to cellular metabolism36. Time-matched parallel screens were performed by in vitro passage or tumor implantation into syngeneic hosts and compared sgRNA abundance through next-generation sequencing (Fig. S1A). Using MAGeCK computational algorithms37,38, we ranked sgRNAs that were depleted or enriched in syngeneic hosts (in vivo) compared to cells grown in culture (in vitro) (Supplementary Dataset 2). The most differentially dependent genes in the in vivo condition compared to the in vitro condition included mediators of the heme biosynthetic pathway (Cpox) or glycosylphosphatidylinositol maturation (Pigv and Gpaa1), consistent with previous findings39–41, validating the robustness of our screens (Figs. 1B and S1B). Notably, Dgat1, encoding diacylglycerol O-acyltransferase 1, an ER-resident enzyme that catalyzes the final step in TAG synthesis from DAG and fatty acids42, was identified as the top-ranked depleted gene in KPCmut tumors implanted in syngeneic hosts relative to in vitro cultured cells (Fig. 1B). In validation experiments, we genetically knocked down Dgat1 with short hairpin RNAs (shRNAs) in three murine PDAC cell lines (KC6141, KPCmut, and KPCflox) derived from KPCmut or KrasG12D;Trp53flox/flox (KPCflox) mouse models43,44. Knockdown of Dgat1 only has a minor or no significant effect on cell proliferation in vitro compared to the scramble control (Fig. S1C–E). These cells were then implanted into both immunodeficient and syngeneic wild-type hosts using subcutaneous or orthotopic models to assess tumor growth. In immunodeficient mice, Dgat1 knockdown in KPCmut resulted in no significant reduction in tumor growth (Fig. 1C). In contrast, in immunocompetent hosts, Dgat1 knockdown markedly impaired tumor growth (Fig. 1D). Similarly, in C57BL/6 mice bearing KC6141 or KPCflox tumors, Dgat1 knockdown significantly retarded tumor growth and extended host survival, whereas in immunodeficient host, tumor growth remained modestly reduced but not significantly different from controls (Figs. 1E and S1F–K). To further validate these findings, we completely knocked out Dgat1 with CRISPR/Cas9 and performed similar experiments (Fig. S1L). Remarkably, Dgat1 knockout profoundly suppressed KPCflox tumor growth and progression, markedly increasing overall survival, with effects surpassing those observed with Dgat1 knockdown (Fig. S1M). These results establish DGAT1 as a critical metabolic driver supporting PDAC progression through immune-dependent mechanisms. To determine how tumor-intrinsic DGAT1 influences the immune landscape, we profiled CD45⁺ leukocytes from KPCflox tumors using scRNA-seq. Analysis of 29,724 immune cells revealed profound remodeling of the TME upon DGAT1 loss. DGAT1-deficient tumors displayed expansion of lymphocytes and NK cells with a concomitant reduction in tumor-associated macrophages (TAMs), while B cells, mast cells, and dendritic cells were largely unaffected (Figs. 1F and S2A). Among T cells, DGAT1 knockdown decreased the proportion of regulatory T cells (Tregs) while increasing conventional CD4⁺ T cells (Fig. 1G). Flow cytometry confirmed enhanced infiltration of CD8⁺ and CD4⁺ T cells, whereas Treg frequencies were unchanged (Fig. 1H). Importantly, CD8⁺ T cells from DGAT1-deficient tumors displayed increased polyfunctionality, with higher co-production of IFNγ and TNFα upon ex vivo stimulation (12.7% vs. 40.2% in controls) (Fig. 1I). Similar results were observed in KPCmut and KC6141 murine PDAC models (Fig. S2B, C). Other populations, including CD103⁺ DCs and neutrophils, were unaffected (Fig. S2D, E), and while TAMs exhibited reduced abundance, their activation marker expression remained largely unchanged (Fig. S2F). These findings suggest that DGAT1 loss alleviates immunosuppression and enhances T cell functionality within PDAC tumors. DGAT1 expression is elevated in PDAC and correlates with poor prognosis To investigate the clinical and pathological relevance of DGAT1 in PDAC progression, we analyzed an in-house cohort of 60 PDAC patients (cohort #1, Supplementary Dataset 3). Examination of DGAT1 expression in this cohort by in situ multi-color immunofluorescence showed that DGAT1 was broadly expressed in epithelial-derived tumor cells. DGAT1 expression was significantly higher in PDAC tissue than in adjacent normal pancreatic tissue and was higher in advanced-stage tumors than in early-stage tumors (Fig. 2A). DGAT1 expression was positively associated with higher histologic grade, greater invasion depth, and lymph node metastasis (Fig. 2B, C). To validate these findings, we examined DGAT1 expression by immunohistochemistry (IHC) in an independent tissue microarray (cohort #2). Although DGAT1 is detectable in the normal pancreatic epithelium, its expression is modestly upregulated in early and late pancreatic intraepithelial neoplasia (PanIN) lesions and markedly elevated in PDAC lesions (Fig. S3A, B). Notably, while DGAT1 expression is present in the stromal region, its expression is significantly higher in malignant PDAC cells, consistent with previous single-cell transcriptomic studies33 (Fig. 2D). DGAT1 expression is elevated in PDAC and correlates with poor prognosis. A–C Representative multiplex immunofluorescence staining and quantification of DGAT1 expression in tissues from human pancreatic tissues (cohort #1) (n = 60). DGAT1 expression was stratified by tumor stage (B) or pathological characteristics (C). Scale bars, 0.5 mm. H-score, histological score. Data represent mean ± SD; two-tailed t-test. D Relative DGAT1 mRNA expression in different cell subtypes in human normal pancreatic tissues and PDAC tumors (GSA: CRA001160). E DGAT1 mRNA expression in normal pancreatic tissues (n = 362) and PDAC tumors (n = 178) in TCGA datasets. Two-tailed t-test. F Violin plot showing the correlation of DGAT1 mRNA expression with PDAC subtypes: basal-like (n = 39) and classical (n = 130). Normalized RNA-seq expression data were obtained from EGAS00001002543. Two-tailed t-test. G Western blot analyses of DGAT1 protein expression in KrasG12D-ON or -OFF AK14837 cell line. KRAS expression was suppressed upon doxycycline withdrawal. H Representative multiplex immunofluorescence staining and quantification of DGAT1 and CD8 expression in tissues from human pancreatic tissues (cohort #1) with low (n = 35) and high (n = 25) expression levels of DGAT1. Scale bars, 100 µm. H-score, histological score. Data represent mean ± SD; two-tailed t-test. I Representative image and quantification of CD8 IHC staining in human PDAC tissues (cohort #2) with low (n = 22) and high (n = 48) expression levels of DGAT1. Two-tailed t-test. Scale bars, 100 µm. Data in (A–C, H, I) are results from one experiment. Data in G are representative results of two independent experiments. Source Data are provided as a Source Data file. Genomic analysis of TCGA datasets further revealed that frequent DGAT1 copy number gain (26.8%, 45/168) or amplification (8.9%, 15/168) in PDAC45 (Fig. S3C), corresponding to significantly elevated DGAT1 expression compared to normal pancreatic tissue (Figs. 2E and S3D). Transcriptomic analysis further showed that DGAT1 expression is elevated in both classical and basal-like PDAC subtypes compared to normal pancreatic tissue46 (Figs. 2F and S3E). Moreover, DGAT1 expression is positively correlated with KRAS mutation status, the primary oncogenic driver of PDAC (Fig. S3F). Using inducible KrasG12D (iKras) PDAC cell lines47, we validated this relationship, demonstrating that DGAT1 expression decreased following KRASG12D suppression via doxycycline withdrawal (Figs. 2G and S3G). To explore the association between DGAT1 overexpression and the immunosuppressive TME, we assessed tumor-infiltrating CD8+ T cells (CD8+ TILs) in cohort #1. Stratifying the cohort into DGAT1high and DGAT1low groups based on DGAT1 intensity and area, we observed that DGAT1high tumors were associated with reduced CD8+ T-cell infiltration (Fig. 2H). This inverse correlation was further validated in an independent cohort assessed by IHC (cohort #2, Supplementary Dataset 4) (Fig. 2I). Together, these findings establish DGAT1 as a clinically relevant metabolic factor in PDAC, with high expression linked to aggressive disease, KRAS signaling, and immune exclusion. DGAT1 inhibition reprograms the TME and promotes stem-like CD8⁺ T cell differentiation Given that T cells showed changes in frequency and function upon Dgat1 knockdown or knockout, we further assessed the role of specific subsets of immune cells in response to DGAT1 inhibition through depleting CD4+ T cells and CD8+ T cells. Neutralizing antibody depletion experiments revealed that loss of CD8+ T cells fully reversed KPCflox tumor suppression induced by Dgat1 knockdown (Fig. 3A–C), identifying cytotoxic CD8+ T cells, but not CD4+ T cells, as the key mediators of this anti-tumor effect. To further examine the impact of DGAT1 depletion on CD8+ T cell subsets, we performed scRNA-seq on CD8+ TILs sorted from syngeneic KPCflox tumors. After stringent quality control, transcriptomes from 32,292 cells were analyzed (Fig. 3D), comparing Dgat1 knockdown and control tumors. Using the uniform manifold approximation and projection (UMAP) algorithm, unbiased clustering analysis identified 6 distinct CD8+ T cell sub-clusters (Figs. 3D and S4A): Cluster 1 expressed markers of actively dividing cells (TMki67), including Mki67, Ccna2, and Top2a. Cluster 3 was characterized by transcripts typical of effector CD8+ T cells (Teff), such as Gzmb, Gzmk, and Ccl5. Cluster 5, identified as terminally exhausted CD8+ T cells (Tex), exhibited high expression of inhibitory receptor genes, including Pdcd1, Lag3, Havcr2, and Tox. Clusters 2/4/6 expressed memory-associated markers (Tcf7, Sell, and Il7r) and progenitor-related transcription factors (Klf2, Foxo1, and Satb1). Cluster 2, termed ISG15+ T cells (TIsg15), expressed markers of type I interferon responses, such as Isg15, Ifit1, and Zbp1. Clusters 4 and 6 were classified as effector memory-like T cells (Tem) and progenitor exhausted T cells (Tpex), respectively, based on the relative expression of effector markers (Ccl5, Socs3, and Bcl2) and progenitor exhaustion markers (Slamf6) (Fig. S4A). DGAT1 inhibition reprograms the TME and promotes stem-like CD8⁺ T cell differentiation. A Schematic illustration of xenograft tumorigenesis assays inoculated with KPCflox cells and application of antibodies for CD8+ and CD4+ T cell depletion. B, C Effect of Dgat1 knockdown on subcutaneously implanted KPCflox tumor growth in C57BL/6 mice treated with isotype IgG, anti-CD4 or anti-CD8 antibodies. Tumor volume and weight were shown. Data represent mean ± SD; n = 6 mice per group. Two-way ANOVA for tumor growth and one-way ANOVA for tumor weight. D scRNA-seq was performed on CD8+ T cells isolated from KPCflox tumors. UMAP plot depicting the distribution of clusters, colored by annotated cell type in both groups. Relative fraction of each cluster was shown on the right. Tumors were pooled from 5 mice per group. E Pseudotime trajectory analysis of CD8+ T cells using scVelo, with cells color-coded by CD8+ T cell subtypes and pseudotime progression. F Visualization of gene signatures scores of memory, progenitor and exhaustion gene sets in CD8+ T cells. G Gene set enrichment analysis for indicated pathways in intratumoral CD8+ T cells isolated from KPCflox tumors. H, I Flow cytometric analysis of memory-associated markers (TCF1+, CD62L and IL-7Rα), progenitor exhaustion (Ly108+TIM3-), and terminally exhaustion (Ly108-TIM3+) markers expression in tumor-infiltrating CD8+ T cells isolated from the KPCflox tumors. Data represent mean ± SD; n = 7 biologically independent samples per group (TCF1⁺, TCF1⁺CD62L⁺, Ly108⁺TIM3-, Ly108-TIM3⁺) and n = 9 biologically independent samples per group (IL-7Rα⁺, CD62L⁺). Two-tailed t-test. J Schematic illustration of coculture assay with tumor-derived conditional media. Flow cytometric analysis of (K) effector (IFNγ+, TNFα+), (L) progenitor (Ly108+TIM3-) and terminally exhaustion (Ly108-TIM3+)-associated markers in CD8+ T cells cocultured with tumor-derived conditional media. Data represent mean ± SD; n = 3 biologically independent samples per group; two-tailed t-test. Data in D–G are results from one experiment. Data in A–C and J–L are representative results of two independent experiments. Data in H and I are representative results of three independent experiments. Source Data are provided as a Source Data file. Notably, clusters 2/4/6 (TIsg15, Tem, and Tpex), which are characterized by high expression of Tcf7, Sell, and Il7r, were significantly expanded, while the Tex and TMki67 clusters contracted in Dgat1 knockdown tumors (Fig. S4B). The proportion of effector CD8+ T cells remained largely comparable between conditions. Clusters 2/4/6 accounted for 53.4% of all CD8+ T cells in Dgat1 knockdown tumors, compared to 19.0% in controls. Trajectory analysis revealed that these clusters developed stepwise into Teff and Tex cells and were enriched in signatures associated with less-differentiated, memory-like CD8+ T cells (Fig. 3E, F). These cells also exhibited reduced expression of genes linked to glycolysis and terminal exhaustion34,48 (Fig. S4C). Moreover, all 3 clusters expressed high levels of Ccr7, suggesting that these cells have the potential to traffic to lymph nodes (Fig. S4D). Similar results were obtained when we sorted CD8+ TILs from KPCflox tumors and performed bulk RNA-seq followed by GSEA analysis (Fig. 3G). To validate the enhanced progenitor- and memory-like phenotype, we profiled immune cell subsets by flow cytometry in syngeneic tumor models. DGAT1 depletion in KPCflox tumor cells led to a significant increase in memory-associated markers (CD62L, IL-7Rα, and TCF1), corroborating our single-cell transcriptomic data (Fig. 3H). Among total antigen-experienced (PD1+CD44+) CD8+ T cells, tumors with Dgat1 knockdown has more progenitor (Ly108+TIM3-) but less terminal (Ly108-TIM3+) exhausted T cells (Fig. 3I). Moreover, co-expression of TCF7 and TIM3 (encoded by Havcr2) was observed in progenitor exhausted cells, with higher TCF7+TIM3- and lower TCF7-TIM3+ populations in Dgat1 knockdown tumors (Fig. S4E). Similar results in subcutaneously implanted KPCmut tumors further validated that tumor-intrinsic DGAT1 depletion promotes a memory-like phenotype in CD8+ TILs (Fig. S4F, G). Given the metabolic interplay between immune and tumor cells in the TME, we hypothesized that tumor-intrinsic lipid remodeling may reshape metabolite composition in TME, thereby enhancing CD8+ T cell function. To test this, we cocultured splenic CD8+ T cells with conditioned media from either wild-type or DGAT1-deficient KPCflox tumor cells. CD8+ T cells exposed to DGAT1-deficient condition media exhibited increased cytokine production (IFNγ and TNFα), and an enrichment of progenitor-like (Ly108+TIM3-) CD8+ T cells over terminally exhausted (Ly108-TIM3+) counterparts (Fig. 3J–L). Similar results were observed when cocultured with conditioned media derived from KPCmut tumor cells (Fig. S4H). Notably, these effects persisted after heat inactivation, suggesting that lipid metabolites, rather than protein factors, underlie the observed T cell reprogramming these effects (Fig. S4I). Collectively, these findings demonstrate that tumor-intrinsic DGAT1 suppresses anti-tumor immunity by driving CD8⁺ TILs toward terminal exhaustion, whereas DGAT1 inhibition promotes the accumulation of progenitor and memory-like CD8⁺ T cells. DGAT1 loss rewires lipid metabolism and promotes lipid uptake in tumor cells We next investigated how DGAT1 inhibition reshapes tumor lipid metabolism to support cytotoxic immunity. DGAT1 catalyzes the esterification of diacylglycerol (DAG) with acyl-CoA to generate triacylglycerol (TAG) stored in lipid droplets. We introduced an alanine substitution (H426A) at the conserved catalytic histidine residue49,50. Dgat1 knockdown markedly reduced tumor burden, which was restored by wild-type DGAT1 but not the H426A mutant, demonstrating that the catalytic activity of DGAT1 underlies its tumor-promoting function (Fig. 4A, B). To assess the impact of DGAT1 depletion on tumor lipid metabolism, we performed high-coverage targeted lipidomics on Dgat1 knockdown and control KPCflox cells. Across 420 lipid species spanning 32 classes, DGAT1 inhibition broadly reshaped lipid composition (Figs. 4C and S5A, and Supplementary Dataset 5). Specifically, the levels of most lipid species, including glycerophospholipids, sphingolipids, and esterified cholesterol, were elevated in DGAT1-depleted cells. We also observed a marked increase in major structural phospholipids, including phosphatidylcholine (PC) and phosphatidylethanolamine (PE), with a more modest increase in lysophosphatidylcholine (LPC). In contrast, several other lipid species were reduced, including phosphatidylserine (PS) and cardiolipin (CL) (Fig. S5B). Although total TAG levels were not decreased, DGAT1 inhibition markedly altered TAG composition. When stratified by the total number of double bonds in their acyl chains, TAG species showed a relative enrichment of polyunsaturated forms accompanied by a reduction in more saturated species (Fig. S5C). Interestingly, pharmacologic or genetic suppression of DGAT1 did not diminish lipid droplet formation, and in some PDAC lines even modestly increased it (Figs. 4D and S5D, E), suggesting compensatory pathways sustain TAG and droplet homeostasis despite impaired DGAT1 activity. DGAT1 loss rewires lipid metabolism and promotes lipid uptake in tumor cells. A, B Effect of murine H426A mutant on subcutaneously implanted KPCflox tumor growth in C57BL/6 mice. Tumor volume (A) and weight (B) were shown. Data represent mean ± SD; n = 6 biologically independent samples. Two-way ANOVA for tumor growth and one-way ANOVA for tumor weight. C Heatmap of representative lipids in AsPC-1 cells analyzed by LC-MS/MS. FFA, free fatty acids; HexCer, hexose ceramides; BMP, Bis(monoacylglycero)phosphate; Cer, ceramide; SM, sphingomyelins; CL, cardiolipins; PE, phosphatidylethanolamines; PS, phosphatidylserines; PG, phosphatidylglycerols; PI, phosphatidylinositols; PC, phosphatidylcholines; CE, cholesteryl esters; DAG, diacylglycerols; TAG, triacylglycerols. D Representative images of lipid droplets in indicated cancer cells stained with BODIPY 493/503, analyzed by fluorescent imaging. Cells were pre-treated with or without DGAT1 inhibitor A-922500 (30 µM) for 48 h. Scale bar 10 µm. E Schematic illustration of isotope tracing experiments. AsPC-1 cells were cultured in DMEM supplemented with 5% lipid-free fetal bovine serum and 0.2 mM [U-13C16]palmitic acid for 6 h. F Relative abundance of 13C palmitoyl-linked lipids in AsPC-1 cells, including phospholipids, TAG and DAG were quantified by LC-MS/MS. Data represent mean ± SD; n = 3 biologically independent samples per group; two-tailed t-test. G Growth curve of subcutaneously implanted KPCflox tumors in C57BL/6 mice. Data represent mean ± SD; n = 6 biologically independent samples; two-way ANOVA. H Flow cytometric analysis of expression of progenitor exhaustion (TCF1+TIM3–) and terminally exhaustion (TCF1-TIM3+)-associated markers in CD8+ TILs isolated from KPCflox tumors. Data represent mean ± SD; n = 6 biologically independent samples; two-tailed t-test. Data in (C, G and H) are results from one experiment. Data in (A, B, E and F) are representative results of two independent experiments. Data in D are representative results of three independent experiments. Source Data are provided as a Source Data file. To investigate FFA utilization under DGAT1 deficiency, we conducted isotopic tracing experiments using [U-13C16]palmitic acid as a lipid tracer51,52. While unlabeled levels of major phospholipid classes increased consistently with DGAT1 inhibition, we observed a concomitant increase in phospholipids containing singly (m + 16) or doubly (m + 32) labeled [U-13C16]palmitic acid (Fig. 4E, F, and Supplementary Dataset 6). Similarly, TAG species with varying numbers of labeled palmitic acid also increased in abundance, albeit to a lesser degree than phospholipids. The enhanced incorporation of the labeled tracer into glycerolipids suggests that exogenous fatty acids are readily utilized for both TAG and phospholipid synthesis upon DGAT1 deficiency. Furthermore, we observed a significant decrease in the levels of two key lipid precursors (DAG and PA) involved in phospholipid biosynthesis with [U-13C16]palmitic acid incorporation (Figs. 4F and S5F, G). This reduction further supports the notion that DGAT1 knockdown accelerates the turnover of glycerolipid precursors, driving increased flux into both storage and structural lipid pools. To elucidate the transcriptional basis of this lipid remodeling, we compared gene expression between DGAT1-deficient and wild-type AsPC-1 cells. GSEA analysis revealed elevated fatty acid uptake and phospholipid metabolism pathways (Fig. S5H) but suppressed de novo fatty acid synthesis (FASN, ACACA), consistent with a hyper-metabolic state. This state is characterized by diversified storage (CE and PUFA-TAG) and accelerated phospholipid biosynthesis. Notably, DGAT1 inhibition triggered marked upregulation of CD36, a scavenger receptor for FFAs and oxidized lipids (Fig. S5I–M). Flow cytometry using fluorescently labeled palmitic acid, cholesterol, and OxLDL confirmed enhanced lipid uptake in DGAT1-deficient cells, which was abolished by Cd36 knockdown or pharmacologic blockade with sulfosuccinimidyl oleate (SSO) (Fig. S5I–M). In vivo, Cd36 silencing in wild-type tumor cells had little effect; however, in DGAT1-deficient tumors it significantly abrogated both tumor growth suppression and the expansion of memory-like CD8+ T cells (Figs 4G, H and S5N). Together, these results demonstrate that loss of DGAT1 rewires tumor lipid metabolism by diverting exogenous FFAs toward phospholipid synthesis, while CD36-mediated uptake prevents accumulation of immunosuppressive lipid pools and supports CD8⁺ T cell persistence in the TME. DGAT1 loss restricts FFA overload and boosts T cell fitness and stemness To investigate how DGAT1 loss promotes stem-like CD8⁺ T cell states, we first profiled transcriptional pathways in CD8⁺ TILs. Gene set enrichment analysis (GSEA) revealed that, in addition to programs involved in immune homeostasis, lipid-associated pathways such as cholesterol and sphingolipid metabolism were enriched, whereas fatty acid metabolism was depleted in CD8⁺ TILs from DGAT1-deficient KPCflox tumors (Fig. 5A). Given DGAT1 catalyzes TAG formation, a central pathway for intracellular lipid storage, we hypothesized that DGAT1 depletion rewires tumor lipid flux and thereby remodels the extracellular lipid milieu. Targeted lipidomics of conditioned media from KPCmut and KPCflox cells53 confirmed broad changes in lipid composition, with FFAs being the only metabolites consistently reduced across both models (Fig. S6A, B, and Supplementary Dataset 7). DGAT1 loss restricts FFA overload and boosts CD8+ T cell fitness and stemness. A GSEA analysis shows the altered pathways for the loss of DGAT1 in tumor-infiltrating CD8+ T cells. NES, normalized enrichment score. B, C Representative fluorescence images, quantification lipid droplets and fluorescent intensity in CD8+ T cells stained with BODIPY 493/503. CD8+ T cells were isolated from Dgat1 knockdown or control KPCflox tumors 3 weeks after inoculation. Data represent mean ± SD; n = 4 biologically independent samples per group; two-tailed t-test. D, E Heatmap and quantification of representative total or individual fatty acids in interstitial fluid derived from KPCflox-tumors, quantified by LC-MS/MS. Data represent mean ± SD; n = 4 biologically independent samples per group; two-tailed t-test. F Schematic illustration of coculture assay with indicated fatty acid supplements. G Flow cytometric analysis of expression of memory (TCF1+, CD62L and IL-7Rα)-associated markers and the ratio of Tpex in CD8+ T cells treated with indicated fatty acid supplements (200 µM). Data represent mean ± SD; n = 3 biologically independent samples per group; one-way ANOVA. H Schematic illustration of adoptive cell transfer (ACT) of CD8+ OT-I T cells with palmitate treatment. I Growth curve and tumor weight of OVA-expressing KPCflox tumors in nude mice following adoptive transfer of CD8+ OT-I T cells. Data represent mean ± SD, n = 6 biologically independent samples per group. Two-way ANOVA for tumor growth and one-way ANOVA for tumor weight. J Flow cytometric analysis of expression of memory (TCF1+, CD62L and IL-7Rα)-associated markers in CD8+ TILs isolated from the ACT models. Data represent mean ± SD; n = 6 biologically independent samples per group; two-tailed t-test. Data in (D, E and H–J) are results from one experiment. Data in B, C are representative results of two independent experiments. Data in F and G are representative results of three independent experiments. Source Data are provided as a Source Data file. Consistent with diminished FFA availability, CD8⁺ TILs from DGAT1-deficient tumors exhibited a marked reduction in neutral lipid droplets, as evidenced by decreased BODIPY fluorescence (Fig. 5B, C). Because CD36 mediates FFA uptake, we next assessed its role using an in vitro chronic stimulation model54–56. Pretreatment of sorted CD8⁺ T cells with the CD36 inhibitor sulfosuccinimidyl oleate (SSO), followed by culture in conditioned media, led to decreased lipid accumulation and an expansion of CD62L⁺, IL-7Rα⁺, TCF1⁺ memory-like T cells (Fig. S6C, D), phenocopying the effect of DGAT1-deficient conditioned media. Interestingly, CD8⁺ TILs isolated from DGAT1-deficient tumors showed downregulated CD36 expression (Fig. S6E), in contrast to the progressive CD36 upregulation typically observed during exhaustion, suggesting an adaptive response to reduced lipid availability. Flow cytometry confirmed diminished uptake of both fluorescent cholesterol and long-chain FFAs in CD8⁺ TILs isolated from DGAT1-deficient tumors (Fig. S6F, G). Similar results were observed in splenic T cells cocultured with conditioned media from KPCflox tumors (Fig. S6H). Moreover, we treated purified splenic CD8⁺ T cells with DGAT1 inhibitor A-922500. DGAT1 inhibition reduced neutral lipid droplet accumulation, as reflected by decreased BODIPY staining, and diminished uptake of fluorescently labeled palmitic acid and cholesterol, without affecting cell viability (Fig. S6I). Functionally, DGAT1 inhibition shifted CD8⁺ T cell differentiation toward a progenitor-like state, characterized by increased Ly108⁺TIM3- cells and reduced Ly108-TIM3⁺ terminally exhausted cells (Fig. S6J). This phenotypic shift was accompanied by elevated expression of FOXO1 and memory-associated markers, including TCF1, CD62L, and IL-7Rα (Fig. S6K, L), suggesting DGAT1 inhibition could exert a T cell–intrinsic effect on CD8⁺ TILs. To determine specific FFAs driving T cell dysfunction, we analyzed the composition of tumor interstitial fluid (TIF), a proxy for the TME’s extracellular milieu, and found reduced levels of multiple FFA species, particularly polyunsaturated fatty acids (PUFAs), in DGAT1-deficient tumors (Fig. 5D, and Supplementary Dataset 8). Among these, palmitic acid (PA, C16:0), oleic acid (OA, C18:1), and arachidonic acid (AA, ω-6 C20:4) were the most abundant fatty acids, and their levels were notably depleted in DGAT1-deficient TIF (Figs. 5E and S6M). Functional supplementation experiments revealed that palmitic acid strongly impaired the expression of memory-associated markers (TCF1, CD62L, IL-7Rα) in CD8+ T cells and the enrichment of progenitor exhausted (TCF1+TIM3-) CD8+ T cells, whereas oleic acid or arachidonic acid exerted only modest effects (Fig. 5F, G). Notably, palmitic acid or oleic acid had little impact on lipid peroxidation (PE/FITC ratio) or viability, while arachidonic acid significantly increased both (Fig. S6N, O). Finally, to assess functional relevance in vivo, we performed adoptive transfers of OT-I CD8⁺ T cells preconditioned with either DGAT1-deficient conditioned media or FFA-supplemented media (Fig. 5H). Cells exposed to DGAT1-deficient media mediated superior tumor control compared to controls; however, this advantage was completely abolished by palmitic acid supplementation (Fig. 5I). Flow cytometry confirmed that the memory-like state induced by DGAT1 deficiency was attenuated by palmitic acid re-exposure (Fig. 5J). Together, these findings demonstrate that DGAT1 loss remodels the lipid-rich TME by depleting immunosuppressive FFAs, particularly palmitic acid and PUFAs. This remodeling limits CD36-dependent lipid uptake, prevents lipid overload, and promotes stem-like CD8⁺ T cell differentiation, thereby fostering durable cytotoxic T cell responses. DGAT1 inhibition preserves stem-like CD8⁺ T cell states via FOXO1 activation We next examined the mechanism by which lipid metabolic remodeling promotes stem-like differentiation of cytotoxic T cells. ATAC-seq profiling of CD8+ TILs isolated from DGAT1-deficient versus control KPCflox tumors revealed ~8,000 differentially accessible chromatin regions, the majority (∼78%) located in intronic or intergenic enhancers and ∼11% at promoters (Fig. 6A). Loci linked to progenitor or memory fate programs (Tcf7, Sell, Ccr7, Il7r) showed increased accessibility, whereas those associated with terminal exhaustion (Pdcd1, Havcr2, Cd244a) were less accessible (Fig. 6A). Transcription factor motif enrichment analysis (chromVAR) across ATAC-seq and scRNA-seq datasets consistently highlighted FOXO1 (Figs. 6B and S7A), a central regulator of memory CD8+ T cell differentiation and restraint of exhaustion57–60. Functional annotation further revealed preferential enrichment of the FOXO1 signature in progenitor-like CD8+ subsets (TIsg15, Tem, Tpex) (Fig. 6B) and multiple genes implicated in memory versus exhaustion balance (Tcf7, Ccr7, Pdcd1, Havcr2) harbored known FOXO1-binding sites (Figs. 6C and S7B)61. Bulk RNA-seq also confirmed FOXO1 target gene enrichment in CD8+ TILs from Dgat1-deficient tumors (Fig. 6D), DGAT1 inhibition preserves stem-like CD8⁺ T cell states via FOXO1 activation. A Chromatin accessibility heatmap grouped by differential accessibility patterns of CD8+ T cells, isolated from KPCflox tumors. Tumors were pooled from 5 mice per group. B Normalized gene expression of FOXO1 regulon genes projected onto the UMAP plot. Violin plots depicts normalized expression of Sell and FOXO1 regulon genes across cell clusters. C Chromatin accessibility tracks for the Tcf7 and Ccr7 gene loci of intratumoral CD8+ T cells, isolated from KPCflox tumors. FOXO1 chromatin immunoprecipitation followed by sequencing (ChIP–seq) peaks are aligned with the corresponding gene loci. D GSEA analysis shows the enrichment of FOXO1 regulon genes in DGAT1-deficient CD8+ TILs isolated from KPCflox tumors. E Flow cytometric analysis of FOXO1 expression in CD8+ T cells treated with indicated fatty acid supplements (200 µM). Data represent mean ± SD; n = 3 biologically independent samples per group; one-way ANOVA. F Immunofluorescence staining of FOXO1 in CD8+ T cells treated with BSA or palmitic acid (200 µM). n = 50 cells per group. Cyto, Nuc and Nuc/Cyto denote FOXO1 expression in the cytoplasm, nucleus and both the nucleus and cytoplasm, respectively. Scale bar, 5 µm. G GSEA plot shows the downregulation of unfolded protein response in DGAT1-deficient CD8+ TILs isolated from KPCflox tumors. H, I Western blot analysis of activated CD8+ T cells treated with palmitic acid (200 µM) with or without PERK inhibitor (GSK2606414) or IRE1α inhibitor (4μ8c). J Schematic illustration of adoptive cell transfer (ACT) of CD8+ OT-I splenic T cells with or without Foxo1 knockdown. K Growth curve of OVA-expressing KPCflox tumors in nude mice following adoptive transfer of CD8+ OT-I splenic T cells. Data represent mean ± SD; n = 6 mice per group; two -way ANOVA. L Flow cytometric analysis of expression of memory (TCF1+, CD62L and IL-7Rα)-associated markers in CD8+ TILs isolated from the ACT models at endpoint. Data represent mean ± SD; n = 6 biologically independent samples per group; two-tailed t-test. Data in A–D, G and J–L are results from one experiment. Data in E and F are representative results of two independent experiments. Data in H and I are representative results of three independent experiments. Source Data are provided as a Source Data file. To validate these findings, we examined FOXO1 levels in CD8⁺ TILs from KPCflox tumors. FOXO1 protein was markedly increased despite only modest changes in transcript abundance (Fig. S7C–E). Notably, T cells from DGAT1-deficient tumors displayed increased mitochondrial mass and membrane potential (ΔΨm), but decreased lipid peroxidation or ferroptosis (Fig. S7F–H). Transmission electron microscopy further revealed more rod-shaped, condensed mitochondria frequently polarized to one side of the cell (Fig. S7I), indicating that DGAT1 blockade enhances mitochondrial fitness and activity in vivo, a hallmark of stem-like CD8⁺ T cells62. Consistent with previous reports63,64, supplementation assays further demonstrated that palmitic acid potently suppressed FOXO1 expression, whereas oleic acid or arachidonic acid exerted only minor effects (Fig. 6E). Similar results were observed in human CD8+ T cells activated and expanded in palmitate-containing media. Palmitate treatment significantly altered CD8⁺ T-cell differentiation, characterized by a reduction in effector memory T cells (TEM: CCR7-CD45RA-) and a concomitant increase in terminal effector memory T cells (TEMRA: CCR7-CD45RA⁺) (Fig. S7J, K). Consistently, TCF1 expression was reduced across most CD8⁺ T-cell subsets following palmitate exposure, with the exception of stem cell–like memory T cells (TSCM: CCR7⁺CD45RA⁺) (Fig. S7L). Functionally, palmitate treatment impaired the enrichment of progenitor-like (TCF1⁺TIM3-) CD8⁺ T cells relative to terminally exhausted (TCF1-TIM3⁺) cells (Fig. S7M). This shift was accompanied by reduced cytokine production (IFN-γ, granzyme B, and TNF-α) and decreased expression of the key transcription factor FOXO1 (Fig. S7N, O). Together, these results demonstrate that excess palmitate similarly drives CD8⁺ T cell dysfunction and terminal exhaustion in human T cells, thereby reinforcing the mechanistic framework established in our murine models. Given that FOXO1 is primarily regulated through post-translational modifications rather than transcriptional control57,62,65,66, we next investigated how palmitate mediates this suppression. Under basal conditions, FOXO1 predominantly localized to the nucleus of CD8⁺ T cells, but supplementation with palmitic acid significantly promoted its cytoplasmic translocation and degradation66, thereby reducing total FOXO1 protein abundance (Figs. 6F and S7P, Q). Gene ontology analysis identified the unfolded protein response (UPR) as prominently attenuated in CD8+ TILs from DGAT1-deficient tumors (Fig. 6G). Indeed, palmitic acid induced phosphorylation of EIF2α and IRE1α in CD8+ T cells, whereas ATF6 was minimally affected (Fig. 6H). Pharmacologic inhibition of IRE1α or PERK restored FOXO1 stability, reduced Ser256 phosphorylation, and increased protein abundance (Fig. 6I). To identify upstream kinases involved in FOXO1 phosphorylation, we pretreated CD8+ T cells with a panel of kinase inhibitors. Notably, inhibition of serum glucocorticoid kinase 1 (SGK1) restored FOXO1 stability, whereas blockade of PI3K/AKT or IKKβ signaling had no effect (Fig. S7R). Moreover, palmitic acid stimulation triggered SGK1 phosphorylation, which was reversed by inhibition of ER stress (Figs. 6I and S7S), indicating that SGK1 is the principal kinase mediating FOXO1 phosphorylation downstream of palmitic acid-induced ER stress. Finally, to test functional relevance, we adoptively transferred Foxo1-silenced or control OT-I T cells into mice bearing OVA-expressing KPCflox tumors (Fig. 6J). DGAT1 depletion in tumor cells significantly restrained tumor growth in mice receiving wild-type OT-I cells, but this benefit was abolished when Foxo1 was silenced (Figs. 6K and S7T). Flow cytometry confirmed that the memory-like CD8+ T cell pool expanded in DGAT1-deficiencient tumors was lost upon Foxo1 knockdown (Fig. 6L). Together, these findings demonstrate that DGAT1 inhibition in tumor cells remodels the tumor lipid milieu to limit palmitate-driven ER stress and FOXO1 destabilization. By preserving FOXO1 activity, DGAT1 loss in tumor cells promotes the persistence of stem-like CD8⁺ T cells in the tumor microenvironment. Pharmacological DGAT1 blockade synergizes with immune checkpoint therapy Given the central role of DGAT1 in sustaining tumor lipid homeostasis, we next tested whether DGAT1 inhibition could sensitize PDAC to ICB. In orthotopic KPCmut models, DGAT1 depletion alone significantly reduced tumor growth and improved survival, resulting in ~50% tumor remission (4 of 7 mice) (Fig. 7A). Consistent with prior findings, α-PD-1 monotherapy had no impact on tumor progression, underscoring the intrinsic resistance of PDAC to immunotherapy. However, the combination of DGAT1 depletion with α-PD-1 dramatically improved outcomes, achieving complete tumor regression in 88.9% of mice (8 of 9) (Fig. 7A). This improvement in immunotherapeutic efficacy was associated with a significantly reduced liver metastasis (Fig. S8A). Similar results were observed in the KPCflox syngeneic mouse model, which mirrors the poor immunogenicity and aggressive progression of human PDAC. While α-PD-1 alone failed to eradicate tumors, Dgat1 knockdown led to ~40% remission (3 of 7 mice) (Fig. 7B). Notably, combining α-PD-1 with DGAT1 depletion resulted in 77.8% complete tumor regression (7 of 9 mice), with long-term survival (>120 days) without evidence of residual disease (Fig. 7C). This improvement in immunotherapeutic efficacy was associated with a significantly increased numbers of intratumoral CD8+ T cells after either DGAT1 depletion alone or combined with α-PD-1 (Fig. S8B, C). DGAT1 inhibition potentiates the efficacy of immune checkpoint blockade. Survival curve of orthotopically implanted KPCmut (A) or KPCflox (B) tumor-bearing mice treated with anti-PD-1 or IgG control antibodies (200 µg per mouse, every 3 days), followed by detecting macroscopic and microscopic tumors after autopsy (C). Pancreas weight at the endpoint is shown on the right. Data represent mean ± SD; n = 7 mice per group, except the combination treatment group (n = 9). Log-rank (Mantel–Cox) test (survival) and one-way ANOVA (pancreas weight). Survival curve of orthotopically implanted KPCmut (D) or KPCflox (E) tumor-bearing mice treated with DGAT1 inhibitor A-922500 (80 mg/kg/day, oral gavage) and/or anti-PD-1 antibodies. Pancreas weight at the endpoint is shown on the right. Data represent mean ± SD; n = 6 mice per group, except in KPCflox (E) mice: A-922500 + IgG (n = 7) and A-922500 + α-PD-1 (n = 9). Log-rank (Mantel–Cox) test (survival) and one-way ANOVA (pancreas weight). F Flow cytometric analysis of expression of memory (TCF1+, CD62L and IL-7Rα)-associated markers in CD8+ TILs isolated from KPCflox tumors. Data represent mean ± SD; n = 6 biologically independent samples per group; two-tailed t-test. G Representative images and quantification of immunofluorescence staining for Pan-CK, DGAT1, FOXO1 and CD8 in human PDAC tissues (pooled cohort #1 and #2). FOXO1+ intratumoral CD8+ T cells between patients with low (n = 57) and high (n = 70) expression levels of DGAT1 was compared. Scale bar, 200 µm. Two-tailed t-test. H Kaplan–Meier analysis of OS and PFS based on DGAT1 expression levels in patients from clinical trials with anti-PD-1 or anti-CTLA-4 therapies. Cancer types include melanoma (n = 423), urothelial (n = 348), head and neck (n = 5), esophageal (n = 103), bladder (n = 73), glioblastoma (n = 28), hepatocellular (n = 22), small cell lung cancer (n = 22), and non–small cell lung cancer (n = 21). Data in G are results from one experiment. Data in A–C and D–F are representative results of two independent experiments. Source Data are provided as a Source Data file. To explore the therapeutic potential of pharmacological DGAT1 inhibition, we treated KPCmut tumors-bearing mice with vehicle control or the DGAT1 inhibitor A-922500 (80 mg/kg/day, by oral gavage), alone or in combination with α-PD-1. Although less potent than genetic depletion, α-PD-1 plus A-922500 significantly reduced tumor burden and extended survival compared to single-agent treatments or controls (Fig. 7D). The combination extended median survival by 18 days compared to A-922500 alone and by 28 days compared to vehicle control, more than doubling the median survival time. Similarly, benefits were observed in KPCflox tumors, where α-PD-1 plus A-922500 significantly prolonged survival and reduced tumor burden (Fig. 7E). Importantly, combined treatment showed no signs of gross toxicity, as evidenced by stable body weight (Fig. S8D, E). Moreover, this improvement in immunotherapeutic efficacy was associated with a significantly disrupted desmoplastic TME (Fig. S8F) and the expansion of stem-like CD8+ T cells, as evidenced by elevated expression of FOXO1 transcriptional factor and memory-associated markers (TCF1, CD62L, IL-7Rα) (Figs. 7F and S8G). In situ multiplex immunofluorescence staining further revealed an inverse correlation between DGAT1 expression in tumor cells and intratumoral CD8+ T cell infiltration, including FOXO1-positive CD8+ T cells, supporting the link between DGAT1 inhibition and enhanced CD8+ T cell memory-like phenotypes (Fig. 7G). Further clinical support for the immunomodulatory function of DGAT1 came from in silico analysis of survival and transcriptome data for patients treated with ICB therapies, across several cancer types (excluding PDAC, due to unavailable data)67. Patients with DGAT1high tumors exhibited significantly shorter overall survival (OS) and progression-free survival (PFS) under ICB therapy (Figs. 7H and S8H). Together, these results establish DGAT1 as a lipid-metabolic checkpoint that enforces immune evasion and demonstrate that DGAT1 inhibition synergizes with PD-1 blockade to overcome immunotherapy resistance in PDAC. Discussion Pancreatic ductal adenocarcinoma (PDAC) is characterized by a profoundly immunosuppressive tumor microenvironment (TME) that underlies its poor responsiveness to immunotherapy2,4. A defining feature of this TME is dysregulated lipid metabolism5,7. While the immunosuppressive effects of excess lipids are increasingly recognized3, effective therapeutic strategies that exploit tumor lipid dependencies to reprogram the TME have been limited. Here, through parallel loss-of-function screens in vitro and in vivo, we identify the tumor-intrinsic lipid enzyme DGAT1 as a central metabolic checkpoint that enforces CD8⁺ T cell dysfunction in PDAC. By esterifying diacylglycerol and fatty acids into triacylglycerols (TAGs), DGAT1 preserves immunosuppressive lipid pools. Inhibition of DGAT1 disrupts this lipid homeostasis, remodels extracellular lipid availability, and promotes CD8⁺ T cell differentiation toward stem-like states via FOXO1 stabilization. These findings establish DGAT1 as a nexus between tumor lipid metabolism and adaptive immunity and highlight its blockade as a therapeutic strategy to sensitize PDAC to immune checkpoint blockade. DGAT1, one of two enzymes catalyzing the terminal step in TAG synthesis, is best known for its role in metabolic diseases such as obesity and type 2 diabetes29. In cancer, its overexpression promotes lipid droplet formation and protects cells from lipotoxic stress, whereas inhibition disrupts lipid storage, elevates ROS, and triggers tumor cell death68–70. Our work extends these observations by showing that DGAT1 inhibition has minimal effects on PDAC growth in vitro or in immunodeficient hosts, but markedly suppresses tumor progression in immunocompetent settings. These findings position DGAT1 as a tumor-intrinsic regulator of adaptive immunity, rather than merely a determinant of tumor cell survival. Mechanistically, DGAT1 blockade rewires tumor lipid metabolism by promoting increased fatty acid uptake and redistribution into multiple lipid pools, thereby depleting immunosuppressive FFAs, particularly PUFAs and palmitic acid. This lipid rewiring alleviates ER stress in CD8⁺ T cells, reduces FOXO1 phosphorylation, and stabilizes FOXO1-driven transcriptional programs that sustain stem-like states. Unlike prior work emphasizing direct lipid uptake by immune cells5,9,18, our findings show that tumor-intrinsic lipid flux dictates the extracellular lipid milieu that ultimately governs CD8⁺ T cell fate and fitness. Among altered lipid species, palmitic acid emerged as a dominant suppressor of memory-like CD8⁺ T cell differentiation. Supplementation with palmitate was sufficient to abolish the benefits of DGAT1 inhibition in vitro and in vivo. Moreover, DGAT1 inhibition additionally shields CD8⁺ T cells from ferroptosis by limiting extracellular PUFA pools, further preserving their persistence in the TME. Together, these findings indicate that tumor-driven lipid metabolism directly shapes the abundance and composition of extracellular fatty acids within the TME, and that limiting FFA availability may be a key prerequisite for sustaining functional CD8⁺ TILs. Recent studies have identified TCF1⁺ progenitor CD8⁺ T cells as a critical reservoir that sustains antitumor immunity and mediates durable responses to immune checkpoint blockade34,71,72. These cells rely on transcriptional programs governed by factors such as FOXO1 to maintain self-renewal capacity and differentiation potential57,61. Our results suggest that DGAT1-dependent tumor lipid metabolism can influence this balance by modulating extracellular lipid availability. Elevated levels of saturated fatty acids, particularly palmitate, suppress FOXO1 activity in CD8⁺ T cells and promote the acquisition of dysfunctional or exhausted phenotypes. Conversely, DGAT1 inhibition reduces extracellular fatty acid burden and alleviates lipid-induced suppression of FOXO1, thereby favoring transcriptional programs associated with progenitor-like CD8⁺ T cells. This model links tumor-intrinsic metabolic programs to the differentiation trajectory of tumor-infiltrating CD8⁺ T cells and highlights metabolic control of progenitor T cell states as a previously underappreciated mechanism of immune regulation in PDAC. DGAT1 loss also revealed divergent effects on lipid-handling pathways. In tumor cells, DGAT1 loss upregulated CD36, consistent with enhanced lipid scavenging, whereas in CD8⁺ T cells, CD36 expression was reduced. This suggests that DGAT1-dependent lipid rewiring drives cell-type-specific adaptations to lipid availability. Deciphering these regulatory circuits will be important for designing interventions that selectively disrupt tumor lipid metabolism while preserving T cell fitness. Importantly, pharmacologic inhibition of DGAT1 phenocopied genetic ablation and synergized with PD-1 blockade to induce complete tumor regressions in immunotherapy-refractory PDAC. This highlights DGAT1 as a druggable metabolic checkpoint with translational potential in immuno-oncology. While DGAT1 inhibitors were originally developed for metabolic disease73–75, our data suggest their repurposing in cancer, particularly in lipid-rich tumors where fatty acid overload enforces immune dysfunction. Additionally, DGAT1 knockout mice are viable, fertile, and normal in size, with no gross physical or behavioral abnormalities76, suggesting that targeting DGAT1 in PDAC could be a feasible therapeutic strategy without significant systemic toxicity. Further optimization of DGAT1 inhibitors, along with combination therapies targeting complementary metabolic vulnerabilities, will be critical for improving therapeutic outcomes. Future studies will focus on refining dosing regimens and delivery methods for DGAT1 inhibitors in patient-derived xenografts and clinical settings. Together, this study establishes a paradigm in which tumor-intrinsic lipid metabolism dictates the immune set-point of the TME. By redirecting lipid flux, tumors impose FFA overload and silence CD8⁺ T cell immunity. Conversely, DGAT1 inhibition relieves this metabolic brake, restores stem-like states, and sensitizes tumors to checkpoint blockade (Fig. S9). Beyond PDAC, our findings underscore the broader principle that targeting tumor lipid metabolism represents a powerful means to rewire T cell differentiation, preserve T cell fitness, overcome immune resistance, and expand the efficacy of immunotherapy. Methods Ethics statement All animal experiments were approved and performed in accordance with guidelines and regulations implemented by the Animal Care and Use Committee of the Zhejiang University School of Medicine (#31411). All mice were bred and maintained in the specific-pathogen-free facilities of the Zhejiang University School of Medicine, with ambient temperature maintained at 22 °C, a relative humidity between 40 and 60% and on a 12 h light/dark cycle. Mice were euthanized upon reaching the institutional criteria for tumor size (20 mm in diameter) and declining health. All experiments involving clinical samples have been approved and performed in accordance with guidelines and regulations implemented by the Ethics Committee of the Second Affiliated Hospital, Zhejiang University School of Medicine (#IR2024398). Clinical specimens Cohort #1 consisted of 60 formalin-fixed, paraffin-embedded (FFPE) PDAC tissues and 10 adjacent normal tissues. Cohort #2 included independent tissue microarrays containing 78 FFPE pancreatic tissues. All tissues were obtained from patients with PDAC who underwent curative surgical resection. Written informed consent was obtained from all patients, and the study was conducted in accordance with the principles of the Declaration of Helsinki. Animal studies Nude mice (BALB/cNj-Foxn1nu/Gpt, 6–8 wks) were purchased from GemPharmatech. C57BL/6J (6–8 weeks) were purchased from Shanghai Laboratory Animals Center. KPCflox (Pdx1-Cre;LSL-KrasG12D/+;Trp53flox/flox)43,44 genetically engineered mouse model was described previously. OT-I (#003831) mice were purchased from the Jackson Laboratory. For the subcutaneous model, approximately 5 × 105 KC6141, 1 × 106 KPCmut, 5 × 105 KPCflox tumor cells were injected subcutaneously in 100 µL PBS. Tumor growth was monitored every 2–3 days using calipers, and volume was calculated by volume = (length × width2)/2. For antibody-based treatment, tumor-bearing mice received a 200 μg pre-treatment of anti-CD4 (clone GK1.5) or anti-CD8α antibody (clone 2.43) 3 days before tumor inoculation, followed by continued treatment every 3 days throughout tumor progression. For the orthotopic model, 2.5 × 105 cells resuspended in 50 μL PBS were surgically implanted into the pancreas. Mice were randomized into four indicated treatment groups once the tumor reached ~5.0 mm in length and treated with either A-922500 (80 mg/kg/day, oral gavage) or PD-1 antibody (200 μg per mouse, every 3 days). Tumor progression was monitored by palpation or ultrasound. At study endpoints, tumor or pancreas weights were measured at the study endpoints. For T cell transduction and adoptive T cell transfer, CD8⁺ T cells were isolated from the spleens of 8–10-week-old OT-I mice and activated for 24 h with OVA peptide (SIINFEKL, 1 mg/mL) or with plate-bound anti-CD3 (2.5 µg/mL) and anti-CD28 (2.5 µg/mL). For Foxo1 knockdown, activated CD8⁺ T cells were transduced twice on consecutive days using retroviral supernatants in the presence of polybrene (10 µg/mL; 2500 rpm, 90 min, 30 °C). After each transduction, viral supernatants were replaced with fresh culture medium, and cells were subsequently expanded. Transduction efficiency was evaluated by flow cytometry 48 h after the final transduction. For FFA treatment, activated CD8⁺ T cells were expanded in fresh T cell media and then pretreated with either BSA or palmitic acid (200 µM) for 3 additional days. For adoptive transfer experiments, mice (n = 6 per group) were injected subcutaneously in the right flank with 1 × 10⁶ OVA-expressing KPCmut tumor cells. Ten days after tumor implantation, tumor-bearing mice received an intravenous infusion of 2.5 × 10⁶ OT-I cells as indicated. Tumor growth was measured every 3 days. Cell lines The human pancreatic cell lines PANC-1, BxPC-3, AsPC-1, MIA PaCa-2 were obtained from the American Type Culture Collection (ATCC). The mouse pancreatic cell lines UN-KC-6141, KPCmut, and KPCflox were provided by Dr. Surinder K Batra (University of Nebraska), Dr. Haoqiang Ying (MDACC) and Dr. Cassian Yee (MDACC). All cell lines were maintained in a humidified atmosphere containing 5% CO2 at 37 °C and were regularly tested and verified to be mycoplasma negative using via conventional PCR. CRISPR-based Screens The metabolism and focused screens were performed as previously described77. Briefly, KPCmut-Cas9 clones with validated Cas9 expression were transduced with sgRNA libraries targeting ~950 mitochondria or metabolism-related genes36 (10 gRNAs per gene) at a MOI of 0.2 with 200-fold coverage. After selection with puromycin, KPCmut cells transduced libraries were either passaged in vitro or implanted subcutaneously in the flanks of C57BL/6 mice. Tumors were grown for 2–3 weeks and gDNAs were extracted with DNeasy Blood & Tissue Kit (QIAGEN). For in vitro culture, 1000× library representation was passaged in triplicate every 3 days for 7 passages (P7/21 days). Nested PCR was used to amplify the sgRNA region, and NGS was performed on Illumina HiSeq to determine sgRNA abundance. After trimming adaptor sequences using Cutadapt, gRNAs were mapped and normalized, and the significantly enriched or depleted sgRNAs between the in vivo (B6) and in vitro (P7) were identified by the MAGeCK algorithm78. A complete list of differential gene scores is provided in Supplementary Dataset 2. shRNA, sgRNA and cloning The shRNAs targeting the human DGAT1, mouse Dgat1, Cd36 and Foxo1 genes were cloned into pLKO.1-puro or pRV-MMLV-GFP lentiviral vectors. SgRNAs specific for the mouse Dgat1 gene were designed using an online CRISPR design tool (F. Zhang lab, MIT). SgRNA sequences were cloned into pLentiCRISPR-v2 using the standard protocol. The cDNA clones expressing human (#Z9248) and mouse (#Mm02130) DGAT1 were obtained from GeneCopoeia. Human or mouse DAGT1 mutants were generated from site-specific mutagenesis of DGAT1 (Vazyme) or via overlapping PCR. Plasmids were sequenced to confirm the mutations. Plasmids were purified by the SANGON Miniprep Kit or NucleoBond Xtra Midi kit (MACHEREY-NAGEL). Specific primers are included in Supplementary Dataset 9. Flow cytometry Tumor tissues were cut into small pieces and digested with collagenase type IV (Worthington #LS004189) and deoxyribonuclease I (Roche) at 37 °C for 30 min with gentle shaking at 150 rpm. The digested tissue was subsequently ground through 70-μm filters into single-cell suspensions. Single-cell tumor samples were resuspended in red blood cell (RBC) lysis buffer and washed twice in cold PBS buffer. Then, single cell suspensions were incubated with Fc receptor-blocking anti-CD16/32 (BioLegend) for at 4 °C 15 min before staining. Cell suspensions were first stained with Zombie Aqua fixable viability kit (BioLegend) following the manufacturer’s instructions. Surface markers were then stained in FACS buffer (PBS containing 2% FBS and 0.1% sodium azide) at 4 °C for 30 min. To detect cytokine production ex vivo, cells were stimulated in the presence of phorbol 12-myristate 13-acetate (PMA) and ionomycin (Sigma-Aldrich) at 37 °C for 4 h. Brefeldin A (Sigma-Aldrich) was added 2 h before staining intracellular cytokines with BD Fixation/Permeabilization Solution Kit. For transcriptional factor staining, cells were fixed and permeabilized with Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher) following the manufacturer’s instructions. Stained cells were analyzed on a BD LSRFortessa flow cytometer, and data were analyzed and illustrated by using FlowJo software (BD Biosciences). Cell sorting was performed on a BD Aria Fusion cell sorter (BD Biosciences), and the purity of the sorted cell populations exceeded 95%. CD8+ T cell isolation and culture To obtain tumor-infiltrating lymphocytes or CD8+ T cells, mouse tumor tissues were minced and digested with Collagenase type IV and DNase I at 37 °C for 30 min. Cells were filtered through 70 μm cell strainers and then washed twice with staining buffer (PBS supplemented with 1 mM EDTA and 2% FBS). Red blood cells were lysed with ACK lysis buffer (Solarbio) for 5 min. CD8+ T cells were first enriched with CD8 MicroBeads (Miltenyi) and then sorted with BD Aria Fusion II cell sorter (BD). Naive CD8+ T cells were obtained by mashing the spleens through a 70 µm nylon cell strainer (BD) and then purified using the Naive CD8+ T Cell Isolation Kit (Miltenyi). For in vitro T cell culture, naive CD8+ T cells were cultured in RPMI 1640 medium containing 10% fetal bovine serum, 1% penicillin-streptomycin, 2mM L-glutamine and 50 U/ml IL-2 (Novus Biologicals) and treated with plate-bound anti-CD3 BioLegend) and anti-CD28 (BioLegend). At 48 h post activation, CD8+ T cells were then treated with anti-CD3 and tumor cell-derived culture supernatant or indicated fatty acid supplements for another 5–10 days. T cells were fed with fresh T cell medium supplemented with either conditional media or fatty acids every 48–72 h and were maintained at a density of 0.5 × 106 cells per ml after feeding. To detect cytokine production in vitro, activated CD8+ T cells were re-stimulated with PMA/Ionomycin and BFA followed by surface and intracellular staining as described above. Fatty acid uptake, neutral lipid content, lipid peroxidation, and mitochondrion staining assays For fatty acid and cholesterol uptake assays, cells were incubated in PBS containing 0.1 mg/mL BODIPY FL C16 or 10 µM NBD cholesterol (Thermo Fisher) for 20 min at 37 °C. Following incubation, cells were washed with FACS buffer before surface staining. For neutral lipid content detection, cells were fixed and permeabilized before staining with BODIPY 493/503 (Thermo Fisher) at a final concentration of 250 ng/mL. For lipid peroxidation measurement, cells were incubated in a humidified chamber at 37 °C with 5% CO₂ for 30 min in culture medium containing the Lipid Peroxidation Sensor (Abcam). After incubation, cells were washed and analyzed by flow cytometry within 2 h. For mitochondrial membrane potential assessment, cells were washed and incubated in PBS containing 100 nM Tetramethylrhodamine (Thermo Fisher) for 15 min. After staining, cells were washed and resuspended in fresh FACS staining buffer before surface marker staining as described above. Transmission electron microscopy Enriched CD8+ T cells were fixed in glutaraldehyde, washed, and incubated in osmium tetroxide. After dehydration in ethanol and propylene oxide, cells were infiltrated with Polybed 812 resin. Ultrathin sections (70–80 nm) were cut, mounted on copper grids, and stained with uranyl acetate and Reynold’s lead citrate. Imaging was performed using a Tecnai G2 spirit transmission electron microscope at 120 kV. Lipid-stripped serum Lipid-stripped serum was prepared following a previously described method79. Briefly, FBS was mixed with 0.8 volumes of di-isopropyl ether and 0.2 volumes of n-butanol and incubated at room temperature for 30 min. Phase separation was achieved by centrifugation at 4000 × g for 15 min at 4 °C. The lower aqueous phase was then extracted again by mixing with an equal volume of di-isopropyl ether for 30 min at RT. After collecting the lower aqueous phase, the solution was stirred under a slow stream of nitrogen gas overnight and subsequently dialyzed against four changes of saline at 4 °C. Control serum containing lipids underwent the same dialysis process. The protein concentrations of both lipid-stripped and control sera were quantified using the Bradford assay, adjusted with saline as needed, and stored at −80 °C. Lipid extraction Lipids were extracted using a chloroform/methanol (2:1, v/v) mixture as previously described80. Briefly, cell pellets were lysed by bead-beating, followed by chloroform and methanol addition. Phase separation was induced with citric acid, and the lower lipid phase was collected and dried using a vacuum concentrator (Labconco). For interstitial fluid (TIF) and conditioned medium, lipids were extracted from 50 µL TIF or 1 mL culture supernatant. TIF was collected as previously described81 by isolating tumors, centrifuging through 70 µm strainers (4 °C, 10 min, 300 × g). The collected liquid was further centrifuged at 10,000 × g for 10 min at 4 °C to remove any tissue debris and then freeze at −80 °C until metabolomics analysis. Lipids measurement by LC-MS/MS Dried lipid extracts were reconstituted in a mixture of isopropanol, acetonitrile, and water (2:1:1, v/v/v). Quality control (QC) samples were prepared by pooling equal volumes of all samples. Lipidomix (Avanti Polar Lipids) was used as an internal standard for phospholipid quantification. Lipid analysis was performed using a targeted LC-MS/MS approach. A 5-µL aliquot of each sample was injected and separated on an Acquity UPLC BEH C18 column (130 Å, 1.7 μm, 2.1 mm × 50 mm). Mobile phase A consisted of 33.3% acetonitrile, 33.3% methanol, 33.4% water, and 5 mM ammonium acetate, while mobile phase B consisted of 100% isopropanol containing 5 mM ammonium acetate. A linear gradient was applied at a flow rate of 0.15 mL/min as follows: 0–1 min, 20% B; 1–3 min, 20–60% B; 3–20 min, 60–98% B; 20.1–25 min, 98–20% B. Lipid detection and quantification were performed using a QTRAP 6500+ System (AB SCIEX) in either positive or negative ion mode, depending on the lipid class. Multiple reaction monitoring (MRM) transitions were used to detect specific lipids, including phospholipids, diacylglycerol (DAG), triacylglycerol (TAG), and sphingomyelin (SM). Peak areas were integrated and quantified using Analyst software (v1.7), with retention times confirmed against appropriate standards. In vitro [U-13C16]palmitic acid tracer experiments Cells were cultured in 6-well plates in DMEM supplemented with 5% lipid-free fetal bovine serum and 0.2 mM [U-13C16]palmitic acid. After 6 h of incubation, media samples were collected and dried using a vacuum concentrator (Labconco). Cells were washed with pre-warmed PBS and harvested by scraping into 0.5 mL of MS-grade methanol containing internal standards (PC 17:0, 1.97 μM; PE 17:0, 2.08 μM; PS 14:0, 2.14 μM). Samples were snap-frozen in liquid nitrogen. After lipid extraction, dried lipid extracts were reconstituted in a mixture of isopropanol, acetonitrile, and water (2:1:1, v/v/v). Quality control (QC) samples were prepared by pooling equal volumes of all samples. Lipidomix (Avanti Polar Lipids) was used as an internal standard for phospholipid quantification. Lipid analysis was performed using a targeted LC-MS/MS approach. A 5-µL aliquot of each sample was injected and separated on an Acquity UPLC BEH C18 column (130 Å, 1.7 μm, 2.1 mm × 50 mm). Mobile phase A consisted of 33.3% acetonitrile, 33.3% methanol, 33.4% water, and 5 mM ammonium acetate, while mobile phase B consisted of 100% isopropanol containing 5 mM ammonium acetate. A linear gradient was applied at a flow rate of 0.15 mL/min as follows: 0–1 min, 20% B; 1–3 min, 20–60% B; 3–20 min, 60–98% B; 20.1–25 min, 98–20% B. Lipid detection and quantification were performed using a QTRAP 6500+ System (AB SCIEX) in either positive or negative ion mode, depending on the lipid class. Multiple reaction monitoring (MRM) transitions were used to detect palmitoyl-linked lipids, including phospholipids, TAG, DAG, and SM, were quantified by LC-MS/MS. A 10-µL aliquot of each sample was injected for analysis. MRM transitions for [U-13C16]palmitoyl-containing lipids were employed, with retention times matched to corresponding unlabeled species using spike-in standards, based on previous studies82. Peak areas were integrated and quantified using Analyst software (v1.7), with retention times confirmed against appropriate standards, based on previous studies82. Single-cell RNA-seq library preparation Tumor-infiltrating CD8+ T cells were enriched using CD8 MicroBeads (Miltenyi) and subsequently sorted using a BD Aria Fusion cell sorter (BD Biosciences). Single-cell suspensions (2 × 10⁵ cells/mL) were pooled and loaded onto a microwell chip using the Singleron Matrix Single Cell Processing System. Barcoding beads were then collected from the microwell chip, followed by reverse transcription of mRNA captured by the beads to generate cDNA, which was subsequently amplified by PCR. The amplified cDNA was fragmented and ligated with sequencing adapters. scRNA-seq libraries were constructed following the GEXSCOPE Single-Cell RNA Library Kit (Singleron) protocol. Individual libraries were diluted to 4 nM, pooled, and sequenced on an Illumina NovaSeq 6000 with 150 bp paired-end reads. ATAC-seq library preparation ATAC-seq libraries were prepared with Hyperactive ATAC-Seq Library Prep Kit (Vazyme) following the manufacturer’s instructions. Briefly, nuclei were isolated using a lysis buffer and nuclei were pelleted and resuspended in the transposition mix at 37 °C for 30 min. DNA was then purified using ATAC DNA Extract Beads. Libraries were barcoded using the TruePrep Index Kit V2 for Illumina and initially amplified for ten cycles. The required number of additional PCR cycles was determined by real-time PCR. The final product was purified and size-selected using ATAC DNA Extract Beads. Libraries were sequenced on a NovaSeq 6000 with 150 bp paired-end reads. Bulk RNA-seq analysis and data processing Raw paired-end reads were trimmed to remove adapter sequences using Trim Galore (v0.6.6). The cleaned FASTQ files were then mapped to the mm10 reference genome using STAR (v2.7.10a). Reads from specific chromosomal regions with annotations were quantified using HTSeq-count (v0.13.5). Differential gene expression analysis was performed using DESeq2 (v1.38.3). Gene set enrichment analysis (GSEA) (https://www.gsea-msigdb.org/gsea/index.jsp) was conducted on datasets from the MSigDB database. Genes were pre-ranked based on RNA-seq expression counts and analyzed using GSEA software, employing the “Signal2Noise” ranking metric. The permutation type was set to “gene set” for smaller sample sizes and “phenotype” for larger sample sizes. Gene set enrichment results were ranked by enrichment score in descending order. Single-cell RNA-seq data processing Digital gene expression matrices were generated for each sample using the CeleScope pipeline. Pre-processing and quality control were performed using Scanpy (v1.10.3) with default parameters. Cells with fewer than 200 detected genes, more than 6000 detected genes, or >10% of reads mapping to mitochondrial RNA were excluded. Genes detected in fewer than three cells were also removed. WT and KD groups were normalized and log-transformed, followed by identification and removal of non-highly variable genes. The effects of total counts per cell and mitochondrial gene percentage were regressed out, and the data were scaled to unit variance. Dimensionality reduction was performed using principal component analysis (PCA), and a neighborhood graph was constructed using 10 neighbors and 40 PCs. Clustering was performed using the Leiden algorithm, and graph embedding was visualized in two dimensions using PAGA and UMAP. For data integration, cluster labels and UMAP coordinates from WT were mapped to KD using WT.obsm[‘X_pca’]. Batch effects between WT and KD were corrected using PCA followed by BBKNN. The final graph was embedded in two dimensions using UMAP, and clustering was refined using the Leiden algorithm. Marker genes for cluster annotation were identified using the Wilcoxon rank-sum test with “tl.rank_genes_groups.” Gene set scoring was performed using “tl.score_genes.” RNA velocity and pseudotime analyzes were conducted using Velocyto and scVelo (v0.3.2) to infer cell state transitions in CD8+ T cells. Transcription factor analysis for Tpex in the KD group was performed using pySCENIC (v0.12.1). Integration of public PDAC scRNA-seq datasets Publicly available scRNA-seq datasets from four independent PDAC studies (PMID: 31273297, 34296197, 37914939, 36781852) were downloaded and processed using Scanpy. After standard quality control filtering, datasets were integrated using batch-correction to minimize inter-study technical variation while preserving biological differences. Major cell types were annotated based on unsupervised clustering and established marker genes. CD8⁺ T cells were extracted and reprocessed independently. Progenitor-like and exhausted transcriptional programs were quantified using previously defined gene signatures derived from LCMV tumor models34. Gene set scores were calculated at the single-cell level using scanpy.tl.score_genes, generating continuous progenitor and exhaustion scores for each CD8⁺ T cell. All CD8⁺ T cells were included in this analysis, without further sub-clustering or manual selection of progenitor or exhausted subsets. Tumor cell pathway activity scoring and correlation analysis Tumor cells were identified based on epithelial markers, and Hallmark pathway activities were quantified using MSigDB gene sets. Pathway scores were calculated at the single-cell level using scanpy.tl.score_genes. For each sample, median progenitor or exhaustion scores across all CD8⁺ T cells and median pathway scores across all tumor cells were computed. Spearman rank correlations were then performed at the sample level between tumor cell pathway activities and CD8⁺ T cell differentiation scores, with statistical significance assessed using two-sided tests. Immunohistochemistry Paraffin-embedded human and mouse pancreatic tissues were sectioned at 5 µm and processed using standard immunohistochemistry (IHC) protocols. Sections were deparaffinized, rehydrated, and subjected to antigen retrieval. After blocking with 5% goat serum (1 h, RT), slides were incubated overnight at 4 °C with primary antibodies in a humidified chamber. Secondary antibody (HRP polymer, anti-mouse/rabbit IgG) incubation followed for 30 min at room temperature. Fluorophore labeling was performed using tyramide signal amplification (TSA), with microwave treatment between sequential antigen labeling steps. Nuclei were counterstained with DAPI, and positive signals were visualized via confocal microscopy (Leica Stellaris 8). Statistical analysis Statistical analyzes were conducted using GraphPad Prism 7. Data are represented as mean ± SD, unless otherwise specified. Comparisons between two independent groups were performed using an unpaired, two-tailed Student’s t-test, assuming equal variance with no sample exclusions. For multiple group comparisons, one-way or two-way ANOVA was used as indicated. Survival analyzes were conducted using the Kaplan–Meier method and assessed with the log-rank (Mantel–Cox) test. A p < 0.05 was considered statistically significant. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Supplementary Dataset 5 Supplementary Dataset 6 Supplementary Dataset 7 Supplementary Dataset 8 Supplementary Dataset 9 Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements We thank S.K. Batra (University of Nebraska), C. Yee and H. Ying (MD Anderson Cancer Center), D. Tuveson (Cold Spring Harbor Laboratory) for providing murine PDAC cell lines. We thank X. Zeng, W. Sheng, P. Qian, Y. Guo, M. Lai, J. Shao, H. Zhang, X. Wang (all from Zhejiang University) for providing experimental materials or technical assistance. We acknowledge the Core Facilities of Zhejiang University School of Medicine (Y. Huang and Y. Li) for technical assistance. Author contributions Y.H.L. conceived the ideas; Y.H.L. and C.Q.Y. designed the experiments and wrote the manuscript; Z.K.L., L.C., T.Z., P.P.M., X.W.Z., Q.Q.S., X.H.Z., R.Y.Z., and R.H.L. performed the experiments and statistical analyses; Y.L.W. performed bioinformatics analysis; T.Z. conducted lipidomics; Y.H.L., C.Q.Y., Q.Z., D.W., J.Y.H., Z.K.L., L.C., T.Z. and Y.L.W. analyzed data; Q.Z. and D.W. provided suggestions and patient samples. All authors reviewed and approved the manuscript for publication. Peer review Peer review information Nature Communications thanks Shailendra Gautam and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Funding YHL discloses support for the research of this work from Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0500300 and 2024ZD0524900), National Natural Science Foundation of China (82072615 and 82273274), and Key Research and Development Program of Zhejiang Province (2024C03143). QZ discloses support for the research of this work from Major Research Plan of the National Natural Science Foundation of China (92474301). DW discloses support for the research of this work from National Natural Science Foundation of China (82303951), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0502900 and 2025ZD0544500) and Key Research and Development Program of Zhejiang Province (2024C03143 and 2025C02197). Data availability The scRNA-seq, ATAC-seq and bulk RNA-seq data for this study have been deposited at National Genomics Data Center under GSA: CRA023668 and HRA018275. Public datasets used for reanalysis in this study are available under the following accession codes: GSE205013; GSE217847; CRA001160; GSE155698 and GSE16515. All other data generated in this study are provided in the Supplementary Information and Source data files. Source data are provided with this paper. Code availability Codes used for the analyses is available at https://github.com/YuliangWang316/LYH_Dgat1. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Zhongkun Liu, Lang Chen, Tong Zhang, Yuliang Wang. These authors jointly supervised this work: Da Wang, Qi Zhang, Cunqi Ye, Yun-Hua Liu. Contributor Information Da Wang, Email: wangda0618@zju.edu.cn. Qi Zhang, Email: qi.zhang@zju.edu.cn. Cunqi Ye, Email: yecunqi@zju.edu.cn. Yun-Hua Liu, Email: yunhualiu@zju.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74315-2. References Associated Data Supplementary Materials Supplementary Information Description of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Supplementary Dataset 5 Supplementary Dataset 6 Supplementary Dataset 7 Supplementary Dataset 8 Supplementary Dataset 9 Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The scRNA-seq, ATAC-seq and bulk RNA-seq data for this study have been deposited at National Genomics Data Center under GSA: CRA023668 and HRA018275. Public datasets used for reanalysis in this study are available under the following accession codes: GSE205013; GSE217847; CRA001160; GSE155698 and GSE16515. All other data generated in this study are provided in the Supplementary Information and Source data files. Source data are provided with this paper. Codes used for the analyses is available at https://github.com/YuliangWang316/LYH_Dgat1.