GPAT4 regulates antitumor immune response through activating the mtDNA-cGAS axis in cancer cells Authors: Zi-Lun Ruan, Wei Liang, Yu-Tong Gong, Wei-Tao Guan, Xuan-Ang Zhou, Kai Jiang, Qing Yang, Ming-Ming Hu Published: 2026 ABSTRACT AbstractActivating tumor-intrinsic innate immune pathways can promote tumor antigen presentation and immune cell recruitment, thereby improving the immunosuppressive microenvironment and overcoming resistance to immune checkpoint blockade (ICB). However, the specific molecular targets remain insufficiently defined. Bioinformatic analysis revealed that Glycerol-3-phosphate acyltransferase 4 (GPAT4) is highly expressed in various tumors, including pancreatic cancer and breast cancer, and is significantly associated with low immunogenicity, poor immune-cell infiltration, and unfavorable prognosis. Syngeneic C57BL/6J and BALB/c tumor-bearing experiments showed that Gpat4-deficient pancreatic and breast tumors grew more slowly and exhibited increased infiltration of CD8+ T lymphocytes and enhanced CD8+ T cell effector-marker expression. Mechanistic studies indicated that GPAT4 deficiency leads to mitochondrial stress and the release of mitochondrial DNA (mtDNA) into the cytoplasm, thereby activating the cyclic GMP-AMP synthase (cGAS) signaling pathway. Interestingly, this process is independent of the metabolic enzyme activity of GPAT4. Single-cell analysis of breast and pancreatic tumors revealed that tumors with low GPAT4 expression exhibit higher levels of interferon response pathway activity and greater infiltration of immune cells, particularly CD8+ T cells and natural killer (NK) cells, which directly mediate tumor killing. In conclusion, our study identifies GPAT4 as a critical innate immune checkpoint molecule in tumor cells. Targeting this molecule can activate tumor-intrinsic innate immune pathways, thereby enhancing antitumor immune responses.Keywords: Antitumor immunity, GPAT4, Mitochondrial DNA, cGAS-mediated innate immunity, Bioinformatics Introduction Immune checkpoint blockade-based tumor immunotherapy has achieved groundbreaking clinical progress in various malignancies in recent years (Havel et al., 2019; Sharma et al., 2017). However, a considerable proportion of patients still fail to benefit due to primary or acquired resistance (Pitt et al., 2016; Sharma et al., 2017). The development of resistance to immunotherapy involves complex multi-dimensional mechanisms, in which tumor-cell-intrinsic factors play a key role. These include epigenetic silencing of the antigen presentation machinery, upregulation of PD-L1 driven by oncogenic signaling pathways, and genetic deficiencies in core components of the IFN-γ signaling pathway (such as JAK1/2 or B2M) (Gao et al., 2016; Kalbasi & Ribas, 2020). These tumor-intrinsic alterations lead to low tumor immunogenicity, making it difficult for the immune system to effectively recognize and eliminate the tumor. Concurrently, factors within the tumor microenvironment, such as T cell exhaustion, enrichment of regulatory T cells, and accumulation of immunosuppressive metabolites, further exacerbate the immune escape state (DePeaux & Delgoffe, 2021; Galluzzi et al., 2025; Togashi et al., 2019; Zhang et al., 2024). Effectively activating the tumor's intrinsic innate immune sensing mechanisms, promoting tumor antigen presentation, and facilitating the infiltration and activation of immune effector cells have become key scientific questions for improving immunotherapy response rates (Binnewies et al., 2018; Schreiber et al., 2011). Tumor-cell-intrinsic innate immune sensing mechanisms, particularly the DNA-sensing cGAS-MITA/STING pathway (Hu & Shu, 2018, 2020, 2023; Ishikawa & Barber, 2008; Zhong et al., 2008), play central roles in recognizing cellular stress, inducing type I interferon responses, and remodeling the immune microenvironment (Bakhoum et al., 2018; Hopfner & Hornung, 2020; Zhang & Zhong, 2022). When cells experience genomic instability, mitochondrial stress, or therapy-induced DNA damage, nuclear DNA or mitochondrial DNA leaks into the cytoplasm and is recognized by cGAS (Li & Chen, 2018; Lin et al., 2023; Liu & Xu, 2025; West et al., 2015). Activation of the cGAS-MITA pathway exerts multiple promoting effects in antitumor immune responses. Type I interferons activate the JAK-STAT pathway through autocrine and paracrine mechanisms, inducing the expression of hundreds of interferon-stimulated genes (ISGs) (Burnette et al., 2011; Zitvogel et al., 2015). At the innate immune level, IFN-I enhances the cytotoxic function of natural killer (NK) cells—mtDNA released by tumor cells can be partially recognized by cGAS, triggering endogenous MITA activation in NK cells, thereby maintaining the antitumor activity of TCF-1+ NK cell subsets with long-term memory potential (Lu et al., 2023; Marcus et al., 2018). Simultaneously, MITA activation enhances the tumor-killing capacity of NK cells by promoting NLRP3 inflammasome signaling (Lu et al., 2023; Sun et al., 2023). At the adaptive immunity level, IFN-I promotes the maturation and antigen-presenting capacity of dendritic cells (DCs), enhances DC cross-presentation of tumor-associated antigens, and thereby efficiently initiates the activation and expansion of tumor antigen-specific CD8+ T cells. Activated cytotoxic CD8+ T lymphocytes induce tumor cell apoptosis by releasing perforin and granzymes (Aliazis et al., 2025; Fuertes et al., 2011). Furthermore, MITA signaling can also promote the differentiation of CD4+ T cells into Th1 and Th9 cells: Th1 cells secrete IFN-γ to directly kill tumor cells, while Th9 cells secrete IL-9 to exert antitumor effects (Benoit-Lizon et al., 2022; Kaplan et al., 2015). At the chemokine level, MITA activation induces the expression of chemokines such as CCL5 and CXCL10, which further recruit CD8+ T cells and NK cells to infiltrate the tumor microenvironment (Mowat et al., 2021; Withers et al., 2023). Therefore, the cGAS-MITA pathway builds a multi-layered antitumor immune response network by bridging innate and adaptive immunity. Glycerol-3-phosphate acyltransferase 4 (GPAT4) is one of the rate-limiting enzymes catalyzing the first step in the synthesis of triglycerides and phospholipids. It is localized on the endoplasmic reticulum (ER) membrane and catalyzes the conversion of glycerol-3-phosphate to lysophosphatidic acid (LPA) (Coleman & Lee, 2004; Nagle et al., 2008; Yu et al., 2018; Yung et al., 2014). Recent studies have revealed multiple roles of GPAT4 in tumor development and progression. In ovarian cancer, high GPAT4 expression is significantly correlated with platinum resistance and poor patient prognosis (Kapper et al., 2024; Yu et al., 2025). The mechanism may involve GPAT4 influencing ferroptosis sensitivity by regulating lipid metabolism, thereby inhibiting lipid peroxidation and iron accumulation, thus protecting tumor cells from ferroptosis-induced killing (Huang et al., 2024; Zhang et al., 2022). In colorectal cancer, downregulation of CPT2 can enhance glycerophospholipid synthesis (such as phosphatidylcholine and phosphatidylethanolamine) mediated by long-chain fatty acids via GPAT4. The accumulation of these glycerophospholipids, which are key components of vesicles and autophagosomes, promotes autophagosome maturation and extension, thereby activating selective autophagy (lipophagy) and accelerating tumor progression (Li et al., 2025). Notably, recent studies in non-tumor fields have found that GPAT4 deficiency can induce endoplasmic reticulum stress, enhance ER-mitochondria communication, lead to mitochondrial dysfunction, and cause the release of mtDNA into the cytoplasm, subsequently activating the cGAS-MITA pathway (Zhao et al., 2025). This discovery provides an important clue for understanding the potential role of GPAT4 in innate immune regulation. However, whether GPAT4 participates in regulating the tumor immune microenvironment and its underlying mechanisms remain unclear. To further investigate the role of GPAT4 in tumor immune regulation and its molecular mechanisms, this study systematically investigated the effects of GPAT4 deficiency on tumor growth, the immune microenvironment, and mtDNA homeostasis using a combination of in vivo and in vitro models, multi-omics analysis, and functional experiments. The study found that GPAT4 deficiency can induce mitochondrial stress, promote the release of mtDNA into the cytoplasm, and subsequently activate the cGAS-mediated signaling axis, enhancing tumor immunogenicity and recruiting CD8+ T cells and NK cells. Notably, this regulatory effect is independent of the metabolic enzyme activity of GPAT4, suggesting it possesses a non-canonical function. Further analysis using single-cell transcriptome sequencing confirmed that low GPAT4 expression correlates positively with intratumoral interferon response activity and the enrichment of immune effector cells. In summary, this study reveals for the first time that GPAT4 functions as a critical innate immune checkpoint molecule in tumor cells, negatively regulating the activity of the cGAS-MITA signaling pathway and suppressing antitumor immune responses. Targeting GPAT4 may represent a novel therapeutic strategy to activate tumor-intrinsic immune sensing capabilities, improve the immunosuppressive microenvironment, and enhance the efficacy of immunotherapy. Results GPAT4 is aberrantly expressed across multiple cancer types and is associated with poor prognosis and an immune-cold phenotype To determine whether GPAT4 is broadly dysregulated in human malignancies, we compared GPAT4 mRNA expression between tumor and normal tissues across a pan-cancer cohort (Fig. 1(A)). GPAT4 was upregulated in tumors from 14 cancer types, including cholangiocarcinoma (CHOL), brain lower grade glioma (LGG), pancreatic adenocarcinoma (PAAD), and breast invasive carcinoma (BRCA), whereas 9 cancer types did not show this pattern (Fig. 1(A)). Thus, although GPAT4 dysregulation was heterogeneous across tumor types, the overall pan-cancer trend favored increased expression in tumors. Pan-cancer analysis of GPAT4 expression, prognostic relevance, cellular dependency, and immune-infiltrationfeatures. (A) GPAT4 mRNA abundance was compared between tumor tissues and matched or reference normal tissues across the indicated pan-cancer cohorts, and the box plots summarize the distribution in each cancer type, with the annotation strip above the panel indicating whether GPAT4 was significantly increased, significantly decreased, or not significantly changed in tumors. (B) Overall-survival associations were evaluated by stratifying patients into high- and low-GPAT4-expression groups within each survival-evaluable cancer type and calculating hazard ratios; points denote hazard ratios and horizontal bars denote 95% confidence intervals. (C) GPAT4 Chronos dependency scores from DepMap were compiled across cancer cell lines from multiple lineages, with each point representing one cell line and the dashed line marking a gene-effect score of 0. (D) Correlations between GPAT4 expression and CIBERSORT-derived immune-cell signatures were calculated across cancer types; dot color indicates the direction and strength of the association, and the red line represents the median correlation between the indicated TIL score and GPAT4 expression. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05. We next evaluated the clinical relevance of this expression pattern by examining the association between GPAT4 expression and the overall survival hazard ratio in individual cancer types. Higher GPAT4 expression tracked with increased mortality risk in 14 cancer types, including PAAD, kidney chromophobe (KICH), BRCA, and adrenocortical carcinoma (ACC), whereas 7 cancer types showed an association in the opposite direction and the remaining cancer types showed no significant relationship (Fig. 1(B), Fig. S1). These findings indicate that the prognostic value of GPAT4 is cancer-type dependent, but in the two disease settings emphasized in this study, PAAD and BRCA, elevated GPAT4 expression consistently pointed to worse outcome. To assess whether the association between GPAT4 and tumor progression mainly reflected a tumor-cell-intrinsic proliferation requirement, we examined the distribution of Chronos gene-effect scores of GPAT4 across cancer cell lines from multiple tissue origins. In most lineages, the dependency score distribution was centered near 0, with no broad or marked shift toward negative values across cancer types (Fig. 1(C)). These data suggest that GPAT4 is not a core survival gene that is widely required across tumor cells, further implying that therapeutic targeting of GPAT4 may not be expected to cause broad toxicity. Next, we assessed the connection between GPAT4 and the tumor immune microenvironment. We compared correlations between GPAT4 expression and CIBERSORT-derived tumor-infiltrating lymphocyte signatures across cancer types. Immune populations with central antitumor functions, particularly CD8+ T cells and activated NK cells, showed a significant negative correlation with GPAT4 expression (Fig. 1(D)). These results indicate that high GPAT4 expression is associated with reduced infiltration of effector lymphocytes and a more pronounced immunosuppressive state across cancers. Together, these pan-cancer analyses show that GPAT4 is aberrantly expressed in multiple tumor types and that its high expression is closely associated with poor prognosis and an immune-cold phenotype. By contrast, GPAT4 does not exhibit broad tumor-cell growth dependency, supporting the possibility that it contributes to tumor progression mainly by shaping the tumor immune microenvironment and highlighting its potential translational value as a therapeutic target. Gpat4 deficiency suppresses pancreatic and breast tumor growth and increases antitumor effector lymphocyte infiltration Given that the pan-cancer analyses linked high GPAT4 expression to an immune-cold phenotype and poor outcome, we next asked whether GPAT4 influences tumor growth and the tumor immune microenvironment. To address this question, we generated control and CRISPR-Cas9-edited Gpat4-deficient Panc02 cells and validated reduced Gpat4 transcript abundance by quantitative reverse-transcription PCR (qRT-PCR) before implantation (Fig. 2(A) and (B)). Because a validated antibody suitable for detecting endogenous Gpat4 in these cells was not available, qRT-PCR was used as a downstream validation readout of CRISPR-Cas9-mediated genome editing rather than as evidence of direct mRNA knockdown. Gpat4 transcript abundance was markedly reduced in the edited population (Fig. 2(B)). Gpat4 deficiency suppresses pancreatic and breast tumor growth and increases antitumor effector-lymphocyte infiltration and function. (A) Schematic of the syngeneic Panc02 transplantation experiments. Control or CRISPR-Cas9-edited Gpat4-deficient Panc02 cells were implanted subcutaneously into C57BL/6J or nude mice, followed by longitudinal tumor-volume monitoring, and endpoint collection for measurement of tumor weight and volume and flow-cytometric analysis. (B) qRT-PCR verification of reduced Gpat4 transcript abundance in Panc02 cells after CRISPR-Cas9 editing before implantation. (C) Tumor-growth curves for subcutaneous Panc02 allografts in immunocompetent C57BL/6J mice. (D) Representative endpoint Panc02 tumors isolated from control and Gpat4-deficient groups. (E, F) Quantification of terminal tumor volume and tumor weight in C57BL/6J mice and nude mice implanted with the same Panc02 derivatives. (G) Flow-cytometric analysis of Panc02 tumor-infiltrating lymphoid populations and CD8+ T-cell functional markers, including CD3+, CD4+, CD8+ and NK-cell fractions as well as Granzyme B+(GZMB), IFN-γ+ and CD69+ CD8+ T cells. (H) Schematic of the syngeneic 4T1 transplantation experiment in BALB/c mice. Control or Gpat4-deficient 4T1 cells were implanted subcutaneously and tumors were monitored until endpoint, and collected for flow-cytometric analysis. (I) Representative endpoint 4T1 tumors. (J) Tumor-growth curves for subcutaneous 4T1 allografts. (K) Quantification of terminal tumor weight and tumor volume in the 4T1 model. (L) Flow-cytometric analysis of live CD45+ cells and CD3+, CD4+ and CD8+ T-cell fractions in 4T1 tumors. n = 8 mice per group for the Panc02 C57BL/6J, Panc02 nude-mouse, and 4T1 BALB/c cohorts. Points represent individual mice or tumors and bars denote mean ± s.e.m.; exact P values are shown where provided and otherwise significance is denoted as ns or by asterisks. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.000 1. We first implanted these cells into immune-competent C57BL/6J mice to assess the overall effect of tumor-cell-intrinsic Gpat4 deficiency on tumor growth. In this cohort (n = 8 per group), tumors derived from Gpat4-deficient cells grew more slowly and were smaller at endpoint, as shown by the growth curve, representative tumor images, and terminal tumor-volume and tumor-weight measurements (Fig. 2(C–E)). Endpoint tumor volume and tumor weight were both reduced in the Gpat4-deficient group (Fig. 2(E)). Flow cytometry further showed that the proportions of CD3+ T cells, CD4+ T cells, CD8+ T cells, and NK cells were increased in Gpat4-deficient tumors (Fig. 2(G)). To determine whether this increase reflected functional effector activation, we analyzed tumor-infiltrating CD8+ T cells and found increased proportions of Granzyme B+, IFN-γ+, and CD69+ CD8+ T cells in Gpat4-deficient tumors (Fig. 2(G)). These data indicate that Gpat4 deficiency not only restrains Panc02 tumor growth but is also accompanied by greater effector-lymphocyte infiltration and stronger CD8+ T cell effector-marker expression. To determine whether this effect depended on host immunity, we repeated the in vivo experiment in nude immunodeficient mice. In the corresponding nude-mouse cohort (n = 8 per group), terminal tumor volume and tumor weight did not differ significantly between the control and Gpat4-deficient groups (Fig. 2(F)). Compared with immune-competent hosts, immunodeficient hosts exhibited a markedly diminished tumor-suppressive effect. This pattern does not support a mechanism driven predominantly by tumor-cell-intrinsic growth impairment and instead indicates an important contribution of host immune participation. To test whether this immune-dependent phenotype extended to another immune-cold tumor model, we performed a parallel syngeneic experiment using CRISPR-Cas9-edited Gpat4-deficient 4T1 breast cancer cells in immunocompetent BALB/c mice. Control or Gpat4-deficient 4T1 cells (1 × 106 cells per mouse) were implanted subcutaneously, tumor growth was monitored from day 8 to day 31, and tumors were collected at endpoint (Fig. 2(H)). Consistent with the Panc02 model, Gpat4-deficient 4T1 tumors grew more slowly and showed lower terminal tumor weight and volume than control tumors (Fig. 2(I–K)). Immune profiling further showed increased infiltration of live CD45+ immune cells as well as CD3+ T cells, CD4+ T cells, and CD8+ T cells in Gpat4-deficient 4T1 tumors (Fig. 2(L)). Gpat4 deficiency showed a modest, non-significant trend toward increased CD8+TIM-3+ cells and NK-cell fractions in the 4T1 model (Fig. S2A). Together, these results support the generality of GPAT4-dependent immune restriction across pancreatic and breast cancer syngeneic models. GPAT4 deficiency activates interferon signaling and enhances cell-surface MHC-I expression in tumor cells Since GPAT4 deficiency suppressed tumor growth in immune-competent hosts and was accompanied by increased T cell and NK cell infiltration, we hypothesized that downregulation of GPAT4 may reprogram tumor-cell-intrinsic immune transcriptional states and thereby increase tumor immunovisibility. To test this possibility, we performed CRISPR-Cas9-mediated Gpat4 depletion in mouse breast 4T1 and mouse pancreatic Panc02 cells followed by bulk-mRNA sequencing (Fig. 3(A)). Principal-component analysis (PCA) of the resulting eight libraries (n = 2 biologically independent libraries per condition) showed clear separation by cell type and perturbation state, and RNA-seq quantification confirmed robust depletion of Gpat4 transcripts in both models (Fig. 3(B)). Gpat4 deficiency activates interferon-related programs and enhances cell-surface MHC-I expression in tumor cells. (A) Experimental workflow for bulk RNA-seq of control and Gpat4-deficient 4T1 and Panc02 cells. Independent cultures were harvested for library preparation and sequencing to define the transcriptional consequences of Gpat4 depletion. (B) Principal-component analysis of the eight RNA-seq libraries together with RNA-seq-based confirmation of reduced Gpat4 expression in the depleted groups. n = 2 biologically independent RNA-seq libraries per condition. (C) Heatmap of representative differentially expressed genes in 4T1 and Panc02 cells after Gpat4 depletion. (D) Gene-set-enrichment analysis showing enrichment of IFN-α response, IFN-γ response and antigen processing and presentation programs after Gpat4 depletion. (E, F) qPCR validation of representative interferon-stimulated genes in 4T1 and Panc02 cells, respectively. (G) Representative flow-cytometry histograms and corresponding quantification of cell-surface MHC-I abundance in control and Gpat4-deficient cells. (H) Pan-cancer HALLMARK-enrichment landscape associated with GPAT4 expression across tumor types, with dot color indicating normalized enrichment score and dot size reflecting enrichment significance. Bars denote mean ± s.e.m., histograms are representative of the distributions summarized in the adjacent quantification, and exact P values are shown where provided and otherwise significance is denoted as ns or by asterisks. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.000 1. We next compared differentially expressed genes across the two models and visualized representative changes in a heatmap (Fig. 3(C)). In both 4T1 and Panc02 cells, a concordant set of genes was upregulated, including Ifi44, Oas family members, Irf7, Ifit3, and Isg15 (Fig. 3(C)). These genes are predominantly associated with interferon-stimulated gene (ISG) and antiviral-response programs, indicating that Gpat4 depletion is accompanied by activation of innate immune transcriptional programs. To validate this trend at the pathway level, we performed gene set enrichment analysis (GSEA) on the RNA-seq data. In both 4T1 and Panc02 cells, IFN-α response, IFN-γ response, and antigen processing and presentation were consistently enriched on the GPAT4-deficient samples (Fig. 3(D)). Consistent with this pattern, supplementary HALLMARK analyses showed that in 4T1 cells, immune-inflammatory pathways, including IFN-γ response, IFN-α response, Inflammatory response, Complement, Allograft rejection, and IL6-JAK-STAT3 signaling, were enriched in the GPAT4-deficient samples, whereas several proliferation- and metabolism-related pathways were enriched in the control groups (Fig. S2B). In Panc02 cells, IFN-α response, IFN-γ response, and Inflammatory response were likewise positively enriched (Fig. S2C). Collectively, these transcriptomic analyses demonstrate that Gpat4 downregulation induces a conserved IFN/ISG-centered antiviral-like program and enhances antigen processing and presentation pathways across both pancreatic and breast cancer models, supporting a tumor-cell-intrinsic role of GPAT4 in restraining innate immune activation. To confirm that these transcriptomic changes were not attributable to sequencing-batch effects, we validated representative genes by qPCR in independent experiments (Fig. 3(E) and (F)). In 4T1 cells, Cxcl10, Ifi44, and Ifnb1 were all upregulated after Gpat4 depletion, whereas Gpat4 remained suppressed (Fig. 3(E)). In Panc02 cells, Cxcl10, Ifi44, and Ifit1 were similarly increased (Fig. 3(F)). Because enhanced IFN signaling is often coupled to improved antigen presentation, we next examined cell-surface major histocompatibility complex class I (MHC-I) protein expression to determine whether the transcriptional changes extended to a functionally relevant phenotype. Flow cytometry showed that cell-surface MHC-I levels were clearly higher in the Gpat4-depleted group than in controls (Fig. 3(G)). These data indicate that Gpat4 downregulation not only activates IFN/ISG transcriptional programs but also enhances antigen-presentation-related phenotypes, which may increase susceptibility to recognition by CD8+ T cells. To determine whether this pattern was consistent with the overall molecular landscape of patient tumors, we further analyzed GPAT4-related HALLMARK enrichment patterns across pan-cancers (Fig. 3(H)). In multiple cancer types, immune-inflammatory pathways, including IFN-α response, IFN-γ response, Allograft rejection, IL2-STAT5 signaling, Inflammatory response, TNFA signaling via NF-κB, and Complement, were more often associated with the low-GPAT4 state, whereas lipid-metabolic or proliferative pathways were more often associated with the high-GPAT4 state (Fig. 3(H)). Thus, clinical expression data further support the conclusion that lower GPAT4 generally corresponds to a more immune-reactive state. Together, these findings indicate that GPAT4 functions as a tumor-cell-intrinsic suppressor of innate immune activation. Gpat4 downregulation activates IFN/ISG programs, enhances MHC-I-associated antigen-presentation phenotypes, and is linked to immune-inflammatory pathway enrichment in patient tumors. GPAT4 suppresses tumor-cell-intrinsic immunity through a cGAS-dependent mechanism linked to mitostress and mtDNA leakage Because GPAT4 depletion consistently activated a tumor-cell-intrinsic IFN response, we next asked whether this effect simply reflected broad disruption of lipid metabolism and whether it depended on the catalytic activity of GPAT4. Mapping GPAT4 together with related lipid enzymes onto the glycerolipid and ether-lipid biosynthetic pathways showed that GPAT4, GPAT3, ABCD3, and GNPAT occupy distinct positions in this network (Fig. 4(A)). Functionally, however, only Gpat4 depletion induced strong upregulation of Cxcl10, Ifi44, and Ifnb1; depletion of Abcd3, Gnpat, or Gpat3 did not produce a comparable response in mouse breast 4T1 cells (Fig. 4(B)). These results indicate that the innate-immune phenotype is selective for GPAT4 rather than a generic consequence of perturbing lipid-anabolic genes. We next tested whether this effect depended on the canonical catalytic motif of GPAT4 (Liu et al., 2024; Nagle et al., 2008). We reconstituted empty vector (EV), wild-type GPAT4 (WT), or an HXXXXD-motif mutant in Gpat4-depleted cells, and found that both WT and mutant GPAT4 comparably suppressed the induction of Cxcl10, Ifi44, and Ifit1, while restoring similar Gpat4 transcript levels (Fig. 4(C) and (D)). Thus, the immunoregulatory function of GPAT4 appears largely independent of its canonical catalytic motif and instead is more consistent with a non-canonical mechanism. Gpat4deficiency drives tumor-cell innate immune activation through the mtDNA-cGAS-dependent axis independently of the canonical catalytic motif. (A) Schematic of glycerolipid and ether-lipid biosynthetic pathways highlighting GPAT4 and related enzymes used for comparison. (B) qPCR analysis of Ifnb1, Ifi44 and Cxcl10 after depletion of the indicated lipid-metabolic regulators. (C) Domain organization of wild-type GPAT4 and the HXXXXD catalytic-motif mutant used for rescue experiments. (D) qPCR analysis of interferon-responsive genes in Gpat4-depleted cells reconstituted with empty vector, wild-type GPAT4 or catalytic-mutant GPAT4 to determine whether the immune phenotype depends on the canonical catalytic motif. (E) qPCR analysis of Ifi44 and Ifit1 in control and Gpat4-depleted cells treated with vehicle or the cGAS inhibitor RU521 to test cGAS dependence. (F) Measurements of mitochondrial membrane potential and reactive oxygen species in control and Gpat4-depleted cells. The corresponding gating workflow for JC-1 and H2DCFDA analysis is shown in Fig. S3 (G, H) Representative transmission electron micrographs showing mitochondrial ultrastructure in 4T1 control (gNC) and Gpat4-deficient (gGpat4) cells. Red arrows indicate mitochondria. White boxes in (G) denote the regions enlarged in (H). Compared with control cells, Gpat4-deficient cells display evident mitochondrial abnormalities, including altered shape and disorganization of cristae. Scale bars: 1 μm (G) and 0.5 μm (H). (I) Quantification of cytosolic mtDNA amplicons shown in the panel, including mtCytb (1), mtCytb (2), mtDloop1, mtDloop2, mtDloop3 and mtND6, together with nuclear DNA Tert as a contamination control, in cytosolic fractions prepared from control and Gpat4-depleted cells. Bars denote mean ± s.e.m., P values are indicated in the respective panels, and exact P values are shown where provided and otherwise significance is denoted as ns or by asterisks. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.000 1. In cancer cells, the cGAS-MITA pathway is responsible for sensing cytosolic DNA and for inducing type I interferon signaling (Hu & Shu, 2018, 2020, 2023). We therefore tested whether the immune phenotype induced by GPAT4 loss required cGAS activity. Treatment with the cGAS inhibitor RU521 largely blunted the increase in Ifi44 and Ifit1 expression caused by Gpat4 depletion (Fig. 4(E)), indicating that activation of innate immune signaling after Gpat4 depletion is cGAS-dependent. We next investigated the upstream signal that might engage cGAS after GPAT4 loss. Gpat4-depleted cells displayed reduced mitochondrial membrane potential (MMP) together with increased reactive oxygen species (ROS) production (Fig. 4(F), Fig. S3), indicating increased mitochondrial stress (mitostress) in Gpat4-depleted cells. Consistent with these findings, transmission electron microscopy (TEM) of 4T1 cells revealed that, relative to control cells, Gpat4-deficient cells more frequently exhibit mitochondrial morphological abnormalities and swelling, accompanied by a reduction or absence of mitochondrial cristae structure, which supports the presence of mitochondrial ultrastructural changes associated with mitochondrial stress (Fig. 4(G) and (H)). Cytosolic fractions from Gpat4-depleted cells also contained increased amounts of the mtDNA amplicons shown in the panel, including two mtCytb amplicons, three D-loop amplicons, and mtND6, whereas nuclear DNA (Tert) remained unchanged (Fig. 4(I)). These data support a model in which GPAT4 deficiency promotes mitostress and mtDNA leakage, thereby activating the cGAS-mediated pathway and downstream IFN programs. Single-cell analysis shows that GPAT4 expression is negatively correlated with IFN signature and antitumor effector lymphocyte infiltration in PDAC Because our in vivo and in vitro data suggested that GPAT4 shapes the tumor microenvironment by regulating tumor-cell-intrinsic immune states, we next examined its cellular source and microenvironmental associations in a pancreatic ductal adenocarcinoma (PDAC) single-cell atlas. We analyzed a PDAC scRNA-seq cohort comprising 234 patients and 726,107 cells (Loveless et al., 2025). Uniform Manifold Approximation and Projection (UMAP) clustering resolved epithelial, endothelial, fibroblast, and major immune-cell populations (Fig. 5(A)). Comparison of GPAT4 percent expression and average expression showed that tumor cells displayed the highest proportion of GPAT4-positive cells together with relatively high mean expression (Fig. 5(B)). Single-cell transcriptomic analysis in PDAC links tumor-cell GPAT4 expression to immune-related programs and microenvironmental composition. (A) UMAP visualization of the integrated PDAC single-cell atlas, colored by major cell populations, generated from n = 234 patients comprising n = 726,107 cells (Loveless et al., 2025). (B) Dot plot summarizing the fraction of GPAT4-expressing cells and average GPAT4 expression across annotated PDAC cell populations. (C) Violin plot showing GPAT4 mRNA expression in GPAT4-negative and GPAT4-positive PDAC tumor cells classified according to detectable GPAT4 transcript signal in the processed single-cell matrix. (D) Violin-plot comparison of IFN-α response and antigen processing and presentation UCell scores between GPAT4-negative and GPAT4-positive tumor-cell states. (E) Workflow for sample filtering and grouping in the PDAC cohort. Starting from 234 PDAC samples, 150 samples passing quality filtering were retained. Tumor samples were then ranked by average GPAT4 expression in tumor cells, and the bottom and top 40% of ranked samples were designated as the GPAT4-low (n = 60) and GPAT4-high (n = 60) groups, respectively. (F) Correlation between average tumor-cell GPAT4 expression and CD8+ T-cell fraction at the sample level. (G) Stacked bar plots showing cell-type composition of individual samples ordered by average tumor-cell GPAT4 expression. (H) Comparison of cell-type proportions between GPAT4-low and GPAT4-high sample groups. Dots indicate group means. Exact P values are shown where provided and otherwise significance is denoted as ns or by asterisks. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. Having established its main cellular source, we next asked whether the tumor-cell-intrinsic GPAT4 expression state was linked to tumor immunogenicity. Within the malignant cell compartment, GPAT4-positive tumor cells were defined as cells with detectable GPAT4 transcript signal in the processed matrix, whereas GPAT4-negative tumor cells were defined as cells without detectable GPAT4 transcript (Fig. 5(C)). IFN-α response and antigen processing and presentation signature scores were both significantly higher in GPAT4-negative tumor cells (P = 1.6 × 10−37 and P = 2.7 × 10−37, respectively; Fig. 5(D)). These results indicate that, within PDAC tumor cells, lower GPAT4 expression is associated with enhanced interferon-response and antigen-presentation programs, consistent with observations from our cell models. Multiple immune-related UCell programs differed consistently between GPAT4-defined tumor cell states, with IFN-α response, IFN-γ response, JAK–STAT signaling, NF-κB signaling and chemokine signaling all reduced in GPAT4-expressing tumor cells (Fig. S4A). In parallel, GPAT4 expression in tumor cells also differed significantly across clinical groups (Fig. S4B): it was lower in treated tumors than in treatment-naive tumors, lower in adjacent normal tissue than in primary tumors, and higher in metastatic lesions than in primary tumors. Collectively, these findings suggest that GPAT4 expression may reflect a tumor-cell state associated with immune attenuation and disease evolution in PDAC, thereby supporting its potential utility as a marker of clinically relevant tumor-cell heterogeneity. To specifically evaluate the relationship between tumor-cell GPAT4 expression and the immune landscape of PDAC, we first filtered the cohort from 234 to 150 samples to retain cases with sufficient tumor-cell information for reliable analysis and then ranked the retained samples by average tumor-cell GPAT4 expression. We defined the bottom and top 40% of the ranked PDAC samples as GPAT4-low and GPAT4-high groups, respectively (n = 60 samples per group; Fig. 5(E)). Patients with higher tumor-cell GPAT4 expression showed a progressive increase in epithelial and endothelial composition together with an overall reduction in immune-cell representation (Fig. 5(G)). This pattern suggests that high tumor-cell GPAT4 expression is associated not only with weaker tumor-cell-intrinsic immune programs but also with a tissue composition characterized by tumor-cell enrichment and restricted immune content. To quantify this association, we next examined the relationship between average tumor-cell GPAT4 expression and the CD8+ T cell fraction across the retained PDAC samples. These variables were significantly negatively correlated (R = −0.27, P = 4.8 × 10−5; Fig. 5(F)). Grouped analysis further showed that GPAT4-low samples (bottom 40%, n = 60) contained significantly more CD8+ T cells, regulatory T cells, monocytes, macrophages, dendritic cells, NK cells, and mast cells, whereas GPAT4-high samples (top 40%, n = 60) showed higher CD4+ T cells fractions. B cells, plasma cells, tumor cells, endothelial cells, and fibroblasts were not significantly different between groups (Fig. 5(H)). Thus, at the PDAC sample level, higher tumor-cell GPAT4 expression is associated with reduced CD8+ T cell and NK-cell infiltration and with broader changes in immune-cell composition. In summary, single-cell analysis indicates that GPAT4 is mainly enriched in malignant cells in PDAC. Within tumor cells, low GPAT4 is associated with stronger IFN-related and antigen-presentation programs, whereas at the sample level, high tumor-cell GPAT4 expression is associated with reduced CD8+ T-cell infiltration and altered immune-cell composition. These patient-level data further support a model in which GPAT4 promotes immune evasion in PDAC by suppressing tumor-cell-intrinsic immune activation and fostering an immune-restricted microenvironment. Single-cell analysis shows that GPAT4 expression is negatively correlated with IFN signature and antitumor effector lymphocyte infiltration in BRCA Since our preceding analyses suggested that GPAT4 promotes tumor immune evasion by suppressing tumor-cell-intrinsic IFN/MHC-I programs and shaping an immune-restricted microenvironment, we next asked whether this mechanism also operated across tumor types and whether it was related to immune checkpoint blockade (ICB)-associated immune states. We therefore extended the analysis to a breast cancer (BRCA) single-cell cohort. We analyzed a BRCA scRNA-seq dataset comprising 119 patients and 236,363 cells (Xu et al., 2024). UMAP analysis resolved the major immune, stromal, and epithelial compartments (Fig. 6(A)). Comparison of percent expression and average expression showed that cancer epithelial cells had a relatively high proportion of GPAT4-positive cells and higher mean GPAT4 expression (Fig. 6(B)). Consistently, UMAP visualization of GPAT4-positive and GPAT4-negative cells showed that positive cells clustered mainly within epithelial regions, especially the malignant epithelial compartment (Fig. 6(C)). Thus, as in PDAC, GPAT4 in BRCA is predominantly enriched in tumor cells. GPAT4 expression and its association with tumor-cell-intrinsic immune programs and microenvironmental composition in the BRCA single-cell atlas. (A) UMAP visualization of the BRCA single-cell atlas generated from n = 119 patients comprising n = 236,363 cells (Xu et al., 2024), colored by cell type. (B) Dot plot showing the fraction of GPAT4-expressing cells and average GPAT4 expression across BRCA cell populations. (C) UMAP projection highlighting cells with detectable or undetectable GPAT4 expression. (D) Dot plot summarizing immune-related genes and programs, including interferon-response, pro-inflammatory-cytokine and MHC-I/antigen-presentation features, across GPAT4-defined tumor-cell states. (E) Violin-plot comparison of GPAT4 expression together with antigen processing and presentation, IFN-α response and IFN-γ response signature scores between GPAT4-negative and GPAT4-positive tumor cells. (F) Workflow for sample filtering and grouping in the BRCA cohort. Starting from 119 BRCA samples, 62 samples passing quality filtering were retained. Based on average GPAT4 expression in tumor cells, samples were stratified into GPAT4-low (bottom 40%, n = 25) and GPAT4-high (top 40%, n = 25) groups. (G) Correlation between average cancer-epithelial-cell GPAT4 expression and CD8+ T-cell fraction at the sample level. (H) Stacked bar plots showing cell-type composition of individual samples ordered by average cancer-epithelial-cell GPAT4 expression, with the per-patient GPAT4-expression track annotated above the bars. (I) Comparison of cell-type proportions between GPAT4-low and GPAT4-high sample groups. Dots indicate group means. Exact P values are shown where provided and otherwise significance is denoted as ns or by asterisks. The statistical methods are described in the Materials and methods section. ns, not significant; ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. We next assessed the relationship between tumor-cell GPAT4 state and immune-related transcriptional programs. In BRCA tumor cells, antigen processing/presentation, IFN-α response, and IFN-γ response signatures were all significantly higher in the GPAT4-negative state (P = 0.021, P = 2.8 × 10−5, and P = 2 × 10−14, respectively; Fig. 6(E)). Likewise, genes including B2M, ISG20, CXCL10, PLSCR1, SP110, EPSTI1, CXCL11, IFIT2, IFIT3, CMPK2, IFIH1, and IFI44L were more highly expressed in GPAT4-negative tumor cells, together with broader increases in pro-inflammatory-cytokine and MHC-I/antigen-presentation features (Fig. 6(D)). Within BRCA tumor cells, IFN-γ response, IFN-α response, and Allograft Rejection were decreased in the GPAT4-positive versus GPAT4-negative tumor cells, whereas Glycolysis, Fatty Acid Metabolism, Estrogen Response Early, and Estrogen Response Late were increased (Fig. S5), indicating that GPAT4 state differences extend beyond immune programs to broader metabolic and hormone-response changes. Together, these findings show that, in BRCA tumor cells, the GPAT4-negative state is associated with enhanced IFN-related and MHC-related transcriptional programs, consistent with the pattern observed in pancreatic cancer and experimental models. To assess whether the tumor-cell GPAT4 pattern observed at the single-cell level was also present at the sample level, we excluded samples with insufficient tumor-cell content and calculated the average GPAT4 expression in cancer epithelial cells for each patient. This filtering workflow reduced the cohort to 62 samples, and the bottom and top 40% of the ranked samples were defined as GPAT4-low and GPAT4-high groups, respectively (n = 25 samples per group; Fig. 6(F)). When samples were ordered by tumor cell GPAT4 expression, those with higher GPAT4 generally showed a greater malignant epithelial contribution and an overall compression of immune-cell components (Fig. 6(H)). Moreover, average GPAT4 expression was significantly negatively correlated with the CD8+ T cell fraction (R = −0.31, P = 0.014; Fig. 6(G)). Grouped analysis further showed that GPAT4-high samples contained significantly fewer CD8+ T cells, CD4+ T cells, NK cells, and plasma cells, whereas B cells, tumor cells, dendritic cells, endothelial cells, fibroblasts, macrophages, mast cells, monocytes, and regulatory T cells were not significantly different (Fig. 6(I)). Thus, at the BRCA sample level, higher tumor-cell GPAT4 is likewise associated with weaker effector-immune infiltration. Collectively, these data show that GPAT4 is enriched in malignant epithelial cells in BRCA, and that higher tumor-cell GPAT4 expression is associated with weaker IFN/MHC-I programs and reduced effector immune-cell infiltration. We therefore further examined whether GPAT4 expression was associated with ICB responsiveness. Association of GPAT4 expression with response to ICI therapy Finally, we examined whether there is a potential relationship between tumor-cell GPAT4 expression and responsiveness to immune checkpoint inhibitor (ICI) therapy. The single-cell analysis was based on the pretreatment breast cancer anti-PD-1 scRNA-seq cohort reported by Bassez et al. (Bassez et al., 2021), which profiled paired pre- and on-treatment biopsies from treatment-naive patients receiving pembrolizumab before surgery. According to post-treatment changes in intratumoral T-cell clonotypes, patients were classified as having no/limited clonal expansion (NE, n = 20) or clonal expansion (E, n = 9), as post-treatment clonal expansion of intratumoral T cells is indicative of a productive antitumor immune response. T-cell subtype analysis showed that responders were enriched for proliferating experienced CD4+ T cells, experienced CD8+ T cells, proliferating experienced CD8+ T cells, and proliferating regulatory CD4+ T cells (Fig. 7(A)). These populations reflect ongoing antigen experience and active cell-cycle entry after ICI treatment, supporting effective T-cell priming and expansion within the tumor microenvironment. By contrast, non-responders were relatively enriched for effector/memory CD8+ T cells, naive CD4+ T cells, tissue-resident memory CD8+ T cells, naive CD8+ T cells, effector/memory CD4+ T cells, and EMRA CD8+ T cells, indicating a T-cell compartment skewed toward non-expanding, resting, or terminally differentiated states rather than actively expanding tumor-reactive states, as previously reported. Association of GPAT4 with clonal expansion-associated immune activation and response to immunotherapy. (A) Differences in T-cell subtype proportions between non-responders and responders in the same cohort. Responders are enriched for proliferating and antigen-experienced CD4+/CD8+ T-cell states. Pre-treatment anti-PD-1 scRNA-seq cohort, stratified according to post-treatment T-cell clonotype expansion into non-responders (NE; no/limited clonal expansion) and responders (E; clonal expansion). (B) Violin-plot comparison of GPAT4 expression, ISG score, and MHC-I score. Tumor cells from non-responders show higher GPAT4 expression, whereas tumor cells from responders display higher interferon-stimulated-gene and MHC-I-related signature scores, linking low-GPAT4 states to increased immune activation. (C) Integrated analysis of public ICI transcriptomic cohorts comparing GPAT4 expression between non-responders and responders: metastatic melanoma tumor specimens from GSE115821, pretreatment FFPE core-needle biopsies from high-risk HER2-negative stage II/III breast cancer in the I-SPY2 trial from GSE173839 with pathological complete response (pCR) versus non-pCR endpoint, nivolumab-treated advanced melanoma tumor biopsies from GSE91061, and advanced non-small cell lung cancer (NSCLC) tumor tissues from patients treated with anti-PD-1/PD-L1 from GSE135222 with durable clinical benefit versus non-durable benefit according to Response Evaluation Criteria in Solid Tumors (RECIST) v1.1 and follow-up duration. GPAT4 expression is significantly higher in non-responders than in responders (non-responder, n = 199; responder, n = 70; P = 0.004 8), supporting an association between elevated GPAT4 and poor immunotherapy responsiveness. (D) Working model summarizing the proposed mechanism whereby GPAT4 depletion increases mitochondrial stress, promotes mitochondrial-DNA leakage into the cytosol, activates cGAS-mediated signaling, induces interferon-stimulated-gene and MHC-I/antigen-presentation programs, and thereby enhances effector-lymphocyte infiltration, reshapes the tumor microenvironment, and improves antitumor immunity. The statistical methods are described in the Materials and methods section. ∗P < 0.05. Comparison of tumor cells from these two groups showed that NE tumors expressed higher GPAT4 (P = 0.046), whereas ISG and MHC-I signature scores were both markedly higher in tumors from responders (both P < 2.22 × 10−16; Fig. 7(B)). Thus, tumor cells with higher GPAT4 more often reside in a state with lower IFN and MHC-I activity, whereas tumor cells associated with clonal expansion, reflecting stronger immune-response potential, display greater immune activation. Together, the tumor-cell and T-cell analyses support that lower GPAT4 is associated with clonal-expansion-linked immune activation and treatment responsiveness. To validate this association, we further analyzed published ICI cohorts from tumor types aligned with the preceding analyses (Listed in Supplemental Table 1). After response annotation and preprocessing, the integrated analysis included 269 samples in total from metastatic melanoma (GSE115821) (Auslander et al., 2018), high-risk HER2-negative stage II/III breast cancer in the I-SPY2 trial (GSE173839) (Pusztai et al., 2021), nivolumab-treated advanced melanoma (GSE91061) (Riaz et al., 2017), and advanced non-small cell lung cancer (NSCLC) treated with anti-PD-1/PD-L1 (GSE135222) (Jung et al., 2019; Kim et al., 2020). Response definitions were harmonized according to the original study annotations, including pCR versus non-pCR for the neoadjuvant breast cancer cohort and durable clinical benefit versus non-durable benefit according to Response Evaluation Criteria in Solid Tumors (RECIST) v1.1 and follow-up duration for the NSCLC cohort. In this analysis, GPAT4 expression was significantly higher in non-responders than in responders (non-responder, n = 199; responder, n = 70; P = 0.004 8; Fig. 7(C)). Collectively, these results support a negative association between GPAT4 expression and responsiveness to ICI therapy. To extend these transcriptomic and immune-infiltration observations to the protein level, we further analyzed Clinical Proteomic Tumor Analysis Consortium (CPTAC) primary tumor proteomics data for PAAD and BRCA (Fig. S6). Samples were stratified according to GPAT4 protein abundance, and selected immune-related proteins were compared between the GPAT4-low and GPAT4-high groups. In BRCA, the GPAT4-high group showed lower CD8 protein abundance and reduced levels of several immune-associated proteins, including MVB12A, C1R, NFKB1, BANK1, TNFAIP6, and LGALS3BP. In PAAD, high GPAT4 protein abundance was likewise accompanied by lower levels of multiple immune- and antigen-presentation-associated proteins. Although not every marker reached significance in both cohorts, the overall protein-level pattern was consistent with the transcriptomic and single-cell analyses, supporting an association between high GPAT4 abundance and an immune-restricted tumor state. Taken together, our findings support a model in which GPAT4 drives tumor immune evasion (Fig. 7(D)). GPAT4 depletion increases mitochondrial stress, as indicated by elevated ROS, reduced mitochondrial membrane potential, and increased cytosolic mtDNA, which in turn activates the cGAS-mediated pathway, induces IFN/ISG programs, and enhances MHC-I-related antigen presentation. Consistent with this mechanism, GPAT4 downregulation in vivo is accompanied by greater effector-lymphocyte infiltration and stronger tumor suppression, while low-GPAT4 states in PDAC and BRCA single-cell cohorts similarly correspond to higher immune activation. Together, these results identify GPAT4 as an important metabolic node that helps maintain an immune-cold tumor state, and they suggest that its inhibition can increase tumor immunogenicity and promote antitumor immunity through mitochondrial-stress-driven cGAS-mediated signaling. Discussion This study reveals for the first time that GPAT4 functions as a tumor-cell-intrinsic innate immune checkpoint molecule, suppressing antitumor immune responses through negative regulation of the mtDNA-cGAS-mediated signaling axis. We found that GPAT4 is highly expressed in various tumors and is significantly associated with reduced immune cell infiltration, diminished interferon signaling pathway activity, and poor patient prognosis. GPAT4 deficiency induces mitochondrial stress, leading to the release of mtDNA into the cytoplasm, activation of the cGAS-mediated pathway, and subsequent enhancement of type I interferon responses and MHC-I expression. This promotes the infiltration of CD8+ T cells and NK cells, ultimately inhibiting tumor growth. At the clinical protein level, analysis of CPTAC primary tumor proteomics data further supported this model: in PAAD and BRCA, tumors with higher GPAT4 protein abundance showed reduced levels of multiple immune-, cytotoxicity-, IFN-, and MHC-I/antigen-presentation-associated proteins. Notably, this regulatory effect is independent of the classical metabolic enzyme activity of GPAT4, suggesting a non-canonical function. Single-cell transcriptomic analysis further confirmed that in pancreatic and breast cancers, tumor cells with low GPAT4 expression exhibit higher interferon signaling activity and enrichment of immune effector cells, which correlates with response to immune checkpoint therapy. In summary, our study establishes a critical role for GPAT4 in tumor immune evasion and provides a theoretical basis for targeting this molecule to activate tumor-intrinsic immunity. GPAT4 is a rate-limiting enzyme catalyzing the synthesis of triglycerides and phospholipids, with previous research primarily focusing on its role in lipid metabolism reprogramming. In ovarian cancer, high GPAT4 expression promotes platinum resistance by regulating ferroptosis sensitivity (Coleman & Lee, 2004; Korbecki et al., 2024; Li et al., 2025; Zhu et al., 2019); in colorectal cancer, GPAT4 is involved in CPT2 downregulation-mediated activation of lipophagy, accelerating tumor progression (Li et al., 2025; Nagle et al., 2008). However, whether GPAT4 influences the tumor immune microenvironment has not been previously reported. This study links GPAT4 with innate immunity, demonstrating that its deficiency activates the cGAS-mediated pathway in an enzyme-activity-independent manner. Notably, we compared the depletion effects of multiple lipid metabolism-related genes (such as ABCD3, GNPAT, and GPAT3), and only GPAT4 depletion significantly induced the expression of interferon-related genes, indicating the specificity of this effect. Furthermore, rescue experiments showed that GPAT4 harboring a catalytic-site mutation still suppressed interferon responses, further supporting its non-canonical function. These results suggest that GPAT4 may possess molecular functions independent of lipid metabolism, such as regulating mitochondrial homeostasis through interactions with other proteins. Recent studies have reported that GPAT4 deficiency can induce endoplasmic reticulum stress and enhance ER-mitochondria communication, leading to mtDNA escape in endocardial cells (Zhao et al., 2025), which is consistent with our observations of decreased mitochondrial membrane potential, increased ROS, and mtDNA leakage in cancer cells. Thus, elevated GPAT4 expression in cancer cells likely utilizes a similar molecular mechanism to sustain endoplasmic reticulum and mitochondrial homeostasis. The ability of the HXXXXD catalytic-motif mutant to suppress ISG induction further raises the possibility that GPAT4 controls tumor-cell-intrinsic innate immunity through a non-enzymatic mechanism. One plausible model is that GPAT4, as an ER-resident membrane protein, may function as a scaffold or organizer of protein complexes that maintain ER-mitochondrial homeostasis, rather than acting solely through its canonical acyltransferase activity. This interpretation is consistent with our observation that depletion of other lipid-metabolic enzymes, including GPAT3, GNPAT, and ABCD3, did not phenocopy GPAT4 depletion, whereas both wild-type GPAT4 and the HXXXXD mutant suppressed ISG induction. Recent work in endocardial cells also showed that GPAT4 deficiency induces ER stress, enhances ER-mitochondria communication, promotes mitochondrial dysfunction and mtDNA escape, and subsequently activates the cGAS-MITA/STING pathway. Mechanistically, that study suggested that GPAT4 may participate in ER proteostasis through protein interactions involving the PERK-eIF2α axis, supporting the broader concept that GPAT4 may have a structural or scaffolding role in addition to its enzymatic function. Therefore, in tumor cells, GPAT4 may limit mtDNA-dependent innate immune activation by preserving ER proteostasis, restraining excessive ER-mitochondria contact, and maintaining mitochondrial integrity. Based on these considerations, several non-mutually exclusive mechanisms could be further investigated in future studies. First, GPAT4 may act as an ER proteostasis-associated scaffold that limits stress transmission from the ER to mitochondria. Second, GPAT4 may help organize ER-mitochondria contact sites, thereby preventing excessive calcium or ROS-dependent mitochondrial injury and subsequent mtDNA leakage. Third, GPAT4 may indirectly function as a mitochondrial DAMP gatekeeper by maintaining the threshold at which stressed mitochondria release mtDNA into the cytosol. These hypotheses could be tested in future work by comparing the interactomes of wild-type GPAT4 and the HXXXXD mutant using co-immunoprecipitation coupled with mass spectrometry or proximity-labeling approaches such as TurboID. Candidate interacting proteins involved in ER proteostasis, ER-mitochondria contact, calcium transfer, or mitochondrial membrane integrity could then be validated by reciprocal co-immunoprecipitation, proximity ligation assays, subcellular fractionation, and functional rescue experiments. Such analyses would help determine whether the catalysis-independent immune-regulatory function of GPAT4 is mediated by conserved protein-protein interactions, by spatial organization of ER-mitochondria interfaces, or by both mechanisms. Although the precise binding partners remain to be defined, our current data support a model in which GPAT4 suppresses tumor immunogenicity not simply as a lipid-metabolic enzyme, but as a non-canonical regulator of organelle homeostasis upstream of mtDNA-cGAS-MITA/STING activation. The cGAS-MITA pathway, as a core mechanism for cytosolic DNA sensing, plays a crucial role in antitumor immunity. In this study, mtDNA release induced by GPAT4 deficiency was accompanied by enhanced interferon responses and effector cell infiltration, suggesting that it may remodel the immune microenvironment through cGAS-MITA pathway activation. Notably, single-cell analysis revealed that tumor cells with low GPAT4 expression not only exhibited enhanced interferon signaling but also showed upregulation of MHC-I-related genes, consistent with our in vitro observation of increased cell-surface MHC-I protein. This enhanced antigen presentation capability may increase the susceptibility of tumor cells to CD8+ T cell-mediated killing, thereby synergistically promoting antitumor immunity. Clinical correlation analysis indicates that high GPAT4 expression is associated with poor prognosis and an immune-cold phenotype in various tumors. In pancreatic and breast cancer single-cell datasets, GPAT4 expression levels in tumor cells were significantly negatively correlated with CD8+ T cell infiltration, and samples with high GPAT4 expression showed an overall reduction in immune cell composition. Furthermore, in immunotherapy cohorts, GPAT4 expression was higher in tumor tissues from non-responders compared to responders, suggesting that GPAT4 may serve as a predictive biomarker for immunotherapy response. These findings are highly consistent with our mechanistic studies, supporting the view that GPAT4 shapes an immune-excluded microenvironment by suppressing tumor-intrinsic immune activity. These findings also suggest a rational combination strategy in which GPAT4-targeting approaches are paired with immune checkpoint inhibitors. By increasing mtDNA-cGAS-MITA/STING activation, IFN/ISG expression, CXCL10 production, and MHC-I-related antigen presentation, GPAT4 inhibition may convert immune-cold tumors into a more inflamed and immunovisible state, thereby sensitizing tumors to PD-1/PD-L1 blockade. This possibility will require future pharmacological development and preclinical testing but provides a clear translational direction for GPAT4-based immune sensitization. In summary, this study reveals GPAT4 as a tumor-cell-intrinsic innate immune checkpoint that promotes immune evasion by inhibiting the mtDNA-cGAS-mediated signaling axis in an enzyme-activity-independent manner. Targeting GPAT4 can activate tumor-intrinsic immunity and improve the immune microenvironment, offering a new strategy to enhance the efficacy of existing immunotherapies. Future research should focus on elucidating the precise mechanisms by which GPAT4 regulates mitochondrial homeostasis and exploring its potential applications in combination therapies. Materials and methods Cell lines and culture conditions Murine 4T1 breast carcinoma cells, Panc02 pancreatic adenocarcinoma cells, and HEK293T packaging cells were cultured under standard conditions at 37 °C in a humidified incubator with 5% CO2. 4T1 and Panc02 cells were maintained in RPMI-1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin, whereas HEK293T cells were maintained in DMEM supplemented with 10% fetal bovine serum. Cell lines were routinely tested for mycoplasma contamination and were used at low passage number for all experiments. Generation of CRISPR-Cas9-edited gene-deficient cells Stable gene-deficient cell populations were generated by lentiviral transduction of a mixture of two independent guide RNAs targeting murine Gpat4, Gpat3, Gnpat, or Abcd3, with a non-targeting sgRNA used as the control. CRISPR-Cas9 was used to edit genomic loci rather than directly modulate mRNA; after DNA repair, indel mutations can disrupt the coding sequence and reduce functional transcript abundance through frameshift-mediated disruption and/or nonsense-mediated decay. Reduced transcript abundance was therefore verified by qRT-PCR before all downstream experiments as a validation readout of the edited cell populations. The gRNA targeting sequences are listed as follows: gmGpat4#1-CAGCGACGGCTACTACGCCA; gmGpat4#1-CAGCGACGGCTACTACGCCA; gmGpat4#2-TAGTAGTTCCCACTACCAAG; gmGpat4#2-TAGTAGTTCCCACTACCAAG; gmGpat3#1-CATGGCGTAACATCCATCGG; gmGpat3#1-CATGGCGTAACATCCATCGG; gmGpat3#2-GTGAGGACAAGCCTTAACCA; gmGpat3#2-GTGAGGACAAGCCTTAACCA; gmGnpat#1-CCTTCGGCTTAGGAACTCCG; gmGnpat#1-CCTTCGGCTTAGGAACTCCG; gmGnpat#2-ATGAGCCACAAACTGCGCAT; gmGnpat#2-ATGAGCCACAAACTGCGCAT; gmAbcd3#1-GCTCACACGGTACCTCTACG; gmAbcd3#1-GCTCACACGGTACCTCTACG; gmAbcd3#2-AACCAGGTACCCGACGACAG gmAbcd3#2-AACCAGGTACCCGACGACAG Plasmid construction Wild-type mouse Gpat4 and a catalytic-motif mutant disrupting the conserved HXXXXD acyltransferase motif were generated using standard molecular-cloning methods. The conserved HXXXXD acyltransferase motif HTSPID was mutated to AAAAAA to obtain Gpat4-Mut. RNA isolation and quantitative RT-PCR Total RNA was isolated using TRIzol (Vazyme) according to the manufacturer's instructions. RNA was reverse-transcribed to cDNA and analyzed by quantitative PCR using SYBR Green chemistry. Transcript abundance was normalized to housekeeping genes, and relative expression was calculated by the 2−ΔΔCt method. Assayed transcripts included Gpat4 and interferon/innate-immunity-related genes such as Cxcl10, Ifi44, Ifnb1, and Ifit1, depending on the experiment. For cytosolic DNA analyses, mitochondrial DNA fragments including mt-Co1, mt-Nd6, mt-Cytb, m-mt-Dloop1, m-mt-Dloop2, m-mt-Dloop3, m-mt-CytB, and m-mt-16s, together with the nuclear DNA control Tert, were quantified by qPCR. Unless otherwise indicated, qPCR data were generated from biologically independent replicate samples and technical replicates were averaged within each experiment. The sequences of qPCR primers are listed as follows: Gpat4-F AATGGAGAGAGGAGCCAAGGA; Gpat4-F AATGGAGAGAGGAGCCAAGGA; Gpat4-R CGGCAAAAGTAGAAGATGTC; Gpat4-R CGGCAAAAGTAGAAGATGTC; Actb-F CATTGCTGACAGGATGCAGAAGG; Actb-F CATTGCTGACAGGATGCAGAAGG; Actb-R TGCTGGAAGGTGGACAGTGAGG; Actb-R TGCTGGAAGGTGGACAGTGAGG; Ifit1-F ACAGCAACCATGGGAGAGAATGCTG; Ifit1-F ACAGCAACCATGGGAGAGAATGCTG; Ifit1-R ACGTAGGCCAGGAGGTTGTGCAT; Ifit1-R ACGTAGGCCAGGAGGTTGTGCAT; Ifi44-F ATGCACTCTTCTGAGCTGGTGG; Ifi44-F ATGCACTCTTCTGAGCTGGTGG; Ifi44-R TCAGATCCAGGCTATCCACGTG; Ifi44-R TCAGATCCAGGCTATCCACGTG; Cxcl10-F ATCATCCCTGCGAGCCTATCCT; Cxcl10-F ATCATCCCTGCGAGCCTATCCT; Cxcl10-R GACCTTTTTTGGCTAAACGCTTTC; Cxcl10-R GACCTTTTTTGGCTAAACGCTTTC; Ifnb1-F TCCTGCTGTGCTTCTCCACCACA; Ifnb1-F TCCTGCTGTGCTTCTCCACCACA; Ifnb1-R AAGTCCGCCCTGTAGGTGAGGTT; Ifnb1-R AAGTCCGCCCTGTAGGTGAGGTT; mt-Co1-F GCCCCAGATATAGCATTCCC; mt-Co1-F GCCCCAGATATAGCATTCCC; mt-Co1-R GTTCATCCTGTTCCTGCTCC; mt-Co1-R GTTCATCCTGTTCCTGCTCC; mt-Nd6-F TTAGCATTAAAGCCTTCACC; mt-Nd6-F TTAGCATTAAAGCCTTCACC; mt-Nd6-R CCAACAAACCCACTAACAAT; mt-Nd6-R CCAACAAACCCACTAACAAT; mt-Cytb-F AGTAGACAAAGCCACCTTGA; mt-Cytb-F AGTAGACAAAGCCACCTTGA; mt-Cytb-R CCGCGATAATAAATGGTAAG; mt-Cytb-R CCGCGATAATAAATGGTAAG; m-mt-Dloop1-F AATCTACCATCCTCCGTGAAACC; m-mt-Dloop1-F AATCTACCATCCTCCGTGAAACC; m-mt-Dloop1-R TCAGTTTAGCTACCCCCAAGTTTAA; m-mt-Dloop1-R TCAGTTTAGCTACCCCCAAGTTTAA; m-mt-Dloop2-F CCCTTCCCCATTTGGTCT; m-mt-Dloop2-F CCCTTCCCCATTTGGTCT; m-mt-Dloop2-R TGGTTTCACGGAGGATGG; m-mt-Dloop2-R TGGTTTCACGGAGGATGG; m-mt-Dloop3-F TCCTCCGTGAAACCAACAA; m-mt-Dloop3-F TCCTCCGTGAAACCAACAA; m-mt-Dloop3-R AGCGAGAAGAGGGGCATT; m-mt-Dloop3-R AGCGAGAAGAGGGGCATT; m-mt-CytB-F GCTTTCCACTTCATCTTACCATTTA; m-mt-CytB-F GCTTTCCACTTCATCTTACCATTTA; m-mt-CytB-R TGTTGGGTTGTTTGATCCTG; m-mt-CytB-R TGTTGGGTTGTTTGATCCTG; m-mt-16s-F CACTGCCTGCCCAGTGA; m-mt-16s-F CACTGCCTGCCCAGTGA; m-mt-16s-R ATACCGCGGCCGTTAAA; m-mt-16s-R ATACCGCGGCCGTTAAA; m-nucDNA-Tert-F CTAGCTCATGTGTCAAGACCCTCTT; m-nucDNA-Tert-F CTAGCTCATGTGTCAAGACCCTCTT; m-nucDNA-Tert-R GCCAGCACGTTTCTCTCGTT m-nucDNA-Tert-R GCCAGCACGTTTCTCTCGTT Bulk RNA sequencing and pathway enrichment analysis Control and Gpat4-deficient 4T1 and Panc02 cells were subjected to bulk RNA sequencing (RNA-seq) to define tumor-cell-intrinsic transcriptional changes induced by GPAT4 loss. Two biologically independent libraries were analyzed for each condition in each cell line. Libraries were prepared from high-quality total RNA and sequenced on an Illumina platform. Reads were aligned to the mouse reference genome GRCm38/mm10 using STAR, and gene-level read counts were generated with featureCounts using the corresponding GENCODE mouse release gene annotation. Principal-component analysis was used to assess sample separation, and gene set enrichment analysis was performed on ranked gene lists to evaluate HALLMARK IFN-α response, IFN-γ response, antigen processing and presentation, and related immune-inflammatory programs. All analyses were performed using R v4.3.1, STAR v2.7.11b, FastQC v0.12.0, MultiQC v1.35, DESeq2 v1.52.0, clusterProfiler v4.20.0, msigdbr v26.1.0, and org.Mm.eg.db v3.23.0. Flow cytometry For cell-surface antigen-presentation analyses, cultured tumor cells were stained with antibodies against mouse MHC-I and analyzed by flow cytometry after exclusion of dead cells and doublets. For immune profiling of tumor tissues, excised tumors were dissociated to single-cell suspensions, stained with antibodies against CD45, CD3, CD4, CD8, and NKP46 (Biolegend), and analyzed to quantify infiltrating T-cell and NK-cell populations. For tumor-infiltrating CD8+ T-cell functional analysis, single-cell suspensions were sequentially gated on live CD45+ immune cells, CD3+ T cells, and CD8+ T cells, followed by quantification of Granzyme B, IFN-γ, and CD69. To assess mitochondrial stress, intracellular reactive oxygen species were measured using H2DCFDA (Solarbio), and mitochondrial membrane potential was measured using JC-1 (MCE) by flow cytometry. For mitochondrial membrane potential and ROS assays, debris was excluded by FSC-A/SSC-A, singlets were selected by FSC-H/FSC-A, intact cells were gated, and JC-1 or H2DCFDA fluorescence was quantified within the final intact-cell population. Flow-cytometry data were analyzed using FlowJo v11.1.1. In vivo tumor studies For syngeneic tumor-growth studies, control or CRISPR-Cas9-edited Gpat4-deficient Panc02 cells (1 × 106 cells per mouse) were implanted subcutaneously into male immunocompetent C57BL/6J mice. Tumor growth was monitored longitudinally by caliper measurement on the indicated days after implantation, and tumor volume was calculated using the standard formula (length × width2)/2. At endpoint, tumors were excised for weighing and immune-cell analysis by flow cytometry. To determine whether the growth phenotype depended on host immunity, the same Panc02 derivatives were implanted into male nude mice using the same implantation dose and monitoring strategy. Each Panc02 cohort contained eight mice per group. For the second syngeneic model, control or CRISPR-Cas9-edited Gpat4-deficient 4T1 cells (1 × 106 cells per mouse) were implanted subcutaneously into female immunocompetent BALB/c mice. Tumor volume was measured on days 8, 12, 16, 21, 27, and 31 after implantation, and tumors were collected on day 31 for endpoint tumor-weight, tumor-volume, and immune-infiltration analyses. The 4T1 cohort contained eight mice per group. Animals were randomly assigned to experimental groups, and all procedures were performed in accordance with institutional guidelines for animal care and use. All experimental procedures involving animals were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Wuhan University Medical Research Institute, and strictly followed the ethical guidelines for animal welfare. Public datasets and bioinformatic analyses Publicly available transcriptomic, clinical, functional-dependency, and immune-infiltration datasets were used to investigate the expression pattern, prognostic significance, tumor-cell dependency, immune relevance, and pathway association of GPAT4. Pan-cancer RNA-seq expression data and matched clinical annotations were obtained from TCGA, and public normal reference datasets were incorporated for tumor-versus-normal comparisons when adjacent normal tissues were unavailable or insufficient. Expression matrices were processed in R (v4.3.1), and RNA-seq values were transformed as log2(TPM + 1) when required for cross-sample comparison. Differences in GPAT4 expression between tumor and normal tissues were evaluated using the Wilcoxon rank-sum test, or the Wilcoxon signed-rank test for paired samples when matched tumor-normal tissues were available. For survival analyses, patients in each cohort were stratified into GPAT4-high and GPAT4-low groups according to the cohort-specific median expression level. Kaplan-Meier survival curves were generated using the survival and survminer packages in R, and significance was assessed by two-sided log-rank tests. Hazard ratios and 95% confidence intervals were estimated using univariate Cox proportional-hazards regression. Additional Kaplan-Meier validation was performed in independent external cohorts with available expression and survival data using the same grouping strategy, and each cohort was analyzed separately to avoid distortion caused by direct merging of datasets from different platforms. Tumor-cell dependency analysis was conducted using Chronos gene-effect scores downloaded from the DepMap portal. GPAT4 dependency scores were extracted across cell lines and summarized by tissue lineage. In this analysis, more negative Chronos scores indicate stronger genetic dependency, and lineage-level distributions were visualized in R using ggplot2. Immune infiltration analyses were performed using deconvolution-derived immune-cell fractions from public bulk-transcriptomic resources, including CIBERSORT estimates. For each cancer type, correlations between GPAT4 expression and the inferred abundance of immune-cell populations were calculated using Spearman's rank correlation in R with the cor.test function. Where multiple immune-cell subsets were evaluated simultaneously, P values were adjusted using the Benjamini-Hochberg method. These analyses were performed within each cancer type independently rather than on directly pooled pan-cancer matrices, in order to minimize confounding from inter-tumor heterogeneity and batch effects. For pathway-level analyses, HALLMARK gene sets and curated immune-related signatures were obtained from the Molecular Signatures Database (MSigDB) or from published studies. Sample-wise pathway activities were quantified using GSVA/ssGSEA implemented in the GSVA package in R, and enrichment differences between predefined groups were compared using the limma package or non-parametric tests as appropriate. In analyses involving multiple public cohorts, each dataset was preprocessed separately, and cohort-specific normalization was performed before any integrated comparison. When only GPAT4 expression was combined across studies, values were first standardized within each cohort using z-score normalization and then merged, thereby reducing cross-platform batch effects while preserving the relative difference between comparison groups within each dataset. Protein-level validation was performed using Clinical Proteomic Tumor Analysis Consortium (CPTAC) primary tumor proteomics data for PAAD and BRCA. GPAT4 protein abundance values were extracted from the processed proteomic matrices, and samples were ranked within each cancer cohort according to GPAT4 protein abundance. The low- and high-GPAT4 protein groups used for display comprised 17 BRCA tumors per group and 35 PAAD tumors per group. Protein abundances were scaled within each cohort before visualization. Immune-, cytotoxicity-, IFN-response-, and MHC-I/antigen-presentation-associated proteins, were compared between the GPAT4-low and GPAT4-high groups. Group differences were assessed using two-sided Wilcoxon rank-sum tests unless otherwise indicated. Published immune checkpoint inhibitor (ICI) cohorts analysis Publicly available ICI-related transcriptomic cohorts were collected from the Gene Expression Omnibus (GEO), including metastatic melanoma tumor specimens from an immune-checkpoint-blockade cohort (GSE115821, 37 samples), pretreatment FFPE core-needle biopsies from high-risk HER2-negative stage II/III breast cancer in the I-SPY2 trial (GSE173839, 105 pretreatment expression profiles, including 71 durvalumab/olaparib cases and 34 FFPE controls), advanced melanoma tumor biopsies collected before and during nivolumab therapy (GSE91061, 109 RNA-seq samples from 65 patients, including 51 pretreatment and 58 on-treatment samples), and advanced non-small cell lung cancer tumor tissues from patients treated with anti-PD-1/PD-L1 (GSE135222, 27 RNA-seq samples). To validate the association between GPAT4 expression and immunotherapy response across cancer types relevant to our analysis, these published ICI cohorts were included for integrated evaluation. Response information was annotated according to the original study-defined clinical categories and then harmonized as responders or non-responders. GSE115821 and GSE91061 response labels were mapped from the published melanoma clinical annotations; GSE173839 used pathological complete response (pCR) versus non-pCR as the neoadjuvant clinical endpoint; and GSE135222 used durable clinical benefit versus non-durable benefit according to Response Evaluation Criteria in Solid Tumors (RECIST) v1.1 and follow-up duration in the original study. Because these cohorts were generated from different cancer types, platforms, and preprocessing pipelines, raw expression values were not compared directly across datasets. Instead, after extraction of GPAT4 expression from each processed expression matrix, values were log-transformed and then standardized within each cohort using z-score normalization. The cohort-wise normalized GPAT4 values were subsequently combined for pooled analysis, which reduced cross-cohort batch effects while preserving the relative difference between responders and non-responders within each dataset. Group differences in integrated GPAT4 expression were then assessed between responders and non-responders. Single-cell transcriptomic analyses of PDAC and BRCA Published single-cell RNA-seq atlases of pancreatic ductal adenocarcinoma (PDAC) and breast cancer (BRCA) (Loveless et al., 2025; Xu et al., 2024) were analyzed using R v4.3.1, Seurat v5.5.0, UCell v2.16.0, msigdbr v26.1.0, and tidyverse v2.0.0. The PDAC atlas included 234 patients and 726,107 cells, and the BRCA atlas included 119 patients and 236,363 cells. Processed expression matrices, metadata, and original cell annotations were obtained from the source datasets listed in the corresponding figure legends and source-data table. Data were imported using Seurat v5.5.0, and the original annotations were used to define malignant/epithelial, stromal, endothelial, and immune compartments. UMAP visualization was generated from the processed Seurat objects using annotated cell identities. For each annotated cell population, the proportion of GPAT4-positive cells, defined as cells with normalized GPAT4 expression greater than zero, and average normalized GPAT4 expression were calculated. For tumor-cell-intrinsic analysis, malignant cells in PDAC and cancer epithelial cells in BRCA were extracted using Seurat subset and stratified into GPAT4-positive and GPAT4-negative states according to detectable GPAT4 expression in the normalized expression matrix. Differential expression of selected immune-related genes between the two states was performed using Seurat FindMarkers with the Wilcoxon rank-sum test. Pathway activity at single-cell resolution was quantified using UCell AddModuleScore_UCell. The analyzed signatures included IFN-α response, IFN-γ response, antigen processing and presentation, JAK–STAT signaling, NF-κB signaling, chemokine signaling, T-cell receptor signaling, NK-cell cytotoxicity, and additional Hallmark pathways when indicated. In the BRCA cohort, tumor-cell interferon-stimulated gene (ISG) and MHC-I/antigen-presentation scores were calculated using MSigDB-derived gene sets obtained with msigdbr. For sample-level tumor microenvironment analysis, samples with insufficient tumor-cell representation were excluded before calculating tumor-cell GPAT4 expression, leaving 150 PDAC samples and 62 BRCA samples. For each retained sample, average normalized GPAT4 expression across tumor cells was calculated as a tumor-intrinsic GPAT4 metric. Samples were ranked by this value, and the bottom and top 40% were defined as GPAT4-low and GPAT4-high groups, respectively, yielding 60 PDAC samples per group and 25 BRCA samples per group. Cell-type composition was calculated as the fraction of each annotated cell population among all cells from the same sample. Associations between tumor-cell GPAT4 expression and CD8+ T-cell fraction were assessed using Spearman's rank correlation with statscor.test. Differences in gene expression, signature scores, and cell-type fractions between groups were evaluated using two-sided Wilcoxon rank-sum tests with wilcox.test. Statistical analysis Data are presented as mean ± s.e.m. unless otherwise indicated. Statistical analyses were performed using GraphPad Prism (version 11.0.0) and R (version 4.3.1). Two-group comparisons were analyzed using two-tailed unpaired Student's t-tests when data met the assumptions of parametric testing; otherwise, appropriate nonparametric tests were used. Comparisons involving more than two groups were analyzed using one-way or two-way ANOVA, as appropriate, followed by multiple-comparison correction. Tumor-growth curves were analyzed using repeated-measures two-way ANOVA. Correlations were evaluated using Spearman's correlation coefficients, as appropriate for data distribution. Survival differences were assessed by Kaplan–Meier analysis with log-rank testing, and hazard ratios were calculated in the corresponding public cohorts where applicable. All tests were two-sided, and P < 0.05 was considered statistically significant. CRediT authorship contribution statement Zi-Lun Ruan: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Wei Liang: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation. Yu-Tong Gong: Writing – original draft, Visualization, Validation, Investigation. Wei-Tao Guan: Visualization, Formal analysis. Xuan-Ang Zhou: Investigation, Formal analysis. Kai Jiang: Writing – review & editing, Supervision. Qing Yang: Writing – review & editing, Writing – original draft, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Ming-Ming Hu: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Data availability statement All the data supporting the findings of this study are available within the article and supplementary information files, or can be obtained from the corresponding author upon reasonable request. Public datasets analyzed in this study are traceable through the accession or source identifiers provided in the Methods, figure legends, and source-data descriptions. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: The authors declare no competing interests. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments We thank the Instrument Sharing Center of the Medical Research Institute of Wuhan University for technical support. This work was supported by grants from the State Key R&D Program of China (Grant Nos. 2021YFA1302400 and 2022YFA1304900), the Fundamental Research Funds for the Central Universities (Grant Nos. 2042025kf0038 and 2042022dx0003), the National Natural Science Foundation of China (Grant Nos. 32470760 and 32170713), and the Natural Science Foundation Exploration Program (Morning Light Program) of Wuhan City (Grant No. 2025040601020204). Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.cellin.2026.100343. Supplementary data The following is the supplementary data to this article: Multimedia component 1 References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement All the data supporting the findings of this study are available within the article and supplementary information files, or can be obtained from the corresponding author upon reasonable request. Public datasets analyzed in this study are traceable through the accession or source identifiers provided in the Methods, figure legends, and source-data descriptions.