Spatiotemporally programmed nanomedicine engineering to resolve conflicting immunosignals in triple-negative breast cancer Authors: Xiuping Guo, Wensheng Zheng, Kaichao Song, Tingting Zhang, Zhigang Luo, Zhouguang Hui, Qingbo Chen, Yuting Qin, Yanan Sun, Chujuan Hu, Xiaolian Tian, Sitong Yang, Ling Ren, Quanyong Yu, Haoyang Yu, Bozhao Li, Yingying He, Yuanbin Li, Mingyu Pan, Yongsheng Che, Guangjun Nie, Jiandong Jiang, Lulu Wang Published: 2026 ABSTRACT In triple-negative breast cancer (TNBC), chemotherapy-induced immunogenic cell death (ICD) often fails to trigger truly effective antitumor immunity. This failure primarily stems from the simultaneous release of damage-associated molecular patterns (DAMPs) and immunosuppressive prostaglandin E2 (PGE2), creating an intrinsic NOT-AND signaling conflict. This barrier hinders efficient immune priming, a response rarely induced by conventional chemotherapy. To address this conflict while minimizing toxicity, R-Gem@Cel-PV, a spatiotemporally programmed nanovesicle, was designed to impose both spatial localization and sequential signal control within the tumor microenvironment. Following preferential accumulation in tumor tissue, enzymatic disassembly of the nanomedicine triggers the rapid release of celecoxib to suppress local PGE2 signaling and alleviate immune suppression. Subsequently, the delayed activation of a phospholipid-gemcitabine prodrug induces DAMP-releasing cell death. This temporal decoupling—unachievable with free drug combinations—converts gemcitabine from a weak ICD inducer into a potent one. In TNBC models, R-Gem@Cel-PV boosted dendritic cell maturation, orchestrated a robust antitumor immune response, and significantly inhibited both primary tumor growth and metastasis. These findings demonstrate that resolving the immunosignal conflict through precise spatiotemporal control is essential for effective immune engagement in TNBC and offer a generalizable strategy for reprogramming the immune response to chemotherapy in immune-refractory tumors. Introduction Traditional anticancer therapies have long viewed tumor cell death as the primary biological and therapeutic endpoint, with chemotherapy serving as a cornerstone for treating unresectable malignancies1,2. While the direct tumoricidal effects of chemotherapeutic agents are well established, the resulting cell death triggers a cascade of events that profoundly reshape the tumor microenvironment (TME), ultimately influencing therapeutic outcomes. One potential immunological benefit of chemotherapy is the induction of immunogenic cell death (ICD)3–5, a regulated form of cell death characterized by the release of damage-associated molecular patterns (DAMPs)6–8. These signals promote dendritic cell (DC) maturation and antigen presentation, linking cytotoxic therapy to adaptive antitumor immunity. However, in many solid tumors, this pro-immunogenic potential is rarely realized6,9. A major contributing factor is the concurrent induction of cyclooxygenase-2 (COX2) expression and subsequent production of prostaglandin E2 (PGE2) by several chemotherapeutic agents9,10. PGE2 is a potent immunosuppressive mediator that impairs DC function11, inhibits T-cell activity12,13, and has been proposed to act as an inhibitory DAMP (iDAMP), counteracting ICD-associated immune activation14. Thus, combining chemotherapy with COX2 or PGE2 inhibition has been explored as a strategy to resolve this signaling conflict. However, such combinations have shown limited and inconsistent benefits in both preclinical and clinical studies15–17. The reasons for this failure remain incompletely understood, representing a critical barrier to the rational design of chemoimmunotherapeutic strategies. Notably, most combination approaches have focused on the presence or absence of immunosuppressive signaling while largely neglecting the temporal coordination between immune suppression relief and ICD induction. Emerging evidence highlights that the temporal sequence of combination therapies is a key determinant of their immunomodulatory outcomes and overall efficacy18,19. For example, the CheckMate 064 study in melanoma demonstrated that initiating treatment with nivolumab followed by ipilimumab resulted in superior outcomes, whereas the reverse sequence induced immunosuppressive T-cell phenotypes, leading to resistance to subsequent nivolumab20,21. Consequently, the efficacy of immunotherapy may be influenced by the preexisting immunological landscape22. In the TME, elevated PGE2 fosters an immunosuppressive state, which is further amplified by chemotherapy-induced PGE2 signaling, impairing the immune system’s response to chemotherapy-induced DAMPs. In such a TME, dominated by immunosuppressive mediators such as PGE2, ICD-associated signals may be functionally silenced, rendering the cytotoxic event immunologically ineffective23–25. These findings support a fundamental immunological premise: the immunogenic potential of chemotherapy-induced ICD depends on a permissive TME in which PGE2-mediated immunosuppression has been alleviated. This issue is particularly pronounced in triple-negative breast cancer (TNBC), an aggressive subtype where cytotoxic chemotherapy remains the primary treatment modality26,27. The TME of TNBC is heavily reliant on PGE2 signaling, contributing to severe immune suppression and resistance to immunotherapy28,29. The disappointing outcomes of combining chemotherapy with celecoxib in TNBC models illustrate the limitations of simple coadministration strategies in this context9,30. These failures suggest that concurrent drug exposure alone is insufficient to resolve the underlying immunosignal conflict. Beyond the immunological barrier, the dense and fibrotic stroma of TNBC further hinders the synchronized delivery of free drugs, which is essential for achieving precise temporal control31. The immune signal conflict, combined with these physical barriers, highlights the urgent need for an advanced delivery platform capable of implementing a precisely coordinated, sequential therapeutic approach. To address these limitations, a spatiotemporally programmed nanomedicine, R-Gem@Cel-PV, was designed to resolve the inherent NOT-AND immunosignal conflict that undermines conventional chemoimmunotherapy (Fig. 1). This system is based on a phospholipase A2 (PLA2)-responsive gemcitabine prodrug (GPC), further functionalized with c(RGDfk) peptide to enhance spatial control. The design first exploits the overexpressed PLA2 in the TME to trigger the initial release of celecoxib, which resolves the NOT condition by preemptively blocking PGE2 synthesis. Subsequently, gemcitabine is activated by intracellular esterases, fulfilling the AND condition required for immune priming. By enforcing this spatiotemporal decoupling—unachievable with free drug combinations—this strategy aligns immune suppression relief with subsequent ICD signaling. This work establishes temporal control of immunosuppression relief as a critical determinant of chemotherapy-induced immune engagement and offers a generalizable framework for engineering nanomedicines capable of reprogramming immune-refractory tumors.Fig. 1Spatiotemporally programmed nanomedicine enables sequentially regulated chemoimmunotherapy. a Schematic of the enzyme-responsive R-Gem@Cel-PV nanomedicine, engineered for the sequential release of celecoxib and a gemcitabine prodrug (GPC) to achieve spatiotemporal control. b The underlying immunosignal conflict model governing the immune response to gemcitabine-induced tumor cell death, highlighting that the concurrent release of immunostimulatory DAMPs and inhibitory PGE2 nullifies the antitumor immune response. c The R-Gem@Cel-PV nanomedicine enforces a spatiotemporal decoupling of these opposing immunosignals. Release of celecoxib suppresses PGE2 and primes the microenvironment. The subsequent activation of gemcitabine then releases DAMPs into this nonsuppressive milieu, allowing them to function as potent immunostimulants. This programmed sequence fundamentally shifts the tumor immune microenvironment (TIME) from suppressive to immunostimulatory, orchestrating robust DC maturation and T-cell activation to drive potent immunity against both primary and metastatic tumors. Figure created with Adobe Photoshop and Adobe Illustrator Spatiotemporally programmed nanomedicine enables sequentially regulated chemoimmunotherapy. a Schematic of the enzyme-responsive R-Gem@Cel-PV nanomedicine, engineered for the sequential release of celecoxib and a gemcitabine prodrug (GPC) to achieve spatiotemporal control. b The underlying immunosignal conflict model governing the immune response to gemcitabine-induced tumor cell death, highlighting that the concurrent release of immunostimulatory DAMPs and inhibitory PGE2 nullifies the antitumor immune response. c The R-Gem@Cel-PV nanomedicine enforces a spatiotemporal decoupling of these opposing immunosignals. Release of celecoxib suppresses PGE2 and primes the microenvironment. The subsequent activation of gemcitabine then releases DAMPs into this nonsuppressive milieu, allowing them to function as potent immunostimulants. This programmed sequence fundamentally shifts the tumor immune microenvironment (TIME) from suppressive to immunostimulatory, orchestrating robust DC maturation and T-cell activation to drive potent immunity against both primary and metastatic tumors. Figure created with Adobe Photoshop and Adobe Illustrator Results Gemcitabine induces a PGE2-mediated immunosuppressive barrier in TNBC To establish a therapeutic benchmark and examine the limitations of conventional chemo-COX2 inhibition strategies, this study evaluated a physical combination of gemcitabine (Gem) and the COX2 inhibitor celecoxib (Cel). This analysis aimed to determine whether gemcitabine elicits a dominant immunosuppressive barrier that restricts the immunogenic potential of chemotherapy in TNBC. Our initial validation confirmed the clinical relevance of the COX2/PGE2 axis in TNBC. Gemcitabine consistently upregulated Ptgs2 expression and PGE2 production across a broad range of preclinical models, including patient-derived xenografts (PDXs) and tumor spheres (Supplementary Fig. 1). Moreover, elevated expression of PGE2 biosynthesis genes correlated with poor patient survival (Supplementary Fig. 2). Despite this strong rationale, the conventional combination in vivo revealed a critical therapeutic dilemma: it achieved only a modest improvement in tumor growth control (Supplementary Figs. 3a and 4) but induced significant systemic toxicity, as evidenced by weight loss (Supplementary Fig. 4d) and clear cardiotoxicity (Supplementary Fig. 5). This outcome highlights the failure of physical coadministration and points to a more complex, unresolved biological issue. It is hypothesized that this failure arose from an underlying spatiotemporal immunosignal conflict, wherein gemcitabine simultaneously triggers opposing immunosignals. On the one hand, gemcitabine induced the release of immunostimulatory DAMPs, including surface calreticulin (CRT), high-mobility group box 1 (HMGB1), and adenosine triphosphate (ATP), effects that were not impaired by celecoxib (Supplementary Figs. 6 and 7). On the other hand, gemcitabine potentiated a counteracting immunosuppressive response, as evidenced by the upregulation of intratumoral COX2 and systemic PGE2 (Supplementary Fig. 8a–d). This was further confirmed by RNA sequencing (RNA-seq) analysis, which revealed the concurrent upregulation of both immunostimulatory and immunosuppressive factors, resulting in a conflicting gene signature with enrichment of immune activation and suppression genes, creating a paradoxical state (Supplementary Fig. 8e). This observation was further supported by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, which highlighted a significant enrichment of immune-related signatures (Supplementary Fig. 8f, g). Consequently, gemcitabine monotherapy impaired DC maturation both in vivo and in vitro (Supplementary Figs. 9 and 10a–c). Although celecoxib addition partially restored DC maturation and shifted intratumoral gene expression from a tolerogenic to an immunogenic profile (Supplementary Fig. 11), the therapeutic benefit of this combination was largely dependent on CD8+ T cells (Supplementary Fig. 12), emphasizing the importance of a robust T-cell response. To provide definitive genetic evidence of the core mechanism behind this conflict, this study further demonstrated that exogenous PGE2 alone was sufficient to impair DC maturation and T-cell-mediated cytotoxicity in vitro (Supplementary Fig. 10d–h). Notably, using Ptgs2 (COX2) knockout 4T1 cells, this study showed that genetic ablation of Ptgs2 completely abolished gemcitabine-induced PGE2 production and restored the ability of gemcitabine-treated cells to promote DC maturation both in vitro and in vivo (Supplementary Fig. 13). These results indicate that while overcoming gemcitabine-potentiated PGE2 signaling is essential for an effective antitumor immune response, it cannot be safely or effectively achieved through the physical combination of free drugs. Design of a spatiotemporally programmed nanomedicine To overcome the intrinsic limitations of free drug combinations, a spatiotemporally programmed nanomedicine, R-Gem@Cel-PV, was engineered to enforce a predefined temporal order between PGE2 blockade and ICD induction. PLA2 expression is typically low in normal tissues but significantly upregulated in cancerous tissues, highlighting its potential role in tumor progression32,33. Leveraging this differential expression, a PLA2-responsive gemcitabine prodrug (GPC) was developed. PLA2 overexpression in murine orthotopic TNBC tumors, compared to normal mammary glands, was first confirmed (Supplementary Fig. 14), validating the premise for spatial control. Based on PLA2’s catalytic mechanism34, which requires substrate flexibility (Supplementary Fig. 15), a C5-linker (glutaric acid) was selected to bridge gemcitabine and the lipid backbone (1-palmitoyl-sn-glycero-3-phosphocholine, LPC), resulting in the creation of GPC. As illustrated in Fig. 2a, GPC was designed for dual responsiveness: PLA2 first hydrolyzes the sn-2 ester bond, releasing glutaryl-anchored gemcitabine (Gem-C5). Subsequently, low-specific esterases abundant in TNBC tumor cells (Supplementary Fig. 16) hydrolyze the ester bond of Gem-C5, further activating the drug.Fig. 2Design and characterization of a spatiotemporally programmed nanomedicine for sequential drug delivery. a Chemical structure of GPC, indicating cleavage sites for PLA2 and intracellular esterase. b HRMS analysis of GPC hydrolysis by PLA2 (5 U/mL), showing spectra for GPC and its product, Gem-C5. c HRMS monitoring of the sequential enzymatic conversion of GPC. GPC was incubated with PLA2 (5 U/mL) for 4 h, followed by the addition of PLE (30 U/mL) at t = 0. Spectra were acquired at the indicated time points. An untreated GPC sample is shown as a control. d Size distribution of R-Gem@Cel-PV determined by dynamic light scattering (DLS) in the presence or absence of enzyme. e Cumulative release of gemcitabine and celecoxib from R-Gem@Cel-PV over 72 h in buffer containing PLA2 (2 U/mL) and PLE (7.5 U/mL). f–h AFADESI-MSI analysis of intracellular drug levels in 4T1 cells after a 4-h treatment. The treatment groups are defined as: 1, free drug mixture of GPC and celecoxib (G + C); 2, Gem@Cel-PV; and 3, R-Gem@Cel-PV. All treatments were administered at a gemcitabine-equivalent dose of 64 μg/mL. f Workflow schematic of AFADESI-MSI analysis for molecular imaging of GPC, celecoxib, and gemcitabine in MRM mode, with their respective mass-to-charge ratios (m/z) listed. Representative ion intensity maps (g) and semi-quantitative analysis (h) of intracellular drug signals. i–l In vivo tumor targeting and biodistribution. Mice bearing 4T1 tumors were intravenously injected with the following formulations: Ctrl, untreated cells; G1, free Cy5 dye; G2, Cy5-labeled Gem@Cel-PV (Gem@Cy5-PV); G3, c(RGDfk)-modified Gem@Cy5-PV (R-Gem@Cy5-PV). i Representative whole-body fluorescence imaging at the indicated time points. j Ex vivo fluorescence imaging of tumors and major organs harvested 12 h post-treatment, with corresponding quantification of fluorescence intensity. k Representative fluorescence images of tumor sections 12 h post-injection, with nuclei stained by DAPI. Scale bar: 50 μm. l Representative CLSM images of tumor sections 12 h after treatment, stained for nuclei (DAPI) and vasculature (CD31). Scale bar: 20 μm. Data are presented as means ± SEM. b–e n = 3 independent experiments; h n = 4 independent experiments; i–l n = 3 biological replicates. Statistical analysis was performed using two-tailed Student’s t-test (for e) and one-way ANOVA with Tukey’s multiple comparisons test (for h, j). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator Design and characterization of a spatiotemporally programmed nanomedicine for sequential drug delivery. a Chemical structure of GPC, indicating cleavage sites for PLA2 and intracellular esterase. b HRMS analysis of GPC hydrolysis by PLA2 (5 U/mL), showing spectra for GPC and its product, Gem-C5. c HRMS monitoring of the sequential enzymatic conversion of GPC. GPC was incubated with PLA2 (5 U/mL) for 4 h, followed by the addition of PLE (30 U/mL) at t = 0. Spectra were acquired at the indicated time points. An untreated GPC sample is shown as a control. d Size distribution of R-Gem@Cel-PV determined by dynamic light scattering (DLS) in the presence or absence of enzyme. e Cumulative release of gemcitabine and celecoxib from R-Gem@Cel-PV over 72 h in buffer containing PLA2 (2 U/mL) and PLE (7.5 U/mL). f–h AFADESI-MSI analysis of intracellular drug levels in 4T1 cells after a 4-h treatment. The treatment groups are defined as: 1, free drug mixture of GPC and celecoxib (G + C); 2, Gem@Cel-PV; and 3, R-Gem@Cel-PV. All treatments were administered at a gemcitabine-equivalent dose of 64 μg/mL. f Workflow schematic of AFADESI-MSI analysis for molecular imaging of GPC, celecoxib, and gemcitabine in MRM mode, with their respective mass-to-charge ratios (m/z) listed. Representative ion intensity maps (g) and semi-quantitative analysis (h) of intracellular drug signals. i–l In vivo tumor targeting and biodistribution. Mice bearing 4T1 tumors were intravenously injected with the following formulations: Ctrl, untreated cells; G1, free Cy5 dye; G2, Cy5-labeled Gem@Cel-PV (Gem@Cy5-PV); G3, c(RGDfk)-modified Gem@Cy5-PV (R-Gem@Cy5-PV). i Representative whole-body fluorescence imaging at the indicated time points. j Ex vivo fluorescence imaging of tumors and major organs harvested 12 h post-treatment, with corresponding quantification of fluorescence intensity. k Representative fluorescence images of tumor sections 12 h post-injection, with nuclei stained by DAPI. Scale bar: 50 μm. l Representative CLSM images of tumor sections 12 h after treatment, stained for nuclei (DAPI) and vasculature (CD31). Scale bar: 20 μm. Data are presented as means ± SEM. b–e n = 3 independent experiments; h n = 4 independent experiments; i–l n = 3 biological replicates. Statistical analysis was performed using two-tailed Student’s t-test (for e) and one-way ANOVA with Tukey’s multiple comparisons test (for h, j). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator The synthesis of GPC was achieved through a two-step process (Supplementary Fig. 17) and characterized using 1H NMR spectroscopy and high-resolution mass spectrometry (HRMS) (Supplementary Figs. 18 and 19). To validate its responsiveness, GPC was incubated with porcine pancreatic PLA2 (ppPLA2). High-performance liquid chromatography (HPLC) analysis demonstrated that PLA2 effectively converted GPC into Gem-C5, with the hydrolytic effect being time-dependent and generalizable to other PLA2 sources (Supplementary Fig. 20a–d). At the molecular level, HRMS analysis corroborated these findings, revealing a time-dependent disappearance of GPC signals (Fig. 2b), alongside the appearance of signals for Gem-C5 and the hydrolysis product LPC (Supplementary Figs. 20e and 21). Following PLA2 activation, the introduction of porcine liver esterase (PLE) led to the hydrolysis of Gem-C5, with HRMS spectra indicating a reduction in Gem-C5 signals and confirming the release of gemcitabine (Fig. 2c and Supplementary Fig. 22). These results validate the successful construction and dual enzyme-responsiveness of the GPC prodrug. Enzyme-responsive prodrug vesicles (R-Gem@Cel-PV), functionalized with the c(RGDfk) peptide for targeted delivery, were synthesized using the thin-film hydration method to coencapsulate the GPC prodrug and celecoxib. Transmission electron microscopy (TEM) revealed an average vesicle diameter of approximately 100 nm (Supplementary Fig. 23a), while dynamic light scattering (DLS) analysis indicated a hydrodynamic diameter of 114.6 nm (Fig. 2d). Vesicle formulation was confirmed by zeta potential measurements (Supplementary Fig. 23b), and R-Gem@Cel-PV maintained colloidal stability in various media for over a week (Supplementary Fig. 24). The successful coencapsulation of both drugs was further validated by Fourier-transform infrared spectroscopy (FTIR), differential scanning calorimetry (DSC), and X-ray diffraction (XRD) analyses (Supplementary Figs. 25–26). To assess enzyme-responsive characteristics, vesicle physical integrity was first monitored. DLS analysis showed a shift from a unimodal to a multimodal size distribution following PLA2 treatment (Fig. 2d), likely due to hydrolysis of the sn-2 acyl bond and subsequent vesicle membrane fusion. Dual enzyme-triggered drug release was then investigated. At low enzyme concentrations, representative of the TME, the vesicles exhibited a sequential release profile: celecoxib was rapidly released, while gemcitabine liberation was delayed (Fig. 2e). This temporal separation arises from their distinct release pathways: celecoxib is released immediately upon vesicle hydrolysis, while gemcitabine requires a subsequent, two-step enzymatic cleavage of the GPC prodrug. In contrast, this sequential release pattern was less pronounced at higher enzyme concentrations, and minimal drug release occurred in enzyme-free media (Supplementary Fig. 27), confirming the enzyme-dependent nature of the platform. These results validate R-Gem@Cel-PV as a TME-responsive nanomedicine capable of sequential drug delivery. Tumor-targeting and intracellular execution of the programmed release Efficient execution of the programmed sequence requires robust tumor accumulation, prompting an examination of the impact of c(RGDfk)-mediated targeting on cellular uptake and intratumoral distribution. Given the high affinity of the c(RGDfk) peptide for integrin αvβ335, which is overexpressed on TNBC cells36, cellular uptake of the nanomedicine was first evaluated. Flow cytometry revealed that R-Gem@Cy5-PV was internalized by 4T1 cells significantly faster and more efficiently than the nontargeted Gem@Cy5-PV and Cy5-free control (Supplementary Fig. 28a). Moreover, R-Gem@Cy5-PV uptake was time-dependent (Supplementary Fig. 28b), and this c(RGDfk)-mediated uptake was further visualized by fluorescence microscopy (Supplementary Fig. 29). To confirm that the nanomedicine could execute its programmed function postinternalization, air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) was employed to visualize intracellular drug metabolism in real-time (Fig. 2f)37. Cells incubated with different formulations for 2, 4, and 6 h were harvested for analysis via multiple reaction monitoring (MRM) mode to track GPC, celecoxib, and gemcitabine. Ion intensity mapping revealed a time-dependent accumulation of the nanomedicine, followed by the enzymatic conversion of GPC to gemcitabine within the cells (Fig. 2g, h and Supplementary Fig. 30). Critically, the R-Gem@Cel-PV group exhibited significantly higher intracellular concentrations of GPC, celecoxib, and gemcitabine than the other formulations (Fig. 2h), highlighting the superior cellular uptake and controlled sequential drug activation. The in vivo tumor-targeting efficacy was then assessed, a prerequisite for achieving localized spatiotemporal control. IVIS imaging showed that R-Gem@Cy5-PV achieved rapid and sustained accumulation at the tumor site within 4 h of injection, whereas the signal from free Cy5 quickly diminished (Fig. 2i). This superior tumor homing was confirmed by ex vivo analysis, which revealed a significantly higher fluorescence signal in tumors from the R-Gem@Cy5-PV group compared to controls (Fig. 2j), indicating both selective accumulation and prolonged retention. Finally, the intratumoral distribution of R-Gem@Cy5-PV was evaluated. R-Gem@Cy5-PV demonstrated significantly improved and more uniform penetration in tumors compared to the nontargeted formulation (Fig. 2k). Furthermore, colocalization analysis with the vascular marker CD31 revealed that R-Gem@Cy5-PV not only targeted tumor vessels but also extravasated into the surrounding tissue (Fig. 2l). In summary, c(RGDfk)-modified R-Gem@Cel-PV vesicles exhibit robust tumor-targeting capabilities, facilitating rapid and efficient accumulation in tumor tissues. Resolution of the immunosignal conflict converts gemcitabine into an effective ICD inducer We next determined whether resolving the PGE2-mediated immunosuppressive barrier could functionally convert gemcitabine into an effective ICD inducer. Given the exceptional enzyme-responsive properties and tumor-targeting potential of R-Gem@Cel-PV, its in vitro antitumor efficacy was subsequently evaluated in 4T1 cells (Fig. 3a). R-Gem@Cel-PV significantly inhibited cell proliferation, demonstrating superior cytotoxicity compared to Gem@PV and a physical mixture of gemcitabine and celecoxib (Supplementary Fig. 31). This enhancement is likely attributed to the coloading of both drugs and the inhibition of the COX2/PGE2 pathway, which is known to promote tumor progression38. Blank vesicles showed no significant inhibitory effects (Supplementary Fig. 32), confirming their biocompatibility. Cell death induction was confirmed by flow cytometry and live/dead staining (Supplementary Figs. 33 and 34). To identify the underlying mechanism, inhibitor-based rescue assays were performed. The ferroptosis inhibitor Fer-1 robustly rescued cells from death, while the pancaspase inhibitor Z-VAD-FMK conferred only a marginal protective effect (Supplementary Fig. 35). These results indicate that ferroptosis is the predominant mode of cell death. Further experiments demonstrated cancer-selective cytotoxicity, as R-Gem@Cel-PV was potent against human MDA-MB-231 cancer cells but had negligible effects on normal MCF-10A epithelial cells (Supplementary Fig. 36). Moreover, R-Gem@Cel-PV exhibited potent antimetastatic activity. Even at subtoxic concentrations, it significantly impeded both the migration and invasion of 4T1 cells in transwell assays (Fig. 3b, c and Supplementary Fig. 37). These findings highlight the potent antitumor efficacy of R-Gem@Cel-PV in vitro, effectively suppressing cell proliferation, inducing ferroptosis-dominant cell death, and inhibiting metastasis.Fig. 3R-Gem@Cel-PV induces robust DC maturation through COX2/PGE2 pathway blockade and ICD induction. a Schematic illustrating the mechanism of R-Gem@Cel-PV in driving DC maturation through COX2/PGE2 inhibition and the release of DAMPs. b, c Inhibition of 4T1 cell migration and invasion following a 24 h treatment (0.31 μg/mL gemcitabine-equivalent). Representative images (b) and quantification (c) of migrated and invaded cells. Scale bar: 50 μm. d, e Assessment of the COX2/PGE2 signaling pathway. 4T1 cells were treated for 24 h with different formulations at a gemcitabine-equivalent dose of 1.25 μg/mL. d Representative CLSM images showing COX2 expression. Scale bar: 50 μm. e PGE2 levels in supernatants determined by ELISA. f–j Analysis of classic ICD markers. 4T1 cells were treated with various formulations for 8 h at a gemcitabine-equivalent dose of 1.25 μg/mL. f–h Analysis of surface CRT exposure. f Representative flow cytometry histograms. g Quantitative analysis of the percentage of CRT-positive cells among the live cell population. h Fold change in CRT mean fluorescence intensity (MFI) relative to the control. Extracellular release of HMGB1 (i) and ATP (j). k–n Evaluation of DC maturation. k Schematic workflow for evaluating DC activation. BMDCs were incubated with conditioned media from 4T1 cells pretreated with the indicated formulations for 24 h. l mRNA expression profile of key markers (Tnfα, Cd40, Ifng, Pdl1, Arg) in BMDCs. Flow cytometry analysis of DC maturation (CD80+CD86+), showing quantification (m) and representative plots (n). Abbreviations for treatment groups: G1, PBS (Ctrl); G2, the physical mixture of gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; G6, R-Gem@Cel-PV. Data are presented as mean ± SEM. b, c, n = 4 independent experiments; d–j, l–n, n = 3 independent experiments. Statistical analysis was performed using one-way ANOVA with Tukey’s multiple comparisons test (for c, e, g–j, l, m). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator R-Gem@Cel-PV induces robust DC maturation through COX2/PGE2 pathway blockade and ICD induction. a Schematic illustrating the mechanism of R-Gem@Cel-PV in driving DC maturation through COX2/PGE2 inhibition and the release of DAMPs. b, c Inhibition of 4T1 cell migration and invasion following a 24 h treatment (0.31 μg/mL gemcitabine-equivalent). Representative images (b) and quantification (c) of migrated and invaded cells. Scale bar: 50 μm. d, e Assessment of the COX2/PGE2 signaling pathway. 4T1 cells were treated for 24 h with different formulations at a gemcitabine-equivalent dose of 1.25 μg/mL. d Representative CLSM images showing COX2 expression. Scale bar: 50 μm. e PGE2 levels in supernatants determined by ELISA. f–j Analysis of classic ICD markers. 4T1 cells were treated with various formulations for 8 h at a gemcitabine-equivalent dose of 1.25 μg/mL. f–h Analysis of surface CRT exposure. f Representative flow cytometry histograms. g Quantitative analysis of the percentage of CRT-positive cells among the live cell population. h Fold change in CRT mean fluorescence intensity (MFI) relative to the control. Extracellular release of HMGB1 (i) and ATP (j). k–n Evaluation of DC maturation. k Schematic workflow for evaluating DC activation. BMDCs were incubated with conditioned media from 4T1 cells pretreated with the indicated formulations for 24 h. l mRNA expression profile of key markers (Tnfα, Cd40, Ifng, Pdl1, Arg) in BMDCs. Flow cytometry analysis of DC maturation (CD80+CD86+), showing quantification (m) and representative plots (n). Abbreviations for treatment groups: G1, PBS (Ctrl); G2, the physical mixture of gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; G6, R-Gem@Cel-PV. Data are presented as mean ± SEM. b, c, n = 4 independent experiments; d–j, l–n, n = 3 independent experiments. Statistical analysis was performed using one-way ANOVA with Tukey’s multiple comparisons test (for c, e, g–j, l, m). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator Next, the mechanism underlying the cytotoxicity and immunomodulatory effects of R-Gem@Cel-PV was investigated. Confocal microscopy revealed that while Gem@PV treatment upregulated COX2 expression in 4T1 cells, R-Gem@Cel-PV effectively reversed this effect (Fig. 3d). Consistent with this, R-Gem@Cel-PV significantly reduced PGE2 secretion compared to that in the Gem@PV group, confirming its inhibition of the COX2/PGE2 axis (Fig. 3e). Building on reports that both celecoxib and gemcitabine can increase reactive oxygen species (ROS)39,40, this study confirmed that R-Gem@Cel-PV induced a substantial increase in intracellular ROS (Supplementary Fig. 38). This ROS surge provides a direct mechanistic link to the observed ferroptosis-dominant cell death, as ROS generation is a critical upstream event in this cell death pathway. The ability of R-Gem@Cel-PV to trigger ICD was then assessed. Treatment with R-Gem@Cel-PV led to significant CRT exposure on the cell surface, as determined by flow cytometry and immunofluorescence (IF) (Fig. 3f–h and Supplementary Fig. 39a). It also promoted the release of HMGB1 (Fig. 3i and Supplementary Fig. 39b) and ATP (Fig. 3j). In summary, the concurrent induction of CRT, HMGB1, and ATP demonstrated the robust capacity of R-Gem@Cel-PV to elicit a classic ICD signature in TNBC cells. While gemcitabine alone can induce DAMPs, the resulting PGE2 release often creates a tolerogenic environment that impedes DC maturation. Thus, the ability of R-Gem@Cel-PV to overcome this limitation was investigated. Conditioned medium from Gem@PV-treated tumor cells induced a tolerogenic DC phenotype, characterized by upregulated Arg and pdl1 expression (Fig. 3k, l). In contrast, medium from R-Gem@Cel-PV-treated cells promoted an immunogenic profile, with increased expression of Tnfα, Cd40, and Ifng. Flow cytometry further supported these findings, revealing a higher proportion of mature DCs upon treatment with conditioned medium from R-Gem@Cel-PV-treated 4T1 and EMT6 cells (Fig. 3m, n and Supplementary Fig. 40). To confirm that PGE2 inhibition was the primary driver of these immunomodulatory effects, rescue experiments were performed (Supplementary Fig. 41). The addition of exogenous PGE2 reversed R-Gem@Cel-PV-promoted DC maturation and impaired subsequent CD8+ T-cell-mediated cytotoxicity. Furthermore, PGE2 supplementation significantly decreased the proportions of IFNγ+ and TNFα+ CD8+ T cells. These results confirm that R-Gem@Cel-PV enhances antitumor immunity primarily by inhibiting the COX2/PGE2 axis, thus resolving a key immunosuppressive conflict of gemcitabine (Fig. 3a). R-Gem@Cel-PV enhances antitumor efficacy in vivo This study further evaluated whether resolving the immunosignal conflict translated into superior therapeutic efficacy in vivo. To validate this strategy, the efficacy of R-Gem@Cel-PV was assessed in an orthotopic TNBC mouse model. The spatiotemporally programmed nanomedicine demonstrated superior tumor suppression compared to the physical combination, attributed to its spatial targeting and temporal release (Fig. 4a, b and Supplementary Fig. 42). Both the nontargeted (Gem@Cel-PV) and the targeted (R-Gem@Cel-PV) formulations significantly outperformed the simple physical mixture, highlighting the advantages of nanovesicle-based codelivery. While celecoxib-containing formulations outperformed Gem@PV, suggesting the critical role of celecoxib, R-Gem@Cel-PV achieved the most pronounced suppression of both primary tumor growth and spontaneous lung metastasis, as quantified by both macroscopic nodule counts and microscopic analysis (Fig. 4c, d and Supplementary Fig. 43).Fig. 4Therapeutic efficacy and COX2/PGE2 pathway inhibition by R-Gem@Cel-PV in an orthotopic breast cancer model. Female BALB/c mice bearing orthotopic 4T1 tumors were intravenously administered saline (G1: Ctrl), a physical mixture of free gemcitabine and celecoxib (G2: G + C), Cel-PV (G3), Gem@PV (G4), Gem@Cel-PV (G5), or R-Gem@Cel-PV (G6). Treatments were administered every three days for a total of six doses, normalized to a gemcitabine dose of 2.5 mg/kg and/or a celecoxib dose of 1.25 mg/kg. All analyses were performed at the experimental endpoint on day 25. a, b Tumor growth kinetics. a Individual tumor growth curve for each mouse. b Mean tumor volume across treatment groups over time (n = 6 mice per group). c Representative images of whole lungs, with metastatic nodules indicated by white arrows. Scale bar: 1 cm. d Quantification of the number of lung nodules. Numbers in parentheses indicate the metastasis incidence (mice with nodules/total mice per group) (n = 6 mice per group). e–g Analysis of CRT exposure on tumor cells in vivo (n = 4 mice per group). Surface CRT expression was measured by flow cytometry. e Representative histograms. f, g Quantification of the percentage of CRT-positive cells among viable cells (f) and the fold change in CRT MFI relative to the control group (g). h, i Analysis of HMGB1 release and translocation. h Plasma HMGB1 levels measured by ELISA (n = 6 mice per group). i Representative confocal images of tumor tissues showing HMGB1 (red) translocation. Scale bar: 10 µm. j, l In vivo validation of COX2/PGE2 pathway blockade. j Representative IHC images of COX2 expression in tumor sections. Scale bar: 50 μm. l PGE2 levels in plasma measured by ELISA (n = 6 mice per group). k, m Apoptosis detection in tumor tissue by TUNEL assay. Representative fluorescence images (k) and quantification of apoptotic cells, presented as fold change relative to the control group (m) (n = 4 mice per group). Scale bar: 50 μm. Data are presented as means ± SEM. Statistical analysis was performed using two-way ANOVA with Tukey’s multiple comparisons test (for b) or one-way ANOVA with Tukey’s multiple comparisons test (for d, f, g, h, l, m). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator Therapeutic efficacy and COX2/PGE2 pathway inhibition by R-Gem@Cel-PV in an orthotopic breast cancer model. Female BALB/c mice bearing orthotopic 4T1 tumors were intravenously administered saline (G1: Ctrl), a physical mixture of free gemcitabine and celecoxib (G2: G + C), Cel-PV (G3), Gem@PV (G4), Gem@Cel-PV (G5), or R-Gem@Cel-PV (G6). Treatments were administered every three days for a total of six doses, normalized to a gemcitabine dose of 2.5 mg/kg and/or a celecoxib dose of 1.25 mg/kg. All analyses were performed at the experimental endpoint on day 25. a, b Tumor growth kinetics. a Individual tumor growth curve for each mouse. b Mean tumor volume across treatment groups over time (n = 6 mice per group). c Representative images of whole lungs, with metastatic nodules indicated by white arrows. Scale bar: 1 cm. d Quantification of the number of lung nodules. Numbers in parentheses indicate the metastasis incidence (mice with nodules/total mice per group) (n = 6 mice per group). e–g Analysis of CRT exposure on tumor cells in vivo (n = 4 mice per group). Surface CRT expression was measured by flow cytometry. e Representative histograms. f, g Quantification of the percentage of CRT-positive cells among viable cells (f) and the fold change in CRT MFI relative to the control group (g). h, i Analysis of HMGB1 release and translocation. h Plasma HMGB1 levels measured by ELISA (n = 6 mice per group). i Representative confocal images of tumor tissues showing HMGB1 (red) translocation. Scale bar: 10 µm. j, l In vivo validation of COX2/PGE2 pathway blockade. j Representative IHC images of COX2 expression in tumor sections. Scale bar: 50 μm. l PGE2 levels in plasma measured by ELISA (n = 6 mice per group). k, m Apoptosis detection in tumor tissue by TUNEL assay. Representative fluorescence images (k) and quantification of apoptotic cells, presented as fold change relative to the control group (m) (n = 4 mice per group). Scale bar: 50 μm. Data are presented as means ± SEM. Statistical analysis was performed using two-way ANOVA with Tukey’s multiple comparisons test (for b) or one-way ANOVA with Tukey’s multiple comparisons test (for d, f, g, h, l, m). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator To confirm that this enhanced efficacy stemmed from our core strategy, the in vivo immunomodulatory effects of the treatment were next examined. Consistent with our hypothesis, R-Gem@Cel-PV treatment effectively dismantled the immunosuppressive axis in vivo. It significantly downregulated intratumoral COX2 expression at both the protein (Fig. 4j and Supplementary Fig. 44) and mRNA levels (Supplementary Fig. 45), and markedly reduced systemic PGE2 levels compared to the Gem@PV group (Fig. 4l). Notably, while suppressing this negative signal, the nanomedicine also amplified the positive immunogenic pathway. This began with a significant surge in intratumoral ROS (Supplementary Fig. 46), a key driver of immunogenic stress, which then translated into a superior ICD signature. Flow cytometry and IF analysis revealed that R-Gem@Cel-PV induced the most potent surface CRT exposure on viable tumor cells and the greatest release of HMGB1 (Fig. 4e–i and Supplementary Figs. 47 and 48). Notably, the targeted formulation showed a significant advantage over the nontargeted Gem@Cel-PV in these key ICD metrics, highlighting the benefit of enhanced spatial control via c(RGDfk)-mediated active targeting. The potent induction of ICD, driven by the nanomedicine’s dual immunomodulatory action, culminated in profound intratumoral cell death, as confirmed by histopathological analysis. H&E staining revealed extensive necrosis in the R-Gem@Cel-PV group (Supplementary Fig. 49a, c), while IHC analysis of Ki-67 indicated a significant reduction in cellular proliferation (Supplementary Fig. 49b, d). Concurrently, TUNEL assays highlighted a marked increase in apoptosis (Fig. 4k, m). In summary, these results highlight that R-Gem@Cel-PV’s therapeutic superiority arises from resolving gemcitabine’s inherent conflict, dismantling the COX2/PGE₂ immunosuppressive barrier while powerfully amplifying the induction of a potent ICD. This approach leads to robust control over both primary tumor growth and metastatic progression. Immunosignal conflict resolution reprograms the tumor immune microenvironment The resolution of the immunosignal conflict was hypothesized to reprogram the tumor immune microenvironment (TIME) and stimulate a robust systemic antitumor immune response. To test this hypothesis, the impact on DCs, key initiators of the immune response, was first assessed. As anticipated, enhanced ICD resulted in significant systemic maturation of DCs. Notably, the nanomedicines (Gem@Cel-PV and R-Gem@Cel-PV) induced a marked increase in mature (CD80+CD86+) and antigen-presenting (MHCIIhigh) DCs across the tumor, lymph nodes (LNs), and spleen, significantly outperforming the physical drug mixture (Fig. 5b–f and Supplementary Figs. 50 and 51). R-Gem@Cel-PV, functionalized with the c(RGDfk) peptide, consistently elicited the highest level of DC activation, highlighting the benefit of spatial targeting. This superior maturation correlated with elevated plasma IL-12p70 levels (Fig. 5g), a key cytokine produced by DCs to prime effective T-cell responses.Fig. 5R-Gem@Cel-PV reprograms the tumor immune microenvironment to activate potent antitumor immunity. All analyses were performed on tissues and immune cells isolated from orthotopic 4T1 tumor-bearing mice at the experimental endpoint (day 25), following the treatment regimen described in Fig. 4. Treatment groups: G1, saline (Ctrl); G2, a physical mixture of free gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; G6, R-Gem@Cel-PV. a Schematic of the proposed immunomodulatory mechanism of R-Gem@Cel-PV, which enhances DC maturation and CTL activation while suppressing immunosuppressive cells through resolving the immunosignal conflict between DAMPs and PGE2. b–f Analysis of DC maturation and function. b–f Quantification of mature DCs (CD80+CD86+ in CD11c+CD45+) in tumors (b), lymph nodes (c), and spleens (d). e Representative flow cytometry plots of mature DCs in tumors. f Percentage of DCs with high expression of MHCII (MHCIIhigh). g Quantification of plasma IL-12p70 levels. h, i Analysis of CTL activation. Representative flow cytometry plots (h) and quantification (i) of TNFα+ cells within CD45+CD8+ T cells. j IF staining of tumor sections showing colocalization of CD8 (green), TNFα (blue), and Granzyme B (Gzmb, yellow). Scale bar: 50 µm. k–n Assessment of immunosuppressive cell populations. k Representative CLSM images of tumors stained for Foxp3 (red) and TGFβ (green). Scale bar: 10 µm. Flow cytometry analysis of Tregs (CD25+Foxp3+ in CD45+CD4+), with representative plots shown in l and quantification in (m). n Flow cytometry quantification of MDSCs (CD11b+Ly6G+ in CD45+). Data are represented as mean ± SEM. c, d n = 5 mice per group; b, f, g, i, m, n n = 4 mice per group. Statistical significance for all quantitative analyses was assessed using one-way ANOVA with Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator R-Gem@Cel-PV reprograms the tumor immune microenvironment to activate potent antitumor immunity. All analyses were performed on tissues and immune cells isolated from orthotopic 4T1 tumor-bearing mice at the experimental endpoint (day 25), following the treatment regimen described in Fig. 4. Treatment groups: G1, saline (Ctrl); G2, a physical mixture of free gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; G6, R-Gem@Cel-PV. a Schematic of the proposed immunomodulatory mechanism of R-Gem@Cel-PV, which enhances DC maturation and CTL activation while suppressing immunosuppressive cells through resolving the immunosignal conflict between DAMPs and PGE2. b–f Analysis of DC maturation and function. b–f Quantification of mature DCs (CD80+CD86+ in CD11c+CD45+) in tumors (b), lymph nodes (c), and spleens (d). e Representative flow cytometry plots of mature DCs in tumors. f Percentage of DCs with high expression of MHCII (MHCIIhigh). g Quantification of plasma IL-12p70 levels. h, i Analysis of CTL activation. Representative flow cytometry plots (h) and quantification (i) of TNFα+ cells within CD45+CD8+ T cells. j IF staining of tumor sections showing colocalization of CD8 (green), TNFα (blue), and Granzyme B (Gzmb, yellow). Scale bar: 50 µm. k–n Assessment of immunosuppressive cell populations. k Representative CLSM images of tumors stained for Foxp3 (red) and TGFβ (green). Scale bar: 10 µm. Flow cytometry analysis of Tregs (CD25+Foxp3+ in CD45+CD4+), with representative plots shown in l and quantification in (m). n Flow cytometry quantification of MDSCs (CD11b+Ly6G+ in CD45+). Data are represented as mean ± SEM. c, d n = 5 mice per group; b, f, g, i, m, n n = 4 mice per group. Statistical significance for all quantitative analyses was assessed using one-way ANOVA with Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator A major consequence of this potent DC activation is the priming of a robust cytotoxic T-cell response. To verify this, the expression of key effector molecules in T cells was assessed. Following R-Gem@Cel-PV treatment, TNFα expression was significantly upregulated in CD8+ T cells within the TME (Fig. 5h, i), with similar increases observed in LNs and the spleen, where CD4+ T cells also exhibited elevated expression (Supplementary Figs. 52 and 53). IF staining revealed a pronounced accumulation of cytotoxic CD8+ T cells coexpressing TNFα and granzyme B (Gzmb) in tumors treated with R-Gem@Cel-PV (Fig. 5j). Moreover, R-Gem@Cel-PV treatment resulted in a marked increase in the proportion of IFNγ-expressing cells among both CD4+ and CD8+ T cells in the TME (Supplementary Fig. 54). This effect extended systemically, with similar trends observed in the spleen (Supplementary Fig. 53) and the highest frequency of IFNγ-secreting CD8+ T cells in LNs (Supplementary Fig. 52). This enhanced cellular activity was reflected in the systemic cytokine profile, where R-Gem@Cel-PV treatment elevated plasma TNFα and IFNγ levels while reducing IL-6 compared to the Gem@PV group (Supplementary Fig. 55). At the transcriptomic level, this cellular and systemic reprogramming was corroborated by increased mRNA expression of immune-activating genes, including Cd40, H2k, Ifng, Tnfα, and Gzmb (Supplementary Fig. 56). Beyond promoting a potent effector response, the impact of R-Gem@Cel-PV on the immunosuppressive network within the TME was also investigated. The effect on regulatory T cells (Tregs) and myeloid-derived suppressor cells (MDSCs), two populations strongly linked to poor clinical outcomes in breast cancer41, was examined. The data revealed a significant reduction in tumor-infiltrating Tregs following treatment with gemcitabine-functionalized prodrug vesicles (Fig. 5k–m). Notably, the codelivery formulations (Gem@Cel-PV and R-Gem@Cel-PV) induced a more pronounced decrease in Tregs (Fig. 5m). This finding was further validated by IF staining, which showed a marked absence of Tgfβ and Foxp3 colocalization in R-Gem@Cel-PV-treated tumors (Fig. 5k). A similar reduction was observed in MDSCs within the TME (Fig. 5n). At the gene level, this effect was reflected by the significant downregulation of immunosuppressive genes, including Pdl1, Ido1, Tgfβ, and Ctla4 (Supplementary Fig. 57). To provide a more comprehensive understanding of the molecular events, RNA-seq analysis of tumor tissues was performed. Gene set enrichment analysis (GSEA) revealed that R-Gem@Cel-PV treatment orchestrated a broad-based immune response, significantly enriching gene sets associated with both adaptive and innate immunity, such as antigen processing and presentation, T-cell proliferation, and interferon-gamma signaling (Supplementary Figs. 58–60). To ensure that these findings were not model specific, the therapeutic and immunomodulatory effects of R-Gem@Cel-PV were validated in an orthotopic EMT6 TNBC model. Consistent with the primary results, R-Gem@Cel-PV treatment demonstrated the most potent antitumor effect, significantly inhibiting tumor growth (Supplementary Fig. 61a–d). This superior efficacy was correlated with robust immune activation, including the highest levels of DC maturation in the tumor, spleen, and LNs (Supplementary Fig. 61e–g). Additionally, it enhanced cytotoxic CD8+ T-cell function, as evidenced by higher IFNγ+ and TNFα+ populations, while simultaneously reducing immunosuppressive Tregs within the TME (Supplementary Fig. 61h, i). In summary, the results from two distinct TNBC models demonstrated that R-Gem@Cel-PV effectively reprograms the tumor landscape. By resolving the inherent signaling conflict of gemcitabine, it unleashes a potent ICD, shifting the balance from immune tolerance to a powerful and systemic antitumor response (Fig. 5a). Antimetastatic and immune activation effects of R-Gem@Cel-PV The immune reprogramming induced by R-Gem@Cel-PV treatment was further examined for its ability to suppress metastatic dissemination and activate a systemic antitumor immune response. To assess this potential, the experimental approach progressed from the initial in vitro and orthotopic TNBC models to an aggressive experimental metastasis model. GFP-luciferase-expressing 4T1 (GFP-luc-4T1) cells were intravenously administered to female BALB/c mice, establishing systemic metastasis (Fig. 6a), which closely mirrors the clinical progression of TNBC 42.Fig. 6Antimetastatic effects and immune activation by R-Gem@Cel-PV in an experimental metastasis model of TNBC. An experimental metastasis model was established by intravenously (i.v.) injecting female BALB/c mice with 4T1 cells engineered to express green fluorescent protein and luciferase (GFP-luc-4T1). On day 3 postinoculation, mice received i.v. treatments every three days for a total of five doses. The treatment groups were as follows: G1, saline (Ctrl); G2, a physical mixture of gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; and G6, R-Gem@Cel-PV (gemcitabine-equivalent: 2.5 mg/kg; celecoxib-equivalent: 1.25 mg/kg). All analyses were performed at the experimental endpoint on day 21. a Schematic diagram of the experimental metastasis model and treatment schedule. b Representative whole-body and ex vivo lung bioluminescence imaging (BLI) on day 21. c, d Macroscopic assessment of lung metastases. c Representative photographs of excised lungs. Scale bar: 2 mm. d Quantification of surface metastatic nodules (n = 6 mice per group). e, f Histological analysis of lung metastases. e Representative H&E staining of lung sections. Scale bar: 200 µm. f Quantification of the metastatic area (n = 6 mice per group). Quantitative analysis of GFP-positive metastatic cells by flow cytometry in lung single-cell suspensions (g) and peripheral blood (h) (n = 5 mice per group). i–k Flow cytometric analysis of LNs (n = 5 mice per group). i Representative plots. Quantification of mature DCs (defined as CD80+CD86+ within the CD11c+CD45+ gate) (j) and CD8+ T cells (k). Data are presented as mean ± SEM. Statistical significance was evaluated by one-way ANOVA with Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator Antimetastatic effects and immune activation by R-Gem@Cel-PV in an experimental metastasis model of TNBC. An experimental metastasis model was established by intravenously (i.v.) injecting female BALB/c mice with 4T1 cells engineered to express green fluorescent protein and luciferase (GFP-luc-4T1). On day 3 postinoculation, mice received i.v. treatments every three days for a total of five doses. The treatment groups were as follows: G1, saline (Ctrl); G2, a physical mixture of gemcitabine and celecoxib (G + C); G3, Cel-PV; G4, Gem@PV; G5, Gem@Cel-PV; and G6, R-Gem@Cel-PV (gemcitabine-equivalent: 2.5 mg/kg; celecoxib-equivalent: 1.25 mg/kg). All analyses were performed at the experimental endpoint on day 21. a Schematic diagram of the experimental metastasis model and treatment schedule. b Representative whole-body and ex vivo lung bioluminescence imaging (BLI) on day 21. c, d Macroscopic assessment of lung metastases. c Representative photographs of excised lungs. Scale bar: 2 mm. d Quantification of surface metastatic nodules (n = 6 mice per group). e, f Histological analysis of lung metastases. e Representative H&E staining of lung sections. Scale bar: 200 µm. f Quantification of the metastatic area (n = 6 mice per group). Quantitative analysis of GFP-positive metastatic cells by flow cytometry in lung single-cell suspensions (g) and peripheral blood (h) (n = 5 mice per group). i–k Flow cytometric analysis of LNs (n = 5 mice per group). i Representative plots. Quantification of mature DCs (defined as CD80+CD86+ within the CD11c+CD45+ gate) (j) and CD8+ T cells (k). Data are presented as mean ± SEM. Statistical significance was evaluated by one-way ANOVA with Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 compared to the indicated groups. Figure created with GraphPad Prism and Adobe Illustrator After a 21-day treatment course, bioluminescence imaging (BLI) demonstrated that R-Gem@Cel-PV treatment significantly suppressed metastatic progression, as evidenced by a marked reduction in lung metastases compared to saline-treated controls (Fig. 6b), further supported by semi-quantitative photon flux analysis (Supplementary Fig. 62). Although the free drug combination had some effect, lungs from R-Gem@Cel-PV-treated mice exhibited minimal BLI signals, which were consistent across both in vivo and ex vivo imaging (Fig. 6b and Supplementary Fig. 62). The antimetastatic impact was corroborated by multiple assessments. R-Gem@Cel-PV treatment substantially reduced both the number of metastatic lung nodules and overall lung weight (Fig. 6c, d and Supplementary Fig. 62). Histopathological analysis via H&E staining revealed a significant decrease in the metastatic area within the lungs (Fig. 6e, f). These findings were further confirmed by IF staining, which showed a notable reduction in metastatic burden and a striking decrease in GFP-positive metastatic foci within the lung sections (Supplementary Fig. 63). Flow cytometry was then employed to precisely quantify tumor dissemination. Analysis revealed a remarkable reduction in GFP-positive cancer cells within both the lungs (Fig. 6g) and circulating blood (Fig. 6h) of R-Gem@Cel-PV-treated mice. These observations highlight the potent efficacy of R-Gem@Cel-PV in limiting both metastatic colonization and systemic spread. Considering the immune system’s pivotal role in regulating metastatic outgrowth43, the immunological mechanisms underlying this therapeutic efficacy were explored. Flow cytometric analysis of LNs revealed that our codelivery formulations induced a robust immunogenic response. Notably, the targeted R-Gem@Cel-PV formulation triggered significantly stronger DC maturation and greater accumulation of CD8+ T cells than the nontargeted formulation (Fig. 6i–k), demonstrating the advantage of c(RGDfk)-mediated targeting in enhancing antitumor immune activation within critical lymphoid tissues. A similar pattern of systemic immune activation was also observed in the spleen (Supplementary Fig. 64), further validating these results. Antitumor vaccination effects of R-Gem@Cel-PV against breast cancer To assess the capacity of R-Gem@Cel-PV to induce an antitumor vaccination effect, a standard vaccination-rechallenge assay was performed. In this experiment, mice were first vaccinated with 4T1 cells pretreated with the indicated formulations and subsequently rechallenged with live 4T1 cells one week later (Supplementary Fig. 65a). As shown in Supplementary Fig. 65b–d, mice vaccinated with freeze-thawed cells (negative control) or those from the sham group displayed unimpeded tumor growth. Vaccination with Gem@PV-treated cells also failed to suppress tumor progression. In contrast, formulations combining celecoxib with gemcitabine significantly delayed tumor growth. Notably, the R-Gem@Cel-PV group demonstrated a potent protective effect comparable to that of the mitoxantrone-positive control, emphasizing the importance of COX2/PGE2 pathway blockade in enhancing antitumor immunity. These results confirm that R-Gem@Cel-PV induces a robust antitumor vaccination effect by amplifying the immunogenicity of chemotherapy-induced cell death. In vivo biosafety evaluation Given the enhanced therapeutic efficacy of R-Gem@Cel-PV, its in vivo biosafety was evaluated. Mice were treated with R-Gem@Cel-PV or saline over a 28-day period (Supplementary Fig. 66a). No significant differences in body weight were observed between the treatment and control groups (Supplementary Fig. 66b), suggesting good general tolerance. Further safety assessments were conducted through histopathological analysis of major organs (heart, liver, kidneys, spleen, and lungs), which revealed no significant abnormalities (Supplementary Fig. 66c). This was further supported by biochemical evaluations, where liver function markers (ALT, AST), kidney function indicators (urea, creatinine), and general damage marker (LDH) all remained within normal physiological ranges. Given the potential cardiovascular risks associated with celecoxib44,45, cardiovascular biomarkers (CK, α-HBDH) were also measured and found to be comparable to the control group, suggesting that the nanomedicine mitigates this side effect (Supplementary Fig. 66d). These results demonstrate that R-Gem@Cel-PV possesses a favorable safety profile in vivo, supporting its potential for clinical application. Discussion A central challenge in chemotherapy-based cancer treatment is that its ability to induce ICD is often coupled with the simultaneous activation of immunosuppressive pathways. This issue is particularly pronounced in TNBC, where gemcitabine induces both immunostimulatory DAMPs and the potent immunosuppressive mediator, PGE2. These opposing signals arise from the same therapeutic event and operate on overlapping spatial and temporal scales, resulting in an intrinsic immunosignal conflict. This conflict, characterized by a NOT-AND signal configuration, requires not only the generation of DAMPs but also the prior relief of PGE2-mediated immunosuppression to effectively prime the immune system (Supplementary Fig. 67). This reasoning explains why the immunogenic potential of chemotherapy is often not fully realized and why simple drug combinations or intensified pathway inhibition frequently leads to inconsistent or disappointing outcomes. Notably, this conflict cannot be resolved through conventional coadministration strategies. When immunosuppressive and immunogenic signals are delivered simultaneously, even effective blockade of PGE2 signaling may fail to restore productive immune activation, as the dominance of immunosuppression can functionally silence ICD-associated signals46. Our findings suggest that the core limitation is not the availability of the appropriate molecular components, but rather the inability of conventional approaches to establish the necessary temporal sequence between immune suppression relief and ICD induction. This insight provides the conceptual foundation for engineering spatiotemporally programmed therapeutic strategies, rather than relying on physical drug combinations with mismatched pharmacokinetics. To overcome this limitation, this study developed R-Gem@Cel-PV, a spatiotemporally programmed nanomedicine designed to decouple these opposing immunosignals via controlled spatial localization and enforced temporal sequencing. By leveraging tumor-associated enzymatic features, R-Gem@Cel-PV preferentially executes its programmed cascade within the TME. The addition of c(RGDfk)-mediated targeting further reinforces tumor-localized execution, enhancing the robustness of spatiotemporal control rather than serving as an independent therapeutic driver. Notably, this design ensures that PGE2-driven immunosuppression is alleviated prior to the initiation of DAMP-releasing cytotoxicity. Using AFADESI-MSI, this study directly visualized this sequential activation in situ, providing mechanistic evidence that precise temporal control—difficult to achieve with free drug combinations—can be reliably implemented at the cellular level. This capacity to enforce immune signal order represents a fundamental distinction between programmed nanomedicine and conventional combination therapy. Moreover, the functional consequence of resolving this immunosignal conflict is a coordinated reprogramming of the TIME. Our study demonstrates that R-Gem@Cel-PV significantly enhances DC maturation and strengthens tumor-specific CD8+ T-cell responses, while simultaneously reducing immunosuppressive cell populations, such as Tregs and MDSCs. These immune alterations led to superior control of both primary tumor growth and metastatic dissemination compared to the physical coadministration of gemcitabine and celecoxib. These findings reinforce the idea that alleviating local immunosuppression is essential for effectively harnessing chemotherapy-induced ICD. Aligning with emerging evidence47–49, this study highlights that therapeutic sequencing, rather than drug identity alone, determines the success of immune-mediated treatment outcomes. Validating that a programmed therapeutic cascade operates as intended within target cells is a central challenge in nanomedicine. Many studies rely on indirect evidence from bulk in vitro assays to infer this behavior. Here, a notable strength of our study is that we moved beyond such inference by using AFADESI-MSI to directly visualize the sequence of molecular events within intact cancer cells. This approach provided unambiguous intracellular evidence of the temporal program, establishing the critical mechanistic link between the design of nanomedicine and its observed therapeutic function. With this mechanistic certainty, we then demonstrated that this precisely controlled sequence translates into broad therapeutic efficacy across multiple TNBC models. While our study supports the preclinical safety and efficacy of R-Gem@Cel-PV in both primary and metastatic TNBC models, several limitations warrant attention. Further research is needed to investigate the temporal dynamics of immune responses in vivo, including the durability of antitumor immunity and the formation of long-term immune memory. Additionally, ongoing optimization of formulation parameters and pharmacokinetic properties will be critical for facilitating clinical translation and ensuring reproducibility and scalability. Beyond its immediate therapeutic implications, this work offers a broader conceptual framework for understanding the inconsistent performance of chemoimmunotherapy combinations. The limited success of COX2 inhibition observed in preclinical and clinical studies may reflect an unrecognized temporal conflict between immunogenic and immunosuppressive signals, rather than inadequate pathway blockade. From this perspective, chemotherapy-induced immunosuppression should be viewed not merely as a parallel adverse effect but also as an active inhibitory signal that must be strategically neutralized before immune activation can occur. This perspective implies that for such combination strategies to succeed, the principles of targeted therapy must evolve, from solely targeting which molecular components are present to precisely controlling when critical signaling events occur. Accordingly, biomarkers for chemoimmunotherapy may benefit from incorporating iDAMPs, such as PGE2, alongside immunostimulatory DAMPs to more comprehensively capture the immune context of the TME46,50. Furthermore, this work highlights the dual utility of our spatiotemporally programmed nanomedicine. It serves not only as a promising platform for delivering established sequential regimens more effectively, but also as a valuable research tool to explore the optimal temporal scheduling required to maximize synergy between different targeted agents. By consolidating complex treatment protocols into a single administration, this approach could enhance efficacy, reduce toxicity, and improve patient compliance. In conclusion, this study identifies an intrinsic immunosignal conflict as a major barrier to effective chemotherapy-induced immune engagement and demonstrates that resolving this conflict requires precise spatiotemporal control, rather than simple drug coadministration. By creating a TME permissive for immune activation through localized alleviation of PGE2-driven immunosuppression, R-Gem@Cel-PV enables a productive antitumor immune response that cannot be achieved through conventional combination therapies. These findings position temporal programming as a critical determinant of chemoimmunotherapy success and provide a generalizable framework for the rational design of programmed nanomedicines to improve therapeutic outcomes in immunologically resistant tumors. Materials and methods Further detailed materials, experimental procedures, and statistical analysis methods are available in the Supplementary Materials section. Ethics approval statements Triple-negative breast cancer specimens were obtained from patients at Anyang Tumor Hospital, Henan University of Science and Technology. All participants provided informed consent in compliance with the Declaration of Helsinki. The entire study protocol received approval from institutional review board of Anyang Tumor Hospital (approval no. AF-SOP-07-1.2-01). All animal experiments received approval from the Animal Experimental Ethics Committee of the Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences, and Peking Union Medical College, with approval number IMB-20231221D102. The mice were accommodated within the institute’s animal facility and were strictly maintained under pathogen-free conditions, adhering rigorously to the committee’s requirements. Materials 1-Palmitoyl-sn-glycero-3-phosphocholine (LPC), 1,2-Dipalmitoylphosphatidylcholine (DPPC), cholesterol, and soy lecithin were obtained from AVT Pharmaceutical Tech Co., Ltd. (Shanghai, China). DPPE-PEG2K-NHS and c(RGDfk) peptides were purchased from Xi’an Ruixi Biological Technology Co., Ltd. (Xi’an, China) Solvents and reagents including dichloromethane (DCM), 4-dimethylaminopyridine (DMAP), N,N-dimethylformamide (DMF), and 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride (EDCI), glutaric anhydride, gemcitabine, and celecoxib were purchased from Aladdin (Shanghai, Beijing). Methanol and acetonitrile were obtained from Thermo Fisher Scientific (MA, USA). Phospholipase A2 (PLA2) from honey bee venom and porcine pancreatic, as well as esterase from porcine, were purchased from Sigma (MO, USA). Chloroform and trifluoroacetic acid (TFA) were purchased from Beijing TongGuang Fine Chemicals Company (Beijing, China). Prostaglandin E2 (PGE2) and mitoxantrone were purchased from MedChemExpress (NJ, USA). Fluorescein diacetate was purchased from Yeasen Biotechnology (Shanghai, China). The PLA2G2A (PA5-102403) antibody was sourced from Invitrogen (CA, USA), while antibodies against iPLA2 (22030-1-AP), CRT (10292-1-AP), HMGB1 (10829-1-AP), GAPDH (10494-1-AP), COX2 (66351-1-Ig), as well as HRP-conjugated secondary antibodies were supplied by Proteintech (IL, USA). cPLA2 (ab227246) was obtained by Abcam (MA, USA). Synthesis of PLA2-responsive gemcitabine prodrug GPC First, the intermediate LPC-COOH was synthesized. To a stirred solution of LPC (496 mg, 1 mmol) and glutaric anhydride (342 mg, 3 mmol) in 5 mL anhydrous DCM was added DMAP (367 mg, 3 mmol) at room temperature. The resulting solution was stirred at 40 °C overnight. After completion of reaction, the solution was concentrated under reduced pressure. LPC-COOH (550 mg, yield 90%) was obtained by recrystallization from ethanol. Subsequently, the final prodrug GPC was prepared by dissolving LPC-COOH (488 mg, 0.8 mmol) and gemcitabine (211 mg, 0.8 mmol) in 4 mL anhydrous DMF. Then EDCI (153 mg, 0.8 mmol) and DMAP (10 mg, 0.08 mmol) were introduced. The mixture was allowed to react overnight with stirring at room temperature. The final product was purified using pre-HPLC system with isocratic elution (methanol/water = 9/1, 0.1%TFA) to obtain pure GPC (200 mg, yield 29%). Characterization of GPC The purity of GPC was determined by HRMS analysis employing a Thermo LCQ Deca XP Maz mass spectrometer (MA, USA) with electrospray ionization (ESI) in positive ion mode. The 1H-NMR spectrum was performed on a Bruker 400 MHz NMR spectrometer (Bruker, Germany) with CD3OD as solvent. Enzyme-responsive assessment of GPC GPC (1 mg/mL) was incubated in a buffer composed of 5 mM HEPES (Pricella Biotechnology, Wuhan, China) and 5 mM CaCl2 (Beyotime, Shanghai, China) in saline. The enzymatic reaction was initiated by introducing PLA2 (5 U/mL). Samples were collected at designed time points (5, 15, 30 min; 1, 2, 4, 6, 8, 12, and 24 h), and subjected to analysis by HPLC (Shimadzu, Japan). The chromatographic conditions were as follows: mobile phase A (0.075% TFA in water) and mobile phase B (acetonitrile) at a ratio of 20: 80 (v/v), with a constant flow rate of 1 mL/min, using a TC-C18 column (5 μm; Agilent, CA, USA) at 35 °C. The injection volume was 10 μl, and detection was conducted at 269 nm. To monitor the molecular conversion of GPC, samples incubated with PLA2 (5 U/mL), or with a combination of PLA2 (5 U/mL) and esterase (30 U/mL) at specific time points were determined by HRMS (Thermo, MA, USA). Preparation of R-Gem@Cel-PV The thin-film hydration method was employed to fabricate R-Gem@Cel-PV. Specially, a mixture of soy lecithin, cholesterol, DPPE-PEG2000, c(RGDfk)-DPPE-PEG2000, celecoxib (Cel), and GPC (at a mass ratio of 10:1:0.6:0.4:0.8:5.2) was dissolved in a chloroform-methanol system (volume ratio 8:2). A thin, lipid-like film was formed by rotary evaporation under reduced pressure. Then, the resulting film was hydrated for 30 min at 40 °C with an appropriate solvent to form macrovesicles. These vesicles were then subjected to an ultrasound-pause cycle (5 seconds of ultrasound followed by a 5-second pause) for 2 min to produce the uniform R-Gem@Cel-PV. The non-targeted vesicles were prepared similarly, but with a component mass ratio of 10:1:1:0.8:5.2 for soy lecithin, cholesterol, DPPE-PEG2000, Cel, and GPC, respectively. Characterization of R-Gem@Cel-PV DLS (Malvern, UK) was utilized to measure the particle size, polydispersity index (PDI), and zeta potential of R-Gem@Cel-PV. For morphology examination, the nanovesicles were negative stained with phosphotungstic acid and observed using a TEM (JEOL, Japan). To quantify the encapsulation efficiency of celecoxib, HPLC (Shimadzu, Japan) was employed under the following conditions: the mobile phase consisting of water with 0.075% TFA (phase A) and acetonitrile (phase B), a flow rate of 1 mL/min with 20% A and 80% B, using a TC-C18 column (Agilent) at 35 °C. Detection for celecoxib and gemcitabine was carried out at 254 nm and at 269 nm, respectively, with a sample injection volume of 10 μl. The stability of R-Gem@Cel-PV was assessed over 7 days in water, PBS, and 10% FBS, by monitoring changes in particle size and PDI using DLS. Enzyme-responsive performance of R-Gem@Cel-PV The enzyme-responsive behavior of R-Gem@Cel-PV was assessed through a 24-h incubation of the samples in PLA2-supplemented buffer. Changes in particle size were recorded. The release profile of gemcitabine from R-Gem@Cel-PV was analyzed by incubation with PLA2 at either high (10 U/mL) or low (2 U/mL) concentrations, along with esterase at high (75 U/mL) or low (7.5 U/mL) concentrations. The release profile of gemcitabine and celecoxib was analyzed utilizing HPLC (Shimadzu, Japan) under the same conditions as described previously. Cell lines The murine cancer cell line, 4T1, B16, CT26, and EMT6, and the human cell lines MDA-MB-231, SW620, HT29, MCF-10A, and A549 were used in this study. Cell lines source was as follows: 4T1, B16, CT26, and MDA-MB-231 were purchased from the American Type Culture Collection (ATCC, VA, USA); SW620 and HT29 were from Procell Life Science & Technology (Wuhan, China). MCF-10A, EMT6, and A549 were generously provided by Professor Ke Li (Institute of Medicinal Biotechnology, CAMS & PUMC). Cellular uptake detection Cellular uptake was assessed in 4T1 cells seeded in 12-well plates. The cells were incubated with Cy5, Cy5-PV, or R-Cy5-PV (0.25 μg/mL Cy5-equivalent) for 1, 2, or 4 h. The incubation was terminated by washing the cells three times with PBS, followed by fixation in 4% PFA. Nuclei staining was performed using DAPI (Beyotime). The uptake of different formulations was monitored using a fluorescence microscope (Zeiss, Oberkochen, Germany). Quantitative analysis was conducted with a CytoFLEX flow cytometer (Beckman, CA, USA). The specific groups are listed in Supplementary Table 1. AFADESI-MSI analysis and data processing 4T1 cells (1 × 105 cells per well), seeded onto cell culture slides in 24-well plates, were incubated with RPMI 1640 medium containing different formulations (equivalent to 64 µg/mL gemcitabine) for 2, 4, and 6 h. After treatment, the samples were subjected to PBS washes and then freeze-dried for 24 h. Intracellular uptake and metabolism of GPC and celecoxib were analyzed by AFADESI-MSI on a QTRAP 5500 mass spectrometer (AB SCIEX, CA, USA) coupled to an AFAI ion source. The spray solution, composed of methanol and water (4:1, v/v, with 0.1% formic acid), was delivered at 10 µL/min. The system utilized nitrogen gas at 2 L/min, a spray voltage of 5000 V, and an extracting airflow of 35 L/min. Multiple reaction monitoring (MRM) analysis was performed to detect GPC (m/z 855 → 184), celecoxib (m/z 382 → 362), and gemcitabine (m/z 264 → 112). Consistent with previous reports37, a dwell time of 200 ms was applied for each transition. Analyst software was used for data acquisition and initial processing, while custom Matlab software handled the final imaging data visualization. Supplementary Table 1 provides a detailed list of the experimental groups. In vitro evaluation of ICD markers 4T1 cells were cultured in confocal chambers (Nest, Wuxi, China) for 24 h before being treated with various formulations at a consistent concentration (equivalent to 1.25 µg/mL gemcitabine) for a further 8 h. Following treatments, cells underwent PBS washes for three times, fixation in 4% PFA, and subsequent blocking of nonspecific sites with 5% bovine serum albumin (BSA). Next, cells were probed overnight at 4 °C using primary antibodies against CRT or HMGB1. After three times subsequent PBS washes, a secondary antibody incubation was performed at room temperature for 2 h. Finally, cell nuclei were counterstained with DAPI for 10 min. All samples were examined using a confocal microscope (Olympus, CA, USA). To quantify surface CRT exposure, treated 4T1 cells were stained with a CoraLite® Plus 555-conjugated anti-calreticulin polyclonal antibody (Proteintech) and the viability dye Sytox Blue (Invitrogen, S34857) before analysis on a CytoFLEX flow cytometer. Representative gating strategies are shown in Supplementary Fig. 68. Additionally, HMGB1 levels in cell supernatants and ATP levels were determined utilizing a commercial HMGB1 ELISA Kit (Fankew Corporation, Shanghai, China) and an ATP Assay Kit (Beyotime), respectively. Animals Female BALB/c mice (aged 6–8 weeks) and female NOD SCID mice (aged 8 weeks) were procured from Vital River Laboratory Animal Technology (Beijing, China). Orthotopic tumor model and antitumor treatment Subcutaneous injections of 100 μL 4T1 tumor cell suspension (5 × 105 per mouse in a 1:1 PBS/Matrigel mixture) were administered into the mammary fat pads of mice. All mice were earmarked and randomly divided into various groups. Therapy commenced when tumor volumes reached approximately 50 mm3. To investigate the antitumor efficiency of the gemcitabine-celecoxib combination (G + C), tumor-bearing mice received administrations of saline, free gemcitabine (2.5 mg/kg), free celecoxib (1.25 mg/kg), or their combination (G + C). All treatments were administered intravenously every three days for six doses (days 7-22). Tumor volumes and body weights were monitored every two days. On day 25, the experiment was terminated, and mice were euthanized for endpoint analyses. Tumor dimensions were assessed using a vernier caliper, and volumes (V) were calculated according to the formula: V = 0.5 × length × width2, where length and width represent the largest and smallest diameters, respectively. The experimental workflow is shown in Supplementary Fig. 3, and the experimental groups are provided in Supplementary Table 2. To evaluate the efficacy of R-Gem@Cel-PV, the mice were randomized into six groups: saline, a physical mixture of G + C, Gem@PV, Cel-PV, Gem@Cel-PV, or R-Gem@Cel-PV. Treatments were administered intravenously every three days for six cycles from day 7 to day 22. All formulations were dosed at an equivalent of 2.5 mg/kg for gemcitabine and/or 1.25 mg/kg for celecoxib. Tumor volumes and body weight were monitored bi-daily. After 18 days of treatment, mice were euthanized and dissected, followed by the quantitative analyses of tumor weights and numbers of lung metastatic nodules. The experimental workflow is shown in Supplementary Fig. 3, and the experimental groups are listed in Supplementary Table 3. Antitumor activity on experimental lung metastasis tumor model To establish the experimental lung metastasis tumor model, GFP-luc-4T1 cells (1 × 105 per mouse) were administered to BALB/c mice through intravenous tail vein injection. Subsequently, mice were earmarked and randomly separated to groups. Three days post-implantation, the mice were commenced treatments with intravenous injections every three days for 5 cycles, receiving saline, a physical mixture of gemcitabine and celecoxib, Gem@PV, Cel-PV, Gem@Cel-PV, or R-Gem@Cel-PV. All formulations were dosed at an equivalent of 2.5 mg/kg for gemcitabine and/or 1.25 mg/kg for celecoxib. After 21 days post-implantation, in vivo bioluminescent living imaging was conducted to evaluate the development of metastatic conditions under varying treatments. Briefly, mice were administered luciferin (Promega, WI, USA) intravenously, anesthetized, and imaged using the IVIS Spectrum optical imaging system (PerkinElmer, OH, USA). Following euthanasia and dissection, ex vivo imaging was performed to assess the metastatic conditions. Lung weights and the number of metastatic nodules were quantified. To analyze the proportion of GFP-positive metastatic nodules or disseminated cancer cells, lung tissues were processed into single-cell suspensions and determined via flow cytometry. Additionally, lung samples were cryo-embedded in optimal cutting temperature (OCT) compound (Solarbio), cryo-sectioned at 8 μm, stained with DAPI, and subsequently imaged. The experimental workflow is shown in Supplementary Fig. 3, and the experimental groups are listed in Supplementary Table 3. Antitumor vaccination assay in vivo To prepare the cellular vaccines, 4T1 cells were treated under several conditions. The experimental groups consisted of cells pretreated for 48 h with the physical mixture of Gem and Cel (G + C), Gem@PV, Gem@Cel-PV, or R-Gem@Cel-PV. For controls, cells treated with mitoxantrone (MIT) served as the positive control, while cells killed by three freeze-thaw cycles served as the negative control. These treated cells were then inoculated into the mammary fat pad of mice (1 × 106 per mouse) for vaccination. A sham-vaccinated group receiving only PBS was also included. One week later, all mice were confirmed to be tumor-free in the left flank. These mice were then challenged with living 4T1 cells (1 × 105 per mouse) in the contralateral mammary fat pad. Tumor volumes were subsequently monitored bi-daily over a period of 18 days. The experimental workflow is shown in Supplementary Fig. 3, and the experimental groups are listed in Supplementary Table 4. Analysis of antitumor immune response by flow cytometry Orthotopic 4T1 tumors and experimental lung metastasis models were established as previously described. At the endpoint, tumors, lymph nodes, and spleens were collected and processed to single-cell suspensions, which were then incubated with a cocktail of antibodies in FACS buffer for 2 h at 4 °C in the dark. For intracellular and intra-nuclear cytokines determination, cells were stimulated with PMA (50 ng/mL; Invitrogen), ionomycin (1 μg/mL; Sigma), GolgiStop (1 μg/mL; BD), and Brefeldin A (1 μg/mL, Biolegend) for 4 h at 37 °C. After stimulation, cells were stained with fluorescent antibodies, including CD45 (catalogue number 103116, clone 30-F11 or catalogue number 157208, clone S18009F; Biolegend), CD4 (catalogue number 100516, clone RM4-5; Biolegend), CD8 (catalogue number 553030, clone 53-6.7; BD Biosciences), CD11b (catalogue number 101206, clone M1/70; Biolegend), CD11c (catalogue number 117318, clone N418; Biolegend), CD80 (catalogue number 104714, clone 16-10A1; Biolegend), CD86 (catalogue number 105008, clone GL-1; Biolegend), CD25 (catalogue number BB515, clone PC61; BD Biosciences), Foxp3 (catalogue number126403, clone MF-14; Biolegend), Ly6G (catalogue number 127616, clone 1A8; Biolegend), IFN-γ (catalogue number 505850, clone XMG1.2; Biolegend), TNF-α (catalogue number 506323, clone MP6-XT22; Biolegend), MHCⅡ (catalogue number 47-5321-82, clone M5/114.15.2, Invitrogen), according to Manufacturer’s instructions and detected by flow cytometry using a CytoFLEX (Beckman, CA, USA). For detection of CRT exposure on the surface of live tumor cells, PI and CoraLite® Plus 555-conjugated anti-calreticulin polyclonal antibody (Proteintech) were stained before analysis on a CytoFLEX. Representative gating strategies are shown in Supplementary Figs. 69–72. qRT-PCR Total RNA was first extracted from either cellular or tumor samples employing TRIzol reagent (Thermo) and a PureLink RNA Mini Kit (Thermo), strictly following the manufacturer’s protocols. Subsequently, a HiFiScript cDNA Synthesis Kit (Cwbio) was employed to reverse transcribe the extracted RNA into cDNA. The qPCR amplification step was performed on a 7500 Fast Real-Time PCR System (Thermo) utilizing FastSYBR Mixture (Cwbio). The specific primer sequences utilized for the PCR are provided in Supplementary Tables 5 and 6. Statistical analysis All statistical analyses were conducted utilizing GraphPad Prism software (Version 10). For assessing differences between two groups, a two-tailed Student’s t-test was applied. In cases involving more than two groups, either one-way or two-way analysis of variance (ANOVA) followed by Tukey’s multiple comparisons test was used, as specified in the corresponding figure legends. Statistical significance was set at p-value < 0.05 (*p-value < 0.05, **p-value < 0.01, ***p-value < 0.001, ****p-value < 0.0001), each denoting increasing levels of statistical confidence. The survival analyses were performed using the Kaplan–Meier method. All quantitative data are presented as the mean ± standard error of the mean (SEM). For in vitro experiments, n refers to the number of independent experiments, while for in vivo studies, n represents the number of mice per group. Supplementary information Supplementary materials Supplementary materials for uncropped blots Supplementary materials Supplementary materials for uncropped blots