Targeting the Golgi apparatus enhances PD-L1 blockade and synergizes with oxaliplatin to improve immunotherapy efficacy.

Results

PCR-Lip disrupts the structure of the Golgi apparatus and impairs its function
To develop a more effective tumor-specific PD-L1 blockade strategy, we engineered a liposomal delivery system modified with CR (chondroitin sulfate [CS] conjugated to RA via an ester bond), previously demonstrated to disrupt the Golgi apparatus (15). These liposomes were further functionalized with PDPPA-1, a bifunctional peptide comprising a PD-L1 inhibitory segment (DPPA-1) and an MMP-2-sensitive linker, yielding the final formulation termed PCR-Lip. As a control, we also prepared PCD-Lip, which lacks Golgi-disrupting capability (Fig. S1). Notably, the DPPA-1-responsive design facilitates the selective release of DPPA-1 in the tumor microenvironment (Fig. S2A), ensuring preferential accumulation in CD44+ tumor cells because of the enhanced affinity of CS for CD44 receptors (Fig. S2, B and C).
To investigate the Golgi-targeting capability and structural interference of PCR-Lip, we treated 4T1 breast cancer cells with a low-cytotoxicity dose of PCR-Lip (Fig. S2D). As expected, CS facilitated efficient Golgi targeting via caveolae/raft-mediated endocytosis (16) (Fig. 1A and Fig. S2E). Transmission electron microscopy (TEM) revealed that PCR-Lip treatment led to disassembly of the Golgi ribbon structure into fragmented ministacks (Fig. 1B). Immunofluorescence analysis further confirmed significant fragmentation of GM130 and Syntaxin-6, markers of the cis- and trans-Golgi networks, respectively (Fig. 1, C and D), indicating profound structural disruption. This disorganization was more severe than that induced by brefeldin A, a known Golgi-disrupting agent. Consistent with structural impairment, the enzymatic activities of two key Golgi-resident enzymes, mannosidase II (Man II) and UDP-galactosyltransferase (GalT), were significantly reduced following PCR-Lip treatment (Fig. 1E). These findings confirm that PCR-Lip effectively delivers RA to the Golgi apparatus, disrupts its structure, and compromises its functional integrity.

PCR-Lip inhibits PD-L1 glycosylation, membrane localization, and exosomal secretion
To assess whether PCR-Lip can comprehensively block PD-L1 at multiple regulatory levels, namely glycosylation, membrane presentation, and exosome-mediated secretion (Fig. 2A), we incubated liposomes with 4T1 cells. PCR-Lip treatment resulted in a marked reduction in PD-L1 glycosylation (Fig. 2B), likely because of interference with the glycan modifications of four conserved asparagine (N) residues (N35, N191, N199, and N218), potentially altered to glutamine (Q) residues (Fig. S3). Notably, the stability of nonglycosylated PD-L1 decreased over time following PCR-Lip exposure (Fig. 2C), suggesting that glycosylation disruption compromises PD-L1 protein stability.
Flow cytometric analysis revealed significantly lower membrane-associated PD-L1 in the PCR-Lip group compared with DPPA-1 and CR-Lip controls (p < 0.01), indicating that DPPA-1 conjugation efficiently inhibits PD-L1 surface expression (Fig. 2D). To evaluate effects on exosomal PD-L1 secretion, we isolated exosomes from treated 4T1 cells (Fig. S4). TEM with PD-L1 immunogold labeling confirmed substantially reduced PD-L1 content in exosomes after PCR-Lip treatment (Fig. 2, E and F). In addition, PCR-Lip markedly decreased the expression of membrane transport proteins Abcd1, Abcd3, and Tap1 (17, 18, 19), which are regulated via exosome-associated pathways (20) (Fig. S5). Collectively, these findings indicate that PCR-Lip simultaneously disrupts multiple steps of PD-L1 regulation, effectively blocking its immunosuppressive functions in tumor cells.

OPCR-Lip enhances antitumor immunity via induction of ICD
To further augment antitumor efficacy beyond PD-L1 blockade, we engineered OPCR-Lip by coencapsulating the chemotherapeutic agent OXA within PCR-Lip to induce ICD. OPCR-Lip displayed a uniform hydrodynamic diameter of 239.0 ± 1.10 nm (Fig. S6A) and exhibited excellent in vitro stability (Fig. S6B), facilitating tumor accumulation through the enhanced permeability and retention effect. Its favorable hemocompatibility was confirmed by low hemolysis rates in rabbit whole blood at OXA concentrations ranging from 1 to 80 μg/ml (Fig. S6C). Moreover, the DPPA-1 peptide was conjugated to an MMP-2-sensitive linker (PDPPA-1), ensuring targeted and efficient release of OXA at tumor sites (Fig. S6D).
By incorporating OXA into PCR-Lip, we aimed to trigger ICD and enhance immune activation. Compared with PCR-Lip and PBS control, OPCR-Lip significantly promoted ICD in 4T1 cells (Fig. 3, A–C). Notably, OPCR-Lip synergistically combined OXA-induced ICD with PCR-Lip–mediated endoplasmic reticulum stress to elevate calreticulin (CRT) exposure (p = 0.01; Fig. 3A). This synergism is crucial, as cancer cells undergoing ICD facilitate dendritic cell (DC) maturation and antigen presentation, thereby activating CD8+ cytotoxic T lymphocytes (CTLs). Indeed, OPCR-Lip markedly promoted DC maturation in vitro upon coculture with pretreated 4T1 cells (Fig. 3, D and E) and in vivo in 4T1 tumor–bearing mice (Fig. S7). In addition, OPCR-Lip–matured DCs secreted higher levels of tumor necrosis factor-α and interleukin (IL)-12, cytokines essential for CTL activation (Fig. 3F) (21, 22).
To assess the immunomodulatory effects of OPCR-Lip in vivo, we compared its performance with PCR-Lip and PBS. OPCR-Lip significantly increased CD8+ CTL infiltration into tumors, spleens, and tumor-draining lymph nodes (TDLNs) (Fig. 3, G and H). Tumors from OPCR-Lip–treated mice also showed elevated levels of proinflammatory cytokines (IL-6, IL-12, and interferon gamma [IFN-γ]) and decreased immunosuppressive cytokines (transforming growth factor-β and IL-10), reflecting a shift toward a highly active antitumor immune microenvironment (Fig. 3I). Similar cytokine trends were observed systemically in mouse serum (Fig. 3J). Collectively, these results demonstrate that OPCR-Lip effectively induces ICD and reprograms the immune microenvironment, promoting CD8+ CTL infiltration and cytokine-mediated antitumor responses.

In vivo antitumor efficacy of OPCR-Lip
To comprehensively evaluate the therapeutic potential of OPCR-Lip, we conducted in vivo studies using 4T1 tumor–bearing BALB/c mice (Fig. 4A). One week postinoculation of 4T1 cells into the mammary fat pad, mice were intravenously administered OPCR-Lip (containing 5 mg/kg OXA) every 2 days for a total of five doses. OPCR-Lip was compared against five control groups: PBS, O-Lip (OXA-loaded liposomes), OCR-Lip (OPCR-Lip lacking PDPPA-1), PCR-Lip, and OPCD-Lip (OXA-loaded, Golgi-inactive liposomes).
Among all groups, OPCR-Lip demonstrated the most potent therapeutic effect, evidenced by robust ICD induction and minimal PD-L1 expression within the tumor microenvironment (Figs. 4B and S8). Unlike prior reports (23), no PD-L1 upregulation was observed with O-Lip, likely because of differences in dosing (Fig. S8A). Notably, OPCR-Lip significantly enhanced CRT exposure compared with OPCD-Lip (p < 0.05), supporting the synergistic role of OXA-induced ICD and Golgi-targeted PD-L1 blockade (Fig. 4B), consistent with Fig. 3A. Histological analyses revealed extensive tumor necrosis, reduced proliferation (via TUNEL assay), and elevated apoptosis (via Ki67 staining) in the OPCR-Lip group (Fig. S9A). These cellular changes translated into markedly reduced tumor growth (Fig. 4, C and D and Fig. S9B) and prolonged survival (Fig. 4E and Table S1).
Importantly, no damage was observed in vital organs during the treatment period, indicating that OPCR-Lip exhibited excellent safety throughout the short-term treatment course. Mice treated with OPCR-Lip showed no significant changes in body weight (p > 0.05) (Fig. S9C) and exhibited no histological damage in vital organs, including heart, liver, lung, kidney, and spleen (Fig. S9D), confirming favorable biosafety profiles.
To further delineate the transcriptional mechanisms underlying the potent antitumor effects of OPCR-Lip, we performed bulk RNA-Seq on tumor tissues harvested from mice treated with OPCR-Lip, OPCR (a free-drug solution of OXA and PCR), or PBS as a control. Notably, OPCR-Lip treatment resulted in the smallest tumor volumes (Fig. S10). Following this observation, tumors treated with OPCR-Lip exhibited significantly greater transcriptional divergence from PBS controls than those treated with OPCR, as evidenced by increased Euclidean distances in the two-dimensional principal component analysis and a higher number of differentially expressed genes (DEGs) (Fig. 4, F and G). These results indicate that OPCR-Lip elicits a more pronounced transcriptional response in the tumor microenvironment.
Functional enrichment analysis of OPCR-Lip–induced DEGs revealed significant upregulation of pathways related to Golgi structural homeostasis (e.g., “intermediate filament organization”) (24), immune surveillance and antigen presentation (e.g., “myeloid leukocyte-mediated immunity”) (25), and the epithelial–mesenchymal transition and metastatic progression (e.g., “epithelial cell differentiation”) (26) (Fig. 4H). These findings confirm that OPCR-Lip exerts broader and more potent effects on Golgi function, immune activation, and metastasis-associated processes compared with OPCR alone.
To validate the functional relevance of these transcriptional changes, we further examined the expression profiles of 21 signature genes associated with the proposed mechanisms of OPCR-Lip (Table S2). As anticipated, all observed gene expression changes aligned with the expected direction of OPCR-Lip action, reinforcing its capacity to enhance tumor immunogenicity and stimulate robust immune responses (Fig. 4I).
Despite the strong antitumor efficacy of OPCR-Lip, marked by stable body weight and absence of pathological damage in off-target tissues, ex vivo fluorescence imaging revealed a modest accumulation of the formulation outside the tumor (Fig. S11). Given the inherent risks of chemotherapy-induced toxicity, we investigated whether OPCR-Lip triggered off-target effects at the transcriptomic level. Bulk RNA-Seq was conducted on six representative organs, along with skeletal muscle and adipose tissue, both of which are essential for metabolic homeostasis and immune modulation during cancer treatment (27, 28). Consistent with histological assessments showing minimal tissue injury (Fig. S11), transcriptomic analyses revealed only minor, nonsignificant expression changes in nontarget tissues (Fig. 4I), highlighting the high tumor selectivity and favorable safety profile of OPCR-Lip.

OPCR-Lip suppresses tumor growth and sustains an antitumor immune response
In cancer immunotherapy, immune cells act as effectors, whereas tumor cells not only serve as immune targets but also actively modulate immune function through diverse mechanisms. To dissect the cellular heterogeneity underlying the therapeutic effects of OPCR-Lip, we performed single-nucleus RNA-Seq (snRNA-Seq) on tumor tissues following treatment. Uniform manifold approximation and projection (UMAP) analysis of 32,087 individual nuclei identified 13 transcriptionally distinct cell clusters (Fig. 5A, Fig. S12, A, C, and D, and Table S3). Among these, two clusters—Top2a+ and Padi4+ malignant cells—exhibited marked copy number variations (CNVs), consistent with their oncogenic identity, as revealed by inferCNV analysis (Fig. S12B). Notably, tumors treated with OPCR-Lip showed a significant reduction in both the abundance and CNV burden of these malignant populations compared with PBS controls (Fig. S12B), suggesting a suppression of chromosomal instability (CIN), a major driver of tumor progression (29).
Simultaneously, OPCR-Lip treatment led to a pronounced expansion of four key immune cell populations—macrophages, DCs, monocytes, and T cells—indicating enhanced recruitment and activation of immune effectors, likely facilitated by immunogenic tumor cell death induced by chemotherapy (30) (Figs. 5C and S12E). This reshaping of the tumor microenvironment supports the observed antitumor efficacy and suggests an immune-permissive milieu conducive to sustained therapeutic response.
Tumor progression is tightly linked to the differentiation and maturation state of malignant cells (31, 32). To evaluate how OPCR-Lip modulates cancer cell plasticity, we applied CytoTrace analysis (33), which revealed significantly lower developmental potential scores for Top2a+ and Padi4+ malignant cells in OPCR-Lip–treated tumors compared with PBS-treated controls (Fig. 5D). Consistent with this, pseudotime trajectory analysis demonstrated a more advanced differentiation state in OPCR-Lip–treated malignant cells (Fig. 5E). These results suggest that OPCR-Lip not only restricts malignant cell proliferation but also pushes them toward terminal differentiation, thereby limiting their adaptability under therapeutic pressure.
Given the central role of CD8+ and CD4+ T cells in orchestrating antitumor immunity (34, 35), we next characterized these immune subsets. CytoTrace analysis showed significantly increased differentiation potential in both CD8+ and CD4+ T cells from OPCR-Lip–treated tumors compared with PBS-treated controls (Fig. 5, F–H and Fig. S12F), indicative of restored immune competence. This suggests that OPCR-Lip reverses T-cell exhaustion and reprograms them into a more functional state, thereby enhancing immune surveillance and therapeutic durability (36, 37).
Together, these findings demonstrate that OPCR-Lip not only effectively suppresses tumor proliferation but also reconfigures the tumor immune landscape, enabling durable immune responses through enhanced T-cell functionality and reduced malignant cell plasticity.

Circadian rhythm–based optimization of OPCR-Lip administration
Mounting evidence suggests that aligning chemotherapy with the body's endogenous circadian rhythms can significantly enhance therapeutic efficacy (38, 39, 40). To explore the circadian characteristics of tumors in 4T1-bearing BALB/c mice and identify an optimal treatment window for OPCR-Lip, we performed a comprehensive temporal transcriptomic analysis over a 24-h cycle. Using bulk RNA-Seq, we generated 34 datasets with four to seven biological replicates collected at 4-h intervals (Table S4). MetaCycle analysis (41) identified 3674 cycling genes within the tumor tissue. These genes exhibited a bimodal phase distribution, with peak expression clustering around ZT5–ZT6 (early afternoon) and ZT16–ZT18 (late night) (Fig. 6, A and B).
Functionally, genes peaking at ZT6 were enriched in pathways related to cardiovascular development, lipid metabolism, neuronal differentiation, and tissue morphogenesis—biological processes typically active at night in healthy mice (Fig. 6C). Compared with nontumor tissues from healthy mice (42), tumors exhibited significant phase shifts in the expression of core circadian regulators, particularly Cry1 and Cry2. These disruptions were especially prominent in metabolically active organs, such as the liver, skeletal muscle, and adipose tissue, as well as in neural tissues, including the hypothalamus, suprachiasmatic nucleus, and dentate gyrus (Fig. 6D). These findings highlight circadian desynchrony in tumor tissues, likely linked to behavioral and physiological dysregulation during tumor progression (43, 44).
Building on this circadian landscape, we designed a mechanism-driven chronochemotherapy strategy for OPCR-Lip administration. DNA damage repair—a key determinant of platinum-based chemotherapy efficacy—is tightly regulated by circadian genes (45, 46, 47). Notably, Per2 enhances both the cytotoxicity and sensitivity to OXA (48, 49), and its rhythmic peak has been shown to optimize therapeutic response in human oral squamous cell carcinoma (50). In our study, Per2 expression peaked around ZT16 in 4T1 tumors (Fig. 6E). Furthermore, OPCR-Lip–targeted pathways—including glycosylation regulation, Golgi-mediated protein transport, and the DNA damage response—also peaked between ZT14 and ZT20 (Fig. 6F). This temporal convergence strongly supports night-time (ZT16) administration of OPCR-Lip to align with peak therapeutic target activity and Per2-mediated sensitivity.
To empirically determine the optimal administration window, we delivered OPCR-Lip to 4T1-bearing mice every 4 h over a 24-h period (Fig. 6G). Consistent with previous chronochemotherapy studies (51), OPCR-Lip exhibited time-dependent efficacy. The ZT16 treatment group demonstrated the most pronounced therapeutic benefit, including significant tumor weight reduction, stable body weight maintenance (Fig. 6, H and I, and Fig. S13), and increased tumor cell necrosis and apoptosis (Fig. 6J). These effects were underpinned by robust immune activation, as evidenced by transcriptomic shifts in the ZT16-treated tumors.
Markers of DNA damage (Chd5, Gadd45b, and Hspb1), ICD (Nfkb1, Dnajc12, and Map3k1), and immune effector activation (Cpm, Ccr8, and Cd40lg) were significantly upregulated following ZT16 administration (Fig. 6K), highlighting the synergistic enhancement of OXA and PD-L1 blockade during this circadian phase. While transcriptomic data elucidate the pharmacodynamic mechanisms driving time-dependent efficacy, circadian fluctuations in pharmacokinetics and drug metabolism may also contribute to the observed outcomes (52).
Collectively, these findings establish ZT16 (night-time) as the optimal administration window for OPCR-Lip, providing a mechanistic and empirical basis for circadian-timed immunochemotherapy aimed at maximizing therapeutic efficacy.

Evaluation of the antitumor efficacy of OPCR-Lip in skin and lung cancer models
To evaluate the broad-spectrum antitumor efficacy of OPCR-Lip against malignancies with variable PD-L1 expression (53), we investigated its effects in melanoma (B16F10) and lung carcinoma (Lewis lung carcinoma [LLC]) models. OPCR-Lip treatment markedly increased tumor necrosis, suppressed cellular proliferation, and enhanced apoptosis in both tumor types, collectively resulting in the most pronounced inhibition of tumor growth among all groups. Importantly, OPCR-Lip exhibited superior antitumor activity compared with OXA alone and OXA-Lip, indicating that its efficacy arises not only from OXA but also from the synergistic effects of other components. (Fig. 7, A–C).
Transcriptomic profiling of tumor tissues treated with OPCR-Lip versus PBS revealed distinct gene expression changes in both B16F10 and LLC models. DEGs were predominantly enriched in pathways related to immune cell proliferation (Fig. S14A). Furthermore, the gene expression signatures induced by OPCR-Lip were highly consistent with remodeling of the tumor immune microenvironment (Fig. S14B), indicating that OPCR-Lip can potentially reprogram immune cell function and restore antitumor immunity.
In addition to its immunomodulatory effects, OPCR-Lip also altered metastasis-related gene expression profiles in these highly metastatic tumors. Specifically, in B16F10 tumors, five genes were significantly modulated, including those linked to primary tumor progression (Cdkn1a (54), Epha3 (55), and Acsl3 (56)) and invasion (Cdh1 (57)) (Fig. S14B). In LLC tumors, three genes associated with epithelial–mesenchymal transition—a key driver of tumor cell invasiveness—were identified: Tgfbi (58), Mmp9 (59), and Mmp10 (60) (Fig. S14B). These findings indicate that OPCR-Lip suppresses metastasis by targeting multiple steps of the metastatic cascade in both tumor models.
Together, these results demonstrate that OPCR-Lip exerts potent antitumor activity in B16F10 and LLC tumors through a dual mechanism: activating immune responses and inhibiting metastatic progression.

OPCR-Lip demonstrates efficacy in both canine and human breast cancer cells
While OPCR-Lip has shown therapeutic efficacy in murine models, a considerable translational gap remains before it can be advanced to clinical application. Bridging this gap requires rigorous preclinical validation in larger animal models, such as canines, followed by well-controlled human clinical trials. To this end, we investigated the ability of OPCR-Lip to disrupt the Golgi apparatus and induce ICD in breast cancer cell lines derived from dogs (CMT-7364) and humans (MCF-7).
Confocal microscopy revealed that OPCR-Lip accumulated within the Golgi apparatus in both CMT-7364 and MCF-7 cells (Fig. 8A), and TEM confirmed that it disrupted both cis- and trans-Golgi compartments (Fig. 8, B–D). Correspondingly, OPCR-Lip treatment significantly suppressed the activity of Man II in CMT-7364 cells and GalT in MCF-7 cells (Fig. 8E), indicating conserved Golgi-targeting specificity across species.
In addition to Golgi disruption, OPCR-Lip triggered the release of key damage-associated molecular patterns, including CRT from both CMT-7364 and MCF-7 cells, and high mobility group box 1 (HMGB1) from CMT-7364 cells. However, no significant ATP release was observed in either cell line, which may be attributed to dose-dependent effects of the ICD inducer (Fig. 8, F–H).
Collectively, these findings demonstrate that OPCR-Lip retains cross-species efficacy in targeting the Golgi apparatus and inducing hallmark ICD events in both canine and human breast cancer cells, supporting its potential as a translationally viable therapeutic candidate for future clinical evaluation.

Discussion
Immune checkpoint inhibitors targeting PD-L1 have revolutionized cancer therapy, demonstrating potent and durable antitumor responses across a wide range of malignancies. Several monoclonal antibodies, such as atezolizumab, durvalumab, and avelumab, have received approval from the US Food and Drug Administration for treating various cancers, with additional indications likely to follow soon (3). Despite these advances, therapeutic resistance remains a significant obstacle in certain patient populations and tumor types, often driven by systemic immunotoxicity, incomplete PD-L1 blockade, or insufficient T-cell responses because of poor tumor immunogenicity (61). When interpreting these promising results, it is important to note that this work did not directly compare the nanoparticle platform with free OXA, a standalone PD-L1 inhibitor, or their unencapsulated combination—an aspect that warrants future investigation to clarify its therapeutic profile. Moreover, the PD-L1 inhibitor used here is a peptide, distinct from clinically approved monoclonal antibodies; thus, a comprehensive efficacy comparison would require parallel assessment of key parameters such as binding affinity and blockade efficiency.
In this study, we developed a liposomal platform (PCR-Lip) codelivering the Golgi-disrupting agent CR and the anti-PD-L1 peptide DPPA, designed to achieve comprehensive and tumor-specific PD-L1 blockade. Given that the synthesis achieved coupling efficiencies of 1.36% and 2.14% (attributable to steric hindrance and the effects of the reaction medium), optimizing these yields will be a critical consideration for scalable production. Although the current yields did not compromise batch consistency, strategies such as phase-transfer catalysis (62) or enzymatic pretreatment (63) offer promising routes for improvement. To enhance tumor targeting, we functionalized the liposomes with CS in an MMP-2-sensitive manner (64), enabling selective accumulation in CD44+ tumor cells. This PD-L1-independent targeting strategy ensures sustained delivery efficacy upon repeated administration. Importantly, the dual-phase release mechanism of the liposomes enables sequential drug delivery in both extracellular and intracellular compartments of the tumor microenvironment, allowing for the inhibition of membrane-bound PD-L1 on the cell surface and intracellular PD-L1 glycosylation and exosomal secretion via Golgi apparatus disruption. To further potentiate antitumor immunity, we integrated OXA, a known inducer of ICD, into the formulation to generate OPCR-Lip. This enhanced construct promoted the maturation of DCs both in vitro and in vivo. In the 4T1 murine tumor model, OPCR-Lip not only inhibited tumor progression but also reprogrammed T cells to sustain robust and durable immune responses. As previously reported, CIN promotes tumor progression through CNVs and genomic instability (29). The ability of OPCR-Lip to reduce CNVs and stabilize the genome may directly counteract CIN-driven tumor evolution. We also conducted preliminary investigations into circadian rhythm–based administration schedules in the 4T1 model to optimize therapeutic outcomes. Moreover, OPCR-Lip demonstrated potent antitumor efficacy in additional models, including B16F10 and LLC tumor–bearing mice. It also exhibited Golgi-targeting capacity and ICD-inducing effects in both human and canine breast tumor cells, highlighting its translational potential across diverse cancer types and species. In summary, our work presents OPCR-Lip as a multifaceted and improved immune checkpoint blockade strategy. It achieves dual-phase PD-L1 inhibition, targeting both extracellular and intracellular PD-L1, and simultaneously enhances tumor immunogenicity, offering broad therapeutic promise for cancer immunotherapy.

Experimental procedures

Mice
Female BALB/c (6 weeks old) and male C57BL/6J (6 weeks old) mice were purchased from Dashuo Biotechnology. All mice were housed under standard laboratory conditions, including a 12-h light–dark cycle, an ambient temperature of 25 ± 2 °C, and a humidity level of 60 ± 10%. Mice were euthanized when tumor volumes reached 2000 mm3. All animal procedures complied with the Regulations for the Administration of Affairs Concerning Experimental Animals and were approved by the Animal Ethics and Welfare Committee of Sichuan Agricultural University (permit no.: 20250678).

Cell lines and cell culture
The 4T1 (murine breast cancer), B16F10 (murine melanoma), LLC (murine lung carcinoma), and MCF-7 (human breast cancer) cell lines were obtained from the Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. Canine mammary carcinoma cells (CMT-7364) were provided by the College of Veterinary Medicine, China Agricultural University. All cells were cultured at 37 °C in a humidified atmosphere with 5% CO2. 4T1 and B16F10 cells were maintained in RPMI1640 medium; LLC and CMT-7364 in Dulbecco's modified Eagle's medium; both media were supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. DCs were isolated from BALB/c mice and cultured in RPMI1640 with 10% fetal bovine serum and antibiotics.

Antibodies and reagents
For flow cytometry assays, CD11c-PE and CD86-FITC antibodies were procured from eBioscience, whereas the anti-CRT antibody (CRT Rabbit Monoclonal) was obtained from StressMarq. Antibodies for CD80-APC, CD3e-APC, CD4-PE, CD8a-BB515, GM130, and Syntaxin-6 were purchased from BD Pharmingen. Additional antibodies, including vascular endothelial growth factor, transforming growth factor-β, PD-L1, IL-10, IFN-γ, Alexa Fluor 594-conjugated, and Alexa Fluor 488-conjugated secondary antibodies, were obtained from Abcam. ATP assay kits were sourced from Beyotime, and ELISA kits were sourced from Elabscience. CS (molecular weight = 6000) and RA were obtained from Sigma–Aldrich. OXA was purchased from Meilune Biological, and the MMP-2-sensitive peptide PDPPA-1 (sequence: CPLGVRG-NYSKPTDRQYHF) was custom-synthesized by Guotai Biotechnology. A Golgi extraction kit was sourced from BestBio. All other chemicals were of analytical reagent grade.

Synthesis and characterization of OPCR-Lip
CR was synthesized according to a previously reported protocol (15). Briefly, CR (0.02 mmol) was dissolved in PBS, followed by the dropwise addition of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide and sulfo-N-hydroxysulfosuccinimide, and stirred at room temperature for 40 min. A PBS solution of N-(2-aminoethyl) maleimide hydrochloride (0.04 mmol) was then added, and the mixture was stirred for 24 h. The reaction product was dialyzed and lyophilized to obtain carboxyl-reactive maleimide (CR-MAL).
To prepare CD-MAL, CS (0.2 mmol) was dissolved in formamide and slowly added to a solution of deoxycorticosterone acetate (481 mg, 1.6 mmol) in formamide. After stirring at room temperature for 48 h, the CD conjugate was isolated and further reacted with MAL to yield CD-MAL using the same procedure as for CR-MAL. The structures of CR-MAL and CD-MAL were confirmed via
1H NMR spectroscopy.
OCR-Lip nanoparticles were fabricated via the film dispersion/postinsertion method. Briefly, Lipoid S100 and cholesterol were dissolved in chloroform, and the organic solvent was evaporated at 55 °C to form a lipid film. The film was hydrated with a 5% glucose solution containing OXA and CR-MAL, followed by sonication for 10 min to obtain OCR-Lip. The liposome solution was then ultrafiltered to remove free OXA. PDPPA-1 was dissolved in 0.01 M phosphate buffer (0.2 ml) and added dropwise to the OCR-Lip dispersion at a molar ratio of peptide to maleimide groups of 18:10. The mixture was stirred at 4 °C for 12 h. The reaction product was dialyzed to remove unreacted PDPPA-1 for 12 h to obtain OPCR-Lip. The coupling efficiencies of PDPPA-1 were obtained by HPLC (14). OPCD-Lip was prepared using a similar procedure.
Nanoparticle size, polydispersity index, and zeta potential were characterized by dynamic light scattering (Zetasizer Nano ZS90) and TEM (JEM-2100Plus). Encapsulation efficiency (%), drug loading (%) of OXA, and DPPA-1 conjugation efficiency were quantified by HPLC (Agilent).

Golgi colocalization and disruption assay
Colocalization of liposomes with the Golgi apparatus was assessed by confocal laser scanning microscopy (CLSM). Cells were seeded in confocal dishes and treated with formulations for 2 h. The Golgi apparatus was stained using BODIPY TR (10 mM Hepes), and nuclei were stained with 4′,6-diamidino-2-phenylindole. Images were captured using CLSM (Nikon A1-90i) and analyzed with ImageJ (National Institutes of Health, Bethesda, MD) using the Colocalization Threshold plugin.
To evaluate Golgi disruption, cells were treated with various formulations for 24 h and immunostained with GM130 and Syntaxin-6 primary antibodies, followed by fluorescent secondary antibodies. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole, and images were obtained using CLSM. Structural changes to the Golgi were also assessed by TEM of ultrathin sections. Enzymatic activities of Golgi-resident enzymes, Man II and GalT, were quantified using enzyme activity assay kits.

PD-L1 glycosylation, surface expression, and exosome secretion
PD-L1 glycosylation in 4T1 cells was analyzed by Western blot, and N-glycopeptides were identified using nano LC–MS/MS. Membrane PD-L1 expression was quantified by flow cytometry after staining with PE-conjugated PD-L1 antibody. To analyze exosomal PD-L1, extracellular vesicles were isolated from treated cell supernatants using a standard differential centrifugation protocol. Exosome size was determined using a Flow NanoAnalyzer N30 E (NanoFCM), and exosomal protein markers were characterized by flow cytometry and TEM.

In vitro ICD induction and bone marrow–derived dendritic cell maturation assay
To evaluate ICD, the release of damage-associated molecular patterns, including CRT, HMGB1, and ATP, was measured following treatment with various formulations containing 20 μg/ml OXA. CRT exposure was detected by immunofluorescence microscopy. HMGB1 localization was similarly assessed. ATP release was quantified using an ATP assay kit, whereas HMGB1 content in the supernatant was measured by ELISA.
For DC maturation assays, immature DCs were isolated from BALB/c mice and cocultured with pretreated 4T1 cells at a 1:1 ratio in 6-well plates. Maturation markers were analyzed by flow cytometry, and cytokine levels (IL-6, IFN-γ, and tumor necrosis factor-α) in the culture medium were determined using ELISA.

LC–MS/MS analysis

LC–MS/MS analysis of N-glycopeptides
Proteins were extracted using radioimmunoprecipitation assay lysis buffer, and the supernatants were collected following centrifugation. For immunoprecipitation, beads were conjugated with the antibody for 2 h at room temperature and washed. Samples were then added to the antibody-conjugated beads, incubated for 2 h, and washed again before bead resuspension. Proteins were reduced with DTT (56 °C, 1 h), alkylated with iodoacetamide (room temperature, dark, 1 h), and digested overnight with trypsin at 37 °C. Peptides were desalted using a C18 tip and dried. Peptides were separated on a Vanquish Nano LC system equipped with a C18 column. A linear gradient from 4% to 95% mobile phase B was run over 131 min at 600 nl/min. Mass spectrometry (MS) analysis was performed on an Orbitrap Fusion Lumos Tribrid mass spectrometer. Full MS scans were acquired, followed by higher-energy collisional dissociation fragmentation of the top 20 precursors. Data were processed using MaxQuant (version 1.6.2.10, Max Planck Institute of Biochemistry, Martinsried, Germany) against a species-specific protein database.

LC–MS/MS-based proteomics

Sample preparation and protein extraction
4T1 cells were lysed in 8 M urea buffer with 1 mM PMSF and 2 mM EDTA, sonicated, and centrifuged at 15,000g for 10 min at 4 °C. The protein concentration of supernatants was measured using a bicinchoninic acid assay.

Tryptic digestion and peptide desalting
Proteins were reduced with 10 mM DTT, alkylated with 50 mM iodoacetamide, and precipitated with cold acetone. The pellet was digested overnight with trypsin (Promega) in 25 mM ammonium bicarbonate at 37 °C. Peptides were desalted using C18 StageTips.

4D-data-independent acquisition MS analysis
Peptides were separated on a nanoElute UHPLC (Bruker) with a 40-min acetonitrile gradient. Analysis was performed on a timsTOF Pro2 (Bruker) in data-independent acquisition–parallel accumulation serial fragmentation mode, covering ∗m/z∗ 400 to 1200 and ion mobility 0.85 to 1.3 Vs/cm2.

Data processing and protein identification
Data were analyzed using DIA-NN (version 1.8.1) against the UniProt Mus musculus database (UP000000589). The search included fixed carbamidomethylation and variable oxidation/N-terminal acetylation. The false discovery rate was set to 1% at both the peptide and protein levels. Protein quantification was performed using MaxLFQ.

Statistical and bioinformatics analysis
Differential protein abundance was assessed by a two-tailed t test (fold change [FC] ≥1.5 or ≤0.667, p < 0.05). Functional enrichment analysis was performed with clusterProfiler (v4.4.4). Protein–protein interaction networks were constructed using STRING.

Biodistribution of OPCR-Lip
A 4T1 tumor model was established via subcutaneous injection of 5 × 106 4T1 cells suspended in PBS into the third mammary fat pad of BALB/c mice. Treatments commenced when tumors reached ∼100 mm3. To assess in vivo biodistribution, formulations labeled with DiD (5 μg/mouse) were administered via tail vein injection. Fluorescent imaging was performed at multiple time points. After the final imaging session, mice were sacrificed, and tumors and major organs were harvested for ex vivo fluorescence imaging.

In vivo antitumor study
Female BALB/c mice were inoculated subcutaneously with 1 × 106 4T1 cells, and male C57BL/6J mice received the same number of B16F10 or LLC cells. Once tumors reached ∼100 mm3, mice were randomized into groups and treated with various formulations (OXA dose: 5 mg/kg), administered every other day for five doses. To assess immune activation, TDLNs, spleens, and tumors were harvested post-treatment. Tumors were dissociated into single-cell suspensions and stained with anti-CD11c, CD80, and CD86 antibodies for flow cytometry. TDLNs, spleens, and tumors were also stained with anti-CD3-EV450, CD8-ER780, and CD4-APC antibodies at 4 °C for 30 min to evaluate T-cell infiltration. For histological analysis, tumors and major organs (heart, liver, spleen, lung, and kidney) were fixed in 4% formaldehyde and subjected to H&E, Ki67, and TUNEL staining.

Chronotherapy with OPCR-Lip
BALB/c mice were housed under a 12-h light/12-h dark cycle (lights on from 8:00 AM to 8:00 PM, corresponding to Zeitgeber time 0–12) with ad libitum access to food and water. Mice were subcutaneously inoculated with 1 × 106 4T1 cells. Once tumors reached ∼100 mm3, OPCR-Lip or PBS was administered at designated circadian time points for 2 weeks. Tumor weights were measured after euthanasia to evaluate time-dependent therapeutic efficacy.

Bulk RNA-Seq and data processing
Breast tumors were isolated from female BALB/c mice; lung and skin tissues were collected from male C57BL/6J mice. Each tissue type had ≥2 biological replicates. Total RNA was extracted using TRIzol reagent (Invitrogen) per the manufacturer's instructions. Libraries were prepared and sequenced on an Illumina NovaSeq 6000 platform using 2 × 150 bp paired-end reads (PE150) by LC-Bio Technology. Transcript quantification was performed using Kallisto against the Ensembl v110 reference genome. Genes with transcripts per million >1 in any sample were considered expressed. Differential expression analysis was conducted using DESeq2, with genes showing |log2FC| >1 and a false discovery rate <0.05 considered significant. Pathway enrichment was performed using Metascape. Circadian gene analysis was conducted using MetaCycle, and enrichment analysis was performed using parametric gene set enrichment analysis.

Single-nucleus RNA-Seq

Library preparation
Tissue samples were minced (∼1 mm fragments) and homogenized in 2 ml of ice-cold Nuclei EZ Lysis Buffer (NUC-101; Sigma–Aldrich) supplemented with protease (Roche) and RNase inhibitors (Promega, Life Technologies) using a Dounce homogenizer. After a 5-min incubation on ice, an additional 2 ml of lysis buffer was added. The sample was further homogenized and gently resuspended. After an additional 6 min on ice, 2 ml of 4% bovine serum albumin (BSA) was added to terminate lysis. The suspension was mixed, centrifuged (300g, 5 min, 4 °C), and resuspended in lysis buffer containing 4% BSA for 3 min. Debris was removed using Miltenyi Debris Removal Solution. The sample was washed with 4 ml buffer, incubated on ice (5 min), centrifuged again, and resuspended in Nuclei Suspension Buffer (1x PBS, 0.07% BSA, and 0.1% RNase inhibitor). The final nuclei suspension was filtered through a 20 μm cell strainer and quantified using a hemocytometer or Countess II Automated Cell Counter.
Single-nucleus suspensions (∼10,000 nuclei) were loaded onto the MobiNova-100 Single-Cell System using a MobiCube Single-Cell 3′ RNA-Seq Kit according to the manufacturer’s protocol. Complementary DNA amplification and library construction followed standard procedures. Sequencing was performed on an Illumina NovaSeq 6000 system (paired-end mode), targeting a minimum of 20,000 reads per nucleus, and conducted by LC-Bio Technology.

Preprocessing, clustering, and annotation
Raw sequencing reads from snRNA-Seq were aligned to the GRCm39 reference genome using MobiVision software (v3.0), generating gene count matrices. A total of 35,790 nuclei were profiled across four libraries derived from four samples. Doublets were identified and removed using the scDblFinder R package (v1.16.0) (65). Nuclei were filtered based on three quality metrics: (i) expression of >300 genes with nonzero counts, (ii) >800 unique molecular identifiers, and (iii) mitochondrial gene content <5% of total reads. Genes expressed in fewer than 20 nuclei were excluded. After quality control, 19,744 genes across 32,206 nuclei remained. Adipocytes were further excluded, resulting in 32,087 high-quality nuclei for downstream analysis.
To integrate datasets across conditions, we used Seurat (v5.1.0) (66), normalizing and scaling each nucleus with the "SCTransform" function after regression on mitochondrial gene ratio, ribosomal gene ratio, and cell cycle scores. Batch effects were corrected using canonical correlation analysis on the top 2000 highly variable genes via the “IntegrateLayers” function. Principal component analysis was then applied, and the first 30 components were used to construct a shared nearest-neighbor graph via the “FindNeighbors” function. Clustering was performed using “FindClusters” at a resolution of 0.5. Dimensionality reduction was achieved with “RunUMAP” and visualized using UMAP.
Cell type annotation was based on canonical markers from previous tissue-specific studies. Marker genes for each cluster were identified using the Wilcoxon's rank-sum test via the "FindAllMarkers" function, selecting markers with avg_logFC >0.25, adjusted p < 0.05, and min.pct >0.20. T-cell subtypes were further resolved using established integration strategies.

CNV analysis
To identify malignant cells, we used the inferCNV R package (v1.18.1). Raw, unintegrated count matrices were input, and only autosomal chromosomes were retained. Endothelial and mesenchymal stromal cells were used as reference cell types. Cells exhibiting substantial chromosomal copy number deviations were defined as malignant, displaying characteristic red-to-blue gradients on CNV heatmaps, representing gain or loss of genomic regions.

Cell type abundance variation analysis
Cell type composition changes were assessed using the local true sign rate, which estimates the probability that the direction of the log-transformed FC is correct. A local true sign rate >0.9 was considered indicative of a statistically significant change in cell abundance between conditions.

Pseudotime and potency score analysis
The differentiation trajectories of malignant cells before and after treatment were inferred using Monocle3 (v1.3.7) (67). Dimensionality reduction was performed for each malignant cell type, and trajectories were constructed in a semisupervised manner. Pseudotime values were calculated using the “pseudotime” function and discretized into integer bins. The proportion of treatment-group malignant cells within each bin was plotted against pseudotime, and a linear regression model was applied to compute R2 and p values.
Cellular potency scores were calculated using the “cytotrace2” function in CytoTRACE2, a deep learning framework that quantifies differentiation states based on single-cell transcriptomic profiles (33).

Gene ontology analysis
Gene Ontology enrichment of DEGs was performed using Metascape (https://metascape.org/) (68).

Statistical analysis
All statistical analyses were performed using SPSS (IBM SPSS Statistics 26). Data are reported as mean ± SD. For comparisons between two groups, unpaired Student’s t tests were used. For multigroup comparisons, one-way ANOVA followed by least significant difference post hoc tests (for normally distributed data) or Kruskal–Wallis tests with Dunn’s post hoc correction (for nonparametric data) were applied. Survival curves were analyzed using the log-rank (Mantel–Cox) test. A p value <0.05 was considered statistically significant.

Data availability
All data are available in the main text or the supporting information. The MS data are available via ProteomeXchange (identifier: PXD073979). The reference genome and gene annotation file (GRCm39, Ensemble release 111) were downloaded from Ensembl (https://ftp.ensembl.org/pub/release-111/). All sequencing data are publicly available: bulk RNA-Seq of multiple tumor/normal tissues (PRJNA1304182), time-series bulk RNA-Seq of breast tumors (PRJNA1304155), and breast tumor snRNA-Seq (PRJNA1304156) via the National Center for Biotechnology Information Sequence Read Archive. The codes used in this study are available at https://github.com/RMIXXS/Mouse-breast-tumor-code.

Supporting information
This article contains supporting information (69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98).

Conflict of interest
The authors declare that they have no conflicts of interest with the contents of this article.