Neutrophil extracellular traps promote liver metastasis by impairing NK cell-dependent tumor surveillance via CCDC25

Abstract

Tumor metastasis requires constant crosstalk between primary tumor cells and immune cells in distant organs, yet how this interaction shapes the evolving immune landscape of metastatic sites remains unclear. Here, using longitudinal single-cell RNA sequencing in a 4T1 orthotopic breast cancer model, we observe a progressive infiltration of neutrophils, robust NETosis and an immunosuppressive niche during metastatic progression in the liver. Through multiple in vivo models, we determine that macrophage-dependent neutrophil extracellular trap (NET) formation serves as a potent driver of liver metastasis. Mechanistically, extracellular NET-derived DNA is sensed by the transmembrane protein coiled-coil domain containing 25 (CCDC25) on natural killer (NK) cells, and the NET-CCDC25 interaction impairs NK cell surveillance by downregulating multiple activating receptors, including NKG2D, NKp46, NKp44, via the CCDC25-ILK-STAT3 axis. In clinical metastatic liver samples, a higher level of NET expression is associated with NK cell dysfunction. Overall, our findings identify CCDC25 as a potential “innate immune checkpoint” on NK cells, revealing a promising therapeutic strategy to reactivate NK cell function to inhibit liver metastasis.

The formation of neutrophil extracellular traps (NETs) contributes to liver metastasis and immune suppression. Here the authors show that NET-derived DNA can be sensed by CCDC25 on natural killer (NK) cells impairing NK cell-dependent tumor surveillance and promoting liver metastasis.

Introduction

The liver is a common site for cancer metastasis, which confers poor therapy responses and an unfavorable prognosis^1,2. Although organ-specific immunity eliminates most disseminated tumor cells, a subset survives through phenotypic adaptations, such as epithelial-mesenchymal transition (EMT) or dormancy^3,4. Moreover, primary tumors systemically prime the liver via soluble factors or exosomes, creating an immunosuppressive niche permissive for colonization^5,6. Nevertheless, current understanding fails to capture the temporal dynamics of immune remodeling in the metastatic liver, which may underlie the limited efficacy of existing therapies.

Neutrophils are highly plastic cell types⁷, whose phenotypes are shaped by tumor-derived factors such as tumor debris and cytokines. Through NETosis, neutrophils release neutrophil extracellular traps (NETs), which are extracellular DNA webs decorated with several granule proteins, including myeloperoxidase (MPO), neutrophil elastase (NE), and cathepsin. NETs facilitate cancer metastasis through multiple mechanisms: disrupting vascular integrity, chemotactically attracting or directly sequestering circulating tumor cells, awakening dormant tumor cells, and inducing tumor cell EMT^8–12. Moreover, NETs could suppress anti-tumor immunity by shielding cancer cells from attack and expanding regulatory T cells through recruiting innate-like B cells^8,13,14. In line with this, our work and that of others have documented significant neutrophil and NET accumulation in the pre-metastatic liver, correlating with metastatic burden and poor patient survival^9,15,16. Nevertheless, the mechanism of liver NET formation and the crosstalk between NETs and other immune cells over time remain elusive.

The pro-tumorigenic functions of NETs have prompted interest in targeting them therapeutically. While NET depletion may suppress cancer metastasis, it may also interfere with anti-tumor immunity, as NET-DNA can be internalized and sensed by intracellular nucleic acid receptors such as the cGAS-STING pathway, thereby promoting dendritic cell (DC) maturation and anti-tumor response^17,18. We previously identified CCDC25 as an extracellular DNA sensor on cancer cells that binds NET-DNA and facilitates liver metastasis⁹. Importantly, inhibiting NET-CCDC25 interaction reduces metastasis without compromising STING-dependent DC activation¹⁹, suggesting that this axis may represent a more selective strategy for targeting the pro-metastatic effects of NETs while preserving NET-associated immune activation. However, although the function of CCDC25 in cancer cells has been characterized, its expression and role within the immune compartment of the metastatic niche remain unclear.

In this work, we perform longitudinal single-cell RNA sequencing (scRNA-seq) in a murine orthotopic breast cancer model to delineate the dynamic immune remodeling of the liver during metastasis, uncovering a robust accumulation of neutrophils and NETs, followed by progressive immunosuppression. We further reveal that tumor-derived debris reprograms macrophages to foster a pro-metastatic environment. More importantly, we identify CCDC25 as an innate immune checkpoint on NK cells. By elucidating this mechanism, our work provides a therapeutic rationale for targeting CCDC25 to restore anti-tumor immunity and inhibit hepatic metastasis.

Results

Dynamic scRNA-seq analysis identifies excessive neutrophil accumulation, NETosis and immunosuppression in the metastatic livers

To longitudinally characterize compositional and transcriptional dynamics of immune cells within the liver metastatic niche, we employed an orthotopic 4T1-luc breast cancer model and performed dynamic scRNA-seq of liver immune cells during metastatic progression. Initially, we estimated the temporal course of the liver metastasis through serial bioluminescent imaging (BLI), histological analyses and immunofluorescent analyses at various timepoints (week 0, 1, 3 and 5), confirming the emergence of small clusters of tumor cells at week 3 and overt metastatic outgrowth at week 5 (Supplementary Fig. 1a–d). Based on this, we classified these timepoints into four distinct stages: normal liver stage (week 0), pre-metastatic niche stage (week 1), early-seeding stage (week 3) and overt metastatic stage (week 5). Mice were sacrificed at the indicated timepoints, and the livers were harvested for scRNA-seq analysis (Fig. 1a). Following droplet encapsulation, sequencing and quality-control filtering, a total of 32,620 high-quality cells from the four groups were acquired and analyzed. Unsupervised clustering and differential gene expression analysis identified diverse immune cell types including neutrophil, T cell, B cell, natural killer (NK) cell, macrophage, common myeloid progenitor cell (CMP), dendritic cell (DC), basophil as well as plasma cell (Fig. 1b and Supplementary Fig. 1e). Comparative analysis of cell composition revealed a progressive decline in T cells, NK cells, and DCs, accompanied by a marked increase in neutrophils from week 0 to week 5 (Fig. 1c). Transcriptional profiling of neutrophils uncovered upregulation of pathways associated with migration and NET abundance (Fig. 1d), with increased representation of neutrophil subsets exhibiting enhanced migratory and NETotic capacities over time (Supplementary Fig. 1f, g). To validate these findings, we performed immunofluorescence (IF) staining for myeloperoxidase (MPO) and citrullinated histone H3 (H3cit), which are commonly used markers for neutrophils and NETs, respectively. In agreement with our scRNA-seq results, both neutrophil infiltration and NET formation were detectable as early as week 1 and continued to intensify afterward (Fig. 1e–g). By contrast, NK cell- and T cell-activation were initially elevated from week 0 to week 1, but were progressively impaired from week 1 to week 5 (Fig. 1h, i). Together, these data delineate the temporal evolution of the immune landscape in the metastatic liver, characterized by early neutrophil infiltration and NET generation followed by a decline in cytotoxic lymphocyte activity at later stages, consistent with prior reports of myeloid-driven immunosuppression in hepatic metastasis²⁰.

Fig. 1. Dynamic scRNA-seq analysis of immune cells from the normal and metastatic livers reveals neutrophil accumulation, NET generation and immunosuppression.

a Schematic of BALB/c mice orthotopically inoculated with 4T1-luc breast cancer cells in the fourth mammary fat pad; livers were collected at weeks 0, 1, 3 and 5 for scRNA-seq (n = 3 mice per group). b t-distributed Stochastic Neighbor Embedding (t-SNE) plot (Left), visualization of liver immune cells and dot plot (Right) of representative DEGs for each cell type. CMP, common myeloid progenitor cell; DC, dendritic cell. c Cell-cluster proportions across time points; Neu %, neutrophil proportion. d Neutrophil migration and NET-abundance signature expression in neutrophils. e–g Representative immunofluorescence (IF) images staining for MPO and H3cit (e) and quantification of neutrophil infiltration (% MPO⁺ cells per FOV) (f) and the MPO-H3cit colocalized area (g) in murine livers (n = 5 mice per group). Scale bar, 10 μm; FOV, field of view. h, i GSVA Z-scores of NK-cell (h, n = 2433 cells in total) and T-cell (i, n = 4937 cells in total) activation pathways. j-k, Cell proportions (j), and neutrophil signature expression (k) in human metastatic liver (ML) and non-metastatic liver (NML) scRNA-seq data (GSE246662, n = 3 patients per group). l–n Representative IF images (l) and quantification of neutrophil infiltration (m) and MPO-H3cit colocalized area (n) in human liver sections (ML, n = 20 patients; NML, n = 6 patients). Scale bar, 20 μm. o, p NK-cell (o, n = 7038 cells in total) and T-cell (p, n = 23802 cells in total) activation scores in human ML and NML scRNA-seq data (GSE246662, n = 3 patients per group). For the dot plot, dot size represents the proportion of cells expressing each gene and dot color indicates the average expression level. Schematic (a) is Created in BioRender. Z, S. (2026) https://BioRender.com/rp8er09. For violin plots in (h, i, o, p), boxes indicate the IQR, center lines indicate medians and whiskers indicate minima and maxima. Data are presented as mean ± SD (f, g) or mean ± SEM (m, n), and statistical analysis was performed using Welch analysis of variance (ANOVA) with Dunnett’s T3 test (f, g), Kruskal-Wallis test with Dunn’s test (h, i), two-sided Chi-square test (j) or two-tailed Mann-Whitney U test (m–p). Source data are provided as a Source Data file.

We next sought to characterize the hepatic immune microenvironment in patients with liver metastases. Analysis of a publicly available human scRNA-seq dataset revealed increased neutrophil infiltration and enhanced NET formation in the metastatic livers (ML) compared to non-metastatic livers (NML) (Supplementary Fig. 1h and Fig. 1j, k). IF staining of liver sections from breast cancer patients with liver metastases further confirmed excessive accumulation of neutrophils and NETs (Fig. 1l–n). In accordance with murine scRNA-seq data, we observed a marked reduction in NK and T cell–mediated anti-tumor immune responses within the metastatic livers of patients (Fig. 1o, p). To further examine the temporal relationship between liver metastasis and NETosis, we established intrasplenic injection liver metastasis models with MC38-luc and EO771-luc tumor cells. NET formation in the liver was evaluated both before and after the emergence of detectable metastases on day 20 post-injection (Supplementary Fig. 1i, j). Consistent with the findings in orthotopic models, NET accumulation was evident in the liver prior to the appearance of overt metastases and significantly increased after metastatic lesions emerged (Supplementary Fig. 1k–n).

Taken together, these findings uncovered abundant neutrophil infiltration and NET formation in the metastatic liver, consistent with our previous report as well as other reports^9,15,16. Moreover, we demonstrated a progressive decline in anti-tumor immunity during metastatic progression, delineating the dynamic immune remodeling of the liver throughout hepatic metastasis.

Neutrophil recruitment during liver metastasis depends on the CXCL1/2-CXCR2 axis

To investigate the mechanism underlying neutrophil recruitment during hepatic metastasis, we primarily focused on chemokine-mediated pathways, which are well recognized as key regulators of neutrophil migration in both physiological and pathological settings²¹. We first conducted a targeted quantitative reverse transcription polymerase chain reaction (RT-qPCR) screening assay on liver tissues from mice bearing orthotopic 4T1 tumors. We assessed the transcription levels of a panel of chemokine and cytokine genes implicated in neutrophil trafficking²², including Cxcl1, Cxcl2, Cxcl9, Cxcl12, Cxcl14, Cxcl16, Ccl1, Ccl2, Ccl5, Ccl7, Ccl8 and Il-33. Among these, the transcription levels of Cxcl1 and Cxcl2 were most significantly and abundantly increased during liver metastasis (Fig. 2a). To validate these findings at the protein level, we measured CXCL1 and CXCL2 concentrations in liver tissue supernatants by enzyme-linked immunosorbent assay (ELISA). Both chemokines showed a progressive increase during metastatic progression (Fig. 2b, c). In light of these results, we assessed the expression of Cxcr2, which encodes CXCR2, a G protein–coupled receptor known to bind both CXCL1 and CXCL2²³. We found that CXCR2 was predominantly enriched in neutrophils compared to other immune subsets in our scRNA-seq dataset (Fig. 2d). These findings implicate the CXCL1/2–CXCR2 axis in neutrophil recruitment to the metastatic liver.

Fig. 2. Neutrophil recruitment to the liver relies on the CXCL1/2-CXCR2 axis.

a RT-qPCR screening of chemokine and cytokine gene expression in liver from BALB/c mice orthotopically inoculated with 4T1 cells for corresponding weeks (n = 3 mice per group). b, c ELISA measurement of CXCL1 (b) and CXCL2 (c) levels in the liver supernatant of the corresponding group (n = 5 mice per group). d t-SNE plot showing Cxcr2 expression across immune cells in murine scRNA-seq data (n = 3 mice per group). e Schematic of anti-CXCL1/2 (αCXCL1/2) neutralizing antibody or isotype treatment to BALB/c mice intrasplenically inoculated with 4T1 cells. f Flow cytometric quantification of hepatic neutrophils among live CD45⁺ cells in the livers from 4T1-bearing mice treated as indicated (n = 5 mice per group). g Schematic of CXCR2 inhibitor SB225002 or vehicle application to BALB/c mice intrasplenically inoculated with 4T1 cells. h Flow cytometric quantification of hepatic neutrophils in the livers from 4T1-bearing mice treated with SB225002 or vehicle (n = 4 mice per group). i Schematic of the intrasplenic injection model on WT and Cxcr2^-/- C57BL/6J mice. j Representative IF images (Left) and quantification of neutrophil (Right) in the livers from tumor-bearing WT and Cxcr2^-/- mice (n = 5 mice per group). Scale bars, 20 μm. k UMAP plots showing CXCL1/2 expression across human hepatic cell clusters (GSE246662, n = 6 patients). l Dot plot showing the expression level of CXCL1/2 in macrophage clusters in the metastatic and non-metastatic livers from patients (GSE246662, n = 6 patients). m–o, Representative IF images (m), and quantification of CXCL1⁺ F4/80⁺ (n) and CXCL2⁺ F4/80⁺ (o) cells per FOV in the normal and pre-metastatic livers (n = 6 mice per group). Scale bar, 20 μm. p, q Quantification of the proportions of CXCL1⁺F4/80⁺ cells among CXCL1⁺ cells (p) and CXCL2⁺F4/80⁺ cells among CXCL2⁺ cells (q) in the pre-metastatic livers (n = 6 mice per group). Schematic (e, g, i) is Created in BioRender. Z, S. (2026) https://BioRender.com/1gbdit2. Data are presented as mean ± SD (a–c, f, h, j, n–q) and statistical analysis was performed using Kruskal-Wallis test with Dunn’s test (a), one-way ANOVA with Dunnett’s test (b, c), Welch’s ANOVA with Dunnett’s T3 test (f), unpaired two-tailed t test (h, j) or Mann-Whitney two-tailed U test (n, o). Source data are provided as a Source Data file.

To further assess the role of CXCL1/2-CXCR2 axis in neutrophil recruitment, we performed intrasplenic injection of 4T1 tumor cells followed by blockade of CXCL1 and/or CXCL2 using neutralizing antibodies (Fig. 2e). Prior to intervention, we first monitored neutrophil infiltration dynamically by flow cytometry (FCM) and observed significant neutrophil infiltration in the liver as early as day 3 post-injection (Supplementary Fig. 2a). Based on this timeline, we administered anti-CXCL1 (αCXCL1), anti-CXCL2 (αCXCL2) or both antibodies daily for 3 consecutive days (Fig. 2e) and observed a marked reduction in liver-infiltrating neutrophils (Fig. 2f and Supplementary Fig. 2b). Similarly, daily treatment with the CXCR2 inhibitor SB225002 to mice for 9 days significantly suppressed hepatic neutrophil infiltration (Fig. 2g, h). Notably, prolonged treatment with either αCXCL1/2 antibodies or SB225002 over 20 days also resulted in reduced hepatic metastatic burdens (Supplementary Fig. 2c–f). In addition to neutrophils, CXCR2 is also reported to be expressed in various tumor types, where it may contribute to tumor cell proliferation and migration^24,25. To distinguish immune-mediated from tumor cell-intrinsic effects of CXCR2 blockade on metastasis, CXCR2-knockout (CXCR2-KO) 4T1 tumor cells were generated and intrasplenically inoculated into BALB/c mice, followed by treatment with the CXCR2 antagonist SB225002 (Supplementary Fig. 2g, h). CXCR2 deficiency in 4T1 tumor cells led to a modest reduction in liver metastatic burden compared with control tumor cells (Supplementary Fig. 2i, j), suggesting that there was a tumor cell-intrinsic contribution. Importantly, the anti-metastatic effect of SB225002 was largely preserved in mice bearing CXCR2-KO tumors (Supplementary Fig. 2i, j), suggesting that the predominant therapeutic effect of CXCR2 blockade in this model is mediated through host immune cells rather than tumor-intrinsic CXCR2 signaling. Consistently, in Cxcr2-knockout (Cxcr2^-/-) mice subjected to the same intrasplenic injection model, we observed a significant reduction in liver-infiltrating neutrophils and liver metastatic burdens compared to wild-type controls (Fig. 2i, j and Supplementary Fig. 2k, l). Collectively, these findings demonstrate that the CXCL1/2–CXCR2 axis plays a critical role in neutrophil recruitment to the metastatic liver microenvironment and thereby promotes liver metastasis.

Next, we sought to investigate the cellular source of elevated CXCL1/2 in the metastatic liver. Analysis of scRNA-seq data from patients revealed that both CXCL1 and CXCL2 were predominantly expressed by macrophages and were significantly upregulated in macrophages from the metastatic livers compared to those from the non-metastatic livers (Fig. 2k, l and Supplementary Fig. 2m). Murine scRNA-seq data revealed a similar pattern of Cxcl1 and Cxcl2 enrichment in the macrophage cluster (Supplementary Fig. 2n, o). Consistently, IF staining showed markedly increased levels of macrophage-derived CXCL1 and CXCL2 in the pre-metastatic livers compared to the normal controls (Fig. 2m–o). Further IF analysis revealed that the majority of CXCL1/2 in pre-metastatic livers was derived from macrophages (Fig. 2p, q). In addition, in the MC38 intrasplenic injection model, there were gradually increasing macrophages that produced CXCL1 and CXCL2 as hepatic metastasis progressed (Supplementary Fig. 2p–r). To further resolve the macrophage subsets contributing to CXCL1/2 expression, we distinguished Kupffer cells from infiltrating bone marrow-derived macrophages (BMDMs) by flow cytometry^26,27. Both populations expressed CXCL1/2 (Supplementary Fig. 2s), with Kupffer cells accounting for a substantial proportion of CXCL1/2-positive macrophages (Supplementary Fig. 2t). Together, these findings indicate that hepatic macrophages, especially resident Kupffer cells, constitute a major source of CXCL1/2 in the metastatic liver microenvironment and may contribute to neutrophil recruitment through the CXCL1/2-CXCR2 axis.

Macrophage-derived complement factor B initiates NETosis

Since CXCR2 ligands have been reported to stimulate NET extrusion¹³, we next examined whether blocking CXCR2 could suppress NET formation induced by pre-metastatic liver supernatant in vitro. As expected, the liver supernatant significantly triggered NETosis; however, CXCR2 blockade did not significantly reduce NET abundance, suggesting that CXCR2 ligands were not the major triggers of NETosis within the metastatic liver microenvironment (Supplementary Fig. 3a). A long-held view has identified the liver as the primary reservoir of complement factors, which are rapidly mobilized in response to infection or tissue damage²⁸. Among these, the anaphylatoxins C3a and C5a are potent inducers of NETosis²⁹. Murine scRNA-seq data analysis revealed that complement-related gene score was significantly upregulated during liver metastasis (Fig. 3a). Enhanced complement induction was also observed in the metastatic livers compared to the non-metastatic livers in patient-derived scRNA-seq datasets (Fig. 3b). These findings led us to hypothesize that complement activation may contribute to NETosis in the metastatic liver.

Fig. 3. Complement factor B (CFB) promotes NET formation in the metastatic liver.

a Violin plot displaying complement induction scores across liver immune cells from BALB/c mice orthotopically inoculated with 4T1 cells (n = 3 mice per group). b GSEA revealing enrichment of the “HALLMARK_COMPLEMENT” pathway in immune cells in human ML compared to NML (GSE246662, n = 3 patients per group). c, d ELISA measurement of C3a (c) and C5a (d) level in liver supernatant (n = 5 mice per group). e, f Representative IF images (e) and quantification of MPO-H3cit colocalized area (f) of murine neutrophils stimulated with metastatic liver supernatant with or without SB290157 and αC5aR antibody in vitro (n = 5 biologically independent experiments). Scale bar, 20 μm. g–i Representative IF images (g), and quantification of neutrophil (h) and MPO-H3cit colocalized area (i) in the livers from EO771-bearing mice treated as indicated (n = 5 mice per group). Scale bar, 50 μm. j, k RT-qPCR screening of complement factor genes (j, n = 3 mice per group) and ELISA measurement of CFB (k, n = 5 mice per group) in the liver tissues. l Dot plot showing CFB expression in macrophages from human ML and NML samples (GSE246662, n = 3 patients per group). m Representative IF images (Left) and quantification of CFB⁺ F4/80⁺ cells per FOV (Right) in the normal and pre-metastatic livers (n = 6 mice per group). Scale bar, 20μm. n Schematic of intrasplenic injection model on WT and Cfb^-/- mice. o, p Representative IF images (o) and quantification of MPO-H3cit colocalized area (p) in the livers (n = 5 mice per group). Scale bar, 20 μm. q, r Representative BLI images and quantification of liver photon flux (q), and representative gross images, H&E staining and quantification of liver metastatic area (r) in mice (n = 5 mice per group). Scale bar, 100 μm. Schematic (n) is Created in BioRender. Z, S. (2026) https://BioRender.com/m3tgn23. Data are presented as mean ± SD (c, d, f, h, i, j, k, m, p–r) and statistical analysis was performed using two-sided GSEA-based permutation test with False Discovery Rate (FDR) adjustments (b), Welch’s ANOVA with Dunnett’s T3 test (c, d, f), unpaired two-tailed t test (h), Welch’s unpaired two-tailed t test (i, q, r), Kruskal-Wallis test with Dunn’s test (j), one-way ANOVA with Dunnett’s test (k), Mann-Whitney two-tailed U test (m) or multiple Welch’s unpaired two-tailed t test (p). Source data are provided as a Source Data file.

To test this hypothesis, we measured C3a and C5a levels in liver tissue supernatants from the orthotopic breast cancer model using ELISA, and observed that both C3a and C5a levels increased progressively with metastatic progression (Fig. 3c, d). We next investigated whether complement activation contributes functionally to NET formation. Pharmacological inhibition of C3a and C5a signaling using a C3a receptor (C3aR) inhibitor (SB290157) and a C5aR-blocking antibody (αC5aR) markedly attenuated NET formation induced by the metastatic liver supernatant in vitro (Fig. 3e, f). Consistently, in vivo inhibition of C3aR and C5aR also significantly reduced NET formation within metastatic livers (Supplementary Fig. 3b and Fig. 3g–i). To more precisely distinguish the relative contributions of CXCL1/2–CXCR2 signaling and complement activation to NETosis in the metastatic liver niche, we collected liver supernatants from mice treated with CXCL1/2-neutralizing antibodies and used these supernatants to stimulate murine neutrophils in vitro. Notably, despite CXCL1/2 blockade, the metastatic liver supernatants retained a strong capacity to induce NET formation (Supplementary Fig. 3c, d), suggesting that CXCL1/2 signaling is not the dominant driver of NETosis in this setting. In contrast, additional inhibition of C3aR and C5aR signaling almost completely abolished NET formation induced by these supernatants (Supplementary Fig. 3c, d). Together, these data suggest that C3a- and C5a-mediated complement activation, instead of CXCL1/2-CXCR2 signaling, plays a key role in driving NET formation in the metastatic liver.

Generally, the complement system could be activated through three major pathways, including the classical, alternative, and lectin pathways³⁰. To determine which pathway is predominantly engaged in the liver metastatic niche, we conducted an RT-qPCR screening assay analyzing the transcriptional level of genes encoding key components of the complement system across all three complement pathways in the liver tissues from the orthotopic breast cancer model. Among these genes, the transcription of complement factor B (Cfb), a key component of the alternative pathway, exhibited the most pronounced upregulation from week 0 to week 5 (Fig. 3j). Consistently, ELISA analysis of the liver supernatants confirmed a gradual increase in CFB protein levels over the same period (Fig. 3k). These findings imply that activation of the alternative complement pathway via CFB may contribute to NET formation in the metastatic liver.

We next sought to elucidate the cellular source of elevated CFB in the metastatic liver. Murine scRNA-seq analysis revealed that Cfb expression was mainly restricted to macrophages (Supplementary Fig. 3e), which was also the major cell type with activated complement pathway (Supplementary Fig. 3f). Longitudinal gene set variation analysis (GSVA) of macrophages showed gradually upregulated complement activation pathway from week 0 to week 5 (Supplementary Fig. 3g), accompanied by a gradual increase in Cfb expression (Supplementary Fig. 3h). In parallel, gene set enrichment analysis (GSEA) of scRNA-seq data from patients demonstrated significantly enhanced complement activation pathway and CFB expression in macrophages from the metastatic livers compared to those from the non-metastatic livers (Supplementary Fig. 3i and Fig. 3l). IF staining further confirmed enhanced production of CFB by macrophages in the pre-metastatic livers (Fig. 3m). Similarly, a progressive increase in CFB-producing macrophages was observed in the MC38 intrasplenic injection model as liver metastasis progressed (Supplementary Fig. 3j, k). Collectively, these data implicate liver macrophages as the primary source of CFB in the metastatic liver.

To functionally assess the role of macrophage-derived CFB in driving NETosis during liver metastasis, we performed longitudinal immunofluorescence analyses to monitor NET formation and neutrophil infiltration in the livers of wild-type (WT) and Cfb-knockout (Cfb^-/-) mice subjected to intrasplenic inoculation of EO771-luc tumor cells (Fig. 3n). Consistent with impaired complement activation, Cfb deficiency markedly reduced the levels of CFB, C3a, and C5a in metastatic liver supernatants (Supplementary Fig. 3l–n). Quantitative immunofluorescence analysis revealed that NETs became detectable by day 10 post-injection and rapidly increased as metastasis progressed in WT mice. In contrast, NET formation was nearly absent in the metastatic livers of Cfb^-/- mice during the course of liver metastasis (Fig. 3o, p). Notably, neutrophil infiltration was only modestly reduced in the livers of Cfb^-/- mice compared to WT mice, and this decrease did not reach statistical significance (Supplementary Fig. 3o). To exclude the possibility that Cfb deficiency indirectly affects macrophage CXCL1/2 expression, we evaluated CXCL1-, CXCL2- and CFB-expressing macrophages in the livers of WT and Cfb^-/- mice. As expected, few CFB-expressing macrophages could be detected in Cfb^-/- mice, while the number of CXCL1/2-expressing macrophages remained comparable to that in WT mice (Supplementary Fig. 3p, q). Importantly, Cfb knockout significantly reduced the liver metastasis burden, as demonstrated by BLI (Fig. 3q), gross liver images and hematoxylin and eosin (H&E) staining (Fig. 3r). Together, these data indicate that macrophage-derived CFB is a critical driver of NETosis and contributes to the pro-metastatic microenvironment in the liver.

Phagocytosis of tumor debris leads to CXCL1/2 and CFB elevation in macrophages within the pre-metastatic liver

We next investigated the mechanism underlying the increased production of CXCL1/2 and CFB by macrophages in the metastatic liver. Tumor cell-derived exosomes have been widely recognized as key mediators of tumor progression and metastasis⁵, yet their role in modulating NETosis remains unclear. To assess this, we educated mice with 4T1 cell-derived exosomes to determine their capacity to induce hepatic neutrophil infiltration and NET formation (Supplementary Fig. 4a). Exosomes were isolated from conditioned medium of 4T1 cells cultured in exosome-depleted FBS as previously described⁵, with their size and typical markers confirmed with NanoSight (the average diameter of exosomes was 114.0 ± 33.9 nm) and western blot assays, respectively (Supplementary Fig. 4b, c). After 2-week exosome education, no significant differences were observed in hepatic neutrophil recruitment or NET formation (Supplementary Fig. 4d, e). Moreover, prior exosome education did not affect hepatic NET generation following intrasplenic injection of 4T1 tumor cells (Supplementary Fig. 4f). To further validate these findings, we disrupted exosome secretion in tumor cells by knocking down Rab27a and Rab27b, which are key regulators of vesicle docking and exosome release³¹. Consistent with previous results, knocking down Rab27a/b in tumor cells did not affect NET generation in the metastatic livers (Supplementary Fig. 4g). These results suggest that tumor-derived exosomes are not the primary drivers of neutrophil recruitment or NETosis in the metastatic liver.

Dynamic scRNA-seq analysis of murine hepatic immune cells from the orthotopic breast cancer model revealed that the phagocytosis-associated pathway was upregulated in macrophages during liver metastasis, a finding corroborated by scRNA-seq data from patients (Supplementary Fig. 4h and Fig. 4a). These results suggest enhanced phagocytic activity of hepatic macrophages during metastatic progression. Previous studies have revealed that shear forces encountered by circulating tumor cells (CTCs) as they traverse pulmonary capillaries result in the generation of abundant tumor-derived microparticles, thereby modulating the phenotype of the resident immune cells in the lung³². Analogously, in the liver—a highly vascularized organ characterized by its specialized sinusoidal microvasculature—CTCs may likewise be susceptible to mechanical disruption during hematogenous dissemination. Cellular debris refers to fragmented cellular material derived from damaged or dying cells that have yet to be cleared by the immune system^33,34. As professional phagocytes, macrophages actively remove cellular debris through phagocytosis, which could induce phenotypic or metabolic reprogramming in macrophages, leading to the formation of a tumor-supportive microenvironment³⁵. In light of the above, we hypothesized that a surge of tumor debris may likewise be generated in the liver sinusoids during early metastatic seeding, subsequently triggering phagocytic clearance by macrophages. To verify this hypothesis, intravital real-time imaging of the pre-metastatic liver was performed at 4 h and 24 h following intrasplenic injection of mCherry-labeled 4T1 cells into BALB/c mice (Fig. 4b). Intriguingly, abundant tumor cell-derived particles were detected within the liver sinusoids as early as 4 h after injection (Fig. 4c). Quantitative analysis further revealed that these particles exhibited an average diameter of 3.9 ± 3.1 μm (Supplementary Fig. 4i), which falls within the reported size range of shear stress–induced tumor-derived microparticles³². At 24 hours post-inoculation, F4/80⁺ macrophages from the pre-metastatic livers had internalized substantial amounts of mCherry⁺ tumor microparticles (Fig. 4d). Flow cytometry analysis further confirmed that the majority of mCherry⁺ tumor signal was present in F4/80⁺ macrophages (Fig. 4e). These findings suggest that macrophages are the principal cells engulfing tumor cell-derived microparticles in the pre-metastatic liver microenvironment.

Fig. 4. Phagocytosis of tumor debris elevates CXCL1/2 and CFB production in liver macrophages.

a GSEA revealing enrichment of phagocytosis pathway in macrophages from human ML compared to NML samples (GSE246662, n = 3 patients per group). b–d Schematic showing mCherry⁺ or EGFP⁺ 4T1 cell intrasplenic injection (b), and representative intravital two-photon images showing mCherry⁺ 4T1-derived microparticles (white arrowhead) with liver sinusoids at 4 h (c) or macrophages at 24 h after injection (d). Scale bars, 20 μm. e Flow cytometric quantification of the proportion of mCherry⁺F4/80⁺ macrophages among total mCherry⁺ cells at 24 hours after injection (n = 4 mice). f Representative IF images staining in the pre-metastatic livers at 24 hours after injection of EGFP⁺ 4T1 cells. Scale bar, 20 μm. g Flow cytometry analysis of CXCL1/2/CFB expression in macrophages from normal or pre-metastatic livers 24 h after mCherry⁺ 4T1 cell injection (n = 4 mice per group). h RT-qPCR analysis showing Cxcl1/2 and Cfb expression in liver macrophages treated as indicated in vitro (n = 4 biologically independent experiments). i, j Representative IF images (i) and quantification of MPO-H3cit colocalized area (j) in the livers from mice treated as indicated (n = 5 mice per group). Scale bar, 20 μm. k–p Schematic showing CytoB treatment to 4T1-bearing mice (k), representative IF images (l, scale bar, 20 μm) and quantification of CXCL1/2/CFB⁺F4/80⁺ macrophages (m), flow cytometric quantification of neutrophils (n), and H&E staining (o, scale bar, 100 μm) and quantification of metastatic area (p) in the livers from mice treated as indicated (n = 5 mice per group). Yellow arrowheads indicate CXCL1/2/CFB⁺F4/80⁺ macrophages; white asterisks mark magnified regions; black dashed outlines indicate metastatic lesions. q–s Representative IF images (q), and quantification of neutrophil (r) and MPO-H3cit colocalized area (s) in the livers from 4T1 tumor-bearing mice treated as indicated (n = 5 mice per group). Scale bar, 20 μm. Schematic (b, k) is Created in BioRender. Z, S. (2026) https://BioRender.com/e33rzar. Data are presented as mean ± SD (e, h, j, m, n, p, r, s) and statistical analysis was performed using two-sided GSEA-based permutation test with FDR adjustments (a), Welch’s ANOVA with Dunnett’s T3 test (h, j), multiple Welch’s unpaired two-tailed t test (m), Welch’s unpaired two-tailed t test (n, p), or two-way ANOVA with Tukey’s test (r, s). Source data are provided as a Source Data file.

We next examined whether phagocytosis of tumor cell-derived materials contributes to macrophage-derived CXCL1/2 and CFB production. In the pre-metastatic livers harvested 24 hours after intrasplenic injection of enhanced green fluorescent protein-tagged (EGFP⁺) 4T1 cells, macrophages that had engulfed EGFP⁺ tumor cell-derived materials exhibited production of CXCL1/2 and CFB (Fig. 4f and Supplementary Fig. 4j). Moreover, macrophages loaded with mCherry⁺ tumor cell-derived materials (mCherry⁺ macrophages) in the pre-metastatic livers expressed markedly higher levels of CXCL1/2 and CFB than those that had not phagocytosed mCherry⁺ tumor cell-derived materials (mCherry^- macrophages) or steady-state macrophages from the normal livers (Fig. 4g). These in vivo data suggest that phagocytosis of tumor cell-derived materials promotes CXCL1/2 and CFB production in macrophages. However, fluorescence-based detection alone cannot definitively determine whether the tumor-derived materials internalized by macrophages represent tumor debris or other forms of tumor-derived materials.

To further distinguish the contribution of tumor debris from other tumor cell-derived materials, we cocultured isolated liver macrophages with either intact tumor cells or experimentally generated tumor debris produced by repeated freeze-thaw cycles in vitro. Notably, phagocytic uptake of tumor debris by macrophages markedly induced the transcription of Cxcl1, Cxcl2 and Cfb, whereas coculture with intact tumor cells did not elicit a comparable response (Fig. 4h). This induction was abrogated by cytochalasin B (CytoB), an actin polymerization inhibitor³⁶ that blocks phagocytosis (Fig. 4h). To directly test whether phagocytosis of tumor debris is sufficient to induce NET formation in vivo, we administered CytoB to mice intrasplenically injected with experimentally generated 4T1 tumor debris (Supplementary Fig. 4k), and found that intrasplenic injection of tumor debris induced robust NETosis in the liver, whereas CytoB treatment markedly reduced NET formation under these conditions (Fig. 4i, j). These results indicate that phagocytosis of tumor debris, rather than other tumor cell-derived materials, is sufficient to induce macrophage expression of CXCL1/2 and CFB, thereby promoting NETosis in the metastatic liver.

To further examine the impact of phagocytosis inhibition on neutrophil recruitment and NETosis in vivo, we administered CytoB to mice subjected to intrasplenic injection of 4T1 tumor cells (Fig. 4k). In line with in vitro findings, CytoB treatment significantly reduced macrophage-derived CXCL1/2 and CFB levels (Fig. 4l, m), accompanied by diminished neutrophil infiltration and NET formation (Fig. 4n and Supplementary Fig. 4l–n). CytoB treatment over 20 days also led to a substantial decrease in hepatic metastatic burden (Fig. 4o, p). To more specifically evaluate the contribution of macrophages, we combined clodronate liposome (Clod)-mediated macrophage depletion with CytoB treatment in vivo (Supplementary Fig. 4o, q). Macrophage depletion alone caused a substantial reduction in both hepatic NET formation and neutrophil infiltration. Importantly, combined treatment with CytoB and clodronate liposomes resulted in only a modest and statistically non-significant additional reduction in NET formation and neutrophil accumulation compared with macrophage depletion alone (Fig. 4q–s), suggesting that the predominant in vivo effects of CytoB on neutrophil recruitment and NET formation are mediated through macrophage-dependent phagocytic mechanisms.

Collectively, these findings demonstrate that phagocytosis of tumor debris by hepatic macrophages enhances CXCL1/2 and CFB production, which in turn promotes neutrophil recruitment and NETosis, facilitating metastatic progression in the liver.

NETs diminish NK cell function through CCDC25

Previous studies have demonstrated that NETs could modulate the tumor microenvironment^8,13. In addition, our murine scRNA-seq data revealed early NET generation followed by a progressively immunosuppressive microenvironment in the liver during metastatic progression (Fig. 1). Nevertheless, whether and how NETs affect the liver metastatic microenvironment remains poorly defined. Numerous studies have identified CCDC25 as a transmembrane DNA sensor on tumor cells that senses extracellular DNA and facilitates metastatic dissemination^9,37,38. Interestingly, our murine scRNA-seq data revealed broad expression of Ccdc25 across various immune cell subsets in the liver (Supplementary Fig. 5a), a pattern similarly observed in scRNA-seq data from patients (Supplementary Fig. 5b). These observations raised the possibility that CCDC25 might also mediate NET-induced immune remodeling within the liver metastatic microenvironment.

To test this hypothesis, we injected parental EO771-luc tumor cells or Ccdc25-knockout (Ccdc25^-/-) EO771-luc tumor cells into the spleens of C57BL/6J mice, followed by treatment with either CCDC25 neutralizing (αCCDC25) antibody or isotype control antibody (Fig. 5a). As expected, administration of αCCDC25 antibody remarkably suppressed hepatic metastasis. Intriguingly, even in Ccdc25^-/- EO771-luc tumor-bearing mice, αCCDC25 treatment further lowered liver metastatic burden compared to IgG control (Fig. 5b, c and Supplementary Fig. 5c), suggesting that CCDC25 exerts a pro-tumor effect beyond tumor cells. Moreover, we employed an intrasplenic injection model on WT, Padi4^-/- (which encodes peptidyl-arginine deiminase 4, an enzyme required for NET formation³⁹) and Ccdc25^-/- mice, respectively (Fig. 5d). Both Padi4^-/- and Ccdc25^-/- mice exhibited remarkably reduced liver metastatic burden compared to WT mice (Fig. 5e, f and Supplementary Fig. 5d). These results indicated that the NET-CCDC25 axis exerted a pro-tumor effect on non-tumor cells within the liver metastatic niche.

Fig. 5. NETs diminish NK cell-mediated tumor surveillance via CCDC25.

a Schematic showing the intrasplenic injection of EO771-WT-luc or EO771-sgCcdc25-luc to C57BL/6J mice, followed by CCDC25 neutralizing (αCCDC25) antibody or Isotype-IgG treatment. b Representative BLI images (Left) and quantification of liver photon flux (Right) in mice on day 30 post-inoculation (n = 5 mice per group). c Representative gross and H&E staining images of the livers of mice from corresponding groups (n = 5 mice per group). Scale bar, 5 mm. d Schematic showing intrasplenic injection of EO771-luc cells to WT, Padi4^-/- and Ccdc25^-/- C57BL/6J mice. e Representative BLI images (Left) and quantification of liver photon flux (Right) in mice on day 30 post-inoculation (n = 5 mice per group). f Representative gross and H&E staining images of the livers of mice on day 30 post-inoculation (n = 5 mice per group). Scale bar, 5 mm. g Box plot showing the GSVA scores of NK activation, T cell activation, B cell activation, DC activation and macrophage activation in whole-liver RNA-seq data of WT, Padi4^-/- and Ccdc25^-/- mice on day 30 post-inoculation (n = 4 mice per group). h–j Experimental design for depletion of CD3⁺ T cells or NK1.1⁺ cells in WT and Ccdc25^-/- mice before and after EO771-luc intrasplenic inoculation (h), with representative BLI images (i) and H&E liver images (j) on day 15 post-inoculation (n = 5 mice per group). k–m Schematic of adoptive transfer of spleen-derived WT or Ccdc25^-/- NK cells into EO771-luc-bearing C57BL/6J recipient mice (k), representative BLI images and liver photon-flux quantification (l), and gross and H&E liver images (m; scale bar, 5 mm) (n = 5 mice per group). Schematic (a, d, h, k) is Created in BioRender. Z, S. (2026) https://BioRender.com/7os6yit. For box plots in (g), boxes indicate the IQR, center lines indicate medians and whiskers indicate minima and maxima. Data are presented as mean ± SD (b, e, l) and statistical analysis was performed using two-way ANOVA with Tukey’s test (b), one-way ANOVA with Tukey’s test (e, l) or one-way ANOVA with Dunnett’s test (g). Source data are provided as a Source Data file.

To determine which cell is primarily affected by the NET-CCDC25 axis, we performed bulk RNA-seq on liver tissues from WT, Padi4^-/- and Ccdc25^-/- mice subjected to the metastatic model described above. We analyzed cell activation signatures covering multiple cell types, including NK cell, T cell, B cell, DC, macrophage, hepatocyte, endothelial cell and hepatic stellate cell. Among these, NK cell was the only cell type exhibiting a marked upregulation of activation pathway in the livers of both Padi4^-/- and Ccdc25^-/- mice (Fig. 5g and Supplementary Fig. 5e). Given prior scRNA-seq data revealing suppressed NK and T cell activation as liver metastasis progressed, we next depleted NK or T cells using αNK1.1 or αCD3 neutralizing antibodies (Supplementary Fig. 5f), respectively, in C57BL/6J WT mice and Ccdc25^-/- mice intrasplenically inoculated with EO771-luc cells (Fig. 5h). Notably, Ccdc25 knockout significantly suppressed liver metastases in isotype control and T cell–depleted mice, whereas this protective effect was largely abolished in NK cell–depleted mice (Fig. 5i, j and Supplementary Fig. 5g, h). To investigate why the NET-CCDC25 axis preferentially affects NK cells, we performed immunofluorescence analysis of metastatic liver sections, which showed that NK cells expressed high levels of CCDC25 and were spatially enriched in NET-rich regions (Supplementary Fig. 5i, j). These spatial and expression patterns suggest that NK cells are particularly susceptible to NET-CCDC25 signaling within the metastatic liver microenvironment, thereby supporting the notion that the NET-CCDC25 axis primarily exerts its pro-metastatic effects through NK cell dysfunction in the hepatic metastatic niche.

To confirm the expression of CCDC25 on NK cell membranes, we performed CCDC25 staining on both human and murine NK cells using flow cytometry without permeabilization and observed substantial surface expression of CCDC25 (Supplementary Fig. 5k). To further verify the direct interaction between NET-DNA and CCDC25 on the membrane of NK cells, we isolated plasma membrane proteins from NK cells and incubated them with biotinylated NET-DNA. Pull-down assay using streptavidin beads revealed that CCDC25 was specifically captured by biotinylated NET-DNA (Supplementary Fig. 5l), confirming its direct binding to extracellular DNA, which is in line with our prior findings^9,40. Functionally, NET treatment significantly impaired NK cytotoxicity in vitro, while either blocking CCDC25 on NK cells or digesting NET-DNA with DNase I significantly restored NK cell function (Supplementary Fig. 5m–o). To further avoid the off-target effect of CCDC25 deletion in vivo, we employed an NK cell intrasplenic adoptive transfer mouse model as previously described^41,42 (Fig. 5k). While transfer of WT NK cells into EO771-luc tumor-bearing mice modestly reduced the hepatic metastatic burden, transfer of CCDC25-deficient NK cells led to a markedly lower liver metastatic burden, suggesting that CCDC25 depletion enhances the anti-tumor function of NK cells (Fig. 5l, m and Supplementary Fig. 5p).

To further exclude the possibility that other CCDC25-expressing cell populations indirectly regulate hepatic NK cell function, we performed reciprocal adoptive transfer experiments in which WT NK cells were transferred into NK-depleted Ccdc25^⁻/⁻ recipient mice (Supplementary Fig. 5q). Under these conditions, hepatic metastatic burdens in NK-depleted Ccdc25^⁻/⁻ recipient mice were comparable to that observed in WT mice (Supplementary Fig. 5r, s). In addition, activating receptor expression and cytotoxic activity of the transferred WT NK cells were not significantly altered relative to those in WT mice (Supplementary Fig. 5t). Collectively, these findings indicate that CCDC25-dependent regulation of NK cell function is predominantly NK cell-intrinsic and support a model in which NETs impair NK cell-mediated tumor surveillance through CCDC25 signaling.

The NET-CCDC25 axis downregulates multiple activating receptors on NK cells

We next investigated how the NET-CCDC25 axis impaired NK cell function. NK cell exerts its tumor surveillance effect through the perforin and granzyme B pathway, interferon-γ pathway, as well as the death receptor ligand pathway⁴³. Different from T cell, NK cell effector function is independent of MHC-mediated antigen presentation and is instead regulated by a complex array of activating and inhibitory receptors⁴⁴. To clarify how NET-CCDC25 axis modulates cytotoxic function of NK cells, we performed an RT-qPCR screening assay on NK cells treated with or without NETs, assessing the transcription levels of genes encoding activating receptors (KLRK1, KLRC2, CD226, NCR1, NCR2, NCR3, CD244, FCGR3A, ITGB2), inhibitory receptors (KIR2DL2, KIR2DL3, KIR3DL1, KIR3DL2, KIR3DL3, KLRC1, TIGIT, LAG3, HAVCR2, KLRB1, KLRG1, PDCD1, CD96), cytotoxic granule proteins (PRF1 and GZMB) and anti-tumor cytokines (TNF and IFNG). Interestingly, NETs induced broad transcriptional downregulation of activating receptors and PRF1, with minimal effects on inhibitory receptors, GZMB, or cytokines (Fig. 6a). Flow cytometry further confirmed that NET exposure reduced the expression of activating receptors (NKG2D, NKp46, NKp44, 2B4, DNAM-1) on both human and murine NK cells (noting that NKp44 is not expressed in mice), and this effect was largely rescued by CCDC25 blockade (Fig. 6b, c). Furthermore, NET treatment also markedly reduced surface CD107a levels on NK cells, whereas CCDC25 inhibition or NET-DNA digestion with DNase I restored surface levels of CD107a (Fig. 6d–g and Supplementary Fig. 6a). These observations indicate that the NET-CCDC25 axis suppresses activating receptor expression and impairs the degranulation capacity of NK cells. Moreover, we evaluated the ligands for these receptors on tumor cell lines, including K562, EO771 and 4T1 cells. Flow cytometric analysis revealed distinct ligand expression profiles across these cell lines. K562 cells expressed high levels of ligands for DNAM-1 and NKG2D, with comparatively lower levels of ligands for NKp46, NKp44 and 2B4 (Supplementary Fig. 6b, c). EO771 cells expressed broadly comparable levels of ligands for 2B4, NKp46, NKG2D and DNAM-1 (Supplementary Fig. 6d, e), whereas 4T1 cells exhibited relatively higher levels of ligands for 2B4, NKp46 and DNAM-1, but lower levels of NKG2D ligands (Supplementary Fig. 6f, g). These ligand expression profiles provide a molecular basis for activating receptor–mediated recognition and killing of tumor cells by NK cells.

Fig. 6. The NET-CCDC25 axis downregulates activating receptors of NK cell.

a RT-qPCR screening assay assessing the transcription of activating receptor, cytotoxic granule protein and anti-tumor cytokine genes (Left), and inhibitory receptor genes (Right) in human NK cells treated with or without NETs (n = 4 biologically independent experiments). b Flow cytometry analysis showing the expression of NKG2D, NKp46, NKp44, 2B4 and DNAM-1 on human NK cells (CD3^- CD56⁺) treated as indicated in vitro (n = 3 biologically independent experiments). MFI, median fluorescence intensity. c Flow cytometry analysis showing the expression of NKG2D, NKp46, 2B4 and DNAM-1 in murine NK cells, untreated or pre-treated with αCCDC25 antibody or IgG Isotype control, followed by NET stimulation (n = 3 biologically independent experiments). MFI, median fluorescence intensity. d–g Flow cytometry analysis showing the frequency of CD107a⁺ human NK cells (d, e) or murine NK cells (f, g) pre-treated with either IgG or αCCDC25 antibody, prior to stimulation with NETs or DNase I-digested NETs (n = 3 biologically independent experiments). h, i, Flow cytometry analysis showing the frequency of CD107a⁺ NK cells (CD3^- NK1.1⁺), the expression of NKG2D (h), and the frequency (i) of the liver-infiltrating NK cells from C57BL/6J mice on day 30 post-inoculation with EO771 cells, with αCCDC25 antibody or Isotype-IgG administration (n = 5 mice per group). MFI, median fluorescence intensity. Data are presented as mean ± SD (a–c, e, g, h, i) and statistical analysis was performed using Welch’s unpaired two-tailed t test (a, NKG2D in h), unpaired two-tailed t test (CD107a in h) or one-way ANOVA with Tukey’s test (b, c, e, g). Source data are provided as a Source Data file.

We next sought to validate these findings in vivo. In mice intrasplenically inoculated with EO771 cells, administration of αCCDC25 antibody increased both the frequency of CD107a⁺ NK cells and NKG2D expression levels in liver-infiltrating NK cells compared with IgG control (Fig. 6h). Notably, this enhancement in NK-cell activation occurred without appreciably altering the overall infiltration of NK cells into the liver (Fig. 6i), indicating that CCDC25 blockade primarily restores NK-cell functionality rather than affecting NK-cell recruitment.

Given the potential translational implications of targeting CCDC25, we further examined major organs from Ccdc25^⁻/⁻ mice, with particular attention to tissues enriched in NK cells, such as the spleen and bone marrow. Histopathological analysis revealed no evident structural abnormalities, inflammatory infiltration, or tissue damage in Ccdc25^⁻/⁻ mice compared with wild-type controls (Supplementary Fig. 6h). Collectively, these results demonstrate that the NET–CCDC25 axis impairs NK cell function by downregulating NK cell activating receptors and suppressing degranulation, while CCDC25 blockade restores NK cell effector activity in vivo.

NET-CCDC25 axis downregulates activating receptors on NK cells through ILK/STAT3 pathway

To explore the mechanism by which the NET-CCDC25 axis downregulates NK cell activating receptors, RNA-seq was performed on human-derived NK cells treated with (“NETs” group) or without (“CTRL” group) NETs. Principal components analysis (PCA) revealed substantial transcriptomic reprogramming in NET-treated NK cells (Supplementary Fig. 7a). GSEA analysis identified upregulation of the Janus kinase–signal transducer and activator of transcription (JAK-STAT) signaling pathway following NET stimulation (Fig. 7a), with pathway enrichment further highlighting significant activation of STAT3-associated signaling (Supplementary Fig. 7b). Consistently, murine scRNA-seq data showed progressive upregulation in STAT3 signaling pathway in NK cells during the course of liver metastasis (Fig. 7b). Analysis of scRNA-seq data from patients similarly demonstrated significantly enriched and upregulated IL-6-JAK-STAT3 signaling pathway in NK cells from the metastatic livers compared to those from non-metastatic livers (Fig. 7c). To further examine the dynamics of STAT3 activation, human NK cells were treated with NETs for varying durations (0, 10, 20, 30 and 40 min) to examine the phosphorylation of STAT3 during the process. Immunoblotting showed that STAT3 phosphorylation at tyrosine 705 residue (Y705) was markedly enhanced within 20 min of NET treatment and sustained up to 40 min (Supplementary Fig. 7c). Together, these data imply that NETs trigger tyrosine phosphorylation of STAT3 in NK cells.

Fig. 7. The NET-CCDC25 axis impairs NK-cell function through the ILK/STAT3 pathway.

a GSEA revealing enrichment of JAK-STAT signaling pathway in NK cells treated with NETs (n = 3 samples per group). b Violin plot showing STAT3 pathway scores in hepatic NK cells from BALB/c mice orthotopically inoculated with 4T1 cells (n = 3 mice per group). c GSEA revealing enrichment of IL-6-JAK-STAT3 signaling pathway in NK cells in human ML compared to NML samples (GSE246662, n = 3 patients per group). d Representative western blot of p^Y705-STAT3 and total STAT3 in human NK cells treated as indicated. GAPDH served as a loading control (n = 3 biologically independent experiments). e, f Flow cytometry analysis of K562 tumor-cell killing by human NK cells transfected with STAT3 or control siRNA before NET stimulation (n = 3 biologically independent experiments). g Schematic showing Napabucasin treatment to EO771-luc tumor-bearing WT and Ccdc25^-/- mice. h Flow cytometry analysis showing CD107a⁺ NK cell frequency, and the expression of corresponding receptors on hepatic NK cells from corresponding groups (n = 5 mice per group). MFI, median fluorescence intensity. i ChIP-qPCR analysis showing enrichment of pSTAT3 at the promoter regions of corresponding genes in NK cells with or without NET stimulation (n = 3 biologically independent experiments). j RT-qPCR analysis showing expression of corresponding genes in NET-treated NK cells with or without Napabucasin pretreatment (n = 3 biologically independent experiments). k Representative western blot of p^Y705-STAT3, total STAT3 and ILK in human NK cells transduced with two ILK-targeting or control sgRNAs before NET stimulation. GAPDH served as a loading control (n = 3 biologically independent experiments). l–n Flow cytometry analysis showing the expression of corresponding receptors (l), K562 tumor-cell killing (m) and CD107a⁺ NK cell frequency (n) in human NK cells treated as indicated (n = 3 biologically independent experiments). MFI, median fluorescence intensity. For violin plots in (b), boxes indicate the IQR, center lines indicate medians and whiskers indicate minima and maxima. Schematic (g) is Created in BioRender. Z, S. (2026) https://BioRender.com/ol5nuwb. Data are presented as mean ± SD (f, h–j, l–n) and statistical analysis was performed using two-sided GSEA-based permutation test with FDR adjustments (a, c), Kruskal-Wallis test with Dunn’s test (b), one-way ANOVA with Sidak’s test (f) or two-way ANOVA with Tukey’s test (h), unpaired two-tailed t test (i), two-way ANOVA with Sidak’s test (j) or one-way ANOVA with Dunnett’s test (l–n). Source data are provided as a Source Data file.

To further elucidate the interplay among NETs, CCDC25 and STAT3 signaling, we blocked CCDC25 on NET-treated NK cells with αCCDC25 antibody and observed a significant reduction in NET-induced STAT3 phosphorylation at Y705 (Fig. 7d and Supplementary Fig. 7d). Consistently, the proportions of p^Y705-STAT3⁺ NK cells were substantially lower in the metastatic livers from Padi4^-/- and Ccdc25^-/- mice compared to WT mice (Supplementary Fig. 7e, f). These findings suggest NET-induced STAT3 Y705 phosphorylation is dependent on CCDC25.

STAT3 phosphorylation has been widely shown to suppress NK-cell antitumor immunity, whereas genetic deletion of STAT3 markedly enhances NK-dependent tumor surveillance across multiple models^45–47. Longitudinal immunofluorescence analyses further confirmed in vivo that NET accumulation (Fig. 1g) temporally coincided with increased STAT3 phosphorylation in NK cells (Supplementary Fig. 7g, h), followed by progressive downregulation of the NK cell activating receptor NKG2D at later stages (Supplementary Fig. 7i). Functionally, pharmacological inhibition of STAT3 phosphorylation with Napabucasin in vitro significantly reversed the NET-induced downregulation of NK cell activating receptors on both human and murine NK cells (Supplementary Fig. 7j, k). In addition, STAT3 knockdown using siRNA significantly enhanced NK cell cytotoxicity under NET stimulation (Supplementary Fig. 7l and Fig. 7e, f), which supports STAT3 phosphorylation as a central mediator of NET-induced NK suppression. To assess the functional relevance of CCDC25-STAT3 signaling in vivo, we treated WT or Ccdc25^-/- mice subjected to intrasplenic inoculation of EO771-luc tumor cells with or without Napabucasin (Fig. 7g). As expected, both Ccdc25 knockout and STAT3 inhibition effectively preserved NK activating receptor expression. Notably, Napabucasin treatment conferred no additional benefit in Ccdc25^-/- mice, indicating that CCDC25 and STAT3 operate within the same pathway to mediate NET-driven NK cell suppression (Fig. 7h). Collectively, these findings demonstrate that the NET-CCDC25 axis suppresses NK cell activity through STAT3 Y705 phosphorylation–dependent downregulation of activating receptors.

To determine the mechanism by which phosphorylated STAT3 (pSTAT3) downregulates the expression of NK activating receptor, sequence analysis was conducted and revealed putative STAT3-binding motifs within the promoter regions of KLRK1, NCR2, NCR1, CD226 and CD244 genes (Supplementary Fig. 7m). Further ChIP-qPCR analysis on NK cells demonstrated increased enrichment of bound pSTAT3 at the promoter regions of KLRK1, NCR2, NCR1, CD226 and CD244 upon NET stimulation (Fig. 7i). Based on this, we hypothesized that pSTAT3 acts as a repressor inhibiting the transcription of NK activating receptor genes upon NET stimulation. To test this, we conducted RT-qPCR analysis using STAT3 inhibitor napabucasin on NK cells prior to NET treatment. As expected, transcriptional levels of these genes were significantly reduced upon NET exposure (Fig. 7j), which is in line with our prior results in Fig. 6a. Importantly, pharmacological inhibition of STAT3 prior to NET stimulation largely restored their transcription (Fig. 7j), indicating that STAT3 activation is required for this suppressive effect. Together, these findings support a model in which pSTAT3 mediates transcriptional repression of NK activating receptors, consistent with a previous report showing that pSTAT3 can act as a transcriptional repressor, such as by binding the Kdm6b locus and suppressing Kdm6b transcription⁴⁸.

To further delineate the signaling cascade linking CCDC25 to STAT3 activation, we next sought to determine whether an adapter acts between CCDC25 and STAT3. Our previous study demonstrated that CCDC25 can intracellularly associate with integrin-linked kinase (ILK) in tumor cells, and that this interaction is enhanced upon NET stimulation⁹. In parallel, ILK has been reported to promote STAT3 phosphorylation at Y705 residue across multiple cell types^49–52, which led us to hypothesize that ILK functions as the adapter bridging CCDC25 to STAT3 phosphorylation in NK cells. To determine whether CCDC25 also interacts with ILK in NK cells, we performed co-immunoprecipitation (Co-IP) analysis on NK cells and observed a clear interaction between CCDC25 and ILK, which was markedly enhanced following NET stimulation (Supplementary Fig. 7n). To further investigate the functional role of ILK, we generated ILK-deficient PBMC-derived NK cells and examined STAT3 phosphorylation in response to NET exposure. Immunoblotting showed that NET-induced STAT3 Y705 phosphorylation was largely abolished in ILK-deficient NK cells (Fig. 7k). Functionally, ILK deficiency not only abolished NET-driven downregulation of activating receptors, but also restored NK cytotoxic function (Fig. 7l–n). Taken together, these findings suggest that the NET–CCDC25 axis impairs NK cell function through ILK-dependent STAT3 signaling, in which phosphorylated STAT3 acts as a transcriptional repressor to suppress the expression of NK cell activating receptor genes.

CXCL1/2 and CFB expression correlates with neutrophil and NET accumulation, and impairs NK-cell activation in human metastatic livers

To clinically validate our findings, we analyzed treatment-naïve histologic specimens of liver metastases obtained from our hospital using quantitative immunofluorescence. In metastatic liver tissues, macrophage-derived CXCL1/2 levels positively correlated with neutrophil infiltration (Fig. 8a–c), whereas macrophage-derived CFB expression showed a positive association with NET formation (Fig. 8d). Notably, NET abundance was inversely correlated with the expression level of NKG2D on NK cells (Fig. 8e). In addition, we also examined treatment-naïve primary breast tumor specimens and observed significantly higher levels of MPO and NETs in the metastatic liver tissues compared to primary tumors (Supplementary Fig. 8a), which is in agreement with our prior reports that NETs are scarce in the treatment-naïve primary breast tumor⁹. Following a similar pattern, elevated CXCL1/2 and CFB levels were also detected in the metastatic liver tissues (Supplementary Fig. 8b). Together, these clinical observations further corroborate our mechanistic findings, supporting the relevance of the CXCL1/2–CFB–NET axis and its association with NK-cell dysfunction in human liver metastases.

Fig. 8. Clinical relevance of neutrophil and NET accumulation in the metastatic livers from patients.

a Representative IF images staining in metastatic liver sections from breast cancer patients. Scale bars, 50 μm. White asterisks mark magnified regions. CD68, CXCL1, CXCL2 and CFB were co-stained on the same histological slide (n = 20 patients). b–e Quantitative IF analysis showing positive correlations between: CXCL1⁺ CD68⁺ cell infiltration and MPO⁺ cell infiltration (b), CXCL2⁺ CD68⁺ cell infiltration and MPO⁺ cell infiltration (c), CFB⁺ CD68⁺ cell infiltration and NET formation (d); a negative correlation between: NET formation and the expression of NKG2D on NK cells (e) (n = 20 patients). f–h Pearson’s correlation analysis of RNA-seq data of clinical liver metastatic samples, revealing positive correlations between: CXCL1 expression and neutrophil chemotaxis (f), CXCL2 expression and neutrophil chemotaxis (g); and a negative correlation between: NET abundance and the expression of NK activating receptors (h) (GSE50760, GSE49355, GSE19279, GSE40367, GSE81558; n = 76 patients). i, j GSEA revealing enrichment of complement and coagulation cascades (i) and the NET formation pathway (j) in metastatic livers from patients with poor prognosis (GSE159216, n = 171 patients). k Multivariable Cox regression analysis for overall survival showing hazard ratios (HRs; dots) and 95% confidence intervals (CIs; horizontal lines) for clinical variables and NK activating receptor expression (GSE159216, n = 171 patients). l Schematic summary of the major discoveries in this study: macrophage phagocytosis of tumor debris induces CXCL1/2 and CFB production, promoting neutrophil recruitment and NETosis; NET-DNA engages CCDC25 on NK cells, enhances CCDC25–ILK interaction and STAT3 Y705 phosphorylation, downregulates NK activating receptors and impairs NK cell-mediated tumor surveillance, thereby accelerating hepatic metastasis. Schematic (l) is Created in BioRender. Z, S. (2026) https://BioRender.com/f59c222. Statistical analysis was performed using two-sided Pearson’s correlation analysis (b–h) or two-sided GSEA-based permutation test with FDR adjustments (i, j). r, Pearson’s correlation coefficient. Source data are provided as a Source Data file.

Furthermore, we externally validated our findings using independent datasets. By integrating bulk RNA-seq data from liver metastases across five independent cohorts (GSE50760, GSE49355, GSE19279, GSE40367 and GSE81558), Pearson’s correlation analyses revealed positive associations between CXCL1/2 expression and neutrophil chemotaxis signatures (Fig. 8f, g), as well as inverse correlations between NET abundance levels and NK-cell activating receptor expression (Fig. 8h), and NET abundance levels and KLRK1 expression (Supplementary Fig. 8c).

We next assessed the prognostic value of our mechanistic findings by analyzing an independent bulk RNA-seq dataset of liver metastases with matched overall survival information (GSE159216). Patients were stratified into poor- (overall survival <60 months) and favorable-prognosis (overall survival ≥ 60 months) groups. Gene set enrichment analysis revealed that pathways related to the complement and coagulation cascades, as well as NET formation, were significantly enriched and upregulated in the poor-prognosis group (Fig. 8i, j). We next performed multivariable Cox regression analysis incorporating key clinicopathological variables. Male sex, presence of extrahepatic metastasis, KRAS mutation, and positive liver margin status emerged as significant risk factors for overall survival (Fig. 8k). Notably, the NK-cell activating receptor expression score remained an independent protective factor after adjustment for these variables (Fig. 8k). Together, these data indicate that activation of NET-related pathways is associated with unfavorable prognosis in patients with liver metastases, whereas preserved NK-cell activating receptor expression confers a significant survival advantage.

Collectively, these clinical data support the proposed interplay among CXCL1/2, CFB, NETs, and NK-cell function and highlight the expression of NK-cell activating receptors as a potential prognostic biomarker in liver metastasis.

Discussion

Liver metastasis portends poor prognosis across multiple cancer types², highlighting the urgent need for more effective therapeutic strategies. A deeper understanding of the molecular events that orchestrate the formation of a permissive hepatic metastatic niche is critical for developing effective interventions. While prior studies have examined the role of immune cells in the liver metastatic niche from various perspectives^4,6, few have investigated the temporal evolution of the immune microenvironment throughout the process of liver metastasis. In this study, we employed a longitudinal scRNA-seq approach in an orthotopic murine breast cancer model to map the immune landscape of the liver metastatic niche during the course of hepatic metastasis. Our findings reveal a progressive infiltration of neutrophils and the formation of NETs in the metastatic liver, corroborating with prior observations^9,15. Importantly, we elucidate the molecular mechanisms underlying neutrophil recruitment and NETosis, and uncover a pathway by which NET-DNA impairs NK cell cytotoxic function in the metastatic liver.

Progressive tumor growth is accompanied by continuous generation of tumor debris as a result of hypoxia, nutrient starvation and cellular stress in the tumor microenvironment^32,35. As circulating tumor cells (CTCs) disseminate, they are exposed to substantial mechanical and biological stresses, including shear stress, endothelial interactions, and immune surveillance^33,53,54. These hostile conditions can induce tumor cell damage or death, leading to the release of tumor debris into the local microenvironment. In addition to spontaneous debris generation during tumor progression, cytotoxic therapies—including chemotherapy and radiotherapy—are also major inducers of tumor debris³³. A recent study reports that macrophages in the tumor-draining lymph node phagocytose chemotherapy-induced tumor debris and subsequently secrete IL-33, which activates regulatory T (T_reg) cells, contributing to therapeutic resistance⁵⁵. Although the experimental identification of tumor debris in vivo remains technically challenging, our findings using experimentally generated tumor debris in combination with pharmacological inhibition of phagocytosis support a model in which macrophage uptake of tumor-derived debris induces CXCL1/2 and complement factor B (CFB) expression, thereby promoting neutrophil recruitment and NET formation to remodel the liver metastatic niche. In addition to tumor cell-derived debris, efferocytosis of liver tissue-resident cell debris resulting from metastasis-induced liver injury also reprograms macrophages to promote liver metastasis⁵⁶. Together, these findings highlight debris-driven macrophage activation as a common pro-tumor mechanism across different contexts, positioning phagocytosis-targeted therapies as a promising strategy to restrain tumor progression. Notably, several inhibitors of macrophage phagocytosis, including MER receptor tyrosine kinase (MERTK) inhibitors such as MRX-2843, ONO-7475, and ARRY-067, have entered clinical trials and shown good efficacy in multiple cancer types⁵⁷. Despite that inhibition of macrophage phagocytosis showed great translational value, the mechanisms through which cell debris reprograms macrophages remain unclear.

Regarding the mechanisms driving neutrophil recruitment and NETosis, prior studies have implicated a range of mediators—including cytokines, chemokines, damage-associated molecular patterns (DAMPs), complement anaphylatoxin C3a and C5a, and urate crystals—in orchestrating this process²¹. Given that the liver is the primary site of complement protein synthesis⁵⁸, NET formation is expected to be particularly prominent in hepatic tissues during tumor progression, as supported by our previous findings⁹. In our study, we demonstrate that liver macrophage-derived CXCL1/2 and CFB are the key mediators that drive neutrophil infiltration and NETosis within the liver metastatic niche. The CXCL1/2-CXCR2 axis has been widely implicated in tumor promotion, primarily through the recruitment of myeloid-derived suppressor cells and neutrophils²³. In addition, CXCR2 has also been reported to be expressed in various tumor types, where it may contribute to tumor cell proliferation and migration^24,25. By combining implantation of CXCR2-deficient tumor cells with systemic CXCR2 blockade, our study demonstrates that tumor cell-intrinsic CXCR2 signaling contributes only modestly to hepatic metastatic progression relative to host immune-mediated CXCR2 signaling. However, future studies employing neutrophil-specific CXCR2 deletion models will be important to more precisely define the relative contribution of neutrophil-intrinsic CXCR2 signaling to liver metastasis and metastatic niche formation. Over the past decade, numerous CXCR2 antagonists, including inhibitors such as SB656933, SCH527123 and AZD5069, have entered clinical trials targeting several inflammatory diseases, including cancer²³. Concurrently, the CFB-mediated alternative complement pathway has been well characterized in renal diseases, with the CFB inhibitor iptacopan (Novartis) showing considerable promise in the treatment of C3 glomerulopathy and IgA nephropathy^59,60. Our findings show that genetic or pharmacological inhibition of CXCR2 or CFB markedly reduces liver metastatic burden, providing a mechanistic rationale for extending CXCR2 or CFB blockade to the prevention of liver metastasis.

In addition to their well-recognized roles in tumor promotion^9–12, NETs also exert profound immune-regulatory effects by directly inducing T cell exhaustion⁶¹, physically obstructing the contact between cytotoxic T cells or NK cells and tumor cells¹³, and promoting T_reg cell expansion via innate-like B cell¹⁴. Here, we uncover a function of NETs in directly impairing NK-cell cytotoxicity. Notably, we show that this effect is dependent on CCDC25, a transmembrane NET-DNA sensor identified in our prior studies^9,19,40. Whereas extracellular DNA is widely known to activate cytoplasmic DNA-sensing cGAS-STING pathway upon uptake—thereby promoting type I interferon expression and enhancing immune cell function^17,62—our findings demonstrate a distinct paradigm in which NET-derived DNA engages CCDC25 on NK cell membrane, triggering integrin-linked kinase (ILK)-dependent STAT3 phosphorylation that subsequently suppresses activating receptor expression. This mechanism positions CCDC25 as a potential innate immune checkpoint and provides a conceptual framework for understanding how extracellular DNA mediates immune suppression within the metastatic microenvironment. Although CCDC25 is detectable across multiple immune cell populations, higher expression levels together with the spatial proximity of NK cells to NET-rich regions may render NK cells particularly susceptible to CCDC25-mediated signaling in the metastatic liver microenvironment. Notably, NK cell function in metastatic liver lesions is regulated by diverse cellular and molecular components of the hepatic niche. For example, activated hepatic stellate cells have been reported to induce NK cell quiescence through CXCL12-mediated signaling, thereby suppressing immune surveillance⁴. To exclude an indirect effect mediated by other CCDC25-expressing cell populations on NK cell dynamics, we performed reciprocal in vivo adoptive transfer experiments, which provided evidence that CCDC25-dependent regulation of NK cell function is predominantly NK cell-intrinsic. Nevertheless, whether the NET-CCDC25 axis also affects other immune cell populations under different pathological contexts remains an important question that warrants further investigation.

Aberrant activation of STAT3 is a well-established driver of tumor progression⁶³. Beyond its canonical roles in promoting tumor cell proliferation, accumulating evidence indicates that tyrosine-phosphorylated STAT3 acts as a central regulator of NK cell–mediated immunosurveillance^46,64. STAT3 has been identified as a central intrinsic node that can reprogram NK-cell receptor and effector programs in a context-dependent manner⁶⁴. Loss of STAT3 could upregulate DNAM-1 and lytic enzymes expression in NK cells in the models of hematological diseases, while STAT3-dependent modulation of activating receptors such as NKG2D and NCRs varies with upstream cues⁶⁴. Furthermore, tumor-derived IL-6–STAT3 signaling induces UBE2S-mediated ubiquitination and degradation of NKp30, providing mechanistic evidence that STAT3 activation can directly downregulate activating receptors and impair NK cell function⁶⁵. In our study, we demonstrate that NETs in the metastatic liver impair NK cell cytotoxicity through STAT3 phosphorylation at the Y705 residue, leading to the broad downregulation of NK cell activating receptors. Pharmacological inhibition of STAT3 restored NK cell function and reduced liver metastasis in vivo. Nevertheless, given the pleiotropic roles of STAT3 signaling across multiple immune and non-immune cell types within the tumor microenvironment⁴⁶, part of the therapeutic effect observed following systemic STAT3 blockade may involve indirect mechanisms beyond NK cells. Therefore, future studies employing NK cell-specific genetic approaches will be important to more precisely define the NK cell-intrinsic role of STAT3 signaling in regulating the metastatic niche. Mechanistically, we further identify ILK as a key adapter linking CCDC25 activation to downstream STAT3 phosphorylation, thereby delineating a NET-CCDC25-ILK-STAT3 signaling axis underlying NK cell dysfunction in the liver metastatic niche and implicating STAT3 inhibition as a potential approach to mitigate liver metastasis. Moreover, our findings suggest that phosphorylated STAT3 (pSTAT3) functions as a transcriptional repressor coordinating the concerted suppression of multiple activating receptors. The precise molecular mechanisms by which pSTAT3 mediates this suppressive program remain to be further elucidated.

Anti-cancer interventions targeting NETs have drawn considerable attention in recent years. Agents such as Disulfiram, Sivelestat, and recombinant human DNase (rhDNase) have entered clinical trials and shown promising potential for cancer treatment⁶⁶. However, given the multifaceted roles of NETs in anti-microbial defenses⁶⁷ and innate immune activation^17,62, many of these trials have encountered challenges related to systemic toxicity and limited efficacy. Our prior work identified the NET–CCDC25 axis as a key driver of tumor-cell chemotaxis⁹ and doxorubicin-induced cardiotoxicity⁴⁰, and demonstrated that CCDC25 blockade enhances chemotherapy efficacy while protecting against cardiac injury⁴⁰. We further developed a membrane-anchored CCDC25 inhibitor, di-Pal-MTO, which suppresses liver metastasis while preserving NET-DNA–mediated dendritic cell maturation¹⁹. In this study, we show that CCDC25 inhibition also reverses NET-induced NK-cell dysfunction, providing a distinct mechanistic rationale for targeting CCDC25 to disrupt metastatic niche formation. Together, these findings highlight CCDC25 as a more precise and feasible therapeutic target than NETs themselves. Consistently, accumulating preclinical evidence indicates that targeting the NET-CCDC25 axis using nanoparticle- or liposome-based strategies effectively reduces liver metastasis^37,68,69, providing a compelling rationale for further development of CCDC25-targeted therapies. Nevertheless, therapeutic translation will require careful consideration of potential immune-related effects. Given that CCDC25 is expressed abundantly in NK cells, therapeutic targeting of CCDC25 may exert immune-related adverse effects. Although our study demonstrated no overt histopathological abnormalities in Ccdc25^-/- mice under steady-state conditions, genetic deletion may not fully recapitulate the impact of systemic pharmacological inhibition. Despite that, long-term safety evaluation and comprehensive immune profiling of CCDC25-targeted therapy, particularly under various inflammatory conditions, will be essential for future translational development.

Collectively, our study dynamically profiles the hepatic immune remodeling during liver metastasis and identifies a mechanism underlying the formation of a liver pro-metastatic niche. We demonstrate that tumor debris reprograms liver macrophages to upregulate CXCL1, CXCL2, and CFB, which in turn promotes neutrophil recruitment and NETosis. NET-DNA next binds to CCDC25 on the NK cell membrane, activating the ILK–STAT3 signaling axis and impairing NK cytotoxicity, ultimately facilitating metastatic outgrowth (Fig. 8l). These findings elucidate the key immunological events orchestrating the formation of liver metastatic niche and uncover an innate immune checkpoint through which extracellular DNA suppresses NK-cell function. More importantly, our results establish CCDC25 as a promising therapeutic target for restraining hepatic metastasis. Nonetheless, given the preclinical nature of this study, further investigation is required to validate the efficacy and safety of CCDC25-targeted strategies in cancer patients.

Methods

Study design

The experiments in this study were designed to examine the underlying mechanism of the pro-metastatic niche formation in the metastatic liver. Primary breast tumor samples (32 cases) and liver metastasis samples (20 cases) from female patients who were diagnosed with invasive breast carcinoma, and non-metastatic liver samples (6 cases) from patients who were diagnosed with benign liver diseases were collected for immunofluorescent staining and analysis. Human mononuclear cells were obtained from healthy donors for in vitro assays, including flow cytometry and western blots. All samples were collected from patients with informed consent, and all related procedures were performed with the approval of the internal review and ethics boards of the Sun Yat-Sen Memorial Hospital under approval number SYSKY-2025-308-01.

We investigated the immunoediting pattern in the metastatic liver in vivo by performing orthotopic or intrasplenic inoculation of murine cancer cells to female BALB/c or C57BL/6J mice. All procedures that involved animals were approved by the Institutional Animal Care and Use Committee (IACUC) of Sun Yat-sen University (SYSU-IACUC-2022-000540) and the Experimental Animal Ethics Committee of Ruiye Model Animal (Guangzhou) Biotechnology Co., Ltd. (RYEth-20250220641). For the orthotopic breast tumor model, primary tumor size was measured with calipers, and tumor volume was calculated using the formula: volume = length × width² / 2. Mice were euthanized at the experimental endpoints or when humane endpoints were reached, including tumor volume exceeding 1500 mm³, maximum tumor diameter exceeding 15 mm, tumor ulceration, impaired mobility, severe distress, poor body condition, body-weight loss exceeding 15%, abdominal distension, or inability to access food or water. These approved limits were not exceeded in any animal experiment performed in this study. In all experiments, mice of similar age and size were used across all groups, and they were randomly assigned to each group. The investigators were unblinded from the group allocation. Experimental data were repeated in at least three independent experiments. No data were excluded from analysis. The details of study design, sample size (which was determined according to previous publications and experimental experience), experimental replicates, and statistics are given in the corresponding figures and figure legends.

Statistical analysis

All statistical analyses were performed using GraphPad Prism version 9 (GraphPad Software, Inc.). Data are shown as mean ± SD or mean ± SEM. Indicated sample sizes shown in figure legends refer to biological replicates (eg, individual patients/animals/in vitro cell experiments). All data, except for the scRNA-seq data, were subjected to statistical analysis based on the following criteria. For comparisons between two groups, normally distributed data were analyzed using an unpaired Student’s t test when variances were equal, or a Welch’s t test when variances were unequal. For comparisons among multiple groups, a one-way ANOVA (for one independent variable) followed by Tukey’s or Dunnett’s post hoc test, or a two-way ANOVA (for two independent variables) followed by Tukey’s post hoc test, was applied to normally distributed data with equal variances. In cases of unequal variances, Welch’s ANOVA followed by Dunnett’s T3 post hoc test was used. When the population was not normally distributed, comparisons were conducted using the Mann-Whitney U test for two groups, or the Kruskal-Wallis test for multiple groups. The statistical methods used for analyzing the scRNA-seq data are described below. A P value less than 0.05 was considered statistically significant. Exact P values are reported in the corresponding figure panels.

Mice experiments

Female BALB/c and C57BL/6J mice of 6–12 weeks old were purchased from GemPharmatech and used in all experiments. The Cfb^-/-, Cxcr2^-/-, Padi4^-/- and Ccdc25^-/- mice in C57BL/6J genetic background were generated by Shanghai Model Organisms Center. All mice were maintained and experiments were conducted in specific pathogen free (SPF) environment according to a protocol approved by the Institutional Animal Care and Use Committee (IACUC) of Sun Yat-sen University and Ruiye Bio-tech Guangzhou Co., Ltd. Laboratory animal facility has been accredited by AAALAC (Association for Assessment and Accreditation of Laboratory Animal Care International) and the IACUC (Institutional Animal Care and Use Committee) of Guangdong Laboratory Animal.

Both the orthotopic and intrasplenic injection models were performed as previously described⁹. For the orthotopic model, 2 × 10⁵ 4T1-luc tumor cells were implanted into the fourth fat pads of BALB/c mice. For the intrasplenic injection model, mice were anesthetized with isoflurane (3% for induction and 1% for maintenance), and then immobilized on a warm pad. After sterilizing the surgery area, a 5 mm incision was made in the left side of the abdomen to expose the spleen. Later, 5 × 10⁵ 4T1, 1 × 10⁶ EO771 or 5 × 10⁵ MC38 tumor cells resuspended in 50 μl phosphate-buffered saline (PBS) were injected to the spleen with a 31 G insulin syringe. To measure the metastatic burden of mouse livers, the area covered by metastatic foci was calculated with ImageJ software or HALO.

For CXCL1/2 blockade, CXCL1 (Clone 48415, Invitrogen, 5 mg/kg), CXCL2 (Clone 40605, Invitrogen, 5 mg/kg) neutralizing antibody or isotype control were administered to mice intraperitoneally (i.p.) daily. For CXCR2 inhibition, CXCR2 antagonist SB225002 (Cat# HY-16711, MCE) was administered i.p. at the dosage of 10 mg/kg daily. As a control, 2% dimethyl sulphoxide (DMSO) dissolved in sterile PBS (vehicle) was administered. For phagocytosis inhibition, phagocytosis inhibitor CytoB (Cat# HY-16928, MCE) was administered to mice i.p. at the dosage of 10 mg/kg 2 h before intrasplenic injection of 4T1 cells and was administered daily. For macrophage depletion in vivo, mice were treated with 200 μL/20 g PBS- or Clodronate-containing liposomes (Cat# CP-005-005, Liposoma) i.p. weekly. For C3aR and C5aR inhibition in vivo, C3aR inhibitor SB290157 (Cat# HY-101502A, MCE, 5 mg/kg) and C5aR neutralizing antibodies (Cat# 135815, BioLegend, 5 mg/kg) were administered i.p. to mice daily.

For tumor cell debris treatment, 100 μl of debris made from 5 × 10⁵ 4T1 cells were injected into the spleens of BALB/c mice, and the livers of mice were collected on day 3 post-injection. For exosome education, mice were injected with 5 μg purified 4T1 cell-derived exosomes retro-orbitally every other day for 2 weeks⁵, followed by euthanasia to harvest livers or continuation with intrasplenic inoculation of 4T1 cells. For T cell and NK cell depletion, 5 mg/kg anti-CD3 (αCD3; clone 17A2, BioXcell) neutralizing antibody or 5 mg/kg anti-NK1.1 (αNK1.1; clone PK136, BioLegend) neutralizing antibody, respectively, was administered i.p. to mice one day prior to intrasplenic inoculation of EO771-luc cells, and every 5 days thereafter. For NK cell depletion in the NK adoptive transfer model, mice received a single intraperitoneal injection of 5 mg/kg anti-NK1.1 antibody 5 days prior to NK cell adoptive transfer, with no further antibody administration thereafter. For CCDC25 blockade, 5 mg/kg monoclonal, neutralizing anti-CCDC25 (αCCDC25) antibody (Sobour, Guangzhou, Biopharmaceutical Co., Ltd.) or isotype-IgG was administered to mice every other day. For application of Napabucasin, 20 mg/kg Napabucasin (Cat# S7977, Selleck) was administered to mice i.p. every other day.

Patients and tissue samples

Paraffin-embedded primary breast tumor samples (32 cases) and metastatic liver samples (20 cases) for immunofluorescence analysis were obtained from treatment-naïve female patients who were all diagnosed with invasive breast carcinoma, and non-metastatic liver samples (6 cases) were obtained from patients who were diagnosed with benign liver diseases between 2008 and 2020 and enrolled into the Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University (Guangzhou, China). After analysis, any remaining tissue samples were stored at Sun Yat-sen Memorial Hospital under the approved institutional protocol.

Cell lines

Murine breast cancer cell line 4T1, colorectal cancer cell line MC38 and human immortalized myelogenous leukemia cell line K562 were purchased from American Type Culture Collection (ATCC), and EO771 was purchased from CH3 Biosystems (New York, USA). All the cells were tested negative for mycoplasma and were cultured in DMEM medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin.

Immunofluorescence

Paraffin-embedded samples sectioned at 4 μm thickness were deparaffinized and rehydrated. Tris-EDTA buffer (pH 8.0 or 9.0, Asegene) was used for heat-mediated antigen-retrieval for 15 min in a pressure cooker. Then, the samples were blocked with 5% BSA solution for 20 min at room temperature to reduce non-specific binding. Subsequently, samples were incubated with goat anti-MPO (1:75, Cat# AF3667, R&D), rabbit anti-H3cit (1:100, Cat# Ab5103, Abcam), rat anti-F4/80 (1:200, Cat# Sc-52664, Santa Cruz), rabbit anti-CXCL1 (1:200, Cat# AF5403, Affinity), mouse anti-CXCL1 (1:200, Cat# Ab89318, Abcam), rabbit anti-CXCL2 (1:200, Cat# P19875, Bioss), rabbit anti-CFB (1:100, Cat# DF6567, Affinity), mouse anti-CD68 (1:200, Cat# Ab201340, Abcam), mouse anti-NK1.1 (1:50, Cat# 108760, BioLegend), rabbit anti-STAT3 (phosphor Y705) (1:200, Cat# EP2147Y, Abcam), mouse anti-CD56 (1:200, Cat# 60238-1-Ig, Proteintech), rabbit anti-CD3 (1:200, Cat# HA720082, Huabio), rabbit anti-NKG2D (1:400, Cat# Ab319162, Abcam), rabbit anti-Luciferase (1:200, Cat# ET1602-44, HUABIO) overnight at 4 °C. After washing three times with PBS, the slides were incubated with fluorescence-conjugated secondary antibodies (1:200, Cat# A21206 or A21432, Thermo Fisher) for 1 h at RT, and DAPI solution (Asegene) was then applied for counterstaining. Multiplex immunofluorescence was performed using a tyramide signal amplification (TSA) kit (Cat# AFIHC037, Aifang Biological, China) according to the manufacturer’s instructions. Images were obtained by laser scanning confocal microscopy (LSM 800, Zeiss) or digital pathology slide scanner (KF-FL-400, KFBIO) and analyzed with ImageJ or HALO image analysis platform (Indica Labs). For quantification of the colocalized area of MPO and H3cit, “Area Quantification” module in HALO was applied to identify the overlapping signals between MPO and H3cit, and the colocalized area was calculated as the percentage of the MPO⁺ H3cit⁺ area relative to the total area of the regions of interest (ROIs). For quantification of the multiplexed immune cells, the “Multiplex” module was applied according to the manufacturer’s instructions.

For immunofluorescence staining of cells cultured in vitro, cells growing on the chamber slides (Thermo Scientific™ Nunc™ Lab-Tek™ II Chamber Slide™) were fixed with 4% paraformaldehyde for 15 min at RT, washed in PBS and permeabilized with 0.1% Triton X-100 in PBS for 5 min. 5% BSA was used to block the non-specific site for 15 min at RT. Afterwards, cells were incubated with primary antibodies against goat anti-MPO (1:75, Cat# AF3667, R&D) and rabbit anti-H3cit (1:100, Cat# Ab5103, Abcam) overnight at 4 °C, followed by incubation with Alexa Fluor secondary antibodies (1:200, Cat# A21206 or A21432, Thermo Fisher) for 1 hour at RT. Then, cells were counterstained with DAPI and images were acquired with laser scanning confocal microscopy (LSM 800, Zeiss) and analyzed with ImageJ or HALO.

Primary human immune cell isolation

Human mononuclear cells were obtained as previously described⁷⁰. Briefly, peripheral blood samples were diluted with PBS of equivalent volume. Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque PLUS (Cat# LTS1077, TBDscience) by centrifugation (37 °C, 800 × g, 30 min, Beckman) with a brake off. NK cells were obtained using CD56 microbeads (Cat# 130-050-401, Miltenyi Biotec) according to the manufacturer’s protocol and were cultured in RPMI-1640 medium (GIBCO) containing 10% FBS, 4 mmol/L L-glutamine, 25 mmol/L HEPES, 25 μmol/L 2-mercaptoethanol, 100 U/ml IL-2 (Cat# 200-02, PeproTech) and 50 U/ml IL-15 (Cat# 200-15, PeproTech) for future use.

Isolation of immune cells from liver

Liver immune cells were isolated through the two-step perfusion as previously described⁷¹. In brief, livers were exposed and the initial perfusion was performed in situ through the inferior vena cava with 15 ml Hanks’ balanced salt solution without Ca²⁺ or Mg²⁺ (D-Hanks) containing EDTA at the speed of 5 ml/min, followed by the second perfusion with 10 ml Hanks’ balanced salt solution (HBSS) containing 25 μg/ml of Liberase^TM (Cat# 05401127001, Sigma) at the same speed. The livers were then excised, rinsed with cold HBSS. Next, the liver sack was ruptured with scissors and single cells were released using a cell lifter. Then, the cell suspension was filtrated through a 70-μm cell strainer. To remove hepatocytes, the cell suspension was centrifuged at 50 × g for 2 min at 4 °C. The supernatant was isolated and centrifuged again at 350 × g for 5 min at 4 °C to acquire immune cells. Red blood cells were lysed with red blood cell lysis buffer (Biolegend) on ice for 2 min and after washing with HBSS, immune cells from the liver were obtained.

Flow cytometry

The isolated liver immune cells from mice were first incubated with anti-CD16/CD32 antibody (1:50, Clone 93, BioLegend) to reduce FcR-mediated non-specific binding. Following standard procedures, murine or human samples were stained with fluorochrome-conjugated antibodies and washed twice before assessment on a four-laser CytoFLEX S Flow Cytometer (Beckman Coulter). Dead cells were excluded using Fixable Viability Dye eFluor™ 780 (1:1000, Cat# 65-0865-14, Thermo Fisher). The antibodies used are listed as follows (all antibodies are used at a 1:20 dilution and purchased from BioLegend unless otherwise indicated): CD45-APC (30-F11), F4/80-eFluor450 (BM8), Ly6G-PE610 (1A8), CD11b-BV605 (M1/70), CXCL1 (Cat# AF5403, Affinity), CXCL2 (Cat# P19875, Bioss), CFB (Cat# DF6567, Affinity), CD3-FITC (17A2), NK1.1-BV421 (PK136), NKG2D-PE (CX5), NKp46-PE (29A1.4), 2B4-PE (m2B4 (B6)458.1), DNAM-1-PE (10E5), CD107a-PE (1D4B), CD3-APC/Cy7 (UCHT1), CD56-APC (HCD56), p^Y705-STAT3-AF488 (13A3-1), NKG2D-FITC (1D11), NKp46-PE (9E2), NKp44-PE (P44-8), DNAM-1-FITC (11A8), CD107a-FITC (H4A3), Perforin-APC (dG9, eBioscience), CCDC25(Cat# 84484-3-RR, Proteintech), Tim4-PE (RMT4-54), CD45-FITC (30-F11), CD11b-APC (M1/70), F4/80-BV421 (BM8). For CCDC25 staining, the anti-CCDC25 antibody was first labeled with Alexa Fluor 488 using the Alexa Fluor 488 Conjugation Kit (ab236553, Abcam) according to the manufacturer’s instructions. Before staining of intracellular cytokines, cells were fixed and permeabilized with fixation and permeabilization buffer (BioLegend) as manufacturer’s instructions after surface markers staining. For staining of non-conjugated CXCL1, CXCL2 and CFB antibodies, cell pellets were first incubated with corresponding primary antibodies at room temperature for 1 h, followed by staining with Alexa Fluor-488-conjugated or Alexa Fluor-555-conjugated secondary antibodies (1:20, Cat# A21206 or A21432, Thermo Fisher) at 4 °C for 45 mins. For staining of CD107a, cells were treated for 4 hours with 2 μM monensin (Cat# 420701, BioLegend), followed by surface and intracellular antigen staining.

Ligands for NK activating receptors on tumor cells are detected as previously described⁷². For K562 cells, 2B4-Fc (Cat# 1039-2B), NKp44-Fc (Cat# 2249-NK), NKp46-Fc (Cat# 1850-NK), NKG2D-Fc (Cat# 1299-NK), DNAM-1-Fc (Cat# 666-DN), and corresponding isotype controls (all from R&D) were biotinylated (ab201796, Abcam) according to the manufacturer’s instructions. Cells were incubated with biotinylated Fc chimera proteins at 5 μg/ml for 30 min at room temperature, followed by staining with PE-Streptavidin (1 μg/ml, Cat# 405203, BioLegend). For murine tumor cells, expressions of ligands for NK activating receptors were detected using 2B4-his (Cat# 58100-M08H, SinoBiological), NKp46-his (Cat# HY-P78331, MCE), NKG2D-his (Cat# HY-P72503, MCE), DNAM-1-his (Cat# HY-P72669, MCE) and corresponding isotype controls at 5 μg/ml for 30 min at room temperature, followed by staining with FITC-anti-His-Tag (1:20, Cat# 362618, BioLegend). All data were analyzed using FlowJo V10 software (Tree Star, Ashland, OR).

Two-photon intravital hepatic imaging and image analysis

4 hours or 24 hours before imaging with two-photon microscopy, an intrasplenic metastatic model was established with 5 × 10⁵ mCherry⁺ 4T1 cells. Immediately prior to two-photon imaging, FITC-anti-F4/80 antibody (10 μg per mouse, Clone BM8, BioLegend), AF488-anti-CD31 (10 μg per mouse, Clone 390, BioLegend) or Hoechst (Invitrogen) was injected into mice via tail vein in a total volume of 100 μl PBS to visualize macrophages, liver sinusoids and cell nuclei, respectively. Mice were anesthetized with isoflurane (3% for induction and 1% for maintenance), and a subcostal incision was made through the dermis and peritoneum. The falciform ligament was resected, and the liver was gently exposed and attached to a customized steel imaging platform. During the whole imaging process, the liver was covered with a moist gauze to avoid dehydration. Images were acquired with an Olympus FVMPE-RS two-photon microscope equipped with the XLPlan 25 × water immersion lens (NA 1.05, Olympus). Fluorescence excitation was provided by two lasers (Insight DS-OL and MaiTai HP-OL, SpectraPhysics) tuned to optimal excitation wavelength for FITC/AF488, mCherry and Hoechst.

Quantitative reverse transcription polymerase chain reaction (RT-qPCR)

Total RNA was extracted from homogenized mouse liver tissues or liver macrophages using TRIzol reagent (Cat# A33251, Thermo Fisher). Briefly, 500 ng of total RNA was reversely transcribed into cDNA using PrimeScript RT Master Mix kit (Cat# RR036A, TaKaRa). Quantitative reverse transcription polymerase chain reaction (RT-qPCR) was conducted with TB Green Premix Ex Taq II (Cat# RR820A, TaKaRa). The primer sequences are provided in Supplementary Table S1. All data were collected and analyzed using a LightCycler 480 instrument (Roche).

Enzyme-linked immunosorbent assay (ELISA)

Liver tissue supernatant was prepared as previously described⁷³. Briefly, liver tissues were plated in 1 ml DMEM medium containing 10% FBS with 1% penicillin-streptomycin for 24 h. Then the supernatant was harvested after centrifugation at 2000 × g, 4 °C for 5 min. All samples were immediately cryopreserved at − 80 °C until use (within 2 months), and frequent freezing and thawing were avoided. All samples were diluted to a suitable concentration according to the different detection ranges of the ELISA kits. For the detection of CXCL1, CXCL2, C3a, C5a and CFB in supernatant, ELISA kits of CXCL1 (Cat# E-EL-M0018, Elabscience), CXCL2 (Cat# E-EL-M0019, Elabscience), C3a (Cat# E-EL-M0337, Elabscience), C5a (Cat# E-EL-M0339, Elabscience) and CFB (Cat# E-EL-M0334, Elabscience) were used, and the assay was carried out following the manufacturer’s instructions.

Liver macrophage isolation and in vitro phagocytosis assay

The liver immune cells from BALB/c mice were isolated as described above. After the red blood cells were lysed, macrophages were enriched by positive selection for F4/80⁺ cells using microbeads (Cat# 130-110-443, Miltenyi Biotec) according to the manufacturer’s instructions. Then the isolated F4/80⁺ cells were cultured in 24-well plates with DMEM supplemented with 20% FBS and 1% penicillin-streptomycin. The macrophages were cultured for 6-7 days before being used for in vitro assay. 4T1 cell debris was generated by 10 cycles of freezing (− 80 °C) and thawing (37°C water bath) of 1 × 10⁴ 4T1 cells. Macrophages were pre-treated with 30 μg/ml CytoB for 2 h before the addition of 4T1 cell debris. After 6-hour 4T1 debris treatment, the culture medium was collected and centrifuged at 2000 × g, 4 °C for 5 min for subsequent ELISA detection of CXCL1, CXCL2 and CFB, while macrophages were lysed in TRIzol reagent for RT-qPCR analysis.

Isolation of murine NK cells

Murine NK cells were isolated and enriched from the spleens of C57BL/6J mice using EasySep^TM Mouse NK Cell Isolation Kit (Cat# 19855, Stemcell) according to the manufacturer’s instructions. Isolated NK cells were expanded in vitro in MEMα media containing 20% fetal bovine serum, 1% Penicillin-Streptomycin and 150 ng/ml IL-15 (Cat# HY-P700193AF, MCE) for subsequent functional assays. NK cells underwent an approximate 10³−10⁴-fold expansion by day 10-12. For the NK cell adoptive transfer model, approximately 1 × 10⁶ CellTracker^TM Deep Red (Cat# C34565, Thermo Fisher)-labeled splenic NK cells in 100 μl PBS were transferred intrasplenically into the recipient C57BL/6J mice 1 day prior to and 2 h within intrasplenic injection of EO771-luc cells. Then the adoptive transfer was conducted every 5 days for 30 days in total^41,42.

Cytotoxicity assay of NK cells

NK cell–mediated cytotoxicity was evaluated using a tumor cell co-culture assay. Briefly, NK cells were co-cultured with target tumor cells (K562 cells for human NK cells; EO771 or 4T1 cells for murine NK cells) at an E:T ratio of 10:1 in complete RPMI-1640 medium supplemented with 10% fetal bovine serum. For cytotoxicity measurement, tumor cells were labeled by CellTracker^TM Orange or Deep Red (Cat# C2927 or C34565, Thermo Fisher) according to the manufacturer’s instructions. Then, NK cells pretreated as indicated were co-cultured with labeled tumor cells for 4 h at 37 °C in a humidified incubator with 5% CO₂. Tumor cell killing was quantified by flow cytometry by measuring the percentage of dead target cells using Fixable Viability Dye eFluor™ 780 (Thermo Fisher) after gating on tumor cells. Basal tumor cell death in the absence of NK cells was used as a control.

For NK cell degranulation assays, NK cells were first pretreated with αCCDC25 antibody or isotype IgG control for 2 h, followed by exposure to 5 μg/ml NETs or DNase I-digested NETs for 12 h. DNase I-digested NETs were generated by incubating NETs with 50 U/ml DNase I for 30 min at 37 °C. After NET treatment, NK cells were washed 3 times with PBS to remove residual NET components. Then NK cells were subjected to a standardized degranulation assay, in which they were co-cultured with target tumor cells at an E:T ratio of 10:1 in the presence of PE-conjugated anti-CD107a antibody (1:20, Clone 1D4B or H4A3, BioLegend). After 1 h of incubation, 2 μM monensin (BioLegend) was added to prevent intracellular protein transport, and the cells were incubated for an additional 3 h. Cells were then harvested and analyzed by flow cytometry to quantify CD107a surface expression.

Mouse neutrophil isolation and in vitro NETosis assay

Briefly, bone marrow was flushed out from the femurs and tibias of BALB/c mice using sterile DMEM solution. After the cell suspension was centrifuged at 400 × g for 5 min, neutrophils were isolated from the sedimental cell pellets using the mouse bone marrow neutrophil isolation solution kit (Cat# P8550, Solarbio) according to the manufacturer’s instructions. The isolated neutrophils were resuspended in RPMI-1640 medium supplemented with 10% FBS and seeded in 24-well plates. Metastatic liver supernatant was prepared as mentioned above. For CXCR2, C3aR and C5aR blockade, CXCR2 neutralizing antibody (10 μg/ml, Cat# MAB2164, R&D), C3aR antagonist SB290157 (20 μM, Cat# HY-101502A, MCE) or C5aR neutralizing antibody (5 μg/ml, Cat# 135815, BioLegend) was applied to neutrophils 1 h before liver supernatant was added. After stimulation, the culture medium was removed, and the adherent NETs or neutrophils were prepared for immunofluorescence staining as described above.

Purification of NETs

NETs were purified as previously described⁹. Briefly, PBMCs were collected from human peripheral blood as mentioned above. After red blood cell lysis of the sedimental layer, neutrophils were collected and seeded in 10-cm plates. The neutrophils were treated with 500 nM PMA for 4–6 h, then the NETs adhered at the bottom were obtained by pipetting 1 ml cold PBS and centrifugation at 1000 × g, 4 °C, for 10 min. The concentration of the DNA component of NETs was assayed by NanoDrop One (Thermo Fisher).

sh/siRNA-mediated silencing and CRISPR-mediated gene knockout

Lentiviruses expressing shRNA were synthesized by GenePharma Inc (Shanghai, China), and the transfection and screening procedures were performed following the manufacturer’s instructions. The shRNA target sequences are listed as follows:

mouse Rab27a, 5´-GCTGCCAATGGGACAAACATA-3´;

mouse Rab27b, 5´-CCCAAATTCATCACTACAGTA-3´;

siRNAs targeting human STAT3 were synthesized by Beijing Tsingke Biotech Co., Ltd. Transduction to NK cells was performed by electroporation as described below. The sense strand sequences of the siRNAs were as follows:

human STAT3 siRNA-1, 5´-UGAUUCUUCGUAGAUUGUG (dT)(dT) −3´;

human STAT3 siRNA-2, 5´- GUCAUUAGCAGAAUCUCAA (dT)(dT) −3´;

human STAT3 siRNA-3, 5´- CAACAAUCCCAAGAAUGUA (dT)(dT) −3´;

The Cas9 lentivirus and gRNA lentivirus for mouse Ccdc25 knockout were synthesized by GenePharma and transduced to EO771 tumor cells. The Cas9 lentivirus and gRNA lentivirus for mouse Cxcr2 knockout were synthesized by Beijing Tsingke Biotech Co. and transduced to 4T1 tumor cells. The Cas9 lentivirus and gRNA lentivirus for human ILK knockout were synthesized by Beijing Tsingke Biotech Co., Ltd. and transduced to PBMC-derived NK cells through electroporation. The gRNA target sequences are listed as follows:

mouse Ccdc25 gRNA, 5´-GAATGCTATTGGCCTTCACA-3´;

mouse Cxcr2 gRNA, 5´-GCGCCGCGATGACATTGACA-3´;

human ILK gRNA1, 5´-CTTGCACTGGGCCTGCCGAG-3´;

human ILK gRNA2, 5´-GCAGGGGGGTGTCATCCCCA-3´.

For transfection, tumor cells were plated in 6-well plates at 5 × 10⁵ per well. The next day, lentiviral vectors were added in the presence of 8 μg/ml polybrene (Cat# TR-1003-G, Sigma-Aldrich). The culture medium was replaced after adding lentiviruses for 24 h. 48 h later, medium with 5 μg/ml puromycin (Cat# A1113803, Gibco) was applied to screen for the successfully transfected cells, and the screening lasted for at least two weeks.

For electroporation of NK cells, PBMC-derived NK cells were pelleted and resuspended in Opti-MEM I medium (Cat# 31985088, Gibco) in a 0.4-cm electroporation cuvette (Bio-Rad) at 5 × 10⁶ cells/ml. The lentiviruses were added at a multiplicity of infection (MOI) of 30, or the siRNAs were added at 100 nM to the cell suspension, and the mixture was electroporated with Gene Pulser Xcell (Bio-Rad) using one pulse of 250 V and 960 μF. NK cells were recovered immediately in warm whole medium and incubated for 48 h.

Immunoprecipitation

1 × 10⁷ PBMC-derived NK cells were stimulated with NETs (5 μg/ml) for 12 h or left untreated. Cells were subsequently washed with ice-cold PBS and gently resuspended in 100 μl ice-cold IP lysis buffer supplemented with a protease and phosphatase inhibitor cocktail (78446, Thermo Fisher), followed by incubation on ice for 30 min. After centrifugation at 14,000 × g for 25 min, clarified lysates were incubated with either an IgG isotype control antibody (50 μg/ml, Cat# 2729, Cell Signaling Technology), an anti-CCDC25 antibody (50 μg/ml, Cat# 84484-3-RR, Proteintech), or an anti-ILK1 antibody (50 μg/ml, Cat# 3856, Cell Signaling Technology) for 1 h at room temperature. Pre-washed Protein A/G magnetic beads were then added and incubated overnight at 4 °C with gentle rotation. The beads were subsequently washed five times with lysis buffer, and bound proteins were eluted by boiling in 1 × SDS loading buffer at 95 °C for 5 min. Immunoprecipitates were resolved by SDS-PAGE alongside 0.5% total input and 12% pull-down sample per lane, followed by immunoblotting analysis. Non-specific binding was blocked with 5% skim milk, after which membranes were incubated overnight at 4 °C with the indicated primary antibodies. Membranes were then washed and probed with the appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies (1:2000, Cat# SA00001-2, Proteintech). Immunoreactive bands were detected by enhanced chemiluminescence (ECL, Cat# 32209, Thermo Fisher).

Chromatin immunoprecipitation (ChIP) -qPCR

The ChIP assay was performed with a ChIP Assay kit (Beyotime) based on the manufacturer’s instructions. Briefly, PBMC-derived NK cells (5 × 10⁶ per immunoprecipitation) were stimulated with NET-DNA (5 μg/ml) for 2 h or left untreated, then cross-linked with 1% formaldehyde for 10 min at room temperature and quenched with 0.125 M glycine. Cells were lysed in Pierce IP Lysis Buffer (Thermo Fisher) supplemented with protease and phosphatase inhibitors, and nuclei were resuspended in nucleus lysis buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA, 1% SDS). Chromatin was sheared by sonication (Sonics VCX130; 90 W, 8 s on/8 s off, 8 min) to yield 200–500 bp fragments. Pre-cleared chromatin was immunoprecipitated overnight at 4 °C with 5 μg anti-phospho-STAT3 (Tyr705) antibody (ab76315, Abcam) or rabbit IgG (Cat# 2729, Cell Signaling Technology). Following sequential washes and elution, cross-links were reversed at 65 °C overnight, and DNA was purified. Quantitative PCR was performed with SYBR Green Master Mix (95 °C 1 min; 40 cycles of 95 °C 10 s, 59 °C 30 s). Related primer sequences were listed in Supplementary Table S2. Enrichment was expressed as fold change of anti-pSTAT3 over IgG (2^ (Ct_IgG-Ct_pSTAT3)).

Western blot

Proteins were extracted from cells with RIPA buffer (Cat# P0013E, Beyotime) containing protease and phosphatase inhibitor cocktail (Cat# P1045, Beyotime), and then quantified using Pierce™ BCA Protein Assay Kits (Cat# 23225, Thermo Fisher) according to standard protocols. After heated in 1 × loading buffer (Cat# AM8547, Thermo Fisher) at 95 °C for 5 min, equivalent amount of protein was separated on 10% SDS-polyacrylamide gel, and transferred to PVDF membrane. Primary antibodies against Alix (1:1000, Cat# 12422-1-AP, Proteintech), CD63 (1:1000, Cat# AF5117, Affinity), CD81 (1:1000, Cat# DF2306, Affinity), horseradish peroxidase (HRP)-conjugated GAPDH (1:1000, HRP-60004, Proteintech), STAT3 (1:1000, Cat# ab68153, Abcam), p^Y705-STAT3 (1:1000, Cat# Ab76315, Abcam), ILK (1:1000, Cat# DF6141, Affinity), CCDC25(1:1000, Cat# 84484-3-RR, Proteintech) and peroxidase-conjugated anti-mouse or -rabbit secondary antibodies (1:2000, Cat# SA00001-1 or SA00001-2, Proteintech) were used. The antigen-antibody reaction was visualized via enhanced chemiluminescence assay (ECL, Thermo Fisher).

NET-DNA pull down assay

NET-DNA pull-down was performed as previously described⁹, with minor modifications. Briefly, membrane proteins from NK cells were extracted using a membrane protein extraction kit (Cat# P0033, Beyotime). NET-DNA was purified using MicroElute DNA Clean Up Kit (Cat# D6296, OMEGA) and biotinylated using Biotin 3' End DNA Labeling Kit (Cat# 89818, Thermo Fisher) according to the manufacturer’s instructions. For the pull-down assay, 500 ng of biotinylated NET-DNA was incubated with NK cell membrane proteins in 400 μl IP lysis buffer (Cat# 87787, Thermo Fisher) at room temperature for 1 h. The mixture was then incubated with 50 μl Pierce™ streptavidin magnetic beads (Cat# 88816, Thermo Fisher) for an additional 1 h at room temperature. After washing 3 times with IP lysis buffer, the bead-bound proteins were eluted and analyzed by immunoblotting.

Exosome experiments

Exosomes were isolated from 4T1 cancer cell culture medium using Exosome Precipitation Solution (ExoQuick-TC, System Biosciences). After isolation, exosomes were examined by western blots and quantified using a NanoSight NS300 instrument (Malvern Instruments) and NTA analytical software (Malvern Instruments).

Bulk RNA-sequencing and bioinformatics analysis

For RNA extraction, total RNA from cells (EO771 or NK cells) and mouse liver biopsies were extracted with TRIzol reagent kit (Invitrogen, Carlsbad, CA, USA). RNA quality was detected using Agilent 2100 Bioanalyzer. Then, the eukaryotic mRNAs were enriched by Oligo (dT) beads and fragmented and reversely transcribed into cDNA. The purified double-stranded cDNA fragments were end repaired, A base added and ligated to Illumina sequencing adapters. The ligation reaction was purified and PCR amplified. The resulting cDNA library was sequenced using Illumina Novaseq 6000 by Gene Denovo Biotechnology Co. (Guangzhou, China).

Following quality control using fastp (v0.18.0), rRNA mapped reads were removed, and the remaining clean reads were further used in assembly and gene abundance calculation. An index of the reference genome was built, and paired-end clean reads were mapped to the reference genome using HISAT (v2.2.4). The mapped reads of each sample were assembled by using StringTie (v1.3.1) in a reference-based approach. For each transcription region, an FPKM (fragments per kilobase of transcript per million mapped reads) value was calculated to quantify its expression abundance and variations, using RSEM software.

Gene expression data were annotated with official gene symbols. Differentially expressed genes (DEGs) were identified using the “Limma” package in R, with a threshold of fold change (FC) > 2 or < 0.5 and P < 0.05. Pathway enrichment was assessed using the “ReactomePA” package.

Single-cell RNA sequencing (scRNA-seq) and bioinformatics analysis

Sample dissociation

Mouse liver cells were isolated as aforementioned, and CD45⁺ cells were further sorted by microbeads (Cat# 130-052-301, Miltenyi Biotec). Sorted cells were washed twice by sterile PBS and assessed for viability with Trypan blue using a Countess® II automated cell counter (Thermo Fisher). Cells were then resuspended at > 1000 cells / μl with a final viability of > 90 %.

Generation of transcriptomic library and sequencing

A 10 × Genomics analysis system was used to prepare single-cell whole transcriptomes. Reverse transcription and library preparation were performed on a Veriti Thermal Cycler with a 96-Deep Well Reaction Module (Thermo Fisher). Amplified cDNA was purified using SPRIselect beads (Beckman Coulter) and sheared to 250-400 bp. Qualification was performed using Qubit 3.0 Fluorometer. All the libraries were sequenced on an Illumina NovaSeq 6000 platform.

ScRNA-seq Data Processing and Downstream Analysis

ScRNA-seq data was processed using 10 × Genomics CellRanger (v3.0.2) pipeline. Subsequent single cell RNA-seq analysis was performed in R package “Seurat” (v5.0.1). In the phase of quality control for mouse scRNA-seq, cells with low quality (< 500 or > 6000 genes per cell and > 20% mitochondrial genes in the cell) were filtered out. For human scRNA-seq (GSE246662), cells with < 200 or > 7000 genes per cell and > 50% mitochondrial genes were discarded. Data were normalized and scaled within each sample, and canonical correlation analysis was performed to integrate data across samples. Cell population deconvolution was performed by the dimensional reduction and shared nearest neighbor (SNN) modularity clustering algorithm. Differentially Expressed Genes (DEGs) in each cluster across different conditions were identified by using the “FindMarkers” function in Seurat. Within each cluster, DEGs between two groups of cells were identified by using a Wilcoxon rank sum test. Adjusted p-values were calculated based on the Bonferroni correction. Genes with a p_val_adj < 0.05 were considered as DEGs. KEGG and GO analysis was performed by using the R package “ClusterProfiler.” GSEA analysis was performed by using the R package “GseaVis”.

Gene signature scores analysis

For scRNA-seq data, module scores for the gene sets “Neutrophil migration”, “NET abundance”, “Complement induction”, “T cell activation”, “NK cell activation” and “STAT3 pathway” (gene lists provided in Supplementary Table S3) were calculated using the AddModuleScore function from the Seurat package (v5.0.1). In parallel, single-cell GSVA scores were computed for Gene Ontology Biological Process (GO-BP) terms — natural killer cell activation (GO:0030101), T cell activation (GO:0042110), and complement activation (GO:0006956) — using the scGSVA package with default parameters.

For bulk RNA-seq data, GSVA scores for the pathways shown in Fig. 5g and Supplementary Fig. 5e were calculated using the GSVA package (v1.50.5) on log2-transformed normalized expression matrices. The corresponding GO-BP gene sets included natural killer cell activation (GO:0030101), T cell activation (GO:0042110), B cell activation (GO:0042113), myeloid dendritic cell activation (GO:0001773), macrophage activation (GO:0042116), hepatocyte apoptotic process (GO:0097284), endothelial cell apoptotic process (GO:0072577), and hepatic stellate cell activation (GO:0035733). For the analyses in Fig. 8f–h, k, GSVA scores were similarly derived from the gene sets listed in Supplementary Table S4. All gene sets were retrieved from the MSigDB C5 collection via the msigdbr R package.

Of note, the NET abundance gene set differs between the scRNA-seq and bulk RNA-seq analyses, reflecting the distinct technical and biological characteristics of the two platforms. For scRNA-seq, which is prone to dropout for lowly expressed transcripts, the NET abundance signature was restricted to robustly detected, neutrophil-intrinsic genes covering the core functional modules of NET biology, including NADPH oxidase subunits (CYBA, CYBB, NCF1, NCF4), pyroptotic/lytic components (GSDMD, CASP1), and upstream signaling and mitochondrial regulators (MAPK3, BNIP3L). For bulk RNA-seq, the gene set was expanded to additionally include several neutrophil infiltration and NET release genes (ELANE, MPO, PADI4, HMGB1), upstream NETosis signaling components (PI3K, MAPK, and NF-κB family members), and NET–platelet/coagulation interaction markers (GP1BA, SELP, ITGB3, FGA), thereby capturing the combined contribution of neutrophil infiltration, NET release, and NET-mediated downstream interactions at the tissue level. The complete gene lists for both analyses are provided in Supplementary Tables S3, S4.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: L.Y., Y.Z., J.Z. and H.L. Methodology: L.Y., Y.Z. and H.L. Investigation: Y.Z., H.L., J.Z., J.L., Y.L., Z.Z., L.X., Z.L., X.Z., S.D., J.W. and Xinyan Liang. Data analysis: Y.Z., J.Z., H.L., Y.L., Xu Liu, J.F., R.W. and X.Z. Project administration: L.Y. and E.S. Supervision: L.Y. and E.S. Writing – original draft: Y.Z., H.L. and J.Z. Writing – review & editing: Y.Z., H.L., L.Y., J.Z., Y.L. and J.L.

Peer review

Peer review information

Nature Communications thanks Robert Eferl and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the grants from: Scientific Research Innovation Capability Support Project for Young Faculty (SRICSPYF-ZY2025120 to L.Y.), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0519800 to E.S.), National Natural Science Foundation of China (82488101 to E.S., 82222055, 32270971 and 82573214 to L.Y.), National Key Research and Development Program of China (2021YFA1300602 to E.S.), Science and Technology Planning Project of Guangdong Province (2023B1212060013 to E.S.), Guangdong Provincial Clinical Research Center for Breast Diseases (2023B110005 to E.S.), Science and Technology Program of Guangzhou (2024B01J1154 to E.S.), Bureau of Science and Technology of Guangzhou (20212200003 to E.S.), Program for Guangdong Introducing Innovative and Entrepreneurial Teams (2019BT02Y198 to E.S.), High-tech, Major and Characteristic Technology Projects in Guangzhou Area (2023-2025) (2023P-ZD14 to E.S.), Fundamental Research Funds for the Central Universities, Sun Yat-sen University (24kxzx001 to E.S.), China Postdoctoral Science Foundation (2024M763778 to H.L.) and Youth S&T Talent Support Program of Guangdong Provincial Association for Science and Technology (SKXRC2025132 to H.L.).

Data availability

The sequencing data generated in this study, including bulk RNA-seq and scRNA-seq, have been deposited in the Genome Sequence Archive for Human (GSA-Human) under accession number HRA012665 and in the Genome Sequence Archive (GSA) under accession number CRA028606, respectively. The GSA data (CRA028606) are publicly available, whereas the GSA-Human data (HRA012665) are available under restricted access due to data privacy and supervision. Access can be obtained by submitting a Data Access Request to the Data Access Committee (DAC) via the GSA-Human portal. Upon approval, access will be granted to qualified researchers for academic research purposes, and requests will be responded to within 15 working days. We acquired the publicly available human scRNA-seq data GSE246662²⁰, human bulk RNA-seq datasets GSE50760, GSE49355, GSE19279, GSE40367, GSE81558, GSE159216)^74–79 from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov/geo/). All other data supporting the findings of this study are available in the article, its supplementary information and source data and/or from the corresponding author on request. Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76459-7.