Inhibition of the lipocalin-2-phosphatidylserine axis restores natural killer cell immune surveillance

Summary

Lipid metabolic reprogramming can facilitate immune escape by promoting a suppressive phenotype in tumor-infiltrating immune cells, although this process remains poorly understood. Here, we identify the lipoprotein, Lipocalin-2 (LCN2), as an essential factor driving natural killer (NK) cell dysfunction and immunosuppressive phenotype. Spatial metabolomics with crystal structure analysis demonstrates that LCN2 binding to phosphatidylserine (PS) and PS enrichment is required for tumor-associated lipid reprogramming. Increased LCN2-PS binding limits IL-15-mediated JAK-STAT pathway activation in NK cells, while inhibiting tumor-infiltrating neutrophil maintenance of anti-tumor potential in NK cells via suppression of IFN-I response. Structure-based drug screening identifies semapimod as an LCN2 inhibitor that blocks interaction with PS, disrupting the LCN2-PS immunosuppressive axis and inducing tumor control. Overall, this study uncovers a lipid metabolic reprogramming mechanism that mediates innate immune evasion and proposes a tumor treatment strategy through enhanced innate immune surveillance.

Graphical abstract

Highlights

• •LCN2 suppresses NK-cell-mediated antitumor immunity across tumor models
• •LCN2 binds phosphatidylserine and promotes its retention within tumors
• •Serine metabolism drives phosphatidylserine accumulation and immune suppression
• •Semapimod disrupts the LCN2-PS axis and restores antitumor immunity

Qin et al. show that Lipocalin-2 binds phosphatidylserine to promote tumor-associated lipid metabolic reprogramming, impair NK cell immunity, and suppress antitumor surveillance. Pharmacological disruption of the LCN2-PS interaction by semapimod restores innate immune function and inhibits tumor growth.

Introduction

Immune therapies based on checkpoint inhibitors or adoptive immune cell transfers have revolutionized cancer treatment.¹ However, most patients with solid tumors, particularly those with metastatic solid tumors, have limited immunotherapy responses,^2,3 highlighting a need for research into immunotherapy resistance and therapeutic approaches. Tumor metabolic reprogramming, along with evolution of the tumor immune microenvironment, are fundamental characteristics of cancer.⁴ These metabolic changes during tumor progression or metastasis drive metabolic alterations in tumor-infiltrating immune cells and immune cells in distant organs, enabling immune evasion.^5,6 The synthesis, uptake, storage, and transport of intratumoral (i.t.) lipid metabolites promote this process.⁷ Previous studies have shown that upregulation of CD36 leads to uptake of oxidized lipids, resulting in CD8⁺ T cell dysfunction while promoting the survival of regulatory T cells (Tregs) and generation of immunosuppressive myeloid cells.^8,9,10,11 Another receptor, Trem2, binds Apolipoprotein E on the neutrophil surface, inducing neutrophil senescence and enhancing their immunosuppressive capacity.¹² In lung cancer, Trem2⁺ macrophage accumulation leads to a paucity and dysfunction of natural killer (NK) cells,¹³ promoting pulmonary metastasis in breast cancer.¹⁴ Over 1,000 representatives, including CD36, APOE, and Trem2, have been annotated as “lipoproteins” with lipid-binding activity based on conserved functional domains. However, whether and how these lipoprotein-encoding genes contribute to the establishment of immune responses remains to be determined and requires extensive characterization.

Given that solid tumor progression is often accompanied by T cell exclusion,^15,16 stimulating innate immune responses may be an alternative strategy for immunotherapy.^17,18 Unlike cytotoxic T cells, NK cells can recognize and destroy diverse tumor cells without sensitization, providing early defense in tumor surveillance and control.^19,20 However, NK cells exhibit limited tumor infiltration,^21,22 restricting their ability to control tumor growth. In contrast, neutrophils are rapidly recruited to tumor sites during early inflammation in tumor formation, and transmit potent immunosuppressive signals that hinder anti-tumor immune responses.^23,24,25 Metabolic and energetic imbalances in the tumor microenvironment (TME) drive these immune landscapes. Nutrient deprivation and the accumulation of metabolic by-products suppress local antitumor immunity, whereas systemically, the host reprograms the distribution of glucose, amino acids, lipids, and even trace elements to fuel tumor growth and establish metastatic niches.²⁶ However, it remains unclear how such macro-environmental perturbations impair innate immune surveillance.

Here, a screen of lipoprotein genes identified Lipocalin-2 (LCN2) as driving the immune activation to immune suppression transition. LCN2-mediated i.t. phosphatidylserine (PS) accumulation disrupts interleukin-15 (IL-15) signaling and IFN-I response. Genetic knockout or blockade of LCN2 could reprogram the immune-suppressive neutrophil phenotype into an NK-activating phenotype, stimulating antitumor immunity.

Results

LCN2 is associated with immunosuppression in lung cancer

To investigate the association between lipoprotein genes and immunosuppression in the transition from immune surveillance to immune evasion, identified by reduced antitumor effector (e.g., CD8⁺ T and NK cells) infiltration and accumulation of immunosuppressive populations, especially myeloid cells,²⁷ we analyzed the expression of 1,187 lipoprotein-encoding genes to identify potential correlations between expression patterns and infiltration of various immune cell subsets in lung cancer transcriptomic data in the Cancer Genome Atlas (TCGA) (Figures S1A–S1H). CIBERSORT (cell-type identification by estimating relative subsets of RNA transcripts) identified 56 genes with expression patterns significantly associated with a decline in antitumor effector cells and a corresponding increase in immunosuppressive cells (Figure 1A; Table S1). We evaluated possible associations between the expression levels of these 56 candidate genes and progression-free survival, quantified as hazard ratios (HRs), in a public dataset from an independent cohort of immune checkpoint blockade-treated patients.²⁸ The prioritized gene, LCN2, showed increased expression in tumors and was consistently associated with clinical response, whereas other differentially expressed genes (DEGs) showed lower fold change or minimal prognostic relevance (Figure 1B; Figure S1I). Therefore, the expression of specific lipoprotein genes could influence dynamic shifts in tumor immune cell infiltration, with LCN2 a prominent candidate associated with infiltration of immunosuppressive subsets and TME establishment in lung cancer.

Figure 1.

LCN2 knockout activates NK cell-mediated anti-tumor immunity

(A) Illustration of the overlap of genes with positive or negative correlations across seven cell types. Genes positively correlated are associated with neutrophils, Treg cells, and M2 macrophages, while those negatively correlated correspond to NK cells, CD8⁺ T cells, CD4⁺ T cells, and M1 macrophages.

(B) Quantitative analysis of gene expression in tumor versus normal tissues, along with its correlation to hazard ratios (HRs) under anti-PD-1 therapy.

(C) Experimental design for constructing in situ lung cancer models in Kras⁺Lcn2^+/+ and Kras⁺Lcn2^−/− mice.

(D) Representative hematoxylin and eosin (H&E) staining images, accompanied by quantification of the tumor node area as a percentage of the corresponding lung lobes in Kras⁺Lcn2^+/+ and Kras⁺Lcn2^−/− mice (n = 25 lobes). Scale bars, 2 mm.

(E) Representative immunohistochemistry images showing NK1.1⁺ (left) and CD8⁺ (right) cells quantified as a percentage of DAPI-positive cells in tumor-bearing lungs of Kras⁺Lcn2^+/+ and Kras⁺Lcn2^−/− mice (n = 25 lobes). Staining, DAPI (white); NK1.1 (green); CD8 (red). Scale bars, 50 μm.

(F and G) Representative immunohistochemistry images and quantification of NK1.1⁺ and CD8⁺ cells as a percentage of DAPI-positive cells in E0771 (n = 35 lobes) and MC38 (n = 25 lobes) tumor-bearing lungs of wild-type (WT) and Lcn2^−/− mice. Scale bars, 50 μm.

(H and I) Flow cytometry analysis of the proportions of IFN-γ⁺ and GZMB⁺ NK cells in E0771 and MC38 tumor-bearing lungs of WT and Lcn2^−/− mice.

(J and K) Survival analysis of the E0771 metastasis model (J) and MC38 metastasis model (K) in WT and Lcn2^−/− mice, treated with isotype antibody (IgG) or anti-NK1.1 depleting antibody (PK136). (n = 7–10 mice per group).

(L) Survival analysis for the E0771 metastasis model treated with IgG or anti-CD8 depleting antibody in WT and Lcn2^−/− mice. (n = 10–16 mice per group).

Each experiment was independently repeated three times. Data are shown as mean ± SEM (D–I). Statistical analyses were performed using Student’s two-tailed unpaired t test (D–I) or log rank (Mantel-Cox) test (J–L).

LCN2 suppresses NK cell-mediated antitumor immunity in vivo

To investigate whether LCN2 affects antitumor immunity, we generated a Cre recombinase-driven Kras^G12D spontaneous lung cancer model in mice in a Lcn2 knockout (Lcn2^−/−) background (Figure 1C; Figure S1J). Following tumor induction via adenovirus-encoded Cre recombinase, lung tumor burden was significantly lower in Kras^G12DLcn2^−/− mice compared to Kras^G12DLcn2^+/+ controls (Figure 1D; Figure S1K). Multiplex immunohistochemistry (mIHC) staining for NK and CD8⁺ T cells revealed that both populations were significantly increased in tumor-bearing lungs of Lcn2^−/− mice compared with wild-type (WT) Kras^G12D controls (Figure 1E). However, no difference in lung NK cell and CD8⁺ T cell numbers was detected across WT and Lcn2^−/− mice under tumor-free conditions, suggesting that Lcn2 deficiency does not affect tissue homeostasis (Figure S1L).

We employed the E0771 and MC38 cell lines to establish lung metastasis models and the LLC (Lewis lung carcinoma) and B16F10 cell lines to generate subcutaneous (s.c.) tumor models. In WT mice, tumor formation in E0771- or MC38-derived lung metastases, and in LLC- or B16F10-established s.c. tumors, was associated with an increase in serum LCN2 levels (Figures S1M and S1N). Immunostaining and flow cytometry indicated that NK, but not CD8⁺ T cell, populations were significantly increased in E0771 and MC38 lung tumors of Lcn2^−/− mice (Figures 1F and 1G; Figures S1O and S1P). Further flow cytometry of activated effector markers, including IFN-γ, GZMB (granzyme B), and CD11b, indicated that their expression was increased in NK cells, while CD49a, Tim-3, and PD-1 all showed reduced expression in E0771 and MC38 lung tumors of Lcn2^−/− mice (Figures 1H and 1I; Figures S1Q and S1R). Similar phenotypes, including increased NK cell numbers and activation, were observed in immunologically “cold” LLC and B16F10 s.c. tumors in Lcn2^−/− mice compared to WT mice (Figures S1S–S1V). Consistent with these observations, tumor growth was significantly slower in Lcn2^−/−mice compared to WT mice across all models (Figures S2A–S2H). These findings suggest that LCN2-mediated suppression of NK-cell antitumor activity extends beyond the tumor-bearing lung niche.

To determine whether this anti-tumor effect was immune-dependent, we depleted NK cells or CD8⁺ T cells using specific antibodies (Figures S2I and S2J). NK cell depletion abrogated the increased survival time phenotype associated with Lcn2 knockout, while depleting CD8⁺ T cells only partially reversed this improved survival in tumor-bearing Lcn2^−/− mice (Figures 1J–1L; Figures S2K and S2L). To define the cellular origin of LCN2 within the TME, we analyzed publicly available single-cell RNA sequencing (scRNA-seq) datasets from non-small cell lung cancer patients.²⁹ LCN2 expression was enriched in epithelial populations, including tumor cells, alongside multiple myeloid cell subsets, suggesting that malignant and non-malignant cellular compartments contribute to the elevated LCN2 levels observed in tumors (Figure S2M)). Consistent with this notion, LCN2 expression was higher in tumor tissues than in adjacent normal tissues. To estimate the contribution of tumor-derived LCN2 to systemic levels, we quantified serum LCN2 in tumor-bearing WT and Lcn2^−/− mice by ELISA. Intact LCN2 expression by tumor cells contributed negligibly to overall circulating LCN2 abundance (Figure S2N), supporting the suitability of global Lcn2-deficient mice as a model of the immunoregulatory role of host-derived LCN2 in vivo. These results indicated that LCN2 promotes tumor progression by inhibiting NK cell immunosurveillance.

LCN2 mediates PS enrichment in the tumor-bearing lung microenvironment

LCN2 promotes metastatic tumor growth through its iron-chelating activity.³⁰ To determine whether restoring NK cell-mediated antitumor immunity in Lcn2^−/− mice results from altered systemic iron homeostasis, we compared serum iron, hepcidin, and transferrin levels between tumor-free Lcn2^−/− and WT mice. No significant differences were observed (Figure S3A). Similarly, iron concentrations in E0771 and MC38 metastatic tumor tissues, as well as corresponding tumor cells, did not differ between Lcn2^−/− and WT mice (Figures S3B and S3C).

In vivo treatment of the E0771 lung metastasis model with transferrin or iron chelators (Figure S3D) revealed that iron chelation, mirroring ovarian cancer,³¹ markedly increased i.t. NK cell infiltration and alleviated immunosuppressive phenotype (Figure S3E). These effects were enhanced in Lcn2^−/− mice (Figures S3E and S3F). In contrast, transferrin administration had no influence on the immune microenvironment. Neither treatment affected the total CD8⁺ T cells, whereas combined iron chelator and Lcn2 deficiency significantly reduced terminally exhausted CD8⁺ T cells (Figure S3G). These findings indicated that Lcn2 deficiency does not limit iron availability in tumor cells, suggesting that altered iron homeostasis is insufficient to explain the observed immunologic phenotype. Given that LCN2 belongs to the lipocalin family of lipid-binding proteins, we investigated whether Lcn2-dependent lipid remodeling in the TME accounts for the observed immune suppression.

We conducted untargeted metabolomic profiling of lung cancer cells and tumor-bearing lung tissues. Lcn2 deficiency produced significant enrichment of the glycerophospholipid metabolism and ferroptosis pathways (Figures 2A and 2B; Tables S2 and S3), implicating LCN2 as a regulator of lipid metabolic reprogramming in the TME. We next performed spatial metabolomic analysis of tumor-bearing lungs from WT and Lcn2^−/− mice to identify the lipid components regulated by Lcn2 (Figure 2C). Spearman’s subsampling clustering classification (SSCC) analysis of total lipid molecules in positive ion mode identified eight distinct clusters (Table S4). Spatial mapping of these metabolites in lung slices (Figure 2D) showed reduced diversity in tumor-bearing lungs of WT mice, with marked enrichment of PS family metabolites (i.e., clusters 5 and 8; Figures 2E and 2F). In contrast, metabolites of the PS family were significantly reduced in tumor-bearing lungs of Lcn2^−/− mice (Figures 2G–2J).

Figure 2.

Phosphatidylserine binds LCN2, resulting in intratumoral enrichment

(A) KEGG analysis of differential metabolites from untargeted metabolomics of E0771-bearing lungs from WT and Lcn2 knockout mice (n = 5 per group).

(B) KEGG pathway analysis of differential metabolites from untargeted metabolomics in LLC cells with or without Lcn2 knockdown (n = 6 per group).

(C) H&E staining of cross-sections from E0771 tumor-bearing lungs in WT and Lcn2^−/− mice. Scale bars, 1 mm.

(D) Spatial metabolomics analysis of tumor cross-sections from (C), colored by SSCC clustering in positive ion mode.

(E and F) Violin plots displaying the expression of four characteristic metabolites across eight clusters.

(G–J) Spatial feature plots of metabolites: PS (37:1), PS (35:1), PS (37:2), and PS (37:4).

(K) Isothermal titration calorimetry (ITC) assays to determine the binding constants of LCN2 to phosphatidylserine (PS), phosphatidic acid (PA), and phosphatidylethanolamine (PE).

(L) Original heat-change recordings of LCN2 binding to PS.

(M and N) Gene set enrichment analysis (GSEA) profiles showing significant enrichment of gene sets associated with “protein-lipid complex binding” (M) and “lipid translocation” (N) in MDA-MB-231 cells treated with siLCN2 or siNC.

(O) Electron density map of the LCN2-phosphoserine complex. The L-O-phosphoserine electron cloud is depicted in green.

(P) Crystal structure of LCN2 in complex with L-O-phosphoserine (PDB: 9JDO), highlighting the residues involved in ionic interactions with L-O-phosphoserine.

Subsequent isothermal titration calorimetry demonstrated that PS and phosphatidic acid (PA) could bind to LCN2 in vitro, with PA exhibiting significantly weaker binding than PS, and phosphatidylethanolamine (PE) showing no detectable binding to LCN2 (Figures 2K and 2L; Figures S3H and S3I). Gene set enrichment analysis (GSEA) of bulk RNA sequencing (RNA-seq) data from small interfering RNA (siRNA)-induced LCN2 knockdown MDA-MB-231 cells (with high LCN2 expression³²) showed significant alterations in multiple lipid metabolism genes and signaling pathways, including downregulation of the protein-lipid complex binding signaling pathway (Figures 2M and 2N; Figures S3J and S3K).

We resolved the structure of LCN2 in complex with a representative PS, L-O-phosphoserine, by X-ray crystal diffraction (Figure 2O). This structure revealed that phosphoserine forms electrostatic interactions and hydrogen bonds with LCN2 through Y72, R92, W99, and K154, which are highly conserved LCN2 sites across mammals (Figure 2P; Figure S3L). Moreover, PS could bind to both human and mouse LCN2 with similar binding constants (Figure S3M). Comparison of the hLCN2-PS complex and mLCN2 protein structures revealed that R92 adopts a different orientation under PS-binding conditions, suggesting it may be required for PS binding (Figure S3N). To validate these structural predictions, we generated LCN2 mutants targeting residues implicated in PS recognition and performed isothermal titration calorimetry (ITC) analyses. Both R92E and Y72E mutations impaired the interaction between LCN2 and PS, supporting the requirement of these residues for PS binding (Figures S3O and S3P). The R92E mutation abolished detectable heat exchange during ITC measurement, indicating a critical role for R92 in mediating the LCN2-PS interaction. To determine whether LCN2 actively regulates PS persistence within tissues, we intranasally (i.n.) administered fluorescently labeled NBD (7-nitrobenz-2-oxa-1,3-diazole)-PS and monitored its lung retention over time. While NBD-PS signals remained detectable in WT lungs, fluorescence rapidly diminished in Lcn2^−/− mice, indicating impaired local persistence of PS in the absence of LCN2 (Figures S3Q and S3R).

Serine promotes PS production to dampen anti-tumor immunity

Given the immune-suppressive effects of this LCN2-PS axis, we investigated the metabolic source of PS. Considering that phosphatidylserine synthase 1 (PTDSS1) and phosphatidylserine synthase 2 (PTDSS2) catalyze phosphatidylcholine or PE conversion to PS in the presence of L-serine^33,34 (Figure 3A), we hypothesized that serine metabolism contributes to i.t. PS production. We administered L-serine by intraperitoneal (i.p.) injection every third day for 24 days following E0771 cell inoculation in WT mice (Figure 3B), as the E0771 lung metastasis model exhibited the highest serum LCN2 levels among tumor models (Figures S1M and S1N). H&E staining showed that although neither serine nor PS affected E0771 cell proliferation in vitro, systemic delivery of exogenous serine accelerated lung tumor growth in vivo (Figure 3C; Figures S4A–S4D), suggesting that serine conferred pro-tumorigenic effects in an immune-dependent manner. Tissue imaging with a PS probe showed that PS contents were significantly increased in serine-treated tumor-bearing lungs compared to vehicle controls, while no PS signal was detected in other organs (Figure 3D; Figure S4E). These results suggested that serine promotes i.t. PS production. Contrary to observations in Lcn2^−/− mice, flow cytometry revealed a reduced number of NK cells with diminished IFNγ and GZMB expression in serine-treated mouse lungs (Figures 3E and 3F). Moreover, serine administration in Lcn2^−/− mice neither altered PS accumulation nor influenced tumor growth within the TME, indicating that LCN2-mediated PS aggregation is essential for immunosuppression (Figures 3G and 3H; Figure S4F). Consistently, elevated PS levels were detected in the periphery of Lcn2^−/− mice, further supporting the critical role of LCN2 in maintaining PS metabolic cycling (Figure S4G).

Figure 3.

Serine treatment accelerates PS accumulation to impair the IFN-I response

(A) Overview of the PS synthesis pathway.

(B) Experimental design for serine metabolism enhancement.

(C) H&E staining of cross-sections and quantification of tumor node areas as a percentage of lung lobe area in E0771 tumor-bearing lungs in the vehicle and serine treatment groups (n = 10 lobes). Scale bars, 2 mm.

(D) Representative imaging and quantification of PS fluorescence intensity in E0771 tumor-bearing lungs.

(E) Flow cytometric quantification of NK cell percentages in the indicated tumors.

(F) Flow cytometric analysis of GZMB and IFNγ expression in NK cells.

(G and H) H&E staining of cross-sections and quantification of tumor node areas as a percentage of lung lobe area in E0771 tumor-bearing lungs from WT and Lcn2^−/− mice, treated with vehicle or serine. Scale bars, 1 mm.

(I) UMAP clustering of tumor-infiltrating myeloid cell subsets from E0771 tumor-bearing lungs in the indicated treatment groups.

(J) Proportion of each myeloid cell subpopulation among CD45⁺ cells across treatment groups.

(K) Gene expression profiles of individual myeloid cell clusters from scRNA-seq data in (I).

(L) Volcano plot showing differential gene expression in neutrophils between the vehicle and serine treatment groups.

(M) GSEA of scRNA-seq data from TINs in the indicated groups. Data are shown as mean ± SEM (C–F and H). Statistical significance was calculated using Student’s two-tailed unpaired t test (C–F and H).

To determine the impact of serine on lung immune cells, we performed scRNA-seq on CD45⁺ cells isolated from tumor-bearing mice lungs with or without serine. GSEA analysis revealed that IL-15 response and cell migration pathways were attenuated among differential gene sets in NK cells from serine-treated mice, likely due to suppression of genes associated with JAK-STAT (Janus kinase–signal transducer and activator of transcription) signaling (Figures S4H–S4L). Dimensionality reduction analysis of myeloid cells revealed that Trem1⁺ neutrophils and Siglecf⁺Trem2⁺ macrophages were significantly increased in the TME following serine treatment (Figures 3I–3K). Further analysis of scRNA-seq data showed that serine treatment reduced IL-15 and interferon-stimulated gene (ISG) expression, but increased Lcn2 expression in tumor-infiltrating neutrophils (TINs) (Figure 3L). Notably, Lcn2 expression also increased in NK cells from serine-treated mice (Figure S4M). To explore whether increased Lcn2 expression observed following serine treatment was associated with remodeling of cytokine networks within the TME, we compared cytokine expression profiles across treatment groups. Il-1β exhibited the highest expression level and was significantly upregulated following serine treatment (Figure S4N). We assessed whether serine-induced inflammatory cues could regulate Lcn2 expression in vitro. Stimulation of lung-derived neutrophils demonstrated that IL-1β, but not serine alone, increased Lcn2 expression (Figure S4O), suggesting that serine-mediated induction of Lcn2 is likely secondary to inflammatory cytokine remodeling rather than a direct metabolic effect. Pathway enrichment revealed that neutrophils of serine group mice exhibited significant inhibition of pathways related to NK cell proliferation and IFN-I response, while glycerophospholipid metabolism and lipid peroxidation were notably upregulated (Figure 3M). Serine-driven PS synthesis could therefore facilitate establishment of immunosuppressive TINs.

Based on our RNA-seq data from TINs isolated from WT and Lcn2^−/− mice, genes involved in one-carbon metabolism were upregulated in TINs of WT mice (Figure S4P). In addition to serine metabolism, other amino acid metabolic pathways were implicated in this tumor-associated reprogramming of TINs (Figure S4Q), suggesting a complex relationship between metabolic and immune networks. These results demonstrated that exogenous serine could establish immunosuppression by promoting PS biosynthesis.

Inhibition of PS production by CRISPR-mediated deletion of PS synthases enhances NK cell-mediated antitumor activity

We examined whether restricting PS production could potentiate NK cell-driven antitumor immunity. Using CRISPR-Cas9, we generated Ptdss1 and Ptdss2 double-knockout E0771 tumor cells and established a lung metastasis model in WT mice (Figure 4A). Loss of Ptdss1/2 suppressed tumor growth alongside increased NK cell infiltration and elevated expression of GZMB and IFNγ (Figures 4B–4D). In contrast, CD8⁺ T cell frequencies remained unchanged, while CD4⁺ T cells and major myeloid cell populations significantly decreased (Figures 4E and 4F). Consistent findings were observed in MC38 metastatic and LLC s.c. tumor models, where Ptdss1/2 deficiency inhibited tumor growth and enhanced NK cell-mediated antitumor responses (Figures 4G–4N). The tumor-promoting effect of PS synthesis was abolished in Lcn2^−/− mice (Figures 4O and 4P). These results establish LCN2 as indispensable for PS-driven immunosuppression within the TME.

Figure 4.

Inhibition of PS production enhances the antitumor activity of NK cells

(A) Validation of Ptdss1 and Ptdss2 knockout efficiency in E0771 cells.

(B) H&E staining images and quantification of tumor node areas in sgNC (n = 30 lobes) or sgPtdss1/2 (n = 35 lobes) E0771 tumor-bearing lungs. Scale bars, 1 mm.

(C) Representative immunohistochemistry images of NK1.1⁺ and CD8⁺ T cells in sgNC or sgPtdss1/2 E0771 tumor-bearing lungs. Scale bars, 50 μm.

(D) Flow cytometry analysis of NK proportions among CD45⁺ cells, and proportions of IFN-γ⁺ and GZMB⁺ NK cells in sgNC or sgPtdss1/2 E0771 tumor-bearing lungs.

(E) Flow cytometry analysis of CD8⁺ T and CD4⁺ T cells proportions among CD45⁺ cells in sgNC or sgPtdss1/2 E0771 tumor-bearing lungs.

(F) Flow cytometry analysis of neutrophils, macrophages, and monocytes among CD45⁺ cells in sgNC or sgPtdss1/2 E0771 tumor-bearing lungs.

(G) H&E staining images and quantification of tumor node areas in sgNC or sgPtdss1/2 MC38 tumor-bearing lungs. Scale bars, 1 mm.

(H) Tumor growth and tumor weight of sgNC or sgPtdss1/2 LLC subcutaneous tumors.

(I–L) Flow cytometry analysis of NK proportions among CD45⁺ cells, and proportions of IFN-γ⁺ and GZMB⁺ NK cells in sgNC or sgPtdss1/2 MC38 tumor-bearing lungs and LLC subcutaneous tumors.

(M and N) Proportions of Tim3⁺ and CD11b⁺ cells among NK cells in sgNC or sgPtdss1/2 MC38 tumor-bearing lungs and LLC subcutaneous tumors.

(O) H&E staining images and quantification of tumor node areas in sgNC or sgPtdss1/2 E0771 tumor-bearing lungs from WT and Lcn2^−/− mice. Scale bars, 1 mm.

(P) Tumor growth of sgNC or sgPtdss1/2 LLC subcutaneous tumors from WT and Lcn2^−/− mice.

Statistical significance was calculated using Student’s two-tailed unpaired t test for (A–O) or two-way ANOVA for (P). Data are shown as mean ± SEM (A–P).

Identification of semapimod as an LCN2 inhibitor

While inhibiting PS production disrupts LCN2-PS axis functions in immunosuppression, it may also cause severe detrimental effects on protein synthesis and homeostasis. Alternatively, our findings strongly suggest that directly targeting LCN2 may provide a safe and effective rational therapeutic strategy. Since LCN2 inhibitors have not yet been reported, we conducted a structure-based virtual screen by docking and scoring experimental drugs in the DrugBank database.³⁵ This analysis identified semapimod, a drug approved for clinical trials in Crohn disease, as a top candidate (Figure 5A). Molecular docking predicted that semapimod occupies the PS-binding pocket of LCN2 and engages residues surrounding R92, a residue identified by structural and mutational analyses as critical for PS recognition (Figure 5B), suggesting that semapimod disrupts the LCN2-PS interaction by targeting the R92-centered binding interface. ITC experiments showed that LCN2 protein could indeed directly bind to semapimod, with a K_d value around 1 μM (Figure 5C). Thermal shift assays indicated that semapimod binding could increase the melting temperature of LCN2 protein (Figure 5D). Moreover, saturation transfer (STD) and water-ligand observed via gradient spectroscopy (WaterLOGSY) peaks of LCN2 interaction with semapimod from nuclear magnetic resonance (NMR) supported the interaction of semapimod with LCN2 (Figure 5E). Notably, fluorescence polarization assays indicated that semapimod could disrupt PS binding to LCN2 (Figure 5F).

Figure 5.

Identification of Semapimod as an LCN2 inhibitor

(A) Two-dimensional structure of Semapimod (Sema).

(B) Molecular docking simulation of LCN2 binding with Sema, showing intermolecular interactions based on the lowest energy conformation (right).

(C) Original heat-change recordings (left) and binding constants (right) between LCN2 and Sema, as determined by ITC.

(D) Stability of LCN2 protein in the presence of 100 μM Sema, assessed by thermal shift assay.

(E) STD and Waterlogsy NMR spectra for 400 μM Sema in the presence of 20 μM LCN2 protein.

(F) Fluorescence polarization (FP) analysis of relative inhibition of the LCN2-PS interaction at 50 nM, 500 nM, and 1 μM semapimod concentrations.

(G and H) Sema enhances ferroptosis inducer-mediated cell death in LLC (G) and MDA-MB-231 (H) cell lines after 8 h of treatment.

(I) IC₅₀ values for Sema’s inhibitory effect on cell viability in LLC cells, assessed over 24 h.

(J) Effects of Sema treatment on lipid droplet production, mitochondrial structure, and mitochondrial reactive oxygen species (ROS) in LLC cells.

(K and L) Differential gene expression upon LCN2 silencing and Sema treatment, with only genes showing p value < 0.05 included.

(K) Scatterplot of log2 fold changes (FC) for LCN2 silencing and Sema treatment relative to matched control samples, with Pearson’s correlation coefficient (R) shown.

(L) Overlap of differentially expressed genes in LCN2 silencing and Sema treatment.

(M–P) GSEA scatterplots of dysregulated signaling pathways: “nuclear division” (M), “cellular response to lipids” (N), “inflammatory response” (O), and “response to type 1 interferon” (P).

(Q) ORA analysis of consistently altered genes between LCN2 silencing and Sema treatment, based on GO and KEGG pathway databases.

Data are presented as mean ± SEM (F, G, H, and J). Statistical significance was assessed using a Student’s two-tailed unpaired t test (F, G, H, and J) or log(inhibitor) vs. response—variable slope (four parameters) for (I).

As LCN2 silencing increases sensitivity to ferroptosis in cell lines in vitro,³⁶ we evaluated whether semapimod treatment could produce a consistent phenotype in cancer cells. Proliferation assays indicated that a low dose (5 μM) of semapimod significantly enhanced the inhibitory effects of erastin and RSL3, in LLC and MDA-MB-231 cells (Figures 5G and 5H), with synergistic effects with erastin in promoting LLC cell death (IC50 = 1.178 μM; Figure 5I). Flow cytometry assays based on various probes indicated that semapimod was associated with increased lipid droplet accumulation, mitochondrial mass, and mitochondrial reactive oxygen species levels in LLC cells compared to vehicle controls (Figure 5J).

To determine the targeting specificity of semapimod, we performed transcriptomic sequencing in MDA-MB-231 cells treated with semapimod or LCN2 silencing. The results showed that MDA-MB-231 cells treated with semapimod or LCN2 silencing had strong correlations in their respective transcriptional changes and considerable overlap in DEGs (Figures 5K and 5L; Table S5). GSEA analysis of common DEGs between them revealed enrichment in “nuclear division,” “cellular response to lipid,” “inflammatory response,” and “response to type 1 interferon” (Figures 5M–5P). Specifically, over representation analysis (ORA) showed that innate immune activation, viral response, and JAK-STAT signaling pathways were all upregulated following semapimod treatment (Figure 5Q). These results illustrated the effects of semapimod in blocking the LCN2-PS axis and further emphasized how LCN2 inhibition could regulate lipid homeostasis and restore IFN-I response.

Semapimod treatment activates NK cell-mediated anti-tumor immunity

We next investigated whether targeting LCN2 with semapimod could serve as an effective anti-tumor therapy in vivo. We administered semapimod (10 mg/kg every 3 days) to mice intravenously (i.v.) injected with E0771 cells. After 30 days of treatment, mice treated with semapimod showed reduced lung tumor burden compared to vehicle controls (Figures 6A and 6B). Imaging by mIHC and flow cytometry showed that semapimod treatment significantly increased tumor-infiltrating NK cells (TINK) numbers and functional molecule expression, with no change in the number of CD8⁺ T cells (Figures 6C and D; Figure S5A). GSEA of bulk RNA-seq data from tumor-bearing lung tissues from semapimod-treated mice showed upregulation of NK-mediated cytotoxicity pathways, type 1 interferon signaling pathways, and innate immune response pathways compared to vehicle-treated control tumors (Figures 6E–6G). Correspondingly, lipid metabolism-related pathways, including the PPAR (peroxisome proliferator-activated receptor) signaling pathway and triacylglycerol synthesis, were downregulated following treatment (Figure S5B). Kaplan-Meier curves indicated that semapimod treatment effectively prolonged the survival of tumor-bearing mice (Figure 6H).

Figure 6.

Semapimod restores NK cell-mediated anti-tumor immunity

(A) Representative luciferase images and quantification of in vivo luciferase activity in mice from the vehicle and Sema-treated groups (n = 14–15 per group) at 32 days post E0771 inoculation.

(B) Representative H&E-stained sections of E0771 tumor-bearing lungs and quantification of tumor nodule area as a percentage of lung lobe area in the vehicle and Sema groups (n = 25 lobes). Scale bars, 1 mm.

(C and D) Representative mIHC images (C) and quantification of NK1.1⁺, CD8⁺, and GZMB⁺ cells as a percentage of DAPI-positive cells (D) in E0771 tumor-bearing lungs from vehicle and Sema-treated groups (n = 15 lobes). Scale bars, 50 μm.

(E–G) GSEA showing upregulation of “natural killer cell-mediated cytotoxicity” (E), “type 1 interferon signaling pathway” (F), and “activation of innate immune response” (G) in E071 tumor-bearing lung tissues from Sema-treated versus vehicle-treated groups (n = 3 per group).

(H) Survival analysis for the E0771 metastasis model treated with vehicle or Sema (n = 11 per group).

(I) H&E-stained sections of MC38 tumor-bearing lungs and quantification of tumor nodule area as a percentage of the total lung area in the vehicle and Sema groups (n = 30 lobes). Scale bars, 1 mm.

(J and K) Tumor growth over time (J, n = 13) and survival analysis (K, n = 10–11) for LLC subcutaneous tumor treated with vehicle or Sema.

(L) Tumor growth and tumor weight in LLC subcutaneous tumor from WT (n = 9) and Lcn2^−/− (n = 6) mice, treated with vehicle or Sema.

(M) H&E images and quantification of tumor nodule areas as a percentage of lung lobe area in E0771 tumor-bearing lungs (n = 35 or 40 lobes) from WT and Lcn2^−/− mice, treated with vehicle or Sema. Scale bars, 1 mm.

(N) Representative H&E staining images and quantification of tumor nodular area as a percentage of the total lung area in Kras-driven spontaneous tumor model treated with anti-PD-1, Sema, anti-PD-1 combined with Sema, or vehicle control (n = 25 lobes).

Student’s two-tailed unpaired t test (A, B, D, I, M, and N), log rank (Mantel-Cox) test (H and K), or two-way ANOVA (J and L). Data are shown as mean ± SEM (A, B, D, I, L, M, and N).

In the MC38 lung metastasis model and s.c. LLC tumors, semapimod treatment consistently slowed tumor growth, prolonged survival, and enhanced NK cell infiltration and function (Figures 6I–6K; Figures S5C–S5F). We established patient-derived xenograft (PDX) models by s.c. implanting human lung tumors into nude mice. Tumor growth and survival analyses demonstrated that semapimod inhibited xenograft progression and extended survival (Figures S5G and S5H). Importantly, the antitumor activity of semapimod was lost in Lcn2^−/− mice (Figures 6L and 6M), indicating that its therapeutic efficacy requires LCN2. These findings, together with the docking, binding, and competition analyses, support that semapimod exerts its antitumor effects primarily through disruption of the LCN2-PS axis rather than through LCN2-independent mechanisms. Furthermore, combining semapimod with anti-PD-1 therapy enhanced antitumor immunity in the primary lung cancer model and the antiPD-1-resistant LLC s.c. tumor model (Figure 6N; Figures S5I and S5J). To assess the tolerability of semapimod at the therapeutic dose used in this study, we monitored body weight throughout the treatment period and performed histopathological evaluation of major organs. No changes in body weight or overt tissue injury were observed at 10 mg/kg (Figures S5K and S5L). These findings suggest that targeting the LCN2-PS axis may be an effective strategy to initiate NK cell-dependent immunotherapy.

PS inhibited NK cell responses to IL-15

We investigated the molecular mechanisms through which LCN2-mediated PS enrichment impairs NK cell-mediated antitumor immunity. PS binds cognate receptors on immune cells, and several receptors have been identified in the PS signaling pathway, including TIM family members (TIM-1, -3, and -4), tyrosine kinase family members (Tyro3, Axl, and MerTK), Stabilin-1/2, CD300 family proteins, and others.³⁷ We examined PS-related receptor expression on human peripheral blood NK cells and mouse splenic NK cells, revealing a broad distribution of PS transporters on NK cells (Figure S6A). Treatment with exogenous PS did not alter receptor expression (Figure S6B). In primary cultured NK spleen cells, treatment with PS significantly inhibited cytotoxicity toward tumor cells in vitro, as well as IFN-γ and GZMB upon cytokine stimulation in fluorescence-activated cell sorting (FACS) assays (Figures 7A–7C), suggesting attenuated NK activation. Alternatively, LCN2 knockout diminished PS-mediated inhibitory effects. We incubated splenic NK cells with fluorescently labeled NBD-PS in vitro and quantified PS accumulation by flow cytometry. Compared with WT controls, NK cells from Lcn2^−/− mice exhibited reduced NBD-PS signals, suggesting impaired PS enrichment in the absence of LCN2 (Figure S6C). However, LCN2 ablation did not affect PS inhibition of NK cells stimulated with PMA (phorbol 12-myristate 13-acetate) /ionomycin (Figure 7D; Figures S6D and S6E). Therefore, LCN2-mediated enhancement of PS inhibitory effects was associated with NK cytokine signaling in the TME.

Figure 7.

Phosphatidylserine restricts IL-15 signaling by disrupting membrane integrity

(A and B) NK cell-mediated lysis of Yac-1 cells assessed by flow cytometry following PS treatment. NK cells isolated from tumor-free mice (WT, white; Lcn2^−/− mice, blue).

(C) The secretion of IFNγ and GZMB by NK cells was analyzed by flow cytometry after treatment with IL-12 and IL-15.

(D) NK cell secretion of IFNγ and TNFα was assessed by flow cytometry following PMA/Ionomycin stimulation.

(E) Gene set enrichment analysis (GSEA) was conducted on scRNA-seq data of tumor-infiltrating NK (TINK) cells sorted from WT or Lcn2^−/− mice.

(F) Flow cytometry of Di-4-ANEPPS staining in NK cells treated with IL-15 or IL-15 combined with PS for 24 h.

(G) Flow cytometry of membrane lipid rafts as determined by CTxB binding in NK cells treated with IL-15 or IL-15 combined with PS for 24 h.

(H) Immunoblot of NK cells treated with IL-15 or IL-15 combined with PS for 3 h.

(I) Immunoblot analysis of human NK cells treated with IL-15 or IL-15 combined with PS for 3 h.

(J) Real-time cell index measurements were performed to assess NK cell-mediated killing of H1975 cells.

(K) Serum PS concentrations in healthy donors (HD, n = 26) and cancer patients (Pat, n = 59).

(L) Serum LCN2 in healthy donors and cancer patients.

(M) Pearson’s correlation between LCN2 and PS concentrations in samples from (K) and (L).

Data are presented as mean ± SEM (B–D, K, and L). Statistical analysis was performed using Student’s two-tailed unpaired t test (B–D, F, G, K, and L) and Pearson’s correlation (M).

We conducted scRNA-seq to characterize the molecular changes in TINKs from WT and Lcn2^−/− mice (Figure S6F). This analysis revealed that NK cells were affected at multiple levels, with enrichment in metabolism, receptors, transcription factors, kinases, and functional molecules in comparisons of Lcn2^−/− vs. WT (Figures S6G–S6J). Consistent with flow cytometry findings, NK cells displayed enhanced production of effector molecules but decreased immune checkpoint expression (Figure S6K). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis showed enrichment in NK responses to interleukin-15 (IL-15) following Lcn2 knockout (Figure 7E). As mTOR signaling is essential for IL-15-mediated NK cell maturation and activation,³⁸ we conducted ORA with our previous scRNA-seq data from tumors of Lcn2^−/− and WT mice. TINKs from Lcn2^−/− mice displayed significantly higher NK cytotoxicity and mTOR signaling, with enhanced chemotaxis, activation, and oxidative phosphorylation pathways (Figure S6L).

To elucidate the mechanism underlying impaired cytokine signaling in NK cells, we investigated PS in membrane organization and signal transduction. Previous studies have demonstrated that disruption of membrane stability diminishes NK cell antitumor activity,³⁹ while lipid metabolism governs lipid raft formation and regulates cytotoxic granule release. We hypothesized that PS suppresses NK cell function by disturbing membrane order and hindering lipid raft assembly. Indeed, stimulation with IL-12 and IL-15 enhanced membrane stability and increased lipid raft formation, whereas PS treatment inhibited these effects (Figures 7F and 7G). Western blot analysis revealed that exposure of IL-15-stimulated NK cells from WT mice to PS (0–20 μM) significantly suppressed activation of the JAK1-STAT3 signaling pathway (Figure 7H), similarly to that under interferon signaling stimulation (Figure S6M). Consistently, PS treatment inhibited IL-15-dependent JAK1-STAT3 activation and reduced cytotoxicity in human NK cells (Figures 7I and 7J), while suppressing lipid raft formation (Figures S6N and S6O).

ELISA of serum samples from our pan-tumor cohort (patients, Pats) and healthy donors (HDs; see Table S6) revealed that PS and LCN2 were significantly elevated in patients (Figures 7K and 7L). Moreover, LCN2 concentrations positively correlated with PS levels (Figure 7M). To explore the clinical relevance of our findings and to provide insights into therapeutic strategies targeting LCN2, we compared LCN2 expression levels with those of NCAM1 (CD56) across various cancers in the TCGA database. Except for lung adenocarcinoma and squamous cell carcinoma, LCN2 expression shared a significant negative correlation with CD56 in colon adenocarcinoma, rectal adenocarcinoma, pancreatic cancer, ovarian cancer, glioblastoma, and renal papillary carcinoma (Figure S6P).

PS enrichment attenuates IFN-I responses and IL-15 production

The LCN2-PS axis reshapes the myeloid immune landscape in the TME, notably through neutrophil enrichment (Figures 3J and 3M). Therefore, we hypothesized that neutrophils contribute to NK cell immunomodulation beyond the direct impact of PS. We examined neutrophil phenotypes in WT and Lcn2^−/− mice. Flow cytometry revealed that Lcn2 deletion enhanced neutrophil maturity across multiple tumor models (Figures S7A and S7B). GSEA showed that “neutrophil_mediated_immunity” and “inflammatory_response” pathways were more highly enriched in TINs from Lcn2^−/− mice (Figure S7C). TINs from Lcn2^−/− mice exhibited elevated ISG and IL-15 expression, along with a more mature phenotype (Figures S7D and S7E). Immunoblot analysis confirmed increased ISG and IL-15 protein levels in TINs from Lcn2^−/− mice, consistent with RNA-seq data (Figure S7F).

We isolated TINs from both WT and Lcn2^−/− mice and co-cultured them with splenic NK cells. TIN-mediated inhibition of NK cytotoxicity in WT mice was abolished in Lcn2^−/− mice (Figure S7G). Flow cytometry analysis showed that TINs from Lcn2^−/− mice stimulated significantly higher Ki67 and CD16 expression on NK cells compared to WT neutrophils or the medium control treatment (Figures S7H and S7I). These Lcn2 KO TINs did not induce strong upregulation of checkpoint molecules, PD-1 and LAG3, compared to TINs from WT mice (Figures S7J and S7K). Neutralization of IL-15 abolished differences in Ki67 and CD16 expression between NK cells co-cultured with WT or Lcn2^−/− TINs, supporting a requirement for neutrophil-derived IL-15 in the enhanced NK activation phenotype (Figures S7L and S7M). Thus, neutrophils with high ISG gene expression promote NK cell-mediated anti-tumor immunity.

We next performed pathway enrichment analysis using transcriptomic data from TINs. The JAK-STAT signaling pathway was upregulated in TINs of Lcn2 knockout mice (Figure S7N). In neutrophils from WT mouse lung, in vitro PS treatment inhibited interferon-stimulated JAK-STAT signaling and significantly suppressed ISG and IL-15 expression, while these effects were abrogated in the absence of Lcn2 (Figures S7O–S7Q). These results suggested that LCN2 enhanced PS inhibition of the JAK-STAT pathway. Similarly, PS treatment exerted comparable inhibition in human neutrophils (Figure S7R). This pattern of cytokine signaling inhibition was paralleled in NK cells, suggesting a shared mechanism underlying LCN2-mediated immune suppression.

We intravenously injected E0771 cells into mice with conditional JAK1 knockout in neutrophils. Subsequent H&E staining showed that neutrophil-specific JAK1 knockout promoted lung tumor development, accompanied by significantly reduced TINK cell counts in mIHC imaging data (Figures S7S and S7T). We next defined a “neutrophil ISG signature” based on the “WP_INTERFERON_TYPE_I_SIGNALING_PATHWAYS” gene set in scRNA-seq data from non-small cell lung cancers in MsigDB²⁹ (Table S7). The “neutrophil ISG signature” was positively correlated with enrichment in mTOR signaling, NK cell-mediated cytotoxicity, glycolysis signaling, and Myc signaling pathways in lung cancer-associated TINKs (Figure S7U). These results demonstrated that interferon-responsive neutrophils perform an essential function in NK proliferation and activation, whereas LCN2-dependent PS enrichment can disrupt this process, promoting NK cell exhaustion.

Discussion

Rapid energy consumption and metabolite production in the TME are associated with immune evasion. Lipid metabolism provides energy for tumor growth and promotes metastasis,⁴⁰ while inducing immunosuppressive cells and impairing anti-tumor effector cells.^{8,41,42,43,44} Here, we define a crucial role of LCN2 in limiting NK-mediated anti-tumor immunity wherein LCN2-mediated enrichment of i.t. PS suppresses IL-15 signaling and IFN-I responses. Conversely, genetic or pharmacological blockade of LCN2 can reactivate IFN-I responses, initiating NK-dependent anti-tumor responses. These findings provide a potentially effective approach to overcoming resistance to immunotherapy.

LCN2 is implicated in inflammation,^45,46 iron homeostasis regulation,⁴⁷ and tumor metastasis.^30,48 Previous studies have focused on LCN2’s role in tumor nutrient acquisition, particularly iron, while its involvement in immune evasion remained underexplored. Lcn2 deficiency did not compromise systemic iron acquisition, likely due to compensatory transferrin activity. Therefore, altered iron availability alone is insufficient to explain the profound immunologic phenotypes observed following Lcn2 deletion. Our findings expand the repertoire of LCN2 beyond its established role in iron transport and highlight lipid remodeling in the tumor immune microenvironment. Combining iron chelation with Lcn2 deficiency elicited an optimal antitumor response. We identified a conserved interaction between LCN2 and PS across mammalian species, with LCN2 promoting the local persistence and enrichment of PS within tumors. This differs from PS being directly recognized by immunosuppressive receptors such as TIM family members, suggesting that LCN2 regulates PS-mediated immunosuppression by controlling its local availability.

Circumventing tumor strategies for T cell evasion is a promising approach for immunotherapy development, especially for tumors that have lost major histocompatibility complex (MHC) expression,⁴⁹ to mobilize the anti-tumor capabilities of NK cells. NK cells can exert immediate effector functions without prior expansion, but require expansion to maintain sufficiently high metabolic activity.⁵⁰ Cytokine-mediated mTOR signaling is essential for metabolism-dependent NK activation.^38,51 Recent studies highlight membrane lipid raft integrity in sustaining cytokine signaling in NK and cytotoxic T cells.^52,53,54 LCN2-driven PS enrichment disrupts NK cell membrane stability, attenuating cytokine responsiveness and suppressing JAK-STAT signaling. In Lcn2^−/− mice, tumor-infiltrating NK cells displayed enhanced mTOR activation and improved tumor infiltration relative to WT counterparts.

While our findings identify NK cells as dominant effectors targeted by the LCN2-PS axis, CD8⁺ T cells in antitumor immunity warrant further consideration. The importance of NK- and CD8⁺ T-cell-mediated responses varies across tumor contexts, reflecting heterogeneity in immune regulation among tumor types. The LCN2-PS axis across tumor models suggests that its immunoregulatory function extends beyond the lung microenvironment.

Neutrophils play multiple roles in tumors,⁵⁵ promoting tumor growth by weakening T cell immunity,⁵⁶ angiogenesis,⁵⁷ and creating an inflammatory environment. Meanwhile, neutrophils enhance antitumor responses. Distinct neutrophil subpopulations enhance the efficacy of immune checkpoint therapies and are indicators in clinical diagnosis and treatment.^58,59,60,61 Neutrophils also establish antigen-independent activation of NK antitumor responses in response to neutrophil expression of IL-15. In response to neutrophil-mediated IL-15 secretion, NK cells exhibit enhanced proliferation and upregulation of CD16, which mediates antibody-dependent cell-mediated cytotoxicity (ADCC). In particular, neutrophils with high ISG expression exhibit stronger stimulatory effects, indicating that IFN-I responses are indispensable for the formation of an immune-activating phenotype in neutrophils. This mechanism is involved in NK-dependent immune surveillance, particularly early in tumorigenesis.

Serine-restricted diets may control tumor growth through a metabolic toxicity-related mechanism in tumor cells,^62,63 as explored through clinical trials.⁶⁴ Our study showed that serine uptake could promote PS production in the TME, limiting IFN-I responses and enhancing the immunosuppressive phenotype of myeloid cells by upregulating LCN2 expression. The upregulation of LCN2 under serine-rich conditions may be driven by IL-1β-mediated inflammatory signaling, suggesting a link between dietary metabolic interventions and cytokine networks within the TME.

Notably, our structural analyses identified an effective but previously unreported LCN2 inhibitor, semapimod, initially discovered as a treatment for inflammatory bowel disease and approved for clinical trials.⁶⁵ Semapimod binds to the LCN2 protein and disrupts its interaction with PS. Transcriptomic sequencing revealed that treatment could restore IFN-I responses and reprogram lipid metabolism in tumors, consistent with genetic disruption of LCN2. Ultimately, semapimod administration significantly inhibits tumor growth and enhances NK cell activation across multiple tumor-bearing mouse models, and may be an effective immunotherapy option. No adverse effects or overt organ toxicity were observed in tumor-bearing mice. However, given its original targets and pleiotropic immunomodulatory properties, optimization will be required to improve specificity toward the LCN2-PS axis.

Overall, our study uncovers a mechanism by which lipid metabolism mediates innate immune evasion. Targeting LCN2 can reprogram the lipid distribution in tumors, activating neutrophil IFN-I responses within the TME to enhance NK-mediated antitumor immunity.

Limitations

Although our findings demonstrate that the LCN2-PS axis suppresses antitumor immunity across multiple tumor models, the extent to which LCN2-dependent lipid remodeling is conserved across tissue microenvironments remains unclear. Given the metabolic heterogeneity among organs, future studies should determine whether LCN2 regulates other lipids and whether LCN2-mediated lipid remodeling exhibits tissue-specific features triggering organ-dependent immune evasion.

Our single-cell analyses identified stromal cells and neutrophils as contributors to i.t. LCN2 expression; however, the relative contribution of individual cellular sources within the TME remains unclear. Future studies employing lineage-specific genetic models should investigate tumor, stromal, and immune cell-derived LCN2 in shaping local lipid composition and antitumor immunity.

Although elevated LCN2 and PS levels were observed in cancer patients, the LCN2-PS axis requires further investigation in human tumor specimens. Characterizing LCN2 and PS within human tumors, determining their relationship with NK-cell dysfunction, and evaluating their role as biomarkers of immune evasion or response to immunotherapy will be key.

Addressing these questions will refine the mechanistic understanding and translational potential of LCN2-PS axis targeting in cancer immunotherapy.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Haiming Wei (ustcwhm@ustc.edu.cn).

Materials availability

All reagents generated in this study (including cell lines and plasmids) are available on reasonable request.

Data and code availability

Atomic coordinates and X-ray diffraction data have been deposited in the Protein DataBank (PDB: 9JDO). Raw sequence data, including scRNA-seq and bulk transcriptomic data, are available in the Genome Sequence Archive⁶⁶ at the National Genomics Data Center,⁶⁷ China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA018843 and HRA008507). These data are accessible at https://ngdc.cncb.ac.cn/gsa and https://ngdc.cncb.ac.cn/gsa-human. The metabolomics data reported in this paper have been deposited in Open Archive for Miscellaneous Data (https://ngdc.cncb.ac.cn/omix; OMIX: OMIX018776). Source data for this paper are provided in the supplementary materials. This paper does not report original code. Any information required to reanalyze the data reported in this paper is available from the lead contact.

Acknowledgments

We thank the BL17B1/BL18U1/BL19U1 beamline staff at SSRF of the National Facility for Protein Science in Shanghai (NFPS), Shanghai Advanced Research Institute, Chinese Academy of Sciences, for providing technical support in X-ray diffraction data collection and analysis. This project was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB0940102), the National Natural Science Foundation of China (grant no. 82530058), USTC Research Funds of the Double First-Class Initiative (grant nos. YD9100002052 and YD9105202615), the Research Funds of Center for Advanced Interdisciplinary Science and Biomedicine of IHM (QYZD20230010), and Students' Innovation and Entrepreneurship Foundation of USTC.

Author contributions

H.W., Y.Z., and J.Q. conceived the project and designed the experiments; J.Q. wrote the manuscript with help from H.W. and Y.Z.; J.Q. and X.H. performed most of the experiments; Y.J. and Y.L. helped construct the s.c. and metastatic tumor models; J.Q. conducted experiments in protein purification and molecular biology with the help of L.W.; L.W. performed the crystal structure resolution; X.H. performed bioinformatics analyses; Z.L. provided guidance on fluorescence imaging; Z.H., X.X., X.D., Z.N., and B.F. provided statistical advice and helped revise the manuscript; K.R. and Z.T. provided theoretical guidance; Y.C. provided clinical samples.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-mouse CD45.2 PE	BioLegend	Cat# 109830; RRID: AB_1186103
Anti-mouse CD3e BV786	BD	Cat# 564379; RRID: AB_2738780
Anti-mouse NK1.1 BV605	BioLegend	Cat# 108740; RRID: AB_2562274
Anti-mouse CD4 FITC	BioLegend	Cat# 100510; RRID: AB_312712
Anti-mouse CD8a APC	BioLegend	Cat# 100712; RRID: AB_312750
Anti-mouse IFN-γ BV421	BD	Cat# 563376; RRID: AB_2738165
Anti-human/mouse Granzyme B PE	BioLegend	Cat# 372208; RRID: AB_2687031
Anti-mouse CD279 (PD-1) PE	BD	Cat# 568261; RRID: AB_3662737
Anti-mouse CD366 (Tim-3) APC	BioLegend	Cat# 134008; RRID: AB_2562997
Anti-mouse/human CD11b PerCP/Cyanine5.5	BioLegend	Cat# 101228; RRID: AB_893232
Anti-mouse CD49a BV786	BD	Cat# 740919; RRID: AB_2740560
Anti-mouse/human CD11b BV421	BioLegend	Cat# 101236; RRID: AB_10897942
Anti-mouse Ly-6G BV510	BioLegend	Cat# 127633; RRID: AB_2562937
Anti-mouse CD170 (Siglec-F) PE	BioLegend	Cat# 155506; RRID: AB_2750234
Anti-mouse CD279 (PD-1) FITC	eBioscience	Cat# 11-9985-82; RRID: AB_465472
Anti-mouse CD223 PE	BD	Cat# 552380; RRID: AB_394374
Anti-mouse Perforin FITC	BioLegend	Cat# 154310; RRID: AB_2910315
Anti-mouse TNF-α BV421	BioLegend	Cat# 506328; RRID: AB_10900823
Anti-mouse IFN-γ BV605	BioLegend	Cat# 505839; RRID: AB_2561438
Anti-mouse/human Ki-67	BioLegend	Cat# 151225; RRID: AB_2941433
Anti-human/mouse Granzyme B Alexa Fluor® 647	BioLegend	Cat# 515406; RRID: AB_2294995
Anti-mouse CD101 APC	eBioscience	Cat# 17-1011-82; RRID: AB_2815082
Anti-mouse CD182 (CXCR2) BV421	BD	Cat# 566622; RRID: AB_2864336
Anti-mouse CD16 PE/Cyanine7	BioLegend	Cat# 158016; RRID: AB_2890721
Anti-mouse CD45.2 PerCP/Cyanine5.5	BioLegend	Cat# 109828; RRID: AB_893350
Anti-mouse NK1.1/CD161	CST	Cat# 39197; RRID: AB_2892989
Anti-mouse Granzyme B	CST	Cat# 44153; RRID: AB_2857976
Anti-mouse S100A9	CST	Cat# 73425; RRID: AB_2799839
Anti-mouse Ly-6G	CST	Cat# 87048; RRID: AB_2909808
Anti-mouse CD8a	Abcam	Cat# ab217344; RRID: AB_2890649
Anti-mouse β-actin	Proteintech	Cat# 66009-1-Ig; RRID: AB_2687938
Anti-mouse/human Jak1	CST	Cat# 3344; RRID: AB_2265054
Anti-mouse/human Phospho-Jak1 (Tyr1034/1035)	CST	Cat# 3331; RRID: AB_2265057
Anti-mouse/human Stat1	CST	Cat# 14994; RRID: AB_2737027
Anti-mouse/human Phospho-Stat1	CST	Cat# 9167; RRID: AB_561284
Anti-mouse/human Stat3	CST	Cat# 9139; RRID: AB_331757
Anti-mouse/human Phospho-Stat3 (Tyr705)	CST	Cat# 9145; RRID: AB_2491009
Anti-rabbit HRP-Goat Recombinant Secondary Antibody	Proteintech	Cat# RGAR001; RRID: AB_3073505
Anti-mouse HRP-Goat Recombinant Secondary Antibody	Proteintech	Cat# RGAM001; RRID: AB_3068333
Anti-mouse CD8α-InVivo	Selleck	Cat# A2102; RRID: AB_3099521
Rat IgG2b isotype control-InVivo	Selleck	Cat# A2116; RRID: AB_3662740
Anti-mouse NK1.1-InVivo	Selleck	Cat# A2114; RRID: AB_3096489
Mouse IgG2a isotype control-InVivo	Selleck	Cat# A2117; RRID: AB_3662739
Anti-mouse PD-1 (CD279)-InVivo	Selleck	Cat# A2122: RRID: AB_3644244
Rat IgG2a isotype control-InVivo	Selleck	Cat# A2123; RRID: AB_3644245
Bacterial and virus strains
E. coli BL21(DE3) Chemically Competent Cell	Transgen	Cat# CD601
Biological samples
Healthy donor blood	First Affiliated Hospital of Anhui Medical University	N/A
Tumor patient blood	First Affiliated Hospital of Anhui Medical University	N/A
Chemicals, peptides, and recombinant proteins
DMEM	VivaCell	Cat# C3103-0500
RPMI-1640	VivaCell	Cat# C3001-0500
Fetal Bovine serum (FBS)	Gibco	Cat# 10091148
Penicillin-Streptomycin Liquid	Solarbio	Cat# P1400
Trypsin-EDTA solution,0.25% (without phenol red)	Solarbio	Cat# T1300
Eosin Y Stain Solution,For HE	Solarbio	Cat# G1100
Hematoxylin-Eosin(HE) Differentiation Solution	Solarbio	Cat# G1862
Bluing Solution	Solarbio	Cat# G1866
Neutral Balsam Mounting Medium	Sangon Biotech	Cat# E675007
Masson Tricolor Staining Kit	Servicebio	Cat# G1006
H&E Staining Kit (Hematoxylin and Eosin)	Servicebio	Cat# G1005
Collagenase IV	Sigma	Cat# C5138
DNase I	Sigma	Cat# D5025
AAV9-TBG-CRE	WZ biosciences	Cat# pAV204014
D-Luciferin,Potassium Salt D	Yeason	Cat# 40902ES03
BD Pharm Lyse™ Lysing Buffer	BD	Cat# 555899
Percoll	Cytiva	Cat# 17-0891-09
Fixable Viability Stain 780	BD	Cat# 565388
CellTrace™ Violet	Thermo Fisher Scientific	Cat# C34557
MitoTracker® Red CMXRos	Yeason	Cat# 40741ES50
MitoSOX Red Mitochondrial Superoxide Indicator	Yeason	Cat# 40778ES50
Cell Activation Cocktail	BioLegend	Cat# 423304
7-AAD	BD	Cat# 559925
Phosphatidylserine	Yuanye Bio-Technology	Cat# S27340
Phosphatidic acid	Yuanye Bio-Technology	Cat# Y35775
Phosphatidylethanolamine	Yuanye Bio-Technology	Cat# Y46794
L-O-Phosphoserine	Macklin	Cat# L859951
Membrane Lipid Strips	Echelon Biosciences	Cat# P-6002
Fatty acid-free bovine serum albumin	Yeason	Cat# 36104ES25
SuperSignal West Atto Ultimate Sensitivity Substrate	Thermo Fisher Scientific	Cat# A38554
PSVue550	Molecular Targeting Technologies	Cat# P-1005
18:1-12:0 NBD PS	Sigma	Cat# 810195P
L-α-phosphatidylserine	Aladdin	Cat# L130318
TRIzolTM Reagent	Thermo Fisher Scientific	Cat# 15596018
RIPA Lysis and Extraction Buffer	Thermo Fisher Scientific	Cat# 89901
Protease Inhibitor Cocktail (EDTA-Free, 100× in DMSO)	Selleck	Cat# B14001
Phosphatase inhibitor cocktail	APEXBIO	Cat# K1012
Immobilon®-P PVDF Membrane	EMD Millipore	Cat# IPVH00010
PageRuler Prestained Protein Ladder	Thermo Fisher Scientific	Cat# 26616
SDS-PAGE Sample Loading Buffer (5×)	Beyotime	Cat# P0015L
ChamQ Universal SYBR qPCR Master Mix	Vazyme	Cat# Q711
Recombinant Mouse IL-2 Protein	R&D systems	Cat# 402-ML
Recombinant Mouse IL-15 Protein	R&D systems	Cat# 448-ML
Recombinant Human IL-2 Protein, CF	R&D systems	Cat# BT-002
Recombinant Human IL-15 Protein, CF	R&D systems	Cat# BT-015
E-Plate Insert 16	Agilent	Cat# 6465382001
Isopropyl-β-D-thiogalactopyranoside	Sigma	Cat# I5502
Nickel-nitrilotriacetic acid	Thermo Fisher Scientific	Cat# R90101
6.5 mm Transwell® 5.0 μm	Corning	Cat# 3421
RSL3	Selleck	Cat# S8155
Erastin	Selleck	Cat# S7242
Semapimod	MCE	Cat# HY-15509
Ferrostatin-1	Selleck	Cat# S7243
L-serine	Sellck	Cat# S9353
Cell Counting Kit-8 (CCK-8)	TargetMol	Cat# C0005
SYPRO™ Orange protein gel stain	Thermo Fisher Scientific	Cat# S6650
Puromycin	InvivoGen	Cat# ant-pr-1
Di-4-ANEPPS	MKBIO	Cat# MX4038
Transferrin His Tag Protein, Mouse	UA BIOSCIENCE	Cat# UA030074
Deferiprone	TargetMol	Cat# T1565
Lipofectamine™ 3000 Transfection Reagent	Thermo Fisher Scientific	Cat# L3000001
Polyethylenimine	Yeason	Cat# 40816ES03
Critical commercial assays
4-plex IHC staining kit	Wisee Biotechnology	Cat# M-D110041-100T
Mouse Lipocalin-2/NGAL Quantikine ELISA Kit	R&D systems	Cat# MLCN20
Foxp3 transcription factor staining buffer set	eBioscience	Cat# 00-5523-00
Cell Navigator® Fluorimetric Lipid Droplet Assay Kit ∗Red Fluorescence∗	AAT Bioquest	Cat# 22735
Human Lipocalin-2/NGAL Quantikine ELISA Kit	R&D systems	Cat# DLCN20
Human PS ELISA KIT	mlbio	Cat# YJ998563
Mouse FPN ELISA Kit	mlbio	Cat# ml324152
Mouse Hepcidin ELISA Kit	mlbio	Cat# ml037452
Mouse TRF ELISA Kit	mlbio	Cat# ml057835
Anti-Ly-6G MicroBeads UltraPure, mouse	Miltenyi Biotec	Cat# 130-120-337
EasySep™ Mouse CD49b Positive Selection Kit	STEMCELL	Cat# 18755
NK Cell Isolation Kit, human	Miltenyi Biotec	Cat# 130-092-657
HiScript III RT SuperMix for qPCR (+gDNA wiper)	Vazyme	Cat# R323-01
Cholera Toxin B subunit, CTB	absin	Cat# 131096-89-4
Deposited data
Crystal structure of LCN2 in complex with L-O-phosphoserine	This study	PDF: 9JDO
RNA-seq data of MDA-MB-231 cell lines treated with siLCN2 and siNC	This paper	GSA-human: HRA008507
RNA-seq data of MDA-MB-231 cell lines treated with Semapimod and vehicle	This paper	GSA-human: HRA008507
RNA-seq data of E0771-bearing lung neutrophils from Lcn2 KO and WT mice	This paper	GSA: CRA018843
RNA-seq data of E0771-bearing lung tissues from mice treated with Semapimod and vehicle	This paper	GSA: CRA018843
Single-cell seq data of E0771-bearing lung tissues from Lcn2 KO and WT mice	This paper	GSA: CRA018843
Single-cell RNA-seq data of CD45+ cells from E0771-bearing lung of mice treated with L-serine and vehicle	This paper	GSA: CRA018843
Metabolomics dataset	This paper	OMIX018776
Public single-cell RNA-seq data	Salcher et al.29	Zenodo7227571 https://zenodo.org/records/7227571
Experimental models: Cell lines
LLC	Shanghai Cell Bank	Cat# SCSP-5252
B16-F10	Shanghai Cell Bank	Cat# SCSP-5233
MDA-MB-231	Shanghai Cell Bank	Cat# SCSP-5043
YAC-1	Shanghai Cell Bank	Cat# TCM28
NCI-H1975	Procell	CL-0298
MC38-Luc	This paper	N/A
E0771-Luc	This paper	N/A
Experimental models: Organisms/strains
Mouse: C57BL/6J	Shanhai SLAC Laboratory Animals	N/A
C57BL/6J-Lcn2em5Cd5686/Gpt	GemPharmatech Co., Ltd.	Strain NO. T014584
C57BL/6JGpt-Krasem1Cin(loxP-stop-loxP-G12D)/Gpt	GemPharmatech Co., Ltd.	Strain NO. T004551
BALB/cNj-Foxn1nu/Gpt	GemPharmatech Co., Ltd.	Strain NO. D000521
C57BL/6JGpt-H11em1Cin(hS100A8-iCre)/Gpt	GemPharmatech Co., Ltd.	Strain NO. T005636
C57BL/6JGpt-Jak1em1Cflox/Gpt	GemPharmatech Co., Ltd.	Strain NO. T018680
Oligonucleotides
Primers for qPCR	See Table S7	N/A
Sense sequence (5′–3′) of siRNA for human LCN2 silencing: GCAUGCUAUGGUGUUCUUCTT	TSINGKE	N/A
Antisense sequence (5′–3′) of siRNA for human LCN2 silencing: GAAGAACACCAUAGCAUGCTG	TSINGKE	N/A
shRNA1 sequence (5′–3′) for mouse Lcn2 silencingGCTTTACGATGTACAGCACCA	TSINGKE	N/A
shRNA2 sequence (5′–3′) for mouse Lcn2 silencing: GCTACTGGATCAGAACATTTG	TSINGKE	N/A
pLV[2CRISPR]-hCas9 carrying a distinct sgRNA targeting Ptdss1 (Ptdss1 sgRNA: CATGGTCGTTTGCCGGTTTC)	VectorBuilder	N/A
pLV[2CRISPR]-hCas9 carrying a distinct sgRNA targeting Ptdss2 (Ptdss2 sgRNA: AGACGCTCATGATCCGTGAC)	VectorBuilder	N/A
Recombinant DNA
LV3 (pGLVH1/GFP&Puro)	GenePharma	N/A
ps.pAX2 vector	Addgene	Cat# 12260
pMD.2G vector	Addgene	Cat# 12259
pET28a hLCN2	This paper	N/A
pET28a mLCN2	This paper	N/A
pET28a hLCN2R92E	This paper	N/A
pET28a hLCN2Y72E	This paper	N/A
Software and algorithms
GraphPad Prism (v.8.0)	Prism	https://www.graphpad.com/
FlowJo vX.0.7	BD	https://www.flowjo.com/
Living Image Software	Perkin Elmer	N/A
ImageJ2 (version 1.54f)	ImageJ	https://imagej.net/software/imagej2/
mzMLConverter and MSiReaderProgenesis QI V2.3 software	Nonlinear, Dynamics, Newcastle, UK	N/A
xCELLigence experiment report (2.0.0.1301)	N/A	N/A
XDS(version Jun 30, 2023)	N/A	N/A
Phenix (v.1.19)	N/A	N/A
MOLREP (v.11.7.03)	N/A	N/A
Python (v.3.11.0)	Python core team	https://www.python.org/
R (v.4.3.3)	R core team	https://www.r-project.org/
clusterProfiler (v.4.10.1)	T Wu#, E Hu#, S Xu, M Chen, P Guo, Z Dai, T Feng, L Zhou, W Tang, L Zhan, X Fu, S Liu, X Bo∗, G Yu∗. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. The Innovation. 2021, 2(3):100141. https://doi.org/10.1016/j.xinn.2021.100141	https://github.com/YuLab-SMU/clusterProfiler
ComplexHeatmap (v.2.18.0)	Zuguang Gu et al., Complex heatmaps reveal patterns and correlations in multidimensional genomic data, Bioinformatics, 2016.	https://github.com/jokergoo/ComplexHeatmap
scanpy (v.1.10.2)	SCANPY: large-scale single-cell gene expression data analysis. F. Alexander Wolf, Philipp Angerer, Fabian J. Theis. Genome Biology 2018 Feb 06. https://doi.org/10.1186/s13059-017-1382-0.	https://github.com/scverse/scanpy
scVI-tools	A Python library for probabilistic analysis of single-cell omics data. Nature Biotechnology 2022 Feb 07. https://doi.org/10.1038/s41587-021-01206-w.	https://github.com/scverse/scvi-tools
scVelo (v.0.3.2)	Bergen, V., Lange, M., Peidli, S. et al. Generalizing RNA velocity to transient cell states through dynamical modeling. Nat Biotechnol 38, 1408–1414 (2020). https://doi.org/10.1038/s41587-020-0591-3	https://github.com/theislab/scvelo
decoupler (v.1.6.0)	Badia-i-Mompel P., Vélez Santiago J., Braunger J., Geiss C., Dimitrov D., Müller-Dott S., Taus P., Dugourd A., Holland C.H., Ramirez Flores R.O. and Saez-Rodriguez J. 2022. decoupleR: Ensemble of computational methods to infer biological activities from omics data. Bioinformatics Advances. https://doi.org/10.1093/bioadv/vbac016	https://github.com/saezlab/decoupler-py

Experimental model and study participant details

Mice

LCN2^−/−, S100a8^cre, Jak1^fl/fl, and BALB/cNj-Foxn1^nu/Gpt mice were purchased from GemPharmatech (Nanjing, China). Kras^LSL-G12D mice were obtained from the Model Animal Research Center of Nanjing University. These mice, along with WT C57BL/6J controls, were bred at the University of Science and Technology of China. Before experiments, 5- to 8-week-old mice were housed for at least 7 days under specific pathogen-free conditions, maintained on a 12-h light-dark cycle at 22°C–26°C. All experimental procedures adhered to the National Guidelines for the Use of Research Animals (China) and were approved by the Animal Ethics Committee of the University of Science and Technology of China (USTCACUC22280122012). Experimental mice were age- and sex-matched.

Cell lines and primary cultures

LLC, B16F10, MDA-MB-231, and YAC-1 cell lines were obtained from the Chinese Academy of Sciences Cell Bank (Shanghai, China), while NCI-H1975 cells were purchased from Procell (Wuhan, China). The MC38 colon adenocarcinoma cell line was generously provided by Professor Yangxin Fu (University of Texas Southwestern Medical Center, Dallas, USA). E0771 cells were purchased from BNCC (Shangcheng, China). MC38-Luc and E0771-Luc cell lines were established by transfecting MC38 and E0771 cells with luciferase-expressing lentivirus.

MC38, MDA-MB-231, B16F10, LLC, and MC38-Luc cells were cultured in DMEM supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin-streptomycin (P/S). E0771-Luc, YAC-1, and NCI-H1975 cells were maintained in RPMI 1640 supplemented with 10% FBS and 1% P/S. All cell lines were confirmed to be Mycoplasma-free.

For in vitro culture of Ly6G⁺ neutrophils, neutrophils were isolated from the lung or bone marrow using anti-Ly6G MicroBeads (Miltenyi Biotec) and cultured in mouse neutrophil complete medium (Procell). NK cells were isolated using a EasySep™ Mouse CD49b Positive Selection Kit (Stemcell). For NK-neutrophil co-culture assays, cells were plated in 96-well round-bottom plates and incubated for 24 h in 200 μL medium before flow cytometry analysis.

Human samples

Serum samples from healthy donors and cancer patients were obtained from the First Affiliated Hospital of the University of Science and Technology of China, with informed consent. The study was approved by the Ethics Committee of the University of Science and Technology of China (2020-research-36). Clinical characteristics of the samples are provided in Table S5.

Tumor models and in vivo therapy

For the Kras^G12D lung cancer model, 2 × 10¹¹ vg/mouse of recombinant AAV9-Cre virus was administered i.n. to age- and sex-matched Kras⁺Lcn2^+/+ and Kras⁺Lcn2^−/− littermates. Lungs were collected for histological analysis 15 weeks post-infection.

In subcutaneous tumor models, 2 × 10⁵ LLC cells, or 1 × 10⁵ B16F10 cells, were inoculated subcutaneously into age- and sex-matched mice. Tumor growth was measured every three days, with volume calculated as 0.5 × length × width² For lung metastasis models, 2 × 10⁵ MC38-Luc or 1×10⁶ E0771-Luc cells were injected intravenously. Mice were anesthetized and i.p. injected with fluorescein potassium salt (Yeasen, 40902ES03) at designated time points. Fluorescence imaging was performed 15 min later using an IVIS Spectrum small animal imager, and results were analyzed with Living Image 4.4 software.

Lung cancer patient-derived xenografts (PDX) were established using the 3#-Ade PDX lung adenocarcinoma model.⁶⁸ PDX tumors were sectioned into ∼20 mm³ fragments and subcutaneously implanted into BALB/cNj-Foxn1^nu/Gpt mice.

For in vivo Semapimod (MCE, HY-15509A) treatment, mice were randomized into two groups on day 3 or 5 post-tumor implantation and received i.p. injections of 10 mg/kg Semapimod or vehicle every three days. For anti-PD-1 therapy, mice bearing LLC tumors were treated with 100 μg/mouse anti-PD-1 antibody (Selleck, A2122, clone RPM1-14) or control IgG (Selleck, A2123, Rat IgG2a) on the indicated days.

Method details

Antibody-mediated cell depletion

For depletion of NK1.1⁺ or CD8⁺ T cells, mice received i.p. injections of 200 μg anti-NK1.1 (Selleck, clone PK136) or 200 μg anti-CD8 (Selleck, clone 2.43) monoclonal antibodies 24 h before tumor challenge. Antibody injections were repeated weekly, and depletion efficiency was verified via flow cytometry.

Real-time cytotoxicity assays

To evaluate the effects of serine or PS on E0771-Luc cell proliferation, 1×10⁴ cells per well were seeded in E-Plate 16 and monitored using the RTCA-MP system, which records cell density as a cell index (CI) curve. After 24 h, varying concentrations of serine or PS were added.

For human NK cell cytotoxicity assays, 1 × 10⁴ H1975 target cells were seeded 24 h before adding NK cells at different effector-to-target ratios. Each condition was tested in triplicate. Cytolysis was calculated as: % cytolysis = (CI (without NK cells) − CI (with NK cells)/CI (without NK cells)).

Tissue processing

To isolate tumor-infiltrating immune cells, tumor tissue was carefully excised, and any surface impurities were removed by rinsing with saline. The tissue was then minced into ∼2 mm fragments. Subsequently, the tissue was subjected to enzymatic digestion in RPMI 1640 medium supplemented with 2 mg/mL collagenase IV and 0.1 μg/mL DNase I for 1 h under continuous rotation. After digestion, the resulting cell suspension was filtered through a 70-μm mesh to remove undigested tissue. Red blood cells were lysed using RBC lysis buffer (BD Biosciences), and the remaining cells were prepared for further analysis.

Flow cytometry

For cell surface staining, antibodies were added to 100 μL of PBS, and cells were simultaneously stained with Fixable Viability Stain 780 (BD Biosciences) to identify live cell populations. For intracellular fluorescent labeling, MitoTracker® Red CMXRos (Yeasen), MitoSOX™ Red Mitochondrial Superoxide Indicator (Yeasen), and Cell Navigator® Fluorimetric Lipid Droplet Assay Kit Red Fluorescence (AAT Bioquest) were diluted as per the manufacturers instructions and incubated with cells at 37°C for 30 min. For intracellular cytokine assays, NK cells were stimulated with Cell Activation Cocktail for 4 h at 37°C. After surface labeling, NK cells were fixed and permeabilized using the Foxp3 transcription factor staining buffer set (eBioscience). Cytokine antibodies were added and incubated at 4°C for 1 h before flow cytometry analysis.

For mouse NK cell cytotoxicity assays, Yac-1 cells were labeled with CellTrace™ Violet (Thermo), and NK cells were incubated with Yac-1 cells for 4 hours at 37°C in a 5% CO₂ atmosphere. Five minutes before flow cytometry analysis, 7-AAD (BD Biosciences) was added to assess cell viability. Flow cytometry was performed using a BD FACSCelesta (BD Biosciences), and data were analyzed with FlowJo vX.0.7 (BD Biosciences) software.

Lentivirus production

Lentiviruses for mouse Lcn2 gene knockdown were packaged using psPAX2, pMD2.G (Addgene), and LV3 (pGLVH1/GFP&Puro) (GenePharma). Briefly, 2 μg of psPAX2, 1 μg of pMD2.G, and 2 μg of LV3 constructs were co-transfected into HEK-293T cells using Polyethylenimine (Yeasen). Lentiviral particles were harvested 48 and 72 h post-transfection. LLC cells were infected with the lentivirus for 24 h, followed by puromycin selection for resistant clones. The shRNA sequence (5′–3′) was GCTTTACGATGTACAGCACCA.

Small interfering RNA (siRNA) knockdown of human LCN2 expression

The MDA-MB-231 cell line was transfected with either siRNA-Lcn2 or siRNA-mock (Tsingke) using Lipofectamine 3000 (Thermo), following the manufacturer’s instructions. After 24 h, cells were collected for RNA extraction and interference efficiency was assessed by qPCR. The sense (5′–3′) and antisense (5′–3′) sequences of the siRNA were GCAUGCUAUGGUGUUCUUCTT and GAAGAACACCAUAGCAUGCTG, respectively.

RNA extraction, cDNA synthesis, and qPCR

Total RNA was extracted from cells using TRIzol (Thermo), and cDNA was synthesized using HiScript II Q RT SuperMix for qPCR with gDNA wiper (Vazyme). Quantitative PCR (qPCR) was performed using ChamQ Universal SYBR qPCR Master Mix (Vazyme). GAPDH and β-actin were used as endogenous controls. All primers used are listed in Table S8.

Histological analyses

Lung tissue was fixed in formalin for 48 h and embedded in paraffin wax. The paraffin-embedded tissue was sectioned into 4-μm slices, which were dewaxed before staining with hematoxylin and eosin (H&E) stain. For quantitative assessment of tumor burden, all lung lobes from each mouse were processed and analyzed independently. Tumor nodular area was measured in each lung lobe and normalized to the total lung lobe area where indicated. Each data point shown in the corresponding scatterplots represents an individual lung lobe. Statistical analyses incorporated measurements from all lung lobes collected from all mice within each experimental group.

mIHC analysis

Dewaxed and rehydrated paraffin sections underwent antigen retrieval in Tris-EDTA buffer (pH 9.0). Endogenous peroxidase activity was blocked using 3% hydrogen peroxide, and sections were incubated at room temperature for 30 min. Primary antibodies were added and incubated overnight at 4°C. After washing, sections were incubated with HRP-conjugated secondary antibodies for 15 min at room temperature. The reaction was amplified using TSA signal amplification buffer (Yuanxi Bio, D110041) for 10 min. This process was repeated for subsequent primary antibodies. For quantification, one random high-power field (40×) per lung lobe was selected, and the proportion of marker-positive cells was determined relative to total DAPI+ cells. All lung lobes from each mouse were analyzed, and statistical comparisons incorporated data from all lobes across all mice within each group.

CRISPR–Cas9-mediated simultaneous knockout of Ptdss1 and Ptdss2

CRISPR–Cas9 was used to generate tumor cell lines lacking both Ptdss1 and Ptdss2. Two separate CRISPR lentiviral vectors (pLV[2CRISPR]-hCas9, VectorBuilder), each carrying a distinct sgRNA targeting Ptdss1 or Ptdss2 (Ptdss1 sgRNA: CATGGTCGTTTGCCGGTTTC; Ptdss2 sgRNA: AGACGCTCATGATCCGTGAC), were used to sequentially infect tumor cells. Lentiviral particles were produced in HEK293T cells by co-transfecting each sgRNA-containing plasmid with packaging plasmids. Following transduction, cells were selected with puromycin and hygromycin to obtain stable polyclonal double-knockout populations.

Enzyme-linked immunosorbent assay (ELISA)

ELISA assays were performed to measure the concentrations of phosphatidylserine (mlBio, YJ998563) and LCN2 (R&D Systems, MLCN20) in human serum; LCN2 (R&D Systems, QK1757) in mouse serum; and FPN (mlBio, ml324152), Hepcidin (mlBio, ml037452), and TRF (mlBio, ml057835) in mouse tissues. Samples were prepared and incubated in wells, allowing target proteins to bind to the surface. Subsequently, HRP-conjugated secondary antibodies and substrate were sequentially added, initiating a catalytic reaction. After 30 min, the reaction was terminated, and optical density was measured to quantify protein concentrations.

Metabolomics analysis

Samples were immediately snap-frozen in liquid nitrogen and pulverized. A solvent mixture of methanol and water (4:1 ratio) was added to the powdered samples. Following a 2-min incubation at −20°C, the mixture was subjected to 60 Hz shaking for 2 min. The samples were then sonicated in an ice-water bath for 10 min, followed by another 30-min storage at −20°C. After centrifugation at 13,000 rpm for 10 min at 4°C, the supernatant was collected. It was dried using a refrigerated centrifugal concentrator and re-solubilized in a solvent mixture of methanol and water (1:4 ratio). After vortexing for 30 s and sonication for 3 min, the samples were stored at −20°C for 2 h. Following another centrifugation step, the supernatant was filtered through a 0.22 μm microfilter and transferred to assay vials. Samples were then analyzed using LC-MS. The metabolic profiles were obtained using an ACQUITY UPLC I-Class plus (Waters Corporation, Milford, USA), coupled with a Q-Exactive mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) equipped with a heated electrospray ionization (ESI) source. Both ESI-positive and ESI-negative ion modes were employed. Raw LC-MS data were processed using Progenesis QIV2.3 (Nonlinear Dynamics, Newcastle, UK) for baseline filtering, peak identification, integration, retention time correction, peak alignment, and normalization. Compound identification was carried out using the human Metabolome Database (HMDB), Lipidmaps (V2.3), Metlin, EMDB, PMDB, and a self-built database based on the precise mass-to-charge ratio (M/z), secondary fragmentation patterns, and isotope distribution. Specifically, for metabolomics analyses, differential metabolites between groups were identified based on fold-change analysis and Student’s t test, followed by Benjamini–Hochberg false discovery rate (FDR) correction. Adjusted p-values were subsequently used to determine statistical significance.

Spatial metabolomics assays and data analysis

Fresh tumor tissues were embedded in a cryosection embedding agent and stored at −80°C. The tissues were cut into continuous sagittal slices approximately 10 μm thick using a cryosectioner (Leica CM 1950, Leica Microsystems, Germany), thawed, and mounted on positively charged desorption plates (Thermo Scientific). The slices were initially dried at −20°C for 1 h, followed by drying at room temperature for 30 min before mass spectrometry imaging (MSI) analysis. Serial sections were retained for H&E staining. MSI was performed using an AFADESI-MSI platform (Beijing Victor Technology Co., Ltd., Beijing, China) connected to a Q-Orbitrap mass spectrometer (Q Exactive, Thermo Scientific, USA). The solvents used were acetonitrile (ACN)/H2O (9:1) for negative ion mode and ACN/H2O (9:1) for positive ion mode. The mass resolution was set to 20,000, with a mass range of 70–1200 Da. MSI data were acquired by scanning the sample surface in the x-direction at a constant rate of 0.2 mm/s, with 50 μm vertical steps in the y-direction. The raw data files were converted to imML format using imzMLConverter, and the imaging files were visualized in MSiReader. The ion images were reconstructed using the Cardinal software package after background subtraction. MS images were normalized per pixel using total ion count (TIC) normalization. Metabolite annotation was carried out using the SmetDB database and the pySM annotation framework.

Protein expression and purification

Both human and mouse LCN2 proteins used for in vitro experiments were expressed in bacteria. DNA sequences corresponding to the human LCN2 (P80188) and mouse LCN2 (P11672) proteins were synthesized and optimized for prokaryotic expression. The Pet28a plasmid carrying the target sequence was transfected into BL21(DE3) cells, which were selected under kanamycin resistance. Monoclonal strains were cultured overnight at 37°C, and the next day, the culture was scaled up to 700 mL of LB medium until the optical density (OD₆₀₀) reached approximately 0.8. After a brief cooling period, protein expression was induced with 0.5 mM isopropyl-β-D-thiogalactopyranoside (IPTG) at 16°C for 18 h. The cells were lysed using high-pressure homogenization in lysis buffer containing 20 mM Tris-HCl (pH 8.0) and 1 M NaCl. The cell lysates were clarified by centrifugation (30 min at 15,000 g, 4°C) and passed through a column containing nickel-nitrilotriacetic acid (Ni-NTA) agarose beads. The beads were washed with a gradient of imidazole and the protein was eluted. Further purification was performed using size exclusion chromatography on a Superdex 75 Increase 10/300 column in buffer containing 20 mM HEPES (pH 7.5) and 150 mM NaCl.

ITC

Binding constants between LCN2 and small-molecule compounds were measured using a Malvern PEAQ ITC at 25°C. Titration experiments were conducted in a buffer containing 20 mM HEPES (pH 7.5) and 150 mM NaCl. The initial titration volume was 1 μL, followed by sequential additions of 2 μL, with a 120-s interval between each addition. The protein solution (20 μM final concentration) was placed in the lower chamber, while small-molecule compounds (200 μM or 400 μM) in the same buffer were titrated from the syringe. The raw data were analyzed using PEAQ-ITC software, and the dissociation constant (K_D) was determined directly from the fitting results.

Crystallization of LCN2-Phosphoserine complex and structure determination

The hLCN2 protein and phosphoserine were incubated at 4°C for 4 h, followed by centrifugation to concentrate the mixture. The concentrated solution was then combined with a reservoir solution (0.2 M Tris-HCl, pH 7.9, 20% PEG4000) at a 1:1 volume ratio and crystallized via vapor diffusion using the sitting-drop method at 20°C. Crystals were cryoprotected by adding 20% glycerol to the reservoir solution and snap-frozen in liquid nitrogen. X-ray diffraction data were collected at the Shanghai Synchrotron Radiation Facility (SSRF) 19U1 beamline. Diffraction data were indexed and integrated with XDS and subsequently scaled using Aimless software. The structure was determined by molecular replacement, using an Alphafold-predicted model as the template, and the solution was refined using Molrep. Structure refinement was performed using PHENIX⁶⁹ and Coot.⁷⁰ The atomic coordinates for the hLCN2-phosphoserine complex have been deposited in the Protein Data Bank under accession code 9JDO. Structural visualization was done using PyMOL (Schrödinger, LLC) and ChimeraX.

Fluorescence polarization (FP) assay

Fluorescence polarization assays were conducted using 18:1 NBD PS (Sigma), which was dissolved in chloroform. Purified hLCN2 or mLCN2 proteins were diluted to a concentration of 50 μM and incubated with 50 nM 18:1 NBD PS in a total volume of 100 μL. Serial dilutions of the proteins were performed, and three parallel experiments were conducted for each dilution. The samples were excited at 485 nm and fluorescence was measured at 525 nm using a SpectraMax M5 plate reader (Molecular Devices) at 20°C. Data were analyzed and fitted using a 1:1 binding model. For competitive binding assays with Semapimod, the drug was added at concentrations of 50 nM or 500 nM to the experimental setup, maintaining the same experimental methodology and analysis.

NBD-PS lung imaging

Mice were i.n. administered NBD-labeled phosphatidylserine (NBD-PS) and sacrificed at the indicated time points after administration. Lungs were harvested immediately, embedded in OCT compound, and processed for frozen sectioning. Lung sections were scanned using a 3D Histech digital slide scanner. To ensure comparability among samples, all sections were imaged using identical acquisition settings, including exposure time, gain, and scanning parameters. Representative fluorescence images were acquired at 40× magnification, and whole-slide overviews were generated using the slide overview function. For fluorescence quantification, one random 40× field was selected from each lung lobe. NBD-PS fluorescence intensity was measured using ImageJ software, and all images were analyzed using identical threshold and quantification settings. Each data point represents an individual lung lobe, and statistical analyses were performed using measurements from all lung lobes across all mice.

Western blot

Cells from various experimental groups were lysed using RIPA buffer (Thermo), supplemented with phosphatase inhibitors (Beyotime, P1081) and protease inhibitors (Selleck, B14001). The protein supernatants were boiled for 8 min with 5× protein loading buffer (Biosharp, BL502B). Equal protein volumes were separated on 4–20% precast SDS-PAGE gels (Genscript, M00656) and transferred onto PVDF membranes. The membranes were blocked with 5% milk for 1 h at room temperature, followed by overnight incubation with primary antibodies (1:1000 dilution). The antibodies used were: β-actin (2D4H5, Proteintech, 66009), JAK1 (6G4, Cell Signaling, 3344), pJAK1 (Cell Signaling, 3331), STAT1 (D1K9Y, Cell Signaling, 14994), pSTAT1 (58D6, Cell Signaling, 9167), STAT3 (124H6, Cell Signaling, 9139), pSTAT3 (D3A7, Cell Signaling, 9145), Erk1/2 (137F5, Cell Signaling, 4695), and pErk1/2 (Cell Signaling, 9101). After three washes with TBST, membranes were incubated for 1 h with HRP-conjugated secondary antibodies (Proteintech, RGAR001 and RGAM001) at room temperature. Following three additional washes, the membranes were developed using SuperSignal West Atto Ultimate Sensitivity Substrate (Thermo, A38554). Images of the uncropped blots are provided in the Supplementary Information.

Protein-lipid binding assay

Lipid binding to LCN2 was assessed using Membrane Lipid Strips (Echelon Biosciences, P-6002). The strips were blocked for 1 h in 1× PBS with 0.1% Tween 20 and 3% fatty acid-free bovine serum albumin (Yeasen, 36104ES25). The strips were incubated overnight at 4°C with 2 μg/mL of His-LCN2 in PBST buffer containing 3% BSA. The following day, the strips were washed three times with PBST, and HRP-conjugated anti-His antibody (Yeasen, 30403ES40) was added for 1 h at room temperature. After washing, the strips were developed with SuperSignal West Atto Ultimate Sensitivity Substrate (Thermo, A38554). His-MBP protein was used as a negative control.

Serine injection and in vivo phosphatidylserine imaging

Tumor formation in the lungs was induced by intravenously injecting the E0771-Luc cell line. Three days after inoculation, mice received an i.p. injection of 100 mg/kg L-serine or the corresponding solvent as a control. On the 20th day, PSVue550 (Molecular Targeting Technologies, P-1005, 10 μM final concentration) was injected i.p. into both groups of mice. Fifteen minutes after injection, the mice were euthanized, and the lungs, livers, kidneys, and spleens were harvested. These tissues were imaged using an IVIS Spectrum small animal imaging system. Fluorescence intensity was quantified using Living Image 4.4 software.

RNA-seq and data analysis

For bulk tumor RNA sequencing, lung tissues from tumor-bearing mice were snap-frozen in liquid nitrogen and ground into a powder. Total RNA was extracted using a Trizol reagent kit (Invitrogen) following the manufacturer’s protocol. For RNA sequencing of cell lines, 5×10⁵ cells from each group were directly lysed, and total RNA was extracted using Trizol. RNA quality was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, USA) and verified by RNase-free agarose gel electrophoresis. mRNA was enriched using Oligo(dT) beads, fragmented, and reverse transcribed into cDNA using a NEBNext Ultra RNA Library Prep Kit (NEB #7530, New England Biolabs, USA). After adapter attachment, cDNA was amplified and sequenced on an Illumina NovaSeq6000 platform.

Reads were aligned to the reference genome (GRCm38 for mouse, GRCh38 for human) using Hisat2 (v.2.2.4). Differential expression analysis was performed using the edgeR R package, with genes exhibiting a false discovery rate (FDR) < 0.05 and fold change ≥2 considered differentially expressed (DEGs). These DEGs were ranked by log₂(fold change) values and subjected to Gene Set Enrichment Analysis (GSEA) using the clusterProfiler (v.4.10.1) package, focusing on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) datasets. Visualization was performed using ComplexHeatmap (v.2.18.0) and ggplot2 (v.3.5.1) in R (v.4.3.3). For LCN2 silencing and Semapimod treatment, genes with p-values <0.05 were selected to generate a Venn diagram and calculate the Pearson correlation coefficient, followed by GSEA and Over Representation Analysis (ORA).

scRNA-seq and data analysis

Single-cell suspensions were prepared as previously described. For the WT or Lcn2^−/− E0771 lung metastasis tumors, CD45⁺ or CD45⁻ cells were sorted by flow cytometry and recombined in a 2:1 ratio. In the case of vehicle or serine-injected groups, CD45⁺ cells were sorted using positive magnetic beads. Each experimental group consisted of two biological replicates. The resulting cell suspensions were processed on a 10× Genomics GemCode single-cell platform to generate single-cell Gel Bead-In-Emulsion droplets. cDNA libraries were then constructed using the Chromium Next GEM Single Cell 3′ Reagent Kit v3.1, followed by high-throughput sequencing on an Illumina NovaSeq 6000 PE150 platform.

Downstream analysis was performed in Python (v.3.11) using the Scanpy (v.1.10.2) library. Cells exhibiting fewer than 100 detected genes, over 20% mitochondrial gene content, or more than 25,000 gene counts were excluded as low-quality, while genes expressed in fewer than three cells were omitted. Additionally, doublets were identified and removed using the Scrublet tool.⁷¹ To mitigate batch effects, we integrated data from multiple samples using the scVI-tools library. The integrated data were based on the top 2,000 highly variable genes (identified using the Seurat v3 methodology), and clustering was performed using the top 50 principal components, visualized via UMAP. Subpopulation annotation was conducted based on Leiden clustering results. Trajectory analysis of neutrophil subclusters was performed using scVelo (v.0.3.2). Enrichment analysis of NK cells was carried out via Over-Representation Analysis (ORA) using the Decoupler (v.1.6.0) tool, with pathways derived from MSigDB.⁷² Differential expression within specific cell populations was performed using the sc.tl.rank_genes_groups function in Scanpy (method = “t test”), with Benjamini–Hochberg correction applied through the built-in multiple-testing correction procedure. Adjusted p-values were used for significance evaluation.

Molecular docking

Docking studies of hLCN2 were performed using Chimera. Missing hydrogen atoms and side chains were added to create a complete receptor model. The coordinates of the small molecules were retrieved from PubChem (CID: 5745214). Molecular docking was conducted using Gnina software, and the binding model exhibiting the lowest binding energy score was selected for further analysis.

Nuclear magnetic resonance (NMR) titration

For NMR titration, the samples consisted of 50% D₂O, 20 μM purified hLCN2 protein, and 400 μM ligand. Saturation Transfer Difference (STD)⁷³ and Water-Ligand Observed via Gradient Spectroscopy (waterLOGSY)⁷⁴ experiments were performed on an Agilent 700 MHz spectrometer equipped with cryo-cooled probes at 25°C. Data were processed and visualized using ACD/Labs software.

Thermal shift assay

Recombinant hLCN2 was mixed with 10× SyproTM Orange (Thermo) in 20 mM HEPES (pH 7.5) and 150 mM NaCl buffer. The mixtures were aliquoted into 384-well reaction plates (Corning). Thermal shift assays were conducted using a Real-Time PCR system (LightCycler 480, Roche). The temperature was gradually increased in 0.1°C increments from 20°C to 98°C, and melting temperatures (T_m) were determined using the LightCycler 480 Analysis Software.

Ferroptosis induction and cell viability assay

Cell viability was assessed using a Cell Counting Kit-8 (CCK8, TargetMol, C0005). Briefly, 10,000 tumor cells were seeded in 96-well plates and allowed to adhere for 24 h. Ferroptosis agonists, RSL3 (1 μM, Selleck, S8155) and Erastin (5 μM, Selleck, S7242), Semapimod, the ferroptosis inhibitor Ferrostatin-1 (10 μM, Selleck, S7243), and a co-treatment group were then added. After 8 h of compound treatment, CCK8 reagent was added and incubated for 2 h at 37°C. Absorbance at 450 nm was measured to assess cell viability. Cell viability was calculated using the following formula: cell viability = light absorption value (with compounds)/mean light absorption value (without compounds).

Public data analysis

Human lipoprotein-related proteins were retrieved from the UniProt Knowledgebase (UniProtKB), filtered for “reviewed” entries, yielding 1,187 proteins. An UpSet plot was generated to visualize the subcellular location distribution of these proteins. The CIBERSORT analytical tool was used to estimate immune cell infiltration (including Neu/Treg/M1 macrophage/NK/CD8⁺ T/CD4⁺ T/M2 macrophage) in the TCGA lung cancer cohort (LUAD and LUSC, 1,121 samples). The correlation between the expression of lipid-related genes and immune cell infiltration was determined using Pearson correlation in Python. Genes identified from the enrichment analysis were further subjected to tumor-versus-normal expression analysis using matched TCGA samples, and relative expression levels were calculated from normalized RNA-seq data. Cox proportional hazards regression was employed to calculate the Hazard Ratio (HR) in a public anti-PD-1 cohort (n = 520), using the survminer (v.0.4.9) and survival (v.3.7-0) packages in R.

Single-cell RNA-seq data from human samples were sourced from public datasets: Zenodo7227571. The Zenodo7227571 dataset, a high-resolution single-cell atlas of non-small cell lung cancer, was provided as a preprocessed and annotated h5ad file.

Following the previous method, ORA was performed on NK cells and neutrophils, calculating the mean enrichment scores for selected pathways from the MSigDB human gene set. Specifically, “WP_INTERFERON_TYPE_I_SIGNALING_PATHWAYS” was used to assess the interferon-stimulated gene (ISG) signature in neutrophils, while “KEGG_NATURAL_KILLER_CELL_MEDIATED_CYTOTOXICITY” and several HALLMARK pathways were used to assess NK cell function. Neutrophils in the Zenodo7227571 dataset were divided into two groups based on median enrichment scores for the 'WP_INTERFERON_TYPE_I_SIGNALING_PATHWAYS.' The top 100 marker genes from the high-enrichment group were defined as the “Neutrophil ISG signature” to represent ISG^high neutrophils. This gene set was subsequently used for ORA in neutrophil populations from other public single-cell RNA-seq datasets. Pearson correlation coefficients and significance levels were calculated and visualized using SciPy and Seaborn libraries in Python.

Quantification and statistical analysis

All statistical analyses were conducted using GraphPad Prism (v.8.0), Python (v.3.11.0), and R (v.4.3.3). Unless otherwise noted, comparisons between two groups were assessed using an unpaired, two-tailed Student’s t test. Data are presented as mean ± standard error of the mean (SEM.). For tumor growth curve comparisons, two-way analysis of variance (ANOVA) was employed. Survival data were analyzed using the log rank (Mantel-Cox) test. p-values of <0.05 were considered statistically significant. Specific p-values are indicated on the corresponding graphs, with significance levels marked as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005, ∗∗∗∗p < 0.0001, and NS denoting non-significance. In animal experiments, each data point represents an individual mouse, while in clinical samples, each data point corresponds to a single donor. Unless otherwise stated, all experiments were performed at least twice, and consistent results were obtained across replicates.

Published: August 28, 2026