Senescent-Like Neutrophils Shape Angiogenic Immunosuppressive Niches in Colorectal Cancer Liver Metastasis

Single-cell and spatial profiling of colorectal cancer liver metastasis identifies a glycolytic, senescent-like BHLHE40⁺ neutrophil subset that drives angiogenesis and recruits immunosuppressive macrophages via CCL3L1–CCR1.

Abstract

Neutrophils are major players in innate immunity. However, their landscape and functions in colorectal cancer liver metastasis (CRLM) remain poorly understood. In this study, using single-cell RNA sequencing and spatial-enhanced-resolution-omics sequencing, we provide a comprehensive transcriptional landscape of tumor-associated neutrophils (TAN) in CRLM. Our analysis reveals that a terminally differentiated protumor neutrophil subset (TAN1), characterized by a glycolysis signature and a senescent phenotype, is significantly enriched in liver metastasis and associated with poor prognosis. Mechanistically, TAN1 arises from other TAN subsets through the upregulation of BHLHE40, driven by glucose deprivation in the metastatic microenvironment. Functionally, TAN1 promotes angiogenesis via VEGFA and recruits immunosuppressive macrophages through potential CCL3L1–CCR1 signaling, thereby fostering an angiogenic immunosuppressive niche. As a result, neutrophil-specific knockout of Bhlhe40 in mice significantly promotes antitumor immunity and suppresses tumor growth. Our data uncover the critical role of BHLHE40⁺ senescent-like neutrophils in shaping the immunosuppressive microenvironment of CRLM.

Significance:

This study identifies a protumoral BHLHE40⁺ senescent-like neutrophil subset in CRLM that orchestrates vascular immunosuppressive niches to drive tumor progression, highlighting a potential target for neutrophil-specific therapeutic intervention.

See related commentary by Droin and Garner, p. 1958

Introduction

Colorectal cancer ranks as the fourth leading cause of cancer-related mortality globally (1). Approximately 50% of patients with colorectal cancer develop colorectal cancer liver metastasis (CRLM), which is responsible for more than 70% of colorectal cancer–related fatalities (2). Although surgery intervention is the curative approach for CRLM, recurrence occurs in more than 50% patients within 2 years (3). Despite the transformative impact of immune checkpoint blockade therapies in various cancers (4), their efficacy in CRLM remains limited (5). Therefore, a deeper understanding of the metastatic microenvironment could provide novel insights for advancing immunotherapy strategies in CRLM.

Neutrophils are a critical component of innate immunity, vastly outnumbering any other immune cell in human blood and playing a pivotal role in swiftly defending against pathogens (6). As an important component of tumor immune infiltrates, tumor-associated neutrophils (TAN) play a dual role in regulating the tumor microenvironment (TME), acting as both defenders and potential contributors to tumor progression (7). For a long period of time, TANs have been considered to contribute to tumor progression through various mechanisms (8), including extracellular matrix remodeling (9), driving angiogenesis (10), and mediating immunosuppression (11–13). However, recent work has shown that neutrophils can also stimulate T-cell responses or produce interferon (IFN)-related immunostimulatory effects (14–16). Recent research highlights that neutrophils play a far more complex role than previously thought (17). Although, in CRLM, a previous study has reported that neutrophils are more enriched in liver metastasis compared with primary lesions (18), given the complexity and heterogeneity of neutrophils, the functional diversity and underlying regulating mechanism of TANs in CRLM remain elusive.

To unravel the landscape and functional diversity of neutrophils in CRLM, single-cell profiling is particularly well suited. However, most single-cell RNA sequencing (scRNA-seq) approaches struggle to achieve high-throughput profiling of neutrophils due to their short lifespan, vulnerability in vitro, and relatively lower mRNA content (19). Deconvoluting the diversity of TANs at single-cell level and linking this intricate information to neutrophil function and patient prognosis remains challenging. In this study, to address this challenge, we used a neutrophil profiling strategy using a microwell-based scRNA-seq platform optimized for capturing low-transcript cells and successfully carried out deep characterization of more than 50, 000 neutrophils from six patients with CRLM. Integrating this with spatial transcriptomics and transgenic mouse models, we delineated the phenotypic landscape, functional diversity, and regulatory circuitry of TANs in CRLM. We identified a terminally differentiated TAN subset with glycolytic and senescent features, enriched in liver metastases and linked to poor prognosis. This pathogenic subset arises via BHLHE40-driven reprogramming under glucose-deprived conditions and orchestrates angiogenic immunosuppressive niches. Neutrophil-specific Bhlhe40 deletion significantly enhances antitumor immunity and inhibits CRLM progression.

Results

Clinical Implications and Functional Role of Neutrophils in CRLM

To systematically evaluate neutrophil infiltration dynamics in CRLM, we first performed a comparative immune profiling of paired primary colorectal cancers and liver metastases. Quantitative analysis revealed a significant enrichment of neutrophils in metastatic lesions compared with primary tumors (Fig. 1A and B), which was further validated by immunofluorescence staining (Fig. 1C and D). Notably, neutrophil infiltration was significantly higher in stage IV colorectal cancer specimens than in earlier stages (I–III; Fig. 1E), with approximately a 1.5-fold increase in neutrophil density observed in metastatic colorectal cancer compared with nonmetastatic colorectal cancer (Fig. 1F). These findings suggest a potential association between neutrophil infiltration and CRLM prognosis. To further assess the prognostic significance of neutrophils in CRLM, we stratified patients (n = 127, detailed clinicopathologic information is presented in Supplementary Table S1) into two groups based on the median myeloperoxidase (MPO) density in liver metastases: high neutrophil infiltration (MPO-high, n = 64) and low neutrophil infiltration (MPO-low, n = 63). Patients with high neutrophil infiltration exhibited significantly worse relapse-free survival and overall survival following radical surgery compared with those with low infiltration (Fig. 1G), as confirmed by multivariate Cox analysis (n = 127; Fig. 1H and I).

Figure 1.

Clinical implications and functional role of neutrophils in CRLM. A and B, Flow cytometry analysis comparing immune cell composition (A) and neutrophil proportions (B) between primary colorectal cancer (n = 6) and CRLM (n = 6). C, Representative immunofluorescence staining of MPO showing neutrophil infiltration in colorectal cancer and CRLM tissues. Scale bars, 100 µm. D, Quantification of neutrophil density in paired colorectal cancer (n = 52) and CRLM (n = 52) tissues. E, Neutrophil density across different colorectal cancer stages: stage I (n = 8), stage II (n = 37), stage III (n = 32), and stage IV (n = 38). F, Comparison of neutrophil density between nonmetastatic colorectal cancer (nmCRC) (n = 77) and metastatic colorectal cancer (mCRC) (n = 72). G, Kaplan–Meier curves of relapse-free survival (RFS) and overall survival (OS) in patients with CRLM with high or low MPO density (n = 127). H and I, Multivariate Cox regression analysis of RFS (H) and OS (I) in patients with CRLM (n = 127). CRS, clinical risk score. J, Schematic of the CRLM mouse model. C57BL/6 mice received intrasplenic and orthotopic rectal injections of mouse colorectal cancer cells (MC38), followed by tail vein injection of PBS (n = 6) or isolated mouse neutrophils (n = 6). Tumors were harvested on day 14. K, Representative images of liver metastases in both groups. L, Comparison of liver weights between the two groups. M, Flow cytometry analysis showing neutrophil fractions in liver metastases from both groups. N, IHC staining of Ly6G to quantify neutrophil density in liver metastases from both groups. Scale bars, 50 µm. O, Representative image of orthotopic rectal tumors in both groups. P, Comparison of orthotopic rectal tumor weights between the two groups. Q, Flow cytometry analysis showing neutrophil fractions in orthotopic rectal tumors from both groups. R, IHC staining of Ly6G to quantify neutrophil density in orthotopic rectal tumors from both groups. Scale bars, 50 µm. P values were calculated by paired Student t test (A, B, and D), unpaired Student t test (E, F, L, M, N, P, Q, and R), or log-rank test (G). n.s., not significant; *, P < 0.05; ***, P < 0.001. CRC, colorectal cancer; NE, neutrophil.

Next, we explored whether neutrophils contribute to CRLM progression. We performed tail vein injection of bone marrow–derived neutrophils in mouse models of CRLM (intrasplenic and orthotopic rectal injection of MC38 cells; Fig. 1J) and found that neutrophil adoptive transfer significantly promoted metastasis of tumor cells to the liver and enhanced tumor growth (Fig. 1K and L; Supplementary Fig. S1A), accompanied by markedly increased neutrophil infiltration at liver metastasis sites (Fig. 1M and N). In contrast, no tumor-promoting effect was observed in the orthotopic rectal implantation model (Fig. 1O and P), with no significant change in neutrophil infiltration in rectal tumors following neutrophil injection (Fig. 1Q and R). Consistent with our observations in the MC38 model, neutrophil adoptive transfer also promoted liver metastasis in mouse models of CRLM established with another mouse colorectal cancer cell line, CT26 (Supplementary Fig. S1B–S1H). Conversely, administration of anti-Ly6G antibody in CRLM mouse models suppressed liver metastasis progression (Supplementary Fig. S1I–S1L), consistent with our observations in the CT26 model (Supplementary Fig. S1M–S1P). To validate the role of neutrophils in another model, we established a liver orthotopic implantation by directly injecting MC38 cells into the liver and found that neutrophil adoptive transfer promoted the growth of established liver tumors (Supplementary Fig. S1Q–S1T). Collectively, these findings highlight a pathogenic role of neutrophils in promoting CRLM.

Enrichment of a Senescent-Like Neutrophil Subset with Enhanced Glycolytic Activity in the CRLM Microenvironment

To comprehensively characterize neutrophil populations in the CRLM microenvironment, we performed microwell-based scRNA-seq on samples from six patients with CRLM, including peripheral blood (PB, n = 6), liver metastases (n = 6), adjacent liver parenchyma (LP, n = 6), and primary colorectal tumors (n = 5; Fig. 2A). Five of these six CRLM samples were synchronous and patient-matched with corresponding primary colorectal cancer tumors, LP tissues, and PB samples. One patient with CRLM is metachronous and has paired liver metastasis, LP tissues, and PB samples. To validate our findings, we additionally performed spatial-enhanced-resolution-omics sequencing (Stereo-seq) on 10 CRLM tissues and 3 colorectal cancer tissues. The detailed clinicopathologic information is provided in Supplementary Table S2. Given the short lifespan, in vitro fragility, and relatively low mRNA content of neutrophils, we isolated neutrophils via fluorescence-activated cell sorting (FACS) prior to scRNA-seq. These neutrophils were then mixed with the remaining viable cells to enrich the neutrophil fraction for the sample. The neutrophil-enriched samples were subsequently used for scRNA-seq. After quality control, 79,285 cells were retained for further analysis. Based on gene expression profiles, we identified 11 major cell types, including 52,198 neutrophils (Supplementary Fig. S2A–S2C). Unsupervised clustering was performed separately on 16,937 neutrophils from tissues samples and 35,261 neutrophils from PB samples (Fig. 2B and C; Supplementary Fig. S2D). Five TAN subsets, namely TAN0_ELL2, TAN1_LGALS3, TAN2_HSPA1B, TAN3_IFIT2, and TAN4_ARG1, were identified across colorectal cancer, LP, and CRLM samples based on their gene expression profiles (Fig. 2D). Additionally, three polymorphonuclear neutrophil (PMN) subsets, namely PMN0_CXCR4, PMN1_S100A12 and PMN2_IFIT1, were identified in PB samples (Supplementary Fig. S2E). PMN subsets exhibited transcriptional similarity and were predominantly enriched in pathways related to granulocyte migration, regulation of inflammatory response, and antiviral defense (Supplementary Fig. S2F and S2G). Due to their transcriptional similarity, we consolidated these PMN subsets into a single group for comparison with TAN subsets. In contrast, TAN subsets displayed greater transcriptional heterogeneity (Supplementary Fig. S2F). Pathway enrichment analysis further revealed distinct functional characteristics among them (Fig. 2D; Supplementary Fig. S2H). TAN0_ELL2, characterized by high expression of SYAP1 and ELL2, was enriched in bacterial response pathways. TAN1_LGALS3, expressing VEGFA, LGALS3, LDHA, and BHLHE40, was associated glycolysis and angiogenesis. TAN2_HSPA1B, with elevated HSPA1B expression, was enriched in protein folding pathways. TAN3_IFIT2, characterized by high ISG15, IFIT3, and IFIT2 expression, was enriched in antiviral response pathways. TAN4_ARG1, with high ARG1 and S100A12 expression, was linked to lipopolysaccharide response pathways. We also compared our TAN subsets with neutrophil clusters from published CRLM scRNA-seq datasets (18, 20) and found that TAN3_IFIT2 corresponds to the IFN-stimulated neutrophils described in these studies, whereas TAN1_LGALS3 was also identified in these datasets (Figure S3A-D). Next, we examine the distribution of TAN subsets and found that TAN1_LGALS3 was significantly enriched in CRLM compared with colorectal cancer or adjacent liver tissues of CRLM (Fig. 2E), which was further validated using the published datasets (Supplementary Fig. S3E and S3F) and spatial transcriptomics analysis (Fig. 2F and G). Moreover, spatial transcriptomics revealed no significant differences in the distribution of any TAN subsets between the invasive margin and tumor core in CRLM (Supplementary Fig. S4A). These results suggest the potential role of TAN1_LGALS3 in CRLM progression.

Figure 2.

Enrichment of a senescent-like neutrophil subset with enhanced glycolytic activity in the CRLM microenvironment. A, Schematic illustrating scRNA-seq of neutrophils from patients with CRLM. B and C, UMAP plots showing five TAN subsets (B) and three PMN subsets (C). D, Heatmap showing marker gene expression across TAN and PMN subsets. E, Comparison of the abundance of each TAN subset in colorectal cancer, adjacent liver tissues (LP), and CRLM tissues. F, H&E staining and spatial expression of the TAN1 signature in representative colorectal cancer and CRLM Stereo-seq samples. G, Quantification of spatial expression of the TAN1 signature in colorectal cancer and CRLM Stereo-seq samples. H, Comparison of glycolysis signature scores across all TAN and PMN subsets, calculated by scMetabolism. I, Comparison of senescence signature scores across all TAN and PMN subsets. J, Comparison of SID5 scores across all TAN and PMN subsets, calculated by SenCID. K, Dot plot showing expression of senescence-associated secretory phenotype (SASP) genes in all TAN and PMN subsets. L, Expression levels of p21 in each TAN and PMN subset. M, Immunofluorescence staining of p21 and CD66b to quantify p21⁺ neutrophil density in colorectal cancer (n = 5) and CRLM (n = 5) tissues. Scale bars, 50 µm. N, Flow cytometry analysis of SA-β-Gal activity in LGALS3⁺ and LGALS3⁻ neutrophils from patients with CRLM (n = 3). O, IHC staining of each TAN subset (TAN0: CD66b and IL1B; TAN1: CD66b and LGALS3; TAN2: CD66b and HSPA1B; TAN3_IFIT2: CD66b and IFIT2; and TAN4: CD66b and ARG1) and Kaplan–Meier curves of relapse-free survival (RFS) in patients with CRLM stratified by the median density of each TAN subset (n = 102). Scale bars, 50 µm. P values were calculated by Wilcoxon rank-sum test (E, G, H, I, J, and L), paired Student t test (M), unpaired Student t test (N), or log-rank test (O). n.s., not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. CRC, colorectal cancer; MFI, mean fluorescence intensity.

To define the characteristics of TAN1_LGALS3, we first analyzed its metabolic profile because TAN1_LGALS3 was closely associated with glycolysis-related pathways as mentioned above. Using the scMetabolism algorithm, we found that TAN1_LGALS3 exhibited a significantly higher glycolysis signature score than other TAN subsets, suggesting enhanced glycolytic activity of TAN1_LGALS3 (Fig. 2H). Notably, TAN1_LGALS3 highly expressed BHLHE40, a transcription factor closely linked to cellular senescence (Fig. 2D). Consistently, TAN1_LGALS3 exhibited a significantly higher senescence signature score than other TAN subsets (Fig. 2I), validated by the SenCID method (Fig. 2J; ref. 21). It also exhibited elevated expression of senescence-associated secretory phenotype genes (IL1A, SPP1, CXCL8, and CSTB; Fig. 2K) and another senescence-regulating gene CDKN1A (encoding p21; Fig. 2L). Further multiplex immunohistochemistry (mIHC) staining confirmed an increased infiltration of p21⁺ neutrophils in CRLM compared with colorectal cancer (Fig. 2M). To further validate the senescent phenotype of TAN1_LGALS3, we assessed senescence-associated β-galactosidase (SA-β-Gal) activity using flow cytometry analysis and found that LGALS3⁺ neutrophils exhibited higher SA-β-Gal activity compared with LGALS3⁻ neutrophils in CRLM (Fig. 2N). These results collectively support the senescent phenotype of TAN1_LGALS3. To assess its clinical relevance, we performed IHC staining of the corresponding TAN subsets and found that only TAN1_LGALS3 was significantly associated with poor prognosis in patients with CRLM (Fig. 2O; Supplementary Fig. S4B). Taken together, these findings highlight TAN1_LGALS3 as a pathogenic neutrophil subset in CRLM, characterized by high glycolytic activity and senescence-like features.

Tissue Specification of Neutrophils in the CRLM Microenvironment

The protumoral role of TAN1_LGALS3 prompted us to investigate how this subset accumulates in the CRLM microenvironment. We first examined chemokine receptor expression in neutrophils and found that TAN1_LGALS3 exhibited high CXCR4 expression but low CXCR1 and CXCR2 levels (Fig. 3A), with significantly more CXCR4⁺ neutrophils infiltrating CRLM than colorectal cancer (Fig. 3B and C). However, cell–cell interaction analysis revealed weak chemokine-related signaling for TAN1_LGALS3, suggesting that this subset may arise through tissue specification rather than active recruitment (Fig. 3D). To explore this possibility, we used CytoTRACE and Monocle 2 to map neutrophil differentiation trajectories and found that PMNs appeared at the initial stage, followed by TAN4_ARG1 and TAN0_ELL2, whereas TAN1_LGALS3 occupied a terminal state in CRLM (Fig. 3E–G). Further analysis of pseudotime-associated genes categorized them into four groups: C4 genes (early-stage) were enriched in neutrophil migration and chemotaxis pathways; C3 genes (mid-stage) were involved in lymphocyte activation; C2 genes (late-mid stage) were mainly enriched in mRNA processing and splicing pathways; and C1 genes, including TAN1_LGALS3 markers BHLHE40, VEGFA, and LDHA, were upregulated in the late-stage and linked to hypoxia response and protein folding in the endoplasmic reticulum (Fig. 3H). In addition, by analyzing our PB neutrophil data along with a public bone marrow neutrophil dataset (22), we found that TAN1-like population was not detectable in either bone marrow or PB (Supplementary Fig. S5A and S5B). These results suggested that neutrophils at the final stage of differentiation in CRLM acquire a TAN1_LGASL3 phenotype. Furthermore, we performed TAN1 similarity scoring for other TAN subsets originating from different tissues and found that TAN subsets from liver metastases exhibited a stronger similarity to TAN1 compared with those from primary colorectal tumors and adjacent liver tissues, indicating a phenotypic shift toward the TAN1_LGASL3 state in CRLM (Fig. 3I). To validate these findings, we cocultured PB-derived neutrophils with the colorectal cancer cell lines DLD-1 and HCT116 in vitro and observed significant upregulation of BHLHE40, CDKN1A (encoding p21), LGALS3 [encoding galectin-3 (Gal3)], VEGFA, and LDHA (Fig. 3J). Flow cytometry further confirmed increased Gal3 expression on the neutrophil surface after coculture (Fig. 3K and L). These findings indicate that PB-derived neutrophils can undergo phenotypic conversion into the TAN1_LGALS3 subset upon interaction with tumor cells. Collectively, these findings support tissue specification as a key mechanism for TAN1_LGALS3 accumulation in the CRLM microenvironment.

Figure 3.

Tissue specification of neutrophils in the CRLM microenvironment. A, Expression levels of CXCR4, CXCR2, and CXCR1 across each TAN and PMN subset. B, mIHC staining of MPO, CXCR1, CXCR2, and CXCR4 in colorectal cancer and CRLM tissues. Scale bar, 50µm. C, Quantitative analysis of the percentage of CXCR1⁺, CXCR2⁺, and CXCR4⁺ neutrophils in paired colorectal cancer and CRLM tissues (n = 5). D, Heatmap showing the mean strength of chemokine–receptor axis for each TAN subset in CRLM tissues. E, Predicted ordering of each TAN subset by CytoTRACE analysis using scRNA-seq dataset. F, Monocle 2 trajectory analysis of TAN subsets differentiation using the scRNA-seq dataset. Cells are colored by pseudotime, ordering predicted by CytoTRACE, and cell types. G, The density of each TAN subset along pseudotime. H, Clustering analysis of gene expression along pseudotime, with pathway enrichment analysis of four gene sets based on pseudotime. I, TAN1 similarity scores of TAN0_ELL2, TAN2_HSPA1B, TAN3_IFIT2, and TAN4_ARG1 across colorectal cancer, adjacent liver tissues (LP), and CRLM tissues. The TAN1 similarity score was calculated based on the top 50 differentially expressed genes in TAN1. J, Relative mRNA expression levels of BHLHE40, CDKN1A (p21), LGALS3, VEGFA, and LDHA as determined by qPCR in control neutrophils or neutrophils cocultured with HCT116 or DLD-1 cells. K and L, Flow cytometry analysis of Gal3 surface expression in neutrophils cultured alone (control) or cocultured with DLD-1 (K) or HCT116 cells (L). P values were determined by paired Student t test (C) or unpaired Student t test (J, K, and L). n.s., not significant; **, P < 0.01; ***, P < 0.001. CRC, colorectal cancer; Ctrl, control; MFI, mean fluorescence intensity.

BHLHE40 Drives the Protumoral Polarization of Neutrophils in the CRLM Microenvironment

We next explored how other TAN subsets transition into the protumoral TAN1_LGALS3 phenotype in the CRLM microenvironment. To identify key regulators of this protumoral phenotype, we applied the SCENIC algorithm, which revealed that TAN1_LGALS3 exhibited high transcriptional regulatory activity of BHLHE40 (Fig. 4A). To validate its role, we established BHLHE40 knockdown and overexpression dHL60 neutrophil-like cell lines and found that BHLHE40 knockdown significantly reduced the expression of LGALS3, VEGFA, LDHA, and CDKN1A (Fig. 4B), whereas overexpression led to a marked increase in their expression (Fig. 4C). Furthermore, chromatin immunoprecipitation (ChIP) assay confirmed the direct binding of BHLHE40 to the promoter regions of these genes (Supplementary Fig. S6A). Collectively, these results suggest that BHLHE40 directly regulates a protumoral transcriptional program in neutrophils.

Figure 4.

BHLHE40 drives the protumoral polarization of neutrophils in the CRLM microenvironment. A, Transcription factor analysis of each TAN subset using the SCENIC algorithm based on scRNA-seq data. B, qPCR analysis of relative gene expression levels of BHLHE40, LGALS3, VEGFA, LDHA, and CDKN1A in dHL60 cells with BHLHE40 knockdown (sh1, shBHLHE40-1; sh2, shBHLHE40-2; sh3, and shBHLHE40-3) compared with negative control (NC). C, qPCR analysis of relative gene expression levels of BHLHE40, LGALS3, VEGFA, LDHA, and CDKN1A in dHL60 cells with BHLHE40 overexpression (OE) compared with NC. D, Bar plot displaying the top five enriched pathways in CRLM tumor cells compared with colorectal cancer tumor cells based on scRNA-seq dataset. KEGG gene sets were used. E, Comparison of glycolysis signature scores between colorectal cancer and CRLM tumor cells, calculated by scMetabolism. F, Comparison of glucose and oxygen uptake scores between colorectal cancer and CRLM tumor cells, calculated by METAFlux. G, Hypoxia signature scores for each TAN and PMN subset. H, TAN subset distribution and spatial expression of hypoxia and glycolysis signatures in a representative CRLM Stereo-seq sample. I, Quantification of TAN subset-enriched spots in high- or low-scoring spots for hypoxia and glycolysis signatures (n = 10). J, Western blot analysis of BHLHE40 protein expression levels in PMNs cultured in control or low-glucose medium. β-Actin served as the loading control. K, Flow cytometry analysis of Gal3 surface expression in PMNs cultured in control or low-glucose medium. L, Measurement of glycolytic rate in PMNs cultured in control or low-glucose medium. M, Heatmap and histogram showing open chromatin regions in PMNs under control or low-glucose conditions, showing fragment mapping intensity across the genome, as determined by ATAC-seq. N, Chromatin accessibility at the BHLHE40 locus in PMNs cultured with control or low-glucose medium, as determined by ATAC-seq. P values were calculated by unpaired Student t test (B, C, and K) and Wilcoxon rank-sum test (E, F, G, and I). ***, P < 0.001. CRC, colorectal cancer; Ctrl, control; MFI, mean fluorescence intensity.

Given the fact that neutrophil phenotypes in the TME can be influenced by tumor-derived signals, we investigated potential mechanisms driving BHLHE40 upregulation in TAN1_LGASL3. scRNA-seq revealed that compared with primary colorectal cancer tumor cells, liver metastasis tumor cells exhibited higher expression of genes involved in glycolysis and cell-cycle pathways, indicative of increased malignancy, proliferation, and metabolic activity (Fig. 4D and E). METAFlux analysis further confirmed that, compared with primary colorectal tumor, liver metastasis tumor cells had greater glucose and oxygen uptake, suggesting enhanced glycolytic activity and increased consumption of glucose from the CRLM microenvironment (Fig. 4F). Consistently, TAN1_LGASL3 displayed a higher hypoxia score than other TAN subsets (Fig. 4G; Supplementary Fig. S6B). Spatial transcriptomic analysis further showed that TAN1_LGASL3 was found at higher frequencies in regions with high glycolysis or hypoxia signature scores compared with other TAN subsets in CRLM (Fig. 4H and I; Supplementary Fig. S6C), suggesting that TAN1_LGASL3 preferentially localizes to regions of hypoxia and glycolytic activity.

To test whether glucose deprivation induced by tumor cells drives BHLHE40 upregulation, we cultured neutrophils from healthy donor blood under either normal or glucose-deprived conditions for 24 hours. Western blot analysis confirmed a significant upregulation of BHLHE40 protein in neutrophils under glucose-deprived conditions (Fig. 4J). Flow cytometry further demonstrated elevated Gal3 protein expression under these conditions (Fig. 4K). Glycolytic rate assay revealed enhanced glycolysis activity in neutrophils exposed to glucose deprivation (Fig. 4L). In addition, Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) analysis showed increased chromatin accessibility in neutrophils under glucose deprivation, particularly in the BHLHE40 promoter region (Fig. 4M and N), which may account for the observed upregulation of BHLHE40. Taken together, these findings highlight the crucial role of BHLHE40 in regulating the TAN1_LGASL3 phenotype in response to glucose deprivation within the CRLM microenvironment.

Proangiogenic and Immunosuppression Phenotype of TANs in the CRLM Microenvironment

Given the protumoral characteristics of the TAN1_LGASL3 subset following metastasis-induced neutrophil reprogramming, we aimed to further investigate its functional role in CRLM progression. Neutrophils are known to modulate the TME through various mechanisms, including promoting angiogenesis, recruiting myeloid cells, mediating immunosuppression, facilitating myeloid-derived suppressor cell (MDSC) differentiation, exerting cytotoxic effects, presenting antigens, producing reactive oxygen species (ROS), and forming neutrophil extracellular traps (NET). To explore these functions, we first performed in vitro assays to assess ROS production, NET formation, and T-cell suppression. However, we observed no significant differences in NET formation, ROS release, or direct inhibition of CD8⁺ T-cell cytotoxicity by HCT116-cultured neutrophils (TAN1-like neutrophils) compared with control neutrophils (Supplementary Fig. S7A–S7D). To specifically examine the functions of TAN1_LGALS3, we analyzed functional gene signatures across neutrophil subsets using scRNA-seq data. TAN1_LGALS3 exhibited significantly elevated scores for angiogenesis and myeloid cell recruitment, accompanied by high expression of key cytokines and chemokines, including VEGFA, CXCL8, CCL3L1, and CCL4 (Fig. 5A–C). To validate the proangiogenic role of this subset, we performed IHC staining of the vascular endothelial marker CD31 in liver metastasis tissues. The expression level of CD31 was found to be positively associated with the abundance of TAN1_LGALS3 but not with other TAN subsets (Fig. 5D and E; Supplementary Fig. S7E). To assess the role of TAN1_LGALS3 in angiogenesis, we performed in vitro assays and found that LGALS3⁺ neutrophils promote tube formation of human umbilical vein endothelial cells (HUVEC) compared with LGALS3⁻ neutrophils (Fig. 5F). Collectively, these findings indicate that TAN1_LGALS3 promotes CRLM progression through proangiogenic mechanisms within the liver metastatic microenvironment.

Figure 5.

Proangiogenic and immunosuppression phenotype of TANs in the CRLM microenvironment. A, Heatmap showing signature scores for angiogenesis, myeloid cell recruitment, neutrophil cytotoxicity, MDSC, and NET across each TAN and PMN subset. B, Comparison of angiogenesis, myeloid cell recruitment, MDSC, and NET signature scores among TAN and PMN subsets. C, Dot plot illustrating expression levels of major cytokines and chemokines in each TAN and PMN subset. D, Representative IHC images of CD31 staining in CRLM tissues. Scale bars, 50 µm. E, Correlation analysis between CD31 and TAN1 density using Spearman’s rank correlation (n = 102). F, Representative images and quantification of tube formation in HUVECs when cocultured with conditioned medium from human LGALS3⁻ neutrophils or LGALS3⁺ neutrophils (n = 6). G, UMAP plot showing six monocyte/macrophage subsets across colorectal cancer, adjacent liver tissues, CRLM, and PB samples. H, Shell-like plot depicting ligand–receptor interaction strength between TAN1_LGASL3 and neutrophil/monocyte/macrophage subset in CRLM. I, Selected ligand–receptor interactions between TAN1_LGASL3 and neutrophil/monocyte/macrophage subset in CRLM. J, Representative mIHC images of CD68, CCL3L1, CD66b, CCR1, and BHLHE40 in CRLM. Scale bars, 50 µm. K, Representative images of macrophage migration assays and effects of conditioned medium from human LGALS3⁺ neutrophils on migration of macrophages (induced from THP-1 cells) in the presence of CCR1 inhibitor (BX471) or anti-CCL3 antibody (n = 6). L, Comparison of immunosuppressive macrophage signature scores between colorectal cancer and CRLM. M, Schematic of the CRLM mouse model. C57BL/6 mice received intrasplenic injections of mouse colorectal cancer cells (MC38). Mice were subsequently treated with IgG control (IgG; n = 6), anti-Ly6G antibody (anti-Ly6G; n = 6), or anti-Ly6G plus anti-CD8a antibodies (anti-Ly6G + anti-CD8a; n = 6). Livers were harvested on day 14. N, Representative images of liver metastases from IgG, anti-Ly6G, and anti-Ly6G + anti-CD8a groups. O, Comparison of liver weights among the three groups. P values were calculated by the Wilcoxon rank-sum test (B and L) and unpaired Student t test (F, K, and O). *, P < 0.05; **, P < 0.01; ***, P < 0.001. CRC, colorectal cancer; Ctrl, control.

Given the fact that TAN1_LGALS3 exhibited elevated scores for myeloid cell recruitment, we further explore the potential cross-talk between TAN1_LGALS3 and other myeloid cell populations in the CRLM microenvironment and reclustered monocytes and macrophages, identifying six distinct subsets: Macrophage_C1QB, Macrophage_CD5L, Macrophage_SPP1, Macrophage_EREG, Monocyte_FCGR3A, and Monocyte_FCN1 (Fig. 5G). Cell–cell interaction analysis using the CellChat algorithm revealed strong communication between the TAN1_LGALS3 subset and several myeloid subsets, including TAN4_ARG1, Monocyte_EREG, Macrophage_SPP1, and Macrophage_C1QB (Fig. 5H). Further analysis indicated that TAN1_LGALS3 potentially recruit these monocytes and macrophages via CCL3L1–CCR1 signaling axis (Fig. 5I), which was validated by mIHC staining (Fig. 5J). To validate the role of CCL3L1–CCR1 signaling in macrophage recruitment, we performed in vitro assays and found that treatment with CCR1 inhibitors (BX471) or anti-CCL3L1 neutralizing antibodies in conditioned medium from LGASL3⁺ neutrophils impaired macrophage chemotaxis (Fig. 5K). In addition to interacting with monocytes and macrophages, TAN1_LGALS3 also exhibited the ability to recruit other TAN subsets via CXCL8–CXCR2 and CXCL8–CXCR1 signaling, suggesting a potential self-reinforcing loop that promotes the accumulation of TAN1_LGALS3. Functional signature scoring further revealed that macrophages in CRLM exhibited elevated immunosuppressive signatures compared with those in colorectal cancer (Fig. 5L), which was accompanied by reduced T-cell activity and cytotoxicity in CRLM tissues (Supplementary Fig. S7F–S7H). To assess the role of T cells in mediating the effects of neutrophil targeting, we performed antibody-mediated depletion of CD8⁺ T cells in the neutrophil depletion mouse model of CRLM and found that neutrophil depletion had a limited effect on liver metastases following CD8⁺ T-cell depletion (Fig. 5M–O; Supplementary Fig. S7I). These findings suggest the role of the TAN1_LGASL3 subset in fostering an immunosuppressive TME.

Given the close interplay between neutrophils, macrophages, and vascular endothelial cells mentioned above, we performed mIHC staining and observed that CD68⁺ macrophages, BHLHE40⁺p21⁺ neutrophils, and CD31⁺ vascular endothelial cells co-localized to form spatial niches in CRLM, which were rarely detected in colorectal cancer tissues (Fig. 6A and B). This finding was further validated by Stereo-seq analysis (Fig. 6C and D). Spatial transcriptomic analysis revealed that regions in close proximity to TAN1_LGASL3 were enriched for immunosuppressive macrophage and angiogenesis signatures (Fig. 6E and F). To directly investigate the role of TAN1_LGALS3, we performed tail vein injections of isolated mouse LGALS3⁺ neutrophils or LGALS3⁺ neutrophils in mouse models of CRLM. Supplementation with LGALS3⁺ neutrophils exerted stronger protumoral effects on liver metastasis compared with LGALS3⁻ neutrophils (Fig. 6G and H). Moreover, IHC staining revealed a significant increase in neutrophils, macrophages, and vascular endothelial cells, accompanied by a marked reduction in CD8⁺ T-cell infiltration (Fig. 6I). Taken together, these results highlight the pivotal role of TAN1_LGALS3 in orchestrating the formation of vascular immunosuppressive niches that support CRLM progression.

Figure 6.

TAN1_LGALS3 promotes immunosuppressive vascular niches in the CRLM microenvironment. A, Representative mIHC images of CD68, p21, CD66b, BHLHE40, and CD31 in colorectal cancer and CRLM tissues. Scale bars, 50 µm. B, Quantification of niche density between colorectal cancer and CRLM (n = 24). A niche was defined as a region within 100 μm containing CD68⁺macrophages, p21⁺BHLHE40⁺CD66b⁺neutrophils, and CD31⁺vascular endothelial cells. C, H&E staining, cell type, and subset distribution in a representative CRLM Stereo-seq sample. D, Quantitative analysis showing changes in proportions of major immune cell types relative to the proximity of TAN1. The spatial resolution between adjacent pseudo-cells was 25 µm. E, Quantitative analysis showing changes in angiogenesis and immunosuppressive macrophage signature scores relative to the proximity of TAN1^high or TAN1^low neutrophils in CRLM. 95% confidence intervals were shown. The spatial resolution between adjacent pseudo-cells was 25 µm. F, Boxplot showing angiogenesis and immunosuppressive macrophage scores of cells adjacent to TAN1^high and TAN1^low neutrophils. Overlapped cells between the two groups were excluded. G, Schematic of the CRLM mouse model. C57BL/6 mice received intrasplenic injections of mouse colorectal cancer cells (MC38), followed by tail vein injection of PBS (Ctrl; n = 6), isolated mouse LGALS3⁺ neutrophils (TAN1; n = 6), or LGALS3⁻ neutrophils (non-TAN1; n = 6). Tumors were harvested on day 14. H, Comparison of liver weights among Ctrl, TAN1, and non-TAN1 groups. I, Representative images and quantification of IHC staining for Ly6G, CD68, CD8, and CD31 in liver metastases from Ctrl, TAN1, and non-TAN1 groups. Scale bars, 50 µm. P values were calculated by paired Student t test (B), unpaired Student t test (H and I), and Wilcoxon rank-sum test (F). *, P < 0.05; **, P < 0.01; ***, P < 0.001. CRC, colorectal cancer; Ctrl, control.

BHLHE40⁺ Neutrophil Depletion Attenuates CRLM Progression

To evaluate whether targeting BHLHE40 in neutrophils could inhibit CRLM progression, we generated a neutrophil-specific Bhlhe40 conditional knockout (cKO) mouse model using the Cre-loxP system (Fig. 7A). The specific downregulation of Bhlhe40 expression in neutrophils was confirmed by flow cytometry analysis (Fig. 7B). Then, we established a CRLM model via intrasplenic injection of the MC38 cell line (Fig. 7C). Mice with neutrophil-specific Bhlhe40 deletion showed significantly reduced tumor weight compared with wild-type (WT) controls (Fig. 7D and E). IHC staining and flow cytometry analysis revealed a marked reduction in macrophages, neutrophils, and vascular endothelial cells, along with a significant increase in CD8⁺ T-cell infiltration (Fig. 7F–I), indicating enhanced T cell–mediated antitumor immunity. Consistent with the intrasplenic injection model, neutrophil-specific Bhlhe40 deletion in the liver orthotopic implantation model also significantly suppressed the growth of established liver tumors and increased CD8⁺ T-cell infiltration (Supplementary Fig. S8A–S8D).

Figure 7.

Neutrophil-specific KO of BHLHE40 suppresses CRLM progression and restores antitumor immunity. A, Schematic illustration of the generation of neutrophil-specific Bhlhe40-KO mice (Mrp8-Cre; Bhlhe40^flox/flox). B, Flow cytometry analysis of Bhlhe40 expression in immune cell populations from WT and Bhlhe40 cKO mice groups (n = 6). C, Schematic of the CRLM mouse model. WT (n = 6) and cKO (n = 6) mice were subjected to intrasplenic injection of mouse colorectal cancer cells (MC38). Liver tumors were harvested on day 14. D, Comparison of liver weights between WT and cKO groups. E, Representative images and quantification of H&E staining of liver metastases in WT and cKO groups. F, Representative images and quantification of IHC staining for Ly6G, CD68, CD8, and CD31 in liver metastases from WT and cKO groups. Scale bars, 50 µm. G–I, Flow cytometry analysis showing the proportions of neutrophil (CD11b⁺Ly6G⁺; G), macrophage (CD11b⁺F4/80⁺; H), and CD8⁺T cell (I) in liver metastases from WT and cKO groups. J, UMAP plots showing major immune cell types in liver tumors from WT and cKO mice. K, Tissue distribution of major immune cell types in WT and cKO groups. L, Comparison of naïve T-cell and T-cell cytotoxicity signature scores between WT and cKO groups. M, Dot plot showing the expression of immunosuppressive macrophage-associated genes in WT and cKO groups. N, UMAP plot identifying three mouse TAN (mTAN) subsets in mouse liver metastases. O, Heatmap showing marker gene expression across the identified mTAN subsets. P, Tissue distribution of the three mTAN subsets in WT and cKO groups. Q, Proportions of each mTAN subsets in WT and cKO groups. R, Comparison of senescence scores, glycolysis signature scores, and Vegfa expression levels among the three mTAN subsets. P values were determined by unpaired Student t test (B, D, E, F, G, H, and I), Wilcoxon rank-sum test (L and R), and χ² test (K and P). *, P < 0.05; **, P < 0.01; ***, P < 0.001. MFI, mean fluorescence intensity.

To further explore the impact of neutrophil-specific Bhlhe40 deletion on the CRLM immune microenvironment, we performed scRNA-seq on tumor tissues from three WT and three Bhlhe40-KO mice. Based on canonical marker expression, five major immune cell populations were identified: neutrophils, monocyte/macrophages, B cells, dendritic cell (DC), and T cells (Fig. 7J; Supplementary Fig. S8E and S8F). Cell composition analysis revealed a pronounced reduction in neutrophils and a concurrent increase in T and B cells in Bhlhe40-KO tumors compared with WT tumors (Fig. 7K), accompanied by enhanced higher T-cell activity and cytotoxicity (Fig. 7L). Additionally, monocytes and macrophages from Bhlhe40-KO tumors displayed diminished immunosuppressive phenotypes, with downregulated expression of Il10, Spp1, and Arg1 (Fig. 7M). Further clustering of neutrophils identified three distinct subsets: mTan0_S100a9, mTan1_Hsp90ab1, and mTan2_Lgals3 (Fig. 7N and O). Notably, the mTan2_Lgals3 subset was specifically enriched in WT tumors (Fig. 7P and Q) and exhibited elevated glycolytic activity, senescence signatures, VEGFA expression, closely resembling the human TAN1_LGASL3 subset observed in CRLM (Fig. 7R). Collectively, these findings demonstrate that neutrophil-specific BHLHE40 deletion disrupts the formation of immunosuppressive neutrophil subsets, reprograms the TME, and restores effective antitumor immunity in CRLM.

Discussion

The complex immune regulation network within the TME is a crucial determinant of tumor development and progression. Innate immunity plays a major role in regulating the TME. However, the role of TANs in CRLM microenvironment has remained unclear. In this study, high-quality single-cell data revealed that TAN1_LGALS3, a terminally differentiated protumor neutrophil subset, with upregulation of BHLHE40, activation of glycolysis-related genes, and a senescent phenotype, was significantly enriched in CRLM compared with primary sites. Spatial analysis showed that TAN1_LGALS3 exhibited proangiogenic activity through VEGFA and recruited immunosuppressive macrophages via the CX3CL1–CCR1 axis, thereby fostering a vascular immunosuppressive niche and promoting CRLM progression. The combination of single-cell multiomics in vitro and in vivo assays provided comprehensive insights into the subset composition, intercellular interactions, and biological functions of neutrophils in CRLM, highlighting their role in mediating immunosuppression and promoting tumor progression.

Neutrophils play critical roles in cancer, exhibiting both protumoral and antitumoral functions (23). Recent studies have been explored neutrophil biology in tumors using single-cell technology. Xue and colleagues (13) identified two protumoral TAN subsets in primary liver cancer: CCL4⁺ TANs, which recruit macrophages, and PD-L1⁺ TANs, which suppress T-cell cytotoxicity. Another study highlighted dcTRAIL-R1⁺ neutrophils with proangiogenic properties that promote tumor growth (10). A pan-cancer analysis uncovered an antitumor, antigen-presenting neutrophil subset capable of triggering both antigen-specific and antigen-independent T-cell responses (17). Additionally, a BHLHE40⁺ neutrophil subset was found to directly suppress CD8⁺ T cells and correlate with poor prognosis in pancreatic cancer (24). However, the diversity and role of neutrophils in metastatic cancers remain poorly understood. In this study, we used a neutrophil profiling strategy combining microwell-based scRNA-seq and flow cytometry enrichment and identified a protumoral BHLHE40⁺ senescent-like neutrophil subset in liver metastases. Of note, this neutrophil subset exhibits proangiogenic activity and fosters immunosuppressive niches with macrophages, facilitating liver metastasis. Our high-resolution neutrophil landscape highlights the pivotal role of BHLHE40⁺ neutrophils in CRLM.

The metabolic interplay between tumor and immune cells significantly influences immune cell phenotypes and shapes the tumor-immune microenvironment (25, 26). Previous study has shown that reduced glucose consumption by hepatocellular carcinoma cells creates a high-glucose microenvironment, enhancing T-cell cytotoxicity (27). In neutrophil, previous study has shown that tumor-infiltrating neutrophils converge into dcTRAIL-R1⁺ reprogrammed neutrophils within hypoxic–glycolytic niches, accompanied by alterations in metabolic and oxidative stress-related genes (10). Another study reported that hypoxia induces a glycolytic switch in neutrophils, enhancing their protumoral functions in pancreatic cancer (24). Additionally, apolipoprotein E secreted by prostate tumor cells binds TREM2 on neutrophils, driving their senescence and promoting an immunosuppressive phenotype (28). Moreover, another study identified that a distinct subset of glycolytic neutrophils, recruited by splenic stromal cells via the CCL9–CCR1 axis, induce glucose deprivation and anergy in T cells, thereby impairing antitumor immunity (29). In our study, enhanced glucose metabolism of tumor cells in liver metastases resulted in a glucose-depleted microenvironment, leading to glucose starvation in TANs. This triggered the upregulation of BHLHE40 in TANs, along with the increased expression of downstream glucose metabolism–related genes and senescence-related genes, ultimately driving a protumoral neutrophil phenotype. These findings suggest that targeting glycolysis in TME may offer a therapeutic approach to suppress protumoral activities in CRLM.

The BHLHE40 has emerged as a key regulator of immunity, influencing cytokine secretion in T cells, macrophage proliferation, and mitochondrial metabolism in tumor-infiltrating lymphocytes (30–32). A previous study using scRNA-seq has shown that BHLHE40 is highly expressed by T_H1-like tumor-infiltrating CD4⁺ T cells in microsatellite-unstable colorectal tumors (33). However, the role of BHLHE40 in neutrophils remains poorly understood. A recent study revealed that BHLHE40 overexpression in neutrophils upregulates genes such as VEGFA, LGALS3, LDHA, PLAU, and PDE4B (24). In our study, glucose starvation–induced upregulation of BHLHE40 in neutrophils led to the increased expression of glycolysis-related genes (e.g., LDHA) and senescence-related genes (e.g., p21), driving a protumoral neutrophil phenotype in liver metastases. Notably, neutrophil-specific KO of BHLHE40 significantly suppressed liver metastases and enhanced antitumor immunity. These findings provide strong evidence supporting targeting BHLHE40 in TANs as a therapeutic strategy for CRLM.

Overall, we identified a protumoral BHLHE40⁺ senescent-like TAN subset in CRLM. Mechanically, this subset, driven by the enhanced glucose metabolism of liver metastasis tumor cells, promotes angiogenesis and fosters immunosuppressive niches with macrophages, thereby accelerating CRLM progression. Our findings highlight the potential of targeting innate immunity for developing cancer therapies.

Methods

Human Specimens

Cohort 1, used for retrospective analysis, included 127 patients with CRLM and 110 patients with colorectal cancer without metastasis who underwent radical resection between 2014 and 2024 at the First Affiliated Hospital of Sun Yat-sen University. All CRLM cases were confirmed by pathologic diagnosis. Patients with colorectal cancer included in this cohort were followed for a minimum of 5 years postoperatively and remained metastasis-free throughout the follow-up period. Patients with extrahepatic metastasis or concurrent malignancies of other origins were excluded. Clinicopathologic data were collected retrospectively from medical records, with detailed information summarized in Supplementary Table S1.

Cohort 2, utilized for scRNA-seq and Stereo-seq, comprised 14 patients with CRLM who underwent radical resection between August 2022 and January 2024 at the First Affiliated Hospital of Sun Yat-sen University. Surgical decisions were made by a multidisciplinary team. Prior to surgery, all patients provided written informed consent. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of the First Affiliated Hospital of Sun Yat-sen University. For scRNA-seq, a total of six fresh PB mononuclear cell (PBMC) samples, six liver metastasis tumor samples, six adjacent liver tissue samples (located >1 cm from the tumor edge), and five primary colorectal cancer tumor samples from six patients were included. For Stereo-seq, 3 primary colorectal cancer tumor samples and 10 liver metastasis tumor samples from 10 patients were utilized. Detailed clinicopathologic information of these patients is provided in Supplementary Table S2.

Cell Lines

The human colon cancer cell lines HCT116 (RRID: CVCL_0291) and DLD-1 (RRID: CVCL_0248), mouse colon cancer cell line CT26 (RRID: CVCL_7254), as well as the human neutrophil cell line HL-60 (RRID: CVCL_0002) were obtained from the American Type Culture Collection. The mouse colon cancer cell line MC38 (RRID: CVCL_B288) was purchased from FuHeng Biology. HUVECs (RRID: CVCL_2959) were obtained from Meisen Chinese Tissue Culture Collections. All cell lines were authenticated via short tandem repeat analysis and confirmed to be Mycoplasma-free. MC38, CT26, and DLD-1 cells were cultured in RPMI 1640 medium (Gibco, C11875500) supplemented with 10% fetal bovine serum (FBS; Pricella, 164210) and 1% penicillin–streptomycin (PenStrep; Gibco, 15140122). HCT116 cells were cultured in McCoy's 5A medium (Pricella, PM150710) supplemented with 10% FBS and 1% PenStrep. HL-60 cells were maintained in Iscove’s modified Dulbecco’s medium (Gibco, 12440053) with 10% FBS and 1% PenStrep. HUVECs were cultured in EGM-2 Endothelial Cell Growth Medium-2 BulletKit (Lonza Bioscience, CC-3162). All cell lines were cultured at 37°C with 5% CO₂ in a culture incubator.

Mouse Strains

Female C57BL/6 mice, 4 to 6 weeks of age, female BALB/c mice, 4 to 6 weeks of age, Mrp8-Cre mice, and Bhlhe40^flox/flox mice were obtained from GemPharmatech, Co., Ltd., and were housed under pathogen-free conditions at Sun Yat-sen University. All animal experiments were conducted in accordance with protocols approved by the Institutional Care and Animal Use Committee of Sun Yat-sen University.

IHC and mIHC Staining

Formalin-fixed, paraffin-embedded (FFPE) tissue blocks from patients with CRLM were sectioned into 4-µm slices. For IHC staining, slides were deparaffinized, rehydrated, and subjected to antigen retrieval in ethylenediaminetetraacetic acid (EDTA) buffer (Solarbio, C1034) via high-pressure treatment. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide. After blocking with 10% goat serum (Solarbio, SL038), slides were incubated overnight with primary antibodies at 4°C. Secondary antibody incubation was followed by 3,3’-diaminobenzidine (DAB) (Aglient Technologies, K5007) detection and hematoxylin counterstaining. Double staining was performed using a double-stain IHC kit (M&R on human tissue, DAB & AP/Red, DS-0004, Zsbio). Slides were scanned using KF-PRO-020 Digital Slide Scanner (KFBIO). The antibodies used for IHC staining are listed in Supplementary Table S3.

mIHC staining was performed according to the manufacturer’s protocol using the TSA Multiplex IHC Assay Kits (TissueGnostics, TGT5C100). Briefly, slides were baked, dewaxed, rehydrated, and fixed in 10% neutral formalin. Antigen retrieval was performed in EDTA buffer using microwave heating. After cooling, protein blocking was performed with 10% goat serum, followed by incubation with primary antibodies for 30 minutes at 37°C. Horseradish peroxidase (HRP)–conjugated secondary antibodies for mouse (Abcam, ab205719, RRID: AB_2755049) or rabbit (Abcam, ab6721, RRID: AB_955447) were incubated for 10 minutes at room temperature. Tyramide signal amplification with Opal fluorophores was applied for 10 minutes. Antigen retrieval was repeated to remove the antibody–tyramide signal amplification (TSA) complex, and the cycle was restarted. After four to five cycles for planed antibodies, the TSA-stained slides were then stained with 4’,6-diamidino-2-phenylindole (DAPI) and mounted using mounting medium (Panovue, 0022001010). Stained slides were scanned using the TissueFAXS SPECTRA imaging system (TissueGnostics). Image processing, as well as cell density and mutual distance analysis, was performed using StrataQuest v.7.0.165 (TissueGnostics) following the manufacturer’s guidelines. The staining sequence, antibodies, and TSA buffers are listed in Supplementary Table S4.

Flow Cytometry

Single-cell suspensions from human or mouse tissues were prepared as previously described (34). To exclude dead cells, samples were stained with a fixable viability dye (BD Biosciences, 565694). Murine cells were labeled with anti-mouse CD45, CD11b, Ly6G, Ly6C, Gal3, F4/80, CD3e, CD4, CD8a, CD19, NK1.1, FOXP3, BHLHE40, CD86, and CD206. Human cells were labeled with anti-human CD45, CD11b, CD15, CD16, CD3, CD8, CD4, CD19, CD68, CD25, CD56, CD14, CD11c, and human leukocyte antigen-DR. After staining, cells were fixed with 1% paraformaldehyde. Data were acquired using a flow cytometer (Cytek Biosciences, Aurora) and analyzed by FlowJo 10.8 software (TreeStar). The antibodies used for flow cytometry are listed in Supplementary Table S3.

Quantitative Reverse Transcriptase Polymerase Chain Reaction

Total RNA was extracted using TRIzol reagent (Invitrogen), and 1 μg RNA was reverse transcribed into cDNA using the Reverse Transcription Master Kit (Takara Bio) according to the manufacturer’s instructions. Quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) was performed using SYBR Green (SYBR Premix Ex Taq II, Takara Bio) on a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Gene expression levels were normalized to β-actin, and relative expression was calculated using the 2^−ΔΔCt method. The primer sequences are provided in Supplementary Table S5.

Western Blot

Total protein was extracted from whole cell lysate using RIPA buffer and quantified using the Pierce BCA Protein Assay Kit (Epizyme, ZJ101) according to the manufacturer’s instructions. Equal amounts of protein (20 μg) were separated by 10% SDS-PAGE and transferred onto polyvinylidene difluoride membranes. The membranes were incubated overnight with primary antibody against human BHLHE40 (17895-1-AP, Proteintech, RRID: AB_2065351), followed by HRP-linked secondary antibodies for signal detection.

Tumor Models and Treatments

To establish the mouse CRLM model, 5 × 10⁵ MC38 or CT26 cells suspended in PBS were injected intrasplenically into C57BL/6 or BALB/c mice. Mice were weighted twice weekly and sacrificed on day 14. For the orthotopic rectal implantation model, 2 × 10⁵ MC38 or CT26 cells suspended in PBS were orthotopically implanted into the rectum of C57BL/6 or BALB/c mice, as previously described (35). Mice were weighted twice weekly and sacrificed on day 14. For the orthotopic liver implantation model, 5 × 10⁵ MC38 cells suspended in PBS were intrahepatically injected. Mice were weighted twice weekly and sacrificed on day 14. For neutrophil adoptive transfer, bone marrow neutrophils were enriched using the EasySep Mouse Neutrophil Enrichment Kit (STEMCELL Technologies, 19762) and cocultured with MC38-conditioned medium for 12 hours. The neutrophils were then washed, resuspended, and counted, and 3 × 10⁶ cells were intravenously injected every 3 days into CRLM mice. For LGALS3⁺ and LGALS3⁻ neutrophil adoptive transfer, the mouse MC38-cultured neutrophils were further sorted into LGALS3⁺ and LGALS3⁻ neutrophils using FACS. The LGALS3⁺ and LGALS3⁻ neutrophils were then washed, resuspended, and counted, and 3 × 10⁶ LGALS3⁺ or LGALS3⁻ neutrophils were intravenously injected every 3 days into CRLM mice. For neutrophil depletion, CRLM mice were treated with 200 μg anti-IgG (intraperitoneally, every 3 days, clone 2A3, Bio X Cell, RRID: AB_1107769) or 200 μg anti-Ly6G (intraperitoneally, every 3 days, clone 1A8, Bio X Cell, RRID: AB_1107721). For CD8⁺ T-cell depletion, 200 μg anti-mouse CD8a (Bio X Cell, BE0061, RRID: AB_1125541) was injected intraperitoneally twice a week. At the end of the experiment, mice were sacrificed, and livers were isolated for flow cytometry and IHC staining.

For the Bhlhe40-KO study, CRLM models were established by intrasplenically or intrahepatically injection of MC38 cells. Neutrophil-specific Bhlhe40-KO mice (Mrp8-Cre; Bhlhe40^flox/flox) and WT mice (Mrp8-Cre; Bhlhe40^wt/wt) were used as the experimental and control groups, respectively. Mice were sacrificed at 2 weeks after injection or at a humane endpoint, with those without tumors excluded from the analysis. At the end of the experiment, mice were sacrificed, and livers were isolated for flow cytometry and IHC staining.

Lentiviral Transduction and Plasmid Construction

Lentiviral vectors expressing pLKO.1 negative control shRNA and shRNA constructs targeting BHLHE40 were obtained from GeneChem, Co., Ltd. For lentivirus production, the vectors were co-transfected with the packaging plasmid pCMVDR8.9 and envelope plasmid pCMV-VSVG into HEK 293T cells using Lipofectamine 3000 reagent (Invitrogen). The resulting lentiviruses were collected and used to infect HL60 cells in the presence of 8 μg/mL polybrene (Solarbio). Stable infected cells were selected with 5 μg/mL puromycin (Solarbio) for 48 hours. For overexpression, full-length open reading frame of the human BHLHE40 gene was cloned into the pEZ-Lv201 vector (GeneChem, Co., Ltd.). The BHLHE40 overexpression and corresponding control vectors were transfected into cells using Lipofectamine 3000 according to the manufacturer’s instructions.

In Vitro Culture Assay

PMNs were isolated from the PB of healthy donors according to the manufacturer’s instructions (P9040, Solarbio) and maintained in DMEM with 10% FBS for further use. For analysis of protein expression, PMNs were cocultured with control medium, HCT116/DLD-1–conditioned medium, or low-glucose DMEM (11885084, Gibco) for 24 hours. After incubation, PMNs were collected for flow cytometry and Western blot analysis.

Tube Formation Assay of Endothelial cells

For the tube formation assay, 100 μL Matrigel was added to each well of a 96-well plate and preincubated at 37°C for 2 hours. HUVECs (1 × 10⁴) were resuspended in 200 μL of conditioned medium from human LGALS3⁺ neutrophils or LGALS3⁻ neutrophils and subsequently added to the Matrigel-coated wells. The cells were incubated at 37°C in a 5% CO₂ cell culture incubator for 4 to 6 hours. Tube formation was analyzed using an IX83 inverted microscope (Olympus) and quantified with ImageJ software.

In Vitro Macrophage Migration Assay

Macrophages (1 × 10⁵, induced from THP-1 cells) resuspended in 200 μL of complete medium containing phorbol 12-myristate-13-acetate (PMA) were seeded into 8-μm pore chambers (Corning, #3422) and placed in 24-well plates (Corning). The bottom chambers were filled with 600 μL of conditioned medium from human LGALS3⁺ neutrophils, with 1 µmol/L BX471 (MedChemExpress, HY-12080) or 1 μg/mL CCL3L1 blocking antibody (Abcam, ab259372, RRID: AB_3094596). The control wells contained conditioned medium from human LGALS3⁺ neutrophils in the bottom chambers. After 24 hours of incubation, nonmotile cells on the top surface of the filter were removed, and the motile cells on the bottom were fixed in 4% paraformaldehyde for 15 minutes, followed by crystal violet staining. Images (five fields per well) were captured using an inverted microscope system (Olympus, BX43) and analyzed with ImageJ software.

Measurement of SA-β-Gal Activity

SA-β-Gal activity was measured using the CellEvent Senescence Green Flow Cytometry Assay Kit, as previously described (36). Human LGALS3⁺ or LGALS3⁻ neutrophils from CRLM tissues were pretreated with 50 µmol/L chloroquine in fresh cell culture medium at 37°C for 4 hours to induce senescence. Subsequently, the cells were fixed and incubated with CellEvent Senescence Green Probe (Invitrogen, C10840) for 2 hours. After incubation, the cells were harvested and analyzed by flow cytometry to detect Senescence Green Probe signal.

In Vitro T-cell Suppression Assay

CD8⁺ T cells were isolated from the PBMCs of healthy donors using the EasySep Human CD8⁺ T Cell Isolation Kit (STEMCELL Technologies, 17953) according to the manufacturer’s protocol. The isolated CD8⁺ T cells were then cocultured with either PMNs or PMNs cultured with HCT116 cells, along with Dynabeads Human T-Activator CD3/CD38 (Thermo Fisher Scientific, 11131D, RRID: AB_3676329) and recombinant IL2 (PeproTech, 200-02) for 3 days. Following incubation, the cells were stained with specific antibodies for flow cytometry analysis.

ROS Detection

The oxidation-sensitive dye dichlorodihydrofluorescein diacetate (DCFDA, MedChemExpress, HY-D0940) was used to measure ROS production in PMNs or PMNs cultured with HCT116 cells. Cells were incubated at 37°C in prewarmed PBS with 10 µmol/L DCFDA for 30 minutes. After incubation, the cells were harvested and analyzed by flow cytometry.

NET Detection

To induce NET formation, PMNs or PMNs cultured with HCT116 cells were seeded in 24-well chamber slides (Corning; 2 × 10⁵ cells per well). Neutrophils were then incubated for 4 hours in cancer cell–conditioned medium or nonconditioned medium. Where indicated, PMA (100 µmol/L) was added to conditioned medium. To detect extruded DNA, PMNs were fixed with 1% paraformaldehyde for 15 minutes and stained with SYTOX Green dye (50 nmol/L) for 10 minutes. Cells were counterstained with DAPI, mounted, and observed under an Olympus FV3000 confocal fluorescence microscope.

Glycolytic Rate Measurement

The glycolytic rate assay was performed using the Agilent Seahorse 96XF device and corresponding kits, following the manufacturer’s protocol (#103344-100, Agilent Technologies). Briefly, 5  ×  10⁵ neutrophils per well (n = 6) were plated onto Cell-Tak (Corning, 354240)–coated XF96 cell plates in XF RPMI assay medium (50 μL, pH 7.4) supplemented with 11 mmol/L glucose, 1 mmol/L pyruvate, and 2 mmol/L glutamine and incubated for 10 minutes at room temperature. Basal oxygen consumption rate and glycolytic rate responses were determined, followed by treatment with 0.5 µmol/L rotenone/antimycin A and 50 mmol/L 2-deoxy-D-glucose (Seahorse XF Glycolytic Rate Assay Kit, #103344-100, Agilent Technologies). The data were analyzed by the Seahorse Wave Software (Agilent Technologies).

ChIP

ChIP assay was performed using the NovoNGS CUT&Tag 2.0 High-Sensitivity Kit (Novoprotein, N259-YH01). Briefly, cells were trypsinized and enriched using concanavalin A magnetic beads. The cell suspension was incubated overnight at 4°C with 2 μg of primary antibody against human BHLHE40 (NB100-1800, Novus Biologicals, RRID: AB_10000524). After washing, cells were incubated for 1 hour with a secondary antibody (10 μg/mL). The beads were then washed in Dig-Hisalt Buffer and incubated with protein A-Tn5 transposase in ChiTag buffer for 1 hour at room temperature. Subsequently, tagmentation was performed in tagmentation buffer for 1 hour at 37°C. The reaction was terminated by adding 10% SDS at 55°C for 10 minutes. Tagmented DNA was extracted using Extract Beads and subsequently used for PCR.

ATAC-seq

Neutrophils (5 × 10⁵) were collected, kept on ice, and resuspended in 25 μL 2× TD Buffer (Illumina NextEra kit), 2.5 μL transposase enzyme (Illumina NextEra kit, 15028252), and 22.5 μL nuclease-free water for a total reaction volume of 50 μL. The reaction was incubated for 1 hour at 37°C. DNA was purified using the QIAGEN MinElute PCR Purification Kit (28004) in a final volume of 10 μL. ATAC-seq libraries were prepared following the Buenrostro protocol (37), and sequencing was performed using NovaSeq X Plus by Gene Denovo Biotechnology.

For ATAC-seq data analysis, raw reads were processed to obtain high-quality clean reads according to the following three filtering criteria: (i) removal of adapter sequences, (ii) removal of reads containing >10% unknown nucleotides, and (iii) removal of low-quality reads with >50% bases having q value ≤20. Clean reads were aligned to the hg38 human reference genome using Bowtie 2 (RRID: SCR_016368; ref. 38) with parameters −X2000 and −m1. Reads aligning to mitochondria or chloroplasts were excluded. Duplicates were removed using Picard, and for paired reads, the alignment was adjusted (+4 bps for the + strand and −5 bps for the −strand). Peak calling was performed using MACS (RRID: SCR_013291; ref. 39) with parameters “-shift -75 -extsize 150 -nomodel -call-summits -nolambda -keep-dup all −P 0.01.” Dynamic Poisson distribution was used to calculate the P value of the specific region based on the unique mapped reads. Peaks were defined when log₂ (fold enrichment above background) ≥ 3 and −log₁₀ (P value) ≥ 3. Peak-associated genes were identified using ChIPseeker (RRID: SCR_021322; ref. 40), and peak distributions across genome regions [promoter, 5′ untranslated region (UTR), 3′UTR, exons, introns, downstream, and intergenic] were analyzed.

Single-Cell Preparation, Library Preparation, and Sequencing

Single cells were isolated from fresh tissues as previously described (41). For human colorectal cancer, CRLM, and adjacent liver tissues, surgically resected samples were cut into small pieces and enzymatically digested using the MACS Tumor Dissociation Kit (Miltenyi Biotec, 130-095-929) or Multi-Tissue Dissociation Kit (Miltenyi Biotec, 130-110-201) at 37°C for 10 to 20 minutes. The dissociated cells were filtered through 70- and 40-μm cell strainers, followed by removal of red blood cells using red blood cell lysis buffer (Miltenyi Biotec, 130-094-183). For human neutrophils, fresh PB was collected preoperatively in EDTA anticoagulant tubes, and red blood cells were removed using the same red blood cell lysis buffer. For mouse MC38 tumors, fresh tissues were mechanically ground in PBS containing 1% FBS and passed through 70-μm cell strainers to obtain single-cell suspensions.

Neutrophil enrichment was performed using FACS. Cells were first stained with fixable viability stain (Fvs; BD Biosciences) to exclude dead cells, followed by staining with human CD45, CD11b, CD15, and CD16 antibodies to identify neutrophils. FACS was performed on a flow cytometer (BD Biosciences). After FACS, single living neutrophils (Fvs⁻CD45⁺CD11b⁺CD15⁺CD16⁺) and living cells (Fvs⁻) were obtained. Then, a neutrophil-enriched sample was used for subsequent scRNA-seq.

Finally, the BD Rhapsody system was used to capture the transcriptomic information of the isolated single cells. Single-cell capture was achieved by random distribution of a single-cell suspension across >200,000 microwells through a limited dilution approach. Beads with oligonucleotide barcodes were added to saturation, so that a bead was paired with a cell in a microwell. The cells were lysed in the microwell to hybridize mRNA molecules to barcoded capture oligos on the beads. Beads were collected into a single tube for reverse transcription and exonuclease I digestion. Upon cDNA synthesis, each cDNA molecule was tagged on the 5′ end (i.e., the 3′ end of a mRNA transcript) with a unique molecular identifier (UMI) and cell barcode indicating its cell of origin. Whole-transcriptome libraries were prepared using the BD Rhapsody single-cell whole-transcriptome amplification (WTA) workflow including random priming and extension (RPE), RPE amplification PCR, and WTA index PCR. The libraries were quantified using a high-sensitivity DNA chip (Agilent Technologies) on a Bioanalyzer 2200 and the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific). Sequencing was performed by Illumina sequencer (Illumina) on a 150 bp paired-end run.

Stereo-seq Experiments

Stereo-seq of colorectal cancer and CRLM tissues was performed as described previously (41). Briefly, surgically resected colorectal cancer and CRLM tissues were collected in optical cutting tissue compound and stored at −80°C. After verifying RNA integrity, two serial sections in 7 µm were selected for hematoxylin and eosin (H&E) staining and spatial transcriptomic analysis. Eligible sections were attached to Stereo-seq chips, fixed in methanol, permeabilized with pepsin, and processed for RNA capture and reverse transcription. Following transcription, sections were digested and incubated overnight at 55°C with cDNA release buffer, then purified using DNA clean beads (Vazyme Biotech, N411), and amplified with KAPA HiFi Hotstart Ready Mix (Roche, KK2601) using 0.8 µm cDNA-PCR primers. A total of 20 ng of cDNA was fragmented using Tn5 transposase and then further amplified and purified as described above. Finally, the purified products were sequenced on a MGI DNBSEQ-T10 sequencer using a 50 + 100 bp single-end strategy.

Preprocessing of scRNA-seq Data and Quality Control

To obtain clean scRNA-seq data, we applied fastp with default parameters to filter adapter sequences and remove low-quality reads (42). Gene expression quantification and generation of the feature-barcode UMI matrix were performed using STARsolo (RRID: SCR_021542; version 2.7.10a) along with the human genome (Ensemble annotation version 104) or mouse genome mm10 (Ensemble annotation version 100; bioRxiv 2021.05.05.442755). The resulting UMI matrix was loaded into the Seurat R package (v4.3.0.1) for downstream analysis. Doublet cells were identified using DoubletFinder (v2.0.4; ref. 43), and cell clusters with conflicting coexpression of canonical genes from more than one cell type were flagged as doublets and excluded from subsequent analysis. Low-quality cells, defined as those expressing fewer than 200 genes or with more than 20% mitochondria gene expression, were also filtered out. In total, 79,285 human cells and 54,248 mouse cells were retained for subsequent analysis.

Stereo-seq Spatial Transcriptomics Data Processing

The raw Stereo-seq data were processed following the Stereo-seq Analysis Workflow (https://github.com/STOmics/SAW) with the human reference genome GRCh38. This pipeline quantified the transcripts captured at each DNA nanoball (DNB) spot and generated raw gene expression profiles. The captured spots had a diameter of 220 nm with a 500 nm distance between two adjacent points. To ensure sufficient number of genes for downstream analysis, we aggregated the expression profiles at near-single-cell resolution using the previously published bin50 strategy (44). This method formed larger pseudo-spots, each containing a 50 × 50 DNB square array (diameter 25 μm). Finally, an aggregated, spatially indexed expression matrix with 181,839 pseudo-spots from three colorectal cancer samples was generated.

Stereo-seq Spatial Transcriptomics Data Analysis

Spotlight (V1.4.1) was used to deconvolute the cell-type proportions for each pseudo-spot (45). The scRNA-seq data generated in this study, including B/plasma cell, T cell, macrophage, neutrophil, endothelial cell, epithelial cell, fibroblast, and hepatocyte, were downsampled to 100 cells per cell type and set as a reference for analyzing the CRLM samples. Notably, hepatocytes were excluded from reference data manually when we annotated the colorectal cancer samples. Finally, we define the cell type with the highest inferred cell proportion as the major cell type of each pseudo-spot. The top 50 differentially expressed genes of TAN1_LGALS3 in scRNA-seq were used to classify TAN1^high and TAN1^low neutrophils in Stereo-seq by AUCell. Neutrophils with scores above or below the median score were defined as TAN1^high or TAN1^low neutrophils, respectively.

To identify tumor areas in spatial transcriptomics data, we draw the borderlines using a semi-supervised strategy based on multiple evidence: (i) H&E staining of adjacent sections, (ii) spatial expression of epithelial cell adhesion molecule (EPCAM) and albumin (ALB), and (iii) unsupervised spots clustering using principal component analysis (PCA) and Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP).

To estimate the cell states, we performed gene set enrichment analyses using Seurat’s “AddModuleScore” function. Detailed gene sets are as follows: (i) immunosuppressive macrophage signature: CCL4, CCL13, CCL18, CCL20, CCL22, CD276, CLEC7A, CTSA, CTSB, CTSC, CTSD, FN1, IL4R, IRF4, LYVE1, MMP9, MMP14, MMP19, MSR1, TGFB1, TGFB2, TGFB3, TNFSF8, TNFSF12, VEGFA, VEGFB, and VEGFC (46). (ii) TAN1_LGALS3 signature was defined as the top 50 marker genes weighted by “avg_log₂ fold change (FC)” in our scRNA-seq analysis. (iii) Glycolysis signature genes were extracted from the Kyoto Encyclopedia of Genes and Genomes (KEGG). (iv) Hypoxia signature genes were extracted from the Molecular Signatures Database (MSigDB) of H (hallmark gene sets). (v) Angiogenesis signature: VEGFA, SND1, MTDH, ITGA5, TNF, CXCL3, ANXA3, HMGB1, HIF1A, SEMA4D, LRG1, and CHI3L1.

For spatial distribution analysis, tissues were divided into three regions: the tumor core, the invasive margin (defined as the areas extending 250 µm on both sides of the tumor border), and the peritumoral liver. The proportion of each TAN subtype spot relative to the total cell count was calculated for each region in every sample. To define high- and low-hypoxic or glycolytic regions, corresponding hallmark gene sets were utilized for enrichment analysis. Spots were then stratified into high- or low-level regions based on the median scores for hypoxia and glycolysis. Finally, the frequency of annotated TAN subtype spots within the high- versus low-scoring regions was determined.

scRNA-seq Data Analysis

Cell Clustering and Annotation

The filtered matrix of patients with CRLM was normalized and scaled using Seurat’s “NormalizeData” and “ScaleData” functions. Then, PCA was performed, and appropriate numbers of principal components were determined for unsupervised clustering based on the “ElbowPlot” function. The clustering results were further visualized using UMAP. The cell clusters were annotated as 11 major cell types based on their canonical genes, including B cell (MS4A1), plasma cell (MZB1), T cell (CD3E), DC (CD1C), macrophage (LYZ and C1QB), mast cell (KIT), neutrophil (CSF3R), endothelial cell (PECAM1), epithelial cell (EPCAM), fibroblast (COL1A1), and hepatocyte (ALB). For neutrophils, those from PBMCs were defined as PMN, whereas those derived from tumor and peritumor tissues were defined as TANs. Harmony algorithm (RRID: SCR_022206; ref. 47) was used to correct the batch effect between samples when we subclustered each major cell type.

Single cells from mouse CRLM models were clustered using a similar two-round Seurat pipeline. First, cells were delineated into seven principal types, including B cells, T cells, DCs, macrophages, neutrophils, stromal cells, and tumor cells. Subsequently, neutrophils were further subclustered into three subsets and named as mTan0_S100A9, mTan1_Hsp90ab1, and mTan2_Lgals3 based on their differential expression genes (DEG), whereas T cells were categorized into Cd4⁺ and Cd8⁺ T-cell subsets.

DEGs and Pathway Enrichment Analysis

DEGs for each cluster were identified using Seurat’s FindAllMarkers function with the parameters setting minpct >0.25 and log FC >0.25. Gene Ontology and KEGG pathway enrichment analyses of DEGs were performed using ClusterProfiler (RRID: SCR_016884; ref. 48).

Tissue Distribution of Cell Clusters

To evaluate the tissue distribution of each cell cluster, we calculated the ratio of observed-to-expected cell numbers (Ro/e) for each cluster across different tissues as previously described (34). The expected number of cells for each cell cluster in a given tissue was determined using the χ² test. A cell type was considered preferentially enriched in a specific tissue if the Ro/e > 1.

Calculation of Cell State Scores

To evaluate the cell states in patients with CRLM, AUCell was used to score gene signatures for individual cells. The gene signatures were defined based on well-known marker genes or extracted from published databases and studies, including (i) neutrophil senescence: CD83, CCRL2, ICAM1, C15orf48, CCL3, CCL4L2, CCL4, and CXCL2 (49); (ii) neutrophil cytotoxicity: NCF1, MYD88, TREM1, TUSC2, CYBB, CYBC1, NCF2, NCF4, RAC1, and RAC2; (iii) NETs: IL8, BST1, FPR2, CYP4F3, CREB5, G0S2, F3, VNN3, S100A12, MPO, HPSE, IL17A, ENTPD4, MME, SLC22A4, CEACAM3, TNFRSF10C, CLEC6A, and SELP; (iv) angiogenesis: VEGFA, SND1, MTDH, ITGA5, TNF, CXCL3, ANXA3, HMGB1, HIF1A, SEMA4D, LRG1, and CHI3L1; (v) myeloid cell recruitment: CCL3, MIF, CXCL14, CSF1, VEGFA, CCL4, and CXCL3; (vi) MDSC score: S100A9, S100A6, CD33, CD14, IFITM1, IL1B, SRGN, PROK2, IFITM2, JUNB, DUSP1, SOCS3, SELL, SELPLG, ASPRV1, IGFBP6, PLA2G7, CSF3R, CXCR2, TPD52, TSPO, GRINA, FABP5, CLEC4D, STEAP4, CCR1, LRG1, ARG2, CD84, and LIMK2; (vii) T-cell cytotoxicity score: GZMK, GZMA, PRF1, NKG7, IFNG, and GZMH; (viii) naïve T-cell score: CCR7, TCF7, LEF1, and SELL; and (ix) immunosuppressive macrophage: CD274, IDO1, LILRB4, LILRB2, TGFB1, TNFRSF18, LGALS9, TNFRSF14, and TDO2. The “hallmark hypoxia” genes were extracted from MSigDB of H (hallmark gene sets). The senescence score (SID5 score) of neutrophils was calculated using SenCID (21). The glycolysis signature score of neutrophils and tumor cells was calculated using scMetabolism (18). To characterize glucose and oxygen uptake levels based on scRNA-seq, we used the METAFlux computational methods to infer metabolic fluxes (50).

The functions of cell populations in mouse CRLM models were scored by AUCell using the following gene sets: (i) glycolysis genes for neutrophils: Eno1, Gapdh, Pkm, Slc16a3, Eno2, Hk1, Pgam1, Ldha, Pfkl, Pfkp, Tpi1, Slc2a1, Hk2, Eno3, Slc16a1, Aldoa, Pgam2, Pfkm, Pklr, and Ldhb; (ii) senescence genes for neutrophils: Igf1, Ccl15, Ccl23, Lyz, Cstb, Ctsd, Ctsl, Ctss, Calr, Creg1, Slpi, Npc2, Cd63, Sdcbp, Psap, Spp1, Il1a, Il1b, Il6, Cxcl8, Cxcl1, Fgh2, Hgf, Mmp1, Mmp3, Mmp10, Mmp13, Tnf, and Tgfb1; (iii) cytotoxic genes for Cd8⁺ T cells: Gzmk, Gzma, Prf1, Nkg7, and Ifng; (iv) naïve genes for Cd8⁺ T cells: Ccr7, Tcf7, Lef1, and Sell; and (v) immunosuppressive macrophage-related genes: Arg1, Cd74, Ptgs2, Vegfa, Ciita, Il10, Mmp12, Mif, Cd274, Mrc1, Msr1, Csf1r, Il1r2, Csf2, and Ctla4 (51).

Pseudotime Trajectory Analysis

Monocle 2 (v2.28.0; RRID: SCR_016339) analysis was used to infer the developmental pseudotime trajectory (52). Genes with high dispersion across cells were selected by “dispersionTable” function and used for dimensional reduction by the DDRTree method. We used CytoTRACE (RRID: SCR_022828; ref. 53) to help determine the root cell of the trajectory.

Cell–Cell Interaction Analysis

CellChat (v1.6.1; RRID: SCR_021946) was applied to infer and visualize cell–cell interaction between neutrophils and other cell subsets (54). Briefly, single-cell gene expression matrices from colorectal cancer, CRLM, adjacent liver, and PB samples were loaded into CellChat. The interaction strengths and communication probabilities were calculated using the “computeCommunProb” and “computeCommunProbPathway” functions. The interactions existed in less than 10 cells were filtered out.

Gene Regulatory Network Analysis

The gene regulatory networks were constructed based on our scRNA-seq data using the R package SCENIC (v1.1.5; RRID: SCR_017247; ref. 55). The matrix containing all neutrophil data was downsampled to 2,000 cells per cell type and used as the input for the SCENIC pipeline, utilizing RcisTarget motifs database (hg19).

Survival Analysis

Survival curves were generated using the Kaplan–Meier method, and statistical comparisons between groups were performed using the log-rank test. Hazard ratios and 95% confidence intervals were estimated using univariate and multivariate Cox proportional hazards models with the R package survival. The survminer R package was used to visualize the survival curves.

Statistical Analysis

Additional analysis and visualization were performed using GraphPad Prism 9.0.0 and R version 4.1.2. Group comparisons were conducted using Student t test or Mann–Whitney U test as appropriate. Categorical variables were analyzed using the χ² test or Fisher exact test. Correlations were assessed with the Spearman correlation coefficient. P values <0.05 were considered statistically significant, indicated as *, P < 0.05; **, P < 0.01; and ***, P < 0.001, with P > 0.05 denoted as not significant.

Supplementary Material

Table S1

Table S1. Clinicopathologic Characteristics of the patients in cohort 1.

Table S2

Table S2. Clinicopathologic Characteristics of the patients in cohort 2, related to Figure 2.

Table S3

Table S3. Antibodies for IHC staining and flow cytometry.

Table S4

Table S4. Schedule of antibodies and fluorophores Opal for mIHC.

Table S5

Table S5. Sequences of the primers used in real-time RT-PCR analysis.

Figure S1

Figure S1. the role of neutrophils in liver metastasis in CRLM mouse models.

Figure S2

Figure S2. Single-cell annotation in CRLM.

Figure S3

Figure S3. Characterization and tissue distribution of neutrophil subsets in CRC and CRLM from public datasets.

Figure S4

Figure S4. Localization of TAN subsets in liver metastasis.

Figure S5

Figure S5. Single-cell analysis of neutrophil markers in peripheral blood and bone marrow.

Figure S6

Figure S6. Analysis of BHLHE40 binding and hypoxia/glycolysis signature in neutrophils from CRLM.

Figure S7

Figure S7. Function of neutrophils and neutrophil-T cell interactions in CRLM.

Figure S8

Figure S8. Role of neutrophil-specific Bhlhe40 knockout in the liver orthotopic implantation mouse model.

Data Availability

Public scRNA-seq and Stereo-seq datasets that were used in this study can be found in the National Omics Data Encyclopedia via the accession number OEP00001756 (https://www.biosino.org/node/project/detail/OEP00001756), the Genome Sequence Archive via accession numbers HRA005549 (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA005549) and PRJCA026987 (https://ngdc.cncb.ac.cn/bioproject/browse/PRJCA026987), and Gene Expression Omnibus via the accession number GSE253355 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE253355). scRNA-seq, Stereo-seq, and ATAC-seq data generated in this study have been deposited at the Genome Sequence Archive at the National Genomics Data Center via the accession number PRJCA060393 (https://ngdc.cncb.ac.cn/bioproject/browse/PRJCA060393). This article does not report original code.

Authors’ Disclosures

No disclosures were reported.

Authors’ Contributions

Z. Chen: Conceptualization, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. X. Ren: Conceptualization, formal analysis, investigation, methodology, writing–original draft. Y. Zhang: Formal analysis, validation, investigation, methodology, writing–original draft. Y. Kang: Formal analysis, validation, investigation, methodology. Y. Yang: Validation, investigation, methodology. Y. Xu: Investigation, methodology. Q. Zhou: Software, formal analysis, investigation, visualization. C. Liao: Software, investigation. Q. Zeng: Software, formal analysis, visualization, methodology. X. Liu: Validation, methodology. R. Lin: Investigation, methodology. Y. Yang: Investigation. F. Luo: Validation, investigation. S. Shen: Validation, methodology. S. Peng: Supervision, validation, methodology. X. Li: Supervision, methodology. J. Long: Supervision, investigation, methodology, writing–review and editing. L. Xu: Conceptualization, resources, supervision, investigation, methodology, project administration, writing–review and editing. M. Kuang: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.