Integrative single-cell and spatial transcriptomic analysis reveals lipid metabolism–mediated macrophage heterogeneity during colorectal cancer liver metastasis progression Authors: Xin Qi, Siyu Hou, Shaozhuo Xie, Jiajia Chen Published: 2025 ABSTRACT AbstractBackgroundColorectal cancer liver metastasis (CRLM) is a leading cause of cancer-related death in patients with colorectal cancer (CRC) and remains a major clinical challenge. Emerging evidence indicates that metabolic reprogramming within the tumor microenvironment profoundly influences immune cell behavior and disease progression, yet the specific role of lipid metabolism in macrophage heterogeneity during CRLM is still unclear.MethodsWe integrated single-cell RNA sequencing and spatial transcriptomics to characterize lipid metabolic states, differentiation trajectories, intercellular communication, and spatial distribution of macrophages across normal colon, primary CRC (pCRC), and CRLM. Lipid metabolic activity was quantified using multiple scoring algorithms, pseudotime analysis was performed with the “Monocle” R package, cell–cell communication was evaluated using the “CellChat” R package, and spatial localization was inferred using the RCTD method and the “MISTy” R package.ResultsMacrophages exhibited the highest lipid metabolic activity among immune cells, and this activity was significantly elevated in CRLM compared with pCRC. Subclustering analysis identified three lipid metabolism–related macrophage subsets (CD1C⁺, CXCL10⁺, CX3CR1⁺) enriched in CRLM, among which CD1C⁺ and CXCL10⁺ subsets displayed distinct differentiation dynamics. Cell–cell communication analysis revealed metastasis-specific remodeling, including diversified VEGF signaling in CD1C⁺ macrophages and enhanced SPP1-integrin and C3–(ITGAX + ITGB2) interactions in CXCL10⁺ macrophages. Spatial transcriptomics analysis showed that lipid metabolism activity was highest in macrophage-rich regions of the normal colon, higher in epithelial and tumor regions of pCRC than in fibroblast regions, and highest in hepatocyte regions of CRLM. Moreover, CD1C⁺ macrophages predominantly localized to tumor regions in both pCRC and CRLM and maintained close associations with other immune cells.ConclusionsLipid metabolism shapes macrophage heterogeneity and function states during CRLM progression, with CD1C⁺ populations emerging as key macrophage subsets linked to metastatic niche remodeling. Introduction Colorectal cancer (CRC) is one of the most prevalent malignancies worldwide and a leading cause of cancer-related mortality. A major contributor to CRC-associated death is distant metastasis, with the liver being the most common site of secondary tumor growth [1]. Colorectal cancer liver metastasis (CRLM) occurs in nearly 50% of CRC patients during the course of the disease and is associated with poor prognosis and limited treatment options [2]. Despite advances in surgical resection, systemic chemotherapy, and targeted therapies, the five-year survival rate for patients with CRLM remains unsatisfactory [3]. Understanding the cellular and molecular mechanisms underlying liver metastasis is therefore critical for the development of more effective therapeutic strategies and for improving clinical outcomes in patients with metastatic CRC. The tumor microenvironment (TME) is a dynamic and complex ecosystem composed of immune cells, stromal components, vascular networks, and extracellular matrix, all of which collectively influence tumor progression and metastasis [4]. Recent studies further show that diverse stromal and immune cells, including lymphocytes, actively modulate tumor–stroma interactions by producing immunoregulatory cytokines that reshape tumor progression [5, 6]. In CRLM, the TME forms a supportive niche that enables disseminated CRC cells to survive, adapt, and proliferate. This metastatic outgrowth is driven by continuous crosstalk between tumor cells and their surrounding microenvironment, which involves immune regulation, tissue remodeling, and metabolic reprogramming [7]. Among the diverse immune cell populations within the TME, macrophages have emerged as critical regulators of metastatic colonization. Recruited and activated by both tumor-derived and liver-resident signals, macrophages can support metastasis through a variety of mechanisms, such as promoting angiogenesis, suppressing anti-tumor immunity, and enhancing extracellular matrix remodeling [8]. However, the molecular and spatial characteristics of macrophages throughout CRLM progression remain poorly understood, requiring deeper investigation using high-resolution approaches. Recent studies have highlighted cellular metabolism, particularly lipid metabolism, as a critical regulator of macrophage function and phenotype [9, 10]. Within the TME, macrophages undergo extensive metabolic reprogramming in response to environmental signals, enabling them to support tumor progression [11]. Changes in lipid uptake, storage, and oxidation can significantly influence macrophage activation states, cytokine production, and immunomodulatory capacity [12]. In CRC, lipid metabolism not only supports the malignant growth and metastasis of tumor cells, but is also influenced by interactions between tumor cells and the surrounding microenvironment, highlighting the pivotal role of lipid metabolic crosstalk in CRC progression [13]. For example, Sun et al. [14] demonstrated that ALKBH5 upregulates CPT1A, the rate-limiting enzyme of fatty acid oxidation (FAO), thereby enhancing FAO in macrophages and driving its M2 polarization, which in turn promotes the progression of CRC. Wu et al. [15] found that oleate-driven long-chain fatty acid metabolism triggers mTOR-dependent lipid droplet accumulation in tumor-associated macrophages in CRC. As a result, these macrophages acquire an immunosuppressive phenotype that supports tumor progression. Beyond solid tumors, disturbances in metabolic pathways are also prevalent in hematologic malignancies and systemic disorders, broadly affecting immune homeostasis and inflammatory regulation [16–18]. Collectively, these findings underscore the widespread relevance of metabolic reprogramming in shaping immune phenotypes and provide a broader biological framework for exploring lipid metabolism–driven macrophage heterogeneity in CRLM. Advances in high-resolution transcriptomic technologies have enabled a more precise characterization of the TME at single-cell and spatial levels. Single-cell RNA sequencing (scRNA-seq) allows for the identification of distinct cellular populations and their transcriptional states, revealing cellular heterogeneity that is often masked in bulk analyses [19]. Spatial transcriptomics complements this by preserving tissue architecture, enabling the mapping of gene expression to specific anatomical locations [20]. Together, these approaches offer a comprehensive strategy to investigate the functional states and spatial organization of macrophages in complex metastatic tissues. Therefore, the goal of this study was to define the lipid metabolic landscape of macrophages in CRLM using integrated single-cell and spatial transcriptomic analyses. In addition, we sought to identify lipid metabolism–associated macrophage subpopulations and to characterize their differentiation trajectories, cell–cell communication networks, and spatial distribution, thereby providing new insights into the immunometabolic mechanisms driving CRC progression. Materials and methods Data collection To investigate the cellular characteristics of CRC, two scRNA-seq datasets comprising both primary tumors and liver metastatic tumors were obtained from the Gene Expression Omnibus (GEO) database. Dataset GSE225857 includes scRNA-seq data from four primary tumor samples and two liver metastatic tumor samples, while dataset GSE178318 contains data from six primary tumor samples and six liver metastatic tumor samples. All data were publicly available and used in accordance with GEO’s data usage policies. To characterize spatial metabolic patterns in CRC, we incorporated two spatial transcriptomics datasets. The normal colon (NC) dataset was obtained from GEO database (GSE158328; sample GSM4797916), and the primary CRC (pCRC) and CRLM datasets were taken from the spatial transcriptomics resource of Wu et al. (samples ST-colon1 and ST-liver2) (http://www.cancerdiversity.asia/scCRLM/) [21]. Single-cell RNA-seq data processing scRNA-seq data were analyzed using the “Seurat” R package (version 5.1.0) [22]. Gene expression matrices and metadata were imported for each dataset. Quality control was performed by filtering out cells with fewer than 500 detected genes, more than 10,000 genes, or greater than 1.5% mitochondrial gene content. Genes expressed in fewer than 3 cells were also excluded. Following quality filtering, gene expression data were log-normalized using Seurat’s NormalizeData function. Highly variable genes were identified using FindVariableFeatures with the “vst” method. The data was subsequently scaled using ScaleData prior to dimensionality reduction. Principal component analysis (PCA) was used for dimensionality reduction, and the number of significant principal components was determined from the elbow plot. Batch effect correction across samples was conducted using the Harmony algorithm. Cells were clustered using the FindNeighbors and FindClusters functions with appropriate resolution parameters. t-distributed stochastic neighbor embedding (t-SNE) was then applied for nonlinear dimensionality reduction and two-dimensional embedding, enabling visualization of cellular distributions and cluster relationships. Cell type annotation Cell clusters derived from scRNA-seq data were annotated based on the expression patterns of canonical marker genes. Specifically, the following markers were used to define major immune cell types: T cells (CD3D, CD3E, CD3G), B cells (MS4A1/CD20, CD79A), Plasma cells (MZB1, JCHAIN, IGKC), Macrophages/Monocytes (CD68, LYZ, C1QB), and Mast cells (TPSAB1, TPSB2). Marker gene expression was visualized using FeaturePlot and DotPlot functions in “Seurat” R package (version 5.1.0), and annotations were manually curated to ensure biological accuracy. Pathway activity scoring for lipid metabolism Lipid metabolism–related genes were obtained as described in a previous study [23]. Specifically, gene sets were retrieved from the C5 Ontology Gene Sets of the Molecular Signatures Database (MSigDB, http://www.gsea-msigdb.org/gsea/msigdb). All gene sets containing the keywords “lipid,” “fatty acid,” “lipoprotein,” “cholesterol,” “triacylglycerol,” or “triglyceride” were selected and integrated to construct a comprehensive lipid metabolism–related gene list, yielding a total of 2,610 unique genes for subsequent analyses. To quantify lipid metabolism activity at the single-cell level, five different pathway scoring methods were implemented in R (version 4.3.2): ssGSEA using the “GSVA” R package (version 2.2.0), UCell with the “Ucell” package (version 2.3.0), AUCell with the “AUCell” R package (version 1.22.0), AddModuleScore via the “Seurat” package (version 5.1.0), and z-score normalization using the ScaleData function in “Seurat”, all under default parameters. These algorithms quantify pathway activity using distinct statistical frameworks: ssGSEA ranks all genes within each cell to calculate an enrichment score for the pathway; UCell and AUCell estimate pathway activation based on the enrichment of lipid metabolism genes among the top-expressed genes in each cell; AddModuleScore computes the average expression of pathway genes after subtracting control genes with comparable expression levels; and z-score normalization assesses relative pathway activity by standardizing gene expression within each cell. Applying these methods in parallel enabled robust and reliable inference of lipid metabolic activity across different computational frameworks. Classification of macrophages based on lipid metabolism activity To further investigate lipid metabolic heterogeneity within macrophages, ssGSEA-based enrichment scores derived from the lipid metabolism–related gene set were used to stratify macrophage cells into high- and low-lipid metabolism groups using the median score as the cutoff. These two groups were subsequently compared in downstream analyses, including differential expression profiling, transcription factor activity, cell–cell communication analysis, and stemness evaluation. Pseudotime analysis of macrophages To characterize the differentiation trajectory of macrophage subpopulations, pseudotime analysis was conducted using the “monocle” R package [24]. Genes expressed in at least 50 cells were first defined as the expressed gene set, from which highly variable genes were selected for trajectory inference. Dimensionality reduction was performed using the DDRTree algorithm, and cells were subsequently ordered in pseudotime using the orderCells() function. The root state was assigned to the macrophage cluster predominantly derived from normal or early-stage tissues, representing the initial differentiation state. The macrophage subsets identified from single-cell transcriptomic data were ordered along a trajectory constructed from gene expression profiles. The dynamic expression patterns of representative marker genes were then examined along the pseudotime axis to reveal transcriptional changes during macrophage state transitions. Cell-cell communication analysis Cell-cell communication analysis was conducted using the “CellChat” R package (version 1.6.1) on the GSE225857 dataset to investigate the intercellular signaling landscape of lipid metabolism-associated macrophage subsets [25]. The normalized expression matrix was converted into a CellChat object, and the human CellChatDB ligand–receptor database was used as the reference. Overexpressed genes and ligand–receptor pairs were identified using identifyOverExpressedGenes() and identifyOverExpressedInteractions(), followed by estimation of communication probability with computeCommunProb() (min.cells = 10) and aggregation at the pathway level using computeCommunProbPathway(). To quantify the directionality of signaling, incoming and outgoing communication roles were evaluated using netAnalysis_signalingRole() and netAnalysis_computeCentrality(), where outgoing signaling reflects a cell group’s ability to send ligands and transmit signals, and incoming signaling represents its ability to receive signals via expressed receptors. Communication networks were constructed separately for primary and metastatic tumor microenvironments to enable direct comparison of signaling pathways, interaction strength, and ligand-receptor pairs between the two contexts. CytoTRACE analysis Stemness analysis was conducted using the “CytoTRACE” R package (version 0.3.3) to estimate the differentiation potential of macrophage subsets with distinct lipid metabolic activity. CytoTRACE is a computational framework designed to predict cellular stemness based on transcriptional diversity inferred from single-cell RNA-sequencing data. For each cell, a CytoTRACE score ranging from 0 to 1 was calculated, with higher scores indicating greater stemness and lower differentiation status [26]. The computed scores were then used to compare stemness levels between high- and low-lipid metabolism macrophage populations. Transcription factor activity analysis Transcription factor (TF) activity was inferred using the “decoupleR” (version 2.0.1) R package [27]. Based on the curated DoRothEA regulon [28], TF activities were estimated from single-cell gene expression profiles using AUCell and consensus scoring methods. The top 25 TFs with the most variable activity between high- and low-lipid metabolism macrophages were visualized with their average activity. Spatial transcriptomics analysis Spatial transcriptomics data were analyzed using the “Seurat” R package (version 5.1.0). Raw count matrices were normalized with the SCTransform method. Dimensionality reduction and unsupervised clustering were performed to group spatial spots with similar expression profiles. Cell type annotations were assigned based on canonical marker genes in each cluster and histological features from H&E staining. Spatial distribution of clusters and gene expression patterns was visualized using SpatialDimPlot and SpatialFeaturePlot functions in Seurat. To infer the spatial distribution of cell types, Robust Cell Type Decomposition (RCTD) implemented in the “spacexr” R package (version 2.2.0), was applied to project scRNA-seq-defined populations onto Visium spatial transcriptomics data [29]. The scRNA-seq reference matrix was generated from the annotated GSE225857 dataset, and only genes shared between the single-cell and spatial datasets were retained. RCTD was run in “doublet” mode (doublet_mode = TRUE) to allow each spot to contain up to two cell types, better reflecting biological mixtures while avoiding overfitting. The maximum likelihood approach implemented in RCTD was used to compute cell-type proportions per spot, and spots were assigned to the most probable cell type or mixture. Only mappings with high confidence (posterior probability > 0.8) were retained for downstream visualization and interpretation. Spatial cell–cell interaction modeling using MISTy Spatial cell–cell interaction modeling was performed using the “MISTy (Multiview Intercellular SpaTial modeling)” R package (version 1.6.0) to infer how local cellular neighborhoods influence gene expression within the TME [30]. For each spatial transcriptomics sample, normalized expression matrices and spot coordinates were used as input. Three spatial “views” were modeled: (1) intraview, representing intrinsic gene expression of each spot; (2) juxtaview, capturing effects from immediate neighboring spots within a predefined radius (default distance = 50 μm); and (3) paraview, modeling influences from broader tissue regions using Gaussian kernel smoothing. MISTy constructs a multi-view random forest model to quantify the contribution of each view to target gene expression, and the view importance scores were computed using get_misty_results(). Only interactions with high stability across bootstraps (stability score > 0.75) and positive standardized importance were considered biologically meaningful. This framework enabled the identification of spatially coordinated niches and macrophage-associated interaction partners across NC, pCRC, and CRLM tissues. Results Lipid metabolic activity of macrophages in pCRC and CRLM To elucidate immune-cell heterogeneity between primary CRC (pCRC) tissues and their liver metastases (CRLM), we analyzed the immune-cell subsets from the GSE225857 single-cell transcriptome dataset. Dimensionality reduction and clustering at a resolution of 1.7 identified 31 cell clusters. Based on the expression patterns of canonical marker genes, these clusters were annotated into five major cell types: T cells (CD3G), Plasma cells (MZB1), Macrophages (CD68), B cells (MS4A1) and Mast cells (TPSAB1) (Fig. 1A). Comparative mapping showed distinct immune-cell distribution between pCRC and CRLM (Fig. 1B). Lipid metabolic reprogramming plays a pivotal role in CRC metastasis by supporting rapid tumor cell proliferation, reshaping the TME, and modulating immune responses [31]. Given the pivotal role of lipid metabolism in tumor-immune interactions, we employed five independent algorithms to quantify lipid metabolic activity across different immune cell types. Among these methods, ssGSEA, UCell, AUCell, and AddModuleScore consistently indicated that macrophages exhibited the highest lipid metabolic activity, whereas the z-score method showed a relatively higher lipid metabolic activity in T cells (Fig. 1C). Notably, macrophage lipid-metabolic scores were significantly higher in CRLM than in pCRC (Fig. 1D). To validate the robustness of these findings, we further analyzed lipid metabolic activity in the GSE178318 dataset. Dimensionality reduction and cell annotation identified eight clusters with distinct distributions between pCRC and CRLM (Figs. 1E-F). Consistent with the discovery cohort, ssGSEA, UCell, AUCell, and AddModuleScore identified macrophages as the cell type with the highest lipid metabolic activity, while only the z-score method exhibited a different trend (Fig. 1G). Importantly, macrophage lipid metabolic activity remained significantly higher in CRLM compared with pCRC (Fig. 1H). Therefore, these findings indicated that macrophages exhibit markedly elevated lipid metabolic activity in CRLM, as supported by the majority of computational methods. Lipid metabolic activity of immune cells in pCRC and CRLM. (A) Dimensionality reduction and clustering (resolution = 1.7) of immune cells in the GSE225857 dataset. (B) Distribution of annotated immune cell clusters across pCRC and CRLM samples in the GSE225857 dataset. (C) Lipid metabolic activity scores of immune cell types in GSE225857, calculated using ssGSEA, UCell, AUCell, AddModuleScore, and z-score algorithms. (D) Comparison of ssGSEA-derived lipid metabolic scores of immune cells between pCRC and CRLM in the GSE225857 dataset. (E) Dimensionality reduction and clustering (resolution = 1.4) of immune cells in the validation dataset GSE178318. (F) Distribution of annotated immune cell clusters across pCRC and CRLM samples in the GSE178318 dataset. (G) Lipid metabolic activity scores of immune cell types in GSE178318, calculated using ssGSEA, UCell, AUCell, AddModuleScore, and z-score algorithms. (H) Comparison of ssGSEA-derived lipid metabolic scores of immune cells between pCRC and CRLM in the GSE178318 dataset Characterization of macrophage subclusters To further investigate the heterogeneity of macrophages in CRC, we extracted macrophages from the GSE225857 dataset and performed clustering at a resolution of 0.5, yielding ten distinct subclusters (Fig. 2A). These subclusters were annotated based on marker genes identified using the FindAllMarkers function. Among them, three clusters displayed high expression of CD1C, CXCL10, and CX3CR1, respectively. Notably, these genes were included in the lipid metabolism–related gene list curated from the MSigDB database, suggesting their potential involvement in lipid metabolic processes. Comparative analysis showed that CD1C⁺, CXCL10⁺, and CX3CR1⁺ macrophage subclusters were more enriched in CRLM than in pCRC (Fig. 2B). To explore their differentiation trajectories, we conducted pseudotime analysis using the “monocle” R package, which showed that CXCL10⁺ macrophages were primarily distributed in the early-to-intermediate stage, CD1C⁺ macrophages mainly in the late stage, whereas CX3CR1⁺ macrophages were relatively few in number and did not exhibit a clear distribution pattern along the trajectory (Fig. 2C). To further characterize the transcriptional dynamics underlying these transitions, we examined the expression patterns of representative marker genes along the pseudotime trajectory. CD1C expression progressively increased toward the late stage, CXCL10 expression declined from the early-to-intermediate to late stages, while CX3CR1 expression remained at a low level across the trajectory (Fig. 2D). These findings suggest that CD1C⁺ and CXCL10⁺ macrophages may play more prominent roles in the metastatic process of CRC. Pseudotime analysis of lipid metabolism-related macrophage subsets in CRC. (A) Dimensionality reduction and clustering (resolution = 0.5) of macrophages in the GSE225857 dataset. (B) Distribution of macrophage subclusters in pCRC and CRLM. (C) Pseudotime trajectory of macrophages constructed by the “monocle” R package. (D) Gene expression patterns of CD1C, CXCL10, and CX3CR1 along pseudotime Cell–cell communication of macrophage subclusters To gain insight into the intercellular communication dynamics of lipid metabolism-related macrophages, we employed the “CellChat” R package to compare the communication patterns of CD1C⁺ and CXCL10⁺ macrophages between the pCRC and CRLM TMEs. As shown in Fig. 3A, both CD1C⁺ and CXCL10⁺ macrophage subsets exhibited extensive interactions with various immune cell types, including T cells, plasma cells, B cells, and mast cells, and the overall number of interactions increased in CRLM. However, despite this broader connectivity, the interaction strengths between CD1C⁺/CXCL10⁺ macrophages and T cells or plasma cells were reduced in CRLM (Fig. 3B). Further ligand–receptor interaction analysis revealed that CD1C⁺ macrophages exhibited active VEGF signaling toward multiple immune cell types in both pCRC and CRLM. Robust ligand–receptor pairs, including VEGFB–VEGFR1 and VEGFA–VEGFR1/VEGFR1R2, were consistently detected under both conditions. In CRLM, however, VEGF signaling became more diversified, characterized by additional interactions such as VEGFA–VEGFR2 and VEGFC–VEGFR3 (Fig. 3C). Comparatively, CXCL10⁺ macrophages showed reduced interaction strength of key VEGF ligand–receptor pairs, including VEGFB–VEGFR1 and VEGFA–VEGFR1/VEGFR1R2, with plasma cells and mast cells in CRLM compared with pCRC. However, the spectrum of VEGF-related interactions was more diverse in CRLM, indicating a shift in signaling dynamics. In addition, CXCL10⁺ macrophages in CRLM displayed enhanced interactions with T cells, characterized by stronger SPP1–(ITGAV + ITGB1) and SPP1–(ITGAV/ITGB5) signaling, as well as increased C3–(ITGAX + ITGB2) complement-mediated communication (Fig. 3D). Together, these findings indicate that lipid metabolism-related CD1C⁺ and CXCL10⁺ macrophages exhibit dynamic and context-specific intercellular communication patterns during CRC progression to liver metastasis. Cell-cell communication patterns of lipid metabolism-related macrophage subsets with other immune cells in pCRC and CRLM. (A) Number of interactions between CD1C⁺ and CXCL10⁺ macrophages and major immune cell types (B cells, plasma cells, T cells, and mast cells) within the TME of pCRC and CRLM. (B) Interaction strength between CD1C⁺ and CXCL10⁺ macrophages and major immune cell types (B cells, plasma cells, T cells, and mast cells) within the TME of pCRC and CRLM. (C) Ligand-receptor interactions of CD1C⁺ macrophages with CXCL10⁺ macrophages, B cells, T cells, and NK cells in the TME of pCRC (left) and CRLM (right). (D) Ligand-receptor interactions of CXCL10⁺ macrophages with CD1C⁺ macrophages, B cells, T cells, and NK cells in the TME of pCRC (left) and CRLM (right) Functional and communication differences between high- and low-lipid metabolism macrophages To explore the functional heterogeneity and intercellular communication patterns of macrophages with distinct lipid metabolic states, macrophages were stratified into high-lipid metabolism (LMhighMac) and low-lipid metabolism (LMlowMac) groups based on the median lipid metabolic activity score calculated using the “ssGSEA” algorithm (Fig. 4A). Pseudotime trajectory analysis showed that LMhighMac and LMlowMac cells span the full differentiation process, with no discrete early- or late-stage enrichment (Fig. 4B). Furthermore, to explore the molecular difference between the two groups, differential expression analysis was performed. Using a threshold of |log2FC| > 1 and P < 0.05, a total of 207 DEGs were identified. Functional enrichment analysis with the “fgsea” R package identified the top 10 significantly enriched KEGG pathways in each group (Fig. 4C). The LMhighMac group was enriched in lipid metabolic and immune-related pathways, such as PPAR signaling, complement and coagulation cascades, ether lipid metabolism and ECM receptor interaction, indicating enhanced metabolic and immunomodulatory activity. In contrast, the LMlowMac group showed enrichment of MAPK signaling, cell cycle, diabetes and ribosome pathways, indicating predominant involvement in basal proliferative and stress-response processes rather than active lipid metabolism. Functional and intercellular communication analysis of high- and low-lipid metabolism macrophages. (A) Macrophages were scored using ssGSEA with a lipid metabolism-related gene set. Based on the median score, they were classified into high- (LMhighMac) and low-lipid metabolism (LMlowMac) groups. (B) Pseudotime trajectory analysis of high- and low-lipid metabolism macrophages using the “monocle” R package. (C) GSEA analysis of high- and low-lipid metabolism macrophages using the “fgsea” R package, highlighting the top 10 differentially enriched KEGG pathways. (D) Cell-cell communication analysis of LMhighMac, LMlowMac, B cells, mast cells, plasma cells and T cells in the TME. (E) Analysis of outgoing and incoming signaling patterns of different immune cell types Cell-cell communication analysis using the “CellChat” R package further demonstrated that both the number and strength of communications in the TME differed markedly between the LMhighMac and LMlowMac groups (Fig. 4D). Furthermore, signaling pattern analysis revealed that LMhighMac exhibited stronger overall outgoing and incoming signaling patterns compared to those in the LMlowMac (Fig. 4E). Stemness and transcriptional regulation features of macrophages with distinct lipid metabolism activity To further explore the intrinsic properties of macrophages with differential lipid metabolic activity, we assessed stemness characteristics using the “CytoTRACE” R package. As shown in Fig. 5A and B, LMhighMac had significantly higher stemness scores than LMlowMac. Moreover, Pearson correlation analysis demonstrated a significant positive correlation between CytoTRACE-derived stemness scores and ssGSEA-derived lipid metabolic scores (Fig. 5C), indicating that enhanced lipid metabolism is closely associated with stronger stemness potential in macrophages. To further uncover the molecular basis linking lipid metabolism and stemness, we identified the top 20 genes most strongly correlated with CytoTRACE-derived stemness scores. As shown in Fig. 5D, positively correlated genes were enriched for HLA-related molecules such as HLA-DPA, HLA-DRB1 and HLA-DQB1, suggesting that antigen presentation pathways may be linked to stemness maintenance. Comparatively, negatively correlated genes were largely chemokine- and stress-related, including ADM, HSPA6, CCL3, BAG3, and CXCL2, indicating potential roles in differentiation or inflammatory responses. TF activity analysis further revealed distinct regulatory patterns: TFs such as TP53, SP3, TCF7L1, SP1, and IRF8 showed higher activity in LMhighMac, whereas HNF4G, FOXA1, TBX18, SIX3, and SOX7 were more active in LMlowMac (Fig. 5E). Collectively, these findings indicate that LMhighMac possess stronger stemness potential and distinct transcriptional regulatory landscapes, which may contribute to their functional plasticity and pro-metastatic roles in the liver metastatic microenvironment of CRC. Stemness and transcriptional regulatory features of macrophages with high and low lipid metabolism. (A) Stemness analysis of LmhighMac and LMlowMac using the “CytoTRACE” R package. (B) Comparison of CytoTRACE stemness scores between LMhighMac and LmlowMac. (C) Correlation analysis between stemness scores and lipid metabolism activity (ssGSEA scores) using Pearson’s correlation. (D) Top 20 genes most strongly associated (positively or negatively) with CytoTRACE stemness scores during differentiation. (E) Transcription factor activity analysis of LMhighMac and LMlowMac using the “decoupleR” R package Spatial distribution of lipid metabolism in CRC progression To characterize the spatial features of lipid metabolism in the progression of CRLM, we performed spatial transcriptomic analyses on NC, pCRC, and CRLM samples. In the NC sample, five major regions were identified based on canonical marker gene expression, including normal epithelium (MALAT1), fibroblasts (COL1A2), monocytes/macrophages (FTH1, CTSD), B/plasma cells (CD74), and lymphoid follicles (IGHA1) (Figs. 6A-B). Lipid metabolism showed clear spatial heterogeneity, with lymphoid follicle and monocyte/macrophage regions exhibiting significantly higher activity than normal epithelium and fibroblast regions (Figs. 6C-D). Similarly, according to the expression pattern of marker genes (Figs. 6E-F), the pCRC sample was partitioned into three major regions, including tumor, fibroblast, and normal epithelium (Fig. 6G), whereas the CRLM sample was divided into tumor, fibroblast, and hepatocyte regions (Fig. 6H). Spatial mapping combined with lipid metabolism scoring revealed that higher activity was observed in normal epithelial and tumor regions compared with fibroblast-rich areas in pCRC (Fig. 6I), whereas hepatocyte regions exhibited markedly elevated lipid metabolism in CRLM (Fig. 6J). Consistently, in pCRC, LMhighMac were predominantly located within the normal epithelial and tumor compartments (Fig. 6I), while in CRLM, they were mainly enriched in hepatocyte region (Fig. 6J). These findings indicate that lipid metabolism displays region-specific spatial organization throughout CRC progression. Spatial transcriptomic landscape of lipid metabolism in NC, pCRC and CRLM. (A) Spatial expression patterns of representative marker genes in the NC sample, including MALAT1 (Normal epithelium), COL1A2 (Fibroblast), FTH1 and CTSD (Monocyte/Macrophage), CD74 (B/Plasma cell), and IGHA1 (Lymphoid follicles). (B) The NC sample was classified into five major spatial regions based on the expression pattern of marker genes. (C) Spatial distribution of spots classified into high- and low-lipid metabolism groups according to the expression levels of lipid metabolism-related genes. (D) Violin plot showing differences in lipid metabolic activity across the five spatially defined tissue regions. (E) Expression patterns of marker genes defining tumor, fibroblast, and normal epithelium regions in the pCRC sample. (F) Expression patterns of marker genes defining tumor, fibroblast, and hepatocyte regions in the CRLM sample. (G) Visualization of spatial segmentation of the pCRC sample, with distinct colors indicating the distribution of tumor, fibroblast, and normal epithelium regions. (H) Visualization of spatial segmentation of the CRLM sample, with distinct colors indicating the distribution of tumor, fibroblast, and hepatocyte regions. (I) Violin plot showing differences in lipid metabolic activity across tumor, fibroblast, and normal epithelium regions (left) and spatial distribution of spots classified into high- and low-lipid metabolism groups (right) in the pCRC sample. (J) Violin plot showing differences in lipid metabolic activity across tumor, fibroblast, and hepatocyte regions (left) and spatial distribution of spots classified into high- and low-lipid metabolism groups (right) in the CRLM sample. NC, normal colon; pCRC, primary colorectal cancer; CRLM, Colorectal cancer liver metastasis. ****P < 0.0001 Spatial distribution and interactions of macrophage subpopulations To characterize the spatial distribution of immune cells, we applied the RCTD algorithm to integrate spatial transcriptomic data with single-cell annotations, thereby mapping immune-cell distributions in NC, pCRC, and CRLM samples (Figs. 7A-C). In the NC sample, lipid metabolism–related macrophage subsets were primarily localized to monocyte/macrophage-rich regions (Fig. 7D). MISTy analysis revealed strong spatial co-distribution, with CD1C⁺, CXCL10⁺ and CX3CR1⁺ macrophages showing close associations with B cells (Figs. 7E-F). In the pCRC sample, CD1C⁺ macrophages were mainly enriched in the tumor region, whereas CX3CR1⁺ and CXCL10⁺ macrophages were concentrated in the fibroblast-rich region (Fig. 7G). MISTy analysis showed that CD1C⁺ macrophages displayed strong spatial associations with CXCL10⁺ macrophages, CX3CR1⁺ macrophages, mast cells, and plasma cells, whereas CX3CR1⁺ macrophages were primarily correlated with T cells and CD1C⁺ macrophages (Fig. 7H–I). In the CRLM sample, CD1C⁺ macrophages were mainly distributed in the tumor region, CX3CR1⁺ macrophages in fibroblast regions, and CXCL10⁺ macrophages in hepatocyte regions (Fig. 7J). MISTy analysis further showed that CD1C⁺ macrophages were closely associated with B cells (Fig. 7K-L). Together, these findings indicate that lipid metabolism-related macrophage subpopulations exhibit distinct spatial distributions and interaction networks across NC, pCRC, and CRLM. Notably, CD1C⁺ macrophages consistently localized to tumor regions and established strong associations with other immune cells, suggesting they may play a particularly important role in shaping the metastatic microenvironment. Spatial distribution and interactions of lipid metabolism–related macrophage subpopulations. (A-C) Spatial distribution of cells in the NC (A), pCRC (B), and CRLM (C) ST samples inferred using the RCTD algorithm based on scRNA-seq-derived cell type annotations. (D) Stacked bar plots showing the proportions of lipid metabolism–related macrophage subtypes (CD1C⁺, CXCL10⁺, and CX3CR1⁺) across different spatial regions in the NC sample. (E-F) MISTy-based spatial co-localization and interaction analysis of different cell types in the NC sample. (G) Stacked bar plots showing the proportions of lipid metabolism-related macrophage subtypes (CD1C⁺, CXCL10⁺, and CX3CR1⁺) across different spatial regions in the pCRC sample. (H-I) MISTy-based spatial co-localization and interaction analysis of different cell types in the pCRC sample. (J) Stacked bar plots showing the proportions of lipid metabolism-related macrophage subtypes (CD1C⁺, CXCL10⁺, and CX3CR1⁺) across different spatial regions in the CRLM sample. (K-L) MISTy-based spatial co-localization and interaction analysis of different cell types in the CRLM sample Discussion Metabolic reprogramming is increasingly recognized as a fundamental driver of immune regulation across diverse disease contexts, linking alterations in cellular metabolism to changes in immune phenotype and function [32, 33]. Among these metabolic processes, lipid metabolism has emerged as a key determinant of macrophage behavior, affecting lipid uptake, fatty acid oxidation, cytokine production, and interaction with surrounding stromal and immune cells [9, 34, 35]. In addition to lipid metabolism, tumors commonly rely on anaerobic glycolysis, with LDH functioning as a key enzyme that enables cellular adaptation to hypoxic conditions [36]. This glycolytic reliance can reinforce the immunosuppressive characteristics of the metastatic niche. Moreover, cytokines are important modulators of cellular metabolism, influencing processes from mitochondrial activity to lipid utilization [37]. Proinflammatory cytokines, in particular, can modify metabolic states in tumor cells and thereby affect tumor progression and treatment response [38]. Given that the liver is a lipid-rich organ and CRLM represents a highly metabolic and immunosuppressive niche, dissecting how lipid metabolism regulates macrophage states is essential for understanding metastatic process. In this study, we systematically investigated the lipid metabolic landscape of macrophages across NC, pCRC, and CRLM by integrating single-cell and spatial transcriptomics. Our analyses revealed that macrophages exhibit the highest lipid metabolic activity among immune populations, with activity elevated in CRLM compared with pCRC. In addition, macrophage subpopulations exhibited distinct differentiation trajectories and communication patterns, highlighting their metabolic and functional reprogramming during metastatic progression. These findings are consistent with previous reports that tumor-associated macrophages undergo metabolic rewiring to sustain energy demands, promote immunosuppression, and support tumor progression [39, 40]. Lipid metabolism, particularly fatty acid uptake and oxidation, has been shown to drive the immunosuppressive polarization and pro-tumor activities of TAMs [41, 42]. Our single-cell analysis supports and extends these observations by demonstrating that macrophage lipid metabolism increases from pCRC to CRLM, highlighting the important role of macrophage metabolic reprogramming in shaping the metastatic microenvironment. Subclustering analysis identified lipid metabolism-related macrophage subsets (CD1C⁺, CXCL10⁺, and CX3CR1⁺) that were more enriched in CRLM than in pCRC. Pseudotime analysis further indicated distinct differentiation patterns among these subsets: CXCL10⁺ macrophages were enriched in the early-to-intermediate stages of the trajectory, CD1C⁺ macrophages accumulated toward the late stage, whereas CX3CR1⁺ macrophages were few in number and did not display a clear trajectory pattern, limiting inferences about their developmental role (Fig. 2C). Intercellular communication analyses further highlighted context-dependent remodeling of signaling networks. CD1C⁺ macrophages exhibited active VEGF signaling in both pCRC and CRLM, but CRLM was distinguished by a broader repertoire of VEGF ligand–receptor interactions, including additional VEGFA–VEGFR2 and VEGFC–VEGFR3 pairs (Fig. 3C), suggesting a more diversified angiogenic communication landscape. Comparatively, CXCL10⁺ macrophages in CRLM displayed enhanced SPP1–integrin (ITGAV/ITGB1 and ITGAV/ITGB5) signaling and increased C3–ITGAM/ITGB2 complement interactions with T cells (Fig. 3D), indicating strengthened engagement in integrin-mediated and complement-mediated immune regulation within the metastatic microenvironment. Together, these results demonstrate that CD1C⁺ and CXCL10⁺ macrophages undergo distinct differentiation trajectories and communication networks, highlighting their functional complexity in shaping CRLM progression. To further capture macrophage heterogeneity, we stratified macrophages into high- and low-lipid metabolism groups (LMhighMac vs. LMlowMac) and compared their functional, transcriptional, and signaling characteristics. Consistent with their elevated metabolic activity, LMhighMac were enriched in lipid metabolic and immune-related pathways, including PPAR signaling and complement/coagulation cascades (Fig. 4C), and demonstrated stronger outgoing and incoming communication signals within the TME (Fig. 4E). In contrast, LMlowMac were enriched in MAPK signaling, cell cycle, and ribosomal pathways (Fig. 4C), reflecting a more basal proliferative and stress-responsive state. Stemness analysis further revealed that LMhighMac possessed significantly higher differentiation potential, with stemness scores positively correlated with lipid metabolic activity (Fig. 5A and B). In addition, TF activity analysis revealed distinct regulatory landscapes, with TP53, SP3, and IRF8 active in LMhighMac, whereas HNF4G and FOXA1 predominated in LMlowMac. Notably, IRF8 has been reported to regulate antigen presentation and T cell exhaustion in tumor-associated macrophages [43], supporting its potential role in shaping immune responses in metastatic contexts. TP53 mutation has also been implicated in controlling tumor stemness and macrophage polarization in CRC [44]. Together, these findings indicate that higher lipid metabolism is associated with increased stemness potential and distinct transcriptional regulation in macrophages. Spatial transcriptomic analysis revealed that lipid metabolism exhibits region-specific organization throughout CRC progression and that macrophage subsets reside within distinct spatial regions. Notably, CD1C⁺ macrophages predominantly localized to tumor regions in both pCRC and CRLM, but their spatial interactions differed between contexts. In CRLM, they showed strong associations with B cells, aligning with recent evidence that B-cell–myeloid cell interactions contribute to shaping the hepatic metastatic microenvironment [45]. Comparatively, CXCL10⁺ and CX3CR1⁺ macrophages were enriched in fibroblast- and hepatocyte-rich regions, reflecting context-dependent roles potentially linked to stromal remodeling and tissue adaptation. These findings suggest that macrophage metabolic state and spatial localization jointly contribute to their functional specialization during metastatic progression. Conclusion In summary, by integrating single-cell and spatial transcriptomics, we found that macrophages are the immune population with the highest lipid metabolic activity, which was further elevated in CRLM. Among the lipid-associated subsets, CD1C⁺ and CXCL10⁺ macrophages were preferentially enriched in CRLM, with CD1C⁺ cells displaying diversified VEGF signaling and CXCL10⁺ cells showing enhanced SPP1-integrin and C3–(ITGAX + ITGB2) interactions. Moreover, macrophages with high lipid metabolic activity exhibited stronger enrichment of lipid metabolic and immune pathways and greater stemness potential compared with their low-activity counterparts. Spatial analyses further revealed region-specific metabolic patterns, with higher lipid metabolic activity in monocyte/macrophage- and lymphoid-rich areas of NC, elevated activity in tumor and epithelial regions of pCRC, and markedly increased activity in hepatocyte regions of CRLM. Moreover, CD1C⁺ macrophages were predominantly localized to tumor regions and closely interacted with other immune cells. Together, these findings highlight lipid metabolism–associated macrophages, particularly CD1C⁺ subsets, as key regulators of the metastatic microenvironment and potential therapeutic targets in CRLM. Author contributions XQ and SH designed the research. SH and SX collected the data and performed the computational analyses. XQ drafted the manuscript. XQ, SH and JC revised the manuscript. All authors read and approved the final manuscript. Funding This work was supported by the National Natural Science Foundation of China (Grant No. 32270705), the Qinglan Project of Jiangsu Higher Education Institutions (Excellent Young Backbone Teacher Program), and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (Grant No. SJCX24_1879). Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Xin Qi, Siyu Hou and Shaozhuo Xie contributed equally to this work and share first authorship. References Associated Data Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.