Phagocytic remodeling in CD74^High tumor-associated macrophages during brain metastasis of lung adenocarcinoma

Introduction

Lung cancer remains a leading cause of cancer-related morbidity and mortality globally (1). Lung adenocarcinoma (LUAD), the most prevalent subtype accounting for approximately 40% of all cases, exhibits a marked propensity for brain metastasis (BM) (2-4). Despite therapeutic advances, patients diagnosed with BM face a dismal prognosis. Emerging evidence indicates that the tumor microenvironment (TME) of BMs displays pronounced immunosuppressive features, characterized by a paucity of tumor-infiltrating lymphocytes, an enrichment of neutrophils, and a discordance between programmed cell death ligand 1 (PD-L1) expression and T-cell inflammatory gene signatures (5). These changes suggest a breakdown in immune surveillance and highlight how the tumor modifies the local environment to survive after spreading to the brain. Consequently, BM is accompanied by systemic TME remodeling, which simultaneously attenuates the antitumor activity of effector cells and creates an immunosuppressive niche conducive to metastatic seeding and expansion.

CD74 [canonically known as the major histocompatibility complex-II (MHC-II) invariant chain] was initially identified as a chaperone protein essential for MHC-II complex assembly. However, subsequent studies have established its role as a functional signaling receptor, particularly as the primary receptor for macrophage migration inhibitory factor (MIF) and amyloid precursor protein (APP), thereby regulating tumor immune responses (6). Evidence suggests that CD74 participates in antigen processing and presentation while simultaneously driving immunosuppression and angiogenesis through MIF binding (7,8). In the context of cancer immunotherapy, CD74 serves as a potential biomarker, with its expression levels positively correlated with an “inflamed” TME phenotype and therapeutic benefits from programmed cell death protein 1 (PD-1)/cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) bispecific antibodies (8,9). Recent findings further demonstrate that APP can inhibit macrophage phagocytosis via the CD74/CXCR4 complex, highlighting the APP-CD74 axis as a promising target for therapeutic intervention (10).

The tumor-associated macrophage (TAM) represents one of the most abundant and highly plastic immune populations within TME, extensively orchestrating immune evasion and metastatic progression (11). Even prior to metastasis, the TAM population facilitates tumor cell invasion and dissemination by suppressing immune responses and promoting angiogenesis, effectively sustaining tumor growth at metastatic sites (12). The TAM drives the growth of new blood vessels by releasing a wide variety of cytokines and growth factors (e.g., VEGF, EGF), ensuring that the metastatic tumor can receive the nutrients and blood supply that it needs. Concurrently, by releasing immunosuppressive factors (e.g., IL-10, TGF-β), TAM attenuates T-cell antitumor functions, aiding tumor cells in escaping immune surveillance (13). Thus, the TAM is not only an active participant in the remodeling of the TME during lung cancer BM but also constitutes a core cell population driving immunosuppression and angiogenesis.

Against this backdrop, we leveraged single-cell RNA sequencing (scRNA-seq) data to systematically dissect the cellular composition and ligand-receptor communication networks of LUAD primary lung tumors (LTs) and BMs. We identified CD74^High TAM as a central receptor hub across organs. Specifically, CD74^High TAM in BM exhibits a distinct functional transition: it shifts from a phenotype characterized by phagocytosis to one dominated by inflammatory responses and metabolic pathway activation. Collectively, in conjunction with established evidence on the MIF-CD74 axis in macrophage remodeling, our findings substantiate that CD74^High TAM is not merely a phenotypic marker but a critical mechanistic node regulating the functional state of the BM microenvironment, providing a biological rationale for targeting this population. We present this article in accordance with the STROBE reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0228/rc).

Methods

Data sources and ethics

In this study, we performed secondary analysis of the publicly available single-cell transcriptomic dataset GSE131907, which comprises 11 primary LUAD and 10 BM samples (14). For external validation, bulk RNA sequencing data of LUAD were obtained from The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/) data portal, and corresponding clinical follow-up information was retrieved from the UCSC Xena platform (https://xenabrowser.net/). For histological validation, formalin-fixed paraffin-embedded (FFPE) pathological sections of LUAD BMs were obtained from The Second Affiliated Hospital of Harbin Medical University. The study protocol was approved by the Medical Ethics Committee of The Second Affiliated Hospital of Harbin Medical University (No. KY2025-221), and written informed consent was obtained from all participating patients. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Single-cell data preprocessing and cell annotation

All single-cell data analyses were conducted within the R statistical environment (version 4.4.3). The Seurat package (v5.2.1) (15) was employed to process the raw expression matrices, encompassing quality control (QC), data normalization, and the selection of highly variable genes. Following dimensionality reduction via principal component analysis (PCA), cells were clustered based on a graph-based approach [shared nearest neighbor (SNN)] and visualized using Uniform Manifold Approximation and Projection (UMAP). Cell type annotation was performed according to canonical marker genes (Table S1). Subsequently, the TAM population was subsetted for downstream analysis.

Copy number variation (CNV) analysis

To identify potential malignant cell populations, we performed the chromosomal CNV analysis on the single-cell transcriptomic data using the inferCNV package (v1.22.0) (16). Raw count matrices served as the input, with human genomic coordinates retrieved and ordered via AnnoProbe (v0.1.7). For the construction of the inferCNV object, Pericytes were initially designated as the reference group to establish the baseline signal. The algorithm was executed with parameters optimized for 10× Genomics data: cutoff =0.1, hclust_method = “ward.D2”, denoise = true, and hidden Markov model (HMM) = false. To ensure a robust estimation of the neutral baseline for quantitative scoring, we subsequently integrated macrophages, fibroblasts, and pericytes into a combined reference set to calculate the mean and standard deviation of the background signal. Based on these statistics, CNV signals were categorized into loss, neutral, or gain statuses and converted into numerical scores to derive a total CNV score for each single cell. Finally, the distribution of CNV scores across distinct cell types was visualized using boxplots.

Inference and comparative analysis of cell-cell communication networks

To systematically dissect the intercellular interaction discrepancies between the LT and BM microenvironments, we constructed independent cell-cell communication networks for each tissue using the CellChat R package (v2.1.2) (17). Initially, communication probabilities between distinct cell subpopulations were inferred based on a curated ligand-receptor interaction database. To facilitate quantitative cross-dataset comparison, we extracted communication pair information from both datasets and implemented a rigorous data normalization strategy. Specifically, communication probabilities were linearly mapped to the [0, 1] interval to mitigate baseline discrepancies arising from technical factors such as sequencing depth. Subsequently, the merged communication probabilities underwent Z-score transformation to evaluate the relative importance of specific signals within the global network.

Based on the normalized communication strengths, ligand-receptor signaling pairs were stratified into three distinct functional categories. Shared signals were defined as interactions conserved in both LT and BM with comparable intensities, exemplified by pairs such as APP-CD74 and COL1A2-CD44. Tissue-specific signals, encompassing “lung-specific” and “brain-specific” groups, were defined as interactions exclusively detected in a single tissue type or exhibiting a significantly higher relative intensity (>2-fold) in one tissue compared to the other. We employed multiple visualization techniques to comprehensively depict the topological landscape of the communication networks. A comparative scatter plot was constructed with LT and BM communication strengths as axes, utilizing sector-based background regions to intuitively demarcate “brain-specific”, “lung-specific”, and “shared” signals, thereby revealing the overall distribution pattern. Additionally, circle plots were utilized to finely characterize the signaling flow and intensity between ligand-sending and receptor-receiving cells.

Multiplex immunohistochemistry (mIHC) validation

FFPE sections were subjected to deparaffinization, rehydration, and antigen retrieval. Following the blocking of non-specific binding, sections were incubated with primary antibodies against CD74 (rabbit mAb, Cat. No. A24027; ABclonal, Wuhan, China) and CD68 (rabbit mAb, Cat. No. A23205; ABclonal). Signal detection was performed using horseradish peroxidase (HRP)-conjugated secondary antibodies and a corresponding detection kit (Cat. No. G1267; Servicebio, Wuhan, China). Nuclear counterstaining and mounting were conducted using 4',6-diamidino-2-phenylindole (DAPI) Fluoromount-G™ (Cat. No. 36308ES20; Yeasen, Shanghai, China). Finally, images were acquired and analyzed using a Leica fluorescence microscopy system.

Construction and scoring of the CD74 co-expression module

To mitigate the inherent dropout effects associated with scRNA-seq and to robustly quantify the transcriptional activity of CD74 within macrophages, we constructed a CD74-specific co-expression gene module. Specifically, Pearson correlation coefficients were calculated between CD74 and all other genes across the normalized expression matrix. A gene set exhibiting a correlation coefficient (R) >0.2 was selected and defined as the “CD74 co-expression signature”. Subsequently, the UCell algorithm (v2.10.1) was employed to compute the CD74 module score for individual cells.

Stratification of TAM subpopulations

To dissect the tissue-specific functions associated with CD74 expression across distinct microenvironments, the integrated TAM population was systematically stratified based on CD74 module scores and tissue origin. Specifically, we employed quartile-based cutoffs as the stratification criteria: cells with CD74 module scores in the top 25% were designated as “high” status, those in the bottom 25% as “low” status, and cells falling within the middle 50% as “intermediate”.

Subsequently, by integrating tissue origin information (LT vs. BM), the TAM population was further categorized into five distinct analytical subgroups: LT_High, LT_Low, BM_High, BM_Low, and the combined intermediate group. Finally, the spatial distribution of these subgroups was visualized via UMAP dimensionality reduction, and the cellular abundance of each subgroup was quantified.

Differential expression analysis and pathway activity scoring

Differential expression analysis between cell subpopulations was performed using the FindMarkers function within the Seurat (v5.2.1) (15), employing a non-parametric Wilcoxon rank-sum test. To ensure biological significance and statistical robustness, the following filtering criteria were applied: (I) a minimum detection percentage of 10% in at least one group (min.pct =0.1); (II) a Benjamini-Hochberg adjusted P value (P.adjust) <0.05; and (III) an absolute log₂fold change (|log₂FC|) >0.25. We specifically focused on the biologically complementary comparison between the LT_High and BM_High groups. Genome-wide expression differences were visualized using volcano plots, with the top 10 upregulated and downregulated genes (ranked by statistical significance and FC) specifically labeled.

To visualize pathway activities, heatmaps were constructed using the ComplexHeatmap package (v2.22.0). To mitigate baseline discrepancies and highlight relative variations between groups, the aggregated AUCell (v1.28.0) score matrix underwent global Z-score standardization. The resulting color gradient (blue-white-red) reflects the normalized relative intensity of pathway activity.

External validation using TCGA-LUAD and survival analysis

Transcriptomic data for LUAD were retrieved from TCGA via the GDC Data Portal (https://portal.gdc.cancer.gov/), specifically utilizing gene expression matrices [log₂-transformed transcripts per million (TPM) or fragments per kilobase of transcript per million fragments mapped (FPKM) formats]. Corresponding clinical follow-up information, including overall survival (OS) and survival status, was obtained from the UCSC Xena platform (https://xenabrowser.net/). Prior to analysis, sample IDs were standardized, and the intersection of transcriptomic and clinical data was retained.

To quantify the activity level of the phagosome pathway in individual patient samples, single-sample gene set enrichment analysis (ssGSEA) was performed using the GSVA R package (v2.0.7). The input gene set comprised the core genes of the phagosome pathway identified in our preceding analysis (Table S2). Enrichment scores were calculated using the gsva function with ssgseaParam, with parameters set to minSize =1 and maxSize =500 to derive a standardized pathway activity score.

Survival analysis was based on these calculated phagosome pathway scores. Patients in the TCGA-LUAD cohort were stratified into high score and low score groups using the median score as the cutoff. Survival objects were constructed using the survival package (v3.8-3), and OS was estimated using the Kaplan-Meier (K-M) method. Statistical significance between survival curves was assessed via the log-rank test, with P<0.05 considered statistically significant. Visualization of survival curves and risk tables was generated using the ggsurvplot function from the survminer package (v0.5.0), with the time axis unified to “years”.

Identification of functional metaprograms via GeneNMF

To identify conserved and robust transcriptomic functional modules across samples in an unsupervised manner, we performed GeneNMF analysis (v0.9.2) (18) following the default parameters outlined in the official documentation. First, the integrated TAM Seurat object was split by sample. Multi-run non-negative matrix factorization (NMF) was executed on each independent biological sample, with the factorization rank (k) scanned from 4 to 9. Low-expression genes were filtered (min.exp =0.05) to minimize noise.

Subsequently, NMF factors extracted from all samples were clustered based on cosine similarity to identify consensus meta programs (MPs). Parameters were set to specificity.weight =5 and weight.explained =0.5 to ensure the identification of highly specific and explanatory gene modules. To guarantee biological robustness, strict filtering was applied based on metaprogram metrics: unstable modules with a silhouette score <0.3 or mean similarity <0.4 (corresponding to MP1, MP2, MP3, MP5, and MP9 in this study) were excluded. Upon examination of the top 10 driver genes, MP6 was identified as a module representing cell cycle contamination or potential TAM-T cell doublets (containing T cell markers) and was therefore excluded from downstream analysis. The remaining high-quality MPs were used to define the core functional states of TAMs.

To construct a cellular landscape based on functional states rather than raw gene expression, we employed a functional dimensionality reduction strategy. The UCell package (v2.10.1) was used to calculate module scores for each retained MP across all single cells. This MP score matrix was then extracted to construct a new dimensionality reduction object, which was stored within the Seurat object. Finally, UMAP was run on this functional score matrix using Euclidean distance. This approach intuitively visualizes the distribution and heterogeneity of the TAM population within a functional space while effectively mitigating batch effects.

Pseudotime trajectory analysis

Pseudotime analysis was conducted using the Monocle3 R package (v1.4.23) (19), adhering to the official documentation. To ensure that the trajectory reconstruction reflected the evolution of cellular functional states rather than stochastic transcriptomic fluctuations, we implemented a Function-driven Trajectory Inference Strategy. A CellDataSet (CDS) object was constructed from the raw count matrix of the subsetted Seurat object. Initial dimensionality reduction was performed using preprocess_cds (num_dim =50), followed by batch correction via align_cds based on sample information. Distinct from traditional gene expression-based UMAP, we directly imported the GeneNMF MP-based dimensionality reduction coordinates (generated in the previous step) into the CDS object. This critical step ensured that subsequent trajectory learning was grounded in robust biological functional modules.

Subsequently, the principal graph was learned on the functional manifold using the cluster_cells and learn_graph functions. Based on biological prior knowledge and cell type annotation, we defined MP8 (lung resident macrophage) as the root node of the developmental trajectory and calculated the pseudotime value for each cell using order_cells. To specifically analyze the dynamic changes in phagocytic function during metastasis, we focused our scope on the CD74^High TAM subpopulation to avoid functional confounding from different CD74 expression states. Using the previously curated phagosome pathway gene set (Table S2), we first depicted the expression dynamics of key phagocytosis-related genes along pseudotime using the plot_genes_in_pseudotime function. To quantitatively assess the phagocytic activity of the CD74^High TAM, we extracted the expression matrix for these core genes within this subpopulation. Given the sparsity of scRNA-seq data, we employed UCell (v2.10.1) to score this gene set at the single-cell level. UCell evaluates the cumulative distribution of target genes within the ranked gene list of each cell, providing a functional activity score robust to sequencing depth and normalization methods. The resulting score was defined as the “phagosome score” (reflecting phagocytic capacity). This functional score was then mapped onto the pseudotime coordinate space and integrated with the MPs defined by GeneNMF to systematically evaluate the trends in phagocytic function of the CD74^High TAM along the trajectory.

Statistical analysis

All statistical analyses were performed in the R statistical environment (version 4.4.3). Single-cell differential expression analysis was conducted using the FindMarkers function in Seurat (Wilcoxon rank-sum test; min.pct =0.1, logfc.threshold =0). P values were adjusted using the Benjamini-Hochberg method, with a significance threshold of false discovery rate (FDR) <0.05; selected results were further filtered by |log₂FC| ≥0.25. For functional enrichment and GSEAs, an P.adjust <0.05 was considered significant. Comparisons of pathway activity scores were performed using the Wilcoxon rank-sum test. Survival analysis employed the K-M method and log-rank test, with Cox proportional hazards models fitted where necessary.

Results

Single-cell transcriptomic atlas reveals the cellular landscape of LT and BM

To characterize the cellular landscape of LUAD BM, we retrieved and re-analyzed the scRNA-seq dataset GSE131907 comprising 11 LUAD primary LTs and 10 BMs samples obtained from treatment-naive patients undergoing surgical resection (14). Following rigorous QC, normalization, and PCA, we performed global dimensionality reduction and UMAP to visualize the cellular heterogeneity. Subsequent manual cell type annotation revealed distinct cellular compositions (Figure 1A). In LT samples, T cells were the predominant population (47.2%), followed by the TAM population (11.9%) and B cells (10.4%). Other identified clusters included epithelial cells (6.9%), mast cells (4.4%), plasma cells (4.3%), fibroblasts (3.9%), dendritic cells (DCs, 3.5%), natural killer (NK) cells (3.4%), endothelial cells (1.5%), neutrophils (2.1%), and pericytes (0.6%). In stark contrast, the BM landscape was dominated by epithelial cells (33.9%), highlighting the expansion of metastatic tumor cells within the brain niche. TAM (20.5%) and T cells (18.9%) constituted the major immune components, followed by NK cells (7.1%), neutrophils (4.2%), plasma cells (3.7%), oligodendrocytes (3.7%), mast cells (3%), fibroblasts (1.1%), and pericytes (1%) (Figure 1B).

Figure 1 Integrated analysis of single-cell transcriptomes of LT and BM, showing the cluster distribution of all subpopulations via UMAP, as well as their CNV profiles across cell types in the TME. (A,B) Circular plots visualizing the single-cell clustering landscape of LT (A) and BM (B). The outer sectors quantify the cell count and proportion of each cluster, while the central panels display the 2D UMAP distribution colored by Seurat-identified subpopulations. (C,D) Box plots showing the CNV profiles of LT (C) and BM (D). The CNV scores (Y-axis) are stratified by Seurat-annotated subpopulations (X-axis and colors). 2D, two-dimensional; BM, brain metastasis; CNV, copy number variation; DC, dendritic cell; LT, lung tumor; LUAD, lung adenocarcinoma; NK, natural killer; TAM, tumor-associated macrophage; TME, tumor microenvironment; UMAP, Uniform Manifold Approximation and Projection.

To validate the malignant identity of the annotated clusters, we employed the inferCNV algorithm to deduce chromosomal CNV across distinct cell subpopulations. CNV scores were calculated for individual cells and stratified by major cell type, with distributions visualized via boxplots (Figure 1C,1D). These analyses confirmed that the epithelial clusters represent LUAD-derived malignant tumor cells. Furthermore, the data demonstrated that these tumor cells retain a high degree of genomic instability and characteristic CNV profiles following their metastatic colonization of the brain.

Comparative cell-cell communication analysis reveals conserved CD74-centered signaling networks across lung and brain microenvironments

To systematically characterize the intercellular communication networks within the TME of LT and BM, we performed an in-depth analysis of the CellChat data. This analysis revealed that while the primary tumor and BM each possess a distinct repertoire of signaling pairs, they also share a subset of highly conserved and significant signals. A detailed examination of these communication patterns indicated that signaling pairs converging on CD74—such as COL1A2-CD44, PPIA-BSG, and APP-CD74—exhibited robust intercellular activity in both the pulmonary and intracranial niches (Figure 2A).

Figure 2 Comparative landscape of intercellular communication networks in the LT and BM microenvironments. (A) Differential signaling pathway analysis predicted by CellChat. The dot plot displays the top five signaling pathways categorized into lung-specific, brain-specific, and shared groups. Dot size corresponds to interaction strength (scale: 0–0.5), while colors denote signal specificity: red for brain-specific, blue for lung-specific, and green for shared pathways. (B,C) Characterization of conserved cell-cell communication networks. Key shared signaling axes—specifically COL1A2-CD44, PPIA-BSG, and APP-CD74—are enriched in both the LT (B) and BM (C) microenvironments. (D,E) Identification of tissue-specific signaling axes. (D) Visualization of Lung-specific interactions, primarily driven by the COL1A1-SDC4 and COL6A2-SDC4 axes. (E) Visualization of brain-specific interactions, highlighting the unique COL1A2-(ITGA10 + ITGB1) and CLDN11-CLDN11 pathways in the metastatic niche. BM, brain metastasis; LT, lung tumor.

Previous studies have elucidated a unique pathological function for CD74 within myeloid cells of BMs. Specifically, the MIF-CD74 axis has been confirmed as a critical mechanism mediating the immunosuppressive TME in BM. It achieves this by impeding the conversion of microglia and macrophages toward an antitumor M1 phenotype, thereby sustaining a pro-tumorigenic M2 polarization (8,20). Given that CD74 serves as the obligatory receptor for this axis, receptor competence is not uniform across all macrophages; only populations with high expression of CD74 possess the capacity to effectively bind APP or MIF and initiate downstream signal transduction (Figure 2B,2C). Consequently, we defined this specific macrophage subpopulation as the focal point of our receptor-side analysis, designating this subpopulation as “CD74^High TAM”.

Furthermore, a comparative analysis of signaling pathways unique to LT vs. BM revealed significant divergence in their regulatory directionality and biological functions (Figure 2D,2E). In LT, the enriched COL1A1-SDC4 and COL1A2-SDC4 signaling axes predominantly represented communication from fibroblasts to tumor epithelial cells. This suggests that in the primary site, stromal cells directly support tumor progression via the collagen-Syndecan-4 axis (21). In contrast, ligands such as CLDN11 and COL1A2 exhibited specific upregulation in BMs, potentially driven by the influence of LUAD. Specifically, CLDN11 mediated tight junction communication between oligodendrocytes and fibroblasts (CLDN11-CLDN11), whereas COL1A2 primarily interacted with integrin receptors (e.g., ITGA10 + ITGB1) on the surface of pericytes. These findings imply a remodeling of the “glial-fibrotic” barrier and the perivascular niche following metastasis (22,23).

Definition of CD74^High TAM and its spatial enrichment in BM

To systematically characterize the spatial distribution and functional landscape of the CD74^High TAM within the TME, we extracted macrophages from both LT and BM samples and integrated them into a unified Seurat object. Following re-normalization and dimensionality reduction via UMAP, we identified 20 distinct cell subpopulations (Figure 3A).

Figure 3 Identification and characterization of CD74^High TAM subpopulations in the LT and BM microenvironments. (A) Circular UMAP visualization displaying the cluster distribution, with outer sectors quantifying cell counts and proportions for each subset. (B,C) Density plots highlighting the expression patterns of the macrophage lineage marker CD68 (B) and CD74 (C), where green intensity indicates high expression levels. (D) Validation of CD74 and CD68 co-localization in BM tissue sections via multiplex immunofluorescence staining. Representative images show the merged and single-channel expression (CD74, CD68, DAPI) at 20× magnification (scale bar: 50 µm, left panels), alongside a high-resolution view of the indicated region at 50× magnification (scale bar: 20 µm, right panels). (E) UMAP visualization of TAMs stratified by tissue origin and CD74 expression levels, categorized into five distinct subsets: LT_High (20.1%), LT_Low (7.5%), BM_High (5%), BM_Low (17.7%), and intermediate (50%). (F) Volcano plot illustrating the differential gene expression analysis between the LT_High and BM_High subsets. Red dots denote genes significantly upregulated in LT_High TAMs, while blue dots represent downregulated genes (top 10 DEGs are labeled). BM, brain metastasis; DAPI, 4',6-diamidino-2-phenylindole; DEG, differentially expressed genes; LT, lung tumor; TAM, tumor-associated macrophage; UMAP, Uniform Manifold Approximation and Projection.

To elucidate the phenotypic identity and distribution patterns of the CD74^High TAM, we first visualized the expression profiles of CD74 alongside the classical macrophage marker CD68 using single-cell transcriptomic data. The analysis revealed striking similarities in their localization patterns, suggesting robust co-expression within the same cellular population (Figure 3B,3C).

To corroborate these transcriptomic findings in situ, we performed mIHC on BM tissue sections. This validation confirmed that CD74 is highly co-expressed and spatially enriched within macrophages residing in the brain metastatic niche (Figure 3D).

Subsequently, we stratified the integrated CD74^High TAM based on tissue origin and expression quartiles. This classification yielded five distinct subgroups: LT_High (20.1%), LT_Low (7.5%), BM_High (5%), BM_Low (17.7%), and the Intermediate group (50%) (Figure 3E). Differential expression analysis comparing the LT_High and BM_High subgroups identified distinct transcriptomic signatures. Specifically, we observed the upregulation of 300 genes in the LT_High group, including SERPINA1, MARCO, and ANXA2. Conversely, the BM_High group was characterized by the upregulation of 356 genes, such as XIST, OLFML3, and GPX1 (Figure 3F).

Transcriptional and functional divergence of CD74^High TAM between LT and BM

To elucidate the functional implications of the transcriptomic signatures identified in the CD74^High TAM, we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis on the upregulated genes specific to the LT_High and BM_High subgroups. Subsequently, we quantified pathway activity across the four stratified subgroups (LT_High, LT_Low, BM_High, and BM_Low) using the AUCell algorithm. The analysis revealed that pathways associated with phagocytosis and immune migration—including phagosome, leukocyte transendothelial migration, and salmonella infection—exhibited significantly higher activity scores in the LT_High subgroup compared to the other three groups. In contrast, pathways related to infectious and inflammatory responses, such as coronavirus disease 2019 (COVID-19), chagas disease, and pertussis, were predominantly enriched in the BM_High subgroup (Figure 4A).

Figure 4 Functional divergence and prognostic significance of CD74^High TAMs in LT and BM. (A) Heatmap illustrating the differential activity of KEGG pathways across four TAM subsets (LT_High, LT_Low, BM_High, BM_Low), quantified by AUCell scores (red indicates high activity; blue indicates low activity). (B) K-M survival analysis based on the phagosome-related gene signature derived from the LT_High subset. High expression of this signature is associated with significantly improved OS (P<0.05). (C) Bar plot visualizing the NES of differentially enriched pathways. Blue bars represent pathways enriched in LT (e.g., ribosome, cytosolic DNA-sensing, NOD-like receptor signaling), while yellow/red bars indicate pathways enriched in BM (e.g., tryptophan metabolism, PPAR signaling, regulation of actin cytoskeleton). (D,E) GSEA plots highlighting distinct immune and metabolic states. (D) Significant negative enrichment of immune-sensing pathways (e.g., cytosolic DNA-sensing) in the BM_High group, suggesting an immunosuppressive or tolerant state. (E) Positive enrichment of metabolic and immune-regulatory pathways (e.g., tryptophan metabolism, PPAR signaling) in the LT_High group. BM, brain metastasis; FC, fold change; GSEA, gene set enrichment analysis; K-M, Kaplan-Meier; KEGG, Kyoto Encyclopedia of Genes and Genomes; LT, lung tumor; NES, normalized enrichment score; OS, overall survival; TAM, tumor-associated macrophage.

Notably, the Phagosome pathway displayed the most pronounced activity differences. Consequently, we extracted the specific gene set enriched within this pathway from the LT_High subgroup (Table S1) to conduct survival analysis. The K-M survival curves demonstrated that patients with high expression of this gene signature exhibited a significantly more favorable prognosis compared to those with low expression, suggesting a potential protective role or a less aggressive phenotype associated with phagocytosis-competent TAMs in the primary setting (Figure 4B).

Furthermore, GSEA unveiled significant disparities between LT and BM tissues regarding metabolic reprogramming and microenvironmental interactions. In the LT_High subgroup, the synergistic upregulation of the PPAR signaling pathway and the ADRB3-UCP1 signaling pathway reflected a profound remodeling of lipid metabolism and energy homeostasis. This metabolic plasticity likely provides critical support for tumor survival (Figure 4C-4E). In marked contrast, the BM_High subgroup was characterized by a high enrichment of pathways governing Ribosome and Translation initiation. This indicates that BMs maintain an elevated rate of protein synthesis to meet the biosynthetic demands required for rapid colonization and expansion within the heterogeneous brain TME. Additionally, the activation of the Toll-like receptor (TLR) signaling pathway highlighted a significant innate immune response, suggesting that brain metastatic lesions may exploit chronic inflammatory microenvironments to facilitate their malignant progression (Figure 4C,4D).

GeneNMF analysis reveals functional subpopulations and state heterogeneity of CD74^High TAM

To dissect the functional heterogeneity shared across LT and BM landscapes, we applied GeneNMF analysis to the integrated TAM population. This approach initially identified 10 functional subpopulations (Figure S1, Tables S2,S3), which were subsequently refined to 5 distinct MPs based on robustness and biological relevance (Figure 5A). We then annotated these MPs by examining their top-ranking driver genes (Figure 5B). MP6 was dominated by T cell markers (e.g., TRAC, CD3D) and proliferation markers (e.g., CDK1, TK1), suggesting that this cluster represents cell cycle-related contamination or TAM-T cell doublets. MP8 was defined as the lung-resident-like TAM, driven by the high expression of canonical alveolar macrophage markers FABP4 and MCEMP1, reflecting the retention of typical tissue-homeostatic characteristics. MP7 was characterized as the Lipid-associated Immunosuppressive TAM, distinguished by the expression of the key immunosuppressive cytokine CCL18 and the iron/lipid metabolism regulator SLC40A1. Distinct from these homeostatic or suppressive states, MP4 represented a unique Inflammatory TAM phenotype, characterized by the significant upregulation of the core inflammatory mediator IL1B and the neutrophil chemoattractant CXCL8. Finally, MP10 was identified as SPP1⁺ Hypoxic TAM, defined by the specific enrichment of the pro-angiogenic factor SPP1 and the hypoxic glycolysis gene SLC2A1, highlighting a phenotype adapted to metabolic stress and vascular remodeling.

Figure 5 Identification of conserved TAM functional meta-programs via GeneNMF analysis. (A) Pearson correlation matrix displaying the transcriptomic similarity among the five retained MPs (MPs: MP4, MP6, MP7, MP8, and MP10). Darker colors indicate higher similarity, with the diagonal representing autocorrelation. The distinct clustering confirms the independence of these functional modules across both LT and BM samples. (B) Heatmap of the top 10 driver genes defining each MP, stratified by tissue of origin (LT vs. BM). Expression levels are color-coded (red: high; blue: low). Selected markers highlight specific functional states, such as immune modulation in MP6 (e.g., CCL8, CD4) and matrix remodeling in MP8 (e.g., COL1A2, FN1), underscoring the conservation of these transcriptional programs across tissues. (C) Functional landscape of TAM MPs. The heatmap illustrates the enrichment of specific biological pathways (derived from KEGG, GO, and Reactome databases) associated with each MP. Red intensity corresponds to the normalized enrichment Z-score, indicating significant pathway activation (P<0.05). BM, brain metastasis; BP, biological process; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; LT, lung tumor; MP, meta program; TAM, tumor-associated macrophage.

To independently validate the biological identities of the identified MPs, we performed functional enrichment analysis using the KEGG, Reactome, and Gene Ontology (GO) databases on the top-ranking feature genes (Figure 5C). This analysis confirmed that each subpopulation possesses distinct signaling and metabolic dependencies. The enrichment results for MP6 pointed to two core functions: cell cycle progression (REACTOME_CELL_CYCLE_MITOTIC) and T cell activation (GOBP_ALPHA_BETA_T_CELL_ACTIVATION), thereby strongly corroborating its classification as technical artifacts (cell cycle contamination or doublets). The lung-resident-like TAM (MP8) showed significant enrichment in the PPAR signaling pathway and Fatty acid metabolic process, reflecting the homeostatic lipid-processing capacity inherent to lung-resident-like macrophages. In contrast, the inflammatory TAM (MP4) exhibited a potent pro-inflammatory phenotype, characterized by extensive activation of TLR, NOD-like receptor, and Cytokine-cytokine receptor interaction pathways, confirming its role in acute immune activation. The lipid-associated immunosuppressive TAM (MP7) displayed a unique convergence of Triglyceride catabolism and Chemokine receptor binding, supporting a metabolically driven immunosuppressive state. Crucially, the BM-enriched SPP1⁺ hypoxic TAM (MP10) showed upregulation of Rho GTPase signaling and leukocyte chemotaxis. Consistent with high SPP1 expression, these pathways suggest that MP10 promotes metastatic colonization through cytoskeleton remodeling.

Function-informed pseudotime reconstruction reveals progressive state transitions of CD74^High TAM during lung-to-BM

To elucidate the dynamic evolution of the TAM population across different developmental and functional states, we performed pseudotime analysis on the GeneNMF-processed TAM to model their biological fate trajectories. Given that MP6 was enriched for cell cycle and T cell markers—indicative of contamination or doublets—it was excluded from subsequent trajectory inference to minimize bias (Figure 6A). Subsequently, we designated the lung-resident-like TAM (MP8) as the root node for pseudotime analysis. This decision was grounded not only in the observation that MP8 retains unperturbed tissue-homeostatic transcriptional signatures, typified by FABP4, but also in established biological models positing that the tissue-resident TAM is a critical component of the TME during initiation, mediating early tumor cell invasion and dissemination (24).

Figure 6 Pseudotime trajectory and functional evolution of TAMs originating from lung-resident-like progenitors. (A,B) Pseudotime trajectory analysis inferred from GeneNMF-derived MPs. (A) UMAP plot displaying the distribution of four MPs, identifying MP8 as the starting lung-resident-like state. (B) Pseudotime projection colored from early to late stages, tracing the evolutionary path. (C,D) Schematic models illustrating the functional remodeling of CD74^High TAMs from the primary LUAD niche to the brain metastasis niche. (C) In the primary LUAD niche, CD74^High TAMs retain relatively active phagosome-related programs and interact with tumor and stromal cells through ligand-receptor pairs including APP-CD74, PPIA-BSG, and COL1A2/COL6A2-CD44. (D) In the brain metastasis niche, CD74^High TAMs undergo transcriptional reprogramming, characterized by attenuated phagocytic programs and enhanced inflammatory, metabolic, and stress-adaptive states. (E,F) Dynamics of phagosome-related genes (E) and phagosome scores in CD74^High TAMs (F) along the pseudotime trajectory. In both panels, individual cells are colored by MP identity, with black curves indicating the progressive downregulation of phagocytic function over time. BM, brain metastasis; LUAD, lung adenocarcinoma; MP, metaprogram; TAM, tumor-associated macrophage; UMAP, Uniform Manifold Approximation and Projection.

Consequently, irrespective of the specific developmental ontogeny of the TAM across organs, this node assignment enabled the reconstruction of a transcriptomic functional continuum. This trajectory captures the evolution of the TAM population from an early “pro-invasive homeostatic state” to a late “pro-metastatic adaptive state”. Specifically, the TAM originates from the lipid-metabolism-characterized lung-resident-like TAM (MP8), transitions through the lipid-associated immunosuppressive TAM (MP7), and ultimately differentiates into two terminal states: the SPP1⁺ hypoxic TAM (MP10) and the inflammatory TAM (MP4) (Figure 6B).

Synthesizing the GeneNMF classification and pseudotime analysis revealed a systematic disparity in the microenvironmental signaling context and functional state of the CD74^High TAM between LT and BM (Figure 6C,6D). In the LT microenvironment, multiple stromal cells (e.g., fibroblasts, endothelial cells) and tumor epithelial cells collectively overexpressed ligands such as COL1A2, COL6A2, APP, and PPIA. These ligands interacted with CD44, BSG, and CD74 receptors on macrophages, coinciding with high activity of phagosome-related pathways, thereby maintaining a relatively robust phagocytic capacity in the CD74^High TAM. In stark contrast, within the BM microenvironment—although these ligands were still expressed by tumor cells, glial cells, and pericytes—the phagosome pathway activity significantly diminished as the TAM evolved along the pseudotime trajectory toward BM-associated states. This loss of function is accompanied by the enhancement of perivascular niche remodeling, pro-inflammatory responses, and metabolic programs. To validate this inference regarding the cross-organ attenuation of phagocytosis, we mapped the expression of the previously identified phagosome gene set (Table S1) onto the pseudotime trajectory. The results demonstrated a significant and continuous decline in the overall expression level of this gene set as the TAM transitioned from LT to BM. This revealed that the evolution of the TAM during lung-to-BM was a process of gradual adaptation to the brain microenvironment, where phagocytic functions were suppressed to accommodate the metastatic niche (Figure 6E).

Finally, to systematically assess the dynamic evolution of phagocytic capacity in the CD74^High TAM during the transition from LT to BM, we quantified the activity of functions associated with the phagosome pathway and tracked their dynamics along the pseudotime trajectory. By integrating phagosome-associated genes into a composite “phagosome score” and mapping this metric onto the trajectory, we observed a distinct overall declining trend in phagocytic activity as pseudotime progressed (Figure 6F). Although the CD74^High TAM population encompasses multiple functional MPs exhibiting a degree of heterogeneity in phagosome score across distinct MPs, the global smoothed fitting curve demonstrated a consistent downward trajectory. Collectively, these results indicate that the phagocytosis-related functions of the CD74^High TAM are progressively attenuated during cross-organ evolution, mirroring the adaptation of the cellular state to the metastatic niche.

Discussion

In this study, by deeply mining single-cell transcriptomic data from LT and BM, and starting from the resolution of cell-cell communication networks within their respective TME, we focused on the CD74^High TAM—a receptor hub occupying a central role across organs. We systematically revealed the phagocytic functional remodeling of TAM during cross-organ metastasis. Specifically, compared to CD74^High TAMs in LT, which retains potential phagocytic activity, CD74^High TAMs in BM exhibited significant downregulation of the phagosome pathway. Notably, high activity of this pathway was significantly positively correlated with a favorable prognosis in LT patients (i.e., “higher phagosome pathway expression predicts better prognosis”). Furthermore, our trajectory analysis precisely delineated the dynamic path of this functional loss: the TAM population originates from a lipid-metabolism-characterized lung-resident-like state (MP8), transitions through a lipid-immunosuppressive state (MP7), and ultimately differentiates into terminal states represented by SPP1⁺ hypoxic TAM (MP10) and pro-inflammatory TAM (MP4). Based on this, we observed the continuous downregulation of core phagosome pathway genes along the differentiation trajectory within CD74^High TAM, confirming the dynamic characteristics of phagocytic remodeling during the BM of LUAD.

Although CD74 maintains high expression in the TAM population across organs, differential gene expression analysis revealed a fundamental shift in the functional connotation of CD74^High TAM within different tissue contexts. In LT, CD74^High TAM retains phagocytosis- and lysosome-related molecular features represented by MARCO, LGMN, CSTB, and LYZ, and is enriched in pathways such as phagosome, antigen processing, and transendothelial migration, suggesting it still possesses certain clearance and immune surveillance potentials (25). In contrast, CD74^High TAM in BM significantly upregulated genes such as SPP1, GPX1, XIST, RNASE1, and FN1, accompanied by the enhancement of ribosome, translation initiation, and stress-related programs. This indicates a functional deviation away from phagocytosis and clearance. This shift suggests that the CD74^High TAM in the brain no longer serves as a “phagocytic executor” but instead assumes functions supporting tumor survival and dissemination.

The inhibition of phagocytic function is one of the core changes in the CD74^High TAM during BM. Pseudotime analysis showed that as macrophages evolved from a lung-resident-like state to a BM-associated state, the phagosome-related gene set was continuously downregulated. This change was not instantaneous but a gradual accumulation. This trend can be glimpsed in the MPs resolved by GeneNMF. Lung-resident-like TAM (MP8) was enriched in lipid metabolism and homeostasis maintenance genes such as FABP4, PPARG, LIPA, and CYP27A1, retaining a state of balanced phagocytosis and metabolism. Conversely, the SPP1⁺ hypoxic TAM (MP10), which dominates in BM, lacked these phagocytosis-related features, replacing them with hypoxia-stress and anabolic-related molecules such as SLC2A1 (GLUT1), BNIP3, ERO1A, and NUPR1. These results collectively indicate that the loss of phagocytic function is not the inhibition of a single signaling pathway, but a systemic degeneration accompanying a switch in overall metabolic and transcriptional programs.

Metabolism-related gene sets further revealed the intrinsic driving force behind the functional transition of the CD74^High TAM. In the primary site, CD74^High TAM and lung-resident-like TAM (MP8) were significantly enriched in PPAR signaling, fatty acid oxidation, and cholesterol metabolism-related molecules (e.g., FABP4, PPARG, LPL, ACOT7), which is highly consistent with the physiological characteristics of alveolar macrophages relying on lipid metabolism to maintain homeostasis (26,27). However, in BM, this feature was markedly weakened and replaced by the enhancement of glycolysis and protein synthesis programs. The co-expression of SLC2A1, ENO1, GALM, and FBP1 in SPP1⁺ hypoxic TAM (MP10) suggests that this TAM is in a high-glycolytic state, alongside the activation of ribosome and translation-related pathways. This reflects a shift in functional focus towards massive protein synthesis and secretion (28). This state is consistent with the previous differential gene analysis results, suggesting that this enhanced adaptation to the high oxidative stress and low-nutrient environment in the brain provides the molecular basis for the sustained survival and pro-tumorigenic role of the TAM population in the BM microenvironment.

Among all subsets of CD74^High TAMs, SPP1⁺ hypoxic TAMs (MP10) represent the most prominent terminal subset in BMs. Co-expression of SLC2A1, ENO1, GALM, and FBP1 indicates that these TAMs exhibit a high glycolytic state accompanied by activated ribosome and translation pathways, reflecting a functional shift toward enhanced protein synthesis and secretion (28). This phenotype is consistent with our previous differential gene analysis and suggests that SPP1⁺ hypoxic TAMs (MP10) enhance adaptation to the high oxidative stress and nutrient-poor brain microenvironment, thereby supporting the survival and protumor functions of TAMs in BMs. Consistent with its role in tissue remodeling, SPP1⁺ hypoxic TAMs (MP10) display strong profibrotic and proangiogenic transcriptional profiles (Table S4). This subset highly expresses key molecules involved in extracellular matrix assembly and focal adhesion formation, including SPP1, FN1, PDPN, TNS1, SDC2, and ADAM8. It also specifically upregulates potent proangiogenic growth factors such as AREG and EREG. Meanwhile, pathways related to Rho GTPase signaling, leukocyte migration, and hypoxic response are significantly enriched. These observations indicate that SPP1⁺ hypoxic TAMs (MP10) actively drive physical remodeling of the brain metastatic microenvironment. Mechanistically, SPP1 functions as a multifunctional extracellular matrix protein and acts synergistically with locally synthesized matrix components including FN1 to promote angiogenesis and matrix remodeling (29,30). In line with previous studies, this transcriptional profile indicates that SPP1⁺ hypoxic TAMs (MP10) directly contribute to the formation of a glial-fibrotic barrier and mediate perivascular niche remodeling (31,32). By establishing this physical barrier, this dominant subset provides critical structural and metabolic support for tumor cells while shielding them from immune surveillance.

Notably, proinflammatory TAMs (MP4) serve as a terminally activated state of SPP1⁺ hypoxic TAMs (MP10) and cooperate with SPP1⁺ hypoxic TAMs (MP10) to shape a tumor-promoting microenvironment in BMs of LUAD (33). Proinflammatory TAMs (MP4) exhibit high expression of genes associated with chronic inflammation and immediate early responses, including IL1B, CXCL8, TNF, and the NR4A family (34), indicating that this cell subset maintains a persistently activated inflammatory state. This phenotype is tightly linked to the hypoxic microenvironment established by SPP1⁺ hypoxic TAMs (MP10) (35). SPP1⁺ hypoxic TAMs (MP10) sustain a highly glycolytic metabolic phenotype via elevated expression of SLC2A1 and related genes, providing sufficient energy support for the survival and functional execution of both themselves and proinflammatory TAMs (MP4). In turn, proinflammatory TAMs (MP4) amplify local inflammatory signals through sustained inflammatory responses, laying a critical foundation for the cooperative prometastatic functions of the two subsets (35). Proinflammatory cytokines secreted by proinflammatory TAMs (MP4), including IL-1β and TNF-α, directly disrupt the structural integrity of the blood-brain barrier (BBB) and significantly increase its permeability, thereby creating favorable conditions for tumor cells to cross the BBB and invade the brain parenchyma (36). In contrast to SPP1⁺ hypoxic TAMs (MP10), which construct a glial-fibrotic barrier and shield tumor cells from immune clearance via SPP1, FN1 and other molecules, proinflammatory TAMs (MP4) mainly enhance the invasive capacity of tumor cells by sustained release of proinflammatory factors and maintenance of a local chronic inflammatory microenvironment. Coupled with the proangiogenic effects mediated by SPP1⁺ hypoxic TAMs (MP10), these processes jointly accelerate tumor cell colonization and proliferation in the brain. This ultimately forms a cooperative functional module in which SPP1⁺ hypoxic TAMs (MP10) construct a prometastatic niche and proinflammatory TAMs (MP4) amplify prometastatic inflammatory signals, collectively driving the initiation and progression of BM in LUAD.

In the process of LUAD metastasis to the brain, CD74^High TAM is not a static immune cell population but undergoes continuous functional reconstruction as the tumor evolves across organs. By combining MPs with pseudotime evolution, we observed that the overall state of TAM shifts from a functional lineage characterized by tissue homeostasis and immune clearance in the primary LT to a phenotypic state more adaptive to the BM microenvironment. This transition is not a single path or a single terminal state but is manifested as the differentiation and parallel existence of multiple MPs along the same evolutionary trajectory. In this context of functional evolution, CD74^High TAM is distributed across different MPs, and its functional characteristics must be understood in conjunction with the overall trajectory. Our results show that although CD74^High TAM exhibits a certain degree of functional heterogeneity across different MPs, its phagocytosis-related functions show a consistent downward trend across functional states and evolutionary stages. This not only suggests that CD74^High TAM undergoes systemic functional remodeling during BM but also indicates that phagocytic function is progressively suppressed throughout the process.

Existing studies on CD74 function have confirmed that the APP-CD74 ligand-receptor interaction acts as a core active regulatory pathway for macrophage phagocytosis inhibition, which can directly initiate the phagocytosis suppression program (10). However, our data demonstrate that this active regulatory mechanism cannot fully explain the preservation of basal phagocytic activity in CD74^High TAMs from LT, despite sustained high expression of CD74. A fundamental shift in the functional regulatory landscape of CD74^High TAMs occurs only after BM of LUAD, and this shift is not driven by the APP-CD74 axis. Our data reveal that the transcriptomic profile of SPP1⁺ hypoxic TAMs (MP10) is predominantly regulated by pathways associated with metabolic stress and hypoxic response. Given that phagocytosis is a highly energy-consuming biological process dependent on ATP supply and dynamic cytoskeletal remodeling, we postulate that within the severely nutrient-deprived and hypoxic niche of BM, CD74^High TAMs undergo fundamental metabolic reprogramming from lipid homeostatic metabolism to hypoxia-adaptive glycolysis to prioritize cell survival. This fundamental shift in metabolic phenotype fails to support the full energy-intensive process of phagosome synthesis, trafficking, and degradation at the level of energy supply, directly eliminating the metabolic basis for the maintenance of phagocytic function. Thus, the ultimate loss of phagocytic function in CD74^High TAMs during the LUAD metastasis to the brain is not solely driven by the previously reported active inhibition mediated by the APP-CD74 axis (10). Instead, the more central mechanism is passive functional degeneration mediated by metabolic reprogramming under the hypoxic and nutrient-deprived stress of the brain metastatic microenvironment. This finding complements the regulatory mechanisms underlying the functional reprogramming of CD74^High TAMs and provides a novel perspective for understanding macrophage functional inactivation in the BM.

Collectively, our study systematically delineates the dynamic spatiotemporal evolutionary trajectory of CD74^High TAMs during the LUAD metastasis to the brain, and this trajectory directly demonstrates that CD74 is not suitable as a prognostic biomarker for patients with LUAD BM. In the LT, CD74 expression is highly correlated with phagosome-related gene signatures, indicating its involvement in antigen presentation and immune surveillance processes, and it can serve as a predictive marker of favorable clinical prognosis. However, this transcriptomic association is completely abrogated in the brain metastatic microenvironment. Following the LUAD metastasis to the brain, intracranial TAMs adapt to the severe metabolic stress of the brain microenvironment, with the concurrent gradual shutdown of their energy-intensive phagosome machinery. Despite this, CD74 expression remains at high levels (Figure S2). In the SPP1⁺ hypoxic TAMs (MP10) subset, this receptor undergoes active functional reprogramming to instead mediate protumor and immunosuppressive signaling. Numerous studies support this alternative oncogenic role of CD74. For example, MIF binding to CD74 promotes angiogenesis and drives macrophage polarization toward a protumor phenotype (7,8), while the APP-CD74 signaling axis actively inhibits immune clearance mechanisms (10). Thus, CD74 cannot serve as a universal surrogate marker of phagocytic activity or favorable clinical outcome across the entire metastatic cascade. Following exposure to microenvironment-mediated metabolic stress in TAMs, its value as a prognostic biomarker for the LUAD metastasis to the brain is completely lost, as the receptor undergoes oncogenic reprogramming and no longer holds independent clinical biomarker utility.

Our study has several limitations. First, our analyses are primarily based on single-cell transcriptomic data, with functional inferences remaining at the transcriptional level and lacking direct experimental validation of phagocytic function or metabolism. Second, TAMs in the BM are essentially a mixed cell subset composed of innate brain-resident tissue microglia and infiltrating bone marrow-derived macrophages originating from peripheral circulating monocytes, which precludes the accurate discrimination of their developmental origins solely based on transcriptomic features. Although distinguishing between these two cell lineages is important, our single-cell transcriptomic analyses revealed significant phenotypic homogenization within the established niche. Using a recently validated lineage-specific gene signature score (Table S5) (37), we did not observe distinct, mutually exclusive subsets (Figure S3). We therefore intentionally evaluated CD74^High TAMs as a holistic, microenvironment-driven functional entity in this study. Future investigations using in vivo lineage tracing models are urgently needed to determine whether this functional reprogramming occurs consistently in cells of both developmental origins. Third, due to the difficulty in obtaining complete long-term clinical follow-up data for patients with BM, we did not perform survival analysis on the BM cohort. We therefore cannot clinically assess whether CD74 expression or phagosome signatures retain their prognostic value observed in primary tumors at the late metastatic stage, which serves as key supporting evidence that CD74 cannot be used as a prognostic biomarker for BM. Fourth, limited by the difficulty in obtaining clinical samples of BM and establishing relevant animal models, we did not perform in vitro and in vivo experimental validation of the causal relationship between CD74-related signaling axes, metabolic reprogramming, and functional transformation. Future studies may integrate multi-omics technologies to systematically analyze the plasticity and reversibility of TAM functional states, thereby more comprehensively elucidating their roles in the brain metastatic microenvironment.

In summary, our study systematically delineates the spatiotemporal dynamic evolutionary trajectory of CD74^High TAMs during the LUAD metastasis to the brain. We confirm that this cell population undergoes functional reprogramming centered on progressive loss of phagocytic function and metabolic rewiring during cross-organ metastatic progression, ultimately transforming from an immune surveillance-competent phagocytic population into a protumor cell subset that supports BM. Our study not only reveals a novel mechanism of macrophage phagocytic inactivation driven by microenvironmental metabolic stress, but also provides a novel theoretical framework for dissecting the mechanisms of immunotherapy resistance and developing precise therapeutic targets for LUAD BM.

Conclusions

This study proposes that CD74^High TAM undergoes functional remodeling centered on phagocytic degeneration and metabolic reprogramming during the BM of LUAD. This change drives the transformation of CD74^High TAM from an immune-phagocytic state to a cell population participating in microenvironment construction and tumor support, providing a new mechanistic perspective for understanding the immune ecology of BM.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0228/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0228/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0228/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0228/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study protocol was approved by the Medical Ethics Committee of The Second Affiliated Hospital of Harbin Medical University (No. KY2025-221), and written informed consent was obtained from all participating patients. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Sperduto PW, De B, Li J, et al. Graded Prognostic Assessment (GPA) for Patients With Lung Cancer and Brain Metastases: Initial Report of the Small Cell Lung Cancer GPA and Update of the Non-Small Cell Lung Cancer GPA Including the Effect of Programmed Death Ligand 1 and Other Prognostic Factors. Int J Radiat Oncol Biol Phys 2022;114:60-74. [Crossref] [PubMed]
3. Villano JL, Durbin EB, Normandeau C, et al. Incidence of brain metastasis at initial presentation of lung cancer. Neuro Oncol 2015;17:122-8. [Crossref] [PubMed]
4. Ali A, Goffin JR, Arnold A, et al. Survival of patients with non-small-cell lung cancer after a diagnosis of brain metastases. Curr Oncol 2013;20:e300-6. [Crossref] [PubMed]
5. Li M, Hou X, Sai K, et al. Immune suppressive microenvironment in brain metastatic non-small cell lung cancer: comprehensive immune microenvironment profiling of brain metastases versus paired primary lung tumors (GASTO 1060). Oncoimmunology 2022;11:2059874. [Crossref] [PubMed]
6. Su H, Na N, Zhang X, et al. The biological function and significance of CD74 in immune diseases. Inflamm Res 2017;66:209-16. [Crossref] [PubMed]
7. Girard E, Strathdee C, Trueblood E, et al. Macrophage migration inhibitory factor produced by the tumour stroma but not by tumour cells regulates angiogenesis in the B16-F10 melanoma model. Br J Cancer 2012;107:1498-505. [Crossref] [PubMed]
8. Ghoochani A, Schwarz MA, Yakubov E, et al. MIF-CD74 signaling impedes microglial M1 polarization and facilitates brain tumorigenesis. Oncogene 2016;35:6246-61. [Crossref] [PubMed]
9. Pang X, Huang Z, Zhong T, et al. Cadonilimab, a tetravalent PD-1/CTLA-4 bispecific antibody with trans-binding and enhanced target binding avidity. MAbs 2023;15:2180794. [Crossref] [PubMed]
10. Ma C, Chen J, Ji J, et al. Therapeutic modulation of APP-CD74 axis can activate phagocytosis of TAMs in GBM. Biochim Biophys Acta Mol Basis Dis 2024;1870:167449. [Crossref] [PubMed]
11. Ghebremedhin A, Athavale D, Zhang Y, et al. Tumor-Associated Macrophages as Major Immunosuppressive Cells in the Tumor Microenvironment. Cancers (Basel) 2024;16:3410. [Crossref] [PubMed]
12. Wang Y, Jia J, Wang F, et al. Pre-metastatic niche: formation, characteristics and therapeutic implication. Signal Transduct Target Ther 2024;9:236. [Crossref] [PubMed]
13. Ji ZZ, Chan MK, Chan AS, et al. Tumour-associated macrophages: versatile players in the tumour microenvironment. Front Cell Dev Biol 2023;11:1261749. [Crossref] [PubMed]
14. Kim N, Kim HK, Lee K, et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat Commun 2020;11:2285. [Crossref] [PubMed]
15. Satija R, Farrell JA, Gennert D, et al. Spatial reconstruction of single-cell gene expression data. Nat Biotechnol 2015;33:495-502. [Crossref] [PubMed]
16. Patel AP, Tirosh I, Trombetta JJ, et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 2014;344:1396-401. [Crossref] [PubMed]
17. Jin S, Guerrero-Juarez CF, Zhang L, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun 2021;12:1088. [Crossref] [PubMed]
18. YerlyLAndreattaMGarnicaJWounding triggers invasive progression in human basal cell carcinoma.bioRxiv:2024.05.31.596823 [Preprint]. Available online: https://www.biorxiv.org/content/10.1101/2024.05.31.596823v2
19. Cao J, Spielmann M, Qiu X, et al. The single-cell transcriptional landscape of mammalian organogenesis. Nature 2019;566:496-502. [Crossref] [PubMed]
20. Liu L, Wang J, Wang Y, et al. Blocking the MIF-CD74 axis augments radiotherapy efficacy for brain metastasis in NSCLC via synergistically promoting microglia M1 polarization. J Exp Clin Cancer Res 2024;43:128. [Crossref] [PubMed]
21. Yang J, Xu Q, Lu Y. Decoding epithelial-fibroblast interactions in lung adenocarcinoma through single-cell and spatial transcriptomics. J Cancer Res Clin Oncol 2025;151:221. [Crossref] [PubMed]
22. Sakamoto S, Inoue H, Takino T, et al. Claudin-11 Enhances Invasive and Metastatic Abilities of Small-Cell Lung Cancer Through MT1-MMP Activation. Cancer Sci 2025;116:1773-84. [Crossref] [PubMed]
23. Su C, Mo J, Dong S, et al. Integrinβ-1 in disorders and cancers: molecular mechanisms and therapeutic targets. Cell Commun Signal 2024;22:71. [Crossref] [PubMed]
24. Casanova-Acebes M, Dalla E, Leader AM, et al. Tissue-resident macrophages provide a pro-tumorigenic niche to early NSCLC cells. Nature 2021;595:578-84. [Crossref] [PubMed]
25. Xing Q, Feng Y, Sun H, et al. Scavenger receptor MARCO contributes to macrophage phagocytosis and clearance of tumor cells. Exp Cell Res 2021;408:112862. [Crossref] [PubMed]
26. Aegerter H, Lambrecht BN, Jakubzick CV. Biology of lung macrophages in health and disease. Immunity 2022;55:1564-80. [Crossref] [PubMed]
27. Schneider C, Nobs SP, Kurrer M, et al. Induction of the nuclear receptor PPAR-γ by the cytokine GM-CSF is critical for the differentiation of fetal monocytes into alveolar macrophages. Nat Immunol 2014;15:1026-37. [Crossref] [PubMed]
28. Metge BJ, Williams L, Swain CA, et al. Ribosomal RNA Biosynthesis Functionally Programs Tumor-Associated Macrophages to Support Breast Cancer Progression. Cancer Res 2025;85:1459-78. [Crossref] [PubMed]
29. Trehan R, Huang P, Zhu XB, et al. SPP1 + macrophages cause exhaustion of tumor-specific T cells in liver metastases. Nat Commun 2025;16:4242. [Crossref] [PubMed]
30. Tan Y, Zhao L, Yang YG, et al. The Role of Osteopontin in Tumor Progression Through Tumor-Associated Macrophages. Front Oncol 2022;12:953283. [Crossref] [PubMed]
31. You H, Baluszek S, Kaminska B. Immune Microenvironment of Brain Metastases-Are Microglia and Other Brain Macrophages Little Helpers? Front Immunol 2019;10:1941. [Crossref] [PubMed]
32. Charles N, Holland EC. The perivascular niche microenvironment in brain tumor progression. Cell Cycle 2010;9:3012-21. [Crossref] [PubMed]
33. Zheng H, Li YQ, Lu X, et al. Senescent SPP1(+) macrophages remodel the tumor microenvironment and promote the progression of early-stage lung adenocarcinoma featured with mixed ground glass nodule. Mol Cancer 2025;24:298. [Crossref] [PubMed]
34. Pei L, Castrillo A, Chen M, et al. Induction of NR4A orphan nuclear receptor expression in macrophages in response to inflammatory stimuli. J Biol Chem 2005;280:29256-62. [Crossref] [PubMed]
35. Liu Z, Wang X, Wang Y, et al. Integration analysis of single-cell transcriptome reveals SPP1(+) and TFF3(+) macrophage subsets contributing to the brain metastasis from lung cancer. Transl Cancer Res 2025;14:5680-93. [Crossref] [PubMed]
36. Shu Y, Peng F, Zhao B, et al. Transfer of patient’s peripheral blood mononuclear cells (PBMCs) disrupts blood-brain barrier and induces anti-NMDAR encephalitis: a study of novel humanized PBMC mouse model. J Neuroinflammation 2023;20:164. [Crossref] [PubMed]
37. Ran X, Zheng J, Chen L, et al. Single-Cell Transcriptomics Reveals the Heterogeneity of the Immune Landscape of IDH-Wild-Type High-Grade Gliomas. Cancer Immunol Res 2024;12:232-46. [Crossref] [PubMed]