POSTN⁺ CAF–SPP1⁺ macrophage crosstalk is associated with immune exclusion and poor prognosis in lung adenocarcinoma

Abstract

Background

Lung adenocarcinoma (LUAD) is a highly lethal malignancy in which the tumor microenvironment (TME) plays an important role in disease progression and therapeutic resistance. However, the spatial organization and molecular basis of stromal–immune interactions in LUAD remain incompletely understood.

Methods

We integrated bulk RNA sequencing, single-cell RNA sequencing, spatial transcriptomics, multiplex immunofluorescence, and in vitro functional assays to characterize TME heterogeneity and investigate mechanisms associated with stromal barrier formation and immune exclusion in LUAD.

Results

POSTN⁺ cancer-associated fibroblasts (CAFs) and SPP1⁺ macrophages were enriched in LUAD, showed marked spatial colocalization, and were associated with an immune-excluded spatial pattern characterized by limited effector T-cell infiltration. Cell–cell communication analysis identified SPP1-centered signaling from SPP1⁺ macrophages to POSTN⁺ CAFs, with CD44 emerging as a prominent receptor candidate and integrin-related receptor pairs suggested as additional candidate branches. In vitro experiments showed that TAM-like macrophage-conditioned medium activated CAF-like HFL-1 cells, increased POSTN, FN1, and COL1A1 expression, enhanced fibroblast adhesion and collagen gel contraction, and restricted primary CD8⁺ T-cell migration across CAF-like stromal barriers. These effects were attenuated by SPP1 neutralization or CD44 knockdown. In addition, CAF-derived C3 was implicated in a feedback loop that may reinforce the SPP1⁺/CD206⁺ macrophage-like phenotype. Pan-cancer analyses further showed that this stromal–myeloid program was conserved across multiple solid tumor types and was associated with poor overall survival, while exploratory cross-cancer immunotherapy analysis suggested a potential association with reduced response to immune checkpoint blockade.

Conclusion

These findings further define a spatially organized POSTN⁺ CAF–SPP1⁺ macrophage niche that may contribute to stromal remodeling, CD8⁺ T-cell exclusion, and potentially reduced immunotherapy benefit in LUAD. The TAM-derived SPP1–CD44 axis and CAF-derived C3 feedback loop may provide a framework for biomarker development and therapeutic targeting in immune-excluded solid tumors.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08746-2.

Introduction

Lung cancer, particularly lung adenocarcinoma (LUAD), remains the leading cause of cancer-related mortality worldwide [1]. Although immune checkpoint inhibitors (ICIs), especially PD-1/PD-L1 blockade, have transformed the treatment of advanced non-small cell lung cancer, their clinical benefit remains limited, with objective response rates of only 20% to 30% and frequent primary or acquired resistance [2, 3]. Current evidence suggests that this limitation is largely attributable to the complex tumor microenvironment (TME). Within the TME, excessive stromal deposition and the accumulation of immunosuppressive cells can establish an “immune-excluded” phenotype, in which cytotoxic T lymphocytes are retained outside tumor nests. This phenotype is a key mechanism of immunotherapy resistance [4].

Cancer-associated fibroblasts (CAFs) and tumor-associated macrophages (TAMs) are two of the most abundant cellular populations in the TME. As central regulators of stromal remodeling and immune suppression, CAFs drive extracellular matrix (ECM) remodeling and desmoplasia, whereas TAMs predominantly exhibit an immunosuppressive M2-like phenotype. Emerging evidence supports bidirectional communication between CAFs and TAMs, and this interaction may exert a stronger effect on tumor progression than either population alone [5–8]. However, the spatial relationship between these two cell types and the precise molecular mechanisms by which they contribute to immunotherapy resistance in LUAD remain incompletely understood.

Both CAFs and TAMs display marked functional heterogeneity, making bulk lineage-level analyses insufficient. Single-cell RNA sequencing (scRNA-seq) has enabled identification of distinct functional subpopulations within these compartments. Among them, Periostin (POSTN)-expressing CAFs are associated with tumor metastasis and therapeutic resistance through enhanced stromal stiffness and collagen cross-linking [8]. Secreted Phosphoprotein 1 (SPP1)-expressing TAMs have been characterized as a subset of M2-like macrophages with pro-angiogenic and potent immunosuppressive properties [9, 10]. Recent studies have reported spatial associations between POSTN⁺ CAFs and SPP1⁺ macrophages in NSCLC and other solid tumors. However, how this interaction generates a functional stromal barrier and whether CAF-derived signals feed back to macrophages remain unclear [11]. Because POSTN and SPP1 are key matricellular proteins involved in stromal remodeling and receptor-mediated cell–cell communication, we hypothesized that POSTN⁺ CAFs and SPP1⁺ TAMs form a spatially coordinated niche that promotes extracellular matrix remodeling, stromal barrier formation, and immune exclusion in LUAD.

In this study, by integrating scRNA-seq, spatial transcriptomics, multiplex immunofluorescence, bulk transcriptomic validation, and in vitro functional assays, we further characterized an immunosuppressive stromal–myeloid niche in LUAD marked by the spatial colocalization of POSTN⁺ CAFs and SPP1⁺ TAMs. This niche was associated with restricted CD8⁺ T-cell infiltration, poor prognosis, and exploratory indicators of reduced immunotherapy benefit. Cell–cell communication analysis suggested SPP1-centered signaling from SPP1⁺ TAMs to POSTN⁺ CAFs, with CD44 emerging as a prominent receptor candidate. In vitro experiments further showed that SPP1 neutralization and CD44 knockdown attenuated TAM-CM-induced ECM-related protein expression, fibroblast adhesion, collagen contraction, and restriction of primary CD8⁺ T-cell migration across CAF-like barriers. In parallel, integrin-associated FAK-related signaling was also examined as an additional candidate branch. In addition, CAF-derived C3 may contribute to a feedback loop that reinforces the SPP1⁺/CD206⁺ macrophage-like phenotype. Together, our findings elucidate a spatial and molecular framework by which the POSTN⁺ CAF–SPP1⁺ TAM niche promotes stromal remodeling and immune exclusion in LUAD.

Methods

Data acquisition and integration

LUAD scRNA-seq data were obtained from the Gene Expression Omnibus database (GSE131907, n = 22). Bulk RNA-seq data and clinical information were obtained from TCGA-LUAD. Seven independent GEO cohorts, including GSE29016, GSE30219, GSE31210, GSE3141, GSE37745, GSE42127, and GSE50081, were used for external validation. Pan-cancer RNA-seq data and clinical phenotypes were downloaded from the UCSC Xena platform. Cross-cancer single-cell validation included 148 samples from 10 solid tumor types. Detailed dataset information is provided in Supplementary Table S1.

scRNA-seq processing and cell annotation

scRNA-seq data were processed using the Seurat R package [12]. Cells with 200–6,000 detected genes, 200–40,000 UMI counts, a mitochondrial gene ratio < 20%, and a red blood cell gene ratio < 5% were retained. Doublets were removed using scDblFinder. Data were normalized using SCTransform while regressing out cell-cycle effects and mitochondrial interference. After batch correction with Harmony, a KNN graph was constructed based on the top 20 principal components, followed by Louvain clustering. Clusters were visualized by UMAP [13–15]. Differentially expressed genes were identified using the FindAllMarkers function, and cell populations were annotated according to canonical marker genes.

Cell subset characterization, tissue preference, and trajectory analysis

The tissue distribution preference of each cell subset was evaluated using the observed-to-expected ratio (Ro/e), with Ro/e > 1 considered indicative of enrichment [16]. Subset-specific marker genes were used to characterize functional heterogeneity. GO enrichment analysis was performed using human gene annotations from org.Hs.eg.db, and the enrichment results and expression heatmaps were visualized using ClusterGVis [17]. Pseudotime trajectory analysis was primarily performed using Monocle 2 [18]. Highly variable genes were used as ordering genes, and dimensionality reduction was performed using the DDRTree algorithm to infer the dynamic progression of fibroblast and myeloid subpopulations. Monocle 3 was used as a supplementary validation of trajectory patterns [18, 19].

Bulk transcriptomic scoring, clinical correlation, and enrichment analysis

The ssGSEA algorithm implemented in the GSVA package was used to estimate the relative enrichment of specific cell subsets in TCGA-LUAD and external validation cohorts based on subset-specific marker genes derived from scRNA-seq data [20]. For the two main stromal–myeloid signatures, POSTN_CAF and SPP1_macro, 15 representative marker genes were selected for each signature to balance subset specificity and robustness. The CD8_Cytotoxic_T signature was constructed using corresponding cytotoxic T-cell marker genes from the single-cell analysis. These scores were interpreted as relative enrichment scores rather than absolute cell abundance. Associations between subset scores and clinicopathological features were evaluated using Wilcoxon tests, correlation analysis, and survival analysis. Optimal cutoff values were determined using the surv_cutpoint function, and Kaplan–Meier curves with log-rank tests were used to evaluate prognostic significance. Co-abundance between POSTN_CAF and SPP1_macro scores was assessed using Spearman correlation analysis.

Patients were stratified using the optimal cutoffs of the POSTN_CAF and SPP1_macro scores. The double-high, discordant, and double-low groups were designated as high-, intermediate-, and low-risk groups, respectively. For four-group analyses, the two discordant subgroups were analyzed separately. Differentially expressed genes between subgroups were identified using the limma package, and GSEA was performed using clusterProfiler with Hallmark, GO, and KEGG gene sets [21].

Spatial transcriptomic analysis and cell–cell communication inference

Spatial transcriptomic data were obtained from the GEO database (GSE307534) and processed using Seurat. Low-quality spots were filtered, and data were normalized using SCTransform. Dimensionality reduction, clustering, and spatial visualization were performed using standard Seurat workflows. A single-cell reference containing refined cell subsets was constructed, and RCTD was used to infer cell-type composition in spatial spots [22]. In addition, ssGSEA-based spatial module scores were calculated for POSTN_CAF, SPP1_macro, and CD8_Cytotoxic_T signatures.

The scRNA-seq dataset was used as the reference for RCTD-based deconvolution and cell-population annotation of each spatial transcriptomic section. Spatial CellChat analysis was then performed separately in samples P6 and P15 using the annotated spatial transcriptomic data [23]. Overexpressed ligand–receptor pairs and spatial communication probabilities were calculated for each sample. Global interaction numbers and strengths were visualized using circle plots. SPP1 and complement signaling pathways were further analyzed to identify major sender, receiver, mediator, and influencer populations. Ligand–receptor pairs from SPP1_macro to POSTN_CAF were visualized using bubble plots, and their spatial communication networks were mapped onto the corresponding tissue sections.

Immune microenvironment and immunotherapy response analysis

The IOBR package was used to evaluate immune infiltration and tumor microenvironment characteristics using multiple algorithms, including CIBERSORT, MCP-counter, quanTIseq, ESTIMATE, and IPS [24]. TMB, MSI, and TIDE scores were included to assess genomic and immune-response-related features [25]. The IMvigor210 cohort was used as an exploratory cross-cancer immunotherapy cohort to evaluate the association between POSTN/SPP1-related scores and immune checkpoint blockade response [26].

Multiplex immunofluorescence staining

The use of human tissue samples was approved by the Ethics Committee of The Fifth Affiliated Hospital of Zhengzhou University (Approval No. KY2026007). Multiplex immunofluorescence staining was performed on paraffin-embedded LUAD tissue sections. After deparaffinization, rehydration, and antigen retrieval, sections underwent sequential incubation with primary antibodies against EpCAM, POSTN, and SPP1, followed by TSA-based signal amplification. Nuclei were counterstained with DAPI. Images were acquired using a fluorescence microscope or panoramic tissue scanner.

Cell culture, conditioned media, and in vitro model induction

A549 human lung adenocarcinoma cells, HFL-1 human lung fibroblasts, and THP-1 human monocytes were obtained from Procell Life Science & Technology Co., Ltd. THP-1 cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum, whereas A549 and HFL-1 cells were cultured in DMEM/F12 containing 10% fetal bovine serum. All cells were maintained at 37 °C in a humidified incubator with 5% CO₂.

A549-conditioned medium was prepared by culturing A549 cells to approximately 70–80% confluence, replacing the medium with fresh complete medium, and collecting the supernatant after 24–48 h. The supernatant was centrifuged and filtered before use. To generate CAF-like HFL-1 cells (CAF-HFL-1), HFL-1 cells were treated with A549-conditioned medium for 48–72 h and validated by western blotting for FAP, POSTN, and α-SMA. To generate TAM-like THP-1 cells (TAM-THP-1), THP-1 cells were treated with 100 nM PMA for 24 h to induce macrophage-like differentiation and then stimulated with A549-conditioned medium for 24–48 h. TAM-like polarization was validated by western blotting for CD206, CD163, and SPP1.

After induction, TAM-THP-1 cells were washed with PBS and cultured in fresh medium for 24 h before supernatant collection. The resulting conditioned medium was defined as TAM-CM, whereas conditioned medium collected in parallel from PMA-THP-1 cells was defined as PMA-CM. For SPP1 neutralization, TAM-CM was pre-incubated with an SPP1 neutralizing antibody before being applied to CAF-HFL-1 cells.

siRNA transfection, RT-qPCR, flow cytometry, and western blotting

CAF-HFL-1 cells were transfected with siRNAs targeting CD44 or C3, or with negative-control siRNA, using a commercial siRNA transfection reagent according to the manufacturer’s instructions. Three independent siRNA sequences were tested for each target, and knockdown efficiency was evaluated by RT-qPCR. siCD44-2 and siC3-2 showed the highest knockdown efficiency and were used for subsequent functional experiments. Primer and siRNA sequences used in this study are listed in Supplementary Table S3.

Total RNA was extracted using TRIzol reagent, and RT-qPCR was performed according to the manufacturer’s instructions. Relative mRNA expression was calculated using the 2^-ΔΔCt method with GAPDH as the internal control. Commercially obtained human peripheral blood-derived primary CD8⁺ T cells were validated by flow cytometry before functional experiments using viability dye and antibodies against human CD45 and CD8. Viable CD45⁺CD8⁺ cells were used to confirm CD8⁺ T-cell purity.

For western blotting, total protein was extracted using RIPA buffer, separated by SDS-PAGE, and transferred to PVDF membranes. Membranes were incubated with primary antibodies against FAP, POSTN, α-SMA, CD206, CD163, SPP1, FN1, COL1A1, CD44, p-FAK, and GAPDH. Protein bands were visualized using enhanced chemiluminescence and quantified relative to GAPDH as the loading control. Information on antibodies and key reagents is provided in Supplementary Table S4.

CAF-HFL-1 functional assays

For the adhesion assay, CAF-HFL-1 cells were assigned to four treatment groups: PMA-CM, TAM-CM, TAM-CM plus SPP1 neutralizing antibody, and siCD44 plus TAM-CM. After treatment, cells were detached, counted, and seeded at equal densities into collagen I-coated 96-well plates. After 30–60 min, non-adherent cells were removed by gentle washing. Adherent cells were fixed, stained with crystal violet, imaged under a light microscope, and quantified by counting adherent cells.

For the collagen gel contraction assay, CAF-HFL-1 cells from different treatment groups were mixed with neutralized collagen I solution and seeded into culture plates. After gel polymerization, gels were detached from the well edges and cultured in the corresponding conditioned medium. Gel contraction was photographed after incubation, and gel area was quantified using ImageJ. Relative collagen gel area was calculated by normalizing the final gel area to the initial gel area.

CAF-HFL-1 barrier-based CD8⁺ T-cell migration and C3 feedback assays

For the CAF-HFL-1 barrier-based migration assay, CAF-HFL-1 cells were seeded into Transwell upper chambers with 8.0-µm pores to establish a stromal barrier and were treated with PMA-CM, TAM-CM, TAM-CM plus SPP1 neutralizing antibody, or siCD44 plus TAM-CM. A no-barrier group without CAF-HFL-1 cells was included as a migration control. Primary human CD8⁺ T cells were then added to the upper chamber, and A549-conditioned medium was added to the lower chamber as a chemoattractant. After incubation, migrated CD8⁺ T cells were collected, stained, imaged, and counted.

For the C3 feedback assay, CAF-HFL-1 cells were transfected with siNC or siC3-2 and stimulated with TAM-CM. Cells were then washed to remove residual TAM-CM and cultured in fresh low-serum medium. After 24 h, the supernatant was collected as CAF-derived conditioned medium. PMA-THP-1 cells were treated with control CAF-CM, siC3 CAF-CM derived from siC3-transfected CAF-HFL-1 cells, or siC3 CAF-CM supplemented with recombinant human C3 for 24 h. Cells were then harvested for western blot analysis of CD206 and SPP1 expression.

For additional recombinant-protein stimulation and antibody co-treatment assays, HFL-1 cells were assigned to the Control, recombinant SPP1 (rSPP1), and rSPP1 plus anti-ITGAV antibody groups, and p-FAK and POSTN protein expression was examined by western blotting. In parallel, HFL-1 cells were treated with recombinant human SPP1, and C3 and POSTN mRNA expression was measured by RT-qPCR. PMA-THP-1 cells were treated with recombinant human C3, and CD206 and SPP1 mRNA expression was measured by RT-qPCR. Key reagents used in these assays are provided in Supplementary Table S4.

Statistical analysis

Statistical analyses were performed using R software and GraphPad Prism. For bioinformatic analyses, two-group comparisons were performed using the Wilcoxon rank-sum test unless otherwise specified. Survival curves were generated using the Kaplan–Meier method and compared using the log-rank test. Correlations were assessed using Spearman or Pearson correlation analysis as appropriate. For in vitro experiments, Student’s t-test was used for two-group comparisons, and one-way ANOVA followed by appropriate multiple-comparison tests was used for multiple-group comparisons. All in vitro experiments were independently repeated at least three times. A P value < 0.05 was considered statistically significant.

Results

Single-cell atlas reveals stromal remodeling characteristics of the lung adenocarcinoma microenvironment

We analyzed single-cell RNA-sequencing data from GSE131907, including 11 primary LUAD samples and 11 distant normal lung samples. After stringent quality control, 81,210 high-quality cells were retained. Adequate mixing of cells from different samples in the UMAP plot indicated effective batch-effect correction (Fig. 1B). Unsupervised clustering of all cells was then performed (Fig. 1A), and cell populations were annotated into eight major lineages based on canonical marker genes (Fig. 1D, E; see Supplementary Table S5 for details), including T/NK cells, B cells, plasma cells, myeloid cells, fibroblasts, endothelial cells, epithelial cells, and mast cells. Tissue distribution and cell proportion analyses (Fig. 1C, F, G) showed that fibroblasts, B cells, and plasma cells were significantly enriched in tumor tissues, whereas the overall proportion of myeloid cells was relatively lower in tumor tissues. These findings prompted further analysis of fibroblast and myeloid heterogeneity.

Fig. 1.

Single-cell transcriptomic landscape of lung adenocarcinoma. (A–D) UMAP plots of 81,210 cells colored by cluster (A), sample (B), tissue type (C), and cell lineage (D). (E) Dot plot showing marker gene expression across eight major cell lineages. (F) Stacked bar plot showing cell lineage proportions in normal lung and tumor tissues. (G) Box plots comparing cell lineage fractions between normal lung and tumor tissues. ns, not significant; **P < 0.01; ***P < 0.001

Tumor-associated POSTN⁺ fibroblasts are associated with lung cancer progression

To further characterize fibroblast heterogeneity, we performed reclustering analysis and identified eight subpopulations with distinct functional states (Fig. 2A, B). Marker gene analysis identified several subsets, including quiescent SFRP1⁺ CAFs, myofibroblastic ACTA2⁺ CAFs, and POSTN⁺ CAFs with prominent stromal remodeling features (Fig. 2D). Tissue distribution analysis revealed a marked population shift (Fig. 2C): SFRP1⁺ CAFs were mainly distributed in adjacent normal tissues, whereas the Ro/e value of POSTN⁺ CAFs was increased in tumor tissues (Fig. 2G), supporting preferential tumor enrichment of this subset. Functional enrichment analysis showed that POSTN⁺ CAFs were associated with pathways related to collagen fibril organization and extracellular matrix organization (Fig. 2E), consistent with stromal remodeling features.

Fig. 2.

Fibroblast subpopulations in lung adenocarcinoma. (A, B) UMAP plots of fibroblast subpopulations in normal lung tissues (A) and tumor tissues (B). (C) Stacked bar plot showing the relative proportions of fibroblast subpopulations in normal and tumor tissues. (D) Dot plot showing marker gene expression across fibroblast subpopulations. Dot size indicates the percentage of cells expressing each gene, and color indicates relative expression. (E) Heatmap of signature gene expression and GO enrichment terms for fibroblast subpopulations. (F) Pseudotime trajectory analysis of fibroblast subpopulation differentiation. (G) Observed-to-expected ratio heatmap showing the distribution preference of fibroblast subpopulations in normal and tumor tissues. (H, I) Comparison of POSTN_CAF scores between normal and tumor tissues (H) and among different clinical stages (I). (J) Kaplan–Meier survival curve and risk table stratified by POSTN_CAF score. ns, not significant; **P < 0.01; ****P < 0.0001

Monocle 2 pseudotime analysis inferred a transcriptional-state trajectory among fibroblast subpopulations (Fig. 2F). SFRP1⁺ fibroblasts occupied an early pseudotime position, whereas one inferred branch extended toward a stromal-remodeling state. POSTN⁺ CAFs were located at the terminal end of this branch, suggesting a terminal fibroblast state with strong extracellular matrix remodeling capacity, and additional Monocle 3 analysis showed a consistent trajectory pattern (Supplementary Fig. S1A). In the TCGA cohort, the POSTN⁺ CAF score was significantly elevated in tumors (Fig. 2H) and increased with advancing TNM stage (Fig. 2I). Survival analysis showed that a high POSTN⁺ CAF score was associated with poor prognosis (Fig. 2J).

Enrichment of SPP1⁺ tumor-associated macrophages is associated with malignant progression

To characterize myeloid-cell heterogeneity, we constructed a refined atlas containing 10 subpopulations (Fig. 3A, B). Marker gene analysis identified key subsets, including resident MARCO⁺ macrophages and tumor-enriched SPP1⁺ macrophages (Fig. 3D). Tissue distribution and Ro/e analysis revealed a marked population shift during tumorigenesis (Fig. 3C, G): the MARCO⁺ subset, which was enriched in normal tissues, gradually declined, whereas SPP1⁺ macrophages expanded prominently in tumor tissues. Functional enrichment analysis showed that, compared with MARCO⁺ macrophages, the SPP1⁺ subset was enriched in pathways related to leukocyte chemotaxis and myeloid cell migration (Fig. 3E), suggesting an active signaling-associated state.

Fig. 3.

Myeloid cell subpopulations in lung adenocarcinoma. (A, B) UMAP plots of myeloid cell subpopulations in normal lung tissues (A) and tumor tissues (B). (C) Stacked bar plot showing the relative proportions of myeloid cell subpopulations in normal and tumor tissues. (D) Dot plot showing marker gene expression across myeloid cell subpopulations. Dot size indicates the percentage of cells expressing each gene, and color indicates relative expression. (E) Heatmap of signature gene expression and GO enrichment terms for myeloid cell subpopulations. (F) Pseudotime trajectory analysis of myeloid cell subpopulation differentiation. (G) Observed-to-expected ratio heatmap showing the distribution preference of myeloid cell subpopulations in normal and tumor tissues. (H, I) Comparison of SPP1_macro scores between normal and tumor tissues (H) and among different clinical stages (I). (J) Kaplan–Meier survival curve and risk table stratified by SPP1_macro score

Monocle 2 pseudotime analysis showed that SPP1⁺ macrophages were enriched at a late pseudotime position along the inferred myeloid-state trajectory (Fig. 3F), and Monocle 3 analysis showed a similar trajectory pattern (Supplementary Fig. S1B). In the TCGA cohort, the SPP1⁺ macrophage score was significantly increased in tumors (Fig. 3H) and continued to rise with advancing TNM stage (Fig. 3I). Survival analysis showed that a high SPP1⁺ macrophage score was associated with poor prognosis (Fig. 3J).

Spatial colocalization of POSTN⁺ CAFs and SPP1⁺ TAMs and the immune-excluded landscape

Co-abundance analysis across eight independent LUAD cohorts revealed a positive correlation between the enrichment scores of POSTN⁺ CAFs and SPP1⁺ TAMs (Fig. 4A). mIF staining further showed that POSTN and SPP1 were distributed in close proximity within the tumor stromal region (Fig. 4B). To resolve this niche at higher spatial resolution, we analyzed multiple LUAD samples using spatial transcriptomics. In representative samples P6 and P15 (Fig. 4C, E), regions with high POSTN_CAF and SPP1_macro scores showed extensive spatial overlap. In contrast, CD8_Cytotoxic_T scores were low in these regions and were mainly localized to the outer boundary of the stromal area (Fig. 4D, F). A similar spatial pattern was observed in independent validation samples P11 and P18 (Supplementary Fig. S2A, B). These findings suggest that the spatial organization of POSTN⁺ CAFs and SPP1⁺ TAMs is associated with an immune-excluded microenvironment and restricted T-cell infiltration.

Fig. 4.

Spatial organization of POSTN_CAF and SPP1_macro in lung adenocarcinoma. (A) Correlation scatter plots showing the association between POSTN_CAF and SPP1_macro scores across eight independent lung adenocarcinoma cohorts. (B) Representative multiplex immunofluorescence images showing DAPI, EPCAM, POSTN, and SPP1 staining in tumor tissues. (C, E) Hematoxylin and eosin staining and corresponding spatial clustering plots of samples #P6 (C) and #P15 (E). (D, F) Spatial score distribution maps of POSTN_CAF, SPP1_macro, and CD8_Cytotoxic_T in samples #P6 (D) and #P15 (F)

High POSTN_CAF and SPP1_macro scores are associated with poor prognosis and immune dysfunction

To evaluate the clinical relevance of the POSTN⁺ CAF–SPP1⁺ TAM niche, TCGA-LUAD patients were stratified into four subgroups according to the optimal cutoff values of the POSTN_CAF and SPP1_macro scores. The POSTN-high/SPP1-high subgroup was designated as the double-high or high-risk subgroup. Kaplan–Meier analysis showed that patients in the double-high subgroup had shorter overall survival than those in the other subgroups (Fig. 5A).

Fig. 5.

Prognostic stratification and functional enrichment based on POSTN_CAF and SPP1_macro scores. (A) Kaplan–Meier overall survival curves and risk tables for four subgroups stratified by POSTN_CAF and SPP1_macro scores. (B–H) GSEA plots showing significantly enriched Hallmark gene sets. (I) GO enrichment bubble plot based on differentially expressed genes. (J) KEGG enrichment bubble plot based on differentially expressed genes

To characterize the molecular features of the high-risk subgroup, we performed GSEA and enrichment analyses. Hallmark analysis showed enrichment of angiogenesis, apical junction, epithelial–mesenchymal transition, interferon responses, UV response down, and TNF-α signaling via NF-κB pathways (Fig. 5B–H). GO and KEGG analyses further showed enrichment of immune-related, stromal, adhesion, and cytoskeletal pathways (Fig. 5I, J). Together, these findings indicate that the POSTN/SPP1 double-high subgroup is characterized by stromal remodeling, inflammatory activation, and immune-related pathway remodeling.

Clinically, the proportion of advanced-stage tumors was higher in the POSTN/SPP1 double-high subgroup, whereas no obvious sex-specific distribution pattern was observed (Fig. 6A, B). TMB and MSI were also higher in the double-high subgroup (Fig. 6C, D). TME analysis showed that this subgroup had higher ESTIMATE, immune, and stromal scores but lower tumor purity, indicating a more prominent non-tumor microenvironmental component (Fig. 6E). The double-high subgroup also showed a higher TIDE score and lower IPS z-score, suggesting a microenvironmental state potentially associated with reduced sensitivity to immune checkpoint blockade. In the IMvigor210 exploratory cross-cancer immunotherapy cohort, higher POSTN⁺ CAF, SPP1⁺ TAM, and combined POSTN/SPP1 signatures were associated with poorer survival after immune checkpoint blockade (Fig. 6F–H). The POSTN/SPP1 double-high subgroup showed a lower proportion of clinical responders and a distinct immune infiltration pattern compared with the other subgroups (Fig. 6I, J). These exploratory findings suggest that the POSTN/SPP1 double-high subtype may be associated with reduced immunotherapy benefit, although LUAD-specific ICB cohorts are still needed for further validation.

Fig. 6.

Clinical and immunological features of POSTN/SPP1 risk subgroups. (A, B) Stacked bar plots showing the distributions of clinical stage (A) and sex (B) across different risk subgroups. (C, D) Box plots showing the distributions of tumor mutational burden (C) and microsatellite instability (D) among different risk subgroups. (E) Violin plots showing the distributions of ESTIMATE score, immune score, stromal score, tumor purity, IPS z-score, and TIDE score across different risk subgroups. (F–H) Kaplan–Meier survival curves and risk tables stratified by POSTN⁺ CAF score (F), SPP1⁺ TAM score (G), and combined POSTN/SPP1 grouping (H) in the immunotherapy cohort. (I) Stacked bar plot showing the proportions of clinical response categories in different POSTN/SPP1 subgroups. (J) Heatmap showing immune cell infiltration estimated by multiple computational algorithms, with age, sex, clinical stage, and subgroup annotations

scRNA-seq and spatial transcriptomics reveal intercellular interactions between POSTN⁺ fibroblasts and SPP1⁺ macrophages

To investigate spatial communication between POSTN⁺ CAFs and SPP1⁺ macrophages, we performed spatial CellChat analysis separately in LUAD spatial transcriptomic samples P6 and P15 after RCTD-based cell-population annotation. Global CellChat analysis showed heterogeneity in both the number of ligand–receptor interactions and communication weights among different cell subsets (Fig. 7A, B, I, J). Pathway-level analysis showed that SPP1⁺ macrophages acted as major signaling senders toward POSTN⁺ CAFs through the SPP1 signaling pathway (Fig. 7C, K). Ligand–receptor analysis identified SPP1–CD44 as a prominent interaction pair, together with several integrin-related SPP1 receptor pairs, in the SPP1_macro-to-POSTN_CAF communication direction (Fig. 7D, L).

Fig. 7.

Spatial CellChat analysis of SPP1 signaling between SPP1_macro and POSTN_CAF. (A–H) SPP1 signaling analysis in spatial transcriptomic sample P6. (A, B) Circle plots showing interaction number (A) and interaction strength (B). (C) Heatmap showing sender, receiver, mediator, and influencer roles in the SPP1 pathway. (D) Ligand–receptor bubble plot of SPP1_macro-to-POSTN_CAF communication. (E) Spatial communication network of the SPP1 pathway. (F–H) Spatial expression maps of SPP1 (F), ITGAV (G), and ITGB5 (H). (I–P) Corresponding SPP1 signaling analysis in independent sample P15, with panel annotations matching (A–H)

In addition to SPP1 signaling, complement pathway analysis suggested a potential feedback communication from POSTN⁺ CAFs to SPP1⁺ macrophages. In representative spatial transcriptomic samples, POSTN_CAF and SPP1_macro showed prominent communication roles within the complement signaling network, supporting a potential complement-related feedback loop between these two cell populations (Supplementary Fig. S3A–F).

To further examine the spatial basis of SPP1-mediated communication, we analyzed representative spatial transcriptomic samples. In sample P6, the spatial SPP1 communication network was mainly localized to stromal regions that corresponded to areas enriched for SPP1_macro and POSTN_CAF signals in the preceding spatial analysis, indicating a spatial interface between SPP1-producing macrophages and POSTN⁺ CAF-rich areas (Fig. 7E). Consistently, SPP1 expression was concentrated in discrete stromal regions, while the representative integrin-related receptor candidates ITGAV and ITGB5 showed adjacent or partially overlapping expression within the same stromal compartment (Fig. 7F–H). A similar spatial pattern was observed in the independent sample P15, supporting the reproducibility of this SPP1-centered stromal communication pattern (Fig. 7M–P). In addition, spatial expression maps of POSTN and CD44 in samples P6 and P15 showed that CD44 expression was located adjacent to or around POSTN-enriched stromal areas, supporting the spatial plausibility that CD44 may participate in signaling within the POSTN⁺ stromal niche (Supplementary Fig. S2C, D). Together with the ligand–receptor analysis, these findings support an SPP1-centered communication program from SPP1⁺ macrophages to POSTN⁺ CAFs, with CD44 prioritized as the main receptor candidate for functional validation, whereas ITGAV and ITGB5 represented additional integrin-related computational candidate branches.

TAM-derived SPP1–CD44 signaling supports CAF activation, stromal barrier formation, and C3 feedback in vitro

To examine the predicted SPP1⁺ macrophage–POSTN⁺ CAF interaction, we established CAF-like fibroblast and TAM-like macrophage models in vitro. A549-conditioned medium increased FAP, POSTN, and α-SMA expression in HFL-1 cells (Fig. 8A; Supplementary Fig. S4D) and CD206, CD163, and SPP1 expression in PMA-treated THP-1 cells (Fig. 8B; Supplementary Fig. S4E), supporting CAF-like and TAM-like phenotypes, respectively. Of three siCD44 sequences, siCD44-2 achieved the strongest knockdown in CAF-HFL-1 cells and was used subsequently (Fig. 8C). Commercially obtained primary human CD8⁺ T cells were validated by flow cytometry before functional assays (Fig. 8D).

Fig. 8.

In vitro validation of the SPP1–CD44 axis and C3 feedback. (A) Western blot analysis of FAP, POSTN, and α-SMA expression in HFL-1 and CAF-HFL-1 cells. (B) Western blot analysis of CD206, CD163, and SPP1 expression in PMA-THP-1 and TAM-THP-1 cells. (C) RT-qPCR analysis of CD44 expression in CAF-HFL-1 cells transfected with siNC or three CD44-targeting siRNAs. (D) Representative flow-cytometry plots showing the cell gate and the proportions of CD8⁺ and CD45⁺ cells in commercially obtained primary human CD8⁺ T-cell preparations. (E) Western blot analysis of FN1, COL1A1, and POSTN expression in CAF-HFL-1 cells under the indicated treatments. (F, G) Representative images and quantification of CAF-HFL-1 adhesion. (H, I) Quantification and representative images of collagen gel contraction. (J, L) Quantification and representative images of primary human CD8⁺ T-cell migration across CAF-HFL-1 barriers. (K) RT-qPCR analysis of C3 expression in CAF-HFL-1 cells transfected with siNC or three C3-targeting siRNAs. (M) Western blot analysis of CD206 and SPP1 expression in PMA-THP-1 cells treated with control CAF-CM, siC3 CAF-CM, or siC3 CAF-CM supplemented with recombinant human C3. Group labels PMA, TAM, TAM + αSPP1, and TAM+siCD44 denote PMA-CM, TAM-CM, TAM-CM plus SPP1-neutralizing antibody, and siCD44-transfected CAF-HFL-1 cells treated with TAM-CM, respectively. CAF-HFL-1, A549-CM-induced CAF-like HFL-1 cells; PMA-THP-1, PMA-differentiated THP-1 cells; TAM-THP-1, PMA-THP-1 cells further stimulated with A549-CM; PMA-CM, conditioned medium from PMA-THP-1 cells; TAM-CM, conditioned medium from TAM-THP-1 cells; CAF-CM, conditioned medium from CAF-HFL-1 cells; siC3 CAF-CM, conditioned medium from siC3-transfected CAF-HFL-1 cells; αSPP1, SPP1-neutralizing antibody; siNC, negative-control siRNA. Quantitative data are presented as mean ± SD. ***P < 0.001. Densitometric quantification of the western blot results shown in panels A, B, E, and M is provided in Supplementary Fig. S4D–G, respectively

Next, we examined whether TAM-like macrophage-conditioned medium could activate CAF-HFL-1 cells through SPP1–CD44 signaling. Compared with PMA-CM, TAM-CM increased FN1, COL1A1, and POSTN expression in CAF-HFL-1 cells, whereas SPP1 neutralization or CD44 knockdown attenuated this effect (Fig. 8E; Supplementary Fig. S4F). Functionally, TAM-CM enhanced CAF-HFL-1 adhesion, while SPP1 neutralization or CD44 knockdown reduced TAM-CM-induced adhesion (Fig. 8F, G). Consistently, TAM-CM promoted collagen gel contraction by CAF-HFL-1 cells, as reflected by reduced relative collagen gel area; this effect was weakened by SPP1 neutralization or CD44 knockdown (Fig. 8H, I).

To determine whether TAM-induced activation of CAF-HFL-1 cells affects T-cell migration, CAF-HFL-1 cells were treated with PMA-CM, TAM-CM, TAM-CM plus an SPP1-neutralizing antibody, or TAM-CM following CD44 knockdown and were subsequently used to establish a Transwell barrier. Primary human CD8⁺ T cells were then added to the upper chamber, and the number of cells migrating to the lower chamber was quantified. Compared with PMA-CM-treated CAF-HFL-1 cells, TAM-CM-treated CAF-HFL-1 cells significantly reduced CD8⁺ T-cell migration, whereas SPP1 neutralization or CD44 knockdown partially reversed this effect (Fig. 8J, L).

Based on the complement signaling prediction, we further examined whether CAF-HFL-1-derived C3 could provide a feedback signal to macrophages. Among three siC3 sequences, siC3-2 showed the strongest knockdown efficiency and was used in subsequent feedback assays (Fig. 8K). Compared with control CAF-CM, siC3 CAF-CM reduced CD206 and SPP1 expression in PMA-THP-1 cells, whereas supplementation with recombinant C3 partially restored their expression (Fig. 8M; Supplementary Fig. S4G). Additional recombinant-protein and antibody-intervention experiments showed that recombinant SPP1 increased p-FAK and POSTN protein expression in HFL-1 cells, whereas co-treatment with an anti-ITGAV antibody attenuated these effects (Supplementary Fig. S4A). Consistently, recombinant SPP1 also increased C3 and POSTN mRNA expression in HFL-1 cells (Supplementary Fig. S4B). In PMA-THP-1 cells, recombinant C3 increased CD206 and SPP1 mRNA expression (Supplementary Fig. S4C). These findings support an ITGAV-associated FAK-related response to SPP1 and further support the reciprocal C3-mediated stromal–myeloid feedback model. Together, these in vitro findings support the involvement of SPP1 and CD44 in TAM-CM-induced CAF-HFL-1 activation, stromal remodeling, and barrier-like function. FAK-related signaling may represent an additional branch of SPP1-induced fibroblast activation, while CAF-derived C3 may contribute to a feedback loop that reinforces the SPP1⁺/CD206⁺ macrophage-like phenotype.

Conservation analysis of POSTN⁺ fibroblasts and SPP1⁺ myeloid signatures in pan-cancer cohorts

To assess whether POSTN⁺ fibroblasts and SPP1⁺ myeloid cells represent conserved tumor-associated stromal–myeloid programs across cancer types, we integrated 148 scRNA-seq samples from 10 solid tumor types to construct a pan-cancer single-cell atlas. Unsupervised clustering separated major stromal, immune, and epithelial lineages, including fibroblasts, myeloid cells, epithelial cells, T/NK cells, B cells, plasma cells, mast cells, endothelial cells, and proliferative cells (Fig. 9A, B).

Fig. 9.

Pan-cancer characteristics and prognostic value of POSTN⁺ CAFs and SPP1⁺ macrophages. (A, B) UMAP plot of the pan-cancer single-cell atlas (A) and dot plot of marker gene expression across major cell lineages (B). (C, D) UMAP plots of fibroblast subpopulations in normal (C) and tumor tissues (D). (E) Relative proportions of fibroblast subpopulations. (F) Dot plot of fibroblast-subpopulation marker genes. (G, H) UMAP plots of myeloid subpopulations in normal (G) and tumor tissues (H). (I) Relative proportions of myeloid subpopulations. (J) Dot plot of myeloid-subpopulation marker genes. (K, L) Density distributions of POSTN_CAF (K) and SPP1_macro (L) scores. (M–O) Kaplan–Meier overall survival curves and risk tables stratified by POSTN_CAF score (M), SPP1_macro score (N), and the combined POSTN/SPP1 signature (O)

Within the fibroblast compartment, reclustering identified multiple CAF subsets, including POSTN_CAF, WNT2_CAF, SFRP1_CAF, ACTA2_CAF, and other fibroblast states (Fig. 9C, D). Marker gene expression and proportional analyses showed that POSTN_CAF and ACTA2_CAF were prominent tumor-associated fibroblast states, whereas SFRP1_CAF was relatively enriched in normal tissues (Fig. 9E, F). Within the myeloid compartment, reclustering identified SPP1⁺ macrophage-related and dendritic-cell-related subsets (Fig. 9G, H). SPP1⁺ myeloid signatures were broadly detectable across multiple tumor types and were enriched in tumor-associated myeloid states (Fig. 9I–L).

In the TCGA pan-cancer cohort, high POSTN_CAF score, high SPP1⁺ myeloid score, and the combined POSTN/SPP1 signature were associated with poorer overall survival (Fig. 9M–O). These results suggest that the POSTN⁺ CAF–SPP1⁺ myeloid cell niche is not limited to LUAD but may represent a conserved stromal–immune program across solid tumors.

Discussion

Dissecting the spatial architecture and intercellular communication of the tumor microenvironment is critical for understanding immune exclusion and therapeutic resistance in LUAD [27]. By integrating single-cell RNA sequencing, spatial transcriptomics, multiplex immunofluorescence, bulk transcriptomic validation, and in vitro functional assays, we further characterized a stromal–myeloid niche marked by the spatial colocalization of POSTN⁺ CAFs and SPP1⁺ TAMs. This niche was associated with stromal remodeling, immune-excluded spatial organization, poor prognosis, and exploratory indicators of reduced immunotherapy benefit. Mechanistically, our computational and in vitro findings support the involvement of SPP1–CD44 signaling in TAM-CM-induced activation of CAF-like fibroblasts and also indicate an integrin-associated FAK-related response. This process may enhance extracellular matrix remodeling, adhesive capacity, collagen gel contraction, and stromal barrier function. In addition, CAF-derived C3 may provide a feedback signal that reinforces the SPP1⁺/CD206⁺ macrophage-like phenotype. These findings extend current understanding of spatial stromal–immune interactions and suggest that the POSTN/SPP1 double-high signature may serve as a potential biomarker for immune-excluded LUAD.

Previous studies have separately reported pro-tumor roles for POSTN⁺ fibroblasts and SPP1⁺ macrophages in multiple malignancies. POSTN has been linked to extracellular matrix deposition, collagen organization, metastasis, and stromal stiffening [8, 28–30], whereas SPP1 has been recognized as an important mediator of macrophage polarization, fibroblast activation, and tumor-promoting inflammation through CD44- or integrin-related signaling. CD44 is a major receptor for SPP1 and is involved in cell adhesion, migration, extracellular matrix interaction, and inflammatory signaling [31]. In the present study, spatial transcriptomic analysis and representative multiplex immunofluorescence imaging showed co-enrichment of POSTN- and SPP1-associated signals within LUAD stromal regions. Regions enriched for both populations exhibited lower CD8_Cytotoxic_T scores, indicating a spatial pattern consistent with T-cell exclusion. Cell–cell communication analysis further identified SPP1 signaling from SPP1⁺ macrophages to POSTN⁺ CAFs. Among the predicted receptor branches, SPP1–CD44 interactions showed a more prominent and consistent communication pattern and were therefore prioritized for subsequent functional investigation, while ITGAV/ITGB5-related interactions were retained as additional candidate routes.

Functional experiments further supported the biological relevance of the SPP1–CD44 branch. TAM-CM increased POSTN, FN1, and COL1A1 expression in CAF-HFL-1 cells, enhanced fibroblast adhesion and collagen gel contraction, and reduced the migration of primary CD8⁺ T cells across a CAF-HFL-1-containing Transwell barrier. These effects were attenuated by SPP1 neutralization or CD44 knockdown, supporting a role for TAM-derived SPP1–CD44 signaling in CAF activation, stromal remodeling, and barrier-like function. In parallel, spatial transcriptomic analysis showed adjacent or partially overlapping expression of ITGAV and ITGB5 within SPP1-associated stromal regions. Recombinant SPP1 increased p-FAK and POSTN expression in HFL-1 cells, whereas anti-ITGAV co-treatment attenuated these responses, suggesting that an integrin-associated FAK-related response may represent an additional signaling branch contributing to SPP1-induced fibroblast activation. Collectively, these findings support the view that POSTN⁺ CAFs and SPP1⁺ TAMs form a spatially organized stromal–myeloid functional unit characterized by TAM-derived SPP1 signaling, CD44-dependent CAF activation and stromal barrier-like remodeling, impaired CD8⁺ T-cell migration, and a potential CAF-derived C3 feedback signal that may reinforce the SPP1⁺/CD206⁺ macrophage-like phenotype.

The C3-mediated feedback mechanism provides an additional layer of stromal–myeloid crosstalk. Complement signaling analysis suggested a potential feedback communication from POSTN_CAF to SPP1_macro, and functional experiments showed that siC3 CAF-CM reduced CD206 and SPP1 expression in PMA-THP-1 cells, whereas recombinant C3 partially restored these markers. These findings suggest that CAF-derived C3 may help maintain a macrophage phenotype characterized by SPP1 and CD206 expression, thereby reinforcing the stromal–myeloid niche. This reciprocal interaction provides a plausible explanation for how a localized stromal barrier may be established and maintained in LUAD. Together, these observations suggest that the POSTN/SPP1 niche may contribute to immune exclusion through combined effects on extracellular matrix remodeling, physical restriction of T-cell migration, macrophage polarization, and tumor-associated tissue remodeling.

This study also has potential clinical implications. Single biomarkers such as TMB or PD-L1 expression often provide limited predictive accuracy for immune checkpoint blockade response [32, 33]. Our findings suggest that combined assessment of POSTN_CAF and SPP1_macro signatures, or spatial detection of POSTN and SPP1 by multiplex immunofluorescence, may help identify patients with an immune-excluded phenotype and potentially limited benefit from immunotherapy. Moreover, because the TAM-derived SPP1–CD44 axis appears to contribute to CAF activation and stromal barrier function, targeting this pathway may help relieve stromal constraints and improve immune-cell infiltration. In this context, therapeutic strategies aimed at disrupting SPP1–CD44 signaling, blocking SPP1 activity, or modulating CAF-derived feedback signals such as C3 may warrant further investigation, particularly in POSTN/SPP1 double-high tumors.

Several limitations should be acknowledged. First, functional validation relied mainly on CAF-like HFL-1 cells and TAM-like THP-1 cells rather than patient-derived CAFs and primary TAMs. Thus, the proposed reciprocal SPP1–CD44/C3 stromal–myeloid circuit requires further validation in patient-derived co-culture systems, organoid models, and in vivo studies. Second, cell-subset annotation and spatial enrichment analyses were based largely on GSE131907-derived annotations and predefined marker signatures, which may introduce cohort-specific, marker-selection, and gene-set-overlap biases despite the use of compact signatures and complementary validation approaches. Third, the clinical analyses were predominantly retrospective, and the IMvigor210 results should be regarded as exploratory cross-cancer evidence rather than LUAD-specific immunotherapy validation. Prospective LUAD cohorts are therefore needed. Finally, communication analysis, spatial expression profiling, recombinant SPP1 stimulation, and anti-ITGAV co-treatment provided preliminary support for an ITGAV-associated FAK-related response; however, direct ITGAV/ITGB5 perturbation and FAK inhibition will be required to further define this branch and its relationship to the more strongly supported SPP1–CD44 pathway.

In summary, this study defines a spatially organized POSTN⁺ CAF–SPP1⁺ macrophage niche in LUAD that is associated with immune exclusion, poor prognosis, and potentially reduced immunotherapy benefit. TAM-derived SPP1 promotes CAF activation and stromal barrier formation mainly through CD44, with an additional integrin/FAK-related response. CAF-derived C3 may further reinforce the SPP1⁺/CD206⁺ macrophage-like phenotype. This conserved stromal–myeloid program may provide a framework for biomarker development and therapeutic targeting in immune-excluded tumors.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ACTA2

Actin Alpha 2, Smooth Muscle

A549-CM

A549-Conditioned Medium

BP

Biological Process

C3

Complement Component 3

CAF-CM

CAF-Derived Conditioned Medium

CAF-HFL-1

CAF-Like HFL-1 Cells

CAFs

Cancer-Associated Fibroblasts

CC

Cellular Component

CD44

Cluster of Differentiation 44

COL1A1

Collagen Type I Alpha 1 Chain

CR

Complete Response

DEGs

Differentially Expressed Genes

ECM

Extracellular Matrix

EMT

Epithelial–Mesenchymal Transition

EpCAM

Epithelial Cell Adhesion Molecule

FAP

Fibroblast Activation Protein

FBS

Fetal Bovine Serum

FN1

Fibronectin 1

GAPDH

Glyceraldehyde-3-Phosphate Dehydrogenase

GEO

Gene Expression Omnibus

GO

Gene Ontology

GSEA

Gene Set Enrichment Analysis

GSVA

Gene Set Variation Analysis

ICB

Immune Checkpoint Blockade

ICIs

Immune Checkpoint Inhibitors

IPS

Immunophenoscore

KEGG

Kyoto Encyclopedia of Genes and Genomes

LUAD

Lung Adenocarcinoma

MF

Molecular Function

mIF

Multiplex Immunofluorescence

MSI

Microsatellite Instability

OS

Overall Survival

PD

Progressive Disease

PD-1

Programmed Cell Death Protein 1

PD-L1

Programmed Cell Death Ligand 1

PMA

Phorbol 12-Myristate 13-Acetate

PMA-CM

PMA-THP-1-Conditioned Medium

POSTN

Periostin

PR

Partial Response

RCTD

Robust Cell Type Decomposition

rC3

Recombinant Human C3 Protein

Ro/e

Ratio of Observed to Expected Values

RT-qPCR

Real-Time Quantitative Polymerase Chain Reaction

scRNA-seq

Single-Cell RNA Sequencing

siRNA

Small Interfering RNA

SPP1

Secreted Phosphoprotein 1

ssGSEA

Single-Sample Gene Set Enrichment Analysis

ST

Spatial Transcriptomics

α-SMA

Alpha-Smooth Muscle Actin

TAM-CM

TAM-Like THP-1-Conditioned Medium

TAM-THP-1

TAM-Like THP-1-Derived Macrophages

TAMs

Tumor-Associated Macrophages

TCGA

The Cancer Genome Atlas

TIDE

Tumor Immune Dysfunction and Exclusion

TMB

Tumor Mutational Burden

TME

Tumor Microenvironment

TSA

Tyramide Signal Amplification

UMAP

Uniform Manifold Approximation and Projection

WB

Western Blotting

PMA-THP-1

PMA-treated THP-1-derived macrophage-like cells

FAK

Focal Adhesion Kinase

ITGAV

Integrin Subunit Alpha V

ITGB5

Integrin Subunit Beta 5

p-FAK

Phosphorylated Focal Adhesion Kinase

rSPP1

Recombinant Human SPP1 Protein

siCD44

CD44-targeting small interfering RNA

siC3

C3-targeting small interfering RNA

Author contributions

Ankang Zhu, Jianyuan Huang, Juan Zhang, and Wenxue Wei contributed equally to this work, and Xiaojie Pan and Xing Lin jointly supervised the study. Ankang Zhu, Jianyuan Huang, and Juan Zhang conceived and designed the study, performed bioinformatic analyses, and drafted the manuscript. Ankang Zhu and Jianyuan Huang conducted in vitro experiments. Wenxue Wei, Ziyan Xu and Haobo Wang contributed to data collection and preprocessing. Shaolin Lin, Wei Wang, Jiguang Zhang, and Jiewei Luo assisted with data analysis and interpretation. Juan Zhang and Xiaojie Pan contributed to data visualization and figure preparation. Xing Lin critically revised the manuscript and provided overall guidance. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No. 82074189) and the Natural Science Foundation of Fujian Province (Grant No. 2023J011164).

Data availability

The datasets analysed during the current study are available in the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/) under the accession numbers GSE131907, GSE29016, GSE30219, GSE31210, GSE3141, GSE37745, GSE307534, GSE42127, and GSE50081. The bulk transcriptomic data and corresponding clinical information for the TCGA-LUAD and Pan-cancer cohorts are available from the UCSC Xena platform (https://xena.ucsc.edu/). All other data supporting the findings of this study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The patient cohorts and associated clinical data analyzed in this study were partly obtained from publicly available databases, including TCGA, GEO, and UCSC Xena. As these data are publicly accessible and de-identified, the requirement for local ethical approval was waived, and their use adhered to the data access policies of the respective databases. For the experimental validation, the collection and use of human LUAD tissue samples were approved by the Ethics Committee of The Fifth Affiliated Hospital of Zhengzhou University (Approval No.: KY2026007). The study was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participating patients prior to sample collection.

Consent for publication

Not applicable.

Declaration of generative AI and AI-assisted technologies in the writing process

ChatGPT (OpenAI) was used solely for language polishing. All content was reviewed and approved by the authors.

Authors’ information

Not applicable.

Competing interests

The authors declare that they have no competing interests.