Single-cell RNA sequencing (scRNA-seq) data were obtained from the GSE103322 (38) and GSE139324 (38,39) datasets acquired from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/). GSE103322 dataset was composed of 18 HNSCC patient samples (containing 5,902 single cells). GSE139324 dataset involved 26 paired samples of peripheral blood mononuclear cells (PBMCs) and tumor-infiltrating immune cells from HNSCC patients 18 human papillomavirus (HPV)-negative and eight HPV-positive), six PBMC samples from healthy donors and five tissue-resident immune cell samples from healthy tonsils (totaling 131,224 cells). The two datasets were merged, resulting in a total of 136,206 single cells.

Data processing and analysis were carried out in R (version 4.4.2; http://www.r-project.org/) using the Seurat package (version 5.3.1; http://satijalab.org/seurat/). Quality control was performed by retaining cells with 300–7,000 detected genes (nFeature_RNA) and less than 10% mitochondrial gene expression (percent.mt). The top 2,000 variable genes were selected for downstream analysis. Nonlinear dimensionality reduction was conducted via Uniform Manifold Approximation and Projection (UMAP), based on 15 principal components and a clustering resolution of 0.5.

Additionally, transcriptomic FPKM data from The Cancer Genome Atlas (TCGA)-HNSCC patients were obtained from cBioPortal (https://www.cbioportal.org/). To correct for batch effects across subgroups, Harmony (version 0.1.0; http://github.com/immunogenomics/harmony) was applied. Subsequent subgroup re-analysis adhered to the standard Seurat workflow. Immune infiltration levels, assessed via single-sample gene set enrichment analysis (ssGSEA), acquired from the TISIDB database (http://cis.hku.hk/TISIDB/download.php).

Dimensionality reduction was first performed on the dataset by applying principal component analysis (PCA) to the top 2,000 highly variable genes. Subsequently, cell clustering was carried out using Seurat's FindClusters function with a resolution of 0.7. Finally, marker genes for the resulting cell subpopulations were identified. All cells annotated based on the conventional markers as previously reported. The following markers were used: Epithelial makers (EpCAM, SFN, KRT14 and KRT5); Fibroblasts markers (FN1, DCN, COL1A1, COL1A2, COL3A1 and COL6A1); T cell markers (CD2, CD3D, CD3E and CD3G); B cell markers (CD19, CD79A, MS4A1, JCHAIN and MZB1); Myeloid markers (CD68, LYZ, CD14, IL3RA, LAMP3, CLEC4C and TPSAB1). Subsequently, the fibroblasts (containing 1,761 single cells) were re-clustered into two subclusters, POSTN⁺ fibroblasts and POSTN⁻ fibroblasts. Myeloid (containing 30,018 single cells) were re-clustered into two subclusters, M1-like macrophage and M2-like macrophage for further analysis.

Gene expression data of TCGA-HNSCC patients (https://portal.gdc.cancer.gov/projects/TCGA-HNSCC) were downloaded through the TCGA biolinks R package (version 2.34.1; http://bioconductor.org/packages/TCGAbiolinks/). Transcripts per million reads (TPM) were transformed to Log₂ (TPM + 1) for further analysis. High and low expression groups were determined according to the median values of POSTN and FAP. The potential response to immune checkpoint blockade ICB was predicted by the tumor immune dysfunction and exclusion (TIDE) algorithm. The expression levels of POSTN and FAP along with immune infiltration levels were quantified using ssGSEA. Based on the ssGSEA enrichment scores of 28 immune signatures, the present study performed unsupervised hierarchical clustering to identify immune-related subtypes. TME was further characterized by calculating the StromalScore, ImmuneScore, ESTIMATEScore using the ESTIMATE algorithm and tumor purity was inferred accordingly. Similarly, molecular subtypes were identified by clustering the entire gene expression dataset. The distinction between POSTN⁺ and POSTN⁻ fibroblasts signatures was visualized using t-distributed stochastic neighbor embedding (t-SNE).

Patient inclusion and exclusion criteria were established to define a homogeneous cohort for analysis. Inclusion criteria comprised: i) a pathological diagnosis of oral squamous cell carcinoma (OSCC); ii) consecutive enrollment from patients undergoing primary surgical resection at Shanghai Stomatological Hospital; iii) availability of adequately preserved formalin-fixed paraffin-embedded (FFPE) tumor tissues; and iv) availability of complete clinical data. Exclusion criteria were: i) Receipt of neoadjuvant therapy prior to surgery; ii) presence of another simultaneous active malignancy; iii) insufficient or poor-quality FFPE blocks unsuitable for sectioning; and iv) incomplete clinical records.

Based on these criteria, paraffin-embedded tissue sections (4 µm) from 40 patients with oral squamous cell carcinoma (OSCC) obtained from Shanghai Stomatological Hospital who underwent surgical resection were included in this study, with 14 normal oral mucosa samples serving as controls. The cohort comprised 26 male and 14 female patients, with 17 patients aged ≤62 years and 23 patients aged ≥62 years. The mean age was 62 years. Clinical and histopathological characteristics of the cohort are summarized in Table SIII and all diagnoses were confirmed by hematoxylin and eosin (H&E) staining.

Clinical and histopathological characteristics of the cohort are summarized in Table SIII and all diagnoses were confirmed by H&E staining.

Immunohistochemical staining was conducted using the SPlink Detection Kit (SP-9000; Beijing Zhongshan Jinqiao Biotechnology Co., Ltd.). Briefly, tissue sections were deparaffinized in xylene, rehydrated through a graded ethanol series and treated with 3% hydrogen peroxide in methanol to quench endogenous peroxidase activity. After blocking nonspecific binding with 10% goat serum (cat. no. C0265; Beyotime Biotechnology, Inc.), the sections were incubated overnight at 4°C with primary antibodies against FAP (cat. no. ab314075; 1:200; Abcam) and POSTN (cat. no. ab14041; 1:100; Abcam). Following this, the sections were incubated with biotinylated goat anti-rabbit IgG (cat. no. ab64256; 1:200; Abcam). at room temperature for 20 min and then washed with PBS at room temperature for 10 min. Immunoreactivity was visualized using 3,3′-diaminobenzidine (DAB) as the chromogen detection, followed by hematoxylin counterstaining at room temperature for 5 min. Negative controls were processed similarly, with the primary antibody replaced by PBS.

To ensure analytical reliability, a rigorous quality assessment was implemented. All stained slides were independently evaluated by two certified pathologists blinded to the clinicopathological data. Slides were included in the final analysis only if they met the following pre-defined criteria: i) adequate control staining (positive/negative controls performing as expected); ii) optimal morphological preservation of tissue structure; and iii) low non-specific background interference. Several slides were excluded during this quality control process due to failure to meet these standards.

Stained sections were air-dried and examined under a Leica Thunder DM6B optical microscope (Leica Microsystems GmbH). Image-Pro Plus software (version 6.0; Media Cybernetics, Inc.) was used to measure the integrated optical density and the area of target protein distribution. The mean density per case was determined by averaging values from at least 10 randomly selected fields.

A total of four representative HNSCC tissue samples were used to capture biological heterogeneity while enabling detailed exploratory analysis. Tissue sections from four HNSCC cases were deparaffinized, rehydrated and subjected to heat-induced antigen retrieval in pH 6.0 citrate buffer (cat. no. C1010; Beijing Solarbio Science & Technology Co., Ltd.) at 95°C for 5 min. Subsequently, the sections were washed with PBS at room temperature for 10 min. After blocking nonspecific binding with 10% QuickBlock Blocking Buffer (cat. no. P0220; Beyotime Biotechnology Inc.), the sections were incubated overnight at 4°C with the following primary antibodies: POSTN (cat. no. ab14041; 1:100; Abcam) and CD163 (cat. no. sc-33715; 1:1,000; Santa Cruz Biotechnology, Inc.). The following day, sections were washed with 1X TBST (containing 0.1% Tween-20) at room temperature for 30 min and then incubated at room temperature for 60 min with species-appropriate fluorescent secondary antibodies: Goat Anti-Rabbit IgG AF488 (cat. no. ab150077; 1:200; Abcam) or Goat Anti-Mouse IgG AF594 (cat. no. ab150116; 1:200; Abcam). The sections were washed with 1X TBST at room temperature for 10 min. Nuclei were stained with DAPI (1:3,000; Thermo Fisher Scientific, Inc.) at room temperature for 5 min and then washed with 1X TBST at room temperature for 10 min. images were acquired using a Leica Thunder DM6B optical microscope (Leica Microsystems GmbH) for subsequent analysis.

The primary cells used in this study-carcinoma-associated fibroblasts (CAFs) derived from human OSCC tissues (CAF-S5 and CAF-S6) and healthy gingival-derived normal fibroblasts (NFs) are identical to those that were isolated and characterized in our previous study (37). CAFs and NFs were maintained in DMEM/F12 medium (Gibco; Thermo Fisher Scientific, Inc.) containing 10% Fetal bovine serum (FBS; ScienCell Research Laboratories, Inc.). Human immortalized monocytic cell line THP-1 was purchased from iCell Bioscience (CVCL_0006) and grown in RPMI 1640 medium (Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% FBS (ScienCell Research Laboratories, Inc.). All cultured cells were supplemented with 100 U/ml penicillin and 100 U/ml streptomycin (Gibco; Thermo Fisher Scientific, Inc.) and maintained at 37°C in a humidified atmosphere with 5% CO₂.

POSTN was silenced by CAFs with concentrated lentivirus particles short hairpin (sh)RNA negative control (shNC; sense:: 5′-GTTCTCCGAACGTGTCACGT-3′; anti-sense: 5′-ACGTGACACGTCGGAGAAC-3′) and POSTN-targeting shRNA (shPOSTN; sense: 5′-CCCATGGAGAGCCAATTAT-3′; anti-sense: 5′-ATAATTGGCTCTCCATGGG-3′). The lentiviral particles and the shRNA constructs were designed and provided by Shanghai GenePharma Co., Ltd. Specifically, the POSTN-targeting shRNA was cloned into the lentiviral vector pHBIV-U6-Scramble-ZsGreen-Puro, and the construct was verified by double enzyme digestion and DNA sequencing.

Human immortalized embryonic kidney cell line 293T (CVCL_0063) was purchased from Shanghai Fuheng Biology and cultured in DMEM High medium (Gibco; Thermo Fisher Scientific, Inc.) supplemented with 10% FBS (ScienCell Research Laboratories, Inc.), 100 U/ml penicillin, and 100 U/ml streptomycin (Gibco; Thermo Fisher Scientific, Inc.) and maintained at 37°C in a humidified atmosphere with 5% CO₂.

When 293T cells reached 85–90% confluence in 10 cm dishes were co-transfected with 10 µg of the lentiviral shRNA vector (pHBIV-U6-Scramble-ZsGreen-Puro), 7.5 µg of the packaging plasmid pSPAX2, and 2.5 µg of envelope plasmid pMD2G using Lipofectamine 3000 (L3000015; Thermo Fisher Scientific, Inc.), following a second-generation system. This combination yielded a plasmid mass ratio of 4:3:1. Viral supernatant was harvested 48 h post-transfection, clarified by centrifugation (3,000 rpm at room temperature for 10 min), and filtered through a 0.45 µm membrane. The viral particles were then concentrated via ultracentrifugation (50,000 × g at 4°C for 2 h) and resuspended in 200 µl of DMEM High medium (Gibco; Thermo Fisher Scientific, Inc.). Subsequently, CAF-S5 and CAF-S6 were incubated overnight with 8 mg/ml Polybrene (cat. no. TR-1003; Sigma-Aldrich; Merck KGaA) and either control lentiviral particles or those carrying POSTN-targeting shRNA at a multiplicity of infection (MOI) of 100. At 48 h post-transfection, media containing 2 mg/ml puromycin (Sigma-Aldrich; Merck KGaA) were used to select stable clones for two weeks.

The human monocytic leukemia cell line THP-1 was grown in a T25 culture flask supplemented with RPMI 1640 medium(Gibco; Thermo Fisher Scientific, Inc.) containing 50 ng/ml of phorbol 12-myristate 13-acetate (PMA; cat. no. P8139; Sigma-Aldrich; Merck KGaA) for 48 h. Then, they were maintained in fresh RPMI 1640 culture medium for 24 h to stabilize their growth.

To confirm that BMP4 promotes macrophage polarization toward the M2 phenotype, THP-1 (2×10⁶ cells) were seeded in 6-well plates and differentiated into macrophages using 50 ng/ml PMA for 48 h. After stabilization, macrophages were assigned to different groups and then either left unstimulated or stimulated for an additional 48 h with recombinant human BMP4 (50 or 100 ng/ml; cat. no. HY-P7007; MedChemExpress). Macrophages were stained with FITC anti-human CD163 antibody (cat. no. 333617; 1:50; Biolegend Inc.) and FITC anti-human CD206 antibody (cat. no. 321103; 1:50; Biolegend Inc.) at 4°C for 15 min and then centrifugation 1,000 × g at 4°C for 5 min. Pellets were resuspended in 1 ml PBS and analyzed the expression of M2 macrophage polarization biomarkers (CD163 and CD206) was then analyzed by Flow cytometry (NovoCyte 2040R; ACEA Biosciences) equipped with NovoExpress software (version 1.6.3; http://explore.agilent.com/ACEA-joins-Agilent). To further to confirm the role of Smad signaling pathway, 10 µM LDN193189 inhibitor (cat. no. HY-12071; MedChemExpress) was added to each well for 48 h during the BMP4 stimulation. Cellular proteins were then harvested and subjected to western blotting to evaluate the expression of markers associated with Smad-related and M2 macrophage polarization.

sEVs isolation methods were performed as previously described (37). EV-depleted FBS was obtained by centrifugation at 100,000 × g at 4°C for 3 h. At 80–90% confluence, CAFs and NFs were washed with PBS three times and cultured in DMEM/F12 medium with 5% EV-depleted FBS for another 72 h. The resulting medium was collected as conditioned medium (CM). Then, CM was differentially centrifuged 500 × g for 10 min, 2,500 × g at 4°C for 20 min, 12,000 × g at 4°C for 30 min, filtered (pore size 0.22 µm), collected by ultracentrifuged at 100,000 × g at 4°C for 70 min. The pellet was diluted in 20 ml PBS and ultracentrifuged at 100,000 × g at 4°C for 70 min. The concentration of sEVs was determined by Nanoflow (N30E; NanoFCM, Co., Ltd.). Purified sEVs in PBS were placed on a formvar carbon-coated 200 mesh grid (cat. no. 200M-Cu; EMCN, Inc.) at room temperature for 20 min, washed twice with PBS and then stained 2% uranyl acetate at room temperature for 60 sec. Analysis of sEVs morphology and structure was performed with a transmission electron microscope (TEM; JEM-2000EX; JEOL, Ltd.) equipped with a CCD camera (Gatan SC1000, Model 832). Images were acquired and analyzed using digital Micrograph software (version 3.6.1; Gatan, Inc.). Positive and negative marker expression of sEVs were confirmed by western blotting. The particles size and concentration of sEVs were determined by Nanoflow (N30E, NanoFCM). sEVs were labeled with PKH26 (cat. no. PKH26PCL, Sigma-Aldrich; Merck KGaA) according to the company's instructions. Labeled sEVs were re-separated by ultracentrifugation at 100,000 × g at 4°C for 70 min.

Purification of sEVs was performed using an OptiPrep density gradient, as previously described (40). In brief, 60, 50, 40, 30, 25, 15, 10 and 5% (w/v) iodixanol solutions were prepared by diluting OptiPrep (60% (w/v) (cat. no. D1556, Sigma-Aldrich; Merck KGaA) with 0.25 M sucrose/10 mM Tris, pH 7.5 in Ultra-Clear tubes (cat. no. 344059; Beckman Coulter, Inc.). Purified sEVs were subjected to density gradient ultracentrifugation. The sample was loaded onto a continuous gradient and centrifuged at 100,000 × g at 4°C for 18 h using an Optima XPN-100 ultracentrifuge with a Ti41 rotor (Beckman Coulter, Inc.), resulting in eight collected fractions. Each fraction was diluted in PBS, pelleted by ultracentrifugation at 100,000 × g at 4°C for 70 min in Ultra-Clear tubes and finally washed and resuspended in PBS.

Purified sEVs were adsorbed onto formvar carbon-coated grids for 20 min. After three washes with PBS, free aldehyde groups were quenched by incubation with 50 mmol/l glycine. Following nine additional PBS washes, nonspecific sites were blocked with 5% BSA for 1 h. Grids were then incubated overnight at 4°C with primary antibodies (cat. no. CD9/POSTN; 10 µg/ml; Abcam). After incubation at room temperature with a 5 nm colloidal gold-conjugated secondary antibody (cat. no. A31565; 1:50; Thermo Fisher Scientific, Inc.) and a wash with PBS containing 0.1% BSA, samples were fixed at room temperature in 2.5% glutaraldehyde for 15 min. Finally, after thorough deionized water rinsing, grids were counterstained with phosphotungstic acid at room temperature for 5 min and imaged using a JEM-2000EX transmission electron microscope (JEOL, Ltd.).

The purified sEVs in 50 µl PBS were fluorescently labeled with 0.4 µl PKH26 Dye (Sigma-Aldrich; Merck KGaA) at room temperature for 20 min. The labeled sEVs were then centrifuged at 100,000 × g at 4°C for 70 min. After discarding the supernatant, equal amounts of sEVs were added to the macrophages at 37°C for 2 h and 6 h, respectively. Macrophages were then fixed in 4% paraformaldehyde at room temperature for 10 min. After centrifugation at 1,000 × g at room temperature for 5 min, the macrophages were resuspended in 1 ml PBS and the percentage of fluorescence was analyzed using a Flow Cytometer (NovoCyte 2040R; ACEA Biosciences).

To confirm whether shNC and shPOSTN sEV-derived CAF-S5/-S6 induce macrophage polarization into M1 and M2 macrophage, sEVs (5×10⁹ particles) were added to macrophage (2×10⁶ cells) grown in 6-well plate for 48 h. To test whether CAF-S5/-S6 sEV-POSTN directly induced M2 macrophage polarization, sEVs (5×10⁹ particles) were incubated overnight 4°C with a POSTN-neutralizing antibody. These pre-treated sEVs were then co-incubated with macrophage (2×10⁶ cells) grown in 6-well plate at 37°C for 48 h. For staining, 100 µl of macrophage (1×10⁶ cells) was incubated with 5 µl of FITC anti-human CD80 antibody (cat. no. 375405; BioLegend, Inc.), FITC anti-human CD86 antibody (cat. no. 374203; BioLegend, Inc.), FITC anti-human CD163 antibody (cat. no. 333617; BioLegend, Inc.) and FITC anti-human CD206 antibody (cat. no. 321103; BioLegend, Inc.) on ice for 15 min and washed twice with PBS. The stained cells resuspended in 1 ml PBS and analyzed on a Flow cytometer (NovoCyte 2040R, ACEA Biosciences) equipped with NovoExpress software (version 1.6.3; http://explore.agilent.com/ACEA-joins-Agilent).

Total RNA was extracted using TRIzol^® (Thermo Fisher Scientific, Inc.) according to the manufacture's protocol. Brief, THP-1-derived macrophages (2×10⁶ cells) treated with PBS (CTRL), shNC sEVs (5×10⁹ particles) and shPOSTN sEVs (5×10⁹ particles) for 48 h. Total RNA was precipitated using isopropanol, washed with 100% ethanol and resuspended in 30 µl of DEPC water. RNA concentration was quantified using a Nanodrop One spectrophotometer (Thermo Fisher Scientific, Inc.). Reverse transcription was performed with the PrimerScript RT reagent Kit (Takara Biotechnology Co., Ltd.) according to the manufacture's protocol, followed by quantitative real-time PCR using the SYBR Premix Ex Taq reagent kit (Takara Biotechnology Co., Ltd.) according to the manufacture's protocol on a LightCycler 480II system (Roche Diagnostics) with the following cycling protocol: Initial denaturation at 95°C for 15 min; followed by 40 cycles of denaturation at 95°C for 10 sec, annealing at 56–60°C for 20 sec and extension at 72°C for 32 sec. GAPDH was used as an internal control. The data for each sample were semi-quantitatively analyzed using the 2^−ΔΔCq method (41). The primers for CD68, CD80, CD86, CD163, CD206, BMP4, TNF-α, IL-6, TGF-β, IL-10 and GAPDH were purchased from Sangon Biotech Co., Ltd. Primer sequences were as following: CD68 (forward): 5′-AGCCACAAAACCACCACTCA-3′, (reverse): 5′-CTAGTGGTGGCAGGACTGTG-3′; CD80 (forward): 5′-GCAGGGAACATCACCATCCA-3′, (reverse): 5′-CACTTCCTTGGTCACGTGGA-3′; CD86 (forward): 5′-GGAAGAAGAAGAAGCGGCCT-3′, (reverse): 5′-CGCTGGGCTTCATCAGATCT-3′; CD163 (forward): 5′-GCGGGAGAGTGGAAGTGAAA-3′; (reverse): 5′-ACCTGCACTGGAATTAGCCC-3′; CD206 (MRC1) (forward): 5′-AGGATGGGTACTGGGCAGAT-3′; (reverse): 5′-CTGGACCTTGGCTTCGTGAT-3′; BMP4 (forward): 5′-GGAGGAGGAGGAAGAGCAGA-3′, (reverse): 5′-TTCTTCGTGGTGGAAGCTCC-3′; TNF-α (forward): 5′-CTTCCAGCTGGAGAAGGGTG-3′, (reverse): 5′-CCCAAAGTAGACCTGCCCAG-3′; IL-6 (forward): 5′-AGTGAGGAACAAGCCAGAGC-3′, (reverse): 5′-GGTCAGGGGTGGTTATTGCA-3′; TGF-β1 (forward): 5′-GCCCTGGACACCAACTATT-3′; (reverse): 5′-AGGCTCCAAATGTAGGGG-3′; IL-10 (forward): 5′-AAGACCCAGACATCAAGGCG-3′, (reverse): 5′-AGGCATTCTTCACCTGCTCC-3′; GAPDH (forward): 5′-GTGAAGGTCGGAGTCAACG-3′, (reverse): 5′-TGAGGTCAATGAAGGGGTC-3′. Each experiment was repeated at least three times.

Total protein was extracted with RIPA buffer (cat. no. R0010; Beijing Solarbio Science & Technology Co., Ltd.) supplemented with protease (cat. no. HY-K0010; MedChemExpress) and phosphatase inhibitors (cat. no. HY-K0021; MedChemExpress) at 4°C for 30 min. The lysates were then subjected to protein concentration measurement with a BCA kit (cat. no. P0009; Beyotime Biotechnology). Proteins (20 µg per lane) were separated by 4–20% SDS-PAGE and subsequently transferred to a PVDF membrane (MilliporeSigma). The membrane was then blocked with 5% fat-free milk for 1 h at room temperature to prevent non-specific binding, followed by an overnight incubation with primary antibodies at 4°C. The primary antibodies included POSTN antibody (cat. no. ab14041; 1:500; Abcam), CD9 (cat. no. ab236630; 1:1,000; Abcam), CD81 (cat. no. ab79559; 1:1,000; Abcam), Calnexin (10427-2-AP; 1:1,000, Proteintech, Wuhan, China), CD80 (cat. no. ab134120; 1:1,000; Abcam), CD86 (cat. no. abs115477; 1:1,000; Absin Bioscience), CD163 (sc-33715; 1:1,000, Santa Cruz Biotechnology), CD206 (cat. no. ab64693; 1:1,000; Abcam), BMPR2 (cat. no. abs147034; 1:1,000; Absin Bioscience), Phospho-Smad 1 (Ser463/465)/Smad5(Ser463/465)/Smad9(Ser465/467) (cat. no. 13820; 1:1,000; Cell Signaling Technology, Inc.), Smad5 (12534; 1:1,000; Cell Signaling Technology, Inc.), Smad9 (cat. no. abs131190; 1:1,000; Absin Bioscience) and GAPDH (cat. no. 10494-1-AP; 1:5,000; Proteintech Group, Inc.), followed by incubation with a HRP-conjugated Goat anti-Rabbit IgG (H+L) as the secondary antibody (cat. no. SA00001-2; 1:3,000; Proteintech Group, Inc.) for 2 h. After washing the membrane three times with 1X TBST, the protein bands were visualized using an ECL detection system. Amersham ImageQuant 800 (Cytiva) and the intensity of the band was quantified using ImageJ software (version 1.54k; National Institutes of Health).

To confirm the presence of POSTN on the surface of CAF sEVs, Human POSTN ELISA kits was purchased (cat. no. H2452c, Elabscience). To assess the content of BMP4 on the surface of sEV-derived CAF-S5/-S6 transfected with shNC and shPOSTN, Human BMP4 ELISA kits was purchased (cat. no. CSB-E17298h; Cusabio Technology, LLC). Serially diluted CAF-derived sEVs were added to each well (100 µl/well) and incubated for 90 min. Subsequently, biotinylated POSTN and BMP4 antibody was introduced and incubated at 37°C for 1 h, followed by sequential additions of horseradish peroxidase-conjugated streptavidin (100 µl, 37°C for 30 min) and 3,3′,5,5′-tetramethylbenzidine substrate (37°C for 15 min). The reaction was terminated with 50 µl of stop solution and absorbance at 450 nm was measured using a Thermo Fisher Scientific, Inc. microplate reader.

For POSTN blocking on the sEV surface, CAF sEVs were incubated overnight at 4°C with a POSTN-blocking antibody (10 µg/ml in 100 µl PBS; Abcam). Non-bound antibodies were removed by washing with 20 ml PBS and ultracentrifugation at 100,000 × g at 4°C for 70 min. All experiments were performed in at least triplicate.

Total RNA was extracted from macrophages treated with CAF sEV-derived shNC and shPOSTN for 48 h using TRIzol^® Reagent (Thermo Fisher Scientific, Inc.) per the manufacturer's protocol. RNA integrity was assessed by Agilent 5300 Bioanalyser and concentration was measured using a Nanodrop ND-2000. High-quality RNA samples were used for library preparation. Subsequently, RNA purification, reverse transcription, library construction and sequencing were conducted by Majorbio Biotechnology on an Illumina NovaSeq/HiSeq Xten (Illumina, Inc.) platform with NovaSeq reagents kits (Illumina, Inc.) according to standard protocol. Raw paired-end reads were processed with Fastp (version 1.0.1; http://github.com/OpenGene/fastp) for adapter trimming and quality control. HISAT2 was used to align the cleaned reads to the reference genome and transcript assembly was performed with StringTie (version 3.0.3; http://ccb.jhu.edu/software/stringtie). Gene expression was quantified in TPM and differential expression analysis was carried out to identify differentially expressed genes (DEGs) between comparative groups. RSEM was used to quantify gene abundances. Essentially, differential expression analysis was performed using the DEGseq (42). Functional-enrichment analysis including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) were performed to identify which DEGs were significantly enriched in GO terms and metabolic pathways at Bonferroni-corrected P-value ≤0.05 compared with the whole-transcriptome background. GO functional enrichment and KEGG pathway analysis were carried out by Goatools and KOBAS (43), respectively.

Single-cell sequencing was performed using RStudio software (version 4.1.1). Data distributions were visualized via histograms generated in R software. Categorical variables were compared using the Chi-squared test. For continuous variables, comparisons between two groups used the unpaired two-tailed Student's t-test or the Wilcoxon rank-sum test. Comparisons among three or more groups were performed using one-way ANOVA test followed by Tukey's post hoc test. Inter-variable relationships were examined using Spearman's rank correlation. Statistical analysis and quantification of western blotting bands were performed using GraphPad prism software (version 7.0; Dotmatics) and ImageJ (version 1.6.0; National Institutes of Health). Results are expressed as mean ± SEM from a minimum of three independent experiments. P<0.05 was considered to indicate a statistically significant difference.

An integrated analysis of the HNSCC single-cell profiling data was conducted using sequencing datasets (Fig. 1A), revealing five clusters that were annotated via Seurat clustering according to specific markers. The gene lists associated with cluster annotation markers are included in Table SI. Existing studies have revealed that POSTN is markedly elevated in tumor-derived fibroblasts compared with in normal tissues (37,44,45). The vlnplot demonstrated that POSTN was mainly expressed in the fibroblasts cluster (Fig. 1B). Subsequently, the fibroblasts were further re-clustered into nine clusters (Fig. 1C) and it was revealed that POSTN was expressed in subpopulations 1, 4 and 5, with its expression was predominantly enriched in subpopulation 1 (Fig. 1D). To determine whether POSTN was expressed in M1/M2 macrophages, the myeloid cluster was further re-clustered into 14 subtypes. Low POSTN expression was detected in five of these subpopulations and it was negligible in the others (Fig. S1A). Based on established biomarkers (Table SII), these subpopulations were categorized into M1-like and M2-like macrophages (Fig. S1B). Notably, low expression of POSTN was present in M1-like macrophages and it was barely detectably in M2-like macrophages (Fig. S1C). Based on UMAP clustering analysis, the fibroblast subpopulations were categorized into POSTN⁻ and POSTN⁺ fibroblasts (Fig. 1E). Heatmap analysis revealed that myofibroblastic CAF (myCAF)-related proteins (ACTA2, MYH11, TAGLN, PDGFA, MCAM and MYLK) and inflammatory CAF (iCAF)-related proteins (IL6, CXCL12, C3 and CFD) were predominantly expressed in POSTN⁻ fibroblasts. By contrast, ECM CAF (eCAF)-related proteins (FAP, FN1, SFRP2, THY1, COL1A1, FBLN1 and LUM) and antigen presenting CAF (apCAF)-related proteins (CD74, HLA-DRA, HLA-DRB5, HLA-DRB1 and HLA-DPA1) were predominantly expressed in POSTN⁺ fibroblasts (Fig. 1F). Vnplots further illustrated the expression levels of eCAF and apCAF-associated proteins in POSTN⁺ fibroblasts and POSTN⁻ fibroblasts (Fig. S1D). Spearman's correlation analysis revealed a significant correlation between the expression levels of POSTN and FAP (Fig. 1G). Moreover, patients with POSTN^high and FAP^high HNSCC exhibited a high poor response to ICB (Fig. 1H). The clinical parameters of 40 OSCC cases and 14 normal controls are listed in Table SIII. H&E and immunohistochemical staining identified higher expression levels of FAP and POSTN in the stroma of OSCC cases compared with those in the normal control samples (Fig. 1I). These findings indicated that POSTN⁺FAP⁺ fibroblasts are localized in the stromal region and associated with a poor response to ICB in patients with HNSCC.

Recent studies have provided evidence for intercellular communication between POSTN⁺FAP⁺ fibroblasts and macrophages (46–48). Additionally, the macrophage infiltration rate is a critical factor in responses to ICB. The ssGSEA algorithm (Fig. 2A) demonstrated that the expression levels of POSTN and FAP were markedly associated with the immune infiltration of macrophages in TCGA-HNSCC. The expression levels of POSTN and FAP exhibited strong positive correlations with markers of M1 (CD80 and CD86) and M2 (CD163 and CD206) macrophages, respectively. However, the correlations with the M2 macrophage markers (CD163, P<2.2×10⁻¹⁶; CD206, P<2.2×10⁻¹⁶) were markedly stronger (Figs. 2B, S2A and B). Immunofluorescence staining revealed the co-expression of POSTN and CD163 in the stromal area in HNSCC cases (Fig. 2C). To delineate the role of POSTN⁺ fibroblasts in the TME, we analyzed 28 immune-related signatures, two distinct immunity subtypes were identified through unsupervised hierarchical clustering: high and low-immunity subtypes (Fig. 2D). The high-immunity subtype was characterized by high stromal, immune, ESTIMATE and POSTN scores coupled with low tumor purity (Fig. 2E). Then, two molecular subtypes of HNSCC-TCGA were identified through unsupervised hierarchical clustering (Fig. S2C and D). Further analysis of the TME revealed that POSTN⁺ fibroblasts were characterized by high stromal, ESTIMATE, immune and POSTN scores, as well as low tumor purity (Fig. S2E). Interestingly, M0 macrophages, M2 macrophages and CD4⁺ T memory cells also highly infiltrated the POSTN⁺ fibroblasts, highlighting the strong positive correlation between the POSTN⁺ fibroblasts and M2 macrophages in HNSCC (Fig. 2F). Collectively, these findings strongly suggest that POSTN⁺ fibroblasts are associated with an immunosuppressive TME in HNSCC, potentially through their strong correlation with and spatial proximity to M2 macrophages.

To further confirm the expression levels of POSTN in CAFs and their sEVs, human primary OSCC-derived CAF-S5/-S6 and healthy gingival-derived NFs were successfully extracted and validated, as previously described (37). The sEVs were isolated from the respective CM of CAF-S5/-S6 and NFs using differential ultracentrifugation (Fig. 3A) and they exhibited characteristic exosomal morphology in TEM images (Fig. 3B). These sEVs tested positive for exosomal markers, including CD9, CD81 and ALIX and tested negative for Calnexin (Fig. 3C). The size distribution and particle concentration of sEVs were examined using a Nanoflow cytometer, revealing an mean size of ~100 nm and a concentration of ~1×10¹⁰ sEVs/1ml of CM (Fig. 3D). The expression levels of POSTN in CAF-S5/-S6 and their sEVs were markedly higher than those in NFs and their sEVs (Fig. 3E). Furthermore, density gradient fractionation analysis revealed the presence of POSTN on CAF-derived sEVs (Fig. 3F). Immunogold labeling confirmed the presence of POSTN on the surface of CAF-S5/-S6 sEVs (Fig. 3G). ELISA further demonstrated that the level of sEV-POSTN from these cells increased in an sEV dose-dependent manner (Fig. 3H), a signal that was markedly reduced by a POSTN-blocking antibody (Fig. 3H). Taken together, these findings identified POSTN as a surface component of CAF-S5/-S6 sEVs.

To determine whether sEV-POSTN can induce macrophage polarization, CAF-S5/-S6 stably transfected with shNC and shPOSTN were established and sEVs were isolated by differential ultracentrifugation. The expression of POSTN in CAF-S5/-S6 and their sEVs was markedly reduced by shPOSTN compared with that in the shNC group (Fig. 4A). The mean sizes of sEVs derived from shNC/shPOSTN-transfected CAF-S5/-S6 were ~95 and 80 nm, respectively and their particles concentration were ~1.9×10¹⁰ and 2.7×10¹⁰ sEV/1 ml CM, respectively (Fig. 4B). Human THP-1 cells were incubated with PMA (50 ng/ml) for 48 h to induce differentiation into M0 macrophages, which exhibited the corresponding adherent morphology (Fig. 4C) and highly expressed the macrophage marker CD68 (P<0.001; Fig. 4D). Flow cytometry revealed that macrophages exhibited no significant difference in their uptake of PKH26-labeled sEVs derived from shNC/shPOSTN-transfected CAF-S5/-S6 (Fig. 4E). RT-qPCR assay revealed that the M2 macrophage markers CD163 and CD206 were markedly downregulated, whereas the M1 macrophage markers CD80 and CD86 remained unchanged, after co-incubation with sEVs derived from shPOSTN-transfected CAF-S5/-S6 compared with those from sEVs derived from shNC-transfected CAF-S5/-S6 (P<0.01; Fig. 4F). Flow cytometry confirmed that the expression levels of M2 macrophage markers (CD163 and CD206) were markedly reduced in macrophages treated with sEVs derived from POSTN-knockdown CAF-S5/-S6 (Fig. 4G). Furthermore, the expression levels of M2 macrophage markers (CD163 and CD206) were markedly downregulated in macrophages treated with sEVs derived from shPOSTN-transfected CAF-S5/-S6 compared with those treated with sEVs derived from shNC-transfected CAF-S5/S6 (Fig. 4H). However, the expression of M1 macrophage markers (CD80 and CD86) showed no significant difference in macrophages treated with sEVs derived from shNC/shPOSTN-transfected CAF-S5/-S6 (Fig. 4H).

RT-qPCR and western blotting analysis of macrophages also showed that the expression levels of M2 macrophage markers (CD163 and CD206) were markedly downregulated in macrophages treated with CAF-S5/-S6 sEVs pre-incubated with a POSTN-blocking antibody, compared with those treated with CAF-S5/-S6 sEVs (Fig. S3A and B). Flow cytometry confirmed that CAF-S5/-S6 sEVs pre-incubated with a POSTN-blocking antibody markedly reduced the expression levels of M2 macrophage markers (CD163 and CD206) (Fig. S3C and D). These results demonstrated that sEVs secreted by POSTN-knockdown CAF-S5/-S6 may inhibit M2 macrophage polarization.

To further investigate the mechanism by which POSTN regulates THP-1-derived macrophages M2 polarization, RNA sequencing was performed on THP-1-derived macrophages stimulated with sEVs derived from CAF-S6 transfected with shPOSTN and shNC. PCA was employed to assess transcriptomics profiles, revealing a clear distinction between macrophages treated with shPOSTN sEVs and CTRL or shNC sEVs (Fig. 5A). A Venn diagram revealed 15 shared genes and 242 unique genes between macrophages treated with shPOSTN sEVs and shNC sEVs (Fig. 5B). Furthermore, a heatmap showed significant upregulation and downregulation of genes in macrophages stimulated with sEVs derived from CAF-S6 transfected with shPOSTN compared with those treated with sEVs derived from CAF-S6 transfected with shNC (Fig. 5C). A volcano plot identified 443 DEGs, including 262 upregulated and 181 downregulated genes, in macrophages treated with shPOSTN sEVs compared with those treated with shNC sEVs (Fig. 5D). Among the top DEGs, BMP4 was identified as a critical growth factor that fosters a pro-tumoral immune environment by polarizing macrophages toward an M2-like phenotype, thereby promoting tumor growth (49).

GSEA revealed that inflammatory response was upregulated in the macrophages treated with shPOSTN sEVs compared with in those treated with shNC sEVs (Fig. 5E). GO analysis revealed that 20 biological processes were markedly altered in macrophages treated with shPOSTN sEVs compared with shNC sEVs, with notable changes in ‘cellular anatomical entity’, ‘binding’ and ‘cellular processes’ (Fig. 5F). Additionally, KEGG pathway analysis revealed significant enrichment of pathways, such as the ‘cytokine-cytokine receptor interaction’ and ‘viral protein-cytokine receptor interaction’, which are closely associated with the inflammatory response during macrophage M2 polarization (50,51) (Fig. 5G). The levels of BMP4 in sEV-derived CAF-S5/-S6 transfected with shNC were higher than those in the shPOSTN group (Fig. S4), average BMP4 level was 2.6 pg/ml in the shNC group and 1.9 pg/ml in the shPOSTN group. Furthermore, RT-qPCR analysis showed markedly higher levels of BMP4 in macrophages treated with sEVs derived from shNC-transfected CAF-S5/-S6 compared with in those treated with sEVs derived from shPOSTN-transfected CAF-S5/-S6 (Fig. 5H). Moreover, the mRNA expression levels of pro-inflammatory factors (TNF-α and IL-6) were markedly downregulated in macrophages with 50 and 100 ng/ml of BMP4 (Fig. 5I). By contrast, the mRNA expression levels of anti-inflammatory factors (TGF-β and IL-10) were markedly upregulated in macrophages with 50 and 100 ng/ml of BMP4 (Fig. 5I). In addition, flow cytometry showed BMP4-induced upregulation of M2 macrophage markers CD163 and CD206 (Fig. 5J). Moreover, the levels of BMPR2 and phospho-Smad1/5/9 were increased in macrophages treated with 100 ng/ml BMP4 for 48 h (Fig. 5K). To assess the role of Smad signaling in BMP4-induced M2 polarization, the BMP type I receptor was inhibited with LDN193189. This treatment suppressed BMP4-triggered Smad1/5/9 phosphorylation and the expression of CD163/CD206 (Fig. 5L). These findings suggested that POSTN-deficient CAF sEVs may decrease BMP4 secretion, which otherwise activates the BMPR2/Smad signaling to drive macrophage M2 polarization, which could offer a therapeutic strategy for targeting TAMs.