Clinical samples

Thirty HNSCC tissue samples were collected, with complete case information, from the Pathology Department of the First Affiliated Hospital of Dalian Medical University (Dalian, China) between January 2017 and December 2020. None of the patients included in the study received preoperative radiotherapy or chemotherapy, and all provided written informed consent. The expression analysis of 520 patients with head and neck cancer (data sourced from The Cancer Genome Atlas, TCGA, www.cancer.gov/) was conducted using the online tool Tumor IMmune Estimation Resource (TIMER, https://cistrome.shinyapps.io/timer/). This study was approved by the Ethics Committee of the Dalian Medical University (2021002).

Cell culture

CAL27, SCC25 (tongue squamous carcinoma cell line), and human umbilical vein endothelial cells (HUVECs) were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA). The human myeloid leukemia mononuclear cell (THP-1) line was obtained from Cellcook (Guangzhou, China). CAL27 and SCC25 cells were often used as typical tumor cells to study of tumor-associated macrophages [19] and cultured in high-glucose Dulbecco’s modified Eagle medium (DMEM; Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS; Gibco). THP-1 cells were cultured in RPMI 1640 medium (Hyclone, Thermo Fisher Scientific) supplemented with 10% FBS. HUVECs were cultured in EC medium (ECM; ScienCell, Carlsbad, CA, USA) containing 5% FBS and 1% endothelial cell growth supplement (ScienCell). Pulmonary HUVECs (pHUVECs) were obtained from ATCC (Manassas, VA, USA). The culture conditions for the pHUVECs were the same as those for the HUVECs. pHUVECs from the 3rd passage to the 8th were used in microfluidic chip experiments. Normal human fibroblasts (NFs) were isolated from patients whose teeth were extracted at Dalian Medical University (Dalian, China, 2022003). Normal gingival tissues were agitated in 10% DMEM/F12 containing collagenase type I (1 mg/mL; Sigma-Aldrich, SCR103, Merck, Germany) for 10 h, and the dissociated tissues were shaken for 5 min. The supernatant was separated, and NFs were collected with DMEM/F12 medium (10% FBS) at 37 ℃ with 5% CO2 in a humidified incubator. All the cells were supplemented with 1% penicillin/streptomycin.

TAM polarization

THP-1 cells were differentiated into M0 macrophages after a 6 h treatment with 100 ng/mL phorbol 12-myristate 13-acetate (MCE, NJ, USA). M0 macrophages were induced using the supernatant of SCC25/CAL27 cells for 24 h and polarized into TAMs. TAM markers (CD163 and CD206) were detected using flow cytometry and western blotting, and mRNA expression of Arg1 and Fizz1 was analyzed by reverse transcription quantitative real-time PCR (RT-qPCR).

Flow cytometry

1 × 106 HUVECs were harvested and washed twice with PBS by centrifugation at 350–500×g for 5 min. After resuspending with 100 µL of PBS, the cells were incubated with the primary antibody including anti-CD206 (10 μL, Proteintech, APC-65155, Wuhan, China) and anti-CD163 (5 μL, Proteintech, CL488-65169) for 40 min at 4 °C in the dark. Then the cells were washed with PBS for 5 min and the supernatant was discarded. A flow cytometer (FACSVerse, BD Biosciences, USA) were used to detect the cells and the data was analyzed by the software Flowlo 10.8.

sEV collection, isolation, and identification

When the polarization of M0/TAMs was achieved, the culture medium was removed, and M0/TAMs were washed thrice with PBS. Complete medium comprising 10% FBS without sEV was added to M0/TAMs. After 48 h, culture supernatants of M0/TAMs were collected. We used the ultracentrifugation method to extract M0/TAM-sEV. Briefly, the supernatants were subjected to various centrifugation steps. First, the supernatant was centrifuged at 500×g for 20 min at room temperature to remove residual cells. Subsequently, centrifugation at 2500×g and 4 °C for 30 min was performed to eliminate cell debris. Then, centrifugation at 12,000×g for 45 min was performed to remove large vesicles. Finally, the supernatant was ultracentrifuged at 120,000×g for 90 min to enrich the sEV. The pellet was rinsed with 20 mL PBS and ultracentrifuged at 120,000×g and 4 °C for 90 min to purify the sEV. For the identification of EVs after purification, three methods were used. The diameter of the sEV was measured using high-sensitivity flow cytometry for nanoparticle tracking analysis (NTA; Nanofcm, Fujian, China). Meanwhile, the M0/TAM-sEV were deposited onto copper grids and negatively stained with 0.2 M phosphotungstic acid for observation under transmission electron microscopy (TEM, JEM-2000EX; JEOL, Tokyo, Japan). Western blot was used to detect the expression of sEV marker proteins, including positive (anti-TSG101, 1:1000, Abcam, ab125011, Cambridge, UK; anti-CD63, 1:1000, Abcam, ab134045; anti-CD9, 1:1000, Abcam, ab263019) and negative (anti-Calnexin; 1:1000, Abcam, ab133615) markers. The pellet containing sEV in 100 μL PBS was stored at −80 °C for further use.

sEV internalization and delivery assays

M0/TAM-sEV labeled with PKH67 (Thermo Fisher Scientific) were incubated with HUVECs for 6 h. To visualize the HUVECs, DAPI (Solarbio, C0065, Beijing, China) was used to stain the nucleus, and rhodamine-labeled phalloidin (Yeasen, 40734ES75, Shanghai, China) was used to label the cell cytoskeleton. The M0/TAM-sEV were detected using a fluorescence microscope (Olympus, Tokyo, Japan). Transwell chambers (Corning, New York, NY, USA) were used to study the intercellular transfer of miRNAs. TAMs were transfected with FAM mimic/NC and seeded in the upper chamber. HUVECs were seeded in the lower chamber, separated by a 0.4 μm porous membrane. The co-culture was maintained for 48 h. The nuclei of HUVECs were stained with DAPI and observed under a fluorescence microscope to detect the PKH67 (green) fluorescence signal.

miRNA mimic transfection

HUVECs (1 × 105) were seeded in a 6-well plate, as described previously [20], and Lipofectamine™ 3000 (Thermo Fisher, L3000001) was used to transfect the cells with 10 nM miRNA inhibitor, miRNA mimic, NC inhibitor, and NC mimic (GenePharma, B03001, B02001, B04003, B04002, Shanghai, China). After a 6 h transfection incubation, the culture medium was replaced with complete ECM for 24 h for subsequent experiments.

Tube formation assay

After 96-well plates (BD, NJ, USA) were coated with Matrigel, HUVECs (2 × 104 cells/well), they were treated with or without M0/TAM-sEV (10 μg/well) for 6 h. Tube-like structures in each well were observed under a light microscope as described previously [21]. To investigate the role of miRNAs in tube formation, HUVECs were transfected with miR-21 mimic or inhibitor for 24 h. The harvested HUVECs were resuspended and seeded in 96-well plates coated with Matrigel for 6 h. Tube formation was observed under a light microscope.

RNA extraction and RT-qPCR

Total RNA was extracted using TRIzol reagent (Vazyme, R411-01, Nanjing, China), and the total RNA of sEV was extracted using the MolPure® Serum/Plasma miRNA Kit (Yeasen, 19332ES50) as described previously [22]. Briefly, total RNA (1 μg) was reverse transcribed using HiScript II Reverse Transcriptase (Vazyme, R201-01). RT-qPCR was performed using Talent qPCR PreMix (SYBR Green, TIANGEN, FP209, Beijing, China) on a Thermal Cycler Dice Real-Time System (Bio-Rad CFX96 Touch, Hercules, CA, USA). The primer sequences are listed in Supplemental Table 1.

Western blot

Total protein was extracted by RIPA lysis buffer mixed with PMSF (Solarbio). A BCA protein assay kit was used to determine the protein concentration (Beyotime, P0009, Shanghai, China). Proteins (20 µg) were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (8% for CD206, CD163, CD68, HIF1α and LATS1, 10% for TSG101, CD63, Calnexin, pYAP1, YAP1 and GAPDH, and 12% for CD9 and VHL) and transferred to a PVDF membrane (Millipore Corporation, MA, USA). After membranes were blocked with 5% nonfat milk (Solarbio) at room temperature, the following primary antibodies were used to incubate membranes overnight at 4 °C (anti-CD68, 1:500, 28058-1-AP, Proteintech; anti-CD206, 1:600, 60143-1-Ig, Proteintech; anti-CD163, 1:100, sc-20066, SCBT, CA, USA; anti-VHL40, 1:200, sc-135657, SCBT; anti-LATS1, 1:5000, 17049-1-AP, Proteintech; anti-YAP1, 1:4000, 13584-1-AP, Proteintech; anti-pYAP1, 1:4000, T55743, Abmart, Shanghai, China; anti-HIF-1α, 1:500, ab308433, Abcam; anti-GAPDH, 1:8000, ab8245, Abcam). The membranes were then incubated with horseradish peroxidase-conjugated secondary antibodies (1:8000, ab205719/ab205718, Abcam) for 1 h, and the protein bands were detected using an ECL detection system (1705061, Bio-Rad).

Immunohistochemistry and immunofluorescence

Paraffin-embedded tissue sections (4-μm-thick) were treated with xylene and rehydrated in gradient alcohol. After treatment with 3% hydrogen peroxide for 20 min and antigen retrieval using sodium citrate buffer (G1202-250ML, Servicebio, Wuhan, China), the sections were blocked with 10% nonimmune goat serum (Gibco, Thermo Fisher) for 1 h. Next, the sections were incubated with the following primary antibodies overnight at 4 °C (anti-CD31, 1:50, ab28364, Abcam; anti-CD206, 1:500, 60143-1-Ig, Proteintech; anti-CD163, 1:50, sc-20066, SCBT). Antibody binding was detected using a peroxidase secondary antibody (1:500, Abcam), followed by counterstaining with hematoxylin. Finally, the integrated optical density (IOD) values of the brown staining were measured using Image-Pro Plus 6.0. The IOD/area ratio was calculated to obtain semi-quantitative values of protein expression.

For immunofluorescence staining, 0.5% Triton-X was used to permeabilize the sections. Next, we blocked the tissues with 10% nonimmune goat serum before incubating them overnight at 4 °C with the following primary antibodies: anti-CD31 (1:20, ab28364, Abcam), anti-CD206 (1:500, 60143-1-Ig, Proteintech), and anti-CD163 (1:200, sc-20066, SCBT). Sections were then incubated with Dylight-488 (1:200; A23220, Abbkine, CA, USA) or Dlight-549 (1:200, A23320, Abbkine) at room temperature for 1 h. Finally, we stained the sections with DAPI and observed them under a fluorescence microscope.

Microfluidic chip assay

The microfluidic chip comprises five channels (C1-C5) with a width of 800 µm. Each channel is separated by pillars and has an inlet and outlet. The interconnected segments between the five channels were all 5.5 mm in length. Channels C1/C2 and C4/C5 were divided by pentagonal microcolumns with a width of 200 µm and a spacing of 90 µm. Channels C2/C3 and C3/C4 were divided by hexagonal microcolumns with a width of 200 µm and a spacing of 90 µm.

The silicon wafer was coated with a SU8-3035 negative photoresist (17,020,067 Microchem Corp., NH, USA) to a thickness of 120 µm. Desired design patterns were created using photolithography. Polydimethylsiloxane (PDMS; Dow Corning Corp, MI, USA) was poured into the prepared photomask and subjected to 30 min of vacuuming to remove air. After the mold was cured at 80 ℃ for 45 min, the PDMS was peeled off from the wafer and trimmed to the desired size. Outlet 2, Outlet 4, Inlet 2, and Inlet 4 were punched using a 4 mm hole punch, whereas Outlet 1, Outlet 3, Outlet 5, Inlet 1, Inlet 3, and Inlet 5 were punched using a 2 mm hole punch. To bond the PDMS membranes to petri dishes (Corning, NY, USA), the membranes were treated with plasma for 1 min. The assembly was placed in a drying oven and baked at 80 ℃ for 12 h, followed by a 12 h UV sterilization.

Fibrinogen (Sigma Aldrich) was dissolved in a 0.9% sodium chloride solution to a concentration of 10 mg/mL. D-PBS (Hyclone) was used to dilute the fibrinogen solution to 4 mg/mL, and then antipain (0.45 TIU/mL, EI3, Sigma-Aldrich) was added. The fibrinogen solution was then filtered through a sterile filter. Thrombin (1 U/mL, 10602400001, Sigma-Aldrich) was mixed with a sterile-filtered fibrinogen solution to induce fibrinogen clot formation. After the fibrinogen mixture was introduced into channels C3 and C5 and mixed with digested NF, the mixture was added to channel C1, where the cell density was adjusted to 2 × 106 cells/mL. The microfluidic chip was kept at 37 ℃ for 15 min to allow fibrinogen gelling. After ECM was added to channels C2 and C4 for 24 h, pHUVECs were introduced into channel C4 at a density of 2 × 106 cells/mL. M0/TAM-EVs were added to channel C2, and different ECM volumes were added to channels C2 and C4 to create a 1 mm H2O liquid-level difference, forming an interstitial flow. To maintain the stability of the interstitial flow, the medium was changed every 12 h and observed under a microscope every 24 h.

Single-cell sequencing data source and processing

The scRNA-seq dataset GSE103322 for HNSCC [23], which includes 5902 cells from 18 patients, was downloaded from the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/). The Seurat package in R software was used to create the Seurat object [24]. Genes detected in fewer than three cells and cells expressing fewer than 200 unique genes were excluded from downstream analyses. Subsequently, we calculated the sequencing depth (within 20,000), number of expressed gene types (300–10,000), proportion of red blood cell genes (<3%), and proportion of mitochondrial genes (<10%) for each cell. High-quality cell data were normalized to find 2000 high-variable genes for subsequent analyses. Based on 2000 highly variable genes, principal component analysis (PCA) was applied to reduce the dimensionality of the data, and clustering was conducted to select 30 principal components (PCs), which were visualized using T-distributed neighbor embedding (t-SNE) [25]. The FindNeighbors function from the Seurat package was then employed to identify neighbors for each cell, and the FindClusters function (within a resolution of one) was applied to the cluster cell types based on the 30 PCs. The t-SNE algorithm was used to visualize the cell clustering results. Additionally, we combined the “SingleR” package with existing marker genes for cell annotation [26], resulting in the identification of 10 cell types. Bubble plots were used to display the expression of marker genes in different cell types. The expression of four exosomal marker genes across different subtypes are derived from normalized datasets and visualized using a violin plot constructed with ggplot2 [27].

Cell–cell interaction analysis

We used CellPhone DB to investigate the interactions between various types of cells [28]. First, we created a Linux subsystem using Ubuntu 22.04.2 and set up a Python 3.6 environment for CellPhone DB. The annotation information and expression matrices of the ten cell types were extracted from the Seurat objects. Cell communication analysis was performed using CellPhone DB 4.0 with 1000 iterations to obtain ligand–receptor pairs involved in the interactions between the 10 cell types. Finally, a heatmap was generated based on the number of ligand–receptor pairs that mediate cell–cell interactions.

Enrichment analysis and cell pathway activity analysis

We used R software (version 4.1.3) and the FindAllMarkers function of the Seurat package, employing the Wilcoxon test as the statistical method to identify signature genes for each cell type. Using the false discovery rate (FDR), only genes with FDR-adjusted p values less than 0.05 were considered significantly dysregulated. In ECs, 1492 signature genes were identified, while 1635 signature genes were identified in macrophages. Enrichment analysis was performed using the clusterProfiler package in R software [29]. Specifically, we conducted Gene Ontology Biological Process (GO.BP, htpps://geneontoloy.org/) and Kyoto Encyclopedia of Genes and Genomes (KEGG, www.genome.jp/kegg/) enrichment analyses separately for 1635 signature genes in macrophages and 1492 signature genes in ECs. Furthermore, to evaluate the pathway activity of VEGF, HIF-1α, and Hippo pathways in all cells, the SCpur package was used to generate an enrichment heatmap [30], visualizing the pathway activity based on the expression levels of pathway-associated genes.

Matrigel-plug angiogenesis assay

We used the Matrigel (Corning, 354248) injection model to detect angiogenesis in vivo. First, 50 μL of PBS, 50 μL M0-sEV (100 μg) and 50 μL TAM-sEV (100 μg) with or without pre-treatment of antagomir (1 OD; B05001, Genepharma, Shanghai, China) was respectively mixed with 450 μL of Matrigel. The mixture was injected subcutaneously into the ventral region of female BALB/c nude mice (4 weeks old, n = 4 per group). After 2 weeks, the mice were sacrificed, and Matrigel plugs were fixed with paraformaldehyde and embedded in paraffin wax. Serial sections were prepared and stained with hematoxylin and eosin (HE) staining and immunohistochemistry. The microvessel area in each sample was quantified using Image-Pro Plus 6.0. All the animal experiments were approved by the Institutional Animal Care and Use Committee of Dalian Medical University (No. AEE23070).

Tumor xenograft assay

To establish the subcutaneous tumor xenograft model, 4-week-old female BALB/c were randomly assigned into four groups (PBS, M0-sEV, TAM-sEV, TAM-sEV/antagomir), and each mouse was injected subcutaneously in the flanks with 1 × 106 SCC25 cells mixed with 100 μL Matrigel. Then, according to the established four subgroups, we started injecting PBS (50 μL), M0-sEV (50 μL, 100 μg), and TAM-sEV (50 μL, 100 μg) with or without antagomir (1 OD, B05001, Genepharma) every 3 days (six times in total). The volume of the tumor (length × width2/2) and weight of the mice were monitored every 2 days throughout the study period. After 1 month, the mice were euthanized, and their tumors were collected. Tumor volumes were calculated using the formula V (mm3) = A × B2/2 (A is the largest diameter and B is the perpendicular diameter). Finally, the harvested tumors were fixed in 4% paraformaldehyde at 4 °C and embedded in paraffin for subsequent histochemical staining.

Statistical analyses

GraphPad Prism (version 7.0) was used for the statistical analysis. The correlation between tumor MVD and TAM markers in tumor tissues was assessed using Pearson’s correlation analysis. To calculate p values, we used Student’s t test and one-way analysis of variance. Statistical significance was set at p < 0.05. We expressed the data as the mean ± SEM of at least three independent experiments.