Tenascin-C Orchestrates an Immune-Suppressive Tumor Microenvironment in Oral Squamous Cell Carcinoma

Head and neck squamous cell carcinomas (HNSCC) are heterogeneous malignancies originating from the mucosal surface of the upper aerodigestive tract. The 5-year survival rate worldwide is around 50% due to disease recurrence and metastasis (1). At least two genetic subclasses of HNSCC can be distinguished, where human papillomavirus (HPV)–negative tumors, representing approximately 65% of HNSCC are caused by chronic exposure to carcinogens including tobacco and alcohol (2). The first-line treatment of HNSCC is surgery followed by radiotherapy and chemotherapy and recently immune checkpoint therapy where long-lasting effects are seen only in a fraction of patients (3, 4).

HNSCC is an immune-suppressive disease where the physiologic microenvironment changes into a protumoral state accompanied by major changes in the extracellular matrix (ECM; refs. 5–8). Tenascin-C (TNC) is one such ECM molecule that impacts the progression of several tumor types through regulation of multiple cancer hallmarks (9–11). In a nontumor context, TNC can serve as a danger-associated molecular pattern (DAMP) molecule, and trigger more severe inflammation through integrin α9β1 and TLR4 (12, 13). Although TNC is mostly absent in normal tissues, TNC is expressed in reticular fibers of lymphoid tissues where it regulates leukocyte maturation (14, 15). In cancer tissue (16–18), TNC is organized in tumor matrix tracks (TMT) that share certain features with reticular fibers and may play a role in immune cell functions in cancer tissue (10, 11, 14). Although TNC is one of the major ECM proteins upregulated in the matrix of HNSCC-associated fibroblasts (19), the precise roles of TNC in this disease have not yet been investigated.

To better understand how immune cells interact with the neoplastic stroma in HNSCC, here, we used the carcinogen 4-Nitroquinoline 1-oxide (4NQO)–driven murine model with abundant or absent TNC. 4NQO applied in the drinking water causes DNA adduct formation thus mimicking the effects of tobacco carcinogens and induces malignant lesions mainly in the tongue and esophagus (20, 21).

We identified TNC as a molecule involved in the immune-suppressive TME in OSCC. Comparison of tumors in wild-type (WT) and TNC knockout (TNCKO) mice allowed us to demonstrate a role for TNC in OSCC progression and lymph node invasion suggesting a mechanism by which the TNC-rich tumor matrix shaped an immune-suppressive, protumoral microenvironment. These results provide relevant information for human HNSCC diagnosis and therapy.

Surgically removed tongue tumors, embedded in paraffin blocks, were retrieved from the archives of the Pathology Department of the Centre Antoine Lacassagne. Informed consent was obtained for all subjects. Patient characteristics are summarized in Supplementary Table S1. Hematoxylin and eosin staining and IHC methods were performed on serial 4-μm deparaffinized tissue microarray (TMA) sections. CD45 staining was performed on a BenchMark Ulter Automated Slide Staining system (Ventana Medical Systems, Inc., Roche Group) using monoclonal anti-CD45 (LCA; clone 2B11+PD7/26) or anti-podoplanin (D2-40) according to instructions of the manufacturer (Cell Marque). For TNC staining, intrinsic peroxidase was blocked by incubating sections with 3% hydrogen peroxide for 15 minutes. Antigen retrieval was performed in EDTA buffer pH 9.0, in a decloaking chamber (Dako, catalog number S2367). Sections were blocked in 4% goat serum for 1 hour, then incubated for 1 hour with mouse monoclonal anti-TNC (clone BC24, Sigma-Aldrich 1/1,000). After rinsing with PBS, sections were incubated with biotinylated secondary antibody (30 minutes) and biotinylated goat anti-mouse IgG (30 minutes) followed by avidin-biotin (Vector Laboratories, VECTASTAIN ABC Kit, catalog number PK-4000). Slides were incubated with 3,3′-Diaminobenzidine developing solution (Vector Lab, DAB, catalog number SK-4100) and hematoxylin before embedding into ProLong Gold antifade reagent (Invitrogen, catalog number P36930). For fluorescence staining, after permeabilization (PBS, 0.1% Triton) cells/tissue were incubated with the primary antibodies (Supplementary Table S2) overnight. Bound antibodies were detected with the appropriate Alexa-labeled secondary antibodies (Supplementary Table S2) prior to nuclear staining with DAPI (Sigma, catalog number D9542) and embedding into ProLong Gold antifade reagent (Invitrogen, catalog number P36930). Fluorescently stained sections were digitalized (40×) using a PerkinElmer Vectra Polaris imaging system and Phenochart software (Akoya Biosciences).

Stained slides were scanned on the Hamamatsu NanoZoomer 2.0-HT Digital slide scanner (40× mode). Scans were viewed and images acquired using the NDP.view2 software. For quantification, we developed a script (based on ImageJ) optimized to be used with interactive surfaces (https://figshare.com/articles/Custom_toolbars_and_mini_applications_with_Action_Bar/3397603/3). The program and the manual are freely available at https://mycore.core-cloud.net/index.php/s/0K61LqHBrnNKShX. Randomly chosen images of noninvasive tumor areas (three per tumor, 5× magnification) were projected on an interactive digital whiteboard. A pathologist determined the regions of interest (ROI) corresponding to tumor cell nest or stroma. These ROIs where extracted after color deconvolution and thresholding to quantify CD45 staining. We then determined the ratio of area containing CD45 (holes were removed as deduced from the hematoxylin image) per image and per ROI type.

Public patient data (GSE27020) were analyzed by the Kaplan–Meier plotter tool (ProggeneV2 prognostic database) as described (22). The cohort was separated by the median of corresponding gene expression as “High” and “Low,” respectively. Gene expression was correlated to relapse-free survival (RFS). The correlation matrix analysis by Corrplot package (R software; https///github.com/taiyun/corrplot, Taiyn and Simko) was performed on gene expression data derived from RNA chromatin immunoprecipitation (chIP) analysis from HNSCC tumors of 68 patients (23). The graphical representation was generated using the R package corrplot. The multiple testing corrections were performed using the pound method (24).

4-NQO (Sigma-Aldrich, catalog number N8141) was administered to 8-week-old WT and TNCKO (KO; ref. 25) mice, which had been bred in house with C57BL/6J mice (Charles River) for more than 10 generations, in the drinking water at a final concentration of 100 μg/mL for 16 weeks (stock 5 mg/mL in propylene glycol). Subsequently, mice were fed with regular water for 4 weeks before sacrifice, where tongue, submandibular lymph nodes, and spleen were collected and prepared for FACS analysis, cryosectioning, mRNA, or protein extraction as described below. During tissue sampling, the general organ appearance and the number of tumors per 4 mice were determined. To assess the roles of CCL21/CCR7 signaling, mice were also subjected to the regular 4NQO protocol as described above. The last 2 weeks before sacrificing the mice, mice were given three intraperitoneal injections of IgG control antibody (200 μg, R&D Systems, catalog number MAB006) or CCR7 antibody (200 μg, R&D systems, MAB3477) as described previously (26). The injections were spaced at least 4 days apart, and the last injection took place 4 days before the sacrifice. All mice were housed and handled according to the guidelines of INSERM and the ethical committee of Alsace, France (Cremeas; Directive 2010/63/EU on the protections of animals used for scientific purposes).

RNA from WT and TNCKO tongue tumors (3 samples per group) was isolated using the RNeasy Mini Kit (Qiagen, catalog number 74104) and RNA integrity was determined with an Agilent Bioanalyzer 2100 (Pico Kit, Agilent Technologies). Total RNA-Sequencing libraries were prepared with SMARTer Stranded Total RNA-Seq Kit v2 - Pico Input Mammalian (TaKaRa, catalog number 634411) according to the manufacturer's protocol. Libraries were pooled and sequenced (paired-end 2*75 bp) on a NextSeq500 using the NextSeq 500/550 High Output Kit v2 according to the manufacturer's instructions (Illumina, catalog number 20024907). Quality control of every sample was assessed with the NGS Core Tools FastQC and sequence reads were mapped using STAR and Bowtie2 (27, 28). The total mapped reads were finally available in BAM (Binary Alignment Map) format for raw read counts extraction. Read counts were found by the HTseq-count tool of the Python package HTSeq (29) with default parameters to generate an abundance matrix. Differential analyses were performed using the DESEQ2 (30) package of the Bioconductor framework. Upregulated and downregulated genes were selected on the basis of the P_adj (<0.10) and the fold change (>±0.8; Supplementary Table S3). Deregulated gene expression analysis was performed by using the PANTHER version 11 (31) and REACTOME software (32).

Tongue tumor pieces from WT and TNCKO mice were resuspended in Laemmli buffer (10 mmol/L Tris pH 6.8, 1 mmol/L EDTA, 5% ß-mercaptoethanol, 5% SDS, 10 glycerol, 1/100 antiproteases). Proteins were extracted for 1 hour upon sonication (four times for 5 minutes). Protein concentration was determined using the RC-DC protein assay (Bio-Rad, catalog number 5000121) following the manufacturer's instructions. Forty micrograms of protein lysate for each sample were heated at 95°C for 5 minutes and stacked in an in-house prepared 5% polyacrylamide SDS-PAGE stacking gel. Gel bands were reduced and alkylated prior to overnight digestion (the ratio of enzyme/protein = 1/50) at 37°C using modified porcine trypsin (Promega, catalog number V5113). The generated peptides were extracted with 60% acetonitrile (ACN) in 0.1% formic acid (FA) followed by a second extraction with 100% ACN. Peptides were resuspended in 100 μL of water and 0.1% formic acid.

Nano-LC/MS-MS analysis was performed on a nanoAcquity UPLC device (Waters) coupled to a Q-Exactive HF-X mass spectrometer (Thermo Fisher Scientific) equipped with a Nanospray Flex ion source. Peptide separation was performed on an ACQUITY UPLC Peptide BEH C18 Column (250 mm × 75 μm with 1.7 μm diameter particles) and an ACQUITY UPLC M-Class Symmetry C18 Trap Column (20 mm × 180 μm with 5-μm diameter particles; Waters). The solvent system consisted of 0.1% FA in water (solvent A) and 0.1% FA in ACN (solvent B). Samples (800 ng) were loaded into the enrichment column over 3 minutes at 5 μL/minute with 99% of solvent A and 1% of solvent B. Peptides were eluted at 400 nL/minute with the following gradient of solvent B: from 1 to 8% over 2 minutes, from 8 to 35% over 77 minutes, and from 35 to 90% over 1 minute. Samples were injected in a randomized order. The MS capillary voltage was set to 2 kV at 250°C. The system was operated in a data-dependent acquisition mode with automatic switching between MS (mass range 375–1,500 m/z with R = 120,000 at 200 m/z, automatic gain control fixed at 3 × 10⁶ ions, and a maximum injection time set at 60 milliseconds), and MS/MS (mass range 200–2,000 m/z with R = 15,000 at 200 m/z, automatic gain control fixed at 1 × 10⁵, and the maximal injection time set to 60 milliseconds) modes. The twenty most abundant peptides were selected on each MS spectrum for further isolation and higher energy collision dissociation (normalized collision energy set to 27), excluding unassigned, monocharged, and superior to seven times charged ions. The dynamic exclusion time was set to 40 seconds, and “Peptide match selection” parameter of the software.

The raw data obtained for each condition were processed with MaxQuant (version 1.6.0.16; ref. 33). Peaks were assigned with the Andromeda search engine with full trypsin (Trypsin/P) specificity against an in-house generated protein sequence database containing all mouse protein entries extracted from UniProtKB-SwissProt (17 007 sequences, taxonomy identifier: 10 090, release 2019–04–09). Carbamidomethylation of cysteines was set as fixed modification, whereas oxidation of methionines and protein N-terminal acetylation were defined as variable modifications. Minimal peptide length was set to seven amino acids and up to two missed cleavage sites were allowed for trypsin digestion. Peptide mass tolerance was set to 20 ppm for the first search and 5 ppm for the main search. The maximum false discovery rate was 1% at PSM, peptide, and protein levels with the use of a target/decoy strategy.

Label-free quantification was done on unique peptides (LFQ min. ratio count of 2) with the match between runs option activated (match window of 2 minutes and alignment window of 10 minutes). Unmodified peptides and those with carbamidomethylated cysteines were used for protein quantification.

After removal of contaminants, reverse entries, proteins only identified with modified peptides and protein groups identified with less than two unique peptides, differential analyses on normalized LFQ intensities were performed using Prostar (version 1.16.6; ref. 34). A Limma t test was performed for the statistical analysis test calibrated with the Pounds method (35). Dysregulated proteins were selected based on the P_adj value (Supplementary Table S4) and further analysis were performed using the PANTHER version 11 (31) and REACTOME software (32).

Proteins were extracted from 5 WT, 5 KO, 4 control antibody and 4 CCR7 antibody tongue tumors in lysis buffer [Triton 1× and protease inhibitors (Roche, catalog number 11697498001) diluted in 1× PBS] following the manufacturer's instructions and protein concentration of tumor samples was determined by optical density measurement (NanoDrop 2000). The expression of immunomodulatory molecules in tumor samples was measured using the Mouse XL Cytokine Array Kit (Biotechne, catalog number ARY028) according to the manufacturer's instruction. After membrane blocking, equal protein amounts from 4 or 5 pooled tumors per group were applied to the membrane overnight. Two membranes were used for each group to have an experimental duplicate. The revelation of each membrane was done by using the Cemi Reagent Mix provided in the kit and a Chemidoc Imager XRS (Bio-Rad). Quantification was done by measuring pixel density with the ImageJ software. The background signal was subtracted with the negative control spots and the positive control spots were used to normalize values of each molecule to compare membranes between each other (Supplementary Table S5).

The OCT-embedded tissue sections (8-μm thick) were incubated in ddH₂O before staining with hematoxylin (Surgipath, catalog number 3801560) for 30 seconds and eosin (Sigma, catalog number HT110132) for 10 seconds, spaced by 1 minute of ddH₂O washes. After the last wash, tissue sections were dehydrated 5 minutes in increasing percentage baths of ethanol (from 70% to 100%) and toluene and then covered with the Eukitt solution (Sigma, catalog number 03989).

For immunofluorescence (IF) staining, unfixed frozen sections of 8 μm or cells fixed with 4% paraformaldehyde were incubated for 1 hour at room temperature with blocking serum (5% normal goat or donkey serum in PBS; Jackson ImmunoResearch, catalog numbers 005–000-121 and 017-000-121, respectively) and overnight directly with the primary antibodies (Supplementary Table S5). Bound antibodies were visualized with goat, rabbit, guinea pig, hamster, or rat secondary antibodies conjugated with Alexa 488, Cy3, or Cy5. DAPI (Sigma) was used to visualize nuclei. After embedding in FluorSave Reagent (Calbiochem, catalog number 345789), sections were examined using a Zeiss Axio Imager Z2 microscope. Pictures were taken with an AxioCam MRm (Zeiss) camera and Axiovision software. Control sections were processed as mentioned above with omission of the primary antibodies. The image acquisition setting (microscope, magnification, light intensity, exposure time) was kept constant per experiment and in between experimental conditions. For quantification of immune cells and positive staining area, the ImageJ software was used. CCL21, CCR7, and gp38 scoring is based on the criteria described in Supplementary Table S6. At least two sections of 5 different tumors/mice were quantified per condition. The number of immune cells was reported in correlation to the total number of DAPI-positive cells.

Frozen and cryopreserved tissue samples were thawed and washed for 15 minutes with distilled water followed by a fixation in 2% (v/v) formaldehyde and 0.25% (v/v) glutaraldehyde in 100 mmol/L cacodylate buffer, pH 7.4, at 4°C overnight. Afterwards, tissue samples were rinsed in PBS, dehydrated in ethanol up to 70% (each step 30 minutes), and embedded in LR White embedding medium (London Resin Company, catalog number 14381-UC) using UV light for polymerization (Leica EM AFS). Ultrathin sections were cut with an ultramicrotome (Leica Ultracut UCT), collected on copper grids (Athene Grids, G202) and negatively stained with 2% uranyl acetate (Serva, catalog number 77970) for 15 minutes. Electron micrographs were taken at 60 kV with a Phillips EM-410 electron microscope using imaging plates (Ditabis).

Raman images were acquired using a WITec alpha300R Raman microscope (WITec). In the upright set-up a 532 nm laser was focused through a 60× dipping objective (NA 1.0) to excite Raman scattering on tissues sections. Tissue areas within the TMTs of 150 × 200 μm were scanned using a pixel size of 1 μm and an acquisition time of 0.08 seconds per pixel. For each tumor, two to three Raman images were generated. The spectral images were then further processed and analyzed using WITec Project Five software (WITec). After cosmic ray removal and background correction, spectra of each pixel were area normalized. The tumor stroma was identified based on a specific spectral pattern, predominantly resembling collagen fibers. For each image all pixels resembling this matrix pattern were averaged. Spectral patterns of stromal matrix in WT and TNCKO were compared using univariate statistics and Principal Component Analysis (PCA). To identify CCL21 in the tumor stroma, a reference spectrum of purified mouse CCL21 (457-6C; R&D Systems) was acquired. Peaks at 757,1030, 1210, 1319, and 1,615 cm⁻¹ are specific for CCL21. This spectrum was used to decompose the ECM spectra in a True Component Analysis (WITec). Here the CCL21 reference was first employed on a CCL21-positive lymph node to identify CCL21 in a physiologic condition. The CCL21 spectrum was extracted from the lymph node Raman scans and employed on the Raman data from tumor stroma. For quantification relative intensities of CCL21 in WT and TNCKO stroma was normalized to the collagenous stroma area.

Recombinant his-tagged human TNC was purified as described previously (36, 37) and used for incubation with cells. Murine strep-tagged TNC was used in negative EM microscopy and treatment of cells. For cloning murine TNC, a PCEP4 expression vector (Invitrogen, catalog number V04450) with TNC (NP_035737.2, aa: 174-2019) from Mus musculus was obtained from R. Chiquet-Ehrismann (FMI, Basel, Switzerland). The coding sequence was modified with a BM40 signal peptide and a N terminal double strep II tag and was confirmed by sequencing (Supplementary Table S7). To generate stable cell lines, HEK293 EBNA cells were transfected with the expression vector using Fugene HD (Promega, catalog number E2311). After 48 hours of transfection, the medium was replaced with 0.5 μg/mL containing DMEM/F12 medium with 10% FCS and the cells were grown to confluency. The protein was then purified from the supernatant by using the Streptactin matrix (IBA, Lifesciences, catalog number 2-1021-001) following the manufacturer's guidelines and was then dialyzed three times against PBS as described previously (37).

Surface plasmon resonance binding experiments were performed on a Biacore 2000 instrument (Biacore Inc.) at 25°C. Recombinant human TNC (36) was immobilized at high surface density (around 7000 resonance units) on an activated CM5 chip (Biacore Inc., catalog number 29149604) using a standard amine-coupling procedure according to the manufacturer's instructions. Soluble molecules were added at a concentration of 10 μg/mL in 10 mmol/L sodium acetate, pH 5.0, and at a flow rate of 5 μL/minute for 20 minutes before addition of 1 mol/L ethanolamine. CCL21 (0.5, 0.87, and 2 μg in 200 μL) was added to the chip at pH 6.0 [10 mmol/L MES, pH 6.0, 150 mm sodium chloride, 0.005% (v/v) surfactant P20], or at pH 7.4 [10 mmol/L HEPES, 150 mmol/L sodium chloride, 0.005% (v/v) surfactant P20], at a flow rate of 10 μL/minute. A blank CM5 chip was used for background correction. 10 mmol/L glycine, pH 2.0, at 100 μL/minute for 1 minute was used to regenerate the chip surface between two binding experiments. A steady-state condition was used to determine the affinity of CCL21 for TNC. The Dissociation constant (K_d) was determined using the 1:1 Langmuir association model as described by the manufacturers (https://www.biacore.com/lifesciences/help/kinetic_model_1_1_binding/index.html).

The interaction of TNC with CCL21 was visualized by negative staining and transmission electron microscopy as described previously (38). Briefly, TNC samples (20 nmol/L) were incubated with a 3 molar excess of CCL21 (457-6C-025 R&D Systems) for 1 hour at 37°C in Tris-buffered saline (TBS), pH 7.4. For visualization in the electron microscope, CC21 was conjugated with 5 nm colloidal gold (39). For inhibition experiments, TNC samples were preincubated with a 10 molar excess of heparin for 1 hour at 37°C. Specimens were examined in a Philips/FEI CM 100 TWIN transmission electron microscope operated at 60 kV accelerating voltage. Images were recorded with a side-mounted Olympus Veleta camera with a resolution of 2048 × 2048 pixels (2k × 2K) and the ITEM acquisitions software. Binding of CCL21 particles to TNC was determined by counting the number of gold particles along the length of the TNC monomer. Number of molecules from 500 randomly picked distinct TNC molecules were determined. As positive control, TGFβ1 was used as it binds in the 5th FNIII repeat of TNC (40). As negative controls EGF (shown not to bind to TNC; ref. 40) and BSA were used, respectively.

All cultured cells were checked for the absence of Mycoplasma (once every 2 months, PlasmoTest, Invivogen catalog number rep-pt1). Lymphatic endothelial cells (LEC) and dendritic cell (DC)–like DC2.4 were purchased from ATCC (HDMVECn, PCS-110-010, 2018) and Merck (SCC142, 2018), respectively, and reauthenticated by determination of LYVE-1 expression (LEC), CD31 (HDMVEC) and CCR7, CD80 and CD86 (DC2.4) by flow cytometry. DC2.4 cells were cultured in DMEM-glucose (Dutscher) complemented with 10% of FBS (Dutscher), 100 U/mL penicillin, 100 μg/mL streptomycin (penicillin/streptomycin, Dutscher), 40 U/mL gentamicin (Thermo Fisher Scientific) and 1× HEPES. LECs were cultured in ECGM with penicillin/streptomycin, gentamicin, and a supplemental growth factor cocktail according to Promocell (catalog number C22110). Fibroblastic reticular cells (FRC) were isolated from the lymph nodes (popliteal, inguinal, brachial, axillary, mandibular, and cervical) of a naïve WT mouse (10 weeks old) as described previously (41). FRCs were cultured in DMEM-glucose complemented with 10% FBS, 1% penicillin/streptomycin (Sigma catalog number P4333) and gentamicin (Dutscher catalog number P06-03100). The OSCC13 cell line was established from a primary 4NQO-induced tongue tumor of a WT mouse. Cells were mechanically dissociated and cultured in DMEM-F12 with 4.5 g/L glucose, 10% FBS, 1% penicillin/streptomycin (Sigma catalog number P4333), gentamycin, and 0.4 μg/mL hydrocortisone (Sigma, catalog number H4001). Cells were cultured for 20 passages and then subcutaneously grafted in the neck of a WT mouse. After two times of grafting in WT C57Bl6-J mice, cells were cultured for 50 passages before use. Silencing of TNC in OSCC13 cells was done by short hairpin (sh)–mediated gene expression knockdown. Briefly, lentiviral particles shRNA vectors (Sigma, catalog number SHCLNV-NM_011607 MISSION shRNA Lentiviral Transduction Particles) encoding specific shRNAs for the knockdown of TNC were used (shTNC:CCGGGCATCAA-CACAACCAGTCTAACTCGAGTTAGACTGGTTGTGTTGATGCTTTTTG). Lentiviral particles encoding a nontargeting shRNA vector were used as control (SHC202V, Sigma). Transduced cells were selected with the previously described DMEM-F12 culture medium supplemented with 10 μg/mL puromycin (Thermo Fisher Scientific, catalog number A1113802) and the selection pressure was kept in all in vitro experiments.

All cell lines were maintained at 37°C in a humidified atmosphere of 5% CO₂. The culture medium was refreshed every 2 to 3 days and passaged into a new dish with trypsin-EDTA (PanBiotech) upon reaching confluency. Cells were starved with DMEM-F12 medium containing 1% FBS overnight before treatment. Cells were treated for 24 hours with purified human or mouse TNC (10 μg/mL) diluted with DMEM medium complemented with 1% FBS, penicillin/streptomycin, and gentamicin. Upon TNC stimulation, the CM was collected, filtered at 0.22 μm, and stored at −80°C for future use. Cells were detached mechanically, concentrated by centrifugation, and lysed in TRIzol reagent (Invitrogen, catalog number 12044977) before storage at −80°C. Before TNC incubation, LECs were pretreated with inhibitors for TGFβRI (GW788388, 10 μmol/L, 45 minutes, Selleckchem, catalog number S2750), TLR4 (Cli95, 1 μg/mL, 6 hours, InvivoGen, catalog number tlrl-cli95), receptor tyrosine kinases (SU6668, 30 μmol/L, 60 minutes, Tocris Bioscience, catalog number 3335), integrin α9β1 [blocking antibody α9Ab, 4 μg/mL, 6 hours, provided by Shigeyuki Kon (42) and α9β1/α4β1, BOP, 1 μmol/L, 45 minutes, Tocris Bioscience, catalog number 6047].

DC2.4 cells were starved with medium containing 1% FBS overnight and pretreated the day after with TLR4 (Cli95, 1 μg/mL, 6 hours) diluted in 1% FBS complemented DMEM. Cells were incubated for 24 hours with 1% FBS complemented DMEM containing lysophosphatidic acid (LPS; 1 μg/mL) or soluble TNC (10 μg/mL). Upon LPS or TNC incubation, cells were detached, lysed in TRIzol (Invitrogen) following the manufacturer's instructions, or stained with anti-CD80-FITC, anti-CD11c-PE, anti-MHCII-APC EF780, and anti-CD86-PE Cy7 from eBiosciences for FACS analysis.

Boyden chamber migration assays on DC2.4 were performed in 5-μm pore-sized polycarbonate membrane transwells (Corning Costar Co, catalog number 3421). The lower surface of the transwells were precoated with Col I (BD Biosciences, catalog number 354236), horse purified fibronectin (FN; ref. 36) and mouse purified TNC at a final concentration of 1 μg/cm², respectively. The bottom chambers of the transwells were filled with DMEM containing mouse or human CCL21 (100 ng/mL, 200 ng/mL, or 400 ng/mL, R&D Systems, catalog number 457-6C-025 and 366-6C-025). To assess the migration of DC2.4 toward the secretome of the LECs, CM from LECs [treated or not with TNC (10 μg/mL) for 24 hours] was placed in the bottom chamber. To block the chemotaxis of DC2.4 cells toward CCL21, cells were incubated 6 hours with CCR7 neutralizing antibodies (10 μg/mL, R&D Systems) diluted in 1% FBS complemented DMEM. DC2.4 (5 × 10⁵) suspended in 150 μL of 1% FBS-complemented DMEM were placed into the top chamber of the transwell system. Cells were incubated for 5 hours (CCL21 in the bottom chamber) or 8 hours (CM in the bottom chamber) at 37°C in 5% CO₂. The number of migrated cells in the bottom chamber was assessed by flow cytometry after the staining of DC2.4 with anti-CD11-PE (eBiosciences).

The DC2.4 chemoretention assays were done with the same set up as described in the migration protocol. After 5 hours (CCL21 conditions) or 8 hours (CM conditions) of migration, the DC2.4 cells attached to the bottom surface of the transwells were fixed in 4% PFA and stained with DAPI. Pictures were taken and analyzed by the ImageJ software. Floating cells were analyzed by flow cytometry (Supplementary Table S8).

Frozen tongue tumors and cultured cells were dissolved in the TRIzol reagent (Invitrogen, catalog number 12044977) for total RNA extraction. RNA quality was confirmed by optical density measurement (OD 260 nm). cDNAs were synthesized from 1000 μg of total RNA using random primers and Moloney murine leukemia virus reverse transcriptase (MultiScribe, Applied Biosystems, catalog number 10117254). The cDNA was used for qRT-PCR in a Mx3005P Real-Time PCR System (Thermo Fisher Scientific). Reactions were carried out in duplicate for all conditions using a SYBR Green Master Mix (Thermo Fisher Scientific, catalog number 4344463) or Fast TaqMan mix (Thermo Fisher Scientific, catalog number 4444557) and expression of mouse or human Gapdh mRNA (Life Technologies, catalog number 433764T) was used as endogenous control in the 2^−ΔΔC_t calculation. Primer sequences used for qPCR determination are listed in Supplementary Table S9.

Tissues or cell lysates were prepared in lysis buffer (50 mmol/L TRIS-HCl pH 7.6, 150 mmol/L NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS) supplemented with a phosphatase inhibitor cocktail (Santa Cruz Biotechnology, catalog number sc-45045) and protease inhibitors (Roche, catalog number 05892970001). The protein concentration of tissue samples and CM was determined by Bradford assay (Bio-Rad, catalog number 5000001) following manufacturer's instructions. Thirty micrograms of protein lysate was loaded in precasted 4% to 20% gradient gels (Bio-Rad, catalog number 4561096), together with Laemmli buffer (Bio-Rad, catalog number 1610737) and separated by SDS-PAGE. The separated proteins were then transferred onto nitrocellulose membranes (Bio-Rad, catalog number 1620113) using the TransBlot Turbo Transfer system (Bio-Rad). Nitrocellulose membranes were then blocked with 5% Blocking-Grade blocker (Bio-Rad, catalog number 1706404) in 0.1% Tween-20 PBS and incubated with the primary antibody (overnight at 4°C) and secondary antibodies (1 hour at room temperature) in 1.5% Blocking-Grade Blocker in 0.1% Tween-20 PBS. Antibodies used are listed in Supplementary Table S2. Protein bands were detected with the Amersham ECL Western Blotting Detection Reagent (GE Healthcare, catalog number RPN2106) or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, catalog number 34095). CCL21 and IFNγ expressions were determined by using the 6-Ckine ELISA kit (Thermo Fisher Scientific, catalog number EMCCL21A) and IFNγ ELISA Kit (Thermo Fisher Scientific, catalog number BMS606), respectively, according to the manufacturer's instructions. The absorbance of each sample and standard was measured with a plate reader (MultiSkan EX, Thermo Fisher Scientific).

Tongue tumors and submandibular lymph nodes were cut into small pieces and inflated with digestion solution containing 1 mg/mL Collagenase D (Roche, catalog number 50-100-3282) and 0.2 mg/mL DNase I (Roche, catalog number 4716728001), 2% inactivated FBS in RPMI, at 37°C for 2 hours. Upon completion of digestion, 92 μL of 54 mmol/L EDTA was added and the samples were vortexed at maximal speed for 30 seconds. The resulting cell suspensions were passed through a 70-μm and 40-μm cell strainer and treated with flow cytometry buffer (PBS, 2% FBS, 1 mmol/L EDTA). After cells were counted, 2 × 10⁶ cells per lymph node/spleen sample or 1 × 10⁶ cells for tumor sample, were stained with Dead Viability dye-efluor 450 (Thermo Fisher Scientific, catalog number 65-0863-18) according the manufacturer's instructions. Cells were then incubated in blocking solution containing 2% FcBlock CD16/CD32 (Thermo Fisher Scientific, catalog number 14-0161-85) in flow cytometry buffer, for 15 minutes at 4°C and then stained 30 minutes at 4°C with a standard panel of immunophenotyping antibodies; solution 1: anti-CD45-FITC, anti-CD11c-PE, anti-B220-APC, anti-MHCII-APC EF780, and anti-CCR7-Percp Cy5; solution 2: anti-CD45-FITC, anti-CD3e-PE, anti-C8a-APC, anti-CD4-APC EF780, anti-Foxp3-PE Cy7, anti-CCR7-Percp Cy5, and anti-CD25-AF700; solution 3: anti-CD45-FITC, anti-Gp38-PE, anti-CD31-APC, anti-F4/80-APC EF780, anti-CCR7-Percp Cy5, and anti-CD11b-AF700 (Supplementary Table S2). Data were acquired with a Beckman Coulter Gallios flow cytometer. Adjustments and data analysis were performed by using the FlowJo software. See Supplementary Table S2 for information on the antibodies and Supplementary Tables S10–S16 for information on the gating strategy.

For all data, Gaussian distribution was tested by the d'Agostino–Pearson normality test. When data followed a Gaussian distribution, statistical differences were analyzed by unpaired t test (with Welch correction in case of unequal variance) or ANOVA one-way with Tukey post test. Otherwise, the Mann–Whitney test or a nonparametric ANOVA followed by Dunn post test were used to verify significance of the observed differences. All statistical analyses were performed using the GraphPad Prism software. Mean ± SEM. P values < 0.05 were considered as statistically significant (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

In contrast to nontumoral human tongue tissue, with weak TNC expression, TNC expression was upregulated in the tongue tumor stroma, in TMT (Fig. 1A; refs. 17, 18). Investigating abundance of TIL (CD45⁺ leukocytes) revealed more TILs in the tumor stroma compared with the tumor nests (Fig. 1B and C; Supplementary Fig. S1B and S1C).

4NQO induced OSCC in the mucosal epithelium of mice (Fig. 1D; Supplementary Fig. S1D and S1E), which recapitulated human OSCC (Supplementary Fig. S1F). Whereas TNC expression was very low in tongue epithelium of nontreated mice, its expression became upregulated in the stroma of the OSCC (Fig. 1E).

To address whether TNC had an impact on tumorigenesis in this model we determined tumor formation in WT and TNCKO mice. TNCKO mice presented a reduced number of tumors per mouse in comparison with WT mice (Fig. 1F). Without TNC, tongue tumors were also significantly smaller than in WT mice (Fig. 1G). TNCKO mice did not develop invasive carcinomas, in contrast with WT mice (Fig. 1H). WT mice developed lymph node metastasis (p63 staining), which was absent in TNCKO mice (Fig. 1I and J).

Malignant tumor cells retained their epithelial (E-cadherin⁺, CK8/18⁺, and vimentin-negative) traits (Supplementary Fig. S2A). Tumor epithelial cell nests (p63⁺) were separated by stromal niches (αSMA⁺ cells), similar to human OSCC (Supplementary Fig. S2A; ref. 43). Tumors were highly vascularized (CD31 and LYVE-1) similar to human OSCC (44). No difference in vascularization, nor survival or proliferation, was seen between tumor genotypes unlike other tumors (Supplementary Fig. S2B–S2G; refs. 5, 16, 45).

Several genes (176) were differentially expressed, 120 up- and 56 downregulated in WT compared with the TNCKO tumors (Supplementary Table S2). Expression of matrisome genes (46) was largely reduced in TNCKO tumors (Supplementary Fig. S2H and S2K; Supplementary Table S3). The analysis of the matrix by Raman microspectroscopy revealed a significant difference in the PC1 score that was below (TNCKO tumors) and above zero (WT tumors; Fig. 2A and B; Supplementary Fig. S2I and S2J). TNC expression was not detected in the TNCKO tumors (Fig. 2C). Collagen networks differed in WT and TNCKO mice, with more parallel oriented and compactly organized collagen fibrils, in WT tumors (Supplementary Fig. S2L). TNC was expressed in TMT together with laminin (LM), fibronectin (FN), Coll IV, and Coll XII (Fig. 2C; Supplementary Fig. S2M and S2N).

There was no difference in the abundance of CD45⁺ leukocytes between the two genotypes (Fig. 2D). However, there were more TILs in the tumor cell nests of TNCKO tumors (Fig. 2E and F). Whereas no difference in the abundance of macrophages (F4/80⁺, CD11b⁺), B cells (B220⁺), CD4 (CD3⁺, CD4⁺), or CD8 T lymphocytes (CD3⁺, CD8⁺) between tumor genotypes was seen, we observed more dendritic cells (DC; MHCII⁺/CD11c⁺) in TNCKO tumors (Fig. 2G; Supplementary Fig. S2O). CD11c⁺ cells resided predominantly in the stroma of WT tumors, whereas more CD11c⁺ cells infiltrated tumor cell nests in TNCKO tumors (Fig. 2H and I). In contrast to DCs, T regulatory cells (Tregs; CD4⁺/Foxp3⁺) were more abundant inside the tumor cell nests of WT tumors (Fig. 2J–L). Macrophages (F4/80, CD206), CD4⁺ and CD8⁺ T lymphocytes, and B cells (B220) were present inside the tumor cell nests and the stroma with no apparent differences between WT and TNCKO tumors (Supplementary Fig. S2R). Thus, TNC appears to orchestrate the intratumoral distribution of some leukocytes, in particular, CD11c⁺ cells and Tregs.

As CD11c⁺ cells are antigen-presenting cells (APC) that play a role in priming T cells in the lymph nodes (47), we investigated the immune cell infiltrate of the local lymph nodes by flow cytometry. We observed more CD45⁺ leukocytes in lymph nodes from TNCKO tumor mice (Supplementary Fig. S2P). Whereas the abundance of macrophages, B cells, and CD8 T cells was similar between genotypes, CD11c⁺ DCs and CD4 T cells were more frequent in lymph nodes from the TNCKO mice (Supplementary Fig. S2Q). Less CD11c⁺ cells in the lymph nodes and a reduced lymph node-to-tumor ratio of these cells in WT (251-fold) compared with TNCKO mice (878-fold) indicated that TNC may have impaired the migration of CD11c⁺ cells toward the draining lymph nodes (Fig. 2D; Supplementary Fig. S3P). A higher proportion of lingual-derived DCs (high expression of MHCII and intermediate expression of CD11c; ref. 48), was observed in TNCKO lymph nodes compared with WT tumor mice indicating that DC homing to lymph nodes was reduced in WT conditions (Supplementary Fig. S2S and S2T).

We observed increased CCL21 (+74%) and CCL19 (+17%) expression in WT tumors (Fig. 3A; Supplementary Table S4). We confirmed higher Ccl21 mRNA and CCL21 protein expression in WT tumors (Fig. 3B and C; Supplementary Table S6). We also determined CCL21 expression in local lymph nodes and observed lower expression in lymph nodes compared with the tumors in WT mice, which could impact DC attraction to the lymph nodes (Fig. 3C). We observed no obvious difference in CCL21 and CD11c⁺ cell abundance and localization within the lymph nodes of WT and TNCKO tumor mice (Supplementary Fig. S3A).

We used Raman microspectroscopy across the whole tumor. On the basis of specific signals of the purified CCL21 protein, we detected a similar spectrum for CCL21 in Raman images of lymph nodes and in WT and TNCKO tumors. Despite a strong background due to collagen-rich matrix, CCL21-specific peaks were identified in lymph nodes (a known source of CCL21) and stroma of WT and TNCKO tumors, whereas they were absent from the tumor cell nests and lung tissue (Fig. 3D and E; Supplementary Fig. S3B). CCL21 was significantly lower in TNCKO compared with WT tumors (Fig. 3F–H). LECs, typically expressing LYVE-1 and CCL21, expressed CCL21 in the OSCC, which was much less pronounced in TNCKO tumors (Fig. 3I). Reduced CCL21 expression was not due to less LECs in TNCKO tumors, as LECs were similarly abundant in TNCKO as in WT tumors (Supplementary Fig. S2B–S2F). Staining for CCL19, the second ligand for CCR7 revealed similar staining intensity and stromal localization in the tumors and no difference in lymph nodes of WT and TNCKO mice (Supplementary Fig. S3C and S3D).

We examined whether TNC induced CCL21 in LECs and FRCs, which reside in tumors (49) and naturally express CCL21 (50). We used human dermal LEC–expressing LYVE-1, gp38, and integrin α9β1 (Supplementary Fig. S3E) and isolated FRC (typically expressing ERTR7 and gp38; ref. 41) from lymph nodes of a naïve WT mouse (Supplementary Fig. S3F). Upon exposure to TNC, there was no difference in FRCs and OSCC13 (isolated from a 4NQO-induced carcinoma, typically expressing p63 and CK8/18; ref. 51; Fig. 3J; Supplementary Fig. S3F and S3G). However, Ccl21 mRNA and CCL21 protein expression largely increased in LECs upon treatment with TNC (Fig. 3K; Supplementary Fig. S3H).

Inhibitors for TGFβRI (GW788388), TLR4 (Cli95), and receptor tyrosine kinases (SU6668) did not alter CCL21 expression upon TNC treatment, but an antagonist for integrins α4β1/α9β1 (BOP) and an integrin α9β1 blocking antibody reduced Ccl21 mRNA and CCL21 protein expression compared with those without induction by TNC (Fig. 3K; Supplementary Fig. S3H–S3J). Thus, TNC induced CCL21 in LECs via integrin α9β1.

Because TNC binds several soluble factors (40), it was crucial to determine whether TNC binds to CCL21. CCL21 bound to several sites within the TNC molecule, whereas uncoated gold particles, or other gold-labeled molecules, not binding TNC (BSA and EGF; ref. 40), did not interact (Fig. 4A and B; Supplementary Fig. S4A–S4C). A major binding site for CCL21 was within the fibronectin type III repeats (FNIII), presumably in the fifth repeat, as CCL21 bound at the same site (fifth FNIII repeat) where TGFβ1 was documented to bind TNC (ref. 40; Fig. 4B; Supplementary Fig. S4D). Heparin blocked binding of CCL21 to the FNIII repeats (Supplementary Fig. S4E). Also, binding of CCL21 to TNC was higher at pH 6 than at pH 7 (Fig. 4C). The TNC/CCL21-binding strength (K_d of 5.8 × 10⁻⁸ mol/L) was lower than CCL21 binding to CCR7 but in the same range (8.4 × 10⁻⁸ mol/L; ref. 52; Fig. 4C).

Using migration assays, we determined whether TNC-bound CCL21 could restrain DC migration (Supplementary Fig. S4F). CCL21 attracted DCs in a concentration-dependent manner, with fewer cells migrating toward TNC compared with FN or Col I (Supplementary Fig. S4G). To determine whether DC2.4 were potentially immobilized on the TNC substratum, we measured cell retention by counting the cells tethered on the surface of the lower side of the insert (Fig. 4D). More DCs were immobilized on TNC (compared with FN or Col I), which occurred in a CCL21 dose-dependent manner and was reduced with a CCR7-blocking antibody (Fig. 4E). CM from TNC-treated LEC caused DC2.4 retention on TNC compared with CM from control LECs and this was abolished by blocking CCR7 (Fig. 4F; Supplementary Fig. S4H).

FRC are a cellular component of reticular fibers of lymphoid tissues producing ECM and soluble factors (41). Also, cells with FRC properties (gp38⁺, ERTR7⁺, LYVE⁻) populate tumors (49). We wanted to know whether TNC impacted the abundance and spatial distribution of FRC. Using gp38 as marker for FRC (with CD31-negative selection) and ERTR7 staining, there were more FRC in WT than TNCKO tumors (Fig. 5A and B; Supplementary Fig. S5A and S5B; Supplementary Table S6).

The cellular crosstalk in lymphatic tissue is regulated by CCR7 signaling (53). There was higher Ccr7 expression and more CCR7⁺ cells, in particular, CCR7⁺ macrophages (CD11b⁺/F4/80⁺), DCs (CD11c⁺/MHCII⁺), and CD8⁺ T cells in WT compared with TNCKO tumors (Fig. 5C–F; Supplementary Fig. S5C–S5E). CCR7⁺CD11c⁺ cells were less prominent in the local lymph nodes of WT tumor mice, again suggesting a potential role of TNC in impairing migration of these cells from the tumor site to the draining lymph nodes yet not within the lymph nodes (Fig. 5G; Supplementary Fig. S5F).

Next, we investigated whether TNC influenced the expression of CCR7. We saw higher CCR7 expression in DC2.4 upon treatment with LPS (positive control) and TNC (Fig. 5H; Supplementary Fig. S5G). As TNC can signal through TLR4 (13), we used Cli95 to inhibit TLR4 (54) and observed that Cli95 abolished induction of CCR7 in DC2.4 (Fig. 5H). Next, we asked whether TNC-induced TLR4 signaling also affected expression of the DC maturation markers CD80 and CD86. Whereas LPS increased expression of both molecules (that was blocked by Cli95), TNC did not affect their expression at the cell surface (Supplementary Fig. S5H and I). As we saw higher Cd80 and Cd86 in WT tumors we considered an indirect effect by the tumor cells (Supplementary Fig. S5K). Therefore, we treated DC2.4 with CM from OSCC13 shC cells, expressing TNC (and shTNC cells with undetectable TNC) and observed higher expression of Cd80, Cd86 (and Ccr7), supporting a paracrine mechanism of TNC action (Fig. 5I; Supplementary Fig. S5J). CD80 and CD86 can be induced by IL6 and TNFα (55, 56) and we observed higher expression of both molecules in WT tumors (Supplementary Fig. S5L).

TNC also robustly increased expression of a group of genes involved in antigen processing and presentation, for example, 15 MhcII genes (H2), β2 microglobulin (B2M), transporter associated with antigen processing 1 (Tap1), and cathepsin S (Ctss), that were higher in WT than TNCKO tumors (Supplementary Fig. S5K).

Tregs (CD4⁺/CD25⁺/Foxp3⁺) and CCR7⁺ Tregs were more frequent in WT than TNCKO tumors (Figs. 2J and 5J). As Tregs express anti-inflammatory cytokines, we observed higher expression of the IL10 pathway (e.g., IL10, IL1ra, IL1a/b) in WT tumors (Supplementary Fig. S5M). We observed a positive correlation between Ccl21 expression and Foxp3 and Il10 expression, respectively, thus, TNC may impact Treg abundance and function through CCL21 (Supplementary Fig. S5N). In addition to Tgfb1, TNC upregulated molecules involved in Treg chemotaxis (e.g., Ccl3, Ccl2, Rantes, Cxcl9, Cxcl10, P-selectin, and Ccl22; Fig. 5K; Supplementary Fig. S5L and S5O). High expression of these genes together with a low number of CCR7⁺ Tregs (6%) could explain that TNC increased Treg abundance, in particular, within the tumor cell nests (Fig. 2K and L).

We wondered whether TNC impacted CTL abundance and activity, as DCs and Tregs can regulate CTL responses (57). Immune-suppressive CD8⁺ Tregs and nonprimed CCR7⁺/CD8⁺ T cells were more abundant in WT tumors (Fig. 5L; Supplementary Fig. S5E). This result suggested a potential impact of TNC on the education of CD8⁺ T cells in the lymph nodes and their impaired activity in tumors. In support of this idea, we saw significantly less Ifnγ, Granzyme b, and Perforin expression in lymph nodes and in tumors from WT mice (Fig. 5M; Supplementary Fig. S5P and S5Q). Also, a majority of “positive T-cell activation”–related genes (24 of 32) were downregulated in WT tumors (Supplementary Fig. S5R). Immune checkpoint inhibitor genes [Pdcd1 (encoding PD1), Cd274 (encoding PDL-1), and Ctla4] and prostaglandin E2—related genes (Ptges2, Ptgs2, and Ptger1) were elevated in WT compared with TNCKO tumors (Supplementary Fig. S5L). Together, these data suggested an immune-suppressive TME in WT tumors.

To investigate whether enhanced CCR7 signaling by TNC was linked to immune suppression and tumor growth, we used a CCR7-blocking antibody. Carcinogen-exposed WT mice were treated with this (and a control) antibody. As shown in Fig. 6A, we observed less tumors. Investigating the numbers of leukocytes and immune subtypes, we did not see any difference between the treated groups (Supplementary Fig. S6A and S6B). Whereas the number of CCR7⁺ leukocytes was not different in local lymph nodes and spleen, the number of CCR7⁺ DCs (CD11c⁺/MHCII⁺), macrophages (CD11b⁺/F4/80), Tregs (CD25⁺/Foxp3⁺), and CD8⁺ Tregs (CD3⁺/CD8⁺/Foxp3⁺) was reduced upon anti-CCR7 treatment (Fig. 6B–E; Supplementary Fig. S6C–S6G). These results precluded a systemic effect of the anti-CCR7 treatment such as a general depletion of CCR7⁺ cells (or other leukocytes). After anti-CCR7 treatment, there were more CD45⁺ and CD11c⁺ cells in the tumor cell nests (Fig. 6F–I), similar to the TNCKO phenotype (Fig. 2G–L).

Whereas anti-CCR7 treatment did not alter CCL21 expression, the expression of many anti-inflammatory molecules was reduced, which was consistent with a lower abundance of Tregs (Fig. 6D and J; Supplementary Fig. S6F, S6H, and S6I), again similar to the TNCKO phenotype (Figs. 2J and 6K). Addressing a potential impact of CCR7 blockade on the abundance of FRC, we observed less FRC (Fig. 6K and L), once more mimicking the TNCKO phenotype (Fig. 5A and B). This was reinforced by a reduced expression of several immune suppression–related genes upon anti-CCR7 treatment, including Mrc1 (encoding CD206) and the immune checkpoint inhibitors Pdcd1, Cd274, and Ctla4 (Supplementary Fig. S6J). The CCR7 antibody treatment also affected ECM-related gene expression in the tumors, notably downregulation of Tnc itself (Supplementary Fig. S6K).

Next we asked whether APC function and priming of CTL potentially was also enhanced. Indeed, we saw higher IFNγ, Granzyme b, and Perforin expression in the local lymph nodes and less nonprimed CD8⁺ T cells (CCR7⁺/CD3⁺/CD8⁺) inside the tumors. Higher expression of genes positively related to T-cell activation upon anti-CCR7 treatment was seen (Fig. 6E; Supplementary Fig. S6L–S6O). Consistently, we saw less cancer cells in the local lymph nodes of anti-CCR7–treated mice (Supplementary Fig. S6P).

To address whether immune suppression through TNC/CCL21/CCR7 is potentially relevant in human OSCC, we investigated human tumors for TNC and the LEC marker podoplanin. As the murine model mimics the early phases of the human disease, we focused on noninvasive OSCC tissue areas. Similar to the murine tumors, LECs were embedded in TNC-rich stroma (Fig. 7A). We costained the tumor tissue for CCL21 and TNC and observed CCL21 expressed in TNC-rich stroma by cells with flat nuclei–forming tubes, likely representing LECs (Fig. 7B).

By investigating publicly available gene expression data (GSE27020), we determined expression of TNC and immune-suppressive markers. High TNC expression (above the median; as well as TGFβi) correlated with shorter time of survival until relapse-free survival (RFS), yet not with overall survival (OS) or metastasis-free survival (MFS; Supplementary Fig. S7A and S7B). Whereas high expression of CCR7, CCL21, Foxp3, IL10, CD206, CTLA4, and PD-1 alone did not correlate with shorter RFS, OS, or MFS, combined high expression of all makers plus TNC (HR = 2.78) or TNC combined with CCR7, CCL21, and IL10 (HR = 2.02) correlated with shorter RFS, thus supporting a potential role of TNC enforcing an immune-suppressive TME in human HNSCC that favors tumor relapse (Fig. 7C; Supplementary Fig. S7C–S7K). This possibility is supported by the study of RNA Affymetrix chip data from 68 patients with HNSCC (23), which shows a positive correlation between the expression of TNC and genes that define the immune-suppressive TME (Fig. 7D).

Altogether, our results showed that TNC promoted a protumorigenic TME with lymphoid properties by impacting FRC, CD11c⁺ cells, Tregs, and CTLs involving integrin α9β1 and TRL4, as well as several chemokines and cytokines, which phenocopies human HNSCC (Fig. 7E and F).

Fewer and smaller tongue tumors arose in absence of TNC in the 4NQO-treated mice and no invasive lesions nor lymph node invasion appeared, indicating that TNC promoted tumor progression, similar to other models (58).

TNC is expressed in TMT, also in OSCC (ref. 16 and this study). Despite similarities to reticular fibers suggesting a potential role in tumor immunity, the roles of TMT were obscure (14, 15). Here, TNC impacted the expression of collagens and several matrisomal molecules indicating that TNC may act as a master orchestrator of TMT.

TNC targets several immune subtypes, such as CTL in models of glioblastoma and prostate cancer, macrophages in breast cancer, and CD11c⁺ cells and Tregs in OSCC as demonstrated here (58–60). Profound differences were observed between WT and TNCKO mice with respect to the immune cell infiltrate and expression of immune-suppressive molecules in tumors and local lymph nodes. The presence of TNC led to less numerous CD11c⁺/MHCII⁺ cells in the tumor nests and enhanced their retention in the stroma. Thus, in a WT tumor, CD11c⁺ cells may be hampered in priming CTL due to poor migration of antigen-bearing DCs to draining lymph nodes, as seen elsewhere (48, 61). We observed less migratory DCs in draining lymph nodes of WT tumor mice, less nonprimed CCR7⁺ CD8⁺ T cells, and more poorly activated CTL in the tumors and lymph nodes of WT compared with TNCKO mice. TNC enforced infiltration of Tregs into the tumor cell nests presumably through elevated expression of Treg-attracting and maturation-promoting factors.

A role of CCL21 signaling in generating a lymphoid immuno-tolerogenic TME has previously been noticed; how this occurs remained unknown, with no link to matrix nor TNC provided (49). In our model, the natural source of CCL21 was LEC and FRC, and not the tumor cells (49). We identified CCL21/CCR7 signaling as a major target of TNC. Through induction of CCL21 in LEC (via α9β1 integrin) and by increasing the number of FRC, a natural source of CCL21, TNC enforced a protumoral TME and inverted the CCL21 gradient between lymph nodes and the tumor. This may have contributed to poor homing of CD11c⁺ cells and poor activation of CTL in the lymph nodes. Inhibition of CCR7 abolished the immune-suppressive properties of the TME and subsequently reduced tumor number, tumor progression, and lymph node metastasis.

Our observations supported a dual function of TNC in tumor immunity, in which its ancient role as DAMP and as a component of reticular fibers may be exploited by tumors (10, 13, 17). CCR7 blockade phenocopied features of the TNCKO supporting a causal link between TNC and CCR7.

In human OSCC, CCL21 induction by TNC in LEC may be relevant as high expression of TNC in conjunction with CCR7, CCL21, and other immune-suppressive markers correlated with shorter RFS. Our results could improve HNSCC diagnosis and therapy such as using Raman microspectroscopy for detection of stromal CCL21. Approximately 80% of patients with combined low expression of TNC and the immune-suppressive markers survived longer than 5 years and may represent a group that would benefit from a less harsh treatment.

DCs were released from the TNC/CCL21 substratum upon CCR7 inhibition suggesting a potential role of CCR7 as coreceptor of β2 integrins expressed on DCs (62). Thus, targeting β2 integrins (63) could be relevant in releasing CD11c⁺ cells from the matrix. Also targeting CCR7 may be useful because of its profound effect on abolishing the immune-suppressive properties of the TME, but not altering general immunity. CCR7 is a target in lymphomas and several metastatic cancers, but not yet in HNSCC (64). Several CCR7-targeting approaches have been developed (65–68) that could be tested in HNSCC. We have shown that targeting CCR7 appears to be safe and efficient.

TNC regulated the crosstalk of immune cells with CCL21, the positioning of TILs, especially CD11c⁺ cells and Tregs, and, subsequently, reduced adaptive immunity, thereby facilitating escape from immunosurveillance. Blockade of CCL21/CCR7 signaling relieved the protumoral immune-suppressive properties of the TME, normalized features of the tumor bed and reduced tumorigenesis and metastasis thus, providing novel targeting opportunities.

E. Van Obberghen-Schilling reports grants from Institut National du Cancer (INCa), Ligue National Contre le Cancer, and Foundation ARC during the conduct of the study. G. Orend reports grants from Institut National du Cancer (INCa) project FITMANET, Institut National du Cancer (INCa)/Ligue contre le Cancer project ECMpact, and Ligue Régionale contre le Cancer Grand Est (CCIR Est) and other from INSERM and University Strasbourg during the conduct of the study, as well as a pending patent. No potential conflicts of interest were disclosed by the other authors.

C. Spenlé: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. T. Loustau: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. D. Murdamoothoo: Investigation. W. Erne: Investigation. S. Beghelli-de la Forest Divonne: Investigation. R. Veber: Investigation. L. Petti: Investigation. P. Bourdely: Investigation. M. Mörgelin: Investigation. E.-M. Brauchle: Investigation. G. Cremel: Investigation. V. Randrianarisoa: Investigation. A. Camara: Investigation. S. Rekima: Investigation. S. Schaub: Software, investigation. K. Nouhen: Investigation. T. Imhof: Resources, investigation. U. Hansen: Investigation. N. Paul: Data curation, software. R. Carapito: Supervision. N. Pythoud: Investigation. A. Hirschler: Investigation. C. Carapito: Supervision, validation. H. Dumortier: Supervision, validation. C.G. Mueller: supervision, validation. M. Koch: Formal analysis, supervision, funding acquisition. K. Schenke-Layland: Supervision, funding acquisition, validation. S. Kon: Resources. A. Sudaka: Resources, validation, investigation. F. Anjuere: conceptualization, formal analysis, supervision, funding acquisition, validation, writing–original draft. E. Van Obberghen-Schilling: Conceptualization, supervision, funding acquisition, validation, writing–original draft. G. Orend: Conceptualization, supervision, funding acquisition, validation, visualization, writing–original draft, writing–review and editing.

The authors thank O. Lefebvre, F. Steinbach, A. Klein, C. Arnold, and the animal facility for technical support; A. Molitor and A. Pichot for RNA sequencing; K. Schlattmann, C. Alampi, M. Chami, and K. Qvortrup for EM imaging assistance; N. Toussan for CD45 immunostaining; and A. Jung for access to gene expression data. This work was supported by grants from INCa, Ligue National Contre le Cancer, and the Foundation ARC (PAIR-VADS11-023, to E. Van Obberghen-Schilling and G. Orend); AAP2017.LNCC/EVO (to E. Van Obberghen-Schilling, F. Anjuere, and G. Orend); the Ligue Régionale contre le Cancer and INSERM and University Strasbourg (to G. Orend); Cancéropôle PACA and LABEX SIGNALIFE program (ANR-11-LABEX-0028-01, to E. Van Obberghen-Schilling and F. Anjuere); Deutsche Forschungsgemeinschaft (EXC 2180, INST 2388/64-1), Ministry of Science, Research and Arts of Baden Württemberg (Az.: SI-BW 01222-91, 33-729.55-3/214-8, to K. Schenke-Layland and E.-M. Brauchle); and fellowship grants from French Ministry of Research MRT (to W. Erne) and Foundation ARC (to D. Murdamoothoo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.