Three human OSCC cell lines established from tumour biopsies with different grades of invasive abilities were used, including OSC-20 cells (low-grade invasive cells), OSC-19 cells (low-grade invasive cells), and TSU cells (high-grade invasive cells). The OSC-20 cell line was derived from a 58-year-old female with tongue cancer metastatic to the cervical lymph nodes (25). OSC-19 was derived from a 61-year-old male with tongue cancer metastatic to the cervical lymph nodes (26). The TSU cell line was established from a patient with gingival squamous cell carcinoma who had developed marked leukocytosis (27). In addition, normal human gingival fibroblasts (HGFs; ATCC no. CRL-2014) obtained from the American Type Culture Collection (Manassas, VA, USA) served as a control. Macrophages and dendritic cells (DCs) were generated from human peripheral blood mononuclear cells (PBMCs), as described previously (28,29). PBMCs were obtained by venepuncture into an 8-ml Vacutainer CPT Cell-Preparation Tube (BD Vacutainer Systems, Franklin Lakes, NJ, USA). Briefly, monocyte-derived macrophages were generated by incubating monocytes (1×10⁶/ml) in RPMI-1640 medium containing 10% fetal bovine serum (FBS), 2 mM glutamine, 25 mM HEPES, 100 U/ml penicillin, 100 μg/ml streptomycin, and 50 ng/ml GM-CSF (PeproTech, Rocky Hill, NJ, USA) at 37°C in a CO₂ incubator for 7 days. Induced macrophages were examined using an anti-macrophage antibody (MAC387; Abcam, Tokyo, Japan). Monocyte-derived DCs were generated by incubating monocytes at 1×10⁶ cells/ml in G4 medium (G4 Dendritic Cell Generation kit; Humanzyme, IL, USA) at 37°C in a CO₂ (5%) incubator for 7 days. Induced DCs were examined using an anti-DC antibody (CD83; Abcam). Carcinoma-associated fibroblasts (CAFs) were generated by the addition of recombinant TGF-β1 to confluent growth-arrested cell monolayers, as previously described (30,31). Recombinant TGF-β1 (R&D Systems Europe Ltd., Abingdon, UK; 50 ng/ml) was prepared in serum-free culture medium. CAFs were examined using anti-αSMA (R&D Systems Europe Ltd.) antibodies. Culture inserts (1-μm pore size) were purchased from Corning Japan (Tokyo, Japan). MyD88 homodimerization-inhibitor peptide and toll-like receptor 4 (TLR4)-inhibitory peptide were purchased from Imgenex (San Diego, CA, USA) and Novus Biologicals (Littleton, CO, USA), respectively, and used in the cell-blocking study.

Experiments using human tissues were approved by the ethics committee of the Kanazawa University Graduate School of Medical Science (IRB no. 2014-004, 352-2), and written informed consent was obtained from patients providing tissue specimens. The subjects included 24 patients with primary OSCC who underwent surgical resection at the Department of Oral and Maxillofacial Surgery at Kanazawa University Hospital between 2000 and 2008. The patients with carcinoma were aged from 43 to 89 years (64.4±11.2 years, mean ± SD). The grade of tumour differentiation was determined according to the criteria proposed by the World Health Organization (32). The mode of tumour invasion was assessed according to the classification by Yamamoto et al (33).

The mRNA expression levels of PD-L1, PD-L2, E-cadherin, N-cadherin, Vimentin, and Snail1 were analysed using a Rotor-Gene Q 2plex System (Qiagen, Hilden, Germany) with FAM/ZEN/IBFQ probes (Integrated DNA Technologies, Inc., Coralville, IA, USA; DNA sequences not opened). Total RNA was extracted using the RNeasy Protect Mini kit (Qiagen), and cDNA was obtained using the PrimeScript first-strand cDNA Synthesis kit (Takara, Tokyo, Japan). All reactions were performed according to the manufacturer's instructions. We amplified 18S rRNA as an internal standard using HEX/ZEN/IBFQ probes (Integrated DNA Technologies, Inc.; DNA sequences not opened). Relative expression levels were calculated using the ΔΔCt method for qPCR (34), which presents the data as fold-differences in expression level relative to a calibrator sample; in this case, the mean expression of 3 experimental measurements of 18S rRNA in control cells or vehicle-treated cells.

The cultured cells were lysed with Pierce RIPA buffer (Thermo Scientific, Waltham, MA, USA). Lysates mixed with sample buffer were electrophoretically separated and transferred onto membranes. Membranes were blocked with Blocking One (Nacalai Tesque, Kyoto, Japan), followed by incubations with an anti-PD-L1, E-cadherin, N-cadherin, Vimentin, or Snail1 antibody (Abcam) and an anti-human β-actin antibody (Cell Signaling Technology, Tokyo, Japan). After washing with Tris-buffered saline (TBS) with 0.05% Tween, membranes were incubated with a horseradish peroxidase-conjugated anti-mouse IgG. After washing with TBS-0.05% Tween, membranes were incubated with the ECL Prime Western Blotting Detection reagent (GE Healthcare, Little Chalfont, UK). Signals were detected and analysed using C-DiGit (M&S TechnoSystems, Tokyo, Japan).

OSC-20 cells were seeded at 70% confluence and cultured for 48 and 72 h in Dulbecco's modified Eagle's medium (Sigma-Aldrich Japan, Tokyo, Japan) with 0.5% FBS (Hyclone, Logan, UT, USA) to induce EMT. Recombinant human TGF-β1 (R&D Systems, Minneapolis, MN, USA) was added to a final concentration of 5 ng/ml.

Each specimen was fixed in 10% buffered formalin and then embedded in paraffin to prepare serial sections. Immunohistochemical staining was performed using antibodies against PD-L1, Snail, Vimentin, CD83, macrophages (MAC387), αSMA, and fibroblast activation protein (FAP) (Abcam). Primary antibodies were detected using EnVision Single reagents (Dako Japan, Tokyo, Japan), Alexa Fluor 488-labeled goat anti-mouse IgG (Cell Signaling Technology), and Alexa Fluor 546-labeled goat anti-rabbit IgG (Thermo Fisher Scientific, Rockford, IL, USA). Staining specificity was confirmed using the Universal Negative Control for IS-Series Rabbit Primary Antibodies reagent (IS600; Dako Japan) as a negative control, instead of the matching primary antibody. After mounting with ProLong Gold Antifade reagent (Thermo Fisher Scientific, Tokyo, Japan), the tissues were subjected to fluorescence microscopy. Three regions were image-captured under a light microscope at ×100 magnification, and the images were evaluated based on number of stained tumour cells, following the method described by Kimura et al (24). PD-L1 expression was evaluated based on the extent of immunolabelling in tumour cell membranes and was classified on a four-point scale: 0 (no labelling, or labelling in <10% of tumour cells); 1 (weak labelling, homogeneous or patchy, in >10% of tumour cells); 2 (moderate labelling, homogeneous or patchy, in >10% of tumour cells); 3 (intense labelling, homogeneous or patchy, in >10% of tumour cells). These scores were subsequently grouped into 2 categories: low expression (0 or 1) and high expression (2 or 3).

Statistical analyses were conducted using JMP 12.0 software (SAS Institute, Inc., Cary, NC, USA). The qPCR data are presented as the mean ± standard error of the mean (SEM). Differences between groups were tested for statistical significance using the 2-tailed Mann-Whitney U test. Differences were considered significant at P<0.05.

PD-L1 and PD-L2 mRNA expression on 3 human OSCC lines with different grades of invasiveness (OSC-19, OSC-20, and TSU) were analysed. The qPCR data are presented as fold-differences relative to the mean values from 3 experiments with HGF cells (Fig. 1A). Among the 3 OSCC cell lines, the PD-L1 mRNA expression level ranged from 5.73 to 20.48-fold higher than in HGF cells (Fig. 1A), while PD-L2 mRNA expression ranged from 0.38 to 1.48-fold higher than in HGF cells (Fig. 1A). In particular, the expression of PD-L1 mRNA in low-grade invasive cells (OSC-20 cells) was highest among the 3 OSCC cell lines. Upregulation of PD-L1 mRNA in high-grade invasive cells (TSU cells) was not as remarkable compared to that in the other 2 low-grade invasive cells (OSC-20 and OSC-19 cells). PD-L2 expression on the 3 different invasive human OSCC lines was much lower, and differences were negligible compared to that of PD-L1. We further analysed PD-L1 expression at the protein level. Both low-grade invasive cells (OSC-20 and OSC-19) had a PD-L1 protein expression level higher than that of TSU cells. HGF cells exhibited negligible PD-L1 protein expression (Fig. 1B).

To evaluate whether PD-L1 expression is associated with EMT features, we analysed the mRNA- and protein-expression levels of E-cadherin, N-cadherin, vimentin, and Snail1 in the human OSC-19, OSC-20, and TSU lines with different degrees of invasiveness. The mRNA expression levels of these EMT-associated genes in OSC-19 and TSU cells are presented as fold-differences relative to those in OSC-20 cells. TSU cells displayed a mesenchymal phenotype manifested by the loss of E-cadherin and acquisition of N-cadherin, vimentin, and Snail compared to the corresponding expression levels in OSC-20 and OSC-19 cells (Fig. 2A and B). OSC-20 cells (epithelial phenotype) were exposed to TGF-β1 to determine whether such exposure could change the mRNA- and protein-expression levels of EMT-associated genes and PD-L1 (Fig. 2C and D). The mRNA-expression levels are presented as fold-differences relative to that of vehicle control-treated cells. Serial examination of EMT markers (loss of E-cadherin and upregulation of N-cadherin, vimentin, and Snail) at 48 and 72 h showed that TGF-β1 treatment resulted in EMT in OSC-20 cells (Fig. 2C). We further confirmed E-cadherin downregulation and N-cadherin, vimentin, and Snail upregulation at the protein level at 72 h after TGF-β1 treatment (Fig. 2D). Strikingly, the total PD-L1 mRNA and protein levels were reduced in OSC-20 cells after EMT induction by TGF-β1 treatment.

To confirm downregulation of PD-L1 on high-grade invasive cells compared to low-grade invasive cells, we further studied immunohistochemical analysis using 3-, 4C-, and 4D-grade invasive OSCC tissues. Previously, we demonstrated that OSC-20 and OSC-19 cells implanted into nude mice successfully reproduced the same types of diffuse invasive tumours (grades 3 and 4C in terms of the mode of invasion, respectively) (35), and repeated inoculation with TSU cells successfully reproduced the formation of grade 4D OSCC tumours in nude mice (36). With the low-grade invasive OSCC lines (grades 3 and 4C, mode of invasion), immunohistochemical analysis demonstrated that PD-L1 was detected in cancer cells, but not in stromal cells (Fig. 3A). However, PD-L1 signals were detected in stromal cells of high-grade invasive OSCC (grade 4D, mode of invasion), whereas PD-L1 was not detected in the cancer cells (Fig. 3A). Table I displays the correlations between PD-L1 expression and the clinicopathological factors. PD-L1 expression in high-grade invasive OSCCs (4D, mode of invasion) was significantly downregulated compared to low-grade invasive OSCCs (3 and 4C, mode of invasion) (P=0.0392). In contrast, PD-L1 expression did not correlate with tumour size (P=0.873), lymph node metastasis (P=0.542), local recurrence (P=0.083), or histological differentiation (P=0.145). Immunohistochemical analysis demonstrated that signals of EMT-related protein (Vimentin and Snail1) were detected in cancer cells in human OSCC tissues classified as grade 4D in terms of the mode of invasion (Fig. 3B). To analyse which cells expressed PD-L1 in the cancer stroma, we investigated cells surrounding the cancer cells by studying several biomarkers. Immunofluorescence analysis demonstrated that most CD83 (DC marker)-positive cells expressed PD-L1 and a fraction of MAC387 (macrophage marker)-positive cells expressed the PD-L1 protein in high-grade invasive OSCC tissues (grade 4D, mode of invasion; Fig. 3C). However, αSMA or FAP (CAF marker)-positive cells did not co-localize with PD-L1 expressing cells (Fig. 3C).

The above immunohistochemical analysis demonstrated that macrophages and DCs expressed PD-L1 in stromal cells in high-grade invasive OSCC tissues. Thus, we performed further in vitro studies to confirm PD-L1 expression in macrophages and DCs in vitro. We found that PD-L1 expression increased significantly (~10-fold) in macrophages or DCs co-cultured with TSU cells, relative to that observed in TSU cells, macrophages or DCs cultured separately. Differences in PD-L1 expression were negligible between TSU cells co-cultured with CAFs and TSU cells or CAFs cultured separately (Fig. 4A). In addition, PD-L1 expression in macrophages or DCs co-cultured with OSC-20 cells, as well as in separate cultures of OSC-20 cells and macrophages or DCs, did not change (Fig. 4A). Differences in PD-L1 expression were negligible between OSC-20 cells co-cultured with CAFs and OSC-20 cells or CAFs cultured separately (Fig. 4A).

Next, macrophages or DCs co-cultured with TSU cells were immunostained using an anti-PD-L1 antibody to determine which cell type expressed PD-L1. PD-L1 signals were detected in macrophages or DCs, but not TSU cells (Fig. 4B). In particular, macrophages or DCs attached to TSU cells that expressed PD-L1. PD-L1 was not detected in the separate cultures of TSU cells, macrophages, or DCs (Fig. 4B). PD-L1 upregulation was not observed after adding the supernatant from a TSU cell culture to macrophages or DCs (Fig. 4C). PD-L1 upregulation was inhibited after adding TSU cells in culture inserts to macrophages or DCs (Fig. 4C). Further, we used a blocking peptide to study the mechanism of toll-like receptors in PD-L1 induction in macrophages or DCs co-cultured with TSU cells. PD-L1 upregulation in macrophages or DCs co-cultured with TSU cells was significantly suppressed by treatment with the MyD88 homodimerization-inhibitor peptide or TLR4-inhibitory peptide (Fig. 4D).