We used four HNSCC cell lines: HO-1-u-1 (floor of the mouth), Sa3 (gingiva), HSC2 (oral cavity), and HSC4 (tongue). For a monolayer culture, these lines were maintained in Roswell Park Memorial Institute (RPMI) 1640 medium (Nissui, Tokyo, Japan) supplemented with 10% fetal bovine serum (Clontech Laboratories, Mountain View, CA, United States), 2 mM L-glutamine, 2.4 g/L sodium bicarbonate, 100 U/ml penicillin, and 100 μg/ml streptomycin. In culture conditions, we incubated the cells at 37°C in a humidified atmosphere containing 5% carbon dioxide in air. SCH772984 (Selleck Chemicals, Houston, TX, United States), an ERK inihibitor, was added to the medium at 5 μM for some experiments.

For the analysis of the GPNMB-positive fraction, we incubated cell pellets with allophycocyanin (APC)-conjugated anti-human GPNMB monoclonal antibodies (303822) (Novus Biologicals, Littleton, CO, United States) at a dilution of 1:11 at 4°C for 15 min. After washing with phosphate-buffered saline three times, we added 2 μg/ml propidium iodide to exclude dead cells. The cells were then filtered through a 40-μm cell strainer (Falcon, USA) and subjected to a FACSAria III cell sorter (BD, Osaka, Japan). To determine the negative fraction, we used APC-conjugated mouse IgG2b isotype antibodies (Novus Biologicals).

We employed the cell sorter to separate the cells into GPNMB-positive and GPNMB-negative fractions. Serum-free semisolid medium was constituted by adding 0.33% agar to 3D Tumorsphere Medium XF (PromoCell GmbH, Heidelberg, Germany). We placed 1 × 10³ scattered cells in 80 μl of the serum-free semisolid medium on 100 μl of solidified serum-free RPMI1640 basal layer containing 0.5% agar in each well of a 96-well plate and cultured them for 20 days. Forms of more than 50 cells were counted as spheres.

For the migration assay, we placed an 8-μm pore filter in a cell culture insert (BD Biosciences Discovery Labware, Bedford, MA, United States) in the lower compartment of a 24-well plate containing 1 ml of serum-loaded RPMI 1640 medium. We seeded 5.0 × 10³ cells suspended in serum-free RPMI 1640 medium in each upper well. After 48 h of incubation, the upper wells were swabbed and stained with hematoxylin-eosin to count the number of cells.

For the invasion assay, we precoated the 8-μm pore filters of the cell culture inserts with 500 μg of Matrigel (BD Biosciences Pharmingen), dried them in an incubator for 24 h, and rehydrated them in RPMI 1640 medium for 1 h before inoculating the cells. This assay was performed according to the same protocol as the migration assay described above, except that the cells were incubated for 72 h (16).

The study examined 174 patients diagnosed with HNSCC who were treated and followed up at Akita University from January 2010 to December 2019 (Table 1). Computed tomography, magnetic resonance imaging, and positron emission tomography-computed tomography were employed to determine the clinical staging, which followed the Union for International Cancer Control criteria. Eighty-six patients underwent chemoradiotherapy with cisplatin or bioradiotherapy with cetuximab (Table 2), and 88 patients underwent surgery (Table 3). The mean observation period was 31.6 months (range: 2–60 months). Informed consent was obtained from all patients, and the study was approved by the Akita University Ethics Committee and conducted in compliance with the Declaration of Helsinki.

Informed consent was obtained from all patients. All procedures used in this research were approved by the Ethics Committee of Akita University Hospital (Approval Number: 2532).

Immunohistochemical staining was performed on paraffin sections using the polymer-peroxidase method. Briefly, the 3-μm thick sections were deparaffinized and rehydrated routinely. Antigen retrieval was performed by boiling the sections in 0.01 M citrate buffer (pH 6.0) for 40 min. After the sections were rinsed in distilled water, the endogenous peroxidase was inactivated with 3.0% hydrogen peroxide in distilled water for 10 min at room temperature. After rinsed in Tris-buffered saline (TBS), pH 7.4, the sections were incubated with 10% normal goat serum for blocking for 60 min at room temperature and then reacted with a primary antibody of interest for 90 min at room temperature. After a wash in TBS, the secondary antibody (EnVision+/HRP; DakoCytomation, Glostrup, Denmark) was added, and the sections were incubated for 60 min at room temperature. To visualize positive signals, the peroxidase reaction was performed using 0.02% 3,3′-diaminobenzidine tetrahydrochloride (DAB) (Sigma-Aldrich, St. Louis, MO, United States) for 3 min at room temperature. The sections were counterstained with Mayer’s hematoxylin, then dehydrated and mounted. The following primary antibodies were used: rabbit polyclonal anti-GPNMB antibody (1:50, 20338-1-AP, Proteintech, Rosemont, IL, United States), rabbit monoclonal anti-SOX2 antibody (1:100, ab93689, Abcam, Cambridge, United Kingdom), rabbit monoclonal anti-Nanog antibody (1:300, ab109250, Abcam), and rabbit polyclonal anti-SNAIL + SLUG antibody (1:900, ab93689, Abcam). The intensity was assessed by at least three board-certified pathologists. The evaluation method was classified by the percentage of GPNMB-positive tumor cells, which were negative, weak, moderate, and strong according to Rietbergen et al. (17); ≤10% of the stained tumor cells were negative (0); 11%–25% were weak (1); 25%–50% were moderate (2); and ≥50% were strong (3). Negative and weak were categorized as the low group, and moderate and strong as the high group.

All experiments were independently repeated at least three times. We employed the Wilcoxon signed-rank test to compare the intensity of the primary lesions and lymph node metastases. Survival was estimated using the Kaplan–Meier method, and statistical significance was tested using log-rank tests. We used Student’s t-test to perform the data analysis and estimate the statistical significance; p-values less than 0.05 were considered statistically significant (*: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001). We performed a multivariate analysis using a Cox proportional hazards model if a significant difference was observed. IBM SPSS version 21 was used for the statistical processing (IBM Inc., Armonk, NY, USA).

The positive rate of GPNMB in the HNSCC cell lines was 13.8% for HO-1-u-1, 17.5% for Sa3, 13.8% for HSC2, and 15.68% for HSC4 (Figure 1). No reports have examined the positivity of GPNMB in HNSCC by flow cytometry or in other carcinomas. In HNSCC cell lines, the GPNMB positivity rate is expected to be approximately 15%.

Similar to normal tissue stem cells, when CSCs are cultured in serum-free medium on non-attachment dishes or in serum-free semisolid medium, they proliferate as spherical cell aggregates and maintain an undifferentiated state without maturing into non-CSCs, allowing for pure CSC cultures (18, 19). We isolated GPNMB-positive and GPNMB-negative cells separately by FACS and cultured them in serum-free semisolid medium for 20 days. The GPNMB-negative cells demonstrated little ability to form spheres, whereas the GPNMB-positive cells showed an efficient ability to form a large number of spheres. There is the possibility that GPNMB-positive cells have CSC characteristics (Figure 2).

We investigated the role of GPNMB in invasion and migration. Compared with the GPNMB-negative cells, the GPNMB-positive cells had greatly enhanced invasive and migratory potential (Figures 3, 4). It has been reported that GPNMB-positive cells can induce EMT (20); our result is consistent with that report.

The biopsy specimens from patients with HNSCC were immunostained with anti-GPNMB antibody. Negative and weak were categorized as the low group, whereas moderate and strong were categorized as the high group (Figure 5). We classified 81 of the 174 patients into the high GPNMB expression group and 93 into the low GPNMB expression group and examined their OS (Figures 6A–C, left) and PFS (Figures 6A–C, right). The OS of the high and low-expression groups was 39.4% and 57.8%, respectively (p = 0.045). The PFS of the high and low-expression groups was 27.6% and 51.6%, respectively (p = 0.013). The low-expression group showed a predominantly good OS and PFS (Figure 6A). We performed univariate and multivariate analyses, which showed that the independent prognostic factors for OS were tumor node metastasis (TNM) stage (hazard ratio [HR] 1.874, 95% CI 1.139–3.084, p = 0.013) and GPNMB (HR 1.722, 95% CI 1.075–2.758, p = 0.024) (Table 4). The only independent prognostic factor for PFS was GPNMB (HR 1.680, 95% CI 1.110–2.545, p = 0.014) (Table 5).

In this study, we classified the 86 patients treated with chemoradiotherapy or bioradiotherapy into either a high GPNMB expression group (42 patients) or a low GPNMB expression group (44 patients) and examined the OS and PFS. The OS of the high and low-expression groups was 32.4% and 56.7%, respectively (p = 0.037). The PFS of the high and low-expression groups was 24.7% and 49.1%, respectively (p = 0.031) (Figure 6B). The results of the univariate and multivariate analyses showed that the prognostic factors for OS were TNM classification (HR 3.374, 95% CI 1.629–6.986, p = 0.001) and GPNMB (HR 2.843, 95% CI 1.445–5.592, p = 0.002) (Table 6). The prognostic factors for PFS were also TNM classification (HR 2.209, 95% CI 1.177–4.143, p = 0.014) and GPNMB (HR 2.136, 95% CI 1.172–3.813, p = 0.013) (Table 7). GPNMB expression was closely related to radiosensitivity.

There was no significant difference in OS (p = 0.569) or PFS (p = 0.225) among the 39 patients in the high-expression group and the 49 patients in the low-expression group (Figure 6C).

In the surgical group, 47 patients were pathologically found to have metastatic lymph nodes. The intensity of GPNMB in the biopsy specimens and metastatic lymph nodes was evaluated from 0 to 3. In these 47 patients, the standard errors of the primary tumor and lymph nodes were compared. The intensity of GPNMB was higher in the metastatic lymph nodes than in the primary tumor (p < 0.0001) (Figure 7). From a pathological viewpoint, it can be assumed that GPNMB is closely related to metastasis. We used the Wilcoxon signed-rank test for analysis.

To compare the expression levels of GPNMB with those of CSC makers and EMT markers on the same specimens, we immunostained serial sections of biopsy specimens with the anti-GPNMB antibody along with the antibodies against CSC markers including SOX2 and Nanog and the antibody against EMT markers including Snail/Slug. As shown in Figure 8, while the HNSCC expressing a high level of GPNMB gave intensive signals of both CSC markers and EMT markers, that expressing no or a low level of GPNMB did not. This result thus indicates that GPNMB behaves as a marker of both CSC and EMT.

It is known that function of GPNMB is mainly mediated by ERK signaling. Then, we examined whether the ability of sphere formation, invasion, and migration is affected by an ERK inhibitor in GPNMB-positive cells. After FACS-isolation of GPNMB-positive fraction from each of Sa3 and HSC4 cell lines, GPNMB-positive cells were subjected to sphere formation, invasion, and migration assays in the presence or absence of the ERK inhibitor SCH772984. All of Sphere-forming ability (Figure 9A), invasive potential (Figure 9B), and migratory potential (Figure 9C) were strongly suppressed (p < 0.0001), suggesting that sphere formation, invasion, and migration enhanced in GPNMB-positive cells (Figures 2–4) should be mediated by GPNMB.