We examined tumor sections from 110 patients with tongue SCC who underwent primary surgery at a hospital at the University of Occupational and Environmental Health, Kitakyushu, Japan, between January 1997 and August 2017. Patients in this study had the following background: Tumor size ≤40 mm; depth of invasion ≤10 mm; no metastasis in any node; surgical margins >5 mm; and had not undertaken neoadjuvant therapy prior to surgery. Lymph node metastasis was monitored using imaging analyses (ultrasonography, computed tomography, or magnetic resonance imaging) in 2- to 6-month intervals post-surgery and verified by pathological diagnosis. Lymph node relapse-free survival (LNFS) and overall survival (OS), as the endpoints of analysis (September 2019), were defined as the time from surgery until lymph node relapse or death had occurred, respectively, or until the last follow-up visit.

Patient samples (females, 32; males, 78) aged 32–89 years [mean ± standard deviation (SD), 64.6±13.3 years] were obtained at diagnosis.

Our study was approved by the Research Ethics Committee of the University of Occupational and Environmental Health (permission nos. H29-212 and H24-6) and was performed in accordance with the guidelines of the Declaration of Helsinki.

After fixation of resected tumor tissues in 10% formalin and embedding in paraffin, sections (thickness, 3 µm) underwent hematoxylin and eosin (H&E) staining.

Histological classification was based on the American Joint Committee on Cancer Staging Manual, 8th edition (27). The WPOI classification system was used to assess the pattern of tumor invasion in patients. Tumor differentiation was divided into three groups: well-differentiated (I), moderately differentiated (II), and poorly differentiated (III; Table I).

An automated immunostainer, Histostainer 36A (Nichirei Biosciences Inc.), was used to process sections that had been fixed in 10% formalin and immersed in paraffin in accordance with the manufacturer's protocol. Antibodies used were: mouse monoclonal anti-human CD68 (clone Kp-1, 1:100) and CD8 (clone C8/144B, 1:100), both from Dako Japan; and rabbit polyclonal antibody to CCL22 (1:100) and goat polyclonal antibody to CCR4 (1:300), both from Abcam. Mouse monoclonal anti-human VEGF-C (clone MM0006-2E65, 1:50) was obtained from Novus Biologicals. Mouse monoclonal anti-human antibody to D2-40 (clone D2-40, 1:1; Nichirei) was used to examine lymphatic vessels.

Five areas were randomly selected and CCL22⁻, CCR4⁻, CD8⁻, and VEGF-C-positive cells were counted using a microscope (magnification, ×400); data was expressed as per mm² surface area. Two independent pathologists, blinded to patients' backgrounds or their prognosis, evaluated all immunochemical and histological slides.

Cell proliferation assays were also performed, as described in our previous study, by using anti-Ki67 (mouse monoclonal, clone MIB-1; 1:100; Dako) (6).

Agreement among observers was excellent (agreement >95%) for all antibodies and according to an interclass correlation coefficient. In the case of a rare disagreement, a third, departmental board-certified pathologist calculated a consensus score.

Thin sections (4 µm) were used for immunofluorescence assays and anti-CCL22, CCR4 and VEGF-C antibodies were used in experiments as described above. In addition, we used rabbit polyclonal anti-human VEGF-C (1:100; Santa Cruz Biotechnology, Inc.), and rabbit polyclonal anti-human signal transducer and activator of a transcription 6 (STAT6; phospho Y641) (1:100; Abcam). Secondary antibodies used were: rhodamine-conjugated donkey anti-rabbit IgG (1:200; Merck) and goat anti-mouse IgG (H+L; 1:200; Merck); and FITC-conjugated goat anti-rabbit IgG (H+L; 1:200; Merck) and donkey anti-goat IgG (1:200; Merck). A nuclear stain, 4′,6-diamidino-2-phenylindole (DAPI; GeneTex), was used on sections, which were then mounted. An ECLIPSE E600 inverted fluorescence microscope (Nikon) was used to examine slides, and images captured and analyzed with LuminaVision software (v. 2.2.2; Mitani Corporation).

Using anti-D2-40, the density of lymphatic vessels (LVD) was measured as previously described (28). LVD was evaluated at the periphery, within 2 mm of the tumor, and next to the invasive front. Five areas that showed many lymphatic vessels were chosen using light microscopy at a ×40 magnification. All stained vessels in each area at a ×200 magnification were counted and the data was expressed as per mm². The mean number of lymphatic vessels was calculated and expressed as LVD in a blinded manner (described above).

A human macrophage cell line (human CD14⁺ monocytes from peripheral blood, single donor; C-12909, http://www.promocell.com/product/human-cd14-monocytes-hmocd14-pb/) from PromoCell GmbH (Heidelberg, Germany) was cultured at a density of 1×10⁶ cells in Monocyte Attachment Medium (PromoCell GmbH). Cells were cultured in M1-Macrophage Generation Medium DXF so that they could differentiate into M1- and M2-like macrophages as in our previous study (19). In addition, these cells were treated with 1, 10 or 20 ng/ml of interleukin-4 (IL-4; R&D Systems) for 24 h. Human monocytic (THP-1) cells were obtained from the American Type Culture Collection and maintained in RPMI-1640 containing 10% fetal bovine serum (Gibco; Thermo Fisher Scientific, Inc.), 1% penicillin-streptomycin (Gibco; Thermo Fisher Scientific, Inc.), and 2 mmol/l of glutamine (Gibco; Thermo Fisher Scientific, Inc.).

For quantitative real-time polymerase chain reactions (qRT-PCR), a TaqMan assay and a 7700 Sequence Detection System (Applied Biosystems; Thermo Fisher Scientific, Inc.) were used as in our previous study (19). A pre-made primer/probe set (Applied Biosystems Japan, Ltd.) containing primers and fluorogenic probes for CCL22 and VEGFC was used. Messenger RNA expression was normalized to that of 18s ribosomal RNA in each sample.

The CCL22 gene was cloned from an upstream gene region using information on chromosome 16q13, and a luciferase assay was performed as described in our previous study (29). Genomic DNA was extracted from monocytes isolated from 15 ml whole blood donated from four healthy volunteer donors. A human CCL22 promoter region spanning between −966 and +32 bp from the transcription start site was then generated by PCR. A pGEM-T Easy vector was used to subclone the PCR product. Serial 5′-deletion constructs (−518/+32, −491/+32, −281/+32) were generated by PCR and digestion with appropriate restriction enzymes. Luciferase reporter genes that included a mutation in the STAT6 site of the CCL22 promoter were generated by site-directed mutagenesis (Stratagene) using the following primers: 5′-ATGTGGACAGCACGAGAAGCCCCAGAT-3′ for STAT6 mutation, where the underlined nucleotides represented the mutated site.

Twenty micrograms of promoterless pGL3-basic vector or CCL22 promoter luciferase constructs were used to transfect THP-1 monocytic cells treated with 20 ng/ml 12-O-tetradecanoylphorbol-13-acetate as in our previous study (29). A pSV-beta-galactosidase control plasmid (Promega Corporation) was used as an internal control. A luminometer (Bio-Orbit Oy) was used to measure luminescence.

CCL22 protein expression via IL-4 (R&D Systems), with or without AS1517499 (Axon Medchem BV) as a selective STAT6 inhibitor, in human macrophage cell line (C-12909; PromoCell GmbH) culture supernatants was measured using an enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems) as in a previous study (29).

Statistical analyses of data between the groups were performed using one-way ANOVA followed by the Tukey-Kramer post-hoc comparison test. Survival curves were plotted according to a Kaplan-Meier method and compared using a log-rank test. Odds ratios (ORs) and corresponding 95% confidence intervals were calculated from logistic regression models. Pearson's correlation coefficient was used to assess the association between lymphangiogenesis and other factors.

A P-value ≤0.05 was considered statistically significant. EZR software (Saitama Medical Center, Jichi Medical University, Saitama, Japan) was used for all analyses. EZR software is a graphical user interface for R, a modified version of R commander (version 1.6–3; The R Foundation for Statistical Computing, Vienna, Austria, version 2.13.0) with statistical functions used in biostatistics (30).

The optimal cut-off values for lymph node relapse (maximizing the sum of sensitivity and specificity) were also analyzed by EZR software using receiver operating characteristic (ROC) curve analysis.

Table I shows clinicopathological features of the 110 patients of this study based on a WPOI classification system. The patients were categorized as follows, and the various parameters were compared: WPOI-1, 19 (17.3%); WPOI-2, 24 (21.8%); WPOI-3, 24 (21.8%); WPOI-4, 29 (26.4%); and WPOI-5, 14 (12.7%).

There was a high correlation between lymphatic invasion and WPOI classification (WPOI-1 vs. WPOI-2 vs. WPOI-3 vs. WPOI-4 vs. WPOI-5; 0.0 vs. 12.5 vs. 50.0 vs. 62.1% vs. 100%, respectively). WPOI-1 and −2 patients with lymph node relapse were 2/43 (4.7%) of the patient cohort. In comparison, 16/67 patients (23.9%) with WPOI-3, −4, and −5 showed lymph node relapse (Table I). The occurrence of lymph node relapse was significant in WPOI-3, −4, and −5 cases (P=0.008; Fisher's exact test) suggesting WPOI-3, −4, and −5 lesions were prone to lymph node relapse. In terms of lymph node relapse rates, WPOI was classified into two distinct groups: WPOI-1 and −2 (WPOI-low); and WPOI-3, −4, and −5 (WPOI-high).

The 5-year LNFS rates in patients in WPOI-low vs. WPOI-high groups were 88.4% vs. 65.7%, respectively; the difference was statistically significant (P=0.001; Fig. 1A). In addition, the 5-year OS rates in patients in WPOI-low vs. WPOI-high groups were 93.6 vs. 76.1%, respectively; the difference was statistically significant (P=0.006) (data not shown).

Ki67 (a marker of tumor proliferation) was expressed only in the margin of the tumor infiltrating region in WPOI-low and WPOI-high cases; it was randomly expressed in tumor cells of the whole infiltrate (Fig. 1B; Ki67). Scattered CD8-positive cells (cytotoxic T lymphocytes; CTLs) were found in the tumor microenvironment and only a few were found in WPOI-high compared to WPOI-low tissues (Fig. 1B; CD8). WPOI-high tissues had significantly more Ki67-positive tumor cells than WPOI-low tissues (P<0.001; Fig. 1C). In contrast, CD8-positive cells were significantly decreased in WPOI-high compared to WPOI-low lesions (P<0.001; Fig. 1D).

In short, the WPOI classification correlated well with the growth potential of the tumor and decreased cellular immunity.

The number of CD68-positive macrophages per area did not significantly differ with WPOI classification, but the mean value tended to be inversely proportional to WPOI classification (Fig. 1E-CD68 and F). In contrast, the numbers of CCL22-positive macrophages per area did not differ in WPOI-1 vs. −2, but increased proportionally from WPOI-2 to −5 (Fig. 1E-CCL22 and G). Fig. 1H shows the CCL22/CD68 ratio (CCL22 ratio). WPOI-high SCC had a significantly high CCL22 ratio (vs. WPOI-low; P<0.001). These results suggest that the CCL22 ratio may be an indicator of WPOI classification.

Fig. 2A shows H&E and CCL22 stains for the same grade of WPOI-high cases, with or without lymph node relapse, during the observation period. CCL22-positive macrophage expression was significantly higher in lymph node relapse cases (Fig. 2B). In comparison, CD68 expression showed a slight low mean and median value in patients with lymph node relapse, but this was not significantly different compared to those without lymph node relapse (Fig. 2C). In cases of lymph node relapse, the expression of CD8 positive cells was significantly lower, whereas the expression of Ki67 was significantly higher (Fig. 2D and E).

For WPOI-high patients, the relationship between lymph node relapse and prognosis was examined. As a result, a difference in the 5-year OS rates between patients, with and without lymph node relapse, was statistically significant (P=0.0014; Fig. 2F).

Lymphatic vessels with D2-40 immunohistological staining are shown in Fig. 3A. A significant relationship between increased LVD and WPOI was observed (WPOI-low vs. -high; P<0.001; Fig. 3B). When LVD was compared in WPOI-high cases, with or without lymph node relapse, it was found to be significantly increased in the lymph node relapse group (P<0.001; Fig. 3C).

Fig. 3D shows the relationship between DOI and lymph node relapse in WPOI-high SCC. DOI was significantly higher in patients with lymph node relapse (without vs. with lymph node relapse; mean, 3.96 vs. 6.26 mm, P=0.0145). Therefore, not only WPOI but also DOI seems to be related to LVD.

Fig. 3E shows the ROC curve of DOI for lymph node relapse [area under the curve (AUC); 0.7524]. DOI could be thus classified into two (<3.4 mm; n=63; and ≥3.4 mm; n=47) expression groups using the Youden index from the ROC curve (sensitivity, 0.8333; specificity, 0.6522). When comparing between WPOI-low and -high for DOI cut-off values (n=5 and n=42), the positive rate of DOI was significantly higher in the WPOI-high group (Fisher's exact test; P<0.001). Additionally, this cut-off value was significantly correlated with LVDs in all patients (Fig. 3F). The expression of CD68 and CCL22 in TAMs showed opposing trends when comparing negative and positive DOI (Fig. 3G and H). As a result, the CCL22-positive/CD68-positive macrophage ratio (CCL22 ratio) was most useful as an index for macrophages with a DOI cut-off value for lymph node relapse (Fig. 3I).

A correlation between LVD and the number of CCL22-positive macrophages or the CCL22 ratio was examined. The increase in the LVD was proportional to the expression of both the number of CCL22-positive macrophages/mm² (R=0.4400; P<0.001) and CCL22 ratio (R=0.5983; P<0.001; Fig. 3J), showing a significantly positive correlation.

These results suggest that both WPOI and DOI may be involved in CCL22-mediated lymph node metastasis. Additionally, a reason for the correlation between CCL22 and LVD was thought to be that CCL22-positive macrophages expressed VEGF-C (Fig. 3K).

WPOI-high patients had a higher lymph node relapse rate. We therefore examined the predictors for lymph node relapse in this patient group.

We were able to divide the patients into two groups according to the CCL22 ratio as follows: low (<0.38; n=72) and high (≥0.38; n=38) based on ROC curve analysis for lymph node relapse. The AUCs for the number of CCL22/mm² and CCL22 ratios for lymph node relapse according to the ROC curve were 0.7811 vs. 0.8210. The AUC was higher for the CCL22 ratio. In patients with high WPOI, predictor variables including grade, DOI, CCL22 ratio, and majority parameters (age, alcohol use, sex, and smoking) were analyzed in a logistic regression model with lymph node recurrence as the dependent variable. Table II shows adjusted OR and p values. The lymph node relapse in patients with WPOI-high was significantly associated with a high CCL22 ratio (P=0.0225) but not with the other variables by univariate and multivariate logistic regression model analyses.

In WPOI-high lesions, the expression of CCR4-positive cells was found to be significantly higher than in WPOI-low lesions (Fig. 4A). In addition, more CCR4-positive cells were observed immunohistologically in patients with lymph node relapse in the same lesion as in Fig. 2A (Fig. 4B). Among WPOI-high cases, CCR4-positive cells were significantly more prevalent in lymph node relapse cases (Fig. 4B) and were found around VEGF-C positive macrophages (Fig. 4C). The aggregation of CCR4-positive cells was proportional to the expression of VEGF-C-positive cells, with a significant correlation noted (R=0.6532; P<0.001; Fig. 4D).

Thus, the function of CCR4-positive cells was also important as a factor in the correlation between CCL22 and VEGF-C.

In vitro, CCL22 expression was significantly higher in M2-like compared to M1-like macrophages, as was VEGF-C expression; in M2-like macrophages, the expression of both was increased by IL-4 in a dose-dependent manner (Fig. 4E and F). The regulation of VEGF-C expression in macrophages is known to be via an IL-4/STAT6 signaling pathway (31); CCL22 was also examined in this study. We found that W1 and W2 promotors, including a STAT6 motif, had significantly higher luciferase activity after IL-4 stimulation. However, there was no significant difference in the luciferase activities of W3 and W4 promoters without a STAT6 motif after IL-4 stimulation. When a STAT6 mutation was inserted into W1 (M1), the luciferase activity significantly decreased (Fig. 4G) and no longer responded to IL-4 stimulation. STAT6 activation also affected the production of CCL22 protein via IL-4 (Fig. 4H). Our results supported the conclusion that IL-4 induces CCL22 expression via an STAT6-dependent pathway. Fig. 4I shows STAT6 activated-cells (green) and VEGF-C-positive macrophages (red). VEGF-C-positive cells accompanied STAT6 activation. In addition, STAT6 activation of almost all carcinoma cells was observed in the deeply invaded part of the tumor. However, in normal tissue around the tumor, STAT6 activation was observed only in basal cells (Fig. 4I normal).

Thus, STAT6 activation was considered to be an important factor in lymph node metastasis of tongue SCC.