One hundred thirty-seven patients diagnosed with laryngeal squamous cell carcinoma at the First Affiliated Hospital of Sun Yat-sen University were recruited for the present study. In detail, immunohistochemistry (IHC) was performed to evaluate the clinical relevance of tumor microenvironment factors in 102 patients, and 35 patients were selected for other in vitro studies. None of these 137 selected patients received palliative surgery or neoadjuvant chemo- and/or radiotherapy before sampling. Clinical staging was classified according to the criteria of the seventh edition of the Union for International Cancer Control (UICC). The freshly resected tumor and adjacent non-tumor tissues were kept in cold PBS for downstream analysis. The adjacent non-tumor tissues were confirmed cancer cells infiltration free by pathological examination. Peripheral blood mononuclear cells (PBMCs) were obtained from 14 healthy donors. All samples were collected after receiving informed consent from the patients, and the Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University approved this study (Approval No. 2012-349).

Paraffin-embedded, formalin-fixed, 5-μm-thick tissue sections ( Table 1 ) were incubated with antibodies against human IL-21 (10 µg/ml, NBP1-02706, Novus), FOXP3 (5 µg/ml, ab20034, Abcam), PD-L1 (1:200, 13684, Cell Signaling Technology), and anti-p16/INK4a (1:250, 10883-1-AP, Proteintech) and then stained using the Dako Envision System (DakoCytomation) according to the manufacturer’s instructions. The procedure for immunohistochemical staining evaluation was described in our previous studies (11). Briefly, for the categorization of samples by IL-21 or PD-L1 expression, specimens with a number of positive cells greater than the median were defined as ‘high’, and those with a number lower than the median were defined as ‘low’. The median level of IL-21⁺ cells was 4.5 cells per field. The expression of PD-L1 was scored semiquantitatively based on the staining intensity and distribution using the immunoreactive score (IRS). The IRS was calculated as staining intensity (SI) × percentage of positive cells (PP) (IRS= SI × PP). The SI was defined as negative (score 0), weak (score 1), moderate (score 2), and strong (score 3). The PP was defined as 0-5% (score 0), 6-25% (score 1), 26-50% (score 2), 51-75% (score 3), and 76-100% (score 4). The cutoff point between low and high PD-L1 expression was 3. P16 was considered positive if there was strong and diffuse nuclear and cytoplasmic staining present in greater than 70% of tumor cells and believed to correlated with Human papilloma virus (HPV). Cells stained with the indicated antibodies were imaged using Zeiss imaging systems (Axio Scan Z1, Carl Zeiss) at 100× and 400× magnification, and at least 5 fields of view per section at 400× magnification were evaluated.

Lymphocytes were isolated from tissue as previously described (17). Briefly, fresh surgical specimens (n=17) were minced into small pieces and subsequently digested with Dulbecco’s modified Eagle’s medium (DMEM) containing 5% fetal bovine serum (FBS), 1 mg/ml collagenase I, 0.5 mg/ml collagenase II, 1 mg/ml hyaluronidase, and 100 U/ml DNase (all from Sigma-Aldrich) at 37°C for 20 mins. After being filtered through 70 µm cell strainers, the resulting cell suspensions were processed with density gradient centrifugation using a lymphocyte separation medium (MP Biomedical).

Single-cell suspensions were stained with antibodies against CD45, CD3, CD4, PD-1, CD25, interferon (IFN)-γ, IL-17, IL-9, IL-4, FOXP3 (eBioscience), and IL-21 (BD Biosciences) according to the manufacturers’ instructions. The dose of each of the above antibodies was 5 µl/test. To detect the cytokine profile, cells were stimulated with Leukocyte Activation Cocktail (2 µl/ml, BD Biosciences) for 5 h before flow cytometric analysis. For intracellular staining, cells were stained with the surface markers listed above and fixed in Fix/Perm Buffer (eBioscience), followed by permeabilization in the presence of antibodies specific for intracellular markers. All the reagents were purchased from BioLegend unless otherwise indicated. The stained cells were acquired on a Cytoflex flow cytometer (Beckman Coulter), and the analysis was performed using FlowJo software (TreeStar).

CD4⁺ T cells were purified from PBMCs from healthy donors using a CD4⁺ T cell isolation kit (Miltenyi Biotec). Following isolation, the cells were plated at 2×10⁵ cells per well in flat-bottomed 96-well plates in RPMI 1640 medium containing 10% FBS, 2 mM L-glutamine, 0.05 mM 2-mercaptoethanol, and 100 U/ml penicillin and streptomycin. The CD4⁺ T cells were activated using a tetrameric complex of anti-CD3 and anti-CD28 beads (ImmunoCult human T cell activator, STEMCELL). Dendritic cell (DC) preparation was performed as previously described (17), Briefly, CD14⁺ cells isolated from PBMCs were cultured at a density of 1 × 10⁵ cells per well in RPMI 1640 medium supplemented with GM-CSF (50 ng/ml, 300-03, PeproTech) and recombinant human IL-4 (20 ng/ml, 200-04, PeproTech) for 6 days.

All statistical analyses were performed using IBM SPSS software version 22.0 (IBM Corporation). The Kaplan-Meier method was used to evaluate survival distributions, and differences between groups were assessed by the log-rank test. The correlation between variables was evaluated using Spearman’s correlation test. Two-sided Student’s t tests and ANOVA were used to analyze data from IHC and flow cytometry experiments, and the Bonferroni adjustment for multiple comparisons was used for between-group comparisons. Data are presented as the mean ± SD. P-values <0.05 were considered statistically significant.

We previously showed evidence that tumor-associated inflammation (TAI) was enriched in the HNSCC tumor environment (10, 17). However, the type of tumor environment inflammation that dominates the regulation of tumor immunity in HNSCC is unknown. In the present study, HNSCC tumor-infiltrating lymphocytes (TILs), which secrete typical cytokines associated with Th1-, Th2-, Th9-, Th17-, and Th21-polarized inflammation (including IFN-γ, IL-4, IL-9, IL-17, and IL-21), were identified using flow cytometric analysis. Our results showed that the frequencies of IFN-γ⁺CD4⁺ (26.2 ± 4.2%) and IL-21⁺CD4⁺ T cells (8.03 ± 4.37%) in tumor tissue were higher than those of IL-4⁺CD4⁺ (0.8 ± 0.5%, P< 0.01 and P< 0.01, respectively), IL-9⁺CD4⁺ (1.2 ± 0.7%, P< 0.01 and P< 0.01, respectively), and IL-17⁺CD4⁺ T cells (3.2 ± 2.6%, P< 0.01 and P< 0.01, respectively), which indicated that IFN-γ and IL-21 may be the dominant secreted cytokines of tumor-infiltrating Th cells ( Figures 1A, B , Supplementary Figure 1 ).

We next compared the differences in the IFN-γ⁺CD4⁺ and IL-21⁺CD4⁺ T cell frequencies between tumor and adjacent non-tumor tissues. Notably, our results showed that despite the high frequency of IFN-γ⁺CD4⁺ T cells in the tumor tissues, the frequency of IFN-γ⁺CD4⁺ T cells was not considerably different between the tumor and adjacent non-tumor tissues (26.2 ± 4.2% vs. 22.8 ± 4.8%, respectively, P > 0.05) ( Figures 1A, B ). However, the frequency of IL-21⁺CD4⁺ T cells was significantly increased in the tumor tissues compared with the paired adjacent non-tumor tissues (8.03 ± 4.37% vs. 2.4 ± 1.4%, respectively, P < 0.01) ( Figures 1A, B ). Moreover, the IL-21-producing T cells in the tumor tissues were FOXP3 negative, and the majority of these cells (84.5 ± 3.3%) were CD3 and CD4 positive ( Figures 1C, D ). Finally, we found that approximately half of the IL-21⁺CD4⁺ T cells in the tumor tissues were IFN-γ positive (45.6 ± 5.2%) and that IL-21⁺CD4⁺ T cells were rarely IL-4 (0.8 ± 0.3%) or IL-17 (1.6 ± 0.5%) positive ( Figures 1E, F ).

We subsequently evaluated the clinical relevance of the increased frequency of tumor-infiltrating IL-21-producing cells in 102 HNSCC patients using IHC. Our results showed that IL-21⁺ cells accumulated in the tumor stroma but not in the tumor nest ( Figure 2A ). The levels of IL-21⁺ cell infiltration in patients with stage IV disease were higher than those with stage I, II or III disease (8.9 ± 3.6 vs. 4.0 ± 2.7, P < 0.001, vs. 4.1 ± 2.9, P < 0.001, vs. 5.6 ± 3.1, P < 0.05, counts per field, respectively), Additionally, those with stage III disease was higher than stage I (5.6 ± 3.1 vs. 4.0 ± 2.7, P < 0.05, counts per field) ( Figures 2A, B ).

To determine whether IL-21⁺ cell infiltration in HNSCC correlates with disease prognosis, one hundred and two HNSCC patients were divided into two groups according to the median value of their IL-21⁺ cell density. The patients with a high level of infiltrating IL-21⁺ cells had worse overall and disease-free survival than the patients with a low level ( Figures 2C, D and Table 2 ). Cox regression analysis revealed that the IL-21⁺ counts and TNM stage could be independent predictive factors for overall survival in the HNSCC patients ( Table 3 ). Taken together, our results identified that the accumulation of IL-21⁺ cells was associated with disease progression and poor prognosis in patients with HNSCC.

Considering that Tregs are known to suppress antitumor immunity and that this suppression results in disease progression, we next determined whether the level of IL-21⁺ cells was related to Treg infiltration. We first examined the prevalence of tumor-infiltrating Tregs and IL-21⁺ cells using immunohistochemical staining for FOXP3 and IL-21. The results showed that FOXP3⁺ and IL-21⁺ cells collocated in the same area in the tumor stroma ( Figure 3A ) and that the increased density of the IL-21⁺ cells was positively correlated with that of the FOXP3⁺ cells ( Figure 3B ). Next, flow cytometric analysis showed that the frequency of tumor-infiltrating IL-21⁺CD4⁺ T cells positively correlated with that of FOXP3⁺CD25⁺CD4⁺ T cells ( Figures 3C, D ), which supported our immunohistochemical results. We further assessed IL-21R expression on Treg cells and conventional T cells (CD45⁺CD3⁺CD4⁺CD8^-CD25^-FOXP3^-), showing that Tregs have remarkably more IL-21R expression ( Supplementary Figure 2 ). These results implied that IL-21⁺ cells may be involved in Treg infiltration in the HNSCC tumor microenvironment.

To determine whether tumor-infiltrating IL-21⁺ cells contribute to Treg generation, recombinant human IL-21 was added to cultured CD4⁺ T cells. However, the unexpected results showed that the frequency of Tregs markedly declined in the CD4⁺ T cells + IL-21 group compared with the CD4⁺ T cells alone group (4.8 ± 0.8% vs. 7.2 ± 0.9%, respectively, P < 0.05) ( Figures 4A, B ), and this result was inconsistent with our clinical data. We speculated that IL-21 may have the effect of antagonizing Treg when treated alone, which was supported by literature demonstrating the pathogenic role of IL-21 in human inflammatory diseases, and may act in concert with other immunosuppressive factors in Treg generation in the tumor microenvironment. As the PD-L1/PD-1 pathway has been recognized as a molecular checkpoint in immune evasion in various cancers (18–22), we next examined the effect of IL-21 on Treg generation from CD4⁺ T cells in the presence of PD-L1. Excitingly, our results showed that the frequency of Tregs was not only higher in the CD4⁺ T cells + PD-L1 group than in the CD4⁺ T cells alone group (13.2 ± 3.2% vs. 7.2 ± 0.9%, respectively, P < 0.05) but also higher in the CD4⁺ T cells + PD-L1 + IL-21 group than in the CD4⁺ T cells + PD-L1 group (17.9 ± 4.1% vs. 13.2 ± 3.2%, respectively, P < 0.05) ( Figures 4C, D ), which meant that IL-21 could enhance PD-L1-induced Treg generation. We next examined whether this Treg generation was PD-1 dependent, and the results showed that compared with PD-L1 (23.5 ± 3.5%, P < 0.05) or IL-21 alone (19.4 ± 5.1%, P < 0.05), the combined use of PD-L1 and IL-21 (32.0 ± 4.4%) markedly upregulated PD-1 expression on CD4⁺ T cells ( Figures 4E, F ), suggesting a synergistic effect of IL-21 and PD-L1 on PD-1 induction. Most importantly, blocking PD-1 with a neutralizing antibody abrogated the effects of IL-21 and/or PD-L1 on Treg generation (vs. IL-21+PD-L1: 5.7 ± 1.5% vs. 17.9 ± 4.1%, P < 0.05; vs. PD-L1: 8.1 ± 1.5% vs. 13.2 ± 3.2%, P < 0.05) ( Figures 4G, H ), which indicated that IL-21 and PD-L1-induced Treg generation was PD-1 dependent.

For the analysis of the functionality of the generated Tregs, Tregs were isolated from different groups and cocultured with CFSE-labeled TAA-specific T cells. Our results showed that the generated Tregs from the CD4⁺ T cells + PD-L1 + IL-21 group (17.8 ± 4.9%, P<0.05) or the CD4⁺ T cells + PD-L1 group (14.3 ± 6.1%, P<0.05) had stronger suppressive effects on TAA-specific T cell proliferation than those from the CD4⁺ T cells alone group (29.0 ± 5.2%), whereas compared with those from the CD4⁺ T cells alone group, the Tregs isolated from the CD4⁺ T cells + IL-21 group showed an impaired suppressive ability (39.7 ± 7.0% vs. 29.0 ± 5.2%, respectively, P<0.05) ( Figures 4I, J ). In addition, TAA-specific CD8⁺ T cells were assessed as well ( Supplementary Figure 3 ). It revealed that PD-L1 and IL-21-induced Tregs show expected suppressive effects as on TAA-specific CD8⁺ T cells. Taken together, schematic figure ( Figure 4K ) illustrates that IL-21 promoted PD-L1-induced Treg generation in a PD-1-dependent manner, and IL-21 and PD-L1-induced Tregs could inhibited TAA-specific T cell proliferation at a greater degree than naturally occurring Tregs.

To test whether the HNSCC tumor environment is capable of potentiating Treg generation through IL-21 and PD-L1, we cocultured CD4⁺ T cells with tumor explants from HNSCC patients with or without IL-21 and/or an anti-PD-1 neutralizing antibody. Accordingly, tumor explants from 10 patients with HNSCC were divided evenly into 2 groups according to the cutoff points for PD-L1 and IL-21 described in the methods section. There were 5 samples with high expression of both IL-21 and PD-L1 (IL-21^high/PD-L1^high) and 5 samples with low expression of both IL-21 and PD-L1 (IL-21^low/PD-L1^low). Our results showed that the frequency of Tregs in the group of IL-21^high/PD-L1^high tumor explants + CD4⁺ T cells was higher than that of the group of IL-21^low/PD-L1^low tumor explants + CD4⁺ T cells (38.8 ± 8.5% vs. 25.8 ± 2.4%, respectively, P < 0.05) ( Figures 5A, B ). The anti-PD-1 antibody showed an inhibitory effect on Treg generation when the CD4⁺ T cells were cultured with the IL-21^high/PD-L1^high tumor explants (31.3 ± 5.0% vs. 38.8 ± 8.5%, P < 0.05). In addition, the combination of IL-21 and the anti-PD-1 antibody had a stronger inhibitory effect on Treg generation than the anti-PD-1 antibody alone (26.6 ± 4.7% vs. 31.3 ± 5.0%, respectively, P < 0.05), whereas blocking IL-21 alone had no significant inhibitory effect on Treg generation (38.1 ± 7.1% vs. 38.8 ± 8.5%, P > 0.05). Conversely, the Treg generation induced by the IL-21^low/PD-L1^low tumor explants failed to be reversed by neutralizing IL-21 (24.4 ± 3.7% vs. 25.8 ± 2.4%, P > 0.05), PD-1 (22.8 ± 2.4% vs. 25.8 ± 2.4%, P > 0.05), or both IL-21 and PD-1 (22.5 ± 3.1% vs. 25.8 ± 2.4%, P > 0.05), indicating that Treg generation in the IL-21^high/PD-L1^high tumor microenvironment may be regulated by mechanisms distinct from those in the IL-21^low/PD-L1^low tumor microenvironment ( Figures 5A, B ). Treg generation induced by IL-21^high/PD-L1^high tumor explant was further validated by cocultured with TAA-specific T cells (both CD4 and CD8) ( Supplementary Figures 4 , 5 ). Using intracellular staining, the cytokine-producing ability of responder CD4⁺ T cells and CD8⁺ T cells was assessed. Our results showed that the stimulated responder CD4⁺ and CD8⁺ T cells exhibited strong proliferation in both, high IFN-γ, IL-2 levels in responder CD4⁺ T cells, high GranzymeB and Perforin levels in responder CD8⁺ T cells. Adding the induced Treg by tumor explants with or without anti-PD-1 or/and anti-IL-21 treatment strongly inhibited the proliferation of responder T cells (both CD4 and CD8) ( Supplementary Figures 4A , 5A ), reduced the IFN-γ, IL-2 levels of CD4⁺ T cells ( Supplementary Figures 4B, D ), and cytotoxicity, GranzymeB, Perforin levels of CD8⁺ T cells ( Supplementary Figures 5B, E ). However, no significant differences were observed between tumor explant with no antibody, with anti-PD-1, with anti-IL-21 group and with the combination anti-PD-1 and anti-IL-21. ( Supplementary Figures 4D, 5E ).There data suggested the Tregs induced by tumor explant with the treatment of anti-PD-1 or/and anti-IL-21 are expectedly suppressive as the Tregs induced by tumor explant alone.

The above data suggested that IL-21 neutralization may enhance the effect of PD-1-targeted tumor immunotherapy in only HNSCC patients with high expression of both IL-21 and PD-L1.

Given that IL-21 plays a positive role in PD-L1-induced Treg generation, we next investigated the relationships between the expression patterns of IL-21, PD-L1, and FOXP3 in tumor specimens. Our results showed that the tumors with high expression of IL-21 or PD-L1 had more FOXP3⁺ cells than those with low expression of both IL-21 and PD-L1 (6.8 ± 3.6 vs. 3.2 ± 2.7 counts per field, respectively, P < 0.05) ( Figures 6A, B ). Furthermore, the tumors with high expression of both IL-21 and PD-L1 had more FOXP3⁺ cells (9.6 ± 4.9 counts per field) than the tumors with low expression of both IL-21 and PD-L1 (3.2 ± 2.7 counts per field, P < 0.05) or the tumors with high expression of either IL-21 or PD-L1 (6.8 ± 3.6 counts per field, P < 0.01) ( Figures 6A, B ). Moreover, in the tumor stage analysis, the PD-L1 expression score in patients with stage IV disease was considerably higher than that in those with stage I, II or III disease (5.8 ± 1.9 vs. 2.0 ± 1.6, P <0.001, vs. 3.1 ± 2.2, P < 0.01, vs. 3.2 ± 1.8, P <0.05, respectively), ( Figure 6C ). Fifteen cases (47%) with simultaneous high expression of IL-21 and PD-L1 were observed among the stage III or IV samples, and 20 cases (29%) were observed among the stage I or II samples, while simultaneous low expression of IL-21 and PD-L1 was observed in 8 cases (25%) among the stage III or IV samples and in 36 cases (51%) among the stage I or II samples. Moreover, high expression of either PD-L1 or IL-21 was observed in 9 cases (28%) among the stage III or IV samples, and 14 cases (20%) were observed among the stage I or II samples ( Figure 6D ). We finally evaluated whether the simultaneous high expression of IL-21 and PD-L1 correlated with the clinical prognosis of HNSCC patients. As expected, the patients with simultaneous high expression of IL-21 and PD-L1 had worse overall and disease-free survival than those with simultaneous low expression of PD-L1 and IL-21 (OS: P < 0.01, DFS: P < 0.01), worse overall survival than high expression of either PD-L1 or IL-21 (P < 0.05, DFS not significant) ( Figures 6E, F ). Overall survival was still significantly different between IL-21^high/PD-L1^high and IL-21^low/PD-L1^low group at stages I + II (P < 0.05) and III + IV (P < 0.05), respectively ( Figures 6G, H ). Disease-free survival was significantly different between IL-21^high/PD-L1^high and IL-21^low/PD-L1^low group at stages I + II (P < 0.05), but not stages III + IV (P > 0.05) ( Figures 6I, J ). Cox regression analysis revealed that the simultaneous high expression of PD-L1 and IL-21 was an independent prognostic marker for overall and disease-free survival in HNSCC patients ( Table 3 ).