In Situ PD-L1 Expression in Oral Squamous Cell Carcinoma Is Induced by Heterogeneous Mechanisms among Patients

The expression of programmed death ligand-1 (PD-L1) is controlled by complex mechanisms. The elucidation of the molecular mechanisms of PD-L1 expression is important for the exploration of new insights into PD-1 blockade therapy. Detailed mechanisms of the in situ expression of PD-L1 in tissues of oral squamous cell carcinomas (OSCCs) have not yet been clarified. We examined the mechanisms of PD-L1 expression focusing on the phosphorylation of downstream molecules of epidermal growth factor (EGF) and interferon gamma (IFN-γ) signaling in vitro and in vivo by immunoblotting and multi-fluorescence immunohistochemistry (MF-IHC), respectively. The in vitro experiments demonstrated that PD-L1 expression in OSCC cell lines is upregulated by EGF via the EGF receptor (EGFR)/PI3K/AKT pathway, the EGFR/STAT1 pathway, and the EGFR/MEK/ERK pathway, and by IFN-γ via the JAK2/STAT1 pathway. MF-IHC demonstrated that STAT1 and EGFR phosphorylation was frequently shown in PD-L1-positive cases and STAT1 phosphorylation was correlated with lymphocyte infiltration and EGFR phosphorylation. Moreover, the phosphorylation pattern of the related molecules in PD-L1-positive cells differed among the cases investigated. These findings indicate that PD-L1 expression mechanisms differ depending on the tissue environment and suggest that the examination of the tissue environment and molecular alterations of cancer cells affecting PD-L1 expression make it necessary for each patient to choose the appropriate combination drugs for PD-1 blockade cancer treatment.


Introduction
Many types of cells, such as lymphocytes, macrophages, myeloid-derived suppressive cells, and fibroblasts, infiltrate into cancer tissues and release many kinds of cytokines and create a complex microenvironment called the tumor microenvironment (TME), which influences tumor development and progression and interferes with the efficacy of immunotherapy [1,2]. Programmed death ligand-1 (PD-L1) expressed in the TME is known to The PD-L1 expression level in HSC3, HSC4, and SAS increased by EGF in a dosedependent manner ( Figure 1A). EGFR and its downstream molecules, STAT1, AKT, and ERK, were phosphorylated for 10 to 120 min after stimulation with EGF ( Figure 1B). These inductions and phosphorylation were inhibited by the EGFR inhibitors, cetuximab, and gefitinib, which suggests that EGF upregulates PD-L1 expression in OSCC cell lines ( Figure 1C,D). The upregulation of PD-L1 by EGF was inhibited by PI3Ki (Omipalisib) and MEKi (GSK1120212), accompanied by the inhibition of phosphorylation of AKT and ERK, respectively, in HSC3 ( Figure 1E). Interestingly, PD-L1 upregulation was inhibited only by PI3Ki, but not by MEKi in HSC4, and, contrarily, it was inhibited only by MEKi but not PI3Ki in SAS. JAK2i (AZD1480) suppressed PD-L1 expression at high concentrations (10 µM) only in HSC3. However, it was not suppressed at low concentrations, and phosphorylation of STAT1 was not inhibited, suggesting the occurrence of non-specific inhibition. These findings suggest that PD-L1 expression was upregulated by EGF through the EGFR/PI3K/AKT, EGFR/MEK/ERK, and/or EGFR/STAT1 pathways. However, the pathways utilized were dependent on the cells ( Figure 2D). EGFR/PI3K/AKT, EGFR/MEK/ERK, and/or EGFR/STAT1 pathways. However, the pathways utilized were dependent on the cells ( Figure 2D). OSCC cell lines were treated with EGF at the indicated dose. PD-L1 expression was tested by flow cytometry. Statistical analysis was performed on the results of EGF 0 ng/mL. (B) Cells were treated with EGF (50 ng/mL) each time, and the phosphorylation of EGFR downstream proteins was evaluated by Western blotting. (C,D) Cetuximab and gefitinib were added into the OSCC cell culture at the indicated dose 60 min before EGF treatment (50 ng/mL). PD-L1 expression and the phosphorylation of EGFR downstream proteins were evaluated after 24 h and 60 min, respectively. (E) Omipalisib, AZD1480, and GSK1120212 were added into the OSCC cell culture at the indicated dose 60 min before EGF treatment (50 ng/mL). PD-L1 expression and phosphorylation of EGFR downstream proteins were evaluated after 24 h and 60 min, respectively. Statistical analysis was performed for the addition of EGF alone. * p < 0.05, ** p < 0.01. Statistical analysis was performed on the results of EGF 0 ng/mL. (B) Cells were treated with EGF (50 ng/mL) each time, and the phosphorylation of EGFR downstream proteins was evaluated by Western blotting. (C,D) Cetuximab and gefitinib were added into the OSCC cell culture at the indicated dose 60 min before EGF treatment (50 ng/mL). PD-L1 expression and the phosphorylation of EGFR downstream proteins were evaluated after 24 h and 60 min, respectively. (E) Omipalisib, AZD1480, and GSK1120212 were added into the OSCC cell culture at the indicated dose 60 min before EGF treatment (50 ng/mL). PD-L1 expression and phosphorylation of EGFR downstream proteins were evaluated after 24 h and 60 min, respectively. Statistical analysis was performed for the addition of EGF alone. * p < 0.05, ** p < 0.01. PD-L1 expression was tested using flow cytometry. Statistical analysis was performed on the results of IFN-γ 0 ng/mL. (B) Cells were treated with IFN-γ (25 ng/mL) each time, and the phosphorylation of downstream proteins was evaluated by Western blotting. (C) AZD1480 was added into the OSCC cell culture at the indicated dose 60 min before IFN-γ treatment (25 ng/mL). PD-L1 expression and the phosphorylation of EGFR downstream proteins were evaluated after 24 h and 60 min, respectively. (D) A graphic summary of in vitro assays. * p < 0.05, ** p < 0.01.

Upregulation of PD-L1 Expression in OSCC Cell Lines by IFN-γ through the JAK2/STAT1 Pathway
The PD-L1 expression levels in HSC3, HSC4, and SAS increased with IFN-γ in a dosedependent manner (Figure 2A), and the phosphorylation of IFN-γ downstream molecules, STAT1, was induced at 10 to 120 min after the stimulation with IFN-γ; however, the EGFR downstream molecules AKT and ERK were not phosphorylated ( Figure 2B). PD-L1 induction and STAT1 phosphorylation were inhibited by JAK2i (AZD1480) ( Figure 2C), which suggests that IFN-γ upregulates PD-L1 expression in OSCC cell lines through the JAK2/STAT1 pathway ( Figure 2D).

MF-IHC Staining of HSC3 Cells and OSCC Tissues
We detected PD-L1 expression and phosphorylation of the downstream molecules, i.e., EGFR, AKT, and STAT1, at 24 h after stimulation with EGF or IFN-γ by MF-IHC using HSC3 paraffin-embedded sections ( Figure 3). The phosphorylation pattern was compared with the results of the Western blot. PD-L1 upregulation was observed after the stimulation of EGF or IFN-γ. Only STAT1 phosphorylation was observed in the stimulation with IFN-γ. In contrast, phosphorylation of all the investigated molecules, i.e., EGFR, AKT, and STAT1, was observed in the stimulation of EGF. These results on phosphorylation patterns were consistent with those of the Western blot, indicating that the analysis of the phosphorylation pattern by MF-IHC can deduce the in situ pathway. Furthermore, the EGF pathway or the IFN-γ pathway was used for PD-L1 expression.

Upregulation of PD-L1 Expression in OSCC Cell Lines by IFN-γ through the JAK2/STAT1 Pathway
The PD-L1 expression levels in HSC3, HSC4, and SAS increased with IFN-γ in a dosedependent manner (Figure 2A), and the phosphorylation of IFN-γ downstream molecules, STAT1, was induced at 10 to 120 min after the stimulation with IFN-γ; however, the EGFR downstream molecules AKT and ERK were not phosphorylated ( Figure 2B). PD-L1 induction and STAT1 phosphorylation were inhibited by JAK2i (AZD1480) ( Figure 2C), which suggests that IFN-γ upregulates PD-L1 expression in OSCC cell lines through the JAK2/STAT1 pathway ( Figure 2D).

MF-IHC Staining of HSC3 Cells and OSCC Tissues
We detected PD-L1 expression and phosphorylation of the downstream molecules, i.e., EGFR, AKT, and STAT1, at 24 h after stimulation with EGF or IFN-γ by MF-IHC using HSC3 paraffin-embedded sections ( Figure 3). The phosphorylation pattern was compared with the results of the Western blot. PD-L1 upregulation was observed after the stimulation of EGF or IFN-γ. Only STAT1 phosphorylation was observed in the stimulation with IFN-γ. In contrast, phosphorylation of all the investigated molecules, i.e., EGFR, AKT, and STAT1, was observed in the stimulation of EGF. These results on phosphorylation patterns were consistent with those of the Western blot, indicating that the analysis of the phosphorylation pattern by MF-IHC can deduce the in situ pathway. Furthermore, the EGF pathway or the IFN-γ pathway was used for PD-L1 expression.

Relationship between PD-L1 Expression, Phosphorylation of EGFR and STAT1, and Lymphocyte Infiltration in OSCC Tissues
We examined the relationship between PD-L1 expression, phosphorylation of EGFR and STAT1, and lymphocyte infiltration in OSCC tissues. PD-L1 expression was observed in 23 of 50 patients (46%). The p-STAT1 score increased significantly in the PD-L1-positive group ( Figure 5A), and the p-EGFR score showed an increasing trend, although this was not significant ( Figure 5B). Moreover, the p-STAT1 score was increased in the "Inflamed" group compared to the "Desert" group ( Figure 5C). Increased PD-L1 expression was observed in the "Inflamed" group compared to that in the "Desert" group, suggesting that PD-L1 expression by IFN-γ derived from infiltrated lymphocytes ( Figure 5D).

Relationship between PD-L1 Expression, Phosphorylation of EGFR and STAT1, and Lymphocyte Infiltration in OSCC Tissues
We examined the relationship between PD-L1 expression, phosphorylation of EGFR and STAT1, and lymphocyte infiltration in OSCC tissues. PD-L1 expression was observed in 23 of 50 patients (46%). The p-STAT1 score increased significantly in the PD-L1-positive group ( Figure 5A), and the p-EGFR score showed an increasing trend, although this was not significant ( Figure 5B). Moreover, the p-STAT1 score was increased in the "Inflamed" group compared to the "Desert" group ( Figure 5C). Increased PD-L1 expression was observed in the "Inflamed" group compared to that in the "Desert" group, suggesting that PD-L1 expression by IFN-γ derived from infiltrated lymphocytes ( Figure 5D). In Figure 5E, the relationship between the p-EGFR and p-STAT1 scores, PD-L1 expression, and lymphocyte infiltration is shown. The p-EGFR and p-STAT1 scores showed In Figure 5E, the relationship between the p-EGFR and p-STAT1 scores, PD-L1 expression, and lymphocyte infiltration is shown. The p-EGFR and p-STAT1 scores showed a positive correlation, suggesting EGFR activation-induced phosphorylation of STAT1 in the OSCC tissues. When the samples were divided to four regions, the following was observed: a p-STAT1 score of 0-1 and a p-EGFR score of 0-1 (I); a p-STAT1 score of 0-1 and a p-EGFR score of 2-4 (II); a p-STAT1 score of 2-4 and a p-EGFR score of 0-1 (III); a p-STAT1 score of 2-4 and a p-EGFR score of 2-4 (IV); and PD-L1-positive (>1%) cases were observed in 20.0% of I (3/15), 0% of II (0/0), 59.1% of III (13/22), and 53.8% of IV (7/13). Most of the PD-L1-positive cases were observed in regions III and IV. The cases in region III and IV were further divided to "Inflamed" cases and "Excluded+Desert" cases. The "Inflamed" cases in region IV indicate that PD-L1 expression is affected by both EGF and IFN-γ. The "Excluded+Desert" cases in region IV indicate that PD-L1 expression is affected by EGF more than IFN-γ, and the "Inflamed" cases in region III indicate that PD-L1 expression is affected by IFN-γ more than EGF. PD-L1-positive cases were observed in all the "Inflamed" cases in region IV (5/5), in 25% of the "Excluded+Desert" cases in region IV (2/8), and in 54% of the "Inflamed" cases in region III (6/11). These results suggest that IFN-γ is more effective than EGF for PD-L1 expression, and EGF might enhance IFN-γ-induced PD-L1 expression. Referring to the TCGA HNSC tumor dataset, we found a positive correlation between IFN-γ-related genes (IFN-γ, CD8A, and STAT1) and PD-L1 (CD274), but not between EGF-related genes (EGFR, AKT1, and MAP2K7), which seems to support the results of this study ( Figure 6). a positive correlation, suggesting EGFR activation-induced phosphorylation of STAT1 in the OSCC tissues. When the samples were divided to four regions, the following was observed: a p-STAT1 score of 0-1 and a p-EGFR score of 0-1 (I); a p-STAT1 score of 0-1 and a p-EGFR score of 2-4 (II); a p-STAT1 score of 2-4 and a p-EGFR score of 0-1 (III); a p-STAT1 score of 2-4 and a p-EGFR score of 2-4 (IV); and PD-L1-positive (>1%) cases were observed in 20.0% of I (3/15), 0% of II (0/0), 59.1% of III (13/22), and 53.8% of IV (7/13). Most of the PD-L1-positive cases were observed in regions III and IV. The cases in region III and IV were further divided to "Inflamed" cases and "Excluded+Desert" cases. The "Inflamed" cases in region IV indicate that PD-L1 expression is affected by both EGF and IFN-γ. The "Excluded+Desert" cases in region IV indicate that PD-L1 expression is affected by EGF more than IFN-γ, and the "Inflamed" cases in region III indicate that PD-L1 expression is affected by IFN-γ more than EGF. PD-L1-positive cases were observed in all the "Inflamed" cases in region IV (5/5), in 25% of the "Excluded+Desert" cases in region IV (2/8), and in 54% of the "Inflamed" cases in region III (6/11). These results suggest that IFN-γ is more effective than EGF for PD-L1 expression, and EGF might enhance IFN-γinduced PD-L1 expression. Referring to the TCGA HNSC tumor dataset, we found a positive correlation between IFN-γ-related genes (IFN-γ, CD8A, and STAT1) and PD-L1 (CD274), but not between EGF-related genes (EGFR, AKT1, and MAP2K7), which seems to support the results of this study ( Figure 6).

Discussion
It is known that PD-L1 expression in cancer cells is controlled by both oncogenic pathways (OncoPath) and immunologic pathways (ImmunoPath) in OSCCs [10]. We confirmed previous findings using OSCC cell lines, i.e., HSC3, HSC4, and SAS in vitro (Figures 1 and 2). EGF is a representative factor driving the OncoPath, and IFN-γ is a representative factor driving ImmunoPath-induced PD-L1 expression via the signaling cascade of EGFR/PI3K/AKT, EGFR/STAT1, or EGFR/MEK/ERK for EGF and of JAK2/STAT1 for IFN-γ, respectively. However, the EGFR downstream cascade differed among the cell lines. The phosphorylation pattern of the downstream cascade molecules in the HSC3 cells stimulated with EGF or IFN-γ using MF-IHC analysis (Figure 3) was consistent with the results of the Western blot, suggesting that phosphorylation analysis in situ using MF-IHC is feasible.

Discussion
It is known that PD-L1 expression in cancer cells is controlled by both oncogenic pathways (OncoPath) and immunologic pathways (ImmunoPath) in OSCCs [10]. We confirmed previous findings using OSCC cell lines, i.e., HSC3, HSC4, and SAS in vitro (Figures 1 and 2). EGF is a representative factor driving the OncoPath, and IFN-γ is a representative factor driving ImmunoPath-induced PD-L1 expression via the signaling cascade of EGFR/PI3K/AKT, EGFR/STAT1, or EGFR/MEK/ERK for EGF and of JAK2/STAT1 for IFN-γ, respectively. However, the EGFR downstream cascade differed among the cell lines. The phosphorylation pattern of the downstream cascade molecules in the HSC3 cells stimulated with EGF or IFN-γ using MF-IHC analysis (Figure 3) was consistent with the results of the Western blot, suggesting that phosphorylation analysis in situ using MF-IHC is feasible.
Our findings ( Figure 5) suggest that the ImmunoPath is important for PD-L1 expression in OSCC tissues, more so than the EGF-induced OncoPath (EGF OncoPath). More than 90% of HNCs overexpress EGFR [22]; this may be related to PD-L1 expression. However, there was a difference between EGFR expression and its activation in HNCs [23,24], as the actual number of cases in which EGFR activation increases PD-L1 may be less than expected. Additionally, our findings also suggest that the PD-L1 expression mechanisms in OSCC tissues are heterogenous and can be classified into five types: both the ImmunoPath and EGF OncoPath are inactive (Type 1, Region I in Figure 5E); the ImmunoPath is dominantly active (Type 2, "Inflamed" cases of Region III in Figure 5E); the EGF OncoPath is dominantly active (Type 3, "Excluded+Desert" cases of Region IV in Figure 5E); both the ImmunoPath and EGF OncoPath are active (Type 4, "Inflamed" cases of Region IV in Figure 5E); and the remaining types (Type 5, "Excluded+Desert" cases of Region III in Figure 5E). The OncoPath excluding EGF is thought to affect PD-L1 expression in Type 1 and Type 5. For example, c-MET activation, PTEN loss, and PI3KCA mutations are candidates [25,26]. Elucidating the mechanism of PD-L1 expression in these types will be a future subject of investigation.
It is known that TME affects the response to immune checkpoint therapies, including PD-1 blockade [14,15,27]. Our classification above may contribute to predictions for PD-1 blockade therapies and the development of combination therapies using the PD-1 blockade and novel immune-checkpoint therapies. Our findings indicate that Type 2 and 4 are effective, but therapeutic effects are not expected in Type 1, 3, and 5. It is reported that EGFR-mutated NSCLC forms an immunosuppressive TME by suppressing effector T-cell infiltration and inducing Treg via the EGFR downstream molecules AKT1 and JNK [28]. The regulation of the EGF OncoPath is thought to be necessary in Type 3 for effector T-cell infiltration [29], and also in Type 4 because the EGF OncoPath might involve the enhancement of ImmunoPath-induced PD-L1 expression. For example, pembrolizumab used in combination with cetuximab has shown great efficacy in a Phase II trial of recurrent or metastatic HNCs and is considered a good candidate [30]. Neither the EGF OncoPath nor ImmunoPath are active in Type 1 and 5; therefore, other therapeutic strategies might be necessary.
There are several limitations to this study. First, this was a retrospective study with a small number of cases. In addition, since we used tissues that were as fresh as possible in consideration of antibody staining, we were not be able to examine the relationship with prognosis. A prospective study with a larger number of cases is necessary in the future. Furthermore, many mechanisms for the regulation of PD-L1 expression are already known, but we investigated the regulation of PD-L1 only at the transcriptional level focused on the EGF and IFN-γ downstream cascade, and we only investigated the phosphorylation of EGFR and STAT1 in situ. Finally, in this study, we only investigated PD-L1 expression in tumor cells. PD-L1 expression in OSCCs is found in stromal immune cells, such as macrophages and tumor cells [5]. Furthermore, in contrast to tumor cells, PD-L1 expression in macrophages has been reported to have a favorable prognosis [31,32]. Further comprehensive research is needed to better understand the expression mechanism and function of PD-L1 in both cancer cells and immune cells.

Image Analysis
The slides were scanned with Vectra at a low resolution, and a region of interest (ROI) was set at the invasive front of the tumor where lymphocytes were accumulated with Phenochart software (PerkinElmer). Then, at least three ROIs were captured with a 20× objective, and the reconstructed multispectral images were obtained and the number of marker-positive cells was counted using Inform software (PerkinElmer). For PD-L1, p-STAT1-and p-EGFR-positive tumor cells were counted as double-positive cells with cytokeratin. To analyze the IHC results, a semi-quantitative method was employed based on previous reports [34,35]. Specifically, the IHC scores for p-EGFR and p-STAT1 were evaluated on a scale of 0-4 (0, <5%; 1, 5-25%; 2, 25-50%; 3, 50-75%; 4, >75%). For PD-L1, a positivity rate of 1% or higher was considered positive.

Database Analysis
Gene expression in OSCC was analyzed using the Gene Expression Profiling Interactive Analysis (GEPIA) online database (http://gepia.cancer-pku.cn accessed on 10 December 2021) [36]. The Cancer Genome Atlas (TCGA) HNSC tumor data were used as the basis for the correlation analysis.

Statistical Analysis
The numerical data were presented as the mean (±SE: standard error) of independent experiments, and differences were examined using the Tukey's test. The Mann-Whitney U test and the Steel-Dwass test were performed to determine the significance of the IHC analysis. The correlation analysis data were determined using the Spearman's rank correlation coefficient. Statistical analyses were performed using the R Statistical Software (version 3.6.3; Foundation for Statistical Computing, Vienna, Austria). p < 0.05 was considered statistically significant.  Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data that support the findings of this study are available from the corresponding author, S.S., upon reasonable request.