High PD-L1 Expression on Tumor Cells Indicates Worse Overall Survival in Advanced Oral Squamous Cell Carcinomas of the Tongue and the Floor of the Mouth but Not in Other Oral Compartments

The markers of the tumor microenvironment (TME) are promising prognostic and predictive factors in oral squamous cell carcinoma (OSCC). The current study aims to analyze the immunohistochemical expression of programmed cell death-ligand 1 (PD-L1) and interleukin-33 (IL-33) in a cohort of 95 chemonaïve OSCCs. PD-L1 and IL-33 were assessed separately in tumor cells (TCs) and tumor-infiltrating lymphocytes (TILs). High PD-L1 expression in TILs was associated with better overall survival (OS) in univariate analysis. Tumors localized in the floor of the oral cavity and tongue tended to have a lower percentage of PD-L1-positive TCs when compared to other locations. PD-L1 expression on TCs had no prognostic significance when the whole cohort was analyzed. However, along with the T descriptor (TNM 8th), it was included in the multivariable model predicting death in carcinomas of the floor of the oral cavity and tongue (HR = 2.51, 95% CI = 1.97–5.28). In other locations, only nodal status was identified as an independent prognostic factor in multivariate analysis (HR = 0.24, 95% CI = 0.08–0.70). Expression of IL-33 had no impact on survival, but it was differently expressed in various locations. In conclusion, the prognostic significance of PD-L1 in oral cancer depends on the tumor site and type of cell expressing immune checkpoint receptor (TCs vs. TILs).


Introduction
Oral squamous cell carcinoma (OSCC) accounts for 95% of all malignancies developing in the oral cavity. It predominantly affects males over 50 years of age and has a causal relationship with tobacco smoking and alcohol consumption [1]. Despite the substantial progress which has been made in OSCC treatment, overall survival (OS) has not improved significantly in the last few decades. The treatment for OSCC is multimodal, with surgery usually being the treatment of choice. Its clinical course mainly depends on the stage of

Study Group
Medical records of 109 patients diagnosed with OSCC, treated at the University Clinical Center Medical University of Gdańsk, Poland, between 2007-2012 were analyzed. The inclusion criteria of this study included histopathologically confirmed OSCC and available treatment-naïve histopathological specimens (biopsy or resection). Patients without survival data were excluded (n = 14). Finally, ninety-five patients (n = 95) were included in the study. Basic demographic and clinicopathological data (age, gender, addictions, treatment methods, cancer localization, grading, staging, recurrence, death, follow-up, and survival rate) were collected. The staging in all cases was determined within one month after the first presentation, and adjusted according to the American Joint Committee on Cancer (AJCC) 8th edition of the TNM classification for the sake of the current study. Patients' data were fully anonymized. The local Bioethical Committee of the Medical University of Gdańsk approved the protocol of the study (approval No NKBBN/59/2016).

Specimen Preparation and Immunohistochemistry
Formalin-fixed paraffin-embedded (FFPE) tissue blocks were collected from tumor resection or, in the case of patients treated with neoadjuvant radiation or chemotherapy, from treatment-naïve biopsy (if applied) after the first presentation. Tissue microarrays (TMA) were prepared with a Manual Tissue Arrayer MTA 1 (Beecher Instruments Inc., Sun Prairie, WI, USA). Two representative cores, both of 0.4 cm diameter, were obtained from each case.
The proportion of positive cells was established by calculating the number of stained tumor cells (TCs) and tumor-infiltrating lymphocytes (TILs) divided by the total number of each type of cells. Two pathologists (AK and RP) experienced in PD-L1 expression evaluation assessed the stainings. When the interpretations differed, the pathologists made decisions by consensus. Only membranous PD-L1 expression was considered positive in TCs, whereas cytoplasmic and/or membranous reaction were considered positive in TILs. Only nuclear IL-33 expression was considered positive. Histologically normal tonsil was used for the positive control. For each patient, the results from two cores were used. The percentage of positively staining cells was estimated in each core and an average score was utilized in the further analyzes. The cut-off for high PD-L1 expression was established depending on the median of the percentage of positively staining cells as >10% in TCs and >20% in TILs ( Figure 1A-D). IL33 was divided into two groups-no expression and positive expression (Figure 2A-D). Subsequently, we compared the agreement between two cores in the binary classification of PD-L1 and IL-33 by Cohen's kappa coefficient to assess the heterogeneity of markers expression.

Statistical Analysis
Statistical analysis was performed using the STATISTICA 13.3 (TIBCO, Palo Alto, CA, USA; licensed to the Medical University of Gdańsk) and R statistical environment [26]. Kaplan-Meier curves were plotted using the "survminer" and "ggsci" packages [27,28]. The associations between analyzed markers and clinicopathological characteristics were assessed by the Mann-Whitney U test for continuous variables. Categorical variables were compared by the chi-square test and Fisher's exact test when applicable. Cohen's kappa coefficient was calculated to assess the reproducibility across the two cores incorporated in the TMA. Kaplan-Meier curves were plotted to assess overall survival (OS) and compared by log-rank test. Hazard ratios were estimated with the Cox proportional hazard regression. The backward selection was employed to create a multivariable model predicting death and to eliminate non-significant variables at p ≤ 0.05. All tests were considered statistically significant as p ≤ 0.05.

Statistical Analysis
Statistical analysis was performed using the STATISTICA 13.3 (TIBCO, Palo Alto, CA, USA; licensed to the Medical University of Gdańsk) and R statistical environment [26]. Kaplan-Meier curves were plotted using the "survminer" and "ggsci" packages [27,28]. The associations between analyzed markers and clinicopathological characteristics were assessed by the Mann-Whitney U test for continuous variables. Categorical variables were compared by the chi-square test and Fisher's exact test when applicable. Cohen's kappa coefficient was calculated to assess the reproducibility across the two cores incorporated in the TMA. Kaplan-Meier curves were plotted to assess

PD-L1 Expression
Forty-four (46.31%) cases demonstrated positive expression of PD-L1 in > 10% of TCs. The mean percentage of positive cells was 21.88%, median 10%. Tumors localized in the floor of the oral cavity and tongue tended to have a lower percentage of PD-L1-positive TCs when compared to other locations (p = 0.019, Mann-Whitney U). There was a trend toward lower PD-L1 expression and the presence of nodal metastases (p = 0.015, Mann-Whitney U). Analogous findings were noted if PD-L1 was assessed as a binary variable (low/high expression). There was no association with gender, T stage, grade, history of smoking, or alcohol abuse. The summary of clinicopathological features with relation to analyzed biomarkers is presented in Table 1. Thirty-one (31.63%) cases displayed PD-L1 positivity in >20% of TILs (high expression). High PD-L1 expression on TILs was associated with the absence of lymph node metastases and lower stage. No association was found between the expression of PD-L1 in TILs and other analyzed clinicopathological variables.
Positive expression of IL-33 in TILs was observed in 18 cases (18.94%). Mean percentage was 0.5% (median 0%, maximum 7%). Expression of IL-33 in TILs was less common in cancers of the tongue and the floor of the oral cavity (p = 0.055, chi-square), but no other association between IL-33 and clinicopathological variables was found (Supplementary Table S1).

Whole Cohort
The mean follow-up was 3.83 years, the minimum was 24 days, and the maximum was 10.87 years. The 5-year OS was 36.65%. The univariate Cox's proportional hazard analysis ( Table 2) demonstrated the association between survival and grade, stage, nodal metastases, and PD-L1 expression on TILs. Expression of IL-33 and PD-L1 in TCs had no impact on survival ( Figure 3A for PD-L1, Supplementary Figures S1-S3 for IL-33). High PD-L1 expression in TILs was associated with better OS (HR = 0.475, 95% CI = 0.281-0.805; Figure 3B), but it was not retained in the multivariate Cox regression model predicting outcomes. Only the presence of nodal metastases was incorporated in the multivariable model predicting death.

Prognostic Significance of PD-L1 and IL-33 Expression in Various Locations
Due to the significant differences in PD-L1 expression in cancers of the tongue and floor of the oral cavity, we decided to create separate multivariable Cox's regression models that predict outcomes and take into consideration cancer location.
In the group of cancers of the tongue and the floor of the oral cavity, two variables were incorporated in the final model: T category and PD-L1 expression on TCs (Table 3). Interestingly, in univariate analysis, PD-L1 expression on TCs and TILs had a statistically borderline impact on survival ( Figure 4). Especially poor outcomes were observed in the group of T3-4 tumors highly expressing PD-L1 on TCs ( Figure 5).
In cancers located in other parts of the oral cavity, only the presence of nodal metastases was incorporated in the multivariable model. Importantly, in univariate analysis, PD-L1 expression on TILs was associated with better outcomes (Figure 6).

Prognostic Significance of PD-L1 and IL-33 Expression in Various Locations
Due to the significant differences in PD-L1 expression in cancers of the tongue and floor of the oral cavity, we decided to create separate multivariable Cox's regression models that predict outcomes and take into consideration cancer location.
In the group of cancers of the tongue and the floor of the oral cavity, two variables were incorporated in the final model: T category and PD-L1 expression on TCs (Table 3). Interestingly, in univariate analysis, PD-L1 expression on TCs and TILs had a statistically borderline impact on survival ( Figure 4). Especially poor outcomes were observed in the group of T3-4 tumors highly expressing PD-L1 on TCs ( Figure 5).       In cancers located in other parts of the oral cavity, only the presence of nodal metastases was incorporated in the multivariable model. Importantly, in univariate analysis, PD-L1 expression on TILs was associated with better outcomes (Figure 6).

Discussion
The oral cavity is in constant contact with the external environment. There are numerous reactions here that are designed to protect the body against harmful factors. The importance of the immune tumor microenvironment and tumor immunology in the prognosis of patients with OSCCs and other head and neck malignancies is becoming increasingly recognized [2,3,29,30]. This is the first study to co-analyze PD-L1 and IL-33 protein expression on TCs and TILs in OSCC. We demonstrated that the level of PD-L1 expression on TCs varies depending on its location-cancers of the tongue and the floor of the oral cavity show lower expression than cancers of other parts. The expression of PD-L1 on TCs was not related to OS in the entire cohort. However, the expression of PD-L1 on TCs appears to have opposite effects in cancers of different locations, which "canceled out" when analyzing the entire cohort. Higher PD-L1 expression on TCs in the carcinomas of the tongue/floor of the oral cavity was associated with a worse OS, especially in cancers of higher T category in TNM. In other locations, higher PD-L1 expression on TCs was associated with a better prognosis (a trend, but not statistically significant). On the other hand, higher PD-L1 expression on TILs was associated with a lower frequency of nodal metastases. Moreover, it was associated with longer OS in the univariate analysis, but this effect was not maintained in the multivariate analysis. Most likely, this is due to the very strong correlation of PD-L1 on TILs and nodal metastases. The latter was identified as the single most important prognostic factor in the whole cohort. These findings are supported by studies on large populations, which demonstrated that the number of metastatic lymph nodes and its characteristics (e.g., the presence of extranodal extension) are critical predictors of survival in OSCC [31,32].
In other works concerning OSCC, PD-L1 expression on TCs was observed with variable frequency ranging from approximately 10 to 90% of cases . The studies differed in the size of the study group, selection of patients, antibody clones used, and the way of assessing PD-L1 expression; the two latter factors in particular significantly influence the final results of the study. In the Supplementary Table S2, we briefly present

Discussion
The oral cavity is in constant contact with the external environment. There are numerous reactions here that are designed to protect the body against harmful factors. The importance of the immune tumor microenvironment and tumor immunology in the prognosis of patients with OSCCs and other head and neck malignancies is becoming increasingly recognized [2,3,29,30]. This is the first study to co-analyze PD-L1 and IL-33 protein expression on TCs and TILs in OSCC. We demonstrated that the level of PD-L1 expression on TCs varies depending on its location-cancers of the tongue and the floor of the oral cavity show lower expression than cancers of other parts. The expression of PD-L1 on TCs was not related to OS in the entire cohort. However, the expression of PD-L1 on TCs appears to have opposite effects in cancers of different locations, which "canceled out" when analyzing the entire cohort. Higher PD-L1 expression on TCs in the carcinomas of the tongue/floor of the oral cavity was associated with a worse OS, especially in cancers of higher T category in TNM. In other locations, higher PD-L1 expression on TCs was associated with a better prognosis (a trend, but not statistically significant). On the other hand, higher PD-L1 expression on TILs was associated with a lower frequency of nodal metastases. Moreover, it was associated with longer OS in the univariate analysis, but this effect was not maintained in the multivariate analysis. Most likely, this is due to the very strong correlation of PD-L1 on TILs and nodal metastases. The latter was identified as the single most important prognostic factor in the whole cohort. These findings are supported by studies on large populations, which demonstrated that the number of metastatic lymph nodes and its characteristics (e.g., the presence of extranodal extension) are critical predictors of survival in OSCC [31,32].
In other works concerning OSCC, PD-L1 expression on TCs was observed with variable frequency ranging from approximately 10 to 90% of cases . The studies differed in the size of the study group, selection of patients, antibody clones used, and the way of assessing PD-L1 expression; the two latter factors in particular significantly influence the final results of the study. In the Supplementary Table S2, we briefly present the previous studies analyzing PD-L1 expression in OSCC with the emphasis on methodology and prognostic effects.
Most studies to date, similarly to ours, are based on the heterogeneous cohorts of tumors located throughout the oral cavity, with a few exceptions focused on certain oral compartments, e.g., squamous cell carcinoma of the tongue. Multiple studies assessing survival in OSCCs demonstrated that high PD-L1 expression on TCs is associated with worse outcomes. Strati et al. [67] demonstrated that PD-L1 overexpression on circulating TCs was associated with inferior progression-free survival and OS in head and neck cancers. Nevertheless, a few studies showed contrary results or no association between PD-L1 expression and survival. The results of our study suggest that these discrepancies may originate from skewed distribution of cancer location in analyzed cohorts, as PD-L1 expression on TCs seems to have different prognostic effects in various compartments of the oral cavity. The site-dependent differences in the TME composition of head and neck carcinomas were previously reported by Green et al. [68], who observed a higher prevalence of TILs in oropharyngeal cancers compared to other locations. Interestingly, PD-L1 is physiologically expressed on the masticatory mucosa of the oral cavity, whereas other epithelia do not constitutively express PD-L1 [69]. Additionally, oral compartments represent ecological niches inhabited by a variety of microbiota modulating the local immune microenvironment [70,71]. For instance, Porphyromonas gingivalis induces the expression of PD-L1 in OSCC cells in vitro [53]. Finally, the immune landscape of HPVand carcinogen-driven head and neck carcinomas differ in some aspects [72]. All these factors may contribute to distinct immune characteristics of OSCCs of various sites.
However, the prognostic significance of PD-L1 expression in TCs may also depend on other factors, especially the TME context in certain parts of the oral cavity [73]. Takahashi et al. [42] demonstrated that patients with high PD-L1 expression and abundant CD4+ T-cells have better outcomes than those with low CD4+ T-cell infiltration. Other research showed that high infiltration by CD4+ and CD8+ and high CD8+/FOXP3+ ratio lymphocytes were associated with positive expression of PD-L1 on TCs [33]. Other factors which influence PD-L1 expression in OSCC include gender, since some studies showed more PD-L1-negative tumors in males than in females [37,39]. Hanna et al. [44] reported that PD-L1 expression was associated with better outcomes in young females with OSCC. In another study, PD-L1 expression on TCs was more common in non-smokers and non-drinkers [52].
Even more perplexing is the role of PD-L1 expression on immune cells in OSCC. Previous translational research demonstrated that PD-L1-positive macrophages induce anergy in CD4+ and CD8+ T-cells in OSCC TME [56,57]. However, the current study demonstrates the more favorable prognosis of OSCC infiltrated by the high number of PD-L1-positive TILs. This effect was strongly correlated with the absence of nodal metastases. These results are consistent with the study by Kim et al. [10] which analyzed a large cohort of head and neck squamous cell carcinomas (including 204/402 oral cancers), and reported that PD-L1 expression on TILs, but not on TCs, was a favorable prognostic factor. Similar associations were noted in laryngeal cancer [74]. Better outcomes and lack of nodal metastases in tumors rich in PD-L1 positive TILs in OSCCs may be related to preexistent anti-tumor adaptive immune response [75].
In our study, IL-33 was rarely expressed in OSCC and had no significant impact on patients' survival. In the only study so far analyzing IL-33 in oral cancer, Ishikawa et al. [24] evaluated IL-33 expression in squamous cell carcinoma of the tongue [24]. The authors showed that high IL-33 expression in tumor cells was associated with local and nodal recurrence. They used a different antibody clone (IL-33, MBS150331, rabbit polyclonal antibody, Medical & Biological Laboratories, Nagoya, Japan) and immunostaining method analysis, which may explain the discrepancies [24]. IL-33 is able to increase PD-1 and PD-L1 expression at the surface of CD8+ T lymphocytes and cancer cells, respectively [22]. The process is most likely driven via enhanced T cell production of IFN-γ. However, we did not find any association between the expression of PD-L1 and IL-33 in our cohort.
Unfortunately, our study has several limitations. The cohort size is suboptimal and the number of cases representing various compartments is low. Thus, it is impossible to draw definite conclusions regarding the survival analysis.

Conclusions
PD-L1 was commonly expressed in OSCC by TCs and TILs in our cohort. However, OSCC immunobiology and prognostic significance of PD-L1 expression vary depending on tumor location. High PD-L1 expression on TILs was strongly correlated with the lack of nodal metastases and, thus, better OS in all locations. On the other hand, PD-L1 expression on TCs seems to have a distinct impact on survival in cancers of the tongue and floor of the oral cavity and other locations. Our findings have the potential to be applied in clinical practice for better post-operative OSCC monitoring. Nevertheless, it is unknown if this factor may influence the response to immune checkpoint blockade. IL-33 expression was rarely observed and had no prognostic significance but its expression on TCs was significantly associated with tumor location. It supports site-specific variations in TME of oral cancer. Thus, future research on the immune landscape of OSCC and its responsiveness to immune therapy should focus on the analysis of cancers from distinct compartments of the oral cavity.
Supplementary Materials: Supplementary materials can be found at https://www.mdpi.com/ article/10.3390/biomedicines9091132/s1, Figure S1: Overall survival probability curves according to IL-33 expression on TILs (A) and TCs (B) in the whole cohort, Figure S2: Overall survival probability curves according to IL-33 expression on TILs (A) and TCs (B) in cancers of the tongue/floor of the oral cavity, Figure S3: Overall survival probability curves according to IL-33 expression on TILs (A) and TCs (B) in cancers of other oral compartments, Table S1: The summary of clinicopathological features with relation to IL-33 expression on TCs and TILs, Table S2: The summary of studies investigating immunohistochemical expression of PD-L1 in OSCC.  Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.

Conflicts of Interest:
The authors declare no conflict of interest.