PD-L1 Expression in Tumor Cells Is an Independent Unfavorable Prognostic Factor in Oral Squamous Cell Carcinoma

The programmed-death 1 receptor (PD-1)/programmed-death ligand 1 (PD-L1, also called B7-H1, and CD274) pathway regulates T-cell responses and the balance between T-cell activation and tolerance, and acts as an immunologic checkpoint (1). In normal tissues, PD-L1 is expressed on T cells, B cells, dendritic cells, macrophages, mesenchymal stem cells, and mast cells, as well as several hematopoietic cells (2, 3). PD-L1 is also expressed in multiple types of cancers (4), including head and neck squamous cell carcinomas (HNSCC) and, among them, in oral squamous cell carcinomas (OSCC; refs. 3, 5, 6). In the tumor microenvironment, PD-L1 expressed in tumor cells binds to the inhibitory receptor PD-1, a member of the B7 family of receptors (7), on activated T cells reaching the tumor [tumor-infiltrating lymphocytes (TIL)], thereby delivering an inhibitory signal to those T cells that ultimately prevents tumor elimination from the immune system (3, 8, 9).

OSCC is the most common type of cancer in the head and neck area. It belongs to the ten most common cancers, and approximately 300,000 new cases are diagnosed annually worldwide (10). OSCC is associated with poor prognosis and higher morbidity when diagnosed at advanced stages (11), and it has been considered to be a very immunosuppressive cancer. Immune tolerance is mediated by inhibitory signaling pathways called immune checkpoints. One of the most important immunologic checkpoints is the axis PD-1/PD-L1 that has been proposed as a potential mechanism facilitating the immunosuppressive phenotype in OSCC (5, 6). Consequently, at least from a theoretical point of view, blockade of the PD-1/PD-L1 pathway could effectively reduce tumor growth and improve survival (6). Positive tumor PD-L1 expression determined by immunohistochemistry (IHC) is thought to be predictive of clinical response (12), and thus, the PD-1/PD-L1 checkpoint inhibitors, such as nivolumab and pembrolizumab have demonstrated clinical efficacy in the treatment of advanced HNSCC (11, 13–16). The expression of PD-L1 is up-regulated in many cancers, such as lung, ovary, colon, skin, brain, kidney, esophagus, stomach and breast cancers (16). Moreover, PD-L1 expression has been associated with poor prognosis in many malignancies, such as nasopharynx, kidney, esophagus, stomach, pancreas, breast and salivary gland carcinomas, as well as malignant melanoma (17–24). Nevertheless, only few studies have to date assessed the clinical significance of PD-1/PD-L1 in OSCC. Hence, comprehensive data concerning PD-1 and PD-L1 expression in OSCC are still lacking, and consequently, the clinicopathologic implications and prognostic relevance need clarification (25, 26).

Therefore, the purpose of the present study was to investigate the expression of PD-1 and PD-L1 in OSCC and to establish correlations between PD-1/PD-L1 expression and clinicopathologic variables, and their impact on patients’ survival.

A retrospective study was designed. Surgical tissue specimens from 125 patients with OSCC who underwent surgical treatment with curative purposes at the Hospital Universitario Central de Asturias between 1996 and 2007 were retrospectively collected, in accordance to approved institutional review board guidelines. All experimental procedures were conducted in accordance to the Declaration of Helsinki and approved by Institutional Ethics Committee of the Hospital Universitario Central de Asturias and by the Regional CEIC from Principado de Asturias (date of approval July 18, 2013; approval number: 81/2013) for the project PI13/00259. Written informed consent was obtained from all the patients. Clinicopathologic data were collected from medical records. Tissue specimens were obtained from the Biobanco del Principado de Asturias, and representative tissue sections were obtained from archival, formalin-fixed paraffin-embedded blocks and the histologic diagnosis was confirmed by an experienced pathologist (VB-L).

Three morphologic representative areas were selected from each individual paraffin tumor block, including both the invasive border and the center of tumor sheets or islands without necrotic areas. In addition, each tissue microarray (TMA) contained morphologically normal oral mucosa samples from non-oncological patients undergoing oral surgery as internal negative controls. To check the histopathologic diagnosis and the adequacy of tissue sampling, a section from each microarray was stained with hematoxylin and eosin and examined by light microscopy.

The TMAs were cut into 3-μm sections and dried on Flex IHC microscope slides (DakoCytomation). The sections were deparaffinized with standard xylene and hydrated through graded alcohols into water. Antigen retrieval was performed by heating the sections with Envision Flex Target Retrieval solution, high pH (Dako). Staining was done at room temperature on an automatic staining workstation (Dako Autostainer Plus, Dako). The following primary antibodies were used: mouse monoclonal PD-1 antibody (clone EH33, 1:100 dilution; Cell Signaling Technology #43248), rabbit monoclonal PD-L1 antibody (clone E1L3N, 1:200 dilution; Cell Signaling Technology #13684), and mouse monoclonal PD-L1 antibody (clone 22C3, 1:200 dilution; PD-L1 IHC 22C3 pharmDx; Dako SK006) by using the Dako EnVision Flex +Visualization System (Dako Autostainer). Counterstaining with hematoxylin was the final step.

PD-1 and PD-L1 immunostainings were manually evaluated by three independent observers (VB-L, JPR, and JMGP), with a high level of interobserver concordance (>95%). PD-1 staining was scored as negative or positive (at least 1% of stained cells). PD-L1 immunostaining was evaluated and initially scored into five different categories according to the percentage of stained tumor cells (0, negative; 1, 1–10%; 2, 10–25%; 3, 26–50%; and 4, more than 50% of stained cells). Positive PD-L1 expression was considered if at least 1% of tumor cells stained positively. Because PD-L1 expression in more than 10% of tumor cells was associated with poorer survival, this was established as a cut-off point for subsequent analyses. Accordingly, for analytical purposes, patients with OSCC were dichotomized into two groups: relevant versus nonrelevant PD-L1 expression, based on the established cut-off point of 10% of stained tumor cells.

χ² and Fisher exact tests were used for comparison between categorical variables. Disease-specific survival (DSS) was determined for the date of treatment completion to death for the tumor. For time-to-event analysis, survival curves were plotted using the Kaplan–Meier method. Differences between survival times were analyzed by the log-rank test. HR with their 95% confidence intervals (CI) for clinicopathologic variables were calculated using the Cox proportional hazards model, both as univariate and multivariate analyses. All tests were two-sided and P values less than 0.05 were considered statistically significant. All statistical analyses were performed using SPSS version 21 (IBM Co.).

The clinical and histopathologic characteristics of the cohort of 125 patients with OSCC under study are shown in Table 1. This cohort was composed of 82 men and 43 women, ranging from 28 to 91 years, with a median age of 57 years, and almost 33% of them were never-smokers. Most of the tumors were well-differentiated, more than 50% of cases were in advanced clinical stages (III or IV), and the most common site of tumor origin was the tongue (41%) followed by the floor of the mouth (30%). Adjuvant radiotherapy was administered to 75 patients (60%), and adjuvant chemotherapy was administered to 14 patients (11.2%). Over a median follow-up of 61 months (range, 1–230 months), 53 deaths occurred.

Positive PD-1 expression was only detected in four (3.2%) of 125 OSCC samples (Supplementary Fig. S1A and S1D). Positive tumor PD-L1 expression (>1% stained cells) was found in 30 (24%) cases when staining was performed using the anti-PD-L1 clone E1L3N (Supplementary Fig. S1B and E), and in 44 (35%) cases with anti-PD-L1 clone 22C3 (Supplementary Fig. S1C and F). There was a strong significant correlation between PD-1 expression and PD-L1 immunostaining using both clone E1L3N (P = 0.003) and clone 22C3 (P = 0.01).

There was no any significant relationship between the PD-1 receptor expression and the various clinicopathologic variables studied (Table 2). Similarly, PD-1 expression had no impact on prognosis (P = 0.63).

Because PD-L1 expression in more than 10% of tumor cells was associated with poorer survival, this was established as a cut-off point and considered relevant PD-L1 expression for subsequent analyses.

Thus, relevant PD-L1 expression was found in 12 (9.7%) OSCC samples with clone E1L3N, and in 18 (14.4%) cases with clone 22C3. Relevant PD-L1 expression either detected with clone E1L3N or clone 22C3 was significantly and inversely associated with the habits of tobacco or alcohol consumption (Table 3), thus being more frequently detected in cancers developed in nonsmokers or nonalcohol consumers. In addition, relevant PD-L1 expression by either clone E1L3N or 22C3 was also consistently associated with tumors that showed recurrence during their evolution (P = 0.03 and 0.08, respectively).

Over a median follow-up period of 61 months, 11 patients showing relevant PD-L1 expression (>10% of tumor cells) with clone 22C3 (21%) and seven cases (13%) with anti-PD-L1 clone E1L3N died due to the index cancer. The Kaplan–Meier analysis showed a significant difference in DSS between the patients distributed according to the relevant PD-L1 expression (>10% tumor cells), although differences only reached statistical significance when the analysis was done with anti-PD-L1 clone 22C3 (P = 0.03; Fig. 1) but not with clone E1L3N (P = 0.104). In this analysis, T and N stages were also significantly associated with DSS (Table 4). A univariate Kaplan–Meier analysis was performed stratifying the clinicopathologic variables according to PD-L1 expression by anti-PD-L1 clone 22C3 (Table 5). T1 tumors, poorly differentiated tumors, and tumors located in the palate or buccal mucosa did not have censored cases in survival analysis when PD-L1 was expressed in more than 10% of tumor cells. Additionally, in tumors located in the retromolar region only one out of six tumors expressed PD-L1 in more than 10% of cells. Consequently mean survival times, 95% confidence intervals, and P values were not calculated in these cases. This analysis showed that male sex (P < 0.0005), alcohol consumption (P = 0.01), tongue subsite (P = 0.01), moderate differentiation grade (P = 0.002), and positive lymph node infiltration (P = 0.04) were associated with increased risk of death in patients with OSCC when PD-L1 expression was higher than 10% of tumor cells. Subsequently, a multivariate analysis performed using the Cox regression model showed that neck node metastasis (HR 2.304; 95% CI, 1.23–4.317; P = 0.009) and PD-L1 expression higher than 10% of tumor cells (HR 2.571; 95% CI, 1.170–5.647; P = 0.01) remained as significant independent factors for poor prognosis in this cohort of patients with OSCC (Table 6).

The aim of this study was to analyze the expression of PD-1 and PD-L1 using a large homogeneous series of 125 resected OSCC, and to ascertain the clinical and prognostic relevance. Oral cancer accounts for 1% to 5% of human malignancies (27), and its overall mortality remains approximately of 50%. There is a considerable interest in discovering novel prognostic biomarkers that may help to improve the limited predictive value of current clinicopathologic markers, ultimately allowing a more adequate risk stratification and treatment selection (28). HNSCC has been defined as one of the most highly immune-infiltrated cancer types, and in fact, the most highly NK-cell and Treg-infiltrated cancer type, although immune infiltration varies depending on several clinical and genetic features, such as HPV infection, tumor location within the head and neck region, molecular subtype, mutational smoking signature, and genomic instability (29). PD-1 is an immunosuppressive receptor expressed on activated T cells, B cells, monocytes, and macrophages (2, 8). PD-L1 is a cell-surface glycoprotein primarily expressed by antigen-presenting cells and tumor cells that induces T-cell anergy and apoptosis by engaging its PD-1 receptor (2). Yu and colleagues (30) undertook a gene expression meta-analysis and verified that the mRNA expression and gene copy number of CD274 and CD279, respectively encoding PD-L1 and PD-1, were significantly increased in HNSCC. In this study, tumor PD-L1 expression was evaluated by IHC using two different monoclonal antibodies (clones E1L3N and 22C3). Relevant PD-L1 expression in more than 10% of tumor cells consistently predicted a poor clinical outcome in terms of DSS and was detected in 10% to 15% OSCC samples depending on the anti-PD-L1 antibody. Binding of PD-L1 to its receptor PD-1 inhibits the proliferation of activated T cells, leading to apoptosis or downregulation of cytotoxic T lymphocytes (CTL), called “T-cell exhaustion” (7), which results in an escape of tumor cells from T-cell–mediated immune surveillance (2). It is assumed that tumor cells upregulate PD-L1 expression to evade the host immune reaction, thereby increasing their survival rate (31). Tumor PD-L1 expression is induced intrinsecally by oncogenic signaling pathways, like the MAPK signaling pathways, and extrinsecally by factors from the tumor microenvironment, such as hypoxia, through the induction of the hypoxia-associated transcription factor HIF1α and cytokine production (32). High expression of PD-L1 has been associated with EBV infection in various EBV-associated malignancies (33). In addition, the PD-1/PD-L1 axis has also been recently associated to HPV infection in HNSCC (34, 35). This possible relationship could not been assessed in the herein studied cohort, because all the cases selected were HPV-negative OSCC.

A total of 125 patients with OSCC were enrolled with a median follow-up of 61 months. To the best of our knowledge, this is one of the largest series investigating the expression of PD-1 and PD-L1 in OSCC. More importantly, this study reveals an inverse association of PD-L1 expression with alcohol and tobacco consumption. This is a noticeable finding because tobacco and alcohol abuse is responsible for 72% of HNSCC (36), which suggests a potential crosstalk mechanism between these carcinogens and the immune scape in oral cancers. Various studies have investigated potential correlations of tumor PD-L1 expression with the clinicopathologic features and survival; however, evidences for a prognostic role of PD-L1 across a wide range of tumor types are inconsistent and ambiguous. Thus, although Maruse and colleagues (37) reported a significant relationship between PD-L1 expression and cervical lymph node metastasis and distant metastasis, Kim and colleagues (38) and Cho and colleagues (39) did not find any prognostic implications. Moreover, Oliveira-Costa and colleagues (28) observed worse prognosis in HNSCC with decreased PD-L1 expression, whereas Lin and colleagues (26) and Straub and colleagues (40) concur about a worse prognosis in OSCC with marked PD-L1 expression.

Using two different anti-PD-L1 antibodies, we consistently found that relevant PD-L1 expression (>10% of tumor cells) was associated with tumor recurrence and lower DSS. In addition, multivariate analysis further revealed that tumor PD-L1 expression was a significant independent prognostic factor in OSCC, independently of other prognostic factors such as tumor stage and neck lymph node metastasis. Specifically, PD-L1 expression (>10% tumor cells) could serve as an independent prognostic marker in male patients, alcohol consumers, patients with tumors arisen in the tongue, moderate or poorly differentiated tumors, and tumors with neck node metastasis. It remains unclear whether sex plays a role in PD-L1 expression of OSCC. Lin and colleagues (26) reported that higher PD-L1 expression was associated with a poor prognosis in males, which is in good agreement to our observations. The inconsistent results regarding the prognostic significance of PD-L1 expression have led to contradictory results in solid tumors, including HNSCC. Thus, Upko and colleagues (41) found that PD-L1 expression in oropharyngeal carcinomas was associated neither with overall survival nor DSS, which reinforces the idea that there are many differences in clinical and biological characteristics between OSCC and other tumor subsites in HNSCC. Furthermore, a number of antibodies to the extracelular and intracellular domains have been developed for IHC evaluation (3), although variable target specificity and the use of different scoring methods could also have contributed to discordance in published data on PD-L1 expression.

The prognostic relevance of PD-L1 was confirmed by multivariate analysis in this study, therefore suggesting that patients with OSCC with high PD-L1 expression might require PD-L1–targeted immunotherapy to improve prognosis and clinical outcome. Thus, even though the mechanisms leading to PD-L1 overexpression are not yet fully understood (36), clinical data from treatment with PD-1/PD-L1 inhibitors have shown response rates ranging from 10% to 50% in HNSCC and other tumor types (42). In a recent meta-analysis, Gandini and colleagues (43) found that the IHC evaluation of tumor PD-L1 expression correlated with clinical response to PD-1/PD-L1 immunotherapy in various cancers, and Topalian and colleagues (44) reported that response to anti-PD-1/PD-L1 targeted therapy was only observed in PD-L1–positive tumors. As stated before, PD-L1 can induce an exhaustion state in T cells, and tumor PD-L1 expression might reflect the presence of ineffective antitumor immune pressure mediated by tumor-infiltrating lymphocytes (28). According to our data, IHC could be a good method for PD-L1 expression assessment in tumor cells, but probably inadequate for detection in immune cells (45). Furthermore, there is a large proportion of cases with PD-L1-positive tumor cells by IHC that do not respond to checkpoint inhibitor therapy, and, on the contrary, there are tumors that do not express PD-L1 on the cell surface and respond to PD-1/PD-L1–targeting antibodies (32, 46, 47). Of additional interest are preclinical data that support a combination of PD-L1 inhibition and radiotherapy as an effective treatment for OSCC (48).

We are aware of some limitations in our study. Tissue microarrays do not represent the whole tumor, and the use of different antibodies for PD-L1 detection together with heterogeneous definition and different cut-offs to define PD-L1–positive expression, as well as different anatomic tumor locations complicate comparisons among studies. Additionally, the underlying mechanism by which sex and alcohol intake history could contribute to the prognostic relevance of PD-L1 is not clear, and finally the study design was retrospective in nature which limits disclosed correlations between the parameters studied.

The cut-off of more than 10% was selected in this study because it was associated with a statistically significant poorer DSS. As an immunosuppressive molecule promoting T-cell exhaustion, increased expression of PD-L1 should lead to a worse prognosis, as documented in several cancers (49–52), and as it was likewise observed in this study. Contrary to this, an association between tumor PD-L1 expression and improved prognosis has also been reported in other cancers (53–55). One potential explanation for this apparent paradox is that the PD-L1 molecule is inducible by cytokines, mainly IFNγ in the tumor microenvironment, and consequently, its expression could reflect an endogenous anti-tumor immune response (12, 53). Thus, the upregulation of PD-L1 may contribute to the cronicity of inflammatory disorders, which frequently precede the development of human cancers (56). Specifically, in the case of OSCC, Tezal and colleagues (57) proved a significant association with periodontitis.

In this study two different monoclonal antibodies against PD-L1 (clones 22C3 and E1L3N) were used. These two antibodies were included together with another two anti-PD-L1 antibodies, 28-8 (Dako Link 48 platform) and SP142 (Ventana Benchmark platform), in a study that compared the performance similarity and interchangeability of all four PD-L1 systems (45), concluding that three of the four assays were essentially equivalent, whereas SP142 antibody identified only about 50% of patients who were detected as PD-L1–positive cases by the other three tests. These other antibodies appear to have minimal differences among them. Differences in PD-L1 positivity detection between clones E1L3N and 22C3 in our study respectively ranged between 9.7% and 14.4%, whereas PD-L1 immunopositivity in HNSCC has been reported to range widely between 18% (58) and 100% (37–41, 59). These varying results could be attributed to the heterogeneity in tumor samples, different stained protocols, and the variability in threshold or cut-off values for defining positive expression in different studies more than a true difference among the biological basis of OSCC in different studies.

In conclusion, patients with OSCC with high PD-L1 expression showed a poor prognosis probably due to the activation of PD-1/PD-L1 pathway that enhances an aggressive tumor phenotype by allowing tumor cells to evade the host immune system, thus establishing a T-cell exhaustion, which ultimately induces a specific tolerance. Accordingly, this group of patients with OSCC emerges as an ideal candidate for clinical trials of anti-PD-L1 targeted therapy.

No potential conflicts of interest were disclosed.

Conception and design: J.C. de Vicente, T. Rodríguez-Santamarta, J.M. García-Pedrero

Development of methodology: J.C. de Vicente, T. Rodríguez-Santamarta, J.P. Rodrigo, E. Allonca

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Rodríguez-Santamarta, J.P. Rodrigo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.C. de Vicente, V. Blanco-Lorenzo, J.M. García-Pedrero

Writing, review, and/or revision of the manuscript: J.C. de Vicente, T. Rodríguez-Santamarta, J.P. Rodrigo, E. Allonca, J.M. García-Pedrero

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.C. de Vicente, T. Rodríguez-Santamarta, J.M. García-Pedrero

Study supervision: J.C. de Vicente, J.M. García-Pedrero

The authors thank the Principado de Asturias BioBank for kindly providing the samples and technical assistance (PT13/0010/0046), financed jointly by Servicio de Salud del Principado de Asturias, Instituto de Salud Carlos III and Fundación Bancaria Cajastur and integrated in the Spanish National Biobanks Network. J.M. García-Pedrero has been awarded two grants from the Instituto de Salud Carlos III (ISCIII) and the FEDER Funding Program from the European Union PI13/00259 and PI16/00280. J.P. Rodrigo has been awarded a grant from the Instituto de Salud Carlos III CIBERONC (CB16/12/00390).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.