PD-L1 predicts poor prognosis for nasopharyngeal carcinoma irrespective of PD-1 and EBV-DNA load

Programmed death-1 (PD-1) is an immunosuppressive receptor functionally bound with programmed death-ligand 1 (PD-L1), which has been reported in various malignancies. However, only a few studies are available for the clinical significance of PD-1/PD-L1 in nasopharyngeal carcinoma (NPC). In this study, we aim to investigate alterations in PD-1/PD-L1 by using immunohistochemistry analysis in a cohort of consecutively enrolled NPC patients (n = 99). To further analyse the correlation between PD-1/PD-L1 and factors involved in clinico-pathology, haematologic biomarkers, EBV-DNA load and outcomes, we collected clinical data for statistical analysis. We observed that lower haemoglobin (HB) and Body Mass Index (BMI) levels were associated with high levels of PD-L1 staining in NPC patients. Importantly, our results suggested that PD-L1 might be a negative indicator for NPC patients. In contrast, a correlation between the PD-1/PD-L1 level and EBV load was not identified. Moreover, PD-1 positivity was suggested to not be significantly correlated with clinical outcomes. Taken together, our results revealed that PD-L1 might be a potential prognostic biomarker for NPC patients. However, further studies are needed to clarify the underlying mechanism of EBV status in the immunosuppression process induced by the PD-1/PD-L1 axis.

of PD-L1 in NPC patients remains inconsistent. For instance, Zhang et al. 16 revealed that PD-L1 protein abundance was correlated with a worse outcome. In contrast, a positive correlation of PD-L1 expression and survival of non-metastatic NPC was delineated by Lee et al. 17 . In addition, Hsu et al. 18 showed that PD-1 on CD8 T cells predicted a poor prognosis in a small size of NPC patients, whereas PD-1 expression might not be a prognostic factor in NPC patients when using a larger sample 16 . Therefore, the prognostic role of PD-1 and PD-L1 in NPC needs to be clarified.
The blockade of the PD-1 and PD-L1 pathway is one of the most promising strategies to activate anti-tumour immunity 19 . There are also ongoing clinical trials evaluating the safety and efficiency of anti-PD1 antibodies in NPC 20 . Clinical data have shown that the treatment response to PD-1/PD-L1 blockade was correlated with PD-1 or PD-L1 detected by immunohistochemistry (IHC) 21,22 . In addition to PD-1/PD-L1 expression, many host-related influences might account for the heterogeneous responses and failures during immunotherapies, such as nutritional status 23 , smoking 24 , and inflammation status 25 . The tumour microenvironment, including hypoxia and immunosuppressive cytokine production, also plays a role in the therapeutic response 26 . Baseline peripheral blood parameters have been reported as predictive biomarkers for the response to PD-1 antibodies 27 . Thus, additional information about the clinico-pathological characteristics and haematologic biomarkers related to PD-1/PD-L1 expression could promote a better understanding of their regulatory role in NPC and might be helpful for improving the response to anti PD-1/PD-L1 treatment.
In addition, PD-1 and PD-L1 could induce T cell deactivation during chronic viral infections 28,29 . The Epstein-Barr virus (EBV) has been reported to be strongly linked with NPC in epidemic areas 30 . EBV-DNA load significantly correlates not only with NPC development but also with treatment failure 31 . Experimental studies 32 have suggested that there is an upregulation of PD-L1 in EBV-positive NPC cell lines. Despite improvement in preclinical research, only a little information is currently available for the correlation between EBV-DNA load and PD-1/PD-L1 expression in NPC.
Taken together, the prognostic value of PD-1/PD-L1 expression in NPC remains largely unknown. Detailed data on potential related biomarkers and EBV infection status with PD-1/PD-L1 are needed to be confirmed by clinical data. Therefore, considering the tremendous therapeutic potential of targeting the PD-1/PD-L1 interaction in NPC, we analysed the prognostic value of PD-1 and PD-L1 in a cohort of NPC patients (n = 99), which were consecutively recruited form a single medical care group over the course of two years. We also evaluated the correlation between PD-1/PD-L1 expression and clinico-pathological parameters as well as potential related haematologic biomarkers, such as the haemoglobin (HB) level. Our study might provide a new strategy for immune checkpoint-blocking therapy.

Results
General information. All of the nasopharynx biopsy samples (n = 99) were collected and histologically diagnosed as NPC by experienced pathologists. The samples were fixed with formalin and embedded into paraffin using a tissue processor. Among the 99 patients enrolled, including 68 males and 31 females, the median age was 47.5 years old (range from 20 to 78 years). All patients were treated with intensity-modulated radiation therapy (IMRT) with or without combined chemotherapy. The average follow-up time was 49.4 months (range from 6.7 to 64.9 months). In this cohort, 18 patients encountered disease progression (recurrence and/or metastasis), and 9 patients died during the follow-up period. All tumours were classified as the undifferentiated non-keratinizing phenotype. The characteristics of the enrolled patients are summarized in Supplementary Table S1.
Clinico-pathologic correlations. PD-1-positive immune cells were present in 44 of the total 99 tumours (44.4%), and PD-L1 staining was detectable in 96 patients (97.0%) and was mainly located at the membrane or in the cytoplasm region (or both) in the tumour cells. PD-L1 was also expressed in tumour-infiltrating lymphocytes (TILs) in a scattered manner. According to the ROC curve analysis for OS, the optimal cut-off value for the H-score was 155 (AUC: 0.780, sensitivity: 1.000, specificity: 0.419). Thus, when the H-score ≥ 155, PD-L1 staining was classified as being at a high level (61 cases). PD-1 positivity was defined as cases with PD-1 staining intensity ≥ 2 in more than 5% of TILs. Representative stainings of PD-1 and PD-L1 in NPC are shown in Figs 1 and 2. The optimal cut-off value for the nasopharynx gross tumour volume (GTVnx) was 29.6 mm 3 (AUC: 0.740, sensitivity: 0.917, specificity: 0.505). In this study, neither PD-1 positivity nor a high level of PD-L1 staining was significantly correlated with the clinico-pathological parameters of age, gender, smoking status, family history of cancer, GTVnx or clinical stage at diagnose. Detailed data are summarized in Table 1 BMI and HB in 99 NPC patients with heterogeneous immunoreactivity of PD-1/PD-L1. Weight and height were collected to calculate the body mass index (BMI), calculated by dividing the weight in kilograms by the square of the height in metres (kg/m 2 ). The BMI of the 99 NPC patients ranged from 16.26 to 33.83 kg/m 2 . We compared the difference in the BMI level of patients with low or high levels of PD-L1 staining, and our results suggested that patients with a high level of PD-L1 staining had significantly lower BMI levels (P = 0.048, Fig. 3c).
Our results also revealed that the HB level was significantly reduced in the patients with a high level of PD-L1 staining (P = 0.029, Fig. 3d). In contrast, we observed no significant difference in the BMI (a) or HB level (b) in patients with positive or negative PD-1 staining. (Fig. 3a,b).

No significant association between PD-1/PD-L1 expression and EBV load.
To evaluate the correlation between EBV viral load and immunoreactivity of PD-1 and PD-L1 in the NPC patients, we collected the qRT-PCR data of EBV-DNA copies from our database. In this cohort of 99 patients, EBV-DNA load was detectable in 77 patients. Non-parametric tests showed no significant difference in the PD-1 (P = 0.853, Fig. 4a H-scores (P = 0.389, Fig. 4b) between patients with a detectable or undetectable EBV-DNA load. Chi-square tests indicated that there was no significant association between the PD-1 (P = 0.914, Table 1) or PD-L1 level (P = 0.439, Table 1) and the EBV level.
Prognostic values related with PD-1 and PD-L1. To evaluate the prognostic values in PD-1-positive tumour-infiltrating lymphocytes (TILs) and tumour cells with PD-L1 expression, we used the Kaplan-Meier survival analysis and log-rank tests. In this consecutively enrolled cohort of 99 NPC patients, our results revealed that a high expression of PD-L1 was correlated with shorter OS (P = 0.015, Fig. 5b) and showed trend of a reduced progression-free survival rate (PFS rate) (P = 0.127, Table 2). The factors significantly correlated with OS by univariate analyses were age (Fig. 5c), the level of PD-L1 staining (Fig. 5b), clinical stage (Fig. 5d), and GTVnx ( Fig. 5e) before treatment. In contrast, we observed that the PD-1 positivity in TILs was not significantly correlated with OS (P = 0.563, Fig. 5a) or PFS (P = 0.616, Table 2). Moreover, in the PD-L1 highly expressed cases, the positivity of PD-1 could not further predict the OS (P = 0.399) and PFS of patients (P = 0.956). Multivariate analyses of prognostic factors were performed using a Cox regression model. T classification was not put in the multivariate analysis for it was found to be significantly correlated with clinical stage by a chi-square test (ϕ correlation coefficient = 0.868, P = 0.000) in this group of NPC patients. Thus, the variables included in the multivariate analysis for OS were age, clinical stage, GTVnx and PD-L1 H-score (Table 3). In the multivariate analysis, only the PD-L1 H-score was suggested to be an independent prognostic factor for OS. Patients with a higher PD-L1 H-score could have a greater risk of death (95% CI, 1.002-1.031, P = 0.028, Table 3).

Discussion
The association of the immunological checkpoint PD-1/PD-L1 and its prognosis of various cancers are currently a research hotspot 19 . PD-L1 has been reported to be overexpressed in most tumours including NPC 16,33 to inhibit T cell-mediated antitumour immunity via PD-1 on TILs 34 . Recent studies have also revealed that PD-L1 was associated with poor prognosis in most epithelial-originated cancers 35 , suggesting an effect of PD-L1 in the induction of tumour progression by interrupting anti-tumour immunity 36 . However, the prognostic significance of PD-L1 as well as PD-1 in NPC has not been clarified yet. We collected biopsies and clinical data from 99 consecutively enrolled NPC patients to analyse the association of PD-L1/PD-1 and related factors. Our results suggested that NPC patients with a high level of PD-L1 had a significantly reduced survival outcome (Fig. 5b), which is consist with the results from 132 NPC patients 16 . However, another publication noted that there was a longer survival rate in non-metastatic NPC patients with a high PD-L1 level 15 , which might be explained by heterogeneous applications of different experimental procedures, scoring criteria, patient samples and survival endpoints. These differences could also illustrate why the PD-L1 expression rate was found to be much lower (25%) in the NPC patients reported by Lee et al. 17 than that reported in our results (97.0%) and Zhang et al. 16 (95.0%).
Furthermore, we also analysed the correlation between PD-1 positivity and the survival outcomes of NPC patients and found that there was no significant association (Fig. 5). Interestingly, previous publications have suggested a worse clinical outcome in NPC patients with PD-1-positive CD8 T cells (total n = 46) 18 . However, in a larger cohort (total n = 132), the impact of PD-1 was not obvious 16 .
As of now, standardized testing procedures and consistent scoring criteria for PD-1/PD-L1 in solid malignancies by IHC still need to be optimized 35,[37][38][39] . The H-score has been applied in many IHC studies for its semi-quantitative characteristic 39,40 . Since no standard scoring criteria for PD-1 and PD-L1 are available 35,3841 , we have used an ROC analysis to determine the cut-off value for PD-1 and PD-L1. However, our results suggested no significant cut-off value for PD-1 based on survival endpoints. The cut-off point was 155 for the PD-L1 H-score by ROC analysis in this study, while an intensity ≥ 2 17 or an H score ≥ 35 16 was chosen to be the cut-off value in previous publications. As reported by Zhang et al. 16 , an H-score > 0 was applied as the cut-off value for PD-1. Consistently, we also found no significant correlation with the pathological features or prognostic role of PD-1 when the cut-off was defined as an H-score > 0. We raised the criteria of PD-1 positivity to a staining intensity ≥ 2 in more than 5% of the TILs (H-score ≥ 10), similar to the methods reported by D'Incecco et al. 24 , to minimize the potential interference of non-specific staining for PD-1.
Our study revealed a non-significant difference between the PD-L1 level and PD-1 positivity (Table 1), suggesting an indirect interaction of PD-L1 and PD-1 in NPC cells. PD-1 belongs to the CD28 family and is characterized as an inhibitory receptor expressed in T cells, dendritic cells (DC), natural killer (NK) cells, macrophages and B cells 42 . The PD-1 ligands, PD-L1 and PD-L2, belong to the B7 superfamily 43 . However, binding targets of PD-L1 and PD-L2 are not restricted to PD-1, since PD-L1 and PD-L2 can also bind to CD80 44 and Repulsive Guidance Molecule b (RGMb) 45 , respectively. In addition, PD-L1 may be expressed in T cells, B cells, myeloid dendritic cells (DC) and in tissue macrophages in the tumour microenvironment 42 . It has also been reported that the predictive significance of PD-L1 can vary depending on if the positivity is defined on tumour cells or TILs 46,47 . We have found an increased level of PD-L1 in TILs (Fig. 2), and further studies are needed to clarify the underlying mechanism. The identification of potential factors that can stratify a response to anti-PD-1 therapies will aid in the selection of NPC therapy. In NPC, experimental research has demonstrated an upregulation of PD-L1 by EBV-induced latent membrane protein 1 (LMP1) in EBV-positive NPC cell lines 32 . Recently, PD-L1 and PD-1 expression were found to be significantly associated with EBV-associated malignancies [47][48][49] , where EBV-encoded RNA was used to test for EBV infection in most studies. However, our results revealed no significant correlations between baseline EBV load and PD-1/PD-L1 expression (Table 1 and Fig. 4). One possible explanation could be that the circulating cell-free plasma EBV-DNA load might be an inactive remnant of a previously active EBV infection 50 . Moreover, our study also showed reduced BMI and HB levels in patients with high levels of PD-L1 staining (Fig. 3). Because of a large overlap in confidence intervals of our statistical results, the clinical significance between these factors still needs to be confirmed in larger samples. Notably, previous studies have confirmed that a low baseline HB is an adverse prognostic factor in patients with locally advanced head and neck squamous cell carcinoma 51 and NPC 52 , by inducing treatment resistance via enhancing tumour hypoxia. Thus, we showed these findings to arouse attention for the possible biologic relationship between HB and PD-L1, which might be triggered by hypoxia, inducing immune deficiency through the PD-1/PD-L1 axis 53 . In addition, we have also compared PD-1/PD-L1 expression with a few baseline haematological parameters, such as neutrophil counts and the lactate dehydrogenase level from peripheral blood in this exploratory analysis, and there was no significant association observed for any of the relevant factors ( Supplementary Fig. S1 and S2).
Our study revealed the prognostic role of age, PD-L1, GTVnx, T classification and clinical stage in the 99 NPC patients by log-rank test. Age, smoking history, EBV-DNA load, HB, GTVnx, and T and N classification as well as clinical stage have been reported to have a prognostic role in primary NPC patients in previous studies 52,54-57 . In our study, age 50 was chosen to be the cut-off value according to previous studies on NPC patients as reported by Chen et al. 54 . The baseline EBV-DNA loads were dichotomized to undetectable and detectable levels as reported by Liu et al. 56 . The optimal cut-off point of GTVnx (29.6 mm 3 ) was determined by an ROC analysis in this study, which was consistent with previous studies in different NPC samples, where 19 mm 3 (Guo et al. 57 ) and 25 mm 3 (Wu et al. 58 ) were determined to be the optimal cut-off value by ROC analysis. Patients with T1-3 classification showed a similar survival rate in this study, which might be due to the satisfactory local control rates of NPC patients raised by IMRT 59 and might be the reason that our multivariate analysis failed to identify the clinical stage as a significant prognostic factor.
In conclusion, PD-L1 and PD-1 are widely expressed in NPC tissue. However, PD-1 positivity could not predict prognosis in our study. Importantly, PD-L1 has been suggested to be a negative prognostic factor for NPC patients. Further studies with a larger sample size are needed to confirm these observations and to evaluate the predictive value of PD-1 and PD-L1 in NPC in the context of PD-1 inhibition as well. Moreover, more studies are warranted to identify the molecular mechanism of the interaction between HB, EBV and PD-1/PD-L1 in NPC.

Materials and Methods
Ethics Statement. The study was approved by the Institutional Review Board of Sun Yat-sen University Cancer Center. As this was a retrospective analysis of routine data, we requested and were granted a waiver  1 Chi-square test. 2 The optimal cut-off points for PD-L1 and GTVnx were determined by the receiver operating characteristic (ROC) curve for overall survival (OS). 3 According to the 7 th Edition of the AJCC/UICC Staging System for Nasopharyngeal Cancer.
of individual informed consent from the ethics committee. Patient records/information was anonymized and de-identified prior to analysis.  Patients with active hepatitis, diabetes, or who failed the follow-up requirements were excluded. Therefore, there were 99 cases that qualified for this study. All medical records were reviewed retrospectively, and all patients were restaged according to the 7 th edition of the UICC/AJCC system. FFPE blocks of NPC were retrieved by fibrescope biopsy from the Department of Pathology at Sun Yat-sen University Cancer Center. The related clinical data of these patients were retrospectively collected from our database or hospital records, including gender, age, body weight, height, clinical stage, GTVnx (gross tumour volume of primary tumour), pre-therapy laboratory counts of neutrophils, lymphocytes, HB, platelets, albumin, lactate dehydrogenase (LDH), and EBV-DNA copies by quantitative PCR test, as well as follow-up records. All of the patients were followed-up for at least 40 months.
Radiotherapy. All patients received a computed tomography simulation scan at the radiotherapy position.
The images included plain and enhanced computed tomography scans. The scope of each scan was from the top of the head to 2 cm below sub-clavicle head (thickness: 3 mm per slice). According to the institutional treatment protocol from Sun Yat-sen University Cancer Center, contouring of targets and organs at risk was performed under the IMRT planning system. The prescribed doses for the NPC patients were applied as follows: (1) Table 2. Univariate analysis of prognostic factors involved in survival. 1 Log-rank test. 2 The optimal cut-off points for GTVnx and the PD-L1 H-score were determined by the receiver operating characteristic (ROC) curve for overall survival (OS). 3 According to the 7 th Edition of the AJCC/UICC Staging System for Nasopharyngeal Cancer. *P < 0.05, **P < 0.01, ***P < 0.001.  Immunohistochemistry staining. The samples were fixed with 4% paraformaldehyde (PFA) containing 2% sucrose in PBS at 4 °C overnight and embedded into paraffin using a tissue processor (EG1150, Leica, Germany). FFPE sections (3 μ m thick) were cut with a rotation microtome (RM2255, Leica, Germany). Dewaxed paraffin sections were rehydrated by alcohol series, treated with 3% H 2 O 2 for 10 min at room temperature and steam-heated for 2.5 min to retrieve the antigen using ethylene diamine tetra acetic acid buffer (PH = 8.0). Subsequent immunostaining was performed with a 50-min incubation period in 37 °C with the monoclonal antibodies for PD-1 and PD-L1 (Supplementary Table S2). The IHC staining of PD-1 and PD-L1 proteins was performed on two different slides. Tonsil tissue was taken as a positive control. Immunoreaction was visualized using a Peroxidase/DAB kit (Cat. K5007, Dako, Denmark). Images were taken with a phase contrast microscope (Eclipse 80i, Nikon, Japan). Details of all reagents with reference to the immunohistochemical staining procedure are listed in Supplementary Table S3.
Evaluation standard of IHC. The percentages of PD-L1-positive tumour cells and PD-1-positive lymphocytes and staining intensity were evaluated by two pathologists who counted 20 sequential high-power fields (0.54 mm field diameter) judged to be representative of the sample, while remaining blinded to clinical information. Staining intensity was scored as follows: 0 for negative or trace; 1 for weak; 2 for moderate; 3 for high. The percentage of stained cells (0-100%) was multiplied by the dominant intensity pattern of staining ranging from 0 to 3. Therefore, the overall semi-quantitative score (H-score) ranged from 0 to 300 (maximum value of 300 corresponding to 100% of tumour cells positive for PD-L1 or TILs positive for PD-1 with an overall staining intensity score of 3).
Statistical analysis. The optimal cut-off point for the expression level of each marker including the PD-L1 H-score and GTVnx were determined by the area under the curve (AUC) of the receiver operating characteristic (ROC) curve at the highest positive likelihood ratio point for overall survival (OS). PD-L1 expression was dichotomized into two groups (High and Low), using a cut-off score of ≥ 155. PD-1 positivity was defined as cases with PD-1 staining intensity ≥ 2 in more than 5% of TILs. OS was measured from the date of diagnosis to the date of death or the date of the last follow-up through April 2016. PFS was measured from the date of diagnosis to the date of death or the date of tumour progression (recurrence and/or metastasis) through April 2016. Patients who were still alive at last contact or died from other causes were treated as censored for OS analysis.
A chi-square test was used to assess the correlation of the expression of PD-1 and PD-L1 with various clinical parameters such as age, clinical stage and EBV-DNA status. A t-test was used to compare the differences between the BMI, HB level, and neutrophil and lymphocyte counts in patients with different PD-1/PD-L1 immunoreactivity after a normality test (Kolmogorov-Smirnov test). A non-parametric Mann-Whitney U test was applied to compare the differences in albumin, LDH and the neutrophil/lymphocyte ratio in patients with different PD-1/PD-L1 immunoreactivities. A non-parametric Mann-Whitney U test was also applied to compare the differences in the PD-1/PD-L1 H-score in patients with different baseline EBV-DNA status. The number that follows the ± sign is the standard error (s.e.m.) in this article. Survival analysis was depicted by the Kaplan-Meier method defined by cut-off points generated from the ROC. Univariate analyses and multivariate analyses were performed with a log-rank test and Cox regression analysis, respectively. A P value < 0.05 was used to denote statistical significance, and all reported P values are two sided. P values were marked as *P < 0.05, **P < 0.01, ***P < 0.001, indicating different level of significance. These statistical analyses were performed with SPSS 16.0 [48][49][50][51][52][53][54][55][56][57][58][59] .