Differentiated tumor immune microenvironment of Epstein–Barr virus-associated and negative gastric cancer: implication in prognosis and immunotherapy

Epstein-Barr virus-associated gastric cancer (EBVaGC) has been proposed to be a distinct subtype with a specific immune microenvironment. Here, we evaluated tumor-infiltrating T-cell subsets and the expression of programmed cell death protein 1 (PD-1) and programmed death-ligand 1 (PD-L1) in 571 gastric cancers (GCs). Tissue microarrays were stained using EBER in situ hybridization for EBV and using immunohistochemistry for CD4, CD8, Foxp3, PD-1 and PD-L1. GCs were categorized into four types based on CD8+ infiltration and PD-L1 expression. The 5-year overall survival (OS) was evaluated according to EBV infection, T-cell subsets, PD-1 and PD-L1 expression and immune types. Thirty-two (5.3%) EBVaGCs were identified, which were more prevalent for CD8+ (p<0.001) and Foxp3+ (p=0.020) cell infiltration than EBV-negative GCs (EBVnGCs), suggesting a better 5-year OS (p=0.003). CD8+ (p=0.001) and Foxp3+ (p=0.018) cell infiltration was associated with better 5-year OS, whereas PD-L1 expression correlated with a poor 5-year OS (p=0.002). EBVaGC and EBVnGC had heterogeneous immune microenvironment, with CD8+ PD-L1− GC exhibiting the best 5-year OS (p<0.001). GC was an immune ignorant dominant tumor and poor to no T-cell infiltration. An immune type classification algorithm can provide prognostic information and a rational basis for immunotherapy.


INTRODUCTION
Epstein-Barr virus (EBV) is a member of the Herpesviridae family that latently infect greater more than 90% of adults worldwide and is associated with several human malignancies, such as Hodgkin's lymphoma, Burkitt's lymphoma, nasal NK/T-cell lymphoma, nasopharyngeal carcinoma and a subset of gastric www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 40), pp: 67094-67103 Research Paper cancer (GC) [1]. Recently, the Cancer Genome Atlas Research Network published a comprehensive molecular characterization of gastric adenocarcinoma and classified EBV-associated GC (EBVaGC) as a distinct GC type that is characterized by recurrent PIK3CA mutations, extreme DNA hypermethylation, and amplification of JAK2 and programmed death ligand-1/2 (PD-L1/PD-L2) [2]. A metaanalysis demonstrated that the prognosis for EBVaGC was better than that for EBV-negative GC (EBVnGC) [3], although, the underlying mechanisms of this effect are not clear. EBVaGC often features extensive lymphocytes infiltration [4], especially CD8 + (cluster of differentiation 8) T cells, which might have a cytotoxic effect on EBVpositive tumor cells [5].
As a "non-self" component, tumor cells can trigger immune responses characterized by the infiltration of various immune cells, which can affect tumor progression. In addition, the tumor develops many strategies to evade an immune response, including immune suppressors such as regulatory T cells (Treg) and myeloid-derived suppressor cells (MDSC) [6], immune checkpoint cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed death protein 1 (PD-1)/PD-L1 [7] and T cell dysfunction [8]. Based on the presence or absence of tumor-infiltrating lymphocytes (TILs) and PD-L1 expression, four tumor immune microenvironment types have been proposed [9], including type I (TILs + PD-L1 + associated with adaptive immune resistance), type II (TILs -PD-L1indicating immune ignorance), type III (TILs -PD-L1 + indicating intrinsic induction) and type IV (TILs + PD-L1indicating other suppressors promoting immune tolerance). To some extent, this stratification has prognostic value and indicates possible cancer immunotherapy strategies.
In this study, we conducted a retrospective analysis to evaluate EBV infection status, tumor-infiltrating T-cell subsets and PD-1/PD-L1 expression in 571 tumor samples. Furthermore, we classified these tumor samples into four cancer types defined by CD8 /PD-L1 status. Overall survival (OS) was analyzed according to EBV infection, T-cell subsets, PD-1/PD-L1 expression and cancer types.

OS, T-cell infiltration subsets and PD-1/PD-L1 expression between EBVaGC and EBVnGC cohorts
Fifteen patients (48.4%) in the EBVaGC cohort and 410 patients (75.9%) in the EBVnGC cohort died during the 5-year follow-up. Five patients in the EBVaGC cohort and 49 patients in the EBVnGC cohort were lost to followup. The log rank test indicated a better OS in the EBVaGC cohort than in the EBVnGC cohort (p=0.003, Figure 2A).
We also analyzed the CD4 + , CD8 + and Foxp3 + cell infiltration and PD-1/PD-L1 expression based on TNM stage. With tumor TNM stage advanced, less CD4 + , CD8 + and Foxp3 + cell infiltration and PD-1 expression was found in the complete cohort (Supplementary Figure 3).

DISCUSSION
In this study, we determined the EBV status of 571 GC patients and profiled tumor-infiltrating T-cell subsets and PD-1/PD-L1 expression patterns. The frequency of EBV infection in GC was 5.3%, and most patients came from northern and western China. The lower frequency of EBV infection in our study might be because of geographical and environmental factors [3,10,11], and EBV infection was associated with younger age.
Several studies have demonstrated that EBVaGC has a better clinical outcome which agrees with the results of our study [3,4,12]. In our study, EBVaGC was predominantly characterized by CD8 + and Foxp3 + cell infiltration, and in the complete cohort, more frequent infiltration of CD8 + and Foxp3 + cells correlated with a better OS. Therefore, we speculated that the adaptive immune response might have a positive effect on EBVaGC prognosis. EBVaGC displays a latency I/II pattern of EBV gene expression, typically characterized by LMP2 and EBNA1 expression [13,14]. EBV-associated proteins and some mutated tumor cells proteins can elicit EBV-specific and tumor antigen-specific cytotoxic T lymphocytes, which can eradicate EBV-positive malignant cells [15][16][17].
PD-1 is expressed on activated T and B cells whereas PD-L1 is generally expressed on many immune cells, including macrophages and dendritic cells, and can be induced by inflammatory cytokines in tumor cells [7]. PD-L1 expression has been observed in various malignancies and is a major factor used to evaluate therapy responses to anti-PD-1 [18][19][20][21]. We found relatively more frequent PD-1 expression and comparable PD-L1 expression in EBVaGC compared with that in EBVnGC. Based on CD8 + cell infiltration and PD-L1 expression, EBVaGCs were primarily type I and type IV cancers, suggesting a favorable immune microenvironment, whereas most EBVnGCs were type II and type III cancers, which indicated that most EBVnGCs did not naturally induce effector T-cell responses. Moreover, we demonstrated that type IV GC had the best 5-year OS, which might explain why EBVaGC has a better prognosis, although immune checkpoint blockade has overall been a disappointing GC treatment. Tumor regression after therapeutic PD-1 blockade requires pre-existing CD8 + T cells that are negatively regulated by PD-1/PD-L1 expression [22]. Hence, immune checkpoint inhibitors treatments in type I cancer might partially relieve adaptive immune resistance and exert a positive effect on clinical outcome. Considering the lack of T cell infiltration in type II and type III GCs, a T-cell inducing agent such as a tumor vaccine combined with immune checkpoint inhibitors might improve GC survival.
A recent study demonstrated that increasing CD8 + T cell infiltration and PD-L1 expression were associated with impaired progression-free survival and OS in 34 gastric adenocarcinomas, which contradicts findings that TILs are associated with better disease outcomes in GC [23][24][25]. In our study, CD8 + cell infiltration predicted a better 5-year OS, whereas PD-L1 expression predicted a poor prognosis. Moreover, we found that Foxp3 + cell infiltration and PD-1 expression were associated with CD4 + and CD8 + cell infiltration, and the same correlation was observed between PD-L1 expression and CD4 + cell infiltration, but not between PD-L1 expression and CD8 + cell infiltration. Furthermore, PD-L1 expression was significantly associated with PD-1 expression. A comparable study illustrated that PD-L1 expression was significantly associated with infiltrating immune cells and PD-1 expression [26]. The expression   of immunosuppressive markers and the presence of immune inhibitory cells do not indicate ineffective antitumor immune response, but are indicators of an immune response [27], because immune suppression usually occurs after immune activation.

PD-1(+) n (%) p # Foxp3(-) n (%) Foxp3(+) n (%)
Our study had several limitations. We had only 32 EBVaGC cases, which restricts T cell subset infiltration and PD-1/PD-L1 expression analysis in this subgroup. We did not find an association between PD-L1 expression and CD8 + cell infiltration, which was observed in other studies, and requires further investigation. Tissue microarray IHC staining as a high throughput technique can facilitate this type of study, but bias might still arise because only a relatively small tissue area was deliberately selected after H&E staining confirmation. Moreover, a tissue microarray with only a 1.5mm diameter was not suitable for quantitative evaluation of lymphocytes infiltrating in tumor and immune stroma, and our results were qualitative. This is a retrospective study, and all of the results are based on formalin-fixed, paraffin-embedded tissue and should be validated on live cells and in clinical trials.
In summary, this study showed that there were more frequent tumor-infiltrating CD8 + and Foxp3 + cells in EBVaGC, which predicted a better OS in this subpopulation. We first categorized GCs into four different tumor microenvironment types based on the presence or absence of CD8 + cells and PD-L1 expression in a large cohort (n=571). EBVaGC and EBVnGC showed heterogeneous immune microenvironments. Approximately 65% of EBVaGCs presented with a type I or type IV microenvironment, suggesting activated adaptive immune responses and a relatively better clinical outcome, whereas more than 70% of EBVnGCs were type II and type III cancers, indicating a lack of immune response and a poor prognosis. A low level of TILs might partially explain the high mortality of GC and restrict the use of antibodies that target immune checkpoints. We hope that this immune microenvironment classification algorithm can provide a rational basis for guiding cancer immunotherapy in future clinical practice.

Patients and specimens
GC specimens were collected from 608 patients who received primary tumor surgical excision at Xijing Digestive Hospital between Jan 2007 and Jan 2010. All of the patients with primary GCs were included unless they had any of following conditions: previous chemotherapy prior to surgery, longstanding autoimmune disease and/ or glucocorticoid treatment, HIV positivity, lymphoma or other sources of gastric metastatic carcinoma. The tissue microarray method was applied (a 1.5-mm diameter column of tissue core). Age, sex, the American Joint Committee on Cancer (AJCC) system for TNM stage (T: depth of tumor invasion, N: lymph node metastasis, M: distant metastasis), tumor location, depth, histological classification and differentiation were assessed.
All patients were followed for 5 years. In analysis, we excluded one gastric lymphoma, and 36 patients were lost at the first follow-up. In the end, 571 patients were analyzed and evaluated.

In situ hybridization (ISH) for EBV-encoded small RNA (EBER)
Tissue microarrays including 608 specimens were used for EBER in situ hybridization (ISH). EBER, a non-coding RNA, is the most abundant RNA in EBVinfected gastric tumor cells. 4μm thick sections were cut, deparaffinized and dehydrated. EBERs in tissue microarrays of primary tumor specimens were hybridized using an Epstein Barr virus Probe ISH Kit, and were detected with the HRP/DAB Detection System (ISH5021, PanPath, Amsterdam, Netherlands) according to the manufacturer's protocols. Cases with strong signals within tumor cells were considered EBV positive. The results were determined by two independent pathologists. If they did not agree the same, then the third pathologist participated in the decision.

Immunohistochemistry
Formalin-fixed, paraffin-embedded tissue microarrays were cut into 4μm thick sections and processed for IHC staining. Sections were deparaffinized and dehydrated with serial passages through changes of xylenes and graded ethanol. All of the microarrays were subjected to hearinduced epitope retrieval for 10 min at 100°C (citric acid sodium citrate solution was used for CD4, CD8, Foxp3 (Forkhead box P3), the pH was 6.0, and universal HIER antigen retrieval reagent was used for PD-L1, Abcam). Endogenous tissue peroxidase was blocked by slide incubation in 3% hydrogen peroxide solution for 10 min at room temperature prior to incubation with blocking serum 30 min at room temperature. Subsequently, the microarrays were incubated with primary antibodies (anti-Foxp3, 1:200, Abcam, which was diluted by antibody diluent, Abcam) overnight at 4°C, 4 hours at 37°C for PD-L1 (10μg/ml, Abcam) and 2 hours at room temperature for CD4 and CD8 with no dilutions. Goat anti-mouse/rabbit secondary antibody and HRP-labeled biotin were used to bind primary antibody. Antigen-antibody binding was visualized via application of 3.3' diaminobenzidine (DAB) chromogen. PD-1 (Polyclonal Goat IgG, 5μg/ml, R&D) IHC was achieved using the anti-goat HRP-DAB Cell & Tissue Staining Kit (brown, R&D) according to the manufacturer's protocols. A negative control was obtained by using a normal rabbit/mouse IgG. Stained sections were counterstained with hematoxylin and coverslipped for review.

IHC analysis
Tissue microarrays were scanned under ×100 and ×200 magnification using Olympus Soft Imaging Solutions GmbH. The levels of CD4, CD8 and Foxp3 expression were defined as follows: cases were scored at 5% intervals for CD4 and CD8, which meant that specimens with≥5% membranous expression were considered positive. The presence of PD-L1 was estimated according to the same criteria [26]. Positive Foxp3 infiltration is defined by a greater than 1% presence of Foxp3 + cell relative to the total cell number. PD-1 expression was positive when≥1% of lymphocytes showed membrane staining. Based on the presence or absence of CD8 + cell infiltration and PD-L1 expression, tumor immune microenvironment was classified as follows: type I (CD8 + PD-L1 + ), type II (CD8 -PD-L1 -), type III (CD8 -PD-L1 + ) and type IV (CD8 + PD-L1 -). The results were determined by two independent pathologists. If they did not agree the same, then the third pathologist participated in the decision.

Statistical analysis
Categorical data were compared using χ2 test or Fisher's exact test with a two-sided p value. Bonferroni correction for multiple comparisons was applied when necessary. Age was described using median and minimum/ maximum values and analyzed using the Mann-Whitney U test. Survival curves were constructed by using the Kaplan-Meier method and compared by using a log-rank test. Hazard ratios (HRs) and 95% confidence interval (95%CI) for 5-year survival by EBV infection and CD4 + , CD8 + , Foxp3 + cell infiltration and PD-1/PD-L1 expression were estimated in Cox regression models. Five-year overall survival (OS) was defined from the date of surgery to the date of death from any cause, or to month 60 if the patient is alive. The Cox proportional hazard regression model was used to define independent prognostic biomarkers such as CD4, CD8, Foxp3 and PD-L1 that inversely impacted OS. Statistical analyses and graphics were obtained using SPSS 21.0 software (Chicago, USA) and Graph pad 5.0. A twotailed p<0.05 was considered statistically significant.