Low Expression of Programmed Death 1 (PD-1), PD-1 Ligand 1 (PD-L1), and Low CD8+ T Lymphocyte Infiltration Identify a Subgroup of Patients With Gastric and Esophageal Adenocarcinoma With Severe Prognosis

Prognosis of gastric and esophageal cancer is poor and treatment improvements are needed. Programmed cell death 1 receptor (PD-1) interaction with its ligand PD-L1 in tumor micro-environment promotes immune tolerance and blocking monoclonal antibodies have entered clinical practice. However, clinical significance of PD-1 and PD-L1 expression in gastric and esophageal adenocarcinomas, particularly in non-Asian patients, is still unclear. Three tissue microarrays including 190 clinically annotated esophageal (n = 31) and gastric (n = 159) adenocarcinomas and 58 paired mucosa specimens, were stained with PD-1, PD-L1, and CD8-specific reagents in indirect immunohistochemistry assays. PD-L1 expression was detectable in 23.2% of cancer specimens. High PD-1 expression was detectable in 37.3% of cases and high CD8+ infiltration in 76%. PD-L1 and high PD1 expression significantly correlated with each other (rs = 0.404, P < 0.0001) and both significantly correlated with CD8+ infiltration (rs = 0.435, P = 0.0003, and rs = 0.444; P = 0.0004, respectively). CD8+ lymphocyte infiltration correlated with improved survival in univariate (P = 0.009), but not multivariate analysis. Most interestingly, multivariate analysis and Kaplan-Meier curves indicate that combined low PD-1/PD-L1 expression and low CD8+ lymphocyte infiltration significantly correlate with poor prognosis. Our data document the clinical significance of a microenvironmental signature including PD-1/PD-L1 expression and CD8+ lymphocyte infiltration in gastric and esophageal adenocarcinomas and contribute to identify a patients' subset requiring more aggressive peri-operative treatments.

Prognosis of gastric and esophageal cancer is poor and treatment improvements are needed. Programmed cell death 1 receptor (PD-1) interaction with its ligand PD-L1 in tumor micro-environment promotes immune tolerance and blocking monoclonal antibodies have entered clinical practice. However, clinical significance of PD-1 and PD-L1 expression in gastric and esophageal adenocarcinomas, particularly in non-Asian patients, is still unclear. Three tissue microarrays including 190 clinically annotated esophageal (n = 31) and gastric (n = 159) adenocarcinomas and 58 paired mucosa specimens, were stained with PD-1, PD-L1, and CD8-specific reagents in indirect immunohistochemistry assays. PD-L1 expression was detectable in 23.2% of cancer specimens. High PD-1 expression was detectable in 37.3% of cases and high CD8+ infiltration in 76%. PD-L1 and high PD1 expression significantly correlated with each other (r s = 0.404, P < 0.0001) and both significantly correlated with CD8+ infiltration (r s = 0.435, P = 0.0003, and r s = 0.444; P = 0.0004, respectively). CD8+ lymphocyte infiltration correlated with improved survival in univariate (P = 0.009), but not multivariate analysis. Most interestingly, multivariate analysis and Kaplan-Meier curves indicate that combined low PD-1/PD-L1 expression and low CD8+ lymphocyte infiltration significantly correlate with poor prognosis. Our data document the clinical significance of a microenvironmental signature including PD-1/PD-L1 expression and CD8+ lymphocyte infiltration in gastric and esophageal adenocarcinomas and contribute to identify a patients' subset requiring more aggressive peri-operative treatments.

INTRODUCTION
Gastric and esophageal cancers are major causes of cancerrelated death. While esophageal cancer is the sixth most common cause of cancer-related death worldwide, gastric cancer is the second, with a high incidence in Asian countries (1). Localized tumors are routinely treated with surgical resection. However, even after potentially curative resection, overall survival (OS) is poor (2,3), and innovative therapeutic approaches are needed.
Responsiveness to immunotherapy critically depends on the characteristics of tumor immune contexture (18,19). Therefore, it is important to investigate CD8+, PD-1+, and PD-L1+ cell infiltration in large numbers of clinical specimens. Yet, the few published studies addressing gastric and esophageal cancer frequently yield conflicting results. Importantly, due to the high incidence of these cancers, most studies are from East Asia. However, gastric and esophageal carcinomas in Asian and non-Asian patients are characterized by different clinical course and immune infiltration (20). Moreover, molecular subtypes may be characterized by different microenvironment composition and expression of checkpoint molecules (21,22). Therefore, studies from different geographic areas are necessary.
Due to limited number of reports from Western countries and their low patient numbers, there is still a paucity of data regarding prognostic relevance, and therapeutic potential of CD8+ and PD-1+/PD-L1+ infiltration in gastric and esophageal cancers. In this study, we addressed these issues by using tissue microarray (TMA) technology.
Since our earlier studies (Däster et al., unpublished data) indicate that ESCC are poorly infiltrated by immune cells, and considering their highly specific molecular features, in this study we focused on the analysis of gastric and esophageal adenocarcinomas.

Tissue Microarray Construction
Three TMA blocks of non-consecutive primary esophageal and gastric cancer and paired non-malignant adjacent tissue specimens were constructed by using TMA-Grand Master R (3DHisteck, Sysmex AG, Switzerland). TMAs contained 1 mm cores of 190 gastric and esophageal adenocarcinomas together with 58 non-malignant mucosa samples. All specimens were part of the Biobank at the Institute of Pathology, University Hospital Basel, Switzerland. For TMA construction, formalinfixed, paraffin-embedded tissue blocks were prepared according to standard protocols. All cancer cores were checked for their percentage of tumor cells and this value was above 50% for all cancer samples.

Data Collection
Clinical and histopathology data were collected retrospectively in a non-stratified and non-matched manner. Data included patients' age, primary cancer site, TNM stage, disease-specific survival, histological subtype, presence of vascular invasion, and tumor diameter and grade. Approval for the use of samples and data was obtained from local ethics committee (Permission 361/12 EKBB).

Statistical Analysis
Data were analyzed using the Statistical Package Software R (Version 3.4.1, www.r-project.org). Descriptive statistics included mean ± standard deviation for parameters with Gaussian distribution or percentage of frequencies for occurrences for discrete variables. Cut-off values used to classify tumors with low or high infiltration of positive cells/histoscores were calculated by regression tree analysis (rpart package). Thereafter, based on the results (high/low) of all three markers, tumors were classified into three groups "PD-1/PD-L1/CD8 high" (high PD-1 and PD-L1 histoscores, high infiltration of CD8+ cells), "PD-1/PD-L1/CD8 low" (low PD-1 and PD-L1 histoscores, low infiltration of CD8+ cells), and "PD-1/PD-L1/CD8 mixed" (the remaining tumors).
Chi-square, Fisher's exact, Mann-Whitney-Wilcoxon, Kruskal-Wallis, and Jonckheere-Terpstra tests were used to determine the association of CD8, PD-1, and PD-L1 positivity and clinical-pathological features. Relationships between continuous markers were calculated using Spearman's correlation.
Survival analysis was carried out by Cox regression analysis and Kaplan-Meier curves were compared by log rank test. Any missing clinical-pathological information was assumed to be missing at random. Multivariate Cox regression analysis and hazard ratios (HR) and 95% confidence intervals (CI) were used to determine prognostic effects on survival. P ≤ 0.05 were considered statistically significant.

Patient and Tumor Characteristics
Clinical pathological characteristics of patients under investigation (n = 190) are reported in Table 1. Tissue samples from 31 esophageal adenocarcinoma to 10 paired non-malignant esophageal biopsies, as well as 159 gastric cancers and 48 non-malignant paired gastric tissue biopsies were evaluated. Four EAC were gastroesophageal junction tumors. Over 75% of tumors were in T2-3 stage and over 70% were in N0-1 stage. A majority of tumors were characterized by a G3 histological grade.

CD8, PD-1, and PD-L1 Expression in Gastric and Esophageal Adenocarcinoma Specimens
Representative examples of CD8, PD-1, and PD-L1-specific staining are shown in Figure 1. Total number of CD8+ lymphocytes detectable within tumor and stroma ranged between 0 and 548 (mean = 95 ± 9.6, median = 66). Tissue specimens were dichotomized as highly (76.2%) or poorly (23.8%) infiltrated, based on a threshold value of 33 CD8+ cells per 1 mm diameter punch, as identified by regression tree analysis. This value also corresponds to the first quartile of infiltrating CD8+ lymphocytes counted in this cohort. The calculated histoscore for PD-1+ (percentage of positive cells multiplied by intensity) in gastric and esophageal adenocarcinoma samples ranged between 0 and 210 (mean = 23 ± 39, median = 4) and a threshold value of 14 was calculated. PD-L1 histoscore, considering all positive cells in each punch, ranged between 0 and 160 (mean = 6.8 ± 21, median and third quartile = 0). Therefore, samples were scored as positive when at least one cell was specifically stained. Thus, a positive signal was detectable in 43/185 tumors (23.2%), but in no normal tissue. Notably, a moderately positive correlation was observed between PD-1+ and PD-L1 histoscores (r s = 0.404; P < 0.0001) and between each of these scores and CD8+ cell infiltration (r s = 0.444; P < 0.0004 and r s = 0.435; P = 0.0003, respectively).
While there was no significant difference in CD8+ infiltration between cancers and normal mucosa samples (P = 0.480), a significantly higher number of cancer samples was characterized by higher PD-1 histoscore, as compared to paired normal tissues (P < 0.0001) (Supplementary Figure 1).

Univariate and Multivariate Analysis of CD8, PD-1, and PD-L1 Expression in Gastric and Esophageal Adenocarcinomas
We initially analyzed the prognostic significance of the expression of individual markers. Univariate Cox regression analysis indicated that CD8+ infiltration was highly significantly (p = 0.009) associated with improved 5 years OS ( Table 2). Instead, higher PD-1 and PD-L1 scores were barely significantly associated "per se" with OS (P = 0.056 and P = 0.05, respectively). Notably however, a tumor microenvironment "signature" including high CD8+ cell infiltration and high PD1/PD-L1 scores was associated with significantly higher OS (P = 0.005). As expectable, pN stage (pos. vs. neg.) appeared to significantly impact on OS (P = 0.01).

Impact of PD-1/PD-L1/CD8 Signature in Gastric and Esophageal Adenocarcinomas
Subsequently, we explored clinical-pathological features in the three subgroups identified by uniformly high or low CD8+ infiltration and PD1/PD-L1 scores (PD-1/PD-L1/CD8 high and PD-1/PD-L1/CD8 low) or mixed results (PD-1/PD-L1/CD8 high and/or low). Complete follow-up data were available for 161 patients, including 133 gastric, and 28 esophageal adenocarcinomas. PD-1/PD-L1/CD8 high signature was detectable in slightly older patients as compared with the mixed or low signature (p = 0.046), but was independent from patients' gender, tumor size, tumor grade, and pN (data not shown). Most importantly, survival analysis indicates that gastric and esophageal adenocarcinomas with PD-1/PD-L1/CD8 "low" signature is characterized by poor long-term prognosis, as compared with the other two subgroups under investigation (p = 0.015, Figure 2). In particular, PD-1/PD-L1/CD8 "low" signature is characterized by poor long-term prognosis, as compared to the "high" (p = 0.008) or mixed signature group (p = 0.03), in spite of an apparent initial overlap of the survival curve of the latter. In contrast, due to relatively low numbers of patients under investigation, difference in survival curves of patients with "high" or "mixed" signature failed to reach statistical significance threshold (p = 0.2). A separate analysis of Kaplan-Meier curves for gastric and esophageal adenocarcinomas is reported in Supplementary Figures 2A,B.

DISCUSSION
PD-1/PD-L1 axis plays a crucial role in inhibiting T lymphocyte functions, allowing cancer cells to escape adaptive immune response (4,5). It has been suggested that "inflamed" tumors with high CD8+ T cell infiltration and high PD-1/PD-L1 expression might most benefit from PD-1/PD-L1 blockade (18).
Due to higher epidemiological relevance, gastric, and esophageal cancers have mainly been studied in Asian cohorts, and there are only few reports focusing on tumor microenvironment in non-Asian patients. However, particularly regarding gastric cancers, distinct gene expression profiles were detected in tumors from Asian and Western cohorts with significantly higher expression of T cell markers in the Western population (20).
In studies focusing on East Asian patients, PD-L1 was shown to be expressed in up to 40% of tumor cells in esophageal cancers, but its prognostic significance is debated (24)(25)(26)(27). In a cohort of Western patients, PD-L1 expression was reported to be associated with a favorable clinical course (28).
In a Western population of 465 gastric cancer patients, PD-L1 expression was detected in 30% of tumor cells and was associated with better outcome (36). However, association with adverse prognosis was also reported by others (37).
Regarding CD8+ T cells, current understanding is that high infiltration correlates with improved outcome, as shown in both esophageal (38,39) and gastric cancers (20,(40)(41)(42) in both Asian and non-Asian patients. However, conflicting data have also been published and a recent report in gastric cancer has surprisingly suggested an association between high CD8+ T cell density and poor survival (37). Importantly, CD8 expression was also associated with PD-L1 expression in that cohort (37).
Interestingly, contrasting data have been reported regarding the extent of CD8+ lymphocyte infiltration in these cancers, as compared to corresponding healthy tissues, possibly reflecting differences related to different tumor subtypes (38,(43)(44)(45)(46) and Helicobacter pylori status. In our study, we did not observe significant differences in cancer, as compared to healthy tissue CD8+ T cell infiltration. However, consistent with previous reports (44)(45)(46), in our series PD-1 expression was increased in cancerous, as compared to normal tissues, whereas PD-L1 expression was undetectable in healthy tissues.
Our data document that although CD8+ T cell infiltration, and, to lower extents, PD-1/PD-L1 expression appear to be associated with improved 5 years OS in univariate analysis, none of these markers significantly correlates "per se" with survival in multivariate analysis. Most interestingly however, a combined immunohistochemical analysis reveals that low PD-1/PD-L1/CD8 expression in gastric and esophageal adenocarcinoma microenvironment is associated with poor prognosis. Therefore, patients bearing these tumors should be eligible for early and more aggressive peri-operative treatments.
Our study has limitations. First, TMA technology does not mirror tumor tissue heterogeneity. However, TMA punches included at least 50% tumor cells and were taken from the center of tumor specimens. Moreover, the numbers of individual gastric and esophageal adenocarcinoma specimens compensate, at least in part, for the heterogeneity of different tumor areas. Secondly, the number of patients suffering from esophageal adenocarcinoma is limited. Subsequently the message of this study is consistent mainly for gastric cancers. Lastly, the retrospective design of the study limits its significance. Nevertheless, these data might help developing further prospective and mechanistic studies and contribute to the identification of subgroups of high-risk patients.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
Approval for the use of samples and data was obtained from local ethics committee (Ethikkommission beider Basel EKBB, Permission 361/12 EKBB).