PD-L1 overexpression in EBV-positive gastric cancer is caused by unique genomic or epigenomic mechanisms

Epstein-Barr virus-positive gastric cancer [EBV (+) GC] is a distinct GC subtype with unique genetic and epigenetic aberrations. Here, we examined resected GC samples and publicly available microarray data and The Cancer Genome Atlas (TCGA) database to identify the mechanism underlying overexpression of PD-L1 in EBV (+) GC. We found that high levels of PD-L1 overexpression in EBV (+) GC were caused by focal amplification of CD274. By contrast, relatively high expression of PD-L1 in tumor tissue and infiltrating immune cells correlated with CD8 lymphocyte infiltration and IFN-γ expression via IRF3 activation. Since we reported previously that PD-L1 expression is associated both with the presence of CD8 T cells in the tumor microenvironment and with IFN-γ expression in GC, we examined a database to see whether IFN-γ-associated overexpression of PD-L1 plays a significant role in EBV (+) GC. Immunohistochemical staining showed that expression of the IRF3 signature in clinical GC samples was higher in EBV (+) than in EBV (−) cases. The data presented herein reveal a unique dual mechanism underlying PD-L1 overexpression in EBV (+) GC: high focal amplification of CD274 or IFN-γ-mediated signaling via activation of IRF3.


Histological examination of PD-L1 upregulation in EBV (+) GC. To investigate the mechanism
underlying PD-L1 overexpression in EBV (+) GC, we first performed IHC staining for PD-L1 in the Fukushima Medical University (FMU) cohort that included 401 GC tumors ( Table 1). The FMU cohort included 27 (6.7%) cases of EBV (+) and 33 (8.2%) cases of deficient mismatch repair (dMMR) GC, confirmed by EBER-ISH and IHC staining for MMR proteins, respectively (Fig. 1a,b and Table 1) 19 . Histological evaluation identified 12 (44%) cases of lymphoepithelioma-like carcinoma, 13 (48%) cases of conventional-type, and two (7%) cases of Crohn's disease-like carcinoma ( Supplementary Fig. S1). This analysis confirmed the previous observation that tumors with EBV (+) or dMMR are mutually exclusively, and that PD-L1 is significantly overexpressed in EBV (+) GC compared with dMMR or pMMR/EBV (−) GC (Fig. 1c). Because 9p24.1 amplification is one of the specific characteristics of genetic aberrations in EBV (+) GC, cases scored highest PD-L1 CPS (> 90) were suggested to be caused by CD274 copy number aberrations. To explore the mechanism underlying PD-L1 overexpression in EBV (+) GC cases lacking CD274 amplification, we attempted to determine whether PD-L1 expression is associated with the presence of CD8 T cells. Because we previously reported that PD-L1 expression by GC cells correlates significantly with the presence of CD8 T cells in the tumor microenvironment and with expression of IFN-γ 9,10 , we first decided to investigate the correlation between PD-L1 expression and the presence of CD8 T cells using histological evaluation focusing on in EBV (+) GC ( Fig. 1a and Supplementary Fig. S2). Analysis of EBV (+) GC (n = 27) revealed that PD-L1 expression in EBV (+) GC correlated positively with CD8 + lymphocyte infiltration, although the result did not reach statistical significance (probably due to the small number of samples in the FMU cohort) (Fig. 1d). These results suggest that PD-L1 overexpression with the presence of CD8 T cells and IFN-γ which observed in GC cells were also observed in EBV (+) GC and, importantly, cases expected to harboring CD274 amplifications were independent of CD8 (+) lymphocyte infiltration.
To further confirm those findings, we analyzed a TCGA stomach adenocarcinoma tissue dataset (n = 269) and found that cases with CD274 amplification showing high expression of CD274 mRNA were common in EBV (+) GC (Fig. 1e). To examine CD274 amplification status, we determined the copy number of representative cancer-related genes that mapped telomeric and centromeric to CD274. Three cases of EBV (+) GC and one case of chromosomal instability (CIN) GC with high CD274 mRNA expression exhibited focal and high level amplification of the segment containing CD274 (Fig. 1f). By contrast, cases with relatively high expression of CD274 mRNA exhibited non-focal and arm-level copy number gain of 9p ( Supplementary Fig. S3). Next, we obtained lymphocyte infiltration data from TCGA pathological reports and attempted to confirm the positive correlation between CD274 expression and lymphocyte infiltration. Consistent with our cohort, cases with CD274 focal amplification lacked lymphocyte infiltration in both EBV (+) and EBV (−) GC, and CD274 expression was marginally associated with lymphocyte infiltration in EBV (+) GC cases without CD274 focal amplification and lymphocyte infiltration (Fig. 1g). However, CD274 expression was not associated with lymphocyte infiltration in EBV (−) GC cases without CD274 focal amplification and lymphocyte infiltration (Fig. 1g). This pathological analysis is in line with our previous studies, and suggests that there are two mechanisms underlying PD-L1 overexpression in EBV (+) GC: one due to focal amplification of CD274 and the other due to lymphocyte (CD8 T cells) infiltration 9,10 .

Histological feature of PD-L1 upregulation in EBV (+) GC
To further explore the mechanisms underlying PD-L1 overexpression in EBV (+) GC, we focused on the role of IFN-γ produced by tumor-infiltrating lymphocytes, including CD8 T cells, during virus infection 9,12 . Based on our recent finding that the IFN-γ gene signature correlates significantly with expression of CD274 mRNA in GC, we next attempted to investigate the effect of the IFN-γ signature on PD-L1/CD274 expression in EBV (+) GC 10 . Because DNA hypermethylation is a unique epigenetic feature of EBV (+) GC, we first analyzed the DNA methylation status of genes comprising the IFN-γ signature; to do this, we used microarray dataset GSE31789, in which CpG site methylation data for EBV (+) GC (n = 11) and EBV (−) GC (n = 43) are available 5 . In general, IFN-γ signature genes were not hypermethylated in EBV (+) GC (Fig. 2a). Consistent with this, de novo DNA methylation in EBV-infected GC cells (GSE31789) did not induce hypermethylation of IFN-γ signature genes (Fig. 2b) 5 . Importantly, expression of the IFN-γ signature was upregulated in EBV-infected GC cells compared to control cells, suggesting that EBV infection possibly stimulate IFN-γ production (Fig. 2c). www.nature.com/scientificreports/ These findings were further supported by gene expression data from microarray dataset GSE51575, which includes EBV (+) GC (n = 12) and EBV (−) GC (n = 14), and the TCGA study, which includes EBV (+) GC (n = 25) and EBV (−) GC (n = 249) 1,20 . The IFN-γ signature was highly expressed in EBV (+) GC compared with EBV (−) GC, with a significant positive correlation between the IFN-γ signature and CD274 mRNA expression in data from the GSE51575 (Fig. 2d) and TCGA databases (Fig. 2e). These observations are in line with our previous finding that IFN-γ upregulates PD-L1 expression and, importantly, that this mechanism plays a significant role in PD-L1 overexpression, particularly in EBV (+) GC 10 .
IFN-γ is upregulated by activation of IRF3 in EBV (+) GC. Next, we addressed the mechanism underlying regulation of IFN-γ production in EBV (+) GC. EBV-encoded RNAs (EBERs) or miRNAs induce type I IFNs in EBV-infected cells via activation of IRF3, IRF7, or NF-κB 12 . Here, we focused on IRF3 because it is a key transcription factor that drives antiviral innate immune responses via production of IFN-γ 16 .
Based on previous studies and our own results, we hypothesized that IRF3 is activated in EBV (+) GC, leading to induction of PD-L1/CD274 expression via IFN-γ. To evaluate this hypothesis, and after we confirmed that IRF3 was not hypermethylated in EBV (+) GC ( Supplementary Fig. S4), we investigated expression of IRF3 using data from the GSE51575 dataset and the TCGA study. Expression of IRF3 mRNA was higher in EBV (+) GC than in EBV (−) GC in the TCGA cohort, and expression of IRF3 correlated positively with the IFN-γ signature; however, the trend was not statistically significant ( Fig. 3a and b). Therefore, we used an IRF3-upregulated gene signature as a read-out to measure IRF3 activity in data from GSE51575 and the TCGA study. After confirming that 14 genes within the IRF3 signature were generally not hypermethylated in EBV (+) GC ( Supplementary Fig. S4), we investigated the correlation between the IRF3 signature and CD274 mRNA expression in both datasets. Consistent with a previous report, we found that the IRF3 signature correlated positively with CD274 mRNA expression in GC (Fig. 3c) 21 . The expression of IRF1, which is activated by IFN-γ resulting PD-L1 overexpression, also correlated positively with CD274 mRNA expression in GC further supported our analysis of public databases (Fig. 3c) 22 . Expression of IRF3 signature genes was higher in EBV (+) GC than in EBV (−) GC, and correlated positively (albeit marginally) with the IFN-γ signature in EBV (+) GC; however, the results did not reach statistical significance, probably due to the small number of samples in the GSE51575 cohort (Fig. 3d). Analysis of the TCGA study revealed higher expression of the IRF3 signature in EBV (+) GC than in CIN or gnomically stable (GS) cases, and a significant positive correlation with the IFN-γ signature in EBV (+) GC (Fig. 3e). These results suggest that overexpression of PD-L1 in EBV (+) GC is induced by high IRF3 activity via IFN-γ.
IRF3 is activated in EBV (+) GC. Finally, to confirm the above findings in clinical samples, we performed IHC staining of representative IRF3 signature genes in samples of FMU GC. NOXA [encoded by phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1)] was expressed more strongly in EBV (+) GC than in dMMR or pMMR/EBV (−) GC in the FMU cohort (Fig. 4a,b and Supplementary Fig. S5). In addition, glucocorticoid receptor (GR) [encoded by nuclear receptor subfamily 3 group C member 1 (NR3C1)] was also expressed more strongly in EBV(+) GC than in dMMR or pMMR/EBV (−) GC in the FMU cohort (Fig. 4a,c and Supplementary Fig. S5). These results strongly suggest that the IRF3 signature is activated in EBV (+) GC.

Discussion
Here, we demonstrated that PD-L1 overexpression in EBV (+) GC is caused mainly by two different mechanisms: CD274 focal amplification and IFN-γ-mediated signaling, including via IRF3 activation. The mechanisms regulating PD-L1 expression in GC cells have been described in several studies; additionally, we recently reported that PD-L1 expression is associated with CD8 T cell infiltration and IFN-γ production within the tumor microenvironment of GC 9,10 . Although a recent study reported that the Epstein-Barr virus nuclear antigen 2 affects PD-L1 expression in diffuse large B-cell lymphoma, the specific mechanism underlying PD-L1 expression in EBV (+) GC remains unclear 23 . Here, we reveal that high expression of PD-L1 by tumor cells in EBV (+) GC is due to high levels of CD274 focal amplification in the absence of CD8 lymphocyte infiltration. By contrast, relatively high expression of PD-L1 in tumor and infiltrating immune cells correlated with IFN-γ production via IRF3 activation. Amplification of CD274, leading to overexpression of PD-L1 in tumor cells, is an interesting characteristic of GC, particularly EBV (+) GC 24,25 . The data presented herein, along with TCGA analysis, reveal that focal and high level amplification of CD274 results in high expression of PD-L1/CD274 in EBV (+) GC tumors. This result is in line with our previous study showing that a small subset of SCLC harbors focal and high level CD274 amplification, resulting in high expression of PD-L1 8 . Our cohort and TCGA data revealed that EBV (+) GC tumors with high CD274 amplification and high PD-L1 expression were not infiltrated by CD8 + lymphocytes. These results indicate that constitutive oncogenic signaling due to chromosomal alterations and amplifications is independent of inflammatory signals in the tumor microenvironment of EBV (+) GC. Different from other GC subtypes, EBV (+) GC is associated with EBV infection; therefore, innate antiviral immune responses in EBV (+) GC contribute to PD-L1 overexpression. By contrast, EBV (+) GC with non-focal, arm-level copy number gain of 9p did not exhibit high CD274 mRNA expression when compared with that in tumors with focal amplification,   8 . Of note, although arm-level 9p gain occurs in EBV (+) GC, this was not the main structural aberration in chromosome 9p in GC; arm-level 9p loss is more common, resulting in loss of CDKN2A 26 .
In contrast to constitutive PD-L1 expression by tumor cells due to CD274 amplification, PD-L1 was induced in tumors and infiltrating immune cells in EBV (+) GC in response to adaptive immune resistance. Through this response, various cytokines (such as interleukins, TNF, and IFNs) are produced or released from infected cells to induce or maintain PD-L1 expression 12 . Among these cytokines, IFN-γ is the most potent inducer of PD-L1 expression in GC tumors 9,10 . Based on these findings, we hypothesized that IFN-γ-mediated signals are induced by IRF3, which could generate cytokines that facilitate viral infection and IFNs that inhibit viral infections. In EBV (+) GC, IRF3 is activated by phosphorylation (to yield pIRF3) through the TLR or RIG-I signaling pathways, which are triggered by EBER1, EBER2, EBV-encoded miRNAs, or other non-coding RNAs 12,13 . The fact that ISH for EBER1 is used commonly as a diagnostic test for EBV infection suggests that EBER1 is expressed at levels sufficient to stimulate the TLR or RIG-I pathways in EBV-infected cells, but expression status of others which could stimulate IRF3 expression were unknown in EBV (+) GC.
The present study focused on IRF3 activity, as measured by transcription and translational production of its target genes, and confirmed that it correlated with CD274 expression 21 . In EBV-infected tumors, IRF3 acts as key downstream transcriptional effector of inflammation and immunity thorough TLR3 and TLR4 signaling in macrophages. Using two independent cohorts, we demonstrated that the IRF3 signature tended to be higher in EBV (+) GC than in other types of GC, and that it correlated positively with the IFN-γ signature and with CD274 mRNA expression 21 . In addition, we further confirmed these findings in clinical samples by showing that NOXA/PMAIP1 and GR/NR3C1, which are components of the IRF3 signature gene set, were expressed at higher levels in EBV (+) GC than in EBV (−) GC. Of note, the IRF3 signature was also upregulated in MSI cases in TCGA (Fig. 3e) and also correlated positively with the IFN-γ signature ( Supplementary Fig. S6a). However, since genes within the IRF3 signature in MSI GC are frequently mutated, accumulation of gene mutations may contribute to activation of the IRF3 signature in MSI GC ( Supplementary Fig. S6b). To support this consideration, it was confirmed that NOXA encoded by PMAIP1 and GC encoded by NR3C1 in which IRF3 signature genes were not strongly expressed in dMMR than EBV (+) GC ( Fig. 4b and c).
Based on the results of the ATT RAC TION-2 study, immune checkpoint blockers targeting the PD-1 axis have been approved for treatment of GC in many countries, including Japan, South Korea, Singapore, and Switzerland 6,27 . Indeed, metastatic GC patients with MSI-high, high PD-L1 CPS, and EBV (+) tumors show a favorable response to anti-PD-1 mAbs used as second-or higher-line treatments 7 . As for GC, patients with other cancers showing high PD-L1 expression and a high mutational load tend to have a favorable clinical course 28,29 . Importantly, since there is a general positive correlation between EBV (+) GC and PD-L1 expression, and these patients show the best response to anti-PD-1 mAbs. However, in this study, we showed the possibility that PD-L1 overexpression associated with the presence high level CD274 focal amplification without lymphocyte infiltration would not show a favorable response to anti-PD-1 mAbs 30 . Additionally, our results recommend an EBV test to detect EBV infection for all patients with GC to prevent loss of treatment opportunity 7 . Of importance, because some EBV (+) GC cases with conventional-type adenocarcinoma exhibit well to moderately differentiated adenocarcinoma without lymphocyte infiltration and lase pattern, it is hard to detect EBV infection histologically (Supplementary Fig. S1).
In conclusion, we have identified the mechanism underlying overexpression of PD-L1 in EBV (+) GC. We show that high expression of PD-L1 in tumor cells is due to high level CD274 focal amplification without lymphocyte infiltration. On the contrary, relatively high level expression of PD-L1 in tumor and infiltrating immune cells is due to IFN-γ-mediated signaling via activation of IRF3.   Microarray data analysis. All microarray data are publicly available from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo). We utilized microarray gene methylation profiles from 54 patients with GC and EBV-infected GC cells, using Akata system of recombinant EBV, deposited as GSE31789 on the basis of Illumina Infinium HumanMethylation27 BeadChip 5 and analyzed according to our previous study 31 . We also utilized mRNA expression profiles from EBV-infected cells, deposited as GSE31787 on the basis of Affymetrix GeneChip Human Genome U133 Plus 2.0 oligonucleotide array 5 . We also utilized the expression profiles from 26 patients with GC deposited as GSE51575 on the basis of Agilent-028004 SurePrint G3 Human GE 8 × 60K Microarray 20 . The normalized expression values were obtained from each dataset and were not pro- TCGA database analysis. Copy number alteration, gene mutation, or expression data of GC patients were obtained from TCGA's cBioPortal database (http://www.cbiop ortal .org/). Those data of CD274, IFN-γ and IRF3 signature genes in GC patients were used for the analyses. For the expression data, RNA-sequencing data that was normalized by RSEM method was used for the analyses. For the clinical tumors, a multi-omics study of 295 GC patients including 269 EBV (−) and 26 EBV (+) GC was selected for the analyses 1 .
Immunohistochemical staining and evaluation. IHC staining of formalin-fixed, paraffin-embedded (FFPE) histological sections (4 μm thick) was performed using a polymer peroxidase method, as previously performed 19,32 . Briefly, after deparaffinization and rehydration, the sections were treated with 0.3% hydrogen peroxide in methanol for 30 min to block endogenous peroxidase activity. After rinsing in PBS, the sections were incubated with anti-PD-L1 antibody (#13684; E1L3N; 1:200 dilution; Cell Signaling Technology, Danvers, MA, USA), anti-CD8 antibody (1:1000 dilution; Cell signaling Technology), anti-NOXA (1:1000 dilution; Novus, Centennial, CO, USA) and anti-GR (1:400 dilution; Cell Signaling Technology) at 4 °C overnight. An additional wash in PBS was followed by treatment with a peroxidase-labeled polymer conjugated to goat anti-rabbit immunoglobulins (ENvision + kit; Dako, Agilent, Santa Clara, CA, USA) as the secondary antibody for 30 min at room temperature. The staining was visualized with diaminobenzidine, followed by counterstaining with hematoxylin. PD-L1 expression was determined by using combined positive score (CPS), which is the number of PD-L1 staining cells among tumor cells, lymphocytes, and macrophages divided by the total number of viable tumor cells. Evaluation of NOXA and GR expressions were defined as positive or negative expression and performed as previously described 19 . IHC staining for MMR protein and integration of EBV by EBER-ISH were performed, as previously described 19 .
Statistical analysis. Mann-Whitney U test were used to determine differences between two variables.
Spearman's correlation was used to evaluate the correlations between levels of expression. Pearson correlation analysis was used to examine the relationships between two parameters. ANOVA test was used for multiple comparisons. All statistical analyses were conducted using GraphPad Prism v7.0 (Graphpad Software Inc., San Diego, CA, USA) or JMP 10 software (SAS Institute, Cary, NC, USA). All P values were two-sided, and P values less than 0.05 were considered statistically significant.