The immune microenvironment features and response to immunotherapy in EBV-associated lymphoepithelioma-like cholangiocarcinoma

Limited data are available for tumor immune microenvironment (TIME) in Epstein–Barr virus (EBV)-associated lymphoepithelioma-like cholangiocarcinoma (EBV-LELCC), a rare subtype of intrahepatic cholangiocarcinoma (IHCC). We aimed to investigate TIME features in EBV-LELCC and the correlation between the components of TIME and the clinical outcomes. Tumor tissues from five EBV-LELCC cases confirmed through EBER in situ hybridization and five stage-matched conventional IHCC (non-EBV IHCC) cases were collected. These samples were used to evaluate genetic alterations, TIME composition, and PD-L1 expression through ion AmpliSeq comprehensive cancer panel, PanCancer immune profiling panel, immunohistochemistry, and immunofluorescence staining. The correlation between clinical outcomes and TIME components was analyzed in the two EBV-LELCC cases receiving anti-PD-1 treatment. The genetic mutations identified in EBV-LELCC were BARD1, CD19, CD79B, EPHA5, KDM5A, MUC6, MUC16, PTEN, RECQL4, TET1, and TNFAIP3. Both CD79B and TNFAIP3 mutations were involved in the NF-κB signaling pathway. PD-L1 was highly expressed in tumor-infiltrating immune cells, especially the T cells and macrophages. The TIME of EBV-LELCC displayed abundant immune cell infiltration with a stronger adaptive immune response. Increased Th1 cells, NK CD56dim cells, and M1 macrophages, decreased M2 macrophages, exhausted CD8 T cell infiltration, and increased T cell activation signatures in TIME were associated with longer survival. Two patients with metastatic EBV-LELCC had good disease control after anti-PD-1 antibody treatment. A significantly larger TIME component made EBV-LELCCs more sensitive to immune checkpoint blockade (ICB). A better understanding of the composition of TIME in EBV-LELCC is critical for predicting the clinical outcomes of ICB treatment.

cell activation signatures in TIME were associated with longer survival. Two patients with metastatic EBV-LELCC had good disease control after anti-PD-1 antibody treatment. A significantly larger TIME component made EBV-LELCCs more sensitive to immune checkpoint blockade (ICB). Conclusion A better understanding of the composition of TIME in EBV-LELCC is critical for predicting the clinical outcomes of ICB treatment.

Introduction
Intrahepatic cholangiocarcinoma (IHCC) is the second most common primary carcinoma of the liver and the incidence of IHCC has increased globally, with the highest incidence in East Asia [1]. The etiological factors of IHCC include biliary stone-related cholestasis and chronic viral infections, such as hepatitis B and C, which initiate the basis for chronic biliary inflammation and damage [2]. Different carcinogenetic pathways result in different histological findings in which chronic hepatitis or cirrhosis induces mass-forming peripheral small duct type IHCC, while perihilar large duct type and intraductal growth type IHCCs are highly related to chronic biliary inflammation [3]. The different morphologies are correlated with different molecular features, including genetic alternations and tumor immune microenvironment (TIME) [4]. Lymphoepithelioma-like cholangiocarcinoma (LELCC), a rare variant of IHCC, is characterized by a unique morphology with syncytial sheets of carcinoma cells and dense tumor-infiltrating lymphocytes (TILs), with no obvious desmoplasia throughout the tumor [5]. The majority of LELCC cases are from Asian populations and mostly associated with Epstein-Barr virus (EBV) infection, within the range of 55 and 73.1% [6,7]. EBV is an oncogenic herpes virus and has been increasingly identified in diverse malignancies including well-known EBV-associated nasopharyngeal carcinoma and gastric cancer (GC) [5,8]. In EBV-IHCC, EBV circular viral genomes are present in tumor cells, but not in the adjacent non-neoplastic epithelium of the bile duct [6].
Programmed death-ligand 1 (PD-L1) is a ligand targeting programmed cell death protein 1 (PD-1), which is present in both tumor cells and TILs in EBV-LELCC with increased expression of the former [6,7]. Chronic inflammatory environments resulting from virus-associated malignancies may promote PD-L1 upregulation and alter the TIME [9]. In pulmonary lymphoepithelioma-like carcinoma (LELC), two of eight top mutated genes are involved in the NF-κB pathway [10,11], which regulates immune functions and immune checkpoint expression. Herein, we presented the genetic background, composition of immune cell infiltration in patients with EBV-LELCC and identified the correlation between the components of TIME and clinical outcomes in two patients who received immunotherapy alone or in combination with chemotherapy.

Patients and data collection
The study was approved by the National Cheng-Kung University Hospital (B-ER-106-224) and Taipei Veterans General Hospital (2015-03-005BC). Five patients that underwent curative surgery with pathological diagnosis of EBV-LELCC during October 2016 and January 2018 were enrolled. For comparison, five cases of conventional IHCC (non-EBV IHCC) were selected randomly by matching with early or late stages, diagnosed during July 2013 and January 2018. Formalin-fixed, paraffin-embedded (FFPE) tumor tissue sections were collected for the following analysis. Electronic medical records (EMRs) were reviewed to obtain information about the baseline characteristics of age, initial tumor size, stage, history of hepatitis B and C, and the date of last follow-up visit. Overall survival (OS) was defined as the duration from the time of pathological diagnosis to the date of death, and disease-free survival (DFS) was defined as the time of cancer recurrence, second cancer, or death from any cause obtained from the EMRs.

Sample processing and Ion AmpliSeq Comprehensive Cancer Panel (IACCP)
Manual macrodissection was performed to isolate tumor tissues from FFPE sections in five cases of EBV-LELCC for DNA extraction using QIAmp DNA FFPE Tissue Kit (Qiagen, Hilden, Germany). DNA was amplified using Ion AmpliSeq Comprehensive Cancer Panel (IACCP, Life Technologies, Carlsbad, CA, USA) consisting of 409 cancerrelated genes. The quality and quantity of extracted DNA and the amplified library were determined using fragment analyzer (Advanced Analytical Technologies, Inc.) and Qubit (Invitrogen, Carlsbad, CA, USA). Sequencing was conducted on ion proton sequencer using an ion P1 chip (Life Technologies) according to the manufacturer's protocol, with a mean sequencing depth of > 800×. Sequencing data of germline mutations from peripheral blood mononuclear cells (PBMCs) were available for two EBV-LELCC cases and applied to filter out putative germline mutations. For three cases without paired PBMC sequencing data, variants with VAF (variant allele frequency) of less than or around 10% were considered as somatic mutations.

RNA extraction, nCounter ® PanCancer Immune Profiling Panel (PIPP), and quantitative real-time polymerase chain reaction (qPCR)
RNA was extracted from the dissected tumor tissues of five non-EBV IHCC and five EBV-LELCCs using RNeasy FFPE Kits (Qiagen) and applied to the NanoString PIPP, a panel assessing expression of 770 genes, according to the manufacturer's instructions. Genes defined as cell type specific were used to calculate immune cell type scores. The qPCR assay was applied for PDCD1 and CD274 mRNA level detection. GAPDH was used as an internal control.

Hematoxylin and eosin (H&E) staining, immunohistochemistry (IHC) staining, and PD-L1 scoring
FFPE tissues of EBV-LELCC were cut into 5-mm-thick sections and stained using H&E or primary antibodies including PD-L1, CK7, CD20, CD56, CD68, CD3, CD4, and CD8 by IHC analysis. PD-L1 expression was determined using tumor-positive score (TPS) as follows: the number of positive tumor cells divided by the total number of viable tumor cells multiplied by 100% as well as using combined positive score (CPS) as follows: the number of positive tumor cells, and mononuclear cells divided by the total number of viable tumor cells multiplied by 100.

Statistical analysis
All analyses were performed using GraphPad Prism version 5.0 (GraphPad Software, La Jolla, CA). A Pearson's correlation test was carried out to compare the concordance analysis of PD-L1 IHC score evaluated by two independent pathologists and relationships between CD274 mRNA and PD-L1 protein levels or mRNA level of PDCD1 and CD274. The student's t test was used for comparison between two groups. The two-way ANOVA analysis were performed for qPCR. Data of all the experiments are expressed as the mean ± standard error. Statistical significance was set at p < 0.05. For further details regarding the materials and technique used, please refer to the supplementary information.

Clinical data and pathological findings
The clinical characteristics and survival status of non-EBV IHCC and EBV-LELCC are summarized in Table 1. For the five EBV-LELCC cases, the mean age was 61.4 years (range, 41-75 years), which was comparable to non-EBV IHCC cases (median age of 60.2 years). All EBV-LELCC patients including three with pathological stage I and two with stage III/IV, according to the American Joint Committee on Cancer 7th edition, underwent surgery. Four EBV-LELCC patients still survived on data cutoff. Only one patient suffered from chronic viral hepatitis B infection. The histological findings of the lymphoepithelioma pattern showed sheets of glandular tumor cells with prominent nucleoli, a fair amount of cytoplasm, and dense lymphoplasma cell infiltration (Fig. 1a). The tumor cells, but not the background inflammatory cells, were positive for EBER in situ hybridization in all EBV-LELCC cases.

Genetic alteration identified by NGS in EBV-LELCC
Due to the diffuse distribution of TILs over the peritumor area and the difficulty of clearly separating glandular tumor cells from TILs, a high proportion of identified variants were germline mutations with more than 50% VAF (Supplementary Table S1 and S2). After germline subtraction and stringent filtration steps, a total of ten somatic mutations were identified in four patients ( Table 2). Among them, one PTEN inactivating mutation (p.Q17*) was identified in case E + 5, leading to increased PI3K/AKT/mTOR signaling. Other deleterious or likely deleterious mutations as predicted by the in silico analysis and affected pathways included CD79B p.D202A (NF-κB signaling), CD19 p.R163C (B cell receptor signaling), TET1 p.R155G (epigenetic regulation), and

Signaling pathways enriched in EBV-LELCC
Five samples each of non-EBV IHCC and EBV-LELCC were extracted for PIPP. As shown in Fig. 1b 137 differentially expressed genes (DEGs) were identified, including 112 upregulated genes and 25 downregulated genes, in EBV-LELCC compared to non-EBV IHCC. The expression levels of T cell markers, including CD3D, CD3E, and CD8A, and adaptive immune resistance-related genes, such as PDCD1 (PD-1) and CD274 (PD-L1), were enhanced in EBV-LELCC. To understand the major pathways involved in EBV-LELCC, we utilized GO, KEGG, Hallmark, and Reactome databases for enrichment analysis of DEGs. The histograms of the top 40 significantly enriched terms are shown in Fig. 1c, in which the pathway adaptive immune response/system was present in both GO and Reactome databases. Furthermore, GSEA of PIPP data revealed that the genes involved in primary immunodeficiency, adaptive immune response, and adaptive immune system were highly positively enriched in EBV-LELCC (Fig. 1d).

Differences in immune cell types between non-EBV IHCC and EBV-LELCC
We further investigated the major components of innate or adaptive immunity in EBV-infected immune responses between non-EBV IHCC and EBV-LELCC by scoring the corresponding cell type-specific genes based on PIPP. For innate immune response, the scores of mast cells (p = 0.8245, data not shown), neutrophils (p = 0.9745, data not shown), dendritic cells (DC), natural killer (NK) cells, NK CD56 dim cells, macrophages, M1 macrophages, and M2 macrophages were determined. Among these, only NK CD56 dim cell (p = 0.0181) and M2 macrophage (p = 0.0304) scores were found to be higher in EBV-LELCC than in non-EBV IHCC ( Supplementary Fig. S1a). In contrast, most immune cell populations involved in adaptive immune responses were generally scored, and were significantly higher in EBV-LELCC than in non-EBV IHCC; such immune cell popula-

Expression of PD-L1 and immune cell markers in EBV-LELCC
Based on PIPP data, EBV-LELCC was associated with significantly larger number of immune components of the tumor microenvironment that were related to adaptive immune resistance than that of non-EBV IHCC. We performed IHC or IF staining of PD-L1, tumor cell marker cytokeratin-7 (CK7), and immune cell markers, including CD20, CD56, CD68, CD3, CD4, and CD8 in continuous slides to verify the significantly increased proportion of immune cells in EBV-LELCC and examine which cell types contribute to the development of an immunosuppressive microenvironment. Figure 2a, b shows that CD20 + B cells, CD56 + NK cells, CD68 + macrophages, and CD3 + , CD4 + , and CD8 + T cells migrated to EBV-LELCC. The representative IHC and IF images revealed that PD-L1 was highly expressed in T cells and macrophages compared to tumor cells or the above-mentioned immune cells (Fig. 2a-c).

Association between EBV status and the expression of PD-1 and PD-L1
PD-L1 IHC-assessed scores by two independent pathologists showed high consistency (r = 0.8358, p = 0.0026; Fig. 3a). Consistent with Fig. 2, the distribution of PD-L1 expression by CPS and TPS results showed high CPS, but low TPS in all five cases with EBV-LELCC (Fig. 3b and c). Indeed, PD-L1 expression is mainly positive in immune cells. Furthermore,  Fig. 3d). CD274 mRNA expression levels were strongly correlated with PDCD1 mRNA levels in both non-EBV IHCC and EBV-LELCC (r = 0.8205, p = 0.0036; Fig. 3e). PDCD1 and CD274 mRNA levels were significantly increased in EBV-LELCC compared to non-EBV IHCC (Fig. 3f), indicating that EBV infection is associated with the expression of PD-1 and PD-L1.

Major components of TIME and clinical outcome in EBV-LELCC
In EBV-LELCC, since the PD-L1 expression was highest in immune cells; abundant in Th1 cells, NK CD56 dim cells, M2 macrophages, and CD8 T cells; and exhausted in infiltrated CD8 T cells, we became interested in finding out whether dynamic interactions between these immune cells and cancer cells could evolve tumor progression and immune evasion. The major cellular components of immunity in the two EBV-LELCC cases (E + 3 and E + 5) showed higher scores for Th1 cells, NK CD56 dim cells, and M1 macrophages. However, lower scores for M2 macrophages and exhausted CD8 T cells were observed in patient E + 5 than in patient E + 3 (Fig. 4a). The mRNA levels of T cell activation signatures including IL-1B, IL-2, IL-12A, IL-12B, TNF, and IFNG [12] were increased in patient E + 5 (Fig. 4b). As compared to patient E + 3, patient E + 5 had a longer DFS time (29.9 vs. 6.2 months, Fig. 4c) and slow growth rate of aortocaval lymph node metastasis after 2 years follow-up of abdominal computed tomography (CT) scan (2.5-2.7 cm, data not shown). Patient E + 3 showed progressive disease with multiple mediastinal lymphadenopathies and bilateral pleural effusion after 6 months by CT scan (the largest nodes were 5 and 2.5 cm in size; Fig. 4d and e, left panel).

Response of EBV-LELCC to anti-PD-1 monoclonal antibody treatment
Patient E + 3 was a 68-year-old woman who underwent operation with a diagnosis of stage IVA EBV-LELCC. Due to tumor recurrence, she started receiving chemotherapy with gemcitabine plus cisplatin. However, multiple enlarged mediastinal lymph nodes with bilateral pleural effusion developed later (Fig. 4d and e, left panel). We changed systemic therapy to nivolumab plus modified gemcitabine and TS-1 (NGS) every 2 weeks [13]. After NGS treatment for 3 months, the multiple metastatic lymphadenopathies obviously decreased bilateral pleural effusion improved which lasted for almost 1 year (Fig. 4d and e, right panel). Patient E + 5 was a 75-year-old man with a pathological stage III EBV-LELCC received chemotherapy after recurrence. Due to enlarged lymphadenopathy in the aortocaval space (Fig. 4f, left panel), he received four cycles of nivolumab monotherapy. Three months later, the post-treatment abdominal CT scan revealed decreased size of the lymphadenopathy with hypodense areas caused by necrosis in the paraaortic and retrocaval regions (4.9-3.8 cm and 3.3-2.8 cm, respectively) (Fig. 4f, right panel). The patient had stable disease under nivolumab treatment for an extended period of 3 or 4 weeks due to a financial issue.

Discussion
The pathological diagnosis of LELCC depends on its morphologic features, including abundant TILs and tumor-associated stroma, but without obvious desmoplastic reaction throughout the tumor [5]. Patients with EBV-LELCC had a significantly longer OS compared to non-EBV IHCC, indicating the unique genetic background and TIME in EBV-LELCC [5]. In this study, we compared EBV-LELCC with non-EBV IHCC in terms of TIME and evaluated the correlation between TIME and prognosis in EBV-LELCC. Higher PD-L1 expression in immune cells, mostly T cells and macrophages, and more activated adaptive immune responses were identified in EBV-LELCC than in non-EBV IHCC. The prognosis of EBV-LELCC appeared to be correlated with the composition and proportion of TIME, in which higher scores of Th1 cells, NK CD56 dim cells, and M1 macrophages and lower scores of M2 macrophages and exhausted CD8 T cells were indicators of a relatively hot tumor. Patients with EBV-LELCC experienced tumor shrinkage after treatment with anti-PD-1 monoclonal antibody. Thus, we can consider promoting immunotherapy for the treatment of EBV-LELCC as monotherapy or in combination with chemotherapy.
Approximately, 12% of all cancers are associated with viral infections worldwide [14]. Several viruses have been reported to be involved in the development of malignancies, Fig. 2 Detection of PD-L1 in cancer cells and immune cells from patients with EBV-LELCC. a, b H&E staining and in situ IHC staining for PD-L1, CK7, CD20, CD56, CD68, CD3, CD4, and CD8 in two patients with EBV-LELCC (E + 4 and E + 5). CK7 was used for identification of cancer cells; CD20 for B cells; CD56 for NK cells; CD68 for macrophages; CD3, CD4, and CD8 for T cells. Original magnification, ×200; scale bars, 100 μm. c IF staining for CD3 or CD68 (green), PD-L1 (red), and nuclear (4′,6-diamidino-2-phenylindole, DAPI; bule) staining in serial sections from patient E + 4. Expression of PD-L1 (red) and co-localization with CD3 (green, middle panel) or CD68 (green, right panel) and distribution is shown by overlapping signal (yellow). The white dotted lines indicate the PD-L1 high-expression regions. Original magnification, ×400; scale bars, 50 μm. EBV Epstein-Barr virus, LELCC lymphoepitheliomalike cholangiocarcinoma, PD-L1 programmed death-ligand 1, CK7 cytokeratin-7, NK cells natural killer cells, H&E hematoxylin and eosin, IHC immunohistochemistry, IF immunofluorescence ◂ including EBV, human papilloma virus (HPV), and hepatitis B/C viruses (HBV/HCV) [15]. In non-EBV IHCC, TP53, KRAS, IDH1, BAP1, ARID1A, and FGFR2 are the most frequently mutated genes [16,17], but none of the above genetic mutations were identified in EBV-LELCC cases of our study. In another study, whole exome sequencing revealed only non-silent mutations of MUC4, DNAH1, GLI2, LIPE, MYH7, RP11-766F14.2, and WDR36, which did not overlap with those of our study [6]. The likely deleterious TNFAIP3 mutation was found in one of our EBV-LELCCs that may cause dysregulated NF-κB signaling pathway as previously reported in pulmonary LELC could partially explain the inflamed and immune-responsive microenvironment [10,11]. The other recognized CD79B mutation also involved in B-cell receptor-dependent activation of NF-κB was frequently reported in diffuse large B cell lymphoma and primary central nervous system lymphoma [18,19], indicating the dominant role of NF-κB signaling in EBV-LELCC. Virus-associated cancers, such as HPVassociated head and neck cancer and EBV-GC, usually show less genetic alternations with a high degree of homogeneity and hypermethylation compared to those without virus infection [8,20]. Chan et al. showed that gene hypermethylation was more frequent in EBV-LELCC than non-EBV IHCC (85.7% vs. 9.1%, p < 0.003) [5]. More frequent pTERT and TP53 mutations were found in EBV-negative LELCC than in EBV-LELCC, supporting that EBV-negative LELCC was more likely to be hepatocellular carcinoma or combined hepatocholangiocarcinoma genetically [21,22]. Altogether, EBV-LELCC harbored a different mutational profiling compared to non-EBV IHCC and EBV-negative LELCC due to different etiologies and pathogenesis.
There are some clinical trials applying anti-PD-1 or anti-PD-L1 monoclonal antibodies for the treatment of BTC after failure to previous chemotherapy, with 5.8-17% of objective response rate [23]. Effective immune biomarkers for  patient selection are important in the era of precision medicine [24]. In GC, PD-L1 expression and EBV positivity have also been identified as potential biomarkers for the benefit of immunotherapy [25]. In EBV-GC, frequent amplification of CD274 and higher immune scores were noted compared to non-EBV GC [8]. There was no CD274 amplification in EBV-LELCC samples by IACCP analysis in our study, and only 5% samples (1 of 20 EBV-IHCCs) were positive as evidenced by FISH in another study [6], suggesting the different mechanisms of PD-L1 overexpression in virus-associated malignancies. Some reports have shown that PD-1 expression is upregulated in T cells during chronic infections and that PD-1 expression is related to T cell exhaustion [26]. Our results revealed PD-L1 expression on both tumor cells and TILs with higher CPS than TPS, indicating increased expression in TILs, especially in T cells and macrophages. TIME was significantly related to EBV infection in GC and IHCC [6,8]. EBV-GC was considered as a hot tumor due to the overexpression of multiple cytokines, chemokines, and HLA genes, which is likely to have inherently strong immune reactions [27]. In EBV-IHCC, a subpopulation of tumor microenvironment types (TMITs), TMIT I, defined by PDL1-tumor + /CD8 high was more common in EBV-IHCC than in non-EBV IHCC (90.0% vs. 12.4%), which correlated with the best OS compared to TMIT III (PDL1-tumor + / CD8 low ) [6]. Here, we found the activation of adaptive immune response in TIME of EBV-LELCC with enriched PD-1 and PD-L1 expression and abundant Th1 cells, NK CD56 dim cells, M2 macrophages, CD8 T cells, and exhausted CD8 T cells infiltration, indicating that EBV-LELCC reflects adaptive immune resistance and may be more sensitive to immunotherapy. In addition, the composition of TIME was associated with DFS in the two EBV-LELCC cases. The patient E + 5 had longer DFS, higher scores for Th1 cells, NK CD56 dim cells, and M1 macrophages, but lower scores for M2 macrophages and exhausted CD8 T cells than patient E + 3, suggesting that the components of TIME may predict clinical course and tumor behavior in EBV-LELCC and can potentially be exploited to enhance the efficacy of immunotherapy.
Because EBV-LELCC is a rare subtype of IHCC, only five samples were enrolled in this study. Further studies with larger sample sizes are needed to validate our results. Furthermore, advanced whole-exon sequencing or wholegenome sequencing is needed to elucidate the pathogenic mechanisms of EBV in carcinogenesis.

Conclusion
EBV-LELCC is a distinct variant of IHCC with dense infiltration of immune cells and different genetic backgrounds and TIMEs. The characteristics of TIME are correlated with clinical outcomes including PD-L1 overexpression in T cells and macrophages, the composition of immune cells, and the degree of T cell activation signatures. Further basket clinical trials of immunotherapy with TIME analysis would be a promising approach to improve the therapeutic outcomes of virus-associated malignancies.