Formalin-fixed paraffin-embedded tissues from patients with EBVaGC (70 cases) and EBVnGC (62 cases) were collected between January 2010 and December 2018 at Third Affiliated Hospital of Sun Yat-Sen University. Clinical data were retrieved from patients’ medical records. Overall survival (OS) time was determined from the date of surgery to the date of death or the last follow-up visit. This study was approved by the Institutional Review Board of Third Affiliated Hospital of Sun Yat-Sen University.

The human GC cell lines AGS and BGC823 were obtained from the Cell Bank of Type Culture Collection of Chinese Academy of Sciences (Shanghai, China). The human BL cell line Akata-EBV-GFP was provided by Prof. Mu-sheng Zeng from Cancer Center, Sun Yat-Sen University. Akata-EBV-GFP cells were modified to produce recombinant EBV, which were resistant to neomycin and could express green fluorescent protein (GFP) (Shimizu et al., 1996). EBV-infected cells (AGS-EBV and BGC823-EBV) were obtained by co-culture AGS or BGC823 with Akata-EBV-GFP using the cell-to-cell infection method as described before (Imai et al., 1998). G418 (400 μg/mL) was used to select and maintain the EBV-infected cells. The infected cells were observed for GFP expression under a fluorescence microscope (Nikon, Japan) and flow cytometry analyses. The above cells were maintained in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS; Gibco). All the cells were cultured at a humidified incubator with 5% CO₂ at 37°C. All cell lines were routinely tested for mycoplasma contamination.

Total RNA was extracted from cultured cells using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. RNA was reverse transcribed to cDNA using RT Premix (Takara, Tokyo, Japan). Then, qRT-PCR was performed on cDNA samples using SYBR Premix Ex Taq (Takara). The relative mRNA levels of target genes were normalized to GAPDH mRNA levels, and the comparative Ct (ΔΔCt) method was used. The sequences of the specific primers used in this study are shown in Supplementary Table S1 .

Total RNA was extracted from cultured cells using TRIzol reagent (Invitrogen). RNA quality was evaluated using the Agilent 2100 bioanalyzer. RNA-seq libraries were prepared using a NEBNext Ultra II Directional RNA Library Prep kit according to the manufacturer’s instructions. The reads were first mapped to the latest UCSC transcript set using Bowtie2 version 2.1.0, and the gene expression level was estimated using RSEM v1.2.15. Trimmed mean of M-values (TMM) was used to normalize the gene expression. Differentially expressed (DE) genes were identified using the edgeR program. Genes showing altered expression with P<0.05 and more than 1.5-fold changes were considered differentially expressed.

GO analysis (http://www.geneontology.org) was performed to set up gene annotations. DE transcriptome was classified into GO terms, including biological process (BP), cellular component (CC), and molecular function (MF). Fisher’s exact test was applied for the GO analysis with significant P-value calculated, and FDR was used to correct the P-values.

KEGG (https://www.kegg.jp) pathway analysis was adopted to describe the genes’ attributes. We turn to the Fisher’s exact test to select the significant pathway, and the threshold of significance was defined by FDR < 0.05.

Total proteins were extracted using RIPA lysis buffer (Beyotime, Haimen, China) according to the manufacturer’s protocol. Twenty micrograms of total proteins were separated on 10% SDS-PAGE gels. The proteins were transferred to PVDF membranes and then probed with primary specific antibodies overnight. Phospho-p65 (Ser536) (3033, 1:1000), p65 (8242S, 1:1000), phospho-IκBα (Ser32) (2859,1:1000), IκBα (4814, 1:1000), β-Actin (4970, 1:2000) were obtained from Cell Signaling Technologies (Beverly, MA, USA). After incubated with HRP-conjugated secondary antibodies (Beyotime) for 1 h, the signals were detected by electrogenerated chemiluminescence (ECL) detection reagents (Millipore). Relative target protein expression levels were normalized to β-Actin and visualized using ImageJ software.

Briefly, cells were cultured in six-well plates at a density of 5 × 10⁵ cells/well, and cell supernatants were harvested at 48 h and microfuged at 1500 rpm for 5 min to remove particles, and the supernatants were frozen at −80°C until use. The concentration of CXCL8 in the cell supernatants was determined by ELISA (R&D Systems, Abingdon, UK), using the standard curve method, according to the manufacturer’s instructions. Absorbance was measured at 450 nm using a microplate reader.

A total of 40 µl Matrigel (BD Biosciences, USA) was added into a 96−well plate and cultured at 37˚C in 5% CO₂ to polymerize. 5 × 10⁴ cells were added to the Matrigel layer and cultured with different treatments. Following 12 h of incubation for GC cells, tubules were photographed under a microscope (Nikon, Japan) and evaluated using ImageJ software. In treated groups, AGS and BGC823 cells were pretreated with BAY 11-7082 (5µM; MedChemExpress, NJ, USA) or BMS345541(5μM; MedChemExpress, NJ, USA) for 6 h.

Cell proliferation was evaluated using the EdU kit (Beyotime) according to the manufacturer’s instructions. Briefly, 1×10⁵ cells were seeded in 12-well plates. Cells were treated with EdU reagent (10 μM) and incubated for 2 h at 37°C. The EdU+ cells were observed under a fluorescence microscope (Nikon, Japan) and evaluated using ImageJ software.

Wound healing migration assay was performed as previously described (Molinie and Gautreau, 2018). Briefly, cells were cultured in a six-well plate at a density of 5 ×10⁵ cells/well. After confluence, the AGS and BGC823 were scratched by a straight line using a sterile pipette tip. After that, PBS was utilized to wash cells for removing cell debris three times. The photographs of the scratch wound were recorded at 12, 24, and 36 hours to investigate and analyze the cell migration ability. An inverted microscope (Nikon, Japan) was used to obtain digital photographs. The scratch area was measured using the ImageJ software.

Immunohistochemical (IHC) staining was performed on 4 μm-thick sections of formalin-fixed paraffin-embedded tissues with an anti-CXCL8 antibody (HPA057179, 1:100; Sigma–Aldrich, MO, USA), anti-CD34 antibody (ab81289, 1:500; Abcam, Cambridge, UK), and anti-Ki67 antibody (9449S, 1:500; Cell Signaling Technology). For CD34/PAS double staining, which was used to examine VM structures, after IHC staining for CD34 described above, the sections were washed with running water for 5 min, incubated with PAS for 20 min, and counterstained with hematoxylin. All IHC slides were analyzed independently by three experienced pathologists. The CXCL8 IHC staining results were scored as follows: staining intensity score, 0 (no staining), 1 (weak), 2 (moderate), or 3 (strong); staining area score, 0 (≤10%), 1 (11–50%), 2 (51–75%), and 3 (≥75%). The staining intensity score and staining area score were then multiplied to yield a final score. Finally, it was divided into two groups: low expression (final score ≤ 3) and high expression (final score > 3).

siRNAs were synthesized by Guangzhou RiboBio Co., Ltd., and the sequences were as follows: siRNA targeting RPMS1#1: 5′-GGCAAGGUCCGGCGUGUCCACdTdT-3′; siRNA targeting RPMS1#2: 5′- UCGUCGACGGGCAAGGUCCGGdTdT-3′; siRNA targeting RPMS1#3: 5′-GACGGGCAAGGUCCGGCGUGUdTdT-3′. siRNA targeting CXCL8#1: 5′-CTTAGATGTCAGTGCATAA-3′; siRNA targeting CXCL8#2: 5′-GTCAGTGCATAAAGACATA-3′. Lipofectamine RNAiMAX (Invitrogen) was used for siRNA transfection according to the manufacturer’s instructions.

In situ hybridization (ISH) assay was performed with a commercially available EBV oligonucleotide probe complementary to EBER-1 (PanPath, Amsterdam, Netherlands), as previously described by Chen et al. (2010). Sections from a known EBER-1-positive NPC tissue were used as the positive controls and a sense probe for EBER-1 was used as the negative control.

Female BALB/c nude mice (4–6 weeks old) were obtained from GemPharmatech Laboratory (Nanjing, China). For the xenograft model, 5 × 10⁶ cells were injected into the subcutaneous tissue of the mouse (n = 5 per group). All mice were sacrificed after four weeks, and the xenografts were fixed with phosphate-buffered formalin and sectioned for H&E staining and immunohistochemical analysis. All animal studies were performed in accordance with the institutional ethics guidelines for the animal experiments, which were approved by the Experimental Animal Ethics Committee of the Third Affiliated Hospital, Sun Yat-sen University.

Statistical analyses were performed using SPSS (version 22.0; IBM Corp., Armonk, NY, USA). Spearman’s correlation test or unpaired Student’s t test was used when appropriate. Survival analyses were performed using the Kaplan-Meier method and a log-rank test. Data with a non-Gaussian distribution were compared with a two-tailed Mann–Whitney test. A p value less than 0.05 was considered statistically significant.

To investigate the role of EBV in gastric carcinogenesis, we first established EBV-infected GC cell lines as previously described (Yue et al., 2019). Two GC cell lines, AGS and BGC823, were infected with recombinant EBV derived from the BL cell line Akata. Upon G418 selection (400 μg/ml), the GFP fluorescence was observed via fluorescence microscopy and flow cytometry analyses ( Figures 1A, B ). EBER in situ hybridization confirmed the presence of EBV in the majority of cells in the EBV-infected cells ( Figure 1C ). To gain mechanistic insights into EBV carcinogenesis, RNA-seq analysis was performed in AGS and AGS-EBV cells. Here, we focused on cytokines expression, so we compared the expression profiles of the 92 known cytokines between AGS and AGS-EBV cells. In total, 21 DE cytokines (fold change ≥ 2) were identified and CXCL8 displayed the high expression in AGS-EBV group ( Figure 1D ). GO enrichment analysis of the DE cytokines between AGS and AGS-EBV cells showed that these cytokines involved in processes including NF-kappaB (NF-κB) signaling and type I interferon signaling pathways ( Figure 1E ). KEGG analysis showed that the DE cytokines were enriched in biological processes of immune response and chemotaxis ( Figure 1F ). These findings indicated that the change of cytokines expression in AGS-EBV might contribute to anti-virus immunity, and NF-κB signaling pathway might drive EBV carcinogenesis.

The expressions of CXCL8 in the paired EBV-positive and EBV-negative cells (AGS-EBV vs. AGS and BGC823-EBV vs. BGC823) were verified by qRT−PCR analysis. CXCL8 mRNA expression was significantly higher in EBV-positive cells than in EBV-negative cells ( Figure 1G and Figure S1 ). Quantitative ELISA showed that CXCL8 concentration in the supernatant of EBV-infection cells was higher than that in parental cells ( Figure 1H ). Western blotting results showed that the expression of CXCL8 is higher in EBV-infective cells than that in parental cells ( Figure S2 ). Moreover, we performed qRT−PCR and ELISA assays using EBV-positive GC cell lines (SNU719 and YCCEL1 cells), the results showed that the level of CXCL8 was significantly upregulated in SNU719 and YCCEL1 cells compared to AGS and BGC823 cells ( Figure S3 ). Taken together, these data showed that EBV infection led to the upregulation of CXCL8 in GC cells.

In addition to the well-known chemotaxis effects on immune cells, CXCL8 has exhibited its promoting effects on angiogenesis in pancreatic cancer (Matsuo et al., 2009; Pausch et al., 2020) and VM formation in glioblastoma (Angara et al., 2018). Therefore, the association between CXCL8 and VM formation in GC was evaluated. As shown in Figure 2A and Figure S4 , the addition of recombinant human CXCL8 (2 ng/ml, 12 h) promoted VM formation in GC cells, similar to the addition of conditioned medium of AGS-EBV (EBV-CM). The decrease of CXCL8 by neutralization CXCL8 antibody (Nab-CXCL8, 1 μg/ml) in GC cells significantly decreased the number of tubes formed on Matrigel. Furthermore, we performed PAS staining to observe the tubular structures between EBV-negative and EBV-positive cells. PAS staining results showed that EBV-positive cells exhibited more VM structures than EBV-negative cells ( Figure S5 ).VM formation is associated with cell proliferation and migration; thus, the effects of CXCL8 on cell proliferation and migration in GC cells was further assessed. The proliferation of AGS and BGC823 cells was increased both in the CXCL8 and EBV-CM groups, but decreased in the anti-CXCL8 group ( Figure 2B ). Wound healing migration assays revealed that the addition of recombinant human CXCL8 or EBV-CM increased the cell migration ability of AGS and BGC823 cells. The effects were reversed by being treated with anti-CXCL8 ( Figure 2C and Figure S6 ). Furthermore, knockdown of CXCL8 with siRNA blocked VM formation as well as cell proliferation and migration ( Figure S7 ). These results indicated that EBV-upregulated CXCL8 promoted VM formation of GC cells.

CXCL8 has been revealed to bind to its receptor and then trigger many downstream signaling cascades, among which NF-κB signaling is associated with tumor VM formation (Zhao et al., 2017). As aforementioned, we demonstrated that the DE cytokines between AGS and AGS-EBV cells were enriched in NF-κB signaling pathway ( Figure 1E ). Therefore, we detected the molecules in the NF-κB singling pathway by Western blotting. As shown in Figure 3A , the expressions of p-IκBα and p-p65 in AGS and BGC823 cells were upregulated with the treatment of recombinant human CXCL8 as well as EBV-CM, whereas the expressions of total IκBα and p65 in AGS and BGC823 cells remained unchanged. When treated with BAY 11-7082(5 μM), an inhibitor of NF-κB, the expression of p−p65 and p-IκBα were downregulated in CXCL8-treated AGS and BGC823 cells ( Figure 3A ). In accordance with these observations, both recombinant human CXCL8 and EBV-CM increased the number of Matrigel VM formation of EBV-positive cells, which was reversed by inhibition of NF-κB signaling ( Figure 3B ). Besides, we investigated the effect of the NF-kB inhibitor BAY 11-7082 on CXCL8 production in YCCEL1, AGS-EBV, and BGC823-EBV cells, and found that the CXCL8 showed a significant reduction in BAY 11-7082 group ( Figure S8 ). We also performed another NF-kB inhibitor (BMS345541) to support the involvement of the NF-kB pathway in the capacity of CXCL8 to regulate VM formation, and found that the expression of p-IKKα/β, p−p65, and p−IκBα was inhibited with the treatment of inhibitor BMS345541 ( Figure S9A ). In addition, the number of tubules in the EBV-CM group was significantly increased compared with the control groups and significantly decreased when treated with BMS345541 ( Figure S9B ). This indicated that NF-κB signaling participated in VM formation induced by CXCL8 in GC. In addition to NF-κB signaling, previous studies also reported that ERK phosphorylation and STAT3 phosphorylation might be involved in the CXCL8 expression (Huang et al., 2017; Bae et al., 2019). Therefore, we also detected the ERK phosphorylation and STAT3 phosphorylation levels in EBV-infected GC cells and their parental cells. The results showed that both levels in EBV-positive GC cells were similar to the EBV-negative control cells ( Figure S10 ). Additionally, EBV-infection did not affect VEGF expression, indicating that VEGF did not contribute to the EBV-induced VM. These findings indicated that NF-κB signaling participated in VM formation induced by CXCL8 in GC.

To further confirm the promoting effects of CXCL8 on VM formation and explore the underlying mechanisms, we investigated whether CXCL8 could regulate the expressions of VM-promoting genes MMP1, MMP9 and Twist. We found that MMP1, MMP9, and Twist expressions were higher in AGS cells treated with recombinant human CXCL8 or EBV-CM ( Figure S11 ).

Having noted the role of EBV-induced CXCL8 in promoting VM, we next tried to identify the viral genes that contribute to VM formation. We first examined the EBV-encoded gene expression in EBV-infected GC cell lines. qRT-PCR results suggested that LMP1, EBNA3C, BHLF1, and BZLF1 were not expressed in AGS-EBV cells, and RPMS1, EBNA1, LMP2A, ebv-circRPMS1, ebv-circLMP2A, and ebv-circBHLF1 were all expressed in AGS-EBV cells, among which RPMS1 transcript was the highest ( Figure 4A ). Among these latent genes, RPMS1, an EBV BART Long Non-coding RNAs, have been implicated in cytokine expression (Verhoeven et al., 2019). Upon transfection of RPMS1 siRNA into AGS-EBV cells, qRT-PCR showed a significant reduction of RPMS1 and CXCL8, but no effect on other EBV latent genes (circRPMS1, EBNA1, and LMP2A) ( Figures 4B, C and Figure S12 ). ELISA also indicated CXCL8 decreased in EBV-positive GC cells following RPMS1-knockdown ( Figure 4D ). More importantly, knockdown of RPMS1 with siRNAs reduced the tube-forming capacities of EBV-infected GC cells, which could be reversed by being treated with CXCL8(2 ng/ml) ( Figure 4E ). Furthermore, RPMS1 knockdown in EBV-positive cells inhibited cell proliferation and migration, which was also reversed by treatment with CXCL8 ( Figures 4F, G , and Figure S13 ). These results suggested an important role of RPMS1 in EBV-induced VM formation.

Then, we investigated how RPMS1 regulates CXCL8 expression. ChIP analysis with anti-H3K27me showed a higher enrichment at the promoter region (−49) of the CXCL8 gene, but this higher enrichment was not observed at the end of the CXCL8 gene (+2918) ( Figure S14 ). Besides, knockdown of RPMS1 resulted in significantly less H3k27me expression ( Figure S15 ). These findings suggested that RPMS1 could downregulate CXCL8 expression by interfering with H3K27me-mediated transcription.

To evaluate the VM formation in vivo, EBV− and EBV+ AGS cells were injected subcutaneously into nude mice. The xenografts images were taken four weeks post-implantation ( Figure 5A ). Compared with the parental cells, AGS cells infected with EBV dramatically exacerbated tumor growth ( Figures 5B, C ). Double staining CD34/PAS showed that the tumors derived from EBV+ cells exhibited more VM structures than those derived from EBV- cells ( Figure 5D ), suggesting that EBV infection promoted the VM formation and tumorigenicity of GC cells.

We also investigated the correlation between VM and CXCL8 formation in GC tissues. The VM channels positive for PAS and negative for CD34 were lined with EBER-positive tumor cells and contained red blood cells ( Figure 6A , top panel). CXCL8 was mainly located in the nucleus of tumor cells. However, cytoplasm and membranous positivity could also be observed. As shown in Figure 6A , EBVaGC showed stronger CXCL8 immunostaining than EBVnGC. In addition, the PAS+CD34− GC tumor cells expressed high levels of CXCL8 in EBVaGC ( Figure 6B ). Furthermore, high CXCL8 expression was associated with worse OS in GC patients ( Figure 6C ). These data confirmed the existence of VM in GC clinical samples and further demonstrated the association between EBV infection, VM formation, and CXCL8 expression.