Decitabine disrupts EBV genomic epiallele DNA methylation patterns around CTCF binding sites to increase chromatin accessibility and lytic transcription in gastric cancer

ABSTRACT

Epstein-Barr virus (EBV) is associated with 10% of human gastric carcinomas, which are distinguished by a CpG island methylator phenotype. In gastric carcinoma tumors and cell lines, the EBV genome also exhibits a high degree of 5-methyl cytosine (5mC) marks, which are propagated by host DNA methyltransferases (DNMT) with each cell cycle. Therefore, we sought to determine the effect of DNMT inhibition by the small molecule Decitabine (DCB) on EBV genomic 5mC and chromatin structure in two tumor-derived gastric cancer cell lines, YCCEL1 and SNU719. Decitabine effects on EBV genomic 5mC, chromatin structure, and viral gene expression were profiled by reduced representation bisulfite sequencing, ATAC-seq, and RNA-seq, respectively. Decitabine treatment resulted in global viral genome hypomethylation and a global increase in open chromatin. The most striking finding resulted from analyzing the methylation pattern from single RRBS sequencing reads, showing that the EBV genome contains a heterogeneous pool of epigenetic states, each of which is eroded upon Decitabine treatment. We observed heterogeneous 5mC epiallele patterns around EBV genomic CTCF binding sites and lytic gene transcriptional start sites. These results highlight the importance of 5mC in maintaining EBV genomic chromatin structure and latency. Furthermore, the presence of 5mC epialleles suggests EBV⁺ gastric cancers harbor transcriptionally distinct EBV episomes, which may exert distinct functional roles in maintaining latency and driving tumorigenesis.

IMPORTANCE

Epstein-Barr virus (EBV) latency is controlled by epigenetic silencing by DNA methylation [5-methyl cytosine (5mC)], histone modifications, and chromatin looping. However, how they dictate the transcriptional program in EBV-associated gastric cancers remains incompletely understood. EBV-associated gastric cancer displays a 5mC hypermethylated phenotype. A potential treatment for this cancer subtype is the DNA hypomethylating agent, which induces EBV lytic reactivation and targets hypermethylation of the cellular DNA. In this study, we identified a heterogeneous pool of EBV epialleles within two tumor-derived gastric cancer cell lines that are disrupted with a hypomethylating agent. Stochastic DNA methylation patterning at critical regulatory regions may be an underlying mechanism for spontaneous reactivation. Our results highlight the critical role of epigenetic modulation on EBV latency and life cycle, which is maintained through the interaction between 5mC and the host protein CCCTC-binding factor.

INTRODUCTION

Epstein-Barr virus (EBV) is a gammaherpes virus that establishes latent infection in the majority (~95%) of the world’s population. Latent EBV infection, in which most of the 80 viral lytic genes are suppressed and infectious virion is not produced, is associated with several malignancies worldwide, including Burkitt lymphoma, Hodgkin lymphoma, and the epithelial cancers nasopharyngeal and gastric carcinoma (GC). Epithelial-associated cancers make up the largest cancer burden of EBV-associated malignancies (1). EBV-associated gastric cancer (EBVaGC) comprises 10% of all GC and shows distinct clinical and molecular characteristics, as compared with EBV-negative subtypes (2, 3). EBVaGC has high levels of DNA methylation and exhibits a CpG island methylator phenotype, which may be a potential target for drug treatment (4, 5).

EBV latency involves the establishment of a minimal viral transcriptional program, in which small numbers of viral genes are expressed. In GC, EBV often expresses a modified type I latency program, such as EBV nuclear antigen 1 (EBNA1), EBV-encoded small ribonucleic acids (EBERs), BamHI-A rightward transcripts (BARTs), and occasionally expresses latent membrane protein 1 (LMP1), LMP2A, BARF1, and BNLF2a (6 – 10). These genes have been implicated in EBV-mediated oncogenesis. The establishment of latency is determined by epigenetic modification of the EBV genome, including 5-methyl cytosine (5mC), histone modification, and looping of the chromatin into an organized 3D structure (11 – 15). 5mC occurs on the fifth carbon of cytosine in cytosine–guanine dinucleotide pairs catalyzed by DNA methyltransferases (DNMT) (16). DNMT3a and DNM3b mediated de novo 5mC, whereas DNMT1 is a maintenance methyltransferase, faithfully establishing the 5mC patterning to newly synthesized daughter DNA strands. 5mC is an important epigenetic mark modulating host and viral transcription and is frequently disrupted in cancer (17 – 19). EBV genomic 5mC heterochromatic regions restrict viral gene expression to only latency genes (18, 20 – 22). Loss of 5mC leads to loss of transcriptional repression and lytic cycle reactivation (23, 24). The host protein CCCTC-binding factor (CTCF) further defines EBV genome chromatin boundaries (12, 25, 26). CTCF is a chromatin architecture protein that facilitates loop formation by bringing together enhancers and promoters, as well as acting as an insulator to demarcate euchromatic from heterochromatin. In this regard, CTCF can prevent transcriptionally repressed chromatin from encroaching into regions of active transcription. CTCF binding of the EBV genome helps maintain the chromatin structure necessary for stable latency while also allowing the virus to respond rapidly to reactivation signals (27, 28). Notably, CTCF preferentially binds to unmethylated DNA (29 – 31).

CpG methylation can be highly heterogeneous in a bulk population, resulting in epialleles with different patterns of CpG DNA methylation. To illustrate this point, consider a genomic region with 50% methylation. The methylation state could exist as (i) two equal distributions of methylation epialleles that are fully methylated or fully unmethylated or (ii) a set of epialleles that are stochastically methylated on 50% of the 5mC sites of all the epialleles. While both would be considered 50% methylated, the latter case results in greater transcription inhibition, as all epialleles recruit transcriptionally repressive 5mC binding proteins (32). Epialleles are alleles with different chromatin and 5mC states that can be inherited and are a source of phenotypic diversity (33). This example highlights the need to assess the epiallele composition about functional effects on transcription, transcription factor binding, and cellular heterogeneity.

EBV genomes persist as multicopy episomes during latent infection largely ranging from 5 to 10 copies per cell in primary GC (34). It is not known whether each viral episome in a single cell and cell population shares an identical epigenetic pattern. Little remains known about the epiallele composition of EBV episomes during latency (35), or for viral genomes more broadly. We sought to address this by using reduced representation bisulfite sequencing (RRBS) to analyze the epiallele pattern of the EBV genome. RRBS identifies site-specific methylation patterns of a single molecule, defining specific epialleles. Here we combined RRBS with chromatin immunoprecipitation sequencing (ChIP) for CTCF, Assay for Transposase-Accessible Chromatin (ATAC-seq), and RNA-seq. We analyzed the effect of the hypomethylating agent Decitabine (DCB) on EBV genome methylation patterning and its association with euchromatin domains determined by ATAC-seq. We show that EBV episomes are comprised of a heterogeneous pool of 5mC epialleles and highlight the critical role of CTCF in maintaining euchromatin regions.

MATERIALS AND METHODS

Cell culture of EBVaGC cell lines

YCCEL1 and SNU719 were grown in RPMI1640 media supplemented with 10% fetal bovine serum, penicillin and streptomycin (50 U/mL), and plasmocin (25 ug/L, InvivoGen). Cells were incubated at 37°C with 5% CO₂ in a humidified chamber. Cell identity was confirmed by short tandem repeat microsatellite testing. Cells were treated with 5-Aza-2′-deoxycytidine, DCB, at 7.5 µM or an equivalent amount of dimethylsulfoxide (DMSO) at the indicated time points.

Reduced representation bisulfite sequencing

DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen, ID 69504), according to the manufacturer’s protocol. RRBS was previously described (36). Briefly, genomic DNA was spiked with lambda phage DNA and digested with MspI restriction enzyme. Fragments were ligated with methylated adapters (NEB) to protect adapter sequences from downstream bisulfite treatment. Digested DNA was bisulfite converted using the EZ DNA Methylation Kit (Zymo). All libraries were prepared using NEBNext adapters and indexed with i7 primers. These methylation-based libraries were spiked with 25% phiX Illumina library to ensure sequence diversity and sequenced 75 bp paired-end on Illumina HiSeq 2500.

Adapter sequences were removed using TrimGalore! (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). RRBS sequence alignment to the human gammaherpesvirus 4 (HHV4) NC_007605.1 was performed by Bismark (27). CpG counts were performed using the bismark_methylation_extractor function. Coverage thresholds for RRBS were set at greater than 10 reads per CpG site allowing for the detection of 7,538 unique CpG sites in all samples. Differential methylation analysis was performed using methylKit in R (28, 29). P values were adjusted for multiple testing using the sliding linear model (SLIM) method (30). Differentially methylated sites were filtered by methylation change >10% and q_adj <0.05, unless otherwise stated.

RNA-seq

RNA was extracted from cell pellets using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol and included the on-column DNAse I treatment. Libraries were prepared using the QuantSeq (Lexogen) library preparation as previously described (37). This was generated by oligo-dT priming to produce strand-specific libraries that were sequenced on NextSeq500 (Illumina) to generate single-end 76bp reads.

CTCF ChIP-seq preparation

Chromatin immunoprecipitation with next-generation sequencing (ChIP-seq) was performed as previously described. Briefly, 25 × 10⁶ cells per immunoprecipitation were collected and fixed with 1% formaldehyde for 15 min and then quenched with 0.25 M glycine for 5 min on ice. After 3 washes with 1× phosphate-buffered saline (PBS), pellets were resuspended in 10 mL each of a series of three lysis buffers before fragmentation in a Covaris ME220 sonicator (peak power 75, duty factor 25, cycles/burst 1,000, average power 18.8, time 720 s) to generate chromatin fragments roughly 200–500 bp in size as determined by DNA gel electrophoresis. Chromatin was centrifuged to clear debris and a 1:20 of this cleared chromatin was kept as standard input for comparison against immunoprecipitations. Chromatin was incubated by rotating at 4°C for 1 h with 8 µg of antibody against CTCF (Active Motif 61311). Chromatin–antibody complexes were precipitated using a 50 µL of Dynabeads Protein A (ThermoFisher, product No. 10001D) incubated by rotating at 4°C overnight. DNA was purified using the Promega Wizard SV Gel and PCR Clean-up Kit (product No. A9285). Libraries for sequencing were made using the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs, product No. E7103) and sequenced on the NextSeq 500 (Illumina).

ATAC-Seq

SNU719 and YCCEL1 cells were treated with DMSO or 7.5 µM DCB for the indicated time points. Then the cells were harvested, and ATAC-seq was performed in two biological replicates according to the Omni-ATAC-seq protocol with modifications. Briefly, 1 × 10⁵ cells (>95% viability) were washed in 50 mL of cold PBS, spun down at 500 × g at 4°C for 5 min, and resuspended in 50 mL of cold ATAC-Resuspension Buffer (RSB) (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, and 3 mM MgCl₂) containing 0.1% IGEPAL CA-630, 0.1% Tween-20, and 0.01% Digitonin. Resuspended cells were kept on ice for 3 min, then washed with 1 mL of cold ATAC-RSB containing 0.1% Tween-20 but no IGEPAL CA-630 or Digitonin. Pellet nuclei were centrifuged at 500 × g and 4°C for 10 min, and the supernatant was removed. The pellet was then resuspended in a 50-mL Tn5 transposase reaction mixture following the manufacturer’s protocol (Illumina Tagment DNA Enzyme and Buffer, Illumina) and incubated at 37°C for 30 min in a thermomixer with 300 rpm mixing. DNA was purified using a MinElute PCR purification kit (Qiagen) and eluted in a 10-mL Elution Buffer for library amplification. PCR amplification of fragmented DNA was done using the NEBNext HiFi PCR Master Mix (New England Biolabs) with a universal forward and sample-specific reverse oligo for sample barcoding using the following PCR conditions: initial incubations of 72°C for 5 min and 98°C for 30 s, followed by five cycles of 98°C for 10 s, 63°C for 30 s, and 72°C for 1 min. An additional number of cycles was determined for each sample through a “side” qPCR using an aliquot of the PCR as a template to determine the number of cycles needed to reach 1/3 of the max fluorescence. PCR products were run on a 1% agarose gel, regions from ~50 bp to ~1 kb were excised, and DNA was extracted using a gel extraction kit (Qiagen). Purified DNA was submitted to the Wistar Institute Genomics core facility for quality analysis and sequencing. All samples were sequenced on NextSeq500 (Illumina) to generate paired-end 2 × 42 bp reads.

CTCF knockdown by siRNA

Cells were plated at 80% confluency the day before transfection. Either scrambled control or CTCF targeting a pool of 10 μM siRNA was transfected with Lipofectamine RNAiMAX Reagent (Invitrogen) in Opti-MEM Medium (Invitrogen), according to the manufacturer’s protocol. The final siRNA concentration was 50 nM. Cells were transfected at 24 and 48 h and collected 48 h after last transfection.

Chromatin immunoprecipitation

For each ChIP assay, 1 × 10⁷ cells were crosslinked with 1% formaldehyde at room temperature for 10 min and the reaction was quenched with 0.125 M glycine for 5 min. Crosslinked cells were harvested and lysed with cell lysis buffer [10 mM Tris-HCl (pH 8.0), 10 mM NaCl, 0.2% NP-40] supplemented with a protease inhibitor cocktail. Nuclei were pelleted and resuspended in nuclear lysis buffer [50 mM Tris-HCl (pH 8.0), 10 mM EDTA, 1% SDS] with a protease inhibitor cocktail. Chromatin was sheared to an average size of 200–500 bp using a Covaris ME220 sonicator (peak power 75, duty factor 25, cycles/burst 1,000, average power 18.8, and time 720 s). Chromatin was diluted with ChIP dilution buffer [0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl (pH 8.1), 167 mM NaCl] with protease inhibitors. For each IP, 10 μL of CTCF antibody (Active Motif) or IgG was added and rotated at 4°C overnight. Protein A Magnetic Beads (Invitrogen) were incubated for 2 h at 4°C with rotation. Beads were washed sequentially with low salt, high salt, LiCl, and TE buffer and then eluted with elution buffer (1% SDS, 0.1M NaHCO₃). Crosslinks were reversed by 65°C incubation with proteinase K followed by RNase A treatment. DNA was purified using the Promega Wizard SV Gel and PCR Clean-Up System (Promega) according to the manufacturer’s protocol. qPCR on the ChIP and input DNA was performed using the SYBR Green Master Mix on a QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific). The relative enrichment was calculated as a percentage of input.

Methylation-specific qPCR

Genomic DNA was extracted using the GeneJET Genomic DNA Purification Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. Bisulfite conversion of 1 μg of DNA was performed using the EZ DNA Methylation-Gold Kit (Zymo Research) following the manufacturer’s instructions. The converted DNA was eluted in 100 μL of elution buffer. A total of 15 ng of bisulfite DNA was added to each reaction. Samples were normalized to a primer sequence specific for bisulfite-converted DNA containing no CpG sites: 84 kb (Fw: AGGAGATTGATTTGGTTTATG and Rv: AACCCTACATTTTTTAATTAATTTTAC). The following primer pairs were designed to detect unmethylated CpG sites around CTCF binding sites on the EBV genome: 6.4 kb (Fw: TTTTTAGAGAGGGTAAAAGGG and Rv: CAAACTACATCACCGTAACA), 89 kb (Fw: TTGGTTAGTAGGGGTTGA and Rv: CCCTAATAAAAACTACCAACC), 91.4 kb (Fw: GGTTAAGTGGTTGGGGTAT and Rv: CCTACCAAAACCAAAAACA), and 138.8 kb (Fw: GTAGAAAATTGATAAGGATTGTG and Rv: ACTCCCACTAATTCCCAC).

Bioinformatic analysis

Reads were mapped against the human gammaherpesvirus 4 (HHV4) NC_007605.1. For ChIP-seq data, reads were mapped to the genome assembly using BWA (38). The 10–40 kb region contains an internal repeat region (IR1) containing multiple copies of a 3-kb repeat, which makes it difficult to find uniquely mapped sequencing reads; therefore, this region is omitted from the analysis. We used MACS2[72,73] software packages to call peaks using input samples as control. deepTools (39) was used for data visualization. RNA-seq data were aligned using STAR (40). RSEM v1.2.12 software was used to estimate read counts (41). Raw counts were used to estimate the significance of differential expression differences between two experimental groups using DESeq2 (42). ATAC-seq data were aligned using bowtie2 NC_007605.1 EBV genome (43). HOMER was used to generate bigwig files and call significant peaks—style factor option (44). Peaks that passed the false discovery rate <5% threshold were considered significant. Normalized signals for significant peaks were derived from bigwig files using the bigWigAverageOverBed tool from the UCSC toolbox with the mean0 option (45). Adapter sequences were removed using TrimGalore! (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). RRBS sequence alignment to the human gammaherpesvirus 4 (HHV4) NC_007605.1 was performed by Bismark (27). CpG counts were performed using the bismark_methylation_extractor function. Coverage thresholds for RRBS were set at greater than 10 reads per CpG site, allowing for the detection of 7538 unique CpG sites in all samples. Differential methylation analysis was performed using methylKit in R (28, 29). P values were adjusted for multiple testing using the SLIM method (30). Differentially methylated sites were filtered by methylation change >10% and q_adj <0.05, unless otherwise stated. Data have been deposited in GEO and can be accessed through the following accession number: GSE239770 and GSE234658.

Digital droplet PCR

Multiplex digital droplet PCR (ddPCR) was performed as previously described (46). DNA was isolated from 1 × 10⁶ cells using the GeneJET Genomic DNA Purification Kit (Thermo Scientific) according to the manufacturer’s protocol. For DNA extraction from supernatant, 200 μL of supernatant was first collected and treated with DNAseI to digest cell-free DNA followed by Proteinase K treatment. A total of 500 ng of cellular DNA was digested with BamHI enzyme (10 U/µL, New England Biolabs) in a total volume of 10 µL for 1 h at 37°C. Digestion was diluted 1:20 in nuclease-free water. An amount of 10 µL of diluted DNA digest was mixed with 12.5 µL of 2× digital PCR supermix for probes (No dUTP) (Bio-Rad), 1.25 µL 20× FAM primers, and 1.25 µL VIC primers for each reaction. The FAM primers sequence for EBV Lmp1 was Fw (5′−3′) AAGGTCAAAGAACAAGGCCAAG, Rv (5′−3′) GCATCGGAGTCGGTGG, and FAM - AGCGTGTCCCCGTGGAGG. Host control primer sequence for Ribonuclease P protein subunit 20 (Rpp30) was Fw (5′−3′) GATTTGGACCTGCGAGCG, Rv (5′−3′) GCGGCTGTCTCCACAAGT, and probe VIC-CTGACCTGAAGGCTCT. 20× primers contain 18 µM PCR primers and 5 µM probes for a final PCR concentration of 900 nM PCR primers and 250 nM probes. Each sample was run in duplicate. The ddPCR plate was sealed with a foil heat seal using the PX1 PCR Plate Sealer (Bio-Rad) at 180°C for 5 s. The plate was vortexed and spun down at 1,000 rpm for 1 min. Droplets were generated using the QX200 Droplet Digital PCR System (Bio-Rad) and the transfer of emulsified samples to a PCR plate was performed, according to the manufacturer’s instructions. The PCR plate containing emulsified droplets was sealed with a foil heat seal. PCRs were performed on the C1000 Touch Therma Cycler (Bio-Rad). The cycling protocol included an enzyme activation step at 95°C for 10 min and cycled 40 times between a denaturing step at 94°C for 30 s and an annealing and extension step at 60°C for 1 min, finally one enzyme deactivation step was performed at 98°C for 10 min. The ramp rate between these steps is set at 2 °C/s. Droplets were then counted using QX200 Droplet Reader (Bio-Rad). The absolute quantity of DNA per sample was determined using the QuantaSoft software.

RESULTS

The hypomethylating agents target CpG sites in EBVaGC

EBV genomes are hypermethylated in EBVaGC with a type II latency pattern. Hypomethylating agents are a potential therapy for EBVaGC as a tool to reactivate. To define the effects of hypomethylating agents on EBV genomic 5mC patterning in GC, we treated two EBVaGC cell lines, SNU719 and YCCEL1, with 5′aza-2′-deoxycytidine or DCB for 72 h. We chose this time point as these cell lines divide approximately once every 72 h, allowing DNMT inhibition effects on viral genomic 5mC to become evident. SNU719 was derived from a primary EBV +GC tumor, whereas YCCEL1 was obtained from a metastatic tumor (47, 48). We then analyzed 5mC site-specific changes by RRBS. RRBS is an efficient high-throughput sequencing technique that allows for the detection of site-specific changes in the 5mC pattern. This approach highlighted an EBV genome-wide hypomethylating effect on DCB treatment of SNU719, and to a lesser extent YCCEL1 (Fig. 1A). It should be noted that while the EBV genome is portrayed linearly, it is a circularized genome. Principal component analysis of 5mC modification demonstrated a clear separation between SNU719 DMSO and DCB-treated samples, and to a lesser extent also for YCCEL1 (Fig. 1B). DCB treatment of SNU719 resulted in hypomethylation (differential 5mC > 10%) for 85% (n = 6344) of EBV genomic CpG sites (Fig. 1C). Less than 0.01% (n = 39) of CpG sites were found to be hypermethylated by DCB treatment. In YCCEL1, 13% of CpG sites (n = 1008) were hypomethylated and less than 0.01% (n = 34) were hypermethylated by DCB (Fig. 1D). A total of 7538 CpG sites were detected in all samples across the EBV genome, many were found to be highly methylated and susceptible to DCB treatment (Fig. 1E). A subset of EBV genomic 5mC sites had low levels of 5mC in both cell lines and were unaffected by treatment. Furthermore, we observed several viral genomic CpG sites with low levels of methylation only in SNU719. Interestingly, high levels of 5mC marks remained unaffected by DCB treatment in both cell lines: SNU719 (n = 56, 0.75%) and YCCEL1 (n = 2412, 32%) (Fig. 1F). The presence of these highly methylated CpG sites resistant to hypomethylation suggests that there is a minimal contribution from newly synthesized EBV genomes, which should be completely hypomethylated. Overall, these results indicated that the global 5mC levels and response to DCB were conserved between the two cell lines (Fig. S1).

Fig 1.

DCB induces global hypomethylation of the EBV episome. (A) EBV genomic tracks showing 5mC before and after DCB treatment in the SNU719 and YCCEL1 cell lines. Representative tracks are an average of two biological replicates. (B) PCA plot based on the 5mC levels of the EBV episome. (C) Volcano plot showing the significance of the percentage change in 5mC after DCB treatment for SNU719 or (D) YCCEL1 cell lines. (E) Heatmap showing the level of methylation from 0% to 100% (yellow to blue, respectively) for all samples. (F) Genomic location of CpG sites with the greatest changes (<25% 5mC change, top green), CpG sites hypomethylated in all samples (<10% 5mC, middle yellow), and CpG sites hypermethylated in all samples (>90% 5mC, bottom blue).

EBV genome hypomethylation is associated with increased chromatin accessibility in EBVaGC

Since 5mC is associated with heterochromatin and silenced gene expression, we sought to define DCB-induced hypomethylation effects on EBV chromatin accessibility. Therefore, we performed Assay for Transposase-Accessible Chromatin (ATAC-seq) to identify regions of open chromatin on control vs DCB-treated cells. EBV genomes from control cells exhibited approximately seven distinct open chromatin peak regions: 6.5 kb/EBER1, 36 kb, 50 kb/Qp, 68 kb, 91 kb/Zp, 144.5 kb, and 166.5 kb/LMP1. These peaks were conserved in SNU719 and YCCEL1 and remained at both 24 and 72 h of DCB treatment (Fig. 2A). By contrast, DCB treatment resulted in a general opening of EBV chromatin, defined by globally increased ATAC-seq signal by 72 h of treatment. Several smaller accessible peaks evident in control cell genomes became more pronounced after 72 h of DCB treatment: 10.5 kb/Cp, 41 kb/oriLyt, and 139 kb (Fig. 2A). In SNU719, we observed several strong ATAC-seq peaks across all treatment conditions in areas of low methylation (Fig. 2B). Areas with minimal ATAC-seq peaks were found in regions with high 5mC. After treatment, these areas show an opening in chromatin with increased ATAC signal and loss of 5mC. Similar effects were observed in YCCEL1 (Fig. 2C). Negative correlation between loss of 5mC and gain of ATAC signal was observed in both cell lines after 72 h of DCB treatment (Fig. 2D and E). Our data correlate DCB-induced hypomethylation with a pronounced effect on relaxing heterochromatin across the EBV genome.

Fig 2.

DCB-induced hypomethylation correlates with euchromatin of the EBV episome. (A) EBV genomic tracks showing changes in 5mC after DCB treatment (top), with ATAC-seq signal peaks for DMSO, 24- and 72-h treated DCB for SNU719 and YCCEL1. Heatmaps centered on 1,000 bp region surrounding the ATAC-seq peak (DMSO, 24-h, and 72-h DCB treatment) and RRBS data (DMSO and 72-h DCB treatment) for (B) SNU719 and (C) YCCEL1. Heatmaps are in descending order with the strongest ATAC peak at the top. A scatterplot shows the correlation between the change in methylation and the log₂(fold change) in the ATAC-seq signal after DCB treatment. Linear regression line is shown in red with Pearson’s correlation coefficient and P value for (D) SNU719 and (E) YCCEL1.

CTCF occupancy defines areas of EBV genome open chromatin in EBVaGC

Since CTCF is an important chromatin architecture protein that regulates host and viral transcription, we next examined how DCB treatment affects EBV chromatin and 5mC marks at CTCF binding sites. To define the CTCF binding landscape in SNU719 and YCCEL1, we performed ChIP-seq. Our analysis revealed 10 EBV genomic CTCF binding sites in SNU719 and 13 in YCCEL1 (Fig. 3A). Common prominent CTCF binding sites occurred at 6 and 10 kb around OriP, 50 kb viral Q promoter, 68 kb within EBNA1, 91 kb within the BZLF1 promoter, and 170 kb at the LMP1 promoter. The 36 kb CTCF site was present in SNU719 but was diminished in YCCEL1. Strikingly, CTCF binding sites overlapped almost exclusively with prominent ATAC-seq signals. CpG sites were not methylated in or around strong CTCF binding sites in either SNU719 or YCCEL1, even before DCB treatment (Fig. 2B and C). Some CTCF binding sites, such as the 10 kb, show weak CTCF binding, lower ATAC-seq signal, and higher levels of 5mC. In SNU719, CpG sites in CTCF binding regions were significantly less methylated than CpG sites outside of CTCF binding sites (Fig. 2D). This contrasted with EBV genomic CpG sites outside of CTCF binding sites, which were highly methylated in control cells and significantly hypomethylated by DCB treatment. To determine the effect of CTCF binding on the 5mC of these regions, we performed CTCF knockdown in the YCCEL1 cells and found a significant reduction in the unmethylated CpG sites around the EBER1 CTCF binding site (Fig. S2). Our data highlight the important functional role of CTCF as an insulator and protector of important transcriptional regulatory regions of the EBV genome from 5mC.

Fig 3.

DCB-induced hypomethylation correlates with euchromatin at CTCF binding sites. (A) EBV genomic tracks for CTCF binding sites, ATAC-seq, and changes in 5mC after DCB treatment. Heatmaps of 2,000 bp regions upstream and downstream of the CTCF peak center for CTCF ChIP-seq, ATAC-seq DMSO, ATAC-seq 24 and 72 h DCB, and RRBS 5mC DMSO and 72 h DCB for (B) SNU719 and (C) YCCEL1. Violin plots show the percentage of 5mC of CpG sites in CTCF binding regions or outside before and after treatment for (D) SNU719 and (E) YCCEL1. The black bar indicates the mean. CpG sites for heavily methylated CTCF binding regions are highlighted: 10 kb (red), 75 kb (orange), and 133 kb (yellow).

Decitabine disrupts DNA methylation epialleles around CTCF binding sites in EBVaGC

To further investigate 5mC patterning at EBV genomic CTCF binding sites, we cross-compared CTCF ChIP-seq with RRBS data from control vs DCB-treated cells. EBV genome CTCF binding sites sat in demethylated pockets even in control cells, tightly flanked by high levels of 5mC, such as observed at positions 6 kb/OriP, 41 kb/, 50 kb/Qp, 68 kb/BMRF1, 91/BRLF1, and 166 kb/LMP1 on the EBV genomic map (Fig. 4A). At the CTCF-bound site at position 36 kb from the left of the EBV genome map, CTCF instead bound to a region lacking surrounding 5mC. The strongest CTCF binding peak was found at the 50 kb EBV genome position, located at the Q promoter (Qp), which drives the expression of the latency gene, EBNA1. 5mC patterning was quantified for groups of four sequential CpG sites either methylated or unmethylated resulting in a total of 16 possible methylation patterns. 5mC patterning located in this region (50,000–50,170 kb) exhibited two distinct forms, a completely methylated (s1111) minor fraction where sequential CpG sites were methylated or a fully unmethylated (s0000) majority (Fig. 4B). In addition to Qp, we found a minor CTCF binding site at the C promoter (Cp), located at ~10 kb, which is one of the only CTCF binding sites containing methylated CpG sites. Notably, Cp is silenced in EBVaGC to block the expression of the latency III program. The majority of EBV genomic 5mC epialleles found between 10 and 11 kb region were methylated, either fully methylated or containing variable 5mC patterning (Fig. 4C). Only a minority were fully unmethylated and this increased after 72-h DCB treatment. The 91-kb CTCF binding site, located upstream of the immediate early gene BZLF1 promoter, had few methylated CpG sites. The 500 bp upstream of this CTCF binding site instead contained highly methylated epialleles (Fig. 4D). DCB treatment reduced the proportion of epialleles containing heterogeneous 5mC patterning in this region.

Fig 4.

DCB disrupts 5mC patterning around CTCF binding sites. (A) EBV genomic track for CTCF-ChIPseq for SNU719. 5mC epiallele patterns (representative sample) around (B) CTCF 10 kb, (C) CTCF 50 kb, and (D) CTCF91kb binding sites for DMSO and DCB treatment with a stacked bar plot quantify the percentage of distribution of epialleles. Methylation patterning is represented by 1 for a methylated CpG or 0 for an unmethylated CpG. Patterns were quantified for groups of four consecutive CpG sites resulting in 16 possible methylation combinations. (E) EBV genomic track for CTCF-ChIPseq for YCCEL1 5mC epiallele patterns (representative sample) around (F) CTCF 10 kb, (G) CTCF 50 kb, and (H) CTCF 91 kb binding sites for DMSO and DCB treatment with a stacked bar plot quantifying the percentage of distribution of epialleles.

Many EBV genomic CTCF binding sites likewise exhibited high levels of flanking 5mC signals in YCCEL1 (Fig. 4E), suggesting that this phenotype may be conserved in EBVaGC. Interestingly, the SNU719 CTCF peak present at the EBV genomic 36 kb region was not found in YCCEL1, and this region exhibited a high degree of methylation in YCCEL1. The CTCF 50 kb peak at Qp is also the most pronounced in YCCEL1, where we again observed a low level of CpG methylation within the CTCF binding region (Fig. 4F). The smaller CTCF 10 kb peak at Cp contains mostly methylated epialleles that were hypomethylated by DCB treatment (Fig. 4G). In addition, CTCF peaks at 6 kb and 41 kb contained a heterogeneous methylated epiallele pattern (Fig. S4). Interestingly, the BZLF1 promoter region upstream of the CTCF 91 kb peak showed predominantly heterogeneous 5mC patterning (Fig. 4H), in contrast to SNU719 epialleles, most of which were fully methylated. Overall, our data suggest that EBV genomic CTCF binding occurs in hypomethylated regions in EBVaGC.

Decitabine-induced hypomethylation causes EBV lytic reactivation in EBVaGC

We next used Quant-seq to define DCB hypomethylation effects on EBV transcription in YCCEL1 and SNU719. Principal component analysis (PCA) highlighted significant differences between each group, with separation between the DMSO- and DCB-treated groups along the PC1 axis explaining 95% of the mRNA variability so that the samples clustering closer together had a more similar transcriptional profile (Fig. 5A). In both SNU719 and YCCEL1, DCB treatment caused a significant increase in the mRNA abundances of many EBV genes (Fig. 5B and C), particularly the viral lytic genes including BZLF1 (Fig. S4). Notably, DCB significantly hypomethylated EBV lytic gene promoters (Fig. 5D), suggesting that 5mC marks are a major mechanism by which latency is maintained in EBVaGC. DCB also upregulated transcripts encoding the EBNA1, LMP1, and LMP2A EBV latency genes, despite the absence of 5mC marks at their promoters in untreated cells, consistent with the EBV lytic cycle driving their increased expression (49 – 51). The EBV-encoded long non-coding RNA RPMS1 was one of the few EBV transcripts that exhibited cell-specific differences, with DCB-induced expression only in SNU719. Unexpectedly, the EBV BARF1 transcript was the only lytic cycle gene downregulated by DCB treatment. In line with the observed transcriptional changes, DCB treatment increased EBV genome copy number per cell, suggestive of productive EBV lytic replication (Fig. 5E). In addition, titrating DCB concentration shows a positive correlation between DCB concentration and EBV genome copy until toxicity is reached at high concentrations above 7.5 μM (Fig. S3). Furthermore, to determine active virion production from abortive lytic replication, we determined the EBV copy number in the supernatant after DCB treatment.

Fig 5.

DCB induces massive expression of the EBV transcriptome and increased EBV lytic transcription. (A) PCA plot of RNA-seq data for SNU719 and YCCEL1 EBV transcriptome. Volcano plot showing significantly changed transcripts for (B) SNU719 and (C) YCCEL1. (D) A heatmap showing the transcriptional changes in the significantly changed genes in both SNU719 and YCCEL1. The change promoter 5mC (500 bp upstream of TSS) was determined by RRBS for both cell lines, where gray indicates no detectable CpG sites in the region. (E) Digital droplet PCR measuring LMP1 of the EBV genome after 72 h of DCB treatment.

Decitabine disrupts DNA methylation epiallele patterning around EBV transcriptional start sites

DCB caused extensive hypomethylation of two EBV genomic regions, located at the end of the BHLF1 coding region and within the origin of lytic replication, oriLyt in the 38–45 kb region (Fig. 6A). BHLF1 was recently implicated in the control of EBV programs, including viral latency in B cells, although its roles in epithelial cells are little studied (52). BHLF1 was not expressed in control GC cells, and the majority of 5mC epialleles were methylated. By contrast, DCB increased BHLF1 locus unmethylated epialleles and upregulated BHLF1 expression in both SNU719 and YCCEL1 (Fig. 6B). Likewise, oriLyt is implicated in a reversal of EBV latency in B cells (27, 53). OriLyt region epialleles were methylated in control cells, although SNU719 had a larger proportion of unmethylated alleles in this region (Fig. 6C). Nonetheless, DCB caused oriLyt epiallele hypomethylation to a greater extent in SNU719. DCB treatment had variable effects on EBV early genes. For instance, the ribonucleotide reductase encoding EBV early lytic cycle gene BORF2 was upregulated by DCB, with concomitant loss of 5mC throughout the BORF2 coding region (Fig. 6D). Many BORF2 promoter epialleles were either fully methylated or unmethylated, in contrast to the epialleles found in the EBV early lytic gene BSLF2/BMLF2 promoter region (Fig. 6E). Nonetheless, the BSLF2/BMLF2 transcript was upregulated by DCB treatment, with concomitant loss of 5mC marks at the BSLF2/BMLF2 locus (Fig. 6F); however, epiallele patterning was more heterogeneous in both SNU719 and YCCEL1 (Fig. 6G). Our data highlight widespread DCB de-repression of the EBV genome in EBVaGC and show that while many epiallele compositions are conserved between the two cell lines, there are nonetheless metastable epialleles.

Fig 6.

Hypomethylation and loss of 5mC patterning are correlated with increased transcription. (A) EBV genomic Quant-seq tracks for forward (+) and reverse (−) transcripts, and 5mC change (RRBS DCB–DMSO) in SNU719 before and after DCB treatment. The region surrounds the BHLF1 (− strand) transcript. (B) Stacked bar plot of the epiallele quantification for two regions upstream and downstream of BHLF1. (C) EBV genomic tracks for BORF1 (+ strand) transcript. (D) Epiallele quantification of a CpG dense region upstream of the BORF1 transcription start site. (E) EBV genomic plots for BSLF2 (− strand) transcript. (F) Epiallele quantification of CpG sites in a region upstream of the BSLF2 transcriptional start site.

DISCUSSION

DNA methylation, histone modifications, and chromatin looping each shape gene expression profiles, although the extent to which they control the EBV genome program in gastric carcinoma has remained incompletely understood. Here, we observed widespread effects of DNA hypomethylation on the viral epigenome and in two tumor-derived EBVaGC cell line models, YCCEL1 and SNU719. Near global hypomethylation triggered increased EBV genome accessibility and lytic reactivation. Our multi-omic analyses highlight a major CTCF insulating role in the control of the EBV genome in EBaGC. Areas of open chromatin corresponded almost identically to CTCF binding sites. Many of these CTCF binding sites sit in an unmethylated region, flanked by regions of hypermethylation, including key viral control elements such as the immediate early gene BZLF1 promoter and the EBNA1 Q promoter. It should be noted that newly synthesized EBV genomes are unmethylated raising the possibility that the hypomethylation of the EBV genome could arise from increased EBV replication during lytic reactivation. However, we detect several regions resistant to DNA hypomethylation after DCB treatment, suggesting that we are detecting chromatinized EBV genomes contained within the cells.

CTCF typically binds to hypomethylated DNA elements (29, 30), although the extent to which this occurs on the EBV genome has remained incompletely understood, particularly in EBVaGC. While we generally observed low levels of DNA methylation at CTCF-occupied regions, as expected, we also observed CTCF binding peaks to occur at regions that contain only a small subset of unmethylated epialleles, such as viral C promoter, EBER, LMP1/LMP2, oriLyt, and BZLF1. CTCF binding can be affected by 5mC (23). In line with this, we observe weaker CTCF binding peaks at regions that contain only a small subset of unmethylated epialleles such as the CTCF 10 kb site at Cp. Our data suggest that only a fraction of EBV genomes bind this site due to the smaller portion of unmethylated epialleles, resulting in a smaller CTCF peak. In response to DCB hypomethylation, viral euchromatin increases around these sites, which may act as transcriptional hubs in lytic reactivation (54). Based on these observations, we are tempted to speculate that CTCF sites evolved at key control elements across the viral genome to suit the needs of EBV to preserve critical regulatory regions from epigenetic silencing through DNA methylation. Such a mechanism would allow the minimal transcription activity necessary for establishing latent infection. Consistent with this hypothesis, EBV episomes carrying mutations at CTCF binding sites usually display altered gene expression patterns and an altered epigenetic landscape (55). For example, EBV episomes carrying mutations that ablate CTCF occupancy at the Qp promoter show increased levels of DNA methylation over time, eventually resulting in Qp silencing (13). More experiments are needed to fully understand the precise role of CTCF binding across the EBV genome; nevertheless, our results indicated the essential role of CTCF binding in organizing the epigenetic landscape of the EBV episome during latent infection.

Host enzyme methylation of the viral genome is thought to be a protective mechanism that silences expression from foreign DNA elements. However, EBV has evolved to circumvent this inhibitory effect by encoding a 5mC-binding transcription factor, Zta encoded by the BZLF1 gene (56 – 59). Indeed, we observed that many EBV lytic genes contain completely methylated epialleles, such as BHLF1 and BSLF2. Interestingly, we observed a heterogeneous pool of epialleles in the BZLF1 promoter, which corresponds to a CTCF binding site at 91 kb. We speculate that heterogeneous 5mC patterning at the BZLF1 promoter may be an underlying mechanism for spontaneous reactivation. Zta works in concert with Rta, and Zta alone may function primarily to facilitate DNA replication at OriLyt (60). In line with this, we observed more stochastic 5mC patterning at the 91 kb region in YCCEL1, which correlated with higher EBV genome copy number, both before and after DCB treatment, relative to SNU719. Overall, there was an increase in lytic transcription which correlated with hypomethylation of many gene promoters. However, it is unclear what the contribution of hypomethylation of the host genome is on increased viral expression.

The Qp promoter is transcriptionally active in EBVaGC, where it is hypomethylated and drives the expression of EBNA1. While we found many 5mC epialleles within the Qp region to be completely unmethylated, there was also a subset that was completely methylated. Given also that EBVaGC contains many copies of the EBV episome, our data suggest that some gastric cancer EBV episomes are silenced at the Q promoter. They may function using another promoter, such as Cp, to drive EBNA1; however, this promoter contained mostly hypermethylated epialleles and was previously shown to be transcriptionally silent. An alternative and more compelling possibility is that there is a transcriptionally dead EBV genomic pool, and that these “zombie” episomes may survive passively due to other transcriptionally functional episomes present. Furthermore, they may serve as a template for replication or provide another unknown function that requires further investigation. How these heterogeneous EBV populations are distributed among tumor cells remains to be characterized. Technological advances, including the ability to detect EBV genomic 5mC patterning at the single-cell level, will undoubtedly yield further insights. Developing sequencing technologies, which offer the promise to analyze which EBV genomic epialleles interact, will provide further insights into the interplay of viral genome 5mC, CTCF, and latency states.

Altogether, these results highlight the critical role of epigenetic modulation in controlling the EBV life cycle. The critical interaction between 5mC and CTCF is disrupted after DCB treatment along with relaxation of the EBV genome and increased lytic transcription. We identified a heterogeneous pool of EBV episomes which may contribute to the specific expression pattern found in Type II latency of gastric cancer. It is possible that “zombie” EBV episomes may be related to the heterogeneous cellular phenotypes displayed among the tumor cells. Since other herpesviruses are epigenetically silenced and establish latency, it will be interesting to cross-compare how 5mC patterning, including epiallele formation and CTCF occupancy, shapes their epigenomes and latency programs.

DATA AVAILABILITY

Data have been deposited in GEO and can be accessed through the following accession numbers: GSE239770 and GSE234658.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/mbio.00396-23.

Fig. S1. mbio.00396-23-s0001.tif.

Conserved response to DCB in SNU719 and YCCEL1.

Supplemental Figure Legends. mbio.00396-23-s0002.docx.

Legends for the supplemental figures.

Fig. S2. mbio.00396-23-s0003.tif.

Increased methylation at CpG sites found in CTCF binding region.

Fig. S3. mbio.00396-23-s0004.tif.

DCB treatment leads to an accumulation of intra and extracellular viral copies.

Fig. S4. mbio.00396-23-s0005.tif.

Epiallele changes at additional CTCF binding sites.

Fig. S5. mbio.00396-23-s0006.tif.

Epigenome of the BZLF1 locus.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.