Epstein-Barr virus evades restrictive host chromatin closure by subverting B cell activation and germinal center regulatory loci

SUMMARY

Chromatin accessibility fundamentally governs gene expression and biological response programs that can be manipulated by pathogens. Here we capture dynamic chromatin landscapes of individual B cells during Epstein-Barr virus (EBV) infection. EBV⁺ cells that exhibit arrest via antiviral sensing and proliferation-linked DNA damage experience global accessibility reduction. Proliferative EBV⁺ cells develop expression-linked architectures and motif accessibility profiles resembling in vivo germinal center (GC) phenotypes. Remarkably, EBV elicits dark zone (DZ), light zone (LZ), and post-GC B cell chromatin features despite BCL6 downregulation. Integration of single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq), single-cell RNA sequencing (scRNA-seq), and chromatin immunoprecipitation sequencing (ChIP-seq) data enables genome-wide cis-regulatory predictions implicating EBV nuclear antigens (EBNAs) in phenotype-specific control of GC B cell activation, survival, and immune evasion. Knockouts validate bioinformatically identified regulators (MEF2C and NFE2L2) of EBV-induced GC phenotypes and EBNA-associated loci that regulate gene expression (CD274/PD-L1). These data and methods can inform high-resolution investigations of EBV-host interactions, B cell fates, and virus-mediated lymphomagenesis.

Graphical Abstract

In brief

Epstein-Barr virus (EBV) persists within memory B cells in most adults and is associated with diverse cancers and autoimmunity. SoRelle et al. identify single-cell chromatin accessibility dynamics governing EBV infection. Successful infection substantially phenocopies activated, germinal center, and effector B cell states. Predicted targets regulating these phenotypes are validated experimentally.

INTRODUCTION

Epstein-Barr virus (EBV) is one of the most successful human pathogens because of its ability to achieve lifelong persistence in memory B cells (MBCs) by co-opting B cell immune responses.¹ More than 95% of adults are EBV-seropositive and typically experience asymptomatic or mild illness.² Notwith-standing, EBV is associated with diverse B cell malignancies including Burkitt lymphoma, post-transplantation lymphoproliferative disease (PTLD), and Hodgkin’s lymphoma,³ in addition to autoimmune disorders including systemic lupus erythematosus (SLE),⁴ and multiple sclerosis (MS).^5,6

EBV engagement of tonsil germinal center (GC) B cell dynamics affords a route to latency within MBCs.^7,8 However, infection yields many cellular fates. Many EBV⁺ B cells incur growth-associated replication defects and subsequent DNA damage response (DDR)-mediated cell-cycle arrest following de novo infection.^9,10 Moreover, sustained infection within in vitro immortalized lymphoblastoid cell lines (LCLs)^11,12 yields a heterogeneous continuum of phenotypes mimicking early activated GC B cells and differentiated plasmablasts.¹³ However, comprehensive understanding of gene regulatory control underlying these distinct outcomes remains unrealized.

Genome-wide sequencing techniques are well suited to address this knowledge gap. Ensemble RNA sequencing (RNA-seq) has revealed that EBV infection rapidly and profoundly reshapes host cell transcriptomes.^14,15 Likewise, bulk assay for transposase-accessible chromatin sequencing (ATAC-seq) and viral transactivator chromatin immunoprecipitation sequencing (ChIP-seq) have provided insight into EBV-driven epigenetic rewiring (e.g., DNA methylation and histone modifications).^16-19 Although these techniques highlight EBV-driven remodeling of host gene regulation, their population-averaged measurements are insensitive to heterogeneous cellular responses that occur asynchronously during infection. To resolve such dynamic biological complexity, single-cell approaches are required.^20-24 Moreover, advances in single-cell multiomics and multimodal integration have facilitated comprehensive studies of hierarchical gene regulation.²⁵ These include techniques for obtaining cell-matched measurements of mRNA transcripts, chromatin accessibility, DNA methylation status, and other molecular levels.^26-29

We recently applied time-resolved single-cell RNA-seq (scRNA-seq) to develop an updated high-resolution model of early EBV infection.³⁰ Here we present early infection single-cell ATAC-seq (scATAC-seq) data, integrate cell-matched gene expression and chromatin accessibility with bulk ChIP-seq data for host and viral proteins, and develop informatic workflows that elucidate phenotype-specific virus-mediated gene regulation. Collectively, this study provides detailed perspectives of host-pathogen dynamics that control the heterogeneous fates of EBV-infected B lymphocytes.

RESULTS

Time-resolved scATAC-seq of primary EBV infection

Single-cell chromatin accessibility (scATAC-seq) libraries were prepared from B cells prior to (day 0) and following de novo infection with B95-8 EBV³¹ at three time points (days 2, 5, and 8). After quality control (QC) filtering and feature selection to include the top 50% of differentially accessible peaks (DAPs), 97,191 cells across two biological donors were analyzed (52,271 and 44,920 cells, respectively). Cells from each time point contained 161,800 ATAC features on average, including >200,000 unique peaks (Figures 1 and S1A-S1D; Table S1). Time course analysis revealed global changes in cellular chromatin landscapes following infection, with the most extensive alterations occurring within two days (Figure 1A). We observed a time-dependent median of ~6,000 to ~12,000 feature peaks per cell and an experiment-wide average of ~25,000 transpositions per nucleus. Cellular chromatin accessibility was transiently but significantly reduced during early infection in both donors, with reductions consistently observed at days 2 and 5 relative to resting B cells (Figure 1B).

Figure 1. Differential single-cell chromatin accessibility from early EBV infection.

(A) Uniform manifold approximation and projection (UMAP) of scATAC-seq for 55,218 cells across in vitro early EBV infection by time point for one donor. Preinfection (day 0) and post-infection (days 2, 5, and 8) samples were captured.

(B) Total transpositions and feature peaks per cell by time point and donor. Significant differences in chromatin accessibility distributions were determined by Kolmogorov-Smirnov (K-S) test (***p < 1e-15.

(C) Chromatin trajectories in early infection. Multi-branch pseudotime (left panel) learned from scATAC-seq data. ATAC signals in pseudotime-ordered cells coded by timepoint and fate trajectory (middle panel). Two branches define EBV⁺ subsets that arise upon de novo infection (resting cells, blue cluster; EBV⁺-low ATAC, green; EBV⁺-high ATAC, gold). R denotes Pearson correlation between ATAC signals and calculated pseudotime. Simplified model of global accessibility changes along EBV⁺ trajectories (right panel).

(D) Multimodal integration coded by cell-matched scRNA-seq early infection phenotypes.³⁰ Cell identity mapping across time points, ATAC pseudotime branches, and model states. Arrow weights approximate the fraction of cells from each state in the sending layer that map to states in the receiving layer.

(E) ATAC signal quantification per cell by multimodal state cluster. Statistical comparisons (one-sided K-S test) of global accessibility in EBV⁺ arrest clusters (c4 and c7) versus all other model phenotypes (*p < 1e-3, **p < 1e-8, and ***p < 1e-15).

(F) Example DAPs with significantly increased accessibility in post-infection phenotypes (top 5 tracks in each panel) versus uninfected phenotypes (bottom 3 tracks). The two panels show DAPs (highlighted regions) upstream of the transcription start site (TSS) for BATF (left panel, −23 kb). Peaks are mapped to hg38.

See also Figures S1 and S2.

We next calculated chromatin accessibility trajectories using pseudotime methods³² (Figures 1C and S1E-S1H). Two main trajectories (branches) developed following infection, one of which had relatively higher net accessibility (EBV⁺-high ATAC) than the other (EBV⁺-low ATAC). Although EBV⁺-low ATAC cells occurred late in pseudotime, they were detectable by day 2. Both branches contained cells from each post-infection timepoint and recapitulated time-resolved trends (Pearson R = −0.32). This bifurcation of global chromatin states corresponded to broad changes in gene regulation following infection. Cis-linked regulatory prediction³³ coupled with gene ontology analysis identified significantly enriched DAPs in EBV⁺-low ATAC and EBV⁺-high ATAC states versus uninfected B cells (Table S1). The top 2,000 ATAC sites with significantly enhanced accessibility in resting B cells versus EBV⁺ branches yielded predicted linkages to >1,800 genes. DAP-linked genes were significantly associated with ~200 Gene Ontology (GO) biological process terms, especially those related to mRNA metabolism, chromatin organization, gene expression, and B cell activation (Table S1). The top 2,000 significant sites with greater accessibility in EBV⁺-high ATAC cells versus resting cells had predicted linkages to nearly 1,500 genes. Significant GO terms for EBV⁺-high ATAC DAP-linked genes were related to lymphocyte activation, defense responses, cytokine production, and programmed cell death (Table S1). By contrast, EBV⁺-low ATAC cells only exhibited 170 genes linked to 174 significant DAPs with enhanced accessibility versus resting B cells. Only one GO term (regulation of immune process, GO:0050776) was enriched in EBV⁺-low ATAC cells versus resting cells. Thus, although EBV⁺-low and EBV⁺-high ATAC states exhibited reduced accessibility versus uninfected cells, only EBV⁺-high ATAC cells gained accessibility at loci regulating B cell activation and immune responses.

We performed multimodal integration of cell-matched scATAC-seq and scRNA-seq data using Signac.³⁴ Clusters identified via weighted nearest neighbor analysis were mapped to early infection phenotypes defined in prior work³⁰ (Figures 1D and S1). These phenotypes include uninfected naive (c8) and memory (c3) B cells, EBV⁺ cells undergoing innate antiviral sensing (c4), EBV⁺ cells exhibiting growth- and DNA damage-induced arrest (c7), EBV⁺ GC dark zone (DZ)-like hyperproliferating cells (c6), EBV⁺ activated precursor/early MBCs (AP-eMBCs; c0), naive-derived EBV⁺ activated cells (c1), EBV⁺ GC light zone (LZ)-like NF-κB activated cells (c2), and EBV⁺ differentiated plasmablasts (c5). Integration enabled cell mapping by timepoint, ATAC trajectory, and biological phenotype. EBV⁺-low ATAC cells with significantly reduced accessibility corresponded to unsuccessful infection (c4 and c7; Figure 1E). By contrast, uninfected (c3 and c8) and hyperproliferative (c6) cells had the most open chromatin. Hyperproliferative cells (c6) exhibited variable accessibility attributable to a composition of proliferating cells (higher accessibility) and those trending toward oncogene-induced arrest (lower accessibility). There were significantly more ATAC peaks in NF-κB activated cells (c2) relative to other activation intermediates (c0 and c1) and differentiated cells (c5). Significant cluster-resolved DAPs, cis-linked genes, and GO enrichment from DAP-linked gene predictions are provided in supplemental information (Table S2).

Because accessibility was recovered between days 5 and 8, state-resolved profiles indicated that EBV-induced chromatin closure is transient during successful infection. Thus, the EBV⁺-low ATAC trajectory defines cell arrest and the EBV⁺-high ATAC trajectory leads to immortalization. Accordingly, significantly depleted accessibility was observed in arrest states, especially c4, at loci such as the AP-1 transcriptional activator BATF (Figure 1F), which is essential for B cell activation and LCL survival.³⁵ GO analysis of DAP-linked differentially expressed genes (DEGs) between resting, EBV⁺ arrest, and EBV⁺ success groups confirmed closure at immune cell activation loci in EBV⁺ arrest and identified accessibility-linked expression of latency establishment mediators (e.g., NFKBIA, RBPJ) in EBV⁺ success states versus resting B cells. B cell activation, antiviral response, and survival genes linked to accessibility in immortalizing but not arresting trajectories included AICDA, BCL2, CD83, IFNG, PRDM1, and TRAF2.

Linked expression and accessibility illuminate phenotype-specific regulation

We summarized gene regulation across different EBV⁺ fate trajectories. Nine hundred fifty-four mRNA features were derived from the top 100 markers for each early infection phenotype. One hundred seventy-seven of these DEGs were significantly linked to 476 DAPs (18.6% of tested genes with potential DAP-linked regulation; Figures 2A and S2). DAP-linked DEG analysis revealed four regulatory patterns: higher accessibility linked to higher expression (+/+), lower accessibility linked to higher expression (−/+), lower accessibility linked to lower exp ression (−/−), and higher accessibility linked to lower expression (+/−). The +/+ and −/− patterns were characteristic of accessible versus inaccessible positive regulatory loci, respectively. The less frequently observed −/+ and +/− patterns were consistent with closure or opening of ATAC sites associated with negative regulation. Genes of interest including CCR7, CXCR4, RUNX3, BACH2, JCHAIN, and PRDM1 provided examples of each pattern and their phenotype-specific variation (Figure S2). For example, seven significant expression-linked DAPs (five positive and two negative regulatory loci) were identified within 20 kB of the transcription start site (TSS) for CCR7. The BACH2 locus contained 13 significant intragenic regulatory loci including two negative sites. As for CCR7 and CXCR4, all positive regulatory loci linked to BACH2 expression exhibited reduced accessibility in EBV⁺ phenotypes. By contrast, RUNX3, PRDM1, and JCHAIN had more open chromatin at positive regulatory sites in EBV⁺ states versus resting B cells, with PRDM1 and JCHAIN specific to EBV-induced plasmablasts (c5).

Figure 2. Cell-matched expression and chromatin accessibility fate trajectories.

(A) Overview of differentially accessible peak (DAP)-linked differentially expressed genes (DEGs) and cluster comparisons for trajectories of interest. DAPs are denoted by shared presence (union, ∪) in one or more states but not in (!) one or more other states. Resting vs. arrested cells, (3 ∪ 8) ! (4 ∪ 7); EBV⁺ activated vs. resting cells, (2) ! (3 ∪ 8); EBV⁺ activated vs. EBV⁺ differentiated cells, (2) ! (5).

(B) Virus-induced arrest responses. State-resolved DAPs, multimodal summaries, and example DEGs and DAP linkages are presented. Horizontal tracks depict cluster-coded accessibility. Boxed regions highlight DAPs (blue, positive linkage to expression; red, negative linkage to expression) in high-resolution insets. Violin plots depict expression. Two (3 ∪ 8) ! (4 ∪ 7) DAP-linked DEGs (CFLAR, NFKB2) are essential for EBV-induced transformation.

(C) Successful infection trajectory state-resolved DAPs as presented in (B). IRF4 and CCND2 are essential for EBV-induced transformation.

(D) EBV-induced B cell activation/differentiation. State-resolved DAPs (c2!c5), multimodal summaries, and trajectory-specific examples are presented as in (B) and (C).

See also Figures S2 and S3.

Because DAP-linked DEGs included numerous B cell fate mediators, we investigated regulatory relationships in distinct trajectories (Figures 2B-2D). In the path from resting to arrested/senescent cells (c4 and c7), 23.4% of DEGs were linked to DAPs that became inaccessible after infection (i.e., [3 ∪ 8] ! [4 ∪ 7] peaks; Figure 2B). DAP-linked DEGs downregulated in this trajectory included CFLAR and NFKB2, both of which are essential in LCLs.³⁵ Similar DAP-linked loss of basal expression from uninfected to arrested states was frequently observed for genes involved in B cell signal transduction (e.g., CD74, FYN, INPP5D, LYN, MAP4K3, PRKCE) and reflected in top GO terms (e.g., intracellular signal transduction, lymphocyte activation, regulation of BCR signaling pathway; Table S3). Accessibility loss in c7 was consistent with induction of senescence-associated heterochromatin foci (SAHF; Figures S3A and S2B).^36-38 Because senescence can arise from diverse mechanisms such as innate sensing or DNA damage, we used higher resolution clustering to refine c7 subsets (7a and 7b). These subsets displayed DEGs involved in the cell cycle and antiviral sensing (Figures S3C-S3F). Different HMGB2 levels between 7a and 7b were notable, as HMGB2 mediates diverse roles in sensing,³⁹ double-stranded break repair,⁴⁰ and p53 downregulation.⁴¹ HMGB2 expression was elevated in 7b (as in hyperproliferative cells) versus 7a (similar to c4, which could precede senescence⁴²). Proliferation markers were also lower in 7a than 7b. Thus, 7a corresponded to EBV⁺ cells that arrested rapidly via innate sensing, while 7b corresponded to later arrest induced by replicative stress.

NF-κB activated EBV⁺ cells (c2) exhibited diminished accessibility at 1,142 sites relative to resting cells (8.7% of resting cell peaks). However, c2 cells possessed 668 peaks absent in resting cells ([2] ! [3 ∪ 8]) linked to 595 genes. One hundred fifty-four of these (25.9%) were DEGs between resting and activated cells (Figure 2C), including 109 upregulated and 45 downregulated genes. Upregulated genes along this trajectory included IRF4 and CCND2, both of which are LCL dependencies.³⁵ The IRF4/DUSP22 locus contained at least nine positive regulatory loci that were more accessible in the EBV⁺ NF-κB activated state versus uninfected cells. Other upregulated genes included mediators of apoptosis and tumor suppression (BCL2A1, TNFRSF8, PDCD1LG2, ST7, IQGAP2, TOPBP1, CD86); proliferation (CDCA7L, MKI67); and B cell signaling (NFKBIA, MAPK6, TNIP1, TRAF3). Collectively, (2) ! (3 ∪ 8) DAP-linked DEGs underscored the importance of host chromatin remodeling for cell activation, proliferation, and apoptotic resistance facilitating immortalization (Table S4).

We next explored DAP-linked DEGs between EBV-induced NF-κB activation (c2) and differentiation (c5; Figure 2D). Activated cells exhibited 999 peaks absent from differentiated cells ([2] ! [5] DAPs), while differentiated cells displayed only 13 exclusive peaks, corresponding to a 15% reduction in accessibility. Notably, (2) ! (5) DAPs were linked to 13.4% of all DEGs between these states identified from scRNA-seq. One example was the TNFAIP3 (A20) locus, which contained ten significant linked positive regulatory loci. Conversely, DAP-linked DEGs upregulated from c2 to c5 included plasmablast-specific transcriptional regulators (e.g., XBP1, PRDM1), translation factors, and protein export machinery (i.e., facilitators of Ig synthesis and secretion; Table S4). CD38, a marker of MBCs, was also a DAP-linked DEG with enhanced accessibility in c5 versus c2. These results demonstrate that successful infection reprograms cellular chromatin not only to support activation, proliferation, and survival but also the subsequent transition to effector states.

Decreased accessibility at EBNA binding sites silences B cell response genes in EBV⁺ arrest

We examined gene silencing associated with inaccessible chromatin in EBV⁺ arrest, particularly at expression-linked DAPs coinciding with EBV nuclear antigen (EBNA) ChIP-seq footprints (Figure 3A). We preliminarily validated reductions in ADAM28 and COBLL1 expression, which are epigenetically suppressed targets of EBNA-3C,^43,44 and found that downregulation of these genes was extensively linked to chromatin closure at EBNA-3C binding sites (Figure S4). Consistent with GO results from DAPs alone, many genes that mediate B cell signal transduction and activation displayed accessibility-linked repression in EBV⁺ arrest. Frequently, expression of these genes was linked to multiple regulatory loci including EBNA-LP, EBNA-2, and/or EBNA-3C ChIP-seq sites (Figures 3B-3F). Expression-linked DAPs at the IL21R locus, including two EBNA-2 sites, were significantly depleted in EBV⁺ arrest (c4 and c7) relative to resting B cells and successful infection (Figure 3B). This finding was noteworthy, since B cell-specific IL21R signaling is essential for GC engagement in vivo.^45-47 Similarly, EBV⁺ arrest states displayed accessibility-linked repression across the CD58 (LFA-3) locus, including an EBNA-LP site (Figure 3C). Although loss of CD58 expression facilitates escape from immune surveillance in B cell lymphoma,⁴⁸ CD58 engagement in vivo with T cell-expressed CD2 is essential for actuating B cell responses to antigen.⁴⁹ EBV⁺ arrest states also exhibited DAP-linked downregulation at EBNA ChIP-seq sites near MAPK8 and PDE4D, which encode B cell receptor (BCR) and Toll-like receptor (TLR) signaling mediators (Figures 3D and 3E).^50,51 Four peaks including three expression-linked DAPs upstream of the Jun N-terminal kinase (JNK)-encoding MAPK8 locus coincided with EBNA-2 sites (Figure 3D). Accessibility-linked MAPK8 loss in EBV⁺ arrest states is predicted to inhibit B cell survival and transformation on the basis of the gene’s essentiality.⁵⁰ The enzyme encoded by PDE4D degrades the secondary messenger cAMP, which, in activated T cells, removes a restraint to TCR signaling.⁵² By analogy, accessibility-linked loss of PDE4D expression in B cells (Figure 3E) may impede signal transduction and NF-κB activation downstream of viral LMP-2A, which mimics BCR signaling,⁵³ via cAMP accumulation.⁵⁴ Finally, EBNA-associated accessibility-linked expression was observed for S1PR2, which encodes a GC B cell-specific surface receptor and helps maintain GC localization and homeostasis.⁵⁵ Although higher S1PR2 expression was observed in c7 and EBV⁺ success states, c4 cells did not exhibit accessibility-mediated S1PR2 upregulation (Figure 3F). A similar pattern was observed at the adjacent DNMT1 locus, whose product is a DNA methyltransferase that restrains viral latency.⁵⁶ These findings suggest that failure of EBNAs to maintain or induce accessibility-linked expression of B cell signaling and adaptive response genes contributes to EBV⁺ arrest outcomes.

Figure 3. Chromatin closure at signaling and proliferation loci in EBV⁺ arrest.

(A) Multimodal strategy to evaluate accessibility-linked silencing in EBV⁺ arrest versus resting and EBV⁺ success states.

(B) Expression-linked accessibility at IL21R. Significant associations between accessibility and expression are depicted by links. Peak tracks are color-coded by resting (blue), EBV⁺ arrest (green) and EBV⁺ success (gold). Violin plots of gene expression are presented by these groups and scRNA-seq model clusters (purple, c8/resting naive; blue, c3/resting memory; seafoam green, c4/EBV⁺ innate arrest; light green, c7/EBV⁺ hyperproliferative arrest; yellow-green, c6/EBV⁺ hyperproliferation; orange, c2/EBV⁺ NF-κB activation). Select expression-linked DAPs emphasized by red arrows are highlighted. Co-localized EBNA ChIP-seq peaks are denoted by stars.

(C) Expression-linked accessibility near CD58, as presented in (B). (D) Expression-linked accessibility near MAPK8, as presented in (B) and (C).

(E)Expression-linked accessibility near PDE4D, as presented in (B)–(D).

(F) Expression-linked accessibility near S1PR2/DNMT1, as presented in (B)–(E).

(G) Expression-linked accessibility near CDCA7, as presented in (B)–(F).

(H) Expression-linked accessibility near MKI67, as presented in (B)–(G).

(I) Expression-linked accessibility near POLA1, as presented in (B)–(H).

(J) Expression-linked accessibility near NEK2, as presented in (B)–(I).

See also Figure S4.

Downregulation of DAP-linked DEGs involved in proliferation also defined EBV⁺ arrest (Figures 3G-3J). CDCA7 (JPO1), a MYC target that promotes lymphoblastic transformation,^57,58 was expressed in successful infection but not resting or EBV⁺ arrested cells (Figure 3G). CDCA7 expression was significantly linked to an upstream EBNA-3C associated DAP. Consistent with prior work dissecting EBV-induced arrest versus proliferation,¹⁰ MKI67 expression was significantly lower in arrest versus success trajectories. Attenuated MKI67 expression in c4 was linked to reduced accessibility at an upstream EBNA-LP site (Figure 3H). An EBNA-2 associated DAP linked to expression of the DNA polymerase (Pol) catalytic subunit POLA1 was closed in arrest phenotypes, implying defective DNA replication (Figure 3I). Another significant EBNA-associated DAP-linked DEG, NEK2, encodes a centrosome-associated kinase that regulates mitosis (Figure 3J). Loss of accessibility and NEK2 expression in c4 further indicated impaired proliferation.

Accessibility-mediated repression of these genes along the arrest trajectory was consistent with inhibited cell activation and proliferation, which contrasts with functional B cell adaptive responses and successful EBV infection. Collectively, these data support an antiviral role for targeted host chromatin closure as well as the necessity for viral oncoproteins to counteract this restriction mechanism to establish latency.

EBV induces GC B cell-like chromatin during successful infection despite BCL6 downregulation

Next, we investigated DAP-linked DEGs between uninfected B cells and EBV⁺ success phenotypes (c6: GC DZ-like; c0: AP-eMBC-like; c2: GC LZ-like; and c5: post-GC memory/plasmablast; Figure 4A). Because EBV induces GC-like expression signatures,^7,30,59-61 we focused our analysis on DAP-linked DEGs that are GC B cell biomarkers and transcriptional regulators and integrated ATAC-seq data from tonsillar B cells⁸ and EBNA ChIP-seq data.

Figure 4. Viral subversion of GC B cell regulatory control in EBV⁺ success trajectories.

(A) Multimodal evaluation of accessibility-linked expression in EBV⁺ success GC phenotypes versus resting B cells.

(B) Expression-linked accessibility near BCL6 in EBV⁺ success versus resting B cells. IGV visualization (top panel) depicts bulk ATAC peaks from LCL, day 0, and day 8; bulk ATAC from tonsillar B cell fractions (two biological replicates); and ChIP-seq peaks for EBNA-LP, EBNA-2, and EBNA-3C. Coverage plots (bottom panel) depict significant DAP-linked expression in the same region within scATAC phenotypes (purple, c8/resting naive; blue, c3/resting memory; yellow-green, c6/EBV⁺ hyperproliferation; light yellow, c0/EBV⁺ AP-eMBC; orange, c2/EBV⁺ NF-κB activation; red, c5/EBV⁺ pre-plasmablast). Stars denote EBNA-associated expression-linked DAPs. Regions outlined in black highlight conserved state-resolved phenotypes between tonsil subsets and EBV-induced states. Regions outlined in red depict sites at which chromatin accessibility of in vivo tonsillar subsets and their EBV-induced analogs deviate. Detail of select DAPs and expression violin plots by model phenotype are included.

(C) Expression-linked accessibility near CXCR4 in EBV⁺ success versus resting B cells, depicted as in (B).

(D) Expression-linked accessibility near LMO2 in EBV⁺ success versus resting B cells, depicted as in (B) and (C).

(E) Expression-linked accessibility near AICDA in EBV⁺ success versus resting B cells, depicted as in (B)–(D).

(F) Expression-linked accessibility near BHLHE40 in EBV⁺ success versus resting B cells, depicted as in (B)–(E).

(G) Expression-linked accessibility near CD83 in EBV⁺ success versus resting B cells, depicted as in (B)–(F).

(H) Expression-linked accessibility near CD86 in EBV⁺ success versus resting B cells, depicted as in (B)–(G).

(I) Expression-linked accessibility near EPHB1 in EBV⁺ success versus resting B cells, depicted as in (B)–(H).

BCL6 is the master regulator of GC B cell responses. Although BCL6 mRNA is expressed in resting naive B cells, BCL6 protein expression in vivo depends on activation and T cell help.⁶² Consistent with BCL6-repressive functions of EBV microRNAs and EBNAs,^63-65 rapid BCL6 loss was observed in all post-infection states (Figure 4B). BCL6 downregulation in EBV⁺ success (c6, c0, c2, and c5) was linked to closure of an upstream locus, which was also inaccessible in tonsillar GC DZ, LZ, and plasmablast subsets. An adjacent EBNA-LP binding site was accessible across tonsillar GC subsets but became inaccessible in EBV⁺ states in vitro; this locus was not significantly linked to BCL6 expression. Further comparisons of de novo EBV ATAC profiles with analogous tonsil phenotypes revealed divergence from conventional GC B cell programming. For example, the GC DZ homing receptor CXCR4^62,66 was downregulated upon infection as found in prior studies.⁶⁷ CXCR4 repression was significantly associated with closure of multiple EBNA-associated DAPs (Figure 4C). Again, reduced accessibility at several of these sites contrasted with open chromatin in tonsillar DZ B cells.

Despite repression of BCL6 and CXCR4, chromatin profiles at many loci encoding GC biomarkers and transcriptional regulators were faithfully recapitulated during EBV infection. For example, c6 (DZ-like) and c2 (LZ-like) EBNA-associated loci near LMO2, an oncoprotein expressed in GCs and GCBDLBCL,⁶⁸ closely resembled accessibility in GC DZ and LZ cells (Figure 4D). Likewise, peaks exclusive to tonsillar DZ and LZ cells near the GC hallmark AICDA^69,70 were also uniquely accessible in c2 and c6 during early infection (Figure 4E). A multi-EBNA site upstream of the AICDA TSS was only accessible in LZ tonsil cells in vivo but was open in EBV-induced DZ-like (c6) and LZ-like (c2) states. The transcription factor BHLHE40, which is required for and restrains the GC reaction in vivo,⁷¹ had an upstream EBNA-associated expression-linked DAP not observed in tonsil subsets. This peak was absent in resting cells in vitro but accessible in c2 (Figure 4F).

The closest EBV-induced mimicry of GC B cell chromatin and expression was observed between c2 and LZ cells. Although expression of the GC LZ marker CD83 was downregulated in the ensemble cell population following infection, its accessibility-linked expression was conserved in c2 (Figure 4G). Although this was consistent with LMP-1 upregulation of CD83 through NF-κB signaling,⁷² significant linkages to EBNA-2 and EBNA-3C DAPs suggest that other latency genes may promote or maintain CD83 expression. CD86, another marker of activated LZ B cells, was likewise enriched in tonsil GCs and c2, with significant linkage to an EBNA-LP DAP (Figure 4H). EPHB1, a marker of LZ MBC precursors in vivo,⁷³ was expressed in c2 and linked to two EBNA-associated DAPs (Figure 4I). One locus was consistently accessible in tonsil DZ, LZ, and plasmablasts, while another was a multi-EBNA binding site inaccessible in tonsils. EBV-induced EPHB1 expression was noteworthy with regard to immune evasion and viral malignancy, as EPHB1 limits T_FH cell recruitment to GCs in vivo ⁷⁴ and is a marker of EBV⁺ Reed-Sternberg cells.⁷⁵

Collectively, this analysis identified extensive virus-driven alterations of host chromatin at genes critical to GC B cell dynamics. Several discrepancies, especially BCL6 downregulation, underscore the dysregulation of EBV-driven dynamics relative to in vivo GCs. However, the extent of EBV-induced GC B cell chromatin and gene expression mimicry reveals viral subversion of host adaptive responses at multiple regulatory levels.

EBV modulates accessibility at transcription factor binding motifs in a dynamic GC-like reaction

To further investigate phenotype-specific regulation, we assayed TF binding site (TFBS; i.e., motif) enrichment by state (Figure S5). We found variable motif enrichment linked to resting cells and among non-arrested EBV⁺ phenotypes (Figure S5A). Variable motif accessibility broadly matched gene expression with respect to antiviral response induction, cell growth, and oncogenesis (Figures S5B and S5C). Activated B cell (c2) ATAC peaks were enriched in binding sites for proto-oncogenic TFs in the REL (cREL, RELA, RELB), AP-1 (FOS, FOSB, JUNB, JUND), and EGR (EGR1–4) families. Enhanced accessibility at NF-κB TFBS in c2 was consistent with the corresponding upregulation of NF-κB targets. Likewise, differentiated cells (c5) were enriched in accessible motifs for IRF4, IRF8, and XBP1. Globally, resting B cells and the innate sensing arrest state shared the greatest motif correlation (R > 0.75) and the lowest correlation to EBV-activated and hyperproliferating cells (0.55 < R < 0.7; Figure S5D).

We examined EBV-induced reprogramming through integrated analysis of motif accessibility and expression (Figure 5). Differentially accessible motifs between uninfected naive versus MBCs were reflected within infected non-GC versus GC (DZ and LZ) phenotypes. TFBS with enhanced accessibility in EBV⁻ MBCs and EBV⁺ LZ cells included motifs for JDP2, FOS/JUN AP-1, and NFE2L2. Conversely, motifs with enriched accessibility in EBV⁻ naive B cells and EBV⁺ non-GC cells included those for POU2F2, BHLHE40, and MEF2C (Figures 5A and 5B). We characterized motif enrichment within EBV⁺ non-GC, LZ, DZ, arrest, and pre-lytic states (Figure 5C) and mapped scRNA-seq clusters onto the motif accessibility landscape (Figure 5D). Pairing this integration with pseudotime analysis resolved dynamic motif accessibility signatures for virus-induced GC entry, engagement, and exit. In addition to increased NF-κB, REL, FOS, and JUN family motif accessibility in LZ cells, we observed anticorrelated YY1 and BHLHE40 motif accessibility in the transition to the LZ state (Figure 5E).

Figure 5. Motif accessibility dynamics during EBV-induced germinal center (GC) engagement.

(A) UMAP of genome-wide transcription factor motif accessibility from day 0 (uninfected) and day 8 by time point (top panel) and phenotype (bottom panel).

(B) Top differentially accessible (DA) motifs between EBV⁺ non-GC and light zone (LZ) states mirror DA between naive and memory B cells. Statistical comparisons evaluated via Kolmogorov-Smirnov (K-S) test D statistic (***p < 2.2e-16).

(C) EBV⁺ phenotype-resolved DA motifs from day 8. Analysis of transcription factors with differential enriched motif between GC-LZ and non-GC states. Cas9/RNP-based MEF2C-KO cells displayed reduced ability to transition from ICAM1^Hi/CD27^Lo (GC-LZ state) into ICAM1^Lo/CD27^Hi (non-GC state). NFE2L2 KO cells interconvert faster to differentiated subsets.

(D) Cell-matched expression and motif accessibility dynamics in EBV⁺ B cells (day 8).

(E) Pseudotime analysis and correlation of phenotype-resolved motif accessibility in EBV⁺ cells (day 8).

(F) Functional analysis of transcription factors with differential motif enrichment between GC-LZ and non-GC states. The transition from sorted CD46⁻/ICAM1^Hi/CD27^Lo fraction into ICAM1^Lo/CD27^Hi was measured for knockouts of MEF2C and BHLHE40 (non-GC states) and NFE2L2 and JDP2 (NF-κB, LZ). Statistical differences were calculated via two-way ANOVA with post-hoc Dunnett method (*p < 0.05% and **p < 0.01; n = 2 technical replicates).

See also Figures S5 and S6.

To assay functional roles of these transcription factors in supporting GC-like transitions, we knocked out each of four TFs whose motifs were either enriched (JDP2 and NFE2L2) or depleted (BHLHE40 and MEF2C) between LZ-like (c2) and non-GC/differentiated (c5) states. We used a fluorescence-activated cell sorting (FACS)-based assay to capture dynamic transition between these states,⁵⁹ where sorting and replating LCLs from the c2/GC LZ state (ICAM1^hi/CD27^lo) enables monitoring of their transition to the c5/differentiated (ICAM1^lo/CD27^hi) state (Figure 5F). CRISPR knockout (KO) of each TF was assessed by co-KO of the inert proxy CD46 ⁷⁶ and validated by sequencing to confirm genetic lesions (Figure S6A). Functional NFE2L2 (NRF2) KO was further validated by western blot (Figure S6B). NFE2L2 KO yielded faster accumulation of differentiated cells, consistent with its predicted role in restraining this transition (Figures 5B and 5F). In contrast, loss of functional MEF2C via MADS DNA binding domain deletion⁷⁷ impeded accumulation of differentiated cells observed in the control CD46 KO (Figures 5B, 5F, and S6). We did not observe significant effects of JDP2 or BHLHE40 KO on activation and differentiation states, suggesting that these were not critical in regulating this transition in LCLs.

Post-infection phenotypes resolved by differential EBV episomal locus accessibility

We also investigated viral genome accessibility during early infection. Twenty-one viral ATAC peaks (20 of 21 in both donors) were detected, including at TSSs for essential latency genes such as the EBNAs and LMP1. Episome ATAC signals in a negligible number of B cells at day 0 (n = 3, <0.02% of cells) were attributed to pre-existing wild-type infection (Figure S7A). EBV⁺ activated (c2) and hyperproliferative cells (c6) had the greatest number of accessible episomal loci relative to other post-infection phenotypes. These sites included the C promoter (Cp) for EBNA1, EBNA2, and EBNA3A–3C; the LMP1 TSS; the TSS for BMRF1, a DNA Pol accessory protein; and the BHLF1 locus, which facilitates latency and immortalization (Figures S7B and S7C).⁷⁸ Innate sensing (c4) and growth arrest states (c7) had the fewest cells with episomal ATAC peaks (Figure S7D).

Informatic inference of phenotype-resolved TF signatures and regulatory elements

The prevalence of DAP-linked DEGs modulated by EBV gene products led us to interrogate genome-wide phenotype TF signatures. We reasoned that ensemble-averaged ChIP-seq data from LCLs would contain TF binding and epigenetic data representing a superposition of EBV⁺ phenotypes at high coverage, thus maximizing chances to identify overlaps with comparatively sparse scATAC-seq data. We further extrapolated that phenotypic variation in TFBS accessibility would underlie distinct infection outcomes. Finally, empirical differential expression measurements of genes identified by cross-referencing cluster ATAC peaks with ChIP footprints of interest could be retrieved from cell-matched scRNA-seq. This tri-modal integration was expected to clarify the heterogeneity of putative regulatory relationships in early infection (Figure S8A). Thus, we employed a bioinformatic workflow to obtain ChIP-seq referenced inference of single-cell phenotypes from scATAC-seq data (“crisp-ATAC”). This process applies genomic range intersection “recipes” between scATAC-seq clusters and bulk ChIP-seq datasets in conjunction with cis-regulatory prediction (Figures S8B). During initial testing, we found that logic-gated peak intersections across multiple ATAC-seq and ChIP-seq datasets frequently exhibited “noisy” short intervals (<250 bases) due to minor offsets in called peaks across input datasets. However, crisp-ATAC consistently recovered “signal” intervals (~1,100 base median) representing DAPs associated with ChIP-seq signals from the applied recipe (Figure S8C). Most noise intervals could be suppressed by incorporating biological replicates. Specifically, DAPs between the same phenotype across replicates were dominated by noise, whereas DAPs between different phenotypes and different replicates contained noise and signal intervals. Logical gating that recovered DAPs present in one phenotype and absent in another across replicates effectively filtered noise (Figure S8D). This allowed comparisons of DAP frequencies and interval length distributions across all pairwise phenotype comparisons (Figure S8E). Thus, replicate-based interval filtering provided essential QC for all crisp-ATAC predictions.

Crisp-ATAC finds TF-linked expression signatures that vary by EBV-induced fate

We used crisp-ATAC to predict EBNA and LMP1-mediated NF-κB accessible sites at promoters, enhancers, and actively transcribed genes in each phenotype from ChIP-seq peaks for EBNAs,^44,79,80 NF-κB/Rel TFs,^81,82 H3K4me1, H3K4me3, H3K27ac, H3K36me3,⁸³ and RNA Pol II^84,85 (Figure S8F). Hyperproliferative (c6), activated (c2), and resting MBCs (c3) exhibited up to 3-fold more enhancers and promoters at EBNA-2 sites relative to naive B cells (c8), other activation intermediates (c0 and c1), plasmablasts (c5), arrested states (c4 and c7), and non-B cells (c9 and c10). Similar patterns were observed for EBNA-3C and EBNA-LP as well as Rel family TFBSs (Figure S8G). Enhancers, promoters, and actively transcribed genes were enriched in c2, c3, and c6 and depleted in c4 and c7, with intermediate levels in c0, c1, c5, and c8.

We assayed regulatory variation associated with key viral transcriptional co-activators, starting with innate sensing arrest (c4) and NF-κB (c2) states (Figure 6A). Peaks were extracted and gated to obtain (2) ! (4) DAPs (n = 1,873) and cis-linked gene predictions (n = 1,514; Figures 6B and 6C). The (2) ! (4) linked gene ontology was enriched for innate defense (inflammation, antimicrobial processes) and EBV-induced responses (lymphocyte activation, regulation of apoptosis; Figure 6D; Table S4). Overall, (2) ! (4) linked gene predictions included 42.5% (71/167) of known EBV super-enhancer (EBVSE) site-linked genes.¹⁸ Consequently, 41%–55% of EBNA-associated DAPs overlapped a peak for the super-enhancer-associated host TF RelA. Of 71 identified EBVSE-linked genes, 19 (27%) were linked to EBNA-LP peaks, 22 (31%) to EBNA-2 peaks, and 15 (21%) to EBNA-3C peaks. EBVSE-linked genes were enriched in EBNA-associated DAP-linked DEGs relative to size-matched random samples of the captured transcriptome (Figure 6E).

Figure 6. crisp-ATAC analysis of DAP-linked DEGs in activated versus innate arrested EBV⁺ B cells.

(A) Schematic of NF-κB activation (c2) and innate arrest (c4) phenotypes.

(B) Multimodal gating to extract c2!c4 DAPs.

(C and D) Prediction of cis-regulatory linkage. All c2!c4 peak intervals (1,873) are input to GREAT³³ to predict c2!c4 DAP-linked genes (1,514).

(E) Occurrence of EBVSE-linked genes identified as c2!c4 DAP-linked DEGs associated with EBNA binding sites relative to overlap frequency by random chance (n = 100 simulation trials, error bars depict mean ± SD). Random samples were size-matched to EBNA-associated gene lists.

(F) Analysis for TRAF1, an EBVSE-linked gene identified as a c2!c4 DAP-linked DEG with phenotype-variable accessibility at multiple EBNA sites.

See also Figures S8 and S9.

We analyzed genes of interest on the basis of EBVSE status, GO process involvement, and/or empirically demonstrated importance to EBV infection. The NF-κB activated gene and signal transducer TRAF1, whose gene product interacts with LMP-1, matched all three criteria above.^86-90 We found (2) ! (4) expression-linked DAPs associated with one or more EBNA at −3, +2, and +37 kb from the TRAF1 TSS (p < 0.05, correlation Z score). Furthermore, these loci exhibited reduced accessibility in arrest (c7), activation intermediate (c1) and differentiated (c5) states versus c2 (Figure 6F).

A grouped comparison ([2 ∪ 5 ∪ 6] ! [3 ∪ 8]) identified changes associated with successful EBV-induced immortalization (Figure 7; Table S5). Despite net accessibility loss after infection, 245 genes linked to 1,824 peaks were present in the EBV⁺ states but not resting cells (Figures 7A-7C). One hundred sixty-six of 245 genes were linked to at least one EBNA site, and 18 genes overlapped with EBVSE targets (7.3% of predicted genes, 10.8% of known EBVSE genes). Only 31 GO terms were shared across the top 100 terms for each EBNA, accounting for 15% of unique terms (Figure 7D). We selected the (2 ∪ 5 ∪ 6) ! (3 ∪ 8) ∩ EBNA targets TNFRSF8 (CD30), CD274 (PD-L1), and PDCD1LG2 (PD-L2) based on their relevance to EBV-associated lymphomas. These included three EBNA-2 sites near the TNFRSF8 TSS (−17, −12, and +16 kb) and a multi-EBNA site −17 kb from the PDCD1LG2 TSS and +43 kb from the end of CD274 (Figure 7E). These loci were enriched for Rel sites and activating histone marks in LCLs. Cas9/RNP KOs were used to investigate the functional role of the putative enhancer located between PDCD1LG2 and CD274 (PD-L1). Relative to an adjacent control region, deletion of this enhancer led to a reduction in PD-L1 expression (Figure 7F), supporting a role for viral oncoprotein-mediated alterations to chromatin accessibility that facilitate immunoevasion.

Figure 7. crisp-ATAC analysis of DAP-linked DEGs along the successful infection trajectory.

(A) Schematic overview of model states used in comparison (c256 vs. c38 peaks, both donors).

(B) Gating to identify c256!c38 DAPs (1,609).

(C) Prediction of DAP-linked genes (1,384).

(D) GO network of c256!c38 DAP-linked DEGs at EBNA-associated sites and overlap with EBVSE-linked genes (18 of 167). The Venn diagram depicts GO process enrichment by EBNA for associated DAP-linked DEGs.

(E) Joint coverage plots for genes of interest (TNFRSF8, CD274, and PDCD1LG2) with EBNA-associated linked sites highlighted.

(F) Functional validation of putative enhancer located between PDCD1LG2 and CD274. Enhancer site deletion was generated using gCD274_dss gRNAs (green). As a control, a nearby region was targeted with gCD274_ctrl gRNAs (yellow). FACS histogram of CD274 expression in CD46⁻ gated LCLs. Statistically significant differences in MFI were calculated using one-way ANOVA with the Dunnett post hoc method (***p < 0.001; n = 4 technical replicates).

See also Figures S8 and S9.

In a final example, we evaluated activated versus differentiated EBV⁺ phenotype DAPs ([2] ! [5]) associated with EBNA binding to explore regulatory control delineating key phenotypes present in LCLs ¹³ (Figure S9A). Five hundred nineteen of 999 identified (2) ! (5) peaks intersected with at least one EBNA site, from which 247 genes were predicted. Twenty-nine of 110 (2) ! (5) ∩ EBNA-2 site-linked genes (26.3%) were DEGs in the scRNA assay, as were 34 of 115 (2) ! (5) ∩ EBNA-LP site-linked genes (29.6%) and 20 of 125 (2) ! (5) ∩ EBNA-3C site-linked genes (16.0%). Twenty genes were identified by including all three EBNAs (Figure S9B), including the EBVSE-linked GPR137B, a regulator of mTORC1 activity and autophagy.^91,92 GPR137B was also a (2) ! (3 ∪ 8) DAP-linked DEG, indicating that loci regulating GPR137B expression after infection were initially inaccessible within resting cells. Two EBNA-associated expression-linked loci were found +14 and +18 kb relative to the GPR137B TSS. One site (+14 kb) coincided with EBNA-LP and EBNA-3C binding sites but did not exhibit a (2) ! (5) DAP. The second site (+18 kb) overlapped with all three EBNAs and was a DAP. Both sites intersected with Rel family TFs (Figure S9C). Other genes involved in lysosome-mediated processes including autophagy and antigen presentation regulation were identified from the (2) ! (5) comparison (TFEB and LAMP3), albeit with modestly elevated but significant expression.^93-95 Other EBNA-associated DAP-linked DEGs involved in immune signaling, apoptosis, and transcription were also identified (Figures S9D and S9E). Collectively, crisp-ATAC analyses of early infection phenotypes refined the genome-wide dynamics of heterogeneous EBV-induced B cell responses.

DISCUSSION

This study provides a detailed view of chromatin accessibility landscapes that develop upon EBV infection. To facilitate high-resolution investigations of EBV-mediated regulatory control, we present annotated infection phenotypes determined from single-cell multiomics. Bioinformatic integration enabled gene- and peak-level interrogation of host and viral epigenetic signature variation between these phenotypes. The combination of resolution, multimodality, and integrative analysis yield a vividly detailed representation of the genome-wide interplay of host and virus.

Because of asynchronous parallel emergence of heterogeneous cell responses, early infection chromatin dynamics cannot be fully resolved with ensemble sequencing. Remarkably, we find transient genome-wide euchromatin-to-heterochromatin transitions (20%–40% reduced accessibility) in some post-infection trajectories. Cell-to-cell variation in chromatin accessibility following infection warrants consideration of Simpson’s paradox. That is, the total number of unique peaks across ensemble populations increases after infection (Table S1; consistent with prior studies), while the number of peaks per cell in fact decreases along several trajectories.^96,97 Although scATAC-seq data sparsity likely contributes to underestimation of accessible loci, variable peak signatures and frequencies among distinct fates suggest the importance of a balance between drivers of epigenetic silencing (e.g., arrest and senescence via antiviral and damage response induction) and activation (e.g., oncogenic EBVSE formation) in determining post-infection cell fate.

In oncogene-induced senescence, DNA damage triggers SAHF formation and epigenetic silencing.³⁷ Our lab found that EBV induces γH2AX foci via ATM/Chk2-mediated DDR signaling ⁹ and H3K9me3 heterochromatin induction¹⁰ in cells that arrest during viral oncoprotein-driven hyperproliferation. Another subset of EBV⁺ cells failed to proliferate but also exhibited DDR induction on the basis of γH2AX and pChk2.⁹ Similarly, here we find substantial loss of accessible chromatin and expression in cells that undergo innate sensing-mediated arrest (c4) or evade this response but arrest due to DDR activation (c7) during hyperproliferation (c6). Our data also indicate a role for ribosome biogenesis in the transition from virus-induced arrest to senescence, likely through a p53-MDM2 axis,⁹⁸ which regulates EBV transformation.⁹⁹ Chromatin closure in EBV-induced arrest was enriched at loci linked to B cell activation, signaling, and proliferation. Many such sites overlap with footprints for EBNAs, several of which (EBNA-1, EBNA-3C) have been shown to directly promote DNA damage through mechanisms including double-stranded break induction.¹⁰⁰ Moreover, EBNA-LP co-localizes with γH2AX foci⁹ and binds to proteins within promyelocytic leukemia nuclear bodies (PML NBs) that sense DNA damage.^101,102 Because ATM-mediated DDR silences genes near DSBs by impeding RNA Pol II-driven chromatin decompaction,¹⁰³ it is conceivable that local damage at EBNA sites contributes to this targeted heterochromatinization in EBV-induced arrest. Although precise mechanisms are unclear, this finding suggests that DDR-induced heterochromatinization in response to EBV oncoprotein function simultaneously enacts antiviral restriction and ablates the infected cell’s adaptive response capacity.

Successful infection is distinguished by increased accessibility near key genes against the backdrop of heterochromatin induction, presumably by avoiding extensive DDR activation.^9,10 Likewise, episomal peak heterogeneity across EBV⁺ phenotypes confirms that latency establishment depends on retained accessibility to viral genes within the repressive host milieu. In successful infection, a substantial number of accessible cellular loci have predicted cis linkages to genes enriched for regulation of apoptosis, tumor suppression, inflammation, and chromatin remodeling. We found that 10 of 87 genes (11.5%) shown to be essential for EBV-induced B cell immortalization³⁵ had significant expression-linked DAPs. Seven of these (BATF, CDK6, CFLAR, IRF2, NFATC2IP, NFKB2, and SP110) exhibited loss of basal accessibility in EBV⁺ arrest relative to EBV⁺ success trajectories. Conversely, DAPs at three genes (CCND2, IRF4, and MDM2) were closed in resting cells and failed to open in EBV⁺ arrest but not EBV⁺ success trajectories. Thus, maintenance of existing accessible loci and induction of accessibility at previously closed regulatory sites constitute essential host epigenomic alterations for infection.

EBV-induced subversion of GC B cell subset regulatory architecture is especially noteworthy. Hyperproliferative cells (c6) mirror GC DZ B cells through elevated levels of LMO2, AICDA, and proliferation genes linked to EBNA-associated accessible chromatin. Likewise, NF-κB activated cells (c2) reflect GC LZ cell chromatin linked to expression of CD83, CD86, and EPHB1. Post-GC effector states (c5) recapitulate in vivo regulatory control of differentiated B cell hallmarks (PRDM1, CD38). Most of these and other GC-associated genes are linked to regions that overlap EBNA footprints. Downregulation of BCL6, a distinctive feature of EBV-induced B cell dynamics relative to GCs in vivo, prevents oncogene repression,¹⁰⁴ thereby facilitating latency and priming malignant potential. Although EBV induces several genes that are repressed by BCL6 (e.g., BCL2, MYC), the extent of EBNA binding at GC regulatory loci indicates a previously unappreciated degree of viral epigenomic manipulation of GC dynamics. EBV-induced EPHB1 and CD274 (PD-L1) expression and CXCR4 downregulation may enable checkpoint surveillance bypass by mitigating potentially suppressive interactions with other immune cells involved in GC progression. Along with immunoevasive functions of EBV latency proteins, these phenotypic characteristics would facilitate viral access to and persistence in MBCs.

Transcription factor motif analysis provides further evidence of virus-induced establishment of GC regulatory dynamics. Among these were significant reductions in YY1 and BHLHE40 motif accessibility in the transition to the LZ state and the adoption of MBC-like chromatin architecture. Differential YY1 motif accessibility between non-GC and LZ states is noteworthy given YY1’s foundation role in gene regulation and cellular identity via promoter-enhancer looping.¹⁰⁵ Likewise, accessibility loss at BHLHE40 sites upon GC engagement is consistent with the protein’s role in restraining GC entry.⁷¹ This multiomic dataset thus provides a powerful resource for future investigations of cell state variation dependent on combinatorial motif accessibility and corresponding TF expression. As a demonstration, we validated the functional significance of TFs with predicted roles in EBV-induced GC regulation from cluster-resolved motif accessibility using reverse genetics. Specific loci predicted to regulate gene expression in cis were similarly validated.

Depending on the phenotype comparison, ~10%–35% of DEGs were DAP-linked. This range was similar to the frequencies of differentially accessible motifs for expressed TFs (23%) and DEGs associated with differentially accessible promoters (25%) identified in dexamethasone-treated A549 cells using sci-CAR.²⁸ These similar frequencies in response to diverse stimuli (e.g., glucocorticoid receptor activation, viral infection) raise intriguing questions regarding the fundamental frequency of cellular response genes regulated by accessibility changes. The observed proportion implies that most DEGs may be regulated by higher order chromatin interactions and differential recruitment of transcriptional activators and/or RNA Pol II.

The crisp-ATAC method provides a simple and flexible approach to map ChIP-seq profiles to phenotypes discovered from scATAC-seq. Thus, it can be used to bootstrap scATAC-seq resolution to other omics datasets. When data from suitable reference samples are available, crisp-ATAC can predict phenotype-resolved regulation by evaluating simple or complex combinations of binding sites and histone modifications across differentially accessible sites. In our case, cell-matched empirical scRNA-seq data can be cross-referenced with crisp-ATAC outputs. We expect this method is generally adaptable for comparing cell phenotypes in contexts such as development and responses to stimuli or drugs. This approach should be particularly powerful for exploring time-resolved TF-associated control of cellular behaviors.

Limitations of the study

This study has several limitations. We did not capture aspects of host-virus dynamics acting at other molecular levels. Examples include DNA methylation, three-dimensional chromatin architecture, changes in protein abundance, post-translational modifications, protein-protein interactions, and signaling activity. Although we present a method for inferring DAP-linked TF binding and epigenetic modifications based on empirical scATAC-seq data, we do not have direct ChIP-seq measurements at the single-cell level.

The crisp-ATAC method also has notable constraints. This approach is limited by the availability of ChIP-seq data from appropriate reference samples. Moreover, regulatory predictions must be empirically tested to validate potential functions in gene expression control. Finally, distance limits imposed for identifying cis-regulatory linkages preclude identification of distal gene regulatory elements formed via long-range three-dimensional (3D) nuclear conformation.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests should be directed to and will be fulfilled by the Lead Contact, Micah A. Luftig (micah.luftig@duke.edu).

Materials availability

Cas9-RNP-generated knockout LCLs used in this study are available upon request from the Lead Contact.

Data and code availability

• Single-cell ATAC-seq data have been deposited in the NIH Sequence Read Archive (SRA: SRP346796) and Gene Expression Omnibus (GEO: GSE189141). These data are publicly available through the accession numbers and identifiers listed in the key resources table.

• Data analysis was performed in R using built-in functions and software packages described in the key resources table. R code used for analysis and visualization is available on Zenodo (https://doi.org/10.5281/zenodo.8125208). Additional information related to data analysis is available upon request.

• Additional information and requests for data resources should be directed to and will be fulfilled by the lead contact.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti CD27-APC	BioLegend (mouse anti-Human CD27, clone M-T271)	Cat. #356410; RRID: AB_2561957
Anti CD274-PE	BioLegend (mouse anti-Human CD274, clone 29E-2A3)	Cat. #329706; RRID: AB_940368
Anti CD46-APC	BioLegend (mouse anti-Human CD46, clone TRA-2-10)	Cat. #352405; RRID: AB_2564356
Anti CD46-PE	BioLegend (mouse anti-Human CD46, clone TRA-2-10)	Cat. #352402; RRID: AB_10895756
Anti ICAM1-Pacific Blue	BioLegend (mouse anti-Human ICAM1, clone HCD54)	Cat. #322716; RRID: AB_893384
Anti NRF2	Invitrogen (rabbit anti-Human NRF2, polyclonal)	Cat. #PA5-27882; RRID: AB_2545358
Anti IgG	Sigma Aldrich (goat anti-Rabbit IgG, polyclonal)	Cat. #12-348
Bacterial and virus strains
Epstein-Barr virus (EBV), B95-8 strain	Luftig Lab (Original publication: Johannsen et al., 2004).31	https://doi.org/10.1073/pnas.0407320101
Biological samples
Whole blood samples	Gulf Coast Regional Blood Center (Houston, TX)	N/A
Chemicals, peptides, and recombinant proteins
TrueCut™ Cas9 Protein v2	ThermoFisher Scientific	Cat. #A36499
Histopaque®-1077	Sigma-Aldrich (Raleigh, NC)	Cat. #H8889
Critical Commercial Assays
BD iMag B cell Negative Selection Kit	BD Biosciences (Franklin Lakes, NJ)	Cat. #558007
10x Chromium Next GEM Single	10x Genomics (Pleasanton, CA)	Prod. #1000285
Cell Multiome ATAC + Gene Expression Kit (2x)
Dual Index Kit TT Set A	10x Genomics (Pleasanton, CA)	Prod. #1000215
Single Index Kit N Set A	10x Genomics (Pleasanton, CA)	Prod. #1000212
10x Chromium Chip J	10x Genomics (Pleasanton, CA)	Prod. #1000230
NovaSeq S2 Reagent Kit v1.5	Illumina (San Diego, CA)	Cat. #20028314
NuPage 4–12% Bis-Tris Gel	ThermoFisher Scientific (Invitrogen)	Cat. #NP0322
PageRuler Pre-stained Protein Ladder, 10–250 kDa	ThermoFisher Scientific	Cat. #26619
Revert 700 Total Protein Stain	Li-COR (Lincoln, NE)	Cat. #926-11011
SuperSignal West Femto	ThermoFisher Scientific	Cat. #34096
Gene Knockout Kit v2 (See Table S3)	Synthego (Redwood, CA)	N/A
Synthetic sgRNA Kit (See Table S3)	Synthego (Redwood, CA)	N/A
Neon™ Transfection System 10uL Kit	ThermoFisher Scientific	Cat. #MPK1096
Deposited data
Donor 1 uninfected B cell scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695143; SRX13167303
Donor 2 uninfected B cell scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695151; SRX13167311
Donor 1 B cells day 2 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695145; SRX13167305
Donor 2 B cells day 2 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695153; SRX13167313
Donor 1 B cells day 5 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695147; SRX13167307
Donor 2 B cells day 5 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695155; SRX13167315
Donor 1 B cells day 8 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695149; SRX13167309
Donor 2 B cells day 8 post-EBV scATAC-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695157; SRX13167317
Donor 1 uninfected B cell scRNA-seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695144; SRX13167304
Donor 2 uninfected B cell scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695152; SRX13167312
Donor 1 B cells day 2 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695146; SRX13167306
Donor 2 B cells day 2 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695154; SRX13167314
Donor 1 B cells day 5 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695148; SRX13167308
Donor 2 B cells day 5 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695156; SRX13167316
Donor 1 B cells day 8 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695150; SRX13167310
Donor 2 B cells day 8 post-EBV scRNA -seq	This study (GEO accession GSE189141; SRA accession SRP346796)	GSM5695158; SRX13167318
GM12878 ATAC-seq	Buenrostro et al., 2013126 (GSE47753)	GSM1155957
GM12878 EBNA1 ChIP-seq	Tempera et al., 201679 (GSE73887)	GSM1905009
GM12878 EBNALP ChIP-seq	Portal et al., 201380 (GSE49338)	GSM1197603
Mutu III EBNA2 ChIP-seq	McLellan et al., 201344 (GSE47629)	GSM1153765
Mutu III EBNA3C ChIP-seq	McLellan et al., 201344 (GSE47629)	GSM1153766
GM12878 IgG control ChIP-seq	Gunnell et al., 2016127 (GSE76869)	GSM2039169
GM12878 RelA ChIP-seq	Zhao et al., 201482; Iannetti et al., 2014 81 (GSE55105)	GSM2628088
GM12878 RelB ChIP-seq	Zhao et al., 201482; Iannetti et al., 201481 (GSE55105)	GSM2628090
GM12878 cRel ChIP-seq	Zhao et al., 201482; Iannetti et al., 201481 (GSE55105)	GSM2628092
GM12878 CTCF ChIP-seq	Lee et al., 2012128 (GSE32883)	GSM489290
GM12878 RNA Pol II ChIP-seq	Raha et al., 201085 (GSE19550)	GSM487431
GM12878 H2AZ ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733767
GM12878 H3K4me1 ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733772
GM12878 H3K4me3 ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733708
GM12878 H3K9ac ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733667
GM12878 H3K9me3 ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733664
GM12878 H3K27ac ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733771
GM12878 H3K27me3 ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733758
GM12878 H3K36me3 ChIP-seq	ENCODE Project Consortium 201283 (GSE29611)	GSM733679
GM12878 Hi-C	Rao et al., 2014129 (GSE63525)	GSE63525_GM12878_dilution_combined_30.hic
Oligonucleotides
GAGAAACAUGUCCAUAUAUA	Synthego sgRNA	sgRNA_CD46_1
AACUCGUAAGUCCCAUUUGC	Synthego sgRNA	sgRNA_CD46_2
UUGCUCCUUAGAGGAAAUAA	Synthego sgRNA	sgRNA_CD46_3
GGCCAAGCACGAGAACACUC	Synthego sgRNA	sgRNA_BHLHE40_1
CGAGACCACCCGGUGGAGGU	Synthego sgRNA	sgRNA_BHLHE40_2
CAGCUCUCCGGCCAAAGGUU	Synthego sgRNA	sgRNA_BHLHE40_3
GGAGGUCGAUGUGUUACACC	Synthego sgRNA	sgRNA_MEF2C_1
GAGCCAGUGGCAAUAGGUUG	Synthego sgRNA	sgRNA_MEF2C_2
GAGACUGGCAUCUCGAAGUU	Synthego sgRNA	sgRNA_MEF2C_3
UAGUUGUAACUGAGCGAAAA	Synthego sgRNA	sgRNA_NFE2L2_1
GCGACGGAAAGAGUAUGAGC	Synthego sgRNA	sgRNA_NFE2L2_2
AUUUGAUUGACAUACUUUGG	Synthego sgRNA	sgRNA_NFE2L2_3
UGGGCAGAUCCCGGACCCUU	Synthego sgRNA	sgRNA_JDP2_1
CCGAGCUGGGGAGCCCGGUC	Synthego sgRNA	sgRNA_JDP2_2
CAAUCAUGGCCCCGAGGUUG	Synthego sgRNA	sgRNA_JDP2_3
AAGUAAGUUGAUCUCACCUA	Synthego sgRNA	sgRNA_CD274_dss_1
UCCACCAAUUAGAGCACGGC	Synthego sgRNA	sgRNA_CD274_dss_2
ACUCAAUGAACCACGAGAGG	Synthego sgRNA	sgRNA_CD274_dss_3
UGGGUCUCUGGAAUCACGAC	Synthego sgRNA	sgRNA_CD274_ctrl_1
GCGGUGUUGCCCAUGAGAUA	Synthego sgRNA	sgRNA_CD274_ctrl_2
TTCAGTGTAATTCTTCTCTGAGCA	IDT	BHLHE40 Forward Primer
AGGAACGGGGAGCTGATCA	IDT	BHLHE40 Reverse Primer
AGGGAGGACGAGCAGAAGTA	IDT	MEF2C Forward Primer
TGGAGTCTAATGAAATGGGGCT	IDT	MEF2C Reverse Primer
CAGGAGGCTGAGGTTGGAAA	IDT	NFE2L2 Forward Primer
AACCATTTGTGACTTTGCCCT	IDT	NFE2L2 Reverse Primer
TGCCTATGATTGGGTCAAGCA	IDT	JDP2 Forward Primer
GCCCAGTGCTTCTCAAGTGT	IDT	JDP2 Reverse Primer
Software and algorithms
CellRanger (cellranger-arc)	10x Genomics (Pleasanton, CA); functions used: cellranger-arc mkref, cellranger-arc mkfastq, cellranger-arc count	v2.0
R/R Studio	Open Source/RStudio, PBC (Boston, MA)	v4.0/v1.4.1106
Seurat (R package)	Macosko et al., 2015108; Satija et al., 2015109; Stuart et al., 2019110	v4.0
Signac (R package)	Stuart et al.,202134	v1.4.0
Monocle3 (R package)	Qiu et al., 2017130	v0.2.3
Bioconductor (R package)	bioconductor.org; R libraries used include: GenomeInfoDB, ensembldb, EnsDb.Hsapiens.v86, BSgenome.Hsapiens.UCSC.hg38, motifmatcher, JASPAR2020, TFBSTools)	v3.14
macs2/macs3	Liu, 2014113	v2.2.7/v3.0
bedtools	Quinlan and Hall, 2010117; functions used: multiinter
GREAT	McLean et al., 201033; http://great.stanford.edu/public/html/	v4.0.4
hicExplorer	Ramirez et al., 2018118; Wolff et al., 2018119; Wolff et al., 2020120	v3.7.2
IGV	Robinson et al., 2011121	v2.9.2
Juicebox	Durand et al.,2016122	v1.5.1
Cytoscape	Shannon et al., 2003123	v3.8.2
Other
RPMI 1640	Sigma-Aldrich (Raleigh, NC)	Cat. #11875093
FBS, heat inactivated	Corning (Durham, NC)	Cat. #35-010-CV
DMSO	Sigma-Aldrich (Raleigh, NC)	Cat. #D2660-100ML
10x Chromium Controller	10x Genomics (Pleasanton, CA)	Prod. #1000204
NovaSeq 6000	Illumina (San Diego, CA)	Cat. #20012850
BD FACS Canto II	BD Biosciences (Franklin Lakes, NJ)	N/A
Sony SH800 Sorter	Sony Biotechnology	N/A
Neon™ Transfection System	ThermoFisher Scientific	Cat. #MPK5000
Human (hg38) reference genome	Genome Reference Consortium	GRCh38.p13
Type 1 EBV reference genome	de Jesus et al. 2003131; FASTA accessed via NCBI	NC_007605.1
hg38+EBV reference genome	Generated via concatenation of the genomes listed above	N/A

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Ethics statement

The experiments herein were conducted with approval from a Duke University institutional review board (eIRB# Pro00006262). All blood was obtained from de-identified donors via Gulf Coast Regional Blood Center (Houston, TX) and thus no data on donor age or sex was obtained for this study. No protected health information (PHI) or HIPAA identifiers were provided with blood samples. Thus, the experiments reported here were classified as non-human subjects research.

Primary cells

PBMCs and B cells were isolated from de-identified human donor whole blood as described in the method details section.

METHOD DETAILS

B cells and primary EBV infection

B cells were isolated from peripheral blood mononuclear cells (PBMCs) prepared from whole blood (50 mL per donor) as previously described.³⁰ Briefly, total B cells (naive and memory) were enriched via negative isolation. Uninfected samples (Day 0) were cryopreserved in freezing media (10% DMSO +90% FBS), and primary infections with EBV were performed on remaining donor cells at a multiplicity of infection (MOI) of 5 to ensure viral genome delivery to virtually all cells.⁹ Samples from each infection timepoint were the cryopreserved as done for uninfected samples. Prior to sequencing, cells were analyzed for B cell enrichment via flow cytometry using anti-CD19 staining (>90% for all timepoints) and successful EBV infection via anti-CD23 staining as previously described.³⁰

Preparation of scATAC libraries

Cryopreserved samples from each early infection timepoint of interest were simultaneously thawed by donor and purified to >90% viable cells by Ficoll gradient separation. Viable cells from each timepoint and donor were then prepared as single-cell matched gene expression (scRNA) and chromatin accessibility (scATAC) libraries by the Duke Molecular Genomics Core staff with the 10x Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Kit (10x Genomics, Pleasanton, CA).^106,107 Briefly, nuclei were isolated from each sample and subjected to transposition at accessible chromatin sites. Next, transposed nuclei, barcoding master mix, and gel beads containing unique barcode sequences were prepared into single-cell GEMs (Gel bead emulsions) using the Chromium Controller and Chip J. Within each GEM, poly-adenylated (poly-A) mRNA transcripts from individual nuclei were captured by barcoded, indexed poly-T primers and reverse transcribed into cDNA. Simultaneously, a separate barcoded sequence containing a spacer and Illumina P5 adaptor sequence was added to transposed regions within the captured nucleus. Barcoded multiomes were then purified, pooled, and pre-amplified by PCR prior to library construction. The scATAC-seq library for each sample is generated by PCR amplification and incorporation of sample index and Illumina P7 adaptor sequences. The scRNA library construction, sequencing, analysis, and results (have been described previously.³⁰

Sequencing, alignment, and count matrix generation

8 paired-end scATAC-seq libraries (4 timepoints per 2 donors) were pooled onto two lanes of an Illumina S2 flow cell and sequenced at a target depth of 25,000 reads per cell on an Illumina NovaSeq (Illumina, San Diego, CA). Sequencing runs were performed by staff at the Duke Center for Genomic and Computational Biology. Raw base calls for each assay were prepared as sample-demultiplexed FASTQ files using cellranger-arc mkfastq (Cellranger, 10x Genomics), a wrapper of the Illumina bcl2fastq function. Next, sample-matched reads were aligned against genomic references to produce multiome count matrices using cellranger-arc count. One set of count matrices was generated by mapping reads to a concatenated genomic reference constructed from the human genome (GRCh38) with the ~172 kB type 1 EBV genome (NC_007605) included as an extra chromosome. These outputs, as well as a second set of count matrices generated by mapping to GRCh38 only were used for downstream joint RNA and ATAC analyses. Compatible reference packages were assembled from the relevant genome (.fa) and annotation (.gtf) files using cellranger-arc mkref.

Data QC and scMultiome analysis

Direct analysis of scRNA- and scATAC-seq data was conducted in R using Signac,³⁴ an extension of Seurat.^108-110 Following read mapping and counting, scATAC-seq data (linked to scRNA measurements) were obtained from between 8,934 and 20,000 cells per sample. After QC filtering, a total of 52,271 and 44,920 cells were analyzed across the infection timecourse for donors TX1241 and TX1242, respectively. Peaks across all samples were normalized using latent semantic indexing (LSI) as previously described.¹¹¹ This process consists of Term Frequency-Inverse Document Frequency (TF-IDF) normalization to reduce bias from sequencing depth variation and enhance rare peak detection followed by singular value decomposition (SVD) of the normalized matrix.³⁴ TF motif enrichment matrices generated as direct outputs from cellranger-arc count were also analyzed. Pseudotime graphs were calculated for scATAC and motif accessibility by preparing the assays as cell datasets, performing dimensional reduction, and using the learn_graph function in Cicero (for peaks) and Monocle3 (for motifs).^32,112 For multimodal analysis, host chromatin accessibility and gene expression were analyzed for each separate timepoint and for a merged object containing Day 0 and Day 8 multiome data. Nucleosome signal and transcription start site (TSS) enrichment were calculated and used for QC filtering (Nucleosome.signal <2, TSS.enrichment >1). ATAC peaks were called using macs2¹¹³ with hg38 annotations. Gene expression data in the joint analysis was processed as described for the RNA-only analysis apart from using Signac’s SCTransform function instead of log normalization for expression counts. Top differential features in each assay (‘peaks’ and ‘SCT’) were determined, and multimodal clustering via weighted nearest neighbors and UMAP dimensional reduction were performed for integrated data visualization. Cluster identities defined in the RNA-only assay were mapped to the merged joint dataset, which contained cells representing all identified subpopulations. Peaks with significant (anti−) correlation (p < 0.05 for Z score of correlation coefficients) to differentially expressed genes were identified using the LinkPeaks function in Signac, which was informed by SHARE-seq.¹¹⁴ Biological zero-preserving read imputation of scRNA data was also performed using a Seurat wrapper for adaptive low-rank approximation (RunALRA) to correct for technical dropout.¹¹⁵ Gene ontology (GO) biological process enrichment for DAP-linked DEGs from phenotype-specific comparisons were generated using Seurat FindMarkers outputs in conjunction with gene list analysis (enrichGO, dotplot) using the clusterProfiler software package.¹¹⁶

crispATAC workflow and reference data curation

ChIP-referenced inference from single-cell phenotype ATAC (crispATAC) was developed to predict subpopulation-resolved gene regulatory features. In a typical workflow, cluster-level chromatin accessibility tracks are cross-referenced against ChIP-Seq (Chromatin Immunoprecipitation Sequencing) profiles for epigenetic marks and TFs of interest measured from a reference cell phenotype (in this study, lymphoblastoid cell lines such as GM12878) via multiple intersection. In this study, cluster-specific called peaks from the multimodal dataset were extracted and prepared as simplified genomic range files (3-column.bed file format). Next, the desired ChIP-Seq datasets for the reference phenotype were downloaded and, where applicable, converted to.bedgraph format to be used as input for peak calling with the macs2 function bdgpeakcall with default parameters. The ChIP-seq (and Hi-C) datasets used for crispATAC in this study are all publicly available from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO). Once all ATAC-seq and ChIP-seq peak files were generated, all were used as inputs to a single call of the bedtools¹¹⁷ function multiinter with default parameters, which output a matrix of all genomic range intersection intervals where at least one input file exhibited a peak. This intersection matrix was imported to R as a data frame and analyzed to identify common and/or differential intervals (matrix rows) among scATAC-seq cluster phenotypes, epigenetic marks, and TFs using Boolean logic gating by dataset (matrix columns). For a given crispATAC recipe (e.g., peaks in cluster 1 not in cluster 2 intersected with EBNA2 ChIP peaks = [c1 ! c2] ∩ EBNA2), the genome intervals matching the gating criteria were returned and converted to.bed files. Lists of differentially accessible TF-associated sites generated in this way were subsequently analyzed with the Genomic Regions Enrichment of Annotations Tool (GREAT) ³³ to identify single-nearest cis-regulated genes within 1 megabase of each query site. As a final step, output lists of potential linked genes were intersected with the top marker genes identified from the corresponding cluster-wise comparison in the scRNA-seq assay using the Seurat FindMarkers function. This process yielded integration of direct scRNA-seq and ATAC-seq measurements with subpopulation-resolved regulatory inferences from ensemble ChIP profiles. In a similar but separate approach, scATAC and ChIP peaks were intersected with topologically associated domain (TAD) boundaries (prepared using hicExplorer ^118-120) and nuclear subcompartments from GM12878 Hi-C data to study differentially accessible TF-associated sites in the context of 3D nuclear architecture.

Visualization of crispATAC outputs, gene ontologies, and networks

Data for genes of interest identified from crispATAC recipes were explored using dimensionally reduced (UMAP) expression maps and cluster-level accessibility tracks (CoveragePlot function in Signac³⁴), called peaks aligned with TFs and epigenetic marks (IGV,¹²¹), and local neighborhoods in Hi-C contact maps (Juicebox¹²²). Cluster-resolved gene ontologies were generated and quantified by GREAT.³³ Top scRNA assay cluster markers and GREAT output gene lists were also visualized as annotated networks using Cytoscape.¹²³

Cas9/RNP-based functional analysis and flow cytometry

To evaluate the functional role of transcription factors whose motif accessibilities differed between the GC LZ NF-κB activated (JDP2 and NFE2L2) and non-GC, more differentiated (BHLHE40 and MEF2C) state, we generated knockouts by transfecting Cas9-ribonucleoprotein (RNP) complexes targeting these genes of interest. Briefly, to assemble the complexes, TrueCut Cas9 protein-v2 (ThermoFisher, cat. #A36498) was incubated with both sgRNA (Gene Knockout Kit v2, Synthego) targeting the gene of interest and CD46, in protein/sgRNA molar ratio of 1:3. Cells to be transfected were resuspended in Buffer R (Neon System Transfection Kit). The transfection was performed using the ThermoFisher Neon Transfection System (10 μL kit) with the following setup: 1 pulse, 1350 mV for 30 ms. Loss of expression of CD46,⁷⁶ measured by FACS, was used as a proxy to identify KO cells. Additionally, KO LCLs were validated by Sanger sequencing. Primers flanking the regions targeted by Cas9 sgRNAs used to generate BHLHE40, JDP2, MEF2C, and NFE2L2 KO LCL (Integrated DNA Technologies, IDT) were used for PCR amplification of sgRNA target regions. Amplified PCR products were sent for sequencing and analysis (Synthego) to confirm successful lesion induction based on the type and predicted frequency of indels.¹²⁴

LCLs were stained using α-CD46-PE (BioLegend cat. #352402), α-ICAM1-Pacific Blue (BioLegend cat. #322716), and α-CD27-APC (BioLegend cat. #356410) antibodies. CD46⁻/ICAM1^Hi/CD27^Lo cells were sorted using a Sony SH800 Cell Sorter. Post-sort checking and tracking of CD46⁻/ICAM1^Lo/CD27^Hi populations was performed on a BD FACS Canto II analyzer system.

To analyze the role of the putative enhancer located between PDCDL1G2 and CD274, we used the same Cas9-RNP approach described above, to generate the deletion of this site. More specifically, multiple sgRNAs (gCD274_dss) targeting this enhancer were used and as a control, a 7kb-apart from the queried peak was targeted with sgRNAs (gCD724_ctrl). CD274 expression was analyzed in CD46 negative cells. The cells were stained using α-CD46-APC (BioLegend cat. #352405) and α-CD274-PE (BioLegend cat. #329706).

Western blot analysis

Based on reduced predicted indel frequency for NFE2L2 (NRF2) replicate LCLs, protein expression levels of NRF2 were evaluated by Western blot. Cell lysates were prepared from WT and NFE2L2 KO LCL replicates with and without hydrogen peroxide treatment (400 mM H₂O₂ for 3 h @ 37°C) to stabilize NRF2 protein, which is normally constitutively targeted for degradation in the absence of oxidative stress.¹²⁵ Lysates were sonicated, total protein was quantified by Bradford assay, and a total of 30 μg protein per sample with added 2-mercaptoethanol and loading buffer were run on 4–12% Bis-Tris gels (Invitrogen NuPage 10-well 1.0–1.5mm gel, ThermoFisher cat. #NP0322BOX). Electrophoresis was performed using 1x MES buffer at a running voltage of 120 V for 90 min, using 10–250 kDa PageRuler protein ladder (Thermo Scientific, cat. #26619) as a size standard. Electrophoresed proteins were transferred to PVDF membranes using a TransBlot Turbo Transfer System at 25 V, 1.3 A, for 12 min (Bio-Rad). After transfer, blots were imaged using a Li-COR Odyssey imager (Li-COR Biosciences, Lincoln, NE) after incubation with total protein stain (Li-COR Revert 700 Total Protein Stain) and washing (1x TBST, 3 washes for 5 min each). After revert imaging, blots were blocked with 5% milk in 1x TBST for 1 h at room temperature, washed with TBST, and incubated overnight at 4°C with 1:2000 primary antibody (rabbit anti-human NRF2, Invitrogen, Thermo Fisher cat. #PA5-27882) in 5% BSA with sodium azide. Primary antibody-incubated blots were washed, incubated with 1:3500 goat anti-rabbit IgG HRP-conjugated secondary antibody (Sigma Aldrich, prod. #12–348) in 1x TBST with 5% milk for 1 h at room temperature, washed, incubated with HRP substrate (Thermo Scientific SuperSignal West Femto, cat. #34096), and imaged for NRF2 and protein ladder detection using a Li-COR Odyssey imager (600 and chemiluminescent channels).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis

Differentially accessible peaks (DAPs) between scATAC-seq cell clusters were identified using the FindMarkers() function in Seurat/Signac, which calculates raw and Bonferroni-adjusted p values from logistic regression differential accessibility testing of peak intensities. Outputs from this analysis for comparisons of interest are provided in supplementary files. DAPs significantly correlated (p < 0.05 for Z score of correlation coefficients) with differentially expressed genes (DEGs) were identified using the LinkPeaks() function in Signac, which was informed by SHARE-seq.¹¹⁴ Significance of differential gene expression between cell clusters was evaluated by Bonferroni-corrected Wilcoxon rank-sum tests and Kolmogorov-Smirnov (KS) tests. Gene ontology (GO) biological process enrichment for DAP-linked DEGs from phenotype-specific comparisons were generated using Seurat FindMarkers() outputs in conjunction with gene list analysis (enrichGO, dotplot) using the clusterProfiler software package.¹¹⁶ For FACS studies of knockout LCLs, statistically significant differences in gated population frequencies and mean fluorescence intensity (MFI) were evaluated via Two- and One-way ANOVA followed by post-hoc Dunnett method, respectively.

Highlights.

• scATAC-seq of B cell and viral chromatin accessibility in primary EBV infection

• Growth-arrested EBV⁺ cells develop heterochromatin at key B cell activation loci

• Successful infection actuates GC-like B cell chromatin dynamics without BCL6

• Gene and regulatory locus multiomic predictions validated in knockout studies