Reconstructing EBV reactivation and DNA damage response kinetics in morphologic pseudotime

Significance

Single-cell assays are essential for studying complex and asynchronous host–virus interactions. We developed single-cell high-content screening (HCS) in morphologic pseudotime to capture the kinetics of Epstein–Barr virus (EBV) lytic reactivation and virus-required host DNA damage response (DDR) factors without live-cell imaging. Screening identified distinct viral and host responses to lytic-inducing treatments and clinically relevant genotoxins. Surprisingly, two lytic-essential DSB factors (γH2AX, 53BP1) became depleted from EBV replication compartments. Along with γH2AX persistence in host chromatin throughout reactivation, these findings support lytic-mediated spatiotemporal DDR dysregulation. Methodologically, we demonstrate HCS and morphologic pseudotime generalizability for future single-cell studies of host–virus dynamics.

Abstract

Epstein–Barr virus (EBV) lytic infection contributes to oncogenesis and autoimmunity and depends on subversion of host DNA damage responses (DDR). We used high-content screening (HCS) to systematically capture single-cell morphologic profiles and pseudotemporal dynamics of EBV reactivation and DDR across common B cell models and lytic induction treatments. We generated an atlas (>850,000 cells) of spatiotemporally distinct phenotypes of immediate-early and late lytic proteins, viral and cellular DNA replication, and double-stranded break (DSB) DDR factors. Cell segmentation, feature quantification, and clustering identified treatment- and model-dependent cell responses and lytic induction. Lytic and latent cells showed distinct genotoxin-induced DDR profiles, and lytic protein localization varied by pharmacologic and physiologic stimuli. Pseudotime trajectories revealed viral replication compartment (VRC) nucleation and expansion alongside concomitant DDR localization. The early DDR marker γH2AX was depleted from VRCs but widespread across host chromatin throughout reactivation. Surprisingly, the lytic-essential late DDR protein 53BP1 was present prior to viral genome replication but subsequently undetected in VRCs and host chromatin, indicating spatial and kinetic DDR dysregulation during EBV reactivation. These data support a model wherein EBV transiently employs host DSB DDR mediators to initiate genome replication while host-targeted DDR is initiated but impaired. We further show biological generalizability and utility of our method across microscope systems. HCS paired with morphologic pseudotime analysis thus provides a powerful approach to recover single-cell host–virus dynamics from snapshot samples.

Epstein–Barr virus (EBV) is an important human pathogen due to its ubiquity (1) and roles in cancers (2) and autoimmunity (3). This gammaherpesvirus infects human B lymphocytes (1), establishes latency in memory B cells (4), and undergoes periodic reactivation (5). Latent and lytic EBV infection each depend on coordinated viral programs that modulate host cell biology, evade immune control, and contribute to disease through distinct mechanisms (6). Single-cell sequencing underscores heterogeneity related to EBV infection phase and strain (7–10), viral gene functions (11), and clinical immune responses (12, 13). Such studies provide genomic perspectives of infection, however complementary single-cell methods are greatly needed to study other molecular aspects, rare phenotypes, and infected cell dynamics at scale.

EBV lytic infection is challenging to study due to its asynchrony and relative rarity in B cells (14). Thus, we developed a high-content screening (HCS) approach to study EBV reactivation across B cell models. Lytic infection is activated by diverse stimuli via distinct mechanisms. B-to-plasma cell differentiation induces reactivation through host transcription factor activation of immediate-early (IE) promoters and other mechanisms (15–17). B cell receptor activation (18) and TGF-β-induced SMAD signaling (19–21) are additional physiologic lytic triggers. Laboratory treatments for lytic induction include phorbol ester (22, 23), butyrate (24), and reactive oxygen species (25). B cell chemotherapeutics induce lytic expression in some models (26, 27) via poorly understood mechanisms. These stimuli activate essential IE proteins, Zta (Z; encoded by BZLF1) and Rta (R; encoded by BRLF1), which activate lytic expression from intranuclear viral episomes (28). Zta binds Z response elements (ZREs) (29) in viral promoters including the origin of lytic replication (oriLyt), early genes essential for viral DNA replication (30), and BZLF1 itself (31). Intriguingly, Zta is an AP-1 family pioneer factor that derepresses viral and host genes via nucleosome eviction (32–35). Rta indirectly activates BZLF1, additional early-kinetic lytic genes (36), and itself (37). While IE and early expression precedes viral DNA amplification, expression of late-kinetic genes encoding tegument, capsid, and envelope glycoproteins depends on viral genome replication (5).

EBV lytic infection depends on host DNA damage response (DDR) pathways (38). Double-stranded break (DSB) repair mediators including phosphorylated ataxia telangiectasia mutated (ATM) kinase (39), phospho-Ser139 in histone H2A.X (γH2AX) (32), homologous recombination (HR) repair complex MRE11-RAD51-NBS1 (MRN) (40), and nonhomologous end joining (NHEJ) mediator 53BP1 (41) are important for EBV reactivation. Prior studies (40, 42, 43) identified these repair factors within intranuclear viral replication compartments (VRCs), which exhibit lytic stage-dependent morphologies and effects on nuclear organization (44–47). Imaging studies of DDR proteins during lytic infection span model systems including nonphysiologic cell types and, in some cases, yield seemingly inconsistent results for DDR protein localization. For example, vector-based expression of Zta (48), BKRF4 (49), or the polymerase processivity factor EA-D (BMRF1) (50) impair 53BP1 focus formation in epithelial lines, yielding diffuse nuclear patterns. By contrast, 53BP1 foci colocalized with BMRF1 in apparent VRCs in Zta-transfected 293 cells harboring EBV genome bacterial artificial chromosomes (BAC) (43). These disparate findings may indicate complex roles in viral reactivation. Like 53BP1, the early DDR biomarker γH2AX plays an important role in lytic infection (32, 51, 52). Direct γH2AX visualization in cells expressing EBV lytic genes has been limited to epithelial Zta overexpression systems (42, 48). Critically, the presence and nuclear localization of DNA damage mediators may depend on model differences, low sample throughput, and missing temporal information accounting for lytic and DDR dynamics. Paired with inherent challenges of heterogeneous infection, these factors underscore the need for systematic quantitative single-cell methods.

We developed high-content screening (HCS) with morphologic pseudotime analyses based on >800 automatically quantified features to address this need. We leverage infected cell population asynchrony to reconstruct single-cell host–virus dynamics from snapshot samples. We resolved treatment-dependent lytic expression and DDR profiles as well as previously unrecognized DDR kinetics indicating spatiotemporal dysregulation during reactivation. We further demonstrate the versatility of HCS and pseudotime analyses to study infected cell biology as well as the method’s technical compatibility across confocal and epifluorescence microscopes. Accordingly, we expect this approach will support high-throughput single-cell virology and be extensible to other intracellular pathogens and cellular dynamics for which live-cell imaging is challenging or technically intractable.

Results

Systematic Single-Cell Analysis across B Cell Lytic Infection Models.

Single-cell responses to EBV lytic induction and progression remain incompletely understood. We developed plate-based methods to systematically investigate B cell model responses to biologic stimuli and genotoxins with HCS (Fig. 1A). Following treatment, cells were subjected to live-cell labeling [typically, a DNA synthesis pulse with 5-ethynyl-2’-deoxyuridine (EdU)]. Labeled cells were washed, transferred to functionalized imaging plates, centrifuged to enhance cell deposition, fixed, permeabilized, stained, and imaged using automated confocal or epifluorescence systems. Images were processed for high-throughput single-cell analyses as described below and in Experimental Methods (SI Appendix).

Fig. 1.

High-content screening assay and study design. (A) Experimental workflow overview. B cell lines were treated with biologic and pharmacologic inducers of EBV lytic infection and assayed for host and viral processes and proteins. Live cell labeling (e.g., pulsed EdU incorporation, mitochondria-selective dyes) was performed prior to harvest, transfer, and fixation on coated 96-well imaging plates. For experiments examining DNA synthesis, EdU was detected via copper-catalyzed click chemistry. Host and viral proteins were detected via immunofluorescence, nuclei were stained, and plates were imaged on automated confocal or epifluorescence microscopes. Cells were automatically segmented, measured, and analyzed. Panel created with BioRender under license. (B) Representative CellProfiler segmentation of nuclei, cytoplasm, and whole cells. (C) Detail of fluorescence signals used for segmentation (region denoted by dashed outline in B). Composite signal includes nuclei (blue), gp350 (cyan), EdU (yellow), and γH2AX (red). (D) Quantitative single-cell morphologic features for clustering and stratified analyses. Spearman correlation for the top 400 variable morphologic features for a representative HCS experiment (Left). UMAP dimensional reduction and clustering of cells by variable morphology (Middle). Experimental factors and metadata used for stratified phenotypic analyses (Right).

B cell models used here (SI Appendix, Table S1) included EBV-negative lines (Akata-, BJAB); EBV-positive Burkitt Lymphoma (BL) lines with lytic defects (Namalwa, Raji); lytic-competent EBV-positive BL lines (Akata-GFP, Daudi, Mutu, Jijoye, and P3HR1-ZHT, which was engineered from a subclone of Jijoye for hydroxytamoxifen-inducible lytic expression); and EBV-positive immunoblastic non-Hodgkin lymphomas (NHL; Farage, IBL1). BJAB and Akata- cells provided EBV-negative immunofluorescence controls. Namalwa cells provided an EBV-positive cell line incapable of reactivation as a negative control. Raji cells can express Zta but fail to progress to late lytic infection due to inactive BALF2, which encodes a lytic-essential single-stranded DNA binding protein (47). P3HR1-ZHT cells treated with 4-hydroxytamoxifen (4HT) served as a positive control for lytic infection and a robust experimental condition with high reactivation frequency. (SI Appendix, Fig. S1). Additional controls confirmed signal specificity and negligible interchannel crosstalk (SI Appendix, Fig. S2). Screens used seven different lytic-inducing treatments encompassing physiologic (TGF-β, anti-IgG), experimental [phorbol ester (TPA) + sodium butyrate (NaB), H₂O₂], and clinical (doxorubicin, etoposide, panobinostat) stimuli plus unstimulated controls.

Multichannel images were analyzed with CellProfiler (53) for segmentation and feature extraction from regions of interest (ROIs): nuclei, cytoplasm, and whole cells (Fig. 1 B and C). Nuclei were segmented by Hoechst staining, whole-cell masks were generated from channel-composite images, and cytoplasmic regions were derived by subtracting nuclei from whole-cell masks. Cells were size-filtered to minimize artifacts and per-cell feature extraction was performed to quantify >800 measurements including intensities, shape, texture, and colocalization. Measurements were annotated with metadata (cell line, treatment, timepoint, well) and standardized preprocessing was performed with PyCytominer (mad_robustize) (54). We performed morphologic feature variance analysis and calculated interfeature Spearman correlation (Fig. 1 D, Left). Top features among top principal components included measurements of nuclear size and texture (compactness, texture-entropy), colocalization of nuclei and DNA damage immunofluorescence [e.g., rank-weighted correlation (RWC)], and lytic protein immunofluorescence (e.g., mean and integrated intensity, texture entropy). High-variance features were used for UMAP dimensionality reduction, and unsupervised Leiden clustering defined morphologic phenotypes. Datasets were analyzed by adapting Seurat v5 (55) to support stratified quantitation (Fig. 1 D, Right). Over 850,000 cells were segmented and quantified across ~12,000 fields of view from confocal imaging experiments (SI Appendix, Figs. S3 and S4 and Table S2 and Dataset S1). Two screens were conducted to investigate lytic infection and corresponding DDR via early and late biomarkers (DNA_EdU_Zta_γH2AX and DNA_EdU_gp350_53BP1, respectively) at a single timepoint (36 h). Two timecourse experiments examined lytic responses and DDR at 24, 48, and 72 h using swapped immunofluorescence combinations to examine late lytic expression alongside an early DDR indicator (DNA_EdU_gp350_γH2AX) and vice versa (DNA_EdU_Zta_53BP1). We also demonstrated the method’s biological generalizability by examining mitochondrial structure and localization during lytic infection stages assayed via Zta (DNA_Mito_Zta) or the late glycoprotein gp350 (DNA_Mito_gp350). Finally, we demonstrated technical generalizability of the HCS approach for widefield epifluorescence systems (DNA_Mito_Ki67_Zta_Viability). Collectively, this technique enabled systematic quantification of response state heterogeneity and spatiotemporal insights into key host and viral processes in reactivating cells.

Differential Lytic Response Phenotypes in B Cell Lines Resolved by Induction Treatment.

EBV predominantly adopts latency in B cells, and many lytic-inducing treatments are inefficient. HCS addresses this by resolving phenotypes at high-throughput. As an initial demonstration, we quantified and characterized cells stained for nuclei, pulsed DNA synthesis with EdU, Zta, and γH2AX across B cell lines and inductions (DNA_EdU_Zta_γH2AX). Twenty-one cell phenotypes were identified following UMAP reduction (56) and Leiden clustering (57) (Fig. 2A). Morphologic feature enrichment was calculated for each cluster (Dataset S2) and identified the highest Zta in cluster 8 (Fig. 2B). This and other clusters with Zta expression above background (7, 18, 19) were almost exclusively composed of EBV-positive lines competent for lytic initiation (Fig. 2C). Additional HCS comparison of isogenic EBV+ and EBV- Akata cells confirmed virus-specific clustering of Zta-positive cells independent of host background (SI Appendix, Fig. S4 A–E).

Fig. 2.

High-throughput quantification of lytic induction and treatment-stratified phenotypes in B cell models. (A) UMAP and clustering of segmented cells stained for nuclei, Zta, EdU, and γH2AX. (B) Cluster-resolved quantification of whole-cell Zta immunofluorescence from experiment in (A). (C) Normalized frequency of model cell lines within clusters depicted in (A and B). (D) Heterogeneity within P3HR1-ZHT cells treated with 4HT to induce lytic infection. One outlined cell (i) exhibiting a nucleus (blue) with low Zta (green) within intranuclear compartments lacking de novo DNA synthesis (yellow) corresponds to cluster 1. Another cell (ii) exhibiting high Zta overlapping low DNA synthesis within intranuclear compartments corresponds to cluster 8. A third cell (iii) exhibiting high-intensity nucleus-wide DNA synthesis in the absence of VRCs or Zta corresponds to cluster 3. (E) Single-cell images from selected conditions with differential Zta expression, VRCs, and DNA synthesis. (F) Cluster-resolved correlation of Zta intensity with nuclear shape and texture. Individual points represent cells colored by cluster as in (A and B). (G) Cluster-and treatment-resolved correlation between Zta texture entropy and nuclear texture entropy. Cells are colored by treatment and split onto axes by cluster.

Individual cell images were recovered from >350 GB of raw data using metadata (well, sample IDs, file names, bounding box coordinates; SI Appendix, Fig. S5). This enabled linkage of cluster quantitative morphology and visual phenotypes to explore response heterogeneity. For example, 4HT-treated P3HR1-ZHT cells exhibited Zta-positive cells in distinct lytic stages and mostly Zta-negative cells interpreted as refractory to reactivation corresponding to different phenotype clusters (Fig. 2D). Cells expressing Zta within intranuclear compartments in the absence of viral DNA synthesis were consistent with replication-independent reorganization of cellular chromatin phenotype I (ROCC I) previously identified in hybrid epithelial/B cell lines (47) (Fig. 2D cell i). Cells with highly colocalized Zta and EdU signal in VRCs corresponded to kinetically later reactivation characterized by active viral replication and progressive host chromatin laminar margination (ROCC II) (47) (Fig. 2D cell ii). EdU intensity and spatial distribution in such lytic cells were distinct from Zta-negative EdU-positive cells, which corresponded to S-phase based on elevated EdU overlapping cellular DNA (Fig. 2D cell iii). ROCC I, ROCC II, and intranuclear viral DNA synthesis were observed within clusters of lytic EBV+ Akata but not EBV- Akata cells (SI Appendix, Fig. S4 E and F).

We examined effects of common lytic induction treatments on replication phenotypes of Zta-positive cells across model lines (Fig. 2E). Zta-positive Raji cells were, with rare exceptions, negative for colocalized EdU signal within intranuclear compartments, consistent with genetically defective viral genome replication (Fig. 2E, first row). Zta-positive Mutu cells with active EBV replication and ROCC were identified in TGF-β treatment and rarely in unstimulated conditions; however, Zta-positive cells induced with the protein kinase C (PKC) activator phorbol 12-myristate 13-acetate (PMA, a.k.a. TPA) and histone deacetylase inhibitor sodium butyrate (NaB) typically lacked EdU signal used to identify viral replication (Fig. 2E second row). Differential responses to TGF-β vs. TPA+NaB treatment were also observed in P3HR1-ZHT (Fig. 2E third row). While some Zta-positive P3HR1-ZHT cells induced by TPA+NaB were EdU-positive, minimal overlap between these signals indicated cellular rather than viral DNA synthesis. Zta-positive Farage cells exhibited a similar response to TPA+NaB, whereas TGF-β did not induce lytic expression (Fig. 2E, fourth row). Whole-cell EdU intensity and intensity distribution analysis revealed TPA+NaB significantly depleted DNA synthesis vs. TGF-β and unstimulated conditions independent of Zta (SI Appendix, Fig. S6A). This effect resembled that of genotoxins with current or prospective antineoplastic applications (doxorubicin, etoposide, and panobinostat), some of which induce expression of EBV lytic genes (27, 58). Notably, reduced DNA synthesis in TPA+NaB-treated cells persisted over time even after washout at 2 h (SI Appendix, Fig. S6B). This effect is consistent with prior studies (59) and highlights a consideration for using TPA+NaB to study EBV replication vs. lytic gene expression per se, as viral DNA replication is essential for early-to-late lytic progression (47).

Feature enrichment in Zta-positive cells (cluster 8) identified proxy metrics for lytic ROCC morphology. Lytic nuclei were comparatively large (low compactness) and porous (high texture entropy) and had specific enriched Zernike features (mathematical descriptors of shape and pattern) (Fig. 2F). Reasoning that cells with VRCs would exhibit entropic nuclei (reorganized chromatin) and entropic Zta (compartmentalized enrichment), we examined these features stratified by cluster and treatment. Overall, Zta-positive cells in cluster 8 exhibited higher nuclear and Zta texture entropy (signal nonuniformity or randomness) in response to TGF-β treatment (and 4HT for P3HR1-ZHT) vs. TPA+NaB, doxorubicin, etoposide, and panobinostat (Fig. 2G). Thus, quantitative morphology delineates qualitatively different cell responses to lytic-inducing stimuli.

Another experiment (DNA_EdU_gp350_53BP1) was performed to assay later reactivation stages across B cell models and treatments. We identified morphologic phenotypes (SI Appendix, Fig. S7 A, Left) including late lytic cells in clusters 17 (high gp350) and 7 (low gp350) (SI Appendix, Fig. S7 A, Right). Cluster cell line composition indicated gp350-positive cells were derived from lytic-competent models (SI Appendix, Fig. S7 B, Left). While low-expressing gp350-positive cells were produced by several treatments, high-expressing gp350-positive cells were largely derived from 4HT-stimulated P3HR1-ZHT and a fraction of TGF-β-treated cells (SI Appendix, Fig. S7 B, Right). Representative gp350-positive P3HR1-ZHT cells confirmed major ROCC in late lytic stages for multiple treatments (SI Appendix, Fig. S7C). While some gp350-positive cells were generated by TPA+NaB, these cells generally exhibited reduced de novo DNA incorporation and limited ROCC. 100× magnification imaging of EBV+ Akata cells stimulated with anti-IgG confirmed perinuclear gp350 accumulation in cells exhibiting ROCC II. This gp350 accumulation correlated with anisotropic cytoplasmic enlargement consistent with virion assembly compartments (SI Appendix, Fig. S8). Consequently, late lytic cell clustering was driven by elevated gp350 intensity and cytoplasmic area features (Dataset S2). Single-cell HCS survey experiments thus revealed line- and treatment-dependent lytic induction and quantitative reactivation morphology metrics.

Distinct DNA Damage Responses in EBV-Positive B Cells Resolved by Treatment and Lytic Expression.

Given the importance of double-stranded break (DSB) repair to dsDNA virus reactivation (38), we investigated the kinetics and localization of lytic antigens alongside DSB biomarkers. Whole-cell γH2AX and Zta measurements highlighted line and treatment-dependent lytic induction and early-kinetic DDR heterogeneity (Fig. 3A). TGF-β and TPA+NaB yielded more Zta-positive cells than doxorubicin, etoposide, or panobinostat, which predominantly induced DDR without lytic protein expression. TPA+NaB induced higher intensity γH2AX than TGF-β in Zta-positive and Zta-negative cells, indicating treatment-associated damage independent of lytic programs. Some P3HR1-ZHT cells treated with etoposide or panobinostat exhibited modest Zta, however this may reflect stabilized expression of the recombinant Zta fusion with murine estrogen receptor hormone-binding domain vs. endogenous Zta. P3HR1-ZHT treated with 4HT provided a positive control for cells copositive for Zta and DDR markers, and reduced γH2AX upon cotreatment with an ataxia-telangiectasia mutated (ATM) inhibitor (KU-55933) confirmed biologic specificity of γH2AX (60). Reduced Zta intensity in P3HR1-ZHT cells cotreated with 4HT and phosphonoacetic acid (PAA) was unexpected since PAA interferes with viral DNA replication as well as expression of leaky and true late genes; however, this may be attributable to low cell recovery from the PAA+4HT condition (n = 177 cells).

Fig. 3.

Genotoxin-induced reactivation and DDR differ quantitatively and qualitatively in B cell models. (A) Scatterplot of line- and treatment-stratified mean cell γH2AX intensity vs. mean cell Zta intensity. (B) Detail of Zta-high cluster replication and DNA damage phenotypes by treatment and line. Merged and single channel images (nuclei in blue, Zta in green, EdU in yellow, γH2AX in red, brightfield in gray) are presented for example phenotypes. (C) RWC between γH2AX and Zta intensities stratified by treatment in lytic cells (Top; cluster 8 in DNA_EdU_Zta_γH2AX) and all cells (Bottom). (D) Heterogeneous γH2AX patterns induced by genotoxins in Zta-negative and Zta-positive cells. Examples are shown for Mutu treated with etoposide (Top two rows), Daudi treated with doxorubicin (Middle two rows), and P3HR1-ZHT treated with panobinostat (Bottom two rows). (E) Scatterplot of correlation between integrated Zta and γH2AX in single cells from DNA_EdU_Zta_γH2AX (color represents treatment). Pearson’s R = 0.38. (F) Scatterplot of correlation between integrated gp350 and 53BP1 in cells from DNA_EdU_gp350_53BP1 (color represents treatment). Pearson’s R = 0.01. (G) Example of 53BP1 staining in gp350-negative (Top row) and gp350-positive (Bottom row) Mutu cells stimulated with TPA+NaB.

Among Zta-positive cells, genotoxin-induced Zta expression and texture entropy were significantly reduced (SI Appendix, Fig. S9). Thus, we explored whether lytic cell DDR phenotypes were likewise treatment-dependent (Fig. 3B). In 4HT-stimulated P3HR1-ZHT cells exhibiting VRCs and ROCC, typical γH2AX staining overlapped host chromatin. This spatially inverse Zta and γH2AX pattern was observed to a lesser degree in TGF-β-treated Raji cells. By contrast, rare Zta-positive cells from genotoxin-treated lines characteristically exhibited pan-nuclear γH2AX and diffuse Zta without obvious VRCs. Quantitatively, this yielded modestly higher RWC (an intensity-dependent measurement of signal overlap) between γH2AX and Zta in cluster 8 (lytic) cells (Fig. 3 C, Top). This RWC trend was reversed across aggregated phenotypes for doxorubicin and etoposide but not panobinostat, likely due to frequent Zta-negative cells with extensive DNA damage in these conditions (Fig. 3 C, Bottom and SI Appendix, Fig. S10). We also investigated whether Zta-positive and Zta-negative cells within the same treated line exhibited distinct γH2AX profiles (Fig. 3D and SI Appendix, Figs. S11 and S12). Pan-nuclear γH2AX accumulation indicating widespread DNA damage (a preapoptotic state in some contexts) (61, 62) was the most common genotoxin-induced DDR phenotype in Zta-positive cells, though other patterns were infrequently observed (SI Appendix, Fig. S11 A–C). Zta-negative cells in equivalent treatments exhibited more diverse DDR phenotypes: γH2AX foci (localized DNA repair), pan-nuclear presence, and annular staining indicating early-stage apoptosis (63). The annular γH2AX phenotype was nearly exclusive to Zta-negative cells, consistent with lytic-induced apoptosis resistance (64, 65). Panobinostat was the most efficient Zta inducer among tested chemotherapeutic agents, particularly in P3HR1-ZHT and Farage. Panobinostat-treated Zta-positive cells generally exhibited pan-nuclear γH2AX staining without EdU incorporation or typical ROCC (SI Appendix, Fig. S11 B and C). We identified linear γH2AX staining in several panobinostat-treated cell lines (particularly IBL1 and Farage) that appeared to correspond to long DNA strand domains (SI Appendix, Fig. S11 C–E). In some cells, linear rays of γH2AX signal appeared to emanate from a shared focus. This may reflect HDACi-mediated nucleosome eviction, possibly in the absence of DSBs (66), and megabase-scale γH2AX spreading (67) due to dysregulated DNA organization. We speculate this linear pattern may indicate early-stage initiation of pan-nuclear γH2AX. Statistically significant morphologic correlation between genotoxin-induced Zta and pan-nuclear γH2AX was manually confirmed for doxorubicin and etoposide as a group (chi-squared P = 1.83e-4, n = 34 cells; SI Appendix, Fig. S12A) and panobinostat (chi-squared P = 1.25e-9, n = 195 cells; SI Appendix, Fig. S12B).

Given the correlation between IE lytic biomarkers and DSB DDR (Fig. 3E), we examined whether later kinetic biomarkers were similarly correlated. However, the late glycoprotein gp350 and the late DDR protein 53BP1 were not correlated (Fig. 3F). Upon closer examination, we found 53BP1 only in gp350-negative cells (Fig. 3G), which prompted further investigation of coordinated lytic DDR kinetics.

Depletion of Cellular DNA Damage Biomarkers γH2AX and 53BP1 in EBV VRCs.

We investigated coexpression of IE lytic with late DDR markers (DNA_EdU_Zta_53BP1) and late lytic expression with early DDR (DNA_EdU_gp350_γH2AX). We screened four cell lines (Namalwa, Mutu, P3HR1-ZHT, and IBL1), four conditions (Unstim, TGF-β, TPA+NaB, and H₂O₂), and three timepoints (24, 48, 72 h) in duplicate (n = 2 wells per combination) per panel. This time-resolved design confirmed TPA+NaB treatment delayed late lytic expression associated with DNA synthesis inhibition (SI Appendix, Figs. S13 and S14). A similar effect was observed for H₂O₂-induced lytic expression (SI Appendix, Figs. S13 and S14). TGF-β treatment yielded lytic induction with minimal impact on EdU incorporation, more rapid accumulation of late lytic cells, and comparatively less DNA damage in latent cells (SI Appendix, Figs. S13–S17). TPA+NaB induced more widespread Zta expression across phenotypically diverse cells, a trade-off for physiological responses elicited by treatments including TGF-β (SI Appendix, Fig. S15A).

53BP1 intensity was depleted in many Zta-positive cells (Fig. 4A and SI Appendix, Fig. S15A; cluster 9 in DNA_EdU_Zta_53BP1). Conversely, γH2AX positively correlated with gp350 (Fig. 4B and SI Appendix, Fig. S15B; cluster 10 in DNA_EdU_gp350_γH2AX). These data indicated early DDR activity in early and late lytic cells while at least one mediator of late DSB repair (53BP1) is diminished during lytic infection. While γH2AX associates with the EBV lytic origin of replication (oriLyt) upon reactivation (52), γH2AX was rapidly excluded from VRCs but remained intense across cellular chromatin (Fig. 4C). This distribution was consistent with histone depletion from VRCs and packaging of epigenetically naïve EBV genomes into virions (68, 69). 53BP1 was likewise depleted in VRCs with active viral replication, whereas Zta-positive cells lacking viral DNA synthesis retained 53BP1 signal coinciding with Zta-positive compartments (Fig. 4 D and E). Mutually exclusive 53BP1 and EdU in Zta-positive cells was evident from treatments with differential effects on DNA synthesis and intrasample phenotypic heterogeneity (Fig. 4 D and E and SI Appendix, Figs. S16A and S17A). These results support 53BP1 recruitment to early-stage VRCs and depletion as EBV replication progresses.

Fig. 4.

DDR biomarkers are depleted within EBV VRCs. (A) Codetection of IE lytic (Zta) and late DDR (53BP1) biomarkers over 72 h. UMAP and clustering of segmented cells stained for nuclei, Zta, EdU, and 53BP1 (Left). Scatterplot of integrated Zta vs. 53BP1 intensity by cluster, Pearson’s R = 0.05 (Right). (B) Codetection of late lytic (gp350) and early DDR (γH2AX) biomarkers over 72 h. UMAP and phenotypic clustering of segmented cells stained for nuclei, gp350, EdU, and γH2AX (Left). Scatterplot of integrated gp350 vs. γH2AX intensity by cluster, Pearson’s R = 0.17 (Right). (C) Detail of γH2AX exclusion from VRCs in a P3HR1-ZHT cell stimulated with 4HT (nuclei in blue, Zta in green, EdU in yellow, γH2AX in red). (D) Treatment-variable 53BP1 in Zta-positive Mutu and P3HR1-ZHT cells. Absence of 53BP1 detection in H₂O₂-reactivated Mutu cell with pan-nuclear Zta expression (Left panel; individual channels in Lower Right). Presence of 53BP1 in Zta-positive EdU-negative cells induced with TPA+NaB (Middle). Absence of 53BP1 in Zta-positive EdU-positive cells exhibiting VRCs induced by TGF-β (Right). Nuclei depicted in blue, Zta in green, EdU in yellow, and 53BP1 in magenta. (E) 53BP1 colocalization with Zta in VRCs and absence associated with onset of viral genome replication. (F) Representative examples host genome-localized γH2AX in late lytic (gp350-positive) Mutu (panels i and ii) and P3HR1-ZHT (panels iii, iv, and v) cells reactivated via TGF-β treatment. Fluorescence intensity plot profiles for each channel correspond to the line depicted in panel v. (G) Absence of 53BP1 signal in late lytic (gp350-positive) P3HR1-ZHT cell without induction (unstimulated). Full-field 60× magnification image (panel i); individual gp350-negative cells with diffuse 53BP1 staining (panel ii), replication-associated 53BP1 foci (panels iii and iv), and damage-associated foci in the absence of DNA replication (panel v) from dashed red boxes in panel i; detail of region including gp350-positive cell (panel vi); depletion of 53BP1 in gp350-positive cells with late VRC(panel vii, detail of the dashed green box in panel vi). (H) Absence of 53BP1 signal in late lytic (gp350-positive) P3HR1-ZHT cell treated with TGF-β. Full-field 60× magnification (panel i); detail of the region in the dashed white box in panel i including late lytic cells and nonlytic cells (panel ii); detail of late lytic (gp-350-positive) cell from dashed green box in panel ii (panel iii); individual channels for gp350-positive cell depicted in panel iii (panel iv).

We observed intense γH2AX accumulation within cellular chromatin but depleted in VRCs for gp350+ cells (Fig. 4F and SI Appendix, Fig. S17B). This DDR pattern was consistent across cells with varying VRC and ROCC morphologies. Interestingly, γH2AX was enriched at interfaces between VRCs and adjacent cellular chromatin and attenuated at increasing distance into cellular chromatin. Despite host-targeted γH2AX, gp350-positive cells exhibited negligible 53BP1 signal (Fig. 4 G and H). Manual quantification of 53BP1 confirmed significant 53BP1 depletion (punctate and diffuse patterns) in TGF-β-treated P3HR1-ZHT (chi-squared P = 5.88e-5, n = 107 cells; SI Appendix, Fig. S18A) and Mutu (chi-squared P = 1.54e-8, n = 125 cells; SI Appendix, Fig. S18B). These data collectively indicate 53BP1 interactions with EBV lytic proteins, particularly Zta (41), are physically partitioned from virus-mediated initiation of an incomplete host-directed DDR evidenced by widespread γH2AX. Moreover, 53BP1 depletion from compartments in Zta- and EdU-copositive cells as well as gp350-positive cells implies its importance during IE and early lytic stages rather than late reactivation.

Pseudotime Trajectories Recover Lytic Kinetics and Support Zta as an Essential Pioneer Factor.

The cellular environment during EBV lytic reactivation has been described as pseudo-S-phase (39, 70); thus, we aimed to understand lytic progression by cell cycle stage in greater detail. We primarily used DNA synthesis pulse labeling to identify EBV VRCs. Along with nuclear staining, EdU labeling of host DNA replication also facilitated cell cycle staging in asynchronous populations with high-resolution S-phase replication timing (Fig. 5A). Cells in early S-phase displayed low EdU intensity limited to internal (presumably, euchromatin-rich) regions. Cells in peak S-phase exhibited intense pan-nuclear EdU. Cells in late S-phase exhibited larger nuclei and limited EdU localized to heterochromatin-rich topological boundaries (nuclear laminae, nucleoli). DNA and EdU intensities in Zta-positive cells were consistent with several cell cycle stages, though cells with intense Zta staining typically exhibited moderate EdU (Fig. 5B). This observation implied Zta expression is attenuated in peak S-phase cells.

Fig. 5.

Pseudotemporal inference of EBV lytic infection and DDR phenotypic trajectories. (A) High-resolution cell cycle staging by DNA and EdU in Namalwa. (B) Zta expression by cell cycle stage. Points represent cells segmented from DNA_EdU_Zta_γH2AX color-coded by Zta intensity. Low DNA intensity with low EdU intensity corresponds to G1 and sub-G1; medium DNA intensity with elevated EdU intensity corresponds to S-phase (and pseudo-S phase for reactivating EBV+ cells); low EdU intensity with high DNA intensity corresponds to G₂/M. (C) Morphologic pseudotime trajectories for DNA_EdU_Zta_γH2AX. Pseudotime values were calculated relative to late reactivation (pt = 0) determined by Zta expression and VRC morphology. Three morphologic state trajectories are denoted by colored lines with arrows indicating directionality. (D) Phenotypic progression inferred along each of three pseudotime trajectories annotated in (C). Green-to-red trajectory (Top row) captures cells in the latent-to-lytic transition. Green-to-gold trajectory (Bottom row, Left) captures cells entering the cell cycle and progressing through S-phase. Blue-to-green trajectory captures cells in late stages of S-phase through G₂/M and subsequent return to G₁. Cell pseudotime (pt) values are provided. (E) Inferred progression (left to right) of nuclear morphology, Zta expression, VRCs, and host genome damage (γH2AX) through IE, early, and late lytic infection for P3HR1-ZHT + 4HT. (F) Morphologic pseudotime trajectory for DNA_EdU_gp350_53BP1. Pseudotime was calculated relative to late reactivation (pt = 0) determined by maximal gp350 intensity and VRC morphology. The morphologic trajectory from latent-to-late lytic infection is denoted by a colored line with the arrow indicating directionality. (G) Phenotypic progression inferred along pseudotime trajectory depicted in (F). The trajectory is initiated from a nonlytic (gp350-negative) cell in G₁ (EdU-negative) exhibiting diffuse 53BP1. Pseudotime values are provided for cells as in (D). A nonlytic cell with 53BP1 foci identified from the same 60× field of view as several depicted lytic cells is included for reference (outlined in black, not part of trajectory).

While EBV lytic reactivation is infrequent, the scale and robustness of HCS enabled capture of thousands of lytic cells in asynchronous phases. In lieu of live-cell tracking (which is incompatible with EdU labeling), thousands of snapshots effectively constitute out-of-order movie frames depicting reactivation dynamics. Originally built to identify developmental trajectories from single-cell transcriptomes, pseudotime analysis (71) is a powerful graph-based approach for organizing heterogeneous cell states into ordered phenotypic progressions. Pseudotime has been used previously in EBV scRNA-seq studies (8, 9). Because scRNA-seq and HCS datasets are structurally similar (cell-by-feature matrices), we applied pseudotime analysis to morphologically resolve dynamic EBV-host dynamics. Moncole3 (72) was used to calculate a contiguous pseudotime graph anchored from the Zta-positive cluster (pseudotime = 0) in the DNA_EdU_Zta_γH2AX survey experiment (Fig. 5C). Cell pseudotime values were calculated based on graph distance from this anchor (Fig. 5C dark purple cells at graph origin), where higher magnitude negative values indicated cells phenotypically distant from lytic cells. We also defined a reference point (cells negative for EdU, Zta, and γH2AX) shared across distinct pseudotime subtrajectories (Fig. 5C black ring and gradient arrows). Segmented cell metadata identified representative cells along each trajectory; for consistency, we highlighted P3HR1-ZHT cells for spontaneous and HT-induced reactivation (Fig. 5D). One trajectory (green-red arrow) captured development of Zta-high cells, along which gradual emergence of punctate Zta signal concomitant with colocalized VRC formation progressed toward uniform high Zta intensity and compartment expansion (Fig. 5 D, Top two rows). This transition was consistent with Zta pioneering functions (73). Two additional trajectories corresponded to G₁-to-S entry (green-orange arrow) and S-phase exit followed by subsequent cell cycle progression (blue-green arrow), consistent with manually curated cell cycle states. Cells at several stages (mostly late S-phase) along these two trajectories exhibited sparse punctate Zta (Fig. 5 D, Bottom row). We also identified progressive changes in nuclear morphology, Zta compartmentalization and intensity, viral DNA replication, and γH2AX profiles in 4HT-treated P3HR1-ZHT cells (Fig. 5E). While untreated cells exhibited moderate perinuclear γH2AX, higher intensity host genome-wide γH2AX was present in all 4HT-treated Zta-positive cells and directly correlated with DNA intensity throughout the ROCC I to ROCC II transition. Persistent EBV-mediated DNA damage has previously been reported in cell lines without external stimulation (74). Additionally, the difference in γH2AX profiles between unstimulated and 4HT-treated cells may reflect a consequence of spontaneous vs. overexpressed Zta. Similar analysis confirmed loss of 53BP1 detection in lytic cells as viral replication commences (SI Appendix, Fig. S19).

Pseudotime analysis also recovered a lytic trajectory from cells stained for DNA, EdU, gp350, and 53BP1 (Fig. 5 F and G). The calculated graph was anchored to cells with maximal lytic signal (pseudotime = 0), and reactivation progression was inferred beginning from a nonlytic reference cell exhibiting diffuse 53BP1 and the absence of gp350 and EdU. 53BP1 signal was virtually absent from phenotypes with discernible VRCs (EdU-positive) or gp350. Early VRC nucleation in pseudotime preceded gp350 expression, which became intense across cells with expanded and fused VRCs as well as cells with fully marginated host chromatin and no EdU. This sequence matched known viral lytic kinetics (75) and the dependence of late-stage ROCC on viral genome replication, thus demonstrating the accuracy and utility of morphologic pseudotime reconstruction to study viral infection dynamics using high-throughput snapshot imaging. We validated pseudotime-inferred lytic reactivation progression by live single-cell tracking of ROCC phenotypes (SI Appendix, Fig. S20 and Movie S1). In real time, the ROCC I to ROCC II transition correlated with increased nuclear volume. These changes corresponded to prior live-cell lytic infection studies (45, 46). Reduced GFP reporter signal and cell motility indicated lytic death in late ROCC II. Analytically, we confirmed cluster relatedness and pseudotime continuity in tSNE space, indicating morphologic relatedness in pseudotemporal trajectories was not a dimensional reduction artifact (SI Appendix, Fig. S21).

HCS Assay Generalizability for Biological Studies.

Plate-based HCS can be adapted readily to study other biological aspects and dynamics of infected cells. To demonstrate the versatility, the same conditions previously surveyed for lytic responses and DDR were assayed for nuclei, mitochondria, and Zta staining (SI Appendix, Fig. S22A). Cells with high Zta expression were identified (clusters 4, 17) and examined for associated mitochondrial features (SI Appendix, Fig. S22 B and C). This revealed differential mitochondrial morphology in Zta-negative cells, Zta-positive cells with ROCC I, and Zta-positive cells with ROCC II (SI Appendix, Fig. S22D). These states were exemplified in P3HR1-ZHT cells lacking stimulation (Zta-negative; SI Appendix, Fig. S22 D, Top row), cells pretreated with PAA to impair lytic progression (ROCC I; SI Appendix, Fig. S22 D, Middle row), and 4HT-induced cells (ROCC II; SI Appendix, Fig. S22 D, Bottom row), respectively. Mitochondrial localization in most cells was anisotropic and compact. Transition from fused filamentous mitochondrial networks in Zta-negative cells (SI Appendix, Fig. S22 D, Top row) toward mitochondrial fission (globular fragments) in Zta-positive cells exhibiting late VRCs (SI Appendix, Fig. S22 D, Bottom row) was suggestive of mitophagy (76) and consistent with morphologic changes seen for Zta overexpression (77). In Zta-positive cells with marginated host DNA, fragmented mitochondrial networks were present and juxtaposed concave nuclear regions (SI Appendix, Fig. S22 D, Bottom row). Mitochondrial and gp350 costaining further supported accumulation of fragmented mitochondria proximal to EBV virion assembly sites in P3HR1-ZHT (SI Appendix, Fig. S22E). This finding was confirmed in TGF-β-treated Mutu cells via gp350, TOM20, nuclei, and pulsed EdU stains (SI Appendix, Fig. S22F). These experiments highlighted the modularity of single-cell HCS to support biologically tailored studies of host–virus interactions.

HCS and Morphologic Pseudotime Generalizability to Widefield Imaging Systems.

While previous experiments employed a laser-based, confocal system, widefield epifluorescence imaging may provide comparative advantages (e.g., fluorophore compatibility, live-cell studies, cost) for some applications at the expense of spatial resolution. Thus, we investigated whether HCS and morphologic pseudotime analyses were extensible to widefield imaging. EBV-positive BL lines were prepared as in prior experiments, stained with a five-color fluorescence panel (nuclei, Zta, mitochondria, the proliferation marker Ki-67, and near-infrared viability dye), and imaged at 40× magnification with real-time computational deconvolution (SI Appendix, Figs. S23 A and B and S24 A and B). Cell segmentation and measurement were performed with a modified pipeline (SI Appendix, Fig. S24C and Code S3) and used to produce dimensionally reduced representations by line, treatment, and cluster (SI Appendix, Fig. S23 C–E). Unsupervised clustering yielded a clear population of lytic cells exhibiting high nuclear texture entropy (as in confocal experiments) and model-dependent reactivation efficiency (SI Appendix, Fig. S23 F and G). Pseudotemporal ordering likewise delineated latent cell states from progressive reactivation evidenced by Zta intensity, host chromatin, and mitochondrial morphology (SI Appendix, Fig. S23H). As for confocal experiments, we observed the gradual accumulation of Zta intensity and nuclear chromatin deformation along the pseudotime reactivation trajectory from widefield data (SI Appendix, Fig. S23H). These data show proof-of-principle that HCS with pseudotime can be successfully adapted across imaging platforms with varying technical specifications.

Discussion

EBV and many other DNA viruses form nuclear VRCs (78), wherein interspecies genome conflicts are manifest (79). Foundational studies using transmission electron microscopy first revealed marginated nuclear chromatin during HSV-1 replication (80), and subsequent fluorescence microscopy studies have underscored the central importance of herpesvirus VRC formation and growth mediated by lytic proteins (81–84). Prior studies of EBV VRCs have been conducted using B95-8 marmoset lymphoblast (40, 44), adherent epithelial (42, 45, 85), and hybrid epithelial/B cell (45, 47) lines with Zta expression vectors. HCS substantially extends insights through low-bias quantitative host–virus kinetics and VRC architecture dynamics in B cells at high throughput. Consistent with viral mutant screening (47), VRC formation in EBV-infected B cells induces partial ROCC I before viral DNA synthesis, which is necessary for subsequent chromatin margination (ROCC II). ROCC was quantifiable by nuclear texture, which supported pseudotemporal reconstruction from compartment nucleation through endpoint reactivation. Full reactivation pseudotemporally originated from G_1, consistent with Zta-mediated CDK inhibitor activation that restrains entry into normal S-phase (86) and favors a pseudo-S-phase environment with impaired checkpoint-mediated arrest (39, 70). Unlike reactivation-competent models, impaired transition to ROCC II in Raji cells is consistent with the line’s mutant BALF2 replication defect (44, 47).

HCS revealed differential timing and efficiency of viral expression and VRC formation for common lytic-inducing stimuli (87). TGF-β induced lytic ROCC I in multiple BL lines but not DLBCL lines (Farage, IBL1). This may stem from EBV-mediated type II TGF-β receptor downregulation (88) and functional interference with SMAD signaling by herpesvirus (and virus-induced cellular) microRNAs (89–93). Though TPA+NaB induces Zta (94) and, indirectly, its transcriptional target early genes (73, 95), even transient TPA+NaB treatment impairs DNA replication and may delay viral genome replication similar to effects of phosphonoacetic acid (96) and consistent with reduced DNA synthesis in TPA-treated Namalwa (59). Reduced colocalization between Zta and EdU suggests TPA+NaB treatment may affect VRC formation. H₂O₂, which induces Zta expression and activity (97, 98), yielded a similar but transient effect on DNA synthesis.

Clinically relevant genotoxins also induce EBV IE and early lytic expression (27). We observed doxorubicin-induced reactivation (26, 27) in small Zta-positive subpopulations (<2% of cells) in multiple models. Rare doxorubicin-treated gp350/EdU-positive cells indicate permissivity for complete lytic cycle unless disrupted by viral DNA synthesis inhibitors (“kick-and-kill”) (26). Etoposide and panobinostat induced Zta-positive cells, although the absence of late lytic phenotypes suggests abortive reactivation. Efficient panobinostat-induced Zta in P3HR1-ZHT may derive from enhanced expression of the engineered fusion construct, however Zta induction in Raji and Farage cells indicates panobinostat-induced endogenous EBV reactivation. Genotoxin-dependent reactivation responses warrant future studies of drug-dependent DNA vs. chromatin damage (99). Doxorubicin and etoposide are anthracycline topoisomerase II poisons (100, 101) that promote chromatin damage via nucleosome eviction (102, 103), however doxorubicin also directly generates DSBs (102, 104). HDAC inhibitors remodel chromatin (105) without direct strand breaks and notably induce break-independent pATM activation (66), which would facilitate EBV reactivation (39, 87, 106). This effect appears consistent with the latent-to-lytic switch upon disruption of lytic-suppressing chromatin insulators (107–111). HDAC inhibitor-driven nucleosome eviction might be kinetically premature for lytic infection since γH2AX association with EBV oriLyt (52) would be lost upon histone H2A.X removal and early lytic transcription occurs from chromatinized viral genomes (69). Thus, mechanistically distinct genotoxins may initiate damage responses conducive or prohibitive to complete EBV reactivation.

Our data provide a unique view of DDR spatiotemporal dynamics during EBV reactivation, including depletion of γH2AX and 53BP1 from active VRCs. The early importance of γH2AX (51, 52) and 53BP1 (41, 43) and their apparent later exclusion from VRCs indicates a changing landscape of DDR components during reactivation stages. This could reflect the dynamic nature of viral DNA lesions and epigenetic topology from initially chromatinized genomes; early nucleosome eviction; strand nicking, ssDNA displacement, and stabilization during rolling circle amplification; and DSB generation as concatemers are processed into individual genomes (69). The observed spatiotemporal dynamics of 53BP1 are intriguing in this context – 53BP1 is present in early compartments when viral genome DSBs are presumably infrequent, yet 53BP1 appears depleted from late compartments in which DSBs are generated by concatemer processing. The lack of 53BP1 detection in late lytic cells might reflect transcriptional downregulation and/or protein degradation. We cannot exclude the possibility that apparent depletion may reflect epitope masking by 53BP1 interactions with viral proteins. Future work should address these possible mechanisms since they may modulate 53BP1 recruitment to viral or host DSBs in γH2AX-independent and γH2AX-dependent manners (112). By contrast, the MRE11-RAD50-NBS1 (MRN) complex [recruited to DSBs independently of γH2AX (113)] associates with newly synthesized viral DNA (40), though spatiotemporal MRN dynamics in lytic cells require further study. While host repair of viral DSBs generated by concatemer processing has been speculated (40), such repair would ostensibly interfere with virion genome packaging. Alternatively, DSB-binding host proteins might protect viral genome ends while downstream damage signaling is subverted.

The γH2AX depletion from VRCs contrasts with ubiquitous host-targeted γH2AX throughout reactivation. Similar γH2AX accumulation occurs during KSHV reactivation (114, 115), suggesting conserved mechanisms among gammaherpesviruses. Pan-nuclear γH2AX induction is also observed during adenovirus infection (116). Considering the absence of 53BP1 foci observed here and by others (42, 48, 114), persistent pan-nuclear γH2AX may perturb genome stability and host transcription while checkpoint-mediated arrest is abrogated. Thus, spatiotemporally dysregulated DDR may directly support viral genome replication while simultaneously stalling host transcription and damage-induced arrest.

Early HCS implementations were used to study filoviruses (117), influenza A (118), and more recently BK polyomavirus (119). However, HCS has not been adopted widely for virology studies. We expect the approach demonstrated herein will be advantageous for studying host–virus interactions and heterogeneous infection outcomes including innate sensing (120, 121), cell death responses (122–124), and persistence (125, 126). Latently infected cells exhibit additional biologic diversity depending on dynamic host–pathogen interactions (111, 127). Lytic replication likewise comprises heterogeneity (128) including abortive vs. complete cycles and single-cell virion burst sizes (129, 130). Notably, infrequent or rare infection states may be key pathogenic determinants (131), highlighting the imperative for high-throughput single-cell approaches to study virus-associated diseases. The HCS workflow offers major practical and biological advantages over conventional techniques to address these challenges and is readily customizable to investigate host–virus interactions that are challenging to assay in live cells. Automated quantification reduces bias in image interpretation and low-throughput sampling; clustering and dimensional reduction support phenotype discovery and streamlined analyses; and pseudotime enables host–virus dynamic inference in asynchronous cells without live-cell tracking. We have provided a template R script with instructions for loading, processing, and analyzing CellProfiler (53) morphologic measurement files with Seurat v5 (55) and monocle3 (71, 72) to help users with less image processing and programming experience (Code S1). Processed R dataset objects (.rds) containing morphologic measurements, UMAP coordinates, and cluster metadata are available as supplementary files (Files S5-S11; see Data, Materials, and Software Availability). By coherently reconstituting many dynamic “attitudes” of cells in motion (132), we anticipate this methodology will accelerate single-cell virology.

Limitations of the Study

The HCS assay is not currently compatible with live-cell suspension imaging. The number, sequence, and complexity of processing steps required for cell segmentation will vary across models and cell types. We speculate successful pseudotime implementations require at least one phenotypically continuous cellular structure or process. Accurate inference of rapid state transitions may be challenging due to reduced probability of detecting sufficient intermediate phenotypes. Accurate morphologic state and dynamics interpretations require manual validation and should be informed by prior knowledge of host–virus interactions if available. Supervised machine learning and methods for trained phenotypic classification may be preferable to unsupervised methods. Ideally, pseudotime trajectories should be validated by live-cell tracking or time-resolved sampling. In the absence of effective isolation strategies (e.g., FACS, magnetic separation), rare phenotype validation and downstream analyses remain challenging. Phenotypic profiling depth is constrained by separable fluorescence channels. Future applications should develop multiplexed detection via spectral techniques or iterative detection.

Materials and Methods

Cell lines used in this study were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS). All pharmacologic and biologic stimuli used for lytic induction were obtained from commercial suppliers. 96-well plate-based imaging was performed on two microscopes: a Yokogawa CV8000 confocal HCS system and a Leica DMi8 Thunder epifluorescence system. Details for cell lines, treatments, fixed and live-cell staining, plate-based preparation, microscopes, image acquisition, processing, data analysis, visualization, and statistical analyses are provided in SI Appendix.

Ethics Statement.

All experiments in this study were conducted with University of Michigan Institutional Biosafety Committee (IBC) approval (#IBCA00002858).

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Code S03 (R)

Code S04 (TXT)

Code S05 (TXT)

Movie S1.

Live-cell imaging of ROCC I to ROCC II transition in Akata-GFP BL cell. GFP reporter expression (green) and nuclear morphology (DRAQ5, blue) in a lytic Akata-GFP BL cell visualized for 3.5 hours in 15-minute intervals at 60x magnification. Live cells were incubated at 5% CO2 and 37°C. The time stamp is in hh:mm format (00:00 corresponds to a lytic cell in ROCC I stage).

Data, Materials, and Software Availability

Single-cell image quantification data are available via Zenodo (10.5281/zenodo.19673764) (133). These files and original images are also accessible via Globus. CellProfiler pipelines and template R code for loading, processing, and analyzing output measurements (.csv) are included herein. Microscopy Images data have been deposited in University of Michigan Globus Endpoint (scEBV_HCS_Lytic) (https://app.globus.org/file-manager?origin_id=89113b26-373c-4933-87c1-670ccf25e7cf&origin_path=/) (134). All other data are included in the manuscript and/or supporting information.