Diffuse large B-cell lymphoma (DLBCL) is the most common aggressive lymphoma worldwide, but not all DLBCL cases are biologically equivalent. A subset, defined by the presence of Epstein-Barr Virus (EBV) within tumor cells and designated EBV+DLBCL, carries a distinctly worse prognosis than its EBV-negative counterpart. In the cohort assembled for this study at The First People's Hospital of Foshan, 23 out of 236 DLBCL cases (9.7%) were EBV-positive based on EBER in situ hybridization, and these patients demonstrated significantly shorter overall survival (median 17 months, range 2 to 33 months) compared to EBV-negative patients (p < 0.05) following R-CHOP-based chemotherapy.
Clinical profile of EBV+DLBCL patients: Among the 23 EBV-positive cases, 60.87% were older than 60 years, 56.52% had elevated lactate dehydrogenase (LDH), and 60.87% had extranodal involvement affecting sites including the spleen (21.74%), tonsil (13.04%), gastric region (8.70%), and nasal cavity (8.70%). By International Prognostic Index (IPI) scoring, 30.43% fell into the high-risk category (4-5 points), and the male-to-female ratio was 1.1:1. This clinical profile is consistent with published literature characterizing EBV+DLBCL as a biologically aggressive disease with limited responsiveness to conventional chemoimmunotherapy regimens.
The core scientific problem: Despite accumulating evidence that EBV infection reshapes the tumor microenvironment (TME), the spatial distribution of immune cells, the specific ligand-receptor interactions driving immunosuppression, and the mechanistic basis for chemotherapy resistance in EBV+DLBCL remain poorly characterized. Prior single-cell RNA sequencing (scRNA-seq) studies have mapped cellular composition but lack spatial context, meaning they cannot reveal where different cell populations are physically located within the tumor tissue and how that geography influences biology.
This study is the first to apply 10X Visium spatial transcriptomics (ST) to EBV+DLBCL, integrating spatial gene expression data with published DLBCL scRNA-seq reference datasets to create a spatially resolved map of TME heterogeneity. The findings implicate PD-1/PD-L1 signaling as the dominant immunosuppressive axis and provide preclinical validation of PD-1/PD-L1 blockade as a therapeutic strategy specifically in EBV+ disease.
The study selected 8 formalin-fixed paraffin-embedded (FFPE) tissue specimens from the 236-patient DLBCL cohort, comprising 4 EBV-positive and 4 EBV-negative cases. Selection criteria required specimens to contain more than 50% tumor cells and less than 10% necrotic tissue, with complete clinicopathological and follow-up data available. Only samples from the most recent 2 years were used for spatial transcriptomics to avoid degradation artifacts inherent to FFPE archiving. Sequencing was performed on a NovaSeq 6000 System (Illumina) using a NovaSeq S4 Reagent Kit (200 cycles), achieving approximately 320 million read pairs per sample at a sequencing depth of roughly 5,000 to 10,000 unique molecular identifiers (UMIs) per spot and 1,000 to 3,000 genes per spot.
Visium 10X platform specifics: The 10X Visium platform uses capture spots of 55 micrometers in diameter, each containing approximately 8 to 20 cells. Each tissue section encompassed up to 5,000 capture spots within a 6.5 mm by 6.5 mm capture area. FASTQ files were processed using Space Ranger 1.3.0 with the Visium Human Transcriptome Probe Set v1.0 (GRCh38-2020-A). After quality filtering for spots with more than 2,000 genes and more than 10,000 UMIs, 34,795 high-quality spots were retained for downstream analysis. All raw data were deposited in the National Omics Data Encyclopedia (NODE; Project ID OEP004742).
Computational integration pipeline: Batch effects across samples were addressed using the Harmony algorithm. Data were merged into a unified Seurat object, normalized via SCTransform selecting the 3,000 most variable features based on residual variance, and then subjected to PCA followed by Harmony correction. The integrated dataset was analyzed using Seurat's standard workflow with the first 40 Harmony-corrected PCA dimensions for both UMAP visualization and Leiden clustering at resolution 0.35. Cell cluster identities were annotated by differential marker expression using a Wilcoxon rank-sum test, with thresholds of 25% minimum expression fraction and a log2 fold-change cutoff of 0.25.
Cell type deconvolution with Cell2location: Because each Visium spot contains multiple cells and cannot resolve single-cell identities directly, the authors used Cell2location, a Bayesian deconvolution framework, to estimate the probabilistic abundance of each cell type within each spot. The reference dataset for deconvolution was a published DLBCL scRNA-seq atlas (Steen et al., Cancer Cell 2021), enabling spatially resolved mapping of B cells, T cell subsets (CD4+, CD8+, TFH, Treg), monocytes, macrophages, and natural killer (NK) cells across the tissue. Ligand-receptor colocalization analysis was conducted using the stLearn package, with Bonferroni-corrected significance thresholds (adjusted p < 0.05) and a requirement for enrichment consistency across at least 2 independent patient samples.
Unsupervised Leiden clustering of all 34,795 spots from the 8 specimens resolved 12 distinct spatial clusters. Critically, the distribution of spot types differed substantially between EBV-positive and EBV-negative tumors. UMAP visualization colored by EBV status showed clear separation of EBV+ and EBV- spots into distinct transcriptional territories, indicating that viral infection fundamentally alters the gene expression landscape at the tissue level, not just in isolated cell subpopulations. The pseudo-bulk comparison between EBV+ and EBV- reference datasets confirmed high compatibility, validating the integration approach.
Spatial mapping of cell types: Using Cell2location-derived cell-type abundance scores, the authors mapped each cluster back to its histological location on the tissue sections. This revealed anatomically organized patterns where distinct cell populations occupied spatially segregated microniches. In EBV+DLBCL tissue, macrophages (monocytes and macrophages, Mo-Macro) were enriched in tumor areas, while in EBV-DLBCL the stromal regions contained relatively higher proportions of immune cells including follicular helper T cells (TFH). The spatial segregation of these populations was not apparent from scRNA-seq data alone, which only captures cellular composition without location.
Functional annotation of tumor clusters: Differential gene expression (DEG) analysis was performed for each identified cluster, followed by Gene Ontology (GO) enrichment analysis focusing on the top 10 GO terms per cancer cell cluster. Using the official GO website to trace parent pathways, clusters were classified by their dominant functional roles within the tumor, including processes related to immune regulation, cell proliferation, metabolic reprogramming, and stromal interaction. Table S2 in the supplementary data catalogs the GO pathway assignments for each cluster.
Immune regulation balance: Gene Set Variation Analysis (GSVA) was applied to immune regulation pathways in the tumor compartment of each sample. The results revealed that both positive and negative immune regulation were stronger in EBV+DLBCL than in EBV-DLBCL. Importantly, in 3 out of 4 EBV+ cases (all except patient P4), negative immune regulation exceeded positive immune regulation. This finding indicates that a majority of EBV+ patients have lost effective immune cell surveillance and are operating under a net immunosuppressive state within their tumor microenvironment.
Differential expression analysis of tumor regions comparing EBV+ versus EBV- specimens, using thresholds of absolute log2 fold-change greater than 1 and adjusted p-value less than 0.05, identified PD-1/PD-L1 immune response signaling as one of the most significantly upregulated pathways in EBV+DLBCL. KEGG pathway enrichment analysis confirmed that PD-1/PD-L1 pathway components were among the top hits in the EBV+ tumor DEG signature. This is consistent with prior literature showing that EBV infection can drive constitutive AP-1 activity that induces PD-L1 expression in lymphoid malignancies.
Immunohistochemistry validation: To confirm that PD-L1 upregulation at the transcriptomic level translates to protein expression, the authors performed IHC staining for PD-L1 on transverse tissue sections from EBV+ and EBV- lymph nodes using a validated PD-L1 antibody (Abcam ab205921, 1:200 dilution). Quantitative scoring of PD-L1 positive cells confirmed significantly higher PD-L1 expression in EBV+DLBCL compared to EBV-DLBCL, providing independent protein-level evidence for the spatially detected transcriptomic signal.
Spatial expression mapping: Spatial feature plots were generated to visualize the expression of PD-1 and PD-L1 genes overlaid directly onto H&E tissue images. In EBV+DLBCL, both PD-1 and PD-L1 showed concentrated, spatially clustered expression in tumor areas. In EBV-DLBCL, the same markers were present but dispersed across the tissue without the focal enrichment pattern observed in EBV+ cases. Co-expression plots (PD-L1 in red, PD-1 in blue) confirmed spatial co-localization of both molecules in EBV+ tumor microniches, directly visualizing the conditions required for PD-1/PD-L1 checkpoint engagement.
Cellular regulators differ by EBV status: Violin plot analysis of cell-type scores within the PD-1/PD-L1-enriched tumor areas revealed that the immune cell populations driving this signaling axis differ between subtypes. In EBV+DLBCL, Mo-Macro (monocytes and macrophages) carry the highest cell-type scores in PD-1/PD-L1-positive tumor clusters, designating them as the primary regulators. In EBV-DLBCL, follicular helper T cells (TFH) are the predominant immune cell in those regions. This subtype-specific cellular architecture explains why EBV+ disease generates a more profoundly immunosuppressive microenvironment, as macrophages are potent sources of co-inhibitory ligands and immunosuppressive cytokines including IL-10.
To functionally validate the spatial transcriptomics findings, the authors conducted cell-based experiments using two DLBCL cell lines: the EBV-positive Farage line (ATCC CRL-2630) and the EBV-negative DB line (ATCC CRL-2289). Both lines were maintained in RPMI-1640 with 10% FBS and treated with BMS202, a small-molecule PD-1/PD-L1 inhibitor, at concentrations of 5 uM, 10 uM, and 20 uM for 24 hours. Cell viability was assessed using the CCK8 (cholecystokinin-8) colorimetric assay, with absorbance measured at 450 nm.
Dose-dependent proliferation inhibition: BMS202 induced dose-dependent inhibition of proliferation and viability in both cell lines, but with significantly stronger effects in Farage (EBV+) cells. In Farage cells, statistically significant growth inhibition was observed at all tested concentrations (5 uM: p < 0.05; 10-20 uM: p < 0.001). In DB cells, only a non-significant downward trend was observed at 5 uM, with significance achieved at higher concentrations (10 uM: p < 0.05; 20 uM: p < 0.01). The authors attribute Farage cells' greater sensitivity to their constitutively higher PD-1/PD-L1 expression, reflecting greater dependence on this pathway for survival signaling.
Apoptosis quantification by flow cytometry: Apoptotic cells were quantified using Annexin V-FITC and propidium iodide (PI) dual staining analyzed on a BD FACS Calibur flow cytometer, with 10,000 events acquired per sample. Populations were defined as early apoptosis (Annexin V+/PI-) and late apoptosis (Annexin V+/PI+). Treatment with 10 uM BMS202 for 24 hours significantly induced apoptosis in both cell types, with Farage cells showing a particularly robust response (Farage: p < 0.01; DB: p < 0.05). The substantially stronger apoptotic response in EBV+ cells further supports the conclusion that EBV+DLBCL has a functionally heightened dependence on PD-1/PD-L1 signaling for cell survival.
mRNA confirmation of pathway blockade: Quantitative RT-PCR was performed on RNA isolated using TRIzol, with first-strand cDNA synthesis using SuperScript III Reverse Transcriptase. qPCR confirmed that BMS202 treatment reduced PD-L1 and PD-1 mRNA expression in both cell lines in a dose-dependent manner, verifying effective pathway blockade at the molecular level. The magnitude of PD-L1 mRNA suppression was greater in Farage cells than in DB cells, consistent with the differential baseline expression and sensitivity observed in the viability assays.
To identify downstream mediators connecting PD-1/PD-L1 signaling to the biological differences between EBV+ and EBV- DLBCL, the authors conducted protein-protein interaction (PPI) network analysis using the STRING database (https://string-db.org). From the PPI network constructed around PD-1/PD-L1 pathway genes, TLR4 (Toll-like receptor 4) emerged as a prominent node, suggesting it functions as a downstream effector in this signaling context. TLR4 is a pattern recognition receptor with established roles in innate immune activation and macrophage polarization, making its identification particularly relevant to the Mo-Macro-dominated immunosuppressive TME observed in EBV+DLBCL.
EBV-status-dependent TLR4 regulation: qPCR analysis of TLR4 mRNA expression following BMS202 treatment revealed strikingly opposite regulatory patterns in the two cell lines. In Farage (EBV+) cells, TLR4 was constitutively highly expressed at baseline. BMS202 treatment significantly downregulated TLR4 mRNA in Farage cells (p < 0.001), demonstrating that PD-1/PD-L1 signaling actively maintains TLR4 expression in EBV+ cells. In DB (EBV-) cells, the reverse was observed: baseline TLR4 expression was lower, and BMS202 exposure significantly upregulated TLR4 mRNA (p < 0.001). This divergent, EBV-status-dependent regulation of TLR4 establishes a functional molecular link between immune checkpoint activity and the distinctive immunosuppressive features of EBV-associated lymphomas.
Biological significance of TLR4 in EBV+DLBCL: The constitutively high TLR4 expression in EBV+ cells, maintained by active PD-1/PD-L1 signaling, may contribute to the immunosuppressive TME through multiple mechanisms. TLR4 activation in tumor-associated macrophages promotes an M2-like, immunosuppressive polarization state associated with IL-10 production and impaired cytotoxic T cell function. The fact that inhibiting PD-1/PD-L1 signaling in EBV+ cells drives TLR4 downregulation suggests that disrupting this checkpoint not only restores T cell activity directly but also modulates innate immune programming in the TME, potentially resensitizing it to immune surveillance.
The authors note that this TLR4 connection extends the mechanistic understanding of why EBV+DLBCL displays elevated IL-10 levels and impaired cytotoxic T cell function compared to EBV-DLBCL, as both are downstream consequences of macrophage TLR4-mediated immunosuppression. TLR4 is proposed as a potential predictive biomarker for clinical validation in future prospective studies.
Resolution limits of Visium spatial transcriptomics: The 10X Visium platform captures gene expression at spot-level resolution of approximately 55 micrometers, each encompassing 1 to 10 cells. This multi-cellular spot resolution obscures single-cell-level heterogeneity, particularly in regions where multiple cell types are spatially intermixed. Consequently, within-spot cellular diversity is averaged rather than resolved, limiting the granularity of cell-type assignments especially in mixed-cell tumor-immune interface zones. The Cell2location deconvolution approach was applied to partially compensate for this limitation, but deconvolution is probabilistic and depends on the completeness and accuracy of the scRNA-seq reference atlas.
Batch effects and sample processing variability: Despite Harmony-based batch correction applied across PCA embeddings, differences in sample processing, FFPE tissue age, RNA quality, and library preparation across the 8 specimens could introduce technical variation that partially confounds biological comparisons. Although high reproducibility was observed across replicates, the authors acknowledge that cross-sample comparisons remain susceptible to residual batch effects that cannot be fully eliminated computationally.
Dropout and low-abundance transcript underrepresentation: A known limitation of spatial transcriptomics is dropout, the stochastic failure to detect low-abundance transcripts, which disproportionately affects rare cell types and weakly expressed genes. This may cause underestimation of certain cell populations in sparse tumor regions and could affect the reliability of cell-type annotation for less common TME constituents. Future platforms with single-cell spatial resolution, such as Xenium or MERFISH, would address this limitation.
Small sample size and limited generalizability: The core spatial transcriptomics analysis was conducted on 4 EBV+ and 4 EBV- specimens, a sample size that, while consistent with published field standards for ST studies, limits statistical power and generalizability to broader patient populations. The rarity of EBV+DLBCL (approximately 10% of DLBCL in this cohort and in published series) and the requirement for high-quality FFPE tissue with matched clinicopathological data constrain the feasibility of assembling larger ST cohorts from routine clinical archives. The in vitro cell line validation, while strengthening mechanistic conclusions, was confined to 2 cell lines and may not fully represent primary tumor biology.
The findings of this study directly support the clinical rationale for PD-1/PD-L1 blockade in EBV+DLBCL. While R-CHOP remains standard first-line therapy for DLBCL with cure rates approximating 60%, roughly one-third of patients experience relapse with poor subsequent outcomes. For the EBV+ subset, the immunosuppressive TME identified here may be a primary driver of the observed inferior response to R-CHOP. Combining PD-1/PD-L1 inhibitors with frontline chemoimmunotherapy, or with novel cellular therapies such as CAR-T cells, could overcome this TME-mediated immunosuppression, particularly given the demonstrated sensitivity of EBV+ cells to PD-1/PD-L1 blockade in vitro.
TLR4 as a predictive biomarker: The identification of TLR4 as a PD-1/PD-L1 downstream effector with EBV-status-dependent regulation introduces a potential molecular biomarker for stratifying patients likely to respond to checkpoint inhibition. Clinical cohort studies measuring baseline TLR4 expression in EBV+DLBCL patients treated with checkpoint inhibitors could validate whether TLR4 level predicts response, durable remission, or toxicity outcomes. Integrating TLR4 assessment alongside EBV status into clinical trial design would provide a prospective test of this hypothesis.
Combination immunotherapy strategies: The presence of multiple immunosuppressive mechanisms in EBV+DLBCL, including macrophage-mediated suppression, elevated IL-10, T cell exhaustion, and alternative checkpoint expression (LAG3, TIM3), implies that single-agent PD-1/PD-L1 blockade may be insufficient to fully reverse TME-mediated immune escape. Combination strategies simultaneously targeting PD-1/PD-L1 and macrophage polarization (via M2 reprogramming agents or CSF1R inhibitors), or pairing checkpoint blockade with bispecific antibodies or CAR-T cell products, may be required to achieve durable remissions in this biologically aggressive subtype.
Advancing spatial resolution and multi-omics integration: The spatial transcriptomics framework established here provides a template for analyzing EBV+DLBCL with next-generation platforms such as 10X Xenium, Vizgen MERSCOPE, or NanoString CosMx, which resolve gene expression at true single-cell or subcellular resolution. Integrating spatial transcriptomics with spatial proteomics (CODEX or MIBI-TOF) would enable simultaneous protein-level characterization of the TME architecture. Additionally, expanding the spatial analysis to multi-site biopsies from EBV+DLBCL patients before and after treatment would capture temporal as well as spatial dynamics of immune remodeling, informing adaptive therapeutic strategies.