PD-L1-linked spatial decoupling of tumour-immune interactions in EBV-positive DLBCL

Technical Summary: PD-L1–linked spatial decoupling of tumour–immune interactions in EBV-positive DLBCL

Problem Statement
Epstein–Barr virus (EBV)-positive diffuse large B-cell lymphoma (DLBCL) is an aggressive malignancy characterized by immune evasion mechanisms involving viral latency proteins and host genomic alterations, particularly PD-L1 overexpression driven by 9p24.1 amplification. While PD-L1 is known to inhibit T-cell function, the specific spatial mechanisms by which PD-L1 genomic gains reshape the tumour microenvironment (TME) in EBV-positive DLBCL remain unresolved. It is unclear whether immune evasion in this context is driven by immune exclusion (lack of infiltration) or by a functional uncoupling where immune cells are present but fail to engage tumour cells effectively.

Methodology
The study employed an integrative multi-omics approach on a retrospective cohort of 34 EBV-positive DLBCL cases (34 patients), with a specific focus on stratifying samples by PD-L1 copy number status (Gain vs. Normal). The methodology included:

• Genomic Profiling: Whole-exome sequencing (WES) and OncoScan copy number arrays to characterize somatic mutations, mutational burden, and 9p24.1 amplification status.
• Bulk Transcriptomics: RNA sequencing to assess global gene expression, immune checkpoint landscapes, and pathway activities, compared against a large EBV-negative DLBCL cohort (GOYA).
• Single-Cell RNA Sequencing (scRNA-seq): Performed on 10 cases (3 PD-L1 Gain, 7 Normal) to resolve cell-type-specific transcriptional states, exhaustion trajectories (using CellRank2), and metabolic signatures.
• Spatial Transcriptomics: 10x Genomics Visium profiling of 18 samples to map cell-type composition, ligand-receptor interactions, and spatial gradients of immune function and metabolic suppression relative to tumour boundaries.
• Spatial Proteomics: CosMx spatial molecular imaging (64-plex protein panel) on 11 samples to validate cell-cell contact, CAF barrier function, and macrophage polarization at single-cell resolution.
• Analytical Framework: The study utilized network propagation (STRING), deconvolution (DWLS), differential expression (DESeq2), gene set enrichment (GSEA/GSVA), and mixed-effects modeling to analyze spatial gradients and functional states.

Key Contributions and Results

1. Genomic and Transcriptional Landscape:

   • EBV-positive DLBCL exhibits a heterogeneous genomic landscape with low mutational burden and minimal alignment with canonical DLBCL molecular subtypes (e.g., LymphGen 2.0).
   • PD-L1 gain (9p24.1 amplification) occurs in ~26% of cases and is associated with distinct transcriptional programs, including enrichment of inflammatory responses, TNFα/NFκB signaling, and hypoxia, contrasting with the proliferative/metabolic signatures of PD-L1 normal cases.
   • EBV latency programs (Type I/II vs. Type III) further stratify tumour cells into proliferation-dominated versus immune-activated states.

2. Spatial Decoupling of Immune Infiltration and Engagement:

   • A central finding is the "spatial decoupling" in PD-L1 gain tumours: T cells accumulate in close proximity to tumour cells but fail to establish direct cytotoxic contact.
   • CAF-Mediated Exclusion: PD-L1 gain tumours show a significant enrichment of cancer-associated fibroblasts (CAFs) at the tumour boundary (mesoscopic scale), creating a physical barrier. Single-cell proteomics confirmed that while T cells are near tumour cells, direct contact (<5 µm) is nearly absent in gain cases, replaced by CAF intercalation.
   • Altered Microenvironment: PD-L1 gain tumours display increased M2 macrophage enrichment and specific immune-immune co-localization (e.g., CD8-M1, neutrophil-M1) but reduced tumour-immune proximity compared to wild-type cases.

3. Functional Exhaustion and Metabolic Suppression:

   • T-Cell Exhaustion: Trajectory inference reveals that PD-L1 gain accelerates CD8 T-cell differentiation toward terminal exhaustion, depleting the progenitor exhausted (Tpex) pool. This exhaustion is spatially persistent, failing to recover even at distances away from the tumour boundary.
   • Metabolic Constraints: PD-L1 gain tumours exhibit a "metabolic suppression barrier" concentrated at the tumour-immune interface. This includes upregulation of adenosine pathway genes (ENTPD1, NT5E, ADORA2A/B) and arginine metabolism, suggesting nutrient deprivation and metabolic competition that further impairs T-cell fitness.

4. Reprogramming of Myeloid and Stromal Compartments:

   • PD-L1 amplification disrupts the spatial polarization of macrophages. While normal tumours show a gradient of M2 enrichment at the boundary, PD-L1 gain tumours show a loss of this gradient and a relative depletion of M2-like macrophages in the immediate tumour-proximal niche, despite global enrichment.

Significance and Claims
The paper claims to define EBV-positive DLBCL not as a classical "immune-excluded" or "immune-desert" tumour, but as an immune-infiltrated yet functionally disengaged ecosystem. The authors propose that PD-L1 amplification acts as a cooperative reinforcement of viral immune evasion, driving a multilayered suppression program that includes:

• Physical exclusion via CAF-mediated barriers.
• Metabolic suppression at the tumour interface.
• Enforced terminal T-cell exhaustion.

This "spatially organized decoupling" explains why immune cells are present but ineffective. The study suggests that therapeutic strategies for EBV-positive DLBCL may require targeting these spatial and metabolic constraints (e.g., combining PD-1 blockade with agents targeting LAG3, CTLA4, or metabolic pathways) rather than relying solely on checkpoint inhibition. The authors note that these spatial features could potentially be inferred from routine histopathology using machine learning, though this requires future validation. The study acknowledges limitations regarding sample size and the cross-sectional nature of the data, which precludes temporal inference of the evolution of these spatial architectures.