Epstein-Barr virus (EBV), a gamma-1 herpesvirus, infects more than 90% of adults worldwide, yet in a subset of patients it actively contributes to lymphoma development and drives an especially poor clinical course. EBV is etiologically linked to multiple non-Hodgkin lymphoma (NHL) subtypes, including natural killer/T-cell lymphoma (NKTCL), follicular lymphoma, and diffuse large B-cell lymphoma (DLBCL), as well as to Hodgkin lymphoma (HL). In each of these contexts, EBV positivity is independently associated with inferior survival compared to EBV-negative disease. Despite advances in immunotherapy, the standard treatment for EBV-positive lymphoma remains radiotherapy and chemotherapy combined with antiviral agents, reflecting the absence of targeted therapeutic strategies developed specifically around EBV biology.
The immune microenvironment connection: A central premise of this work is that EBV does not simply reside passively within lymphoma cells - it actively reshapes the immune microenvironment (IME) to suppress antitumor immunity and promote disease progression. The IME of lymphomas contains diverse infiltrating immune cells including T cells, B cells, macrophages, natural killer cells, myeloid-derived suppressor cells, and dendritic cells. Tumor-associated macrophages (TAMs) are of particular interest: M2-polarized macrophages in the tumor IME are consistently associated with poor prognosis in NHL, while M1-polarized macrophages were traditionally viewed as antitumor. More recent evidence has complicated this picture, showing that M1 macrophages exposed to apoptotic lymphoma cells can paradoxically promote lymphoma growth. EBV may modulate macrophage polarization via macrophage migration inhibitory factor (MIF), a cytokine that drives M2 polarization and whose deficiency spontaneously restores M1 states.
Study rationale: At the time of this study, comparisons between EBV-positive and EBV-negative lymphoma had focused predominantly on non-coding RNA differences, with relatively little attention to mRNA-level differences or their downstream effects on the IME. The authors aimed to systematically identify mRNA-level differential genes between EBV+ and EBV- lymphoma using public gene expression datasets, link those genes to immune cell infiltration patterns via computational deconvolution, validate findings in clinical specimens, and ultimately propose novel therapeutic targets for EBV+ NHL patients.
This study employed a layered computational and experimental design, drawing on eight publicly available gene expression datasets alongside a prospectively collected clinical cohort. The primary differential gene discovery was performed using three GEO microarray datasets: GSE38885 (65 immunocompetent NHL patients including 31 EBV+ and 34 EBV-), GSE34143 (3 EBV+ peripheral T-cell lymphoma patients), and GSE13996 (18 EBV+ and 33 EBV- Hodgkin lymphoma patients with lymph node samples). It is notable that GSE38885 also included 40 immunocompromised post-transplant lymphoma patients, a specific subpopulation with distinct immune biology that required separate handling in the analysis.
Prognostic analysis datasets: To determine whether the identified differential genes carried prognostic significance, the authors used three survival-annotated datasets: GSE4475 (159 NHL patients including 123 DLBCL, 22 atypical Burkitt lymphoma, 9 aggressive B-NHL unclassifiable, and 5 Burkitt lymphoma), GSE39133 and GSE39134 combined (58 HL patients), and the TCGA DLBC cohort (48 DLBCL patients). Two additional datasets, GSE132929 (227 germinal center B-cell-like DLBCL patients) and GSE58445 (147 peripheral T-cell lymphoma patients), were used specifically for immune infiltration analysis. Batch effect correction across datasets was performed using the sva package via the Sangerbox online platform.
Bioinformatics workflow: Differential gene identification used the limma algorithm in R v4.0.5. Genes were considered EBV-related differentially expressed if they met a false discovery rate-adjusted p-value below 0.05 and an absolute log2 fold change greater than 1. Prognostic filtering was then applied via univariate Cox proportional hazard analysis, retaining genes with p values below 0.05 for survival associations. Functional enrichment of the differential genes used the Metascape platform for GO and KEGG annotation. Immune cell infiltration was estimated using CIBERSORT with the LM22 signature matrix (covering 22 immune cell subtypes) and 1,000 permutations. Targeted drug identification was performed against the DrugBank and PubChem databases using the hub genes as query targets.
Clinical validation: To ground the computational findings in real tissue, the authors collected formalin-fixed paraffin-embedded (FFPE) tissue from 30 NHL patients (15 EBV+ and 15 EBV-) at the Third Xiangya Hospital under IRB approval. RNA extraction from FFPE used the AmoyDx FFPE RNA Kit, with quantification on a NanoDrop 2000. RT-qPCR was performed on a LightCycler 480 II system using HiScript II U+ One Step RT-qPCR Probe Kit, with gene expression normalized to the ribosomal protein RPL13A housekeeping gene. Immunohistochemistry on FFPE sections used antibodies against PLA2G2D and TMEM163 to complement the PCR findings.
The initial differential analysis comparing EBV+ and EBV- samples in the combined lymphoma datasets (GSE38885 and GSE13996) identified 35 differentially expressed genes when NHL and HL samples were analyzed together. The limited overlap led the authors to separately analyze NHL and HL cohorts, correctly reasoning that EBV likely affects these two lymphoma types through distinct biological pathways given their very different pathological characteristics, malignancy grades, and survival curves. In GSE38885 (NHL samples), 21 up-regulated and 14 down-regulated genes were identified in the EBV+ group meeting the stringent criteria of |log2FC| greater than 1 and FDR-adjusted p below 0.05. In GSE13996 (HL samples), a separate set of differential genes was obtained.
CCL3 as the IME bridge: A Venn diagram analysis comparing differential gene lists from the NHL (GSE38885) and HL (GSE13996) cohorts identified only one overlapping gene: CCL3, also known as macrophage inflammatory protein 1-alpha (MIP-1a). This chemokine is known to recruit macrophages to sites of inflammation. The singular overlap of CCL3 across both lymphoma types, combined with its established role as a macrophage chemoattractant, immediately focused the analysis on the IME's macrophage compartment as a shared mechanism by which EBV affects prognosis across lymphoma subtypes. High CCL3 expression was confirmed in EBV+ immunocompromised lymphomas (30 EBV+ versus 10 EBV- in GSE38885), further validating its relevance to EBV biology.
Four prognostic hub genes identified in NHL: Focusing on the 65 NHL patients in GSE38885 (comparing EBV+ and EBV- groups), a separate differential analysis yielded 151 genes (79 up-regulated, 72 down-regulated) in EBV+ NHL. Functional enrichment of these 151 genes pointed to viral response, immune system regulation, and immune cell-related pathways. Univariate Cox survival analysis applied to this gene list, with validation against TCGA-DLBC (48 patients) and GSE4475 (159 NHL patients excluding germinal center B-cell-like DLBCL), identified four down-regulated genes consistently associated with poor prognosis in NHL: CHIT1 (chitinase 1), SIGLEC15 (sialic acid-binding Ig-like lectin 15), PLA2G2D (phospholipase A2 group IID), and TMEM163 (transmembrane protein 163). High expression of each of these four genes predicted longer overall survival.
To understand how EBV shapes the cellular composition of the tumor immune microenvironment, CIBERSORT was applied to five datasets (GSE38885, GSE34143, GSE132929, GSE58445, and GSE13996) to estimate the relative proportions of 22 immune cell subtypes from bulk gene expression data. The LM22 reference matrix was used, which provides gene expression signatures for 22 human immune cell types including seven T cell subsets, B cells, NK cells, and macrophage subtypes M0, M1, and M2. At the broadest level, macrophages constituted the dominant immune infiltrate in lymphoma samples across all datasets analyzed, with M0 macrophages (resting, undifferentiated macrophages) showing the highest infiltration levels among macrophage subtypes.
M0 macrophage differential between EBV groups: A critical finding emerged when EBV+ and EBV- NHL samples in GSE38885 were compared: M0 macrophage infiltration was significantly higher in the EBV- group than in the EBV+ group. Given that M0 macrophages represent a resting, polarization-capable state, and that their abundance is positively correlated with favorable prognosis in lymphoma, a reduction in M0 macrophages in EBV+ patients suggests that EBV may be shifting the macrophage pool away from this protective resting state. Beyond M0 macrophages, CIBERSORT analysis also revealed significant differences in T cell follicular helper (Tfh) cells and regulatory T cells (Tregs) between EBV+ and EBV- groups across the combined lymphoma samples, indicating broader immune dysregulation.
CCL3 and macrophage correlations: Correlation analysis in the 65 NHL patients from GSE38885 showed that CCL3 was negatively correlated with M0 macrophage infiltration but positively correlated with M1 macrophage infiltration - a finding also confirmed in the 51 HL patients from GSE13996. In immunocompetent NHL patients from GSE132929, CCL3 again correlated positively with M1 macrophages. Combined with recent literature showing that M1 macrophages exposed to apoptotic lymphoma cells actually promote tumor growth, this pattern suggests that M1 macrophages in the lymphoma context represent a pro-tumor polarization state, contrary to their conventional antitumor role in solid tumors. EBV may thus be promoting a shift toward M1 macrophage polarization through CCL3 upregulation, paradoxically worsening prognosis.
Four hub genes and macrophage correlation: Correlation analysis between the four hub genes (CHIT1, SIGLEC15, PLA2G2D, TMEM163) and macrophage infiltration levels in GSE38885 revealed a consistent pattern: all four genes were positively correlated with M0 macrophage infiltration and negatively correlated with M2 macrophage infiltration. These correlations were replicated in the immunocompetent NHL datasets GSE132929 and GSE58445. Since M0 macrophages are the precursors to both M1 and M2 states, the interpretation offered is that higher expression of these genes promotes M0 accumulation and inhibits their polarization into the functionally active M1 or M2 states - preserving the protective resting macrophage pool and thereby improving prognosis.
CHIT1 (Chitinase 1): CHIT1 is widely regarded as a marker of activated macrophages, expressed primarily in macrophages rather than in tumor cells. It plays regulatory roles in macrophage maturation and polarization: during monocyte-to-macrophage differentiation, CHIT1 is significantly upregulated in both M1 and M2 macrophages, and changes in its expression level correlate with the direction of macrophage polarization. In this study, CHIT1 expression was positively correlated with M0 macrophage infiltration and negatively correlated with M2 macrophage infiltration in NHL patients. Two EBV+ patients who died during the study period showed the lowest CHIT1 expression across all samples, establishing a direct empirical link between very low CHIT1 and fatal outcomes. The authors note that the lymphoma immune microenvironment likely modifies the normal regulatory relationship between CHIT1 and macrophage polarization.
SIGLEC15 (Sialic Acid-Binding Ig-Like Lectin 15): SIGLEC15 is emerging as a tumor-associated immune checkpoint molecule distinct from PD-L1/PD-1. Pan-cancer analyses have linked higher SIGLEC15 expression to improved overall and relapse-free survival in multiple tumor types including bladder urothelial carcinoma, breast invasive carcinoma, head and neck squamous cell carcinoma, thyroid carcinoma, and uterine corpus endometrial carcinoma. This study identified SIGLEC15 as down-regulated in EBV+ NHL patients and associated with poor prognosis - a relationship not previously described in lymphoma. The two deceased EBV+ patients in the clinical cohort also showed the lowest SIGLEC15 expression, suggesting potential utility as both a prognostic biomarker and therapeutic target via expression upregulation strategies.
PLA2G2D (Phospholipase A2 Group IID): This secreted phospholipase is involved in lipid metabolism and inflammatory signaling. Its expression was associated with favorable prognosis in survival analyses, but notably its high baseline expression in lymphatic tissue itself complicated experimental validation - PCR and IHC results in clinical samples were inconsistent with the computational predictions, likely because endogenous PLA2G2D expression in lymphoid tissue masked the EBV-related differences. This illustrates a common challenge in validating tissue-expressed markers in biopsies from the same tissue type.
TMEM163 (Transmembrane Protein 163): TMEM163 encodes a zinc transporter with functional similarity to the SLC30 family, which transports zinc ions from the cytoplasm to the extracellular space or organelle lumen. Zinc ion transport plays important roles in macrophage function and innate immunity. The hypothesis offered is that TMEM163 regulates macrophage subtypes by modulating intracellular zinc availability - potentially including induction of apoptosis in M2 macrophage subpopulations, analogous to how loss of SLC39A10 (a zinc importer) causes macrophage apoptosis through p53 stabilization. RT-qPCR results for TMEM163 in the clinical cohort showed the expected lower expression in EBV+ compared to EBV- patients, consistent with computational predictions.
The computational findings were tested in a prospective clinical validation cohort of 30 NHL patients from the Third Xiangya Hospital: 15 EBV+ and 15 EBV-. The cohort composition reflected real-world EBV+ NHL epidemiology, with the EBV+ group dominated by NK/T-cell lymphoma (12 of 15 patients), which is the NHL subtype most strongly associated with EBV infection, plus 2 Burkitt lymphoma and 1 DLBCL. The EBV- group was more heterogeneous, including 8 DLBCL, 3 peripheral T-cell lymphoma, 1 mantle cell lymphoma, 1 chronic lymphocytic leukemia/small lymphocytic lymphoma, 1 enteropathy-associated T-cell lymphoma, and 1 plasmablastic lymphoma. This distribution is epidemiologically appropriate but introduces subtype heterogeneity as a source of biological noise in the gene expression comparisons.
RT-qPCR results: CHIT1 mRNA expression was lower in EBV+ than EBV- patients, in the expected direction, but the difference did not reach statistical significance (unpaired t-test with Welch's correction, p not reported as significant). TMEM163 mRNA was similarly lower in EBV+ patients, again directionally consistent but not statistically significant. PLA2G2D mRNA appeared higher in EBV- versus EBV+ patients, consistent with its predicted down-regulation in EBV+ disease, but again without a statistically significant difference. SIGLEC15 showed a rising trend in the EBV- group but no significant difference. When GC-DLBCL cases were excluded from the comparison (reducing to 13 EBV+ and 13 EBV-), similar non-significant directional trends were maintained.
Interpretation of non-significance: The authors appropriately contextualize the lack of statistical significance as a power issue rather than a contradiction of the computational findings. The sample size of 15 per group is limited, particularly when the EBV+ group contains predominantly NKTCL patients while the EBV- group contains mostly DLBCL - introducing a confound between EBV status and lymphoma subtype. Despite this, the directional consistency between computational predictions and clinical measurements strengthens confidence in the hub gene signatures. The two patient deaths, both in the EBV+ group, showed the lowest expression of both CHIT1 and SIGLEC15 across all 30 samples, providing a clinically compelling if anecdotal validation of the extreme-end predictions.
A drug repurposing and target identification analysis was performed using DrugBank and PubChem databases, querying each hub gene as a molecular target to identify existing or candidate compounds. This type of analysis is an increasingly common step in computational biology studies, as it connects newly identified disease-relevant genes to the existing pharmacological landscape without requiring de novo drug development. The rationale is straightforward: if down-regulation of CHIT1, SIGLEC15, PLA2G2D, and TMEM163 drives poor prognosis through macrophage dysregulation in EBV+ NHL, then therapeutic strategies that upregulate or activate these targets could improve the immune microenvironment and patient outcomes.
Compounds identified: For CHIT1 (Chitinase 1), DrugBank returned two compounds: DB03109 (2-acetylamino-2-deoxy-b-D-allopyranose) and DB03539 (2-(Acetylamino)-2-Deoxy-6-O-Methyl-Alpha-D-Allopyranose). These are sugar analogs with structural similarity to chitin hydrolysis products. For PLA2G2D (Phospholipase A2 Group IID), DrugBank identified DB03193 (Stearic acid) and DB04404 (Lauric acid), both fatty acids known to interact with phospholipase binding sites. For SIGLEC15, PubChem returned two compounds identified by CID numbers 155553698 and 155518853, described as glycan-based structures (C40H67N5O27 and C42H70N6O27 respectively, with full names provided in supplementary materials). No druggable candidates were reported for TMEM163.
Translational context: The authors are careful to note that the functional activity of these identified compounds against the hub gene targets in the lymphoma context remains unclear and that the compounds are research-stage candidates rather than clinical drugs. SIGLEC15 is a particularly interesting target because it is an immune checkpoint receptor whose blockade is being explored in solid tumors (NC318, a SIGLEC15-blocking antibody, has entered clinical trials for advanced solid tumors). Whether anti-SIGLEC15 blockade or SIGLEC15 agonism would be beneficial in EBV+ NHL requires additional mechanistic investigation, as the directionality of its immune checkpoint function in lymphoma may differ from solid tumor biology.
Dataset limitations: A key methodological challenge is that the initial differential gene discovery necessarily relied on datasets including immunocompromised post-transplant lymphoma patients (GSE38885), because immunocompetent EBV+ lymphoma samples are scarce in public repositories. Post-transplant lymphoproliferative disorders have distinct immune biology driven by profound iatrogenic immunosuppression, which could skew the differential gene landscape compared to immunocompetent EBV+ NHL. The authors addressed this by subsequently validating all four hub genes in immunocompetent NHL cohorts (TCGA-DLBC, GSE4475, GSE132929, GSE58445) for both survival and immune infiltration analyses, demonstrating consistent associations. They conclude the hub genes retain general relevance to NHL biology regardless of immune status, but the limitation is acknowledged as a source of potential bias in the initial discovery phase.
Clinical cohort limitations: The 30-patient validation cohort faces several issues: the sample size is underpowered for statistical significance; the subtype composition differs substantially between EBV+ (predominantly NKTCL) and EBV- (predominantly DLBCL) groups, creating a subtype-EBV confound; and the number of deaths available for extreme-case analysis is only two. The authors note that future studies should increase clinical sample sizes and use subtype-matched cohorts to cleanly separate the effects of EBV infection from subtype-specific biology. Single-cell sequencing is proposed as the next methodological advance to identify the specific macrophage subpopulations mediating the hub gene effects.
M0 macrophage biology: A notable conceptual contribution of this study is the reframing of M0 macrophages - previously viewed simply as an undifferentiated precursor state - as a functionally relevant IME component whose abundance correlates positively with prognosis. The authors propose that higher expression of CHIT1, SIGLEC15, PLA2G2D, and TMEM163 promotes M0 macrophage accumulation and simultaneously inhibits their polarization into the pro-tumor M1 or M2 states, effectively preserving a macrophage resting pool that is associated with better outcomes. This model, if validated by functional experiments, would represent a significant refinement of the macrophage polarization paradigm in lymphoma immunology.
Future therapeutic directions: The study identifies several translational priorities: functional characterization of the four hub genes in macrophage polarization systems; mechanistic dissection of how TMEM163-mediated zinc transport affects M2 macrophage survival; clinical testing of SIGLEC15 as a lymphoma immune checkpoint target; and development of strategies to increase expression of these down-regulated genes in EBV+ NHL patients. The authors also highlight CHIT1 as a candidate prognostic biomarker distinguishing immunocompromised from immunocompetent NHL patients, warranting further investigation in this underappreciated clinical distinction.