Prognostic model for diffuse large B-cell lymphoma based on an HBV-associated gene signature and immune microenvironment insights Authors: Guanhua Zhao, Sheng Gao, Yalu Liu Published: 2026 ABSTRACT AbstractDiffuse large B-cell lymphoma (DLBCL) is the most common malignant lymphoma and exhibits significant clinical and molecular heterogeneity. Hepatitis B virus (HBV) infection has been associated with distinct clinicopathological features and an unfavorable prognosis in DLBCL. However, the molecular mechanisms underlying HBV-associated DLBCL and the remodeling of its immune microenvironment remain poorly understood. Through in silico analysis on transcriptomic data from the GEO and TCGA databases, we identified a set of prognostic HBV-associated differentially expressed genes (DEGs) in DLBCL and constructed a prognostic model based on these genes. A total of 72 HBV-associated DEGs were identified. From these, a 13-gene risk signature was constructed from univariate Cox regression and stepwise Akaike Information Criterion (stepAIC) optimization. This model successfully stratified patients into high- and low-risk groups, with differing survival analyses (GSE10846: risk HR = 2.38, p < 0.001; 3-year AUC = 0.753). A combined nomogram integrating clinical data with the HBV-derived signature maintained excellent discriminative ability (3-year AUC = 0.794) and calibration in the external datasets. Immune infiltration analysis revealed that compared to high-risk patients, low-risk patients had a greater total stromal and total immune cells, lower tumor purity, and a higher proportion of cytotoxic T and NK cells. In contrast, high-risk patients exhibited higher counts of M2 macrophages and immune checkpoint-expressing suppressive cells. The 13-gene model, linked to HBV biomarkers, establishes a novel computational framework for risk stratification in DLBCL and highlights key alterations in the tumor immune microenvironment associated with HBV infection. These findings provide a rationale for exploring immunotherapeutic strategies and elucidate the pathogenesis of HBV-related lymphomas.Keywords: HBV-associated DLBCL, Tumor immune microenvironment, Prediction model, TCGA, GEO Introduction Diffuse large B-cell lymphoma (DLBCL) is one of the most common histologic subtypes of malignant lymphoma, accounting for approximately one-third of non-Hodgkin lymphomas worldwide. Despite progress in the molecular subclassification of DLBCL and the development of treatment strategies, the disease remains clinically heterogeneous, with highly variable responses to standard immunochemotherapy and diverse genetic backgrounds [1, 2]. Although several novel therapies, particularly those based on the monoclonal antibody rituximab, have improved survival rates, 30–40% of patients still experience relapsed or refractory disease due to a range of genetic, molecular, and microenvironmental factors [3, 4]. The oncogenic potential of viral infections has been well-documented across various malignancies, with distinct risk profiles observed across different populations and geographical regions [5–7]. Notably, hepatitis B virus (HBV), a non-cytopathic hepatotropic virus, is known to cause chronic infections that can progress to cirrhosis and hepatocellular carcinoma [8]. Epidemiological evidence has also linked HBV to an elevated risk of other malignancies, including lymphoma. Recent investigations have unveiled a biological association between HBV and B-cell lymphomas, especially DLBCL [9]. HBV-associated DLBCL exhibits distinct clinical characteristics, including advanced disease stage, suboptimal response to conventional immunochemotherapy, and inferior prognostic outcomes [10, 11]. At the molecular level, these cases frequently demonstrate aberrant activation of the nuclear factor kappa B pathway and mutations in critical oncogenes such as BCL-6 and MYC [11]. The pathogenic mechanism may involve HBV-induced chronic inflammation, promoting the accumulation of genetic and epigenetic alterations across multiple oncogenic signal pathways [12, 13]. Although HBV is recognized as a major risk factor contributing to the pathogenesis of HBV-associated DLBCL, there remains no consensus on a standardized or highly effective treatment protocol for this subtype. Emerging evidence underscores the pivotal role of immune-related genes and the tumor immune microenvironment (TME) in modulating tumor progression and therapeutic responsiveness [14, 15]. In the context of DLBCL, numerous studies have highlighted the significance of the TME, particularly immune status, in disease pathogenesis [16, 17]. Thus, accurately characterizing the TME is essential for developing effective treatment strategies for DLBCL patients. The integration of high-throughput sequencing with bioinformatics has established integrative transcriptomic modeling as a key approach for identifying prognostic biomarkers and elucidating disease mechanisms. In this study, we systematically evaluated gene expression profiles associated with HBV in DLBCL to construct a comprehensive prognostic framework. Using datasets from the Gene Expression Omnibus (GEO), The Cancer Genome Atlas (TCGA), and other sources, we identified a 13-gene HBV expression signature that effectively stratifies patients by relative survival and delineates deviations in the survival-specific immune microenvironment. This approach enhances our understanding of the molecular heterogeneity in HBV-related DLBCL and provides a conceptual foundation for prognostic modeling and therapeutic design. Methods Data acquisition and preprocessing The RNA sequencing data along with clinical annotations for patients diagnosed with DLBCL were retrieved from the Gene Expression Omnibus (GEO). Three datasets were comprised in this study cohort: GSE56315 (35 DLBCL and 33 normal samples) and GSE10846 (417 DLBCL samples) for training, and GSE87371 (223 DLBCL samples) for validation. All datasets were generated using the GPL570 platform. Additionally, we retrieved 162 HBV-related genes from the KEGG pathway database (hsa05161). Differential expressed and functional analysis To identify HBV-related genes differentially expressed between DLBCL and normal samples in GSE56315, we employed the R package “limma”. Genes were identified as differentially expressed if they met the thresholds of a false discovery rate (FDR) < 0.05 and an absolute log2 fold change |(FC)| > 1. Subsequently, enrichment analyses for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) terms were performed on these HBV-related differentially expressed genes (DEGs) using the “clusterProfiler” package in R. Statistical inference was conducted using a critical threshold of 0.05 for adjusted P values. Development of the HBV scoring system We constructed a prognostic model through a systematic approach. First, we identified the intersection between DEGs and genes present in GSE10846. Using expression data from GSE10846, we performed univariate Cox regression analysis to assess the prognostic value of the dysregulated HBV genes. The prognostic gene set was further refined using the stepwise Akaike information criterion (stepAIC) method implemented in the MASS package. For each patient, we calculated a risk score using the following formula, which incorporated normalized expression levels (Expi) and corresponding regression coefficients (Coei) of the candidate genes: DLBCL patient stratification into high- and low-risk groups was performed using median risk scores as the threshold. Afterwards, the prognostic value of our gene signature was evaluated through receiver operating characteristic (ROC) curves and Kaplan-Meier survival analysis, implemented through the “survminer,” “survival,” and “survivalROC” packages We assessed the signature’s independent prognostic utility through univariate and multivariate Cox regression analyses, incorporating relevant clinical parameters. Based on these findings, we developed an integrated nomogram combining molecular and clinical factors, which underwent validation through calibration curves and decision curve analysis (DCA). Analysis of tumor microenvironment features Immune landscape characterization was conducted through dual computational approaches. Single-sample gene set enrichment analysis (ssGSEA) was applied to estimate the abundance of 28 immune cell types, referencing marker genes identified by Charoentong et al. Additionally, the relative proportions of 22 types of immune cells were estimated by the CIBERSORT R package. To further evaluate TME characteristics, we calculated immune, stromal, and ESTIMATE scores for each sample using the ‘ESTIMATE’ R package. We then conducted differential gene expression analysis focusing on immune checkpoint and HLA-related genes to evaluate immune infiltration patterns. The therapeutic response potential was assessed through tumor immune dysfunction and exclusion (TIDE) scoring to predict immunotherapy outcomes. Stemness signatures analysis We evaluated stemness characteristics using 26 curated gene sets obtained from StemChecker (http://stemchecker.sysbiolab.eu/), a comprehensive database incorporating stemness signatures derived from multiple sources including transcription factor (TF) target gene sets, RNAi screens, gene expression profiles, computational summaries, and literature reviews. We quantified the enrichment of 26 stemness gene signatures across DLBCL samples via ssGSEA implementation in GSVA, then conducted differential expression analysis between groups. Statistical analysis Statistical analyses were conducted using R v4.3.1. Non-parametric tests (Wilcoxon and Kruskal-Wallis) were applied for group comparisons, while survival outcomes were evaluated using Kaplan-Meier estimates with log-rank tests. Throughout all analyses, we considered P values less than 0.05 as statistically significant. Statistical significance was defined at four levels: p < 0.05, 0.01, 0.001, and 0.0001, denoted by increasing asterisk numbers. Results HBV-related gene expression profiles in DLBCL Comparative analysis between DLBCL and normal samples revealed 72 differentially expressed HBV-related genes with distinct transcriptomic signatures (Fig. 1A). The expression patterns of these DEGs across the GEO cohort are illustrated in Fig. 1B. Analysis of protein interaction networks highlighted five principal regulators-TP53, MAPK3, TNF, STAT3, and JUN-orchestrating HBV-related pathological mechanisms (Fig. 1C). KEGG pathway enrichment analysis demonstrated significant enrichment in Hepatitis B, Lipid and atherosclerosis, Toll-like receptor signaling, Epstein-Barr virus infection, and HIV-1 infection pathways (Fig. 1D). Gene Ontology biological process (BP) analysis revealed predominant enrichment in cytokine-mediated signaling pathways, stress-activated MAPK cascade, immune response regulation, stress-activated protein kinase signaling, and cellular response to abiotic stimulus (Fig. 1E). (A) HBV differential expression genes (DEGs) between the DLBCL and Normal groups. (B) Expressions of the 72 DEGs in the GEO cohort. (C) Five potential hub genes may play important roles in the pathological process of HBV. (D) Major enriched pathways of DEGs indicated by KEGG enrichment analyses. (E) Major enriched pathways of DEGs in biological process (BP) analysis Construction of the HBV related prognostic signature in DLBCL Univariate Cox regression analysis of the GSE10846 dataset identified 28 HBV-related genes with prognostic value (p < 0.05). Subsequent stepAIC analysis refined this to a 13-gene prognostic model (Fig. 2A), defined by the equation: Risk score = (-0.2602) * MMP9 + (-0.1934) * STAT1 + (0.1878) * TP53 + (0.4237) * MAPK8 + (-0.1747) * CCNE2 + (-0.3189) * PRKCA + (-0.2040) * AKT3+ (0.2791) * CREB3L2 + (0.2067) * MAPK12 + (0.2422) * TLR4 + (0.3227) * TGFB1 + (-0.1733) * EGR3+ (0.2409) * CCNE1. We used the median risk score as a cutoff point to divide patients into low-risk (n = 208) and high-risk (n = 209) groups. The GSE10846 cohort showed longer median overall survival (OS) in the low-risk group (86.1 vs. 42.3 months in high-risk; p < 0.0001, Fig. 2B). This pattern was confirmed in GSE87371 (45.5 vs. 37.4 months; p = 0.00014, Fig. 2C). Predictive performance was demonstrated by AUC values at 3 and 5 years: 0.753 and 0.745 in GSE10846 (Fig. 2D), and 0.599 and 0.735 in GSE87371 (Fig. 2E), respectively. Survival status and risk score distribution for both cohorts are presented in Fig. 2F-G. (A) 13 HBV-related genes with coefficients to construct the prognostic model. (B) Patient numbers at high and low risk groups and their overall survival (OS) rate in the GSE10846 cohort. (C) Patient numbers at high and low risk groups and their overall survival (OS) rate in the GSE87371 cohort. (D) AUC of the ROC curve for 3-, and 5-year survivals in the GSE10846 cohort. (E) AUC of the ROC curve for 3-, and 5-year survivals in the GSE87371 cohort. (F) Distribution of risk scores and survival status for the GSE10846 cohort. (G) Distribution of risk scores and survival status for the GSE87371 cohort Nomogram construction and validation Univariate Cox regression identified gender, age, subtype, stage, ECOG, LDH, and risk score as significant prognostic factors. Multivariate analysis confirmed age (HR: 1.030, 95% CI: 1.013–1.046, p<0.001), stage (HR: 1.251, 95% CI: 1.032–1.516, p = 0.022), ECOG (HR: 1.457, 95% CI: 1.188–1.786, p<0.001), LDH-ratio (HR: 1.100, 95% CI: 1.032–1.173, p = 0.004), and risk score (HR: 2.143, 95% CI: 1.656–2.774, p < 0.001) as independent prognostic indicators (Fig. 3A-B). An integrated prognostic model was constructed as a nomogram to estimate survival probabilities at 1, 3, and 5 years (Fig. 3C). This comprehensive model demonstrated improved risk discrimination compared to the gene-only signature (p < 0.001, Fig. 3D), with robust predictive performance (3-year AUC: 0.829; 5-year AUC: 0.814) (Fig. 3E). Calibration curves confirmed the nomogram’s predictive accuracy (Fig. 3F), while DCA analysis demonstrated its clinical utility across different survival time points (Fig. 3G). These results validate the nomogram’s robust predictive power and practical clinical applicability for DLBCL prognosis assessment. (A) Factors indicated by univariate Cox regression analysis which associated with OS in patients with DLBCL. (B) Factors indicated by multivariate Cox regression analysis which associated with OS in patients with DLBCL. (C) Nomogram developed to predict survival at 1-, 3- and 5-year for patients with DLBCL based on clinicopathological features and HBV -related signature. (D) Prognosis calculated for the low- and high-risk groups by nomogram model. (E) AUC of 1-, 3- and 5-year survival by nomogram model. (F) Calibration curve demonstrated the predictive accuracy of nomogram model. (G) 1-, 3- and 5-year survival compared by decision curve analysis and nomogram model Immune characteristics of HBV-related prognostic subgroups We explored the association between the HBV-related prognostic signature and immune status within the GSE10846 cohort. By applying the ESTIMATE algorithm to assess the immune microenvironment, we observed that low-risk patients had higher stromal (p < 0.001), ESTIMATE (p < 0.001), and immune scores (p = 0.2), alongside reduced tumor purity (p < 0.001) (Fig. 4A). These findings suggest an association between the risk model and immunosuppression. Analysis of immune checkpoint gene expression revealed distinct patterns between risk groups (Fig. 4B). High-risk patients showed elevated expression of CD274 and PDCD1, while CD28 and NRP1 were predominantly expressed in the low-risk group. Additionally, most HLA-related genes demonstrated notably higher expression in low-risk patients (Fig. 4C). A comprehensive immune microenvironment evaluation identified 15 differentially enriched immune cell populations between risk groups. Low-risk patients showed higher enrichment scores for multiple immune cell populations, including NK cells, activated CD8 T cells, and immature B cells. Conversely, high-risk patients displayed higher levels of monocyte and activated dendritic cells (Fig. 4D). Notably, gamma delta T cells and regulatory T cells (Tregs) were significantly reduced in high-risk patients, concurrent with increased M2 macrophage (Fig. 4E). The TIDE score was significantly lower in low-risk patients, suggesting enhanced potential responsiveness to immunotherapy (Fig. 4F). Analysis of cancer stem cell characteristics revealed that 22 of 26 stem cell gene sets showed higher enrichment scores in the low-risk group (Fig. 4G), implying distinct stemness patterns between risk groups. (A) Immune score, stromal score, ESTIMATE score and tumor purity in high- and low-risk groups of HBV-relevant risk score. (B) Immune checkpoint expressions in high- and low-risk groups. (C) Expressions of human leukocyte antigens in high- and low-risk groups. (D) Enrichment scores of infiltrated immune cells in high- and low-risk groups. (E) Composition of tumor microenvironment cells in high- and low-risk groups. (F) Tumor immune dysfunction and exclusion (TIDE) score in high- and low-risk groups. (G) Enrichment scores of stem cell gene sets in in high- and low-risk groups Biological functions associated with HBV-related prognostic signature Differential expression analysis between risk groups identified 227 significantly altered genes (Fig. 5A). GO and KEGG pathway analyses revealed enrichment in multiple biological processes, including negative regulation of phosphorylation, signaling receptor ligand precursor processing, immune system regulation, blood circulation, NF-kappaB transcription factor activity, interleukin-6 production, extracellular matrix organization, DNA-binding transcription factor activity, immune effector processes, and tumor necrosis factor superfamily cytokine production (Fig. 5B). GSVA analysis of MSigDB gene sets demonstrated distinct pathway enrichment patterns between risk groups. DNA damage response, p53-mediated signal transduction, HPV E6/Notch signaling, macrophage and natural killer cell proliferation, and T-helper 2 cell differentiation pathways were significantly enriched in low-risk patients. Conversely, IL10/Jak-STAT signaling, DNA catabolic process, and meiotic cell cycle phase transition were enriched in the high-risk group (Fig. 5C). (A) Volcano plot of differential expression genes (DEGs). (B) Biological functions of differential expression genes (DEGs) indicated by GO/KEGG analysis. (C) Biological functions in high- and low-risk groups by GSVA analysis Discussion DLBCL remains the most prevalent lymphoma entity, and while treatment advances have improved survival rates, significant challenges persist. Despite the availability of standardized treatments, relapsed/refractory DLBCL patients who undergo salvage chemotherapy with autologous hematopoietic stem cell transplantation face discouragingly low cure rates of approximately 10%. Recently, chimeric antigen receptor T (CAR-T) cell therapy has offered durable remission for some patients, yet high costs, lengthy preparation, and adverse reactions hinder its use [18]. Similarly, novel antibody-based therapies, such as antibody-drug conjugates, monoclonal, and bispecific antibodies, face implementation barriers due to their substantial costs. Emerging evidence has established a compelling biological connection between HBV and DLBCL [9, 19]. Persistent HBV infection and genomic integration stimulate chronic inflammatory processes, fostering genetic and epigenetic modifications within key oncogenes-CREBBP, MYC, and BCL-6 [20–22]. This process involves multiple signaling pathways, particularly NF-κB, JAK/STAT3, and BCR pathways, establishing HBV as a major risk factor for DLBCL development. Notably, HBV-associated DLBCL patients typically present with advanced disease stages, demonstrate poor responses to first-line treatments, and experience worse prognoses [23]. The absence of standard first-line immunochemotherapy protocols underscores the urgent need for refined clinical prognostic models to guide personalized therapeutic approaches. Our study identified 72 HBV-related DEGs between normal tissues and DLBCL, revealing distinct transcriptomic signatures. Through comprehensive GO and KEGG enrichment analyses, we elucidated the regulatory mechanisms underlying these DEGs. Univariate Cox regression analysis, followed by StepAIC optimization, yielded a robust 13-gene prognostic model. Several components of our risk score model have been independently validated in previous studies, particularly STAT1, TP53, and MAPK8. TP53, a critical tumor suppressor, plays a fundamental role in maintaining aggressive cancer phenotypes, with mutations correlating with poor survival rates and suboptimal responses to standard R-CHOP therapy in DLBCL patients [24–26]. Similarly, STAT1—a key mediator of the JAK-STAT pathway—has been reported to be upregulated in DLBCL [27]. The prognostic model also includes matrix metalloproteinases (MMP), with MMP-9 being of particular interest. MMP-9 is a zinc-dependent proteolytic metalloenzyme involved in extracellular matrix degradation and other pathophysiological processes; its dysregulation has been observed in several malignancies, including lymphoma, where inhibition of MMP-9 activity represents a potential therapeutic strategy [28–31]. The landscape of prognostic models for DLBCL risk stratification has evolved significantly since the introduction of the International Prognostic Index (IPI) in 1993, which remains the primary instrument for patient selection and risk assessment in clinical trials [32]. However, contemporary treatment regimens have diminished the IPI’s prognostic significance, necessitating more refined prognostic tools. Comprehensive retrospective registry analyses comparing thirteen prognostic models revealed superior performance of the NCCN-IPI and DLBCL-PI, while highlighting limitations in the aaIPI’s discriminative ability [33–35]. Laboratory-based models, while promising, require thorough validation given the substantial influence of clinical parameters on disease outcomes. Our study addresses these limitations by developing an integrated prognostic model that combines patient age and disease staging with a novel HBV-related gene signature. This approach demonstrated superior prognostic accuracy compared to existing models. The model’s robust performance was validated across training, testing, and validation datasets, with consistent results in our clinical cohort. Risk stratification based on the median signature value effectively distinguished between high-risk and low-risk groups, as confirmed by Kaplan–Meier analysis. The derived nomogram for predicting 1-, 3-, and 5-year survival, integrating both HBV-related signature and clinicopathological features, showed enhanced prognostic performance compared to the gene signature alone. The lymphoma TME, consisting of immune cells, adipocytes, fibroblasts, and endothelial cells, plays a fundamental role in tumor development and metastatic progression. Recent research has demonstrated increased M2 macrophage infiltration and reduced T-cell presence in high-risk DLBCL patients [36]. Our findings align with and extend these observations, revealing significant reductions in HBV-related gene expression, gamma delta T cells, and Tregs in high-risk patients, concurrent with increased M2 macrophage infiltration. These results suggest that therapeutic strategies targeting M2 macrophage polarization might effectively inhibit DLBCL progression. In conclusion, our study makes several important contributions to the field. First, we have developed a prognostic signature that not only predicts survival outcomes but also provides insights into the immunological status of DLBCL patients. Second, our integrated model offers improved prognostic accuracy compared to existing tools, potentially enabling more precise risk stratification. Finally, the correlation between our risk scores and specific immune cell populations suggests potential therapeutic targets for personalized treatment approaches. However, we acknowledge certain limitations. While our findings show promise for improving patient outcomes and personalizing therapeutic strategies based on risk scores, further experimental and clinical validation studies are necessary. Acknowledgements We thank Siyuan Zhao and Long Hu for their hard work and efforts. Abbreviations Diffuse large B-cell lymphoma Hepatitis B virus Tumor immune microenvironment Gene Expression Omnibus The Cancer Genome Atlas Differentially Expressed Gene Gene Ontology Kyoto Encyclopedia of Genes and Genomes Stepwise Akaike information criterion Decision curve analysis Gene set variation analysis Single-sample gene set enrichment analysis Transcription factor Corresponding regression coefficients Biological process Overall survival Tumor immune dysfunction and exclusion Chimeric antigen receptor T Matrix metalloproteinases Extracellular matrix International Prognostic Index Author contributions Guanhua Zhao: conceived and designed the experiments, performed the experiments, analyzed the data, prepared figures and tables, authored or reviewed drafts of the article, and approved the final draft. Sheng Gao: conceived and designed the experiments, performed the experiments, analyzed the data, prepared figures and tables, and approved the final draft. Yalu Liu: conceived and designed the experiments, performed the experiments, analyzed the data, prepared figures and tables, authored or reviewed drafts of the article, and approved the final draft. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability The datasets used in this study are available in the following public repositories: TCGA database (https://portal.gdc.cancer.gov) and the GEO database (https://www.ncbi.nlm.nih.gov, GSE10846, GSE56315, and GSE87371). Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Guanhua Zhao and Sheng Gao contributed equally to this work. References Associated Data Data Availability Statement The datasets used in this study are available in the following public repositories: TCGA database (https://portal.gdc.cancer.gov) and the GEO database (https://www.ncbi.nlm.nih.gov, GSE10846, GSE56315, and GSE87371).