Precision medicine in diffuse large B-cell lymphoma: Integrating molecular biomarkers, targeted therapies, and prognostic tools

Abstract

Diffuse large B-cell lymphoma (DLBCL) has undergone a profound transformation in its therapeutic landscape over the past decade, driven by advances in molecular characterization and the emergence of targeted immunotherapies. Although standard immunochemotherapy with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone achieves high initial remission rates, approximately 30%-40% of patients experience refractory or relapsed disease, underscoring the urgent need for refined risk stratification and innovative treatment approaches. This review synthesizes the current state of precision medicine in DLBCL, emphasizing the integration of molecular biomarkers—including genetic subtypes, tumor microenvironment features, and circulating biomarkers such as circulating tumor DNA—into clinical decision-making. We critically evaluate emerging targeted therapies, including Bruton tyrosine kinase inhibitors, BCL-2 inhibitors, antibody-drug conjugates, and bispecific T-cell engagers (BiTEs), alongside the expanding role of chimeric antigen receptor T-cell therapy. Central to this discussion is the development of robust prognostic models designed to identify high-risk patients most likely to benefit from novel agents, highlighting recent advances in nomogram-based systems derived from Chinese cohorts. Ultimately, we propose that the future of DLBCL management lies not merely in the availability of novel agents, but in the rational sequencing and combination of these modalities based on dynamic molecular monitoring and individualized risk assessment—translating molecular insights into actionable clinical pathways.

Key Words: Diffuse large B-cell lymphoma; Precision medicine; Molecular biomarkers; Targeted therapy; Prognostic models; ctDNA; Chimeric antigen receptor T-cell therapy; Genetic subtyping; Tumor microenvironment; Immunochemotherapy

Core Tip: Diffuse large B-cell lymphoma (DLBCL) is a biologically heterogeneous disease requiring precision medicine approaches beyond standard immunochemotherapy with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone. This review synthesizes current evidence on integrating molecular subtyping (cell-of-origin and genetic subtypes), circulating tumor DNA monitoring, and tumor microenvironment characterization into clinical decision-making. We critically evaluate emerging targeted therapies including Bruton tyrosine kinase inhibitors, BCL-2 inhibitors, antibody-drug conjugates, bispecific T-cell engagers, and chimeric antigen receptor T-cell therapy. Prognostic models have evolved from the International Prognostic Index to dynamic, integrative nomograms incorporating molecular biomarkers. Challenges in standardization, cost-effectiveness, and equitable access are discussed alongside future directions for personalized DLBCL management.

INTRODUCTION

Diffuse large B-cell lymphoma (DLBCL) is the most common type of non-Hodgkin lymphoma in adults. Over the past decade, substantial progress has been made in elucidating the biological heterogeneity of DLBCL. Rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) is a relatively old but highly effective chemotherapy regimen that remains widely used in clinical practice, and it is curative for approximately two-thirds of patients. However, the median survival of patients with primary refractory disease is less than six months[1-3]. Clinically, two patients may present with the same disease, at the same stage, and with the same International Prognostic Index (IPI) score; yet one may recover normally, whereas the other may fail to achieve a favorable response. Several factors may account for these differences, and such observations have laid the foundation for research on DLBCL over the past decade.

Biological heterogeneity in DLBCL is reflected in the identification of various molecular subtypes by gene expression profiling (GEP), such as the activated B-cell-like (ABC) and germinal center B-cell-like (GCB) subtypes, which differ in cell of origin and clinical outcome[4,5]. However, these broad categories do not fully capture the diverse manifestations of the disease. More recently, genomic studies have identified additional categories of DLBCL based on recurrent somatic mutations, copy number alterations, and structural variations. For example, LymphGen is a molecular classifier that divides DLBCL cases into genetic clusters with distinct oncogenic drivers and therapeutic vulnerabilities, thereby offering new avenues for understanding disease biology[6]. These advances suggest that molecular diagnostics will soon be incorporated into the general workflow for personalized medicine.

The tumor microenvironment (TME) is known to modulate DLBCL development and response to treatment. Immune cell infiltration, stromal components, and immune checkpoint expression within the TME have been shown to influence prognosis and treatment response[7]. Increasingly, immune cells in the vicinity of lymphoma are no longer regarded as passive bystanders; rather, they actively participate in disease progression and influence treatment efficacy, making them attractive therapeutic targets. Liquid biopsy, particularly the analysis of circulating tumor DNA (ctDNA), has recently emerged as a noninvasive approach for studying disease indicators and the molecular characteristics of DLBCL. A single blood test can simultaneously detect tumor-associated mutations and assess the impact of treatment on those mutations[8,9].

The therapeutic landscape of DLBCL is evolving rapidly and moving beyond R-CHOP-based therapy. Targeted agents such as Bruton tyrosine kinase (BTK) inhibitors and BCL-2 antagonists have demonstrated efficacy in molecularly defined subgroups, including the MYD88 (L265P)/CD79B-mutated (MCD) subtype and non-GCB DLBCL. CD19-targeted chimeric antigen receptor T-cell (CAR-T) therapy has achieved favorable clinical outcomes in relapsed/refractory DLBCL and is now clinically available; however, challenges such as toxicity, high cost, and limited patient eligibility remain to be addressed. Newer immunotherapeutic approaches, such as bispecific antibodies targeting CD20 and CD3, have expanded the therapeutic armamentarium and are now being evaluated in earlier lines of treatment. Prognostic tools have evolved from the IPI to incorporate additional factors, including molecular and immune-microenvironment data. Nomogram models and immune-related gene classifiers have shown improved predictive performance for survival, thereby enabling individualized risk assessment.

This review aims to synthesize the key aspects of precision medicine in DLBCL and to examine its current status, ongoing challenges, and future prospects for integrating molecular biomarkers, targeted therapy, and prognostic indicators. We have included a dedicated section on human immunodeficiency virus (HIV)-associated DLBCL, which, although systematically excluded from most pivotal clinical trials, represents a clinically distinct entity. We also present a representative computational model and, based on the above evidence, propose a conceptual three-stage management pathway.

This is a narrative review based on a comprehensive literature search of the PubMed, Web of Science, and Scopus databases for papers published between January 2000 and December 2025. Search terms included “diffuse large B-cell lymphoma”, “molecular subtyping”, “tumor microenvironment”, “circulating tumor DNA”, “targeted therapy”, “CAR-T”, “bispecific antibodies”, and “prognostic models”. Priority was given to landmark genomic studies, influential translational research, and pivotal clinical trials that have substantially advanced the biological understanding and clinical management of DLBCL.

MOLECULAR HETEROGENEITY AND PRECISE CLASSIFICATION OF DLBCL

Cell-of-origin classification and its clinical relevance

For the past two decades, the first molecular subtypes of DLBCL have been classifications based on cell of origin (COO). Studies using gene expression profiling (GEP) to divide DLBCL into two types have identified the GCB and ABC subtypes, which differ in cell of origin, signaling requirements, and clinical outcomes[4,5]. The ABC subtype is associated with a worse prognosis. Constitutive activation of nuclear factor kappa B (NF-κB) is associated with shorter progression-free survival (PFS) and overall survival (OS) in patients treated with R-CHOP. In contrast, the GCB subtype generally has a favorable prognosis. Hans and colleagues developed a practical immunohistochemistry (IHC)-based algorithm using CD10, BCL6, and MUM1. Although the accuracy of this algorithm is approximately 80% and therefore not optimal, it is economically feasible and easy to implement[10].

The COO classification also has limitations. ABC-DLBCL is associated with chronic B-cell receptor (BCR) signaling and NF-κB activation, and thus with reduced sensitivity to chemotherapy[11]. GCB-DLBCL is linked to BCL2 translocations and mutations in epigenetic modifiers. Consequently, the two subtypes have distinct therapeutic vulnerabilities. However, relapse or refractoriness cannot be predicted by COO to the same extent. Indeed, double-hit status and the double-expressor phenotype are more prominent in the relapsed state[11]. COO remains suitable for first-level risk assessment, but guiding treatment after relapse requires further investigation.

The prognostic value of COO classification in the relapsed or refractory (R/R) setting is relatively limited. Although COO has been identified by both IHC and GEP, neither approach has been associated with improved OS after salvage therapy with high-dose chemotherapy and autologous stem cell transplantation. Other molecular features with stronger prognostic value in this context include double-hit lymphoma status and the double-expressor lymphoma phenotype[12].

Mutation-based genetic subtype systems

Next-generation sequencing (NGS) has identified several distinct genetic subtypes that enhance the prognostic value of COO classification and reveal additional treatment targets. Schmitz and colleagues identified four genetic subgroups—MCD, BN2, N1, and EZB—each characterized by distinct mutations and translocations[13]. Chapuy et al[14] independently constructed a five-cluster model using whole-exome sequencing. Wright and colleagues developed LymphGen, a probabilistic classifier that has been widely used in clinical research to date[6]. More recently, Chapuy et al[15] proposed DLBClass, which further improves accuracy through a neural network–based approach.

Principal genetic subgroups: The MCD subtype is classified on the basis of whether MYD88 L265P and CD79B mutations occur together; these mutations activate BCR and Toll-like receptor signaling pathways, leading to prolonged NF-κB activation, and the subtype is therefore dependent on BTK for survival. Both preclinical and clinical studies have demonstrated sensitivity to BTK inhibitors in this context[14,16]. A retrospective analysis of the PHOENIX trial revealed that patients with MCD-subtype disease also benefited from the addition of ibrutinib to R-CHOP[17].

The BN2 subtype is associated with BCL6 rearrangements and NOTCH2 mutations. NOTCH signaling is weakly active and is thus associated with an intermediate prognosis. Notably, BN2 tumors respond to BET inhibitors because their survival depends on BET-mediated transcriptional programs, and disruption of this pathway inhibits growth[14].

The EZB subtype is associated with EZH2 mutations and BCL2 translocations. Mutations in EZH2 cause H3K27me3 hypermethylation, suppress tumor suppressor genes, and inhibit differentiation, whereas BCL2 translocation promotes cell survival. Most patients with EZB have a favorable prognosis; however, high BCL2 expression may contribute to acquired resistance to therapy.

The N1 subtype is caused by mutations in NOTCH1 that constitutively activate NOTCH signaling; this subtype has a poor prognosis and is less responsive to chemotherapy[13].

The A53 subtype harbors TP53 mutations together with numerous complex chromosomal abnormalities and is regarded as a relatively poor-prognosis subtype. TP53 mutations occur more frequently in patients with relapsed or resistant disease than in newly diagnosed patients, are closely associated with genomic instability, and contribute to treatment resistance[14]. Although no targeted therapies for TP53-mutant DLBCL have yet been approved, p53-restoring compounds such as APR-246 are being investigated in early-phase clinical trials[18].

Table 1 summarizes the main attributes of all the classification systems. Although these systems share some similarities, they are not identical. Schmitz’ classification has a strong biological foundation and therapeutic potential. Chapuy’s unsupervised method identified several previously unrecognized groups[14,15]. LymphGen is reproducible and readily available, whereas DLBClass represents a typical deep learning approach in this field. Genetic subtyping has expanded COO classification from two categories to five or more biologically and therapeutically significant groups in clinical practice. However, several challenges remain in the clinical application of genetic subtyping: Sequencing is still relatively expensive, and the classification algorithms have not been standardized.

Table 1 Comparison of diffuse large B-cell lymphoma molecular classification systems.

Classification system	Year	Technology platform	Subtypes	Classification basis	Major advantages	Major limitations
COO classification	2000	GEP	2 (GCB/ABC)	Gene expression profiling	Biologically well-defined; established the foundation for molecular subtyping	GEP-dependent; difficult for routine clinical implementation
Hans et al[10], algorithm	2004	IHC	2 (GCB/non-GCB)	CD10/BCL6/MUM1	Simple operation; applicable in routine pathology	Accuracy approximately 80%; low sensitivity for ABC subtype identification
Schmitz et al[13], classification	2018	WES + translocations	4 (MCD/BN2/N1/EZB)	Mutations + chromosomal translocations	Mechanistically clear; provides explicit therapeutic guidance	WES-dependent; high cost
Chapuy et al[14], classification	2018	WES clustering	5 (C1-C5)	Whole-exome clustering	Independently validated; reveals additional subgroups	Clinical significance of some subtypes unclear
LymphGen	2020	Probabilistic classification	7	Naïve Bayes algorithm	Reproducible probabilistic classification; most widely applied	Relies on WES data quality
DLBClass	2025	Deep learning	Probabilistic	Neural network	Higher accuracy	High technical barrier; challenging for clinical implementation

COO: Cell-of-origin; GEP: Gene expression profiling; GCB: Germinal center B-cell-like; ABC: Activated B-cell-like; IHC: Immunohistochemistry; WES: Whole-exome sequencing; MCD: MYD88 (L265P)/CD79B-mutated subtype; BN2: BCL6 fusion/NOTCH2-mutated subtype; N1: NOTCH1-mutated subtype; EZB: EZH2-mutated/BCL2-translocated subtype; C1-C5: Clusters 1-5 in Chapuy classification; LymphGen: A probabilistic classification tool for genetic subtypes of DLBCL; DLBClass: A neural network-based molecular classifier for DLBCL.

Immune landscape of the TME

The TME of DLBCL is highly complex, and malignant B cells, immune cells, stromal cells, extracellular matrix (ECM), and other components collectively contribute to disease development, immune evasion, and treatment resistance[7]. Central to the TME is the composition and function of tumor-infiltrating lymphocytes (TILs), including CD8+ cytotoxic T cells, regulatory T cells, and natural killer (NK) cells. CD8+ cytotoxic T lymphocytes are the main effectors, and a higher density of CD8+ TILs generally indicates a better prognosis[19]. Regulatory T cells expressing TIM-3 and programmed death 1 (PD-1) suppress antitumor immunity through the release of IL-10 and other cytokines. NK cells remain relatively understudied; when NK cells fail to function normally, tumors can evade the immune system[20]. The balance between effector and regulatory T cells ultimately determines the strength of the immune response.

Macrophages exhibit remarkable plasticity, with M1 (antitumorigenic) and M2 (protumorigenic) polarization states. In DLBCL, M2-like tumor-associated macrophages (TAMs) are associated with poor prognosis, as they secrete immunosuppressive cytokines, promote angiogenesis, and remodel the ECM. CSF1R is a key pathway for macrophage recruitment and polarization and is being explored as a novel therapeutic target. Stromal components—specifically, fibroblastic reticular cells—are reprogrammed by tumors, upregulate FAP expression, alter chemokine secretion, impair TIL migration, and suppress CD8+ cell cytotoxicity. The net result is an immunosuppressive niche that supports lymphoma survival and impairs immunotherapy efficacy. Collagen remodeling of the ECM increases stiffness and alters mechanotransduction, thereby promoting survival and drug resistance[21].

PD-1/programmed death ligand-1 (PD-L1), LAG-3, and TIM-3 are immune checkpoint molecules that contribute to immune evasion in DLBCL. High PD-L1 expression in tumor cells and TAMs is associated with immunosuppression and poor prognosis in patients with Epstein-Barr virus (EBV)-positive DLBCL[22]. Coexpression of multiple checkpoint molecules on TILs indicates an exhausted T-cell phenotype and is associated with poor prognosis[20]. Therapeutic blockade of these checkpoints has shown promise in restoring T-cell function, and various combinations are currently being explored[22,23].

Recent advances in spatial transcriptomics and single-cell RNA sequencing have revealed that immune cells in the DLBCL TME are heterogeneous and spatially organized, leading to the identification of recurrent cellular neighborhoods with distinct immune cell compositions and interactions that affect patient prognosis. For instance, the proximity of CD8+ T-cell-rich areas to immune-poor regions is associated with a good prognosis, whereas close interactions between PD-L1+ B cells and CD8+ T cells predict a poor prognosis[24]. Extracellular vesicles in the TME carry molecular cargo that regulates the immune system and can serve as liquid biopsy markers. Metabolic reprogramming—glycolysis and lactate production—also affects immune cells. Immune-infiltrated tumors contain numerous immune cells and are generally associated with a good prognosis; however, immune-desert and immune-excluded tumor types do not share the same favorable outcomes.

In summary, the TME of DLBCL reflects the balance between effector and suppressive factors. Many agents now target PD-1/PD-L1, CSF1R, and FAP, but their effectiveness varies greatly because of the high heterogeneity of the TME among patients[21,23]. Recently, the application of TME features in patient stratification has begun to emerge.

CIRCULATING BIOMARKERS: APPLICATIONS OF LIQUID BIOPSY IN DLBCL

ctDNA as a tool for tumor burden and dynamic monitoring

ctDNA has emerged as a novel biomarker for DLBCL that can be used to assess tumor burden and disease changes during treatment noninvasively. There are two main types of ctDNA analysis: (1) Tumor-based sequencing, which involves sequencing tissue samples to identify patient-specific single-nucleotide variants (SNVs) and determining whether these SNVs have changed in the blood; and (2) Immunoglobulin heavy chain V(D)J rearrangement analysis, which is particularly suitable for DLBCL because of its characteristic clonal pattern of B cells[8,9]. The clinical application pathway for initial diagnosis, end-of-treatment (EOT), and positron emission tomography/computed tomography (PET/CT) is shown in Figure 1.

Figure 1 Clinical application pathway of circulating tumor DNA in diffuse large B-cell lymphoma. This figure illustrates the clinical application of circulating tumor DNA (ctDNA) across the disease continuum of diffuse large B-cell lymphoma. Baseline assessment: At diagnosis, tissue biopsy for genetic profiling and blood draw for baseline ctDNA are performed. Baseline ctDNA levels correlate with tumor burden, International Prognostic Index, lactate dehydrogenase, and maximum standardized uptake value. High baseline ctDNA predicts inferior progression-free survival and overall survival [pooled hazard ratio (HR) = 2.50, 95% confidence interval: 2.15-2.90]. On-treatment monitoring: Early molecular response (≥ 2-log reduction after cycle 1) and major molecular response (≥ 2.5-log reduction after cycle 2) predict favorable event-free survival. Persistent ctDNA positivity is associated with a high risk of progression (HR = 4.0). End-of-treatment (EOT) assessment: Integration of EOT positron emission tomography/computed tomography (PET/CT) and ctDNA results enables four-quadrant risk stratification: PET-/ctDNA- (dual remission) → routine follow-up; PET-/ctDNA+ (high-risk subgroup, specificity 90.8%) → consider consolidation therapy; PET+/ctDNA- (possible inflammatory lesions, negative likelihood ratio = 0.15) → avoid unnecessary biopsy; PET+/ctDNA+ (true residual disease) → intensified therapy. Post-treatment surveillance: Serial ctDNA monitoring detects molecular relapse 3-6 months before radiographic progression and identifies resistance mutations (e.g., BCL2 G101V, CD79B Y197*/E196*), enabling preemptive intervention with chimeric antigen receptor T-cell therapy, bispecific antibodies, or novel targeted agents. Timeline: Diagnosis → cycle 1-2 → cycle 3-6 → EOT (PET/CT) → follow-up (every 3-6 months). Figure 1 was created by the authors using Microsoft PowerPoint. No third-party copyrighted materials were used. R-CHOP: Rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone; pola-R-CHP: Polatuzumab vedotin combined with rituximab, cyclophosphamide, doxorubicin, and prednisolone; ctDNA: Circulating tumor DNA; PET: Positron emission tomography; SUVmax: Maximum standardized uptake value; IPI: Prognostic index; LDH: Lactate dehydrogenase; EMR: Early molecular response; MMR: Major molecular response; MRD: Minimal residual disease; EOT: End-of-treatment; CAR-T: Chimeric antigen receptor T-cell.

Baseline ctDNA levels are related to tumor size and PET/CT features such as the maximum standardized uptake value. The baseline concentration is an independent prognostic factor for PFS and OS that is not explained by the IPI or lactate dehydrogenase (LDH) level[25]. A meta-analysis of 53 studies revealed that a high baseline ctDNA concentration was associated with an increased risk of disease progression [pooled hazard ratio (HR) = 2.50, 95%CI: 2.15-2.90]; however, these results are mainly based on observational cohorts and require prospective validation[25].

Dynamic ctDNA monitoring during treatment can guide therapeutic adjustments. Patients who show a rapid decrease in ctDNA during the first few cycles have better PFS and OS than those who do not[8,26]. Among patients who achieved a complete metabolic response on PET/CT, ctDNA positivity was still associated with a higher risk of relapse. Therefore, ctDNA is associated with minimal residual disease and occult resistant clones that cannot be identified by imaging alone[8,25]. Early molecular response (≥ 2-log reduction after cycle 1) and major molecular response (≥ 2.5-log reduction after cycle 2) predict better event-free survival. A meta-analysis reported a combined HR of 4.0 for interim ctDNA positivity; the HR for positivity at the end of treatment is significantly greater, at 13.69[25], and ctDNA clearance occurs approximately 97 days before radiographic changes.

Integration of ctDNA with PET/CT enables risk stratification. Among patients with negative EOT PET results, ctDNA positivity had a specificity of 90.8% for subsequent relapse and a positive likelihood ratio of 5.5. Therefore, a patient with a negative PET scan but a positive ctDNA test is at increased risk of recurrence. Moreover, among patients with a positive EOT PET result, a negative ctDNA result had a negative likelihood ratio of 0.15 and was therefore associated with a low risk of relapse; thus, unnecessary consolidation therapy or invasive biopsy could be avoided. Table 2 shows the four-quadrant risk stratification combining EOT ctDNA and PET/CT. A Chinese cohort study using an immunoglobulin heavy chain rearrangement-based decision tree model [area under the curve (AUC) = 0.85] further validated this approach[27]. The above results are based on observational cohorts, and prospective trials are needed to determine the effect of ctDNA-guided treatment.

Table 2 Risk stratification by integrating end-of-treatment circulating tumor DNA and positron emission tomography/computed tomography findings.

EOT PET/CT	EOT ctDNA	Clinical implication	Recommended management
Negative	Negative	Molecular + radiographic dual remission	Routine follow-up
Negative	Positive	High-risk subgroup (specificity 90.8%)	Consider consolidation therapy or intensified surveillance
Positive	Negative	Possible inflammatory or inactive lesions (NLR 0.15)	Avoid unnecessary consolidation therapy/invasive biopsy
Positive	Positive	True residual disease/progression	Intensified therapy

EOT: End-of-treatment; ctDNA: Circulating tumor DNA; PET: Positron emission tomography; CT: Computed tomography; NLR: Negative likelihood ratio.

ctDNA changes reflect the effect of treatment and can predict early relapse in patients with relapsed/refractory DLBCL. Clearance of the baseline mutation during salvage chemotherapy is associated with a good prognosis, whereas the presence or reappearance of ctDNA mutations indicates treatment failure and disease progression[28,29].

ctDNA detection of resistance mutations and molecular relapse

Early treatment resistance or molecular relapse can be detected by ctDNA. Known resistance-associated mutations, such as BCL2 G101V and CD79B Y197*/E196*, can be detected in plasma before clinical or radiographic progression and serve as early warning signals of loss of response to targeted therapy. Mutations in CD79B have been linked to resistance to BTK inhibitors, and BCL2 mutations can confer resistance to BCL2-targeted agents.

Molecular relapse is defined as the reappearance of tumor-specific mutations in ctDNA after initial remission, and it generally occurs approximately 3-6 months before radiographic relapse; this interval therefore provides a therapeutic window for preventive treatment, such as CAR-T-cell therapy or novel targeted agents[25,29]. Ultradeep NGS and digital polymerase chain reaction are highly sensitive for detecting low-frequency mutations and copy number variations. However, standardized protocols for ctDNA processing, sequencing depth, bioinformatics pipelines, and interpretation criteria have not yet been established.

Other circulating biomarkers

Circulating extracellular vesicles: These include exosomes, which contain microRNAs, long noncoding RNAs, and proteins that carry information about tumor development. Specific exosomal miRNA signatures have demonstrated diagnostic and prognostic value, with high diagnostic accuracy and AUC values as high as 0.90[30]. Notably, miR-451a expression was significantly correlated with PFS and OS, and, in combination with the IPI, it improved prognostic stratification[30].

Serum cytokines: Increased levels of proinflammatory cytokines, such as IL-6, IL-10, and soluble CD25, are associated with tumor burden, disease stage, and systemic inflammatory responses[21]. However, these cytokines have low specificity for DLBCL and are also elevated in many other inflammatory and malignant diseases.

Cell-free RNA (cfRNA): cfRNA analysis can detect changes in gene expression and the TME during disease development. cfRNA includes both coding and noncoding transcripts and thus reflects transcriptional and other cellular alterations. However, the development of cfRNA biomarkers still faces several challenges, including RNA instability, low abundance, and technical variation.

THERAPEUTIC STRATEGIES TARGETING THE BCR SIGNALING PATHWAY

Application of BTK inhibitors in DLBCL

BTK inhibitors have emerged as a pivotal class of targeted therapies for DLBCL, particularly in R/R patients and in molecular subtypes with active BCR signaling, such as the ABC subtype. Ibrutinib is a first-generation BTK inhibitor that has shown single-agent efficacy in R/R ABC-DLBCL, with an overall response rate (ORR) of approximately 30%-35% and a complete remission (CR) rate of only approximately 10%-15%[16,31]. However, the phase III PHOENIX trial did not demonstrate a survival advantage for ibrutinib in combination with R-CHOP in the general non-GCB population. Retrospective analysis revealed that patients with MCD/N1 subtype disease improved after the addition of ibrutinib to R-CHOP[17,32]. This modest efficacy and subtype-specific benefit underscore the need for improved therapeutic strategies.

Second-generation BTK inhibitors, such as acalabrutinib, zanubrutinib, and orelabrutinib, offer improved selectivity and reduced off-target toxicity. Zanubrutinib has demonstrated encouraging efficacy in non-GCB DLBCL, with an ORR of approximately 29% and a CR rate of nearly 17% in the R/R setting[31].

In recent years, combinations of BTK inhibitors with immunomodulatory agents and monoclonal antibodies have been developed and used for genetically defined subgroups, such as MCD.

Acquired resistance to BTK inhibitors remains a significant challenge. Mechanisms include mutations in the BTK binding site (notably C481S), mutations in downstream signaling molecules such as PLCG2, and compensatory activation of alternative signaling pathways. To address resistance, novel noncovalent BTK inhibitors and combination regimens, including BTK inhibitors combined with BCL-2 inhibitors, are under investigation[33].

Downstream targets of the BCR signaling pathway

Spleen tyrosine kinase inhibitors: Spleen tyrosine kinase (SYK) is a key proximal kinase in the BCR signaling pathway. SYK inhibitors such as fostamatinib and entospletinib have been shown to inhibit downstream BCR signaling; however, owing to a lack of single-agent efficacy, they have not been widely adopted in clinical practice.

PI3Kδ inhibitors: Idelalisib and copanlisib are PI3Kδ inhibitors that block the AKT/mammalian target of rapamycin pathway. Copanlisib has shown clinical efficacy in relapsed/refractory DLBCL, but its use is limited by toxicities such as hepatotoxicity, as well as an increased risk of infection.

PKCβ inhibitors: Enzastaurin is a selective PKCβ inhibitor that failed to meet its primary endpoint in a phase III trial; this suggests that survival pathways are redundant and that a combination approach is therefore needed.

CD79B as a therapeutic target

Antibody-drug conjugates (ADCs) targeting CD79B have become leading therapies for DLBCL. Polatuzumab vedotin is a CD79B-targeted ADC linked to monomethyl auristatin E (MMAE); it binds to CD79B on lymphoma cells, is internalized, releases MMAE, disrupts microtubule function, and thereby induces apoptosis[34]. A phase III trial, POLARIX, established polatuzumab vedotin combined with rituximab, cyclophosphamide, doxorubicin, and prednisone (pola-R-CHP) as a new frontline option and demonstrated an improvement in 2-year PFS over R-CHOP (76.7% vs 70.2%; HR = 0.73; P = 0.02)[35]. OS at 2 years was similar between the two groups, but the PFS benefit was more pronounced in patients with the ABC subtype and those with an IPI score of 3 or higher. Compared with the control treatment, polatuzumab vedotin combined with bendamustine and rituximab achieved a complete response (CR) rate of approximately 40% in the relapsed/refractory setting and significantly improved PFS and OS[36]. Peripheral neuropathy is a relatively common adverse event and is generally mild to moderate.

The molecular determinants of polatuzumab vedotin response are currently being investigated. Mutations in CD79B, especially those in the ITAM domain, are relatively frequent in MCD-genotype DLBCL. CD79B glycosylation status and its regulation by the E3 ubiquitin ligase KLHL6 also affect polatuzumab vedotin sensitivity[34]. However, low or absent CD79B expression does not preclude a response, suggesting that other factors may influence ADC effectiveness.

STRATEGIES TARGETING BCL-2 AND ANTI-APOPTOTIC PATHWAYS

Venetoclax: Mechanism of action and resistance

Overexpression of BCL-2 confers a survival advantage and therefore represents a therapeutic target. Venetoclax is a BCL-2 inhibitor that increases mitochondrial outer membrane permeability and thereby triggers caspase-dependent apoptosis. In the phase II CAVALLI trial, venetoclax combined with R-CHOP improved the CR rate in BCL-2-positive patients, albeit with increased hematologic toxicity[37]. A distinct subset of DLBCL termed BCL-2 superexpressors exhibits uniform and strong BCL-2 staining, is associated with poor survival, and may be particularly responsive to BCL-2-targeted therapy[38].

Resistance to venetoclax remains a significant clinical challenge. Acquired resistance mechanisms include mutations in the BCL2 gene (notably G101V), compensatory upregulation of other antiapoptotic proteins such as BCL-XL and MCL-1, and hyperphosphorylation of BCL-2 family proteins.

Inhibitors targeting MCL-1 and BCL-XL

MCL-1 inhibitors such as S63845 and AMG176 have shown strong preclinical efficacy in DLBCL cell lines that are dependent on MCL-1 for survival[39]. Navitoclax, a BCL-XL inhibitor, has been limited by dose-limiting thrombocytopenia, because BCL-XL is required for platelet survival[40]. Combinations targeting multiple antiapoptotic proteins show synergistic cytotoxicity but require careful management of overlapping toxicities.

Combined signal pathway inhibition

Combined inhibition of the BCL-2 and BTK or PI3K pathways interrupts survival signaling and activates apoptosis. ViPOR (venetoclax, ibrutinib, prednisone, obinutuzumab, and lenalidomide) resulted in longer remission in heavily pretreated patients and was particularly effective in non-GCB and high-grade B-cell lymphomas[41]. Venetoclax can also be combined with CAR-T cell therapy.

ADC AND BISPECIFIC ANTIBODIES

Polatuzumab vedotin

As discussed in Section 3.3, polatuzumab vedotin is now used as a first-line option (pola-R-CHP) and in the relapsed/refractory setting (Pola-BR)[35,36].

Other novel ADCs

Loncastuximab tesirine, an anti-CD19 ADC, has shown favorable results in heavily pretreated patients, including those who have failed CAR-T cell therapy, with an ORR of approximately 48% and a CR rate of nearly 20%[42]. Adverse events, including edema, rash, and myelosuppression, require careful management. CD22-targeted ADCs, such as inotuzumab ozogamicin, are being evaluated in DLBCL clinical trials.

Developments in bispecific T-cell engagers

Bispecific T-cell engagers (BiTEs) represent a new generation of immune-enhancing agents. Glofitamab, epcoritamab, and mosunetuzumab (all CD20×CD3 bispecific antibodies) have shown favorable clinical efficacy in heavily pretreated DLBCL, with ORRs of approximately 50%-60% and CR rates of approximately 40%[43]. These “off-the-shelf” agents avoid the production delays and access barriers associated with CAR-T cell therapy. The STARGLO trial revealed that, compared with R-GemOx, glofitamab plus GemOx extended OS in transplant-ineligible patients[44]. Subcutaneous formulations of epcoritamab allow outpatient administration with shorter treatment times.

Resistance to BiTEs can take many forms, including CD20 antigen loss (due to MS4A1 gene alterations), genomic reprogramming of TP53 and MYC, and T-cell exhaustion characterized by an increase in inhibitory receptors such as PD-1, LAG-3, and TIM-3[43].

PRECISE APPLICATION OF CAR-T THERAPY

Efficacy and limitations of approved CAR-T products

CD19-targeted CAR-T cell therapy has changed the course of treatment for R/R DLBCL. ZUMA-1 evaluated axicabtagene ciloleucel (axi-cel), which achieved an 83% ORR, a 58% CR rate, and a 5-year OS of 42.6%[45]. TRANSFORM revealed that lisocabtagene maraleucel (liso-cel) is superior to high-dose chemotherapy followed by autologous stem cell transplantation in the second-line setting[46]. BELINDA did not demonstrate an advantage of tisagenlecleucel over standard treatment[47]. Thus, not all CAR-T cell products are equivalent, highlighting product-specific differences.

The safety profile includes cytokine release syndrome and immune effector cell-associated neurotoxicity syndrome, both of which require specialized management[48]. Common mechanisms of relapse include loss of CD19 antigen and T-cell exhaustion[49]. Table 3 summarizes the efficacy of the main novel regimens for R/R DLBCL.

Table 3 Efficacy of major novel therapeutic regimens in relapsed/refractory diffuse large B-cell lymphoma.

Therapy type	Regimen	Key trial	ORR, %	CR rate, %	PFS/OS	Target population
ADC	Polatuzumab + BR	Pola-BR	NA	40 (vs 18)	PFS 10 months vs 4 months; OS 12 months vs 5 months	Transplant-ineligible R/R DLBCL
Anti-CD19 mAb + IMiD	Tafasitamab + lenalidomide	L-MIND	58	40	5-year OS 81% in CR patients	CAR-T intolerant
Bispecific (IV)	Glofitamab	NP30179	52	39	18-month sustained CR 67%	Heavily pretreated
Bispecific (SC)	Epcoritamab	EPCORE NHL-1	63	39	NA	Heavily pretreated
CAR-T (third-line)	Axi-cel	ZUMA-1	83	58	5-year OS 42.6%	≥ 3 prior lines
CAR-T (second-line)	Liso-cel	TRANSFORM	NA	74 (vs 43)	EFS significantly superior to ASCT	Early relapse in the second-line setting

ADC: Antibody-drug conjugate; mAb: Monoclonal antibody; IMiD: Immunomodulatory drug; BR: Bendamustine + rituximab; IV: Intravenous; SC: Subcutaneous; CAR-T: Chimeric antigen receptor T-cell; Axi-cel: Axicabtagene ciloleucel; Liso-cel: Lisocabtagene maraleucel; ORR: Overall response rate; CR: Complete response; NA: Not available; PFS: Progression-free survival; OS: Overall survival; EFS: Event-free survival; ASCT: Autologous stem cell transplantation; R/R: Relapsed/refractory.

Patient selection and predictive biomarkers

Patient selection is critical for optimizing CAR-T cell outcomes. Elevated baseline LDH levels, bulky tumor burden, and poor Eastern Cooperative Oncology Group (ECOG) performance status are associated with a poor response[50]. Dynamic monitoring of ctDNA after CAR-T cell infusion is now used to predict long-term remission: Rapid clearance is associated with persistent CR, whereas persistent ctDNA indicates disease relapse[51]. Characterization of the tumor immune microenvironment, with a focus on immunosuppressive cell populations and T-cell exhaustion markers, may provide further insight into CAR-T cell persistence and durability.

New CAR-T strategies

New CAR-T cell strategies include dual-targeting CAR-T cells (CD19/CD22 or CD19/CD20) to address antigen escape, immune checkpoint inhibitors to reverse T-cell exhaustion, and allogeneic off-the-shelf products to overcome manufacturing delays[52].

SYNERGISTIC STRATEGIES OF IMMUNE MICROENVIRONMENT AND IMMUNE CHECKPOINT THERAPY

Expression and function of the PD-1/PD-L1 pathway in DLBCL

Overexpression of PD-L1 in DLBCL is frequently associated with 9p24.1 amplification and EBV positivity[53]. Although biologically plausible, PD-1 inhibitor monotherapy yields a low response rate of approximately 10% in patients with DLBCL[54]. Classical Hodgkin lymphoma and primary mediastinal B-cell lymphoma are more frequently associated with high levels of PD-L1 expression and are highly responsive to PD-1 inhibitors.

Immune checkpoint inhibitor combination therapy strategies

Combination strategies are being employed to improve clinical outcomes. Pembrolizumab plus lenalidomide achieved a complete response rate of approximately 31% in R/R DLBCL[55]. In the context of CAR-T cell therapy, PD-1 blockade may reinvigorate exhausted CAR-T cells and extend their lifespan[56]. LAG-3, TIM-3, and TIGIT are now emerging as checkpoints, and dual blockade has shown synergistic antitumor effects in preclinical studies.

Assessment and regulation of TME functional status

Multiplex immunohistochemistry and single-cell RNA sequencing have been used to quantify the abundance, functional state, and spatial distribution of immune cells in the TME[57]. CSF1R inhibition can reprogram M2-like macrophages into tumor-killing M1-like macrophages[58]. Vascular normalization by VEGF inhibitors can increase T-cell infiltration and enhance the effects of immunotherapy[59].

PROGNOSTIC MODELS: FROM IPI TO MOLECULAR NOMOGRAMS

Traditional IPI and its deficiencies

The IPI includes age, ECOG performance status, LDH level, extranodal involvement, and Ann Arbor stage. The 5-year OS rates of the four risk groups are 73%, 51%, 43%, and 26%, respectively[60]. The R-IPI reduced this to three grades, with 4-year OS rates of 94%, 79%, and 55%[61]. The NCCN-IPI refined age stratification (> 40 years, > 60 years, > 75 years) and LDH grading (> 1 × the upper limit of normal, > 3 × the upper limit of normal), with a C-index of approximately 0.632 and a 5-year OS ranging from 92% (low risk) to 49% (high risk)[62]. However, these indices do not account for the molecular heterogeneity of DLBCL and thus have differing prognostic value within the same IPI risk group.

Integration of molecular prognostic biomarkers

Integration of molecular biomarkers significantly improves prognostic accuracy. Both double-expressor lymphoma (MYC and BCL2 coexpression) and double-hit lymphoma (MYC and BCL2 and/or BCL6 rearrangements) are associated with poor prognosis[63]. TP53 mutations, NOTCH1 mutations, and KMT2D mutations have also been identified as adverse prognostic factors[64]. Gene expression signatures of immune cell infiltration also provide prognostic information: High M2-like TAM infiltration is associated with poor prognosis, whereas a high density of CD8+ T cells predicts longer survival[21]. Dynamic ctDNA monitoring can detect changes in tumor cells at any time point and may thus improve patient prognosis[65].

Nomogram-based individualised prognostic systems

A nomogram-based prognostic model predicts an individual’s risk of all-cause mortality on the basis of several patient factors. Models developed in China have incorporated clinical characteristics and molecular indicators to create a lymphoma impact and response score[66]. Multinational nomograms that incorporate biochemical markers, performance status, and molecular subtypes have been used to divide patients into several risk groups[67]. Clinically, these models still need to be integrated with electronic health record systems, externally validated, and made accessible in low-resource settings[68].

HIV-ASSOCIATED DLBCL: MOLECULAR CHARACTERISTICS AND PRECISION THERAPY

Epidemiology and pathogenesis

HIV-associated DLBCL is a distinct clinical entity with unique molecular features and specific treatment requirements. This patient population has been systematically excluded from most major clinical trials. Given the high prevalence of HIV and the elevated risk in certain populations, HIV testing may be recommended for patients with DLBCL on the basis of epidemiological factors and clinical conditions. In a Chinese cohort of 63 patients with HIV-associated DLBCL patients, 74.6% had a GCB origin, 56.9% had an IPI score of 3-5, 78% had Ann Arbor stage III-IV disease, and 56% of patients were newly diagnosed with HIV and lymphoma simultaneously[69]. HIV is not directly oncogenic; rather, the reduction in CD4+ T cells impairs immune surveillance and permits EBV reactivation in 90%-100% of immunoblastic cells. The HIV proteins gp120, p17, and Tat directly promote B-cell proliferation and genomic instability[70]. Table 4 summarizes the genomic differences between HIV-positive and HIV-negative DLBCL.

Table 4 Genomic differences between human immunodeficiency virus-positive and -negative diffuse large B-cell lymphoma.

Gene/feature	HIV+ DLBCL	HIV- DLBCL	P value
TP53 mutation	More frequent, diverse	Less frequent	< 0.05
MYD88 mutation	Significantly reduced	Common (especially in MCD subtype)	< 0.05
CD79B mutation	Significantly reduced	Common	< 0.05
PIM1 mutation	Significantly reduced	Common	< 0.05
MYC mutation (SNV)	More frequent	Less frequent	< 0.05
LRP1B/TYK2 mutation	Higher frequency	Lower frequency	< 0.05
LymphGen ‘unclassified’ proportion	Higher	Lower	< 0.05

HIV: Human immunodeficiency virus; DLBCL: Diffuse large B-cell lymphoma; SNV: Single nucleotide variant; MCD: MYD88 (L265P)/CD79B-mutated subtype; LymphGen: A probabilistic classification tool for genetic subtypes of DLBCL.

Molecular landscape

Compared with HIV-negative DLBCL, HIV-associated DLBCL has a distinct genomic landscape, characterized by significantly lower frequencies of MYD88 and CD79B mutations, higher frequencies of TP53 and MYC mutations, higher proportions of LRP1B/TYK2 mutations, and a higher proportion of LymphGen “unclassified” cases[71]. A single-cell immune atlas revealed a dual immune evasion strategy: Malignant B cells markedly reduced MHC class I expression, and CD8+ T cells displayed senescence-like dysfunction characterized by impaired cytolytic granule polarization[71].

Treatment and outcomes

With the addition of antiretroviral therapy, standard chemoimmunotherapy (R-CHOP, DA-EPOCH-R) has become feasible and effective, achieving CR rates ranging from 69% to 91%[72,73]. In a Chinese study, the 1-, 3-, and 5-year OS rates were 65.0%, 47.1%, and 43.5%, respectively[69]. Independent adverse prognostic factors were age ≥ 60 years (HR = 2.784), IPI score 3-5 (HR = 2.814), ECOG-PS score 2-4 (HR = 4.015), and fewer than 4 cycles of chemotherapy (HR = 0.290). Unboosted integrase strand transfer inhibitors, including dolutegravir and bictegravir, are preferred to avoid drug interactions with chemotherapy[70]. Most key immunotherapy trials systematically exclude HIV-positive patients; therefore, there is an urgent need to advocate for the inclusion of individuals with virologically suppressed HIV in clinical trials[74].

CHALLENGES AND FUTURE DIRECTIONS OF PRECISION MEDICINE IMPLEMENTATION

Standardization and accessibility of biomarker testing

GEP and targeted sequencing are relatively expensive and require fresh or frozen tissue samples; therefore, their clinical application is limited. IHC-based surrogate assays are more convenient but have inconsistent accuracy[75]. Liquid biopsy is noninvasive; however, its preanalytical and analytical stages need to be standardized. Cross-validation and participation in external quality assessment programs are essential to ensure reliable results[76].

Cost-effectiveness and resource allocation

New therapies are costly; consequently, CAR-T cell therapy and bispecific antibodies are not widely available. Cost-effectiveness analyses have supported the use of these approaches in certain patient groups in high-income countries, but affordability remains a global challenge[77]. Resource-constrained settings should focus on biosimilar-based regimens and simplified molecular diagnostics.

Artificial intelligence and multimodal data integration

Artificial intelligence and machine learning can be used to integrate multimodal data from imaging, molecular profiles, and clinical records. Deep learning applied to digital pathology has advanced the prediction of immunochemotherapy response in patients with DLBCL[78]. Machine learning models that combine features from PET/CT images and clinical data can improve prognostic accuracy[79]. However, challenges related to data standardization, reproducibility, and biological interpretation remain unresolved[18].

Innovations in clinical trial design

The molecular heterogeneity of DLBCL requires novel clinical trial designs. Basket trials can investigate multiple molecularly defined patient subgroups treated with the same therapeutic agent concurrently. Platform trials can evaluate multiple investigational agents simultaneously within the same infrastructure. Minimal residual disease-driven adaptive designs dynamically adjust therapy on the basis of serial ctDNA measurements to accelerate new drug development.

CONCLUSION

Precision medicine in DLBCL has transitioned from a conceptual framework to the initial stages of clinical implementation. Distinct biological features of DLBCL cells can be identified through molecular classification, after which targeted therapy can be selected on the basis of these differences.

A pivotal advance is that, instead of the traditional fixed-value prognostic index, the IPI, a dynamic and comprehensive model incorporating ctDNA, TME features, and genetic subtypes has been proposed. Together, these biomarkers offer a more nuanced and real-time reflection of the disease and its response to treatment. Molecular classification and ctDNA-guided approaches are promising but have not yet been prospectively validated.

With the emergence of agents targeting specific molecular vulnerabilities, such as BTK inhibitors, BCL-2 inhibitors, ADCs, and bispecific antibodies, treatment options for high-risk DLBCL have gradually expanded. CAR-T cell therapy can achieve durable remission in patients with refractory or relapsed DLBCL, but issues related to accessibility and response rates remain to be addressed. The TME and emerging immune checkpoint targets have become fertile ground for investigations aimed at overcoming therapeutic resistance. Prognostic nomograms, such as those developed in China based on patient cohorts, exemplify this trend toward personalized survival prediction and risk-adjusted therapy.

Challenges related to standardization in molecular diagnostics, therapeutic resistance, and equitable access to personalized medicine remain unresolved. In the future, further research will be needed to develop affordable molecular diagnostic tools and to integrate multiple data sources adaptively for low- and middle-income countries. Ultimately, the future direction of DLBCL treatment will require multidisciplinary collaboration to incorporate molecularly targeted therapy, immunotherapy, and advanced prognostic indicators into a comprehensive, patient-oriented treatment plan.

Based on the evidence reviewed, we propose a conceptual framework for integrating precision medicine into DLBCL treatment. Initial risk assessment would consider the NCCN-IPI and COO classification, as well as baseline ctDNA analysis when available. Serial ctDNA monitoring could be performed during treatment to evaluate early response and identify patients at high risk of relapse; however, this approach has not yet been prospectively validated. At the time of relapse or progression, CAR-T cell therapy, bispecific antibodies, ADCs, and novel targeted agents should be selected on the basis of previous treatment history, genetic subtypes, and resistance mutation patterns. On the basis of the current evidence, this framework is proposed as an evolving model, and its application is limited by cost, availability, and the need for standardized assays.