Abstract
Background: Diffuse large B-cell lymphoma (DLBCL) is clinically and genetically heterogeneous, with 20–40% of patients relapsing after frontline therapy. Existing genetic classifiers LymphGen and DBclass are complex in their logic and limited in their ability to classify diverse DLBCL subtypes, presenting a challenge in patient stratification for clinical trials and treatment selection. To address this challenge, we created Lymphly, an evidence-based classifier that integrates pathway-relevant genomic alterations—such as mutations and copy number changes—to refine DLBCL subtyping.
Methods: Lymphly constitutes a nine-step hierarchical algorithm that assigns DLCBL subtypes based on core (clearly defined) and extended (broad) genomic features. We analyzed over 300 publications and extracted 850 genomic alterations associated with previously reported DLBCL genetic subtypes. DLBCL samples (total n=840, from internal (BostonGene) and open-access (NCICCR) cohorts, including genetically composite cases, were classified into six distinct subtypes—EZB, MCD, BN2, N1, JS3, and JS6—based on their whole-exome sequencing data. Subtyping using a hierarchical decision framework prioritized the number of mutated genes and mutation burden within core and extended genomic features, designed to assign even genetically complex cases to specific subtypes. Mixed subtypes were designated exclusively during hierarchical classification where preference was given to clinically actionable entities such as EZB, MCD, and BN2. TP53+ and MYC+ statuses were also assigned to designate high-risk cases. Finally, Lymphly was benchmarked against LymphGen (subtypes EZB, MCD, BN2, N1, ST2) and DLBclass (clusters C1–C5) to evaluate its performance.
Results: Lymphly successfully assigned subtypes to >96% of samples analyzed. Lymphly EZB overlapped substantially with LymphGen EZB and DLBclass C3, indicative of GCB (Germinal Center B-cell like) phenotype and epigenomic disruption. Lymphly MCD resembled DLBclass C5 and LymphGen MCD, consistent with an ABC (Activated B-cell-like) phenotype and BCR/TLR activation. Notably, JS3 and JS6 are two distinct, genetically stable subtypes that further subdivide the broader LymphGen ST2 and DLBClass C4 categories. JS3 corresponds to ABC-like DLBCLs with hyperactivated NF-κB signaling, often driven by autocrine STAT3–IL6/IL10 loops. These tumors often show PI3K pathway activation, comparable to AML-like myeloid biology and exhibiting resistance to BCR targeting therapies. In contrast, JS6 represents a GCB-like, PMBL-like (Primary mediastinal B-cell lymphoma) subtype enriched for PDL1 amplification, STAT6 gain-of-function mutations, and a CD4+ T-cell–dominant immune microenvironment. These features suggest immune evasion and JAK/STAT-driven oncogenesis, highlighting opportunities for immunomodulatory therapies.
Lymphly also identified 285 TP53+ cases across all subtypes that partially overlapped with LymphGen A53 (n=81) and DLBclass C2 (n=117) while uniquely capturing 97 TP53-altered tumors missed by other classifiers. Also absent from existing classifiers is the MYC+ status defined by MYC translocations and activating mutations. Lymphly identified 59 MYC+ cases among our dataset that were consistently associated with the DZsig transcriptional signature, reflective of dark zone features in DLBCL.
Lymphly EZB showed the best overall survival (OS) followed by BN2, while MCD and N1 showed the worst OS. The prognostically unfavorable JS3 subtype showed an OS resembling that of MCD, whereas the favorable JS6 subtype showed an OS comparable to that of EZB/BN2. Co-occurrence of TP53+ and MYC+ markedly worsened the prognosis in GCB samples, resulting in an OS comparable to that of ABC cases.Conclusions: Lymphly enables robust, transparent classification of DLBCL based on discrete genetic events, capturing canonical and composite subtypes alongside biologically relevant TP53+ and MYC+ cases. Specifically, it refines LymphGen and DBclass classification by incorporating deeper biological insights, enabling a more comprehensive characterization of DLBCL heterogeneity. The identification of Lymphly JS3 and JS6 also underscores the limitations of current models and uncovers new avenues for targeted therapy for ABC and GCB classes. By reducing the unclassified rate and supporting modular data input, Lymphly improves patient stratification for clinical trials and treatment decision-making.
