Relapsed diffuse large B-cell lymphoma is shaped by molecular subtypes and genomic alterations Free

Introduction: Newly diagnosed (ND) diffuse large B-cell lymphoma (DLBCL) is a genetically heterogeneous disease that includes at least 5 molecular clusters (C1-C5) defined by distinct genetic signatures and responses to standard induction therapy. The clusters with inferior responses to frontline R-CHOP include: ABC-enriched C5 DLBCLs, characterized by frequent MYD88^L265Pand CD79B mutations, 18q copy number gain, and extranodal tropism; GCB-predominant C3 tumors with BCL2 translocations, including a subset with concurrent MYC translocations, and mutations in chromatin modifiers, B-cell transcription factors, and PI3K pathway components; and cell-of-origin independent C2 DLBCLs with biallelic TP53 alterations and associated genomic instability (AGI). In contrast, C1 and C4 DLBCLs have more favorable responses to R-CHOP. We recently developed a neural network-based probabilistic molecular classifier, DLBclass, to prospectively assign tumors to their respective clusters (Chapuy et al., Blood 2025).

Despite the interest in matching targeted therapies with molecularly defined subsets of ND DLBCL, novel agents are initially evaluated in patients (pts) with relapsed (R) tumors that have incompletely characterized genetic signatures. Herein, we investigate the molecular substructure of R DLBCL using the DLBclass framework.

Methods: We performed whole exome sequencing and obtained the comprehensive genetic signatures (mutations [SNVs and indels], somatic copy number alterations [SCNAs], structural variants [SV] including translocations) and DLBclass assignments for 122 first-relapsed DLBCLs including 30 central nervous system (CNS) R, and a subset (42) of matched diagnostic (Dx) specimens from pts who were treated with anthracycline-based induction chemo-immunotherapy and had full clinical annotation.Our previously characterized cohort of ND DLBCLs and the NIH series with defined genetic signatures, DLBclass calls,and known clinical outcomes were used for comparison.

Results: We used MutSig2CV to identify recurrent mutations in R DLBCL and found that ~80% of the alterations were DLBclass-defined features of ND DLBCL. This suggested that the propensity to relapse may be largely predetermined at diagnosis and provided the rationale for applying DLBclass in the R setting. In comparison to ND DLBCLs, R DLBCLs were enriched for C2 tumors (28.5% [ND] vs 41.8% [R], p=0.007), with a paucity of C1 lymphomas (15.5% [ND] vs 5.7% [R], p=0.004).

Extranodal (EN) Rs, including CNS Rs, were more likely than nodal (N) Rs to be C5 DLBCLs (32.2% [EN] vs 3.65% [N], p=0.002). In contrast, N Rs were more likely to be C3 tumors (11.8% [EN] vs 30.8% [N], p=0.01). These findings link specific patterns of tissue tropism with distinct genetic programs. Of interest, late R (>24 mos) were enriched for C5 DLBCLs in comparison to early R (<12 mos) (35.3% vs 18.9%, p=0.06).

Among pts with paired Dx and R samples, 31/42 (74%) retained their original molecular cluster IDs at R, whereas 8 of the remaining 11 acquired features of increased TP53-AGI and C2 cluster IDs. The higher frequency of TP53-AGI in R DLBCLs prompted us to further characterize the genes perturbed by focal CNAs in this setting. The more frequent focal CNAs at R included known immune response modulators and targets of inactivating mutations – CD70 (19p13.3del), B2M (15q15.3del), MHCI (6p21.33del), CD58 (1p13.1del), and FAS (10q23.31del). These findings potentially link TP53-AGI with additional CNA-dependent mechanisms of immune evasion at R.

We also identified Dx features associated with an increased risk of R using the previously characterized ND DLBCLs from our earlier analyses, the NIH cohort and this series, annotated for outcome (R [250 tumors] vs no R [249 DLBCLs]). Chromosome 17pdel, TP53mut, and 6p23.33del (MHCI) were significantly more frequent in Dx tumors of pts who subsequently relapsed, further emphasizing the importance of TP53-AGI and MHCI loss. In a multivariate Cox model adjusted for DLBclass IDs, concurrent MYC and BCL2 SVs (HR=3.39) and MYC SVs (HR=2.33) retained adverse prognostic significance, highlighting the additional impact of these alterations on outcome.

Conclusions: Relapse in DLBCL is largely driven by genetic programs present at diagnosis and defined by DLBclass. These signatures influence the clinical course of relapse, including tissue tropism, timing and immune escape, providing a framework for improved risk assessment and therapeutic targeting.