Introduction Detection of measurable residual disease (MRD) at end-of-treatment (EOT) using ultrasensitive circulating tumor DNA (ctDNA) assays is predictive of relapse risk (Roschewski et al., ASH 2024; Wang et al., ASCO 2025). Prior studies have shown that ctDNA-based molecular response assessment early during therapy also holds promise for risk stratification (Kurtz et al., JCO; Nat Biotechnol). These advances have enabled ctDNA-guided treatment escalation and de-escalation strategies in ongoing prospective DLBCL trials (Nijland et al., ICML 2025; Cherng et al., ASCO 2025). However, comprehensive real-world data capturing the full longitudinal dynamics of ctDNA in large B-cell lymphomas (LBCLs) are lacking.

Methods We analyzed ctDNA in patients enrolled in HOVON-902, a prospective, multicenter, real-world study of blood-based response monitoring in LBCL patients receiving R-CHOP or DA-EPOCH-R induction therapy. Plasma samples were collected at up to 11 time points: baseline (C1D1), at interim induction cycles (C2D1, C3D1, C4D1), at EOT (defined as end of induction), and during surveillance (every 3 months in year 1 and every 6 months in year 2). Patient-specific phased variants (PVs) were identified using PhasED-Seq from FFPE or baseline plasma with matched germline DNA (Foresight CLARITY, Boulder, CO). PVs were tracked longitudinally to detect MRD at key treatment milestones and to assess previously defined molecular response kinetics (early molecular response [EMR], major molecular response [MMR]) in relation to best observed response (BOR) according to Lugano 2014 response criteria, PFS and OS.

Results We monitored ctDNA MRD in >1000 serial plasma samples from >150 patients with LBCLs (94% DLBCL-NOS, 6% HGBCL), using PVs successfully identified from either FFPE or baseline plasma. The median age was 68 years (19-88), 66% were male, 80% had advanced-stage disease (Lugano III/IV), and 20% had Low, 30% Low Intermediate, 28% High Intermediate, and 22% High IPI risk. At a median follow-up of 37 months (1.4-55.4), the 3-year PFS and OS were 81% and 88%, respectively.

Overall, more patients cleared MRD with each successive cycle during 1L therapy. Specifically, MRD detection decreased over time during induction, with preliminary on-treatment MRD clearance rates of 28%, 51%, 64%, and 81% at C2D1, C3D1, C4D1, and EOT, respectively. MRD status was prognostic at each landmark (P<0.05 for all), but with increasing prognostic value over time. Consistent with the therapeutic effect of additional treatment cycles, outcomes for MRD+ patients at each landmark worsened throughout therapy. For example, initial results show a 3-yr PFS of MRD+ patients after 1 cycle of 69% (at C2D1), as compared to 11% at EOT. Outcomes were also correlated with early ctDNA response kinetics: when considering quantitative response thresholds, EMR (2-log10 decrease in [ctDNA]) and MMR (2.5-log10 decrease) were both significantly associated with 3-year PFS. For example, preliminary results show lack of EMR associated with significantly inferior 3-year PFS (EMR, HR 3.2, 95% CI 1.7-5.9, p < 0.001).

We also considered the role of disease surveillance after the end of induction therapy. With the addition of surveillance samples, initial results show that 91% of patients with confirmed DLBCL relapse were MRD+ at either the EOT or during follow-up, representing a 17% increase in clinical sensitivity for detecting relapse. Conversely, among patients who did not relapse, 96% had negative EOT MRD, and 99% had either negative EOT MRD or achieved MRD clearance during follow-up. Interesting and distinctive temporal and kinetic patterns of MRD detection as a function of anatomic site of relapse, relapse histology, and response to consolidative or secondary therapies will be presented, along with corresponding lead times relative to current surveillance practice strategies and estimates of potentially avoidable radiographic studies.

Conclusions This is the first real-world study of ctDNA dynamics across serial time points in a uniformly treated LBCL cohort. Our findings confirm the prognostic value of on-treatment MRD and molecular response assessment. Post-EOT ctDNA monitoring reflects the temporal relationship between tumor burden detection and relapse, highlighting its potential for early detection. Our study provides a comprehensive overview of ultrasensitive MRD detection and how it might inform personalized surveillance and treatment strategies for LBCL.

This content is only available as a PDF.
2025
Sign in via your Institution