Remission Assessment by Circulating Tumor DNA in Large B-Cell Lymphoma

Abstract

Purpose: Large B-cell lymphomas (LBCLs) are curable, but patients with residual disease after therapy invariably experience progression. Ultrasensitive methods to detect circulating tumor DNA (ctDNA) as minimal residual disease (MRD) may improve the determination of remission.

Methods: We integrated data from five prospective studies of frontline anthracycline-based chemotherapy in patients with LBCL. Tumor-specific phased variants were identified from pretreatment samples and monitored at landmark time points. Serial plasma specimens were blindly analyzed for detectable ctDNA as MRD. MRD status was compared with conventional response criteria for prognosis of progression-free survival (PFS).

Results: We studied ctDNA-MRD in 137 patients by monitoring 409 plasma specimens over time. Detectable ctDNA rates decreased during therapy with 55% and 78% of patients achieving undetectable ctDNA after two cycles and at the end of therapy, respectively. After a median follow-up of 37 months, the 2-year PFS for patients with detectable versus undetectable ctDNA after two cycles was 67% versus 96% (P = .0025; hazard ratio [HR], 6.9) and after therapy was 29% versus 97% (P < .0001; HR, 28.7), respectively. Ninety-two (94%) patients with undetectable ctDNA at the end of therapy remained alive without progression, while 19 (68%) patients with detectable ctDNA progressed or died. MRD status at the end of therapy had greater prognostic utility than conventional lymphoma response criteria using positron emission tomography (PET) scans (HR, 3.6 for positive PET and 28.3 for detectable ctDNA).

Conclusion: Ultrasensitive ctDNA detection after frontline LBCL therapy is more prognostic than conventional radiographic response criteria. A refined definition of remission with ctDNA-MRD may improve clinical and psychological outcomes for patients with LBCL.

Trial registration: ClinicalTrials.gov NCT04002947 NCT04134936 NCT04231877 NCT00398177 NCT02529852.

FIG 1.

PhasED-Seq. (A) The figure depicts the key features of phased variants. As opposed to single-nucleotide variants, phased variants are multiple alterations that can be observed together on a single DNA molecule. The concordant observation of multiple alterations simultaneously significantly lowers the background error profile for MRD detection. (B) PhasED-Seq uses phased variants to track many phased variants simultaneously. Since the amount of cell-free DNA in plasma is limited, this is essential to enable detection to the parts per million range from a standard blood plasma collection. ctDNA, circulating tumor DNA; MRD, minimal residual disease; PhasED-Seq, Phased Variant Enrichment and Detection by Sequencing; PV, phased variants; SNV, single-nucleotide variants.

FIG 2.

Flow diagram for pooled cohort inclusion. The diagram summarizes the enrollment and sample availability across five prospective clinical trials included in the pooled ctDNA-MRD analysis. A total of 163 patients were initially considered. Patients were excluded because of absence of a baseline sample (n = 2), failure to identify phased variants (n = 9), lack of post-treatment plasma samples (n = 13), incorrect histologic diagnosis (n = 1), or previous systemic therapy (n = 1), resulting in a final evaluable cohort of 137 patients. Sample availability at MRD landmark time points (cycle 2 day 1, cycle 3 day 1, and end of therapy) is shown by trial. ctDNA-MRD, circulating tumor DNA-minimal residual disease; FL, follicular lymphoma; MDACC, University of Texas MD Anderson Cancer Center; MRD, minimal residual disease; NCI, National Cancer Institute; PV, phased variants; UW, University of Washington.

FIG 3.

ctDNA kinetics during treatment. (A) Dot plot illustrating variant allelic levels across treatment courses, stratified by PFS status during the follow-up period—red indicates patients who experienced a PFS event, and blue represents those who remained event-free. Variant allelic level is expressed as the fraction of molecules harboring lymphoma-specific phased variants among all informative molecules analyzed. Horizontal lines indicate median values within each group. Patients with undetectable ctDNA are annotated, and the proportion of patients with undetectable ctDNA is displayed at the bottom of the plot. (B) Bar graph depicting the percentage of patients with undetectable ctDNA at each profiled time point during treatment. C2D1, cycle 2 day 1; C3D1, cycle 3 day 1; ctDNA, circulating tumor DNA; EOT, end of treatment; PFS, progression-free survival.

FIG 4.

Stratification and performance metrics of end of therapy ctDNA profiling. (A) Kaplan-Meier curve for PFS stratified by ctDNA detection status at the end of therapy. HR and P value from Cox proportional hazards regression are shown. Donut plots display the proportion of patients with detectable ctDNA by PFS status, distinguishing cases with relatively high ctDNA burden (≥0.01%, ie, ≥10⁻⁴) in dark red from cases with lower ctDNA levels (<0.01%, ie, <10⁻⁴) in light red. (B) Kaplan-Meier curve for FFP stratified by ctDNA detection status at the EOT. HR and P value from Cox proportional hazards regression are shown. Donut plots display the proportion of patients with detectable ctDNA by FFP status, distinguishing cases with relatively high ctDNA burden (≥0.01%, ie, ≥10⁻⁴) in dark red from cases with lower ctDNA levels (<0.01%, ie, <10⁻⁴) in light red. (C) Sensitivity (blue) and NPV (purple) of end-of-treatment ctDNA profiling for predicting 24-month PFS, evaluated across analytical thresholds ranging from 10⁻⁶ to 10⁻⁴. ctDNA, circulating tumor DNA; EOT, end of treatment; FFP, freedom from progression; HR, hazard ratio; MRD, minimal residual disease; NPV, negative predictive value; PFS, progression-free survival.

FIG 5.

Prognostic value of EOT ctDNA detection compared with PET-CT and clinical risk factors. (A) Kaplan-Meier curve for PFS stratified by EOT PET-CT status in patients eligible for analysis of both modalities. HRs and P values from Cox proportional hazards regression are displayed. (B) Kaplan-Meier curve for PFS stratified by EOT ctDNA detection status in the same patient cohort. HRs and P values from Cox proportional hazards regression are shown. (C) Kaplan-Meier curve for PFS stratified by EOT ctDNA detection status within the PET-negative subset. HR and P value from Cox proportional hazards regression are shown. The donut plot illustrates the proportion of patients with detectable ctDNA within the PET-negative subset. (D) Kaplan-Meier curve for PFS stratified by EOT ctDNA detection status within the PET-positive subset. HR and P value from Cox proportional hazards regression are shown. The donut plot illustrates the proportion of patients with detectable ctDNA within the PET-positive subset. (E) Forest plot showing HRs, 95% CIs, and P values derived from multivariable Cox proportional hazards regression for PFS, incorporating EOT ctDNA detection, EOT PET-CT, IPI, COO, and histology subtype (ie, DLBCL v PMBL or HGBL) as covariates. COO, cell-of-origin; ctDNA, circulating tumor DNA; CT, computed tomography; DLBCL, diffuse Large B-cell lymphoma; EOT, end of treatment; GCB, germinal center B cell-like; HGBL, high-grade B-cell lymphoma; HR, hazard ratio; IPI, International Prognostic Index; PET, positron emission tomography; PFS, progression-free survival; PMBL, primary mediastinal B-cell lymphoma.