Introduction Odronextamab, a CD20×CD3 bispecific antibody, showed compelling efficacy and generally manageable safety in patients (pts) with R/R B-NHL in ELM-2 (NCT03888105). Minimal residual disease (MRD), an exploratory endpoint of ELM-2, was initially assessed by ctDNA using the duplex CAPP-Seq method (AVENIO Oncology Assay NHL Test). Here, we evaluate MRD by ctDNA detected using phased variant tracking to determine whether this method more accurately predicts outcomes.

Methods Raw AVENIO ctDNA sequencing data from plasma samples of pts with R/R follicular lymphoma (FL) and R/R diffuse large B-cell lymphoma (DLBCL) in ELM-2 were re-analyzed to identify phased variants per Kurtz et al. (Nat Biotech 2021). Baseline (BL) tumor biopsies were not used to determine variants, reflecting realistic clinical sample availability. Phased variants were defined as 2 somatic mutations on the same DNA strand <170 bp apart. Triplet variants were not analysed. Identified BL phased variants had an allele frequency (AF) >1% (AF >0.2% for samples with no variants with AF >1%), but were filtered out if found in the germline sequence or in the healthy volunteer (HV) cell-free (cf)DNA blacklist (cfDNA was sequenced from 5 HVs). Remaining variants were “reporter phased variants”. The first on-treatment sample was taken at Week (Wk) 12 and used as a landmark to determine assay power in predicting clinical progression.

Results In evaluable pts with R/R FL (n=65), the median number (range) of duplex reporter variants per pt (102 [0–323]) correlated moderately (rho=0.61) with the number of phased reporter variants (42 [1–9943]) at BL. In evaluable pts with R/R DLBCL (n=86), the observed correlation was similar (rho=0.59), although a lower median number (range) of reporter variants was detected by duplex (148 [0–529]) vs phased (680 [1–9765]) at BL. The relative amount of ctDNA (mutant molecules per mL [MMPM]) correlated significantly between the 2 methods in both FL and DLBCL (rho=0.65 and 0.83, respectively). In 3 BL samples, ctDNA was undetectable by duplex, but detectable by phased (with 8, 24, and 92 reporter variants).

FL BL MMPM was highly predictive of progression-free survival (PFS) by both duplex and phased (HR: duplex 0.36 [95% CI 0.16–0.78] vs phased 0.33 [0.15–0.73]). This BL predicitivity was not observed in DLBCL.

Concordance between the 2 methods at Wk 12 was 66.7% in FL and 66.3% in DLBCL. Undetectable ctDNA at Wk 12 was strongly associated with longer PFS in FL by both duplex and phased, with similar HRs (0.31 with both methods) and median PFS (42.4 vs 41.2 mos from Wk 12, respectively). In DLBCL, phased slightly outperformed duplex (HR in pts with undetectable ctDNA: 0.32 vs 0.42, respectively).

When using Wk 12 ctDNA status to predict progressive disease (PD) (by Lugano criteria or given as reason for treatment discontinuation) the phased method showed a higher false negative rate in FL, with a sensitivity (true PD/all actual PD cases) of 0.52 (12/23) vs 0.83 (19/23) by duplex. Sensitivity was similar for both methods in DLBCL (duplex 0.80 [28/35] vs phased 0.77 [27/35]). Specificity (true non-PD/all actual non-PD cases) was slightly higher by phased (0.84 [31/37]) vs duplex (0.76 [28/37]) in FL, with this observation more pronounced in DLBCL (phased 0.7 [31/44] vs duplex 0.48 [20/42]) indicating a lower false positive rate with duplex.

Beyond Wk 12, 2 pts with FL had undetectable ctDNA by duplex yet detectable ctDNA by phased at all timepoints, while remaining progression-free at 2 and 4 yrs. In contrast, 2 pts with detectable ctDNA by duplex but undetectable ctDNA by phased were progression-free after 2 yrs. In DLBCL, discordance was observed in 3 pts who remained progression-free (1 pt with undetectable ctDNA by duplex but not by phased, and 2 pts with undetectable ctDNA by phased but not by duplex).

Conclusions ctDNA quantification by phased and duplex variant analysis correlated strongly, but ctDNA determination by phased did not consistently improve the power to predict PFS or PD. Phased method showed greater specificity in DLBCL, by better identifying pts with cleared ctDNA who did not progress, while the duplex method had greater sensitivity to identify FL pts who were ctDNA-positive and would later progress. Each approach has specific benefits, but both appear appropriate for MRD assessment, notably in pts with DLBCL. Further analyses of these methods across B-NHL subtypes are required to discern their relative advantages.

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