Comparative evaluation of NGS-based MRD detection in AML using longitudinal matched bone marrow and blood samples

Abstract Background: Measurable residual disease (MRD) detection is a critical prognostic tool in acute myeloid leukemia (AML), guiding treatment decisions and relapse monitoring. While flow cytometry remains the clinical standard, next-generation sequencing (NGS) offers enhanced sensitivity and broader mutation profiling. However, clinical implementation of NGS-based MRD testing is limited by assay variability, lack of standardization, and logistical constraints. Methods: Takeda conducted a pilot study involving 23 AML procured patients with matched bone marrow aspirate (BMA) and whole blood (WB) samples collected at three timepoints (T1–T3) including before and after treatment. We examined two NGS-based MRD assays for AML, Assay A and Assay B, from different diagnostic providers for their ability to detect MRD, variant concordance, and longitudinal tracking. MRD positivity was defined by the presence of pathogenic or likely pathogenic variants with variant allele frequency (VAF) ≥0.25%, excluding CHIP-associated genes. Results: MRD status showed high concordance between matched BMA and WB samples: 83.8% for Assay A and 86% for Assay B. Assay A identified a higher MRD positivity rate (57%) compared to Assay B (33%) under standard criteria. Variant-level concordance between BMA and WB was strong for Assay A (80%) and modest for Assay B (45%), with VAFs showing high correlation (R=0.84 and R=0.99, respectively). Longitudinal analysis revealed stable MRD status in most patients, though a subset showed dynamic VAF changes over time. Assay A and Assay B disagreed on the most frequently mutated genes, with TP53 being the only overlap. Assay A most frequently detected TP53, ASXL1, and CEBPA mutations, while Assay B identified SF3B1, TP53, and NRAS. A subset of 23 variants was detected by both assays, with Assay A demonstrating greater sensitivity in identifying shared variants. A total of 97% of samples were identified as MRD positive using a broader Assay B definition (any mutations with VAF between 0.5% and 40%, excluding CHIP genes), with a concordance rate of 93% between BMA and WB. Assay A showed 97% concordance under similar criteria. However, baseline MRD negativity (based on standard criteria) in newly diagnosed patients raised concerns about assay sensitivity and variant filtering strategies. Both assays failed to consistently exclude CHIP and/or germline mutations, emphasizing the need for standardized MRD definitions and variant interpretation frameworks. Conclusions: This study demonstrates that both Assay A and Assay B can reliably detect MRD in AML using either BMA or WB, with high inter-sample concordance. However, differences in assay sensitivity, panel design, and MRD definitions contribute to discordant results and false negativity. These findings support the feasibility of blood-based MRD monitoring and highlight the importance of assay selection and standardization in clinical and research settings. Broader adoption of NGS-based MRD testing in AML will require harmonized reporting criteria, standardized definitions for MRD, improved assay validation, and integration into clinical workflows.