Background:NPM1 mutations (NPM1m) are a well-characterized biomarker for minimal residual disease (MRD) detection in acute myeloid leukemia (AML). Based on our retrospective meta-analysis across multiple trials and prior published studies, we are evaluating the utility of MRD complete response as a surrogate biomarker for clinical benefit in the AGILITY study (NCT05020665) to support a potential accelerated approval. Currently, measurement of NPM1m RNA transcripts (relative to ABL1) using reverse transcriptase quantitative polymerase chain reaction (RT-qPCR) is considered the gold standard for monitoring of MRD; however, DNA-based next generation sequencing (NGS) provides several logistical and technical advantages over RT-qPCR in a global registrational trial setting. NGS allows for detection of all NPM1m variant alleles in a single assay (more closely approximating the variant allelic fraction [VAF] from leukemic cells vs NPM1m transcripts). Furthermore, longer stability of DNA during transit alleviates logistical challenges related to sample processing across different labs and transport from sites worldwide to a central testing site. Purpose: To develop a standardized and highly sensitive DNA-based NGS assay for detection of MRD in NPM1m AML patients at the end of 2 cycles of chemotherapy in the AGILITY study and to assess its performance against an established RT-qPCR method. Methods: Assay performance was assessed by serially diluting NPM1 synthetic DNA or NPM1m genomic DNA (gDNA) extracted from OCI/AML3 cells in a background of wild-type bone marrow (BM) gDNA (range: 0.008%-0.0005% VAF) using a fixed sample input of 400,000 gene copies. DNA sequencing (median read depth: 5.5×106) analysis based on the Inivata MRD platform was performed on the Illumina NextSeq 550 instrument using a proprietary bioinformatic pipeline. A set of follow-up AML samples consisting of peripheral blood (N = 31) and BM (N = 55) from patients enrolled in the UK NCRI AML Study Group AML-19 trial (ISRCTN 2014-002195-90) were tested for comparison of MRD status by NGS with an orthogonal NPM1 RT-qPCR method. This sample cohort included both negative (n = 16) and positive (n = 70) samples as defined by RT-qPCR analysis. To assess the level of agreement between both platforms, we use Pearson's correlation statistics including 95% CIs. Assay performance was evaluated by determining the area under the receiver operating characteristic curve (ROCAUC). RT-qPCR was performed as previously described (Ivey A. N Engl J Med. 2016). NGS results are expressed as percent variant allele frequency (%VAF) and RT-qPCR results as NPM1m transcript copies per 10e4ABL transcripts (normalized copy number [NCN]). Results: The NPM1m NGS assay demonstrated high sensitivity and specificity with a limit of detection with 95% confidence (LoD95) of 0.002% VAF when using serially diluted NPM1 synthetic DNA or NPM1m gDNA extracted from OCI/AML3 cells. In the AML19 set, MRD was detectable in 52/70 (74.3%) samples by both platforms. The level of MRD detected by NGS varied between 0.00025%-46% VAF (median: 0.0042%) and by RT-qPCR between 0.01-1.65x105 NCN (median: 8.03), showing a statistically significant correlation when considering all positive samples (R = 0.84, P < 9.8e-15). Despite this agreement, in 18/70 (25.7%) samples NPM1m was not detected by NGS but detected by RT-qPCR, with NCN between 0.03 and 20.5 (median: 0.37), confirming the expected higher sensitivity of RT-qPCR. Conversely, MRD was detected in 5/16 samples by NGS but not detected by RT-qPCR (VAFs were < technical LoD95 in all cases). The NGS assay could reliably detect MRD positivity in almost all samples with an NCN > 10 copies (23/24 samples, 95.8%) and there was a statistically significant correlation (R = 0.78, P < 1.4e-5). Using a cutoff of > 10 copies, ROC analysis was performed, again demonstrating good concordance (AUC: 91.9%) and identifying an optimal cutoff of 0.003% VAF (specificity = 85.5%, sensitivity = 87.5%). Conclusion: Our DNA-based NGS-MRD assay demonstrates high sensitivity and specificity with a technical LoD95 in serial dilution of 0.002% VAF. It is sufficiently robust to reliably identify MRD positive samples above an RT-qPCR threshold of 10 copies / 10e4ABL (ie, 0.01% NPM1/ABL ratio) and notably, this is tenfold below the provisional European LeukemiaNet definition for high-level MRD of > 0.1%.