AUTHORS
Morgan Shannon, MD, Ritika Dutta, MD2, Henning Stehr, MD3, Tian Zhang, MD PhD4, Gabriel Mannis I, MD5
AFFILIATIONS
1 Stanford Health Care, Stanford, California, United States
2 Stanford Health Care, Stanford, California, United States
3 Stanford Health Care, Stanford, California, United States
4 Stanford Health Care, Stanford, California, United States, Stanford University School of Medicine, Stanford, California, United States
5 Stanford Health Care, Stanford, California, United States, Stanford University School of Medicine, Stanford, California, United States
ABSTRACT
Introduction:
Measurable residual disease (MRD) assessment carries well-established prognostic and therapeutic significance in NPM1-mutated acute myeloid leukemia (AML). Although the co-mutational landscape of NPM1-mutated AML is known to evolve at relapse, co-mutational MRD dynamics remain poorly understood. Heme-STAMP (Stanford Actionable Mutation Panel for Hematolymphoid Malignancies) MRD is a novel, bioinformatically optimized variation of the 203-gene Heme-STAMP next generation sequencing (NGS) panel designed to detect ultra-low variant allele frequencies (VAFs). To determine whether the co-mutational MRD landscape provides prognostic value beyond NPM1 MRD alone in NPM1-mutated AML, we retrospectively analyzed clinical outcomes in NPM1-mutated AML patients treated at Stanford Health Care from 2019 to 2026 with Heme-STAMP MRD testing.
Methods:
Adult patients with NPM1-mutated AML at diagnosis and at least two Heme-STAMP MRD tests performed were included in this retrospective cohort. The Heme-STAMP MRD assay sequences samples to ~5000x depth before deduplication and error correction using unique molecular identifiers. Its clinically validated 95% confidence limit of detection is 0.125% VAF for an average patient with 3 tracked variants, at a specificity of 98%. DTA variants were excluded from all MRD analyses. Variants were classified as cleared (detectable at diagnosis, clearing at subsequent timepoint), persistent/re-emergent (detectable at diagnosis and continuously present, or cleared then reappearing; P-RE), or newly emergent (undetectable at diagnosis but appearing on follow-up; NE). Overall survival (OS) and relapse-free survival (RFS) were estimated by the Kaplan–Meier method and compared using the log-rank test. All analyses were performed in R (v4.3.1), with p<0.05 considered statistically significant.
Results:
The analysis cohort included 54 patients with diagnostic Heme-STAMP, MRD assessment, and adequate clinical data. Of these, 34 (63%) cleared NPM1 MRD by Heme-STAMP, of whom 31 (91%) were also negative by Invivoscribe. Consistent with prior reports, patients with P-RE NPM1 MRD (by Heme-STAMP or Invivoscribe) had worse OS (15.2 months vs not reached [NR], p=0.001) and RFS (11.5 months vs NR, p<0.001) than those who cleared NPM1 MRD.
Focusing on co-mutational MRD status, both OS (40.0 months vs NR, p=0.013) and RFS (15.9 months vs NR, p<0.001) were significantly worse in patients with P-RE or NE non-NPM1 variants compared to those who cleared all variants. To test the relevance of ultra-low VAF detection, we reclassified variants below VAF of 2% (the traditional threshold) as undetected; the decreased RFS in patients with P-RE or NE variants persisted (13.4 months vs NR, p=0.002), whereas the OS difference did not (68.3 months vs NR, p=0.097). Importantly, this signal persisted even among the 31 patients who cleared NPM1 MRD. Patients with P-RE or NE co-mutations had significantly worse OS (40.0 months vs NR, p=0.047) and RFS (39.8 months vs NR, p=0.018) than those who cleared all co-mutations. At the gene level, patients with poor outcomes had persistence or emergence of secondary-type (SF3B1, RUNX1, SETD2, U2AF1), signaling (NRAS, CBL, FLT3, PTPN11), and/or TP53 mutations.
Finally, grouping variants by six functional categories rather than by AML ontogeny, clearance differed by category: chromatin/cohesion and signaling variants cleared most efficiently, whereas splicing and DNA damage repair variants persisted across all patients regardless of induction intensity. Interestingly, transcription factor variants persisted more frequently after low-intensity than high-intensity induction. This was driven largely by RUNX1 and CEBPA variants, consistent with prior reports implicating RUNX1 in venetoclax resistance in AML.
Conclusions:
To our knowledge, this is the first study to characterize co-mutational MRD dynamics with an ultra-sensitive, multigene NGS panel in NPM1-mutated AML and to show such co-mutations carry independent prognostic significance. Patients with P-RE or NE non-NPM1 variants had elevated risk of relapse and death, including among those who cleared NPM1 MRD and would otherwise be considered to have a favorable response. Clearance was further shaped by variant function and treatment intensity. Together, these findings suggest co-mutational MRD provides prognostic information beyond NPM1 MRD alone and may refine risk stratification and guide therapeutic decision-making in NPM1-mutated AML.
DISCLOSURES
Morgan Shannon, MD: No relevant conflicts of interest to declare.