High-Sensitivity Genomic Minimal Residual Disease Detection Reveals Multiclonal Hematopoiesis and Is Associated with Survival in Adult AML

Introduction: Minimal residual disease (MRD) measurement is emerging as a prognostic marker of relapse-free survival (RFS) and overall survival (OS) in acute myelogenous leukemia (AML). In order to improve the technical feasibility of MRD detection and to enable broad, ultra-sensitive mutational profiling of residual AML, we optimized a droplet-based digital PCR (ddPCR) approach capable of detecting mutated alleles in most recurrently mutated genes in AML in as few as 5 in 100,000 cells. We subsequently analyzed bone marrow specimens of AML patients in early complete remission (CR) and correlated MRD levels with outcome.

Methods: Fifty-two consecutively enrolled patients with AML at a single center who had paired diagnostic and CR bone marrow specimens were studied. First, genomic DNA from FACS-purified leukemic cells of diagnostic samples underwent Sanger resequencing of NPM1, IDH1, IDH2, NRAS, KRAS, FLT3, TP53, DNMT3A, TET2, ASXL1, BCORL1, CEBPA, and RUNX1. For cases with <2 mutations identified, an additional 36 recurrently mutated genes in AML were sequenced using droplet-based library preparation and massively parallel sequencing. Second, allele-specific fluorescent oligonucleotide probes using minor groove-binding or locked nucleic acid chemistries were developed for each mutation and optimized for allele discrimination using quantitative PCR. ddPCR (RainDance Technologies) was then performed on replicate DNA samples extracted from Ficoll-enriched bone marrow aspirates obtained in CR (median time from induction: 36d) to determine variant allele frequency (VAF) for all mutations.

Results: In CR bone marrow samples, we detected residual gene mutations in 88% of cases (46 of 52) using a detection threshold of 0.005% VAF. Notably, DNMT3A and TET2 mutations persisted at >1% VAF in 85% (11 of 13) (range 1.6-34.6%) and 50% (5 of 10) (range 2.3-15.1%) of cases, respectively, including in patients with subsequent durable remissions (>1000 days). Conversely, 94% (15 of 16) and 100% (5 of 5) of NPM1 and NRAS mutations, respectively, were reduced to <1% VAF, and 44% (7 of 16) of NPM1 mutations and 60% (3 of 5) of NRAS mutations were undetectable.

In cases for which ≥2 mutations were measured (N=28), two distinct patterns of residual allele burden were identified. In 32% (9 of 28) of cases, the VAFs of all genes measured were concordant (<1 log difference in VAF), but importantly, in 68% (19 of 28) of cases one or more VAFs were discordant (≥1 log difference), indicative of separate residual clones. In discordant cases, DNMT3A was the most common mutated gene of highest VAF (10 of 19, 53%), while NPM1 was the most frequent gene mutation of lowest VAF (10 of 19, 53%).

Kaplan-Meier analysis of all patients in first CR (N=48) was performed using a cutoff of 0.01% VAF to define MRD_high (N=33) and MRD_low/neg (N=15) groups. Due to frequent high-level persistence, DNMT3A mutations were excluded from the analysis. In cases where ≥2 mutations were measured, the highest VAF was used to assign that case. MRD burden was associated with RFS (MRD_high, median RFS 9.9 mo; MRD_low/neg, median RFS not reached; p <0.01) and OS (MRD_high, median OS 17.8 mo; MRD_low/neg, median OS not reached; p <0.01). In the exploratory subset of patients who proceeded to allogeneic stem cell transplantation in first CR (N=13), all patients who were MRD_low/neg (N=5) remained in a long-term remission, while 75% of MRD_high (6 of 8) patients relapsed (median RFS 14.8 mo; p =0.01).

Conclusions: High sensitivity measurement of residual clonal hematopoiesis with digital PCR is feasible for the majority of recurrently mutated genes in AML. Most DNMT3A and some TET2 and IDH1 mutations persist at high levels in remission even in patients with subsequent durable clinical remissions, further refining the hierarchical acquisition model of gene mutations in AML and implying they are present in pre-leukemic or non-blast differentiated cell populations. Due to high MRD levels, mutations in DNMT3A and possibly TET2 and IDH1 may not be suitable for use in MRD-based prognostication. The majority of cases with multiple measured gene mutations demonstrate discordance in VAF suggesting the simultaneous persistence of chemosensitive, chemoresistant and pre-leukemic populations. Ongoing work aims at identifying which mutation-defined subclones are responsible for relapse. Finally, in this cohort mutation-based MRD measured at initial CR is associated with RFS and OS.