Circulating Tumor DNA (ctDNA) Enables Superior and Universal Measurable Residual Disease (MRD) Monitoring in Acute Myeloid Leukemia (AML) Highly Predictive of Relapse Free and Overall Survival

Introduction: Standard-of-care methods for AML MRD detection (multiparameter flow cytometry and single gene or structural variant molecular assays) are hampered by (1) inadequate lower limits of detection (LOD), (2) few recurrent mutations across AML genomes to allow universal molecular monitoring including clonal evolution, and/or (3) need for invasive bone marrow biopsy. Herein we further develop AML-CAPP-Seq: a noninvasive ctDNA assay using a Cancer Personalized Profiling by Deep Sequencing (CAPP-Seq) approach to both quantify MRD and evaluate clonal dynamics in AML. Methods: To improve upon the above limitations, we designed patient-specific hybrid capture panels based on somatic variants identified by whole exome sequencing (WES) at time of AML diagnosis and combined them with a shared panel of 58 recurrently mutated (canonical) AML genes as a composite approach to generate personalized AML-CAPP-Seq assays for MRD detection. We applied AML-CAPP-Seq to track single nucleotide variants (SNVs) in 1043 biological samples from 56 AML patients over time (median follow up duration of 17.5 months and 5 timepoints per patient) and 21 healthy controls. We compared the performance of MRD detection in cellular (bone marrow mononuclear cell/BMMC, peripheral blood mononuclear cell/PBMC) and cell-free (PB-ctDNA and BM-ctDNA) compartments against standard-of-care (SOC) clinical MRD assays and patient outcomes. Results: We observed a median of 32.5 SNVs (IQR 21-42) per patient from WES compared to 2.5 (IQR 1-4) from the canonical panel alone. Increased SNV detection enabled sensitive MRD monitoring in 100% (56/56) of patients with a LOD of <0.01% tumor fraction. If limited to canonical genes, only 84% of patients had at least 1 SNV to allow MRD detection. Notably, just 32% of patients in this cohort had a feature amenable to SOC molecular MRD methods. Comparing quantitative levels of MRD in each biological compartment, tumor burden in PB-ctDNA, BM-ctDNA, PBMC and BMMC samples were highly correlated (r > 0.91) and overall similar. However, PB-ctDNA contained quantitatively more tumor DNA than corresponding PBMC and BMMC samples (p < 0.02) at low tumor burdens (median 3.8- and 2.9-fold enrichment when mean tumor allele fraction in PBMCs or BMMCs was < 1% respectively), most pronounced in patients with primarily extramedullary disease (n = 4). These results suggest that PB-ctDNA effectively captures total body molecular disease in AML and may serve as a superior analyte for MRD monitoring compared to both PBMCs and BMMCs. AML-CAPP-Seq detected residual disease in PB-ctDNA in 98.5% (208/211) of samples known to contain measurable disease by SOC marrow-based methods (100% (211/211) by flow, 88.9% (24/27) by single gene assay). AML-CAPP-Seq also identified +MRD in PB-ctDNA in 47.4% (64/135) of samples with undetectable MRD by SOC assays. Of these, 46.8% (30/64) were from patients who ultimately relapsed and 50% (32/64) were from patients yet to relapse but who subsequently attained MRD negative levels in PB-ctDNA after additional therapy. Among patients who achieved CR1 (n = 48), +PB-ctDNA MRD was identified in 89.3% (25/28) of patients who later relapsed, compared to 46.4% (13/28) by any SOC MRD method or 35.7% (10/28) by flow MRD alone. Accordingly, MRD in PB-ctDNA robustly predicted relapse free survival (RFS) (hazard ratio (HR) = 11.9, p = 7e-8) versus SOC (HR = 3.4, p = 5e-4). Moreover, attaining an MRD negative CR in PB-ctDNA at any timepoint was strongly prognostic for overall survival (HR = 19.7, p = 5e-9) compared to SOC MRD (HR = 5.2, p = 4e-6). For patients (n = 27) receiving allogeneic stem cell transplant (SCT), risk stratification based on not only pre-SCT +MRD in PB-ctDNA but also post-SCT MRD clearance significantly predicted RFS (p = 4e-4), suggesting that ctDNA kinetics may be a valuable clinical metric in the peri-transplant setting. Finally, by tracking dozens of tumor variants per patient including indels and copy number variants, AML-CAPP-Seq allowed time-series resolution of complex clonal architecture dynamics in both cell-free and cellular compartments during treatment. Conclusions: AML-CAPP-Seq enables ultrasensitive detection of ctDNA-MRD superior to standard MRD methods without need for invasive BM assessment. Personalized SNV tracking enhances assay performance and allows universal MRD tracking for all AML patients while permitting concomitant profiling of clonal evolution in ctDNA.