Accurate and Sensitive Analysis of Minimal Residual Disease in Acute Myeloid Leukemia Using Deep Sequencing of Single Nucleotide Variations

The prognostic importance of minimal residual disease (MRD) for treatment response evaluation in acute myeloid leukemia (AML) is evident both after conventional chemotherapy and before and after stem cell transplantation. The genetic landscape of AML is characterized by numerous somatic mutations, which constitute potential markers for MRD. We have recently demonstrated that a patient-tailored MRD analysis can be performed through identification of leukemia-specific mutations with exome sequencing at diagnosis, followed by targeted deep sequencing during follow-up (Malmberg, E et al, Eur J Haematol 2017;98:26-37). The development of next generation sequencing based MRD analysis in AML has been hampered by limited sensitivity, mostly due to general sequencing errors as well as sequence-specific errors. The objective of this study was to optimize, validate and clinically test the targeted deep sequencing assay for leukemia-specific substitution mutations (single nucleotide variations, SNVs). By applying a statistical model that minimizes sequencing-specific errors using a reference sample, we achieved a limit of detection (LOD) for SNVs with a variant allele frequency (VAF) of 0.018% (15 SNVs in normal samples). The LOD inversely correlated to the sequencing depth, with 0.018% reached at 5.0x10⁵ reads. Precision, accuracy and linearity were determined using a commercial standard with set levels of mutations (VAF 1 and 8%). The median coefficient of variation (CV) was 4.1% (range 2.2-5.7%, 4 SNVs, n=3) at VAF 1% and 13.3% (8.8-19.4%) at VAF 0.1% with a bias of 7.9% (2.5-15.3%, VAF 1%). At VAF 8%, the CV was 3% and bias 2.8% (1 SNV, n=24). Linearity in the range of MRD (VAF 0.03%-1%) was confirmed for 5 SNVs. The deep sequencing method was applied to a series of 34 bone marrow (BM) samples from 6 children with AML and compared to 8 color multiparameter flow cytometry (MFC) according to guidelines in the study protocol NOPHO-DBH AML-2012, and when applicable to quantitative reverse transcription-polymerase chain reaction (RT-qPCR). Leukemia-specific somatic mutations (2-3/case) were identified in diagnostic samples using exome sequencing of sorted leukemic cells and lymphocytes and then assessed in each sample using deep sequencing. When we compared deep sequencing results with results obtained with MFC, there was a strong correlation (R_s 0.90, p < 0.001). When we dichotomized the results into MRD⁺ and MRD^- results according to each method, 18 of 27 samples (and 53/70 mutations) showed concordant results with 12 samples (31 mutations) MRD⁺ with both methods, 6 samples (22 mutations) MRD^- with both methods. Nine samples (17 mutations) were MRD⁺ with deep sequencing whilst MRD^- with MFC. No samples/mutations were MRD^- with deep sequencing but MRD⁺ with MFC. When comparing results from deep sequencing with results from RT-qPCR (RUNX1-RUNX1T1, KMT2A-MLLT10), 13 of 14 samples showed concordant MRD assignments. To explore the applicability of the method for MRD detection in peripheral blood, we analyzed blood samples from the same time points as MRD⁺ BM samples (27 mutations in 12 samples, VAF in BM 0.018%-5%). There was a significant correlation between VAF in blood and BM (R_s 0.65, p < 0.001), at least one mutation was detected in 10 out of 12 investigated blood samples and all mutations with VAF>0.1% in BM (19/27) were MRD⁺ in blood. In conclusion, targeted deep sequencing of single nucleotide variations in recurrently as well as non-recurrently mutated genes enables accurate and precise detection of low levels of residual leukemic cells with a superior sensitivity to MFC. Introduction of this method in the care of patients with AML will allow for sensitive leukemia surveillance in virtually every patient with AML.