Mathematical Modeling of Molecular Relapse Kinetics in NPM1c+, PML-Rara+, RUNX1-RUN1T1+, and CBFB-MYH11+ Acute Myeloid Leukemias

Introduction: Closely monitoring minimal residual disease (MRD) in acute myeloid leukemia (AML) patients in complete remission (CR) holds the promise of detecting relapses in time to initiate cytoreduction prior to a full-blown hematological relapse. While the literature has demonstrated that fluorescence-based quantitative PCR (RQ-PCR) protocols are excellently suited for this purpose, standardized measures for determining sampling intervals and the relative value of peripheral blood (PB) v. bone marrow (BM) in detecting MRD are still lacking. AIM: To quantify key parameters related to relapse kinetics and through mathematical models enable comparisons between different molecular markers as well as between BM and PB. PATIENTS AND METHODS: We analyzed 838 patients (pts) with the four most prevalent RQ-PCR markers and selected patients who fulfilled the following criteria:

• pts should achieve CR,

• pts should experience a hematological relapse (HR),

• pts should not have received treatment upon molecular relapse (MR) before the development of the HR, and

• samples should be available from within the last 12 months prior to HR.

Sixty-eight pts were included with the following characteristics: 31 NPM1c+(32 relapses and 70 CR samples), 14 PML-RARA+ (14 relapses and 47 CR samples), 12 RUNX1-RUNX1T1+ (13 relapses, 46 CR samples) and 13 CBFB-MYH11+ (13 relapses, 59 CR samples). To analyze these data we devised a mathematical model, which exploits the information from both positive and negative samples taken prior to relapse. The model determines the relationship between sampling intervals and the likelihood of finding a patient positive prior to HR on the one hand, and the median time from MR to HR (t_m) on the other.

RESULTS: We observed strikingly different relapse kinetics in AML subgroups. Thus, MR was detected at the earliest in CBFB-MYH11+ pts with 50% of the pts positive in BM 8 months prior to HR. By contrast, 50 % had MR 5 months prior to HR in NPM1c+ pts, 4 months in RUNX1-RUNX1T1+ pts, and only 3 months prior to HR in PML-RARA+ pts. Thus, with BM sampling every third month, very different relapse detection frequencies (RDFs) and t_m’s are achieved, ranging from 70% and 45 days for PML-RARA+ relapses to 100% and 200 days for CBFB-MYH11+ relapses. Of importance, PB and BM were mostly comparable for these four gene alterations. This contrast to data on MRD detection based on over-expression of the Wilms’ tumor 1 gene (WT1), where WT1 analysis in BM (3 months RDF 100%, t_m 85 days) is clearly more predictive than PB analysis (3 month RDF 65%, t_m 40 days) (

CONCLUSION: The quantitative description of relapse kinetics provided in this data set should be useful in cost-benefit calculations regarding MRD monitoring implementation, in clinical decision-making in the individual patient, and in statistical power calculations when designing clinical trials.

Guidelnes for MRD follow-up using CBFB-MYH11, RUNX1-RUNX1T1, PML-RARA, NPM1c or WT1 as molecular markers