Single Cell Sequencing of Pediatric Acute Myeloid Leukemia Reveals Clonal Evolution to Relapse on Combination Chemotherapy with Sorafenib

Introduction: Relapse of pediatric acute myeloid leukemia (AML) remains a leading cause of childhood cancer mortality, and leukemias with activation of the Fms-like tyrosine kinase 3 (FLT3) are particularly susceptible to relapsed disease. Risk-directed therapy to prevent relapse is based both on genetic changes known to drive drug resistance, and measurable residual disease (MRD) at the end of induction therapy (EOI). In adult AML, resistance to type II FLT3-inhibitors, like sorafenib, is primarily driven by on-target FLT3 kinase domain (KD) mutations. However, the resistance mechanisms for pediatric leukemias, which are treated on combination therapies, have not been fully elucidated. MRD is considered the among the most predictive markers of future relapsed disease. It has been assumed that the major clone at the time of MRD assessment will predict the majority clone at relapse. However, this assumption has not been proven. The definition of the most specific genetic and MRD markers of relapse are essential to prognosticate and personalize therapy to prevent relapsed disease.

Methods: We performed single cell sequencing (SCS) with a high-throughput DNA sequencing platform, Mission Bio Tapestri, on bone marrow or peripheral blood samples from 24 samples from 8 pediatric patients treated on COG AAML1031 with serial samples from diagnosis, EOI, and relapse.

Results: We analyzed a total of 94,833 cells from 8 pediatric patients (median cells per patient 12,428) all treated on AAML1031. SCS revealed a sensitive and specific description of clonal evolution on the combination of sorafenib with cytotoxic chemotherapy. The FLT3 internal tandem duplication (ITD) was controlled by the therapy in only half of the patients. In five of the patients, the FLT3-ITD was present in multiple clones. The FLT3-ITD co-mutated with additional mutations (NRAS, SH2B3, WT1, TET2, or NPM1) in half of the patients. However, the presence of a co-mutation did not necessarily correlate with whether or not the ITD-containing clone persisted at the time of relapse. Of the leukemias whose relapse was not driven by FLT3, the most likely mutational driver of resistance was NRAS. Notably, however, despite the fact that FLT3 KD mutations make up the bulk of mutational resistance to type II FLT3i such as sorafenib in adult patients, there were no on-target FLT3 mutations found in any of these pediatric patients. Further, SCS allows for an unprecedented depth of analysis of the genetic complexity of pediatric AML. Phylogenic analysis revealed that the same mutations may arise independently in different cells (NPM1 W288fs, NRAS G60E). Additionally, the same gene may be mutated twice within the same cell (WT1, TET2). These data, consistent with our prior work, suggest that some leukemias may have a predilection to mutations within specific loci. Finally, although there is a standing assumption that the dominant MRD population will proliferate into relapsed disease, in 3/8 patients, the dominant MRD clone did not predict the dominant relapse clone.

Conclusions: SCS allows for direct measurement of clonal hierarchy and evolution, phylogeny, co-mutational status, and zygosity, which can only be inferred through traditional bulk NGS. The mutational mechanisms of resistance seen in adult leukemias treated with sorafenib monotherapy are not necessarily relevant to the pediatric population; rather than on-target FLT3 mutations, off target mutations including NRAS are found. This corroborates prior findings that off-target RAS pathway mutations may drive resistance to FLT3i. Non-RAS off-target mutations found in this cohort do not necessarily predict sorafenib resistance, so may be passenger mutations. The lack of consistent resistance mutations suggests that other mechanisms of resistance such as epigenetic modifications may also drive resistance to combination chemotherapy with FLT3i in pediatric leukemia. Further, SCS exposes more genetic complexity in pediatric AML than has previously been appreciated: the same mutation may independently arise in more than one cell or the same cell may have multiple mutations within the same gene. Finally, the sensitivity of SCS reveals that the major clone at the time of MRD assessment is not necessarily the major clone at relapse. This suggests a benefit of more frequent MRD monitoring to track clonal evolution in real time.