A Longitudinal Single-Cell Atlas of Treatment Response in Pediatric AML

Introduction:

Pediatric acute myeloid leukemia (pAML) is a heterogeneous disease in terms of driver alterations, treatment response and patient outcomes. Although approximately 90% of pAML patients respond to initial treatment, tumours relapse frequently. Despite large-scale genomic sequencing efforts, it remains elusive why tumors relapse. Previously, mutations have been identified that affect transcriptional regulation, indicating that epigenetic processes may contribute to treatment resistance in pAML. We therefore studied how tumor epigenetic patterns, transcriptional regulation and cell populations change over the course of treatment using single-cell ATAC sequencing (scATAC-seq) and single-cell RNA sequencing (scRNA-seq) in pediatric populations.

Methods:

We profiled 330,047 cells using scRNA-seq and 353,984 cells using scATAC-seq derived from 28 uniformly treated pAML patients, driven by either MLL fusions, CBFB fusions, RUNX1 fusions, FLT3-ITD or other aberrations (i.e. CEBPA mutations). All patients were enrolled in the AAML1031 trial (https://childrensoncologygroup.org/aaml1031), a randomized phase 3 clinical trial that investigated the addition of Bortezomib (BTZ) and Sorafenib to standard treatment modalities. Samples were obtained serially at diagnosis, at remission and at relapse and single-cell profiles were compared between these stages.

Results:

We identified malignant and non-malignant populations in the scATAC-seq and scRNA-seq data, showing that in all cases, malignant cells from patients at diagnosis were distinct from those taken at relapse. Major differences between diagnosis and relapse were associated with differentiation, and by inferring distinct cell populations we identified that relapsed cell populations appeared to shift towards a more primitive state. The extent of this shift was dependent on the genetic subtype and the effects were more pronounced in MLL-rearranged tumors. To identify the candidate mechanisms and transcription factors involved in this shift we studied the monocytic, lymphocytic and erythrocytic lineages in non-malignant cells and compared these to the tumor at diagnosis and relapse. Based on motif enrichment we found that across all different subgroups, transcription factors (TFs) that drive monocytic differentiation (i.e. CEBPA) have a lower activity while TFs driving erythrocytic and lymphocytic differentiation (i.e. TCF3) have a higher activity. Analysis of hematopoietic stem-like and progenitor-like cells (HSPC-like) confirmed this change, indicating that the “priming” towards other lineages is not only a result of shifts in cell populations but also a result of rewiring of early progenitors. To investigate this rewiring, we analysed regions of open chromatin that were enriched in malignant HSPCs compared to normal HSPCs across different subgroups. We found that those regions were highly specific to the driving alteration and were enriched for MEF2C binding sites in MLL-driven tumors and AP-1 binding sites in other tumors. Overall, upon relapse MEF2C expression and inferred TF activity appeared to increase, coinciding with an apparent increase in lymphoid priming and early lymphoid factors including LMO2, LYL1 and TCF3, among others. Two cases in the cohort appeared to switch lineage from a monocytic to lymphocytic phenotype and were enriched in CD19 and CD79A positive malignant cells upon relapse. Both cases shared the characteristic that MEF2C was upregulated upon relapse and showed upregulation of typical B-lymphoid transcription factors such as PAX5.

Discussion:

By generating a large dataset of pediatric AML patient samples collected at diagnosis, remission and relapse, we have identified changes in cellular hierarchies and transcriptional networks that occur upon relapse. Our study shows that shifts in cell populations readily occur upon treatment and are dependent on subgroup-defining alterations. Particularly MLL-rearranged tumors show a high level of plasticity that could explain the poor outcome of this group. Furthermore, we show that tumors appear to adapt their transcriptional programs towards a more “stem-like” state, with a likely higher propensity to differentiate towards other lineages such as the lymphoid lineage. We have identified MEF2C as a possible regulator of these processes, opening a new perspective to target the plasticity of these tumors.