Multimodal Single-Cell Profiling of Minimal Residual Disease in B Acute Lymphoblastic Leukemia Reveals Genetic and Non-Genetic Factors of Early Drug Resistance

Introduction

Cure rates for pediatric B-cell acute lymphoblastic leukemia (B-ALL) have dramatically increased due to improved treatment regimens. However, patients who relapse continue to have poor outcomes. Risk of relapse is highly correlated with the level of detected minimal residual disease (MRD) after induction therapy. Despite MRD level being the strongest prognostic indicator in B-ALL, how leukemic blasts evade chemotherapy is poorly understood. In this study, we used single-cell targeted DNA sequencing (scDNA-Seq) and multiome sequencing (scRNA-Seq + scATAC-Seq) to characterize genomic, transcriptomic, and chromatin accessibility changes after treatment and identify drivers of chemoresistance in B-ALL.

Methods

Bone marrow aspirates were obtained from 41 pediatric patients across 4 high-risk B-ALL subtypes: KMT2A-rearranged, Ph+, Ph-Like, and iAMP21. Paired initial diagnosis (IDX) and MRD timepoint samples from these patients were used to produce simultaneous scRNA-Seq and scATAC-Seq data. A subset of 24 patients were profiled using scDNA-Seq targeted at 355 recurrently mutated sites in pediatric B-ALL.

Results

In the multiome dataset, we identified 357,235 leukemic blasts from 442,710 total cells using multiple lines of evidence, including marker expression and copy number variant calling. To understand how induction therapy affected arrest states, we projected cells onto a healthy pediatric bone marrow reference to identify patient-specific developmental states. Across the cohort, we found that HSPC-like and Mature-B-like leukemic blasts increased post-induction while most blasts maintained a Pre-pro-B-like state. To determine whether these changes occurred uniformly, we clustered patient arrest state changes and found four patterns defined by proportions of cell states along the hematopoietic trajectory. Surprisingly, these patterns were uncorrelated with MRD level or genetic subtype. To understand how specific cell types respond to treatment within patient clusters, we conducted differential gene expression analysis across timepoints. We found that upregulated genes at the MRD timepoint in cell states with increased abundance were enriched for cluster-specific pathways, displaying differing phenotypic responses to induction treatment. scATAC-seq analysis of the same blasts revealed that IRF and NFKB transcription factor activity corresponds with arrest states increasing after treatment, identifying potential regulators of the chemoresistance phenotype.

Our cell state analysis also revealed that nearly half of patients displayed significant increases in myeloid-like blasts following treatment not limited to KMT2A-rearranged patients where myeloid lineage switch is well-documented. Myeloid-like blasts lowly express canonical B-cell immunophenotypic markers, including CD19, CD179b, and CD24. In contrast, these blasts express a unique set of markers with therapeutic relevance in acute myeloid leukemia, including CD83 and CD98. This increase in CD19-low, myeloid-like blasts following treatment suggests that induction therapy may promote expansion of populations that drive CD19-negative relapse after CAR-T therapy.

To understand how mutagenesis may drive chemoresistance, we analyzed a median of 2,346 cells at IDX and 1,828 cells at MRD to identify high-confidence single nucleotide variants using targeted scDNA-Seq. Mutations were validated orthogonally through variant calling using the multiome dataset. For certain patients, mutations known to be enriched at relapse, including NRAS, NR3C1, MSH6, and PMS2, were observed at diagnosis and post-induction, demonstrating the potential for these mutations to enable chemoresistance early in treatment. However, overall mutagenesis rates across the cohort were limited which reinforces the lack of a common mutational mechanism driving chemoresistance in B-ALL.

Conclusion

Our study presents the first comprehensive multiomics characterization of pediatric B-ALL minimal residual disease. Complementary multiomics data indicates that while mutagenesis is uncommon after induction in B-ALL, leukemic blasts undergo drastic transcriptomic and epigenomic shifts following induction therapy. Given the patient-specific nature of these responses to treatment, our findings underscore the need for more personalized treatment regimens for high-risk B-ALL patients.