MRD Detection in Standard-Risk Pediatric AML Using Single-Cell RNA Sequencing

Introduction: Despite advancements in chemotherapy, targeted therapies, and hematopoietic cell transplantation, pediatric acute myeloid leukemia (pAML) continues to have unacceptable outcomes. One challenge is accurately risk-stratifying patients to ensure they receive appropriately intense treatments. While genetic profiling has allowed for the classification of some children into high- or low-risk groups, patients without identifiable markers, known as "standard-risk," often do not benefit from tailored therapy upfront. For these patients, determining treatment strategy relies heavily on detecting and quantifying measurable residual disease (MRD) at the end of induction (EOI) chemotherapy. We hypothesized that single-cell RNA sequencing (scRNAseq) could more accurately identify MRD at EOI compared to the traditional method of multiparameter flow cytometry (MFC) and that the gene expression profile of MRD at EOI would predict relapse.

Methods: We studied 26 pediatric patients with standard-risk AML who had cryopreserved leukemia samples from both diagnosis and EOI. Of these, 13 patients had detectable low-level MRD at EOI by MFC, while the remaining patients were MRD negative. Samples were thawed, and those with less than 60% viability were subjected to dead cell removal. Seven samples (3 diagnostic, 4 EOI) had low viability (<50%) and cell number and could not be captured. Cells were stained with 141 CITE-seq antibodies and a hash antibody for sample identification. Diagnosis and EOI samples were mixed in a 1:10 ratio for genetic demultiplexing. Cells were captured on the 10X Genomics Chromium X platform using 5’ HT or GEM-X 5’ reagents, and libraries were prepared and sequenced according to protocol. Data analysis was performed using Seurat and Monocle workflows. Following QC, our data consisted of 475,301 cells for downstream analysis. UMAP visualization showed that healthy cells cluster together, while suspected leukemic cells from diagnostic samples formed distinct clusters. A reference of bulk RNA sequencing from 736 ALL and AML cases was used as a reference to annotate leukemic cells and predict subtypes. Using this approach, we found that the gene expression of suspected leukemic cells corresponded with the known molecular aberrations of each patient (KMTA fusions, FLT3 status, etc), consistent with their identification as leukemic disease. MRD estimates were then calculated based on the proportion of EOI cells in leukemic clusters.

Results: Across the cohort, single-cell RNA sequencing detected significantly more MRD than MFC (paired Wilcoxon, p = 0.0007). In patients with known low-level MRD by MFC, scRNAseq detected more disease in 8 of 10 patients, finding a median of 8.9 times more disease (MRD range 0.06-26%). For patients who were MRD negative by MFC at EOI (n=12), scRNAseq identified a median of 0.08% MRD, with a median of 0.02% in patients who experienced long-term remission (n =9, range 0-26.6%), 2.27% in patients who experienced relapse, (n = 2, range 0.03-4.5%), and 2.8% in 1 patient who died in CR1. Gene expression analysis revealed that the LSC17 “stemness” score was significantly higher (p = 2.2*10^-16) in residual disease from patients who experienced relapse. Additional gene and proteomic markers associated with relapse were identified, including genes CLEC11A, LYE6, HCST, LYZ, and FABP5 (logfold change 2.5-3.8, p = 0), and proteins CD31 and CD41 (logfold change 2.2, p = 0).

Conclusions: Single-cell RNA sequencing proved more effective than MFC in detecting MRD in pediatric AML with standard-risk genetic features. The technology also identified gene expression and proteomic profiles associated with relapse, suggesting that scRNAseq holds promise for identifying novel biomarkers that predict relapse in pAML.