Session: 618. Acute Myeloid Leukemias: Biomarkers and Molecular Markers in Diagnosis and Prognosis: Poster I
Hematology Disease Topics & Pathways:
Research, Fundamental Science, Translational Research, Bioinformatics, Computational biology, Technology and Procedures, Machine learning
To facilitate this interpretation, the sorted CD34+ bone marrow mononuclear cells (BMMCs) from one pAML with MLLr and one pAML with FLT3-ITD aberrations were initially processed over the 10x Genomics Chromium platform, respectively. The generated Gel Bead-in-Emulsions (GEMs) were divided into two fractions: one was sent for high-throughput droplet-based 5′-single cell RNA-sequencing (scRNA-seq) process, while the other was reserved for Oxford Nanopore long-read library construction and sequencing. To provide a reference framework, single-cell transcriptomic profiles from CD34+ BMMCs of two healthy donors (HDs) were employed and annotated with 14 differential subclusters, based on stage-specific marker genes.
By projecting the transcriptional signatures from short-read scRNA-seq onto the previously annotated differential roadmap, CD34+ BMMCs from the two pAML patients predominantly exhibited progenitor and promonocyte/monocyte (Promono/Mono) clusters, aligning with the established understanding differentiation blockade in AML. Noteworthy was the differential inclination across the two subtypes: the FLT3-ITD pAML demonstrated a marked propensity towards stem and progenitor cells, whereas the MLLr subtype did not display this tendency. This distinction underscores the heterogeneity in cellular hierarchies and lineage commitment disruptions accross different genetic subtypes of AML.
To confirm the leukemic identities with genetic variations, we analyzed single-cell long-read sequencing data from the second fractions of GEMs, which shared barcodes with cells used for scRNA-seq from the first fractions. In leukemic cells of MLLr pAML, we identified the ELL::KMT2A rearrangement with translocation from breakpoints of the exon 1 of ELL and exon 10 of KMT2A. Meanwhile, leukemic cells from the FLT3-ITD pAML exhibited a tandem repeat of 27 bases within the exon 14 of FLT3 gene. Utilizing 80% of all isolated cells as the training set and the remaining 20% as the test set, we constructed a machine learning model by employing the molecular characteristics of captured leukemia to build a classifier to distinguish leukemia cells from cells without variation identified. Notably, the area under the curve (AUC) of the MLLr leukemic classifier and FLT-ITD altered leukemic classifier both satisfactorily exceeded 0.99, demonstrating their high discriminative performance and reliability in identifying leukemia cells with these specific genetic alterations. Based on internal analysis of the top hits in the classifiers from clinical follow-up data, high MT-ND1 expression in MLLr pAML, and CD99 as a leukemic stemness marker in FLT3-ITD pAML, were further validated to be closely associated with significantly inferior event-free survival and overall survival rate from external TARGET-AML cohort.
Our study, leveraging the long-read and short-read single cell sequencing strategies, had profiled full-length transcripts and simultaneously facilitated the reliable detection of gene variations within individual cells, providing a more robust platform for elucidating the heterogeneity and pathogenic mechanisms of different leukemic subtypes. This integration enabled the construction of accurate classifiers to identify leukemia cells based on their subtype-specific molecular characteristics, such as upregulated MT-ND1 and CD99, proposing insights for prognostic assessments and targeted therapies in pAMLs in further investigations.
Disclosures: Yang: Takeda: Honoraria.