3261 – HIGH RESOLUTION IDENTIFICATION OF LEUKEMIC CELL STATES USING A MULTIMODAL MANUAL CURATION FRAMEWORK

Abstract

Acute myeloid leukemia (AML) is a heterogeneous disease of the hematopoietic system characterized by the accumulation of dysfunctional stem and progenitor cells in the bone marrow. Relapse is driven by minimal residual disease, largely composed of leukemic stem cells (LSCs) that regenerate the leukemic hierarchy and confer chemotherapy resistance. Recently, tandem duplications (TD) of the upstream binding transcription factor (UBTF) gene have emerged as a common genomic alteration in pediatric AML and have also been identified in adults. The mechanisms driving UBTF-TD AML remain poorly understood and are further complicated by cellular heterogeneity, which shapes therapy resistance. High resolution characterization of UBTF-TD AML cell states is therefore required. We benchmarked Seurat and single-cell ANnotation using Variational Inference (scANVI)-based reference mapping in healthy hematopoiesis and AML as a performance baseline. Reference mapping reliably identified differentiated lineages but showed limited accuracy for progenitor populations, especially in AML. To address this, we designed a multimodal manual framework that integrates data-driven clustering, gene set enrichment, and profiling of surface markers. When applied to a cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) cohort of six patients with AML UBTF-TD (31,256 cells) spanning diagnosis and relapse, this framework enabled the resolution of leukemic progenitor cell states beyond the resolution achievable by reference mapping alone. Notably, an antigen-presenting hematopoietic stem cell population and a transcriptionally distinct pediatric AML-like subpopulation were characterized. Compositional analysis uncovered that AML UBTF-TD cellular composition is more primitive at relapse than at diagnosis, which is consistent with the known role of LSCs in disease recurrence. Compared with reference mapping based annotation, the manual curation recapitulated compositional patterns and cell states while providing additional biological resolution.

Acute myeloid leukemia (AML) is a heterogeneous disease of the hematopoietic system characterized by the accumulation of dysfunctional stem and progenitor cells in the bone marrow. Relapse is driven by minimal residual disease, largely composed of leukemic stem cells (LSCs) that regenerate the leukemic hierarchy and confer chemotherapy resistance. Recently, tandem duplications (TD) of the upstream binding transcription factor (UBTF) gene have emerged as a common genomic alteration in pediatric AML and have also been identified in adults. The mechanisms driving UBTF-TD AML remain poorly understood and are further complicated by cellular heterogeneity, which shapes therapy resistance. High resolution characterization of UBTF-TD AML cell states is therefore required. We benchmarked Seurat and single-cell ANnotation using Variational Inference (scANVI)-based reference mapping in healthy hematopoiesis and AML as a performance baseline. Reference mapping reliably identified differentiated lineages but showed limited accuracy for progenitor populations, especially in AML. To address this, we designed a multimodal manual framework that integrates data-driven clustering, gene set enrichment, and profiling of surface markers. When applied to a cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) cohort of six patients with AML UBTF-TD (31,256 cells) spanning diagnosis and relapse, this framework enabled the resolution of leukemic progenitor cell states beyond the resolution achievable by reference mapping alone. Notably, an antigen-presenting hematopoietic stem cell population and a transcriptionally distinct pediatric AML-like subpopulation were characterized. Compositional analysis uncovered that AML UBTF-TD cellular composition is more primitive at relapse than at diagnosis, which is consistent with the known role of LSCs in disease recurrence. Compared with reference mapping based annotation, the manual curation recapitulated compositional patterns and cell states while providing additional biological resolution.