ICASA: An Interactive Comprehensive Atlas of Single Cells in AML from 82 Bone Marrow Samples

Background: Acute myeloid leukemia (AML) is a heterogeneous disease characterized by arrested differentiation and clonal expansion. While certain efforts have been devoted to integrating single-cell data in AML, there is a clear need to build an interactive atlas fully annotated with clinical, demographic, molecular, cytogenetic and treatment data, providing comprehensive sources and customized functions for analysis. To mend this gap, we developed an interactive Comprehensive Atlas of Single cells in AML (iCASA) platform which can enhance the understanding of the cellular communication, transcriptomic regulation, immune repertoires, and cell population changes over the course of AML evolution in various treatment settings. Methods: Single cell RNA profiling (scRNAseq) of 73 bone marrow (BM) samples from 43 AML patients and 9 healthy BM samples including our inhouse previously published and newly generated data were included. Batch effects were evaluated by Harmony. Cell types were annotated by integration of canonical markers, flow cytometry, conventional cytogenetics, inferCNV, FISH, and manual revision. Reference mapping of AML cells was projected by Symphony. Data-sharing platform Shiny App was built in R for interactive use. Results: AML BM samples included 22 newly diagnosed patients who received venetoclax-based therapy, 13 newly diagnosed patients who received frontline therapy without venetoclax and 8 relapsed/refractory patients. Of those, 19 patients had paired samples (2 to 3) pre and post therapy, providing longitudinal assessment of AML-immune cell interactions. The median age of the population at time of diagnosis was 71 years (range 38 - 89). The most frequent mutation was TP53 (39%), followed by IDH2 (23%), and DNMT3A (23%) and 44% patients had del7/7q. After preprocessing, 404,819 high-quality cells were retained. To facilitate the public utility of our resource and drive future discoveries, the iCASA features were divided into 8 tabs: 1) Dimensionality reduction visualizes distribution of cells by UMAP or tSNE where pathways of interest can be projected to assess biological functions in different cellular states. 2) Differential expression analysis identifies differential genes across cell types, normal versus tumor, with interactive visualization such as heatmaps and volcano plots. 3) Correlation analysis performs pairwise gene correlation analysis at pseudo-bulk level for modules of interest. 4) Signature Scoring of user-provided gene sets or predefined pathway signatures enable users to look at pathway activities across different cell types and clinical subgroups. 5) Survival analysis provides overall survival curve based on gene expression, pathway scores, or gene regulator modules (including transcription factor activity) 6) Cell-cell communication predicts ligand-receptor interaction between cell types by CellChat. Visualization includes circle plot, chord diagram, and heatmap. 7) Trajectory analysis determines the dynamic process of cell differentiation. 8) Immune repertoires (scTCR and scBCR) paired with immune cell subtypes to examine markers of immune response. The iCASA platform will be available on a free web-access portal for users to explore. The interactive interface makes it accessible even to those without prior experience in computational biology. Additionally, users can download the raw data for use on local machines. As an application of the portal, we identified 12 broad cell clusters and 38 specific cell states spanning hematopoietic stem cells to differentiated cells. T cells were the dominant lineage in the niche (56%). AML cells were mapped by a linear mixture model to hematopoietic cells from 9 healthy donors to determine their cellular states. The overall cellular states showed significant diverse distribution across patients (p < 2.2x10-16), indicating high intra-tumoral heterogeneity of AML patients. The web portal is now in demo mode and expected to go live in January 2025. Conclusions: We provided a large scRNAseq resource of hematopoietic differentiation in AML patients. This integrated information is a valuable resource to characterize dynamic changes in AML patients with chemotherapy, understand response to combined therapies. Ongoing analyses to derive biological insights include identifying meta clusters associated with cellular states, evaluating dynamic changes in different myeloid subtypes.