Spatial single-cell multiomics of the bone marrow microenvironment in NPM1-mutant Acute Myeloid Leukemia Free

Background:NPM1-mutant acute myeloid leukemia (AML) is the most common molecular subtype of AML and generally responds well to intensive chemotherapy. However, patients with refractory/relapsing disease continue to show dismal outcomes, emphasizing the need for improved biomarkers for stratification and the discovery of key resistance mechanisms. While signatures of immature hematopoietic stem cell-like cell populations and monocytic differentiation have been linked to treatment resistance on the transcriptomic level, it remains unclear how the spatial interactions of individual leukemic subpopulations, immune cells and stromal elements in the bone marrow microenvironment (BMME) impact treatment responses and outcomes.

Methods: We established a 60-plex antibody panel for CO-Detection by indEXing (CODEX) highly multiplexed microscopy of formalin-fixed, paraffin-embedded (FFPE) BM trephine biopsies. To map the BMME in NPM1-mutant AML, we included markers for different hematopoietic lineages, lymphocyte subsets, stromal cells and a mutant-specific anti-NPM1 antibody to visualize leukemic cells across differentiation states. Diagnostic BM biopsies from 45 NPM1-mutant AML patients and 39 matched controls with normal BM (NBM) were collected and tissue microarrays compiled for high-throughput analysis. A total of 1.246.585 cells were imaged, segmented and annotated. After filtering, 1.155.110 high-quality cells remained in the dataset. To identify cell types with high confidence, we combined machine-learning approaches including reference-based classification with unsupervised clustering and manual annotation. In addition, we performed spatial transcriptomics analysis using Multiplexed Error-Robust Fluorescence In Situ Hybridization (MERIFSH) technology on selected cases from the same FFPE sample cohort previously analyzed by CODEX. We analyzed 7 AML cases using a targeted panel of 815 genes implicated in immune response, inflammation and oncogenic signaling. Data analysis was conducted using Merscope Visualizer software and custom Python-based pipelines to integrate single-cell gene expression profiles with cell type identification and cellular neighborhood analyses.

Results: In the CODEX dataset, we identified 35 healthy cell subtypes, including myeloid, lymphoid, erythroid, vascular and stromal cells, megakaryocytes, and hematopoietic stem/progenitor cells, and 10 different leukemic subpopulations. There were striking differences in the differentiation state of leukemic subpopulations between and within individual patients, such as phenotypically immature, monocytic and granulocytic subtypes. Spatial analysis of NBMs revealed 12 cellular neighborhoods (CNs) corresponding to known hematopoietic niches, tertiary lymphoid structures, and erythroblastic islands. The leukemic BMME showed a severe loss of structural organization with fragmentation of NBM CNs and expansion of leukemia-specific CNs. Among the latter, we identified CNs characterized by various degrees of myeloid blast differentiation state, the type and level of immune infiltration (e.g., CD4+FOXP3+ regulatory T cells, CD8+ T cells), and the association with mesenchymal stromal cells. Leukemia-specific CNs enriched for suppressor cells and undifferentiated leukemic cells were associated with poor treatment response and clinical outcomes independently of clinical risk stratification. Orthogonal MERFISH analysis of 7 AML cases identified 27-34 cell types per case (average 31). Single-cell gene expression and spatial analyses of these cell types further characterized the differentiation state of leukemic subpopulations, as well as the activation state and cytokine profiles in situ. These results will be compared and integrated with CODEX data to provide a comprehensive characterization of the BMME in NPM1-mutant AML.

Conclusions: Using highly multiplexed spatial proteomics and transcriptomics, we captured the spatial landscape of the BMME in NPM1-mutant AML and identified leukemia-specific and prognostically relevant CNs in a real-world cohort of AML patients.