Multi-omics single-cell spatial profiling of paediatric and adult acute myeloid leukaemia links spatial organization, molecular genetics and microenvironmental interactions

Abstract

Abstract While acute myeloid leukaemia (AML) urgently requires novel treatment strategies, immunotherapy has so far shown limited efficacy, despite its success in other haematological malignancies. Although single-cell RNA sequencing (scRNA-seq) has advanced our understanding of the AML immune microenvironment, the spatial organization and cellular interactions that underpin immune evasion within the bone marrow (BM) remain poorly defined. To address this gap, we employed multi-omics single-cell spatial profiling to characterize the AML BM in situ. We constructed three tissue microarrays from 148 formalin-fixed paraffin-embedded (FFPE) BM biopsies from AML patients (59 children, 26 adults) and non-leukemic controls (15 children, 14 adults). Spatial transcriptomics (ST) was performed using 10x Genomics Xenium with the pre-designed Human Multi-Tissue and Cancer panel, supplemented with probes for 100 genes to better characterize BM-resident cells, and probes for the RUNX1::RUNX1T1 fusion to aid identification of malignant cells, totalling 477 genes. Complementary spatial proteomics data were generated via imaging mass cytometry (IMC) using a 37-plex panel. Of the 148 biopsies, 126 (85%) met the quality criteria. Cell segmentation and ST quality control yielded 675,235 high-quality single-cell transcriptomes. Using canonical markers and 2 scRNA-seq reference datasets, 33 cell types were annotated. The RUNX1::RUNX1T1 fusion was specifically detected in all 7 patients with t(8;21) AML, with highest expression in immature, granulocytic, and eosinophil progenitor-like blasts, consistent with the bifurcated differentiation seen in this AML subtype (Derevyanko et al., bioRxiv 2025). Notably, both ST and IMC identified macrophages (median 2%) and stromal cells (median 4%) at substantially higher frequencies than in typical BM aspirate-based analyses, which often fail to capture macrophages and typically detect less then 0.3% stromal cells. Further characterization of the stromal compartment (26,798 cells) revealed 5 distinct stromal subtypes, including a perivascular stromal population almost exclusively present in a subset of paediatric AML. This rare subtype colocalized with sinusoidal endothelial cells and expressed markers of endothelial-mesenchymal transition, previously described only in regenerating BM (Kenswil et al., Cell Stem Cell 2021). Transcriptome-based neighbourhood analysis identified 10 distinct cellular neighbourhoods (CNs) highlighting stark differences between patients and controls and among patients. Three CNs (late myeloid; erythroid and B cells; and a diverse microenvironmental CN) were shared between patients and controls, while the other 7 were AML-specific. These AML blast-enriched CNs varied in immune and stromal content, with some CNs being lymphoid-depleted, and others enriched for T cells and macrophages. CN proportions partially correlated with molecular subtypes, e.g., a lymphoid-depleted CN was unique to KMT2A-rearranged AML. IMC-based CN analysis supported these findings. To further explore cellular crosstalk between leukemic cells and the microenvironment, we performed spatially informed cell-cell communication analysis, which revealed Galectin-9, MHC-II, and PECAM-1 as key signalling pathways mediating interactions between blasts and macrophages, stromal cells and T cells. These findings were validated using whole transcriptome 10x Flex scRNA-seq on FFPE BM biopsies, which also recapitulated the cell type distributions from the Xenium data. Notably, these interactions had greater signalling strength in CNs with higher immune and stromal diversity. These pathways are currently under functional investigation. In conclusion, we generated a comprehensive multi-omics single-cell spatial atlas of the AML BM, allowing in situ identification of rare cell types including a potentially novel stromal subtype. Our integrated spatial analysis reveals distinct CN architectures and uncovers key niche-dependent interactions between blasts and their microenvironment that may represent novel therapeutic targets.