Redefining the topology of the human bone marrow using augmented spatial transcriptomic analysis

Abstract

The bone marrow (BM) is the main site of haematopoiesis in adult life. Our understanding of the pathogenesis of BM-derived blood cancers is limited by lack of spatial contextualisation. While emerging spatial transcriptomic (ST) platforms offer unprecedented opportunities for spatially-resolved cellular phenotyping, we recognise and incorporate the power of AI-based tissue feature detection to enhance ST workflows. We perform ST analysis to define the topology of the normal bone marrow (BM) and BM in myeloproliferative neoplasms (MPNs), profiling 5,104,452 cells across 30 human BM samples. Following rigorous histology-based QC, we identify spatially-restricted trajectories of haematopoiesis and extend our understanding of the haematopoietic stem cell (HSC) niche. We find that BM fibrosis in MPN is associated with expansion of distinct immune and stromal co-enriched cell neighbourhoods. We then present a machine learning (ML)-based model trained on ST data that quantifies BM microenvironmental deviation, identifying heretofore unrecognised inter- and intra-individual sample heterogeneity in MPN. Our study demonstrates the potential for AI-based augmented ST analysis, and redefines our understanding of human BM topology.