Multimodal Spatial Transcriptomic Profiling Elucidates Niche-Specific Dynamics in Medullary and Extramedullary Acute Myeloid Leukemia

Background: Acute myeloid leukemia (AML) is a heterogeneous disease characterized by variable leukemic states that primarily originate in the bone marrow (medullary) and can also occur in extramedullary (EM) tissues, indicating its adaptability to different environments. The spatial dynamics of medullary and extramedullary AML remain underexplored. Understanding these dynamics is crucial for identifying the microenvironments in AML and improving diagnostic approaches.

Methods: We conducted Visium spatial transcriptomics (ST), suspension single cell RNA, and GeoMx-based Digital Spatial Profiling (DSP) on AML medullary and extramedullary patient samples. For Visium ST, we applied two assay versions (v1 and CytAssist v2) on paired bone marrow (BM) and EM samples from two AML patients (n=8 total samples). We compared Visium assays on these decalcified and non-decalcified leukemic tissues and adapted median absolute deviation (MAD) for spatial spot filtering. Spatial deconvolution was performed using in-house generated BM suspension scRNA-seq reference from 16 samples (9 healthy, 7 AML patients). Findings were validated using GeoMx-based DSP in three independent AML patients, providing a comprehensive spatial map of medullary and extramedullary leukemia.

Results: Visium ST patients: Patient 1 (PT1) had medullary and EM leukemia, while Patient 2 (PT2) had isolated EM leukemia with no BM involvement. RNA fragmentation on these samples varied, measured as the DV200 values before library preparation (range 24-62%; mean DV200 = 44.5, standard deviation = 16.6). Post-library DNA traces for v1 were lower in BM samples (24-29%) compared to v2 (89-91%), while EM tissues showed similar post-library traces for both assays (v1: 87-83%, v2: 88-91%). V2 assays detected more oligonucleotides (mean: 11072.37) and genes (mean 4735.67) compared to v1 (mean oligonucleotides: 1292.76, and mean genes detected: 955.19).

Considering the variability in each tissue, we adapted MAD to filter spots based on mitochondrial content and expression profiles. We next performed shared nearest neighbor (SNN) modularity optimization-based unsupervised clustering, revealing overlap between determined clusters and hematopathologist-identified histopathological tissue annotations. To spatially delineate spot-level cellular information in our BM samples, we deconvolved individual spots using probabilistic label transfer workflow with our scRNA-seq BM reference. High leukemic score (HLS) spots defined with increased AML population and showed co-localization with monocytes and granulocyte-monocyte progenitors (GMP) in BM and macrophages in EM tissues.

Spatial cell-cell communication and pathway analysis indicated CXCL12-CXCR4 and PI3K/AKT/mTOR pathway promoted trans-differentiation in inflammatory niches. Additionally, deconvolution of AML cells showed that primitive-like AML cells were located near the endosteal niche, while committed-like cells were spatially distant from the bone. These patterns were validated with GeoMx-based DSP, suggesting that AML cells at different states of differentiation locate in distinct niches within the BM.

Conclusions: To our knowledge, this is the first study applying Visium ST to both medullary and extramedullary AML tissues from paired samples to delineate biologic niche-specific differences. Our study elucidates the spatial niches of AML using multimodal ST, identifying distinct inflammatory niches with significant involvement of the CXCL12-CXCR4 axis, promoting trans-differentiation via the PI3K/AKT/mTOR pathway, aiding in AML migration and infiltration into EM sites. Spatial analysis revealed that primitive-like AML cells predominantly locate near the endosteal niche in BM, while committed-like and GMP-like populations situate more distally. All our data will deposit in an interactive public website that allows other researchers to navigate and explore. Spatial analysis is ongoing for 17 additional BM AML patients with concomitant proteomics profiling.