As a proof-of-concept, we organized 16 bone marrow biopsies from 7 adult AML patients into 2 tissue microarrays using 1.5 mm diameter cores with a 4 µm section thickness. Each TMA was then imaged using the COMET multiplex IF platform with a 28-plex panel covering canonical markers for immune cells (n=4), AML cells (n=12), and functional markers (n=12). Using the same slide, we then performed IMC with a 25-plex panel containing 16 overlapping markers to assess concordance between methodologies for a total of 34 unique proteins imaged. Image alignment was guided by tissue structure and co-expression of shared markers onto a reference H&E-stained serial section, which provided structural components such as trabecular bone. Cells were segmented using the convolutional neural network U-Net and manually phenotyped by mean intensity thresholds. Unbiased clustering over each cell’s spatial K function was performed to identify 5 distinct regions of cellular organization. We then validated the results of our spatial workflow by staining adjacent sections using the Opal 7-color multiplex IF protocol. Gene set enrichment was performed on an integrated AML transcriptomic dataset consisting of patients from TCGA, MDACC, and BEAT-AML studies (n=480) using a published TLS signature (Cabrita et. al. Nature 2020) to assess prognostic ability.
Clustering over the spatial K function revealed 5 regions enriched in AML cells, monocyte/macrophage lineage cells, lymphocytes, unlabeled cells, and one region that contained a relatively even mix of all cell types. Interestingly, we observed the lymphocyte enriched region to only appear in 3 of the 16 tumor cores, taking the form of a dense aggregate of B and T cells, similar in structure and organization to tertiary lymphoid structures (TLS) in solid cancers. B cells in these aggregates displayed increased expression of Ki-67, suggesting proliferating B cells analogous to germinal centers. Furthermore, in tumor cores where these aggregates appeared, we found significant enrichment of monocyte and macrophage lineage cells (p=0.0164), as well as increased CD4 (p=0.017) and CD8 (p=0.041) T cell infiltration into areas of high AML cell density. Finally, we were able to validate the presence of these TLS-like aggregates by staining adjacent sections of the two TMAs using the Opal 7-color multiplex IF protocol.
We sought to further investigate the presence of TLS-like aggregates in AML by querying large public datasets with a previously validated TLS signature (Cabrita et. al. Nature 2020). Interestingly, the TLS signature strongly correlated to hypoxia, inflammation, and angiogenesis hallmark gene sets, aligning with the propensity of TLSs to form in inflammatory conditions and promote vessel growth in solid cancers. Furthermore, when dichotomizing patients by median TLS score, we found that the signature significantly correlated with overall survival (p=0.016).
We find that by utilizing the strengths of both high-plex IF and IMC in our integrative sample-to-analysis workflow, we can reliably spatially characterize the immune, tumor, and structural elements in the bone marrow microenvironment, allowing us to uncover novel cellular structures such as TLS-like aggregates and characterize their effect on the leukemic bone marrow microenvironment. Further, our study’s methodologies can be adapted for other bone marrow diseases where decalcification and autofluorescence present challenges.