A multidimensional analysis reveals distinct immune phenotypes and the composition of immune aggregates in pediatric acute myeloid leukemia

Raw GeoMx Data generated forhttps://www.nature.com/articles/s41375-024-02381-w Paper abstract:Because of the low mutational burden and consequently, fewer potential neoantigens, children with acute myeloid leukemia (AML) are thought to have a T cell-depleted or ‘cold’ tumor microenvironment and may have a low likelihood of response to T cell-directed immunotherapies. Understanding the composition, phenotype, and spatial organization of T cells and other microenvironmental populations in the pediatric AML bone marrow (BM) is essential for informing future immunotherapeutic trials about targetable immune-evasion mechanisms specific to pediatric AML. Here, we conducted a multidimensional analysis of the tumor immune microenvironment in pediatric AML and non-leukemic controls. We demonstrated that nearly one-third of pediatric AML cases has an immune-infiltrated BM, which is characterized by a decreased ratio of M2- to M1-like macrophages. Furthermore, we detected the presence of large T cell networks, both with and without colocalizing B cells, in the BM and dissected the cellular composition of T- and B cell-rich aggregates using spatial transcriptomics. These analyses revealed that these aggregates are hotspots of CD8+ T cells, memory B cells, plasma cells and/or plasmablasts, and M1-like macrophages. Collectively, our study provides a multidimensional characterization of the BM immune microenvironment in pediatric AML and indicates starting points for further investigations into immunomodulatory mechanisms in this devastating disease. GeoMx methods:5 μm thick FFPE BM biopsy sections from six pediatric AML cases with an immune-infiltrated BM and two non-leukemic controls were put on three different slides and prepared for GeoMx Digital Spatial Profiling (DSP; NanoString), as previously described. Slides were simultaneously incubated with immunofluorescent antibodies and GeoMx Whole Transcriptome Atlas profiling reagents. SYTO13 (S7575, Thermo Fisher,) was used for identification of nuclei, CD45 (NBP2-34528, Novus) for leukocytes, and CD3 (NBP2-54392AF647, Novus) for T cells. Stained slides were loaded onto the GeoMx instrument and scanned. ROIs were selected using the above-mentioned antibodies in combination with overlayed images of CD20, CD34, CD3-CD4 (duplex), and CD117 (IHC). Then, UV-photocleaved oligonucleotides were collected in separate wells and sequenced on the Nextseq2000 (Illumina, San Diego, CA, USA). Raw data were normalized using Quartile 3 count (Q3) normalization in R (V.4.2.1) as per NanoString’s recommendations (code is available in the vignette of the Geomxtools package:https://bioconductor.org/packages/release/workflows/vignettes/GeoMxWorkflows/inst/doc/GeomxTools_RNA-NGS_Analysis.html). Adjusted code is available with the download. Batch correction was performed using Combat-seq¹⁷. Spatial Deconvolution was performed using the safeTME reference (SpatialDecon package; cell reference profiles are available via https://github.com/Nanostring-Biostats/CellProfileLibrary/blob/archive/safeTME-for-tumor-immune.csv). Furthermore, we retrieved single-cell RNA-sequencing data from pediatric tonsillar B cells (sample BC005 was chosen since it had the highest number of cells, as done previously) and adult AML bone marrow CD8+ T cells (all patients), and converted these two additional reference profiles using R (V4.2.1). Deconvoluted abundance scores were normalized for ROI-size and, in case of immune aggregates, for the ROI-area covered by these aggregates. The ‘12chem’, ‘Tfh’, and ‘TLS imprint’, and M2-predominance signatures were applied to Q3-normalized data and further normalized as mentioned above. For questions please contact [e-mail] or [e-mail].