Fig. 3

Spatial Organization and Molecular Identities of Cells and Proteopathy in ADD hippocampus is Revealed with Single-Cell and Object Segmentation. a Cartoons show examples of cells (astrocytes, microglia, vessels, neurons, and immune cells) and possible planar views in CNS tissue (dashed lines numbered 1 through 4). The representative photomicrograph (IHC-DAB) and MIBI-TOF (pseudo-colored spectral images) visually exemplify the corresponding focal planes (numbered 1 through 4) for the different cells, which are identified by their pan-markers (GFAP: astrocytes, Iba1-CD45: microglia, CD31-CD105-MCT1: vasculature, MAP2: neurons, CD45: immune cells). b Conceptual overview of nuclear and object segmentation approaches. With cellular segmentation, nuclear associated features (nuclei interiors, nuclei borders) are obtained using a pixel-based convolution neural network (DeepCell neural network) and custom pixel expansion around the nuclei, to generate cell segmentation masks for cell lineage assignments. Object segmentation was used to capture microglia and astrocytic processes, vascular boundaries, Aβ plaques, and PHF1-TAU NFTs and NTs. Masks for both discretized biological units are integrated to reveal their relationships in silico. c Photomicrograph (on the conductive gold slide, left panel) and corresponding MIBI-TOF spectral image (HH3 for nuclei, right panel) of human ADD hippocampus. White box highlights the re-rastered region that was used for analysis. d UMAP visualization of all segmented objects across all hippocampal FOVs, colored by lineage and proteopathy identities. Inset shows only Aβ plaques, and NFT-NTs overlaid on the UMAP. e Cell Phenotype Maps (CPM) of nuclear and object segmented masks labeled by their phenotype. f Composite CPM of all segmented masks with six inserts for a zoomed in view of cell identity overlaid onto segmentation masks for the hippocampal region. Regions 1–6 represent zoomed in images representative areas of CPM’s in e