3D reconstruction of skin and spatial mapping of immune cell density, vascular distance and effects of sun exposure and aging

Mapping the human body at single cell resolution in three dimensions (3D) is important for understanding cellular interactions in context of tissue and organ organization. 2D spatial cell analysis in a single tissue section may be limited by cell numbers and histology. Here we show a workflow for 3D reconstruction of multiplexed sequential tissue sections: MATRICS-A (Multiplexed Image Three-D Reconstruction and Integrated Cell Spatial - Analysis). We demonstrate MATRICS-A in 26 serial sections of fixed skin (stained with 18 biomarkers) from 12 donors aged between 32–72 years. Comparing the 3D reconstructed cellular data with the 2D data, we show significantly shorter distances between immune cells and vascular endothelial cells (56 µm in 3D vs 108 µm in 2D). We also show 10–70% more T cells (total) within 30 µm of a neighboring T helper cell in 3D vs 2D. Distances of p53, DDB2 and Ki67 positive cells to the skin surface were consistent across all ages/sun exposure and largely localized to the lower stratum basale layer of the epidermis. MATRICS-A provides a framework for analysis of 3D spatial cell relationships in healthy and aging organs and could be further extended to diseased organs.

AF image of all tissues in one serial section Virtual H&E images of all tissues in one serial section Micro CT image of all samples in the block Supplementary Figure 4 -Micro CT imaging workflow. a) Samples were embedded in a single paraffin block, which then underwent micro CT imaging prior to block sectioning; b) shows an example section with AF images; c) the same section but shown as a virtual H&E and each sample is annotated as region 1, 2, etc.; d-e) zoomed in example of the Micro CT slice for region 6; f) Corresponding virtual H&E section for region 6.
Virtual H&E of region 6 after sectioning supervised and unsupervised model where a supervised DL model was used for DAPI/nuclei segmentation. First, an encoder-decoder based deep learning (DL) model was trained on a small sample (194 DAPI image patches selected from 30 images from 10 patients) of manually annotated nuclei created using the annotation function in QuPATH. Multiscale Laplacian of Gaussian (LOG) was introduced along with the DAPI images as separate channels to our encoder-decoder based DL model. The LOG feature detects blob like structures (which correspond to nuclei shape and boundary) in the DAPI images, and thereby provides contextual information to the DL model. Use of multiple channels allowed us to train an accurate DL model from a small sample of the manually annotated DAPI images. The depth of the encoder-decoder DL model was set to 4 and binary cross entropy was used as a loss function.  Model. An unsupervised GMM was used for automatic probabilistic segmentation of immune cell types: T killer (CD8), T reg (FOXP3), T helper (CD4), macrophages (CD68), as well as endothelial cells (CD31), markers of proliferation (Ki67), DNA damage (p53), and DNA repair (DDB2). GMM with two clusters was used to obtain a probabilistic segmentation of the cell-type and functional markers and background. Pixels with high signal for cell type and functional markers were automatically assigned high probability values in one cluster (positive class) and pixels with low signal intensity were assigned high probability values in the second cluster (backgroundnegative class). Union of probabilities obtained from the positive class of GMM model and nuclei segmentation (from the DL model) were then fused. Probability values were used to automatically scale (between 0-1) and quantify positive cells (i.e., for each marker of interest) and determine the percentage overlap between the markers and the segmented nuclei. Low percentage overlap was further used to remove imaging artifacts, debris, and cells with low/background marker intensity.  Figure 6 -Normalized 3D immune cell counts analyzed by age and sun exposure in the entire sample (dermis and epidermis) (a-g) and correlations (h); There was a non-significant trend for a positive relationship between T Helper/T Killer cell counts and age (p=0.01, adj. p=0.07) (see panel h).   Figure 7 -Immune cell counts within 100 µm of endothelial cells and correlations with age/skin exposure to UV (a-d) and correlations (e); (iii) There was a trend for an inverse relationship between T Killer cell counts within 100 µm of endothelial cells and age (adj. Region 9 (upper arm, marked sun exposure, 60 years) shows a higher distribution of p53 and Ki67 positive cells associated with hair follicle cells and up to 1600 µm deep with a reference virtual H&E section from the same sample and highlighted hair follicle. For interactive visualization go to: https://hubmapconsortium.github.io/vccfvisualization-release/html/epidermis_entire/epidermis_region_9.html (i) (ii) 500 µm c Supplementary Figure 9 -Normalized epidermis (AE1/stratum basale layer) cell counts for DNA damage (DDB2 and p53) and proliferation (Ki67) (a-d); (e) correlations with age and skin exposure to UV. There was a significant inverse correlation between DDB2 positive cells and age (adj. p=0.05) (see panel e).  Figure 9 -Normalized epidermis (CK26/entire epidermis) cell counts for DNA damage (DDB2 and p53) and proliferation (Ki67) (f-i); (J) correlations with age and skin exposure to UV. There was a significant inverse correlation between DDB2 positive cells and age (adj. p=0.04) (see panel j   Figure 11 -Example images for pan cytokeratin PCK26 in earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. DAPI -Blue, AE1 -Cyan Normal skin 2.7 D/P Cy3 at 5 µg/mL Supplementary Figure 11 -Example images for pan cytokeratin AE1 in earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for AE1 are from normal skin.

Ki67 (Abcam ab196907, clone EPR3610
Lymph node -Commercial conjugate Cy5, 10 µg/mL Supplementary Figure 11 -images for KI67 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for Ki67 are shown from lymph node and skin.

P53 (DAKO M7001, DO-7)
Stomach adenocarcinoma -1.5 D/P Cy5 at 2 µg/mL Supplementary Figure 11 -images for p53 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for p53 are shown from stomach adenocarcinoma and skin.

Region of interest with p53 staining in multiplexed skin sample
DAPI -Blue, p53 -Yellow 100 µm 100 µm d

DDB2 (Abcam ab181136, clone EPR9811)
Normal bladdersecondary detection with Cy3 at 6.2 µg/mL Supplementary Figure 11 -Example images for DDB2 in earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for DDB2 are shown for normal bladder and skin.

Region of interest with DDB2 staining in multiplexed skin sample
DAPI -Blue, DDB2 -Magenta 100 µm 100 µm e Na+K+ATPase (Abcam ab167390, clone EP1845Y) Breast carcinoma -4.2 D/P Cy3 at 5 µg/mL Supplementary Figure 11 -Example images for Na+K+ATPase from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for Na+K+ATPase are shown for breast carcinoma and skin.

Region of interest with Na+K+ATPase staining in multiplexed skin sample
DAPI -Blue, Na+K+ATPase -Yellow 100 µm 100 µm f Pan Cadherin (1-13, 15-20, 22-24) Figure 11 -Example images for pan-cadherin from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for pan-cadherin are shown for normal liver and skin.

2.38)
Reactive lymph node 2.5 D/P Cy5 at 5 µg/mL Supplementary Figure 11 -Example images for CD3 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for CD3 are shown for breast lymph node and skin.

Region of interest with CD3 staining in multiplexed skin sample
DAPI -Blue, CD3 -Yellow 100 µm 100 µm h CD4 (Abcam ab181724, clone EPR6855) DAPI -Blue, CD4 -Magenta Lymph node, 5.1 D/P Cy3 at 5 µg/mL Supplementary Figure 11 -Example images for CD4 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for CD4 are shown for lymph node and skin.

Region of interest with CD4 staining in multiplexed skin sample
100 µm 100 µm i

FOXP3 (Biolegend 320114, Clone 206D)
Reactive lymph node, Alexa 647 commercial conjugate, 10 µg/mL Supplementary Figure 11 -Example images for FOXP3 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for FOXP3 are shown for lymph node and skin.

Region of interest with FOXP3 staining in multiplexed skin sample
DAPI -Blue, FOXP3 -Yellow 100 µm 100 µm j CD8 (DAKO M7103, Clone C8144B) Stomach Adenocarcinoma 2.6 D/P Cy5 at 5 µg/mL Supplementary Figure 11 -Example images for CD8 from earlier validation studies and in the multiplexed skin study. As described in methods, a multi-step process was utilized and is available on protocols.io Cell DIVE™ Platform | Antibody Characterization for Multiplexing (protocols.io). A multi-organ slide (Pantomics MTU391) containing tumor and corresponding uninvolved tissues as controls is used for screening antibodies and testing dye conjugates and concentrations. Example images for CD8 are shown for stomach adenocarcinoma and skin.   Figure 11q UCHL1 (PGP 9.5) Neuron 13C4 Santa cruz sc-58594 4 Cy3 secondary AB_793620 Supp. Figure 11r Supplementary Table 3 Antibody, dye and staining information for each marker included in the study