| Popis: |
Additional File 2: Figure S1. A �� pathology in the temporal pole cortex. Description: Immunohistochemistry for A�� (mouse monoclonal antibody, clone 6F/3D, Agilent, #M0872, 1:600) with peroxidase/DAB was performed in nearly-adjacent sections to those used for cyclic multiplex fluorescent immunohistochemistry in a Leica BOND-III automated stainer. Sections were counterstained with hematoxylin. Scale bars: 5 mm, insets 200 ��m. Figure S2. Phospho-tau pathology in the temporal pole cortex. Description: Immunohistochemsitry for phospho-tauSer202/Thr205(mouse monoclonal antibody, clone AT8, Thermo-Scientific, #MN1020, 1:10,000) with peroxidase/DAB was performed in nearly-adjacent sections to those used for cyclic multiplex fluorescent immunohistochemistry in a Leica BOND-III automated stainer. Sections were counterstained with hematoxylin. Scale bars: 5 mm, insets 200 ��m. Figure S3. Expression levels of selected markers across astrocytic and microglial subclusters from public single-nuclei RNA-seq studies. Description: Bubble plots illustrate the percent of nuclei (bubble size) and the gene expression levels (z-scores, color bar) of the astrocytic and microglial markers used in our cyclic multiplex fluorescent immunohistochemistry protocol across the astrocytic and microglial subclusters rendered by several published single-nuclei RNA-seq data sets. Note that our set of markers discriminates some of these transcriptomic subclusters. Figure S4. Characterization of astrocytes and microglia in AD vs. CTRL by cortical layer. Description: Box and whisker plots illustrate the distribution (box: median and interquartile range [IQR]; whiskers: 1.5 �� IQR) of mean gray intensity (MGI) z-scores for (a) each astrocytic marker and (b) each microglial marker across the CTRL and AD groups by cortical layer. Only layers II to VI were included in this study. Figure S5. Characterization of astrocytic and microglial states by cortical layer. Description: Box and whisker plots show the distribution (box: median and interquartile range [IQR]; whiskers: 1.5 �� IQR) of mean gray intensity (MGI) z-scores for each astrocytic (a) or microglial (b) marker across the three phenotypes by cortical layer. Only layers II to VI were included in this study. Figure S6. Effects of proximity to AD neuropathological changes on astrocytic and microglial phenotypes from two CTRL subjects with abundant A�� plaques. Description: (a) Representative high-plex image of astrocytes from a CTRL subject with abundant A�� plaques; note the differences with AD astrocytes in Fig. 5a. For clarity, only ALDH1L1, EAAT2, and GFAP markers are shown together with A��. Scale bar: 100 ��m, insets a1���a3: 10 ��m. (b) Histograms show the proportion of each astrocyte phenotype in n=2 CTRL subjects with abundant A�� plaques relative to all their astrocytes as a function of their distance (��m, x axis) to the nearest A�� plaque. Note that there are equal numbers of astrocytes within 25 ��m from the nearest A�� plaque classified as homeostatic, intermediate, or reactive. (c) Representative high-plex image of microglia from the same field of the same CTRL with abundant A�� plaques; note the differences when compared to AD microglia in Fig. 5c. For clarity, only IBA1, TMEM119, and CD68 markers are shown together with A��. Scale bar: 100 ��m, insets c1���c3: 10 ��m. (d) Histograms indicate the proportion of each microglial phenotype in n=2 CTRL subjects with abundant A�� plaques relative to all their microglial profiles as a function of their distance (��m, x axis) to the nearest A�� plaque. Note that most microglia in the vicinity of A�� plaques were classified as homeostatic, suggesting that their phenotypic transition to intermediate and reactive had not yet occurred. Figure S7. Differences in neuritic component of A�� plaques from CTRL and AD subjects. Description: Representative images of A�� and phospho-tau (PHF1) immunohistochemistry corresponding to the same fields of the AD and CTRL subjects shown in Fig. 5 and Fig. S6, respectively. Note the differences in the PHF1+ neuritic changes between CTRL and AD A�� plaques. Scale bar: 100 ��m, insets a1 and b1: 10 ��m. Figure S8. Gradient boosting machine models accurately discriminate between glial phenotypes. Description: Receiver operating characteristic (ROC) curves demonstrate the high discriminative power of the gradient boosting machine (GBM) models to discern between states (i.e., homeostatic vs. intermediate vs. reactive) of (a) astrocytes and (b) microglia based on mean gray intensity (MGI) data from thousands of high-plex single-cell profiles. Rankings of the variable importance scores shown in the horizontal bar plots reveal the most relevant markers for each classification task, respectively. Figure S9. Application of deep learning model interpretability functions to astrocytes with extreme classification probabilities. Description: Examples of the convolutional neural network (CNN) model interpretability functions applied to astrocytes with extreme classification probabilities (i.e., confident and correct predictions). Columns 1 and 5 show DAPI and all astrocyte markers of the high-plex image of a single astrocyte cell body from a CTRL and an AD subject, respectively, after performing the CNN normalization steps described (i.e., segmentation, interpolation, channel-level z-score). Hence, the signal intensity is represented by dynamic range rather than by pixel intensity. Columns 2���4 and 6���8 show the saliency (2 and 6), integrated gradient (3 and 7), and GradCAM (4 and 8) maps, which illustrate the pixels of each marker that the CNN considered most important for the classification of these two astrocytes as CTRL or AD. Figure S10. Application of deep learning model interpretability functions to microglia with extreme classification probabilities. Description: Examples of the convolutional neural network (CNN) model interpretability functions applied to microglia with extreme classification probabilities (i.e., confident and correct predictions). Columns 1 and 5 show DAPI and all microglial markers of the high-plex image of a single microglial cell from a CTRL and an AD subject, respectively, after performing the CNN normalization steps described (i.e., segmentation, interpolation, channel-level z-score). Hence, the signal intensity is represented by dynamic range rather than by pixel intensity. Columns 2���4 and 6���8 show the saliency (2 and 6), integrated gradient (3 and 7), and GradCAM (4 and 8) maps, which illustrate the pixels of each marker that the CNN considered most important for the classification of these two microglia as CTRL or AD. |
