Fig. 4: Plaque heterogeneity across major brain regions.

a Anatomical annotations for brain regions studied and optical image showing location of region of interest (ROI) for high-spatial-resolution imaging. The left-hemisphere image is derived from the Allen Mouse Brain Atlas (CCFv3), atlas.brain-map.org. b–f (left) Ion image at 5 µm spot size and pitch of plaque-associated lipids and (right) their relative amounts in various brain regions determined from whole-brain-slice imaging at 50 µm spot size and pitch. Each point is averaged from all sections imaged for a given animal (n = 6 per group). Box plots are shown with center line as median, boxes across the 25th–75th percentile (the interquartile range, or IQR), and whiskers extending to the lowest and highest non-outlier points, where outliers are points more than 1.5 times the interquartile range. Error bars in box plots represent the minimum and maximum non-outlier point. Statistics, including exact p values are provided in the Source data. g Performance evaluation of models for predicting region-of-origin for plaque-associated lipids. Each dot is from one five-fold cross validation using data from all animals (n = 6 per group). Box plots are shown with center line as median, extending to the IQR, with whiskers to the highest and lowest non-outlier points. h Confusion matrix from the random forest model, showing prediction power for different brain regions. i Effect of retained lipid features on the random forest model performance. The shaded area denotes ±1 standard deviation from 5-fold cross validation. j The first two dominant features from t-SNE analysis of lipid profiles from all plaque-localized pixels showing regional heterogeneity, color-coded by brain region. k 14 “plaque-pixel” clusters were identified via DBSCAN. l Heatmaps comparing normalized feature intensity (with each feature scaled between −1 and 1) of plaque-associated lipids organized by cluster ID (left) and brain region (right). Data in box plots, t-SNE, and heat maps are obtained from n = 6 animals per group, with 2–3 sections imaged per animal. *p < 0.01, **p < 0.001 as determined from two-sided, unpaired t-tests. Source data are provided as a Source data file.