Elucidating Molecular Features of White Matter Hyperintensities in Alzheimer’s Disease through Multimodal Imaging and SHAP Analysis

Abstract

White matter hyperintensities (WMHs) are a common feature of Alzheimers disease and are associated with cognitive decline, yet their molecular composition and spatial heterogeneity remain incompletely defined. Here, we identify distinct lipid signatures in human AD WMHs compared to matched normal-appearing white matter (NAWM) from the same donors. Using an integrated multimodal approach combining magnetic resonance imaging, matrix-assisted laser desorption/ionization imaging mass spectrometry (MALDI IMS), histological staining, and complementary liquid chromatography-tandem mass spectrometry, we resolve spatially localized lipid alterations within tissue sections while preserving spatial context. This approach reveals region-specific heterogeneity in WMH lipid composition across anterior and posterior brain regions that may be obscured by bulk lipidomics alone. Machine learning-based analysis using Shapley additive explanations (SHAP) identified lipid features that contribute to WMH classification, with sulfatide, hexosylceramide, and phosphatidylinositol species emerging as key discriminators. In anterior brain regions, WMHs were associated with differential abundance and depletion of specific sulfatide and hexosylceramide species (SHexCer 44:2;3O, SHexCer 42:2;3O, HexCer 41:1;3O, SHexCer 42:2;2O), whereas posterior WMHs were characterized by reduced phosphatidylinositol species (PI 36:1), demonstrating that AD-associated white matter pathology is governed by region-specific, heterogeneous lipid remodeling rather than uniform global degradation.