Function-oriented microglial transcriptional scores in Alz-heimer's disease: resolving terminal functions beyond activa-tion states

Abstract

Alzheimer's disease (AD) staging based on amyloid-{beta}, tau, and neurodegeneration does not capture heterogeneous microglial functions. Existing transcriptomic scores indicate microglial activation or state similarity but do not resolve post-activation functions and are not directly comparable across scales. Herein, we develop a function-oriented framework comprising seven harmful-and seven adaptive-associated gene modules, selected from predefined mechanisms and direct functional evidence. Using SEA-AD micro-glia/perivascular macrophage nuclei, we compute do-nor-brain-region pseudobulk scores on a fixed P10-P90 reference scale and validate them against matched protein, activity, and tissue endpoints. H3 is related to cytokine/chemokine proteins; H4 and A5 are related to the same isoprostane readout in opposite directions; and A3 is related to dense-core plaque fraction. Benchmark comparison shows modest incremental fit but distinct out-come-specific information. In external GSE146639, eight modules remained FDR-significant, although harmful and adaptive arms are highly collinear; H7 and A2 are replicated in external proteomics. Downsampling, leave-one-gene-out, and alternative scoring analyses supports technical stability, and single-cell reconstruction supported a state-resolved basis. These scores summarize relative engagement of damage-related and adaptive programmes, enabling mechanistic comparison across donors and brain regions, but remain observational, are not intended for clinical decision-making, and require further validation.