A Spatial Risk Model Reveals a 25 μm Transcriptomic Event Horizon for Plaque-Associated Myeloid Activation in Alzheimer’s Disease

Background Spatial transcriptomics has enabled high-resolution mapping of Alzheimer’s disease (AD) pathology; however, the precise spatial boundary at which homeostatic microglia and perivascular macrophages (PVMs) transition into a disease-associated state remains poorly defined. Existing trajectory-based approaches generally model activation as a gradual continuum and may fail to capture abrupt spatial state transitions. Methods We analyzed 195,620 combined Microglia-PVM cells from the Seattle Alzheimer’s Disease Brain Cell Atlas middle temporal gyrus MERFISH dataset. A Gaussian Mixture Model defines an activation threshold using a composite CD74+CTSS score. Plaque-proxy anchors were derived from the 500 most activated myeloid cells, verified for spatial dispersion across the parenchyma. A sparsity-aware XGBoost classifier was trained on a 138-gene feature set (excluding CD74 and CTSS) to model the spatial probability of activation. Spatial discontinuity was evaluated using LOESS and piecewise regression with an exhaustive knot-sweeping analysis, while SHAP identified key regulatory drivers. Results The model achieved a ROC-AUC of 0.896 on an independent test set. Automated knot-sweeping piecewise regression demonstrated strong statistical support for a discrete spatial threshold at precisely 25 μm from plaque-proxy anchors (ΔAIC=329.9 relative to LOESS). The threshold remained stable across alternative anchor definitions (24.8–25.7 μm). SHAP analysis revealed LRRK1, L3MBTL4, and PDE4B as major positive predictors of disease-associated activation, whereas TACR1 and GRIP2 showed protective homeostatic signatures. Pre-threshold analysis revealed early induction of SNTB1, SLC24A2, and PEX5L within the 30–40 μm warning zone. Conclusions Our findings identify a localized 25 μm transcriptomic event horizon surrounding plaque-associated inflammatory niches in AD. This spatial boundary provides a quantitative framework for understanding myeloid activation dynamics and highlights early warning biomarkers for intervention before irreversible neurotoxic conversion occurs.