Cell‐Based Genetic Subtypes Predict Neuroimaging Biomarkers and Clinical Conversion to Alzheimer’s Disease

Abstract

AbstractBackgroundThe complex etiology and slow progression of AD have complicated efforts to establish robust patient subtypes resulting in high attrition rates among late‐stage clinical trials. We developed a novel procedure to stratify at risk individuals to develop AD using network‐based polygenic risk scores from brain cell‐type specific transcriptome signatures.MethodWe previously computed cell‐based polygenic risk scores (cbPRSs) in the Alzheimer’s Disease Neuroimaging Initiative (Sahelijo et al. 2022). Recently, we computed cbPRSs in 8,481 Framingham Heart Study (FHS) participants using the same genetic markers used in ADNI. We defined low‐ and high‐risk genetic subtypes using the first and the fourth quartiles of the cbPRS distributions, respectively. We conducted association tests with domain specific cognitive tests and neuroimaging biomarkers using subtype risk status as a binary outcome in cognitively normal (CN), mild cognitive impairment (MCI), AD, and all subjects from ADNI. We compared conversion rates from CN and MCI to AD between subtypes using a Cox proportional hazard regression in ADNI and FHS. Using brain‐wide gray matter volume maps with voxel‐based morphometry we compared brain atrophy patterns between subtypes in ADNI corrected for multiple comparisons using Standard FDR at q = 0.05.ResultGenetic subtypes for astrocytes (Ast‐M2) and oligodendrocytes (Oli‐M45) showed significant association with memory and executive function impairment, decreased hippocampal volume, increased amyloid deposition, and decreased glucose metabolism using all subjects from ADNI (P<1.39E‐3). These associations for memory impairment, global amyloid deposition, and decreased glucose metabolism remained significant only in MCI subjects (best P<7.23E‐09). Conversion rates for Ast‐M2 and Oli‐M45 genetic subtypes were significantly different in both ADNI (P<8.67E‐5) and FHS (P<3.90E‐9). These differential conversion rates for Ast‐M2 and Oli‐M45 in FHS were not fully explained by presence of APOE ɛ4 allele (P<0.05 in ɛ4 non‐carriers). Voxel‐wise whole brain analysis identified significant atrophy in bilateral hippocampi, entorhinal cortex, and amygdala for Ast‐M2 (P = 0.0007) and Oli‐M45 (P = 0.001).ConclusionWe demonstrate that polygenic risk scores informed by cell type specific brain transcriptomic networks can stratify at risk subjects with AD related changes in imaging biomarkers and higher chance of progression to AD. These findings provide insight into early detection of cell‐type specific at‐risk individuals for AD.