Alzheimer's Imaging Consortium

Variability of brain amyloid burden across individuals suggests that biological factors may play a role. This study investigates the association between polygenic risk scores (PRS) and amyloid burden in a preclinical Alzheimer's disease (AD) cohort to identify genetic determinants.

PRS for AD susceptibility (Kunkle et al., 2019) and for CSF-Aβ42 and ptau181 (Jansen et al., 2022) were computed for 867 non-demented AMYPAD participants. PRS were calculated with and without the APOE region (19:44.4-45.5Mb) under three p-value inclusion thresholds (pT=5x10⁻⁸, 1x10⁻⁵, 0.1). Linear regression models tested associations with amyloid burden (Centiloid) adjusting for age, sex, and education (years). Model performance was evaluated using 10-fold cross-validation (R²). SNPs from the top-performing PRS for each brain region were mapped to genes using biomaRt, followed by enrichment analysis with enrichGO. Separate linear regressions for males and females assessed sex differences.

Participants were predominantly cognitively unimpaired (CDR=0, 85.1%), with low amyloid burden (CL<10, 60%), and 42.4% carried at least one APOE-ε4 allele. The strongest predictive ability was observed for PRS_amyloid at pT=1x10⁻⁵ for frontal amyloid burden (R²=0.18, Figure 1A). PRS_amyloid and PRS_Kunkle at pT=5x10^-8 and 1x10^-5 performed comparably across brain regions (R²>0.14). At pT=1x10^-5, PRS_amyloid had 23 of 38 SNPs mapped to genes, enriched for acylglycerol (p = 0.006) and triglyceride (p = 0.005) homeostasis. PRS_Kunkle had 50 of 99 SNPs mapped to genes, enriched for negative-regulation-of-amyloid-beta-formation (p = 1.9x10^-6) and amyloid-precursor-protein-catabolic-process (p = 2.8x10^-6). Sixteen mapped genes overlapped between the two scores. PRS at pT=0.1 consistently performed worse (R²<0.12), as did PRS_noAPOE (R²<0.11). Stratification by sex showed higher R² for males (N = 354) than females (N = 512), with similar observations as above: PRS_amyloid and PRS_Kunkle at pT=5x10^-8 and 1x10^-5 showed comparable performance (Figure 1B). Overall, predictive performance was highest in the frontal cortex, comparable with the global composite in men, and lowest in the occipital cortex.

PRS prediction was strongest in the frontal cortex, with PRS_amyloid and PRS_Kunkle performing similarly at stringent SNP inclusion thresholds. PRS_amyloid performed slightly better despite different biological enrichment. These findings, stronger in men, enhance our understanding of the genetic basis of amyloid pathology and support the potential of genetic profiling for identifying at-risk individuals, with potential sex differences.