[Preprint] Dissecting Alzheimer’s disease heterogeneity by cross-trait polygenic prediction

In this study, led by Bill Li, we introduce cross-trait polygenic prediction as a strategy for using phenome-wide PGS libraries in deeply phenotyped cohorts to dissect genetic heterogeneity in Alzheimer’s disease.

The study integrates 713 UK Biobank-derived polygenic scores with 36 deeply characterized Alzheimer’s disease phenotypes across 1,678 ROSMAP participants. The analysis identifies 268 significant associations between 12 prioritized PGS and Alzheimer’s disease phenotypes including cognition, amyloid, and neurofibrillary tangles. The prioritized PGS include blood lipid measurements, inflammatory biomarkers, and cancer traits, and 49 associations persist when APOE is excluded from the scores.

Multi-score predictive models improve prediction of amyloid and cognition compared with Alzheimer’s disease PGS or APOE alone. The approach also identifies six individual-level Alzheimer’s disease polygenic subtypes supported by distinct pathological patterns, providing a model for stratifying heterogeneous disease-focused cohorts using genomics.