Longitudinal plasma proteomics separates diagnostic differences from progression-linked changes in Alzheimer's disease.

Most plasma proteomic studies in Alzheimer's disease (AD) compare cases and controls cross-sectionally, leaving unresolved which AD-associated proteins mark diagnostic states and which are linked to disease progression. Using longitudinal SomaScan profiling from the Global Neurodegeneration Proteomics Consortium (13,449 participants, 17,269 samples, 7,362 aptamers), we separated baseline AD differences from AD-specific change over time. Linear mixed-effects models requiring concordant baseline and AD-by-time effects defined a 30-protein signature. We prioritized proteins across five evidence domains: clinical progression, AD biomarker alignment, cerebrospinal fluid concordance, independent prospective replication in UK Biobank and genetic support from Mendelian randomization and rare-variant burden. Thirteen proteins were supported in two or more domains and six in three. EDA2R, HPGDS, ITGAV and CLEC3B converged across clinical, biomarker and prospective evidence. Signature proteins aligned more strongly with tau and neuronal-injury markers than with Aβ42/40. ANTXR1 showed direction-concordant plasma pQTL Mendelian randomization and nominal rare-variant burden signals, supporting its prioritization within the longitudinal AD signature. By distinguishing diagnostic-state markers from progression-linked changes, this longitudinal, multi-domain approach prioritizes proteins for validation as markers of AD progression and for mechanistic and therapeutic follow-up.