Metabolic brain network reorganization precedes clinical conversion in Alzheimer's disease

Abstract

Structured AbstractO_ST_ABSIntroductionC_ST_ABSBrain glucose hypometabolism is a hallmark of Alzheimers disease (AD), yet conventional [18F]-fluorodeoxyglucose (FDG) positron emission tomography (PET) analyses have limited sensitivity in preclinical stages. Metabolic brain network approaches may better capture early vulnerability preceding clinical conversion. MethodsCognitively unimpaired individuals (n = 127) from the ADNI cohort with baseline FDG-PET and amyloid (A) and tau (T) status were classified as clinically stable or converters over an average longitudinal follow-up of 5.8 years. Baseline brain FDG uptake patterns were analyzed at the regional, voxel, and network levels across AT profiles. Network density was quantified globally and within functional networks. AD biomarkers and cognitive performance were also examined. ResultsConventional FDG-PET SUVr analyses failed to distinguish cognitively stable individuals from clinical converters at baseline, either at the regional or voxel levels. AT(N) biomarkers and neuropsychological performance likewise did not differ significantly between groups. In contrast, clinical converters exhibited hyperconnected metabolic networks at baseline, including within the default-mode network. These effects were consistent across A-T-, A+T-, and A+T+ groups, with network density higher in clinical converters than in cognitively stable individuals. Conversely, network density among stable individuals declined with AT progression, pointing to divergent network trajectories. DiscussionMetabolic network organization analysis revealed early AD-related vulnerability beyond regional hypometabolism, even before detectable amyloid positivity, and may reflect divergent trajectories of resilience and pathological propagation preceding clinical conversion. By leveraging existing FDG-PET datasets, this framework offers a valuable opportunity to identify individuals at risk of clinical progression at scale.