Multiscale Metabolic Covariance Networks Uncover Stage-Specific Biomarker Signatures Across the Alzheimer's Disease Continuum.
BackgroundConnectomics studies analyze neural connections and their roles in cognition and disease. Beyond regional comparisons, recent research has revealed inter-regional brain relationships via graph theory of brain network connectivity. Within these networks, path length measures a networks efficiency in communication. These connections can be quantified as inter-subject covariance networks related to functional connectivity, with alterations reported in neurodegenerative diseases. MethodsRetrospective analysis of ADNI 18F-FDG PET images using metabolic covariance analysis and hierarchical clustering was used to assess regional brain networks in subjects from cognitively normal (CN) to AD. We evaluated AD stage changes by calculating whole brain entropy, connection strength, and clustering coefficients. Additionally, estimates of shortest path for positive and negative correlations as a measure of network efficiency. We also developed a novel region set enrichment analysis (RSEA) to detect brain functional changes based on metabolic variations. Results were aligned with transcriptomic signatures and clinical cognitive assessments. FindingsIn AD subjects, whole brain metabolic connectivity revealed an increase in entropy, connection strength, and clustering coefficients, which indicates brain network reorganization as compensatory mechanisms of pathological disruption. As AD advances, path lengths between brain regions decrease from CN to MCI; however, path lengths significantly increased in AD. RSEA indicated functional changes in motor, memory, language, and cognition functions related to disease progression. InterpretationMetabolic covariance analysis of whole brain, and regional connectomics, track with AD progression. Moreover, path lengths permitted AD stages determination via alterations in brain connectivity. Furthermore, RSEA facilitated the identification of functional changes based on metabolic readouts. FundingNIH grant T32AG071444