Metabolic Connectivity and human brain function at the network-level: methodological innovations and applications across health, disease, and cognition

Over the past few decades, connectomics has revolutionized our understanding of brain organization, highlighting that cognition and behaviour emerge from dynamic interactions among spatially distributed regions, coordinated into large-scale networks. This perspective has gained momentum as research increasingly links connectivity alterations to a wide range of neurological, neuropsychiatric, and oncological disorders. Although progress has been made in mapping structural and functional connectivity (MRI-based techniques), these approaches largely overlook the molecular dimension of inter-regional communication. This thesis addresses this gap by advancing the use of [18F]FDG PET to investigate metabolic connectivity (MC), defined as the relationship between metabolic measurements of different brain regions. Specifically, no gold standard currently exists for computing MC. Positioned within this evolving and heterogeneous landscape, we focused on strengthening the applicability of the Euclidean similarity-based approach, which, being directly derived from raw time-activity curves, is highly generalizable and does not require additional signal preprocessing or normalization steps. The thesis is structured around three major axes of investigation. First, we assess the clinical applicability of MC in glioma patients, showing that it captures widespread network alterations even in structurally unaffected regions. Compared to conventional SUVR-based analysis, our approach offers a more holistic view of disease burden, providing interesting additional insights into glioma-related alterations. Second, based on the hypothesis that metabolic and functional dynamics are complementary dimensions of brain organization, the thesis introduces a multimodal framework to investigate the interplay between cerebral glucose metabolism and functional communication at rest. We used multivariate Partial Least Squares Correlation to examine these relationships at both regional and network levels. This thesis extends also conventional functional connectivity by incorporating effective connectivity, decomposed into symmetric partial covariance (true dependencies) and antisymmetric differential covariance (directional flow). Clinically, we show that disruptions in effective–metabolic coupling depend on tumor location and enable novel patient classification schemes based on whether the decoupling is local or network-level. This underscores the added value of MC in clinical neuroscience and supports its role as a biomarker of brain dysfunction. Third, we extend our analysis to a task-based cognitive paradigm in healthy subjects, exploring how metabolic networks reconfigure in response to externally imposed cognitive demands. Here, we demonstrate that MC is sensitive to task-related network dynamics, particularly in regions associated with higher-order cognitive processing, while also revealing novel roles of specific brain networks. Crucially, integrating MC with functional connectivity enhances the prediction of individual behavioural performance, underscoring the complementary nature of these modalities and the benefits of a multimodal analytical approach. In conclusion, this thesis demonstrates the applicability and versatility of metabolic connectivity as a powerful framework for studying brain network organization across diverse physiological and pathological conditions. By developing and applying innovative methodologies in both clinical and experimental settings, including resting-state paradigms and cognitively demanding tasks, this work showcases the unique contribution of PET-derived connectivity, whether used independently or in synergy with functional measures, to the broader field of network neuroscience.