Investigating the brain’s ‘dark energy’ through the complex coupling of [18F]FDG PET and resting-state functional MRI

The human brain has a remarkable metabolic budget, and most of its glucose and oxygen consumption happen during rest. However, the precise factors that control resting-state metabolism across different brain regions are still unknown. Two functional imaging tools that can provide a window into the complex mechanisms of brain metabolism and spontaneous activity are positron emission tomography (PET) and functional MRI (fMRI). In particular, the PET radio- tracer [18F]fluorodeoxyglucose ([18F]FDG) allows to track the first steps of glucose metabolism in the brain in vivo; resting-state fMRI (rs-fMRI), on the other hand, has a offered a powerful non-invasive tool for assessing proxies of spontaneous brain activity through blood oxygenation, as well describing a large-scale brain organization into ‘functional connectivity’ (FC) networks, composed of brain regions whose rs-fMRI signals fluctuate in synchrony. Trying to disentangle both the redundancy and the complementarity in the information coming from these two imaging modalities is extremely relevant for both neuroscientific and technical questions, e.g., 1) to characterize the functional drivers of local glucose consump- tion, 2) to better understand the somewhat unclear physiological and metabolic bases of the rs-fMRI signal, 3) to describe the large-scale functional network architecture of the resting brain both in hemodynamic and in metabolic terms, 4) to provide reliable fMRI-based proxies of glucose metabolic consumption to use as biomarkers of disease etc. In this thesis work, organized into three main parts, we have broadened the horizon of [18F]FDG PET vs. rs-fMRI integration on multiple levels. First, we assess the relationship between [18F]FDG standard uptake value ratio (SUVR), a relative and semiquantitative proxy of glucose metabolism, and a large range of fMRI-derived variables (= 50) to understand if the metabolic information probed by [18F]FDG was more related to the fMRI signal local activity and coherence, or large-scale static and time-varying FC, expanding on previous assessments based only on a handful of fMRI features. Also, we develop a new methodological framework (including multiple regression and multilevel hierarchical modelling) to explore whether a combination of rs-fMRI variables could meaningfully explain more of the regional metabolic variability than simple pair- wise associations. Then, we expand our assessment by exploring the details of metabolic physiology thanks to full kinetic modelling of [18F]FDG dynamic PET data: in particular, we move away from SUV R by estimating parametric maps of the [18F]FDG delivery (K1 [ml/cm3/min]) and phosphorylation (k3 [min−1]), and evaluate their peculiar regional distribution, never previously described at this level of spatial resolution. We proceed by assessing how these parameters, including the tracer uptake rate (Ki [ml/cm3/min]), interact not only with rs-fMRI features, but also with regional cerebral blood flow (CBF) and metabolic rate of oxygen (CMRO2), to have the most complete vision possible of these complex metabolic and hemodynamic relationships. Finally, we try to understand if a closer match between [18F]FDG and rs-fMRI information can be attained at the large-scale network level by obtaining a single- subject ‘metabolic connectivity’(MC) estimate, i.e., a PET counterpart to fMRI FC. To do so, we provide a completely new methodological framework for single- subject MC estimation, by employing a distance-based (and not a correlation- based) metric, and using kinetic modelling to differentiate MC matrices based on tracer inflow vs. metabolic events. These individual MC estimates are then com- pared to traditional across-subject covariation matrices of [18F]FDG parameters, and both are related to fMRI FC to understand which approach has a higher level of similarity.