Brain Molecular Connectivity in Neurodegenerative Diseases: Recent Advances and New Perspectives Using Positron Emission Tomography

Abstract

Positron emission tomography (PET) represents a unique molecular tool to get in vivo access to a wide spectrum of biological and neuropathological processes, of crucial relevance for neurodegenerative conditions. Although most PET findings are based on massive univariate approaches, in the last decade the increasing interest in multivariate methods has paved the way to the assessment of unexplored cerebral features, spanning from resting state brain networks to whole-brain connectome properties. Currently, the combination of molecular neuroimaging techniques with multivariate connectivity methods represents one of the most powerful, yet still emerging, approach to achieve novel insights into the pathophysiology of neurodegenerative diseases. In this review, we will summarize the available evidence in the field of PET molecular connectivity, with the aim to provide an overview of how these studies may increase the understanding of the pathogenesis of neurodegenerative diseases, over and above "traditional" structural/functional connectivity studies. Considering the available evidence, a major focus will be represented by molecular connectivity studies using [18F]FDG-PET, today applied in the major neuropathological spectra, from amyloidopathies and tauopathies to synucleinopathies and beyond. Pioneering studies using PET tracers targeting brain neuropathology and neurotransmission systems for connectivity studies will be discussed, their strengths and limitations highlighted with reference to both applied methodology and results interpretation. The most common methods for molecular connectivity assessment will be reviewed, with particular emphasis on the available strategies to investigate molecular connectivity at the single-subject level, of potential relevance for not only research but also diagnostic purposes. Finally, we will highlight possible future perspectives in the field, with reference in particular to newly available PET tracers, which will expand the application of molecular connectivity to new, exciting, unforeseen possibilities.

Keywords: FDG–PET; amyloid PET; brain networks; connectivity; multivariate analysis; neurodegenerative diseases; neurotransmission; tau PET.

FIGURE 1

Schematic representation of the most common analytical approaches for molecular connectivity modeling. (A) Seed correlation analysis: molecular connectivity estimation is performed from a specific seed of interest, selected by the researcher. Here, the seed corresponds to a cluster encompassing the precuneus/posterior cingulate cortex; the resulting connectivity map corresponds to the default mode network. (B) Independent component analysis: whole-brain tracer uptake signal is decomposed into multiple statistically independent components, at voxel-wise level. The number of extracted components is set by the researcher. Here, N = 20 components are extracted (only six are shown for visualization purposes). By comparing the topography of the identified components with known anatomo-functional networks, the researcher can then select the components of interest. Here, three selected components are shown for visualization purposes, corresponding to the primary visual network, executive network, and default mode network. (C) Partial correlation analysis: molecular connectivity estimation is performed from ROIs selected by the researcher. Here, the ROIs comprehensively cover the whole brain. If the number of subject is smaller than the number of ROIs, sparse inverse covariance estimation (SICE) is used to estimate a connectivity matrix. This is then translated into a network, where nodes are represented by ROIs and edges by molecular connections. Here, a weighted connectivity matrix and weighted networks are shown, with edges computed at three different network densities. A wide array of graph theory metrics can then be estimated. BrainNet Viewer (http://www.nitrc.org/projects/bnv/) was used for rendering (Xia M. et al., 2013). ROIs, regions of interest.

FIGURE 2

Whole-brain metabolic connectome in Dementia with Lewy bodies. (A) Brain connectivity graphs in healthy controls and patients with dementia with Lewy bodies. A global connectivity reconfiguration is evident in patients with dementia with Lewy bodies, with metabolic connectivity decreases mainly affecting occipital cortex, thalamus, and cerebellum. Only the strongest connections (density ≈ 3%) are shown (yellow edges). The size of each node depends on the node total number of connections, whereas the color indicates its anatomical localization. (B) The T-score matrix reports T-test statistics, derived from the direct comparison of the number of metabolic connections (within and between each macroarea) between patients and healthy controls, following a bootstrapping procedure. Connectivity decreases are indicated by negative T-scores (green); connectivity increases by positive T-scores (violet). BrainNet Viewer (http://www.nitrc.org/projects/bnv/) was used for rendering (Xia M. et al., 2013). Modified from Caminiti et al. (2017c). F, frontal; PCL, paracentral lobule; MCC, median cingulate cortex; ROL, rolandic operculum; P, parietal; O, occipital; T, temporal; In, insula; Th, thalamus; BG, basal ganglia; BS, brainstem; Cbl, cerebellum.

FIGURE 3

Metabolic connectivity targeting cholinergic pathways. Left panel: selection of ROIs for assessment of metabolic connectivity within the major cholinergic pathways. The first row shows regions supplied by basal forebrain (Ch1-2; Ch3) and brainstem (Ch5-6) nuclei. The second row shows the three cholinergic pathways projecting from the nucleus basalis of Meynert (Ch4). Right panel: histograms show the prevalence of regional connectivity decreases in dementia with Lewy bodies, in each cholinergic pathway. Prevalence of metabolic connectivity decreases in each pathway was computed as the number of regions, within the pathway, presenting with significantly decreased metabolic connectivity, divided by the total number of regions belonging to that pathway. Only connections within each pathway (i.e., between regions innervated by the same cholinergic nucleus) were taken into account to compute the prevalence of connectivity decreases within each pathway. Prevalence of metabolic connectivity decreases was higher in regions supplied by Ch1–Ch2 nuclei (100% of regions presenting with significantly decreased metabolic connectivity within this pathway) and Ch5–Ch6 nuclei, with additional involvement of the medial and lateral projection (capsular subdivision) of the nucleus basalis of Meynert (Ch4). BrainNet Viewer (http://www.nitrc.org/projects/bnv/) was used for rendering (Xia M. et al., 2013). Modified from Caminiti et al. (2017c). Ch4m, Ch4 medial projections; Ch4lc, Ch4 lateral projections, capsular subdivision; Ch4lp, Ch4 lateral projections, perisylvian subdivision.