Predictors of coupling between structural and functional cortical networks in normal aging

Abstract

Understanding how the mammalian neocortex creates cognition largely depends on knowledge about large-scale cortical organization. Accumulated evidence has illuminated cortical substrates of cognition across the lifespan, but how topological properties of cortical networks support structure-function relationships in normal aging remains an open question. Here we investigate the role of connections (i.e., short/long and direct/indirect) and node properties (i.e., centrality and modularity) in predicting functional-structural connectivity coupling in healthy elderly subjects. Connectivity networks were derived from correlations of cortical thickness and cortical glucose consumption in resting state. Local-direct connections (i.e., nodes separated by less than 30 mm) and node modularity (i.e., a set of nodes highly interconnected within a topological community and sparsely interconnected with nodes from other modules) in the functional network were identified as the main determinants of coupling between cortical networks, suggesting that the structural network in aging is mainly constrained by functional topological properties involved in the segregation of information, likely due to aging-related deficits in functional integration. This hypothesis is supported by an enhanced connectivity between cortical regions of different resting-state networks involved in sensorimotor and memory functions in detrimental to associations between fronto-parietal regions supporting executive processes. Taken collectively, these findings open new avenues to identify aging-related failures in the anatomo-functional organization of the neocortical mantle, and might contribute to early detection of prevalent neurodegenerative conditions occurring in the late life.

Keywords: FDG-PET; aging; cortical thickness; functional connectivity; functional-structural coupling; large-scale cortical networks; structural connectivity.

Figure 1

Analysis pipeline followed in this study. A. Cortical thickness estimation (left panel). T1‐MR images were segmented with Freesurfer to obtain individual cortical thickness maps that were further averaged to establish the cortical parcellation scheme. Cortical glucose consumption estimation (right panel). Partial volume effects were corrected on individual FDG‐PET cerebral images to obtain glucose consumption maps. B. Cortical parcellation. By applying a backtracking algorithm [Romero‐Garcia et al., 2012], the cortical surface was divided into 599 regions (with a surface area of 250 mm² each). Structural and functional cortical networks were based on thickness and glucose consumption, respectively, and adjacency matrices were obtained from this cortical scheme. C. Properties of connections. We calculated separately the influence of different structural connections (local direct, local indirect, global direct, and global indirect) on F‐S coupling for all network nodes, hubs (based on node degree and betweenness), and cortical regions grouped into modules. D. Left panel. Schematic representation of different types of hubs (local and global) and network modules (classification criteria are detailed in Materials and Methods). Right panel. Example of regional distribution of modular hubs in cortical networks. The color scale represents the proportion of sparsities for one specific region that reached the hub criterion.

Figure 2

Topological properties of structural (dark‐grey circles) and functional (light‐grey circles) cortical networks. Small‐worldness (σ), clustering coefficient (γ _g), and path length (λ _g) of networks computed with positive partial correlations. Note that small‐worldness was slightly enhanced in structural when compared with functional cortical networks. This enhancement was due to the natural trend of structural connectivity networks to form segregated clusters (middle panel). The integration ability of functional networks was slightly higher than that in structural networks at lower sparsities, although the two networks behave randomly at higher sparsities (right panel).

Figure 3

Overlapping of local and global hubs in structural and functional cortical networks. Overlapped hubs (based on node degree) were split into those with the highest number of direct connections (left column) and the highest number of indirect pathways (right column) in structural and functional networks, separately. The color scale represents the proportion of sparsities showing overlapped local/global hubs in both cortical networks. Note that overlapped local hubs were mostly allocated over unimodal association areas, whereas global hubs mainly appeared over heteromodal association areas. L = left; R = right.

Figure 4

Regional distribution of cortical modules overlapping local (blue scale) and global hubs (red scale) in structural and functional networks. The color scale represents proportions for a modular node reached the hub criterion (node degree) across sparsities. Those modular nodes overlapping both local and global hubs are represented in green. L = left; R = right.

Figure 5

Functional‐structural (F‐S) coupling as a function of node attributes and intrinsic properties of structural network connections. F‐S coupling was displayed for direct (dark‐grey circles) and indirect connections (light‐grey circles) for cortical regions communicated through both local (A) and global interactions (B) across different levels of sparsity in the structural network. This analysis was performed for all nodes in the network, and for those nodes with either high centrality (based on node degree or betweenness) or high modularity. Dashed lines represent the statistical threshold (p <0.05) for each condition. Values above the dashed line represent significant F‐S correlations. Note that local direct structural connections were excellent predictors of F‐S coupling in cortical networks supported by local (A), but not by global elements (B).

Figure 6

Functional‐structural (F‐S) coupling as a function of node attributes and intrinsic properties of functional network connections. F‐S coupling was displayed for direct (dark‐grey circles) and indirect connections (light‐grey circles) for cortical regions communicated through both local (A) and global interactions (B) across different levels of sparsity in the functional network. This analysis was performed for all nodes in the network, and for those nodes with either high centrality (based on node degree or betweenness) or high modularity. Dashed lines represent the statistical threshold (p <0.05) for each condition. Values above the dashed line represent significant F‐S correlations.