A more randomly organized grey matter network is associated with deteriorating language and global cognition in individuals with subjective cognitive decline

Abstract

Objectives: Grey matter network disruptions in Alzheimer's disease (AD) are associated with worse cognitive impairment cross-sectionally. Our aim was to investigate whether indications of a more random network organization are associated with longitudinal decline in specific cognitive functions in individuals with subjective cognitive decline (SCD).

Experimental design: We included 231 individuals with SCD who had annually repeated neuropsychological assessment (3 ± 1 years; n = 646 neuropsychological investigations) available from the Amsterdam Dementia Cohort (54% male, age: 63 ± 9, MMSE: 28 ± 2). Single-subject grey matter networks were extracted from baseline 3D-T1 MRI scans and we computed basic network (size, degree, connectivity density) and higher-order (path length, clustering, betweenness centrality, normalized path length [lambda] and normalized clustering [gamma]) parameters at whole brain and/or regional levels. We tested associations of network parameters with baseline and annual cognition (memory, attention, executive functioning, language composite scores, and global cognition [all domains with MMSE]) using linear mixed models, adjusted for age, sex, education, scanner and total gray matter volume.

Principal observations: Lower network size was associated with steeper decline in language (β ± SE = 0.12 ± 0.05, p < 0.05FDR). Higher-order network parameters showed no cross-sectional associations. Lower gamma and lambda values were associated with steeper decline in global cognition (gamma: β ± SE = 0.06 ± 0.02); lambda: β ± SE = 0.06 ± 0.02), language (gamma: β ± SE = 0.11 ± 0.04; lambda: β ± SE = 0.12 ± 0.05; all p < 0.05FDR). Lower path length values in precuneus and fronto-temporo-occipital cortices were associated with a steeper decline in global cognition.

Conclusions: A more randomly organized grey matter network was associated with a steeper decline of cognitive functioning, possibly indicating the start of cognitive impairment.

Keywords: Alzheimer's disease; MRI; cognition; connectivity; graph theory; grey matter network; longitudinal; mild cognitive impairment; subjective cognitive decline.

Figure 1

(a) example of single‐subject grey matter network extraction. (Step 1) Grey matter segmentations are divided in regions of interest (ROI) of 3 × 3 × 3 voxels. (Step 2) Connectivity is defined statistical similarity between two ROIs as computed with the Pearson's correlation of grey matter intensity values across corresponding voxels in the ROIs. (Step 3) All similarity values are collected in a similarity matrix. (Step 4) ROIs are connected when their similarity value exceeds a statistical threshold determined with a random permutation method. Here a toy model is shown with an example connectivity density of 23% (i.e., 7 out of 30 possible connections present). (b) Schematic representation of network parameters. A node represents a ROI, and an edge the connection between nodes. The degree is the number of edges of a node, in this example the degree of the green node is 5. Path length is the minimum number of edges between a pair of nodes, in this the path length between the green and orange nodes is 3. Clustering coefficient quantifies to what extent neighbors of a given node are connected, which is 1/3 for the green node as 1 from the 3 possible connections exists. Betweenness centrality is the proportion of paths that run through a node, which is maximal for the green node, and zero for all other nodes. (c) Example of a network with a small‐world organization (left) and with a random organization (right) [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 2

(a) Associations between gamma (i.e., normalized clustering) values and memory changes over time. (b) Associations between lambda (i.e., normalized path length) values and global cognitive changes over time. (c) Associations between clustering values and language changes over time. (d) Associations between small world network values and language changes over time. Predicted changes over time (fixed effects) were obtained with the fitted linear mixed models on the original data. (e) Surface plot of AAL areas where lower path length values were associated with steeper decline of global cognition (p < 0.05 FDR‐corrected) [Color figure can be viewed at http://wileyonlinelibrary.com]