Hubs of long-distance co-alteration characterize brain pathology

Abstract

It is becoming clearer that the impact of brain diseases is more convincingly represented in terms of co-alterations rather than in terms of localization of alterations. In this context, areas characterized by a long mean distance of co-alteration may be considered as hubs with a crucial role in the pathology. We calculated meta-analytic transdiagnostic networks of co-alteration for the gray matter decreases and increases, and we evaluated the mean Euclidean, fiber-length, and topological distance of its nodes. We also examined the proportion of co-alterations between canonical networks, and the transdiagnostic variance of the Euclidean distance. Furthermore, disease-specific analyses were conducted on schizophrenia and Alzheimer's disease. The anterodorsal prefrontal cortices appeared to be a transdiagnostic hub of long-distance co-alterations. Also, the disease-specific analyses showed that long-distance co-alterations are more able than classic meta-analyses to identify areas involved in pathology and symptomatology. Moreover, the distance maps were correlated with the normative connectivity. Our findings substantiate the network degeneration hypothesis in brain pathology. At the same time, they suggest that the concept of co-alteration might be a useful tool for clinical neuroscience.

Keywords: Alzheimer's disease; VBM; network degeneration; pathology spread; physical distance; schizophrenia; topological distance; transdiagnostic.

FIGURE 1

PRISMA flowchart of the selection of studies

FIGURE 2

Illustration of the methods. (a) For each experiment included in a given meta‐analysis (decreases or increases), a modeled alteration (MA; or activation) map is created placing a Gaussian distribution of probability around each reported focus of alteration (or activation). (b) For each experiment, each region of the atlas is considered to be altered if the 20% of its voxels is covered by the MA corresponding to that experiment. Thus, for each experiment, we obtain a vector of dichotomous values that represent the status of every region. (c) The Patel's κ is calculated between each one of such vectors and all the others, obtaining a network of co‐alterations (or co‐activations). Here, the co‐alteration matrix and the network of the voxel‐based morphometry decreases are shown. (d) Any network analysis can be computed on the resulting matrix. For instance, a measure of distance can be associated to each edge x _i,j to calculate the mean distance of each node. Otherwise, a centrality measure such as the degree can be derived. (e) A statistical map can be produced assigning to the volumes of the atlas the values calculated for each node

FIGURE 3

Parametric mapping of the mean distances of co‐alterations, divided for decreases and increases. Higher values indicate increasing mean distance. Axial slices are in radiological convention

FIGURE 4

Top left panel: Surface mapping of the co‐activation degree centrality and mean Euclidean distance. Top right panel: Radar chart comparing the average values of mean Euclidean distances of decreases and increases, mean Euclidean distance, and degree centrality of co‐activation for each of the following networks: visual networks 1, 2 and 3 (V1, V2, and V3), orbitofrontal cortex (OFC), cerebellum, dorsal attentional network (DAN), thalamus and basal ganglia, auditory network, premotor network, salience network, default mode network (DMN), ventral attentional network (VAN), and sensorimotor network. Middle left panel: Surface mapping of the mean fiber‐length and structural topological distances. Middle right panel: Radar chart comparing the average values of mean fiber‐length and topological distances of decreases, increases, and structural connectivity for each canonical network. Bottom left panel: Surface mapping of the network‐betweenness (between‐network edges/total number of edges. Bottom right panel: Mean network‐betweenness for each canonical network

FIGURE 5

Two‐dimensional representation of the decreases and increases co‐alteration networks, organized according to the belonging of each node to each one of the following resting‐state networks: visual networks 1, 2, and 3 (V1, V2, and V3), orbitofrontal cortex (OFC), cerebellum, dorsal attentional network (DAN), thalamus and basal ganglia, auditory network, premotor network, salience network, default mode network (DMN), ventral attentional network (VAN), and sensorimotor network

FIGURE 6

Top panel: Scatter plots of co‐activation and co‐alteration values of each area of the Talairach Daemon atlas. Bottom panel: Parametric mapping of the voxels' contribution to correlation analysis between the functional meta‐analytic degree centrality map and the functional meta‐analytic mean distance map with the mean distance co‐alteration maps of decreases and increases. Positive values indicate areas of concordance between the maps, negative values indicate discordance. Axial slices are in radiological convention

FIGURE 7

Top panel: Surface mapping of the transdiagnostic variance of the mean Euclidean distance maps of decreases and increases. Higher values indicate higher variance, that is, the voxels whose value in the mean distance map is more variable across pathologies

FIGURE 8

Top panel: Parametric mapping of the mean Euclidean distance of co‐alterations in schizophrenia, divided for decreases and increases. Higher values indicate increasing mean distance. Bottom panel: Parametric mapping of the anatomical likelihood estimation of decreases and increases of schizophrenia. Axial slices are in radiological convention

FIGURE 9

Top panel: Parametric mapping of the mean Euclidean distance of co‐alterations in Alzheimer's disease, divided for decreases and increases. Higher values indicate increasing mean distance. Bottom panel: Volumetric mapping of the anatomical likelihood estimation of decreases and increases of schizophrenia. Axial and coronal slices are in radiological convention