Age-related differences in network controllability are mitigated by redundancy in large-scale brain networks

Abstract

The aging brain undergoes major changes in its topology. The mechanisms by which the brain mitigates age-associated changes in topology to maintain robust control of brain networks are unknown. Here we use diffusion MRI data from cognitively intact participants (n = 480, ages 40-90) to study age-associated differences in the average controllability of structural brain networks, topological features that could mitigate these differences, and the overall effect on cognitive function. We find age-associated declines in average controllability in control hubs and large-scale networks, particularly within the frontoparietal control and default mode networks. Further, we find that redundancy, a hypothesized mechanism of reserve, quantified via the assessment of multi-step paths within networks, mitigates the effects of topological differences on average network controllability. Lastly, we discover that average network controllability, redundancy, and grey matter volume, each uniquely contribute to predictive models of cognitive function. In sum, our results highlight the importance of redundancy for robust control of brain networks and in cognitive function in healthy-aging.

Fig. 1. Study outline.

a Diffusion MRI data from 480 subjects from the HCP Aging dataset were used in our study. b We constructed structural networks using the functional Schaefer local-global parcellation with 17 networks and 400 ROIs. c For each subject, we calculated network controllability, a measure of a node’s ability to steer the brain into easy to reach states. d We studied the relationship between controllability and network redundancy in aging, testing the extent to which redundancy influences the relationship between age and network controllability. e We hypothesized that redundancy would mitigate the effects of age-associated differences in topology on average controllability in brain networks. f Finally, we investigated the extent to which grey matter volume, network controllability, and redundancy, can jointly predict age-associated variance in cognitive function.

Fig. 2. Hub and average network controllability are impacted by aging.

a The affiliations of hubs of average controllability in middle-aged subjects (ages 40–65) were predominately within the default mode network. Percentages were corrected by network size, which equalizes the probability of hubs falling within each network. b Distributions of average controllability for each hub, for middle- and old-aged participants (ages 65–90). Two hubs in the default mode network exhibited less average controllability in old-aged participants. c Mean network average controllability was negatively associated with age in the default mode network (DefaultA), control network (ContB), and limbic network (LimbicB). In the group comparisons (Panel b) and the rank-correlations (Panel c), participant education was included as a covariate. The Bonferroni method for correction for multiple comparisons was applied to correct for the number of hubs analyzed (16) (Panel b). and the number of networks (17) (Panel c). *corrected P < 0.05, **corrected P < 0.001.

Fig. 3. Multi-step connectivity (redundancy) mediates relationships between age and mean network average controllability over and above the effects of degree.

a Average network degree was negatively associated with age for 15 of 17 large-scale networks. b Changes in degree mediated the relationship between age and mean network average controllability for 14 of 17 networks. c Average network redundancy also showed age associated declines, but for all networks examined. d Average network redundancy mediated relationships between age and average controllability for 3 of 17 networks when including degree as a covariate. These networks included the visual (VisPeri), dorsal attention (DorsAttnB), and default mode (DefaultB) networks. In each analysis participant education was included as a covariate. We used the Bonferonni method to correct for multiple comparisons. In each panel we corrected for the number of networks analyzed (17). In panels b and d, the mediation was significant if the confidence intervals did not cross 0 when the α = 0.05/17 to correct for multiple comparisons. Significant mediations are indicated by black confidence intervals, while non-significant mediations are indicated by grey confidence intervals.

Fig. 4. Associations between grey matter volume (GM), mean network average controllability, and redundancy, and processing speed.

a Mean average controllability in the frontoparietal control (ContB), and default mode (DefaultB) networks was positively associated with processing speed. b Processing speed was positively associated with redundancy in 5 of 17 networks (all p_bonf.’s < 0.05). c Total hippocampal volume did not significantly change until around the age of 67, after which it showed a negative association with age. d For participants older than 66.92, IC volume-adjusted total hippocampal volume was positively associated with processing speed. e Performance of a general linear models when predicting processing speed with measures of GM volume, average controllability, and redundancy. For GM, we used IC-volume-adjusted measures of hippocampal volume, subcortical volume, and cortical volume. f The z-scored predicted processing speed versus real z-scored processing speed for the best model shown in e. In panels a and b, we used the Bonferroni method to correct for multiple comparisons based on the number of networks analyzed (17). For panels d and e, measures of processing speed and GM volume were z-scored. In the rank-correlations performed in panels a and b, participant education was included as a covariate.