Age-related reorganization of functional network architecture in semantic cognition

Abstract

Cognitive aging is associated with widespread neural reorganization processes in the human brain. However, the behavioral impact of such reorganization is not well understood. The current neuroimaging study investigated age differences in the functional network architecture during semantic word retrieval in young and older adults. Combining task-based functional connectivity, graph theory and cognitive measures of fluid and crystallized intelligence, our findings show age-accompanied large-scale network reorganization even when older adults have intact word retrieval abilities. In particular, functional networks of older adults were characterized by reduced decoupling between systems, reduced segregation and efficiency, and a larger number of hub regions relative to young adults. Exploring the predictive utility of these age-related changes in network topology revealed high, albeit less efficient, performance for older adults whose brain graphs showed stronger dedifferentiation and reduced distinctiveness. Our results extend theoretical accounts on neurocognitive aging by revealing the compensational potential of the commonly reported pattern of network dedifferentiation when older adults can rely on their prior knowledge for successful task processing. However, we also demonstrate the limitations of such compensatory reorganization and show that a youth-like network architecture in terms of balanced integration and segregation is associated with more economical processing.

Keywords: aging; functional connectivity; graph theory; language production; semantic memory.

Fig. 1

Neuropsychological results and experimental design. a) Test scores were z-transformed. Higher z-values signify better performance. STW, spot-the-word test; DSST, digit symbol substitution test; and TMT trail-making test. ^***P < 0.001, ^**P < 0.01, and ^*P < 0.05. b) The fMRI experiment consisted of task blocks of overt paced semantic fluency and counting, which were presented in a pseudo-randomized order and separated by rest periods. An example for each task is shown. Participants were instructed to produce exactly one exemplar for a category or to say one number when the fixation cross turned green and to pause when the cross turned red. If they could not think of an exemplar, they were instructed to say “next.” Each task block contained 9 trials of the same semantic category/counting task, which were separated by jittered interstimulus intervals.

Fig. 2

Behavioral results. a) Both groups performed better for counting than semantic fluency. Although there was no effect of age for accuracy in either task, older adults performed slower than young adults during semantic fluency but not during counting. b) Both groups performed better for easy than difficult semantic categories, and young adults better than older adults for easy semantic categories. Furthermore, young adults were generally faster in responding during semantic fluency than older adults, independent of difficulty level. Points show mean response times with 2 standard deviations, ^*P < 0.05.

Fig. 3

ICA-derived networks and their overlap with cognitive networks. T-scores from 1-sided t-tests (FWE-corrected P < 0.05 at peak level) are displayed for the 7 selected component maps with their respective network label according to spatial similarity analysis. Overlaps between the thresholded component map and the spatially most similar cognitive network according to the Jaccard index are outlined on the surface of the brain. The areas of overlap were used for subsequent network analyses.

Fig. 4

Functional coupling between task-relevant networks in young and older adults and their behavioral relevance. a) Chord diagrams display significant results of functional coupling between ICA-derived networks. Connectivity values are partial correlations. The color intensity and width of a connection indicate its correlational strength. Higher z values indicate positive coupling and negative values indicate decoupling between networks. Chord diagrams of each age group are based on cPPI-derived significance values, whereas age differences were assessed using permutation testing in network-based statistics (cluster-forming threshold P = 0.01, FWE-corrected significance threshold P = 0.025 with 10,000 permutations). b) Network connections that showed significant age differences were probed for their behavioral relevance. Plots show significant 2-way interactions between age and the respective network pair for accuracy and response time data. Connectivity values were mean-centered for interaction analyses. Results were corrected for multiple comparisons using the Bonferroni–Holm method at P = 0.05. VAN, ventral attention network; DAN, dorsal attention network.

Fig. 5

Age-related differences in whole-brain segregation and integration and their behavioral relevance. a) For each participant, a task-related brain network graph was constructed using 121 nodes. The nodes were based on significant global and local peak maxima of the 7 networks derived from the ICA (see Supplementary Table S3 for exact locations of nodes). b) Spring-embedded graphs depicting age differences in the modular organization of the brain. Graphs are based on average connectivity in each age group. Stronger segregation is reflected by higher within- and lower between-network correlations. In comparison, young adults show stronger segregation than older adults for most networks. For visualization purposes, graphs are displayed at 5% graph density. c) Brain-wide system segregation was higher for young adults and had distinct effects on behavior for each age group with young adults profiting from increasing segregation. d) A different picture emerged for global efficiency. Global efficiency was calculated for individual orthogonal minimum spanning trees (OMST), which were based on weighted correlation matrices. The graphs of young adults showed stronger global efficiency than older adults. While increasing global efficiency was associated with better performance in both age groups, it predicted slower performance in young and faster performance in older adults. Note that segregation and global efficiency values were mean-centered for analyses with behavior.

Fig. 6

Segregation of individual networks is associated with distinct behavior of older and young adults as a function of system type. a) Individual networks’ segregation values by age. All networks showed stronger segregation in young adults. b) Generalized linear mixed-effects models for accuracy revealed significant interactions with age and network segregation for 2 systems, whereas c) linear mixed-effects models for response time showed significant interactions for 3 networks. For most networks, increasing segregation was associated with better and faster performance in young adults and worse and slower reactions in older adults. d) Significant correlations between network segregation and neuropsychological measures. For young adults, we detected a positive correlation of increasing segregation of VAN-B with executive functions, whereas for older adults, a negative correlation of increasing segregation of DMN-B with semantic memory was found. Note that segregation and global efficiency values were mean-centered for analyses with behavior. VAN, ventral attention network; DAN, dorsal attention network.

Fig. 7

Topology of network hubs in young and older adults. a) The normalized participation coefficient (PC) was calculated for individual orthogonal minimum spanning trees (OMST). Graphs display the PC of each node for the average OMST in each age group (top). For visualization purposes, the strongest 5% of connections are shown. Stronger PC values are reflected by color and node size. The higher the PC, the more a node is connected with nodes from other communities. The node with the highest PC value in each age group is extracted and displayed with its neighboring nodes colored by community (middle). Note that these connector hubs are connected to many different communities. Connector hubs were defined in each age group via PC values at least 1SD above the mean. In both groups, connector hubs were detected in frontal, parietal, and temporal regions (bottom). b) A linear model with age as predictor revealed nodes with stronger PC only in older adults. The top and middle graphs were plotted using the ForceAtlas2 algorithm. The force-directed layout causes nodes of the same community to cluster together and diversely connected hubs (connector hubs) to appear in the center of the graph.