Age-related disintegration in functional connectivity: Evidence from Reference Ability Neural Network (RANN) cohort

Abstract

Aging is typically marked by a decline in some domains of cognition. Some theories have linked this decline to a reduction in distinctiveness of processing at the neural level that in turn leads to cognitive decline. Increasing correlations with age among tasks formerly considered independent have been posited, supporting dedifferentiation, although results have been mixed. An alternative view is that tasks become more, and not less, independent of one another with increasing age, suggesting age-related differentiation, or what has also been termed disintegration. In the current study, we investigated if the aging process leads to a loss of behavioral and neural specificity within latent cognitive abilities. To this end, we tested 287 participants (20-80 years) on a battery of 12 in-scanner tests, three each tapping one of four reference abilities. We performed between-task correlations within domain (pertaining to convergent validity), and between domain (pertaining to discriminant validity) at both the behavioral and neural level and found that neural convergent validity was positively associated with behavioral convergent validity. In examining neural validity across the lifespan, we found significant reductions in both within- and between-domain task correlations, with a significant decrease in construct validity (convergent or discriminant) with age. Furthermore, the effect of age on total cognition was significantly mediated by neural construct validity. Taken together, contrary to a hypothesis of dedifferentiation, these correlation reductions suggest that tasks indeed become more independent with advancing age, favoring a differentiation/disintegration hypothesis of aging.

Keywords: Cognitive aging; Dedifferentiation; Disintegration; Reference ability neural networks; Task-based functional connectivity; Within-subjects fMRI.

Figure 1.. Behavioral correlations of within-domain versus between-domain task performance.

Left panel: Fisher’s Z correlation matrix of all behavioral task pairings. Y-labels denote the behavioral task and X-labels denote the corresponding reference ability. Square boxes indicate within-domain tasks. Centered values within each square indicate the average within-task correlation not considering a task’s perfect correlation with itself. Asterisks represent the between-task correlations that were used to later compute discriminant validity. Only between-domain correlations belonging to the same session were considered in order to control for session effects that could have inflated within-domain correlations. Right panel: Scatterplots of Fisher's Z coefficients divided by task pairing (convergent validity = within-domain; discriminant validity = between-domain). For convergent validity, correlation coefficients are organized by domain; for discriminant validity, correlation coefficients are organized by the session from which they were calculated (see legend). "M&F" indicates between-domain correlations for Memory and Fluid Reasoning and "S&V" indicates between-domain correlations for Speed and Vocabulary. The expanse of the box represents one standard deviation, the shaded middle region represents the standard error of the mean (SEM) for the 95% confidence interval, and the black line represents the mean.

Figure 2.. Neural correlations of within-domain versus between-domain task performance.

Left panel: Fisher’s Z correlation matrix of all task pairings of the functional connectivity values. Y-labels denote the task and X-labels denote the corresponding reference ability. Square boxes indicate within-domain tasks. Centered values within each square indicate the average within-task correlation not considering a task’s perfect correlation with itself. Asterisks represent the between-task correlations that were used to later compute discriminant validity. Only between-domain correlations belonging to the same session were considered in order to control for session effects that could have inflated within-domain correlations. Right panel: Scatterplots of Fisher's Z coefficients divided by task pairing (convergent validity = within-domain; discriminant validity = between-domain). For convergent validity, correlation coefficients are organized by domain; for discriminant validity correlation coefficients are organized by the session from which they were calculated (see legend). "M&F" indicates between-domain correlations for Memory and Fluid Reasoning and "S&V" indicates between-domain correlations for Speed and Vocabulary. The expanse of the box represents one standard deviation, the shaded middle region represents the standard error of the mean (SEM) for the 95% confidence interval, and the black line represents the mean.

Figure 3.

Scatterplot of neural validity plotted against behavioral validity. Lightly-dashed upper line is the least-squares line for convergent validity and the thicker-dashed lower line is the least-squares line for discriminant validity. As can be observed, there is a positive relationship between neural and behavioral convergent validity such that neural convergent validity significantly increases with increasing values of behavioral convergent validity (Z = .767, p = 0.024) as demonstrated via permutation analysis.

Figure 4.

Scatterplot of neural convergent and discriminant validity plotted per subject as a function of age. Lightly-dashed upper line is the least-squares line for convergent validity and the thicker-dashed lower line is the least-squares line for discriminant validity. As can be observed, there is a negative relationship between validity measures and age.

Figure 5.

Scatterplot of the relationship between neural validity and total cognition. Each dot represents one subject and the size of the dot is scaled to represent the age of the participant, with larger size indicating greater age. The black line represents the least-squares fit (Z = .3077), indicating that, as neural validity increases, total cognition increases. Furthermore, as can observed, increasing age is associated with lower scores on neural validity and total cognition.