The search for the underlying mechanisms of age-related cognitive decline is motivated by our goal to preserve the benefits of increased longevity through reducing functional losses. Task-free fMRI provides information about the integrity of several highly reproducible intrinsic connectivity networks (ICNs) and is well suited for characterizing age and ageing related changes in brain function as participant input is minimized ¹.

Three ICNs are particularly relevant to studying age-related cognitive loss: the default mode network (DMN), the executive control network (ECN), and the salience network (SN), which interact to balance internally and externally driven cognitive processes ². Cross sectional studies of older adults have highlighted the loss of functional specialization evidenced by decreased intra-network FC in these ICNs ³ as well as changes in functional segregation evidenced by changes in inter-network FC between ICNs ⁴. These findings need to be verified in longitudinal studies because extrapolating cross-sectional findings to predict the effects of ageing is not always appropriate ⁵. To this end, we examined the longitudinal intra- and inter-network FC changes within and between the three ICNs and their relationships with cognitive performance in a cohort of relatively healthy older adults.

Seventy-eight relatively healthy Chinese older adults from the Singapore-Longitudinal Ageing Brain Study ⁶ visited the center twice or thrice in a span of four years. Each visit comprised an 8-min eyes-opened task-free fMRI scan with fixation and neuropsychological assessment on five cognitive domains (processing speed, attention, verbal memory, visuospatial memory, and executive functioning).

Functional and structural images were preprocessed using a pipeline based on FSL and AFNI ⁷. Global signals were removed from the functional images. Seventy-four cortical ROIs corresponding to DMN, ECN, and SN were identified ⁸. For each participant and at each time point, functional connectivity (Fisher’s z transformed Pearson’s correlation) was computed between all ROI pairs and averaged according to which ICN (intra-network FC) or ICN pair (inter-network FC) the FC measures belonged to. Longitudinal changes in FCs and cognitive performance were then modeled using the linear mixed model ⁹, in which FC or cognitive performance was predicted by the longitudinal ageing effect (years since first visit) that was moderated by age (age at first visit), i.e., ageing-by-age interaction. Finally, the brain-cognition associations in their longitudinal trends were evaluated between FCs and cognitive domains that showed an ageing effect by regressing the predicted longitudinal change in cognitive performance on the predicted longitudinal change in FC and age.

There were significant longitudinal decreases in intra-network FC within DMN (p = 0.007) and ECN (p = 0.044) (Figure 1), and a marginal decrease within SN (p = 0.054). These indicate reduced functional specialization within these higher-order cognitive networks ³. Additionally, there was a significant ageing by age interaction involving ECN-DMN inter-network FC (p = 0.032). The aggregate of individual trajectories of ECN-DMN inter-network FC was u-shaped with respect to age (Figure 2), indicating that younger elderly was able to maintain or elevate functional segregation, an ability that was progressively lost in older elderly. This suggests initial compensatory efforts ¹⁰ accompanied by brain network reorganization ¹¹ that end with declining functional segregation of networks in older age ⁴.

Regarding cognitive performance, only processing speed, a robust marker of age-related cognitive decline, showed unequivocal decline with ageing (p = 0.002). Importantly, its longitudinal change was associated with the longitudinal change in inter-network FC between the ECN and DMN p = 0.03) such that faster decline in inter-network anti-correlation (ECN-DMN) was associated with more rapid decline in processing speed (Figure 3). This relationship may indicate that degradation in functional segregation compromises the balance between task-positive and task-negative activities, or links to the failure to differentiate goal related and task-irrelevant information ¹⁰.