By 2050, the total number of older adults (i.e. individuals ≥ 60 years old) worldwide is expected to reach 2.1 billion (Prince et al., 2015). While improvements in life expectancy is a significant achievement of the 21st century, population aging represents a major societal challenge (World Health Organization, 2015). This is because older adults are often at a greater risk of developing certain health conditions than their younger counterparts. For instance, an individual aged 90 or older has 25 times the risk of developing dementia compared to an individual in their late 60s (Yip et al., 2006; Brayne, 2007). Whole population rates of dementia, such as Alzheimer’s disease, are projected to triple by 2050 as a consequence of population aging (Patterson, 2018). Consequently, there has been growing scientific interest in identifying modifiable factors that may reduce the risk of developing dementia and contribute to “better” brain health in late life.

Leisure activities (e.g., visiting friends and family, reading, and going to the cinema or museum) represent one set of modifiable factors that potentially support healthy cognitive aging, by promoting brain plasticity (Stern, 2012). For example, higher activity participation has been systematically linked to better cognitive performance, higher regional and global gray matter (GM) volume, fewer volumetric measures of WM lesions and less decline in the quality of white matter (WM) tracts (Fratiglioni et al., 2004; Wang et al., 2012; Sexton et al., 2013; Erickson et al., 2014; Yates et al., 2016; Anatürk et al., 2018; Chan et al., 2018; Evans et al., 2018; Matyas et al., 2019; Wassenaar et al., 2019). Variability in functional connectivity measures also appears to be partly accounted for by activity engagement, particularly for physically demanding activities (Stillman et al., 2019). However, many of these studies have employed composite measures of leisure activities, providing limited insights into the specific activities that need to be targeted to promote brain health in older individuals. Targeting non-optimal activities may in part explain the limited efficacy of current randomized-controlled trials (RCTs) on cognitive and neural outcomes (Mortimer et al., 2012; Stephen et al., 2019).

A small number of epidemiological studies have begun to shift away from the composite approach when examining the link between leisure activities and the aging brain, finding that not all activities equally contribute to the risk of cognitive impairment (Krell-Roesch et al., 2017, 2019; Fancourt et al., 2018). For example, Fancourt et al. (2018) examined data collected from 3,911 participants enrolled in the English Longitudinal Study of Aging and found that adults who visited museums, art galleries and exhibitions on a regular basis, had a lower incidence of dementia over the course of 10 years. The association between cultural activities and dementia incidence appeared to be robust, as it remained significant after adjusting for how frequently these individuals were involved in community-based activities (e.g., social clubs, volunteering, sports clubs), alongside sociodemographic (i.e., age, sex, marital status, education, employment status, wealth, and previous occupational classification), and health-related co-variates (i.e., eyesight, depression, hearing, and existing cardiovascular health conditions). Importantly, a study that aims to evaluate whether different activities relate to markers of brain health requires a comparably larger number of univariate tests, relative to when a single composite measure of leisure activity is of interest. A range of key factors also need to be appropriately adjusted for when investigating the association between leisure activities and the aging brain, due to the bias that these confounding variables can otherwise introduce in the results and conclusions of a study, including inflated effect sizes or even spurious findings. For example, education and socio-economic status (SES) are determinants of how often a person engages in cultural activities such as visiting a museum, art gallery and exhibition (Grisolía et al., 2010; Mak et al., 2020). As low levels of education and SES are linked to an increased risk of dementia (Cadar et al., 2018; Livingston et al., 2020), demonstrating an independent relationship between cultural activities and the dementia risk strengthens the argument that these activities potentially contribute to maintaining brain health in older ages. The results of such correlational studies can then help guide randomized controlled trials (RCTs) to focus their intervention programs on a specific subset of activities, to determine whether these reported associations translate to direct effects on the aging brain. With a sample that currently exceeds several thousand well-characterized individuals, the United Kingdom Biobank (UK Biobank) study offers the statistical power to employ a more fine-grained approach to investigate the relationship between activities and markers of brain health, after adjusting for a wide range of confounders. This cohort study also offers measures of activity levels at more than one timepoint, allowing us to capture how longitudinal patterns of activities relate to metrics of brain health.

The aim of this study was to investigate whether individual leisure activities relate to MRI measures of GM volume, WM microstructure, WM lesions, resting-state functional connectivity, and cognitive function in late life. We examined six leisure activities that were available in the UK Biobank: going to a pub or social club, undertaking a religious activity, attending adult education classes, going to a sports club or gym, visiting friends and family and leisure-time computer use. Based on their self-reported activity levels at two timepoints, individuals were divided into one of four groups: (1) stable weekly participation, (2) stable low participation, (3) low to weekly participation, and (4) weekly to low participation. We predicted that weekly activity participation (both stable and increased participation over time) would associate with higher global cognitive function (Brown et al., 2012; Mitchell et al., 2012; Wang et al., 2012; Yates et al., 2016; Evans et al., 2018). We further expected that consistently high or increased participation in each leisure activity would correlate with greater structural integrity, including greater regional GM and higher WM integrity (i.e., higher fractional anisotropy (FA), lower mean diffusivity (MD; Anatürk et al., 2018). Given the limited evidence investigating activity-specific effects on resting-state functional connectivity, no predictions were made regarding this modality, so any results would be exploratory.

We present the largest multi-modal study to examine whether patterns of engagement in leisure activities associate with markers of brain structure, functional connectivity, and cognition. We found that stable weekly participation in a sports club or gym correlated positively with connectivity of the sensorimotor network. Additionally, weekly visits to or from family and friends was associated with higher volumetric measures of GM in the occipital pole. However, apart from consistent leisure time computer use, no other activity was significantly linked to global cognition. Each of these results are discussed, in turn.

One of our main findings was that consistent, weekly visits to a gym or sports club correlated with greater absolute connectivity of the sensorimotor network and cerebellar network, relative to infrequent attendance over time. Similar activation patterns at rest and during task execution indicate that both the sensorimotor and cerebellar networks play a role in motor tasks (Smith et al., 2009), potentially interacting with one another to support motor behavior. Subdivisions of the cerebellum are typically found within the sensorimotor network (Habas et al., 2009; Dobromyslin et al., 2012), an observation that was also replicated in our study (Supplementary Figure 3). The present findings additionally identified a link between connectivity of the sensorimotor and lateral visual network in individuals reporting consistent gym and sports club attendance across time, suggesting that sustained activity engagement contributes to enhanced visual-motor coupling. Potentially underlying this relationship is the importance of visual information in regulating and updating current behavior during physical exercise (Mallek et al., 2017). Our findings are in line with a previous study reporting strengthened connectivity between regions involved in the sensorimotor network, cerebellar and visual network, following a 13-week cognitive and motor skills training intervention (Demirakca et al., 2016). Task based fMRI has also highlighted a greater BOLD response in regions involved in these networks (Domingos et al., 2021). For example, the supplementary motor areas, precentral gyrus, angular gyrus and supramarginal gyrus were among the areas to demonstrate greater BOLD activation during a famous face recognitions task, in a group of adults (age range = 65 to 85) reporting high physical activity levels (n = 34) relative to a low physical activity group (n = 34, Smith et al., 2011). However, stronger connectivity between resting-state networks in an aging sample is not entirely straightforward to interpret. For instance, older adults often exhibit less segregated sensorimotor networks (i.e., lower within-network connectivity and higher between-network connectivity) when compared to younger adults, with these increases in connectivity linked to poorer task performance (e.g., reaction time, dexterity; Cassady et al., 2019). Additionally, physical activity levels did not account for the associations observed across the sensorimotor-visual-cerebellar networks, suggesting that other factors (e.g., physical fitness or socially interacting with others) may support these associations.

A difficulty in interpreting our findings is that while gym/sports club attendance was linked to greater connectivity between the visual and sensorimotor networks, no parallel associations were found between this activity and cognition, despite an established link between physical activity participation and cognitive function (Kelly et al., 2014; Erickson et al., 2019). The absence of a relationship with global cognition may be partly explained by the lower statistical power present in the analyses of cognitive function relative to the MRI analyses, due to the smaller sample size (1,734 individuals versus 7,152 individuals in the MRI sample). As we focused our analyses on global cognition, our study may have also failed to detect specific cognitive domains that are sensitive to gym/sports club participation (e.g., executive function (Erickson et al., 2019). Future studies examining cognitive subdomains could therefore provide novel insights into the cognitive correlates of these types of activities. Additionally, we were only able to generate a crude measure of activity participation based on the measures available in the UK Biobank study, which potentially hampered our ability to detect a link between gym/sports club attendance and cognitive function. Specifically, the granularity of our activity measure may have prevented us from capturing sufficient inter-individual variation in activity levels, such as between individuals who attended a gym/sports club once a week to those who attend it on a more regular basis (e.g., five times a week). Further work would therefore benefit from using more fine-grained assessments of activities (e.g., duration and frequency), as this approach could offer enhanced sensitivity in detecting associations between gym/sport club attendance and cognitive function. A final and more general consideration is that attending the gym or sports clubs is unlikely to serve as an accurate proxy of overall physical activity levels, as many activities can be performed outside of these settings (e.g., running and cycling). This could explain why our study failed to replicate well-established links between physical activity, GM and WM outcomes (Cheng, 2016; Sexton et al., 2016; Wassenaar et al., 2019).

Another of our key findings was that consistently visiting friends and family on a weekly basis associated with higher GM volume in the occipital pole, relative to low and stable patterns. This finding offers partial support for our second hypothesis. Two previous studies have also reported a relationship between higher social activity levels and higher GM volume within this region (James et al., 2012; Arenaza-Urquijo et al., 2016), consistent with the “brain maintenance” (Nyberg et al., 2012) or “brain reserve” (Stern, 2012) hypotheses. However, other studies have not replicated this association (Foubert-Samier et al., 2012; Seider et al., 2016; Anatürk et al., 2020), or otherwise report associations across frontal, parietal and temporal regions (Valenzuela et al., 2008; Arenaza-Urquijo et al., 2015; Seider et al., 2016). The discrepant findings may be due to the use of a cross-sectional assessment of activities in most of these studies or alternatively a lack of assessment into activity-specific relationships. While the frontal and temporal regions are considered to be the most age-sensitive, GM in the occipital cortex also demonstrates a negative relationship with age (e.g., Allen et al., 2005; Fjell and Walhovd, 2010; Lemaitre et al., 2012; Raz et al., 2005). Sustained interpersonal interactions could potentially prevent age-related neuronal or synaptic loss or otherwise promote synaptic or dendritic plasticity (Anatürk et al., 2018). Animal studies offer support for this interpretation (e.g., Briones et al., 2004; Okuda et al., 2009; Zhao et al., 2011; Jung and Herms, 2014), with one study reporting an increase in the synaptic density of the rat visual cortex after housing in an enriched environment for a month (Briones et al., 2004). At the same time, the selectivity of the occipital cortex needs to be interpreted with caution, given the “blunt” nature of our activity measure. Overall, our results suggest that encouraging middle-aged and older adults to stay engaged with their existing social network could potentially be an important avenue for supporting individuals through their later years of life.

Individuals who consistently reported weekly computer use had higher global cognitive performance, relative to those who consistently reported less frequent participation over time. Despite the small number of individuals in the low and stable group (i.e., reference group; n = 197) this association was one of the few to survive FDR corrections. These results are in line with prior findings indicating a link between frequent computer (Tun and Lachman, 2010) and internet use (Berner et al., 2019) and better cognitive function. Our findings further complement the observation that computerized cognitive training programs lead to improvements in trained cognitive domains, with some findings suggesting transfer to untrained domains (e.g., Kueider et al., 2012; Nguyen et al., 2019). While speculative, the pathway linking computer use to cognition could be through an increased exposure to novelty (e.g., reading articles online), a greater demand on psychomotor skill (through mouse use and typing) and/or the opportunity to engage several domains of cognition at once (e.g., attention and memory when playing computer games; Tun and Lachman, 2010). Nevertheless, we cannot rule out the explanation that individuals who frequently use the computer are more accustomed to using technology and may have had an advantage to non-users on the computerized cognitive assessments administered in the UK Biobank. Reverse causation could alternatively explain our results, such that individuals with higher levels of cognitive function are inclined to spend more of their time using computers relative to individuals with lower levels of global cognition. While no structural or functional correlates were found for this activity, a link with GM volume in the putamen was detected at trend-level (β = 0.22, 95%CI = [0.12, 0.33], FDR q = 0.07), Supplementary File: Results). The putamen is implicated in both movement preparation and execution and non-motor functions, (e.g., executive control, working and episodic memory and category fluency, Ell et al., 2011) and although the trend reported here needs to be independently validated by future studies, it could represent a potential region of interest for studies investigating potential mediators of the computer-cognition association. Taken together, our results suggest that improving computer use level among middle-aged adults represents an important aim of future RCTs, which will also concurrently establish the directionality of the associations reported here.

Otherwise, contrary to our hypotheses, none of the other activities examined (i.e., going to the pub or social clubs, attending an adult educational class, undertaking religious activities) were associated with markers of GM, WM microstructure, functional connectivity or cognitive function, after FDR corrections. While prior meta-analytic investigations have identified associations with brain structure (Anatürk et al., 2018) and cognition (Kuiper et al., 2016) when composite measures of activities were used, our results suggest that comparatively speaking, they do not uniquely contribute to brain health in early late life. The alternative explanation is that the crude measure of activity levels used in our study potentially attenuated associations between these activities and neural/cognitive outcomes. Therefore, an important next step for this line of work is to examine whether our results are replicated when assessing other dimensions of engagement (i.e., duration and frequency). Considering that the present findings imply dissociable effects between activities in brain-cognition associations, our results are in favor of an approach sensitive to these inter-activity differences when the statical power of a study allows for it. However, it also needs to be considered that cumulative engagement over separate activities is likely necessary for sustained improvement in brain function, especially as only weak associations were found for the individual activities investigated in our study. Accordingly, we repeated our results with the inclusion of cumulative leisure activities (i.e., number of activities engaged in on a weekly basis across the two timepoints) and found that although cumulative activities were not significantly linked to any MRI or cognitive outcomes, a trend (i.e., β = 0.03, 95%CI = [0.02, 0.05], FDR q = 0.066, Supplementary File: Results [Posthoc MRI results]) was detected for higher cumulative activity and stronger connectivity between the medial visual and posterior default mode network, which was not observed for any specific activity. Hence, activity-specific associations could serve as complementary to that of composite leisure activity scores. Overall, our results are informative to clinicians and researchers planning intervention studies as we highlight several activities that may play a role in maintaining brain health in older adult populations.

We found that sustained sports club/gym attendance was linked to greater absolute connectivity of the sensorimotor connectivity although no parallel associations with cognition were found. Conversely, consistent family and friend visits over time were associated with higher volumetric measures of GM in the occipital cortex. Only weekly leisure time computer use over time was linked higher levels of global cognition. Overall, this study demonstrates selective associations between different leisure activities, highlighting several that may be relevant for RCTs aiming to promoting cognitive health in late life.

The data analyzed in this study is subject to the following licenses/restrictions: Data access to the UK Biobank will need to requested through a standard data access procedure. Requests to access these datasets should be directed to http://www.ukbiobank.ac.uk/register-apply.

The studies involving human participants were reviewed and approved by NHS National Research Ethics Service (Ref 11/NW/0382). The patients/participants provided their written informed consent to participate in this study.

SMS provided the overall scientific strategy for UK Biobank brain imaging. MA planned and conducted the analyses and prepared the manuscript, including all tables and the figures. SS, SMS, CES, and KPE provided feedback and comments on all versions of the manuscript. All authors contributed to the article and approved the submitted version.

This research was conducted using the UK Biobank Resource under the approved application of 45301. MA was supported by the Clarendon Trust DPhil Fellowship and HDH Wills 1965 Charitable Trust (1117747). SS reports funding from the Academy of Medical Sciences/the Wellcome Trust/the Government Department of Business, Energy and Industrial Strategy/the British Heart Foundation/Diabetes UK Springboard Award (SBF006\1078). SMS receives support from the Wellcome Trust (098369/Z/12/Z, 203139/Z/16/Z). KPE reports support from the UK Medical Research Council (G1001354, MR/K013351/), the HDH Wills 1965 Charitable Trust (1117747), Alzheimer’s Research UK (PPG2012A-5), and the European Commission (Horizon 2020 grant “Lifebrain,” 732592). SS and CES were supported by the NIHR Oxford Biomedical Research Center located at the Oxford University Hospitals NHS Trust and the University of Oxford, the NIHR Oxford Health BRC. The Wellcome Center for Integrative Neuroimaging was supported by core funding from the Wellcome Trust (203139/Z/16/Z). CES is now a full-time employee of the Alzheimer’s Association.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.