Introduction
Existing research has focused on characterizing age-related differences in spatial navigation, with evidence of particular deficits in processes dependent upon the hippocampus (Lester et al., Reference Lester, Moffat, Wiener, Barnes and Wolbers2017). However, it is important to also consider factors that may contribute to the deficits, particularly ones that may be modifiable. While several factors may contribute to hippocampal aging and related impairments in spatial navigation, the role of age-related disruption in sleep quality remains unclear. Thus, the current study focused on understanding the role of sleep disturbance in age effects on spatial navigation.
The ability to navigate an environment to a goal location involves two separable strategies and neural systems (Burgess, Reference Burgess2008; Maguire, Reference Maguire1998; Wolbers et al., Reference Wolbers, Weiller and Büchel2004). Allocentric navigation strategies involve a flexible, world-centered (object-to-object) representation (i.e., a “cognitive map”) and are hippocampal dependent. Egocentric navigation strategies involve environmental information represented from the perspective of the navigator (i.e., route based) and are dependent on the caudate nucleus and medial parietal cortex. In dual-solution paradigms, there is individual variability in which strategy is predominately used, paralleling structural and functional variance in the hippocampus and caudate (e.g., Bohbot et al., Reference Bohbot, Lerch, Thorndycraft, Iaria and Zijdenbos2007; Iaria et al., Reference Iaria, Petrides, Dagher, Pike and Bohbot2003; Marchette et al., Reference Marchette, Bakker and Shelton2011). While both strategies may be used during navigation, there is also evidence of competitive interactions between the systems (Packard & McGaugh, Reference Packard and McGaugh1996; Poldrack & Packard, Reference Poldrack and Packard2003).
While advancing age is associated with deterioration of the neural substrates of both navigation strategies (Raz & Rodrigue, Reference Raz and Rodrigue2006), the magnitude of hippocampal atrophy exceeds that of caudate atrophy (Raz et al., Reference Raz, Ghisletta, Rodrigue, Kennedy and Lindenberger2010; Walhovd et al., Reference Walhovd, Westlye, Amlien, Espeseth, Reinvang, Raz, Agartz, Salat, Greve, Fischl, Dale and Fjell2011). Consistent with this, there is greater age-related decrement on tasks assessing the formation, retention, and use of a cognitive map relative to learning a route (Head & Isom, Reference Head and Isom2010; Iaria et al., Reference Iaria, Palermo, Committeri and Barton2009; Lester et al., Reference Lester, Moffat, Wiener, Barnes and Wolbers2017). Additionally, older adults exhibit a shift from allocentric to egocentric strategies in dual-solution paradigms consistent with competitive interactions between the systems (Rodgers et al., Reference Rodgers, Sindone and Moffat2012; Wiener et al., Reference Wiener, de Condappa, Harris and Wolbers2013). Lastly, associations between hippocampal volume and allocentric tasks have been observed in older adults (Daugherty et al., Reference Daugherty, Yuan, Dahle, Bender, Yang and Raz2015; Head & Isom, Reference Head and Isom2010).
Several lines of evidence support poor sleep quality as a potential contributor to these age effects on allocentric navigation. Total sleep deprivation is associated with reduced hippocampal plasticity, a shift from allocentric to egocentric strategy use and a differential deficit in allocentric compared to egocentric navigation task performance in rodents (Hagewoud et al., Reference Hagewoud, Havekes, Tiba, Novati, Hogenelst, Weinreder, Van der Zee and Meerlo2010; Murata et al., Reference Murata, Oka, Iseki, Mori, Ohe, Mine and Enjoji2018; Soto-Rodriguez et al., Reference Soto-Rodriguez, Lopez-Armas, Luquin, Ramos-Zuñiga, Jauregui-Huerta, Gonzalez-Perez and Gonzalez-Castañeda2016). Sleep fragmentation, a measure of brief arousals during a bout of sleep, is more ecologically valid and common than sleep deprivation. Experimentally induced sleep fragmentation negatively impacts hippocampal plasticity and hippocampal-dependent learning in rats (Guzman-Marin et al., Reference Guzman-Marin, Bashir, Suntsova, Szymusiak and McGinty2007; Lee et al., Reference Lee, Katsuyama, Duge, Sriram, Krushelnytskyy, Kim and de la Iglesia2016; Sportiche et al., Reference Sportiche, Suntsova, Methippara, Bashir, Mitrani, Szymusiak and McGinty2010; Tartar et al., Reference Tartar, McKenna, Ward, McCarley, Strecker and Brown2010). Furthermore, sleep fragmentation is associated with disruptions in electrophysiological signals that support hippocampal-dependent memory processes (Astori et al., Reference Astori, Wimmer and Lüthi2013; Fogel & Smith, Reference Fogel and Smith2011; Ondze et al., Reference Ondze, Espa, Dauvilliers, Billiard and Besset2003). The incidence of sleep fragmentation rises with advancing age (Ohayon et al., Reference Ohayon, Carskadon, Guilleminault and Vitiello2004) and is modestly associated with impaired cognitive performance in older adults (André et al., Reference André, Tomadesso, de Flores, Branger, Rehel, Mézenge, Landeau, de la Sayette, Eustache, Chetelat and Rauchs2019; Blackwell et al., Reference Blackwell, Yaffe, Ancoli-Israel, Redline, Ensrud, Stefanick, Laffan and Stone2011; Kaneshwaran et al., Reference Kaneshwaran, Olah, Tasaki, Yu, Bradshaw, Schneider, Buchman, Bennett, de Jager and Lim2019; Luik et al., Reference Luik, Zuurbier, Hofman, Van Someren, Ikram and Tiemeier2015). However, much of this work was conducted with clinical populations or limited to general psychometric measures. Thus, targeted research exploring the impact of sleep fragmentation on spatial navigation processes associated with the hippocampus in healthy older adults is necessary.
Circadian rest-activity rhythm fragmentation, in the form of daytime napping and/or nighttime arousals, may also contribute to age effects on allocentric navigation. The intradaily variability, an actigraphy-derived measure of rest-activity rhythm fragmentation, increases with older age (Huang et al., Reference Huang, Liu, Wang, Van Someren, Xu and Zhou2002; Luik et al., Reference Luik, Zuurbier, Hofman, Van Someren and Tiemeier2013; Musiek et al., Reference Musiek, Bhimasani, Zangrilli, Morris, Holtzman and Ju2018; Oosterman et al., Reference Oosterman, Van Someren, Vogels, Van Harten and Scherder2009). Furthermore, higher intradaily variability has been associated with medial temporal lobe atrophy and dysfunction across the lifespan (Van Someren et al., Reference Van Someren, Oosterman, Van Harten, Vogels, Gouw, Weinstein, Poggesi, Scheltens and Scherder2019). The impact of circadian disruption on navigation strategies remains unclear.
Relevant to our central goal of understanding the effects of sleep fragmentation on allocentric navigation is consideration of other lifestyle factors. Exercise and stress have both been shown to regulate hippocampal plasticity, with exercise conferring benefits and stress imparting a negative impact in rodents and humans. (Erickson et al., Reference Erickson, Voss, Prakash, Basak, Szabo, Chaddock, Kim, Heo, Alves, White, Wojcicki, Mailey, Vieira, Martin, Pence, Woods, McAuley and Kramer2011; Gianaros et al., Reference Gianaros, Jennings, Sheu, Greer, Kuller and Matthews2007; Kleemeyer et al., Reference Kleemeyer, Kühn, Prindle, Bodammer, Brechtel, Garthe, Kempermann, Schaefer and Lindenberger2016; Lucassen et al., Reference Lucassen, Meerlo, Naylor, van Dam, Dayer, Fuchs, Oomen and Czéh2010; Machida et al., Reference Machida, Takamiya, Amagasa, Murayama, Fujiwara, Odagiri, Kikuchi, Fukushima, Kouno, Saito, Yoshimine, Inoue and Shobugawa2021; Niemann et al., Reference Niemann, Godde and Voelcker-Rehage2014). As an initial step in building models of how these factors may play a role in age effects on hippocampal-dependent navigation, we considered interactive effects of stress and exercise with sleep fragmentation. That is, neurophysiological changes with aging could make older adults’ hippocampal system simultaneously more vulnerable to cumulative damage from sleep fragmentation and stress as well as more likely to benefit from protective factors like physical activity and exercise (Graham et al., Reference Graham, Christian and Kiecolt-Glaser2006; Miller & O’Callaghan, Reference Miller and O’Callaghan2005).
The current study collected one-week measures of sleep quality, perceived stress, and physical activity as assays of trait-like characteristics to address several novel aims. First, we examined associations between sleep disturbance and spatial navigation. We hypothesized that sleep fragmentation would have a particularly negative impact on allocentric navigation performance and be associated with a shift from allocentric strategy use to egocentric strategy use. A second goal was to determine the role of sleep disturbance in age-related differences in spatial navigation. We hypothesized that sleep fragmentation would mediate age-related effects on allocentric navigation. Our third aim was to determine whether perceived stress or physical activity moderates the effects of sleep disturbance. We hypothesized that individuals with greater perceived stress would evidence greater negative effects of sleep fragmentation, and that this moderating effect would be greater for older adults. Additionally, we hypothesized that greater physical activity would mitigate the negative effects of sleep fragmentation, and this moderating effect would be greater for older adults. Total sleep time was also examined as a secondary measure of sleep quality, with the hypotheses that shorter sleep duration would serve a similar role as greater sleep fragmentation.
The fourth aim was to determine whether circadian rest-activity rhythm fragmentation was associated with spatial navigation in post-hoc analyses. We hypothesized that intradaily variability would be negatively associated with allocentric navigation performance and strategy use and would mediate age effects on these tasks.
Methods
Participants
Younger adult participants (aged 18–25) were recruited via an undergraduate participant pool in the Department of Psychological & Brain Sciences at Washington University or Volunteers for Health (VFH), a research volunteer registry at Washington University. Older adult participants (aged 60–89) were recruited via VFH. Participants were compensated monetarily or with course credit. All participants were right-handed with normal or corrected vision. Participants with a current or past psychological disorder or neurological disorder (e.g., stroke, dementia, traumatic brain injury), current sleep disorder, reduced mobility, or use of sleep-promoting medications were excluded. In total, 80 participants were enrolled (42 younger adults (27 females); 37 older adults (19 females); see Supplemental Table 1). There was a non-significant trend for higher education in older adults (p = .08). One older individual was excluded due to sleep medication use. Participants provided written consent in accordance with Washington University Human Research Protection Office guidelines.
Experimental procedures
General
Participants had two lab visits approximately one week apart. During the first visit, participants were given an actigraphy watch and instructions on completing a sleep diary. During the second visit, participants completed experimental tasks. A psychomotor vigilance task was first. Spatial navigation task order was counterbalanced. Participants completed stress, physical activity, computer expertise, health, and sense of direction questionnaires via REDCap before or after the second visit. A summary of study measures is in Table 1 and descriptive statistics are in Supplemental Table 2.
Table 1. Summary of key tasks and variables
Measures
Sleep fragmentation, total sleep time, and intradaily variability
Sleep measures were obtained using Actigraph GT9X watch (ActiGraph; Pensacola, FL) with a tri-axial accelerometer. With the exception of bathing or swimming, participants continuously wore the watch on the non-dominant wrist for three–eight consecutive nights (82% with 5+ days; mean = 6.2; SD = 1.4). Movement sampling frequency was set to 30 Hz and binned in 60-s epochs without the low frequency extension filter (Feito et al., Reference Feito, Hornbuckle, Reid and Crouter2017). A wear-time validation algorithm was applied before further processing (Choi et al., Reference Choi, Liu, Matthews and Buchowski2011). Sleep periods were determined in ActiLife (Pensocola, FL) using the Cole–Kripke algorithm (Cole et al., Reference Cole, Kripke, Gruen, Mullaney and Gillin1992). If entire nights of sleep were not detected by the Cole–Kripke algorithm but were indicated in participant’s sleep diary, self-reported sleep and wake times were manually entered (n = 23, no age group difference, p < .92). The sleep fragmentation index was computed as the sum of the movement index (total scored awake minutes/total time in bed in hours × 100) and the fragmentation index (total one-minute scored sleep bouts/total number of sleep bouts of any length × 100) (ActiLife 6 User’s Manual, 2012). Total sleep time was the average minutes of sleep per 24 hr.
Intradaily variability, a measure of rest-activity fragmentation, was calculated as a ratio of the variability between successive hours of activity and overall variability of activity in 1 day (Witting et al., Reference Witting, Kwa, Eikelenboom, Mirmiran and Swaab1990). Greater intradaily variability is consistent with greater fragmentation of rest-activity cycles. Intradaily variability was derived from full days of raw activity counts binned in 60-s epochs using the R package “nparACT” (Blume et al., Reference Blume, Santhi and Schabus2016).
Sleep diary
Participants completed the Consensus Sleep Diary (Maich et al., Reference Maich, Lachowski and Carney2018) as soon as possible each morning via REDCap or a paper form.
Psychomotor vigilance
This task was administered using Psychology Experiment Building Language (Mueller & Piper, Reference Mueller and Piper2014). Participants attended to a crosshair centered on the screen and quickly pressed the spacebar when a red dot appeared at randomly varying time intervals. Response time appeared after each trial. The index of performance was a standardized composite score of average reciprocal of response times, average response times for the slowest 10% of trials, and number of trials where response time exceeded 500 milliseconds (Basner & Dinges, Reference Basner and Dinges2011).
Spatial navigation tasks
Spatial navigation tasks were administered in Unity (San Francisco, CA) on 24-inch monitors (see Supplementary Figure 1). Participants first practiced using a joystick in a virtual environment of a long hallway with alternating turns. Next, visuomotor expertise was tested by requiring participants to traverse the environment within 60 seconds. All participants passed this test.
Cognitive mapping
For this task, the indoor environment consisted of interconnected hallways with twelve landmarks (Allison et al., Reference Allison, Fagan, Morris and Head2016). During study, participants freely explored a virtual environment for four minutes and were instructed to discover as many objects as they could and to pay attention to the objects and their locations. For the test phase, participants indicated the locations of the landmarks by making an X on a two-dimensional map. These study-test phases were repeated for four trials. Cognitive mapping performance was quantified as the average number of correctly placed landmarks across trials.
Route learning
During study, participants followed a path marked by arrows in an environment consisting of interconnected hallways and twenty-six landmarks (Allison et al., Reference Allison, Fagan, Morris and Head2016). For the test phase, participants were instructed to accurately traverse the route as quickly as possible without the arrows. These study-test phases were repeated for four trials. Route learning performance was quantified as average time to traverse the route across trials in seconds.
Spatial navigation strategy use
Participants were administered a dual-solution task (Marchette et al., Reference Marchette, Bakker and Shelton2011). First, participants studied pictures of each landmark with its associated name two times for familiarization. Next, participants encoded a route through an outdoor environment that included the twelve landmarks, interconnecting pathways, and four unique surrounding mountains by viewing a video across nine trials. To ensure sustained attention, five matching trials were randomly interspersed with experimental route trials. Matching trials were in a different environment from the experimental route environment with no landmarks, but instead with colored spheres along a route. Participants answered yes/no questions about number, color, or sequence in which spheres were presented.
For the test phase, participants were placed at one landmark in the maze and asked to locate a target landmark indicated at the top of the screen by name. Twenty-four combinations of starting landmarks and target landmarks were administered in random order. For 16 of these trials, participants could either find the target by following the path originally shown or by taking a shortcut route.
We used a protocol to classify trials similar to past work (Furman et al., Reference Furman, Clements-Stephens, Marchette and Shelton2014; Marchette et al., Reference Marchette, Bakker and Shelton2011). Shortcut routes were defined as paths that were shorter than the original path and with more than half of the path deviating from the original path. Original routes were defined as having more than half of the participants’ trajectory on the original path. Inefficient routes were defined as being longer than the original path. Unsuccessful routes were defined as not reaching the target location. The navigation strategy index was defined as the proportion of trials in which a shortcut was taken (0 = no shortcut use; 1 = shortcut used on all trials). Unsuccessful and inefficient routes were not included when calculating the strategy index.
Shortcut use is considered an allocentric strategy because it requires knowledge of object locations and their relative distances irrespective of viewpoint. Conversely, following an original route is considered an egocentric strategy, where the navigator uses knowledge of local landmarks and associated body turns from a first-person perspective. Therefore, higher values on the navigation strategy index, indicating greater use of shortcuts, is interpreted as reflecting greater allocentric strategy use. Greater shortcut use on this task correlates with relative activation of the hippocampus versus caudate (Furman et al., Reference Furman, Clements-Stephens, Marchette and Shelton2014; Marchette et al., Reference Marchette, Bakker and Shelton2011).
Perceived stress
Psychological stress over the prior month was measured with the ten-item version of the Perceived Stress Scale (Cohen et al., Reference Cohen, Kamarck, Mermelstein, Cohen, Kessler and Underwood Gordon1994). This scale has demonstrated strong internal consistency, strong test–retest reliability, and moderate criterion validity in both younger and older adults (Lee, Reference Lee2012).
Physical activity
Physical activity was measured via actigraphy. Acquisition parameters and wear-time validation were the same as for sleep measures (Choi et al., Reference Choi, Liu, Matthews and Buchowski2011). ActiLife software was set to indicate wrist-worn watch, which scales the scoring algorithms accordingly via an internal algorithm (ActiLife 6 User’s Manual, 2012). The Freedson algorithm (Freedson et al., Reference Freedson, Melanson and Sirard1998) was used for cut-points to classify intensity levels into sedentary (<100 counts per epoch), light (100–1952), moderate (1952–5723), and vigorous activity (>5724) categories. Physical activity was quantified as average minutes in moderate to vigorous physical activity per valid wear day. Minutes of physical activity only included bouted activity, defined as at least 10 minutes of continuous physical activity allowing for two minutes of interruption (Physical Activity Guidelines Advisory Committee, 2008).
Computer experience
Participants rated computer experience, experience playing computer games, and experience with computer games involving virtual reality on a 0–7 Likert scale (higher number = more experience).
Subjective sleepiness
The Stanford Sleepiness Scale (Hoddes et al., Reference Hoddes, Zarcone, Smythe, Phillips and Dement1973) was completed after the experimental tasks. Participants rated their current state on a 7-point scale (higher number = greater sleepiness).
Analytic approach
Covariates
Gender, computer experience, visuomotor expertise, subjective sleepiness, psychomotor vigilance, and number of nights of actigraphy measurement were considered as potential covariates. Variables significantly associated with age were included as covariates. Computer experience, subjective sleepiness, and number of measurement nights were significantly different between age group (ps < .05; see Table 1).
Outliers
Outliers were defined as data points > 2.5 standard deviations from the mean and with Cook’s D value greater than 4/n (Cook, Reference Cook1977). Results are presented with outliers removed (see Supplemental Tables 2–5 for results with outliers included).
Aims 1 and 2 analyses: associations amongst age, sleep characteristics, and spatial navigation
For these analyses, we used age group as the predictor, navigation measures as the outcomes, and sleep fragmentation (primary) and total sleep time (secondary) as mediators. All analyses were conducted in Stata 16.0 (StataCorp) using the “sem” command and full information maximum likelihood estimation to address missing data (see Table 1). The first step examined whether the necessary conditions for mediation were met (i.e., the predictor and outcome variable were significantly related to the mediator). Based on current recommendations, a significant association between predictor and outcome was not required to assess for mediation (MacKinnon et al., Reference MacKinnon, Lockwood, Hoffman, West and Sheets2002; Rucker et al., Reference Rucker, Preacher, Tormala and Petty2011). Mediation analyses implemented a regression-based approach to estimate effects in conjunction with bootstrapping. The total effect represents the association of age group with spatial navigation, including both direct and indirect effects. The direct effects indicate the degree of association between: (a) age group and the mediator; (b) the mediator and spatial navigation controlling for age group; and (c) age group and spatial navigation controlling for the mediator. Indirect effects indicate the degree to which age group influences spatial navigation via the mediator. Indirect effects were examined using 5,000 bootstrapping samples and 95% percentile confidence intervals (CIs).
Aim 3 analyses: stress and physical activity as moderators of sleep associations with spatial navigation
Regression analyses were conducted to examine moderating effects. Analyses were conducted in Stata 16.0 (StataCorp) using the “sem” command and full information maximum likelihood estimation to address missing data (see Table 1). Continuous variables were standardized using a Z-transformation prior to creating interaction terms. In all analyses, covariates, age group, the sleep variable, perceived stress (or physical activity), and the primary 2-way interaction of interest (sleep variable × perceived stress (or physical activity) were entered in the first step. Remaining 2-way interactions with age group (age group × sleep variable; age group × perceived stress (or physical activity) as well as the 3-way interaction of primary interest were entered in the second step. Standardized coefficients and standard errors (SE) are presented.
Aim 4 post-hoc analyses: associations amongst age, intradaily variability, and spatial navigation
Analyses were conducted as described under Aim 1, except intradaily variability was the mediator.
Results
Aims 1 and 2: associations amongst age, sleep characteristics, and spatial navigation
Age-related differences
The age groups did not differ in sleep fragmentation, but older age was associated with reduced total sleep time (sleep fragmentation index: β = .183 [SE = .129], p = .153; total sleep time: β = −.386 [SE = .104], p < .001). There were age group differences in cognitive mapping, route learning, and navigation strategy use. Older adults were less able to form a cognitive map, recreate a learned route, and use shortcuts (β = −.732 [SE = .072], p < .001; β = .620 [SE = .091], p < .001; β = −.373 [SE = .117], p = .001, respectively).
Sleep fragmentation
Sleep fragmentation was not associated with cognitive mapping, route learning, or navigation strategy use (β = −.041 [SE = .068], p = .547; β = .094 [SE = .108], p = .35; β = .131 [SE = .103], p = .203, respectively).
However, we did observe that the association of sleep fragmentation with cognitive mapping differed by age group (see Tables 2 and 3) in the moderation models. The interaction remained significant when examined independently from stress or physical activity (stress model: β = −.265 [SE = .113], p = .019; physical activity model: β = −.298 [SE = .111], p = .008; independent model: β = −.250 [SE = .103], p = .015). Specifically, greater sleep fragmentation was associated with worse cognitive mapping performance among older adults but not younger adults (older: β = −.344 [SE = .156], p = .022; younger: β = .203 [SE = .168], p = .227; see Figure 1). However, sleep fragmentation did not mediate age effects on cognitive mapping within the older adults in a post-hoc analysis (β = −.155; b = −.032; 95% CI: −.088–.009).
Table 2. Stress as a moderator of sleep characteristics
SFI = sleep fragmentation index; TST = total sleep time.
Note. Data represent standardized path coefficients with standard errors in parentheses. Higher scores on the strategy index indicate greater relative shortcut use. Higher scores on cognitive mapping indicate better performance. Higher scores on route learning indicate worse performance.
* p < 0.05.
** p < 0.01.
*** p < 0.001.
Table 3. Physical activity as a moderator of sleep characteristics
MVPA = moderate-to-vigorous physical activity; SFI = sleep fragmentation index; TST = total sleep time.
Note. Data represent standardized path coefficients with standard errors in parentheses. Higher scores on the strategy index indicate greater relative shortcut use. Higher scores on cognitive mapping indicate better performance. Higher scores on route learning indicate worse performance.
* p < 0.05.
** p < 0.01.
*** p < 0.001.
Figure 1. Sleep fragmentation association with cognitive mapping. Scatter plot of the significant age group × sleep fragmentation interaction with separate regression lines for younger adults (gray squares; β = .203 [SE = .168], p = .227) and older adults (black circles; β = –.354 [SE = .154], p = .022). See text for details.
We also examined whether sleep fragmentation was more strongly associated with cognitive mapping relative to route learning within the older adults. While the association between sleep fragmentation and route learning was not significant (β = .123 [SE = .216], p = .570), the magnitude of the association was not significantly lower than the association with cognitive mapping (χ2 = 2.27, p = .132).
While there was also evidence of an age × sleep interaction for the strategy index in the physical activity model, the interaction was not significant when physical activity was removed from the model (β = .624 [SE = .196], p = .001; β = .271 [SE = .152], p = .075, respectively).
Total sleep time
Total sleep time was not associated with cognitive mapping, route learning, or the spatial navigation strategy index (β = .084 [SE =.075], p = .258; β = .119 [SE = .104], p = .252; β = −.199 [SE = .117], p = .087, respectively).
Aim 3: stress and physical activity as moderators of sleep associations with spatial navigation
Sleep fragmentation
There were no significant interactions involving either stress or physical activity with sleep fragmentation predicting navigation outcomes (ps > .113; see Tables 2 and 3).
Total sleep time
There were no significant interactions involving stress and total sleep time for any of the navigation outcomes (ps > .250; see Table 2). Physical activity did not moderate the association of total sleep time with cognitive mapping (ps > .894; see Table 3).
There was a significant total sleep time × physical activity interaction for route learning (β = .270 [SE = .098], p = .006). However, the interaction reflected that longer sleep duration was associated with worse performance at higher physical activity levels but was not associated with performance at lower physical activity levels (high PA: β = .238 [SE = .099], p = .016; low PA: β = .062 [SE = .201], p = .758).
We also observed an age group × physical activity × total sleep time interaction for navigation strategy that was inconsistent with hypotheses (β = .506 [SE = .219], p = .021; see Table 3). For older adults, less sleep was particularly associated with greater shortcut use (i.e., greater allocentric strategy use) at higher physical activity compared to lower physical activity levels (interaction: β = .484 [SE = .190], p = .011; high PA: β = −.239 [SE = .109], p = .028; low PA: β = −.924 [SE = .096], p < .001). The physical activity × total sleep time interaction was not significant for younger adults (β = −.200 [SE = .140], p = .155).
Post-hoc analyses were conducted to better understand these interactions. Considering evidence of an inverted U-shape relationship between sleep duration and cognitive outcomes (Mohlenhoff et al., Reference Mohlenhoff, Insel, Mackin, Neylan, Flenniken, Nosheny, Richards, Maruff and Weiner2018; van Oostrom et al., Reference van Oostrom, Nooyens, van Boxtel and Verschuren2018), we assessed for quadratic (i.e., nonlinear) associations for our navigation outcomes in the full sample by including the quadratic (squared) term for total sleep time in the regression models. There were significant quadratic associations of total sleep time with navigation strategy use and route learning (Supplemental Figure 2) (β = .275 [SE = .099], p = .005); β = .223 [SE = .101], p = .027, respectively). Thus, it is conceivable that observed interactions with physical activity are spurious (Belzak & Bauer, Reference Belzak and Bauer2019; Cohen et al., Reference Cohen, Dahmen and DeVore2003).
Aim 4: Associations amongst age, intradaily variability, and spatial navigation in post-hoc analyses
Older adults evidenced lower intradaily variability compared to younger adults indicating reduced rest-activity fragmentation, which is opposite from the expected age-related increase (β = −.287 [SE = .112], p = .015). Greater intradaily variability was associated with lower cognitive mapping but was not related to route learning or strategy use (β = −.235 [SE = .074], p = .001; β = .134 [SE = .089], p = .129; β = −.095 [SE = .112], p = .413, respectively).
Within the older adult sample, there was the expected increase in intradaily variability with advancing age (β = −.449 [SE = .172], p = .009; β = −.389 [SE = .174], p = .026, respectively). In addition, greater intradaily variability was associated with lower cognitive mapping (see Figure 2), but not with route learning (β = −.437 [SE = .175], p = .005; β = .252 [SE = .153], p = .100, respectively) among older adults. However, intradaily variability was not differentially associated with cognitive mapping relative to route learning (χ 2 = .91, p = .341). Lastly, intradaily variability did not mediate age effects on cognitive mapping within the older adults (β = −.170; b = −.035; 95% CI: −.099–.006).
Figure 2. Circadian rest-activity fragmentation association with cognitive mapping. Scatter plot of the significant association of circadian rest-activity fragmentation with cognitive mapping. Younger adult data are depicted with gray squares and older adult data depicted with black circles. The separate regression lines represent the association in the full sample (gray line; β = –.235 [SE = .074], p = .001) and in the older adults (black line; (β = –.437 [SE = .175], p = .005). See text for details.
Discussion
The current study replicated age differences in spatial navigation tasks (Lester et al., Reference Lester, Moffat, Wiener, Barnes and Wolbers2017; Wiener et al., Reference Wiener, Kmecova and de Condappa2012) and extended this literature in several ways. The novel goals of this study were to examine: (a) associations between sleep disturbance and spatial navigation; (b) whether sleep disturbance mediated age differences in spatial navigation; (c) whether perceived stress and physical activity moderated sleep associations with spatial navigation; and (d) associations amongst rest-activity fragmentation, spatial navigation, and age in post-hoc analyses. We discuss the results for the each these aims below.
Associations amongst age, sleep characteristics, rest-activity fragmentation, and spatial navigation
Novel convergent findings were observed regarding the fragmentation measures and cognitive mapping. There was a significant age-related increase in sleep fragmentation within the older adult sample, which is consistent with previous findings (Boulos et al., Reference Boulos, Jairam, Kendzerska, Im, Mekhael and Murray2019; Ohayon et al., Reference Ohayon, Carskadon, Guilleminault and Vitiello2004). Sleep fragmentation has been linked to neurophysiological changes in the hippocampus including reduced neurogenesis (Guzman-Marin et al., Reference Guzman-Marin, Bashir, Suntsova, Szymusiak and McGinty2007), metabolic dysregulation (Nair et al., Reference Nair, Zhang, Ramesh, Hakim, Kaushal, Wang and Gozal2011; Williams et al., Reference Williams, Perland, Eriksson, Carlsson, Erlandsson, Laan, Potter, Frediksson, Benedict and Schiöth2016), and reduced long-term potentiation (Tartar et al., Reference Tartar, Ward, McKenna, Thakkar, Arrigoni, McCarley, Brown and Strecker2006). Consistent with these adverse physiologic effects, greater sleep fragmentation was associated with reduced cognitive map formation in older adults. Circadian rest-activity fragmentation also exhibited an age-related increase within the older adults in post-hoc analyses, similar to past observations (Luik et al., Reference Luik, Zuurbier, Hofman, Van Someren and Tiemeier2013; Musiek et al., Reference Musiek, Bhimasani, Zangrilli, Morris, Holtzman and Ju2018; Oosterman et al., Reference Oosterman, Van Someren, Vogels, Van Harten and Scherder2009; but see Park et al., Reference Park, Lee, Byun, Yi, Lee, Jeon, Hwang, Yoon, Choe, Kim, Shin, Suk and Lee2021). Importantly, greater circadian rest-activity fragmentation was also related to lower cognitive mapping performance. This is consistent with evidence linking circadian rest-activity fragmentation with hippocampal activity and atrophy (Sherman et al., Reference Sherman, Mumford and Schnyer2015; Van Someren et al., Reference Van Someren, Oosterman, Van Harten, Vogels, Gouw, Weinstein, Poggesi, Scheltens and Scherder2019). Neither the sleep nor circadian fragmentation measure significantly mediated age effects on cognitive mapping within the older adults in post-hoc analyses. This may be attributable to the modest effect sizes combined with a smaller sample size.
We also hypothesized specific negative effects of fragmented sleep and circadian rhythms on hippocampal-dependent navigation. However, we did not observe significant differential associations of either sleep fragmentation or circadian rest-activity fragmentation with cognitive mapping relative to route learning. In addition, there were no indications of fragmentation-related shifts from allocentric to egocentric strategies. Thus, it is relevant to consider that sleep fragmentation and circadian rest-activity fragmentation might instead lead to broad rather than selective deficits. Both types of fragmentation have been associated with reduced processing speed, memory, visuospatial abilities, executive functions, and general cognitive status to varying degrees (André et al., Reference André, Tomadesso, de Flores, Branger, Rehel, Mézenge, Landeau, de la Sayette, Eustache, Chetelat and Rauchs2019; Lim et al., Reference Lim, Yu, Costa, Leurgans, Buchman, Bennett and Saper2012; Luik et al., Reference Luik, Zuurbier, Hofman, Van Someren, Ikram and Tiemeier2015; McSorley et al., Reference McSorley, Bin and Lauderdale2019; Oosterman et al., Reference Oosterman, Van Someren, Vogels, Van Harten and Scherder2009; Swanson et al., Reference Swanson, Hood, Hall, Kravitz, Matthews, Joffe, Thurston, Butters, Ruppert and Harlow2021). This literature has generally used neuropsychological tests and no consistent pattern has emerged in terms of domain specificity. Thus, larger-scale studies using more targeted experimental tasks, such as those employed here, may be beneficial for determining the degree to which certain cognitive domains are differentially sensitive to sleep or circadian rest-activity fragmentation.
In terms of total sleep time, duration was not associated with any of the spatial navigation measures, except within the context of interactions with physical activity. However, linear associations were in the opposite directions from those hypothesized for route learning and navigation strategy. One potential contributing factor is the presence of nonlinear associations between total sleep time and both tasks in post-hoc analysis, which can lead to spurious interactions (Belzak & Bauer, Reference Belzak and Bauer2019; Cohen et al., Reference Cohen, Dahmen and DeVore2003). Such nonlinear associations are observed in the literature particularly for self-report measures of sleep (Devore et al., Reference Devore, Grodstein and Schernhammer2016). However, findings are also mixed in terms of whether such quadratic associations are observed in younger versus older adults, and in terms of differences across cognitive domains (e.g., Devore et al., Reference Devore, Grodstein, Duffy, Stampfer, Czeisler and Schernhammer2014; Lo et al., Reference Lo, Groeger, Cheng, Dijk and Chee2016; Mohlenhoff et al., Reference Mohlenhoff, Insel, Mackin, Neylan, Flenniken, Nosheny, Richards, Maruff and Weiner2018; van Oostrom et al., Reference van Oostrom, Nooyens, van Boxtel and Verschuren2018; Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014). We observed the best route learning performance associated with a sleep duration of approximately seven hours, which is consistent with prior literature implicating short and long sleep duration with poorer cognitive outcomes (Lo et al., Reference Lo, Groeger, Cheng, Dijk and Chee2016). In contrast, short and long sleep durations were associated with more optimal performance for the navigation strategy task (i.e., taking shortcuts reflective of an allocentric strategy). Considering this heterogeneity in findings, large-scale, systematic investigations are necessary to enhance mechanistic understanding of conditions, cognitive domains, and populations in which nonlinear effects are most likely to be observed.
Stress and physical activity as moderators of sleep associations with spatial navigation
Surprisingly, neither perceived stress nor physical activity moderated associations of sleep disruption with spatial navigation tasks (Minkel et al., Reference Minkel, Moreta, Muto, Htaik, Jones, Basner and Dinges2014; Van Reeth et al., Reference Van Reeth, Weibel, Spiegel, Leproult, Dugovic and Maccari2000). In terms of stress, the null effects may be related to its measurement, which was a single retrospective self-report of the past week. Self-reported stress has previously exhibited mixed findings in terms of associations with the stress hormone cortisol (O’Brien et al., Reference O’Brien, Tronick and Moore2013). Whereas modulation of navigation strategy use by experimentally induced acute stress has been more consistently associated with elevated cortisol (Diamond & Rose, Reference Diamond and Rose1994; Schwabe et al., Reference Schwabe, Schächinger, de Kloet and Oitzl2009). Furthermore, stress exposure may affect specific aspects of sleep physiology, such as slow wave activity (Sanford et al., Reference Sanford, Suchecki and Meerlo2015), which would not be captured here. Lastly, studies have indicated that acute and chronic sleep deprivation lead to increased hypothalamic-pituitary-adrenal (HPA) axis reactivity to experimentally induced stressors (Suchecki et al., Reference Suchecki, Tiba and Tufik2002). Thus, the ecological sleep changes observed here may not be potentiated by stress. Overall, it may be important to systematically incorporate polysomnography to determine sleep architecture, measures of chronic stress exposure, and HPA axis reactivity via cortisol.
Another consideration is that lower physical activity and greater perceived stress may instead be directly associated with greater sleep fragmentation and circadian rest-activity fragmentation (Hooghiemstra et al., Reference Hooghiemstra, Eggermont, Scheltens, van der Flier and Scherder2015; Lambiase et al., Reference Lambiase, Gabriel, Kuller and Matthews2013). Furthermore, there are age-related reductions in physical activity and complex age effects on stress appraisal (Watson et al., Reference Watson, Carlson, Gunn, Galuska, O’Connor, Greenlund and Fulton2016; Surachman & Almeida, Reference Surachman and Almeida2018). Thus, these two lifestyle factors may also instead directly mediate age effects on spatial navigation processes, rather than interact with sleep quality as examined here. Future work in larger samples should test more complex models (e.g., multiple mediator and/or moderated mediation).
Limitations
One potential limitation is the use of measures derived from wrist-worn actigraphy devices. Wrist actigraphy may be less sensitive to sleep fragmentation compared to polysomnography (de Souza et al., Reference de Souza, Benedito-Silva, Pires, Poyares, Tufik and Calil2003), and may underestimate physical activity levels in older adults relative to hip placement (Kwan et al., Reference Kwan, Liu, Lee, Tse and Lee2020). These factors may have influenced the lack of expected age group differences in sleep fragmentation (Boulos et al., Reference Boulos, Jairam, Kendzerska, Im, Mekhael and Murray2019; Ohayon et al., Reference Ohayon, Carskadon, Guilleminault and Vitiello2004), as well as the lack of modulation by physical activity. In addition, the finding of greater circadian rest-activity fragmentation in younger compared to older adults may be due to the recruitment of college students, as this group self-reports irregular sleep-wake patterns (Lund et al., Reference Lund, Reider, Whiting and Prichard2010). Lastly, although we excluded individuals with a known neurological condition, including dementia, we did not screen older adults for global cognitive status.
Conclusions
In conclusion, we provide novel evidence of associations of both sleep fragmentation and circadian rest-activity fragmentation with hippocampal-dependent spatial navigation in a non-clinical population. These results demonstrate that increased sleep fragmentation and circadian rest-activity fragmentation are associated with reduced cognitive mapping performance in older adults. There was no support for moderating influences of physical activity or psychosocial stress. These and other factors should be further examined to determine mediators and moderators of sleep fragmentation and circadian rest-activity fragmentation on spatial navigation.
Supplementary material
The supplementary material for this article can be found https://doi.org/10.1017/S1355617722000303
Acknowledgements
We thank Brandon Davis, Alex Nelson, Cheyenne Parson, and Chidera Agwu for assistance with data collection.
Author contributions
DH and HM participated in the design of the study, data analysis and data interpretation. HM participated in data acquisition and wrote the draft of the manuscript. DH participated in manuscript revisions. BJAP participated in providing actigraphy equipment, data interpretation, and manuscript revisions. All the authors read and approved the final manuscript.
Funding statement
This work was supported by the McDonnell Center for Systems Neuroscience (DH) and National Institutes of Health R01AG057901 (BJAP).
Conflicts of interest
None.
