Sleep spindles and slow waves are physiological markers for age-related changes in gray matter in brain regions supporting problem-solving skills

Abstract

As we age, the added benefit of sleep for memory consolidation is lost. One of the hallmark age-related changes in sleep is the reduction of sleep spindles and slow waves. Gray matter neurodegeneration is related to both age-related changes in sleep and age-related changes in memory, including memory for problem-solving skills. Here, we investigated whether spindles and slow waves might serve as biological markers for neurodegeneration of gray matter and for the related memory consolidation deficits in older adults. Forty healthy young adults (20–35 yr) and 30 healthy older adults (60–85 yr) were assigned to either nap or wake conditions. Participants were trained on the Tower of Hanoi in the morning, followed by either a 90-min nap opportunity or period of wakefulness, and were retested afterward. We found that age-related changes in sleep spindles and slow waves were differentially related to gray matter intensity in young and older adults in brain regions that support sleep-dependent memory consolidation for problem-solving skills. Specifically, we found that spindles were related to gray matter in neocortical areas (e.g., somatosensory and parietal cortex), and slow waves were related to gray matter in the anterior cingulate, hippocampus, and caudate, all areas known to support problem-solving skills. These results suggest that both sleep spindles and slow waves may serve as biological markers of age-related neurodegeneration of gray matter and the associated reduced benefit of sleep for memory consolidation in older adults.

Sleep is important for good physical and mental health as well as memory (for review, see Rasch and Born 2013), including declarative memory (Plihal and Born 1997; Gais et al. 2000), procedural motor skills (Smith and MacNeill 1994; Nader and Smith 2003; Fogel et al. 2007, 2015), and cognitively complex procedural skills necessary for problem-solving (van den Berg et al. 2019). However, the benefit of sleep for offline memory consolidation is impacted by reductions in both quantity and quality of sleep with increasing age (Fogel et al. 2012; Mander et al. 2017; Sergeeva et al. 2020). For example, as we age, the total amount of slow wave sleep (SWS) is dramatically reduced (Feinberg et al. 1967; Blois et al. 1983; Bliwise 1993; Carrier et al. 2001; Ohayon et al. 2004; Cajochen et al. 2006). Aging also impacts the electrophysiological signatures of nonrapid eye movement (NREM) stage 2 sleep (NREM2); most notably, sleep spindles (Carrier et al. 2011; Lafortune et al. 2012; Martin et al. 2012). Spindles are brief (<3-sec) bursts of oscillatory activity (∼11–16 Hz) generated from rhythmic depolarization of thalamocortical neurons (Steriade 2006). In addition, age impacts the production of other oscillations in NREM sleep, including slow waves (Feinberg et al. 1967; Blois et al. 1983; Bliwise 1993; Carrier et al. 2001; Ohayon et al. 2004; Cajochen et al. 2006). Slow waves occur within the broader delta band (e.g., <4 Hz) and are typically characterized by high-amplitude, low-frequency (e.g., ∼0.5- to 2.0-Hz), synchronized, discrete phasic oscillations (McCarley 2007; Bersagliere and Achermann 2010; Helfrich et al. 2018). These marked decreases in spindles and slow waves begin as early as middle age and become progressively reduced with increasing age (Feinberg et al. 1967; Blois et al. 1983; Bliwise 1993; Carrier et al. 2001; Ohayon et al. 2004; Cajochen et al. 2006).

Significant progress has been made in terms of understanding the structural correlates of sleep-related declarative memory consolidation (Mander et al. 2013; Helfrich et al. 2018; Muehlroth et al. 2019) and procedural motor skills (Fogel et al. 2017b; Muehlroth et al. 2019). Moreover, important dissociations have been identified in terms of the unique contributions that sleep makes to different forms of memory (e.g., declarative vs. procedural) (Walker and Stickgold 2004; Walker 2005; Diekelmann and Born 2010; Born and Wilhelm 2012; Rasch and Born 2013; Stickgold 2013; Peigneux et al. 2015). Indeed, even within the category of procedural memory, sleep contributes in different ways to motor procedural skills as compared with problem-solving skills that require the acquisition of novel cognitive strategies. It remains to be determined whether distinct neurobiological substrates and their electrophysiological correlates support sleep-related consolidation of newly acquired problem-solving skills. This is the main aim of the current study.

Both sleep and memory issues are the most common complaints reported in aging populations. Age-related changes in sleep have a negative impact on sleep-dependent memory consolidation (for reviews, see Mander et al. 2017; Sergeeva et al. 2020). Simultaneous EEG-fMRI brain-imaging studies in healthy, young adults have revealed that spindles are associated with increased hemodynamic responses in the thalamic nuclei, paralimbic areas, frontal cortex, and cortical areas involved in sensorimotor processing, as well as in the hippocampus (Schabus et al. 2007). Moreover, brain areas that are recruited during both declarative (Bergmann et al. 2012; Jegou et al. 2019) and procedural (Fogel et al. 2017a; Vahdat et al. 2017) learning are subsequently reactivated during sleep, time-locked to spindles. The extent of the memory trace reactivation was associated with offline gains in performance, whereby greater postlearning spindle activity is associated with stronger memory reactivation and the related offline gains in performance.

Spindles and slow waves form a temporally related complex of events that also include hippocampal ripples (Mölle et al. 2006; Mölle and Born 2009; Latchoumane et al. 2017). These events are believed to work in concert to support memory reactivation and replay during sleep, providing a benefit to the consolidation. Recent studies have shown that spindles and slow waves become decoupled with increasing age (Helfrich et al. 2018) and that this may be related to a reduced benefit of sleep for memory consolidation. Thus, age-related changes in spindles and slow waves may serve as electrophysiological markers of the earliest indicators of age-related cognitive decline and the associated neurodegeneration.

Age-related changes in sleep are associated with a reduced enhancement of overnight memory consolidation for procedural skills (Spencer et al. 2007; Brown et al. 2009; Pace-Schott and Spencer 2011). Importantly, functional neuroimaging studies in young adults have identified the functional neuroanatomical substrates that support problem-solving skills (for review, see Unterrainer and Owen 2006). These studies show that the dorsolateral, medial prefrontal cortex, posterior hippocampus (Dagher et al. 2001), supplemental motor cortex, posterior parietal, cerebellum, and striatum, and in particular the caudate nucleus (Baker et al. 1996; Owen et al. 1996; Dagher et al. 1999, 2001; Rowe et al. 2001; Unterrainer and Owen 2006; Goldfine et al. 2013), support the acquisition of a novel cognitive strategy (one that involves problem-solving and planning). However, fMRI studies on the role of sleep for supporting motor procedural skills (Fogel et al. 2014) and complex problem-solving skills (Toor et al. 2022) have shown that older adults do not derive the same benefit from sleep for distinct types of procedural memory consolidation.

Recently, Toor et al. (2022) investigated differences in brain activation from training to retest in young versus older adults as a function of sleep versus wake. Following a period of sleep, young but not older adults displayed a reduced change in activation from training to retest in both the hippocampus and dorsolateral prefrontal cortex (structures thought to be involved in the consolidation process of problem-solving skills) and increased activation of the premotor and parietal cortex (structures involved in planning and mental and visuospatial imagery). Furthermore, sleep enhanced performance and transformed the memory trace in young adults in a pattern consistent with hippocampal–neocortical transfer. However, this pattern was not observed in older adults. This suggests that sleep contributes to the consolidation process in young, but not to the same extent in older, individuals for a set of brain areas that support problem-solving skills, including the hippocampus, prefrontal regions, and caudate.

Spindles are an electrophysiological marker of the extent of the age-related reduction in memory consolidation (Fogel et al. 2012, 2014, 2017a). The structural changes in the brain that underlie these functional consequences are beginning to be identified and understood. For example, age-related structural changes in brain regions such as the striatum, hippocampus, frontal cortices, and motor cortical regions are associated with altered sleep physiology in older age (Dubé et al. 2015; Fogel et al. 2017a). Furthermore, age-related gray matter loss has been linked to reductions in spindle and slow oscillation generation (Mander et al. 2013; Fogel et al. 2017a; Muehlroth et al. 2019). In young adults, measures of slow wave characteristics, including the amplitude, density, and spectral power of slow waves, are strongly associated with structural gray matter density and volume in the prefrontal cortex (Saletin et al. 2013). In older adults, gray matter atrophy is associated with reduced slow wave activity (Mander et al. 2013). In addition, Saletin et al. (2013) reported differences in gray matter density within the insula and auditory cortex that were related to slow spindles (∼11–13.5 Hz), while gray matter density in the hippocampus was associated with fast spindles (∼13.5–16 Hz). The extent of the SWA reduction in older adults was associated with reduced gray matter and mediated overnight declarative memory consolidation.

Recently, we identified sleep spindles to be biological markers of age-related neurodegeneration of gray matter in the hippocampus and memory consolidation deficits for motor procedural skills (Fogel et al. 2017a), thus suggesting that age-related changes in sleep might mediate cognitive decline due to neurodegenerative processes. However, this study only investigated the impact of age-related changes in sleep for cognitively simple motor procedural skills. It remains to be investigated whether sleep is similarly related to neurodegeneration of gray matter that underlies age-related changes in the consolidation of newly learned cognitive strategies required for problem-solving.

We hypothesize that (1) there will be gray matter differences in specific brain structures that support strategy learning in young compared with older adults (e.g., the hippocampus, prefrontal regions, and caudate), and (2) gray matter density in brain structures that support strategy learning will be correlated to a different extent in young versus older adults with both (a) offline changes in performance for a newly acquired cognitive strategy and (b) spindle/slow wave characteristics. We aim to investigate whether spindles and slow waves might serve as biological markers for neurodegeneration of gray matter and the related reduced benefit of sleep for memory consolidation in older adults.

Results

Behavioral data

First, to ensure that learning did take place in both young and older participants, an age (young or older) × trial (trails 1–8) ANOVA revealed a significant increase in speed over the course of the training session overall (F_(7,476) = 36.71, P < 0.0001, η² = 0.35), which was observed in both the young (F_(7,273) = 35.17, P < 0.0001, η² = 0.47) and older (F_(7,203) = 12.22, P < 0.0001, η² = 0.30) groups. There was no significant interaction with age (F_(7,476) = 1.17, P = 0.32, η² = 0.02), suggesting that both young and older adults improved on the task with practice and did so to the same extent. Again, a generally similar pattern was observed for accuracy in terms of overall improvement with practice at training (F_(7,476) = 5.70, P < 0.0001, η² = 0.08). Young participants significantly improved in terms of accuracy with practice (F_(7,273) = 7.30, P < 0.0001, η² = 0.16), whereas older adults did not (F_(7,203) = 1.41, P = 0.202, η² = 0.05). However, it should be noted that there was no significant difference in accuracy between young and older participants by the end of the training session (t₍₆₈₎ = 0.06, P = 0.960, d = 0.01), suggesting that, importantly, by the end of the training session, young and older participants did learn the task to the same extent.

To investigate the age- and sleep-related changes in performance improvements, a sleep/wake condition (nap or wake) × age (young or older) group ANOVA revealed a significant interaction for percentage increase in speed (F_(1,66) = 64.39, P < 0.0001, η² = 0.49) and accuracy (F_(1,66) = 15.73, P < 0.0001, η² = 0.19) (Fig. 1). Follow-up tests showed that the young nap (YN) condition had significantly greater improvement than the young no nap (YNN) condition (speed: t₍₃₈₎ = 8.86, P < 0.0001, d = 2.80; accuracy: t₍₃₈₎ = 2.36, P = 0.024, d = 0.75) and older nap (ON) condition (speed: t₍₃₃₎ = 12.13, P < 0.0001, d = 4.14; accuracy: t₍₃₃₎ = 4.54, P < 0.0001, d = 1.55). There was no difference between the YNN and older no nap (ONN) conditions (speed: t₍₃₃₎ = −1.16, P = 0.256, d = −0.40; accuracy: t₍₃₃₎ = −0.57, P = 0.573, d = −0.19). However, individuals in the ON condition performed significantly worse than the ONN condition (speed: t₍₂₈₎ = −2.87, P = 0.008, d = −1.05; accuracy: t₍₂₈₎ = −4.04, P < 0.0001, d = −1.47).

Figure 1.

ToH performance. Mean percent improvement for speed and accuracy on the ToH from training to retest in young and old participants. Young adults benefited from a daytime nap as compared with wake, whereas older adults did not. Error bars represent standard error of the mean. (Figure adapted with permission from Toor et al. [2022].)

It is important to note that mean speed (i.e., 1/RT) in PVT performance did not change over the course of the testing sessions as a function of age or sleep/wake condition (F_(3,192) = 2.13, P = 0.120). No other main effects or interaction effects were significant, thus suggesting that a daytime nap did not differentially impact vigilance in young and older adults over the course of the experimental protocol.

Sleep architecture

As expected, young and older participants differed in terms of sleep characteristics (Table 1). Older participants spent significantly more time awake (minutes: t₍₃₃₎ = 4.00, P = <0.001; %TST: t₍₃₃₎ = 4.00, P = <0.001). Furthermore, it took older adults significantly longer to initially fall asleep at lights out (t₍₃₃₎ = 4.75, P = <0.001), and they displayed reduced total sleep time (t₍₃₃₎ = −2.52, P = 0.017) relative to younger adults. Older participants exhibited significantly less time spent in SWS (minutes: t₍₃₃₎ = −2.46, P = 0.019; %TST: t₍₃₃₎ = −2.48, P = 0.018) and had significantly more time in lighter, NREM2 sleep (%TST: t₍₃₃₎ = 2.16, P = 0.038) relative to younger adults. Last, there was significantly more REM sleep (minutes: t₍₃₃₎ = −2.96, P = 0.006; %TST: t₍₃₃₎ = −4.18, P = <0.001) in young compared with older adults. However, it should be noted that this was largely due to a floor effect, whereby 87% of older participants had no REM sleep at all, compared with only 25% of young participants. Due to the variable duration of NREM2 sleep and an insufficient quantity of SWS across participants during a daytime napping protocol, NREM2 and SWS were combined into a total measure of NREM sleep for spindle and slow wave detection and subsequent analyses.

Table 1.

Mean (and standard error [SE]) sleep parameters recorded during the 90-min daytime nap retention interval in both young and older participants

It has been previously shown that as little as 5 min of sleep is sufficient to afford a measurable benefit on the ToH and related procedural tasks (Milner et al. 2006; King et al. 2013; Fogel et al. 2014; Fang et al. 2021; van den Berg et al. 2021; Toor et al. 2022). In the current study, young participants had on average 68.79 min ± 3.19 min of sleep and older adults had 51.53 min ± 6.83 min of sleep; which is in line with or greater than previous studies and would be sufficient to lead to the hypothesized effect of sleep (Leong et al. 2022). In addition, sleep duration did not correlate with offline gains in performance in either young (r₍₂₀₎ = −0.300, P = 0.199) or older participants (r₍₁₅₎ = 0.371, P = 0.173). Thus, it is unlikely that sleep duration on its own could explain differences in memory consolidation.

Sleep spindles and slow waves

Younger participants had significantly more spindles per minute (slow: t₍₃₃₎ = −2.74, P = 0.010; fast: t₍₃₃₎ = −2.77, P = 0.009) and significantly larger spindles (slow: t₍₃₃₎ = −3.67, P = <0.001) relative to older adults (Table 2). Younger adults had significantly more slow waves per minute (t₍₃₁₎ = −3.57, P = 0.001) and significantly larger slow waves (t₍₃₁₎ = −3.60, P = 0.001) relative to older adults (Table 2).

Table 2.

Density and size of slow (11- to 13.5-Hz) and fast (13.5- to 16-Hz) spindles and slow waves during the 90-min daytime testing nap intervals in young and older participants

Structural MRI results

Gray matter differences in young vs. older adults

As expected, greater gray matter density was observed in young versus older adults in a widespread set of brain regions, including the cortico–striatal–hippocampal and cortico–cerebellar regions. There was greater gray matter density in the cerebellum, cingulate cortex, and precentral gyrus bilaterally. Other regions such as the midtemporal, medial frontal gyrus, visual associative, temporal lobe, superior frontal, anterior prefrontal cortex, and primary motor cortex also displayed greater gray matter density in young relative to older adults (Table 3; Fig. 2). Importantly, greater gray matter density was observed in young adults in areas important for acquisition of novel cognitive strategies such as the striatum, including the putamen and the caudate, as well as bilaterally in the hippocampus. Finally, the widespread gray matter differences observed frontally included the prefrontal and dorsolateral prefrontal cortex in young versus older adults.

Table 3.

Brain regions with significantly greater gray matter in young versus older participants

Figure 2.

Gray matter differences in young versus older adults. Several areas that support procedural problem-solving skills differed between young and older adults, including motor cortical sensorimotor areas, the mPFC, the precuneus, the cingulate cortex, the caudate nucleus, the putamen, the hippocampus, and the cerebellum. Clusters thresholded at P < 0.005, FWE. X, Y, and Z coordinates are given in MNI coordinates, displayed in radiological orientation.

Gray matter correlates differently with ToH performance, spindles, and SW in young vs. older adults

To follow up the age-related differences in gray matter between young and older adults, a series of regressions was conducted by including ToH performance, spindle, and slow wave parameters (e.g., density and size) into the GLMs as covariates of interest. Here, we controlled for age, sex, and depression scores. Gray matter in young and older individuals correlated to a different extent in young (i.e., more positive) versus older (i.e., more negative) adults with improvements in ToH speed in the brainstem (the pons in particular), putamen, hippocampus, cerebellum, temporal lobe, parietal lobe, and prefrontal cortex (Table 4A). In addition, gray matter correlated to a different extent in young (i.e., more positive) versus older (i.e., more negative) adults with spindles in the parieto–occipital sulcus (a region important for planning), the premotor, the supplementary motor area, and the pons, and bilaterally in the primary motor cortex (Table 4B). Finally, gray matter in young and older individuals correlated to a different extent in young (i.e., more positive) versus older (i.e., more negative) individuals with slow waves in the pons, the cerebellum, and both the anterior and posterior hippocampus (Table 4C). In addition, to determine whether total sleep duration was significantly related to gray matter in young versus older adults, we also entered total sleep time into the model. However, there was no significant difference in the relationship between gray matter and sleep duration in young versus older adults, thus suggesting that the duration of the nap itself is unlikely to explain the above findings.

Table 4.

Brain regions where relationship between gray matter and the covariates of interest (e.g., performance improvements, slow waves, and spindles) differed significantly between young (i.e., more positive correlation) and older (i.e., more negative correlation) adults

Conjunction between ToH performance gains and sleep in young and older adults

The next step was to perform conjunction analyses, which allowed us to colocalize differences in gray matter that correlated to a different extent in young and older participants with both offline changes in performance and sleep oscillation characteristics (e.g., spindles and slow waves). Conjunction analyses (Fig. 3) revealed regions where gray matter was correlated to a different extent in young versus older adults with both ToH performance gains and sleep parameters. Gray matter correlated with both ToH gains and slow waves to a different extent in young versus older adults in the anterior cingulate cortex, hippocampus (Fig. 3F,G and inset), cerebellum, caudate (Fig. 3C,D), putamen, and thalamus (Table 5B). Again, the same overall pattern was observed in which gray matter was positively correlated in young individuals and negatively correlated in older individuals.

Figure 3.

(A) Significant clusters of gray matter positively correlated with offline changes in performance in conjunction with slow wave size in the hippocampus, caudate, putamen, and cerebellum in young versus older adults (Table 5). The relationship between gray matter and the conjunction of the covariates of interest (e.g., performance improvements and slow waves) differed significantly between young (i.e., more positive correlation) and older (i.e., more negative correlation) adults. (B,C) Mean gray matter density in the caudate was higher (P < 0.005, FWE) on average in young versus older adults (B) and was more negatively correlated with offline changes in performance in older individuals than in young participants, who showed a positive relationship between changes in performance and gray matter (C). (D) A similar pattern of differences between correlations in young versus older adults was observed between gray matter and slow wave size. (E–G) The same overall pattern was observed for the hippocampus. (Inset) Sagittal view at X = 28. X, Y, and Z coordinates are given in MNI coordinates, displayed in radiological orientation.

Table 5.

Brain regions where relationship between gray matter and the conjunction of the covariates of interest (e.g., performance improvements, slow waves, spindles) differed significantly between young (i.e., more positive correlation) and older adults (i.e., more negative correlation)

In addition, gray matter correlated with both ToH gains and spindles to a different extent in young versus older adults in the superior temporal, cerebellum, brainstem (in particular the pons), somatosensory, prefrontal, orbitofrontal, anterior cingulate, and parietal cortex. Similar to previous work (Fogel et al. 2017a), for all regions, the same overall pattern was observed in which positive correlations were observed in young adults, while negative correlations were observed in older adults (Table 5A), which significantly differed from one another.

Discussion

Here, we investigated whether age-related changes in gray matter intensity relate to key features of NREM sleep (spindles and slow waves) that are known to be important for memory consolidation and are reduced with age. Consistent with the extant literature, we observed reduced gray matter density in a widespread network of brain areas within the cortico–striatal–hippocampal and cortico–cerebellar regions in older compared with younger adults (Table 3; Fig. 4). We followed up these analyses using a conjunction analysis approach, which revealed gray matter density correlated with both offline improvements in ToH performance and sleep spindles to a different extent in young versus older adults in primarily cortical areas. These included the superior temporal lobe, parietal cortex, somatosensory cortex, orbitofrontal cortex, and prefrontal cortex. In contrast, ToH performance gains and slow wave characteristics correlated with gray matter density to a different extent in young versus older adults almost exclusively in subcortical areas that support learning and offline improvement for newly learned cognitive strategies needed for problem-solving. These areas included the anterior cingulate cortex, hippocampus, thalamus, cerebellum, putamen, and caudate nucleus. The overall pattern of the relationships between gray matter and improvements in ToH performance and sleep characteristics was consistent, where gray matter density was generally more positively correlated (with both offline gains and sleep EEG) in the young and, by comparison, generally more negatively correlated in older individuals. These results suggest that sleep spindles and slow waves serve as markers for age-related changes in gray matter in regions that support sleep-related memory consolidation of newly learned cognitive strategies, in which spindles may be primarily related to the cortical aspects and slow waves may be primarily related to subcortical regions. These brain areas are critical for the consolidation of novel problem-solving skills.

Figure 4.

Study design. Overview of experimental protocol in the nap and wake conditions. After an initial screening and acclimatization polysomnography (PSG), participants were divided into young nap (YN), young no nap (YNN), older nap (ON), and older no nap (ONN) conditions. Nap condition: Participants arrived at the sleep laboratory at 9:00 a.m. and were trained and tested on the ToH at 10:00 a.m. They then had a nap opportunity from 1:00 p.m. to 2:30 p.m. while their sleep was recorded via PSG. Upon awakening, but after sleep inertia dissipation, participants were retested at 5:00 p.m. Wake condition: The same procedure was used in the eake condition, except that instead of a daytime nap opportunity, individuals remained awake and read in a supine position with polysomnographic recording.

In terms of behavioral improvements, despite slower performance consistently reported in older participants compared with younger participants (Spencer et al. 2007; Brown et al. 2009; Pace-Schott and Spencer 2011), older participants displayed the same pattern of increasing performance over the course of the training session as the young participants did on the ToH, thus suggesting that initial learning of the strategy was not significantly impacted by age. Despite intact learning, older adults benefited less from retention intervals that contain sleep as compared with young adults. It is worth noting that psychomotor vigilance did not differ in young and older adults over the course of the experimental protocol. Thus, changes in ToH performance are unlikely to be due to the impact of a nap on psychomotor vigilance. Together, these results suggest age-related deficits in latent/offline learning processes (Spencer et al. 2007; Pace-Schott and Spencer 2011; Mander et al. 2013; Fogel et al. 2014, 2017a).

Consistent with well-known age-related reductions in gray matter as a part of the normal aging process (Gunning-Dixon et al. 1998; Fjell and Walhovd 2010; Raz et al. 2015), we observed the most pronounced changes in gray matter in the thalamus, hippocampus, striatum, frontal cortex, motor cortical regions, and cingulate cortex. These regions support procedural memory and are critical for planning, learning new strategies, and problem-solving skills (Baker et al. 1996; Owen et al. 1996; Dagher et al. 1999; Lazeron et al. 2000; Rowe et al. 2001; Unterrainer and Owen 2006). Our results revealed that gray matter correlated with both performance gains and slow waves to a different extent in young (i.e., more positive) versus older (i.e., more negative) adults in primarily subcortical regions (e.g., the anterior cingulate cortex, hippocampus, cerebellum, caudate, putamen, and thalamus) and, in contrast, largely in cortical regions (e.g., the superior temporal, somatosensory, parietal, prefrontal, and orbitofrontal cortex) for spindles. The differential association between gray matter and both spindles and offline gains in performance in the parietal, temporal lobe, and somatosensory cortices are in line with PET and fMRI studies in healthy young adults using similar tasks requiring planning and strategy (Baker et al. 1996; Owen et al. 1996; Dagher et al. 1999; Lazeron et al. 2000; Rowe et al. 2001; Unterrainer and Owen 2006). Our results suggest that spindles are a marker of age-related neurodegeneration of gray matter that relates to the extent of the lost benefit of sleep for the consolidation of newly acquired cognitive strategies.

The differential association between gray matter and both slow waves and offline gains in performance in the hippocampus and the caudate in particular are in line with studies showing that slow waves are associated with the degree of increasing hippocampal independence during postsleep memory retrieval (Takashima et al. 2006). The hippocampus is responsible for encoding newly acquired information and later transferring it to the cortex for long-term storage (Dudai 1996; Squire and Zola-Morgan 1996; Dudai et al. 2015). As consolidation progresses, information is integrated into neocortical circuits and reorganized so that it can ultimately be retrieved independently of the hippocampus (Takashima et al. 2009). Previous behavioral (Spencer et al. 2007; Brown et al. 2009; Pace-Schott and Spencer 2011), EEG (Fogel et al. 2014; Debarnot et al. 2017), and fMRI (Fogel et al. 2014; Toor et al. 2022) studies have shown that older adults do not derive the same benefit from sleep for procedural memory consolidation.

The present study indicates that slow waves are a marker of the age-related neurodegeneration in the hippocampus that impacts memory consolidation for newly acquired cognitive strategies. Moreover, a similar pattern was observed in the caudate nucleus. The caudate supports higher-order reasoning, goal-directed behavior, novel learning (e.g., strategies), and problem-solving skills (Owen et al. 1996; Dagher et al. 1999), all of which are required to perform the ToH. Activation of the caudate nucleus is observed during active planning of a novel action versus continuous rule following (Monchi et al. 2006), suggesting a role for novel and potentially higher-order learning. Our results suggest that slow waves are an electrophysiological marker of neurodegeneration of the caudate, which relates to reduced sleep-dependent memory consolidation in older adults.

Limitations

The current study was limited in terms of the use of a daytime nap protocol. A nap protocol is preferable to control for time of day effects when comparing young versus older adults but is limited in terms of sleep duration and architecture. Recent studies have shown that the coupling between slow waves and sleep spindles is reduced with age and impacts memory consolidation (Helfrich et al. 2018; Muehlroth et al. 2019). However, due to the relatively short duration of the nap recordings, these data are insufficient for this type of analysis. There was also limited REM sleep, which precluded further analysis of REM-related covariates. This type of procedural memory consolidation has been found to be related to the features of REM sleep (Conway and Smith 1994; Smith and Smith 2003; Fogel et al. 2015). Future studies using overnight sleep recordings could address these important issues.

The ToH is widely considered to be an archetypal implicitly learned strategy task; however, it is possible that participants could have memorized the strategy. This possibility is unlikely given the limited number of trials used at training and the use of the relatively complex five-disk version of the ToH, which requires a minimum of 31 moves (62 key presses in total) needed to solve the puzzle. In a previous study (Viczko et al. 2018), we found no evidence of explicit awareness of a complex series of motor movements in a procedural memory task consisting of only 13 moves—far less than what is required to arrive at the optimal solution to the ToH. Thus, the optimal sequence of movements was not likely to have been memorized or learned explicitly; rather, the underlying cognitive strategy needed to arrive at the optimal solution to the ToH was most likely acquired implicitly, possibly through trial and error. This interpretation is consistent with several other studies investigating problem-solving skills (e.g., Baker et al. 1996; Dagher et al. 1999, 2001; Ronnlund et al. 2001; Rowe et al. 2001; Beauchamp et al. 2003; Smith and Smith 2003; Anderson et al. 2005; Boghi et al. 2006; Unterrainer and Owen 2006; Brand et al. 2010; Ashworth et al. 2014; Fogel et al. 2014, 2015, 2017a; Balachandar et al. 2015; Nielsen et al. 2015; van den Berg et al. 2019).

Another limitation was an imbalance in the number of participants per group (i.e., N = 15/group in the older age groups, N = 20/group in the younger groups). This was handled in two different ways in the analyses: (1) As recommended in the FSL VBM manual, the study-specific template was generated using equal numbers of participants (randomly selected) from each group so as to not bias the template and the downstream analyses to one age group or another, and (2) the statistical analyses are permutation-based nonparametric statistics, which involve randomly shuffling labels for participants over 10,000 iterations. This procedure is appropriate given that there is no assumption of equal sample sizes for permutation-based tests. In addition, for all groups, the target sample size of N = 15, determined a priori, was achieved. Thus, all hypotheses were tested with the required statistical power.

All participants were within the target age range and below the cutoff for depression and other related inclusion criteria. However, participants were not perfectly matched for age, sex, and depression scores across groups. In order to control for any potential confounds, we entered these variables into the model as variables of noninterest. It should be noted that inclusion of these control variables did not alter the overall outcome or conclusions and, in fact, strengthened the effects of interest. It is also worth noting that for daytime nap protocols such as this, it is not possible to control total sleep time. Consequently, we entered sleep duration as a variable of interest to ascertain whether the duration of the nap was related to gray matter in a manner similar to spindles and slow waves. This analysis revealed that there was no significant relationship between total sleep time and gray matter in young versus older adults. Thus, it is unlikely that nap duration alone could account for the observed pattern of results.

Conclusions

In conclusion, the results of the present study suggest that age-related changes in sleep spindles and slow waves are differentially related to gray matter density in young and older adults in brain regions that support sleep-dependent memory consolidation for problem-solving skills. Both sleep spindles and slow waves may serve as biological markers of age-related neurodegeneration of gray matter and the associated memory consolidation deficits in older adults.

Materials and Methods

Ethics statement

All participants provided informed consent and were financially compensated for their participation. This research was approved by the University of Ottawa's Research Ethics Board, as well as the Research Ethics Board at The Royal's Institute of Mental Health Research.

Participants

A priori power analyses, based on studies using the closest experimental designs, techniques, and methodology, suggested that N = 15 is sufficient for the study to be adequately powered to test our main hypotheses. Previous studies (Fogel et al. 2014, 2017b) indicated that we would need a bare minimum of N = 15/group to obtain 80% power (Mumford 2012) to detect age effects for the behavioral and MRI data and N = 12/group for spindle effects at 95% power. This is based on an observed effect size of f = ∼0.7 in young (age 20–35) versus older (age >65) participants. A recent meta-analysis (Leong et al. 2022) has shown that the effect of a nap on procedural tasks is Cohen's D = 0.5 (CI₉₅ = 0.3–0.7, or medium to large effect sizes).

Participants were asked to complete a preliminary screening questionnaire to determine their eligibility, which confirmed that they were in good health, right-handed, nonshift workers; free from chronic pain and seizures; did not take medications known to interfere with sleep; had normal or corrected-to-normal vision, a body mass index <30, no previous history of head injury or any hand mobility problems; considered themselves to be nonsmokers; and normally consumed less than two to three caffeinated beverages per day. In addition, all participants were required to have no previous experience with the Tower of Hanoi (ToH) or similar tasks. Eligible participants were then asked to complete additional screening questionnaires to rule out depression, anxiety, and any sleep disorders. To be included in the study, participants were required to score <10 on the Beck depression (Beck and Beamesderfer 1974) and Beck anxiety (Beck et al. 1988) inventories, in addition to having no signs of sleep disorders as indicated by the Sleep Disorders Questionnaire (Douglass et al. 1994). To rule out signs of mild cognitive impairment or dementia, older participants were also required to score ≥26 on the Montreal Cognitive Assessment (Nasreddine et al. 2005) and >2 on the Everyday Cognition Scale (Tomaszewski et al. 2008).

For the duration of the study, all participants were required to limit caffeine intake (no more than one beverage in the morning) and abstain from alcohol and nicotine consumption during the experiment. To ensure compliance with the study protocol, participants were required to wear a Motionlogger (Ambulatory Monitoring, Inc.) actigraph, a wrist-worn accelerometer that measures sleep–wake-related limb movements. In addition to actigraphy, participants were also asked to complete a log of their daily activities and sleep habits, keep a regular sleep–wake cycle (bedtime: 10:00 p.m.–1:00 a.m., wake time: 6:00 a.m.–9:00 a.m.), and abstain from daytime naps. Participants were excluded from the study if the results of the actigraphy or sleep log indicated noncompliance with the study protocol.

Participants who met the initial eligibility criteria were then asked to complete a polysomnographic recording, which served to acclimate them to sleeping in the laboratory, to confirm that they were able to sleep in these conditions as well as to screen for signs of sleep disorders. During the screening polysomnography (PSG), electroencephalography (EEG) was recorded from six scalp electrodes (i.e., Fz, Cz, Pz, Oz, M1, and M2), placed according to the 10–20 system (Jasper 1958). In addition to the EEG, eye movements were recorded via electrooculography (EOG) from electrodes placed on the outer canthus of the eyes. Bipolar chin muscle activity and heart rhythm were recorded via submental electromyography (EMG) and electrocardiography (ECG), respectively. All electrodes were referenced to Fpz online with Afz ground and impedances kept <5 kΩ. In addition, an electrode was placed on the anterior tibialis muscle of each leg to screen for restless leg syndrome and periodic limb movement disorder. To screen for sleep apnea, blood oxygen saturation was recorded via pulse oximetry, nasal airflow was recorded via a thermistor, and both thoracic and abdominal respiratory efforts were recorded via respiratory belts. Based on the results of the polysomnographic screening, participants were included in the final sample if they had a sleep efficiency >80%, a periodic limb movement (PLM) index <10 events per hour, and an apnea–hypopnea index less than five events per hour.

Following polysomnographic screening, all participants were randomly assigned to either the nap or wake (i.e., “no nap”) condition, yielding four groups: young nap (YN), young no nap (YNN), older nap (ON), and older no nap (ONN). To minimize time of day effects for learning and memory performance in young versus older adults, a nap protocol was used. For the participants that met the initial eligibility criteria, seven older participants and five young participants did not meet the inclusion criteria for keeping a regular sleep schedule within the prescribed hours. Furthermore, three older participants did not meet the polysomnographic prescreening criteria due to not achieving any sleep during the nap opportunity. Finally, one younger participant did not meet the polysomnographic prescreening criteria for bruxism, and another did not meet the prescreening criteria for periodic leg movements. The final sample consisted of 40 healthy young adults aged 20–35 yr old (23 females; mean age = 23.85 yr, SD = 3.86) and 30 healthy older adults aged 60–85 yr old (20 females; mean age = 68.00 yr, SD = 7.62). The final groups included N = 20 (12 females; mean age = 23.05 yr, SD = 3.0) young participants in the nap condition (YN), N = 20 (11 females; mean age = 24.65 yr, SD = 4.50) young participants in the wake condition (YNN), N = 15 (nine females; mean age = 66.13 yr, SD = 6.97) older participants in the nap condition (ON), and finally, N = 15 (11 females; mean age = 68.07 yr, SD = 8.41) older participants in the wake condition (ONN).

The data analyzed in this study were part of a larger polysomnographic, behavioral, and MRI protocol using various techniques intended to address a variety of research questions. The behavioral results have been partly reported previously (van den Berg et al. 2021; Toor et al. 2022). These previous studies investigated functional brain activity in relation to sleep and memory, whereas the current study is focused on structural gray matter in relation to sleep and memory. The structural MRI data and VBM analyses have not been reported elsewhere and are unique to this study. See Toor et al. (2022) for additional methodological details about the experimental protocol specific to that study. All methods relevant to the current investigation are reported here, and methods used in related studies are reported in the Supplemental Material.

Experimental design

One week following the screening process, all eligible participants returned to the sleep laboratory for ToH testing with MRI structural scans in addition to PSG (Fig. 4). Participants completed the training session of the Tower of Hanoi (ToH) at 10:00 a.m. (see Toor et al. 2022 for corresponding behavioral and functional neuroimaging results in these participants). At 1:00 p.m., the nap participants engaged in a 90-min daytime nap opportunity, whereas participants in the wake condition remained awake while reading and lying in a supine position under dim ambient lighting. Both nap and wake conditions were monitored via polysomnographic recording (in part to ensure the participants in the wake conditions did not fall asleep). At 5:00 p.m., >30 min after the end of the nap opportunity (to avoid sleep inertia effects), participants completed the retest session of the ToH. The Stanford Sleepiness Scale (Maclean et al. 1992) and the Psychomotor Vigilance Test (PVT) (Dinges and Powell 1985) were administered prior to and following the training and retest sessions to obtain subjective and objective measures of sleepiness across the testing protocol.

Behavioral tasks and analysis

A computerized version of the ToH task was used, coded in Matlab R2016a (Mathworks, Inc.) using the Psychophysics Toolbox extension v3.0.12 (Brainard 1997; Kleiner et al. 2007) for Windows (Microsoft). The ToH task included five disks of increasing size stacked in ascending order that could be moved to and from one of three pegs by pressing the corresponding key on the button pad. The main goal of the task was to move all disks from the start (far-left peg) position to the goal (far-right peg) position in the same ascending order while abiding by the following constraints: (1) Disks could only be moved one at a time, (2) each move required taking an uppermost disk from one of the stacks and placing it on another peg, and (3) only a smaller disk could be placed on top of a larger disk. The task can be solved through trial and error in order to learn to use recursive logic, breaking down the overall optimal solution into smaller components and recursively performing this sequence of smaller solutions several times. The optimal number of moves to complete the ToH is determined algebraically by 2^N − 1, where N is the number of disks, or 31 moves for the five-disk task.

In total, participants performed eight training trials (to ensure adequate practice) and four retest trials of the ToH, where each trial was followed by a 20-sec rest period. During the rest, participants were shown the three pegs with no disks and were instructed not to press any buttons. A trial of the ToH ended when either the participant solved the task (i.e., all five disks were in ascending order on the far-right peg) or the maximum number of moves was reached prior to solving the task. The maximum number of moves was set to two times the optimal number of moves (2^N − 1, where N is the number of disks; i.e., 2⁵ − 1 = 31) or, in this case, 62 moves. ToH performance was measured in terms of speed (time to complete each ToH trial) and accuracy (mean absolute percentage error [MAPE]). The MAPE for n trials was calculated as follows:

$$\mathrm{MAPE}=100-\left({\displaystyle \frac{1}{n}}\right)\times \sum \left({\displaystyle \frac{\left|\mathrm{number}\mathrm{of}\mathrm{moves}-\mathrm{optimal}\mathrm{number}\mathrm{of}\mathrm{moves}\right|}{\mathrm{number}\mathrm{of}\mathrm{moves}}}\right)\times 100.$$

The MAPE can be interpreted as the percentage of perfect performance (e.g., 31 moves), in which 100% represents perfect performance (i.e., the optimal solution), whereas 50% reflects making twice as many moves as perfect performance, and 0% theoretically represents an infinite number of moves. For all measures, the percent improvement for both speed and accuracy was calculated as follows:

$$\text{\%}\mathrm{improvement}=\left({\displaystyle \frac{\mathrm{retest}-\mathrm{training}}{\mathrm{training}}}\right)\times 100.$$

Behavioral analysis

SPSS Statistics version 27 (IBM) was used to carry out the behavioral statistical analyses. To ensure that learning did take place on the ToH task in both the young and older groups, 2 × 8 age (young or older) × trial (trails 1–8) ANOVAs were conducted for speed and accuracy. Simple effects repeated measures ANOVAs in each group were used to follow up any significant effects.

To explore whether sleep facilitated performance on the ToH task in young and older groups, 2 × 2 age (young or older) × condition (nap or wake) ANOVAs were conducted for percent change in speed and accuracy. Independent samples t-tests were used to follow up any significant group effects.

Polysomnographic recording and analysis

Recording parameters

Embla N7000 amplifier systems (Natus) were used for in-laboratory polysomnographic recordings. All signals were sampled at 500 Hz. EEG (F3, Fz, F4, C3, Cz, C4, P3, Pz, P4, Oz, M1, and M2) and EOG (placed on the outer canthus of each eye) were referenced online to Fpz with a ground electrode placed at Afz. EEG and EOG were rereferenced offline to average mastoid derivations (M1 and M2), placed according to the international 10–20 system (Jasper 1958). Submental EMG channels were recorded as bipolar derivations. EEG was filtered offline from 0.3 to 35 Hz, EOG was filtered from 0.3 to 10 Hz, and EMG was filtered from 10 to 50 Hz.

Manual sleep stage scoring was conducted by an expert PSG technologist according to the American Academy of Sleep Medicine criteria (Iber et al. 2007) in 30-sec epochs using Hume analysis software (https://www.jaredsaletin.org/hume) for EEGlab (Delorme and Makeig 2004). Movement artifacts were automatically detected using custom Matlab (The Mathworks, Inc.) scripts. This method uses a variance-based (i.e., first-derivative) transformation of the EMG channels, and then movement event markings are visually inspected by an expert to validate the automatic detection algorithm before being used to exclude EEG from further analyses.

Sleep spindle detection

A well-established (Fogel et al. 2014, 2017a,b; Albouy et al. 2015; Fang et al. 2017, 2019) and validated (Ray et al. 2015) method implemented in EEGlab using the “Detect_Spindles” plug-in (https://github.com/stuartfogel/detect_spindles) was used to detect spindles from Fz and Pz in movement artifact-free NREM sleep. For more details, see Ray et al. (2015). Spindles were identified in the 11- to 16-Hz range (Landolt et al. 1996; Zeitlhofer et al. 1997; Fogel et al. 2007, 2014; Ferrarelli et al. 2010) and identified as either slow (11–13.5 Hz) or fast (13.5–16 Hz) at the sites where they were maximal (Terrier and Gottesmann 1978; Anderer et al. 2001; De Gennaro and Ferrara 2003); that is, at Fz and Pz, respectively. The number of spindles during NREM sleep, density (number per minute), and size (sum of absolute amplitude) were extracted from these analyses.

Slow wave detection

Slow waves were detected using period amplitude analysis (PAA), adapted from previously published methods (Feinberg et al. 1978; Geering et al. 1993; Bersagliere and Achermann 2010), using the EEGlab (Delorme and Makeig 2004) “PAA” plug-in (https://github.com/stuartfogel/Period-Amplitude-Analysis) written for Matlab (The MathWorks, Inc.). The slow wave data were extracted from movement artifact-free NREM sleep epochs. In line with previous research (Bersagliere and Achermann 2010), the detection method used bandpass filtering to detect waves between 0.5 and 2.0 Hz. Derivations Cz, Fz, Oz, and Pz were bandpass-filtered at 0.46 Hz (64th-order Chebyshev type II high-pass filter, −80-dB stopband attenuation) and 2.15 Hz (32nd-order Chebyshev type II low-pass filter, −80-dB stopband attenuation) to achieve minimal attenuation in the band of interest (0.5–2.0 Hz) and good attenuation at neighboring frequencies. The filters were applied in the forward and reverse directions to achieve zero phase distortion, resulting in doubling of the filter order. Half-waves were determined as deflections (negative or positive) between two consecutive zero crossings in the bandpass-filtered signal. In line with scoring rules for slow waves, the peak-to-peak amplitude threshold of 75 µV (Iber et al. 2007) was applied to neighboring half-waves. Slow wave detection was visually verified following automated detection. As described in Bersagliere and Achermann (2010), half-wave number, density (number per minute), and area (square microvolts) were extracted for each participant where slow waves are typically maximal (Fz). Exploratory bivariate correlations between behavioral performance improvements on the ToH (accuracy and speed) and sleep oscillations (fast spindles, slow spindles, and slow waves) during NREM sleep in young and older adults are presented in Supplemental Table S1.

MRI acquisition and analysis

Recording parameters

Functional magnetic resonance imaging (fMRI) data were collected on a Siemens Biograph mMR 3.0 Tesla MRI whole-body scanner (Siemens) using a 12-channel head coil. Anatomical images were acquired using a standard 3D Multislice MPRAGE sequence (TR = 2300 msec, TE = 2.98 msec, TI = 900 msec, FA = 9°, 176 slices, FoV = 256 × 256 mm², matrix size = 256 × 256 × 176, voxel size = 1 × 1 × 1 mm³) to obtain high-resolution T1-weighted images for all participants and were used for subsequent VBM analyses.

Preprocessing

Structural data were analyzed with FSL (Douaud et al. 2007) using an optimized VBM protocol (Good et al. 2001; Smith et al. 2004). Similar to a previously published study using this method (Fogel et al. 2017a), the following steps were used: (1) Structural images were brain-extracted and gray matter was segmented before being registered to MNI 152 standard space using nonlinear registration (Andersson et al. 2007). (2) A study-specific gray matter template was created using the FSL-VBM protocol. (3) All brain-extracted images were segmented into gray matter, white matter, and cerebrospinal fluid (CSF). (4) Gray matter images were then affine-registered to the MNI gray matter ICBM-152 template, concatenated, and averaged. (5) The gray matter images were then rereregistered to this first-pass “affine” gray matter template using nonlinear registration, concatenated into a 4D image, averaged, and flipped along the X-axis. (6) Both mirror images were then averaged to create the final symmetrical, study-specific nonlinear gray matter template at 2 × 2 × 2-mm³ resolution in standard space. (7) All native gray matter images were nonlinearly registered to the study-specific template generated in the previous step and modulated to correct for local expansion due to the nonlinear component of the spatial transformation (Good et al. 2001). (8) Last, all images were then smoothed with an isotropic Gaussian kernel with a sigma of 3 mm.

Statistical analysis

To investigate age-related differences in gray matter, voxel-wise General Linear Modeling (GLM) using FSL VBM was applied to investigate gray matter for the linear contrasts for (1) young, (2) older, and (3) young-older participants using threshold-free cluster enhancement (TFCE) permutation-based nonparametric testing using 10,000 permutations (Nichols and Holmes 2002; Winkler et al. 2014). This procedure generated maps of t-statistics, the corresponding uncorrected P-value maps, and family-wise error (FWE)-corrected P-value maps for young versus older groups of participants. Total intracranial volume, age, sex, and depression scores were mean-centered and included in the GLM as variables of noninterest in the model. To follow up age-related differences in gray matter in young versus older adults, the offline changes in (1) ToH performance, (2) spindles, (3) slow waves, and (4) total sleep time were entered as covariates of interest into the GLM. Contrast images were generated to identify brain regions where gray matter correlated with the covariates of interest to a different extent in young versus older individuals.

Finally, to colocalize differences in gray matter in young versus older participants that correlated with both offline changes in performance and sleep characteristics, the conjunction of performance-related gray matter t-maps and spindle- or slow wave-related gray matter t-maps was taken as the minimum t-statistic using the global null hypothesis (Friston et al. 2005). These conjunction maps were able to identify brain regions where gray matter was correlated to a different extent in young and older (and young vs. older) participants with offline changes in performance, spindles, or slow waves over (1) the performance-correlated statistical parametric maps in young and older participants with (2) each spindle-correlated map (e.g., density and area) in young and older participants (e.g., young, older, and young vs. older) and (3) each slow wave-correlated map (e.g., density and area) in young and older participants (e.g., young, older, and young vs. older). The conjunction maps were thresholded so that results were significant at a combined P-level of P < 0.001. Note that a significant conjunction does not necessarily mean that all the contrasts were individually significant (i.e., a conjunction of significance). Instead, it indicates that the effects were consistently high across all conditions included in the conjunction, such that they were jointly statistically significant (Friston et al. 2005). Note that the minimum t-values do not have the usual Student's t-distribution, and small minimum t-values can be highly significant.