Variability in objective sleep is associated with Alzheimer’s pathology and cognition

Abstract

Both sleep duration and sleep efficiency have been associated with risk of Alzheimer’s disease, suggesting that interventions to promote optimal sleep may be a way to reduce Alzheimer’s disease risk. However, studies often focus on average sleep measures, usually from self-report questionnaires, ignoring the role of intra-individual variability in sleep across nights quantified from objective sleep measures. The current cross-sectional study sought to investigate the role of intra-individual variability in accelerometer-based objective sleep duration and sleep efficiency in relation to in vivo Alzheimer’s disease pathology (β-amyloid and tau) using positron emission tomography imaging and cognition (working memory, inhibitory control, verbal memory, visual memory and global cognition). To examine these relationships, we evaluated 52 older adults (age = 66.4 ± 6.89, 67% female, 27% apolipoprotein E4 carriers) with objective early mild cognitive impairment. Modifying effects of apolipoprotein E4 status were also explored. Less intra-individual variability in sleep duration was associated with lower β-amyloid burden, higher global cognition and better inhibitory control, with a trend for lower tau burden. Less intra-individual variability in sleep efficiency was associated with lower β-amyloid burden, higher global cognition and better inhibitory control, but not with tau burden. Longer sleep duration was associated with better visual memory and inhibitory control. Apolipoprotein E4 status significantly modified the association between intra-individual variability in sleep efficiency and β-amyloid burden, such that less sleep efficiency variability was associated with lower β-amyloid burden in apolipoprotein E4 carriers only. There was a significant interaction between sleep duration and apolipoprotein E4 status, suggesting that longer sleep duration is more strongly associated with lower β-amyloid burden in apolipoprotein E4 carriers relative to non-carriers. These results provide evidence that lower intra-individual variability in both sleep duration and sleep efficiency and longer mean sleep duration are associated with lower levels of β-amyloid pathology and better cognition. The relationships between sleep duration and intra-individual variability in sleep efficiency with β-amyloid burden differ by apolipoprotein E4 status, indicating that longer sleep duration and more consistent sleep efficiency may be protective against β-amyloid burden in apolipoprotein E4 carriers. Longitudinal and causal studies are needed to better understand these relationships. Future work should investigate factors contributing to intra-individual variability in sleep duration and sleep efficiency in order to inform intervention studies.

Fenton et al. report that less variability in sleep duration and efficiency is associated with lower amyloid burden and better cognition in older adults. The association between sleep efficiency variability and β-amyloid was modified by APOE4 status, suggesting that consistent sleep may be protective against amyloid burden in APOE4 carriers.

Graphical Abstract

Graphical Abstract.

Introduction

Alterations in sleep observed in some older adults exceed those changes attributed to normal aging, with evidence suggesting that the most marked changes may occur in those at risk of developing Alzheimer’s disease (Ad).^1,2 Indeed, numerous studies have reported significant associations between various sleep measures and current and future levels of β-amyloid (Aβ)^3–16 and tau pathology^4,5,11,17,18 and risk of cognitive decline.^14,19,20 However, several questions remain unanswered. Many studies have characterized sleep using self-report measures, which have been shown to differ from objectively measured sleep^4,21–27 with the greatest discrepancy occurring in individuals with higher levels of Aβ.⁴ Additionally, the literature has largely focused on average measures of sleep quality and quantity, ignoring the potential role of intra-individual variability. One aging cohort study that examined the association between variability in sleep measures and risk of cognitive decline found that individuals with greater variability in sleep efficiency and sleep duration had increased risk of conversion to mild cognitive impairment (MCI) and dementia.²⁸ However, it remains unclear how Aβ or tau deposition may have contributed to the relationship, as those measures were not collected.

The relationship between sleep and Ad pathology in individuals with the apolipoprotein E4 (APOE4) allele (the E4 allele being the largest contributor to the genetic risk for late-onset Ad) also warrants further investigation. It is well established that APOE4 carriers tend to exhibit higher levels of Aβ and tau pathology than non-carriers.²⁹ APOE4 carrier status has also been associated with objective³⁰ and subjective³¹ sleep disturbances and with sleep disorders.^32,33 Importantly, recent evidence suggests that the negative impact of APOE4 carrier status on sleep occurs through mechanisms independent of Ad pathology.³¹ This supports the potential utility of behavioural sleep interventions to decrease Ad pathology in APOE4 carriers. In line with this idea, one prior study found evidence for an attenuating effect of sleep consolidation on tau but not Aβ pathology (measured post-mortem) in APOE4 carriers.³⁰ However, the relationship between objectively measured sleep (means and variability) and Aβ and tau pathology measured in vivo in APOE4 carriers is unexplored.

In sum, research is needed to elucidate the associations between both averages and variability of objectively measured sleep and outcome measures of in vivoAd pathology and cognition and to determine whether sleep may protect individuals at increased genetic risk for Ad (i.e., APOE4 carriers).

The current study sought to address these questions by investigating relationships between objectively measured sleep duration and sleep efficiency (averages and variability), Aβ positron emission tomography (PET), tau PET and cognition in a cohort of older adults with objective early mild cognitive impairment (MCI). Modifying effects of APOE4 status on these relationships were also explored. We hypothesized that individuals with higher sleep efficiency and less variability in sleep measures would have lower levels of Aβ and tau burden and perform better on neuropsychological tests of global cognition, memory and executive function. Additionally, we hypothesized that APOE4 status would modify the relationships between sleep quality, quantity and Ad pathology.

Materials and methods

Participants

This study included data from a sample of 52 older adults with objective early MCI (67% female, mean age = 66.4, age range = 55–80), characterized by performance at least one standard deviation below normative levels on tests of attention, executive function or memory. The study included all participants who had actigraphy, neuropsychological and brain imaging data available. Participants were recruited from the Greater Los Angeles Area and had no current or prior history of any major psychiatric illness, organ failure, epilepsy or hydrocephalus. The study was performed in line with the principles of the Declaration of Helsinki. Study procedures were approved by the Institutional Review Board at USC, and informed consent was obtained from all participants.

Actigraphy sleep measures

Following an initial screening visit, participants were given a physical activity monitor (GENEActiv), which was worn for approximately 1 month (avg = 30.86 days) prior to brain imaging and neuropsychological testing. Data from the accelerometer was processed using GGIR version 1.8, an R-package used to analyse movement data, including sleep metrics.^34,35 A heuristic algorithm designed to estimate sleep in the absence of a sleep diary was used to detect the sleep period time window (SPT window), defined as the time elapsed between the start of sleep onset and waking time, for each participant.³⁶ In our study, sleep duration was defined as the total time asleep within a given SPT window, and sleep efficiency was defined as the percentage of time asleep within a given SPT window. Intra-individual variability in sleep duration and sleep efficiency were characterized using an intra-individual coefficient of variation, calculated as standard deviation/mean × 100. This quantification adjusts for the increased likelihood of greater variability in individuals with higher mean scores and has been described as the preferred measure of variability in sleep research.^37–40

Neuroimaging measures

MRI acquisition

MR images were acquired on a Siemens 3 T Prisma scanner. The following parameters were used to obtain structural T1-weighted (T1w) MPRAGE images: repetition time (TR)/echo time (TE), 2400/2.2 ms; field of view (FOV), 176 × 240 × 256 mm; and resolution, 1.0 mm³ isotropic.

PET image acquisition

Each participant received PET scans with Neuraceq [florbetaben F 18 (FBB)] to measure Aβ and [¹⁸F]flortaucipir (FTP) to measure tau. IV injections of the tracers (FBB, −8.11 6 0.6 mCi; FTP, 10.6 ± 1.2 mCi) were done outside of the scanning room. For FBB scans, 4 × 5-min frames were acquired beginning 90 min after injection. For FTP scans, 6 × 5-min frames were acquired beginning 75 min after injection. Further details on PET image acquisition have been published elsewhere.⁴¹

Data processing

An automated in-house processing pipeline, the details of which have been published previously,⁴¹ was used for all data processing. A general description of the processing pipeline is described below.

T1 MR templates were created from T1w scans using the Advanced Normalization Tools (ANTs) package template building tool.^42,43 FreeSurfer version 6.0.0 (https://surfer.nmr.mgh.harvard.edu/) was used to generate subject-specific regions of interest (ROIs).

Each frame of dynamic PET images was aligned to an average image to correct for motion. Motion-corrected frames were then averaged and co-registered to the T1w image. They were next moved into template space and smoothed with an 8-mm Gaussian kernel. Inferior cerebellar grey matter was used as a reference region for FTP scans.⁴⁴ The whole cerebellum was used as a reference region for FBB scans. The PET signal in each voxel was divided by the average signal in the respective reference region to create standardized uptake value ratio (SUVR) images. Composite FTP scores consisted of previously defined brain regions consistent with Braak stages I–VI.⁴⁴ Composite FBB scores consisted of frontal, parietal, lateral temporal and cingulate cortices.⁴⁵

Neuropsychological measures

All participants underwent a 2-hour neuropsychological battery of tests designed to measure working memory, inhibitory control, verbal and visual memory and global cognition. The battery was administered by a trained research associate. The cognitive variables for the current study included performance on Letter–Number Sequencing (LNS) as a measure of working memory;⁴⁶ the flanker task as a measure of inhibitory control;⁴⁷ the California Verbal Learning Test, Second Edition (CVLT) as a measure of verbal memory;⁴⁸ the Complex Figure (CF) test as a measure of visual memory;⁴⁹ and the Montreal Cognitive Assessment (MoCA) as a measure of global cognition.⁵⁰ For the flanker task, inhibitory control was characterized as the difference in reaction time for the congruent and incongruent trials, with lower scores indicating better performance. For the CVLT, verbal memory was characterized as the total long delay free recall score. For the CF, visual memory was characterized as the total score on the long delay trial.

Statistical analyses

Multivariable linear regression analyses were conducted to examine associations between sleep measures, PET SUVR and neuropsychological test scores. All models were adjusted for age, sex and APOE4 status. Models with cognition as an outcome included education as a covariate. Two participants with missing measures of visual memory were excluded from models with visual memory as an outcome variable. One participant with a missing measure of inhibitory control was excluded from models with inhibitory control as an outcome variable.

To explore modifying effects of APOE4 status on these relationships, multivariable linear regressions including an interaction term for the sleep variables and APOE4 status were conducted for each of the outcome measures. These models were adjusted for the same covariates described above. All analyses were run in R [R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL: https://www.R-project.org/], with a P-value < 0.05 indicating statistical significance. Regression assumptions were checked through visual inspection. Participants were assumed to be independent from one another a priori.

Exploratory, post hoc analyses were conducted to investigate potential mediating effects of Aβ and tau on relationships between sleep and cognition. In univariate models, significant associations between sleep measures, Aβ and tau burden and cognitive outcome variables (a, b and c’ pathways of mediation analysis) were identified using Pearson correlations. If associations between Aβ or tau approached significance (P < 0.10) with both a sleep variable and a cognitive outcome variable, mediation models were tested. All mediation models were conducted in R using the lavaan package.⁵¹ Models were further adjusted for age, sex, APOE4 status and education. To assess reverse causality, we interchanged the exposure and mediator variables and tested whether the sleep variable in our mediation models mediated the effect of Ad pathology on the cognitive outcome variable.

Exploratory, post hoc analyses were also conducted to investigate relationships between variability in sleep and Aβ burden in individual ROIs (frontal, parietal, lateral temporal and cingulate cortices) of the overall composite FBB measure.

Results

Sample characteristics

Participant characteristics are summarized in Table 1. The entire sample consisted of 52 older adults (67% female, mean age = 66.4, age range = 55–80). There were 38 APOE4 carriers and 14 non-carriers. Sample characteristics of carriers versus non-carriers did not differ significantly, with the exception of sex and ethnicity. APOE4 carriers had a greater proportion of females and Hispanics compared to non-carriers. One participant did not have a measure of inhibitory control available, and two participants did not have measures of visual memory available. These missing data points were due to technical issues (inhibitory control) and non-completion (visual memory) during testing.

Table 1.

Participant characteristics

Sample characteristics (N = 52)a
	Entire sample (N = 52)	APOE4 carriers (N = 14)	APOE4 non-carriers (N = 38)	Carriers versus non-carriers (P-value)b
Female (%)	67%	85%	61%	P = 0.05
Age, M (SD)	66.4 (6.89)	64.5 (7.57)	67.11 (6.58)	P = 0.27
Years education, M (SD)	17.25 (2.42)	17.39 (1.9)	16.33 (4.67)	P = 0.63
APOE4 carriers (%)	27%	–	–
MoCA, M (SD)	26.08 (2.74)	26.64 (2.41)	25.87 (2.85)	P = 0.34
Sleep efficiency (percentile), M (SD)	86% (6%)	85% (6%)	86% (5%)	P = 0.62
Sleep duration (h/day), M (SD)	6.68 (1.07)	6.23 (1.08)	6.85 (2.85)	P = 0.08
Sleep efficiency variability, M (SD)	7.16 (14.7)	7.05 (2.23)	7.2 (2.34)	P = 0.83
Sleep duration Variability, M (SD)	20.78 (7.07)	22.53 (6.49)	20.13 (7.25)	P = 0.53
Ethnicity
Caucasian, % (n)	79% (41)	71% (10)	82% (31)	P = 0.43
African American, % (n)	10% (5)	0% (0)	13% (5)	P = 0.15
Hispanic, % (n)	12% (5 Mexican, 1 Cuban)	29% (4)	5% (2)	P = 0.02
Asian American, % (n)	6% (3)	7% (1)	5% (2)	P = 0.80
Unknown, % (n)	5% (3)	21% (3)	0% (0)	P < 0.01
Medical/health history
Body mass index, M (SD)	27.34 (4.83)	28.56 (4.73)	26.89 (4.85)	P = 0.27
Treated sleep apnoea, % (n)	4% (3)	7% (1)	5% (2)	P = 0.82
Psychiatric history, % (n)	32% (17)	36% (n = 5)	32% (12)	P = 0.79
Blood pressure medication, % (n)	29% (15)	14% (2)	34% (13)	P = 0.12
Anti-depressants, % (n)	15% (8)	14% (2)	16% (6)	P = 0.90
Sleep medication, % (n)	2% (1)	7% (1)	0%	P = 0.34

^aMeans and standard deviations or percentiles and counts are shown.

^bDifferences between carriers and non-carriers were assessed using t-tests for continuous variables and chi-square tests for categorical variables.

Variability in sleep

Variability in sleep duration and sleep efficiency were positively correlated (r = 0.37, P = 0.01).

Variability in sleep duration

Less variability in sleep duration was associated with lower Aβ burden (b = 0.01, SE = 0.003, P = 0.04, Fig. 1A), higher MoCA scores (b = −0.14, SE = 0.06, P = 0.02, Fig. 1B) and better inhibitory control (b = 0.01, SE = 0.003, P = 0.04, Fig. 1C). There was a trend towards an association between less variability in sleep duration and lower tau burden (b = 0.003, SE = 0.002, P = 0.07). No significant associations were observed between variability in sleep duration and working, verbal or visual memory. Exploratory analyses revealed significant relationships between less variability in sleep duration and lower Aβ burden in all ROIs (all P < 0.05) with the exception of the lateral temporal region (P = 0.07).

Figure 1.

Associations between sleep measures, composite Aβ and cognitive variables. (A) Less variability in sleep duration is associated with lower Aβ burden (b = 0.01, SE = 0.003, t = 2.12, P = 0.04). (B) Less variability in sleep duration is associated with higher MoCA scores (b = −0.14, SE = 0.06, t = −2.48, P = 0.02). (C) Less variability in sleep duration is associated with better inhibitory control (higher scores = worse performance) (b = 0.01, SE = 0.003, t = 2.1, P = 0.04). (D) Less variability in sleep efficiency is associated with lower Aβ burden (b = 0.03, SE = 0.01, t = 3.4, P < 0.01). (E) Less variability in sleep efficiency is associated with higher MoCA scores (b = −0.44, SE = 0.17, t = −2.81, P = 0.01). (F) Less variability in sleep efficiency is associated with better inhibitory control (b = 0.02, SE = 0.01, t = 2.31, P = 0.05). (G) Longer sleep duration is associated with better visual memory (b = 1.20, SE = 0.52, t = 2.37, P = 0.02). (H) Longer sleep duration is associated with better inhibitory control (b = −0.06, SE = 0.02, t = −2.88, P < 0.01). IIV: intra-individual variability (standard deviation/mean × 100). SUVR: standardized uptake value ratio.

Variability in sleep efficiency

Less variability in sleep efficiency was associated with lower Aβ burden (b = 0.03, SE = 0.01, P < 0.01, Fig. 1D), higher MoCA scores (b = −0.44, SE = 0.17, P = 0.01, Fig. 1E) and better inhibitory control (b = 0.02, SE = 0.01, P = 0.05, Fig. 1F). No significant associations were observed between variability in sleep efficiency and working, verbal or visual memory or between variability in sleep efficiency and tau burden. Exploratory analyses revealed significant relationships between less variability in sleep efficiency and lower Aβ burden in all ROIs (all P < 0.05).

Mean levels of sleep

Means levels of sleep duration and sleep efficiency were not significantly correlated (r = 0.16, P = 0.24).

Mean sleep duration

Longer sleep duration was significantly associated with better visual memory (b = 1.20, SE = 0.52, P = 0.02, Fig. 1G) and better inhibitory control (b = −0.06, SE = 0.02, P < 0.01, Fig. 1H), but not with verbal memory, working memory or global cognition. No significant associations were observed between mean sleep duration and Aβ or tau burden.

Mean sleep efficiency

No significant associations were observed between mean levels of sleep efficiency and Aβ or tau burden or any cognitive outcome variables.

Modifying effects of APOE4 status

There was a significant modifying effect of APOE4 status, such that lower sleep efficiency variability was associated with lower Aβ burden in APOE4 carriers only (b = 0.08, SE = 0.02, P < 0.01, Fig. 2A). There was an interaction between sleep duration and APOE4 status, suggesting that longer sleep duration is more strongly associated with lower Aβ burden in APOE4 carriers relative to non-carriers (b = −0.08, SE = 0.05, P = 0.08, Fig. 2B).

Figure 2.

Interactions between sleep measures and APOE4 status. (A) Less variability in sleep efficiency is associated with lower Aβ burden in APOE4 carriers (b = 0.08, SE = 0.02, t = 5.003, P < 0.01). (B) Longer sleep duration is associated with lower Aβ burden in APOE4 carriers (b = −0.08, SE = 0.05, t = −1.82, P = 0.08). IIV: intra-individual variability (standard deviation/mean × 100). SUVR: standardized uptake value ratio.

Mediation analyses

Exploratory mediation models (Tables 2 and 3) revealed a significant indirect effect (a*b pathway in Fig. 3) of variability in sleep duration on CVLT scores, via tau (b = −0.07, SE = 0.03, P = 0.03), indicating a mediating role of global tau burden on the relationship between variability in sleep duration on verbal memory. With covariates included in the model, the effect was slightly reduced (b = −0.06, SE = 0.03, P = 0.07). There was also a trend towards a significant indirect effect (a*b pathway in Fig. 4) of variability in sleep duration on MoCA scores, via tau (b = −0.04, SE = 0.02, P = 0.06), suggesting that global tau burden may be a mechanism through which variability in sleep duration affects global cognition. However, because the MoCA contains several items related to verbal memory, it is possible that this relationship was driven by performance on memory items. The effect was slightly reduced when adjusting for covariates (b = −0.03, SE = 0.02, P = 0.14). Models to test reverse causality (unadjusted for covariates) revealed no significant indirect effects of tau on CVLT scores, via variability in sleep duration (b = −0.06, SE = 0.22, P = 0.79). There was a trend towards an indirect effect of tau on MoCA scores, via variability in sleep duration (b = −0.37, SE = 0.22, P = 0.09).

Table 2.

Associations between variability in sleep duration and CVLT mediated by global tau burden

Adjusted for covariatesb
Effect	b	SE	95% CIa	P
a. Sleep dur variability → tau	0.29	0.13	(0.03, 0.54)	0.027
b. Tau → CVLT	−0.21	0.07	(−0.33, −0.08)	0.002
c’. Sleep dur variability → CVLT	−0.04	0.07	(−0.17, 0.09)	0.556
a*b. Sleep dur variability → tau → CVLT	−0.06	0.03	(−0.12, 0.01)	0.071
c. Sleep dur variability → CVLT	−0.10	0.08	(−0.25, 0.06)	0.208
Unadjusted for covariates
Effect	b	SE	95% CI	P
a. Sleep dur variability → tau	0.38	0.13	(0.13, 0.62)	0.003
b. Tau → CVLT	−0.18	0.06	(−0.29, −0.07)	0.001
c’. Sleep dur variability → CVLT	−0.02	0.07	(−0.15, 0.12)	0.798
a*b. Sleep dur variability → tau → CVLT	−0.07	0.03	(−0.13, −0.01)	0.025
c. Sleep dur variability → CVLT	−0.09	0.07	(−0.23, 0.06)	0.240

^aCI: Confidence interval.

^bCovariates: Age, sex, APOE4 status and education.

Table 3.

Associations between variability in sleep duration and MoCA mediated by global tau burden

Adjusted for covariatesb
Effect	b	SE	95% CIa	P
a. Sleep dur variability → tau	0.29	0.13	(0.03, 0.54)	0.027
b. Tau → MoCA	−0.11	0.05	(−0.20, −0.02)	0.018
c’. Sleep dur variability → MoCA	−0.11	0.06	(−0.22, 0.02)	0.085
a*b. Sleep dur variability → tau → MoCA	−0.03	0.02	(−0.07, 0.01)	0.136
c. Sleep dur variability → MoCA	−0.14	0.07	(−0.28, 0.00)	0.056
Unadjusted for covariates
Effect	b	SE	95% CI	P
a. Sleep dur variability → tau	0.38	0.13	(0.13, 0.62)	0.003
b. Tau → MoCA	−0.11	0.04	(−0.19, −0.03)	0.010
c’. Sleep dur variability → MoCA	−0.11	0.06	(−0.23, 0.01)	0.072
a*b. Sleep dur variability → tau → MoCA	−0.04	0.02	(−0.08, 0.00)	0.062
c. Sleep dur variability → MoCA	−0.15	0.07	(−0.29, −0.01)	0.034

^aCI: Confidence interval.

^bCovariates: Age, sex, APOE4 status and education.

Figure 3.

Indirect effect of variability in sleep duration on CVLT scores via global tau burden. Results of the exploratory mediation analysis investigating the effect of variability in sleep duration on CVLT scores, via global tau burden. In the model unadjusted for covariates, there was a significant indirect effect (a*b pathway) of variability in sleep duration on CVLT scores, via tau burden. CVLT: California Verbal Learning Test.

Figure 4.

Indirect effect of variability in sleep duration on MoCA scores via global tau burden. Results of the exploratory mediation analysis investigating the effect of variability in sleep duration on MoCA scores, via global tau burden. In the model unadjusted for covariates, there was a trend towards an indirect effect (a*b pathway) of variability in sleep duration on MoCA scores, via tau burden. MoCA: Montreal Cognitive Assessment.

Discussion

The current study suggests that variability in both objectively measured sleep duration and sleep efficiency provides valuable information beyond that of mean-level measures and is associated with Aβ burden and cognition in older adults at risk for Ad. When looking at objective mean measures of sleep, longer sleep duration, but not better sleep efficiency, was significantly associated with better visual memory and inhibitory control. The relationship between sleep variability and Aβ burden differed by APOE4 status, suggesting that more consistent sleep efficiency and longer sleep duration may attenuate the increased risk of Ad conferred by the presence of at least one APOE4 allele.

The objective measurement of sleep is a major strength of this study, as prior research has demonstrated discrepancies between self-reported and objectively measured sleep,^4,21–27 with the greatest differences in individuals with high Aβ burden.⁴ These findings, particularly if extended to larger and more diverse cohorts, have the potential to inform public health recommendations for those at risk for Ad and, importantly, to help clarify whether it is average targets (e.g., 8 h of sleep with 85% sleep efficiency) or specific within-person targets (e.g., remain within 30 min of your typical sleep duration) that may be most beneficial for brain health.⁵²

Variability findings

This study is novel in investigating the associations between variability in objectively measured sleep duration, sleep efficiency and global in vivoAd pathology. One prior cross-sectional study identified associations between day-to-day variability in sleep fragmentation and grey matter volume within the thalamus, and a trend towards an association between variability in sleep fragmentation and Aβ burden in the left rectus gyrus in cognitively normal older adults.⁵³ Our study adds to the sleep and Ad literature by showing that, unlike our average measures of sleep, intra-individual variability in both sleep duration and sleep efficiency was significantly related to Aβ burden, MoCA scores and inhibitory control. These findings suggest that intra-individual variability is capturing unique information that is unappreciated when measures are averaged across days and supports the utility of analysing day-to-day patterns in sleep. Additionally, the positive correlation between variability in sleep duration and sleep efficiency suggests shared causal factors, highlighting the potential for interventions to simultaneously improve consistency in both sleep duration and sleep efficiency.

There are several mechanisms which may explain the relationships observed between intra-individual variability in sleep measures and Aβ burden and cognition. On a cellular level, increased intra-individual variability in sleep efficiency and duration may reflect disruption in circadian rhythms, which are typically synchronized to an approximately 24-h cycle and help to regulate physiological processes such as the sleep–wake cycle. Prior work has identified increased disruption of circadian rhythms in individuals with neurodegenerative disease,⁵⁴ and disruptions in circadian rhythms between days (i.e., increased day-to-day variability in the 24-h activity/rest signal) have been associated with increased risk of conversion to MCI and Ad.⁵⁵ Therefore, increased intra-individual variability in sleep measures may be driven by disrupted circadian functioning (e.g., disruption in the production and timing of melatonin release). Notably, however, one prior study that investigated associations between day-to-day circadian rhythms and Ad pathology found that circadian rhythms were ‘more’ consistent across days in amyloid positive individuals.⁵⁶ More research on the associations between circadian rhythm functioning and macroscopic sleep measurements such as variability in sleep duration and sleep efficiency is warranted.

From a macroscopic perspective, variability in sleep efficiency and duration across days may reflect psychological and/or lifestyle factors either associated with or caused by Ad pathology. For example, inconsistent day-to-day sleep patterns may be reflective of a more unstable and unpredictable lifestyle or situational circumstances. Indeed, increased intra-individual variability in sleep duration measured via actigraphy has been associated with a higher number of stressful life events and, in those with high negative affective, higher levels of norepinephrine.⁵⁷ Additionally, increased intra-individual variability in sleep duration has been associated with lower levels of subjective well-being,⁵⁸ suggesting that depressive symptoms and mood disturbances may be associated with variability in sleep patterns. In addition to these psychological factors, increased intra-individual variability in sleep quantity has also been associated with metabolic abnormalities, cardiovascular conditions and poorer self-rated health.^59,60 It remains to be seen whether intra-individual variability in sleep patterns is driven by these micro- and/or macroscopic mechanisms, contributing to them, or whether a third variable (e.g., Aβ burden, neurodegeneration) may be at play. Given the lack of significant indirect effects of intra-individual variability in sleep duration and efficiency on cognition through Aβ burden, variability in sleep may be associated with cognition and Aβ burden via distinct pathways. Interestingly, exploratory mediation analyses in the present study suggested that tau burden may be a mechanism through which variability in sleep duration affects cognition, and thus, the mediating role of Ad pathology on the relationship between sleep and cognition may be more evidenced in later disease states.

Mean-level findings

Neither average sleep duration nor average sleep efficiency was significantly associated with Ad pathology. The lack of association between sleep duration and Ad pathology is consistent with prior research. Prior studies have identified significant relationships between both longer and shorter self-reported sleep duration and Aβ burden,^9,14 suggesting a U-shape in the relationship between sleep quantity and Aβ pathology. Therefore, our lack of significant findings may be due to our investigation of sleep duration as a continuous variable within our entire sample, as opposed to splitting the group into ‘short’, ‘typical’ and ‘long’ sleepers, which would result in small sample sizes in each group.

The lack of significant association between sleep efficiency and Ad pathology was more surprising, as several prior studies have reported associations between decreased objectively measured sleep efficiency and increased levels of Aβ^3,15,16 and tau⁵ pathology. This discrepancy in findings between the current study and prior research may be a result of differences in methodology. The studies cited above used sleep diaries, which allowed for the inclusion of sleep onset latency (the total time between getting into bed to fall asleep and falling asleep) in the quantification of sleep efficiency. Therefore, the current study’s characterization of sleep efficiency (percentage of time asleep within a given SPT window) may reflect different pathophysiological underpinnings.

Longer sleep duration, but not increased sleep efficiency, was significantly associated with better visual memory and inhibitory control. This was surprising given earlier findings showing that objectively measured disturbances in sleep are more related to cognition than total sleep time.⁶¹ However, our study does align with prior findings of associations between short sleep duration, impaired executive functioning and visual memory abilities in individuals with insomnia.⁶² Given the lack of association between sleep duration and Ad pathology in the current study, sleep quantity may influence executive functioning and visual memory processing through mechanisms unrelated to Aβ or tau deposition.

Modifying role of APOE4 findings

Our finding that lower intra-individual variability in sleep efficiency, i.e., more consistent sleep quality from night to night, was associated with lower Aβ burden in APOE4 carriers adds to literature suggesting that consistent sleep quality may attenuate the effects of APOE4 carrier status on Ad pathology. Prior longitudinal work has identified an interaction between APOE4 carrier status and sleep consolidation, in which better sleep consolidation attenuated the relationship between APOE4 carrier status and tau, but not Aβ burden.¹⁷ The difference in these findings may be explained by differences in sleep measures (sleep consolidation versus intra-individual variability in sleep duration and efficiency) and/or differences in neuropathological assessment. In the study by Lim et al.¹⁷ quantification of amyloid and tau was conducted at autopsy when participants were significantly older than those in the current study. Taken together, the findings suggest that sleep quality may attenuate the effects of APOE4 status on Aβ burden earlier in the disease course and on tau burden later in life. Therefore, the window in which sleep quality is important for reduction of Ad risk in those with the genetic vulnerability may be wide.

Limitations

There are several limitations to the current study. Participants in the current study did not complete sleep diaries in addition to wearing the accelerometer. The lack of a sleep diary prevents cross-validation of accelerometer-derived sleep measures, weakening the validity of the sleep measures obtained and limiting our ability to compare findings in the current study to prior work. It is likely that error was introduced when quantifying motionless time in bed versus time asleep. However, past studies have validated sleep data obtained via accelerometer against measures of self-report sleep and polysomnography, providing evidence to support the use of accelerometer-derived sleep data in the absence of sleep diaries.³⁶ The small sample size and demographic characteristics of the current study limits the generalizability of our findings. Given that the current sample included only 14 APOE4 carriers, the modifying effect of APOE4 status on the relationship between variability in sleep efficiency and Aβ burden should be interpreted cautiously. Additionally, it is possible that the observed effect may differ between single versus double APOE4 allele carrier status, which we were unable to reliably measure with the assay used in the current study. Future research investigating associations between variability in sleep measures, in vivoAd pathology and cognition in a larger and more diverse sample is necessary. Investigating the potential modifying effect of APOE4 carrier status and APOE4 homozygosity versus heterozygosity is also important. Finally, the cross-sectional nature of the current study limits our ability to infer causality in any of the relationships we investigated. For example, it is possible that the accumulation of Ad pathology results in sleep impairment, rather than sleep impairment leading to increased Ad pathology. While current evidence suggests a bi-directional relationship between sleep and Ad pathology,^63–66 future work is needed to investigate causal mechanisms.

Conclusion

In conclusion, results from the current study provide evidence that lower variability (i.e., more consistency) in both sleep duration and sleep efficiency across nights is associated with lower levels of Ad pathology and better cognition. In contrast, neither mean sleep duration nor mean sleep efficiency was associated with Ad pathology. Longer mean sleep duration, but not mean sleep efficiency, was associated with better cognition. Findings from the current study also suggest that longer sleep duration and more consistent sleep efficiency across nights may be protective against Aβ burden in APOE4 carriers.

Abbreviations

Ad =

Alzheimer’s disease

APOE4 =

apolipoprotein E4

Aβ =

β-amyloid

CF =

Complex Figure

CVLT =

California Verbal Learning Test

FBB =

florbetaben F 18

FTP =

[¹⁸F]flortaucipir

LNS =

Letter–Number Sequencing

MCI =

mild cognitive impairment

MoCA =

Montreal Cognitive Assessment

SUVR =

standardized uptake value ratio

Funding

This work was supported by the National Institute of Health (NIH) (PI: Pa, R01AG046928).

Competing interests

The authors have no conflicts of interest to report.

Data availability

The data that support the findings of this study will be made available on request to the corresponding author in accordance with the data-sharing agreement with the National Institute on Aging.