Suboptimal self‐reported sleep efficiency and duration are associated with faster accumulation of brain amyloid beta in cognitively unimpaired older adults

Abstract INTRODUCTION This study investigated whether self‐reported sleep quality is associated with brain amyloid beta (Aβ) accumulation. METHODS Linear mixed effect model analyses were conducted for 189 cognitively unimpaired (CU) older adults (mean ± standard deviation 74.0 ± 6.2; 53.2% female), with baseline self‐reported sleep data, and positron emission tomography‐determined brain Aβ measured over a minimum of three time points (range 33.3–72.7 months). Analyses included random slopes and intercepts, interaction for apolipoprotein E (APOE) ε4 allele status, and time, adjusting for sex and baseline age. RESULTS Sleep duration <6 hours, in APOE ε4 carriers, and sleep efficiency <65%, in the whole sample and APOE ε4 non‐carriers, is associated with faster accumulation of brain Aβ. DISCUSSION These findings suggest a role for self‐reported suboptimal sleep efficiency and duration in the accumulation of Alzheimer's disease (AD) neuropathology in CU individuals. Additionally, poor sleep efficiency represents a potential route via which individuals at lower genetic risk may progress to preclinical AD. Highlights In cognitively unimpaired older adults self‐report sleep is associated with brain amyloid beta (Aβ) accumulation. Across sleep characteristics, this relationship differs by apolipoprotein E (APOE) genotype. Sleep duration <6 hours is associated with faster brain Aβ accumulation in APOE ε4 carriers. Sleep efficiency < 65% is associated with faster brain Aβ accumulation in APOE ε4 non‐carriers. Personalized sleep interventions should be studied for potential to slow Aβ accumulation.

• Across sleep characteristics, this relationship differs by apolipoprotein E (APOE) genotype.
• Sleep duration <6 hours is associated with faster brain Aβ accumulation in APOE ε4 carriers.
• Sleep efficiency < 65% is associated with faster brain Aβ accumulation in APOE ε4 non-carriers.
• Personalized sleep interventions should be studied for potential to slow Aβ accumulation.

BACKGROUND
][12][14][15][16][17] While cognitive impairment and dementia due to AD pathology are typically diagnosed in later life, the accumulation of brain Aβ begins decades earlier while individuals are cognitively unimpaired (CU). 180][21][22][23][24] However, only one longitudinal study has investigated the capacity of poor night-time sleep characteristics to predict brain Aβ dynamics.The authors of this longitudinal study of CU older adults (N = 32) reported that objectively measured sleep efficiency and proportion of slow wave sleep, below 1 Hz, predicted the subsequent rate of brain Aβ accumulation. 4Characteristics of sleep quality may therefore be viewed as a prospective biomarker of brain Aβ accumulation while also representing a potentially modifiable risk factor for AD.However, to our knowledge, no study to date has investigated whether self-reported sleep quality characteristics are associated with brain Aβ accumulation, a knowledge gap the current study addresses.6][27] CU older adults represent a group with intact cognition who are therefore better able to accurately self-report sleep quality, and ultimately represent a potential target population for sleep improvement interventions aimed at preventing the onset of cognitive decline due to AD.
The apolipoprotein E (APOE) ε4 allele is a strong genetic risk factor for brain Aβ deposition, influencing Aβ clearance and aggregation. 28,29 AD patients, APOE ε4 non-carriage was found to be associated with greater deterioration in sleep quality with advancing age. 30Moreover, a small body of cross-sectional research suggests that APOE genotype potentially interacts with sleep quality to determine risk of AD-related pathological change (including Aβ pathology, tau pathology, and gray matter volume), though the evidence is inconsistent, potentially due to the heterogeneity of study samples and methodologies.

Study population
This

Sleep measures
The PSQI, a 19-item, valid, reliable measure of self-reported sleep quality over the past month, frequently adopted in aging research, was used to derive self-report sleep characteristics for analysis. 25

Brain imaging
9][40] The CapAIBL quantification algorithm was used to calculate a Centiloid (CL) value for each PET image, providing a single continuous variable representing brain Aβ burden. 41For each participant included in the current analysis, brain Aβ PET imaging was conducted at a minimum three AIBL timepoints.

APOE genotyping
Participants donated fasted blood samples, from which DNA was extracted and analyzed to determine APOE genotype, per standard protocols described elsewhere. 36APOE ε4 allele carrier status (carrier or non-carrier) was included in data analysis.

Statistical analysis
Statistical analyses were carried out using R version 4.1.0.Baseline demographics, brain imaging, and sleep characteristics were determined, and between-group differences assessed using independent samples t tests (continuous variables), chi-square (χ 2 ) analysis (categorical variables), or the Kruskal-Wallis test (ordinal variable and median difference; Table 1).A linear mixed effects model (LMM), including ran-

RESULTS
Baseline demographic, brain imaging, and sleep characteristics for AIBL participants (N = 189) included in the current study are presented in

Sleep duration and rate of brain Aβ accumulation
As presented in Table 2, for the whole cohort (β = −0.67 ± 0.26, P = 0.012) and for APOE ε4 non-carriers (β = −0.66 ± 0.25, P = 0.010), longer total sleep time was nominally associated with slower brain Aβ accumulation over time, after Bonferroni correction, whereby for each hour of sleep, 0.67 fewer CL units of amyloid were accumulated per year.At baseline, longer sleep duration in APOE ε4 carriers was nominally associated with lower brain Aβ burden (β = −9.17± 3.57, P = 0.011).To determine whether the relationship between sleep duration and brain Aβ accumulation reflected a U-shaped curve, as is reported for sleep and cognition, we analyzed a categorical sleep duration variable (see Table 2, Table S2 in supporting information, and Figure 2).For APOE ε4 carriers (β = 4.33 ± 1.58, P = 0.007), but not for the whole cohort or APOE ε4 non-carriers, baseline sleep duration of < 6 hours was associated with a faster rate of brain Aβ accumulation compared to optimal sleep duration of 6 to 8 hours (see Table 2 and Table S2).Figure 2 demonstrates that APOE ε4 carriers with < 6 hours of sleep accumulated approximately 23% more brain Aβ over a 6-year period than ε4 carriers with 6 to 8 hours of sleep.Inspection of the y intercepts of Figure 2 show the absence of a U-shaped curve for the relationship between sleep duration and brain Aβ.Instead, this figure shows that in both APOE ε4 non-carriers and carriers, those with > 8 hours of sleep had a lower baseline brain Aβ burden compared to groups with shorter sleep duration, while those with < 6 hours of sleep presented at baseline with higher brain Aβ load, though these main effects were not significant.

Sleep efficiency and rate of brain Aβ accumulation
As shown in Table 2, after Bonferroni correction, higher baseline sleep efficiency was nominally associated with slower brain Aβ accumulation over time, within the whole cohort (β = −0.05± 0.02, P = 0.025), and in APOE ε4 non-carriers (β = −0.05± 0.02, P = 0.021), but not in APOE ε4 carriers.Analysis of categorical sleep efficiency (see Table 2 and Figure 3) revealed that baseline sleep efficiency of < 65% was associated with faster brain Aβ accumulation over time relative to those with sleep efficiency > 85%, in the whole cohort (β = 2.91 ± 0.93, P = 0.002), and in APOE ε4 non-carriers (β = 2.86 ± 0.89, P = 0.001), but not in APOE ε4 carriers (see Table S2).Figure 3 shows that those APOE ε4 noncarriers with sleep efficiency < 65% accumulated ~57% more brain Aβ over 6 years compared to APOE ε4 non-carriers with sleep efficiency of > 85%.Additionally, visual inspection of the y intercepts in Figure 3 reveals that, at baseline, both APOE ε4 non-carriers and carriers with < 65% sleep efficiency had a higher brain Aβ burden than those with sleep efficiency of 65% to 85%, or > 85%, though these main effects were not significant.

Self-reported sleep quality category and rate of brain Aβ accumulation
For APOE ε4 carriers (β = 3.42 ± 1.64, P = 0.038), but not for the whole cohort or APOE ε4 non-carriers, self-reported "poor" sleep quality was nominally associated with a faster rate of brain Aβ accumulation

F I G U R E 3
Plots for the relationship between baseline sleep efficiency category and the trajectory of brain Aβ burden, in cognitively unimpaired APOE ε4 allele non-carriers (left) and carriers (right).Brain Aβ burden is presented as Centiloid value (95% Winsorized).Sleep efficiency category was derived from PSQI-determined sleep efficiency, which represents the percentage of time in bed spent asleep, with < 65% sleep efficiency categorized as "very poor", 65% to 85% sleep efficiency as "suboptimal", and > 85% sleep efficiency as "optimal".The horizontal line, at a Centiloid value of 20, represents the threshold for high brain Aβ burden.Colored bands represent 95% confidence intervals.Sleep efficiency category < 65% was associated with greater brain Aβ burden at baseline and a steeper slope of brain Aβ accumulation in APOE ε4 allele non-carriers.Comparison of the baseline Centiloid value (y intercept) and final Centiloid value for APOE ε4 non-carriers with < 65% sleep efficiency and those with > 85% sleep efficiency indicated that those with < 65% sleep efficiency accumulated ~57% more brain Aβ over time.Aβ, amyloid beta; APOE, apolipoprotein E; PSQI, Pittsburgh Sleep Quality Index.compared to those reporting "very good" sleep quality (see Table S2 and Figure S1 in supporting information), after Bonferroni correction.
Moreover, the y intercepts of Figure S1 suggest that APOE ε4 carriers reporting "poor" sleep quality have greater baseline brain Aβ burden compared to APOE ε4 carriers with "very good" or "fairly good" selfreported sleep quality, though these main effects were not significant.

Sleep onset latency and rate of brain Aβ accumulation
Analysis of sleep onset latency variables, both continuous and categorical, revealed no significant relationships, for any sub-group comparison, with respect to rate of brain Aβ accumulation (see Table S2 and Figure S2 in supporting information).However, Figure S2 clearly shows that APOE ε4 carriers with sleep onset latency of ≥ 60 minutes had higher brain Aβ levels at both baseline, and over time, compared to APOE ε4 carriers with sleep onset latency < 60 minutes, though the rate of brain Aβ accumulation did not differ.The main effect for sleep onset latency of ≥ 60 minutes (compared to < 30 minutes) was approaching nominal significance (β = 46.77± 24.61, P = 0.059).

Sleep disturbance score and rate of brain Aβ accumulation
After Bonferroni correction for multiple comparisons, a nominally significant three-way interaction was observed among moderate/severe sleep disturbance, APOE ε4 carriage, and time (β = 2.35 ± 1.17, P = 0.046), with respect to rate of brain Aβ accumulation.While a trend toward nominal significance was observed in APOE ε4 carriers with moderate/severe sleep disturbance (compared to mild sleep disturbance; β = 1.96 ± 1.06, P = 0.065), sleep disturbance score produced no significant results for any sub-group with respect to rate of brain Aβ accumulation (see Table S2).

PSQI global score and rate of brain Aβ accumulation
Analyses of the PSQI global score (quantitative indicator of overall sleep quality based on the sum of PSQI component scores) produced no significant results, for any sub-group, with respect to rate of brain Aβ accumulation (see Table S2).

DISCUSSION
The A longitudinal analysis conducted by Blackman et al. 43 using PSQI component scores and cerebrospinal fluid (CSF) measures of Aβ did not find any relationship between short sleep duration and poor sleep efficiency, and changes in CSF Aβ over time.However, this study was of shorter duration (follow-up 1.5 years ± 0.5) and > 80% of participants provided only one sample of CSF. 43Nevertheless, the authors did find that greater sleep disturbance was associated with faster decline in CSF Aβ42 levels (proposed to represent faster brain Aβ deposition). 43 the current study, no relationship between sleep disturbance and rate of brain Aβ accumulation was observed, although a trend toward significance was noted in APOE ε4 carriers.
While no prior longitudinal study has considered the relationship between self-reported night-time sleep quality and accumulation of brain Aβ, Carvalho et al. 44 reported that individuals self-reporting excessive daytime sleepiness (EDS) demonstrated faster accumulation of brain Aβ compared to those without EDS.This is consistent with Spira et al., 45 who reported that those with EDS were 2.75 times more likely to have high brain Aβ at follow-up (15.7 years ± 3.4).While these studies did not investigate night-time sleep duration or sleep efficiency, short and/or inefficient night-time sleep may contribute to the EDS reported.
While past cross-sectional analyses, using PET imaging, have reported a relationship between higher brain Aβ burden and both shorter self-reported sleep duration 6,21 and poorer objectively measured sleep efficiency, 20  The novel findings of this study, conducted within a large sexbalanced cohort of older adults, suggest that markers of poor sleep measured by the PSQI (an accessible, cost-effective, simple-toadminister tool), may act as biomarkers for brain Aβ accumulation.
Nevertheless, these findings are not without limitations.Specifically, collection of baseline sleep data and brain Aβ PET imaging were not completed on the same day.However, the slow process of brain Aβ deposition occurs over many years, 50 while sleep habits are typically chronic in the age group studied.The protracted process of brain Aβ deposition may also mean that the median follow-up time of 4.3 years was insufficient to observe an effect for all sleep variables.Additionally, while diagnosed obstructive sleep apnea (OSA) represents an exclusion criterion for the AIBL Study, it is possible that some participants have undiagnosed OSA: this is relevant as OSA has previously been associated with higher brain Aβ burden. 51Finally, the current study used a self-report sleep measure, namely the PSQI.While the study included participants who were CU at the time of sleep assessment to circumvent recall bias due to objective memory impairment, selfreport sleep measures still rely on the accuracy of the participant, and, unlike polysomnography, do not provide important information on sleep microarchitecture.Nonetheless, the PSQI has been widely adopted and found to be a valid, reliable measure with good internal consistency for use in older adults, while offering the benefit of greater clinical utility and cost effectiveness. 26,27 summary, our findings demonstrate that self-reported sleep duration < 6 hours in APOE ε4 carriers, and sleep efficiency < 65% in the whole cohort and APOE ε4 non-carriers, was associated with faster accumulation of brain Aβ.This study requires validation in independent samples, and extension to mid-life cohorts to further elucidate these relationships.Future intervention studies, using both objective and self-report sleep measures are required to determine whether improvements in poor sleep lead to a slowing of the rate of brain Aβ longitudinal investigation (up to 6 years' follow-up) used data from 189 participants drawn from the AIBL Study of Ageing.Included participants (see Figure 1) were aged ≥ 60 years at study enrollment, were classified as CU at baseline, and underwent brain Aβ PET imaging at baseline then at a minimum two further AIBL timepoints (AIBL assessments occur at 18-month intervals; for the current sample, median 3.0 PET scans, median absolute deviation 0.0).Included individuals had also completed the PSQI 34 within 18 months of their baseline PET scan.Data were collected from July 2012 to July 2019.The AIBL Study, including recruitment methods, screening criteria, and diagnostic classification procedures, has been described in detail elsewhere.

F I G U R E 1
Participant selection flowchart.AIBL, Australian Imaging, Biomarkers and Lifestyle Study of Ageing; PET, positron emission tomography; PSQI, Pittsburgh Sleep Quality Index.
dom slopes and intercepts, interaction for APOE ε4 allele status, and time, adjusting for sex and baseline age (confirmed by model fit analysis as appropriate), was used to model change in brain Aβ over a minimum of three PET scan collections, as predicted by each sleep parameter.Given the previous findings that brain Aβ accumulation begins ~15 years earlier for APOE ε4 carriers 42 compared to non-carriers, we tested the LMM framework for both a two-way interaction with APOE ε4 carriage and time, and a three-way interaction among sleep, APOE ε4 carriage, and time, with brain Aβ the dependent variable.A significant interaction for APOE ε4 carriage and time (P = 0.001) and a trend toward significance for the three-way interaction (moder-ate/severe sleep disturbance, P = 0.046, sleep duration category < 6 hours, P = 0.085, and sleep efficiency category 65%-85%, P = 0.070) confirmed modeling stratified for APOE ε4 carriage was warranted.This model included random slopes, random intercepts, interaction effects for time, and APOE ε4 allele carriage, to model against the 95% Winsorized CL value.Covariates excluded through this process were an age-sex interaction, cardiovascular risk score, body mass index (BMI), depressive symptomology (Geriatric Depression Scale [GDS]), education category (≤ 12 years/> 12 years) and the use of sleep-related medications.Statistical significance was set at P < 0.05, with a Bonferroni-corrected threshold of 0.008 (0.05/6) applied to account for multiple comparisons.Further statistical methodology can be found in the Statistical Methods section in supporting information.

F I G U R E 2
Plots for the relationship between baseline sleep duration category and the trajectory of brain Aβ burden, in cognitively unimpaired APOE ε4 allele non-carriers (left) and carriers (right).Brain Aβ burden is presented as Centiloid value (95% Winsorized).Sleep duration category was derived from PSQI Question 4, self-report hours of actual sleep, with < 6 hours of sleep duration categorized as "short", 6 to 8 hours of sleep duration as "normal", and >8 hours of sleep duration as "long".The horizontal line, at a Centiloid value of 20, represents the threshold for high brain Aβ burden.Colored bands represent 95% confidence intervals.Sleep duration category < 6 hours was associated with greater brain Aβ burden at baseline and a steeper slope of brain Aβ accumulation in APOE ε4 allele carriers.Comparison of the baseline Centiloid value (y intercept) and final Centiloid value for APOE ε4 allele carriers with < 6 hours of sleep and those with 6 to 8 hours of sleep indicated that those with < 6 hours of sleep accumulated ~23% more brain Aβ over time.Aβ, amyloid beta; APOE, apolipoprotein E; PSQI, Pittsburgh Sleep Quality Index.

Table 1 .
There were no significant differences for demographic factors Baseline demographic, brain imaging, and sleep characteristics for the cognitively unimpaired sample who completed the PSQI at baseline and underwent Aβ PET imaging at a minimum three AIBL study timepoints.Results of linear mixed models examining the associations among baseline total sleep time, sleep duration category, sleep efficiency, sleep efficiency category, and brain Aβ burden over time.Models include PSQI-derived sleep measure, APOE ε4 allele carrier status (±), time from baseline PET scan until final PET scan, baseline age, and sex as main effects, and three-way sleep x APOE ε4 status x Centiloid interaction.Beta coefficients (β) ± SE from the LMM are shown, with bold indicating nominal significance (P < 0.05) and # indicating significance after Bonferroni correction (P < 0.008).Brain Aβ burden in Centiloid units is derived from PET scans, conducted at a minimum three time points, with a median follow-up period of 51.55 months (MAD = 13.01).Sleep duration category, derived from PSQI Question 4, self-report hours of actual sleep, with < 6 hours of sleep duration categorized as "short," 6-8 hours of sleep duration as "normal," and > 8 hours of sleep duration as "long."Sleep efficiency category, derived from PSQI-determined sleep efficiency, which represents the percentage of time in bed spent asleep, with < 65% sleep efficiency categorized as "very poor," 65% to 85% sleep efficiency as "suboptimal," and > 85% sleep efficiency as "optimal."Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E gene; LMM, linear mixed model; MAD, median absolute deviation; PET, positron emission tomography; PSQI, Pittsburgh Sleep Quality Index; SE, standard error.
between APOE ε4 non-carriers and carriers.As expected, compared to non-carriers, APOE ε4 carriers had higher brain Aβ at baseline (mean difference = 21.4CL, P < 0.001), and a greater proportion (% difference = 30.4) of individuals with high brain Aβ burden (CL ≥ 20; P < 0.001).APOE ε4 non-carriers reported shorter sleep duration compared to APOE ε4 carriers (mean difference = 24 minutes, P = 0.021), TA B L E 1 a p-values for independent sample t tests, χ 2 tests or Kruskal-Wallis tests comparing APOE ε4 non-carriers and carriers.b BMI was calculated as weight in kilograms divided by height in meters squared.c Baseline to final PET scan interval.d Threshold for high brain Aβ burden is ≥ 20 Centiloid.e Good sleepers were defined based on a PSQI global score of ≤ 5. f PSQI sleep medication component score: "During the past month, how often have you taken medicine (prescribed or 'over the counter') to help you sleep?" TA B L E 2 a Compared to 6-8 hours of ("normal") sleep duration.b Compared to > 85% ("optimal") sleep efficiency.and when participants were categorized as either "good" or "poor" sleepers (poor sleeper = PSQI global score > 5), there was a greater proportion (18.9%) of "poor" sleepers (P = 0.022) in the APOE ε4 non-carrier group compared to APOE ε4 carriers.
current study investigated the relationship between self-reported sleep characteristics and rate of brain Aβ accumulation, over up to 6 years, in CU older adults.It is the first study to demonstrate an asso- 507,20ngitudinal analyses showed that sleep duration and sleep efficiency as continuous variables were not significantly associated with brain Aβ accumulation, after correction for multiple comparison, suggesting that this relationship is more complex.Winer et al.4did not consider the effect of APOE ε4 carriage on the relationship between sleep and brain Aβ accumulation.In line with our expectations that the effects of poor sleep would be magnified in carriers of the APOE ε4 allele, we found that sleep duration < 6 hours was associated with a positive slope of brain Aβ accumulation in APOE ε4 carriers, but not in the whole sample or in APOE ε4 non-carriers.However, counter to our expectations, we observed a significant relationship between poor sleep efficiency (< 65%) and the subsequent positive slope of brain Aβ accumulation in APOE ε4 non-carriers, but not in carriers.Past research reporting the effect of APOE genotype on the relationship between sleep and brain Aβ burden has been scant and inconsistent.Hwang et al.33found, in cross-sectional analysis of CU older adults from Korea, that APOE ε4 carriage moderated the relationship between the sleep-wake cycle and brain Aβ deposition.carriers, this relationship did not survive correction for multiple comparisons.Additionally, we found that PSQI global score, a measure of overall sleep quality, while highly correlated with sleep duration and efficiency, was not associated with the rate of brain Aβ accumulation in any group.Moreover, while prior research has reported a crosssectional relationship between longer sleep onset latency and higher brain Aβ burden, we found that in the current study sleep onset latency was not associated with the rate of brain Aβ accumulation.3,7,20However,consistentwithpastresearch,longer sleep onset latency was associated with higher brain Aβ at baseline (and higher Aβ burden over time in the current study), compared to shorter sleep onset latency, particularly in the APOE ε4 carrier group (FigureS2), though the main effects for these results did not survive correction for multiple comparisons.Given that brain Aβ accumulation is a protracted process50it is possible that the duration of the current study was insufficient to observe associations among PSQI global score, self-reported sleep quality, and sleep onset latency and rate of brain Aβ increase.It is also conceivable that the characteristics of longer sleep onset latency and poor overall sleep quality make a greater contribution to the rate of brain Aβ accumulation prior to older adulthood.Future research is warranted to address these knowledge gaps.
6,7,19ε4 non-carriers) can progress to high brain Aβ burden and subsequent increased risk of AD.Moreover, APOE ε4 non-carriers with < 6 hours' sleep duration, or APOE ε4 carriers with < 65% sleep efficiency (Figures2 and 3) had greater brain Aβ load at baseline, which remained higher at follow-up, though the rate of accumulation did not differ across sleep duration or efficiency categories.Future research is required to understand at which stage of life this difference in brain Aβ burden begins to manifest.Such information may inform when sleep interventions aimed at improving short sleep duration or poor sleep efficiency, ideally targeted to APOE genotype, may be of the greatest benefit with regard to slowing brain Aβ accumulation and thus maintaining brain health.With respect to the results of the current study, it is possible that APOE ε4 carriers may have overestimated their self-reported sleep quality.We found that APOE ε4 non-carriers reported significantly shorter sleep duration and had a significantly higher proportion of "poor sleepers" compared to APOE ε4 carriers.However, a study by Past cross-sectional studies have reported a relationship between measures of suboptimal sleep quality (defined based on multi-question questionnaires)6,7,19and higher PET-measured brain Aβ burden.While we found that for APOE ε4 carriers self-reported "poor" sleep quality (from a single Likert scale response) was nominally associated with faster brain Aβ accumulation, but not in the whole sample or in APOE ε4 non-