Insomnia symptoms and markers of preclinical Alzheimer’s disease in the community

Abstract

Background

Brain amyloid production increases during wakefulness before being cleared during sleep. Insomnia is one of the most common sleep disorders, yet its relationship to core early markers of preclinical Alzheimer’s disease remains unknown. We investigated the cross-sectional relationship between insomnia symptom severity and cerebrospinal spinal fluid (CSF) concentrations of Alzheimer’s disease biomarkers in a non-demented community sample aged 43-67 years.

Method

66 participants from the Healthy Brain Project Biomarker sub-study (age = 58±6 years; MMSE = 29±7; 68% women) completed a lumbar puncture, two weeks of actigraphy to measure sleep and wake patterns, and the Insomnia Severity Index (ISI) to measure insomnia symptom severity over the past two weeks. Difficulty initiating sleep (prolonged sleep onset latency) and difficulty maintaining sleep (wake after sleep onset [WASO] and number of awakenings) were measured by actigraphy as two of insomnia’s core features. Linear regression was used to estimate the associations between the insomnia variables and CSF amyloid-beta 42 (Aβ42), phosphorylated tau 181 (p-tau181), total-tau, and neurofilament light chain protein (NfL), adjusting for age and sex.

Result

Higher ISI score was associated with greater average levels of CSF Aβ42 (per point: 17.6 pg/mL, 95% CI: 0.65, 34.6, p = 0.042), as was higher WASO (per 10 min: 77.1 pg/mL, 95% CI: 18.0, 136, p = 0.012) and more awakenings (per 5: 68.9 pg/mL, 95% CI = 23.3, 115, p = 0.004). There was no clear evidence that difficulty initiating sleep was associated with CSF Aβ42, nor that any of the insomnia features were associated with p-tau181, total-tau, or NfL levels. The relationship between difficulty maintaining sleep and CSF Aβ42 may be modified by APOE e4 polymorphism, with a stronger relationship amongst e4 allele carriers.

Conclusion

In this relatively young sample, symptoms and features of insomnia were associated with higher CSF Aβ42 levels. These findings may reflect increased amyloid production due to acute sleep disruption. Longitudinal studies are needed to determine whether sustained sleep disruption associates with amyloid accumulation over time.

Introduction

Sleep plays a fundamental role in maintaining brain health, including synaptic consolidation and memory formation[1]. Of relevance to Alzheimer’s disease (AD) and dementia, slow wave sleep facilitates the removal of neurotoxic waste, including amyloid beta[2]. However, relatively little is known about how sleep disorders may contribute to dementia risk.

Insomnia is one of the most common sleep disorders, experienced in various forms by 1 in 3 people[3]. Insomnia involves a frequent and chronic complaint of dissatisfaction with sleep quantity or quality, associated with difficulty initiating sleep, difficulty maintaining sleep, or early morning awakenings[4]. Several studies have found associations between insomnia and an increased risk of dementia or poorer cognitive function[5–9]. By definition, insomnia causes significant distress or impairment in everyday functioning. Thus, cognitive impairments in insomnia could stem from decreased alertness and increased fatigue rather than underlying neurodegenerative processes. Better understanding these mechanisms could clarify the role of sleep dysfunction in dementia.

Accordingly, this study aimed to examine insomnia symptom severity and related objective markers of sleep quality (difficulty falling asleep and maintaining sleep on actigraphy) in association with cerebrospinal fluid (CSF) amyloid, tau, and neurodegeneration (ATN) biomarkers of preclinical AD in a community-based cohort of cognitively healthy middle-aged adults. We hypothesized that greater insomnia symptom severity would be associated with greater evidence of preclinical AD. Poor sleep and neurodegeneration likely share a bidirectional relationship. Thus, to limit reverse-causation, we performed the current investigation in a cognitively healthy middle-aged sample at risk of AD but who were unlikely to have clinically significant neuropathology that could impact sleep.

Methods

Participants

Participants were recruited from the Healthy Brain Project (HBP)[10]. The HBP (healthybrainproject.org.au) is a longitudinal community-based online cohort of middle-aged adults (approximately 8,500 participants) aged between 40 and 70 years at baseline. The study was designed to investigate the biological, environmental, and psychological factors that affect cognitive aging and dementia risk. Participants were free from clinical cognitive impairment and significant neurological disease at study entry.

From this cohort, a subset of participants were recruited for an in-person biomarker sub-study which took place at the Royal Melbourne Hospital in Melbourne, Australia. The subgroup was recruited based on ability to travel to the hospital for the assessment and without conditions that would prevent completing an MRI brain scan and lumbar puncture procedure. The study sample was enriched with APOE e4 carriers to study a sample at high risk of late-onset AD. In total, 82 participants completed the biomarker sub-study; 77 completed a lumbar puncture, and of these, 70 participants completed the sleep assessments. Four participants were excluded for having less than 10 days of actigraphy, leaving an analysis sample of 66 (see Figure 1). Four participants did not have APOE e4 genotype data available. The Melbourne Health ethics board approved the study and all participants provided written informed consent. Data collection occurred between November 2018 and February 2020.

Figure 1. Study Flow Diagram.

HBP= Healthy Brain Project, ISI= Insomnia Severity Scale. Note: Two participants from the final sample had missing APOE e4 data due to sample processing errors.

Measures

CSF biomarkers

We measured CSF biomarkers that mapped onto the NIA-AA Amyloid (Aβ42), Tau (ptau181), Neurodegeneration (t-tau, NfL) (ATN) Framework[11]. CSF samples were obtained by lumbar puncture in the L3/L4 or L4/L5 interspace, with most samples collected between 13:00 and 14:30 hours. CSF samples were transferred for processing on wet ice following well-established guidelines. Samples were spun at 2000 x g at +4 °C for 10 minutes. Supernatant was pipetted off to a new polypropylene tube and gently inverted a few times to avoid possible gradient effects. Samples were then aliquoted in 0.5mL portions into screw-cap polypropylene tubes and stored at −80 °C pending biochemical analyses. CSF concentrations of amyloid beta 42 (Aβ42), total tau (t-tau), and tau phosphorylated at threonine 181 (ptau181) were measured by immunoassay (Roche Elecsys^®) and CSF concentrations of neurofilament light chain protein (NfL) were measured using ELISA (UmanDiagnostics, Umeå, Sweden). All analyses were conducted at the National Dementia Diagnostics Laboratory (The Florey Institute, University of Melbourne, Australia). 25 (30%) participants had amyloid scores above the maximum limit of detection and were assigned the top range score (1700 pg/mL).

Sleep Measures

Insomnia Severity Index (ISI).

Self-reported insomnia symptom severity over the last two weeks was measured with the 7-item ISI. The ISI was scored on a continuous scale with higher scores indicating more severe insomnia symptoms.

Activity Monitoring

We measured core features of insomnia from wrist actigraphy (Phillip Respironics Actiwatch Spectrum Plus), including difficulty initiating sleep (measured as sleep onset latency) and difficulty maintaining sleep (measured using both wake after sleep onset [WASO] and number of awakenings). While polysomnography is the gold standard for sleep assessment, actigraphy demonstrates high precision in assessing sleep parameters compared with polysomnography and provides a more naturalistic measure of habitual sleeping patterns[12]. Participants were required to wear wrist actigraphy for 17 consecutive days, which measured 24-hour activity levels in 30-second epochs. The analysis of actigraphy data was performed according to a standardized protocol by the same trained independent researcher using specialized software (Phillips Respironics Actiware 6.0.9). The first two nights of data were excluded due to the possibility of sleep disruptions because of the lumbar puncture, which occurred the day before the first day of actigraphy. Participants also completed a sleep diary, which was used to corroborate bed and rise times generated from actigraphy. Sleep and wake times used in analysis were derived from the actigraphy. However, the sleep diary time was reported when there was a discrepancy of more than one hour between the time on the diary and the time on the Actiwatch. The Actiware 6.0.9 autonomic wake threshold was used to calculate WASO and awakenings. The wake threshold was calculated by dividing the sum of activity (calculated using an accelerometer) by mobile time (when the number of activity counts recorded in that epoch is greater than or equal to the epoch length in 15-second intervals), multiplied by 0.88888. We averaged all actigraphy measures over at least 10 days to ensure stability in sleeping patterns (i.e., participants with less than 10 days of recording were excluded).

Demographics

Basic demographic information was collected, including age and sex. Participants were also asked if they used any sleep medications or have any medical diagnoses. APOE genotype was determined through TaqMan genotyping assays (Life Technologies).

Statistical Methods

We examined the relationships between the sleep metrics and CSF biomarkers using linear regression. All exposure and outcome variables were included as continuous variables. Models were adjusted for age and sex as these variables are known determinants of sleep characteristics and AD biomarkers. Additionally, an ANCOVA model was performed for the ISI in which the ISI score was split into two groups (normal vs. high), where a score of 8 or more indicated high insomnia symptomatology, as utilized previously[13]. The ANCOVA model was age and sex adjusted. Least-square means were used to calculate Cohen’s d to quantify the magnitude of effect when comparing normal and high ISI groups. To examine whether the associations between sleep and the CSF biomarkers were modified by APOE ε4 carriage (ε4 carrier vs. non-ε4 carrier), interaction terms were added to the above models.

Conditional on age and sex, missing outcome data was assumed to missing completely at random. In all models, participants with missing data in predictor or outcome variables were excluded (i.e., we used a complete case analysis).

No p-value adjustments for multiple comparisons were made. Results were interpreted within the context of the larger set of study results and in light of the multiple tests conducted[14] All analyses were conducted using SPSS Statistical Software (version 26) and R version 4.1.0[15]. R analysis code is available in the Supplementary Material.

Results

Sample Overview

Table 1 displays the demographic characteristics of the sample. On average, participants were 58 (SD, 6) years old and 60% were female. Of the sample, 28 (42%) participants reported insomnia symptom severity above the threshold. These participants had longer average WASO, more awakenings, and shorter total sleep time. Surprisingly, they also had shorter sleep onset latency. Additionally, two participants used medications to improve their sleep and two participants reported having sleep apnoea. 11 (18%) participants in the sample had CSF amyloid levels <1000pg/mL that were indicative of amyloid positivity[16].

Table 1.

Sample demographics stratified by the degree of insomnia severity

		Insomnia Severity Scale score*
Variable	Overall (N = 66)	Low (N=38)	High (N=28)
Age, years	58.43 (6.47)	59.28 (6.02)	57.27 (6.98)
Education, years	16.68 (3.58)	16.93 (3.75)	16.34 (3.36)
Females, N (%)	45 (68.2)	29 (76.3)	16 (57.1)
APOE ε4 carrier, N (%)	26 (41.3)	16 (43.2)	10 (38.5)
Aβ42, pg/mL	1387.94 (334.53)	1319.12 (354.32)	1487.04 (281.74)
T-tau, pg/mL	191.09 (62.76)	190.04 (69.22)	183.94 (52.61)
P-tau181, pg/mL	15.93 (5.89)	16.44 (6.51)	15.19 (4.91)
NfL, pg/mL	596.60 (258.12)	610.60 (313.73)	576.43 (149.36)
Sleep onset latency, min	18.14 (14.46)	20.10 (17.16)	15.34 (8.98)
WASO, min	37.72 (13.80)	36.82 (14.27)	39.00 (13.31)
Awakenings, number	32.46 (8.91)	30.61 (7.90)	35.11 (9.75)
Total sleep time, hrs	7.06 (0.85)	7.15 (0.70)	6.92 (1.04)

Only participants with ISI data available (n = 66) are included in the table. All values are presented as mean (SD) unless stated otherwise; Highest sample sizes are reported. Reading data columns left to right, the sample sizes for APOE ε4 were 63, 37, 26, the sample sizes for Aβ42, t-tau, p-tau-181, and NfL were 61, 36, 25, and the samples sizes for sleep onset latency, WASO, awakenings and total sleep time were 56, 33, 23. ISI= Insomnia Severity Scale, Aβ42= Amyloid beta 42, T-tau= total tau, P-tau181= Tau phosphorylated at threonine 181, NfL= Neurofilament light chain, WASO= wake after sleep onset.

^*Low ISI= normal score on the ISI, defined as any score of less than 8; High ISI= high score on the ISI, defined as any score of 8 or more.

Association between insomnia symptoms and AD biomarkers

There was moderate evidence that higher ISI score was associated with greater average levels of CSF Aβ42 (per point: 17.6 pg/mL, 95% CI: 0.65, 34.6, p = 0.042), adjusted for age and sex (Figure 2). Similarly, there was moderate-strong evidence that participants with higher average WASO (per 10 mins: 77.1 pg/mL, 95% CI: 18.0, 136, p = 0.012) and more average awakenings (per 5: 68.9 pg/mL, 95% CI = 23.3, 115, p = 0.004) had greater mean levels of CSF Aβ42.

Figure 2. A) Association between total ISI score (continuous variable) and Aβ42 B) Association between ISI score (high vs low) and Aβ42.

Models are adjusted for age and sex. ISI= Insomnia severity scale; Aβ42= Amyloid beta 42.

There was little or no evidence that sleep onset latency was associated with CSF Aβ42, nor that any of the sleep characteristics were associated with average levels of CSF T-tau, p-tau-181, or NfL (Table 2).

Table 2.

Associations between sleep variables and CSF biomarkers

Variable	N	Aβ42, pg/mL		T-tau, pg/mL		P-tau 181, pg/mL		NfL, pg/mL
		β (95% CI)	p	β (95% CI)	p	β (95% CI)	p	β (95% CI)	p
ISI, per point	61	17.6 (0.65, 34.6)	0.042	−1.21 (−4.31, 1.90)	0.44	−0.14 (−0.43, 0.16)	0.36	−0.12 (−11.8, 11.6)	0.98
Sleep latency, per 10 min	52	−17.6 (−85.9, 50.7)	0.61	−7.80 (−19.7, 4.10)	0.29	−0.83 (−1.97, 0.31)	0.23	−2.45 (−52.6, 47.6)	0.96
WASO, per 10 min	52	77.1 (18.0, 136)	0.012	−0.32 (−11.3, 10.7)	0.95	−0.31 (−1.37, 0.75)	0.56	−4.69 (−50.3, 40.9)	0.84
Awakenings, per 5	52	68.9 (23.3, 115)	0.004	0.53 (−8.08, 9.14)	0.90	−0.17 (−1.00, 0.65)	0.68	−9.46 (−45.1, 26.2)	0.60

Models adjusts for age and sex. Betas are unstandardized. Note: For these analyses, all sleep scores and biomarker outcomes are measured on a continuous scale. ISI= Insomnia severity scale; Aβ42= Amyloid beta 42; T-tau= total tau; P-tau181= Tau phosphorylated at threonine 181; NfL= Neurofilament light chain; WASO= Wake after sleep onset (minutes). Higher scores on the ISI indicate higher insomnia symptom severity. Higher onset latency, WASO, and awakenings indicate poorer sleep quality.

In the ANCOVA analysis, participants with a high ISI score (ISI>7) had greater average levels of CSF Aβ42 (mean difference = 192 pg/mL, 95% CI = 16.1, 367, p = 0.033, Cohen d = 0.58) when comparing to the remainder of the sample (Figure 2).

Interactions with APOE e4

There was moderate evidence that the relationship between average WASO and average awakenings was modified by APOE e4 carriage (p interaction = 0.039 and p interaction = 0.017, respectively). The relationship between average WASO and average awakenings and CSF Aβ42 levels was stronger in e4 allele carriers (Figure 3). There was weak or no evidence that APOE-e4 carriage modified the association between the remaining sleep characteristics and preclinical AD biomarkers.

Figure 3. Effect modification by APOE-e4 status.

A: Estimates of interaction terms (APOE-e4 carriage x exposure). Interactions represent the difference in the estimated relationship between exposure and biomarker variable between APOE-e4 allele carriers and non-carriers. Positive interactions indicate that the relationship between the exposure and biomarker is larger (more positive) amongst e4 allele carriers.

B: Association between average WASO and average awakenings with Aβ42, stratified by APOE-e4 carriage. Models are adjusted for age and sex. ISI= Insomnia severity scale; Aβ42= Amyloid beta 42; T-tau= total tau; P-tau181= Tau phosphorylated at threonine 181; NfL= Neurofilament light chain; WASO= Wake after sleep onset (minutes). Higher scores on the ISI indicate higher insomnia symptom severity. Higher onset latency, WASO, and awakenings indicate poorer sleep quality.

DISCUSSION

In a middle-aged sample, insomnia symptoms were associated with higher levels of CSF Aβ42. Difficulty maintaining sleep, one of insomnia’s core features, was also associated with higher Aβ42, whereas there was no such evidence for difficulty initiating sleep. Additionally, preliminary evidence suggests that the strength of the relationship between WASO and number of awakenings and Aβ42 may be stronger in APOE e4 carriers, who are genetically predisposed to a higher risk of AD, compared to non-carriers. There was no evidence that poor sleep quality was associated with other CSF AD biomarkers (t-tau, p-tau-181, or NfL). Overall, these data show that in middle-aged adults at risk of dementia but without cognitive impairment, insomnia symptoms are associated with higher CSF amyloid, though similar evidence was not found for the other dementia biomarkers.

We hypothesized that insomnia symptoms would be associated with greater evidence of AD but found a positive association between insomnia symptoms and CSF Aβ42. Since low CSF Aβ42 indicates amyloidosis and higher amyloid burden in individuals with AD[17], our results may appear counterintuitive at first. Our finding, however, may not be unexpected. Neuronal activity increases amyloid production, which is then cleared during sleep. Thus, greater time spent awake and greater sleep disruption may increase Aβ42 production or reduce Aβ42 clearance in persons at risk of dementia. In the absence of plaques, which would sequester CSF Aβ42 leading to abnormally low levels in the CSF, in this healthy and relatively young sample, this may result in an increased Aβ42 production being detectable in the CSF. Indeed, several studies have shown that acute sleep disruption increases levels of CSF Aβ42[18, 19]. For our study, however, we examined sleep patterns over two weeks. Therefore, our results might suggest that more sustained sleep disturbance can also increase Aβ42 levels in persons at risk of dementia. In support of this suggestion, one study of middle-aged adults reported that CSF levels of Aβ42 were higher in 23 patients with chronic insomnia as compared to 23 controls[20]. However, this study lacked other relevant dementia biomarkers, and objective sleep assessments were not examined. With respect to other sleep disorders, the treatment of obstructive sleep apnoea (OSA) has been shown to result in increased slow-wave sleep which was, in turn, significantly correlated with lower CSF Aβ42 after treatment[21]. Together, these findings suggest that sleep disorders like insomnia and OSA may increase CSF Aβ42, perhaps through the effects of sleep disruption on amyloid production or clearance.

Our results should be interpreted in the context of our middle-aged, cognitively healthy sample. Although we link insomnia symptoms to higher CSF amyloid in midlife and others have linked insomnia to a higher risk of dementia[5, 7, 8], it has not been proven that these two processes are linked. In our study, sleep was not associated with any other dementia biomarkers. Therefore, the significance of the association between insomnia symptoms and amyloid for eventual risk of dementia is unclear. It has been hypothesized that high CSF Aβ42 in cognitively healthy individuals may represent Aβ42 overproduction[22]. In the context of AD, CSF Aβ42 may then begin to fall only once Aβ accumulates in the form of plaques. Indeed, in autosomal dominant AD, there is a trend for increased CSF Aβ42 levels before an eventual (and much larger) fall closer to symptom onset[23]. Therefore, those with fragmented sleep and higher amyloid load may be more susceptible to plaque formation, as a mismatch between production and clearance emerges. In this sense, longitudinal CSF, imaging, and clinical measures are needed to determine the temporal relationships between sleep disturbance and the progression of AD over time.

We observed an interaction with APOE e4 carriage, whereby the positive relationship between sleep fragmentation, as measured by WASO and number of awakenings, and CSF Aβ42 was pronounced in e4 carriers compared to non-carriers. These findings dovetail with other recent findings showing that insomnia symptom severity was more strongly associated with poorer cognition in APOE e4 carriers[6]. APOE is the strongest known genetic risk factor for late-onset AD and is involved in the clearance and aggregation of amyloid beta[24]. APOE e4 carriers with sleep disturbance may experience a double-hit, whereby sleep disruption and the effects of APOE e4 both impair the glymphatic clearance of amyloid. This may lead to higher CSF levels in healthy individuals and possibly increase the risk of plaque formation and future cognitive impairment. Nevertheless, we cannot rule out that this effect modification was a chance finding, considering the number of interaction terms assessed, and external replication would be valuable.

Strengths of our study include both the objective and subjective assessment of sleep and gold-standard AD biomarkers that map onto the ATN biomarker framework[11]. Moreover, since we explored sleep-AD biomarker relationships in a cognitively healthy middle-aged sample, we limited the potential impact of more advanced neurodegeneration on sleep (and thereby limited reverse-causation). However, our study is not without limitations. Firstly, our data were cross-sectional, which precluded investigation into whether insomnia symptoms were associated with the progression of preclinical changes leading to dementia. Moreover, we choose CSF to measure AD biomarkers over PET imaging since CSF AD biomarkers become abnormal earlier in the AD course and permit the measurement of multiple biomarkers simultaneously. However, the absence of amyloid position emission tomography imaging in our study means that we could not quantify regional brain amyloid burden. Lastly, our small and homogenous sample precluded sub-group analysis (e.g., by sex or race).

In conclusion, this study found that in a middle-aged sample, symptoms and features of insomnia were associated with higher CSF Aβ42 levels but no other CSF biomarkers. These findings may reflect increased amyloid production due to sustained sleep disruption, an effect that may be magnified in APOE e4 carriers, who have impaired amyloid clearance. Our results suggest a possible pathway linking insomnia to a higher risk of cognitive impairment, reported previously[6]. Prospective studies are needed to ascertain if insomnia is associated with AD progression over time. Such research could inform whether the treatment of insomnia could help protect against the development of dementia.