Associations Between Amyloid Burden,Hypoxemia,Sleep Architecture,and Cognition in Obstructive Sleep Apnea

INTRODUCTION

Obstructive sleep apnea (OSA) is characterized by cyclic apneas and hypopneas throughout sleep, leading to repeated episodes of hypoxia, followed by subsequent arousals and a return to normal breathing. The prevalence of OSA (defined as an apnea-hypopnea index [AHI] of ≥5) is estimated to be 22% in men and 17% in women, with the risk increasing with age and obesity [1]. OSA is a major public health concern due to its association with higher mortality and an increased risk of cardiovascular disease, including stroke [1, 2].

Persons with untreated OSA commonly exhibit impairments in the cognitive domains of vigilance, psychomotor speed, attention and executive function; meta-analyses are not entirely in agreement as to whether memory and language are also affected [3, 4]. While the underlying mechanisms have not been fully delineated, the cognitive impairments observed in OSA may be due to disrupted sleep, leading to reduced slow wave sleep and excessive daytime sleepiness, and/or intermittent hypoxemia, which is associated with dysfunction and death of neurons [5, 6]. These changes are associated with a range of downstream processes affecting inflammation and cerebral perfusion [7].

A potential mechanism underpinning the cognitive impairment in OSA is the accumulation of intracerebral amyloid-β (Aβ), a peptide associated with Alzheimer’s disease (AD) pathogenesis. High cortical burdens of Aβ in apparently cognitively healthy older people are associated with reduced cognitive function when compared to similar persons without a high Aβ burden [8], with impairments being found in episodic memory, working memory, and language [9–12]. Older individuals with OSA experience higher rates of cognitive decline and have an increased risk of developing AD [13–15]. Furthermore, human studies have demonstrated associations between sleep disruption, including OSA, and increased Aβ accumulation in the blood, cerebrospinal fluid (CSF) [16], and brain [17–19].

A community-based study of cognitively unimpaired older adults demonstrated that those with sleep-disordered breathing had higher levels of cerebral Aβ, which was associated with nocturnal hypoxemia [20]. In healthy older cohorts, poor quality sleep (i.e., lower sleep efficiency and a lower proportion of slow-wave activity) [21] and excessive daytime sleepiness [22] were associated with the subsequent deposition of greater amounts of Aβ, as determined by PET imaging. However, neither sleep quality nor Aβ burden were correlated with cognitive performance [20] or a decline in cognitive performance over time [21]. These findings suggest that while sleep quality and Aβ are linked, it is not a simple linear relationship. Research using animal models indicates that the relationship between Aβ burden and sleep quality is bidirectional. Sleep deprivation, particularly slow-wave sleep disruption, is associated with increased accumulation of Aβ, whereas the presence of Aβ plaques is linked to subsequent impairment in slow-wave activity during sleep [23–25].

Beyond older age, the strongest risk factor for high Aβ burden and memory decline is the expression of the ɛ4 allele of the apolipoprotein E gene (APOE ɛ4) in both older adults with late onset AD, and in those who are cognitively normal [11]. While some early reports suggested that APOE ɛ4 carriers may also be at higher risk of developing OSA, this was not supported by meta-analysis [26]. However, when ɛ4 carriers do have OSA they tend to have greater levels of cognitive impairment than counterparts lacking APOE ɛ4 [27]. It is not known whether OSA carriers of APOE ɛ4 have heavier burdens of Aβ.

To summarize, a strong body of evidence has established that OSA is associated with impaired cognition and more extensive accumulation of Aβ [15, 28]. However, it is not known whether the Aβ burden of middle-aged individuals with OSA is associated with more extensive cognitive impairments, nor is it known whether Aβ burden in OSA is associated with the extent of nocturnal hypoxemia or sleep disruption. An understanding of these relationships would help to elucidate how OSA increases the risk of AD as well as point to potential therapeutic options. The present study aimed to investigate these relationships by determining whether:

the severity of nocturnal hypoxemia and/or disrupted sleep (arousals and reduced slow wave sleep) are correlated with Aβ burden;

We hypothesized that: 1) greater severity of nocturnal hypoxemia will be associated with higher Aβ burdens in OSA; 2a) Aβ burden will be significantly associated with impairments in verbal and working memory, and information processing speed; and 2b) Aβ burden will continue to be associated with impaired cognitive performance, even after accounting for the independent effects of slow wave sleep, nocturnal hypoxemia, and arousal index.

METHODS

35 patients with moderate-to-severe untreated OSA (AHI > 15; 16 females) and 12 healthy controls (6 females) participated in the study. Participants were eligible if they were aged 35–75 years, had sufficient English to undertake cognitive testing, and were able to provide informed consent. OSA participants were recruited from the Department of Respiratory and Sleep Medicine, at Austin Health, Melbourne. Eligible OSA participants included consecutive patients who had completed a recent diagnostic sleep study with confirmed OSA (apnea-hypopnea index (AHI) >10 events/hour) and were OSA treatment-naïve. Healthy controls were asymptomatic for sleep disorders (confirmed via polysomnography), and recruited from the community through advertisements, and involvement in previous research studies, using a similar age range and educational attainment as OSA participants. The research was approved by the Austin Health Human Research Ethics Committee (reference number: HREC/16Austin/134) and the study conformed to the Declaration of Helsinki. All participants provided written informed consent.

Exclusion criteria for all participants included major psychiatric conditions (but not a history of depression or anxiety if well managed), a history of neurological conditions that may affect cognition, uncontrolled hypertension, learning disorder, pregnancy, substance abuse, diagnosed cognitive impairment or dementia. An additional exclusion criterion for the control group was having a sleep disorder (i.e., AHI > 10).

Participants underwent a telephone screening interview and were then sent questionnaires to complete on demographics, health, sleep, and mood. They were scheduled to have: i) a neuropsychological assessment; ii) a PET scan; iii) a blood test. Control participants completed a night of laboratory-based or at-home polysomnography to confirm the absence of any sleep-disordered breathing (AHI < 5).

OSA diagnosis (or confirmation of the absence of OSA for controls) was made following clinical overnight polysomnography which was conducted either in the laboratory using the Compumedics E-Series (Abbottsford, Victoria, Australia) or at home using the Compumedics Somté (31% of the OSA group and 25% of the control group conducted in-home studies). Acquired signals were electroencephalogram (F4-M1, C4-M1, O2-M1), left and right electrooculogram, chin electromyogram, electrocardiogram, piezo movement or electromyogram sensors, a nasal pressure cannula, an oronasal thermistor, chest and abdominal respiration bands, body position sensor, and pulse oximetry. Sleep stages, respiratory events and arousals were scored according to American Academy of Sleep Medicine v2.3 criteria and the 2010 Australasian Commentary [29, 30]. Polysomnographic variables derived from the clinical diagnostic report included AHI, non-Rapid Eye Movement (NREM) AHI, REM AHI, oxygen saturation (SaO₂) nadir, oxygen desaturation index 3% (ODI 3%), Arousal Index, Stage 1 Sleep (N1 min and %), Stage 2 Sleep (N2 min and %), Slow-wave sleep (SWS min and %), REM sleep (min and %), and REM onset latency (min). ODI was calculated as the number of oxygen desaturations ≥3% ×60 / total sleep time (min). Sleep studies were staged and scored using the same method and scoring criteria for both groups by a single clinical sleep technician at the Austin Hospital.

PET scans were acquired on a Philips TF64 PET/CT scanner (Philips Healthcare) at the Molecular Imaging Department at Austin Health, Melbourne. Participants received intravenous injection with 185 MBq of the amyloid plaque imaging tracer ¹⁸F-NAV4694. After resting for 50 min, the participant was placed in a supine position in the PET scanner while a 20-min image was acquired. PET scans were analyzed using Computational Analysis of PET from AIBL (CapAIBL) software [31], which allows quantitative PET measurements without relying on magnetic resonance imaging. A continuous measure of global neocortical amyloid burden expressed as the average cortical standardized uptake value to cerebellar cortex ratio (SUVR) was recorded. Participants blood was immunoassayed to determine presence of APOE ɛ4.

The Mini-Mental State Examination was used as a measure of global cognitive status [32]. The Wechsler Test of Adult Reading provided a measure of estimated premorbid intellectual function [33]. The Digit Span Backwards subtest [33] was used as a measure of working memory span. The Logical Memory task was used as a measure of verbal episodic memory [33]. The Rey Complex Figure Test [34] was used as a measure of visual memory. The 30-min delayed score was used as the outcome for both memory tasks. Autobiographical memory was measured using the Autobiographical Memory Test [35], number of specific memories recalled. Information processing speed was assessed by Digit Symbol Coding task. Delis-Kaplan Executive Function System Verbal Fluency Test provided a measure of letter fluency and semantic (category) verbal fluency and switching [36]. Trail Making Test (B minus A) provided a measure of set shifting ability [37]. A 30-min computerized CogState (Cogstate Ltd., Australia) battery included the Groton Maze Test (executive functioning, error monitoring and processing speed), Identification Task (attention), One Card Learning (complex working memory and visual learning), One Back Working Memory Task (simple working memory), and Continuous Paired Associate Learning Task (visuo-spatial associative memory).

Sleepiness, fatigue, and mood were assessed using the Epworth Sleepiness Scale (ESS) [38], the Samn Perelli Fatigue Scale [39], and the Hospital Anxiety and Depression Scale [40], respectively. Sleepiness and mood scales were administered to the OSA participants on the evening prior to the PSG study (as per clinical protocols) and the fatigue scale was administered at the end of the cognitive test battery. All three questionnaires were administered to the control group on the cognitive testing day.

RESULTS

OSA and control participants were equivalent in age, gender, and estimated premorbid intellectual function (Table 1). There were no group differences in global cognitive function, with both groups performing close to ceiling (OSA M (SD) = 28.9 (1.5); healthy controls M (SD) = 29.5 (0.5)). All participants performed at or above the established dementia cut-off of 24, although there was more spread in the OSA group (range = 24–30) compared to the healthy control group (range = 29–30). In contrast, OSA participants had greater BMI, AHI, and ESS, as expected as a function of their diagnosis, and they had a slightly lower education. This group also reported more fatigue following the cognitive test battery and reported higher levels of depressive symptoms. OSA participants had a significantly higher ODI 3% and arousal index, and lower REM% and SWS% when compared to controls (Table 1).

The PET scan results from one participant were compromised by high tracer retention in the cerebellar cortex reference region and were removed from the analyses, leaving N = 34 OSA participants for the SUVR analyses. There was no mean difference in SUVR between groups. Using an SUVR cut-off for early Aβ accumulation of 1.25 [44], two healthy controls and 14 OSA participants met this criterion. Details of the SUVR results have been reported previously [19]. For the APOE ɛ4 results, data for 8 of the OSA participants was unavailable, leaving N = 27 in the APOE ɛ4 analyses. Seven OSA and 5 controls were APOE ɛ4 positive (Table 1).

Participants with OSA had significantly poorer performance on digits backward (p = 0.003), logical memory delay (p = 0.001), RCFT long delay (p = 0.043), autobiographical memory test (p = 0.001), letter fluency (p = 0.045); category fluency (p = 0.041), and identification speed (p = 0.01), but not on the Trail Making Task or any of the other CogState computerized measures (p > 0.05; see Table 2). Given the lack of group differences in all but one of the computerized CogState measures, these measures were not included in the regression analyses.

DISCUSSION

The present study investigated a sample of middle-aged persons with moderate-to-severe untreated OSA and compared them to age-matched healthy controls, with the aim of determining whether nocturnal hypoxemia or indices of sleep quality can account for variance in cortical Aβ burden. The extent of nocturnal hypoxemia accounted for a significant proportion of the variance in Aβ burden, after controlling for age, whereas sleep quality and daytime sleepiness were not predictive of Aβ burden. We also investigated whether the extent of impairment on cognitive tests is associated with Aβ burden or indices of sleep quality. Our findings demonstrated that Aβ burden was associated with impairments in verbal episodic memory, information processing speed and set shifting. Aβ burden continued to account for variance in information processing speed and set shifting performance after controlling for sleep variables. In addition, ODI was associated with poorer autobiographic memory performance. Variance in working memory, verbal fluency or visual memory were not associated with Aβ burden or any indices of sleep quality. Previous studies of OSA have reported that the APOE ɛ4 allele is linked to poorer spatial working memory [45], verbal memory and executive function performance, but this relationship was not observed in those without the allele [27, 46]. Twenty-seven OSA participants were screened for the APOE ɛ4 allele in the current study, and it was present in only 7 OSA patients and 5 controls.

Studies of elderly persons without OSA who suffer from disrupted sleep have reported increased levels of Aβ in the CSF [47, 48], however, those studies did not examine brain burdens of Aβ. The present study demonstrated that nocturnal hypoxemia is associated with Aβ burden. This supports our first hypothesis and is consistent with a postmortem study which demonstrated that Aβ burden in the hippocampus of OSA patients is correlated with ODI [49]. Interestingly, animal studies have reported an equivocal relationship, where chronic intermittent hypoxia has been reported to increase [50] or decrease [51] cerebral and hippocampal burdens of Aβ in transgenic mouse models of AD. Similarly, slow-wave sleep was not a predictor of Aβ burden in the current study, contrary to animal studies that have reported Aβ accumulation during SWS deprivation (for review, see [25]). This difference could be explained by the OSA group experiencing levels of sleep deprivation or SWS restriction that are not as severe as achieved in animal models, and therefore, the association may only be observed in the extreme.

Our study demonstrated that Aβ burden was associated with impairments in verbal episodic memory, information processing speed and set shifting. These findings support our second hypothesis, and are consistent with a meta-analysis of 38 studies which found that Aβ burden in cognitively normal older adults is associated with poorer episodic memory, reduced speed of information processing and impaired executive function [12]. Our study extended these findings by examining the contribution of sleep to these associations. When controlling for these aspects of sleep, verbal episodic memory performance was no longer predicted by Aβ burden; however, Aβ burden remained significantly associated with set-shifting performance and information processing in these adjusted models.

A further novel element of this study was the assessment of episodic autobiographical memory, which was found to be associated with ODI when accounting for sleep variables and Aβ burden. Impaired autobiographical specificity has been previously observed in both OSA [52] and individuals with cognitive impairment [53]. Hippocampal function has been implicated in autobiographical memory retrieval [54]. Given that OSA patients show grey matter atrophy in this brain region [55], coupled with the susceptibility of the hippocampus to hypoxia [56], it is perhaps not surprising that the measure of hypoxemia in the current study was uniquely predictive of impaired autobiographical memory.

A limitation of the present study was the lack of information on the chronicity of OSA. This limitation is common to most studies because individuals can suffer from OSA for many years without being aware of their condition and they typically obtain a clinical diagnosis only when their symptoms have begun to impact on the activities of daily living, although some algorithms to estimate age of onset are emerging [57]. We included standard measures of sleep and respiratory events derived from the PSG, however there are limitations to the robustness of some outcomes, such as the arousal index as a measure of sleep fragmentation. In addition, EEG recording in the at-home remote setting may be less reliable than that recorded in-laboratory, and thus combining both in-laboratory and at-home PSG in the same study is not without its limitations. There may also be some biases arising from comparing community-based controls to a clinical sample; however, clinic-based controls can also carry limitations and biases if they have presented to the laboratory with a sleep issue. We did not examine tau burden, which may play an important role in the degeneration of wake-promoting neurons in AD [58]; nor biomarkers of neurodegeneration which are features of the latest diagnostic criteria for AD [59]. Additionally, the cross-sectional nature of this study precludes us from determining the direction of the associations observed; for instance, nocturnal hypoxemia was associated with Aβ burden, but we cannot be certain whether hypoxemia causes Aβ deposition or whether Aβ deposition increases the severity of nocturnal hypoxemia. Both the nocturnal hypoxemia and sleep disturbances in OSA are treatable, and consistent treatment has been shown to improve cognitive performance and memory deficits. It will be important to investigate whether long-term treatment of OSA can slow or halt the accumulationof Aβ.

The findings of the current study, while preliminary, offer important insights into the relationship between OSA and AD, and highlight hypoxemia as an important correlate of Aβ burden. While sleep disturbances in OSA are treatable, whether hypoxemia and reduced slow-wave sleep are causally related to higher Aβ burden and lower cognition is an important question for future research as the field moves toward targeted interventions for modifiable risk factors to slow the early pathogenesis of AD and the associated declines in cognition.