Sleep characteristics, cognitive performance, and gray matter volume: findings from the BiDirect Study

Abstract

Study Objectives

Sleep is essential for restorative metabolic changes and its physiological correlates can be examined using overnight polysomnography. However, the association between physiological sleep characteristics and brain structure is not well understood. We aimed to investigate gray matter volume and cognitive performance related to physiological sleep characteristics.

Methods

Polysomnographic recordings from 190 community-dwelling participants were analyzed with a principal component analysis in order to identify and aggregate shared variance into principal components. The relationship between aggregated sleep components and gray matter volume was then analyzed using voxel-based morphometry. In addition, we explored how cognitive flexibility, selective attention, and semantic fluency were related to aggregated sleep components and gray matter volume.

Results

Three principal components were identified from the polysomnographic recordings. The first component, primarily described by apnea events and cortical arousal, was significantly associated with lower gray matter volume in the left frontal pole. This apnea-related component was furthermore associated with lower cognitive flexibility and lower selective attention.

Conclusions

Sleep disrupted by cortical arousal and breathing disturbances is paralleled by lower gray matter volume in the frontal pole, a proposed hub for the integration of cognitive processes. The observed effects provide new insights on the interplay between disrupted sleep, particularly breathing disturbances and arousal, and the brain.

Introduction

Sleep is an indispensable physiological process of the brain important for maintaining cognitive functions [1] and restorative metabolic changes [2–4]. Population-based studies investigating self-reported sleep quality suggest that 32.1%–36% of adults are poor sleepers [5, 6]. Biological correlates of impaired sleep can be examined by monitoring physiological functions with polysomnography (PSG), including the assessment of apnea and hypopnea events, periodic limb movements, cortical arousal, the architecture of sleep stages, and many more.

Prior studies used PSG recordings primarily to define patient groups or utilized individual PSG variables, for example, breathing-related disturbances or slow-wave sleep, to analyze associations with gray matter volume [7–11]. Particularly, frontal slow-wave sleep activity was associated with a higher gray matter volume in the medial prefrontal cortex [9], sleep spindle frequencies were related to a higher gray matter volume in the hippocampus and also in regions involved in sensory perception [11], and oxygen desaturation during sleep was paralleled by a smaller hippocampus volume [8]. However, analyzing individual PSG variables alone does not take into account the full spectrum of sleep characteristics that is provided by multiple PSG recordings altogether. Identifying and aggregating shared variance from multiple PSG recordings into specific components has recently emerged as useful for the prediction of clinical outcomes [12]. This technique has, however, not yet been applied to investigate associations between PSG recordings and gray matter volume.

The aim of the present study was to analyze associations between aggregated measures of overnight PSG recordings and cerebral gray matter volume. Moreover, we analyzed the potential associations between sleep characteristics and cognitive functioning. It was hypothesized that characteristics indicative of poor sleep would be associated with lower gray matter volume as well as lower cognitive performance.

Methods

Participants

The overnight PSG was an add-on to the first follow-up examination of the BiDirect Study [13]. The BiDirect Study comprised two patient cohorts and a cohort of community-dwelling participants, aged between 35 and 65 years during baseline enrollment. The sample of community-dwelling adults was randomly drawn from the population registry of the city of Münster (Germany) and invited to participate in BiDirect. At baseline, this cohort included 911 participants, 800 of whom returned for the first follow-up examination. Participants underwent an extensive examination program, including a face-to-face interview with regard to medical history and medication intake. This was followed by a cognitive assessment and a magnetic resonance imaging of the brain. In a last step, participants residing within a distance of not more than 30 km away from the city of Münster were invited for an overnight PSG. Of those, 241 participants underwent PSG and were selected for the present analysis. Participants with missing magnetic resonance imaging (n = 37), scanning artifacts (n = 1), or severe brain abnormalities (n = 5) were excluded. The same holds true for those with a total sleep time less than 4 h or missing values on any PSG variable (n = 8), yielding a final sample of 190 participants. Study participation was voluntary and without any financial compensation. The BiDirect Study was approved by the ethics committee of the University of Münster and conducted according to the Declaration of Helsinki. Written informed consent was signed by all participants.

Image acquisition and preprocessing

The 3D T1-weighted images of the head were acquired on a 3 T scanner (Philips Intera, Best, Netherlands) as follows: 160 sagittal slices (2 mm thickness, reconstructed to 1 mm), TR = 7.26 ms, TE = 3.56 ms, 9° flip angle, matrix dimension 256 × 256, FOV = 256 mm × 256 mm.

The SPM12 r7219 (www.fil.ion.ucl.ac.uk/spm/software/) and CAT12 r1363 (www.neuro.uni-jena.de/cat/) toolboxes running on Matlab r2016a were used for the multistage preprocessing of the T1-weighted images. The images were corrected for bias-field inhomogeneities (using full affine preprocessing), segmented into gray matter, white matter, and cerebrospinal fluid with an Adaptive Maximum A Posterior technique. The segmented tissue images were normalized using the Diffeomorphic Anatomic Registration Through Exponentiated Lie (DARTEL) algebra algorithm with a resulting voxel resolution of 1.5 mm × 1.5 mm × 1.5 mm. Finally, the modulated, normalized gray matter images were smoothed with SPM12 using a Gaussian kernel with a size of 8 mm full-width half-maximum.

Polysomnography

The overnight PSG sleep assessment took place at the former Department of Sleep Medicine and Neuromuscular Disorders of the University Hospital Münster (Germany) and was acquired with Somnoscreen (Somnomedics, Randersacker, Germany) or Polysmith (Nihon Kohden, Tokyo, Japan) devices. The recordings were started at 10:00 pm and continued until 06:00 am. All PSG measurements were performed by certified personnel according to the criteria established by the American Academy of Sleep Medicine [14] (AASM). The overnight PSG included the following measurements: electroencephalogram (F4-M1, C4-M1, C3-M2, and O4-M1), electrooculogram (two electrodes), electrocardiogram (two electrodes), surface electromyogram (electrodes on the chin and both anterior tibialis muscles), pulse oximetry, nasal pressure, respiratory inductive plethysmography, tracheal microphone, and body position sensors. Limb movements were assessed with bipolar surface electrodes from the electromyogram, which were attached along the anterior tibialis muscle.

The PSG recordings were analyzed by sleep experts and the sleep stages, arousals, apnea–hypopnea events, the number of limb movements, periodic limb movements during sleep (PLMS), sleep efficiency, the number of awakenings, minimum blood oxygen saturation, and minimum heart rate were assessed according to the updated AASM guidelines [14, 15]. The apnea–hypopnea index was calculated as the total number of apneas and hypopneas per hour of sleep. The PLMS index and the arousal index were computed analogously.

Cognitive assessment

A battery of cognitive tests was administered to the participants, including the trail making test (TMT) and the Stroop Test as measures of executive functioning [16, 17], and the Animal Naming Test (ANT) for the assessment of semantic fluency [18].

The total time needed to complete each of the two parts of the TMT was recorded. TMT-A measures visual search and psycho-motor speed, whereas TMT-B additionally demands cognitive flexibility and switching between interfering stimuli [19]. In order to obtain a clean measure of cognitive flexibility, a time difference score (TMT-B − TMT-A) was calculated, a procedure which is frequently applied and constitutes established practice [20]. The distribution of this time difference score was slightly right-skewed and therefore transformed using a log₁₀ transformation (y_i = log₁₀[TMT-B − TMT-A]).

The time needed to complete each Stroop subset was recorded and the time difference between the third (color words printed in incongruent ink) and the second subset (color patches) was calculated. The rationale for the time difference score is that the completion of the second subset relies primarily on visual recognition and psycho-motor speed, whereas the third subset additionally requires selective attention and the ability to inhibit cognitive interference [21]. Calculating a time difference score [17] results in a cleaner measure of selective attention and inhibition. Due to the slight right-skew of the time difference score, a log₁₀ transformation was subsequently applied (y_i = log₁₀[Stroop3 − Stroop2]).

For the ANT, participants were instructed to name as many animals as possible within 60 s. The number of correct and nonrecurrent namings was recorded.

Additional variables

Age and sex were collected during the face-to-face interview, and the body mass index (kg/m²) was measured during the physical examination program of the first follow-up of the BiDirect Study [13]. Total intracranial volume (TIV) was estimated with the CAT12 toolbox. Age, sex, body mass index, and TIV were considered as potential confounders and therefore included in the respective statistical models investigating associations between PSG recordings and gray matter volume or cognitive functioning.

Statistical analysis

For the purpose of deriving informative components of PSG characteristics, the following 13 variables were fed into a principal component analysis (PCA) with subsequent promax rotation: the duration of stage 1, stage 2, stage 3, and rapid eye movement (REM) sleep, sleep efficiency, apnea–hypopnea index, the arousal index, PLMS index, the number of nonperiodic limb movements, the number of awakenings, minimum oxygen saturation, and minimum and maximum heart rate. A few participants had extreme values on six variables (apnea–hypopnea index, arousal index, PLMS index, the number of limb movements, the number of awakenings, and the maximum heart rate). Such extreme values can potentially distort the estimation of principal components [22, 23]. Before these variables were used for the PCA, each of them was first transformed as follows: y_i = log₁₀(x_i + 1). The number of components to be extracted was determined by parallel analysis [24], which compares the eigenvalues of the principal components with those of an equally sized random data matrix derived from a Monte-Carlo simulation. Components with eigenvalues greater than those of the simulated data matrix were retained. Statistical analyses of all non-imaging data were conducted using R version 3.5.2.

Associations between the individual extracted component scores and gray matter volume were analyzed using a voxel-wise general linear model, performed with FSL randomize [25] version 2.9. Each of the component scores was individually modeled as independent variable, adjusted for age, sex, TIV, and body mass index. The additional adjustment for body mass index was necessary due to its previously observed associations with gray matter volume [26] as well as polysomnographic sleep characteristics [27, 28]. The analyses were conducted within a gray matter mask, which was produced by averaging all gray matter images, thresholding (intensity > 0.1) and dichotomizing the resulting image. Threshold-free cluster enhancement with 5,000 permutations (2 mm variance smoothing) was used and family-wise error corrected statistical maps were thresholded at p < 0.05.

Furthermore, we conducted multiple linear regression analyses to examine associations between cognitive performance (ANT, TMT, and Stroop Test) and those PSG component scores that were related to gray matter volume alterations. The individual PSG component scores served as independent variables, adjusted for age, sex, and body mass index.

In a last step, separate multiple regression models were used to analyze potential associations between cognitive performance as outcome (TMT, Stroop Test, and ANT) and extracted clusters of PSG-related alterations of gray matter volume as independent variable, adjusted for age, sex, TIV, and body mass index. For the analyses of cognitive data, six participants had to be excluded due to language difficulties (n = 1), color vision deficiency (n = 2), missing (n = 2), or measurement error (n = 1). The level of significance was set to α = 0.05 (two-tailed).

Results

Demographics and sleep characteristics

The demographic details and sleep characteristics of the sample are presented in Tables 1 and 2. The mean age for the sample was 56.39 years, and average total sleep time was 6.16 h (standard deviation = 0.87). Pearson correlations among the PSG variables are given in Figure 1. For the PCAs of the 13 PSG variables, the number of components to be extracted was estimated with parallel analysis and yielded 3 principal components. These three components together accounted for 52.6% of the variance in the data (component 1 = 22.3%, component 2 = 17.1%, and component 3 = 13.2%), and the component loadings for each PSG variable are shown in Table 3. Component 1 was particularly characterized by high arousal-, PLMS-, and apnea–hypopnea indices as well as more stage 1 sleep, less stage 3 sleep, and lower minimum blood oxygen saturation. Component 2 was primarily associated with more stage 2 sleep, REM sleep, and sleep efficiency. The third component was related to a lower minimum heart rate, more awakenings, and a higher maximum heart rate.

PSG components and gray matter volume

The voxel-wise analyses of cerebral gray matter volume and PSG components revealed a significant association between component 1 and lower gray matter volume in the left frontal pole (Brodmann Area 10; Peak MNI coordinates: x= −22.5, y = 64.5, z = −4.5, cluster size = 168 voxel) as displayed in Figures 2 and 3. With regard to PSG components 2 and 3, no significant associations with gray matter volume were detected.

Cognitive functioning

The apnea-related PSG component 1 (β = 0.05; p = 0.008; model p < 0.001; model R² = 0.11) was associated with lower performance (i.e. more time needed) on the log₁₀-transformed TMT difference score, after controlling for age, sex, and body mass index. An increase in PSG component 1 (β = 0.034; p = 0.028; model p < 0.001; model R² = 0.15) was likewise associated with lower performance on the log₁₀-transformed Stroop score. The association between PSG component 1 (β = −1.28; p = 0.009; model p = 0.06; model R² = 0.05) and performance on the ANT did not reach significance on the model level.

In subsequent analyses, the extracted PSG-related gray matter volume in the frontal pole (β = −0.95; p = 0.009; model p < 0.001; model R² = 0.11) was associated with a lower log₁₀-transformed TMT difference score, after controlling for age, sex, body mass index, and TIV. This cluster (β = −0.76; p = 0.01; model p < 0.001; model R² = 0.16) was also associated with a lower log₁₀-transformed Stroop difference score, and higher scores (β = 25.55; p = 0.007; model p = 0.02; model R² = 0.07) on the ANT.

Discussion

The study at hand aimed to investigate associations between aggregated measures of sleep characteristics, gray matter volume alterations, and cognitive functioning. We found that the PSG data were summarized by three components, whereby the first component was mainly defined by sleep-disordered breathing, cortical arousal, more superficial stage 1 sleep, and less deep, slow-wave sleep (stage 3). This component possibly reflects fragmented sleep, where the transition to slow-wave sleep was repeatedly disrupted by arousal and breathing disturbances, which consequently caused a return to lighter levels of sleep. The second component was defined by REM sleep and sleep efficiency, describing long and restful sleep. The third component was primarily associated with the number of awakenings. The PSG component 1, which was primarily defined by breathing disturbances and cortical arousal, was significantly associated with gray matter volume, indicating that breathing disturbances and arousal during sleep are related to lower gray matter volume in the left frontal pole. Prior studies that analyzed associations between individual sleep characteristics and cerebral gray matter revealed that mean oxygen saturation [29] as well as slow-wave sleep [9–11] are associated with alterations in the frontal lobe. It is important to acknowledge that sleep-disordered breathing is not exclusively characterized by apnea–hypopnea events alone, but is also associated with cortical arousal [30, 31], shortened stage 3 sleep, and prolonged stage 1 sleep [32], which is in agreement with our results from the PCA. The identification of shared variance from PSG data may provide a more complete depiction of breathing-related disturbances and its consequences for brain structure.

A comprehensive theory of frontal lobe functioning and sleep has been advanced, where the prefrontal cortex plays a key role in the transition from wakefulness to slow-wave sleep [33]. According to this framework, activity in the frontal lobe decreases during this transition, before becoming reactivated when REM sleep emerges. Arousal, breathing disturbance, and blood oxygen desaturation may interfere with the transition to restful slow-wave sleep and possibly give rise to the observed lower gray matter volume in the frontal pole. Studies using proton magnetic resonance spectroscopy suggest that the hippocampus and the frontal lobe are particularly prone to exhibit metabolic modifications following hypoxia [34, 35]. Other metabolic changes observed in the frontal lobe relate to the function of slow-wave sleep, which may foster glymphatic amyloid-β clearance [2, 36]. When slow-wave sleep and thus glymphatic clearance are disrupted, for example, by cortical arousal and breathing disturbances, amyloid-β might accumulate and then ultimately affect brain metabolism, structure and function. The relevance of slow-wave sleep has also been highlighted in previous studies [9, 10], which showed that age-related fragmentation of slow-wave sleep is associated with lower cognitive performance and gray matter alterations in the frontal lobe.

The analyses of cognitive data revealed a significant association between the breathing disturbance and arousal-related PSG component 1 and lower performance on the TMT and Stroop Test, while the association with the ANT did not reach statistical significance in a dichotomous sense (on the model level). However, it has been argued [37, 38] that a classical dichotomous decision of significance based on p-values falls short of taking into account factors like the previous evidence, study design, data quality, and others. A lower cognitive performance across several domains would generally be in line with meta-analyses [39, 40] and reviews [41, 42] showing lower cognitive performance in patients with sleep apnea.

Furthermore, we extracted the previously identified gray matter cluster related to arousal and breathing disturbance and observed significant associations with all three cognitive tests. This finding is of particular interest because it highlights the important role of the frontal pole (Brodmann area 10) in both sleep and cognitive functioning. Evidence from nonhuman primates suggests that the frontal pole has several efferent [43] (e.g. cingulum and superior temporal gyrus) and afferent [44] connections (e.g. amygdala, thalamus, and basal forebrain). These connections support the theory that the frontal pole may serve as a gateway for integrating cognitive processes across multiple domains, which has been suggested by a meta-analysis [45] and a review [46] of task-related imaging studies.

Limitations of the present findings include the hospital-based PSG, which took place in a completely different setting compared with the patients’ home bedrooms. Such an unfamiliar hospital setting may compromise some sleep characteristics to a certain degree [47], along with the set-up of the patient monitoring cables before the PSG session. Due to logistical constraints, it was not possible to include a habituation night. Furthermore, the scoring guidelines of the AASM have changed over time, which hampers the comparison of studies [48, 49]. The present results should not imply that a single night of sleep disrupted by cortical arousal and breathing disturbances is associated with a lower gray matter volume. It is rather assumed that the observed lower volume could be a cumulative long-term effect of disrupted sleep. However, the present study is cross-sectional and thus does not allow conclusions regarding the causal relationship. Furthermore, the age range of the BiDirect Study does not cover very young or very old adults in whom sleep patterns tend to change [50]. Major strength of the present analysis is the large, population-based sample and the comprehensive examination program including PSG as well as magnetic resonance imaging, which is a rare combination in the field of sleep medicine.

In conclusion, the presented results advance our understanding of disrupted sleep, particularly with regard to breathing disturbances, cortical arousal, and its associations with lower gray matter volume. Furthermore, these sleep characteristics were also associated with lower cognitive performance. Additional research is demanded to further investigate the utility of aggregated sleep characteristics using dimensionality reduction techniques.

Funding

The present work was supported by grants from the German Federal Ministry of Education and Research (BMBF; grants FKZ-01ER0816 and FKZ-01ER1506).

Disclosure statement

Financial disclosure: none.

Nonfinancial disclosure: none.

References