Regional Neocortical Gray Matter Structure and Sleep Fragmentation in Older Adults

Participants

We studied 141 participants from the MAP, a community-based longitudinal study of the chronic conditions of aging, which began in 1997.²² Actigraphy was added to the assessment protocol in 2005,²³ and a subset of the MAP cohort has undergone structural brain magnetic resonance imaging (MRI) beginning in 2009.²⁴ For these analyses, we examined data from participants with at least 7 consecutive days of actigraphy and a brain MRI obtained in the same year. We excluded participants with mild cognitive impairment (MCI) or dementia of any cause, participants with a history of stroke, and participants taking sedative or hypnotic medications. At the time of these analyses, a total of 1,178 MAP participants had at least one actigraphic recording with at least 7 consecutive days of usable data (see next paragraphs for quality control procedures). Of these participants, 272 had an MRI at the same study year as the acti-graphic recording, 69 had MCI or dementia, 43 had a history of stroke, and 19 were taking sedative or hypnotic medications at the time of actigraphy, leaving 141 participants whose data were included in these analyses.

Statement of Ethics Approval

This study was approved by the institutional review board of the Rush University Medical Center. Written informed consent was obtained from all participants.

Assessment of Sleep Fragmentation

Sleep fragmentation was assessed by actigraphy. Actigraphs (Actical, Philips Respironics, Bend, OR, USA) were set to record in 15-sec epochs and placed on participants' nondominant wrist by study staff. Participants were instructed to leave the device on their wrist until study staff returned to remove them after 10 d. Periods of suspected actigraph removal were flagged in a two-stage process. First, actigraphic records were visually examined by trained technicians, and periods of suspected removal were flagged. Then, using automatic algorithms, periods of 4 h or greater with no movement whatsoever were also flagged. We analyzed only recordings in which there were at least 7 consecutive days of recording without any flagged periods. When more than 7 d of recording were available, we analyzed only the first 7 consecutive days, ensuring the presence of exactly 5 weekdays and 2 weekend days in each recording.

Sleep fragmentation, the extent to which sleep is interrupted by repeated awakenings, was quantified using the metric k_RA as described and validated elsewhere.^25–27 Briefly, k_RA is the probability per 15-sec period of having an arousal, as indicated by movement (a nonzero activity count) after a sustained (≥ ∼5 min) period of rest (i.e., sleep). A higher k_RA indicates greater sleep fragmentation. Although k_RA is conceptually distinct from other more common metrics of sleep fragmentation such as sleep efficiency (SE) or wake time after sleep onset (WASO) that summarize the total amounts of wake and sleep across the night rather than describing their temporal distribution, we have previously shown that k_RA is moderately correlated with both WASO (r = 0.4) and SE (r = 0.6).²⁷ In the current study, participants were not asked to keep diaries tracking time in bed. Because calculation of WASO or SE is dependent on knowing time to bed and out of bed, WASO and SE could not be calculated in this study.

Self-Report Sleep Measures

Self-report sleep measures were also obtained. Participants were asked about their estimated average daily total sleep time. As well, participants were asked how often they were troubled by waking up in the night, with responses on an ordinal scale ranging from 0 (“never”) to 5 (“very often”).

Magnetic Resonance Imaging

We obtained T1-weighted anatomical data using a 1.5 Tesla MRI scanner (General Electric, Waukesha, WI, USA). We used a three-dimensional magnetization-prepared rapid acquisition gradient echo sequence with parameters echo time (TE) = 2.8 msec, repetition time (TR) = 6.3 msec, preparation time = 1000 msec, flip angle 8°, field of view 24 × 24 cm, 160 slices, 1-mm slice thickness, a 224 × 192 acquisition matrix reconstructed to 256 × 256, and two repetitions.

The MRI data were averaged and automatically segmented with FreeSurfer 5.0 (http://surfer.nmr.mgh.harvard.edu). The intracranial volume, total gray matter volume, total cortical gray matter volume, total subcortical gray matter volume, total white matter volume, and volumes of 34 cortical and 9 sub-cortical structures²⁸ were obtained. For the cortical and sub-cortical regions, we considered the left and right hemispheres separately. We flagged the volume of regions the algorithm was unable to reliably segment due to the lack of sufficient image contrast. Thirty-six scans had at least one flagged volume, and a total of 80 flagged volumes were excluded from analysis. Regions with flags in more than two scans included the insula (25 scans), the superior temporal gyrus (five scans) and the temporal pole (five scans). The remaining regions were flagged in two or fewer scans. For secondary analyses, we also measured mean regional cortical thickness (in mm) and mean regional cortical surface area in cm² normalized to total intracranial volume (ICV).

Other Covariates

Age was computed from the self-reported date of birth and the date of actigraphy. Sex and education were recorded at the baseline interview. Presence or absence of congestive heart failure, peripheral vascular disease, diabetes, smoking, and hypertension were determined by history. The presence/absence of a history of stroke was determined by a clinician considering all available clinical information including physical examination. Vascular diseases were summarized as the sum of congestive heart failure, peripheral vascular disease, and stroke. Vascular risk factors were summarized as the sum of diabetes, smoking, and hypertension. Medications were inspected and coded using the Medi-Span system (Wolters-Kluwer Clinical Drug Information, Alphen aan den Rijn, Netherlands). Depressive symptoms were assessed with a 10-item version of the Center for Epidemiologic Studies Depression Scale.²²

As previously described, a composite measure of global cognitive function was computed on the basis of 19 cognitive tests as administered annually by trained technicians.²⁹ This measure is scaled such that 0 represents the mean score of all MAP participants at study baseline, positive scores indicate better performance, and 1 unit represents approximately 1 standard unit of performance. Individuals were classified as having Alzheimer disease or not according to National Institute of Neurological and Communicative Diseases and Stroke/ Alzheimer's Disease and Related Disorders Association criteria³⁰ using all available cognitive and clinical data as previously described.³¹

Statistical Analysis

All 68 regional gray matter volumes (34 regions, left and right) as well as total cortical and subcortical gray matter volume, and total white matter volume, were corrected for ICV and expressed as %ICV. Each volume was then standardized to the mean and standard deviation (SD) for that region across all scans. Sleep fragmentation measured by k_RA was also standardized to the mean and SD for all participants. For total cortical gray matter volume, as well each cortical regional gray matter volume, we considered a linear regression model, adjusted for age, sex, and education, with normalized k_RA as the outcome, and standardized regional gray matter volume as the predictor. To account for multiple comparisons, we generated 10,000 permuted null datasets by permuting scans between participants. For each region, we calculated an empirical P value, corrected for multiple comparisons, by determining the proportion of null datasets with at least one P value smaller than the observed P value for that region. We also estimated the false discovery rate as follows. For each observed P value, we first determined the number of regions per null dataset with P values less than or equal to the observed P value, and we then averaged them across all 10,000 permuted null datasets. The mean estimate gave the expected number of false positives. The false discovery rate (FDR) was then calculated as the expected number of false positives divided by the number of observed positives, i.e., the number of regions in the observed dataset with P values less than or equal to the P value under consideration. In secondary analyses, we repeated the aforementioned analysis for the 18 subcortical regions (nine regions, left and right).

For cortical regions associated with k_RA at an estimated FDR < 0.05, we examined the regional specificity of these associations by considering models augmented with a term for total cortical gray matter volume.

For regions associated with k_RA at an estimated FDR < 0.05, we assessed the effect of a number of potential confounders in augmented models. Depression may be related to both sleep fragmentation and to regional cortical gray matter volume. To examine for potential confounding by depression, we augmented the models above with a term for number of depressive symptoms, and repeated the above analysis. Medical co-morbidities, for example, vascular risk factors such as diabetes, smoking, and hypertension, or diseases such as coronary artery disease and peripheral vascular disease, may plausibly affect both cortical gray matter volumes and sleep. To examine this possibility, we augmented the aforementioned models with terms for number of vascular diseases, or number of vascular risk factors, and repeated the aforementioned analyses. Cognitive function has been shown to be related to both regional gray matter volume,³² and to sleep fragmentation.²⁶ To explore this further, we considered models augmented with a term for composite global cognition as measured previously. Finally, we considered models augmented with terms for all covariates together.

In secondary analyses, we performed an identical analysis to that previously described, but relating k_RA to regional cortical thickness and surface area instead. As well, to examine associations between regional gray matter volumes and self-reported sleep duration and frequency of nighttime awakenings, we repeated these analyses for the self-reported sleep measures, using linear regression models for self-reported sleep duration, and ordinal logistic regression models for frequency of nighttime awakenings.

All statistical analyses were performed using R programming language.³³ Visual inspection of diagnostic residual plots confirmed model assumptions.

Characteristics of the Study Participants

Data from a total of 141 participants were included in this study. Their clinical characteristics at the time of actigraphy are summarized in Table 1. Participants were mostly female, had a median age of 83 y, a median Mini-Mental State Examination score of 29, and a global cognition score above 0. They were also relatively healthy, with a median of no depressive symptoms, no vascular diseases, and one vascular risk factor (which for most was hypertension).

Total Cortical and Subcortical Gray Matter Volume

We first examined the relationship between total gray matter volume and sleep fragmentation measured by the actigraphic metric k_RA. In a linear regression model adjusted for age, sex, and education, lower total gray matter volume was associated with higher sleep fragmentation (estimate +0.23 SD of sleep fragmentation per 1 SD lower total gray matter volume, SE 0.11, P = 0.037). In contrast, total white matter volume was not associated with sleep fragmentation (estimate −0.07 SD of sleep fragmentation per 1 SD lower total white matter volume, SE 0.08, P = 0.35). We then examined the separate associations of subcortical and cortical gray matter volumes with sleep fragmentation. Although total subcortical gray matter volume was not associated with sleep fragmentation (estimate −0.01 SD of sleep fragmentation per 1 SD lower subcortical gray matter volume, SE 0.09, P = 0.90), lower total cortical gray matter volume was associated with higher sleep fragmentation (estimate +0.29 SD of sleep fragmentation per 1 SD lower cortical gray matter volume, SE 0.11, P = 0.0096).

Regional Cortical Gray Matter Volumes

Given the significant associations between total cortical gray matter volume and sleep fragmentation, we next examined the relation between sleep fragmentation and gray matter volume in individual cortical regions. Lower gray matter volume in 26 of the 68 cortical regions was nominally associated with higher sleep fragmentation (Table 2) at a threshold of P < 0.05. Of these 26 regions, four (the right and left inferior frontal gyrus pars orbitalis, and the right and left lateral orbitofrontal gyrus; Figures 1 and 2) were associated with sleep fragmentation at a FDR of less than 0.05. All four regions remained statistically significant when outliers (volumes more than 3 SD from the mean) were removed (Table S1, supplemental material). In contrast to regional gray matter volumes, only two of the 18 sub-cortical regions were nominally associated with higher sleep fragmentation (the left and right nucleus accumbens; Table S2, supplemental material) at a threshold of P < 0.05. Of these two regions, neither was associated with sleep fragmentation at a FDR of less than 0.05.

To assess the regional specificity of the cortical findings, we augmented the models with a term for total cortical gray matter volume. The associations between k_RA and regional volumes in both inferior frontal gyri pars orbitalis and left orbitofrontal gyrus remained significant in these models (right inferior frontal gyrus pars orbitalis: estimate +0.42 SD of sleep fragmentation per 1 SD lower gray matter volume, SE 0.15, P = 4.9 × 10⁻³; left inferior frontal gyrus pars orbitalis: estimate +0.26 SD of sleep fragmentation per 1 SD lower gray matter volume, SE 0.11, P = 1.7 × 10⁻²; left orbitofrontal gyrus: estimate +0.40 SD of sleep fragmentation per 1 SD lower gray matter volume, SE 0.16, P = 1.1 × 10⁻²) indicating that these associations were independent of global effects on cortical volumes. In these models, the association between k_RA and gray matter volume in the right lateral orbitofrontal gyrus was somewhat attenuated (estimate +0.27 SD of sleep fragmentation per 1 SD lower gray matter volume, SE 0.15, P = 8.5 × 10⁻²), suggesting that the association in this area may be in part reflective of global effects.

Effect of Clinical Covariates

For the four regions whose gray matter volumes were associated with sleep fragmentation at an FDR threshold of 0.05 (the left and right lateral orbitofrontal gyrus, and left and right inferior frontal gyrus pars orbitalis), we examined the effect of potential confounders (Table 3). Depression is plausibly related to both regional cortical volume, and to sleep. In models augmented with a term for number of depressive symptoms, the associations between gray matter volumes in these four regions and sleep fragmentation were not substantially changed (Table 3, Model B). Medical health can also plausibly affect both cortical gray matter structure, and sleep. However, addition of terms for vascular diseases (coronary artery disease, stroke, peripheral vascular disease) and vascular risk factors (smoking, hypertension, diabetes) did not substantially alter the associations between sleep fragmentation, and gray matter volumes in these regions (Table 3, Models C and D). Cognition can be associated with both regional cortical gray matter volume,³² and with sleep fragmentation.^26,27 However, the associations between sleep fragmentation and cortical gray matter volume in these four regions remained statistically significant at P < 0.05 after adjustment for a composite cognition measure (Table 3, Model E). Finally, after addition of terms for all of these covariates together, the associations between sleep fragmentation and cortical gray matter volume in these four regions remained statistically significant (Table 3, Model F)

Cortical Thickness and Cortical Surface Area

In secondary analyses, we examined associations between sleep fragmentation k_RA and regional cortical thickness and surface area using an identical approach to that used for regional cortical volumes. Lower cortical thickness in 26 of the 68 regions was nominally associated with higher sleep fragmentation (Table S3, supplemental material) at a threshold of P < 0.05. This list included the left and right lateral orbito-frontal cortex and the right inferior frontal gyrus pars orbitalis. However, only one region, the right inferior frontal gyrus cortex pars orbitalis, had an estimated FDR < 0.05. Lower cortical surface area in two of these 68 regions was nominally associated with higher sleep fragmentation (Table S4, supplemental material) at a threshold of P < 0.05 including the left inferior frontal gyrus pars orbitalis. However, none had an estimated FDR < 0.05.

Self-Report Sleep Measures

In multivariable linear regression models adjusted for age, sex, and education, none of the 68 regions was associated with self-reported total sleep time at an FDR < 0.05. Similarly, in multivariable ordinal logistic regression models adjusted for age, sex, and education, none of the 68 regions was associated with self-reported frequency of nocturnal awakenings at an FDR < 0.05.