The association between white matter and sleep spindles differs in young and older individuals

Abstract

Study Objectives

Sleep is a reliable indicator of cognitive health in older individuals. Sleep spindles (SS) are non-rapid eye movement (NREM) sleep oscillations implicated in sleep-dependent learning. Their generation imply a complex activation of the thalamo-cortico-thalamic loop. Since SS require neuronal synchrony, the integrity of the white matter (WM) underlying these connections is of major importance. During aging, both SS and WM undergo important changes. The goal of this study was to investigate whether WM integrity could predict the age-related reductions in SS characteristics.

Methods

Thirty young and 31 older participants underwent a night of polysomnographic recording and a 3T magnetic resonance imaging acquisition including a diffusion sequence. SS were detected in NREM sleep and EEG spectral analysis was performed for the sigma frequency band. WM diffusion metrics were computed in a voxelwise design of analysis.

Results

Compared to young participants, older individuals showed lower SS density, amplitude, and sigma power. Diffusion metrics were correlated with SS amplitude and sigma power in tracts connecting the thalamus to the frontal cortex for the young but not for the older group, suggesting a moderation effect. Moderation analyses showed that diffusion metrics explained between 14% and 39% of SS amplitude and sigma power variance in the young participants only.

Conclusion

Our results indicate that WM underlying the thalamo-cortico-thalamic loop predicts SS characteristics in young individuals, but does not explain age-related changes in SS. Other neurophysiological factors could better explain the effect of age on SS characteristics.

Statement of Significance

Neuronal activity during sleep spindles impacts brain long-term functionality, including memory and learning. The strong spindle reduction observed in older individuals is linked to significant functional consequences, and the underlying mechanisms remain unknown. Since spindle generation relies on thalamo-cortical connections, we investigated whether lower white matter (WM) integrity may explain age-related spindle modifications. We showed that higher integrity between the thalamus and the frontal cortex predicted higher spindle amplitude in young individuals but not in older participants. These results indicate that in young participants, better WM connections contribute to the higher neuronal synchrony associated with spindles. Since WM does not explain lower spindles in older individuals, further studies should investigate cerebral and metabolic factors possibly underlying these age-related modifications.

Introduction

The sleep-wake cycle is an indicator of cognitive health in older individuals and is a potential biomarker of cognitive decline [1–4]. Demystifying the links between sleep microarchitecture and cerebral changes in aging is necessary for the development of new diagnostic tools and preventive strategies, and for the identification of therapeutic targets. Sleep spindles (SS) are among the most studied electroencephalographic (EEG) patterns that occur during sleep. They represent bursts of 12–16 Hz EEG oscillatory brain activity in non-rapid eye movement (NREM) sleep, predominantly in centro-parietal regions [5, 6]. Growing interest in SS is due to its well-documented role in cognitive functioning. Neuronal activity during SS modulates neural responses to stimuli, allowing the sleeping brain to be more or less permeable to them, and impacts the long-term functionality of the brain, including memory and learning [6–10].

Several studies showed a strong reduction of SS density, amplitude, and duration with age [11, 12]. Research indicates that, compared to younger individuals, middle-aged and older participants have lower SS density, amplitude, and duration [11, 12]. Moreover, age-related differences in SS density and amplitude are more prominent in anterior derivations, whereas age-related effects on SS duration are stronger in more posterior derivations [12]. Recent studies indicated that age-related decrease in SS characteristics are associated with poorer memory and cognitive functions [13, 14]. For instance, one study reported that the reduction in SS in older individuals is related to lower sleep-dependent motor memory consolidation and decreased cortico-striatal network activity [14]. Furthermore, higher SS density predicts better cognitive performance in healthy older individuals, whereas lower SS density and amplitude are associated with dementia development in Parkinson’s disease patients [15, 16].

SS generation relies on a complex activation of the thalamo-cortico-thalamic loop (see ref. [17] for a review). Specifically, anatomical and physiological studies showed that activation of the GABAergic neurons of the thalamic reticular nucleus (TRN) causes highly synchronized and rhythmic burst of inhibitory firing in the thalamo-cortical (TC) neurons mainly in dorsal thalamic nuclei [18]. Post-inhibitory rebound activity in TC neurons generates glutamatergic excitatory potentials in both the cortex and the TRN, preparing them for the next firing burst [6, 19, 20]. Although thalamic nuclei are central to SS generation, the entire thalamo-cortico-thalamic loop is necessary for synchronization and termination of SS [21]. Indeed, as successive volley of excitatory TC input increases cortical activity, it also increases cortico-thalamic feedback to the TRN and dorsal thalamic nuclei. This excitatory feedback synchronizes successive cycles of activity between TRN and TC neurons [22], but also increases firing rate in dorsal thalamic nuclei desynchronizing firing of cortico-thalamic projections. Ultimately, this disruption of intrathalamic oscillations causes SS waning [23–25].

The connections between the thalamus and the neocortex rely on white matter (WM) projections. Diffusion tensor imaging (DTI), a magnetic resonance imaging (MRI) technique assessing the diffusion of water molecules in each voxel, is a sensitive measure of underlying WM microstructure [26]. Piantoni and colleagues [27] evaluated the relationship between WM properties and SS variables (SS power and density). In a group of 15 healthy young participants, they found that higher SS power was associated with higher axial diffusivity (AD: higher diffusion of water molecules along the fiber orientation) in major frontal WM tracts and in diffuse regions throughout the brain, including subcortical connections with the thalamus. Their results also indicated that higher SS density (number of SS per minute of NREM sleep) was associated with lower AD only in the temporal cortex. These results suggest that SS density and power not only reflect the dynamic of functional connectivity between brain neocortical regions but are also positively correlated with markers of WM integrity [27, 28].

It is well established that normal aging is associated with changes in WM microstructure suggesting lower WM integrity [29–32]. Age-related changes in WM appear gradually during adulthood (30 to 50 years old) and accelerate between 60 and 65 years old [32–34]. Compared to young individuals, older participants showed lower fractional anisotropy (FA: a measure of the coherence of water molecules along a specific axis) and higher mean diffusivity (MD: a global measurement of the diffusivity of water molecules) in all major WM tracts of the brain, with the frontal areas showing a particular vulnerability to age-related changes, and the caudal brain being the least affected [31, 34–40]. Studies also showed an increase in AD and radial diffusivity (RD: water diffusion transverse to the fiber orientation) with aging in less diffuse regions, although an opposite relationship with age has also been reported in relatively isolated cerebral regions [31, 38, 41, 42].

The goal of this study was to investigate whether WM integrity, as measured with DTI, could explain age-related changes in SS variables. We hypothesized that widespread age-related WM changes will be associated with a decline in SS characteristics (density and amplitude). This association will be stronger for regions involved in the generation and the propagation of spindles such as the TC tract and frontal cortex, due to the fact that these regions exhibit the most prominent age effects.

Methods

Participants

Sixty-one participants completed the research protocol: 30 young (20 to 30 years old; 22.93 ± 2.76 years old; 14 women) and 31 older (50 to 70 years old; 59.80 ± 5.40 years old; 18 women). The two groups did not differ on years of education (young = 15.43 ± 2.21 years; older = 15.45 ± 3.30 years) or on the number of men and women in each group (X²(1) = 0.794, p > 0.05). Information related to exclusion criteria was collected with the help of an in-house questionnaire and a semi-structured interview. Exclusion criteria included smoking, body mass index (BMI) > 27, using drugs or medication known to affect the sleep-wake cycle or the nervous system, complaints about the sleep-wake cycle or cognition, habitual sleep duration <7 hours or >9 hours, night shifts or transmeridian travels within 3 months prior to the study, and history of neurological or psychiatric illness. Participants with a sub-clinical score (>13) on Beck Depression Inventory [43], or on a Beck Anxiety Inventory (>7) [44], were also excluded. Potential cognitive impairment or dementia diagnosis, as measured by neuropsychological assessment, was ruled out for each participant. The complete neuropsychological evaluation thoroughly assessed the cognitive functions known to decline during normal aging including working memory, attention, executive functions, episodic memory, visuospatial functions, as well as processing speed [45]. Premenopausal women reporting regular menstrual cycles (25–32 days) during the year preceding the study and menopausal women without hormonal therapy were included. No woman reported vasomotor complaints (e.g. hot flashes, night sweats). Each participant underwent one adaptation and screening polysomnographic (PSG) night at the laboratory. This PSG evaluation included EEG, leg electromyogram (EMG), thoracoabdominal plethysmograph, and oral/nasal cannula. Participants were excluded if they showed an index >10 per hour of sleep for sleep apneas/hypopneas or for periodic leg movements index associated with microarousals. The Hôpital du Sacré-Coeur de Montréal Institutional Review Board and the Unité de Neuroimagerie Fonctionnelle (UNF) research ethics mixed committee approved the protocol. All participants gave written consent prior to the study and received monetary compensation for their participation.

PSG recordings, spectral analysis and spindle detection

After the screening night, all participants underwent a night of PSG recording. A Grass Model 15A54 amplifier system (Natus Neurology, Warwick, Rhode Island) recorded EEG, chin EMG, electrooculogram, and electrocardiogram signals. EEG included 20 EEG derivations (Fp1, Fp2, Fz, F3, F4, F7, F8, Cz, C3, C4, Pz, P3, P4, Oz, O1, O2, T3, T4, T5, T6) referred to linked earlobes (10–20 international system; EEG: gain 10000; bandpass 0.3–100 Hz; −6 dB). Signals were digitalized at a sampling rate of 256 Hz using commercial software (Harmonie, Stellate Systems, Montreal, Quebec, Canada). A sleep technician visually scored the sleep stages (N1, N2, N3, and REM) according to standard AASM criteria [46] using 30-second epochs. Artifacts were detected by an automatic algorithm [47] and visually verified by a trained technician.

A Fast Fourier Transform (cosine tapering) was performed on each derivation, on 5-second artifact-free sections to calculate an averaged spectral power in the sigma band (11–15 Hz) during NREM sleep. SS were automatically detected on artifact-free NREM epochs in parasagittal derivations (Fp1, Fp2, F3, F4, Fz, C3, C4, Cz, P3, P4, and Pz) using a previously published detection algorithm [12, 15]. Specifically, EEG data were bandpass filtered between 11.1 and 14.9 Hz using a linear phase finite impulse response filter (–3 dB at 11.1 and 14.9 Hz). Data were forward and reverse filtered to obtain zero-phase distortion and to double the filter order. The root mean square of the filtered signal was then calculated with a 0.25-second time window and thresholded at the 95th percentile [48]. A SS was detected when at least two consecutive root mean square time points exceeded the threshold duration criterion (0.5 second). SS were detected in NREM N2 sleep stage. SS density (number/minute of N2 sleep) and mean SS amplitude (expressed in µV) were calculated. Sigma and SS variables were averaged over frontal (Fp1, Fp2, F3, F4, and Fz) and posterior (C3, C4, Cz, P3, P4, and Pz) electrodes. Two-way ANOVAs with two independent factors (two age groups: young and older; two sex groups: men and women) were performed to investigate interactions between age and sex on sleep variables (PSG, sigma power and SS characteristics). Simple effects were decomposed for significant interactions.

Diffusion tensor imaging

MRI acquisitions were performed using a 3T Siemens Trio MRI scanner (Siemens Medical Systems, Erlengan, Germany) at the UNF of the Research Centre of the Institut Universitaire de Gériatrie de Montréal. Diffusion MRI data were acquired using an echo planar imaging (EPI) sequence with the following parameters: repetition time (TR) = 12700 ms, echo time (TE) = 100 ms, bandwidth = 1302 Hz/Px; 128 × 128 acquisition matrix, 75 slices, antero-posterior acquisition; one reference image at b = 0 s/mm² and 64 diffusion-weighted images at b = 700 s/mm²; 256 × 256 × 150 mm FOV and 2 mm isometric voxel size.

DTI data processing

WM diffusion variables were assessed using a DTI model. First, data denoising was carried out using a non-local means algorithm [49]. Data were then upsampled at 1 mm isometric for further analysis and were corrected for motion artifacts and eddy current distortions using the FMRIB Software Library (FSL) tools (FSL 5.0 [50]). Moreover, diffusion data was visually inspected by a trained research assistant in order to detect any major brain distortion or data abnormalities. The amount of absolute displacement caused by motion, and eddy current distortions were retrieved for each subject. The two age groups did not significantly differ on the amount of displacement corrected in our preprocessing (t(1,59) = −1.368, p = 0.176) indicating that young and older participants were comparable. Moreover, in order to assure that the amount of registration would not contribute to the differences found between young and older individuals, the same regression analyses were carried out using our original diffusion data in their native space. To do so, we used limited linear transformations to align the brain of all participants. Similar significant statistical patterns of results were found, suggesting that data registration during the preprocessing of our data did not influence the differential results between young and older individuals. Diffusion tensors were fitted to diffusion data in each voxel using the same toolbox. Maps were then computed for the diffusion variables of interest to our study. Diffusion tensors can be characterized with three orthogonal eigenvectors and three eigenvalues, from which are calculated the diffusion metrics (FA, MD, AD, and RD) used for the analyses. In each voxel, FA is defined as the normalized variance of all three orthogonal vectors and represent how coherent is the water molecule diffusion along a specific orientation. MD is defined as the mean of the three eigenvalues. AD is defined as the tensor’s largest eigenvalue whereas RD is defined as the mean of the two others eigenvalues. In diffusion terms, MD represents the intensity of the diffusion tensor while AD and RD designate respectively the diffusion parallel and perpendicular to the principal axis. Voxelwise statistical analyses were then carried out using tract-based spatial statistics (TBSS) [51]. TBSS allows whole-brain group analyses by first aligning and registering FA maps to a standard MNI152 template using nonlinear registration. Then, a WM skeleton of only major WM fiber bundles is created by averaging individual maps including only voxels with FA > 0.3. Finally, individual DTI variables (FA, MD, AD, and RD) are then projected back onto this WM skeleton allowing further voxelwise group analyses. TBSS analyses were processed separately in each group of interest. The WM skeleton contained 96118 voxels and 91146 voxels, respectively for the young and the older groups. In order to get an idea of the extent of the results, significant WM results are also presented as a percentage of the TBSS WM skeleton.

DTI statistical analyses

WM statistics were carried-out using FSL tool Randomise with 10000 permutations, a general linear model for nonparametric inferences [52]. Preliminary analyses were carried out for testing the importance for controlling for sex and intracranial volume (ICV) in the variability of WM measures. When significant correlations with WM variables were found, sex and ICV were added to the statistical model as covariates. The first set of analyses assessed age group differences for each WM diffusion metric. The design matrix for this analysis considered age as a categorical variable and added the adequate corrections (sex and/or ICV). For the second set of analyses, correlations between SS variables and WM diffusion metrics were computed in each age group separately. The design matrices included a column of ones to capture the intercept and a column for the SS variables transformed in z-scores. For both sets of analyses, threshold-free cluster enhancement (TFCE) correction was used to correct for multiple comparisons [53]. In short, this approach enhances areas of signal exhibiting spatial contiguity which allow to better discriminate the voxels within cluster-like regions from the background noise. Cluster extent of 125 contiguous voxels was considered to filter significant clusters and voxelwise significance threshold of p < 0.05 was used in interpreting WM results. FSL toolbox and JHU White-Matter Tractography atlas were used for tract identification with a probability of region overlap threshold of 2%.

Moderation analyses

Absence of significant correlations between diffusion metrics and SS variables in each age group prevented us to compute mediation analyses. Based on this, we investigated a post-hoc moderating effect. More precisely, we performed moderation analysis to test the differences in the relationship between diffusion variables and SS characteristics in young and older individuals [54]. Thereby, a third set of analyses used hierarchical linear regressions to test interactions between age and WM diffusion metrics in the prediction of SS variables, i.e. does the relationship between WM and SS variables significantly differs in the young and the older groups? Moderation analyses were performed for SS posterior density, frontal amplitude, and sigma spectral power since they showed a significant correlation with diffusion metrics in at least one age group. Specific regions of interest were defined as WM clusters showing significant correlations with SS frontal amplitude or sigma power. These clusters were used to select voxels from the original diffusion data for each subject. Averaged values for diffusion metrics across each region of interest were then used in hierarchical linear regression models. All variables used in the model were mean-centered for regression analyses to predict SS variables. In step 1 of the hierarchical regression model, sex and ICV were entered as controlled variables. Age group and diffusion metric were then entered in step 2. Finally, the interaction between age and diffusion metric was entered in step 3. Significant interactions were decomposed according to previously described procedures [55].

Supplementary analyses

Using the FSL tool Randomise, supplementary voxelwise model of analyses was performed to test group differences with continuous covariate interaction. This statistical model is corresponding to an ANCOVA and test whether the linear relationship between the WM diffusion metrics and SS amplitude differs between young and older individuals by comparing the regression slopes between both groups. A first contrast evaluated which voxels showed a larger slope in the young as compared to the older individuals. A second contrast evaluated which voxels showed a larger slope in the older as compared to the younger participants.

Results

Age and sex effects on SS variables

Age group and sex differences on SS variables are presented in Table 1. Significant interactions between age and sex were found for frontal (F(1,57) = 8.1, p < 0.01) and posterior (F(1,57) = 10.5, p < 0.01) SS density. Both older men and women showed lower frontal SS density than young men and women but the age effect was more prominent in men (F(1,57) = 50.6, p < 0.001) than in women (F(1,57) = 11.0, p < 0.01). Posterior SS density was significantly lower in older men as compared to younger men (F(1,57) = 38.2, p < 0.001) but no significant age effect was found for the women. Amplitude of frontal SS and posterior SS was significantly higher in young as compared to older individuals (fontal SS amplitude: F(1,57) = 33.1, p < 0.001; posterior SS amplitude: F(1,57) = 17.1, p < 0.001) and in women as compared to men (frontal SS amplitude: F(1,57) = 20.2, p < 0.001; posterior SS amplitude: F(1,57) = 13.8, p < 0.001) with no significant interaction between age and sex. Similar effects were found for frontal and posterior sigma spectral power: young participants showed higher power than older participants (fontal sigma power: F(1,57) = 20.9, p < 0.001; posterior sigma power: F(1,57) = 16.0, p < 0.001) and women showed higher power compared to men (fontal sigma power: F(1,57) = 14.7, p < 0.001; posterior sigma power: F(1,57) = 8.9, p < 0.01). Age and sex effects on PSG variables are presented in Supplementary Table 1.

Table 1.

Characteristics of sleep spindles in young and older participants

	Young	Older	Main effect (p values)			Effect
	N = 30	N = 31	Age	Sex	Interaction
Spindle density (nb/minute)
Frontal	3.8 ± 0.4	3.2 ± 0.3	N/A	N/A	0.006	(M) YO > OL (W) YO > OL
Posterior	3.8 ± 0.4	3.3 ± 0.3	N/A	N/A	0.002	(M) YO > OL (W) n.s.
Spindle amplitude (μV)
Frontal	36.2 ± 11.4	24.4 ± 8.3	0.0001	0.0001	n.s.	YO > OL W > M
Posterior	39.7 ± 10.4	31.0 ± 10.1	0.0001	0.0001	n.s.	YO > OL W > M
Sigma power
Frontal	18.0 ± 11.9	9.0 ± 5.9	0.0001	0.0001	n.s.	YO > OL W > M
Posterior	22.1 ± 11.6	13.3 ± 8.0	0.0001	0.004	n.s.	YO > OL W > M

Data expressed as mean ± SD. p values were considered significant at p < 0.05. M, men; N/A, non-applicable; n.s., non-significant; OL, older participants; W, women; YO, young participants.

Age differences on WM diffusion metrics

Significant age differences for the four diffusion metrics are shown in Figure 1. As compared to older participants, young individuals showed significantly higher FA (45.6% of skeleton, p < 0.05) and AD (43.6% of skeleton, p < 0.05) in all major WM tracts of the brain. More specifically, young individuals showed higher FA in forceps minor and bilateral inferior fronto-occipital fasciculus, and higher AD in forceps minor, forceps major, and right inferior fronto-occipital fasciculus as compared to older individuals. Young participants also showed lower AD in a circumscribed area of the right inferior fronto-occipital fasciculus (0.3% of the skeleton). Young individuals showed lower RD (22.2% of skeleton, p < 0.05) and MD (2.9% of skeleton, p < 0.05) compared to older individuals. Age-related RD differences were found in the main frontal and parietal WM tracts overlapping bilateral anterior thalamic radiation, forceps minor, bilateral inferior fronto-occipital fasciculus, bilateral superior longitudinal fasciculus, right inferior longitudinal fasciculus, and right cingulum and parts of the cingulate gyrus. MD age differences were more limited, and included left anterior thalamic radiation, forceps minor, left inferior fronto-occipital fasciculus, left uncinate fasciculus, and parts of the left cingulum and cingulate gyrus. Clustered WM differences between young and older participants are presented in Table 2.

Figure 1.

White matter differences between young and older participants. This figure shows thresholded significant voxels (1-p > 0.95 corrected) overlaid on group-specific white matter skeleton in dark gray and standard MNI152 template. According to the convention, the right hemisphere is represented on the left side. Warm colors represent voxels where young participants present higher diffusion metric than older participants whereas cold colors represent voxels where older participants had significantly higher diffusion measures compared to young participants. Given the small size of the cluster, voxels where older participants showed higher AD than young individuals were circled in white.

Table 2.

Clustered white matter differences between young and older participants

				Max coordinates (mm)
	Effect	Voxels	1-p Max	X	Y	Z	Locations (>2% probability)
FA
Cluster 1	YO > OL	42381	0.999	−15	18	−16	Forceps minor (3.8%), Left IFOF (2.2%), Right IFOF (2.5%)
MD
Cluster 1	OL > YO	1291	0.969	−31	35	7	Left ATR (9.5%), Forceps minor (9.1%), Left IFOF (11.1%), Left UF (5.0%)
Cluster 2	OL > YO	840	0.955	−25	−53	27	Left Cingulum/Cingulate gyrus (2.3%)
Cluster 3	OL > YO	367	0.952	−13	33	39	Left ATR (2.0%), Forceps minor (15.3%)
Cluster 4	OL > YO	125	0.951	−22	21	−8	Left IFOF (8.3%), Left UF (10.4%)
AD
Cluster 1	YO > OL	37373	0.999	−34	1	−34	Forceps Minor (3.6%)
Cluster 2	YO > OL	2993	0.997	−27	−53	−40	Right corticospinal tract (2.2%)
Cluster 3	YO > OL	259	0.955	20	−82	23	Forceps Major (7.3%), Right IFOF (3.3%)
Cluster 1	OL > YO	231	0.981	27	−47	24	Right IFOF (3.6%)
RD
Cluster 1	OL > YO	12491	0.999	−18	39	4	Left ATR (2.6%), Forceps minor (8.4%), Left IFOF (3.9%), Left SLF (2.4%)
Cluster 2	OL > YO	4816	0.993	20	−22	38	Right IFOF (8.0%), Right ILF (3.9%), Right SLF (3.8%)
Cluster 3	OL > YO	3159	0.996	21	46	10	Right ATR (6.6%), Forceps minor (12.3%), Right IFOF (6.8)
Cluster 4	OL > YO	154	0.952	9	−41	31	Right Cingulum/Cingulate gyrus (17.2%)

All coordinates are defined in MNI152 space. 1-p values were considered significant at 1-p > 0.95. ATR, anterior thalamic radiation; IFOF, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus; OL, older participants; SLF, superior longitudinal fasciculus; UF, uncinate fasciculus; YO, young participants.

Association between WM characteristics and spindle variables

SS density

Correlations between WM characteristics and spindle variables were carried out independently in each age group. No significant correlations between frontal and posterior SS densities and WM diffusion metrics were found in the young group. In the older group, no significant correlations were found for frontal SS density, but MD was positively correlated with posterior SS density in bilateral SLF (18.1% of skeleton, p < 0.05). Since significant interactions between age and sex were found for frontal and posterior densities, correlations were also performed independently in men and women (Supplementary Material).

SS amplitude

Significant associations between WM characteristics and SS amplitude were found only in the young group (Figure 2). Clustered results are presented in Table 3. More precisely, in young participants, FA was positively correlated to frontal SS amplitude in bilateral inferior fronto-occipital fasciculus, superior longitudinal fasciculus, uncinate fasciculus, right temporal part of the superior longitudinal fasciculus, left anterior thalamic radiation and right corticospinal tract (0.3% of skeleton, p < 0.05) whereas RD, AD, and MD were negatively correlated with frontal SS amplitude in the same regions, and also in left temporal part of the superior longitudinal fasciculus, forceps minor, and forceps major (p < 0.05; RD: 17.5%, AD: 1.7%, MD: 26.4% of skeleton). No significant correlation was found between WM diffusion metrics and SS amplitude in the posterior part of the brain.

Figure 2.

Correlations between white matter diffusion metrics, frontal sleep spindle amplitude, and frontal sigma spectral power in young participants. This figure shows thresholded voxels with significant correlations between WM diffusion metrics, frontal SS amplitude, and frontal spectral power in young participants (1-p > 0.95 corrected). Results are overlaid on group-specific white matter skeleton in dark gray and standard MNI152 template. According to the convention, the right hemisphere is represented on the left side. Warm colors represent positive correlation and cold colors negative correlation.

Table 3.

Clustered white matter correlations with frontal sleep spindle amplitude in young participants

				Max coordinates (mm)
	Correlation sign	Voxels	1-p Max	X	Y	Z	Locations (>2% probability)
FA
Cluster 1	Positive	1478	0.965	34	5	2	Right IFOF (9.1%), Right SLF (9.3%), Right UF (4.0%), Right temporal part of SLF (3.1%)
Cluster 2	Positive	1191	0.972	−29	−1	15	Left IFOF (16.1%), Left SLF(2.4%), Left UF (9.4%)
Cluster 3	Positive	438	0.961	−21	5	17	Left ATR (35.7%), Left IFOF (9.3%), Left UF (3.7%)
MD
Cluster 1	Negative	23334	0.997	−6	0	26	Forceps minor (3.3%), Left SLF (3.2%), Right SLF (3.1%)
AD
Cluster 1	Negative	1652	0.972	9	−38	17	Forceps major (3.1%), Forceps minor (6.8%)
RD
Cluster 1	Negative	9754	0.989	−21	5	16	Left ATR (4.5%), Left IFOF (5.2%), Left SLF (6.7%), Left UF (2.7%), Left temporal part of SLF (2.6%)
Cluster 2	Negative	7082	0.981	27	−7	28	Right corticospinal tract (4.2%), Right IFOF (4.5%), Right SLF (6.9%), Right temporal part of SLF (2.5%)

All coordinates are defined in MNI152 space. 1-p values were considered significant at 1-p > 0.95. AD, axial diffusivity; ATR, anterior thalamic radiation; IFOF, inferior fronto-occipital fasciculus; SLF, superior longitudinal fasciculus; UF, uncinate fasciculus.

Sigma spectral power

Significant correlations were found between WM characteristics and sigma spectral power in the frontal derivations in the young group only (Figure 2). In the young group, RD and MD in bilateral superior longitudinal fasciculus, left anterior thalamic radiation, and left inferior fronto-occipital fasciculus, were negatively correlated with sigma power (p < 0.05; 1.1% and 19.9% of skeleton respectively) (Figure 2 and Table 4). No significant correlation was found between WM diffusion metrics and sigma power in the posterior part of the brain.

Table 4.

Clustered white matter correlations with frontal sigma power in young participants

				Max coordinates (mm)
	Correlation sign	Voxels	1-p Max	X	Y	Z	Locations (>2% probability)
MD
Cluster 1	Negative	19167	0.986	−6	1	26	Left SLF (3.7%), Right SLF (3.4%)
RD
Cluster 1	Negative	1036	0.958	−23	13	15	Left ATR (13.4%), Left IFOF (2.1%)

All coordinates are defined in MNI152 space. 1-p values were considered significant at 1-p > 0.95. ATR, anterior thalamic radiation; IFOF, inferior fronto-occipital fasciculus; SLF, superior longitudinal fasciculus.

The association between WM and spindle variables significantly differs in young and older individuals

Hierarchical regression models showed no significant difference between the two age groups in the relationship between spindle density and WM. However, significant interactions between age group and WM diffusion were observed in the models predicting SS amplitude and sigma power in the frontal part of the brain (see Table 5 and Figure 3). Age group was found to significantly moderate the association between all four WM diffusion metrics and frontal SS amplitude (AD: β = 8.0, p < 0.01; FA: β = −6.2, p < 0.01; MD: β = 5.7, p < 0.01; RD: β = 3.8, p < 0.01). These moderating effects explained between 5.6% and 12.6% of the variance of SS amplitude in the brain frontal region. Significant interactions were decomposed and showed that higher AD, MD and RD, and lower FA were associated with lower SS amplitude among young participants (AD: β = −0.6, p < 0.01; FA: β = 0.7, p < 0.01; MD: β = −0.5, p < 0.01; RD: β = −0.6, p < 0.01). However, no significant link was found in the older group. These effects accounted for 23.3% to 39.0% of the variance in the young participants. Similar results were found for sigma spectral power in the anterior part of the brain (Table 5). Interactions between age group and two of the four WM diffusion metrics were significant in the model predicting sigma spectral power (MD: β = 6.5, p < 0.01; RD: β = 3.3, p < 0.05), accounting for 5.5% and 6.9% of sigma power variance. Decomposition of significant interactions demonstrated that among the young group only, increased diffusion metrics were associated with decreased sigma power explaining respectively 16.1% and 13.6% of the variance (MD: β = −0.4, p < 0.05; RD: β = −0.4, p < 0.05). However, this significant effect was not observed in the older group.

Table 5.

Hierarchical regression analyses with age and white matter diffusion variables as predictor of sleep spindle amplitude and sigma spectral power in frontal region of the brain

	Frontal spindle amplitude			Frontal sigma spectral power
Predictors	β	R 2	R 2 change	β	R 2	R 2 change
WM diffusion – AD
Step 1		14.6%*
Sex	0.444*
ICV	0.098
Step 2		50.9%**	36.3%**
Age group	−0.668**
AD	−0.241*
Step 3		61.6%**	10.7%**
Age group × AD	7.962**
WM diffusion – FA
Step 1		14.6%*
Sex	0.444*
ICV	0.098
Step 2		51.8%**	37.2%**
Age group	−0.581**
FA	0.238*
Step 3		64.4%**	12.6%**
Age group × FA	−6.153**
WM diffusion – MD
Step 1		14.6%*			11.2%*
Sex	0.444*			0.017
ICV	0.098			0.346
Step 2		52.2%**	37.6%**		42.4%**	31.2%**
Age group	−0.591**			−0.541**
MD	−0.256**			−0.235*
Step 3		57.7%**	5.6%**		49.2%**	6.9%**
Age group × MD	5.739**			6.530**
WM diffusion – RD
Step 1		14.6%*			11.2%*
Sex	0.444*			0.017
ICV	0.098			0.346
Step 2		52.8%**	38.2%**		42.1%**	30.9%**
Age group	−0.536**			−0.471**
RD	−0.264**			−0.228*
Step 3		58.7%**	5.9%**		47.6%**	5.5%*
Age group × RD	3.820**			3.301*

*p < 0.05, **p < 0.01.

Figure 3.

Scatterplots of the association between white matter diffusion metrics and frontal sleep spindle amplitude in young and older participants. This figure shows scatterplots of the association between WM diffusion metrics and frontal SS amplitude in young and older participants. The diffusion metrics were extracted from the mask used for the moderation analysis (cluster showing a significant association between WM diffusion and SS frontal amplitude). Young participants are represented in light green whereas older individuals are represented in dark blue.

Results from supplementary voxelwise model of ANCOVA

When correcting for multiple comparisons, linear relationships between WM and SS did not significantly differ between young and older individuals. However, when exploring the uncorrected maps (p < 0.05), we observed significant differences in the regression slopes between young and older individuals in the same regions as reported with our moderation results which corroborate our current findings (Supplementary Material).

Discussion

The goal of this study was to investigate whether WM integrity explained age-related reductions in SS amplitude and density. Our main results showed that diffusion metrics correlated with SS amplitude and sigma power in tracts connecting the thalamus to the frontal part of the brain in the young but not in the older group. Moderation analyses indicated that age group significantly interacted with WM diffusion in the model predicting SS amplitude and sigma power, i.e. the relation between WM and SS amplitude differs significantly between the two age groups. WM diffusion metrics explained between 14% and 39% of SS amplitude and sigma power variance in the young participants only. Our results support the fact that WM is a contributing factor to SS amplitude and sigma power in the young adult brain, whereas WM does not appear to explain SS and sigma power modifications in aging.

Age-related changes in WM microstructure

Compared to the older group, FA and AD were higher in young individuals across most of the WM skeleton, although AD was lower in a more circumscribed region which included a part of the right inferior fronto-occipital fasciculus. MD and RD showed the expected age-related changes, a significant increase being found in older participants, predominantly in the frontal and parietal WM tracts. Our results are in line with the previous literature showing a general trend of decreased FA and AD, and increased MD and RD in older participants compared to young ones in all major WM tracts of the brain [31, 34–40]. Furthermore, the restricted age-related AD increase found in our study matches the literature, which showed a variable pattern of change in normal aging [31, 34, 38]. Changes in diffusion metrics are generally referred to as changes in WM integrity. A better integrity would reflect a more anisotropic diffusion alongside WM fiber orientation and less diffusivity across such tract. In diffusion terms, better integrity would be predominantly demonstrated by increased FA and AD, representing, respectively, coherence along the specific axis and diffusion alongside the tensor’s first vector. Better WM integrity would also be reflected by decreased RD representing diffusion across the axon membrane and decreased MD, a measure of averaged diffusivity expressing the extent of diffusion in all directions. Axial and radial diffusion metrics were found, respectively, to be sensitive markers of axonal [56–58] and myelin integrity [30, 59], and are thought to reflect demyelination, decreased axonal packing attenuation [38, 60–62], axonal degeneration, and axonal loss [63, 64]. Animal research, including early studies on Wallerian nerve fiber degeneration, demonstrated that axonal degeneration leads to myelin sheaths breakdown [65]. However, myelin degeneration can also occur even when the axon is intact. Studies also supported a remyelination process in the aging brain that creates a thinner myelin sheath with shorter internodes suggesting less efficient connections which could, in turn, affect the timing of the neuronal circuitry [66, 67]. Therefore, conduction delays, changes in action potential refractory times, or dispersions of impulses associated with myelin breakdown, may have a significant impact on synchrony of nerve impulses, on which normal brain functions rely [68, 69]. These findings, thus, confirm age-related changes of WM integrity, which suggest the implication of processes involved in axonal and myelin breakdown, and are further corroborated by our results showing decreased FA and AD, and increased MD and RD in older participants. The frontal localization of these last results is in line with previous research showing a more pronounced myelin breakdown in vulnerable later myelinated regions such as the frontal lobe [70]. Finally, the hypothesis that myelin degeneration can occur in neurons with intact axons could also explain the circumscribed AD increase in older individuals. Importantly, this WM degeneration is already prominent in a relatively young and particularly healthy group in this study given the extent of our exclusion criteria.

WM microstructure predicts SS amplitude only in young participants

Before testing the moderation model, this study investigated the link between WM diffusion, and SS characteristics and sigma independently in each group. In the young group, higher WM integrity in the anterior part of the brain, as indicated by higher FA, and lower AD, MD, and RD was associated with higher frontal SS amplitude. Similar associations were observed between frontal sigma power and MD and RD. The only other study that focused on the association between WM and SS, found that higher SS power and density in frontal electrodes were associated with higher AD in anterior WM tracts and in temporal WM tracts respectively, whereas no relationship was found between FA and SS variables [27]. The association between WM diffusion and SS was corroborated by our study, although not in an identical manner, as can be seen in our results showing that lower AD in a small cluster overlapping the major and minor forceps is associated with lower SS amplitude; and that higher FA in the frontal WM tracts is linked to higher SS amplitude. The difference in results in these two studies may be due to the fact that several studies in healthy individuals across the lifespan have demonstrated that AD changes vary in different directions depending on a complex interaction between several biological factors including axonal packing, axonal caliber, and fiber coherence [71–75]. This variability between studies illustrates the importance of precisely assessing all potential diffusion possibilities in order to understand the physiological phenomena underlying WM diffusion changes, and how these could translate in changes in sleep EEG. Our study was, therefore, the first to characterize the diffusion pattern differences using the global metrics such as FA and MD in addition to using also more precise metrics assessing the axial and the radial diffusivities.

The most widespread effect seen in our results is the reduction in MD and RD, correlated with higher SS amplitude and sigma power in the anterior part of the brain. These results support the hypothesis that better WM integrity, as shown by less diffusivity across the neuron membrane, could enhance neuronal synchrony in young individuals. Indeed, it is largely supported by existing literature, that in SS generation, the TRN sends its inputs through dorsal thalamic relays up to the cortex using several frontal WM tracts including the anterior thalamic radiation, which specifically connects the mediodorsal and anterior thalamic nuclei to the prefrontal and anterior cingulate cortices [76]. Differences in WM diffusion underlying these connections could impact various SS characteristics by disrupting the efficiency of the communication between the thalamic dorsal nuclei and the cortex. This could, in turn, affect cortical feedback of the thalamo-cortico-thalamic loop leading to disruption of spindle oscillations. Hence, our findings suggest that increased integrity of the underlying WM fibers, connecting the thalamus to the frontal cortex, could enhance neuronal synchrony in the loop, which could, in turn, have an impact on the EEG and increase SS amplitude and sigma power in the anterior part of the brain, at least in young participants.

Our moderation model showed that age interacts with WM diffusion in the models predicting frontal SS amplitude and sigma power, meaning that the relation between WM and SS variables differs in young individuals compared to older group. Indeed, decomposition of significant moderation effects showed that WM account for a significant amount of the variance of SS variables in young participants, whereas in older participants WM measures do not significantly add any explained variance. These results were also corroborated by a different statistical model showing significant voxels in the uncorrected maps for which the regression slopes between WM diffusion metrics and frontal SS amplitude differed between the young and the older groups in the same regions as reported with our moderation results. Altogether, these results support the notion that in older individuals, significant physiological changes including cellular, molecular, and metabolic factors, as well as brain oedema, could affect SS without having a direct effect on WM. It is, then, possible that these factors conceal the relationship between diffusion metrics and SS characteristics in our older group. For instance, vascular and metabolic factors affect the aging brain by reducing microvascular plasticity and modulating the microvasculature response to variation in metabolic demand [77]. The aging brain is also particularly vulnerable to cellular and molecular risks such as inflammatory reactions, oxidative stress, and natural selection deleterious gene mutations [77–81]. Finally, it is possible that the relationship between the aging brain and SS could rely on other age-related changes, such as gray matter or brain plasticity. Indeed, a nap study showed that SS duration and amplitude correlated with gray matter atrophy in the hippocampus, parietal, cerebellum, and supplementary motor area in older participants and, more specifically, hippocampus atrophy predicted the decrease in frontal SS density in those same participants [82]. Modification of SS could also arise from age-related cerebral plastic changes as a wealth of evidence have linked SS with neurophysiological processes implicated in brain plasticity including long-term potentiation, learning, and memory consolidation [13, 83].

Limitations and future studies

Our study evaluated young and relatively healthy older participants who could also belong to the middle-aged range. Elderly individuals (>70 years old) could show a different pattern of association between WM diffusion and SS variables and further studies evaluating elderly individuals are clearly needed. In addition, specific WM tracts should be assessed with a streamline-based or bundle-specific approach to quantify precise streamlines between subcortical nuclei such as the thalamus and the frontal cortex. This type of advanced methods usually relies on different models of diffusion MRI such as the fiber orientation density function (fODF) which are known to better represent the diffusion signal in areas with abundant crossing fibers [84]. Moreover, future research using a diffusion MRI pipeline which includes more non-weighted images (B₀) to reduce the signal-to-noise ratio should be planned to replicate our results. Finally, a larger sample of participants and further methodological developments should aim at statistically testing the moderation effect within each brain voxel in order to better understand the implication of WM fiber bundles in the prediction of SS variables.

Conclusion

WM integrity, as assessed through diffusion characteristics in the frontal area, including tracts underlying the TC loop, explains a significant proportion of the variability in SS amplitude and sigma power in young participants. However, age-related changes in WM diffusion metrics do not explain SS characteristics in the older brain. Key factors explaining age-related changes in human SS characteristics still need to be uncovered.

Supplementary material

Supplementary material is available at SLEEP online.

Funding

This study was supported by the Canadian Institutes of Health Research (CIHR – grant number 190750) and the Fonds de Recherche du Québec – Santé (FRQS).

Notes

Notes

Conflict of interest statement. None declared.

This work was performed at the Center for Advanced Research in Sleep Medicine of the Hôpital du Sacré-Coeur de Montréal (Address: 5400 Gouin West Blvd. H4J 1C5, Montréal, Québec, Canada).