The association between inadequate sleep and accelerated brain ageing

Abstract

Numerous studies indicate large heterogeneity in brain ageing, which can be attributed to modifiable lifestyle factors, including sleep. Inadequate sleep has been previously linked to gray (GM) and white (WM) matter changes. However, the reported findings are highly inconsistent. By contrast to previous research independently characterizing patterns of either GM or WM changes, we used here linked independent component analysis (FLICA) to examine covariation in GM, and WM in a group of older adults (n = 50). Next, we employed a novel technique to estimate the brain age delta (difference between chronological and brain age assessed using neuroimaging data) and study its associations with sleep quality and sleep fragmentation, hypothesizing that inadequate sleep accelerates brain ageing. FLICA revealed a number of multimodal components, associated with age, sleep quality, and sleep fragmentation. Subsequently, we show significant associations between brain age delta and inadequate sleep, suggesting 2 years deviation above the chronological age. Our findings indicate sensitivity of multimodal approaches and brain age delta in detecting link between inadequate sleep and accelerated brain ageing.

Keywords: Ageing; Brain age; Gray matter; Magnetic resonance imaging; Sleep; White matter.

Fig. 1

Overview of the brain age prediction model. A recently published approach described by Smith et al. (2019) was used to estimate brain age delta in an unbiased manner. (A) The structural and diffusion MRI data were preprocessed using the UK Biobank pipeline. (B) The T1 imaging derived phenotypes (T1IDPs) were extracted from the parcellations of cortical and subcortical GM volumes based on the Harvard-Oxford structural atlases. (C) The diffusion imaging derived phenotypes (dMRI IDPs) were extracted using predefined protocols for identifying major WM tracts as described by the FSL's XTRACT tool. (D) The structural and microstructural IDPs were represented using an imaging matrix X (NxM) such that N = 50 participants and M = 176 (demeaned) imaging features. The head scaling factor was used as an additional structural IDP and introduced as a confound variable in the brain age prediction model. Next, a matrix Y was created to represent the (demeaned) chronological age for N = 50 participants. A matrix Y₂ was also computed to account for quadratic ageing processes which was subsequently demeaned and orthogonalized with respect to Y. The initial brain age prediction model is Y_B = X$\beta$₁ + $\delta$_b such that $\beta$₁ = X⁺Y and the uncorrected (biased) delta which is typically used in brain ageing studies is $\delta$_b = Y_B – Y. The corrected (unbiased) brain age prediction model is then computed as follows: $\delta$_b = Y₂$\beta$₂ + $\delta$_q such that $\beta$₂ = Y₂⁺$\delta$_b. The linearly-corrected estimate of delta is $\delta$ = $\delta$_b – Y$\beta$₂ and the quadratically-corrected estimate of delta is $\delta$_q = $\delta$_b – Y₂$\beta$₂. (E) The relationships between chronological age (Y) and brain age (uncorrected [Y_B]; linear [Y_L]; quadradic [Y_Q]) were assessed using Pearson's correlation coefficient. The relationships between brain age delta (uncorrected [$\delta$_b], linear [$\delta$], quadratic [$\delta$_q]) and sleep measures (PSQI, WASO) were also evaluated using Pearson's correlation coefficient.

Fig. 2

VBM: Whole-brain morphologic alterations in the ageing brain. Reduced GM volumes are significantly associated with increasing age. The colored voxels show widespread GM changes in the following regions: Default mode network: posterior cingulate cortex (PCC), precuneus (Pre), parahippocampal gyrus (PHG); Primary motor cortex: precentral gyrus (PreG); Primary somatosensory cortex: postcentral gyrus (PosG); Auditory cortex: Heschl's gyri (HG), superior temporal gyrus (STG), planum polare (PP); Visual areas: lingual gyrus (LG), fusiform gyrus (FG), inferior temporal gyrus (ITG); Language areas: angular gyrus (AG), inferior frontal gyrus (IFG), insular cortex (INC), middle temporal gyrus (MTG); Prefrontal regions: superior frontal gyrus (SFG), middle frontal gyrus (MFG), frontal polare (FP); Limbic regions: hippocampus (HIP), amygdala (AMYG), orbitofrontal cortex (OFC); Basal ganglia: dorsal striatum (caudate (CAU), putamen (PUT), ventral striatum (nucleus accumbens [NAcc]), pallidum (Pa). The spatial map is FWE-corrected for multiple comparisons set at p < 0.05.

Fig. 3

TBSS: Whole-brain skeletonized microstructural differences in older adults. (Top) Reduced FA is significantly associated with increasing age. (Bottom) Increased MD is significantly associated with increasing age in older adults. Changes were significant in a number of WM fiber bundles, including: Association fiber bundles: Inferior Longitudinal Fasciculus (ILF), Inferior Fronto-Occipital Fasciculus (IFOF), Superior Longitudinal Fasciculus (SLF), Uncinate Fasciculus (UF). Projection fiber bundles: Acoustic Radiation (AR), Anterior Thalamic Radiation (ATR), Posterior Thalamic Radiation (PTR), Corticospinal Tract (CST), Optic Radiation (OR). Limbic fiber bundles: Cingulum Gyrus part of Cingulum (Cing), Hippocampal part of Cingulum (CBH), Fornix (FX). Commissural fiber bundles: Forceps Major (FMA), Forceps Minor (FMI). The spatial maps are FWE-corrected for multiple comparisons set at p < 0.05.

Fig. 4

FLICA: Strongest independent components (ICs) that capture different modes of covariations in VBM and WM microstructure in addition to their relationships with age (A-B), PSQI (C-D) and WASO (E). Note that the x-axis shows the demeaned demographic (i.e., age) and behavioral (i.e., PSQI, WASO) values. The effect magnitude is represented by R². The significant relationships between the IC subject loadings and demographic and behavioral measures are Bonferroni-corrected for multiple comparisons set at p < 0.05.

Fig. 5

FLICA: Spatial maps of the independent components (ICs) that reveal GM volumetric and WM microstructural changes measured by FA and MD in Age (Top Panel), PSQI (Middle Panel) and WASO (Bottom Panel). The spatial maps are Bonferroni-corrected for multiple comparisons set at p < 0.05.

Fig. 6

Relationships between brain age delta and sleep measures using the unimodal (T1) and multimodal (T1+dMRI) IDPs. (A) $\delta$ was significantly associated with PSQI using the unimodal T1 IDPs. (B) $\delta$ was not significantly associated with WASO after applying FDR correction when the unimodal T1 IDPs were used. (C) $\delta$ was significantly related to PSQI with the multimodal (T1+dMRI) IDPs. (D) $\delta$ was not significantly associated with WASO when the multimodal (T1+dMRI) IDPs were used. All relationships between brain age delta and sleep measures are FDR-corrected for multiple comparisons set at p < 0.05. Note that the raw PSQI and WASO scores are square root and log-transformed, respectively (see Section 2.2).