Aims of the study

In this follow-up study we aimed to assess whether our previous findings on the correlation between adolescents’ sleep habits and regional brain grey matter volumes in the mPFC remained present at the age of 16, and to extend these findings by including the hippocampus and amygdala brain regions and examining the relationships with psychological functioning.

Participants

Participants were recruited from schools near Paris, France, based on their age and absence of any major somatic condition. Written consent was obtained from all subjects in this study. The study was approved by the regional ethics committee (Comité de Protection des Personnes [CPP] Ile-de-France 7). The adolescents participated in a larger multi-centre study (http://www.imagen-europe.com/en/the-imagen-study.php) [22] at age 14 (baseline), and were followed up at age 16. Only the French adolescents were assessed for their sleep habits and were thus eligible for this study. Details of the sample at baseline have been previously reported [8]. This study focuses on the sample at age 16, at which time point written informed assent and consent to study participation were obtained from a total of 138 adolescents and their parents, respectively. We excluded participants who did not complete the sleep questionnaires, those whose Magnetic Resonance brain images did not pass the quality control of the raw or the segmented images, participants with brain lesions, and those with marked alcohol consumption (alcohol use disorders identification test (AUDIT) total score >7 [23]). In this study, we present data from the remaining 101 adolescents (mean age = 16.83 years, SD = 0.61; 56 girls; Table 1). At the time of the study, none of the participants were followed in the psychiatric care system and all participants were attending school regularly.

Table 1

Clinical, behavioral characteristics, and brain volumes in community adolescents at age 16.

Variable		Mean or %	SD
Demographic variables	Age (years)	16.83	0.61
	Sex	55% (n = 56) female
Sleep variables (N = 101)	Wake up time WD	7:03	0:43
	Wake up time WE	10:01	1:15
	Bed time WD	22:42	0:43
	Bed time WE	00:05	1:10
	WE delay wake up time	2:59	1:17
	WE delay bed time	1:23	1:04
	Time in bed WD	8:18	0:59
	Time in bed WE	9:56	1:00
	Sleep debt	1:40	0:59
Performance scores (N = 82)	Delay discounting large amounts	-2.25	0.68
	Delay discounting medium amounts	-2.03	0.73
	Delay discounting geomean	-2.06	0.66
	Spatial working memory	7.59	7.80
Behavioural problems (N = 74)	SDQ internalising	4.34	2.40
	SDQ externalising	7.36	2.68
Global brain measures (N = 101)	Total grey matter volume	742.02	68.99
	Total white matter volume	477.22	49.28
	Total CSF	395.55	40.89
	Volume scaling factor	1.32	148.21

WD = weekday; WE = weekend; CSF = cerebrospinal fluid; SDQ = Strenghts and Difficulties Questionnaire.

Sleep assessments

Sleep habits were assessed by asking the adolescents their usual bed times and wake up times during weekdays (WD) and weekends (WE). Time in bed was approximated by calculating the number of hours between bed time and wake up time, separately for WD and WE. WE delay in wake up time and bed time (“social jet lag”) was defined as the difference in wake up and bed times between weekdays and weekends, respectively. Sleep debt was defined as the difference in time in bed between weekdays and weekends.

Psychological functioning

Symptom assessment was performed using the Strengths and Difficulties Questionnaire (SDQ) [24], a child and adolescent self-report questionnaire used to identify internalising and externalising problems. It consists of 25 items, five items for each subscale: conduct problems, hyperactivity, emotional problems, peer problems, and prosocial behaviour. The items are scored 0 to 2, reflecting the answers “not true”, “somewhat true” or “certainly true”. The scores are then summed, generating five scale scores ranging from 0 to 10 with higher scores reflecting more problems in the first four scales or more prosocial behaviour in the last scale. In low-risk samples, conduct problems and hyperactivity are best combined into an ‘externalising’ subscale, and emotional problems and peer problems into an ‘internalising’ subscale [25].

Other behavioural measures included the Kirby Delay-Discounting Questionnaire [26], a monetary choice questionnaire assessing cognitive impulsivity through delay discounting by having participants choose between smaller immediately available rewards and larger delayed rewards, and the spatial working memory task, a subtest of the computer-administrated Cambridge Automated Neuropsychological Test Battery (CANTAB) measuring executive functioning [27]. The spatial working memory test has been widely used in typically developing and clinical populations of children and adolescents [28]. The error score was used as the outcome variable.

MRI data acquisition and processing

MRI was performed on a 3T scanner (Siemens Trio). High-resolution anatomical MR images were obtained using a standardised 3D T1-weighted magnetisation prepared rapid acquisition gradient echo (MPRAGE) sequence based on the ADNI protocol (http://adni.loni.usc.edu/methods/mri-analysis/mri-acquisition/). The parameters were as follows: repetition time = 2,300 ms, echo time = 2.8 ms, flip angle = 8°, 256x256x170 matrix, 1.1x1.1x1.1 mm voxel size.

The images were processed using Statistical Parametric Mapping 8 (SPM 8) using Voxel-Based Morphometry (VBM) [29]. The “unified segmentation” algorithm was used to normalise and segment the T1-weighted images into grey matter, white matter and cerebrospinal fluid. Home-made tissue probability maps were used instead of the standard template of SPM based on fully grown and developed brains of adults, who have larger brain volumes. The modulated images were smoothed with a 10-mm full-width at half-maximum isotropic Gaussian kernel. Head size was measured by the volume scaling factor, which is based on the affine transformation performed during spatial normalisation (https://surfer.nmr.mgh.harvard.edu/fswiki/eTIV).

Statistical analyses

The outcome measures for the Kirby Delay-Discounting Questionnaire were calculated using an automated calculator [30]. A logistic regression that allowed for continuous estimates of k [31], and a logarithmic transformation to normalise the distribution of k-values were applied. The geometric mean of the k-values is bounded by the lowest implied indifference k-value at which subjects chose the larger delayed reward and the highest indifference k-value at which the subject choses the immediate reward.

Group comparisons for socio-demographic and clinical data, and global brain volumes were performed within the framework of the general linear model (GLM) using R software (http://cran.r-project.org). Voxel-wise comparisons were carried out within the GLM framework of SPM8. Sleep variables were the main factors while age, sex and volume scaling factor were entered as confounding variables. Pearson partial correlations between sleep and psychologicial function variables were calculated with age and sex as confounder variables.

A priori masks were used to examine the hippocampus, amygdala, and medial prefrontal region with a region of interest (ROI) approach. The hippocampus and amygdala masks were created with the Anatomical Automatic Labeling (AAL) atlas in SPM. The medial prefrontal region was marked with a 10 mm radius from coordinates [–2, 34, –14] taken from our previous results on an analogous sample at age 14, which indicated a strong relationship between sleep habits and GMV in this region [8]. Subsequently, it was investigated whether psychometrics measures (spatial working memory, delay discounting, internalising and externalising problems) were associated with grey matter volumes in these regions of interest.

At the voxel-level, statistical significance was set to p<0.05 Family Wise Error (FWE) corrected for multiple comparisons. Brain locations were reported as x, y, and z coordinates in Montreal Neurologic Institute (MNI) space.

In addition, it was examined whether the sleep assessment variables that were significant in the ROI VBM analyses, were correlated with psychological functioning. Subsequently, causal mediation analyses were performed to determine whether the grey matter clusters could mediate the relation between sleep and psychological functioning variables. These analyses were performed if there was a significant relationship between sleep and psychological functioning variables and GMV of the same ROI. The mediation analyses were performed with an algorithm using a set of general linear models to derive the mediation and direct effects from the total effect [32]. The psychological functioning measures were entered as dependent factor, and sleep variables as independent factor within a linear regression model. For the mPFC ROI analysis, raw volume was extracted from the smoothed, normalized, and modulated images and entered as mediator variable. The hippocampus and amygdala volumes were extracted from the same images using the AAL masks instead of the significant clusters in order to better approximate the complete volumes. Volume scaling factor and age were entered as confounding variables. Gender was added for analyses without a gender interaction between sleep and psychological functioning scores. If there was a gender effect, we performed the mediation analysis separately for boys and girls. This mediation model was performed using 5000 Monte Carlo draws for nonparametric bootstrap. In causal mediation analysis, a significant mediating effect is defined as a 95% confidence interval that does not include zero.

Participant characteristics

On average, boys had later WD bed times (boys = 22.53 ± 0.47, girls = 22.33 ± 0.33; F(99) = 2.41, p = 0.02) and WE wake up times (mean wake up time: boys = 10.23 ± 1:17, girls = 9:44 ± 1:10; F(99) = 2.63, p = 0.01) as compared to girls. As shown in Table 1, there were no gender differences in any of the other sleep variables, nor in delay discounting, spatial working memory or internalising and externalising problems.

Clinically relevant externalising symptoms (SDQ score ≥10), were present in 15 out of 81 adolescents (18.3%) and 10 out of 81 (12.2%) showed clinically relevant internalising symptoms (score ≥8).

Sleep and regional grey matter volumes

WE wake up time correlated negatively with GMV in the bilateral amygdalae and the mPFC (p = 0.011, p = 0.041, p = 0.020, resp.; Table 2, Fig 1). WE delay in wake up time correlated negatively with GMV in the right hippocampal region and the right amygdala (p = 0.025, p = 0.008, resp.; Table 2). No statistically significant correlations between WE bed time, WE delay in bed time, or sleep debt and regional grey matter volumes were found.

Fig 1

Later wake up time during the weekend associated with reduced grey matter volumes in the (A) left amygdala and (B) medial prefrontal cortex (mPFC).

Corrected p FWE <0.05.

Table 2

Grey matter volume correlations with sleep measures in community adolescents at age 16 using regions-of-interest.

Sleep measure	Amygdala		Hippocampus		mPFC
	T-value	Peak (x,y,z)	T-value	Peak (x,y,z)	T-value	Peak (x,y,z)
Weekend wake up time	3.51*	21,-6,-12			3.11*	2,42,-11
	3.04*	-12,-3,-15
Weekend delay in wake up time	3.23*	20,-3,-12
	4.23**	20,-25,-23

* = p<0.05,

** = p<0.01,

*** = p<0.001 FWE corrected.

- = no significant results, Montreal neurological Institute coordinates are given for the voxel of maximal statistical significance; analyses are covaried for volume scale factor,age, and gender.

Sleep and psychological functioning

Later WE wake up time and WE delay in wake up time correlated with measures of externalising problems (p<0.001), whilst a larger WE delay in wake up time was correlated with lower internalising problems (p<0.001) (Table 3).

Table 3

Correlations between sleep variables and psychological functioning in community adolescents at age 16 adjusted for age and sex.

	WE wake up time	WE delay wake up time
Delay discounting large	r = 0.02	r = 0.07
Delay discounting medium	r = 0.05	r = 0.12
Delay discounting geomean	r = 0.05	r = 0.14
SDQ externalising	r = 0.51***	r = 0.61***
SDQ internalising	r = -0.14	r = -0.42**
Spatial working memory errors	r = -0.20	r = 0.12

* = p<0.05,

** = p<0.01,

*** = p<0.001.

WE = weekend; SDQ = Strengths and Difficulties Questionnaire.

Grey matter volumes and psychological functioning

Smaller grey matter volumes in the mPFC region were associated with increased delay discounting (T = 3.28, p<0.01) as well as internalising (T = 4.99, p<0.001) and externalising (T = 6.03, p<0.001) problems (Table 4). Additionally, smaller grey matter volumes in the amygdala (most significant cluster T = 10.57, p<0.001), and hippocampal regions (T = 11.01, p<0.001) were associated with internalising problems. There were no other significant relations between grey matter volumes and psychological functioning measures (Table 4).

Table 4

Grey matter volumes correlations with psychological functioning in 138 community adolescents at age 16.

Psychological functioning	Amygdala		Hippocampus		mPFC
	T-value	x,y,z	T-value	x,y,z	T-value	x,y,z
Delay discounting large	-	-	-	-	2.97*	-6,42,-15
Delay discounting medium	-	-	-	-	3.29*	-8,40,-17
Delay discounting geomean	-	-	-	-	3.28**	-8,40,-17
Spatial Working Memory	-	-	-	-	-	-
SDQ externalising	-	-	-	-	6.03***	-2,44,-14
SDQ internalising	10.57***	-30,-3,-29	11.01***	-28,-7,-32	4.99***	4,30,-8
	4.40***	36,0,-24	7.34***	21,-45,-3
			6.13***	40,-15,-23

* = p<0.05,

** = p<0.01,

*** = p<0.001 FWE corrected.

- = no significant results, Montreal neurological Institute coordinates are given for the voxel of maximal statistical significance; analyses are covaried for volume scale factor,age, and sex. SDQ = Strengths and Difficulties Questionnaire.

Mediation analysis

Causal mediation analyses showed that variability in mPFC volumes accounted for 30.5% (p<0.001) of the total effect between wake up time during the WE and externalising problems (Fig 2).

Fig 2

Causal mediation analysis on the relationship between sleep variables and externalising problems with medial prefrontal cluster volumes as mediator.

Estimate of the size of the effect (95% confidence interval of the point estimate); * = p<0.05, ** = p<0.01, *** = p<0.001; SDQ = Strengths and Difficulties Questionnaire.