Structural Alterations in Large-scale Brain Networks and Their Relationship with Sleep Disturbances in the Adolescent Population

Although sleep disturbances are highly prevalent in adolescents, neuroimaging evidence on the effects of sleep disturbances on their developing brains remains limited. Therefore, we explored gray matter volumes (GMVs) at the whole-brain level and investigated their relationship to sleep disturbances in a sample of Korean adolescents in the general population. We recruited participants from one middle school and high school. All participants and their legal guardians gave informed consent before participating in our study. We used component 5 of the Pittsburgh Sleep Quality Index to measure sleep disturbances and conducted a voxel-based morphometry-DARTEL procedure to measure GMVs. We performed partial correlation analyses to examine whether the GMVs were associated with sleep disturbances. A total of 56 adolescents participated in this study. Our results revealed that GMVs in multiple global regions were negatively correlated with sleep disturbances. Moreover, most of these identified regions belong to large-scale brain networks categorized by functional neuroimaging studies. We found an association between regional GMVs in multiple global regions involved in large-scale networks and the severity of sleep disturbances in the adolescent population. Based on this evidence and previous neuroimaging evidence, we suggest that structural alterations in the networks may be linked to sleep disturbances.

partial correlation values and the significance values, including both uncorrected and FDR-corrected p-values, in Table 3. Figure 1 shows the correlations between the rGMVs and severity of sleep disturbances in our study population.

Discussion
In the present study, the assessment was not solely based on a single region of interest; rather, it focused on the whole brain. Using VBM-DARTEL, we found an association between the rGMVs in global brain regions and the severity of sleep disturbances in a sample of Korean adolescents in the general population. Its magnitude remained strong, even after we controlled for relevant covariates (i.e., age, sex, TIV) and type I errors. Interestingly, we also found that these identified global regions are components of large-scale brain networks categorized by a large body of functional magnetic resonance imaging (fMRI) studies with adult study populations in clinical settings and nonclinical settings 32,[35][36][37][38][39][40][41][42][43] . The networks include the default mode network (DMN), central executive network (CEN), visual network (VN), somatosensory network (SSN), and salience network (SN). In recent years, altered functional connectivity in these networks has increasingly been revealed by many neuroimaging studies on different domains of sleep problems [44][45][46] . Comparably, the relationships between the structural alterations in some core regions of these networks and different domains of sleep problems have been increasingly reported by magnetic resonance imaging (MRI) studies conducted in adult populations 15,[26][27][28] . Based on all of this evidence, we discuss in this section altered GMVs and their relationship with sleep disturbances from this large-scale brain network perspective. Table 3 lists the multiple brain regions that are involved in the DMN. In our study, adolescents with higher levels of sleep disturbances had decreased volumes in the right opercular inferior frontal gyrus, the right triangular inferior frontal gyrus, the right medial superior frontal gyrus, and the left inferior parietal lobule. Among these regions, the right inferior frontal gyrus had the strongest association with sleep disturbances. The inferior parietal lobule is regarded as one of the core regions in this network 47 . Several functional and structural neuroimaging studies and an electroencephalogram (EEG) study conducted in adolescent populations have also revealed associations between sleep problems, including insufficient sleep and poor sleep quality, altered GMVs and atypical activation in multiple brain regions that are implicated in the DMN 16,23,[48][49][50][51] . Similarly, a smaller GMV in the right middle superior frontal gyrus has also been shown to be correlated with a later bedtime on weekends in 14-year-old community adolescents 23 . The DMN is conventionally regarded as a network that is active when the brain is at rest, yet there is compelling evidence that the DMN also plays an important role in sleep as well as in other activities 37,52 . A growing body of structural and functional neuroimaging studies have demonstrated that altered GMVs and functional connectivity within the DMN are linked not only to sleep problems, but also to other symptoms of sleep problems, including poor cognitive functions 30,45,53 . Specifically, the prefrontal cortex is www.nature.com/scientificreports www.nature.com/scientificreports/ active not only during the deactivation of rapid eye movement (REM) sleep and the transition between wake and non-REM sleep but also in social cognitive processes such as decision making and self-regulation 37 .
As shown in Table 3, the right middle frontal gyrus, the angular gyrus, and the right inferior orbitofrontal gyrus are implicated with the CEN, which is responsible for decision making, working memory, judgment, and goal-oriented planning 35,37 . The orbitofrontal cortex is known to be one of the key nodes of this network 53 . Although we described here the angular gyrus as a region in the CEN, this region is also regarded as a hub for other networks (i.e., DMN, attention network) involved in various functions (e.g., self-processing, attention, semantic information processing)0 54 . Our findings corroborate those of a structural neuroimaging study that first demonstrated robust evidence on how sleep habits are associated with rGMVs and average academic performance during early adolescence 23 . A shorter time in bed during the weekdays has been shown to be strongly associated with a smaller GMV in the middle frontal gyrus 23 . Our findings also align with the results of a previous study that revealed smaller GMVs in the left orbitofrontal and parietal cortices in a sample of patients with chronic insomnia 26 .  Table 3. Results of the partial correlation analyses between sleep disturbances and gray matter volumes   www.nature.com/scientificreports www.nature.com/scientificreports/ Our findings also showed that the right superior occipital gyrus, which belongs to the VN, which is responsible for visual information processing, is also associated with sleep disturbances 36 . The left middle occipital gyrus and left lingual gyrus are known to be key nodes of the VN 44,55 . We found that the GMV of the right superior occipital gyrus was negatively correlated with the severity of sleep disturbances. A recent fMRI study reported altered emotion-dependent functional connectivity between the occipital cortex and other brain regions and its positive association with sleep duration in a sample of school-aged children 56 . In contrast, a recent neuroimaging study conducted in a sample of primary insomnia patients found an increased volume and altered functional connectivity in the left superior occipital gyrus 40 . Earlier fMRI studies demonstrated hyperarousal activation in the bilateral occipital gyrus during sleep deprivation and in individuals with primary insomnia 37,44,55,57 . This discrepancy in previous results might be due to the heterogeneity of the study populations as well as the scarcity of available data. Therefore, these results should be corroborated by additional investigations with larger samples sizes of adolescents in the general population.

Brain regions Network
Similarly, we found decreased volume in the right postcentral gyrus, a main receptive region for external stimuli, which is located in the primary somatosensory cortex 28 . The SSN plays a critical role in processing somatosensory input and integrating sensory and motor signals for skilled movement 58 . Consistent with our findings, an earlier structural MRI study demonstrated a significant reduction in gray matter in the bilateral postcentral gyrus in patients with primary insomnia 28 . Specifically, this result was related to prolonged sleep latency as well as difficulty in sleep initiation. This study suggested that the association may be explained by a decreased capacity to disengage from external information processing 28 .
In our study, decreased GMVs in the bilateral middle orbitofrontal cortex and the right insula also had strong associations with more severe sleep disturbances. These regions are considered key nodes of the SN, a network that plays a major role in detecting, filtering, and processing external and internal stimuli as well as in attentional filtering and orienting and emotional processing 42,43 . Our findings support the existing robust evidence that this region is closely linked to sleep disturbances 26,28 . The first VBM study 26 that compared the rGMVs of 24 adults with chronic primary insomnia with those of control subjects without sleep problems and showed a strong negative correlation between GMV in the left orbitofrontal cortex and subjective severity of insomnia. Furthermore, our findings pertaining to the insula support established neuroimaging evidence on the strong link between this region and different domains of sleep problems (e.g., poor sleep quality) 37,59-62 . A recent study 63 found a significant correlation between sleep quality and depression/anxiety among 370 college students. Moreover, the mediating role of the GMV in the right insula in this correlation was revealed, which raises the possibility that depression/anxiety may affect sleep quality through volumetric variation in the right insula. Collectively, fMRI studies have provided compelling evidence of altered functional connectivity in this network in individuals with sleep problems 39,43 . Furthermore, the SN appears to play a crucial role in alternately activating the CEN and the DMN 39 . A recent systematic review on obstructive sleep apnea (OSA) and insomnia revealed altered functional connectivity in the SN as well as its link to the DMN and the CEN, and it highlighted the functional integration of the large-scale brain networks 39 .
In the current study, by assessing the whole brain in a sample of adolescents in the general population, we found that structural alterations in multiple regions of the large-scale brain networks categorized by fMRI studies were associated with sleep disturbances. From a neuroimaging perspective, it has been postulated that these networks interact with one another simultaneously 64,65 . Conversely, disruptions in these networks may contribute to a wide range of problems 65 . An established body of both functional and structural neuroimaging research has revealed altered functional connectivity and GMVs of the networks in adults and patients with diverse sleep problems 37,40,42,66 . Liu et al. 37 made the first attempt to characterize the topological architecture of whole-brain functional networks in individuals with primary insomnia, using a seed-based functional connectivity approach. The authors revealed a pattern of functional deficits in the DMN, CEN, and SN.
Furthermore, the strength of the association between rGMVs in our identified regions of the networks and levels of sleep disturbances remained strong, even after we adjusted for age, sex, and TIV. This result is noteworthy since our findings suggest that rGMVs in broad regions throughout the brain might be affected by sleep disturbances prior to adulthood. This result is consistent with that in a recent systematic review on the effects of sleep on developing brain functions and structures in the pediatric and adolescent population 15 . In this review, multiple epidemiological studies have reported an association between sleep and gray and/or white matter volumes in school-aged children and adolescents. For instance, smaller GMVs were associated with later bedtimes on the weekends in adolescents 23 . Likewise, high levels of subjectively measured sleep variability and its association with low white matter integrity were observed during mid-to late-adolescence, indicating that high sleep variability may impair white matter development 67 .
Based on these results and previous structural and functional neuroimaging evidence, we suggest that GMVs in multiple regions of the large-scale networks may already be affected by sleep disturbances prior to early adulthood. Moreover, it is possible that the relationship between structural alterations in the large-scale networks and sleep disturbances may be involved in a wide variety of symptoms (e.g., poor cognitive performance), as these networks are known to be responsible for a broad range of functions. Considering the limited neuroimaging evidence on this relationship in adolescents, the results of our study may provide new insights into determining the structural manifestations of sleep disturbances in the developing brains of adolescents in the general population as well as developing a comprehensive understanding of sleep disturbances and their relationships with the brain structures of adolescents. In accordance with our findings, the seriousness of sleep problems and the importance of healthy sleep should be encouraged and prioritized in the adolescent population. Health care providers and clinicians should emphasize healthy sleep habits and sufficient time for sleep, as sleep may be essential for developing brain structures and functions during adolescence.
Despite the novelty of the current study, our findings have some limitations that should be taken into account. First, we measured the severity of sleep disturbances with a single component of the PSQI, which was a major limitation of our study and may limit the generalizability of our findings. Although this component has been validated and has been shown to have high specificity and sensitivity, a previous study showed a nonsignificant correlation between the PSQI components and actigraphy results in a nonclinical sample 68 . Second, our findings may not be generalizable to other adolescent populations because sampling bias may be present due to the method of participant recruitment that we used (i.e., participants were recruited from one middle school and one high school).
Third, there may be other potential confounders (e.g., depression, anxiety) that might have contributed to the magnitude of the association between the rGMVs and sleep disturbances 69 . Last, our findings were exploratory and we did not present any specific hypotheses, so our findings may have limited contributions toward gaining a better understanding of the association of GMVs with sleep disturbances. We therefore recommend that future studies with VBM group analyses between adolescents with sleep disturbances and those without sleep disturbances and correlation analyses are conducted to elucidate the relationships between structural alterations in the large-scale networks and sleep disturbances. A seed-based structural covariance analysis may also be helpful to examine these relationships in adolescent populations. In addition to this analysis method, we recommend that additional studies are conducted to examine whether GMVs in our identified regions become relatively smaller with age in a sample of adolescents with sleep problems versus those without sleep problems.
To our knowledge, at present, there is not enough data of the relationship between sleep disturbances and structural alterations in brain regions in adolescent populations. Therefore, we made an attempt to investigate whether there is an association between rGMVs and the severity of sleep disturbances in a sample of Korean adolescents from the general population. Our results revealed that rGMVs in multiple global regions were negatively correlated with the severity of sleep disturbances. The magnitude of this association remained strong, even after controlling for multiple covariates. Combined with previous structural and functional neuroimaging evidence, we suggest that the structural changes to gray matter in multiple regions that are involved in the large-scale networks may be linked to sleep disturbances. Moreover, we recommend that an investigation of both functional and structural alterations in the regions of the major networks may be beneficial to tackle underlying neurological mechanisms of sleep disturbances.

Methods
Study population. We recruited adolescents aged from 12 to 18 years from one middle school and one high school in Seoul, South Korea. After the school principals approved our study, we visited the students and teachers at the schools to explain the study's objectives and guaranteed confidentiality of their information. We also mailed letters to the parents of potential participants containing brief information about the study along with the contact information of the principal investigator. The letter also stated that they would be informed of the results of our study following the completion of the analyses. All participants and their parents or legal guardians provided informed consent before taking part in the study.
All participants who participated in our study (a) were capable of fully understanding the description of and following the instructions of the present study; (b) had no possibility of pregnancy prior to the study; (c) did not consume drugs that could significantly affect their sleep and waking conditions; and (d) showed no expected problems in brain imaging and psychological tests. Participants with any clear history of an acquired brain injury such as cerebral palsy, neurological disorders such as convulsive disorder, psychiatric disorders (including schizophrenia, bipolar disorder, or pediatric psychosis), developmental disorders (including autism or intellectual disabilities), learning disabilities, language impairments, or uncorrected sensory impairment were excluded from the analyses in our study. This study was approved by The Institutional Review Board for Human Subjects at Seoul National University Hospital and conducted in accordance with the Declaration of Helsinki.

Study variables. Pittsburgh Sleep Quality Index (PSQI).
The PSQI is a self-rated questionnaire that assesses sleep quality and sleep disturbances over the past month 70 . It has been well validated and is widely used in both clinical and nonclinical settings 71,72 . It has demonstrated high levels of internal consistency and construct validity, with a Cronbach's alpha of 0.8 73  For the present study, we used a single component, component 5, since we focused on sleep disturbances; component 5 measures the severity of sleep disturbances. Regarding the other components, we concluded that these components did not accurately reflect the sleep quality of Korean adolescents, considering there is a difference between their weekday sleep and weekend sleep due to a variety of external factors (e.g., early school start times, extra classes or private lessons at night) that could profoundly restrict their sleep duration 6,74 . Moreover, as the participants in our study did not take sleeping medications, component 6 was not used in our study. A recent study validated the single-factor scoring structure and psychometric properties of the PSQI in a sample of community-based adolescents 75 . Its finding of a single-factor model was consistent with other models used in previous studies conducted in adults and adolescents. The Cronbach's alpha of each PSQI component was higher than 0.6 (α = 0.71 for component 5).
Component 5 consists of a set of 10 items that ask respondents how often they had trouble sleeping 70 . For each statement (e.g., I cannot get to sleep within 30 minutes, I wake up in the middle of the night or early morning), the response options were 'not during the past month, ' 'less than once a week, ' 'once or twice a week, ' and 'three or more times a week. ' The scores of sleep disturbance ranged from 0-27. Higher scores indicate more severe sleep disturbances. www.nature.com/scientificreports www.nature.com/scientificreports/ Voxel-Based Morphometry (VBM) Analysis. To prevent circularity problems during the process of registration (i.e., tissue classification, spatial normalization, spatial smoothing), a refined VBM method has been introduced and implemented 31,76,77 . This method includes a new registration method called diffeomorphic anatomical registration involving exponentiated Lie algebra (DARTEL) 31 . This method provides clearer segmentation and better registration with regards to boundaries between gray matter and white matter compared to optimized VBM 78,79 . With this method, we conducted a VBM analysis using the SPM12 VBM-DARTEL procedure (SPM12, http:// www.fil.ion.ucl.ac.uk/spm/,Wellcome Trust Centre for Neuroimaging, London, UK) 80 .
No abnormalities from motion or other artifacts were found in the T1-weighted images, which were inspected by a well-trained physician. The procedure for preprocessing the T1-weighted images included (i) manual reorientation to the anterior commissure, (ii) gray matter segmentation based on a standard tissue probability map provided from SPM, (iii) the creation of a study-specific template, spatial normalization with DARTEL to normalize individual images to the DARTEL template, modulation to adjust for volume signal changes during spatial normalization and (iv) spatial smoothing of the gray matter partitions with a Gaussian kernel of 8 mm full-width at half maximum. After preprocessing, the rGMV for each area was extracted by averaging the values in 116 brain regions from the AAL atlas 81 .
For this particular study, we used the standard adult SPM template instead of an age-specific template. While this application remains controversial 82 , several neuroimaging studies on developing brains have not only shown neuroanatomical differences between adults but also demonstrated valid neuroimaging results for children using the adult template 71,[83][84][85] . Moreover, the standard adult SPM template enables us to compare or combine the results of previous studies conducted in adults or across different age groups 72,86,87 . Statistical analysis. All statistical analyses were performed using MATLAB-based custom software (MathWorks, Sherborn, MA, USA) and SPSS 20.0 for Windows (SPSS Inc., Chicago, IL, USA). Partial correlation analyses were carried out to assess which relevant covariates established by the previous literature were associated with sleep disturbances. The factors include the participant's age, sex, TIV, and caffeine consumption 15,17,74 . As a next step, we performed partial correlation analyses (covariates; age, sex, and TIV) to investigate associations of rGMVs with the severity of sleep disturbances. A false discovery rate (FDR) threshold of <0.2 was determined to be significant for addressing multiple comparison issues 78 . FDR thresholding controls the expected proportion of false positives only among brain regions showing significance 78 . Although conservative levels of FDR (e.g., 0.01-0.05) can be used in neuroimaging studies, FDR control levels in the range of 0.1-0.2 are originally and practically known to be acceptable and have been applied in several neuroimaging studies 33,78,79,88,89 . FDR corrected p-values were calculated by using spm_P_FDR.m with all regional p-value inputs, which is a MATLAB code included in the SPM toolbox.