Functional Connectivities in the Brain That Mediate the Association Between Depressive Problems and Sleep Quality

Key Points

Question  Is it possible to identify brain functional connectivities that mediate the association of depressive symptoms with poor sleep quality and advance understanding of the differences in brain functional connectivity in individuals with higher scores on the Depressive Problems scale?

Findings  In 1017 participants in the Human Connectome Project, brain areas with increased functional connectivity associated with both sleep and Depressive Problems scores included the lateral orbitofrontal cortex, dorsolateral prefrontal cortex, anterior and posterior cingulate cortices, insula, parahippocampal gyrus, hippocampus, amygdala, temporal cortex, and precuneus. A mediation analysis showed that these functional connectivities underlie the association of depressive problem scores with poor sleep quality.

Meaning  In this study, the increased functional connectivity between these brain regions provides a neural basis for the association of depression with poor sleep quality; in this general population from the United States that was not selected for depression, Depressive Problems scores were correlated with functional connectivities in the brain, including the lateral orbitofrontal cortex, which has implications for the treatment of depression and poor sleep quality.

Importance  Depression is associated with poor sleep quality. Understanding the neural connectivity that underlies both conditions and mediates the association between them is likely to lead to better-directed treatments for depression and associated sleep problems.

Objective  To identify the brain areas that mediate the association of depressive symptoms with poor sleep quality and advance understanding of the differences in brain connectivity in depression.

Design, Setting, and Participants  This study collected data from participants in the Human Connectome Project using the Adult Self-report of Depressive Problems portion of the Achenbach Adult Self-Report for Ages 18-59, a survey of self-reported sleep quality, and resting-state functional magnetic resonance imaging. Cross-validation of the sleep findings was conducted in 8718 participants from the UK Biobank.

Main Outcomes and Measures  Correlations between functional connectivity, scores on the Adult Self-Report of Depressive Problems, and sleep quality.

Results  A total of 1017 participants from the Human Connectome Project (of whom 546 [53.7%] were female; age range, 22 to 35 years) drawn from a general population in the United States were included. The Depressive Problems score was positively correlated with poor sleep quality (r = 0.371; P < .001). A total of 162 functional connectivity links involving areas associated with sleep, such as the precuneus, anterior cingulate cortex, and the lateral orbitofrontal cortex, were identified. Of these links, 39 were also associated with the Depressive Problems scores. The brain areas with increased functional connectivity associated with both sleep and Depressive Problems scores included the lateral orbitofrontal cortex, dorsolateral prefrontal cortex, anterior and posterior cingulate cortices, insula, parahippocampal gyrus, hippocampus, amygdala, temporal cortex, and precuneus. A mediation analysis showed that these functional connectivities underlie the association of the Depressive Problems score with poor sleep quality (β = 0.0139; P < .001).

Conclusions and Relevance  The implication of these findings is that the increased functional connectivity between these brain regions provides a neural basis for the association between depression and poor sleep quality. An important finding was that the Depressive Problems scores in this general population were correlated with functional connectivities between areas, including the lateral orbitofrontal cortex, cingulate cortex, precuneus, angular gyrus, and temporal cortex. The findings have implications for the treatment of depression and poor sleep quality.

Many individuals with depression report poor sleep quality¹ and disturbances of sleep, including insomnia and less slow-wave sleep.² What is the association between depression and sleep? What are the brain systems associated with both depression and sleep quality? Understanding the answers to these questions may lead to better treatments for depression and may improve sleep quality.

Significant correlations between depression and poor sleep quality have been found in a meta-analysis.¹ Poor sleep quality is typically measured with a high value on the Pittsburgh Sleep Quality Index (PSQI),³ which is closely correlated with insomnia.⁴ The widespread use of the PSQI has shown that subjective sleep disturbances are associated with depression.^1,5-7 Moreover, there is a genetic component to the association between poor sleep quality and depression,^2,7 and this makes it especially interesting to search for possible neural mechanisms that may mediate the association. Previously, it has been shown that sleep deprivation affects brain systems that are involved in emotion, such as the amygdala,⁸ and the anterior cingulate cortex has been described as a region in which genes that control circadian rhythms are dysregulated in depression⁹ and in which neurons increase their activity during sleep and disengagement from tasks.¹⁰

Depression is ranked by the World Health Organization as the leading cause of years of life lived with disability.^11-13 In most countries, the percentage of people with depression during their lifetimes falls within an 8% to 12% range.^11,13 There is a major personal burden to these individuals and their families of depression as well as a major economic burden.¹⁴ Thus, advances in our understanding of the neural bases of depression, and how they may be associated with poor sleep quality, are key areas for investigation.

In this study, we analyzed the neural association between depression and sleep quality by measuring the associations between functional connectivity between brain areas in the resting state and both poor sleep quality and depressive problems and used a mediating analysis to assess the underlying mechanisms. Resting-state functional connectivity between brain areas, which reflects correlations of activity, is a fundamental tool in augmenting understanding of the brain regions with altered connectivity and function in mental disorders.¹⁵

The data set used for this investigation was selected from the March 2017 public data release from the Human Connectome Project (HCP; N = 1200) from the Washington University–University of Minnesota (WU-Minn HCP) Consortium. The participants included in this sample were scanned on a 3-T connectome-Skyra scanner (Siemens). The WU-Minn HCP Consortium obtained full informed consent from all participants, and research procedures and ethical guidelines were followed in accordance with Washington University institutional review board approval. The resting-state data were obtained and preprocessed by the HCP with its standardized method,^18-21 with details provided in eFigure 1 in the Supplement. We emphasize that the participants in this investigation were from the Human Connectome Project and were not selected for their existing symptoms or diagnosis of depression; rather, measures of depressive feelings and behavior were available and were used in these analyses.

After preprocessing, the gray matter of the whole brain was parcellated into 250 regions of interest using the Shen atlas,²² which has been validated in resting-state functional magnetic resonance imaging (fMRI) studies^23,24 as described in the eMethods in the Supplement. Then the time series were extracted in each region of interest by determining the mean of the signals of all voxels within that region. Pearson cross-correlations between all pairs of regional blood oxygen level–dependent signals were calculated for each participant, followed by z transformation to improve normality, and the whole-brain functional connectivity network (250 × 250 regions with 31 125 edges) was constructed using 2 scans, as described in the eMethods in the Supplement. Anatomical regions are listed in eTable 1 in the Supplement.

The sleep phenotype of participants in the HCP was assessed by the self-reported PSQI total score, which combines 7 component scores: subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleeping medication, and daytime dysfunction in the past month (with further details in eFigure 1 in the Supplement), and which is considered to be a good measure of insomnia.⁴

The measure of depressive problems in the HCP, the Adult Self-report DSM-IV Depressive Problems raw score, is based on the Achenbach Adult Self-report for Ages 18-59.²⁵ Specifically, the 123 items from Section VIII were administered (of which only 2 were on the topic of sleep quality). Then items associated with depressive symptoms were used to generate the raw score. In addition, this stage of data collection included obtaining information from participants on whether they had ever been diagnosed with depression via the question, “Has the participant experienced a diagnosed DSM-IV Major Depressive Episode over his/her lifetime?”

To investigate the association between functional connectivity and sleep quality (PSQI total score, a very widely used measure of sleep quality that is a good measure of insomnia^1,4-7), linear regression analysis was used, after removing 14 confounding variables: age, sex, years of education, race (categorized as white or other), handedness, head motion (mean framewise displacement), body mass index (calculated as weight in kilograms divided by height in meters squared), blood pressure (diastolic and systolic), total gray matter volume, total white matter volume, alcohol use, tobacco use, and marijuana use. To take into account the family relationships of the participants in the HCP dataset, the significance for all association analyses in this study was estimated by the permutation analysis of linear models technique²⁶ in accordance with prior studies.^20,27 To address the problem of multiple comparisons, the network-based statistic (NBS),²⁸ a well-validated method for brain network association analysis that has previously been used widely in neuroimaging studies,^28-31 was used to identify and ascribe significance to any connected subnetworks found in the set of significantly altered links found in the group with poor sleep quality (eTable 2 in the Supplement). We considered P < .05 (corrected for multiple comparisons) to be significant.

We first investigated the association between the poor sleep quality measure and the depressive measure (ie, Adult Self-report DSM-IV Depressive Problems raw score) by partial correlation analysis after removing the 14 confounding variables described. The correlation was highly significant (r = 0.371; P < .001; Figure 1B). We note that only 2 of the 123 items in the Adult Self Report questionnaire, which are used to calculate the Depressive Problems Score, are on the topic of sleep quality, and so these are unlikely to account for the correlations in this study. Although participants of the HCP data set were not selected to have a psychiatric disorder and most are healthy control participants, the Depressive Problems raw score was a useful indicator of depression, in that 92 participants who reported having been diagnosed with a major depressive episode also had significantly higher scores on this measure.

We used a separate large data set (from the UK Biobank) to test whether the analyses of the HCP data are robust. This data set included 8718 participants.

The sample used in these analyses included 1017 participants (of whom 546 [53.7%] were female; age range, 22-35 years). Data on the participants, including their sleep and depression scores, are shown in Table 1. Further details are in eTable 2 of the Supplement.

Notably, 92 participants (9.0%) had been diagnosed with a major depressive episode (as defined by the DSM-IV) over their lifetimes. These participants had a significantly mean (SD) higher score (7.55 [4.83] points) than the other participants (3.70 [3.07] points; t₉₈₂ = −10.8; P < .001) on the Depressive Problems scale.

There were 162 significant links after NBS correction. Table 2 shows the 60 functional connectivity links that had associations of greatest significance with the PSQI score, including names, supporting statistics, and P values. All are significantly positively correlated with the poor sleep quality assessment. All links that are correlated with the PSQI score are shown in Figure 2. Many significantly correlated links involved the precuneus, the anterior cingulate cortex, the lateral orbitofrontal cortex, the hippocampus and parahippocampal cortex, and some motor areas (Figure 2; eFigure 2 in the Supplement). The correlation between the sleep quality and the mean strength of these significant functional connectivities is shown in Figure 1A (r = 0.181; P = 6.7 × 10⁻⁹).

In the correlation analysis between functional connectivity, poor sleep quality, and depressive scores, we found that the PSQI score was positively correlated with the Depressive Problems raw score (r = 0.371; P < .001; Figure 1B). Then a permutation test was performed, which showed that the 162 significant links identified in the correlation analysis with the PSQI score were also significantly correlated with the Depressive Problems score (Table 2). Figure 2D shows 39 significant individual links identified in the sleep analysis that are also significantly correlated with Depressive Problems scores. The distribution of brain areas associated with these significant links included the lateral orbitofrontal cortex, the anterior cingulate cortex, and the precuneus, as shown in Figure 2C and eFigure 3 in the Supplement. Furthermore, we found that the mean strength of each of the 39 functional connectivities involved in both sleep quality and the Depressive Problems score (shown in Figure 2D) was also positively correlated with the Depressive Problems raw score (r = 0.133; P = 2.1 × 10⁻⁵; Figure 1C).

eFigure 4 and eTable 2 in the Supplement show correlations between the functional connectivities across the whole brain and the Depressive Problems score. Key areas shown include the lateral orbitofrontal cortex, the dorsolateral prefrontal cortex, the anterior and posterior cingulate cortices, and the insula. If the 92 participants who had ever been diagnosed with depression were removed from this analysis, the correlations of some of these areas with the Depressive Problems score were still significant, including the anterior cingulate gyrus and middle frontal gyrus (middle frontal gyrus with right anterior cingulate cortex: r = 0.115; P = .001; left cuneus with left superior frontal gyrus, dorsolateral: r = −0.109; P = .002; right inferior parietal gyrus, excluding supramarginal and angular gyri with left middle cingulate and paracingulate gyri: r = 0.059; P = .01; right middle frontal gyrus with left anterior cingulate cortex: r = 0.059; P = .01; left postcentral gyrus with left inferior temporal lobe: r = 0.059; P = .01; eTable 2 in the Supplement). This provides evidence that some areas can be associated with depressive symptoms even in individuals who had never been diagnosed with depression. However, all of the links involving the lateral orbitofrontal cortex became nonsignificant if this subgroup of 92 individuals was removed (eTable 3 in the Supplement). This was supported by the finding of correlations between the depressive score and many functional connectivities involving the lateral orbitofrontal cortex when only this subgroup of 92 were included. In particular, in this subgroup of 92, correlations were found for the Depressive Problems score for 6 links involving the lateral orbitofrontal cortex (to the left middle frontal gyrus: r = 0.410; P < .001; left supramarginal gyrus: r = 0.377; P < .001; left middle cingulate and paracingulate gyri: r = 0.385; P < .001; left precuneus: r = 0.430; P < .001; left insula: r = 0.348; P < .001; and to 2 ROIs within the right middle cingulate and paracingulate gyri: r = 0.368 and r = 0.402; P < .001 for both), 7 links involving the anterior cingulate cortex (to the left supplementary motor area: r = 0.446; P < .001; right postcentral gyrus: r = 0.368; P < .001; temporal pole [middle temporal gyrus]: r = 0.338; P < .001; left inferior temporal lobe: r = 0.365; P < .001; left middle frontal gyrus: r = 0.321; P < .001; left middle occipital lobe: r = 0.329; P < .001; inferior frontal gyrus, opercular part: r = 0.373; P < .001), and 1 between the lateral orbitofrontal cortex and anterior cingulate cortex (r = 0.365; P < .001). The evidence from the HCP data set was thus that high lateral orbitofrontal cortex functional connectivities were especially associated with the Depressive Problems score, with this association being especially significant in people who at some point in the past had been diagnosed with depression. In the same 92 participants, other areas with high correlations (all with r values greater than 0.32 and corresponding P values less than .001) with the depressive problems score included the angular gyrus, anterior cingulate cortex, parahippocampal gyrus, and temporal cortex (eTable 3 in the Supplement).

Given that the strengths of the functional connectivities were significantly associated with the PSQI score (β = 5.459 [95% CI, 3.10-7.95]; P = 1.2 × 10⁻⁴) and Depressive Problems score (β = 0.0025 [95% CI, 0.00-0.01]; P = 6.4 × 10⁻⁵), we assessed whether the functional connectivities significantly mediated³² this association. The mean strength of the 39 significant functional connectivities shown in Figure 2D involved in both poor sleep quality and Depressive Problems scores significantly mediated the relationship between the PSQI score and depressive problems score (Figure 1D) (4.8% of the total effect size; β = 0.0139 [95% CI, 0.01-0.03]; P = 2.1 × 10⁻⁵). We further found that all 39 links significantly mediated the association between the PSQI score and the Depressive Problems score (eTable 4 and eFigure 5 in the Supplement).

The association patterns between functional connectivities and sleep duration were very similar in both the HCP and the Biobank data sets (Figure 3A). Figure 3B shows a high positive correlation (r = 0.524; P < .001) between the r values for the correlation between the functional connectivity links for the whole brain and sleep duration in the HCP and Biobank data sets. More details are provided in the eTable 5 in the Supplement.

This is the first study, to our knowledge, to examine the neural mechanisms underlying the association of depression with sleep, a topic of great interest, with a large sample (n = 1017). The results show that both poor sleep quality and depressive problems are significantly positively correlated with functional connectivities involving the lateral orbitofrontal cortex, the dorsolateral prefrontal cortex, the cingulate cortex, and the precuneus. Further, a mediation analysis showed that the functional connectivity links between the brain areas identified played a significant role in the association of depressive problems with poor sleep quality. Much smaller associations were found in the reverse direction; that is, the associations of sleep quality with depressive problems mediated by these links were less significant. These findings provides a neural basis for understanding how depression is associated with poor sleep quality, and this in turn has implications for treatment because of the brain areas identified.

Areas in which functional connectivity was associated with depressive problems and their effects on sleep quality included the lateral orbitofrontal cortex, precuneus, and angular gyrus. There is increasing evidence that the lateral orbitofrontal cortex is a key brain region associated with depression. The theory is that it has increased sensitivity in depression of attractor networks activated by not receiving expected rewards (which can lead to sadness and depression).^16,34 Consistently, the lateral orbitofrontal cortex has increased functional connectivity with a number of brain regions, including the precuneus.¹⁷ In depression, the precuneus itself also has increased functional connectivity with prefrontal cortex areas involved in short-term working memory (W. Cheng, PhD, written communication, June 21, 2018). The precuneus is implicated in representations of the sense of self and space and in autobiographical memory.³⁵ These findings support the theory that the nonreward system in the lateral orbitofrontal cortex has increased influence on areas in which the self is represented, which results in low self-esteem, with the increased connectivity to the prefrontal cortex short-term memory system contributing to the rumination symptoms characteristic of depression.^16,17

Part of the importance of the present investigation is that it provides strong support for a role of the lateral orbitofrontal cortex in depression (eFigure 4 and eTables 2 and 3 in the Supplement). The previous findings¹⁷ were in hundreds of patients with major depressive disorder and healthy control participants in China. The present findings were not on patients selected because of their existing depression but instead were on a general population in the United States in which a tendency to have depressive symptoms could be assessed. Interestingly, when the patients who had at some time been diagnosed with depression were removed, the links involving the lateral orbitofrontal cortex no longer exhibited the threshold level of significance. Moreover, if only the 92 participants who had had a diagnosis of depression were considered, then relatively high correlations with functional connectivities involving the lateral orbitofrontal cortex were found. This helps to cross-validate the findings in the data sets from China in which all these areas have been identified in a network of brain areas with different functional connectivity in depression.^17,44 This important cross-validation with participants from the United States provides support for the theory that the lateral orbitofrontal cortex is a key brain area that might be targeted in the search for depression treatments.^16,34,36 Consistent with this, repetitive transcranial magnetic stimulation and inhibition of the lateral orbitofrontal cortex may be useful in the treatment of depression.³⁷

This study found that another brain region strongly associated with the effects of depressive problems on sleep quality is the precuneus. The precuneus is implicated in representations of the sense of self and space and in autobiographical memory.^35,38,39 Acute depression is associated with impaired self-referential processing.⁴⁰ On the bases of the findings of this study and others,^16,17 we propose that the low self-esteem in depression is associated with increased functional connectivity between the precuneus (which is involved in the representation of self) and the lateral orbitofrontal cortex, which is involved in nonreward and therefore implicated in depression.

Another brain region strongly associated with the correlations between depressive problems and sleep quality is the anterior cingulate cortex. Interestingly, the supracallosal part of the anterior cingulate has especially increased functional connectivity with depressive problems and sleep quality and also poor sleep quality, because the supracallosal anterior cingulate has activations associated with punishment and negative emotional stimuli.^41,42 Through separate analyses, we are assessing the implications that the supracallosal part of the anterior cingulate cortex may contribute to the negative and nonreward emotions in depression and their association with sleep quality and that the supracallosal anterior cingulate cortex has high functional connectivity with the convexity part of the lateral orbitofrontal cortex.

Strengths of the present investigation are the large number of participants (1017), leading to robust findings; cross-validation with an independent data set; the mediation analysis, which links the findings to recent advances in understanding brain mechanisms associated with depression^16,17; and the inherent interest of the findings to a wide readership.

Several limitations are considered. Most participants were healthy controls, and the depressive score was used as an indicator of depressive symptoms, not formal diagnosis; however, we did demonstrate a high correlation between the Depressive Problems score and whether a person had at some time been diagnosed with depression. We note that there is a strong association between depression and poor sleep, in that the correlation found in this study between the Depressive Problems score and PSQI score was high (r = 0.371, P < 1.0 × 10⁻¹⁰, Figure 1B); however, this does not address the directionality of the association, which could be in both directions. Indeed, we note that poor sleep quality also may have an effect on depression, as shown by the mediation analysis. We note that the causal relations between sleep and depression is an important topic that deserves much further investigation.²

Although previous studies have found changes in patients with depression in some of the regions of interest identified in this study, including the precuneus and angular gyrus,⁴³ this is to our knowledge the first study to examine the neural mechanisms underlying the associations of depressive problems with sleep quality and to associate sleep quality with functional connectivity in a large sample of participants. A strength of this investigation is that the effects are likely to be very robust, given the large sample of participants with resting-state fMRI and the cross-validation with the UK Biobank data set with a sample of 8718 participants in one of its first uses of resting-state fMRI data. The understanding that we developed in this study is consistent with areas of the brain involved in short-term memory (the dorsolateral prefrontal cortex), the self (precuneus), and negative emotion (the lateral orbitofrontal cortex) being highly connected in depression, which results in increased ruminating thoughts that are at least part of the mechanism that impairs sleep quality.^16,17,44

Accepted for Publication: June 5, 2018.

Corresponding Author: Jianfeng Feng, PhD, Institute of Science and Technology for Brain-inspired Intelligence, School of Mathematical Sciences, Fudan University, Shanghai 200433, China (jianfeng64@gmail.com).

Published Online: July 25, 2018. doi:10.1001/jamapsychiatry.2018.1941

Author Contributions: Drs Cheng and Feng had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Cheng and Rolls contributed equally to this work.

Concept and design: Cheng, Rolls, Feng.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Cheng, Rolls, Feng.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: All authors.

Obtained funding: Cheng, Feng.

Administrative, technical, or material support: Cheng, Feng.

Supervision: Cheng, Rolls, Feng.

Conflict of Interest Disclosures: None reported.

Funding/Support: This work is supported by the key project of Shanghai Science & Technology Innovation Plan (grants 15JC1400101 and 16JC1420402, Dr Feng) and the National Natural Science Foundation of China (grants 71661167002 and 91630314, Dr Feng), the Shanghai AI Platform for Diagnosis and Treatment of Brain Diseases (grant 2016-17), Base for Introducing Talents of Discipline to Universities (grant B18015), grants from the National Natural Sciences Foundation of China (grants 81701773 and 11771010, Dr Cheng), the Shanghai Sailing Program (grant 17YF1426200, Dr Cheng), and the Research Fund for the Doctoral Program of Higher Education of China (grant 2017M610226, Dr Cheng), and Natural Science Foundation of Shanghai (grant 18ZR1404400, Dr Chang).

Role of the Funders/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Use of the Human Connectome Project (https://www.humanconnectome.org/) dataset and the UK Biobank (http://www.ukbiobank.ac.uk/) dataset is acknowledged.