Association between abnormal default mode network homogeneity and sleep disturbances in major depressive disorder

Participants

Twenty-four patients with MDD and chief complaint of sleep disturbance symptoms were enrolled as the Pa_s, and 33 patients with MDD without sleep disturbance symptoms were included as Pa_ns. All patients were recruited from the Second Xiangya Hospital of Central South University, China. Moreover, 32 HCs were recruited from the local community. The inclusion criteria for the patients were as follows: (1) first major depressive episode; (2) 17-item Hamilton Rating Scale for Depression (HAMD-17) total scores ≥18; (3) no other accompanying conditions that may result in sleep disturbances. The severity of depressive symptoms was assessed by the scores of HAMD-17. Anxiety/somatisation severity was evaluated by items 10–13, 15 and 17. Retardation symptoms were evaluated by items 1, 7, 8 and 14. The severity of cognitive disturbances was evaluated by items 2, 3 and 9. The severity of weight loss was evaluated by item 16. The anxiety state was assessed by the Beck Anxiety Inventory (BAI). Sleep disturbance scores were computed by summing the scores of items 4, 5 and 6 of HAMD-17. If sleep disturbance scores are greater than 4 and characterised by chief complaints of sleep disturbance symptoms, the patients with MDD would be assigned to the Pa_s group. If sleep disturbance scores are 4 or lower without chief complaints of sleep disturbance symptoms, the patients with MDD would be assigned to the Pa_ns group20(figure 1).

Request permission

All the participants were aged between 29 and 38 years, right-handed and educationally matched. Patients were diagnosed with MDD independently by two psychiatrists using the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). The participants were excluded if they had other psychiatric disorders except MDD, any severe physical diseases, brain injury, a history of electroconvulsive therapy use or were pregnant. For HCs, they were excluded if they had a family history of mental illness, substance abuse or psychotic symptoms.

The study was conducted in accordance with the Helsinki Declaration. Each participant has submitted written informed consent before enrolment.

Image acquisition

All images were obtained using a 3.0 T GE scanner (General Electric, Fairfield Connecticut, USA). Participants were asked to close their eyes, remain awake and quiet during the scan. Parameters were as follows: repetition time/echo time=2000/30 ms, 30 slices, 64×64 matrix, 90° flip angle, 24 cm field of view (FOV), 4 mm slice thickness, 0.4 mm gap and 250 volumes (500 s).

Data preprocessing

Data were preprocessed by the data processing assistant for resting-state functional magnetic resonance imaging (fMRI) with MATLAB (MathWorks). To mitigate initial signal instability and early adaptation time of subjects, the first 10 original images were discarded. The primary processes included slice timing and normalisation. Participants who had more than 2° of angular rotation in each axis and 2 mm of translation in the x, y or z direction were excluded. After normalisation, the imaging data were temporally band-pass filtered at 0.01–0.08 Hz and linearly detrended. Several covariates were removed, including Friston-24 head motion parameters acquired by rigid body correction, signals from a ventricular region of interest, and signals from a ventricular region centred in the white matter. The global signal was retained as indicated previously.21 After data preprocessing, the DMN mask was constructed using Group Independent Component Analysis of fMRI toolbox (GIFT). The analysis processes were as follows: data reduction, independent component separation and back reconstruction. The DMN components were chosen based on the templates provided by GIFT. For each component, a statistical map was generated by voxel-wise one-sample t-tests (Gaussian random field (GRF) theory corrected at p<0.05 with voxel significance at p<0.001 and cluster significance at p<0.05), and masks were generated to apply in the following NH analysis.

The NH analysis was conducted using MATLAB, which calculated correlation coefficients between each voxel and other voxels within the DMN mask. The resulting NH maps were smoothed with a Gaussian kernel of 8 mm full width at half-maximum.

Statistical analyses

For the differences in the three groups, the χ² test was applied to describe qualitative variables. The analysis of variance was used to compare quantitative variables, such as age, years of education, HAMD-17 total score and the five-factor scores across the groups.

The NH analysis was performed by an analysis of covariance across the three groups, followed by post hoc t-tests. Age, sex, level of education and the average frame-wise displacement were used as covariates. The p value was set at 0.05 and corrected using the GRF (voxel significance at p<0.001 and cluster significance at p<0.05).

The correlation between NH abnormalities and clinical variables, such as HAMD-17 total scores, was analysed using partial correlation analysis.

The receiver operating characteristic (ROC) curve was employed to distinguish between Pa_s and Pa_ns. Subsequently, the optimal cut-off maximising the combined sensitivity and specificity was computed.

Demographic characteristics and clinical information

There were no group differences in age and years of education among the three groups. For the two patient groups, a significant difference was found in sex ratios but not for illness duration. Both patient groups showed higher HAMD-17 total and factor scores than HCs. Additionally, the Pa_s group showed higher BAI scores, HAMD-17 total scores and sleep disturbances than those of the Pa_ns group (table 1).

NH differences across groups

Intriguingly, the NH values showed notable differences in the DMN, mainly in the left superior MPFC and right Pcu across the three groups (figure 2). Compared with Pa_ns, Pa_s showed significantly decreased NH in the left superior MPFC (T=−3.99, p<0.001) and increased NH in the right PCu (T=4.08, p<0.001). Compared with HCs, Pa_s displayed decreased NH in the left superior MPFC (T=−4.43, p<0.001) and increased NH in the right PCu (T=4.97, p<0.001). Between the pa_ns and HCs groups, no difference in NH values was found in the DMN (figure 2 and table 2).

Request permission

Correlations between NH and clinical characteristics

For all patients, there was a negative correlation between the NH values of the left superior MPFC and BAI scores (r=−0.37, p=0.005), a positive correlation between the NH values of the left superior MPFC and retardation (r=0.28, p=0.038) and a negative correlation between the NH values of the left superior MPFC and sleep disturbances (r=−0.42, p=0.001). There was a positive correlation between the NH values of the right PCu and BAI scores (r=0.35, p=0.010), a positive correlation between the NH values of the right PCu and weight (r=0.36, p=0.005), a negative correlation between the NH values of the right PCu and retardation (r=−0.33, p=0.013) along with a positive correlation between the NH values of the right PCu and sleep disturbances (r=0.41, p=0.002). However, only the correlations between the NH values of the left superior MPFC and the right PCu and the sleep disturbances for all patients survived the Bonferroni correction (figure 3).

Request permission

For Pa_ns, there was a positive correlation between the NH values of the right PCu and weight (r=0.37, p=0.035). However, this correlation could not survive the Bonferroni correction. For Pa_s, there was no significant correlation between abnormal NH and the HRSD-17 total scores as well as the five-factor scores.

ROC curves in MDD with and without sleep disturbances

We then used abnormal NH values in DMN to classify Pa_s from Pa_ns. We achieved 81.06% accuracy for the right PCu and 78.03% for the left superior MPFC. Sensitivity reached 83.33% for both the right PCu and the left superior MPFC. Specificity stood at 78.79% for the right PCu and 72.73% for the left superior MPFC. The ROC analysis yielded an area under the curve of 0.84 for the left superior MPFC and 0.83 for the right PCu (online supplemental figure 1).

Main findings

In this study, we demonstrated that the Pa_s group displayed more serious depressive and anxious symptoms than the Pa_ns group. Importantly, Pa_s showed decreased NH in the left superior MPFC and increased NH in the right PCu compared with Pa_ns and HCs. The NH values of the left superior MPFC were negatively correlated with sleep disturbances, while the NH values of the right PCu showed a positive correlation with sleep disturbances. Moreover, abnormal NH values in the DMN can successfully distinguish pa_s from pa_ns.

The superior MPFC is a core region of the DMN, which is important for sleep rhythm maintenance. It is also tightly linked with affective-limbic regions (such as the amygdala and the hippocampus), executive control, as well as emotional processing regions (such as the orbital frontal cortex and anterior cingulate cortex). Through these connections, MPFC may be involved in the modulation of emotional behaviour and self-referential processing.22 Dysfunction of this region, such as decreased NH in the left superior MPFC, can affect the balance between the cognitive system and the ventral emotional system. Therefore, decreased NH in the left superior MPFC may reflect a decrease in emotional regulation capabilities and more severe MDD symptoms.

PCu is usually involved in awareness and conscious self-perception.23 Based on the concept of ‘default mode’, it is the most active cortical region during silent rest.24 PCu activity during conscious resting states contributes to the self-referential ‘thought’ processing, which has the highest metabolic activity in the baseline resting state and consumes about 35% more glucose than other areas of the cerebral cortex.24 Recent functional imaging studies23 also revealed an obvious inactivation of states of consciousness presented by the PCu and adjacent posteromedial cortical regions. Because PCu and prefrontal cortices were less active than the rest of the brain during different sleep cycles,25 26 PCu dysfunction may be associated with sleep disturbances and contribute to abnormal NH in the right PCu.

Numerous studies have consistently demonstrated increased DMN activity in patients with MDD, which may be associated with the prevalent negative self-referential thinking, rumination and difficulties in emotion regulation. Notably, previous studies have indicated a close correlation between abnormal DMN activity and sleep disturbances in individuals with MDD.27 Sleep disturbances may lead to enhanced DMN activity, thus forming a vicious cycle that exacerbates the severity of depressive symptoms. Moreover, excessive DMN activity may relate to decreased rapid eye movement sleep and altered deep sleep stages in patients with MDD.8 These changes can cause poor sleep quality and disrupted sleep patterns. Additionally, the abnormal functional connectivity between the DMN and dorsal attention network may correlate with sleep duration in depression.27 In line with these observations, we have found that patients with MDD and sleep disturbances exhibit abnormal DMN activity, which may contribute to more severe depressive symptoms compared with those without sleep disturbances.

Besides, the functional connectivity between Pcu and MPFC has been reported to significantly correlate with both sleep quality and circadian preference.28 Patients with MDD showed increased activity in the right Pcu, which was associated with disease severity and poor social cognition. Additionally, patients with MDD had significantly altered regional homogeneity in the bilateral Pcu.29 The above results collectively reflect the decreased neural activity in the Pcu of patients with MDD. Consistently, our study confirmed increased NH in the right PCu and decreased NH in the left superior MPFC in Pa_s compared with Pa_ns and HCs. Sleep restriction can inhibit MPFC activation and cause activation in the PCu for the most difficult task condition. The result demonstrated that the two brain regions would make compensatory functional responses. In this study, we found that higher NH values in the MPFC of patients with MDD corresponded to milder sleep disturbances, whereas higher NH values in the PCu indicated more severe sleep disturbances. This could suggest that the MPFC region is still in a functional compensation state, whereas the PCu region has lost this compensation. The increased functional connectivity between the MPFC and PCu within the DMN indirectly supports this possibility.28 Besides, antidepressant treatment can improve the functional connectivity of DMN,30 which suggests that the DMN itself may play a significant role in the pathology of MDD. Therefore, the DMN may play a crucial role in the neurobiological mechanisms of MDD with sleep disturbances.

Our study has several highlights. First, all recruited patients with MDD in this study were first-episode and drug-naive, avoiding the influence of medication and illness duration on NH values. Second, we compared the differences in brain functional connectivity using the NH method between patients with MDD with and without sleep disturbances to reveal NH differences within the DMN. Third, abnormal NH values in DMN successfully differentiated pa_s from pa_ns.

This study is not without its limitations. First, the sample size was relatively small. Second, the age range of the sample was relatively narrow. Third, our study only focused on the abnormal function of the DMN, which may overlook the functions of other brain regions. Finally, we have not analysed other specific factors for sleep problems, such as sleep deprivation and sleep duration, which could potentially bias our results.