Major depressive disorder (MDD) is a complex psychiatry disorder that includes three categories of clinical manifestations: disorders of mood, cognitive function, and neurovegetative functions (i.e., sleep and appetite function) (Lisanby, 2007), which severely affects a patient’s quality of life and can even lead to suicide. The prevalence of depression is about 2–4% worldwide and 1.7–2% in China alone (Gu et al., 2013). Hence, effective treatment of depression is a major public health challenge (Sartorius, 2001). Nowadays, compared to some commonly available treatments, such as cognitive behavioral therapy (Siegle et al., 2006; Ritchey et al., 2011) and antidepressant medication (Spronk et al., 2011), electroconvulsive therapy (ECT) (Bouckaert et al., 2016; Yrondi et al., 2018; Takamiya et al., 2019) has long been considered to be the most effective and rapid treatment for depression, and around 70–90% of the depressed patients showed an improvement when treated with ECT (Kellner et al., 2014). However, the mechanisms underlying the action of ECT are still not known.

Mounting evidence suggests that MDD is a system-level disorder that is associated with the dysfunction of neuronal network activity across multiple brain regions (Mayberg et al., 2005). Some structural and functional neuroimaging studies have investigated ECT-induced changes in the structural or functional connectivity of depression by conducting a longitudinal study (Abbott et al., 2013; Bouckaert et al., 2016; Wei et al., 2018). Functional connectivity refers to the degree of temporal correlation between neural signals coming from two different regions (Fingelkurts et al., 2005). Electroencephalogram (EEG), owing to its advantages such as high temporal resolution, non-invasive nature, and a relatively inexpensive technique, becomes a preferable choice for computing functional connectivity, which can be widely applied in clinical and research fields (Sakkalis, 2011; van Diessen et al., 2015).

Existing studies indicated that ECT regulated resting-state EEG oscillatory patterns in the frontal lobe (Farzan et al., 2014) or central nodes of the default mode network (DMN) (Takamiya et al., 2019). Some studies have found that ECT decreased EEG complexity in the responded patients based on the multi-scale entropy method (Okazaki et al., 2013; Farzan et al., 2017). In addition, other EEG studies using low-resolution electromagnetic tomography (LORETA) analysis revealed that ECT-induced theta changes in the subgenual anterior cingulate cortex (ACC) (Reischies et al., 2005) were correlated with decreased depressive scale scores (McCormick et al., 2009). However, compared to neuroimaging studies, the number of studies examining the alteration in ECT-induced functional connectivity networks based on EEG is still small. Only a few EEG studies based on functional connectivity methods reported that ECT changed phase synchronization in the beta/delta frequency band (Deng et al., 2015; Takamiya et al., 2019). Currently, graph theory analysis based on functional connectivity is a simple and straightforward approach to evaluate the topological structure of complex brain networks (He and Evans, 2010; Rubinov and Sporns, 2010). Moreover, graph measures have also been applied to reveal the changes in network topology following seizure therapy. One EEG study that used the graph theory technique reported a decrease of functional connectivity in the beta frequency band and deviation from the small-world architecture, including increased characteristic path length and decreased clustering coefficient, in depressive patients who received a single session of treatment for seizure therapy [either ECT or magnetic seizure therapy (MST)] (Deng et al., 2015). Similarly, another study of EEG based on graph theory analysis suggested an increased clustering coefficient and decreased characteristic path length in the theta frequency band and decreased clustering coefficient in the beta frequency band in depressive patients who received a series of MST (Hill et al., 2021). These results indicate that seizure therapy has the potential to modulate the functional topology of the brain, but larger studies are needed to evaluate the effects of single therapy alone to confirm these preliminary findings. Hence, to expand current research, this study will analyze the change mode of the whole brain topology of depressive patients pre- and post-ECT treatment from the perspective of a functional brain network based on EEG signals.

In this prospective, longitudinal study, we used resting-state EEG signals to examine the functional brain networks in patients with MDD (N = 24) at two time points: (1) prior to ECT and (2) within 7–10 days of completion of ECT series. The first objective of this study was to explore the differences between brain network topology structure pre- and post-ECT in patients with MDD, particularly based on the following three aspects: changes in functional connectivity, changes in graph theoretical measures (e.g., local/global efficiency), and changes in hubs distribution. The second objective of this study was to investigate whether the changes in these network measures correlated with clinical response.

This study used functional connectivity and graph theory analysis to explore the topological changes in resting-state EEG functional brain network of depressive patients pre- and post-ECT treatment. For clinically remitted patients with a HAMD-17 score ≤ 7 and no relapse within 3 months, our main findings were as follows: (1) ECT marginally significantly decreased functional connectivity in the alpha frequency band. (2) ECT increased GE, EBC, LE, and meanD values, which revealed that depressive patients tended to have a superior network structure after ECT treatment. Also, changes in these network metrics had significantly moderate correlations with clinically remitted depressive symptoms in repeated measures correlation. (3) ECT changed the distribution of hubs, mainly in frontal and occipital lobes. Collectively, these findings provided insight into the role of the functional brain network in the underlying mechanism of action of ECT for the treatment of depressive patients.

In this pilot study, when Coh was used to calculate the functional connectivity value, there was a marginal but significant decrease in functional connectivity in the alpha frequency band after ECT treatment. Although normal subjects were not included in this study, the obtained results could be verified with the previous literature. Leuchter et al. (2012) used quantitative EEG coherence to examine differences in the resting-state functional connectivity between patients with MDD and healthy controls. Significantly higher overall coherence in the delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), and beta (12–20 Hz) frequency bands were found in the subjects with MDD. Despite the contrary findings, most studies on depressive patients over the age of 30 years have reported increased functional connectivity (Bohr et al., 2012; Li et al., 2015). The increased functional connectivity of MDD patients indicated that the depressive patients had abnormally activated or overloaded brain activity (Rotenberg, 2004), which might result from the low efficiency of network organization. Moreover, the most significant increases in coherence were found in the alpha frequency band, which could be explained by the failure of the top-down control imposed by rhythmic alpha activity (Leuchter et al., 2012). In this study, post-ECT depressive patients had marginally significantly reduced functional connectivity in the alpha frequency band compared to the pre-ECT depressive patients. Hence, we could speculate that ECT treatment could modulate the abnormal brain pattern in depressive patients and improve the efficiency of brain information communication. Furthermore, a resting-state functional magnetic resonance imaging (fMRI) ECT study had proposed the “hyper-connectivity” hypothesis, which suggested that treatment responses during depressive episodes might be related to decreased seed-voxel connectivity within some brain networks (e.g., cognitive control network and DMN) (Sheline et al., 2010; Perrin et al., 2012). However, another fMRI ECT study showed that the functional connectivity of the left angular gyrus significantly increased after ECT in depression patients, which is accompanied by improved mood (Wei et al., 2018). A related EEG ECT study found increased functional connectivity in the theta frequency band between the posterior cingulate cortex and the anterior prefrontal cortex, and decreased functional connectivity in the beta frequency band between the posterior cingulate cortex and the temporal regions (Takamiya et al., 2019). So, from the inconsistent research results, we consider that the ECT treatment-induced increased or decreased functional connectivity might depend on the brain regions. Future studies should focus on the effect of ECT on functional connectivity of the whole and local brain regions to validate our findings and those of previous studies.

To objectively represent the topological structure of complex brain networks, we adopted two novel binarization methods, CST and MST, to construct the binary brain networks based on the functional connectivity matrix calculated by Coh, ICoh, and PLI, and then extracted network metrics to evaluate the topological characteristics of the networks. It was found that network metrics constructed using the Coh coupling method and CST binarization method showed more significant differences between pre- and post-ECT treatments. However, the network metrics obtained from ICoh/PLI and CST only showed marginally significant differences between pre- and post-ECT patients, which might be affected by the epoch length of EEG signals (Bonita et al., 2014; Fraschini et al., 2016). Though the Coh coupling method was influenced by the artifacts of volume conduction, relevant studies reported that voltage correlation due to volume conduction would be unimportant for electrodes at intervals of 4 cm or more (Doesburg et al., 2005), and a small amount of sparsely distributed electrodes could reduce the effect of volume conduction (Kim et al., 2013). In addition, the network sparsity of CST was significantly higher than that of MST, and the network sparsity of CST was different for every subject (Supplementary Figure 4). The results demonstrated that CST captured differences at low and high thresholds, which might make various network measures more sensitive (Sun et al., 2019). However, MST resulted in a highly sparse network where important edges might be removed, leading to insensitivity to subtle differences in the brain network (Smith et al., 2016). So, for the 19-channel EEG systems, we concluded that combining the coupling method Coh with the binarization technique CST could facilitate the construction of a valid functional brain network.

Based on this combination method, our results found that in the alpha frequency band, network measures (GE, EBC, LE, and meanD) of post-ECT patients with MDD were significantly higher than those of pre-ECT patients with MDD. In addition, increased GE and LE values indicated a general increase in network efficiency. Previous research indicated that compared to normal controls, the shorter characteristic path length and greater GE, as well as lower clustering coefficient and LE, were found in patients with MDD (Li et al., 2015; Li H. et al., 2017). These studies concluded that functional brain networks in subjects with MDD were closer to a randomized network architecture, which might prove that ECT could effectively modulate neuronal network organization and optimize the topological structure of the brain networks in patients with MDD. The results of the evaluation of the network metrics (GE, LE, NBC, EBC, and meanD) in normal controls might reveal that ECT treatment could modulate the abnormal brain pattern of depressive patients tended to normalization, and then improve the efficiency of brain information communication (described in detail in Supplementary Section 6). A recent resting-state fMRI study has indicated that patients with MDD had significantly increased normalized clustering coefficient, normalized path length, and normalized small-worldness after ECT treatment (Sinha et al., 2019). However, an ECT study based on the functional brain network of EEG had reported an opposing conclusion. They found that the brain network of MDD patients deviated from the small-world structure after an ECT session, with a significant increase in the characteristic path length and a significant decrease in the clustering coefficient post-treatment (Deng et al., 2015). However, analysis of only one ECT session might not be able to predict the cumulative neurocognitive changes after multiple ECT treatment sessions, and hence the inconsistent conclusion was also reasonable. Moreover, in Figures 1–5, we can observe that there are individual differences among depressive patients, which may translate into the variability of derived indicators of brain network measures. Hence, a larger sample is required to confirm and generalize our results.

Furthermore, by analyzing the distribution characteristics of hubs, it was found that the hubs mainly existed in the right frontal lobe of patients with MDD before the ECT treatment in the alpha frequency band. These results might reflect the unsuccessful effort of MDD to overcome dysfunction (Rotenberg, 2004). However, after the ECT treatment, the hubs were mainly distributed in the frontal and occipital brain regions of patients with MDD. One relevant study suggested that ECT induced slow-wave oscillations in the frontal brain region (Farzan et al., 2014). Another study found that ECT increased theta-current source density in the ACC and increased connectivity between the anterior frontal cortex and the posterior cingulate cortex regions (Takamiya et al., 2019). Therefore, it is tempting to speculate that ECT might act to rebalance hemispheric activity by regulating the connectivity between frontal and occipital brain regions, thereby alleviating the depressive symptoms.

The results of repeated measures correlation suggested that within the individuals, an increase in the GE, LE, EBC, and meanD values of the alpha frequency band is moderately correlated with a decrease in HAMD-17 scores. From Figure 7, it can be observed that the changes in network metrics and HAMD-17 scores between pre- and post-ECT treatments showed the same trend in most of the patients with MDD, that is, the value of network metrics increased and the depression score decreased. Therefore, we could consider that the increase in network metrics between the pre- and post-ECT treatments correlates with the remission of depressive symptoms. In contrast to repeated measures correlation, the results of Pearson’s correlation showed no association between these changed network metrics and changed HAMD-17 scores. These inconsistent results of the two statistical methods were consistent with the previous studies (Nuninga et al., 2020; Gbyl et al., 2021; Takamiya et al., 2021). In contrast to Pearson correlation analysis, repeated measures correlation estimated the common regression slope, that is, the correlation shared among individuals. Moreover, repeated measures correlation utilized the longitudinal data measured before and after ECT treatment, which greatly improves the statistical accuracy (Bakdash and Marusich, 2017). Therefore, future research studies should try to use repeated measures correlation to reproduce our results.

Several issues need to be addressed further. First, the sample size was relatively small, and there was a lack of comparison with the normal controls under similar experimental conditions and equipment. Hence, larger populations were required to validate and generalize our results. Second, this study used a 19-channel EEG system at the sensor level to investigate functional brain networks. However, to bridge the gap between EEG and fMRI findings of subjects with MDD, high-density EEG (128 and 256 channels) recording combined with source localization technique should be adopted in future studies, as high-density EEG recording could improve topology accuracy (Holmes et al., 2010). Finally, to determine whether the network metrics can act as effective biomarkers to evaluate the efficacy of ECT treatment, future studies may be required to adopt data mining and deep learning algorithms to classify the remission and non-remission patients with MDD after ECT treatment.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Jining Medical University. The patients/participants provided their written informed consent to participate in this study.

BH, GL, and XL originated the study and revised the manuscript. SS and PY conducted data analysis and drafted the manuscript. SS, PY, and SJ collected the relevant data. SS, HC, and XS conducted pre-processing of data. All authors contributed to and have approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.