Background
Major depressive disorder (MDD) is a common mental illness with a high lifetime prevalence [
1]. The symptoms of MDD significantly impair patients’ daily functions [
1]. Selective serotonin reuptake inhibitors (SSRIs) are recommended as the first-line treatment for depression [
2]. However, the treatment efficacy with SSRIs varies widely across patients, and the clinical response rate ranges between one-third to two-thirds [
3,
4] due to the heterogeneity in clinical presentations and genetic predisposition [
5]. Previous studies using genetic and serum biomarkers did not capture large variation in treatment response, neither reflect brain signals directly [
6,
7]. In comparison, non-invasive and safe electroencephalogram (EEG) is easier to capture real-time and direct brain signals. The EEG biomarkers might help dissect the biological underpinnings of clinical manifestations and tailor treatment prescription [
8].
EEG can reveal oscillations emanating from the brain in characteristic frequency bands, such as the power values of the theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz). EEG signals were found to distinguish MDD patients and healthy controls [
9]. Moreover, previous EEG studies using spectrum analysis (SA) reported predictive value for treatment response [
9,
10]. However, the potential signals from spectrum EEG analysis could not provide a robust prediction at the patient level [
11,
12]. On the other hand, MDD is increasingly recognized as a disorder with dysregulated neural networks rather than a local brain disorder [
13]. Power SA is considered to reveal the strength of the local signals but does not sufficiently reflect distributed networks related to mood.
One recent systematic review has suggested using functional connectivity (FC) analysis to help reveal the pattern changes of different activities in a depressed patient [
14]. FC analysis refers to the observed connection between interconnected brain areas [
15,
16]. Studies used complex network analysis has its origins in the mathematical study of networks, known as graph theory [
17]. A graph is an abstract representation of a network and it consists of a set of nodes and connections (edges) [
17]. Brain connectivity datasets comprise networks of brain regions that are connected by anatomical tracts or functional associations [
15,
17,
18]. Dissecting functional network topologies among patients reveal the presumed connectivity abnormalities in neurological and psychiatric disorders compared with healthy controls [
19,
20].
A resting EEG study found that FC between the prefrontal cortex and posterior cingulate cortex is elevated in remitted MDD, suggesting EEG FC as a neural marker of depression [
21]. One clinical study found a negative relationship between FC and depressive severity [
22]. A previous magnetic resonance imaging (MRI) study showed FC changes following antidepressant medication, for which increased connectivity between frontal and limbic brain regions was reported [
23]. Another EEG study on Alzheimer’s disease showed the advantages of the FC analysis over the traditional SA method [
24]. Therefore, EEG FC may have the potential to evaluate its predictive power for the treatment response of antidepressants in MDD. Delineating the FC for depression would advance a neurobiological understanding of treatment response and assist in identifying patients who benefit from medication. So far, it lacks studies investigating the predictive performance of EEG, which can compare the results between SA and FC for MDD diagnosis and treatment response. There were several aims in the present study. First, we explored the differences in FC between MDD patients and healthy controls. We also evaluated the differences in FC among MDD patients before and after treatment (the combined effects of antidepressants and therapeutic effects). Second, previous EEG studies have suggested that early EEG changes may correlate with clinical responses [
25]. We hypothesized that the changes in graph-theoretical brain FC in the first week of treatment could serve as markers for evaluating the effectiveness of antidepressants treatment at the week4. Third, we investigated the differences in FC between responders and non-responders (therapeutic effects). Fourth, we explored the correlation between depressive severity at baseline and EEG FC at baseline. Finally, the discriminative ability of MDD diagnosis and treatment response was investigated using band power values and FC analyses.
Discussions
Our results suggest that the greater the FC improvement in the first week, the more reduction of the depressive scores over the four weeks of the treatment period. The FC changes in MDD patients before and after treatment were mainly in the low-frequency (delta) band rather than in the high-frequency band. It’s also noted that the differences in FC between patients before treatment and controls were in the delta band. For the differences between responders and non-responders, we found consistently higher signals in high-frequency waves (alpha and beta bands) in responders than in non-responders over the initial stage of treatment.
Low-frequency band alterations, especially the delta band, have constantly been characterized in depression. One previous meta-analysis of EEG studies showed that MDD patients under an eyes-closed state had significantly increased delta band and theta band activity [
33]. Using the network-based-statistic approach, a recent EEG study showed that differences in the delta band exhibit the most discrimination ability for the diagnosis of MDD [
34]. Comparing MDD with healthy controls, significantly reduced resting brain connectivity was observed in the delta band in the depressed patient [
35]. In addition, synchronization likelihood in the delta frequency bands differentiated depressive patients from controls, with the former exhibiting lower synchronization likelihood than the latter [
36]. It was reported that the global phase synchronization index of the depressive patients had a much lower value than controls [
37]. Taken together, in the current study, controls and MDD patients exhibited different FC patterns, and altered FC in delta bands characterized MDD patients. Our results in the present study are well supported by previous EEG studies, and FC of the delta band may be a promising marker for assisting the clinical diagnosis of MDD. This study found apparent changes in FC among patients who received medication treatment, which are the combined effects of antidepressants treatment and therapeutic effectiveness. To evaluate the contributions of antidepressants use or therapeutic effects, we investigated FC differences between responders and non-responders. At the initial EEG examination, there were significant differences in FC in high-frequency waves (alpha and beta bands) between the responsive and the non-responsive groups. One early study reported that patients who responded to fluoxetine had more significant EEG alpha signals than non-responders [
38], which echoes our findings. Moreover, FC analysis of one week before and after treatment showed that differences in FC persisted in high-frequency bands. These results indicate that higher signals in the alpha and beta band in responders than in non-responders are relatively stable over the initial stage of treatment (i.e., at baseline before treatment and one week after receiving treatment). This is in accordance with the observation of no significant changes in FC before and after treatment among MDD patients in the alpha and beta bands in the present study. An early study also reported no significant EEG changes of alpha power during 12 weeks of treatment with fluoxetine [
38]. One early study found that the difference in alpha power activity between treatment responders and non-responders would not change during the treatment course and suggested that alpha power activity represents state-independent characteristics of treatment [
38]. High alpha power has been found in recovered depressed patients in a euthymic state, which led Pollock and Schneider [
39] to hypothesize that it reflects a specific marker to identify a subgroup of depressed patients with better treatment outcomes. There is an inverse relationship between alpha power and cortical activity [
40]. Increased alpha power was evidence of reduced brain activity in depressed patients. Our findings supported this hypothesis, and these features may help differentiate a subgroup of depressed patients who respond to a SSRI. It is known that serotonergic activity is closely related to arousal. In an awake state, serotonergic cells in raphé nuclei display a constant pattern of discharge that decreases in firing rate as arousal decreases to a sleep state [
41]. It is possible that increased alpha in depressed patients who respond to an SSRI reflects low arousal associated with low serotonergic activity. Some people may worry about whether the co-existing anxiety will affect the results. We have further put the symptoms of anxiety into the regression model for adjusting, which has not changed the original conclusion of this study. Moreover, the beta power showed a similar difference between responders and non-responders. Previous studies found a positive correlation between beta-band activity and attentional performance [
42]. Meanwhile, patients with better cognitive function showed a better response to SSRIs, which provides potential link between beta power and treatment response [
43]. Because patients in this study were allowed to take BZD and BZD can influence effects of beta wave, a more complete drug-free study is needed to further explore the role of beta power in predicting treatment response.
A question remains as to why the clinical improvement of depressive severity in SSRI responders did not normalize their alpha power. Although a common serotonergic mechanism might underlie both depression and EEG abnormalities in responders, they may not have the same pharmacological properties. A preclinical study found that the spontaneous firing of serotonin neurons in the dorsal raphé of rats was not altered after two weeks of escitalopram administration [
44]. In contrast, combined treatment with escitalopram plus bupropion resulted in a marked increase in firing rates. Moreover, the persistence of alpha abnormalities in treatment responders is compatible with an endophenotypic vulnerability marker to MDD [
45]. On the other hand, our findings showed that responders significantly increased FC in the delta band than the non-responders. Further investigation of the physiological roles of the delta band is warranted.
We further investigated the correlation between FC and severity of depression at baseline. A previous EEG study showed a significant negative correlation between FC parameters (degree, efficiency, and betweenness) and HAM-D scores [
46]. However, the correlation was not significant in the present study despite the similar magnitude of the correlation coefficient, which is likely due to the moderate sample size to achieve significance. The amount of FC changes is positively correlated with improvement in depression in mean BC, though. The significance level did not pass multiple corrections, which may require further expansion of the sample size to verify the results. Further, this study found that the power SA in the delta band could slightly differentiate the healthy group from the depressive group, but not in other band frequencies. SA cannot distinguish consistently between patients who respond to SSRI treatment and those who do not. The discrimination ability of FC is better than that of the SA in the current study in terms of treatment response. Comparing with FC, which explores the connectivity between nodes, SA targets on the amplitude strength of brain wave. FC is more in line with the functional characteristics of our brain [
47]. Because MDD is a mental illness that affects brain function, it is plausible that FC has better discriminative ability than SA.
There are some limitations in the current study. First, the sample size is relatively small in the present study. We may not have sufficient power to detect EEG biomarkers with smaller effect sizes. Second, the benzodiazepine was allowed to use at baseline, which may confound the correlation between EEG marker and treatment response. Third, the background noise was difficult to filter completely by manipulation. While the performance of the denoise and artifact removal function is still limited, we examined the EEG to guarantee that muscle movement, head motion, or channels with poor signal were not involved and selected EEG sections with relatively good quality for further processing and analysis. Fourth, patients in the present study were treated mainly with escitalopram. Therefore, the results may not be inferred for the treatment response in patients treated with other antidepressants. Last, although differences between responders and non-responders in alpha and beta power represent stable, state-independent traits, their biological basis is unknown.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.