Abnormal functional connectivity of EEG gamma band in patients with depression during emotional face processing.

Clinical Neurophysiology xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph

Abnormal functional connectivity of EEG gamma band in patients with depression during emotional face processing Yingjie Li a,⇑, Dan Cao a, Ling Wei a, Yingying Tang b, Jijun Wang b,⇑ a b

School of Communication and Information Engineering, Shanghai University, Shanghai 200444, China Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai 200030, China

a r t i c l e

i n f o

Article history: Accepted 31 December 2014 Available online xxxx Keywords: Depression EEG Emotion processing Regular network Negative bias

h i g h l i g h t s Abnormally increased brain functional connectivity was found in patients with depression by

calculating EEG coherence. The brain networks of both the depressed group and healthy controls in gamma oscillation presented

regular network characteristics during emotional processing, but the depressed group showed randomization trends. Healthy controls showed significantly negative bias in the gamma band during emotional processing, while the bias was not detected in patients with depression.

a b s t r a c t Objective: This paper evaluates the large-scale structure of functional brain networks using graph theoretical concepts and investigates the difference in brain functional networks between patients with depression and healthy controls while they were processing emotional stimuli. Methods: Electroencephalography (EEG) activities were recorded from 16 patients with depression and 14 healthy controls when they performed a spatial search task for facial expressions. Correlations between all possible pairs of 59 electrodes were determined by coherence, and the coherence matrices were calculated in delta, theta, alpha, beta, and gamma bands (low gamma: 30–50 Hz and high gamma: 50–80 Hz, respectively). Graph theoretical analysis was applied to these matrices by using two indexes: the clustering coefficient and the characteristic path length. Results: The global EEG coherence of patients with depression was significantly higher than that of healthy controls in both gamma bands, especially in the high gamma band. The global coherence in both gamma bands from healthy controls appeared higher in negative conditions than in positive conditions. All the brain networks were found to hold a regular and ordered topology during emotion processing. However, the brain network of patients with depression appeared randomized compared with the normal one. The abnormal network topology of patients with depression was detected in both the prefrontal and occipital regions. The negative bias from healthy controls occurred in both gamma bands during emotion processing, while it disappeared in patients with depression. Conclusions: The proposed work studied abnormally increased connectivity of brain functional networks in patients with depression. By combing the clustering coefficient and the characteristic path length, we found that the brain networks of patients with depression and healthy controls had regular networks during emotion processing. Yet the brain networks of the depressed group presented randomization trends. Moreover, negative bias was detected in the healthy controls during emotion processing, while it was not detected in patients with depression, which might be related to the types of negative stimuli used in this study. Significance: The brain networks from both patients with depression and healthy controls were found to hold a regular and ordered topology. Yet the brain networks of patients with depression had randomization trends. Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

⇑ Corresponding authors at: P.O. Box 98, 99 Shangda Road, Baoshan District Shanghai University, Shanghai 200444, China. Tel./fax: +86 21 66137258 (Y. Li). 600 Wan Ping Nan Road, Shanghai Mental Health Center, Shanghai Jiao Tong University, School of Medicine, Shanghai 200030, China (J. Wang). E-mail addresses: [email protected] (Y. Li), [email protected] (J. Wang). http://dx.doi.org/10.1016/j.clinph.2014.12.026 1388-2457/Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

Please cite this article in press as: Li Y et al. Abnormal functional connectivity of EEG gamma band in patients with depression during emotional face processing. Clin Neurophysiol (2015), http://dx.doi.org/10.1016/j.clinph.2014.12.026

2

Y. Li et al. / Clinical Neurophysiology xxx (2015) xxx–xxx

1. Introduction With more and more research conducted on functional brain imaging in major depressive disorder (MDD), it is indicated that observed psychopathology might be related to the distributed property of large-scale cortical systems with a number of functionally connected cortical regions (Damasio, 1994; LeDoux, 1996; Rippon et al., 2001; Lagopouplos et al., 2012). Recent studies provided further evidence that individuals with MDD tended to have altered brain networks (Berman et al., 2011). Furthermore, emotional processing is also completed by cooperating among different brain regions (Galderisi and Mucci, 2002). However, how MDD, as a disorder characterized by a distinct change of mood, organizes their emotional brain network still remains unclear. Most studies indicated that the synchronization of oscillations has been used to identify networks of interacting brain regions (Ertl et al., 2013), and scalp electroencephalogram (EEG) is used to investigate neural oscillations generated in cortical brain structures. To date, many EEG studies have revealed that oscillations in different frequency bands have individual connections with emotional processing in different ways (see review of Güntekin and Basar, 2014). In this review, the related studies presented that beta and gamma oscillatory responses reflected fast and automatic processing of negative stimuli, while the alpha band gave inconsistent results from various groups. The authors concluded that the search for functional correlates of brain oscillations has been an important trend in neuroscience. EEG coherence is an efficient method to calculate the linear-dependent interaction of EEG signals between two channels or brain regions on the frequency domain without restrictedly priori assumptions (Andrew and Pfurtscheller, 1996; Pfurtscheller and Andrew, 1999). Compared with other methods such as partial directed coherence and phase synchronization estimating the function connectivity, EEG coherence can be used conveniently to build network frameworks in the EEG frequency domain (Di and Rao, 2007). A report based on an auditory oddball task showed that most patients with schizophrenia had abnormal patterns of coherence in the temporal lobe (Calhoun et al., 2011). Meanwhile, Leuchter pointed out that patients with MDD in the resting state had significantly higher overall coherence than controls in the frequency range of 0.5–20 Hz (Leuchter et al., 2012). Therefore, one may come to a conclusion that coherence is available to characterize the interaction in brain dynamics. On the other hand, EEG coherence only indicates the interconnection between two structures or brain regions, which does not reveal the global network property. Graph theoretical analysis (GTA) provides a framework for understanding brain global network topology. GTA has been widely used in the research of brain networks (Stam and Reijneveld, 2007; Bullmore and Sporns, 2009). This approach offers an unique window into the balance of local and distributed interactions occurring in the brain (Fingelkurts et al., 2005). The graph theory enables the detection of so-called small-world network architecture, which combines the high clustering of the regular network with short path lengths of the random network (Sporns and Zwi, 2004; Bassett and Bullmore, 2006; Uehara et al., 2013). Reports on many neuropsychiatric diseases, such as Alzheimer’s disease (AD) (Stam et al., 2009, 2007), schizophrenia (Jalili and Knyazeva, 2011; van den Heuvel et al., 2010), and epilepsy (Bernhardt et al., 2011; Ponten et al., 2007), as well as on patients with depression have consistently showed a randomization of network topology and a disruption of the small-word network architecture. A report based on a functional magnetic resonance imaging (fMRI) study showed that patients with MDD exhibited reduced negative blood oxygenation level-dependent responses in the core cortical midline regions of the default-mode network during processing of emotion stimuli (Grimm et al., 2009). Further fMRI study demonstrated that both

MDD and healthy controls had small-world architecture in resting-state brain networks, while MDD had a shift toward randomization in their brain networks (Zhang et al., 2011). Meanwhile, researchers reported that healthy controls exhibited neuronal networks closer to the ordered part of the rewiring scale, while patients had brain networks closer to the random part of the scale during sleep (Leistedt et al., 2009). However, little evidence is available about whether the complex network connectivity of depression during cognitive tasks has small-word properties. In addition, studies of facial emotion processing play an important role in the research of emotion and cognition in MDD. Most of the neuroimaging research show abnormalities in patients with MDD in a common face-processing network, indicating mood-congruent processing bias of hyperactivation to negative stimuli and hypoactivation to positive stimuli, particularly in the amygdala, insula, parahippocampal gyrus, fusiform face area, and putamen (see review of Stuhrmann et al., 2011). In particular, negativity bias, a well-known concept in psychology, was reported by many theorists in that negative experience or fear of bad events might have a greater impact on people than neutral experiences or even positive experiences might (Baumeister et al., 2001; Vaish et al., 2008). Researchers then assumed that it is the reason for the bias that unpleasant stimuli can produce stronger emotional effects than pleasant stimuli (see review of Olofsson et al., 2008). Large-scale studies provided a reasonably consistent evidence that negative response bias towards sadness existed in individuals with major depression, so that positive (happy), neutral, or ambiguous facial expressions were evaluated as more sad or less happy compared with healthy controls (see review of Bourke et al., 2010). Therefore, we would like to uncover if the underlying attentional bias is different in a healthy person compared to a person with depression. Gotlib et al. (2004) found that patients with depression exhibited specific bias to the emotion of sadness, while not to the angry or happy faces. As both angry faces and sad faces are negative stimuli, the mechanism of negative bias for normal person and participants with depression should be different. However, it is still uncertain whether this bias affects the whole brain networks. In our previous study, we used a face-in-the-crowd task to induce attention-modulation processing in depression, and we found hypoactivity in response to the positive face in the left frontal region and hyperactivity in response to the negative one in the right frontal region (Tang et al., 2011). However, we could not recognize the stimuli as sadness or anger as schematic faces were used in the study. Therefore, we applied a similar visual search for real facial expressions task for patients with MDD and healthy controls in this study. In this work, we used EEG coherence to measure the correlations between all pairs of electrodes. The brain functional networks were established in individuals with depression and healthy controls, and the GTA was applied to study the network characteristics. Our hypothesis was that patients with depression, as the other neuropsychiatric diseases, might have an abnormal brain functional connectivity during emotional face processing compared with healthy controls. We therefore investigated the EEG’s specific network features in different frequency bands, and we confirmed that the functional brain networks of depression were characterized by a loss of small-world features. 2. Materials and methods 2.1. Participants The depressed group included 16 right-handed outpatients with depression (male/female = 6/10, 37.75 ± 14.19 years old, 12.06 ± 2.91 years of education) recruited from the Shanghai Mental Health Center (SMHC). All participants with depression



fulfilled the CCMD-3 (the third revision of Chinese Mental Disorder Classification and Diagnosis Standard) diagnosis criteria of MDD (current episode of depression) and had no history of manic episode. The control group included 14 right-handed healthy participants (male/female = 4/10, 40.86 ± 12.29 years old, 11.54 ± 3.75 years of education) with no personal history of neurological or psychiatric illness. All patients were unmedicated or had not taken any medicine for at least 1 month. None of the controls were under psychoactive medication. All participants including healthy controls had no history of any substance or alcohol abuse. Each subject had normal or corrected-to-normal vision. Before experiments, all participants participated in an interview in which the Hamilton Rating Scale for Depression (HAMD) was administered. The Self-rating Anxiety Scale (SAS) and Self-rating Depression Scale (SDS) were self-rated. The scores of the normal group were in the normal range showing no mood disorder (see Table 1). All participants signed the informed consent form (ICF) before the experiment, and they were paid after the experiment. The study protocol was approved by the Institutional Review Board (IRB) of the Shanghai Mental Health Center. 2.2. Materials and procedure

3

2.3. EEG recording and data preprocessing EEG was recorded from 64-channel surface electrodes mounted in an elastic cap (QuickCap™, Brain Products Inc., Gilching, Bavaria, Germany). Fifty-nine electrodes were selected from 64 electrodes that covered the whole scalp: Fp1, Fpz, Fp2, AF7, AF3, AF4, AF8, F7, F5, F3, F1, Fz, F2, F4, F6, F8, FT7, FC5, FC3, FC1, FC2, FC4, FC6, FT8, T7, C5, C3, C1, Cz, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPz, CP2, CP4, CP6, TP8, P7, P5, P3, P1, Pz, P2, P4, P6, P8, PO7, PO3, POz, PO4, PO8, O1, Oz, and O2. Data recording was referenced to the tip of the nose. EEG was continuously recorded at a sampling frequency of 1000 Hz, and the interelectrode impedance was kept below 5 kX. Artifacts from vertical and horizontal eye movements and blinks were removed off-line by an ocular correction algorithm using a Brain Vision Analyzer (Brain Products Inc., Gilching, Bavaria, Germany). The artifact-free data were band-pass-filtered between 0.05 and 100 Hz. Data were segmented from 200 ms before stimulus onset to 1000 ms post stimulus, and they were baseline-corrected to the first 200 ms of the epoch. Segmentations with artifacts (>±100 lV) or leading to incorrect answers were excluded. 2.4. Network analysis

The face-in-the-crowd task was constructed according to the experiments of Mark A. Williams et al. (Williams et al., 2008; White, 1995). Specifically, the stimuli consisted of six human faces, which were selected from the Ekman emotion database (Ekman and Friesen, 1976). There were three kinds of expressions (positive, negative, and neutral) without hair, glasses, beard, or other facial accessories. All images were software-edited using Adobe Photoshop and converted to gray scale. The stimulating faces appeared in the six vertices of the hexagon randomly with one different target face. The experiment contained four blocks and each had 144 trails (i.e., 72 positive + 36 negative + 36 neutral faces). Each trail was displayed for 1500 ms with a black background. Then an interstimulus interval (ISI) of 1000 ms was presented when a fixation cross appeared alone in the center of the screen. The sequence of events in a trial is shown in Fig. 1. The experiment was GO-NOGO paradigm and programmed by E-Prime (Version 1.0). The participants were comfortably seated 80 cm from a 17-inch LCD-screen, and they were instructed to look at the cross in the center of the screen. All participants were asked to judge whether the present image contains the target face, positive or negative ones, during the stimulus onset asynchrony. They were asked to press button ‘‘1’’ if positive face stimuli were found, and button ‘‘5’’ for negative face stimuli. There was a break period of 1 min between blocks, and the whole experiment took about 30 min for each subject.

Table 1 Demographic and affective characteristics of normal controls and patients with depression (mean ± SD). Normal controls

Patients with depression

Cases (n) Handedness (left/right) Age (years) Education (years) HAMD score SAS score SDS score

14 0/14 40.86 ± 12.29 11.54 ± 3.75 7.27 ± 6.94 35.5 ± 5.13 0.48 ± 0.09

16 0/16 37.75 ± 14.19 12.06 ± 2.91 24.5 ± 7.40 61.3 ± 9.74 0.89 ± 0.08

Gender Male Female

4 10

6 10

Functional brain networks can be explored using the graph theory through the following steps: (1) defining 59 nodes in the networks, which represent the 59 electrodes and cover the whole scalp, (2) using EEG coherence to measure the association between nodes, and then constructing brain network models, and (3) calculating the complex network parameters in the graphical model of brain networks. 2.4.1. Coherence analysis Coherence is defined as the spectral cross-correlation between two signals normalized by their power spectra. With the two given EEG signals x and y considered, the magnitude-squared coherence (MSC) or simply coherence is calculated for a particular frequency f by taking the square of the cross-spectral density function and then normalized by their individual auto-spectral density functions (Sakkalis, 2011).

Cxy ¼

jSxyðf Þj2 Sxxðf ÞSyyðf Þ

ð1Þ

The estimated MSC ranges between 0 and 1, where 0 indicates indicated no coupling between two signals and 1 indicates maximum linear interdependence between two signals. In this study, the coherence was calculated between each possible pair of EEG channels with respect to each single frequency. For each segment (1000 ms after the stimulus presented), a square 59 59 matrix was obtained (59 was the number of EEG channels), and the coherence matrix was defined with each element as the mean value of all segments from each subject. We also calculated the global coherence defined as the mean value of all elements in the coherence matrix. The global coherence indicated the level of interdependence of the whole brain. To study the level of interdependence in specific physiological frequency ranges, we considered a frequency-averaged coherence value in the following interest bands: delta band (1–4 Hz), theta band (4–8 Hz), alpha band (8–13 Hz), beta band (13–30 Hz), low gamma band (30–50 Hz), and high gamma band (50–80 Hz). 2.4.2. Complex networks analysis According to the graph theory, the brain can be viewed as a complex functional network basically represented as graphs. Applying GTA to coherence matrices is to convert the matrix into


4


Fig. 1. Sequence of events in a typical trial from the task. The stimulus consisted of six human faces, which were selected from the Ekman emotion database. There were three kinds of expressions (positive, negative, and neutral) without hair, glasses, beard, or other facial accessories. Additionally, images were all software-edited using Adobe Photoshop and converted to gray scale. The stimulating faces appeared in the six vertices of the hexagon randomly with one different target face. Each trail was displayed for 1500 ms with a black background. Then an interstimulus interval (ISI) of 1000 ms was presented when a fixation cross appeared alone in the center of the screen.

a binary undirected graph. Binary represents the edges between nodes. The coherence matrix can be converted to a graph by considering a threshold T. An edge exists if the coherence value between a pair of channels exceeds T, absent otherwise. The choice of T had a significant influence on the constructed graphs in that fixed low T value generated densely connected networks, whereas networks based on large T values were sparse. As there was no generally accepted approach to choose the appropriate threshold, we researched the whole range of thresholds (0.05 < T < 1, with increments of 0.05) and repeated the full analysis for each value of T. The topological properties of the graph were characterized once the conversion of the coherence matrix to a graph was completed. A graph could be characterized by clustering coefficient C and characteristic path length L. These two indexes correspond to the two basic brain functional organizing principles, that is, functional segregation and integration (Friston, 2009). Functional segregation is the ability of information processing in a specialized manner. The clustering coefficient C is a potential measure of the functional segregation. C quantifies the number of connections that exist between the nearest neighbors of a node as a proportion of the maximum number of possible connections (Strogatz, 2001). The clustering coefficient of the whole network is equal to the average clustering coefficient of all nodes. It is a measure of the local interconnectedness of the graph, and it ranges from 0 to 1. Functional integration is an ability of the brain to combine information in different brain regions. The characteristic path length L is a good way to measure the functional integration. L is defined as the average minimum number of edges that must be traversed to go from one node to another. It represents the overall connectedness and indicates how well the nodes are interconnected in the graph. To establish network topological characteristic, we calculated the ratios C/Cs and L/Ls as a function of threshold T where Cs and Ls denote the values of C and L for matched random reference graphs. The randomized network defined by Maslov and Sneppen (Maslov and Sneppen, 2002) had the same nodes and connectivity as in the original one, whereas the choice of their interaction nodes was totally random. The actual data were considered as an undirected network, and the randomized version of network was constructed by randomly reshuffling links, which kept the in- and out-degree of each node. Fig. 2 shows the details of the algorithm performing the randomization. After that, 20 such random networks were generated and the average Cs and Ls were calculated for a comparison with the actual networks. Small-world network organization is evident when values of C/Cs are significantly greater

than 1 while the values of L/Ls are near the value of 1 (Stam and Reijneveld, 2007). The software used in this study was Matlab for off-line calculation of the coherence value, and Brain Connectivity Toolbox (http://www.stanford.edu/~dgleich/programs/matlab_bgl/) for the calculation of two graph theoretical measures. 2.5. Statistical analysis The condition with positive/negative target (with one emotional face among five neutral faces) was considered in the study. Repeated measures analysis of variance (ANOVA) was performed to test the significant changes of coherence values and network characteristics when a within-group factor Emotion (positive and negative) and a between-group factor Group (the depressed group and the normal) were considered as fixed factors. The simple effect analysis was performed if any interaction between factors was found. All analysis was conducted at the 0.05 level of significance. 3. Results We divided EEG signals into common frequency bands as described in Section 2.4.1. The results of EEG coherence had statistically significant discrepancies only in the low gamma band and high gamma band. 3.1. The global coherence The coherence matrices of both groups and both emotions showed a complex but rather similar pattern, with various regions of high and low levels of interdependence. Fig. 3 shows these matrices in the high gamma band as an example. Compared with healthy controls, the patients with depression had more strengthened coherence values in the frontal and occipital region. The global coherence was calculated to measure the mean value of all elements in the coherence matrix. We compared the global coherence values of positive and negative emotional stimuli within group, and the differences between groups. A significant Group effect was observed on the global coherence in the high gamma band (F(1.28) = 4.552, p = 0.042). The interaction effect of group emotion was significant in the low gamma band (F(1.28) = 3.422, p = 0.075). The simple effect analysis was performed afterwards. As shown in Fig. 4, the Emotion effect was found in healthy controls (low gamma: F(1.13) = 5.122, p = 0.041; high gamma:



5

Fig. 2. The flowchart of the algorithm for generating the randomized network.

Fig. 3. The coherence in the high gamma band of patients with depression (the first row) and healthy controls (the second row) under the positive (the first column) and negative (the second column) facial stimuli. The size of the coherence matrix was 59 59. Each element represented the coherent value between two channels, in which the values of the main diagonal were 0. The horizontal and vertical axis denoted 59 channels.


6


Fig. 4. The statistical results of global coherence on groups and emotions in both gamma bands. (A) In the low gamma band, the global coherence in the positive condition for the depressed group was higher than that for healthy controls. (B) In the high gamma band, the global coherence of patients with depression was significantly higher than healthy controls in emotion processing. (C) In the low gamma band, the global coherence in the negative condition was significantly higher than that in the positive condition for normal controls. In the high gamma band, the global coherence in the negative condition tended to be significantly higher than that in the positive condition. ⁄⁄Denotes the value of p to be significantly smaller than 0.05; ⁄denotes the value of p to be between 0.05 and 0.1.

F(1.13) = 4.044, p = 0.066). The global coherence in the negative condition was higher than in the positive condition. However, there was no Emotion effect in the depressed group (low gamma: F(1.15) = 0.333, p = 0.573; high gamma: F(1.15) = 0.188, p = 0.670). Moreover, in the high gamma band, the Group effect was found such that the depressed group had a higher global coherence than the healthy group (F(1.28) = 4.552, p = 0.042). On the other hand, in the low gamma band, the statistic result indicated that the trended Group effect (F(1.28) = 3.166, p = 0.086) was only in the positive condition. 3.2. Network parameters 3.2.1. The mean ratio C/Cs and L/Ls with the change of thresholds T Fig. 5 shows the mean ratio C/Cs and L/Ls as a function of threshold T for two groups. The values of C/Cs for both groups were significantly greater than 1, while the values of L/Ls for both groups were greater than 1. This stated that both the brain networks of the depressed group and healthy controls presented regular network organizations. For those of T value points indicated by asterisks, the values of C/Cs in healthy controls were greater than those in the depressed group. The values of L/Ls in the healthy group were significantly higher than that in the depressed group at most of the T value points (indicated by asterisks) through two gamma bands. This indicated that the brain networks of the depressed group tended to be a shift towards randomization compared with healthy controls.

3.2.2. The topographic maps of clustering coefficient Combining the brain network attributes in a certain time with the locations of brain regions, the measure was adopted by plotting the clustering coefficient of each node by topographic maps. The topographic maps were plotted by the threshold 0.3, which can clearly indicate the difference of two groups in both gamma bands according to our above results. Fig. 6 shows the topographic maps of the clustering coefficient for the depressed group and healthy controls in both gamma bands under positive and negative conditions. Table 2 shows the statistical analysis of the mean clustering coefficient for each electrode in different conditions, which indicates the statistical discrepancies for the topographic maps. In the low gamma band, the group effect for the positive condition was mainly found in the frontal region (Fp1, Fp2, Fz, FT7, FC3, and FC1). Similarly, the group effect for the negative condition was also found in the frontal region (Fp1, Fp2, F7, and F1). In the high gamma band, the group effect for the positive and negative conditions was mainly found in the frontal region (Fp1, Fpz, Fp2, AF8, F5, F1, F8, FC3, FC1, FC4, and FC6), and partly in other regions (T7, T8, C1, CPz, CP6, P2, P4, P6, PO3, and O2). These results demonstrated that the depressed group had significantly higher clustering coefficient in the frontal region than healthy controls in both gamma bands under positive and negative conditions. However, there was little statistical discrepancy for the Emotion effect between the depressed group and healthy controls in both gamma bands. Finally, the number of comparisons tested suggests caution in evaluating what are clearly exploratory statistical inferences.



7

Fig. 5. The mean ratio C/Cs and L/Ls as a function of threshold T for two groups in two gamma bands under both emotion conditions. The values of C/Cs in the depressed group were significantly smaller than those in healthy controls in both gamma bands. Furthermore, the values of L/Ls in the depressed group were also significantly smaller than those in healthy controls in both gamma bands. The horizontal axis presents different thresholds. The values under asterisks (⁄) denoted the values of C/Cs or L/Ls at different thresholds when the Group effect was significant at these thresholds. ⁄⁄Denotes the value of p to be smaller than 0.05; ⁄denotes the value of p to be between 0.05 and 0.1.

Fig. 6. The topographic maps of the clustering coefficient in the depressed group and healthy controls in positive and negative conditions in both gamma bands for T = 0.3. The first row represents the maps of the depressed group in the positive and negative conditions in low gamma and high gamma bands, and the second row represents the maps of healthy controls in the same conditions.

3.2.3. The topology structures of the networks in three specific thresholds T According to the above results, the topological structures of functional networks were plotted between the depressed group and healthy controls at different thresholds in the high gamma band under positive and negative conditions. Specifically, the connection edges were plotted existing in the brain network of the

depressed group as well as healthy controls with conditions of three specific threshold values: 0.15, 0.3, and 0.45 according to the network results (see details in Fig. 5). As the topological difference between two frequency bands was not obvious, Fig. 7 only showed the networks in the high gamma band. Fig. 7 was drawn by subtracting one coherence matrix from another one. The difference 1 or 1 meant the existence of connected edges in one


8


Table 2 The statistical effects of the mean clustering coefficient for each electrode under different conditions. Condition

Electrode

F, P

Low gamma:positive_group effect

Fp1 Fp2 Fz FT7 FC3 FC1

F = 22.600, P = 0.000 F = 5.070, P = 0.032 F = 3.196, P = 0.085 F = 4.012, P = 0.055 F = 4.266, P = 0.048 F = 5.400, P = 0.028

Low gamma:negative_group effect

Fp1 Fp2 F7 F1

F = 7.887, F = 3.348, F = 3.741, F = 3.601,

High gamma:positive_group effect

Fp1 Fpz Fp2 AF8 F5 F8 FC3 FC1 FC4 FC6 T8 CPz CP6 P2 P4 PO3 O2

F = 11.855, P = 0.002 F = 11.062, P = 0.002 F = 5.875, P = 0.022 F = 4.220, P = 0.049 F = 4.225, P = 0.049 F = 5.448, P = 0.027 F = 3.780, P = 0.062 F = 8.446, P = 0.007 F = 4.753, P = 0.038 F = 3.736, P = 0.063 F = 4.014, P = 0.055 F = 3.431, P = 0.075 F = 3.323, P = 0.079 F = 5.234, P = 0.030 F = 4.132, P = 0.052 F = 3.661, P = 0.066 F = 6.133, P = 0.020

High gamma:negative_group effect

Fp1 Fpz Fp2 F1 F8 FC3 FC1 FC4 T7 C1 CPz CP6 P2 P4 P6 PO3 O2

F = 7.252, F = 8.464, F = 4.138, F = 3.121, F = 4.714, F = 7.258, F = 3.788, F = 4.489, F = 4.130, F = 3.491, F = 3.993, F = 4.297, F = 3.581, F = 4.147, F = 3.580, F = 3.931, F = 5.871,

P = 0.009 P = 0.078 P = 0.063 P = 0.068

P = 0.012 P = 0.007 P = 0.051 P = 0.088 P = 0.039 P = 0.012 P = 0.062 P = 0.043 P = 0.052 P = 0.072 P = 0.055 P = 0.047 P = 0.069 P = 0.051 P = 0.069 P = 0.057 P = 0.022

network or not. The difference 0 meant the existence of the edges in both networks and nothing, and the connected edges with 0 were not shown in the topological map. As shown in Fig. 7, we found the following: (1) when the depressed group processed the emotional stimuli, the connected edges of brain networks were more than those of brain networks in healthy controls; (2) when the threshold increased, the connected edges of brain networks in healthy controls declined sharply, but slowly in the depressed group; and (3) the connected edges of brain networks in the negative condition were more than those in the positive condition when healthy controls processed the emotional stimuli. 4. Discussion Prior work has documented the altered brain functional connectivity in MDD, especially in sleep and resting state (Sheline et al., 2010; Greicius et al., 2007), whereas it is necessary to characterize the brain networks while patients with MDD are processing emotion information. The principal finding of the present study is that patients with depression indeed have an abnormal strengthened functional connectivity in gamma bands. The brain networks of

the patients and healthy controls presented a regular network organization while the networks of the depressed group tended to be a shift towards randomization according to the results of C/Cs and L/Ls only in the gamma bands. 4.1. Why did we find the abnormality only in gamma oscillations? In this paper, we did not find any significance in other frequency bands but in the gamma band. Gamma oscillation about a spatiotemporal magnetic field was interpreted as a coherent sweep of activity due to a continuous phase shift over the hemisphere (Llinás and Ribary, 1992). After a large number of studies on gamma oscillations, they were widely regarded as a crucial part in integrating distributed neural processes into highly ordered cognitive functions, such as emotional processes (Basßar et al., 2001; Keil et al., 2001). However, there is no consistent conclusion on emotion-related processing in other bands (Güntekin and Basar, 2014). In our study, we reconfirmed the sensitiveness of gamma oscillation to emotion processing. We found that global gamma coherence presented a significant Emotion effect indicating that negative stimuli induced higher coherence than positive stimuli for the normal group. Furthermore, the negativity bias was found only in normal controls in this study, which is discussed in detail in Section 4.5. In addition, patients with depression showed specific gamma activity according to the presented research. The global coherence of the depressed group was significantly higher than that of healthy controls in gamma bands under the positive stimuli. A recent report was published that individuals with depression displayed sustained and increased gamma band (35–45 Hz) EEG power to negative emotional words (Siegle et al., 2010), which is consistent with ours. In 2003, Kay first reported that gamma oscillations in a local field potential were of two types, which included high gamma (65–100 Hz) and low gamma (35–65 Hz). His research proved that high gamma activity was prominent during exploratory behavior, and low gamma activity was prominent during alert immobility (Kay, 2003). Study on ERS also showed that the temporal and spatial characteristics of low and high gamma activity were distinct, suggesting relatively independent neurophysiological mechanisms (Crone et al., 1998). For example, processing both negative facial emotions and negative feedback elicited significant high gamma (50–150 Hz) responses (Jung et al., 2011). In our study, the searching task needed subjects to explore the target all the time. Therefore, our significant results appeared primarily in the high gamma band (50–80 Hz). 4.2. The hyperactivity of brain functional connectivity of patients with depression In the present study, increased global coherence of the depressed group was found significantly existing in the high gamma band, while the global coherence of the depressed group only increased in the low gamma band under the positive condition. A typical report indicated that an increase in coherence could fulfil the criteria required for the formation of Hebbian cell assemblies, which were bound together with parts of the brain (Miltner et al., 1999). Moreover, when subjects were in a depressive state, their brain became informationally activated or overloaded (Rotenberg, 2004). We therefore interpreted the increased coherence in the depressed group as the abnormal activated or overloaded communication in depressive brains. A recent study based on quantitative electroencephalography (qEEG) showed that patients with MDD had significantly higher overall coherence compared with controls at 0.5–20 Hz, which suggested a broad loss of selectivity in functional connections in MDD (Leuchter et al., 2012).



9

Fig. 7. The brain network topological structures of two groups for three thresholds in the high gamma band with positive and negative conditions. (a1) and (a2) show the networks of both groups in the positive condition. We drew the networks through subtracting the matrix of healthy controls by the matrix of the depressed group. (a1) represents the connected edges only in the network of the depressed group in the positive condition (the difference value = 1) and (a2) represents the edges only in the network of healthy controls (the difference value = 1). (b1) and (b2) show the networks of both groups in the negative condition. (b1) shows the connected edges only existing in the network of the depressed group (the difference value = 1) and (b2) shows the connected edges only existing in healthy controls and not in the depressed group. In order to demonstrate the difference between two emotional conditions, we used the same method to draw the networks in the normal group as above. (c1) shows the connected edges only in the positive condition (the difference value = 1), and (c2) only in the negative condition (the difference value = 1). The values of T from left to right: T = 0.15; T = 0.3; and T = 0.45.

Other studies based on fMRI also demonstrated that there were significant increases in resting-state cortical functional connectivity in MDD (Sheline et al., 2010; Zhou et al., 2010; Greicius et al., 2007). Our results suggested that patients with MDD had abnormally increased connectivity during cognitive tasks, which were consistent with the results of functional connectivity in resting state in MDD. What is more, a previous study showed that depressed individuals tended to see even positive information as negative because it was associated with personally relevant negative information (Siegle, 1999). It was supported by our results that, in the low gamma band, the global coherence of the depressed group was higher than that of healthy controls in the

positive condition, which was interpreted as this abnormal activation to positive stimulus in depression perhaps helping maintain the depressive state. In addition, we found the Emotion effect in healthy controls but not in the depressed group. The global coherence of negative stimulus was higher than that of positive stimulus in healthy controls in both gamma bands. A more effective interaction was reported to exist between and across cortical regions during negative emotional processing (Sporns and Zwi, 2004). Moreover, this negativity bias in attention allocation based on emotional processing has also been proved by a range of event-related potential (ERP) studies (Smith et al., 2003; Pegna et al., 2008). Our former study also


10


confirmed the existence of the negativity bias in a facial expression-searching task (Ma et al., 2012). 4.3. The regular network characteristics in both groups We found that the values of C/Cs for both groups were significantly greater than 1, and the values of L/Ls for both groups through the whole thresholds were greater than 1. The regular network has a high clustering coefficient and long characteristic path length (Sporns and Zwi, 2004). Our results revealed that both patients with MDD and healthy controls showed the characteristics of regular networks during processing of positive emotion stimuli. Given the studies of recent years on functional networks in depression, most studies focused on brain activities in the resting state or during sleeping in depression. One resting-state functional magnetic resonance imaging (R-fMRI) study investigated the topological organization of intrinsic brain networks in patients with MDD. They found that both patients with MDD and healthy controls showed small-world architectures in brain functional networks, while lower path length and higher global efficiency in patients with MDD implied a shift toward randomization in their brain networks (Zhang et al., 2011). Leistedt et al. (2009) reported the loss of small-world characteristics in the sleep functional brain networks in MDD compared with healthy controls, suggesting a disruption of topological organization caused by this disease. However, the small-world characteristics were not found for either the depressed group or healthy controls in our study. Therefore, we assumed that the brain networks in cognitive processing did not present small-word characteristics. In addition, we found that the clustering coefficient C/Cs of the depressed group was smaller than healthy controls, and the characteristic path length L/Ls was also lower than healthy controls at most of the T value points through two gamma bands. The patients with MDD showed a significantly decreased path length and clustering coefficient in their brain networks compared with healthy controls, which supported the increased global integration and decreased functional segregation. This suggested that functional brain networks in MDD were closer to a randomized configuration. Notably, a previous EEG study reported that patients with depression showed a significantly lower path length in the theta and delta frequency bands compared with healthy controls (Leistedt et al., 2009), providing further support for our findings. This randomization process has been observed in brain functional networks in other neuropsychiatry diseases. In an early EEG study, functional brain networks of patients with schizophrenia were shown to have lower normalized clustering and path length compared with healthy controls (Micheloyannis et al., 2006). An EEG study based on a network analysis of the level of nonlinear coupling also showed more random network topology in patients with schizophrenia (Rubinov et al., 2009). A decrease of the normalized clustering coefficient and path length in AD could be confirmed in a more recent EEG study (Stam et al., 2009). Our findings of a shift toward randomization in their brain networks with MDD further provided evidence that MDD was a disorder with disrupted neuronal network organization and deficient cognitive processing. 4.4. The abnormal network topology of the depressed group between the prefrontal and occipital regions Our exploratory results in network topological structures demonstrated that the functional disorders in prefrontal and occipital regions were the main features in patients with MDD. The frontal cortex is clearly a significant component of the distributed system related to major depression (Matsubara et al., 2014). Research showed that function obstacle of the frontal area was

the main reason for the deficient cognitive performance in patients with depression (Austin et al., 1999; Fossati et al., 1999), and an MRI study showed that the rostral anterior cingulate (rACC) and dorsomedial prefrontal cortex (dmPFC) in MDD as nexus sites in the dysfunctional regulation of cognitive–affective state played a role in vulnerability to recurrence/relapse within this disorder (Nixon et al., 2013). This conclusion was also proved in our study. With a lower value of threshold (T = 0.15 and 0.3), the activated edges of the depressed group was presented in the frontal regions. The frontal cortex is not the only ‘‘location’’ of the complex neuronal processes associated with the depressive state. The top occipital lobe cortex is also thought to be associated with cognitive processes. The parietal region and the right occipitotemporal cortex are the locations of the amygdala and hippocampus. Amygdala is the region of emotion, which is closely related with the terror process. Furthermore, it is related to cognitive processes, such as attention, associative learning, working memory, and so on (Salzman and Fusi, 2010). Hippocampus is also thought to be related with emotion. Research showed that the hippocampal volume of patients with depression was significantly decreased than healthy controls; in addition, the hippocampal neurons of these patients atrophy and drop (Lee et al., 2002). Meanwhile, the right hemisphere has been demonstrated to be able to produce a much broader network of associations than the left hemisphere (Beeman et al., 1994), and the physiological over activation of the right hemisphere reflected the unsuccessful effort to overcome its functional insufficiency in depression (Rotenberg, 2004). In healthy controls, this process did not require any additional activation. Our results showed that the depressed group had additional edges with a lower threshold (T = 0.15) and higher thresholds (T = 0.3 and T = 0.45) in the prefrontal and occipital regions for positive emotional stimulus. These are also in accordance with the previous results. 4.5. Negative emotion bias in depression Negative bias has been demonstrated to be a common phenomenon in emotion processing. Several studies indicated that negative events elicited more rapid and prominent responses than neural or positive events (Mogg et al., 2000; Cacioppo and Gardner, 1999). An ERP study reported that P200, an attention-related component, showed higher amplitudes and shorter latencies in response to negative stimuli than to positive stimuli (Carretié et al., 2001). A study on the brain oscillation also found that the amplitudes of alpha responses upon angry face stimulation were significantly higher than upon presentation of the happy faces at posterior locations (Güntekin and Basar, 2007). Therefore, a kind of interpretation showed that negative entities were more contagious than positive entities (Rozin and Royzman, 2001). Our results of topological structures provided new evidence that healthy controls had negative bias in terms of the functional connectivity in brain networks. On the other hand, negative bias was different in patients with depression compared to normal controls. Gotlib et al. (2004) reported that the attentional bias in patients with depression was specific to the emotion of sadness, but not to the angry or happy faces. Further, Olofsson and colleagues (2008) reported that the negativity bias would reflect rapid activity by amygdala processing of aversive information, so that attentional resources were engaged more readily for unpleasant stimuli relative to neutral or pleasant stimuli (Olofsson et al., 2008). Patients with depression showed greater amygdala activation than controls when processing sad faces (Willner et al., 2013). However, the sad faces were not used as the negative stimuli in our study. This might lead to the reason that negative bias was not found in patients with depression. According to the above results, we supported that the negative bias in participants with depression was directed to



depression-relevant (i.e., sad) faces. In the future, we will focus on studying the distinction of different types of negative stimuli (i.e., sad, fearful, angry, and threatening faces) to functional connectivity in patients with depression and healthy controls. In conclusion, abnormally increased EEG gamma coherence of the depressed group was found in both gamma bands, especially in the high gamma band during emotion processing. The results showed that brain networks appeared regular during processing from both patients with depression and healthy controls based on the GTA. Conversely, compared with normal brain, the networks of patients with depression tend to show a shift towards randomization. Moreover, negative bias was found in healthy controls from the results of coherence and topological structures. However, the negative bias in patients with depression was not detected. Additionally, GTA opened up a new way for obtaining the characteristics of functional network, along with the strength of connectivity or the number of significantly different links. This method might be used to study different cognitive disturbance caused by neuropsychological diseases. As the present results were based on EEG data at the sensor level, our future study will focus on detecting the network features from actual sources. Acknowledgments This study was supported by the National Natural Science Fund of China (61171032 and 61102020), and Natural Science Creation Program of Shanghai Municipal Education Commission (12ZZ099). We would like to thank Manfang Ma for data preprocessing, Xuanhong Zhang for clinical support, and Xiaoli Yang for English correction. We would also like to thank the careful work of anonymous reviewers who gave us helpful suggestions. Conflict of interest statement: None. References Andrew C, Pfurtscheller G. Event-related coherence as a tool for studying dynamic interaction of brain regions. Electroencephalogr Clin Neurophysiol 1996;98:144–8. Austin MP, Mitchell P, Wilhelm K, Parker G, Hickie I, Brodaty H, et al. Cognitive function in depression: a distinct pattern of frontal impairment in melancholia? Psychol Med 1999;29:73–85. Basßar E, Basßar-Eroglu C, Karakasß S, Schurmann M. Gamma, alpha, delta, and theta oscillations govern cognitive processes. Int J Psychophysiol 2001;39:241–8. Bassett DS, Bullmore ED. Small-World Brain Networks. Neuroscientist 2006;12:512–23. Baumeister RF, Bratslavsky E, Finkenauer C, Vohs KD. Bad is stronger than good. Rev Gen Psychol 2001;5:323–70. Beeman M, Friedman RB, Grafman J, Perez E, Diamond S, Lindsay MB. Summation Priming and Coarse Semantic Coding in the Right Hemisphere. JCN 1994;6:26–45. Berman MG, Peltier S, Nee DE, Kross E, Deldin PJ, Jonides J. Depression, rumination and the default network. Soc Cogn Affect Neur 2011;6:548–55. Bernhardt BC, Chen Z, He Y, Evans AC, Bernasconi N. Graph-theoretical analysis reveals disrupted small-world organization of cortical thickness correlation networks in temporal lobe epilepsy. Cereb Cortex 2011;21:2147–57. Bourke C, Douglas K, Porter R. Processing of facial emotion expression in major depression: a review. Aust N Z J Psychiatry 2010;44:681–96. Bullmore E, Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009;10:186–98. Cacioppo JT, Gardner WL. Emotion. Annu Rev Psychol 1999;50:191–214. Calhoun VD, Sui J, Kiehl K, Turner J, Allen E, Pearlson G. Exploring the psychosis functional connectome: aberrant intrinsic networks in schizophrenia and bipolar disorder. Front Psychiatry 2011;2:1–13. Carretié L, Mercado F, Tapia M, Hinojosa JA. Emotion, attention, and the ‘negativity bias’, studied through event-related potentials. Int J Psychophysiol 2001;41:75–85. Crone NE, Miglioretti DL, Gordon B, Lesser RP. Functional mapping of human sensorimotor cortex with electrocorticographic spectral analysis. II. Eventrelated synchronization in the gamma band. Brain 1998;121:2301–15. Damasio A. Descartes’ error: emotion, reason, and the human brain. New York, NY: G.P. Putnam’s Sons; 1994. Di X, Rao HY. The review of human brain functional connectivity. Acta Bioch Bioph Sin 2007;5:5–12. Ekman P, Friesen WV. Pictures of Facial Affect. Palo Alto, CA: Consulting Psychologists Press; 1976.

11

Fingelkurts AA, Fingelkurts AA, Kähkönen S. Functional connectivity in the brain—is it an elusive concept? Neurosci Biobehav Rev 2005;28:827–36. Ertl M, Hildebrandt M, Ourina K, Leicht G, Mulert C. Emotion regulation by cognitive reappraisal—the role of frontal theta oscillations. Neuroimage 2013;81:412–21. Fossati P, Amar G, Raoux N, Ergis AM, Allilaire JF. Executive functioning and verbal memory in young patients with unipolar depression and schizophrenia. Psychiat Res 1999;89:171–87. Friston KJ. Modalities, modes, and models in functional neuroimaging. Science 2009;326:399–403. Galderisi S, Mucci A. Psychophysiology in psychiatry: new perspectives in the study of mental disorders. World Psychiatry 2002;1:166–8. Gotlib IH, Krasnoperova E, Yue DN, Joormann J. Attentional biases for negative interpersonal stimuli in clinical depression. J Abnorm Psychol 2004;113: 127. Greicius MD, Flores BH, Menon V, Glover GH, Solvason HB, Kenna H. Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus. Biol Psychiatry 2007;62:429–37. Grimm S, Boesiger P, Beck J, Schuepbach D, Bermpohl F, Walter M, et al. Altered negative BOLD responses in the default-mode network during emotion processing in depressed subjects. Neuropsychopharmacol 2009;34:932–943. Güntekin B, Basar E. Emotional face expressions are differentiated with brain oscillations. Int J Psychophysiol 2007;64:91–100. Güntekin B, Basßar E. A review of brain oscillations in perception of faces and emotional pictures. Neuropsychologia 2014;58:33–51. Jalili M, Knyazeva MG. EEG-based functional networks in schizophrenia. Comput Biol Med 2011;41:1178–86. Jung JL, Bayle D, Jerbi K, Vidal JR, Henaff MA, Ossandon T, et al. Intracerebral gamma modulations reveal interaction between emotional processing and action outcome evaluation in the human orbitofrontal cortex. Int J Psychophysiol 2011;79:64–72. Kay LM. Two species of gamma oscillations in the olfactory bulb: dependences on behavioral state and synaptic interactions. J Integr Neurosci 2003;2:31–44. Keil A, Müller MM, Gruber T, Wienbruch C, Stolarova M, Elbert T. Effects of emotional arousal in the cerebral hemispheres: a study of oscillatory brain activity and event-related potentials. Clin Neurophysiol 2001;112:2057–68. Lagopouplos J, Wang L, Hermens DF, Hickie IB. A systematic review of resting-state functional-MRI studies in major depression. J Affect Disord 2012;142:6–12. LeDoux J. The emotional brain. New York: Simon & Schuster; 1996. Lee AL, Ogle WO, Sapolsky RM. Stress and depression: possible links to neuron death in the hippocampus. Bipolar Disord 2002;4:117–28. Leistedt SJ, Coumans N, Dumont M, Lanquart JP, Stam CJ, Linkowski P. Altered sleep brain functional connectivity in acutely depressed patients. Hum Brain Mapp 2009;30:2207–19. Leuchter AF, Cook IA, Hunter AM, Cai C, Horvath S. Resting-state quantitative electroencephalography reveals increased neurophysiologic connectivity in depression. PLoS One 2012;7:e32508. Llinás RR, Ribary U. Rostrocaudal scan in human brain: a global characteristic of the 40-Hz response during sensory input. In: Basar E, Bullock TH, editors. Induced Rhythms in the Brain. Boston: Birkhäuser; 1992. p. 147–54. Ma MF, Li YJ, Xu ZZ, Tang YY, Wang JJ. Small-world network organization of functional connectivity of EEG gamma oscillation during emotion-related processing. In: 5th International Conference on Biomedical Engineering and Informatics (BMEI). China: Chongqing; 2012. p. 597–600. Maslov S, Sneppen K. Specificity and stability in topology of protein networks. Science 2002;296:910–3. Matsubara T, Matsuo K, Nakashima M, Nakano M, Harada K, Watanuki T, et al. Prefrontal activation in response to emotional words in patients with bipolar disorder and major depressive disorder. Neuroimage 2014;85:489–97. Micheloyannis S, Pachou E, Stam CJ, Breakspear M, Bitsios P, Vourkas M, et al. Smallworld networks and disturbed functional connectivity in schizophrenia. Schizophr Res 2006;87:60–6. Miltner WHR, Braun C, Arnold M, Witte H, Taub E. Coherence of gamma-band EEG activity as a basis for associative learning. Nature 1999;397:434–6. Mogg K, McNamara J, Powys M, Rawlinson H, Seiffer A, Bradley BP. Selective attention to threat: a test of two cognitive models of anxiety. Cogn Emot 2000;14:375–99. Nixon NL, Liddle PF, Worwood G, Liotti M, Nixon E. Prefrontal cortex function in remitted major depressive disorder. Psychol Med 2013;43:1219–30. Olofsson JK, Nordin S, Sequeira H, Polich J. Affective picture processing: an integrative review of ERP findings. Biol psychol 2008;77:247–65. Pegna AJ, Landis T, Khateb A. Electrophysiological evidence for early non-conscious processing of fearful facial expressions. Int J Psychophysiol 2008;70:127–36. Pfurtscheller G, Andrew C. Event-Related changes of band power and coherence: methodology and interpretation. J Clin Neurophysiol 1999;16:512–9. Ponten SC, Bartolomei F, Stam CJ. Small-world networks and epilepsy: graph theoretical analysis of intracerebrally recorded mesial temporal lobe seizures. Clin Neurophysiol 2007;118:918–27. Rippon G, Rolls E, Davidson RD. The brain and emotion. J Psychophysiol 2001;15:208–10. Rotenberg VS. The peculiarity of the right-hemisphere function in depression: solving the paradoxes. Prog Neuropsychopharmacol Biol Psychiatry 2004;28:1–13. Rozin P, Royzman EB. Negativity bias, negativity dominance, and contagion. Pers Soc Psychol Rev 2001;5:296–320.


12


Rubinov M, Knock SA, Stam CJ, Micheloyannis S, Harris AW, Williams LM, et al. Small-world properties of nonlinear brain activity in schizophrenia. Hum Brain Mapp 2009;30:403–16. Sakkalis V. Review of advanced techniques for the estimation of brain connectivity measured with EEG/MEG. Comput Biol Med 2011;41:1110–7. Salzman CD, Fusi S. Emotion, cognition, and mental state representation in amygdala and prefrontal cortex. Annu Rev Neurosci 2010;33:173–202. Sheline Y, Price JL, Yan Z, Mintun MA. Resting-state functional MRI in depression unmasks increased connectivity between networks via the dorsal nexus. Proc Natl Acad Sci USA 2010;107:11020–5. Siegle GJ. A neural network model of attention biases in depression. Prog Brain Res 1999;121:407–32. Siegle GJ, Condray R, Thase ME, Keshavan M, Steinhauer SR. Sustained gamma-band EEG following negative words in depression and schizophrenia. Int J Psychophysiol 2010;75:107–18. Smith NK, Cacioppo JT, Larsen JT, Chartrand TL. May I have your attention, please: Electrocortical responses to positive and negative stimuli. Neuropsychologia 2003;41:171–83. Sporns O, Zwi JD. The small world of the cerebral cortex. Neuroinformatics 2004;2:145–62. Stam CJ, de Haan W, Daffertshofer A, Jones BF, Manshanden I, van Cappellen van Walsum AM, et al. Graph theoretical analysis of magnetoencephalographic functional connectivity in Alzheimers disease. Brain 2009;132:213–24. Stam CJ, Jones BF, Nolte G, Breakspear M, Scheltens P. Small-world networks and functional connectivity in Alzheimer’s Disease. Cereb Cortex 2007;17:92–9. Stam CJ, Reijneveld JC. Graph theoretical analysis of complex networks in the brain. Nonlinear Biomed Phys 2007;1:1–19. Strogatz SH. Exploring complex networks. Nature 2001;410:268–76.

Stuhrmann A, Suslow T, Dannlowski U. Facial emotion processing in major depression: a systematic review of neuroimaging findings. Biol Mood Anxiety Disord 2011;1:10–26. Tang Y, Li Y, Wang N, Li H, Li H, Wang J. The altered cortical connectivity during spatial search for facial expressions in major depressive disorder. Prog Neuropsychopharmacol Biol Psychiatry 2011;35:1891–900. Uehara T, Yamasaki TK, Okamoto T, Kan S. Efficiency of a ‘small-world’ brain network depends on consciousness level: a resting-state fMRI study. Cereb Cortex 2013;342:213–24. Van den Heuvel MP, Mandl RC, Stam CJ, Kahn RS, Hulshoff PH. Aberrant frontal and temporal complex network structure in schizophrenia: a graph theoretical analysis. J Neurosci 2010;30:15915–26. Vaish A, Grossmann T, Woodward A. Not all emotions are created equal: the negativity bias in social-emotional development. Psychol Bull 2008;134:383–403. White M. Preattentive analysis of facial expressions of emotion. Cogn Emot 1995;9:439–60. Williams MA, McGlone F, Abbott DF, Mattingley JB. Stimulus-driven and strategic neural responses to fearful and happy facial expressions in humans. Eur J Neurosci 2008;27:3074–82. Willner P, Scheel-Krüger J, Belzung C. The neurobiology of depression and antidepressant action. Neurosci Biobehav Rev 2013;37:2331–71. Zhang JR, Wang JH, Wu QZ, Kuang WH, Huang XQ, He Y, et al. Disrupted brain connectivity networks in drug-naive, first-episode major depressive disorder. Biol Psychiatry 2011;70:334–42. Zhou J, Greicius MD, Gennatas ED, Growdon ME, Jang JY, Rabinovici GD, et al. Divergent network connectivity changes in behavioural variant frontotemporal dementia and Alzheimer’s disease. Brain 2010;133:1352–67.


Functional connectivity of negative emotional processing in adolescent depression.

Asymmetric engagement of amygdala and its gamma connectivity in early emotional face processing.

Reduced beta connectivity during emotional face processing in adolescents with autism.

Abnormal cortico-limbic connectivity during emotional processing correlates with symptom severity in schizophrenia.

Abnormal EEG Complexity and Functional Connectivity of Brain in Patients with Acute Thalamic Ischemic Stroke.

Aberrant Neural Connectivity during Emotional Processing Associated with Posttraumatic Stress.

Abnormal Functional Connectivity of Ventral Anterior Insula in Asthmatic Patients with Depression.

Abnormal neural processing during emotional salience attribution of affective asymmetry in patients with schizophrenia.

Unreliability of putative fMRI biomarkers during emotional face processing.

Abnormal whole-brain functional connectivity in patients with primary insomnia.

Effective connectivity during visual processing is affected by emotional state.

Functional connectivity for face processing in individuals with body dysmorphic disorder and anorexia nervosa.

Patterns of abnormal processing of emotional information in panic disorder and major depression.

Beta-Band Functional Connectivity Influences Audiovisual Integration in Older Age: An EEG Study.

Aberrant network connectivity during error processing in patients with schizophrenia.

High-gamma band fronto-temporal coherence as a measure of functional connectivity in speech motor control.

Resting-state EEG oscillatory dynamics in fragile X syndrome: abnormal functional connectivity and brain network organization.

Classifying different emotional states by means of EEG-based functional connectivity patterns.

Altered insular activation and increased insular functional connectivity during sad and happy face processing in adolescent major depressive disorder.

Emotional face processing in pediatric bipolar disorder: evidence for functional impairments in the fusiform gyrus.

Abnormal functional connectivity density in Parkinson's disease.

Visual scanning behavior during processing of emotional faces in older adults with major depression.

Increased Resting-State Gamma-Band Connectivity in First-Episode Schizophrenia.

Functional connectivity in major depression: increased phase synchronization between frontal cortical EEG-source estimates.