When subjects are asked to lay or sit still in an MRI, PET, EEG, or MEG setup and to let their minds wander, the activity that arises is one that speaks of the fundamental organization—or disorganization—of the brain [e.g., Ref. (26–29)]. Resting states can be categorized into several networks (on the order of 7 ± 1): the sensorimotor network, the primary and extrastriate visual network, the auditory network, lateralized frontoparietal networks, the temporoparietal network, the central executive network (CEN), and the, most extensively studied, default mode network (DMN) (27, 30–35). The DMN englobes primarily the medial prefrontal cortex (mPFC), the posterior cingulate cortex, and the precuneus cortex, as well as the inferior parietal cortex, the lateral temporal cortex, and the subgenual anterior cingulate cortex (ACC) (27, 34). A large amount of evidence shows the DMN to be deactivated when one is engaged in a cognitive or sensorimotor task and active during rest or meditative tasks (7, 36–39). Of particular interest, disruptions in this network have been linked to the occurrence of psychopathological symptoms (e.g., depressive, manic, or psychotic episodes) (27, 34, 40–42).

The literature on neural network connectivity, particularly graph theory, suggests that the purpose of a node is guided by how it is connected to other nodes in a given network and that its function is a consequence of the action of its integral network (43). Hence, when resting-state activity is observed for a few minutes, the spontaneous oscillatory behaviors form consistent and reliable functional networks [e.g., Ref. (44)]. The efficiency of the connections within and between these networks appears to rely on at least two main factors: epigenetics and experience (45). The first factor pertains to the interaction between genes and the growth of brain structures, whereas the second pertains to intrinsic neural activity and activity-dependent changes in synaptic strength (e.g., learning) elicited by a person’s interaction with their environment.

Three types of connectivity are generally examined: anatomical, functional, and effective. First, anatomical connectivity pertains to the physical connection between brain regions. It is typically examined using MRI-based DTI analysis of white matter, axonal, and tracts (46). Second, functional connectivity is used to describe the statistical dependency of time-series activity arising from two brain areas (9, 47, 48). It can be measured using linear and non-linear tools such as correlations, coherence, phase-lag index, and mutual information (49–52). It is important to note that although evidence from both human and animal studies shows a close relationship between structural and functional connectivity (5, 53–55), direct anatomical linkage is not necessary for functional connections to take place (54, 56). Finally, effective connectivity speaks of the direct or indirect influence of one brain system on another based on neuronal coupling (9, 44) and can be measured using metrics such as Granger causality and direct transfer functions (57).

In this review, we focus on functional connectivity abnormalities across MDD and BD.

It has been proposed that local and long-range neural synchrony patterns speak of the inherent organization of the brain (43, 58), and thus, an exploration of oscillatory rhythms could help us understand the fundamental neural functioning of different populations. However, confusion can emerge for investigators who are new to the discussion on neural network connectivity. This confusion is entangled in the lack of consistency in the vocabulary employed to describe the two different processes of local and long-range synchrony. It has been argued that when neural populations synchronize, it is a phenomenon that expands across multiple temporal and spatial scales, from local integration of information within the areas that specialize in the same functions to long-range connections that connect different modalities of an object (10). Generally speaking, local synchrony is what is captured by power estimation from a single brain signal (e.g., data from one channel or cortical source). By contrast, long-distance synchrony is captured by estimating the coupling between data from two brain signals.

Specifically, local power modulations of a neural population reflect the activity of a small spatial area of neurons on the order of 1 cm [based on experiments in visual networks, e.g., Ref. (59)]. The measure of spectral power is taken as a reflection of the amplitude of oscillations at different frequencies (60). Neurophysiological studies have underlined the importance of examining local synchronization to observe the different types of information that are carried by different frequency bands (10, 61). As for long-range connectivity, it reflects the functional coordination and synchronization of time series from two brain regions that may or may not have direct structural linkage (e.g., through myelinated white matter tracts). This type of connectivity bridges brain areas between and within different neural networks (43, 58, 60, 62).

Although both local and long-range synchrony can be measured at sensor and source levels during EEG and MEG studies, caution should be taken when measuring the statistical or coherence differences between two recording (sensor) sites. Indeed, what may first be thought to be coordinated time series reflecting connectivity between two brain areas (63) may in fact be spurious coupling arising from volume conduction or field spread [e.g., Ref. (48, 64)]. Different methods have been proposed to overcome this challenge. A solution that can be applied to reduce the impact of this linear mixing limitation is to use coupling measures that are not overly affected by field spread and perform the connectivity estimations in MEG source space [cf Ref. (48, 52, 65, 66)] [see Contrasting Controls and Patients: Differences in Artifacts and Signal-to-Noise Ratio (SNR)].

Given the DMN’s role in self-referential behaviors (7, 27), this network, particularly the mPFC, is recurrently noted as a region important for the discrimination of depression from normal population (34, 75–77). Specifically, fMRI studies show depressed patients to display increased connectivity between certain nodes of the DMN [for instance, between the subgenual ACC and the posterior cingulate cortex (78)], which could be detrimental to cognitive processes (4, 79). These findings are supported by reviews that also highlight enhanced patterns of connectivity within these areas of the DMN (3, 80). In addition, studies have observed altered connectivity between nodes of the DMN and the nodes of the CEN. For instance, one study found enhanced connectivity between the dorsolateral prefrontal cortex (PFC) and the subgenual ACC (81), whereas a review found decreased connectivity between the inferior/superior parietal DMN and dorsal CEN (3).

Moreover, a number of original studies and reviews have reported atypical functional connectivity within areas key to emotional processing. Indeed, limbic regions [amygdala, insula (82, 83)], parts of the DMN [mPFC (77, 83)], and long-range connections between the DMN and the limbic system [e.g., thalamus and posterior cingulate cortex (79)], as well the CEN and the limbic system [e.g., dorsolateral PFC and amygdala (84–86)], appear reduced in patients compared to controls. However, connectivity between the saliency network (i.e., anterior insula, dorsal ACC) and the anterior DMN (34), as well as between the insula and the amygdala (87) seems enhanced in depressed individuals when compared with controls.

All in all, within and between network connectivity of the DMN and limbic system appear to be altered in MDD patients, particularly with respect to projections involving the subgenual ACC. The atypical resting-state organization of their brain appears to correlate with their cognitive and emotional symptoms (88).

Multiple fMRI studies have attempted to untangle the connectivity anomalies observed in these patients, with somewhat contradicting results. Indeed, depending on the analytical method used to extract connectivity, whether it be independent component analysis (ICA) or seed-based/ROIs, wavering conclusions have been made on BD patients’ corticolimbic connectivity patterns [e.g., Ref. (108)].

Although DMN activity has been closely linked to mind wandering and interoceptive thoughts (7, 40), it has also been shown to be germane to social cognition, which is known to be impaired in psychiatric disorders (109–111). A number of fMRI studies have summarized that most bipolar patients and their unaffected relatives have decreased connectivity within the nodes of the DMN (36, 37, 112), between the mPFC and the insula compared with controls (113). Hyperconnectivity has been noted between the DMN and the CEN (114), the DMN and the temproparietal network (36, 37), and the DMN and the visual and the auditory networks (36, 37, 115) compared to controls. Finally, the predominant result of this research in BD, based on reviews of fMRI and DTI studies, shows altered connectivity between parts of the DMN (e.g., mPFC) and the limbic system [e.g., amygdala (42, 85, 108, 114, 116–118)].

With respect to the ventrolateral PFC, contradicting findings have been reported, with some estimating enhanced connectivity with the amygdala (108, 118), while others diminished (119) compared to controls. In addition, a number of studies have investigated the influence of psychotic symptoms on network connectivity patterns during resting state. For instance, one paper observed hypoconnectivity within the PFC and between the dorsolateral PFC (CEN) and amygdala (120), which was predominantly present in BD patients who presented with a history of psychosis, and not in non-psychotic BD or controls.

All in all, models proposed by many researchers suggest that disruption between these keys areas, particularly connectivity involving the amygdala, underlie the occurrence of manic symptoms and inefficient emotional management (108, 118).

The recent article by Jiang et al. (128) compared the oscillatory activity of MDD patients with those of age- and education-matched control subjects. Depression correlated with power decrease in theta (4–8 Hz) and alpha (8–14 Hz) frequency bands in the frontal and parietal areas, respectively, as well as with enhanced power in beta frequency band (14–30 Hz) oscillations in the DMN. Similar to EEG findings [see EEG Power Modulations (Local Synchronization)], no significant difference was found between the two populations across higher frequency bands [>30 Hz (128)].

Moreover, Li et al. (129) examined MEG signals in treatment-resistant MDD individuals who received 10 daily repetitive transcranial magnetic stimulation (rTMS) in the region of the dorsolateral PFC for two consecutive weeks. The authors normalized the spectral amplitude of five frequency bands (delta, 2–4 Hz; theta, 4–8 Hz; alpha, 8–13 Hz; beta, 13–30 Hz; and gamma, 30–50 Hz) by the mean power across all bands to obtain a relative amplitude index for each oscillatory band. Moreover, they measured frontal alpha asymmetry (FAA) in all their subjects, as FAA had been previously associated with symptom severity in depression (130). This article, however, found no significant difference between patient and control subjects in terms of FAA, similar to the inconclusive results of previous electrophysiological studies on FAA, e.g., Ref. (15, 131).

Table 1 summarizes the details of the four studies on MEG resting-state connectivity in MDD patients. The article by Nugent et al. (132) demonstrated that resting-state networks are altered in MDD patients compared with controls in beta frequency band (14–30 Hz). Specifically, based on temporal ICAs and correlations in source space, they found patients to have altered connectivity between nodes of the DMN and the limbic system. Of note, long-range connectivity between the subgenual ACC and the hippocampus was diminished in patients (132). Moreover, the authors observed MDD patients to show enhanced connectivity between the right insular-temporal region and parts of the limbic system (i.e., amygdala, thalamus), and the left insular-temporal region and the angular gyrus in the parietal lobe and the precentral gyrus, which is part of the posterior region of the frontal lobe (132).

In a follow-up MEG study (133), the same authors sought out to examine the effect of ketamine on long-range synchronizations in MDD patients. The source-space connectivity patterns that were uncovered were similar to the disrupted areas found in their earlier article (132). In the beta frequency band (14–30 Hz), 0.5 mg/kg ketamine restored the abnormal hyperconnection between amygdala and insula-temporal regions to normal levels. Interestingly, the authors noted that ketamine appeared to decrease all connectivity patterns across all the regions of the brain, regardless of the subjects’ baseline activity.

Pathak et al. (134) recently used the magnitude-squared coherence to estimate long-range connectivity in depressed individuals before and after rTMS in the dorsolateral PFC at 10 Hz. Their source-space findings (via minimum-norm estimate) reveal that symptom improvement after 4 weeks of treatment correlated with changes in the connectivity within the DMN. Post-TMS, MDD patients found increased coherence between the dorsolateral PFC and the amygdala and the pregenual cingulate cortex in the delta frequency band, as well as decreased coherence in the gamma band between the dorsolateral PFC and the subgenual ACC before treatment (134). The findings of this article could imply that baseline connectivity patterns in MDD involve diminished coherence between dorsolateral PFC-amygdala and dorsolateral PFC-pregenual cingulate cortex, along with enhanced coherence between dorsolateral PFC-subgenual ACC. Moreover, the outcome of this study underlines the importance of analyzing and reporting the type of treatment received by patients as it directly affects the neural network organization.

Finally, Li and colleagues’ (129) longitudinal study explored connectivity in alpha frequency band oscillations in the PFC (via MEG) and glucose metabolism in the thalamus (via PET), which is typically underactive in MDD patients. For the analysis of PFC and thalamus connectivity, patients were divided into binary categories of responders and non-responders to 2-week treatment of rTMS at 10 Hz. Patients were categorized based on their symptom ratings on the Hamilton Depression Rating Scale in the eighth week of the study. This type of antidepressant treatment was able to rescue the disrupted functional connection in responders 14 weeks after the start of the study, while this did not succeed in non-responders. Thus, according to the authors, their sensor-space finding could be seen as additional evidence that the strength of prefrontothalamic connectivity could be an index of depressive symptoms, as previously observed in fMRI studies (79).

The main finding of these MEG papers speak of altered long-range connectivity between the DMN and the CEN. In particular, the resting-state MEG literature supports previous fMRI studies that have demonstrated the implication of the subgenual cingulate cortex, the dorsolateral PFC, and the thalamus in illness severity and symptomatology in MDD [e.g., Ref. (79)]. Importantly, the subgenual cingulate cortex is typically targeted for DBS and rTMS in treatment-resistant depressed individuals and is thus critical to the understanding of their resting-state neural networks (97, 98, 135, 136).

This section explores task-based MEG studies that corroborate connectivity results from resting-state MEG studies in MDD.

A number of studies have investigated the long-range connectivity patterns that emerge during affective and cognitive tasks in psychiatric patients. Among their findings, the diminished long-range synchronization between the dorsolateral PFC and the amygdala, observed recently in resting-state MEG by Pathak et al. (134), was also observed by Lu et al. (137). Indeed, the authors explored effective connectivity within the prefrontal-limbic system circuit using dynamic causal modeling analysis (137). During the affective task, subjects viewed 3-s clip of faces who were eating, neutral, happy, or sad and then indicated by button-press if the expression was sad or not. Under the most optimal model, patients had decreased connectivity from the dorsolateral PFC to the amygdala compared with controls. The authors hypothesized that this could explain part of the dysfunction observed in MDD patients with respect to the integration of both affective and cognitive information for overt behavior.

Other connectivity alteration in MDD patients have also been noted in task-based MEG studies. Specifically, a measure of wavelet coherence has shown enhanced connectivity between the ACC and the amygdala in the gamma (30–48 Hz) and in the delta (below 4 Hz) frequency bands (138). Moreover, enhanced connectivity between the amygdala and the inferior frontal gyrus, as well as between the amygdala and the ACC, in patients, was found to be highly discriminative features during the aforementioned affective task to differentiate MDD and control subjects (139). These connectivity alterations have yet to be observed in resting-state MEG findings. It may be the case that these differences in long-range synchronizations are due to the nature of the task or, rather, that they are best detected by these emotion-based paradigms.

Finally, Salvadore et al. (140) used the widespread working-memory task, N-back, to investigate how the connectivity patterns of MDD patients change before and after a single ketamine infusion. During this task, subjects were asked to keep in mind a stimulus (number between 1 and 4) and to respond when it was matched to the observed stimulus either right away or one or two trials previously. By using source-coherence analysis, the authors observed that the connectivity strength between the pregenual ACC and the left amygdala correlated negatively with the effect of treatment. Indeed, stronger coherence between these two regions prior to ketamine infusion correlated with improvement in symptoms. Similar findings about altered long-range synchronization between the subgenual ACC and the amygdala have been observed in intracranial EEG/DBS studies (97, 98). Given the reoccurring report of involvement of these brain regions, future resting-state MEG studies could clarify whether this pattern of connectivity alteration pertains to the nature of the task or to the intrinsic neural organization in MDD.

An important limitation that connectivity studies might display is that of being conducted at sensor level rather than source level. Although most of the reported MDD studies were conducted in source space, the article by Li et al. (129) was in sensor space, where only 26 gradiometers (out of a possible 306 channels) from the frontal region were used. However, the article had a major strength of employing a multimodal approach to studying connectivity (MEG, PET, and TMS), which included an anatomical T1 from MRI to obtain anatomical precisions. This allowed access to a richer set of information than what is provided using a single neuroimaging tool.

Next, exploring specific frequency bands can also be a limitation. Indeed, in the article by Nugent et al. (132), only the beta band frequency range was explored, while Li et al. (129) reported only on alpha frequency band oscillatory behavior. A strength of Nugent et al. (132) is that the authors took additional quality control steps (to verify the reliability of their ICA estimates), they tested the reliability of their results by comparing it with a second data set (unmatched groups) and, importantly, they included unmedicated MDD subjects. By doing so, the authors allowed to examine intrinsic connectivity prior to pharmaceutical effects.

Furthermore, the metric of coherence and correlations can raise questions about the spatial accuracy of the long-range synchronizations observed due to potential field spread effect (see Challenges, Pitfalls, and Methodological Recommendations for Future Studies for further details).

Finally, small sample size can be problematic in the interpretation of findings. In their recent study, Nugent et al. (133) explored the effect of ketamine on long-range synchronization and on symptoms scores as assessed by the Mania and Depression Rating Scale in a subset of patients from their previous study (132). However, this was performed in a small number of subjects. Non-parametric statistical tests were applied to compensate for their cohort of patients. Similarly, while Pathak et al.’s (134) study was important to evaluate the longitudinal effect of repetitive TMS on neural network connectivity, the small sample of patients (n = 5) and lack of multiple comparisons put the findings of this paper at risks of type I error, a fact acknowledged by the authors. While it is important to evaluate promising treatments, it would be interesting to evaluate its effect on connectivity patterns in a larger cohort of patients to increase reliability.

Al-Timemy et al. (141) were able to successfully classify BD and control populations using MEG resting-state spectral features within the delta (1.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and gamma (30–40 Hz) frequency bands. Relative power modulations in delta and theta frequency bands in the posterior region of the brain were observed to be significantly different between patients and healthy controls (141). However, the direction in these differences was not specified. The authors of this study also explored median frequency (MF), described as the frequency that divides the area under the curve of the power map (1.4–40 Hz) into two. Their analysis found MF of BD patients to range between 9.92 and 12.54 Hz depending on the examined brain region. Furthermore, unlike healthy control subjects who demonstrated a positive correlation between their MF and age, BD patients had a negative correlation between MF and age (141).

Table 2 summarizes the details of the relevant study that have been published on BD. Chen et al. (142) had an interesting, although small, pool of euthymic (no overt depressive or manic symptoms) BD-I patients that was compared with matched healthy controls. The authors focused on the frontal cortex activity, and thus, oscillatory modulations in only 11 of their 306 MEG channels were reported. Their spectral analysis across pairs of channels was performed using a derivative of the similarity index (SI) framework used by Arnhold and colleagues (143). Differences between patients and controls were noted based on global SI of channel pairs: patients displayed an increase in the synchronization of delta band (2–4 Hz) frequencies and a decrease in beta band (12–24 Hz) frequencies within nodes of the frontal cortex (142). While there is a number of MEG studies that have examined alterations of spectral power in BD during tasks [e.g., Ref. (144, 145)], as it stands, and to our knowledge, the article by Chen et al. (142) is so far the only resting-state MEG study that has evaluated the functional connectivity in BD population.

To the best of our knowledge, no MEG task-based study has explored long-range synchronizations in BD. However, EEG resting-state studies, such as one by Kim et al. (124), have observed disrupted connections within the PFC of BD patients compared with controls. Future MEG studies, with and without tasks, could help elucidate more specific neural network patterns that are either specific to the neural organization of BD patients, or, alternatively, to altered patterns of information processing.

Although of important value, the resting-state MEG study by Chen et al. (142) exploring long-range synchronization had a number of limitations. For instance, although significant information can be gathered by exploring euthymic patients (i.e., with no overt depressive or manic symptoms), the pool of subjects was relatively small (n = 10). Moreover, in addition to being conducted in sensor space, the authors focused on the frontal cortex activity, with oscillatory modulations of only 11 of their 306 MEG channels being reported. The investigation of the neural network connectivity pattern of BD using MEG is clearly still in its early days, and more studies are needed to elucidate the key connectivity patterns that define this illness.