Altered functional activity in bipolar disorder: A comprehensive review from a large‐scale network perspective

Abstract Background Growing literature continues to identify brain regions that are functionally altered in bipolar disorder. However, precise functional network correlates of bipolar disorder have yet to be determined due to inconsistent results. The overview of neurological alterations from a large‐scale network perspective may provide more comprehensive results and elucidate the neuropathology of bipolar disorder. Here, we critically review recent neuroimaging research on bipolar disorder using a network‐based approach. Methods A systematic search was conducted on studies published from 2009 through 2019 in PubMed and Google Scholar. Articles that utilized functional magnetic resonance imaging technique to examine altered functional activity of major regions belonging to a large‐scale brain network in bipolar disorder were selected. Results A total of 49 studies were reviewed. Within‐network hypoconnectivity was reported in bipolar disorder at rest among the default mode, salience, and central executive networks. In contrast, when performing a cognitive task, hyperconnectivity among the central executive network was found. Internetwork functional connectivity in the brain of bipolar disorder was greater between the salience and default mode networks, while reduced between the salience and central executive networks at rest, compared to control. Conclusion This systematic review suggests disruption in the functional activity of large‐scale brain networks at rest as well as during a task stimuli in bipolar disorder. Disrupted intra‐ and internetwork functional connectivity that are also associated with clinical symptoms suggest altered functional connectivity of and between large‐scale networks plays an important role in the pathophysiology of bipolar disorder.


| INTRODUC TI ON
Bipolar disorder is a neuropsychiatric disorder that is characterized by severe disturbance and fluctuation in mood (Rowland & Marwaha, 2018). According to the latest edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) (American Psychiatric Association, 2013), bipolar and related disorders entail affective episodes of mania and depression as primary clinical symptoms (Severus & Bauer, 2013) along with symptoms of cognitive deficit (Douglas et al., 2018), together which may lead to an overall reduced quality of life (Saarni et al., 2010). Considering the high lifetime prevalence of bipolar spectrum disorder of 2.4% across the globe (Mathers et al., 2003;Merikangas et al., 2011), elucidating the pathophysiology of bipolar disorder is an urgent matter among other psychiatric illnesses (Ferrari et al., 2016).
Furthermore, recent literature has also taken functional magnetic resonance imaging (fMRI) data to identify specific brain regions that undergo significant alteration in functional activation patterns in bipolar disorder (Chen et al., 2011;Marchand & Yurgelun-Todd, 2010). Collective findings from these studies may suggest the potential roles of certain cortical and subcortical regions of the brain as neural correlates of bipolar disorder. However, the associations between the multiple brain regions identified to be significantly altered as well as the clinical symptoms of bipolar disorder remain rather inconclusive. A recent review on the classification and prediction of brain disorders using functional connectivity discusses the advantages to the investigation of brain connectivity according to regions that are functionally co-activated . This is in alignment with other previous literature which notes the importance of identifying the functional connectivity in a multitude of brain regions that are known to have functional connections with a key anatomical brain region involved in the disorder (Pannekoek et al., 2014). Moreover, considering that a single brain region can belong to a number of overlapping brain networks (Xu et al., 2013) and its functional activation at a regional level may not fully explain for the functional alterations at a wholebrain level, investigating the functional connectivity between multiple, anatomically remote yet functionally relevant brain regions may reveal more widespread information on the neural underpinnings of the disorder. In the case of bipolar disorder, a study also reported the significance of a particular brain region and its related functional connectivities as a significant marker that distinguishes bipolar disorder from healthy controls, particularly the connectivity between the basal ganglia to regions that make up the default mode network and the central executive network (Teng et al., 2013). Therefore, further investigation using a broader perspective, such as the observation of functional alterations from a network-based approach, may expand the scope of our understanding on the neurological changes in bipolar disorder.
Recently, growing evidence suggest the significant roles of functional brain alterations in bipolar disorder, such as the functional connectivity within and between brain regions that are responsible for cognition and emotion (Cerullo et al., 2012;Han et al., 2019;Whalley et al., 2012). For instance, studies demonstrated that emotional instability, a primary symptom in bipolar disorder, is significantly associated with the functional connectivity among brain networks involving the limbic regions (Blond et al., 2012;Townsend and Altshuler, 2012). Moreover, a recent study suggests that the pattern of functional activation among particular brain regions may not only act as neural correlates of a key symptom in bipolar disorder, but also a potential biomarker that classifies bipolar disorder from other illnesses with similar clinical symptomatology such as major depressive disorder (Manelis et al., 2020) and schizophrenia spectrum disorders (Palaniyappan et al., 2019). This highlights the potential significance of exploring the functional connectivity of a particular network of brain regions as a reliable approach to further understanding bipolar disorder. In addition, considering that bipolar disorder includes a broad range of symptoms even within the domains of emotion and cognition such as affective instability, impulsivity, executive function, and memory (Phillips and Vieta, 2007)-all of which involve the activation of a number of brain regions-investigating brain regions associated with the symptoms of bipolar disorder from a network-based perspective may provide a more comprehensive way to understanding the pathology of the disorder.
In recent years, large-scale functional brain networks have become a growing interest in neuroimaging studies, as typically measured using signaling obtained through functional magnetic resonance imaging (fMRI) (Riedl et al., 2016). One of the key implications of recent network-based studies is the identification of specific brain regions that are different in functional activity in bipolar disorder compared healthy controls. In addition to this, understanding the functional alterations of the brain at a network level that are characteristic of bipolar disorder may provide insight with regards to the major challenges in the current clinical field, such as difficulty in accurate disease diagnosis, delayed diagnosis due to inadequate characterization of symptoms, and the frequent comorbidity found in bipolar disorder (Angst & Cassano, 2005). Considering that distinct functional connectivity patterns at a network level have been identified in a depression model (Kaiser et al., 2015;Mulders et al., 2015), a similar approach may be taken to characterize bipolar disorder from a neurological perspective.
Among the many ways to categorize brain regions according to large-scale networks, recent literature implicates a growing interest in the functional connectivity in intrinsic resting-state networks (RSNs). Currently, functional connectivity of RSNs have been mostly explored to differentiate between bipolar disorder and other affective disorders such as unipolar depression  or psychotic disorders like schizophrenia (Whalley et al., 2012). Also, since the recent review of the functional connectivity among intrinsic RSNs in bipolar disorder (Vargas et al., 2013), a growing number of original researches have been conducted, which may warrant for an updated review on this perspective. Findings on the relationship between intrinsic RSNs may improve the current limitations to bipolar disorder research, as it may help interpret the multiple brain regions that were previously suggested as individual neural correlates of bipolar disorder in a more comprehensive manner. The characterization of bipolar disorder acquired at a network level may then allow a detailed prognosis of the disorder, leading to an early onset of preventive measures, as well as the promotion of targeted treatment strategies to further explore treatment response.
Furthermore, overseeing the multi-regional alterations in functional connectivity in bipolar disorder may provide additional information regarding the neurological pathways that underlie bipolar disorder, which may then impart insight toward understanding the pathology of the disorder in a broader perspective.
The aim of this study was to review major findings from recent studies on bipolar disorder that investigated the brain functional networks of bipolar disorder. The review of a network-based perspective may elucidate distinct intra-and internetwork connectivity alterations in bipolar disorder that can explain inconsistencies often found using more targeted approaches such as voxel-based studies (Bora et al., 2010). For this review, we focused on original research articles that have been conducted within the past 10 years in a systematical assessment method. The study will summarize current findings on the network-based alterations of bipolar disorder, as well as assess how these findings relate to recent findings on region-based functional connectivity in bipolar disorder.

| Literature search strategy
A literature search was conducted on two electronic databases: PubMed and Google Scholar. The search was performed under a filter where studies published from January of 2009 through August of 2019 were included only. The following keywords were used as search terms: bipolar disorder, affective disorder, mood disorder, brain imaging, magnetic resonance imaging, network, and functional network. Specifically, an advanced search strategy was conducted with the above terms used either alone or in combination in the search.
Following the literature search, the records identified were reviewed where duplicates were removed, and the titles and abstracts of the remaining records were assessed for eligibility as according to the following eligibility criteria.

| Eligibility criteria
All studies included in the current review were eligible original research that satisfied the following criteria: (a) in English language, (b) primary target population of individuals with bipolar disorder, (c) brain functional magnetic resonance imaging data as one of the outcome variables, and (d) evaluates the functional connectivity or functional network of the brain. Any review papers, conference abstract, short communications were excluded.

| Methodology of analysis for brain functional connectivity
For the current review, functional connectivity was defined as follows: (a) spatial or temporal alterations in blood oxygen leveldependent (BOLD) signals (Lee et al., 2013), (b) measures of connectivity between two cortical or subcortical regions using path length, participation index, or clustering coefficient (Roberts et al., 2017), or (c) the amplitude of regional neuronal activity as measured by amplitude of low-frequency fluctuations (ALFF) as previously described (Zou et al., 2008). Studies with designs that fell within these definitions were included as part of the literature review, including those that measured the functional connectivity using resting-state BOLD signals, as well as those that have investigated task-based BOLD signals. For analytical approaches to measuring functional connectivity, the current review includes some of the primary methodologies that are most common in recent literature. In particular, hypothesisdriven approaches that are reviewed are seed-based or region-ofinterest (ROI) analysis and ALFF analysis. ROI analysis calculates the temporal correlation of a priori selected seeds with other regions of the brain, while ALFF measures the ratio of low-frequency fluctuations to that of the whole frequency within a specific a priori brain region in order to estimate the regional neuronal activity level (Zou et al., 2008). The review also includes studies using the data-driven approach of independent component analysis (ICA), which divides the whole brain into a number of independent components (ICs), resulting in the extraction of functional maps across time (Beckmann & Smith, 2004).

| Ethical statement
We confirm that all authors have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

| Results from literature search
A total of 536 articles were identified after the literature search.
First, duplicates were removed, followed by the screening of titles and abstracts of the studies. The screening process excluded 306 studies as according to the eligibility criteria mentioned above. A full-text review was then conducted, where a total of 61 articles were excluded after careful consideration. Specifically, 39 articles were removed because the scope of the studies did not include a network-based approach, 8 articles were excluded due to having a sample population of either a pediatric bipolar disorder patients or healthy relatives of bipolar disorder patients, and 14 articles were removed as they only evaluated the structural networks of the brain.
A total of 49 studies were identified and included in this review, including 4 studies that were obtained through cross-reference. Of these, 30 were ROI-based studies that investigated the functional connectivity among two or more regions of the brain, 16 were ICAbased studies that investigated the alteration in functional connectivity within and between functional brain networks, and 3 were ALFF-based studies which observed the activity in slow fluctuations between brain regions that are representative of specific RSNs. A detailed summary of the literature search process in identifying original articles eligible for the current review is shown in Figure 1.  Table S1.

| Core functional networks altered in bipolar disorder
Of the 49 studies reviewed, 40 studies acquired resting-state fMRI data, and therefore primarily explored the functional activity of brain regions that make up the major resting-state networks. Brain regions that make up the default mode network (DMN), central executive network (CEN), and salience network (SAN) were most frequently used as a priori seeds in ROI-based studies, and the same networks were also shown to undergo significant alteration in the majority of the ICA studies. Others that underwent significant alteration in functional connectivity included brain structures that make up the limbic network (LIM) and the cerebellar network (CBL).
The remaining 9 studies explored brain functional connectivity in bipolar disorder through task-based fMRI. Research on task-based fMRI focused on exploring the affective and cognitive symptoms of F I G U R E 1 Flowchart of literature search and identification of eligible articles. The initial database search resulted in 113 records from PubMed and 423 records from Google Scholar, together which resulted in 412 records after the exclusion of duplicates. The first screening of titles and abstracts excluded 306 records for the following reasons: 133 records were not an original research paper, 157 records did not target bipolar disorder, 9 records did not include MRI data, and 7 records were not in English. A second screening of the 106 full-text articles led to the exclusion of 61 records as a result of the following: 39 records did not use a network-based approach, 8 records were based on pediatric or relatives of bipolar disorder, and 14 records evaluated structural brain networks. An additional four records were also included from cross-referencing, resulting in a total of 49 records for this review. MRI, magnetic resonance imaging. Records excluded (n = 306) -Not an original research paper (133) -Target group is not bipolar disorder (157) -Not an MRI-based study (9) -Not in English language (7) Full-text articles assessed for eligibility (n = 106) bipolar disorder through the selection of ROIs of the frontal cortex, specifically regions that make up the CEN. The nature of the tasks in these studies primarily included cognitive tasks such as those targeting working memory or affective tasks of negative and positive emotional stimuli.
Among the literature reviewed, studies that report altered brain functional connectivity in bipolar disorder using detailed anatomical brain regions as according to the Montreal Neurological Institute (MNI) template and cluster size are summarized in Table   S2.

| Network alteration according to ROIbased studies
For studies that used the hypothesis-driven approach of ROI analyses, functional alteration of large-scale brain networks were assessed based on the networks to which the a priori ROIs belong, as well as the brain regions that demonstrated significant temporal correlations with the selected seeds as assessed through fMRI data.
A total of 30 studies underwent seed-based ROI analyses as to in- (also coined with the cingulo-opercular network due to consisting of similar seeds). A detailed summary of the ROI-based studies reviewed is provided in Table 1. large-scale RSNs at rest for individuals with bipolar disorder, compared to the respective healthy control group, is shown in Figure 2.

| Intranetwork functional connectivity
Among the large-scale networks, the functional alteration of regions that make up the DMN were the most prevalent across studies.

Specifically, some of the key regions that demonstrated significantly
altered functional connectivity were the medial prefrontal cortex For the CEN, an overall decreased functional connectivity between regions of the CEN (also described as the frontoparietal network in some studies) was reported (Baker et al., 2014;Mamah et al., 2013). Specific regions that demonstrated decreased functional connectivity were the connection between the dorsolateral prefrontal cortex (DLPFC) and the intraparietal sulcus (Mamah et al., 2013) as well as the lateral parietal cortex (Baker et al., 2014).
Similar to the DMN, when performing a task, the functional con- Specifically, functional connectivity was decreased between the anterior insula and ventrolateral prefrontal cortex (VLPFC) (Ellard et al., 2018;Pang et al., 2018), between the subgenual ACC and the inferior temporal gyrus (ITG) (Gong et al., 2019;Magioncalda et al., 2015) as well as between the perigenual ACC to the VLPFC (Magioncalda et al., 2015). Functional connectivity among the subregions of the ACC was reduced also, where reduced functional connectivity between the perigenual ACC to the supragenual ACC was reported (Magioncalda et al., 2015). In addition, reduced Note: The altered functional connectivity of large-scale functional brain networks in bipolar disorder as reported in the ROI-based studies reviewed. The name of the large-scale functional brain network was determined either by the authors themselves within the study and/or based on the major brain regions that were used as a priori seeds. The direction of alteration for each study is in reference to the respective healthy control group. Functional connectivity between brain regions were reported under the term "decrease" for both positive and negative connectivities, including cases where the polarity of the functional connectivity was reversed. Network-based statistics studies that compared the between-group differences in functional connectivity strength of an a priori region with other regional connections and networks.

TA B L E 1 (Continued)
intranetwork functional connectivity of the cingulo-opercular network (CON), which consists of brain regions that also make up key regions of the SAN, was reported in one study (Mamah et al., 2013). Furthermore, reduced global and local efficiency of the CON was found in one study, which was also significantly associated with cognitive ability (Sheffield et al., 2017). Among these, three reported that the anticorrelated relationship of CEN-DMN weakened in the case of individuals with bipolar disorder as compared to healthy control (Chai et al., 2011;Favre et al., 2014;Roberts et al., 2017), while two studies reported findings that were quite the opposite (Marchand et al., 2014;Townsend et al., 2013).
Second, hyperconnectivity of the SAN-DMN was reported.
Specifically, increased functional connectivity was found between the insula of the SAN and mPFC of the DMN (Chai et al., 2011), and between the anterior cingulate cortex of the SAN and superior frontal gyrus of the DMN (Oertel-Knöchel et al., 2015). Furthermore, the functional connectivity from the anterior insula cortex toward the middle frontal gyrus was demonstrated to be increased at rest (Li, Tang, et al., 2018), as well as between the ventral anterior insula and precuneus (Chai et al., 2011), and between the subgenual ACC and PCC (Rey et al., 2016). Similar to the internetwork functional relationship between the CEN-DMN, the direction of alteration in functional connectivity of the SAN-DMN reversed when performing a task, where functional connectivity among the ACC, PCC, medial frontal gyrus, and VLPFC (as well as the DLPFC of the CEN) decreased during emotion down-regulation (Townsend et al., 2013).
Third, significantly altered functional connectivity of the SAN-CEN was found (Pompei et al., 2011), where the polarity of the functional connectivity between the insula of the SAN and inferior parietal lobule (IPL) of the CEN was reversed in an anticorrelated manner (Ellard et al., 2018). In addition, this reduction was significantly associated with greater impairment in perceived emotion control, as well as impaired inhibition (Ellard et al., 2018), which are characteristics of mania (Christodoulou et al., 2006). In contrast, contradictory findings of increased functional connectivity between the dorsal anterior insula of the SAN and inferior IPL of the CEN was also reported at rest by one study (Pang et al., 2018). However, one study reported opposite findings where the functional connectivity between the SAN-LIM was decreased in bipolar disorder compared to control (Anand et al., 2009). In addition, significant alteration in functional connectivity between the mPFC of the DMN and amygdalae was reported in two studies, but the direction of alteration were inconsistent (Anticevic et al., 2013;Favre et al., 2014). The functional relationship between the task-negative network, DMN, and the CBL, which is associated with cognition, was also explored in many studies. An overall reduced functional connectivity between the DMN-CBL was reported in individuals with bipolar disorder compared to the respective healthy control group Luo et al., 2018;Chen et al., 2019). In particular, reduced functional connectivity was reported between the mPFC and various cerebellar regions Luo et al., 2018). Decreased connectivity of the CEN-CBL was also reported by one study (Luo et al., 2018).  Table 2.

| Intranetwork functional connectivity
Five studies reported altered functional connectivity within the DMN. (2012)

F I G U R E 3
Internetwork alteration in functional connectivity at rest based on current articles reviewed. Internetwork alteration in functional connectivity in bipolar disorder based on the triple network model. Red represents major regions of the salience network, green represents major regions of the central executive network, and blue represents major regions of the default mode network. The size of the nodes represents hyperconnectivity, hypoconnectivity, or statistically similar (neurotypical) functional connectivity between the respective region and its nearby subregional structures, such as the functional connectivity between the dorsal and ventral part of the region, as compared to the healthy control group. In summary, the results from the 49 studies reviewed indicates a significant increased connectivity between the salience network and default mode network. Decreased functional connectivity was reported between the salience network and central executive network, as well as between the central executive network and the default mode network. It is noteworthy that increased functional connectivity was also found between the right dorsolateral prefrontal cortex functional connectivity was also reported in Das and colleagues (2014) between the precuneus and the rest of the DMN in bipolar disorder, which was associated with emotional awareness. In contrast, Meda and colleagues (2014) used a para-ICA approach targeting the DMN to report reduced global functional connectivity in regions of the DMN including the mPFC, PCC, and precuneus as consistent with the ROIbased study findings, and further suggest selective nodes within the DMN that may be heritable through genetic associations. The reduced functional connectivity of the DMN at rest was also supported in studies (Öngür et al., 2010;. (2014)   . Findings also showed that weaker intranetwork functional connectivity was prevalent for all three of the triple network model (Wang et al., 2019a), where more time was spent in the phase where all 3 networks have sparse functional connections as compared to the healthy control group.

Brandt and colleagues
Other large-scale functional network that underwent intranetwork functional connectivity alterations in bipolar disorder included the increased functional connectivity of the sensorimotor network (Skåtun et al., 2018), which was associated with symptoms of perceptual and cognitive distortion in the case of individuals with bipolar disorder.

| Internetwork functional connectivity
Significant altered functional connectivity between the three large-  (Wang et al., 2019b). However, another study reported stronger functional connectivity among the CEN-DMN, specifically between the DLPFC of the CEN and angular gyrus of the DMN .
In addition, contrary to the ROI-based findings, decreased functional connectivity of the SAN-DMN at rest was reported also . Here, it is noteworthy that the SAN and the meso/paralimbic network were identified as a single component, and the CEN was referred to as the frontoparietal network.
For the internetwork connectivity between networks of the triple network model and other large-scale networks, He et al. (2018) showed decreased functional connectivity between independent components that make up the DMN (precuneus), CEN (DLPFC), and SAN (subgenual ACC), respectively, and the CBL, a finding which was performed using ROI-based analysis then validated through an ICA approach. However, in contrast, Wang et al. (2019b) reported increased functional activity between the SAN-CBL. In addition, the CEN functional activity was reported as more strongly connected with the caudate in individuals with bipolar disorder type 1, while those with bipolar disorder (type unspecified) showed greater functional activity of the sensorimotor network with the precentral gyrus of the CEN (Thomas et al., 2019). Furthermore, Meda et al. (2012) identified decreased functional connectivity between the anterior DMN and fronto-occipital network as well as increased functional connectivity between the LIM and frontotemporal network, which in turn correlated with negative clinical symptoms as according to the Positive and Negative Symptom Scale (PANSS) (Meda et al., 2012). Yip et al. (2014) also reported significant functional connectivity alteration in bipolar disorder type 2 participants, where increased coherence across the insula of SAN and regions of the LIM including the putamen was reported at rest. Other networks with significant alteration included the significant alteration between thalamic regions, which make up the LIM and the somatomotor network (SMN) (Skåtun et al., 2018).

| Network alteration according to ALFF-based studies and others
Three studies using ALFF as outcome measures to investigate functional brain network alteration were selected for review. Of these, Li, Liu, et al. (2018) reported decreased functional connectivity within the LIM, specifically through higher ALFF levels in the amygdala and supplementary motor area in individuals with bipolar disorder compared to the healthy controls. Liu and colleagues (2012) reported that those with depressive bipolar disorder had significantly higher ALFF in the left insula of the SAN along with other temporal regions, as well as the CBL. In contrast, Liu et al. (2012) demonstrated lower ALFF in the parahippocampal regions in bipolar disorder. Lastly, Wang X and colleagues (2019) showed significantly greater fractional ALFF in the superior frontal gyrus of the DMN, a finding that also correlated with decreased gray matter volumes in the respective regions.
Apart from the triple network model, Jimenez et al. (2019) used nine key RSNs using the Smith atlas to generate subject-specific maps and showed significant reduced functional connectivity in the medial visual network of bipolar disorder compared to control which also correlated with emotion management. TA B L E 2 Summary of the large-scale functional brain networks altered in bipolar disorder according to ICA and ALFF-based studies Note: The altered functional connectivity of large-scale functional brain networks in bipolar disorder as reported in ICA-based studies, as well as ALFF-based studies. For ICA-based studies, the altered direction refers to the alteration in overall functional connectivity within a large-scale functional network or between multiple large-scale functional networks. For ALFF-based studies, the direction of alteration refers to the increase or decrease of frequency fluctuation, and does not necessarily indicate the strength of the functional connectivity. Unspecified altered direction indicates an alteration in functional connectivity that is statistically significant as compared to the respective healthy control group, without the provision of whether the alteration is increased or decreased.

| Summary of main findings
The aim of the current systematic review was to examine the current trends in existing literature on network-based study findings in bipolar disorder. Through a systematic search of literature, a total of 49 studies were reviewed, of which the methodologies and key findings were explored. A majority of the studies reviewed demonstrated altered functional connectivity of functional brain networks through the hypothesis-driven approach of a priori ROI analyses, where brain regions that are most representative of a specific large-scale network were selected as seeds. Other methodologies exploring the functional connectivity of the brain in bipolar disorder included the ICA, ALFF, and network-based statistics.
In general, in terms of intranetwork functional activity, patients with bipolar disorder generally show reduced functional connectivity within each of the triple network model which are major RSNs as according to previous literature (Menon, 2011). Reduced functional connectivity was reported between regions that make up the DMN (Öngür et al., 2010;Anticevic et al., 2013;Meda et al., 2014;Magioncalda et al., 2015;Gong et al., 2019;Liu et al., 2019), CEN (Baker et al., 2014;, and SAN (Ellard et al., 2018;Gong et al., 2019;Karcher et al., 2019;Magioncalda et al., 2015;Mamah et al., 2013;Pang et al., 2018;Sheffield et al., 2017) at rest. Specific regions showed increased functional connectivity at a subregional level also, such as between the medial and dorsolateral parts of the SFG of the DMN (Marchand et al., 2014) and the dorsal and ventral regions of the anterior insular cortex of the SAN (Li, Tang, et al., 2018;. In particular, the altered within-network functional connectivity of the intrinsic brain networks such as the DMN in bipolar disorder was associated with the symptoms of depression (MADRS) (He et al., 2016) and lifetime psychosis history (Anticevic et al., 2013). These findings are in alignment with previous literature, which emphasized the clinical significance of disrupted DMN in mood disorders (Mohan et al., 2016;Zhong et al., 2016), as well as in anxiety disorders (Coutinho et al., 2016;Lanius et al., 2010). Specifically, these studies noted that disruption in the DMN functional activity was associated with symptoms of depression, anxiety, or suggested as reliable predictors of related clinical symptom severity. In addition, the significant relationship between the functional activity of the DMN and clinical symptoms in mood disorders was also validated through the significant normalization of DMN function in depressed individuals after antidepressant treatment (Qin et al., 2015). Studies reviewed here support previous findings of disruption in particular large-scale networks of the brain in bipolar disorder, and in particular, networks that make up the triple network model.
The studies reviewed in the current article also provide supportive evidence for previous voxel-based neuroimaging studies that demonstrated significant structural and functional alteration in bipolar disorder. In particular, brain regions that indicated significant altered functional connectivity between its subregional structures were similar to regions that underwent significant gray matter alteration in previous literature. To name a few, significant gray matter volume reduction was reported in the mPFC (Liu et al., 2012), bilateral cingulate cortex (Lochhead et al., 2004), posterior parietal cortex and parietal lobule (Adler et al., 2005), and dorsolateral prefrontal cortex (Tost et al., 2010), as well as increased volume in the insula (Lochhead et al., 2004).
Among these, the functional connectivity of the mPFC was particularly of interest across the studies reviewed. The mPFC has been described to play a crucial role in self-reflection (Passingham et al., 2010), and is associated with important cognitive and affective functions such as emotional facial recognition (Keener et al., 2012).
The collective findings from the studies reviewed demonstrated reduced functional connectivity within the mPFC and increased functional connectivity between the mPFC and insula in individuals with bipolar disorder as compared to healthy controls. One of the ways to interpret this alteration may be based on a previous study, which describes increased functional connectivity as a functional compensatory response in bipolar disorder . The hyperconnectivity between the mPFC of the DMN and insula of the SAN as reported in the current studies reviewed may indicate a compensation for the lack of mPFC connectivity with other regions in the case of bipolar disorder, and may enable adequate functioning of the brain at rest as well as salience detection.
Another interpretation to the reduced connectivity within the mPFC and hyperconnectivity of the SAN-DMN may follow the paradigm of attentional shift as described in the triple network model (Menon, 2011). According to the triple network model, dynamic interactions are present between networks in order to regulate the shift between attention and access to cognitive resources, and this shift is supported by greater activity in the frontal region (Menon, 2011). Considering that reduced functional connectivity within the mPFC have been suggested to represent disruption in emotional processing (Etkin & Wager, 2007), the reduction in functional connectivity of the mPFC may be a neural characteristic of affective disorders including bipolar disorder. Furthermore, studies have also noted that increased activation of the insula may be a misattribution of emotional salience and lead to increased symptoms of anxiety, neuroticism, and in schizophrenia, psychosis (Menon & Uddin, 2010). Dysfunctional saliency processing has also been noted to lead to diminished cognitive capabilities (Menon, 2011).
Therefore, the current findings may suggest reduced functional connectivity within the mPFC of the DMN and the increased activation of the insula as characteristics of disrupted emotional processing and severe mood in bipolar disorder.
From an internetwork perspective, the articles reviewed indicate a significant disruption across functional networks in bipolar disorder. Specifically, many of the studies demonstrated hyperconnectivity between the DMN and SAN at rest, while the relationship between the CEN-SAN and CEN-DMN indicated hypoconnectivity.
Considering the anticorrelation nature of the CEN and DMN in normal healthy conditions (Menon, 2011), the weakened or reversed functional connectivity between the two networks as reported in the studies currently reviewed may indicate an imbalance in the brain's ability to segregate between the state of rest and a task-positive state in the case of bipolar disorder. The increased functional activity involving the CEN during the performance of a cognitive or emotion-related task may also indicate the increased need to compensate for the decreased ability to enable emotional and/or cognitive processing at rest in bipolar disorder (Rey et al., 2016) due to the ongoing disruption of the brain involving affective disorientation. Das et al. (2014) has also noted the increased functional coupling between the DMN-SAN as a compromise of the weakened social salience detection and self-referential processing in individuals with affective problems such as in bipolar disorder. Second, it is noteworthy that the current search strategy led to the inclusion of two main approaches to the investigation of functional alteration of the brain which are the ROI analyses and ICA analyses. While the two methods of analyses represent both the hypothesis-driven and data-driven approaches to exploring the functional connectivity of the brain, it is noteworthy that very few articles using other methods such as NBS of ALFF were included.

| Limitations
Through ROI-based studies, precise alteration among a priori seeds were precisely reported, highlighting the advantage of exploring functional deficits in regions that are known to be associated with the disorder. In contrast, ICA-based analyses allow an overall view of brain connectivity (Vargas et al., 2013), which enables a more exploratory approach to the investigation of neural correlates of bipolar disorder. However, other techniques that are noted as promising methods of research should also be noted and discussed in future reviews. For instance, researchers have described the applications of graph theory which enables the precise computation of networks according to nodes and edges (Peyron & Fauchon, 2019), or global brain connectivity, which examines the functional connectivity at a voxel level (Hahamy et al., 2014) as important techniques. Future reviews that explore various techniques and measures of functional alteration of the brain in bipolar disorder is warranted. In addition, it is also noteworthy ICA approaches allow the overlap between networks to account for the brain regions that belong to more than one network (Pletzer et al., 2016). The overlap of cortical structures as a result of the respective regions participating in a number of functions across networks has been noted previously (Haller & Bartsch, 2009), making it difficult to interpret the altered functional connectivity between regions at a network level and its clinical implications. For instance, the CEN is composed of the anterior insula, operculum, dorsal anterior cingulate cortex, and thalamus, all of which may overlap with a part of the LIM or SAN (Sadaghiani & D'Esposito, 2014). Therefore, further discussion is warranted in the development of a fixed guideline as to the large-scale networks to which each seed belongs.
Third, the current systematic review did not include other methods to investigate altered brain functional connectivity such as regional homogeneity or expansion of ROI-based approaches using various atlases for parcellation of seed selection. The addition of supportive evidence through other methods as mentioned above may provide more conclusive patterns of altered intra-and internetwork functional connectivity in bipolar disorder.
Fourth, it is noteworthy that 9 of the 49 studies currently reviewed investigated the functional connectivity of the brain in bipolar disorder using a task-based fMRI study design. While this may be a small proportion compared to the majority of the studies reviewed which explored the neurological alterations of bipolar disorder through a resting-state fMRI study design, research has shown that the two MRI acquisition techniques have very distinct functional characteristics . The merging of findings for studies that explored the functional network of the brain at rest with those during a task-based stimuli may challenge the current summary of results. Here, it is also noteworthy that, even within task-based fMRI studies, additional factors such as the difficulty of the task have also been considered as important variables to consider (Manelis et al., 2020). Similarly, growing literature also demonstrates that dynamic functional connectivity provides more informative results than static functional connectivity, and notes the potential shortfalls of the underlying assumption in fMRI studies that functional connectivities of brain networks are stationary (Du et al., 2017;Zhi et al., 2018). Future research that narrows the scope of the review to one of the two fMRI techniques or provides a more extensive literature review including studies that explored the dynamic functional connectivity of large-scale brain networks in bipolar disorder is warranted.
Lastly, the current review article does not discuss the key findings while considering other additional factors associated with bipolar disorder, such as the segregation of results according to the state and mood of bipolar disorder, as well as medication status of the participants. In particular, it is noteworthy that while the majority of the studies discussed in this review provided some sort of information regarding the medication status and history of the bipolar disorder participants, only 4 of the studies reported medication status in a detailed manner where medication load was calculated in the form of ordinal data or composite scoring. While the state and mood of bipolar disorder as well as medication status may be important factors to consider in developing a more extensive and comprehensive understanding of bipolar disorder from a neurological perspective, they have not been actively discussed in this review due to the limited number of studies that provided the relevant data.
For instance, previous research reported significant differences in the functional networks altered according to the type of bipolar disorder (Thomas et al., 2019), while another discussed the potential limitations of interpreting findings on brain activity due to disparity in medication status or the complexity of the medication regimen (Chepenik et al., 2010). Therefore, future studies that also review the state and mood of bipolar disorder as well as the presence of concomitant pharmacotherapy may provide more consistent findings at a network level. Furthermore, the consideration of other important factors that contribute to the complexity of bipolar disorder including genetic polymorphisms (Meda et al., 2014;Oliveira et al., 2015), age of onset (Oliveira et al., 2015), and environmental factors such as maltreatment at an early age (Agnew-Blais & Danese, 2016) is warranted.

| CON CLUS IONS
In conclusion, the current review suggests the functional activity of the triple network model at rest and at task-based stimuli as potential neural correlates for bipolar disorder as compared to healthy individuals. While the current literature review was construed under the assumption that functional connectivity between two brain regions imply functional interactions between two brain networks to which each brain region belongs, the present approach enables the identification of potential networks of interest in revealing the neural correlates of bipolar disorder.
The collective findings of recent literature indicated that individuals with bipolar disorder have overall weakened function of the task-negative state, salience detection, and cognitive control at rest, which may result to the dysfunction in alternating between task-positive and task-negative states and lead to difficulties in affective and cognitive function. Furthermore, such weakened activity across the triple network model may potentially lead to a compensatory response between the task-negative and salience networks at rest, and hyperconnectivity of the cognitive network when performing a task. Studies that used the hypothesis-driven approach of ROI analysis were most prevalent and easily inter-

This research was supported by the National Research
Foundation of Korea (NRF) (2020R1A6A1A03043528 and 2020M3E5D9080555 to IKL). We thank Shinhye Kim for her technical support and contribution toward the preparation of the tables and figures.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N
IKL developed the original idea of the study and supervised the study. SY and TDK retrieved the data. SY, TDK, and JK abstracted and analyzed the data. All the authors wrote and approved the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analyzed in this study.