Dynamic functional connectivity in schizophrenia and bipolar disorder: A review of the evidence and associations with psychopathological features

Alterations of functional network connectivity have been implicated in the pathophysiology of schizophrenia (SCZ) and bipolar disorder (BD). Recent studies also suggest that the temporal dynamics of functional connectivity (dFC) can be altered in these disorders. Here, we summarized the existing literature on dFC in SCZ and BD, and their association with psychopathological and cognitive features. We systematically searched PubMed, Web of Science, and Scopus for studies investigating dFC in SCZ and BD and identified 77 studies. Our findings support a general model of dysconnectivity of dFC in SCZ, whereas a heterogeneous picture arose in BD. Although dFC alterations are more severe and widespread in SCZ compared to BD, dysfunctions of a triple network system underlying goal-directed behavior and sensory-motor networks were present in both disorders. Furthermore, in SCZ, positive and negative symptoms were associated with abnormal dFC. Implications for understanding the pathophysiology of disorders, the role of neurotransmitters, and treatments on dFC are discussed. The lack of standards for dFC metrics, replication studies, and the use of small samples represent major limitations for the field


Introduction
Schizophrenia (SCZ) and bipolar disorder (BD) are severe psychiatric disorders that share a significant overlap in many features, including genetic susceptibility (Lee et al., 2013a;Smoller et al., 2013) and clinical manifestations . Interestingly, both disorders have been associated with neural changes, mainly involving the fronto-thalamostriatal and limbic regions Chen et al., 2011;Leroy et al., 2020;Minzenberg et al., 2009;Wu and Jiang, 2020). Studies using functional magnetic resonance imaging (fMRI) in these disorders have found altered functional connectivity also at rest (rs-fMRI), suggesting a role for intrinsic alterations of brain wiring (Lee et al., 2013b). At rest, several regions of the brain show synchronous low-frequency oscillations of the fMRI signal (Fransson, 2005) that suggest a high level of functional coupling or functional connectivity (FC) between them (Menon, 2011;Raichle, 2011;Smith et al., 2009). Importantly, these sets of connected areas, referred to as resting-state networks, show correspondence with brain networks recruited during the performance of a goal-directed task (Biswal et al., 1997).
Alterations in FC of resting-state networks, particularly the default mode network (DMN), the salience network (SAL), and the executive network (EXE), have been reported in association with several clinical features both SCZ and BD (Hare et al., 2019a;Lee et al., 2018;Menon, 2019;Sambataro et al., 2021a;Whitfield-Gabrieli and Ford, 2012). According to the "triple network" model (Menon, 2011), these networks interact to support cognition, affective functions, and goal-directed behaviors. In particular, the EXE is active in high-order cognition, the DMN is temporally anti-correlated with the EXE and is thought to contribute to vigilance, rumination, self-processing, and learning (Buckner et al., 2008), and the SAL mediates the switching between these networks (Menon, 2011). In addition to these, sensorimotor (SM), visual (VIS), auditory (AUD), language, emotional, and basal ganglia networks have been consistently described at rest in healthy and neuropsychiatric samples (Jimenez et al., 2019;O'Donoghue et al., 2017).
Early resting state studies were based on the assumption that the FC had spatial and temporal stationarity, thus supporting the notion that the average FC could have been representative of the connectivity of the brain. However, brain activity changes dynamically depending on demands, e.g., sleep, sedation, tasks, etc. (Bharath et al., 2017;Harvey et al., 2011), and this also holds for rest, where multiple mental activities can occur . Therefore, while studies operating under the assumption of stationarity have helped to identify rs-fMRI networks, they have not been able to capture their complex dynamic changes (Hutchison et al., 2013). Accordingly, it has been proposed that the study of time-varying aspects of FC, the so-called dynamic functional connectivity (dFC), may provide greater insight into the properties of brain networks Hutchison et al., 2013).
Several measures can be used to characterize the properties of dFC, including functional connectivity strength (FCS), which is a measure of dynamic connectivity calculated as the time-varying sum of connections between a brain voxel and all other voxels and describes the magnitude of signal coupling between brain regions or networks over time in a specific state (Yu et al., 2013). Additionally, measures of stability and predictability of dFC, such as variability, flexibility, entropy, and global efficiency, are also commonly employed. In particular, the FC variability of a specific brain region reflects its dynamic change over time within brain states and is generally estimated by the overall variance of the dFC between networks/regions . Differently, flexibility reflects the dynamic reconfiguration of functional connections between different brain areas that occurs over time and for different tasks (Garcia et al., 2018;Harlalka et al., 2019). Such flexibility could be measured in the context of entropy, which is an index of complexity that characterizes nonlinear properties of resting-state signal (Sokunbi et al., 2011;Wang et al., 2014). Lastly, global efficiency measures the efficiency of information exchange over time in a temporal network (Dai et al., 2016).
In this framework, the overarching goal of this review was to systematically summarize and analyze the extant literature on dFC at resting state in SCZ and BD to identify disease-associated changes.

Article selection and classification
In February 2022 we conducted a systematic search of the literature on PubMed, Web of Science, and Scopus without any language restriction, in accordance with the Meta-analysis of Observational Studies in Epidemiology guidelines (MOOSE, see Supplementary Materials) (Stroup et al., 2000). A combination of the following keywords was used: "dynamic functional connectivity" OR "dynamic functional network connectivity" OR dFC OR "dynamic network connectivity" OR "dynamic brain network" OR "dynamic brain functional network" OR "dynamic brain connectivity" AND schizophrenia OR "bipolar disorder" OR BD or psychosis. We also included relevant studies appearing in the Studies were included if they: 1) estimated dFC; 2) investigated a clinical population affected by BD or/and SCZ; 3) included a healthy control (HC) comparison group. Longitudinal studies were excluded if the baseline dFC was not evaluated.
All selected papers were independently assessed by the authors (GC and FM) and evaluated against inclusion and exclusion criteria. The initial search resulted in 411 articles. The number of duplicates was 198 studies. After reviewing the abstracts of these articles, 123 studies were selected for full-text reading and 43 studies were further excluded because they did not meet the inclusion criteria. Finally, a total of 77 studies were selected (see Fig. 1 for the selection process).

Data extraction
We used a systematic data extraction procedure to individually determine the main characteristics of the included studies with respect to five categories of variables: 1) population (sample size, age, sex); 2) psychiatric diagnosis; 3) control group (sample size, age, sex); 4) experimental design (methodology, diagnostic tools, dFC pipeline); 5) outcomes (dFC alterations, associations between dFC and psychopathological or cognitive variables).

Studies characteristics
All but four studies Lottman et al., 2017;Wang et al., 2021;Zhang et al., 2021) had a cross-sectional design. Overall, 54 studies were carried out in SCZ, 16 in BD, and seven in both disorders. Within SCZ studies, a study was conducted in patients with first-episode psychosis (FEP) (Briend et al., 2020), while two investigations included early-stage SCZ with a duration of the illness of approximately two years Mennigen et al., 2019). Three investigations explored alterations in dFC in SCZ and in their unaffected relatives (Braun et al., 2016;Guo et al., 2018;Su et al., 2016) and two studies also included individuals at high clinical risk of psychosis (CHR) Mennigen et al., 2019). Most studies were conducted at rest, and only in 5 studies dFC was calculated during the performance of a task (Braun et al., 2016;Gifford et al., 2020;Sakoglu et al., 2010;Yue et al., 2018). Eleven studies were conducted on the same sample Faghiri et al., 2021;Fateh et al., 2020;Fu et al., 2021;Fu et al., 2018;Miller et al., 2016aMiller et al., , 2016bRahaman et al., 2021;Salman et al., 2019;Sendi et al., 2021aSendi et al., , 2021b) (see Supplementary Material).
Psychopathological evaluations were carried out using different clinical diagnostic instruments, including the Brief Psychiatric Symptom Scale (BPRS) (Overall and Gorham, 1962), the Hamilton Anxiety Scale (HAM-A) (Hamilton, 1959), the Hamilton Depression Scale (HAM-D) (Hamilton, 1960), the Positive and Negative Affect Scale (PANAS-N) (Watson et al., 1988), the Positive and Negative Symptoms Scale (PANSS) (Kay et al., 1987), the Scale for the Assessment of Negative Symptoms (SANS) (Andersen, 1989), the Scale for the Assessment of Positive Symptoms (SAPS) (Andersen, 1984), the Sign and Symptoms of Psychiatry illness (SSPI) (Liddle et al., 2002), the Structured Interview for Prodromal Syndromes (SIPS) (Miller et al., 2003) and the Young Mania Rating Scale (YMRS) (Young et al., 1978) (see Supplementary Material for a complete list of the clinical scales). Some studies also included functioning and cognitive assessments (see Supplementary Material for details).
The Image acquisition protocols, analytic methods, and clinical assessment tools of the studies are summarized in Tables 1, 2, 3.

dFC techniques
First, fMRI time series were preprocessed, and then dFC analysis was performed with the following steps: a) signal extraction to obtain meaningful metrics in terms of raw time series, low-frequency oscillations, regional connectivity, etc.; b) dFC calculation, where several timedependent connectivity matrices are obtained across the whole timeseries; and finally, 3) the estimation of recurring and stable patterns of dFC states at the individual and group level (see Table 4).

Signal extraction
The most widely used technique to obtain time series was spatial group independent component analysis (ICA) (Calhoun et al., 2001). Furthermore, 19 studies explored FC with a seed-based approach that detects the univariate pairwise correlation of one or more a prioriselected seeds or regions of interest (ROI) and other areas of the brain, thus producing seed-based FC maps (Wu et al., 2018). One study investigated the dynamics of regional homogeneity (ReHo) (Dong et al., 2019), which studies the similarity of the time series of a particular voxel with the time series of neighboring voxels, providing a measure of localized FC (Zang et al., 2004). Spontaneous brain activity at rest can be measured not only using the correlation between time series but also by exploring changes in the frequency domain, which is the analysis of the power spectrum that allows the study of specific frequencies of the signal. In particular, the amplitude of low-frequency fluctuations (ALFF) and the fractional amplitude of low-frequency fluctuations (fALFF) detect the intensity of spontaneous low-frequency fluctuations of the BOLD signal in the whole brain (Turner et al., 2013). Here, we included studies that explored the dynamics of ALFF (dALFF) and dynamic fALFF (dfALFF), defined as recurring patterns of ALFF and fALFF variability over time calculated with the sliding window approach and clustered in states Fu et al., 2018;He et al., 2021;Liang et al., 2020;Luo et al., 2021;Nyatega et al., 2021). Moreover, Yang et al. (2020) used voxel mirrored homotopic connectivity analysis to investigate the temporal variability of interhemispheric functional connectivity between homotopic areas ) (see Supplementary Material).

dFC calculation
Several techniques were used to study time-varying changes in FC. One of the most widely used was the sliding window (SW) approach, which partitions the time course of the fMRI signal into several fixed temporal windows (that may partially overlap), where pairwise correlations between regions/networks are computed until reaching the end of the time courses itself (Hutchison et al., 2013;Rashid et al., 2014). Then, to assess the frequency and structure of reoccurring FC patterns, a clustering algorithm for windowed covariance matrices is commonly used (Lloyd, 1982) (Fig. 2). In addition, other methods have been used, including: 1) network flexibility, which is a measure of how often a brain area changes its allegiance to a community of nodes over time (Braun et al., 2016); 2) quasi-periodic patterns (QPP), which reflect the spatiotemporal patterns of signal oscillations in the infra-low frequency range and are supposed to underlie functional connectivity (Briend et al., 2020); 3) filter-banked connectivity, an approach that does not make a priori assumptions about connectivity frequency and performs frequency tiling in the connectivity domain (Faghiri et al., 2021); and 4) dynamic directional functional domain connectivity, a method that operates at a dimensional scale sufficient to capture multiplexed dynamical relationships within and between functional domains (Miller et al., 2016b). Four studies explored the effects of global signal regression on dFC findings and three investigations examined frequencyrelated changes in dFC (see Supplementary Material).

Estimation of connectivity states
k-means clustering is one of the most widely used methods to modularize windowed connectivity patterns. Briefly, k-means clustering is an unsupervised technique that automatically partitions a data set into a predefined number (k) of clusters, typically spanning from 2 to 20 (Shakil et al., 2014;Supekar et al., 2019). In this context, each state is         Positive correlation between the temporal regional efficiency in the left orbitofrontal and PANSS positive scores.
Negative correlation between the temporal regional efficiency in the precuneus and left temporal pole and PANSS negative scores. Positive correlation between the temporal regional efficiency in the left orbitofrontal and PANSS general scores. Negative correlation between the temporal regional efficiency in the amygdala and left temporal pole and PANSS overall scores.   mutually exclusive and the time spent in a specific connectivity state is defined as dwell time. Also, the dFC was measured using spatio-temporal meta-state analysis. Recently, cross-domain mutual information (CDMI) that uses mutual information (i.e., mutual dependence between pairs of variables adopted from a measure from information theory) within the brain networks belonging to the same functional domain has been used to estimate dFC thus including linear and nonlinear relationships . See the Supplementary Material for other dFC techniques.
In the following paragraphs, for each network we will use this approach: first, we will describe the magnitude of the dFC (FCS), then its variability, and its interaction with other networks. Changes in dFC in SCZ and BD in terms of FCS and variability are illustrated in Fig. 3.

Global connectivity
A general pattern of dynamic dysconnectivity between brain networks was reported in SCZ (Long et al., 2021). Consistent with this, both increases and decreases in dFC were described in different frequency bands, mainly distributed in the triple network, cerebellum, VIS, SM, and the subcortical network . Importantly, several studies showed that SCZ spent less time in globally coherent and subcortical-centered states Espinoza et al., 2019;Sanfratello et al., 2019;Zarghami et al., 2020) and in states with high within-and between-FC of sensory networks (Weber et al., 2020), while they dwelled longer in states characterized by strong FC within networks (in particular, the DMN and the language network) (Weber et al., 2020). Differently, Plis et al. (2018) showed that SCZ made significantly more transitions to states characterized by weaker connectivity within most brain networks (subcortical, AUD, VIS, SM, EXE, DMN, and cerebellum) (Plis et al., 2018). A reduction in time-varying connectivity patterns in the whole-brain networks was reported (Miller et al., 2016c;Rabany et al., 2019), particularly in patients with more severe hallucinations (Miller et al., 2016b). Moreover, SCZ presented increased entropy and reduced cross-domain mutual information, which is a measure of dependence across sets of related brain areas grouped for anatomical and functional associations, indicating reduced dynamic changes in brain connectivity (Salman et al., 2019;Salman et al., 2017). Stepwise functional network reconfiguration (sFNR), a measure reflecting the global ability to rewire brain networks, was increased in large-scale brain networks, including SM, VIS, EXE, and DMN, thus reflecting an increased temporal variability of the networks and, therefore, their instability (Fu et al., 2021). Finally, when selectively investigating the low-frequency bands, SCZ had more occurrences of states characterized by weaker widespread dALFF patterns and fewer occurrences of strong dALFF states in most brain networks, particularly the AUD, SM, VIS, and subcortical networks .

Default mode network
Within-DMN dFC was reduced (Du et al., 2016;Luo et al., 2020;Salman et al., 2019;Sendi et al., 2021a), although posterior DMN (i.e., right medial parietal cortex) showed increased temporal global efficiency (Sun et al., 2019). The variability of DMN was reduced (Dong et al., 2019). Furthermore, synchronizability, modularity, recurrence, and consistency of the statelets in the DMN were decreased, suggesting that SCZ exhibit more erratic and less efficient communication between the DMN and other brain networks (Rahaman et al., 2021) (see Supplementary Material for the definition of statelets). Between-network dFC revealed that SCZ dwelled in or switch to a state with high positive connectivity between DMN and EXE (Bhinge et al., 2019).

Executive and attention network
Patients presented either a general reduction (Long et al., 2021) or an increase  in within-EXE functional FCS, and spent more time in states with weaker FCS in this network (Zarghami et al., 2020).
The contribution of QPP to FCS was greater in EXE in FEP (Briend et al., 2020), suggesting a greater impact of QPP on intrinsic brain activity in these subjects. In contrast, the frontal cortex had a lower state-specific FCS in all the states  and higher temporal nodal efficiency, which assesses the efficiency of information transfer between nodes in a temporal network (Sun et al., 2019). Additionally, patients with SCZ demonstrated a decrease in state-specific FCS between EXE and the cerebellar motor cluster (He et al., 2019) and between EXE and VIS  Regarding variability measures, higher flexibility scores were reported in SCZ in the EXE (Gifford et al., 2020), along with increased voxel-wise, region-wise, and network-wise FC variability in the attention network (Dong et al., 2019).

Salience network
SAL FCS and within-network connectivity were reduced in different frequency bands in SCZ . Between-network dynamic interactions of SAL-centered cross-networks within the triple-network model were significantly reduced, less persistent, and more variable in patients (Supekar et al., 2019;Wang et al., 2016).

Sensory-motor network
FCS was increased in the motor network (Du et al., 2021a), but showed high variability and reduced interaction with other networks. In particular, flexibility and variability were higher in the cerebellar, subcortical, and thalamic areas in SCZ (Gifford et al., 2020). Conversely, FCS between the motor and the EXE, DMN, and SM (He et al., 2019), as well as between the SM and the VIS and AUD (Faghiri et al., 2021), was reduced. Additionally, SCZ dwelled less in states with the predominance of sensory and motor networks (Faghiri et al., 2020;Sendi et al., 2021a). Lastly, the synchronizability, modularity, recurrence, and consistency of the statelets were reduced (Rahaman et al., 2021).

Visual networks
FCS was reduced in VIS (Sheng et al., 2021) and also between VIS and the EXE , AUD and SM networks , and the mirror system network . A higher sample entropy was observed in the right middle occipital gyrus (Jia and Gu, 2019), while the lateral occipital cortex showed an increased interaction with EXE and the thalamus at rest, and with DMN during task switching . Synchronizability, modularity, recurrence, and consistency in VIS networks were reduced (Rahaman et al., 2021) and FC variability was increased in dorsal VIS (Deng et al., 2019).

Emotional network
The FC variability was reduced within the emotional network (Deng et al., 2021), and increased between the amygdala-prefrontal network in SCZ (Yue et al., 2018).

Subcortical and other networks
Higher flexibility scores (Gifford et al., 2020) and temporal global efficiency (Sun et al., 2019) were reported in subcortical areas. Decreased FCS was also reported between the olfactory cortex and the hippocampus, and this may be part of altered sensory integration patterns in this disorder (Du et al., 2021a).

dFC alterations in relatives of SCZ and CHR
Mixed results in small samples have been reported in unaffected siblings of SCZ. A small study found dysconnectivity within DMN, SAL and VIS (Su et al., 2016), with a general and nondomain-specific increase in network flexibility (Braun et al., 2016). Other studies investigating whole-brain FC in relatives (Guo et al., 2018) and general dFC , and transitions (Mennigen et al., 2019) between states in clinical high-at-risk individuals (CHR) did not find differences between individuals at risk and HC.

Global connectivity
A heterogeneous picture of alterations in global dFC was observed in BD. The FCS between the right anterior insula and the right middle occipital gyrus and the left inferior parietal lobule was increased (Pang et al., 2018). Furthermore, in patients with BD in the depressive phase, the dynamic interhemispheric connectivity, defined as the dFC between a given voxel and the corresponding homologous voxel in the contralateral hemisphere, was reduced in the superior parietal lobule, the angular gyrus, the precuneus, and increased in the cerebellum, orbitofrontal cortex, postcentral gyrus, superior temporal gyrus and supplementary motor area. Notably, increased dynamic interhemispheric connectivity in the postcentral gyrus was associated with a greater number of depressive episodes . When affective status was considered, depressed BD switched more between states and dwelled more in a state characterized by a negative correlation between the SAL, cerebellum, and the subcortical network and the SM, AUD, and VIS, and less in a state characterized by negative correlations between the DMN and other functional networks . Compared to MDD, unmedicated patients with BD-II showed greater variability in dFC between the dorsal striatal putamen and sensory-motor regions (i.e., left supramarginal area) and the ventral rostral putamen and the parietal cortex (i.e., right inferior parietal lobule), similarly to MDD, and between the dorsocaudal putamen and the motor regions (i.e., precentral gyrus) compared to MDD and HC . Lastly, a study conducted on euthymic BD reported an increased number of transitions between a high-level cognitive state and a low-level sensory state in BD (Du et al., 2021b).

Default mode network
BD was associated with decreased network switching rate in the DMN (Han et al., 2020). In particular, reduced dFC was present in posterior DMN in depressed patients with BD , and specifically BD-I (Liang et al., 2020). Also, FCS between DMN (middle temporal gyrus and the postcentral gyrus) and SM (superior temporal gyrus) was reduced during depression relative to euthymia in BD . In BD-I, the FCS between the two hubs of the DMN (medial prefrontal cortex and posterior cingulate cortex) was less variable over time, indicating greater rigidity and this was associated with reduced cognitive performance (Nguyen et al., 2017).

Executive network
The dFC in the frontal-striatal-thalamic circuit was increased in euthymic BD  and in depressed BD relative to HC (Tang et al., 2022) and MDD and HC (Pang et al., 2020).

Salience network
Euthymic BD showed increased dFC variability of the right anterior insula. Notably, BD shared a reduced variability between the right ventral anterior insula and the ventrolateral prefrontal cortex with MDD and had the greatest variability of the dFC of the right dorsal anterior insula with temporo-occipital regions compared to MDD and HC (Pang et al., 2018).

Sensory-motor network
A state-dependent increase of FCS between SM and DMN, which was greater in depressed BD relative to euthymic BD relative to euthymic BD and HC, was reported by one study .

Emotional network
Depressed BD was associated with changes in between-network FCS of the limbic system and precisely increased amygdala-cerebellar and decreased amygdala-postcentral gyral dFC, respectively (Fateh et al., 2020). In addition, depressed BD showed reduced dynamic regional phase synchrony, a measure of instantaneous coherence, in fronto-   striato-limbic areas (Tang et al., 2022).

dFC differences between SCZ and BD
Studies comparing SCZ and BD indicated greater dysconnectivity in SCZ relative to BD, with a pattern of decreased within-network dFC in VIS, SM, SAL and EXE, increased dFC between the VIS and the EXE, SAL and limbic networks, and decreased dFC between the SAL and EXE, DMN and SM, and EXE and DMN (Li et al., 2021). SCZ had more widespread dFC changes relative to BD, involving increased FC variability in the SM, VIS, attention, limbic and subcortical areas at the regional and network levels, as well as decreased regional FC variabilities in the DMN areas (Long et al., 2020). In line with this, a similar aberrant FC pattern was reported in DMN, VIS, SM, and EXE in SCZ and BD, with a greater magnitude of changes in SCZ relative to HC (Rashid et al., 2014). Parieto-parietal inter-hemispheric network dFC was greater in both SCZ and BD in the right hemisphere, and in BD only in the left hemisphere, respectively, compared to HC (Das et al., 2020). Furthermore, an increase in functional stability in VIS (i.e., calcarine sulcus) was reported in BD relative to SCZ, indicating a higher concordance of dynamic FC over time in these patients . When compared within the bipolar-schizophrenia spectrum, a reduced dFC fronto-parieto-cerebellar circuit with increased dFC in corticothalamic networks was observed, and the magnitude of this dysconnectivity increased from HC to BD, schizoaffective disorder (SAD), and SCZ. SCZ, BD, and SAD shared a decrease in FCS between the thalamus and cerebellum and an increase in FCS between the postcentral gyrus and the thalamus . A follow-up study showed that BD and SCZ had similar connectivity changes between VIS (i.e., cuneus) and the insula, the putamen, and the supramarginal gyrus (Du et al., 2020).

Brain-behavior correlations
3.6.1. PANSS positive In SCZ, the PANSS positive score was associated with the variability of dFC and cross-domain mutual information (Dong et al., 2019;Salman et al., 2019) and sample entropy (Jia and Gu, 2019) of the VIS, in addition to dynamic time-varying measures of SAL Supekar et al., 2019). Furthermore, a correlation was observed between PANSS positive scores and FCS of the left thalamus ) and temporal regional efficiency in the left inferior orbitofrontal gyrus (Sun et al., 2019). In BD, SAD, and SCZ, hypoconnectivity between postcentral and frontal gyri was negatively correlated with PANSS positive scores .

PANSS negative
In SCZ, the PANSS negative scores were correlated with the variability of dFC and temporal regional efficiency (Deng et al., 2019;Sun et al., 2019), and the entropy (Jia and Gu, 2019) of VIS and abnormal FC variability (Dong et al., 2019). Additionally, an association was also observed between PANSS negative and FCS in the right insula and the left orbital inferior frontal gyrus  and the left cerebellum crus 1 (Wang et al., 2019a). Moreover, negative symptom severity was associated with the probability of transition from a state with predominant anterior-to-posterior DMN (lower precuneus/posterior cingulate cortex and higher anterior cingulate cortex) FC relative to a state with reverse pattern (higher precuneus/posterior cingulate cortex and lower anterior cingulate cortex) (Sendi et al., 2021b). Dwelling longer in a state characterized by sparse and weak connectivity predicted PANSS negative scores, with reduced DMN and VIS dFC predicting greater attention domain impairment .
In a study conducted in a small sample of adolescent-onset SCZ, reduced dFC between the left middle temporal gyrus and the left extrastriate visual area predicted increased emotional withdrawal evaluated with item 2 of PANSS negative . Lastly, hypoconnectivities linking postcentral and frontal gyri were negatively correlated with the PANSS negative scores in BD, SAD, and SCZ (Du Table 4 Processing steps for the calculation of dynamic functional connectivity.

ANSS total
The FCS of the cortico-thalamic circuits (i.e., the bilateral insula, left thalamus, and left paracentral lobule) , of temporal (i. e., right amygdala and left temporal pole) (Sun et al., 2019), and striatoparietal networks (i.e., right supramarginal gyrus and right putamen) (Wang et al., 2019a), reduced dALFF of the SAL-EXE connection , and increased variability of dFC of the frontal-amygdala connection (Yue et al., 2018) were associated with a higher PANSS total score, thus supporting the dysconnectivity hypothesis of SCZ (Yue et al., 2018). Additionally, the overall symptom severity was associated with the greater probability of transitioning from a state with predominant anterior-to-posterior DMN (lower precuneus/posterior cingulate cortex and higher anterior cingulate cortex) FC relative to a state with reverse pattern (higher precuneus/posterior cingulate cortex and lower anterior cingulate cortex) (Sendi et al., 2021b). The FC variability of VIS (Deng et al., 2019) and VIS, SM, and thalamus (Dong et al., 2019) was associated with a higher PANSS total score.
The correlations between PANSS scores and dFC measures are summarized in Table 5.

Other symptom scales
In SCZ, trait hallucination proneness over one year showed a significant association with dwell times in a state characterized by strong positive FC within the DMN and negative FC between the DMN and the insula (Weber et al., 2020), while hallucination severity measured with BPRS was positively correlated with the temporal instability of lateral occipital cortex connectivity . Additionally, illness duration was associated with the entropy of the VIS (i.e., left superior occipital gyrus) (Jia and Gu, 2019), cortico-limbic networks (i.e., right amygdala, right superior orbital frontal gyrus, and left inferior parietal gyrus) (Jia et al., 2017). In depressed BD, depression severity (HAMD score) was positively correlated with the dFC between the right anterior insula and inferior parietal lobule (Pang et al., 2018) and with dwelling in a state with decreased FC between DMN, SAL, and EXE (Wang et al., 2019b). Moreover, in depressed BD, the abnormal dynamic FCS in the frontal-striatum-thalamic circuit predicted anhedonia measured with the Snaith-Hamilton Pleasure Scale (Pang et al., 2020). Disorganization evaluated with the SSPI was associated with dFC in the SCZ, but not in BD (Das et al., 2020). The correlations between clinical scales and dFC measures are summarized in Table 5.

Cognitive performances
Four studies explored the relationship between dFC measures and cognitive performance in SCZ and BD (Table 6). In SCZ, dwell time in a state with positive FC within the middle temporal gyrus and between the middle temporal gyrus with other regions predicted visual learning memory (Sendi et al., 2021a). The variability of FC in cortico-limbic circuits (i.e., amygdala-medial prefrontal cortex) was associated with poorer performance on the digit symbol coding task (Yue et al., 2018), while the temporal instability of the lateral occipital cortex connectivity predicted higher switching costs during task performance in SCZ . In BD, reduced connectivity variability within the DMN was associated with slower processing speed and impaired set-shifting (Nguyen et al., 2017).

Effect of medications on dFC
Four studies explored the effects of second-generation antipsychotics on dFC in SCZ and they all described a normalizing effect on dFC and clinical symptoms. Unfortunately, the sample size of these studies was considerably smaller compared to most cross-sectional studies. Two longitudinal trials focused on risperidone Lottman et al., 2017). The first employed risperidone at a dosage of 4-6 mg/day for 8 weeks in antipsychotic-naïve first-episode patients with SCZ and observed a normalization of the dFC variance of the abnormal connections , while the second showed a normalization of mean dwell times in a sparsely connected state with a dosage of 4.4 mg/ day after 6 weeks (Lottman et al., 2017). In both studies, the treatment also resulted in clinical improvement. Wang et al. (2021) explored dFC after 12 weeks of treatment with various atypical antipsychotics, including olanzapine, risperidone, paliperidone, ziprasidone, quetiapine, amisulpride, and aripiprazole in monotherapy (62.5%) or in combination (37.5%) and showed that, compared with HC, SCZ presented more unstable brain states, which normalized to some extent after antipsychotic treatment. Furthermore, in this case, antipsychotic treatment was associated with a decrease in PANSS scores . Lastly, after 8 weeks of various antipsychotic treatments, including paliperidone, clozapine, risperidone, olanzapine, aripiprazole, and quetiapine, a significant increase in the symptomatic improvementrelated occurrence of a dFC state characterized by greater inter-network integration was observed. Furthermore, the reduction in symptoms was correlated with increased FC variability in the connections within the DMN and between the AUD, EXE, and cerebellar network to other networks .

Discussion
Our systematic review aimed at summarizing all available evidence on dFC alterations at resting state in SCZ and BD and their association with psychiatric symptoms and behavior. We found a global alteration of dFC in SCZ, while a more heterogeneous picture of altered dFC was observed in BD. However, in both disorders, dysfunction of the triple network involved in the performance of goal-directed behavior emerged. A direct comparison between SCZ and BD confirmed a predominant pattern of dysconnectivity in the triple network in SCZ. Psychopathological measures showed an association with dFC metrics in almost all the studies on SCZ, with positive and negative symptoms demonstrating an association with abnormal dFC. Remarkably, dFC alterations were normalized after antipsychotic treatment in responders.

Schizophrenia
Overall, the findings of our review show a consistent pattern of dFC alterations in SCZ compared to HC, involving abnormal FCS and an increased dwell time and a number of transitions to states characterized by weaker connectivity within and between all major resting-state networks.
Significant progress in the neuroimaging field in recent decades has provided robust evidence to the so-called "dysconnectivity" theory, postulated to explain the core psychopathological characteristics of SCZ. First described in the 1990s, this theory was based on the observation of abnormal functional integration between anatomically distinct brain regions (Friston and Frith, 1995;Stephan et al., 2009) at the core of symptomatology in SCZ. Importantly, SCZ is characterized by both global dysconnectivity, as demonstrated by global signal abnormalities, and alterations at the topographic level in lower-order sensory and higher-order cognitive regions that may underlie sensory and cognitive symptoms (Yang et al., 2014;Zhang and Northoff, 2022). Accordingly, a consistent pattern of dFC alterations in the triple network has been suggested to play a prominent role in the pathogenesis of SCZ (Dong et al., 2018;Menon, 2011). Interestingly, structural and functional alterations in SAL have been commonly associated with impaired attribution of salience to stimuli, which, in turn, is associated with delusions and hallucinations in SCZ (Palaniyappan et al., 2011). Furthermore, altered FC between SAL, EXE, and DMN has been associated with positive and negative symptoms (Hare et al., 2019b;Manoliu et al., 2014). In our review, alterations in dFC involving areas of the triple network appeared to be associated with psychiatric symptoms in SCZ (Dong et al., 2019;He et al., 2021;Luo et al., 2020;Salman et al., 2017;Sun et al., 2019). Among these, Supekar et al. (2019) showed a positive association between the lack of dynamic engagement of the SAL with the EXE and DMN and disorganized thought (Supekar et al., 2019). Overall, our results suggest that patients with SCZ present a reduction in dynamic connectivity metrics in the triple network, which may underlie psychotic symptoms for altered salience attribution, negative symptoms for altered DMN persistence, and cognition for impairment of EXE connectivity. In addition, abnormalities in dFC were reported in sensorimotor circuits, particularly in the VIS (Deng et al., 2019), AUD (Geng et al., 2020), and SM (Sambataro et al., 2021b), suggesting that altered FC metrics in these areas could be associated with deficits in the processing of external stimuli, which may lead to psychotic symptoms (Kubera et al., 2019;Thoma et al., 2016). In particular, abnormalities in the VIS and AUD pathways have been commonly reported in SCZ (Harvey et al., 2011;Kaufmann et al., 2015), and appear to be associated with hallucinations and negative symptoms (Orliac et al., 2017).
Interestingly, several studies have shown a relationship between changes in SM network dynamics and psychopathological measures, such as PANSS total (Deng et al., 2019;Dong et al., 2019), general (Dong et al., 2019;Jia and Gu, 2019) and negative scores (Deng et al., 2019;Jia and Gu, 2019;Wang et al., 2019a). Furthermore, changes in dFC in sensory networks showed a correlation with positive symptoms evaluated with PANSS (Dong et al., 2019;Jia and Gu, 2019;Salman et al., 2017;Sun et al., 2019), as well as with hallucination severity measured with BPRS . These results align with the spatiotemporal model of psychopathology proposed by Northoff and Duncan (Northoff, 2015;Northoff and Duncan, 2016), according to which temporal and spatial changes in spontaneous brain activity affect cognitive and affective processing in SCZ. In particular, abnormalities in the SM and the sensory networks dFC could be associated with altered perceptions of spatial relationships with respect to the body and the environment in patients with SCZ, which might lead to delusions and hallucinations. Furthermore, as previously demonstrated in depression (Northoff, 2016), affective and cognitive symptoms such as anhedonia could be the phenotypic manifestation of spatiotemporal disturbances of the activity of the resting state that in SCZ appear to be prevalent in the VIS network, frontal areas, insula, and cerebellum.
Interestingly, dFC abnormalities have also been reported in individuals at genetic risk for SCZ (Braun et al., 2016). Meta-analytic evidence from task-based fMRI studies has shown that unaffected relatives of SCZ present a pattern of functional abnormalities involving the cortico-striato-thalamic network , while the few available rs-fMRI studies showed alterations in the prefrontal, thalamic, limbic, and SAL networks (Li et al., 2015;van Leeuwen et al., 2021;Xi et al., 2020). Abnormal dFC in subjects at risk for SCZ suggests that these alterations are not related to the pathology itself but may be a risk phenotype of the disorder (intermediate phenotype).
Finally, studies investigating the effect of antipsychotics on dFC in SCZ showed a normalization of the metrics of dFC after treatment, which was accompanied by a symptomatic improvement Lottman et al., 2017;Wang et al., 2021;Zhang et al., 2021). It is plausible that dFC abnormalities may reflect disorganized patterns of neuronal activity that could result in the inability of patients to reside in globally coherent states, leading to an impaired ability to perceive, process, and filter out external information. Antipsychotic medications, which decrease neurotransmitter hyperactivity, might attenuate aberrant brain dynamics and result in a decrease in symptoms (see below).

Bipolar disorder
A more heterogeneous picture is derived from studies conducted in patients with BD. Here, previous findings at rest commonly reported topographical alterations in the motor cortex and hippocampus that vary with mood phase and reflect behavioral and cognitive symptoms, while the global signal does not appear to change (Zhang and Northoff, 2022). The abnormalities of the dFC involved a wide range of cortical and subcortical areas, including frontal areas, limbic lobe, basal ganglia, and thalamus, along with large brain networks, such as DMN, EXE, SAL, and SM. Our results are in line with static rs-fMRI investigations that showed that BD was characterized by hypo and hyperconnectivity within the DMN, affective, EXE, ventral attention, SM and thalamic networks . In particular, in BD we found that the anterior insula, which is a key node of the SAL, had greater connectivity to the inferior parietal cortex, a node of the EXE, and reduced connectivity to the right ventrolateral FPC, which is another important region of this network for the control of cognition and impulsivity. Additionally, the DMN showed reduced integrity and modulation both in terms of lower network switching and reduced connectivity between its subnetworks, reduced dALFF, and altered interplay with anticorrelated networks, including EXE and SM. Abnormal thalamocortical connectivity may be a part of EXE dysfunction and may contribute to emotional dysregulation (Ramsay, 2019), which is a prominent feature of this disorder . Altered connectivity of the SAL can result in impaired cognition-emotion interaction and therefore contribute to the well-known mood and cognition impairments reported in BD (Ellard et al., 2019). Furthermore, altered amygdala connectivity has been extensively studied in BD for its role in emotional processing and for its widespread interaction with brain networks (Rey et al., 2021).
Abnormal connectivity of the amygdala may contribute to the pathogenesis of the emotional and behavioral symptoms that are present in BD  by: 1) increased connectivity with the cerebellum, which has been implicated not only in sensorimotor function but Table 5 Associations between clinical scales and dFC measures in schizophrenia and bipolar disorder. also in emotion and motivational processing and in several psychiatric disorders (Phillips et al., 2015); 2) reduced connectivity with the somatosensory cortex (postcentral gyrus) that could be responsible for the interaction between emotion and motor control and its subjective experience (Toschi et al., 2017). Remarkably, these functional coupling changes were also present in studies that focused only on BD-I, which is more closely related to SCZ, suggesting partial shared pathophysiological mechanisms for these disorders (Trevisan et al., 2022). These heterogeneous results could be explained by the manifold clinical characteristics of patients with BD, both in terms of mood state (i.e., depression, euthymia, mania), presence/absence of psychotic symptoms and duration of the disease. Interestingly, half of the studies that explored the correlations between dFC metrics and psychopathology observed an association with depressive symptoms evaluated with HAMD (Pang et al., 2018;Wang et al., 2019b) and the severity of anhedonia (Pang et al., 2020), while the others did not.

Disorder-specific changes
Investigations comparing dFC in SCZ and BD showed that these disorders present some commonalities, however FC alterations twere more pronounced in SCZ compared to BD. Notably, studies exploring the association between psychopathology and dFC in BD and SCZ showed that dynamic FC parameters in SCZ were correlated with the disorganization evaluated with the SSPI scale in the SCZ group, while no correlations were observed in the BD group (Das et al., 2020). Interestingly, a correlation was observed between PANSS scores and dFC in BD, SAD, and SCZ . Furthermore, dFC metrics were also correlated with cognitive performance in SCZ and BD, suggesting that brain dynamics could be involved not only in the development of psychopathology but also in cognition.

Altered connectivity and signaling pathophysiological models
At the level of brain circuits, the dynamics of functional connectivity can arise from changes in local cortical states that can interact with remote regions within large-scale networks (Hutchison et al., 2013). Additionally, subcortical circuits, including subthalamic and brainstem regions, could affect the reconfiguration of these brain networks by modulating neurotransmitter signaling systems. Dopamine has been associated with the dynamics of brain networks (Sambataro et al., 2009) for its role in stabilizing cortical responses through the modulation of cortical pyramidal neurons and GABA-inhibitory interneurons. Furthermore, GABA can modulate the frequency of membrane oscillations and result in increased synchronization within large-scale networks (Seamans and Yang, 2004). Increased presynaptic dopamine signaling has been implicated in SCZ, in the so-called "dopamine hypothesis", and similarly, albeit of a small magnitude, increased D2/D3 availability and striatal dopamine amino transporter levels have been reported in BD (Ashok et al., 2017). The normalizing effects of antipsychotics that are mostly D2 antagonists seem to corroborate these results.
Furthermore, glutamate signaling (particularly N-methyl-D-aspartate, NMDA) has also been implicated in modulating brain dynamics and in SCZ. Braun et al. (2016) showed that, during working memory processing, dextromethorphan, an NMDA-receptor antagonist, can increase network flexibility, a measure of the ability to reconfigure a node within a network, which suggests temporal disorganization of the community structure of the brain (Braun et al., 2016). Similar hyperflexibility was also found in SCZ in the same study. In particular, altered glutamatergic signaling with hypoactivity of the NMDA system in excitatory pyramidal cortical cells and in fast-spiking GABA inhibitory interneurons can affect the synchrony of brain oscillations and their discharge, ultimately translating into reduced stability of brain networks (Uhlhaas and Singer, 2010), and can result in positive, negative, and cognitive symptoms of SCZ (Merritt et al., 2013). Moreover, converging evidence from preclinical and clinical studies suggests an increased activity of NMDA in BD, with mood stabilizers modulating the glutamatergic signaling (Fountoulakis, 2012). Pharmacological studies with NMDA antagonists, including ketamine, memantine, and magnesium, have also shown some efficacy in BD depression, although further studies are needed to confirm these results (Delfino et al., 2020). Finally, antipsychotics can modulate NMDA activity and this effect can contribute to their clinical effects (Choi et al., 2009). In general, alterations in dopamine and glutamate signaling can alter dFC and contribute to the pathophysiology of SCZ and BD.
Overall, brain connectivity at rest is not static but oscillates over time across several brain states, which can be defined as spatial patterns of signals that are stable for a certain period of time. The study of dFC is complex as the time scale of the phenomena that occur in the brain can be highly variable, the number of states is unknown, and they can be intertwined and interact with each other. Furthermore, cross-frequency coupling may drive the self-organized dynamics of brain states with lowfrequency oscillations modulating the synchronization patterns of faster rhythms (Vanhatalo et al., 2004). Recent approaches have tried to unravel the interaction between brain states and have considered the coexistence of multiple states at a specific time point rather than an allor-nothing phenomenon (Miller et al., 2016c). These achievements have contributed to a better understanding of SCZ and BD in terms of brain dynamics. Future studies on the physiology, reliability, and replicability of dFC indexes are needed to create gold standard measures for this novel field, thus allowing the comparability across studies, and more thorough analyses of the molecular and electrophysiological correlates of these phenomena (Hutchison et al., 2013).

Limitations
The results of this review must be interpreted in light of some limitations. First, there was considerable heterogeneity in the image acquisition parameters and dFC techniques used by the included studies. Second, the characteristics of the patients differed between the studies, increasing the ecological validity of this study, but, at the same time, contributing to the heterogeneity of our findings. Third, the majority of patients were taking psychotropic medications, which could have influenced our results.

Conclusions
In conclusion, a pattern of abnormal dFC was observed in SCZ, involving mainly the EXE, SAL, DMN, and sensorimotor circuits, and these alterations are associated with psychopathological features such as Table 6 Associations between cognition and dFC measures in schizophrenia and bipolar disorder.

Group
Cognitive tests Associations between cognition and dFC measures SCZ visual learning memory -dwell time in a state with positive FC within the middle temporal gyrus and between the middle temporal gyrus with other regions predicted visual learning memory performances (Sendi et al., 2021a) digit symbol coding task -variability of FC in cortico-limbic circuits was associated with poorer performance on the digit symbol coding task (Yue et al., 2018) switching costs -temporal instability of lateral occipital cortex connectivity predicted higher switching costs during task performance ) BD processing speed and set-shifting -reduced FC variability within the DMN was associated with slower processing speed and impaired set-shifting (Nguyen et al., 2017) BD: Bipolar Disorder, DMN: Default mode network, FC: Functional connectivity, SCZ: Schizophrenia.
hallucinations and delusions. In BD, a mixed picture of altered dFC was observed, with only some studies reporting an association with affective symptoms. Alterations in dFC have also been observed in unaffected relatives of SCZ, but not in individuals at clinical risk of psychosis. Lastly, antipsychotic treatment, when effective in relieving psychiatric symptoms, seems to play a normalizing role in dynamic abnormalities, thus suggesting a potential avenue for developing effective treatments.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.