Major depressive disorder (MDD) is the leading cause of disability worldwide and is characterized by high morbidity and recurrence rates (Eaton et al., 2008; Vos et al., 2017). After remission, a substantial number of individuals with MDD are susceptible to recurrences, likely due to cognitive vulnerabilities and underlying neurobiological mechanisms (Holtzheimer and Mayberg, 2011). The cognitive vulnerability model suggests that depressive episodes can be triggered by the interaction of premorbid vulnerabilities (e.g. cognitive risk factors like rumination and neural alterations) and psychosocial stressors (Beck, 2008; Ingram et al., 1998). Given the high rate of recurrence, the identification of trait-like brain functional abnormalities in individuals with first-episode, medication-naïve current major depressive disorder (cMDD) and remitted major depressive disorder (rMDD) is crucial to identify targets for prevention.

Traditional (i.e., static) resting-state functional connectivity (RSFC) analyses have shown that depression is characterized by overall alterations in the coordination of large-scale brain network function (i.e., across the entire scan) (Fischer et al., 2016; Kaiser et al., 2015). These alterations include hyperconnectivity of the default mode network (DMN), involved in self-referential thought (Andrews-Hanna et al., 2014), and hypoconnectivity of the frontoparietal network (FPN), involved in executive functions and stress regulation (Kaiser et al., 2015; Snyder, 2013). Based on this evidence and subsequent empirical work, imbalance among intrinsic networks and impairment of DMN–FPN interactions have been strongly implicated in the pathophysiology of MDD (Li et al., 2018; Wang et al., 2020a). Integrating findings from at-risk, depressed, and formerly depressed populations (Chai et al., 2016; Li et al., 2013; Nixon et al., 2014), it has been proposed that alterations of dynamic interactions among intrinsic functional network are critically implicated in increased risk for future MDD (Beevers, 2005; Wang et al., 2016). However, to the best of our knowledge, no prior report has investigated brain dynamic network alterations in individuals with current or past MDD in the same study (Yang et al., 2021a). Although useful, “static” RSFC analyses cannot capture the dynamic fluctuations of MDD-related brain functional abnormalities, especially with regards to cross-network interaction and maladaptive dynamics of particular network relationships.

To address this limitation, recently developed dynamic functional network (dFN) analyses capture fluctuating functional coordination patterns as networks form and dissolve as well as changes in the spatial organization of transient network patterns over time (Bray et al., 2015; Hutchison et al., 2013). Differing from static-based resting state network approaches (e.g. seed-based functional connectivity) that examine the average functional connectivity of large-scale brain networks across an entire resting state scan, growing evidence suggests that resting brain activity is not stable during scanning, but slowly wanders through a series of time-varying, reoccurring states of brain region coupling (Bolton et al., 2020a; Cabral et al., 2017), yielding various metrics that capture such dynamic properties (Calhoun et al., 2014). Among the dFN methods, co-activation pattern (CAP) analysis is a novel, data-driven frame-wise level approach that enables the identification of states of strong interregional co-activation across time series and the classification of each timepoint into one of those states using a clustering algorithm (Chen et al., 2015; Liu and Duyn, 2013). Compared to the widely used sliding window dFN approach, CAP analyses avoid the potential confounds that may be introduced by cleaving data into arbitrary sliding windows and sidesteps the compromise between statistical power and specificity (Chen et al., 2015). Thus, by examining the spatial or temporal features of extracted network states, CAP analyses provide a window on time-varying changes in exact spatial patterns of resting-state network (Liu et al., 2018).

Disruptions in resting-state dFN captured by diverse approaches (in particular CAP analyses) have been associated with depressive disorders, clinical symptoms, and behavioral and cognitive performance. Recent findings suggest that aberrant dFN patterns related to depression involve the DMN (Kaiser et al., 2022; Sendi et al., 2021; Wise et al., 2017), FPN (Demirtaş et al., 2016; Kaiser et al., 2019; Wang et al., 2020a), and sensorimotor network (SMN) (Wang et al., 2020b). Using various CAP metrics, depressive symptomatology and rumination in adolescents were shown to be associated with more time spent in a hybrid frontal insular default configuration and increased frequency of transition between a hybrid FPN/DMN and typical DMN state (Kaiser et al., 2019). In another study conducted in community sample of adults, elevated depressive symptom levels were associated with increased DMN state frequency and dwell time and less frequent entries into the hybrid FPN/DMN network (Goodman et al., 2021). Thus, CAP analyses offer a powerful means for characterizing altered dynamic resting-state network patterns over time, and can provide new insight into depression pathophysiology.

Previous traditional FC analyses in both remitted and current MDD point to state-independent (trait-like) effects of MDD on salience/attention (Ji et al., 2021; Servaas et al., 2017), and limbic (Dong et al., 2019a; Jacobs et al., 2016), or cognitive control systems (Dong et al., 2019b; Jiao et al., 2020; Liu et al., 2021; Stange et al., 2017; Zhang et al., 2022). However, most recent dFN approaches including CAP analyses focused on single depressive groups – such as remitted MDD (Figueroa et al., 2019) or different active depressive group (Goodman et al., 2021; Kaiser et al., 2019; Wang et al., 2020b). Neglecting the relationships between brain dFN alternations and clinical states (episode and remission) prevents distinguishing of state-independent and -dependent effect of depression on dynamic network pattern configurations, and thus identifying potential neural markers of MDD risk. Therefore, a systematic comparison of cMDD, rMDD, and healthy subjects in their dynamic resting-state functional network with CAP analyses is needed.

In the current study, we collected resting-state functional magnetic resonance imaging (fMRI) data from medication-naïve individuals with first-episode cMDD, unmedicated and medication-naïve individuals with rMDD, and healthy controls (HCs). Using a k-means clustering approach, we extracted recurring network states and examined group differences in dynamic CAP metrics to probe putative trait- and state-like temporal variations in resting-state brain network abnormalities. We hypothesized that increased time spent and entries in network states implicating the DMN and decreased dominance of network state involving FPN would be linked to current MDD, whereas state-independent dynamic network characteristics would be associated with imbalance among intrinsic brain networks, including alternations of network states transitions (e.g., FPN-to-DMN) (Kaiser et al., 2019; Scalabrini et al., 2020; Wang et al., 2016). Potential group differences were further evaluated for possible relationships with cognitive vulnerability factors, such as trait rumination, which has been previously linked to hyperconnectivity within the DMN (Tozzi et al., 2021; Zhou et al., 2020).