Whole-Brain Functional Dynamics Track Depressive Symptom Severity

Abstract

Depressive symptoms are reported by 20% of the population and are related to altered functional integrity of large-scale brain networks. The link between moment-to-moment brain function and depressive symptomatology, and the implications of these relationships for clinical and community populations alike, remain understudied. The present study examined relationships between functional brain dynamics and subclinical-to-mild depressive symptomatology in a large community sample of adults with and without psychiatric diagnoses. This study used data made available through the Enhanced Nathan Kline Institute-Rockland Sample; 445 participants between 18 and 65 years of age completed a 10-min resting-state functional MRI scan. Coactivation pattern analysis was used to examine the dimensional relationship between depressive symptoms and whole-brain states. Elevated levels of depressive symptoms were associated with increased frequency and dwell time of the default mode network, a brain network associated with self-referential thought, evaluative judgment, and social cognition. Furthermore, increased depressive symptom severity was associated with less frequent occurrences of a hybrid brain network implicated in cognitive control and goal-directed behavior, which may impair the inhibition of negative thinking patterns in depressed individuals. These findings demonstrate how temporally dynamic techniques offer novel insights into time-varying neural processes underlying subclinical and clinically meaningful depressive symptomatology.

Introduction

Major depressive disorder (MDD) is a leading cause of functional disability worldwide (Vos et al. 2017). In the United States of America, MDD remains a persistent public health concern, with 10.4% of people experiencing depression every year and 20.6% experiencing depression at some point in their life (Hasin et al. 2018). An estimated 20% of people experience a significant degree of current depressive symptoms, many of whom do not receive a formal diagnosis and consequently remain untreated (Shim et al. 2011). The prevalence of depression in the United States of America has risen in recent years, with rates increasing faster for young and elderly adults, and those of lower socioeconomic status (Weinberger et al. 2018).

Clinical scientists have advocated for a shift to investigating the dimensional nature of psychopathology rather than classifying complex psychiatric phenomena into discrete diagnostic categories (Cuthbert and Insel 2013). This perspective is a reaction to evidence that comorbidities occur in more than half of individuals with one psychiatric condition (Kessler et al. 1994) and psychiatric symptoms are often nonspecific to a particular disorder (Caspi et al. 2014). Moreover, there has been a lack of alignment between diagnostic categories and neuroscientific research findings (Henderson et al. 2020). To address these issues, the National Institute of Mental Health launched the Research Domain Criteria (RDoC) initiative, emphasizing that psychiatric research should shift to focus on the continuums on which symptoms occur as well as their underlying neurobiological correlates (Cuthbert and Insel 2013; Insel 2014). Neuroscientific research from such a dimensional perspective is necessary to understand neurocircuitry underlying depressive symptoms broadly, both in healthy individuals and across diagnostic classifications.

Traditionally, research on the neural correlates of depression has utilized case–control designs in which group differences in neural circuits are compared across healthy controls and clinically depressed patients. This perspective has uncovered significant deviations in the neurocircuitry of depressed individuals, including altered functional connectivity of large-scale brain networks (Kaiser et al. 2015; Fischer et al. 2016). The default mode (DMN) (Raichle et al. 2001) or medial frontoparietal network (M-FPN) has key nodes in medial prefrontal and posterior cingulate cortices and is involved in internally directed attention and self-referential thought (Uddin et al. 2007). This network is frequently implicated in depressive disorders and associated symptomatology such as rumination (Berman et al. 2011; Zhou et al. 2020). Within-network hyperconnectivity of DMN nodes is associated with depression across the lifespan (Alexopoulos et al. 2012; Zhou et al. 2020). Additionally, increased connectivity between the DMN, subgenual prefrontal cortex (PFC), and thalamic regions appears to underlie clinical MDD (Greicius et al. 2007; Hamilton et al. 2015). Hypoconnectivity of the lateral FPN, involved in executive functions and emotion regulation, has also been observed in depressed individuals (Kaiser et al. 2015; Fischer et al. 2016). Moreover, DMN hyperconnectivity may disrupt the functioning of other brain networks, contributing to pathology (Fischer et al. 2016). Although relationships between MDD, abnormal functional connectivity, and network dysregulation are well established in the literature, past research has emphasized case–control comparisons in the study of depression rather than the continuum on which depression occurs. Furthermore, comparisons between healthy individuals and those with MDD may exclude subclinical and mild depressive symptoms, which affect a significant portion of the population (Shim et al. 2011). Consequently, further neuroscientific research, which investigates depressive symptoms continuously and in a community sample of adults, is warranted.

Many of these past studies rely on static assessments of functional connectivity and compute average patterns of interregional connectivity over time. Compared with traditional, static analyses of resting-state fMRI data, time-varying, or dynamic analyses have allowed for more nuanced investigations of brain network activity. Whereas static analyses assume that the relationship between neural regions remains constant over time, dynamic analyses explore temporal variations in these relationships (Hutchison et al. 2013). In sliding window dynamic connectivity analyses, an arbitrary fixed window length is typically determined prior to analyses. Functional connectivity matrices are then calculated for observations within that window, the window slides along the timeseries, and states are clustered based on the matrix calculated from each window (Allen et al. 2014). In this sliding window approach, longer window lengths increase statistical power and produce more stable functional connectivity relationships, but smaller fluctuations in the data may be obscured. Coactivation pattern (CAP) analysis is an approach for examining brain dynamics that identifies states of strong interregional coactivation across the time series and classifies each time point into one of those states (Liu and Duyn 2013; Chen et al. 2015). By clustering every observation in the time series, CAP sidesteps the compromise between statistical power and specificity inherent in sliding window analyses, the most common method of dynamic connectivity (Chen et al. 2015).

Despite this methodological innovation, early studies of dynamic functional connectivity and depressive symptom severity have produced ambiguous results. One study revealed differences in static functional connectivity based on depressive presentation but found no temporal fluctuations, suggesting that connectivity between brain regions remains stable irrespective of symptom severity (Maglanoc et al. 2019). Other studies have demonstrated differing dynamic patterns as a function of depressive symptoms. While overall brain connectivity was similar in healthy and depressed individuals, depressed individuals spent more time in a loosely connected DMN configuration, and dynamic network dysfunction was associated with symptom severity (Zhi et al. 2018). Similarly, depressed individuals demonstrated decreased variability of DMN connectivity, supporting evidence that MDD is characterized by hyperconnectivity and enhanced stability of the DMN (Kaiser et al. 2016). More variability in the connectivity of frontal regions was associated with greater severity within depressed individuals (Kaiser et al. 2016). Using CAP analysis, it was shown that adolescent depressive symptomatology, irrespective of diagnosis, is associated with greater time spent in a hybrid FPN/DMN configuration and not time spent in a prototypical DMN state (Kaiser et al. 2019).

Taken together, many of the previous studies investigating both static and dynamic neural correlates of depressive disorders and symptomatology have relied on case–control designs, often in small sample sizes. Few studies have implemented the dimensional perspective of psychopathology emphasized by RDoC to characterize brain–behavior relationships underlying depressive symptom severity in adequately powered samples. Therefore, the current study sought to address this gap in the literature by applying a categorical-dimensional approach (Nomi 2019). Through this perspective, continuous relationships between depressive symptomatology and underlying brain relationships are preserved while also allowing for changes to occur at meaningful points in the depressive symptom continuum. This study applied CAP analysis to a large sample of participants (n = 445) from the Nathan Kline Institute’s publicly available database (Nooner et al. 2012) to examine the patterns of association between dynamic CAPs and depressive symptom severity in a community sample of adults with and without current psychiatric diagnoses.

Materials and Methods

Participants

A sample of 801 unrelated adults was drawn from the Enhanced Nathan Kline Institute’s Rockland Sample (Nooner et al. 2012). Participants completed multiple self-report measures, cognitive tasks, and a structured clinical interview (Structured Clinical Interview for DSM-IV-TR Axis I Disorders—Non-Patient Edition [SCID-I]) over 1 or 2 days. Participants also completed a resting-state fMRI scan during the visit. Each participant’s scan was evaluated by trained researchers for imaging artifacts; 70 subjects were removed, as their average head motion exceeded a framewise displacement of 0.50 mm (Power et al. 2012). The final sample of participants with clinical and demographic data available included 445 participants (65% female) ranging from 18 to 65 years of age. Demographics and descriptive statistics are reported in Table 1; 19% of the sample met criteria for at least one current psychiatric diagnosis (see Supplementary Tables 1 and 2 for diagnosis and medication information).

Table 1.

Demographics for the final sample (n = 445)

	M	(SD)
Age	37.58	(14.19)
BDI-II	6.21	(7.03)
	n	(%)
Female	288	(64.7)
Left-handed	33	(7.4)
Hispanic/Latinx	71	(16.0)
Ethnicity
White	302	(67.9)
Black/African American	82	(18.4)
Asian American/Pacific Islander	30	(6.7)
Other	19	(4.3)

Table 2.

Regression coefficients for regressions regressing CAP on depressive symptoms (Schaefer Parcellation)

Depressive symptoms	Dwell1				Dwell2			Dwell3			Dwell4			Dwell5
	b	SE	β	b	SE	β	b	SE	β	b	SE	β	b	SE	β
BDI-II < 14	−0.011	0.008	−0.07	0.002	0.008	0.02	−0.020*	0.008	−0.12	0.019*	0.008	0.12	0.008	0.009	0.04
BDI-II ≥ 14	0.019	0.013	0.08	−0.006	0.013	−0.03	0.027*	0.013	0.11	−0.028*	0.012	−0.12	−0.031*	0.014	−0.12
	Freq1				Freq2			Freq3			Freq4			Freq5
	b	SE	β	b	SE	β	b	SE	β	b	SE	β	b	SE	β
BDI-II < 14	−0.001	<0.001	−0.08	<0.001	<0.001	0.01	−0.001*	<0.001	−0.11	0.001*	<0.001	0.10	0.001	<0.001	0.08
BDI-II ≥ 14	0.001*	<0.001	0.12	<0.001	<0.001	0.01	0.002*	<0.001	0.13	−0.002*	0.001	−0.11	−0.002*	<0.001	−0.13

Note. Betas (β) indicate the standardized regression weight. ^*P < 0.05.

Image Acquisition

One 10-min resting-state fMRI scan collected from each participant was used in our analysis. Imaging was performed on a Siemens Trio 3.0 T scanner that collected a T₁ anatomical image and multiband (factor of 4) EPI sequenced resting-state fMRI data (2³mm, 40 interleaved slices, TR = 1.40s, TE = 30 ms, flip angle = 65°, FOV = 224 mm, 404 volumes). Participants were instructed to keep their eyes open and fixate on a cross centered on the screen (http://fcon_1000.projects.nitrc.org/indi/enhanced/mri_protocol.html).

Image Preprocessing

Resting state fMRI data were preprocessed using FSL, AFNI, and SPM functions through DPARSF-A in DPABI (Yan et al. 2016). The first five images were removed to allow the MRI signal to reach equilibrium. Data were despiked using AFNI 3dDespike, realigned and normalized with DPARSF-A into 3-mm MNI space, and then smoothed to 6 mm with AFNI 3dBlur. The ICA-FIX classifier was trained on hand-classified independent components separated into noise and nonnoise categories on the data from 24 subjects (random sampling by choosing subjects separated by ~10 years of age) by visually identifying noise and nonnoise independent components from 24 subjects randomly chosen across the lifespan (Griffanti et al. 2014; Salimi-Khorshidi et al. 2014). The resulting component classifications were then fed into FMIRB’s ICA-FIX classification algorithm (Griffanti et al. 2014) to classify noise and nonnoise components from individual subject data before conducting nuisance regression of classified noise components from the resting-state scans in MNI space. Next, the Friston 24 motion parameters (Friston et al. 1996) and linear trends were regressed out of the data, before the application of a band-pass filter (0.01-0.10 Hz) to isolate low-frequency fluctuations that characterize resting-state BOLD signals (Damoiseaux et al. 2006).

Whole Brain Parcellation and Region-of-Interest Selection

To ensure robustness of the results across multiple brain parcellation schemes, three independent whole-brain parcellations were applied to the data to derive regions of interest (ROIs): two parcellations developed for resting-state functional connectivity analyses (Gordon et al. 2016; Schaefer et al. 2018) and a parcellation previously utilized by Kaiser and colleagues (Kaiser et al. 2019), which included a 17-network functional parcellation of the cortex (Yeo et al. 2011) and striatum (Choi et al. 2012), and an anatomical atlas of the amygdala (AAL; Tzourio-Mazoyer et al. 2002). ROIs are displayed in Supplementary Figure 1.

Dynamic CAP Analyses

CAPs were calculated via k-means clustering on the concatenated time series of all subjects and all ROIs separately for each of the three parcellations (Fig. 1). CAPs derived from this technique represent the periods of shared activation of ROIs across the resting-state scan. More specifically, k-means clustering identifies a specified number of discrete, recurring activity patterns across all ROIs. Each timepoint is then classified as one of the identified clusters based on how similar the activation of all ROIs is at that timepoint to the centroid of each cluster. Testing values of k = 2–20, the optimal value of k = 5 was determined using the elbow criterion by applying a least-squares fit line to the cluster validity index (Allen et al. 2014; Nomi et al. 2016; Nomi et al. 2017), defined as the ratio of within-cluster to between-cluster differences (Supplementary Fig. 3). To further ensure robustness of the results, analyses were conducted using the 5-cluster solution in the Schaefer parcellation (Schaefer et al. 2018) and replicated in 4- and 6-cluster solutions (Supplementary Figs 5 and 6). CAPs of the 5-cluster solution in the Schaefer parcellation (Schaefer et al. 2018) are displayed in Figure 2. Three subject-level metrics of CAPs were derived from each solution: dwell time, state frequency, and transitions (Hutchison et al. 2013). Dwell time indicates the average consecutive timepoints an individual resided in each state. Frequency indicates the percentage of time each brain state occurred throughout the scan. Transitions represent the total number of switches between brain states across the scan. As each of the CAP metrics are inherently intercorrelated, CAP metrics were investigated as dependent variables to circumvent concerns resulting from excessive collinearity.

Figure 1 .

CAP conceptual schematic. The top panel (a) demonstrates the clustering of timepoints from a given ROI into each brain configuration. The bottom panel (b) illustrates the calculation of “dwell time” and “frequency” for each brain state, as well as “transitions” between states (brain activation images visualized with “BrainNet Viewer”; Xia et al. 2013).

Figure 2 .

CAP states derived from k-means clustering (k = 5). Percentages indicate the average amount of time spent in each state across all participants.

Clinical Measures

The Beck Depression Inventory—II (BDI-II; Beck et al. 1996) is a self-report assessment of depressive symptoms. Participants respond to 21 items assessing affective, behavioral, cognitive, and physiological symptoms commonly experienced in MDD. Responses were coded on a four-point scale, with higher scores indicating greater symptomatology. Items demonstrated excellent interitem reliability (α = 0.91) and were summed to create total scores. Total scores ranged from 0 to 34 (M = 6.21, SD = 7.03, Median = 4.00; see Supplementary Fig. 2). Based on the originally proposed ranges of depressive symptom severity (Beck et al. 1996), 386 participants (86.7%) were not depressed, 33 participants (7.4%) were mildly depressed, 18 participants (4.0%) were moderately depressed, and 8 participants (1.8%) were severely depressed.

Statistical Analyses

Regression was used to regress state dwell time and frequency, as well as overall transitions, on depressive symptoms while controlling for motion, age, gender, and psychotropic medication use (0 = No; 1 = Yes; see Supplementary Table 2). As the relationship between symptoms and CAP metrics were revealed to be nonlinear, piecewise regression was used. Cut points at ~14 have demonstrated optimal sensitivity and specificity for identifying depressed individuals (Beck et al. 1996; von Glischinski et al. 2019); therefore, the knot point of each regression was set at BDI-II scores of 14.

Results

Cluster Analysis Yields Five Recurring CAPs

The CAP states derived from clustering were consistent across all three whole-brain parcellations (Fig. 2 and Supplementary Fig. 4), as were the brain regions involved in each state. Slight variations in the frequency of states and the intensity of brain regions emerged across parcellations, with the Schaefer parcellation (Schaefer et al. 2018) demonstrating the clearest delineations between states.

CAP topography

State 1 was characterized by the coactivation of the medial PFC, posterior cingulate cortex, inferior parietal lobule, middle temporal gyrus, and superior and inferior temporal sulci. The superior parietal lobule and precuneus were active across regions, although it was most prominent in the Kaiser parcellation (Kaiser et al. 2019). This aligns well with regions of the M-FPN, or DMN, which is involved in updating associations, self-referential thought, and social cognition (Uddin et al. 2019).

State 2 included the lateral and medial PFC and lateral parietal cortex. Specifically, the middle frontal gyrus, inferior frontal sulcus, inferior parietal lobule, and medial superior frontal gyrus demonstrated strong coactivation in this state. This aligns closely with the lateral frontoparietal network (L-FPN), often also referred to as a cognitive control or central executive network, and is most typically implicated in executive function and goal-oriented activity (Menon 2011).

State 3 involved the coactivation of the lateral and medial occipital lobe broadly and clearly represented a visual network (VN).

State 4 resembled a pericentral network (PN) involved in motor processes as well as somatosensory and auditory processing, which included precentral and postcentral gyri, the central sulcus, the juxtapositional lobule, and the superior temporal gyrus.

State 5 included notable coactivation across subregions of the prefrontal, temporal, and parietal lobes. The most active areas included the lateral and medial portions of the dorsal and orbital PFC, inferior parietal lobule, the posterior cingulate, middle and inferior temporal gyri, and the inferior temporal sulcus. This state appears to represent a hybrid of L-FPN and DMN states, which is a consistent finding with previous CAP analyses (Kaiser et al. 2019).

CAP Topography and Depressive Symptoms

The dwell time and frequency of each state were regressed on depressive symptoms. The zero-order correlations between depressive symptoms and head motion (r = −0.01, P = 0.856), age (r = −0.01, P = 0.766), sex (r < −0.01, P = 0.964), and psychotropic medication use (r = 0.08, P = 0.095) were all nonsignificant. The relationship between each CAP metric and symptom severity was notably nonlinear (Fig. 3); the directionality of each relationship appears to differ at low versus high depressive symptoms, with inflection points at moderate symptom severity. Consequently, piecewise regression with a knot-point at a validated threshold which best discriminates between healthy and depressed individuals (BDI-II ≥ 14) was used to examine nonlinear relationships between symptom severity and CAP metrics. This knot-point fit is consistent with past psychometric analyses (Beck et al. 1996; von Glischinski et al. 2019). Several nonlinear relationships were consistent across all three parcellations (see Supplementary materials and Supplemental Tables 3–11). States with significant effects are displayed in Figure 3. There were no significant relationships between state transitions and depressive symptoms (P’s ≥ 0.532), regardless of parcellation. Model assumptions were met, and residuals were normally distributed; density plots of residuals are displayed in the supplementary materials (Supplementary Fig. 7).

Figure 3 .

Piecewise regressions predicting dwell time and frequency from depressive symptoms. Blue lines indicate slopes for participants with minimal depressive symptoms below the knot of 14; red lines indicate slopes for participants with elevated depressive symptoms above the knot of 14. ^*P < 0.05.

State 1: DMN coactivation

The slope for those with elevated depressive symptoms above the knot-point was significant and positively associated with greater frequency of the DMN state, indicating that for individuals with significant depressive symptoms, more frequent occurrences of the DMN were associated with greater symptom severity. There was no relationship between symptom severity and DMN CAP metrics for individuals with low symptomatology.

State 2: L-FPN coactivation

There was no relationship between CAP metrics of the L-FPN state at either minimal or elevated depressive symptoms.

State 3: VN coactivation

With respect to the VN, the slope of depressive symptoms for dwell time and frequency was negative at low symptomatology but positive at high symptomatology; severity was associated with less time in the VN until symptoms reached an elevated degree, at which point depressive symptoms were associated with more time in the VN state.

State 4: PN coactivation

Dwell time and frequency of the PN state were significantly related to both minimal and elevated depressive symptoms. At low symptom severity, greater levels of depressive symptoms were associated with increases in PN dwell time and frequency, whereas elevated symptoms were associated with less dwell time and frequency.

State 5: L-FPN/DMN hybrid state coactivation

The slope for individuals with elevated depressive symptoms was negative for both dwell time and frequency of the hybrid L-FPN/DMN state. At elevated depressive symptoms, higher symptom severity was associated with decreased dwell time and frequency of the L-FPN/DMN state.

Discussion

This study investigated the temporal patterns of brain activity underlying subclinical and mild depressive symptomatology in a community sample of adult participants. Whereas static functional connectivity analyses are useful in elucidating relationships in overall connectivity between brain regions across a scan, dynamic analyses allow for investigating more nuanced changes in brain function, which can both compliment and compete with results reported in the static functional connectivity literature (Hutchison et al. 2013). Dynamic metrics and their relation to psychiatric symptom severity may be overlooked by static approaches, which aggregate brain activity across the scan. Consequently, dynamic approaches may offer a more nuanced representation of the temporal variability between neural regions. Moreover, piecewise regression at diagnostically informed knot-points permits the evaluation of continuous relationships while recognizing CAP state changes aligned with categorical classifications of depression.

Results revealed significant differences in CAP state organization as a function of depressive symptomatology, with notable changes occurring in individuals exhibiting an elevated degree of symptom severity (Beck et al. 1996; von Glischinski et al. 2019). Generally, the relationship between symptom severity and each of the CAP metrics was in opposing directionality for more versus less depressed individuals. This broad pattern suggests that notable neurobiological changes occur in more depressed individuals and may provide insight to neural patterns underlying illness trajectories and treatments. Additionally, this study compliments recent advancements in clinical neuroimaging research which advocate for a combined categorical-dimensional perspective (Nomi 2019), providing a more nuanced investigation into neuropsychiatric processes.

At low symptom severity, dwell time in a DMN state was unrelated to depressive symptoms. The DMN is active during normative cognitive processes, including autobiographical memory (Spreng and Grady 2010), spontaneous cognitions (Andrews-Hanna et al. 2014), reflection and prospection (Szpunar et al. 2014), and navigating social interactions (Buckner et al. 2008). This study affirms that time spent in a DMN configuration is not exclusively indicative of depressive symptomatology; however, this state becomes more dominant in those experiencing elevated depressive symptoms. This suggests that individuals with greater symptom severity may be stuck in a repetitive ruminative process underwritten by overactivity of the DMN (Berman et al. 2011; Burrows et al. 2017; Zhou et al. 2020), and is consistent with the wealth of static functional connectivity research implicating dysregulation of the DMN as underlying depressive symptomatology (Greicius et al. 2007; Hamilton et al. 2015). Moreover, these results serve as a compliment to static analyses by demonstrating DMN dysregulation is not only limited to increased connectivity between nodes, but also increased time spent in a DMN state.

Elevated depressive symptom severity was associated with decreased time spent in a hybrid L-FPN/DMN state. The L-FPN is most typically implicated in executive functions including working memory, goal-directed cognition, and cognitive control. Furthermore, the L-FPN works in tandem with the DMN specifically during tasks of autobiographical planning of real-world goals (Spreng et al. 2010), a process which is impaired during depressive episodes. Greater dynamic connectivity between nodes of the DMN and L-FPN has previously been associated with more severe symptomatology in depressed individuals (Kaiser et al. 2016). Furthermore, a hybrid L-FPN/DMN state was observed in past CAP analyses of adolescent depression; however, that study found that increased dominance and persistence were associated with greater symptom severity (Kaiser et al. 2019). The present study relied on a community sample, largely experiencing subclinical depressive symptoms, whereas past research focused on more severely depressed individuals. These sample characteristics may help to explain discrepancies in the directionality of relationships between brain states and symptomatology. Given static connectivity has identified these brain regions in studies of both cognitive control and evaluative thought, we propose that decreased dominance of this L-FPN/DMN state may result in an impaired ability to reign in negative, maladaptive thoughts. This is consistent with cognitive theories positing depression is characterized by uncontrollable negative thinking (Clark and Beck 2010; Beck and Bredemeier 2016) and top-down, attentional and cognitive control theories supported by neural circuitry in the dorsolateral PFC (Fales et al. 2008; Sheline et al. 2009; Grahek et al. 2019). Furthermore, we are unaware of static functional connectivity analyses which have revealed such a hybrid network with respect to depressive symptomatology, suggesting dynamic analyses may provide unique insights into brain correlates of depression.

Dwell time and frequency of the VN state were positively associated with elevated depressive symptoms, whereas the opposite was found for the PN state. Although infrequently researched, preliminary research found disruptions in resting-state visual and sensory network functional connectivity in depressed adults (Eyre et al. 2016; Zhang et al. 2020). Furthermore, disruptions to functional sensory networks may underlie regulation difficulties in psychiatric disorders (Wu et al. 2017); however, prior research in the integration of sensory networks and depressive symptoms appears limited to case–control designs. Moreover, relationships between PN and VN state metrics and depressive symptoms were less robust to replication across parcellations. Further research is warranted to parse the roles of the VN and PN in psychiatric symptomatology.

Further longitudinal research is necessary to understand the interplay between brain state processes in the brief period during which they are measured and more long-term depressive symptom expressions, which may last for weeks or years. These findings have important implications for understanding the neurobiological mechanisms underlying depressive symptomatology and advancing our understanding of psychiatric illnesses as disorders of the brain networks (Insel and Quirion 2005; Ross et al. 2015), and serve as a steppingstone to understanding longitudinal relationships between brain functioning and psychiatric conditions. Unifications between psychiatric and neuroscientific research can inform the etiology and maintenance of psychiatric conditions and have led to the development of more efficacious treatment through both psychopharmacological and psychotherapeutic interventions (Reynolds et al. 2009; Arbuckle et al. 2017; Ross et al. 2017). Additional research should also consider brain dynamics underlying more specific sequalae of depressive disorders, such as rumination and executive function deficits. Furthermore, given the transdiagnostic presentation of depressive symptoms across psychiatric disorders, further research on the relationship between brain states and broad psychiatric distress is warranted.

Strengths and Limitations

To our knowledge, this study is the first to use dynamic CAP analysis to dimensionally explore depressive symptom severity in a community-based sample of adults. This approach allows for both better applicability of findings to the public as well as generalizability across psychiatric conditions, as these results are not limited to individuals meeting criteria for MDD. CAP analysis may provide a unique technique to understanding changes in brain activity in response to psychopharmacologic and psychotherapeutic treatments.

One limitation of note is that this study relies on a community sample who are largely experiencing subclinical depressive symptoms, and much of the variance in depressive symptoms is contributed by a subset of more severely depressed individuals. While investigations of subclinical depressive symptoms are important, as up to 20% of individuals experience such symptoms (Shim et al. 2011), the relationships between brain states and symptom severity may change in more severely depressed individuals. A second potential limitation of the study is the use and interpretation of a cut point delineating between healthy and depressed individuals. Utilizing cut scores to group persons into diagnostic categories may result in separating individuals with scores close together into differing groups; however, relying on cut scores determined from appropriate psychometric analyses (von Glischinski et al. 2019) allows for a combined categorical-dimensional approach (Nomi 2019), which may provide more insights into neurobiological processes underlying psychiatric conditions than purely categorical or dimensional perspectives otherwise could. Lastly, the neural mechanisms underlying functional connectivity, static or dynamic, remain ambiguous (Reid et al. 2019). Future directions for the field include the incorporation of computational models to further elucidate the neural mechanisms contributing to empirical findings in the dynamic resting-state fMRI analysis literature (Deco et al. 2008; Heitmann and Breakspear 2018).

Conclusions

Using temporally dynamic neuroimaging techniques and a combined continuous-categorical perspective of psychopathology, this study elucidated several brain states which relate to the degree to which individuals experience depressive symptoms, irrespective of clinical diagnoses. These findings lay a foundation for understanding how neurobiological processes over time contribute to psychiatric conditions and symptomatology, and the extent to which these neurobiological changes align with treatment theory. Finally, this study may allow for better treatment of individuals, as continued research investigates techniques to target brain network dysfunction directly.

Notes

Conflicts of Interest: The authors declare no competing financial interests.

Funding

National Institutes of Health (T32-HL007426 to Z.T.G.); National Institute of Mental Health (R03MH121668), NARSAD Young Investigator Award to J.S.N.; National Institute of Mental Health (R01MH107549), Canadian Institute for Advanced Research, University of Miami Gabelli Senior Scholar Award to L.Q.U.