Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Nov 1.
Published in final edited form as: Behav Res Ther. 2021 Sep 2;146:103960. doi: 10.1016/j.brat.2021.103960

Resting State Functional Connectivity Subtypes Predict Discrete Patterns of Cognitive-Affective Functioning across Levels of Analysis among Patients with Treatment-Resistant Depression

Mary L Woody 1, Benjamin Panny 1, Michelle Degutis 2, Angela Griffo 1, Rebecca B Price 1,3
PMCID: PMC8653528  NIHMSID: NIHMS1760407  PMID: 34488187

Abstract

Resting state functional connectivity (RSFC) in ventral affective (VAN), default mode (DMN) and cognitive control (CCN) networks may partially underlie heterogeneity in depression. The current study used data-driven parsing of RSFC to identify subgroups of patients with treatment-resistant depression (TRD; n=70) and determine if subgroups generalized to transdiagnostic measures of cognitive-affective functioning relevant to depression (indexed across self-report, behavioral, and molecular levels of analysis). RSFC paths within key networks were characterized using Subgroup-Group Iterative Multiple Model Estimation. Three connectivity-based subgroups emerged: Subgroup A, the largest subset and containing the fewest pathways; Subgroup B, containing unique bidirectional VAN/DMN negative feedback; and Subgroup C, containing the most pathways. Compared to other subgroups, subgroup B was characterized by lower self-reported positive affect and subgroup C by higher self-reported positive affect, greater variability in induced positive affect, worse response inhibition, and reduced striatal tissue iron concentration. RSFC-based categorization revealed three TRD subtypes associated with discrete aberrations in transdiagnostic cognitive-affective functioning that were largely unified across levels of analysis and were maintained after accounting for the variability captured by a disorder-specific measure of depressive symptoms. Findings advance understanding of transdiagnostic brain-behavior heterogeneity in TRD and may inform novel treatment targets for this population.

Keywords: fMRI, person-specific resting state functional connectivity, community detection, transdiagnostic positive valence systems, transdiagnostic negative valence systems, treatment-resistant depression

There is well-known heterogeneity in depressive disorders, as defined by the Diagnostic and Statistical Manual of Mental Disorders (DSM) (APA, 2013). Thus, a growing movement within psychiatry seeks to parse neurobiological substrates in order to predict variability in discrete features of depressed mood (Beijers, Wardenaar, van Loo, & Schoevers, 2019; Lynch, Gunning, & Liston, 2020). These efforts may transform our understanding of brain-behavior relations that transcend diagnostic boundaries and allow for the development of precision medicine interventions that elegantly convert individual heterogeneity from nuisance parameter to treatment predictor. Such an approach may illuminate person-specific interventions for treatment-resistant depression (TRD), which is critical given that an estimated 50% of patients with this condition do not respond to currently available treatments (Rush, 2007) and may present with distinct neurobiological patterns compared to their treatment-responsive peers (Akil et al., 2018).

Data-driven methods have emerged to classify neurobiological heterogeneity across individual patients (Beltz, Moser, Zhu, Burt, & Klump, 2018; Clementz et al., 2016; Karalunas et al., 2014; Yang et al., 2014). For example, a recent review of depression-focused studies demonstrated that data-driven subgroups of resting state functional connectivity (RSFC) can help classify clinical features such as internalizing diagnoses, symptom severity, and recurrence (Beijers et al., 2019). An important next step for this work is to link neurobiological subtypes within a diagnostic group to key markers of functioning that cut across diagnostic boundaries. Federal initiatives like the NIMH Research Domain Criteria (RDoC) and the BRAIN Initiative were created with the explicit goal of connecting classification of psychopathology to advances in genetics and neuroimaging, which often indicate core features or influences that cut across traditional diagnostic boundaries (Cuthbert & Insel, 2013; Litvina et al., 2019). Such transdiagnostic frameworks posit that aberrations in neural circuits may be a key component of parsing heterogeneity within clinical “constructs”, which are defined across multiple levels of analysis occurring within a dimensional matrix (Kozak & Cuthbert, 2016). For example, for patients reporting low levels of positive affect and high levels of negative affect, which is often true of depressed patients, brain-behavior measurements within both transdiagnostic positive and negative valence systems may be most appropriate (Williams, 2016; Woody & Gibb, 2015).

Neural circuits relevant to transdiagnostic positive and negative valence systems fall largely within three interconnected networks, which are also highlighted in depression research more broadly (Kaiser, Andrews-Hanna, Wager, & Pizzagalli, 2015; Williams, 2016): ventral affective [VAN; e.g., amygdala, insula, ventrolateral prefrontal cortex (VLPFC), subgenual anterior cingulate cortex (sgACC), nucleus accumbens], default mode [DMN; e.g., perigenual anterior cingulate cortex (pgACC), posterior cingulate cortex (PCC)], and cognitive control [CCN; e.g., dorsolateral prefrontal cortex (DLPFC), dorsal anterior cingulate cortex (dACC), posterior parietal cortex] networks. Given myriad potential connections that exist within and between these regions and networks, data-driven approaches are needed to parsimoniously identify connectivity-based subgroups that might predict discrete patterns of functioning in related levels of analysis (genes, molecules, cells, physiology, behavior, and self-report). RSFC is a strong candidate for classification as it provides an index of coordinated activation across brain regions in the absence of task demands and is thought to remain stable over time with high levels of individual specificity (Finn et al., 2015; Gratton et al., 2018). Meta-analyses indicate that depression is associated with RSFC alterations within and between VAN, DMN, and CCN circuits (Kaiser et al., 2015). However, connectivity findings are often inconsistent, likely due to the heterogeneity within depression and the lack of “one-size-fits-all” brain-behavior associations (Hasler & Northoff, 2011; Kaiser et al., 2015).

Future research is still necessary to determine whether RSFC subgroups, as identified through data-driven approaches, can be used to predict patterns of depression-relevant functioning as described within transdiagnostic positive and negative valence systems. Such an approach is well-suited to describe brain-behavior relations that are otherwise masked by case-control comparisons (e.g., comparing depressed patients to healthy controls). Further, because RSFC has been posited as a promising neurocognitive target for intervention among patients with TRD (Price & Duman, 2020), using data-driven approaches to understand translational links between RSFC and other levels of analysis within positive and negative systems would be informative for the development of novel interventions for TRD. Therefore in the current study, given the relevance to positive and negative valence systems and TRD, we classified RSFC in and across a network of 15 VAN, DMN, and CCN regions used in prior research by our group (Price et al., 2020; Price, Gates, Kraynak, Thase, & Siegle, 2017; Price, Lane, et al., 2017) and others (Chahal et al., 2020) using Subgroup Group Iterative Multiple Model estimate (S-GIMME) (Gates, Lane, Varangas, Giovanello, & Guiskewicz, 2017). For each patient, this approach can determine the presence and the direction of connectivity among pairs of regions (i.e., does A predict B after controlling for all other network-wide influences including B’s influence on itself?) with dramatically improved accuracy and power compared to more traditional connectivity metrics (Gates & Molenaar, 2012). Research validating S-GIMME shows it can reliably recover underlying subgroups even in small samples (n = 25 before subgrouping) and that subgroups remain stable even after randomly perturbing the similarity matrix of connectivity weights in order to simulate random fluctuations across individual data points (Gates et al., 2017; Lane, Gates, Pike, Beltz, & Wright, 2019).

We hypothesized that data-driven identification of RSFC subgroups derived from S-GIMME could be used to predict functioning across related levels of analysis (self-report, behavioral, and molecular) among patients seeking intervention for moderate-to-severe TRD (defined in the current study as patients with clinically elevated unipolar depressive symptoms who have experienced at least one failed course of FDA-approved antidepressant medication within the current episode). We chose to focus on assessments that captured transdiagnostic cognitive-affective functioning across multiple levels of analysis (self-report, behavioral, neural, and molecular), given that altered functional integration across VAN, DMN, and CCN networks is thought to be related both to ‘upstream’ neuronal/molecular alterations, as well as to ‘downstream’ rigid and inflexible cognitive-affective biases (e.g., affective information processing, self-reported affect) observed in depression (Price & Duman, 2020). Given the paucity of prior research examining links between RSFC heterogeneity and dimensional transdiagnostic measures in depression and our explicitly data-drive analytic approach, we chose to focus our a priori hypotheses on the likelihood of generalization from data-driven RSFC classification to relevant patterns of cognitive-affective functioning as captured by external measures, rather than predicting specific patterns of RSFC (e.g., DMN hypoconnectivity) that might be associated with performance on these measures. Finally, we chose to examine these processes in a sample of patients with TRD given the imperative need for novel interventions for this underserved clinical population (Rush, 2007) and because the treatment-resistant nature of their current depressive episode increased the probability of entrenched brain-behavior connections that could elucidate heterogeneous links between RSFC and other measures of cognitive-affective functioning (Akil et al., 2018).

At the self-report level of analysis, by its definition, depression is characterized by patient reports of intractable sadness and/or reductions in the experience of pleasure. Not surprisingly, this pattern is also borne out by case-control comparisons showing that depressed patients display chronically higher levels of negative affect and lower positive affect that often do not respond flexibly to changing environmental inputs (e.g., the absence of expected shifts in positive and negative mood when engaging in pleasant activities) (Rottenberg, 2017). However, even amongst chronically depressed patients, there is well-known heterogeneity in self-reported positive and negative affect, and research suggests that RSFC may partially underlie such variability (Beijers et al., 2019; Goldstein-Piekarski et al., 2020; Price, Lane, et al., 2017; Williams & Goldstein-Piekarski, 2020). Yet, even when taking a dimensional approach, most past research in this area has examined links between RSFC and depression-relevant affect using disorder-specific reports of depressive symptoms [e.g., Beck Depression Inventory (BDI); Hamilton Depression Rating Scale (HAMD)], which may fail to capture some aspects of cognitive-affective functioning that cut across transdiagnostic positive and negative valence systems. To address this gap in the current study, we examined whether RSFC-based subgroups would discriminate depressed patients’ levels of transdiagnostic positive affect and negative affect (i.e., sadness, anger, and anxiety), above and beyond the variability captured by a disorder-specific measure of depressive symptoms.

At the behavioral level, case-control comparisons have demonstrated that patients with depression display disruptions in performance-based measures of affective information processing, leading to rigid and inflexible biases in the way depressed individuals respond to and recall emotional stimuli (Price & Duman, 2020). Critically, these biases are linked to reductions in goal-directed behaviors crucial to adaptive functioning, in part because flexible control of responses elicited by emotional stimuli is necessary for emotion regulation strategies such as savoring a happy memory when feeling down or inhibiting responses to distracting emotional information in favor of task-relevant demands. Yet, although inflexible responses to both positive and negative affective stimuli are a well-replicated feature of depression (LeMoult & Gotlib, 2019), studies examining its neural substrates have been mixed, leading theorists to suggest that neurobiological subtyping may be a key next step in understanding the neural correlates of affective information processing within the context of depression (Grahek, Everaert, Krebs, & Koster, 2018). Therefore, we predicted that RSFC-based subgroups would capture differences in patient performance on computer tasks assessing affective information processing, including the ability to upregulate positive mood during a positive mood induction and to flexibly inhibit responses to emotional stimuli (sad, happy, and neutral facial expressions) during an affective Go/NoGo task.

Finally, at the molecular level, case-control comparisons show that depressed patients exhibit reduced tonic striatal dopamine activity (Dunlop & Nemeroff, 2007; Grace, 2016), which is related to reduced functional connectivity between VAN, DMN, and CCN networks and may partially underlie impairments in motivation and reward-driven behaviors that perpetuate inflexible responding (Dubol et al., 2018; Hamilton et al., 2018; Zhang et al., 2019). However, because traditional assessments of striatal dopamine activity in depressed patients utilize specialized procedures and equipment, dimensional investigations of the relation between RSFC and dopaminergic functioning in depression have been limited (Price et al., 2021). To address this gap, the current study capitalized on recently developed neuroimaging indices that use normalized T2*-weighted (nT2*w) imaging to estimate tissue iron concentration in the dopamine-rich striatum, which is made possible by the paramagnetic properties of tissue iron that disrupt the T2* signal and provide for a psychometrically reliable, readily attainable estimate of tissue iron concentration (Larsen & Luna, 2015). Previous research has demonstrated that striatal tissue iron concentration levels are strongly related to dopamine receptor D2 (D2R) expression, dopamine transporter (DAT) function, and dopaminergic neuron excitability in striatal regions (Beard & Connor, 2003; Erikson, Jones, & Beard, 2000; Haacke et al., 2005; Larsen & Luna, 2015) and that the nT2*w signals accords fairly well with more specialized measures of tissue iron (e.g., quantitative susceptibility mapping), thereby providing a putative fMRI-based proxy of dopaminergic functioning (Larsen, Olafsson, et al., 2020; Peterson et al., 2019). Critical to our conceptual model, striatal tissue iron can be characterized as a transdiagnostic biological substrate of flexible responding to dynamic environments, such that decreases in striatal tissue iron concentration are associated with reduced capacity to engage goal-directed responses and inhibit task-irrelevant ones (Adisetiyo et al., 2014; Larsen, Bourque, et al., 2020). Therefore, congruent with our hypotheses at other levels of analysis, we hypothesized that RSFC-based subgroups would predict patients’ levels of tissue iron concentration in the dopamine-rich dorsal (caudate, pallidum, putamen) and ventral (nucleus accumbens) striatum, given previously-established links between striatal dopamine activity and functional connectivity in depression (Dubol et al., 2018; Hamilton et al., 2018; Zhang et al., 2019).

Method

Participants

Patients were 70 adults with clinically elevated depressive symptoms [Montgomery-Åsberg Depression Rating Scale (MADRS) score ≥ 25 (Montgomery & Åsberg, 1977)] and one or more failed course of FDA-approved antidepressant medication within the current depressive episode (full inclusion/exclusion criteria and handling of missing data in Supplement). See Table 1 for demographic and clinical characteristics. This study was approved by the local Institutional Review Board and informed consent was obtained from all patients. The current analyses utilized baseline data from a larger, ongoing treatment study (clinicaltrials.gov: NCT03237286).

Table 1.

Demographic and Clinical Characteristics

Group (n=70) Subgroup A (n=34) Subgroup B (n=19) Subgroup C (n=17)
Age 36.70 (10.87) – – –
Median Household Income $58,000 – – –
Female 57% – – –
White 87% – – –
Black/African American 4% – – –
Asian American 4% – – –
Multiracial 4% – – –
Hispanic/Latinx 4% – – –
Current Use of Psychotropic Medication 81% – – –
MADRS 34.43 (4.91) – – –
Number of Treatment Failures 2.56 (1.56) – – –
Anxiety 65.65 (6.28) – – –
Depression 67.75 (7.04) – – –
Anger 56.96 (8.48) – – –
Positive Affect 39.77 (4.25) 39.87 (4.04) 37.49 (3.76) 42.13 (4.05)
PMI Average Affect 707.91 (133.25) – – –
PMI Affect Variability 92.26 (45.12) 80.42 (36.02) 93.51 (44.81) 113.84 (54.88)
GNG D-prime 1.58 (.68) 1.67 (.62) 1.75 (.75) 1.26 (.65)
nT2*w Caudate .85 (.07) .84 (.06) .84 (.06) .89 (08)
nT2*w Pallidum .57 (.05) .58 (.04) .54 (.05) .60 (.06)
nT2*w Putamen .68 (.06) .68 (.06) .66 (.06) .71 (.07)
nT2*w Nucleus Accumbens .63 (.08) .62 (.08) .62 (.06) .65 (.08)
Open in a new tab

Note. Subgroup means and standard deviations are only presented for variables that were significantly discriminated by S-GIMME subgroups; All individuals who identified as Hispanic/Latinx also identified as White; MADRS = Montgomery-Åsberg Depression Rating Scale; PMI = Positive Mood Induction; GNG = Go/NoGo Task; nT2*w = normalized T2*-Weighted Tissue Iron; Possible PMI affect ratings ranged on a scale from 140 to 880; Smaller GNG D-prime values correspond to worse response inhibition; Larger nT2*w values correspond to lower levels of tissue iron concentration.

Measures

fMRI methods.

Data were acquired during a 7-min eyes-open resting state block. T2*-weighted images depicting BOLD contrast were acquired on a 3Tesla Siemens PRISMA scanner using Human Connectome Project sequences (multi-band factor=8; TR=800ms; TE=37; flip angle=52°; 72 slices; FOV=200×200; 2mm isotropic voxels). Standard preprocessing steps were applied using Analysis of Functional Neuroimaging (AFNI) using the afni_proc.py pipeline. The following preprocessing steps were applied: slice time correction, 6-parameter motion correction, spatial distortion correction utilizing forward- and reverse-direction Spin Echo fieldmaps and AFNI’s “blip” step, cross-registration of functional data to a high-resolution structural scan acquired in the same fMRI session (axial MPRAGE: TR=2400; TE=2.22; 208 slices; flip angle=8°; 0.8mm isotropic voxels), 32-parameter nonlinear warping to the Montreal Neurological Institute Colin-27 brain data set, spatial smoothing [6-mm full width half maximum], scaling to percent change [this final step was applied prior to timeseries analyses, but was not applied for the nT2*w (tissue iron) analyses].

To provide timeseries data for connectivity analyses, fifteen ROIs were selected for consistency with prior literature using S-GIMME in depression. As described previously (Price, Gates, et al., 2017; Price, Lane, et al., 2017), network nodes were defined by combining anatomical masks based on standardized (MNI and Talairach) atlases. ROIs included regions from the ventral affective [VAN; e.g., left and right amygdala, left and right insula, left and right ventrolateral prefrontal cortex (VLPFC), subgenual anterior cingulate cortex (sgACC), left and right nucleus accumbens], default mode [DMN; e.g., perigenual anterior cingulate cortex (pgACC), posterior cingulate cortex (PCC; coordinates from (Greicius, Krasnow, Reiss, & Menon, 2003)], and cognitive control networks [CCN; e.g., left dorsolateral prefrontal cortex (DLPFC; functionally defined as in Siegle, Thompson, Carter, Steinhauer, & Thase, 2007), dorsal anterior cingulate cortex (dACC), left and right posterior parietal cortex (PPC; meta-analytic coordinates from Kaiser et al., 2015)]. Mean preprocessed time-series data were extracted for each participant for each ROI. Timepoints with incremental/rotational movement >=.5 mm or .5 degrees (on average 1.71% of data; no more than 7% of timepoints per patient) were marked as missing and skipped by the algorithm to protect against confounding motion-related connectivity. This approach is similar to “scrubbing” (i.e., deleting timepoints) but with the added benefit of maintaining temporal ordering of scans.

To calculate estimates of tissue iron concentration, following all preprocessing steps described above, normed T2*-weighted (nT2*w) signal values were calculated as described by prior research (Larsen & Luna, 2015; Vo et al., 2011). In brief, each TR was initially scaled to its own mean across the 7-min resting state block, which served to eliminate scan-specific noise and facilitate comparison of T2* values across participants. Next, the median of the normalized T2* signal was calculated across all TRs in the resting state block (525 TRs), voxel-wise, which created one nT2*w value per participant. Finally, using a priori striatal masks defined anatomically in AFNI via the MNI atlas, regional means were calculated bilaterally for the dorsal (caudate, pallidum, putamen) and ventral (nucleus accumbens) striatum.

S-GIMME directed connectivity and subgrouping.

S-GIMME employs unified structural equation models (Kim, Zhu, Chang, Bentler, & Ernst, 2007) and a Bayes net formulation. First, it detects whether there are any lagged or contemporaneous directed connections from one ROI to another that occur for ≥ 75% of the sample. Then, it creates subgroups by using the individual-level estimates of the group-level connections in addition to anticipated estimates for candidate connections and utilizes an “unsupervised” community detection algorithm (Walktrap). Any shared patterns are categorized based on their sign (positive/negative), significance (p<.05 vs. p>.05, after Bonferroni correction for the number of subjects), direction of influence (e.g., A predicts B), and temporal pattern (contemporaneous or lagged). If any subgroups are detected, S-GIMME identifies contemporaneous or lagged directed connections that occur for ≥ 50% of the subgroup. Finally, the program detects if there are any individual-level connections. All connections are detected based on LaGrange Multiplier Equivalents, which determine which connections (if added to a map) will maximize the model fit.

The GIMME approach has been validated using both simulated (Gates et al., 2017; Gates & Molenaar, 2012; Gates, Molenaar, Iyer, Nigg, & Fair, 2014; Lane, Gates, Pike, Beltz, & Wright, 2019) and empirical data (Nichols, Gates, Molenaar, & Wilson, 2014; Price et al., 2020; Price, Gates, et al., 2017; Price, Lane, et al., 2017). Large-scale simulations show that S-GIMME is a valid and reliable connectivity mapping approach particularly when timeseries are long (as is the case for fMRI data) and lagged autoregressive effects are modeled (as was done in the current study) (Gates et al., 2017; Lane et al., 2019). Critically, GIMME is one of the only approaches that has successfully and reliably recovered the true models in a benchmark dataset compiled by Smith and colleagues (Smith et al., 2011). The GIMME algorithm can recover individual-level directed paths in simulations that are modeled after conditions often seen in fMRI studies (i.e., nonstationarity, increased noise in selected ROIs). At the group-level, the same group-level paths are reliably identified when the algorithm is applied to smaller subsets of participants (Nichols et al., 2014), which demonstrates robustness to sample perturbations. These are clear advantages of this approach, which have been identified by reviews of causal search algorithms for fMRI by the Gates group and independent sources (Henry & Gates, 2017; Mumford & Ramsey, 2014). Specific to the S-GIMME algorithm used in the current study, simulated and empirical data have shown that subgroup classification is robust to perturbations in the degree to which individuals are similar. For example, subgroups remain stable even after randomly perturbing the similarity matrix of connectivity weights in order to simulate random fluctuations across individual data points (Gates et al., 2014). Further, validation tests show that S-GIMME reliably recovers underlying subgroups even in small samples (n = 25 before subgrouping) (Gates & Molenaar, 2012), which is also reflected by the sample sizes of the empirical validation studies (n=60–118) (Nichols et al., 2014; Price et al., 2020; Price, Gates, et al., 2017; Price, Lane, et al., 2017).

In the current study, directed paths (i.e., establishing which of a pair of ROIs statistically predicts the other) were derived for each individual in a data-driven fashion using S-GIMME (Gates et al., 2017; Lane, Gates, & Molenaar, 2015). Paths could be contemporaneous (marking prediction at the same timepoint) or lagged (marking prediction from one timepoint to the next). For each participant, S-GIMME generated a connectivity map (and associated beta weights) with group-level, subgroup-specific, and individual-level connections. To understand the nature of the connectivity patterns exhibited by each subgroup, the unique subgroup-level connections were inspected, and subgroups were compared across self-report, behavioral, and molecular levels of analysis to test within-sample generalization.

Patient-Reported Outcomes Measurement Information System (PROMIS) Computer Adaptive Tests (CATS).

PROMIS CATS are a collection of adaptive computer-based measures designed to reliably assess patient-reported outcomes (Cella et al., 2019) by using an algorithm that utilizes prior item responses to detect an appropriate successive item. The measure ends when a fixed measurement index (standard error<3.0) or specific number of items is reached. Each measure produces a T-score, where a score of 50 reflects the general population mean (SD=10). To index positive and negative affect, patients used the REDCap Assessment Center Application Programming Interface (API) to complete the PROMIS Anxiety, Anger, Depression, and Positive Affect CATS, with higher T-scores indicating greater levels of each respective mood state.

Positive Mood Induction.

Consistent with previous research (Horner et al., 2014), patients idiographically selected their favorite piece of music from a list of nonlinguistic pieces that have been found to elicit happy mood. They constructed a short paragraph about a vivid, extremely positive memory of one of the best times of their lives [at least a 7 on a scale of 1 (neutral) to 9 (the happiest they have ever been)]. During the 7-min mood induction, patients listened to the selected happy music and viewed their script on a screen. Patients continuously rated their affect by moving a mouse left for more negative and right for more positive. A visual scale was displayed to anchor ratings, from left to right in equidistant intervals ranging from “very sad”, “somewhat sad”, “neutral”, “somewhat happy” to “very happy”. Consistent with calculations conducted by Horner and colleagues (2014), from these continuous ratings we calculated indices of sustained affect (mean affect rating) and overall affective variability (SD of mean affect).

Affective Go/NoGo Task.

Consistent with past research (Tottenham, Hare, & Casey, 2011), this task instructed patients to make a button press when a particular facial expression target was presented on a computer screen (i.e., happy, sad, neutral). Faces were presented one-at-a-time, and patients were asked to make a button press as fast as possible when the target expression was presented on “Go” trials, occurring on 70% of trials. Patients were asked to withhold a button press for “NoGo” facial expressions. For each trial, a neutral expression was paired with an emotional expression (happy or sad), and depending on the block, the emotional expression either acted as the “Go” or “NoGo” target. Following practice trials, there were four blocks of “Go/NoGo” pairs (sad-neutral, neutral-sad, happy-neutral, and neutral-happy), consisting of 30 randomized trials each. Stimuli were presented for 500ms. False alarm rate (i.e., an error of commission by pressing the button during a “NoGo” trial) was calculated as the proportion of total false alarms to total “NoGo” trials. Hit rate was calculated as the proportion of correct responses to total “Go” trials. Consistent with prior research (Tottenham et al., 2011), to index ability to inhibit responses to emotional stimuli, a “d-prime” (d’) value for each block was calculated by subtracting the z-transformed false alarm rate from the z-transformed hit rate. Of note, consistent with traditional d’ calculations, this transformation calculates the z-transform of each hit and false alarm probability using ordinates from the probability density function of the standard normal distribution (μ = 0, σ = 1).

Results

Circuit Level of Analysis.

Group-level:

At the group level, contemporaneous and lagged connectivity paths, depicted in Figure 1A, were present, in addition to lagged autoregressions at every ROI. ROIs behaved as a strongly interconnected network, including numerous bilateral, ipsilateral, and within-network connections.

Figure 1.

Figure 1.

Figure 1.

Figure 1.

Figure 1.

Open in a new tab

(A) Functional regions of interest represented as nodes in roughly illustrated anatomical space. Nodes of the ventral affective network (VAN) are presented in blue; default mode network regions in green; and cognitive control network (CCN) regions in purple. Group-level directed connectivity paths between regions are depicted with arrows. Black, solid arrows represent positive, contemporaneous paths; red, dashed arrows represent negative, lagged paths. Not shown: positive, lagged autoregressive paths were also present for every region. (B) Directed connectivity paths unique to subgroup A (green/solid = positive, contemporaneous path; green/dashed = positive, lagged path; red/solid = negative, contemporaneous path; red/dashed = negative, lagged path), superimposed on group-level connectivity map (in grey). (C) Directed connectivity paths unique to subgroup B (green/solid = positive, contemporaneous path; green/dashed = positive, lagged path; red/solid = negative, contemporaneous path; red/dashed = negative, lagged path), superimposed on group-level connectivity map (in grey). (D) Directed connectivity paths unique to subgroup C (green/solid = positive, contemporaneous path; green/dashed = positive, lagged path; red/solid = negative, contemporaneous path; red/dashed = negative, lagged path), superimposed on group-level connectivity map (in grey).

Subgroup-level.

Based on an unsupervised search for the optimal number of subgroups, three subgroups emerged (Figures 1B–D). Subgroup A contained 49% of patients (n=34), Subgroup B contained 27% (n=19), and Subgroup C contained 24% (n=17). Pathways unique to each subgroup increased monotonically: Subgroup A exhibited 6 additional pathways, Subgroup B exhibited 9, and Subgroup C exhibited 10. Of its 9 additional pathways, subgroup B included one contemporaneous VAN→DMN negative pathway and two lagged VAN→DMN negative pathways that duplicated contemporaneous positive pathways between the same two regions. While Subgroups A and B did not share any additional pathways, Subgroup C included 4 pathways also seen in Subgroup B (VAN→VAN; CCN←→VAN; DMN→VAN) and 1 pathway (DMN→VAN) also seen in Subgroup A. Pathways unique to Subgroup C included additional VAN→CCN, VAN→VAN, and DMN→CCN connections.

S-GIMME subgroup was not associated with patient age, sex assigned at birth, racial or ethnic distribution, current use of psychotropic medication, treatment failures, or clinician-rated MADRS scores (ps≥.117). In addition, although our fMRI preprocessing steps buffer against motion confounds, we conducted analyses to ensure subgroup was unrelated to motion and other data quality issues. Across 12 motion parameters calculated for each participant (maximum absolute change from baseline and maximum incremental movement across each of 6 movement planes: roll, pitch, yaw, right-left, front-back, up-down), no motion index differed as a function of connectivity subgroup (ps>.307). The number of TRs ‘scrubbed’ for micromovement also did not differ by subgroup (p=.200).

Self-Report Level of Analysis.

A 3 (S-GIMME Subgroup: A, B, C) × 4 (Affect: Depressed, Positive, Angry, Anxious) mixed-model analysis of variance was performed on the dependent measure of PROMIS CATS T-scores. Although the main effect of S-GIMME was non-significant, F(2,67)=.637, p=.532, n2p=.019, there was a significant S-GIMME × Affect interaction, F(6,67)=2.678, p=.016, n2p=.074. Follow-up analyses revealed that, although there were no significant group differences in self-reported negative affect (anger, anxiety, depression) (ps≥.108), S-GIMME subgroup predicted differences in self-reported positive affect, F(2,67)=6.135, p=.004, n2p=.155, which was maintained after statistically controlling for the influence of patient’s clinician-rated depressive symptoms (p=.004). Posthocs revealed that Subgroup B exhibited the lowest levels of positive affect, compared to Subgroup A (p=.040; 95% CI [−4.645, −.106]) or C (p=.001; 95% CI [−7.280, −1.990]), and Subgroup A exhibited marginally lower levels compared to Subgroup C (p=.060; 95% CI [−4.612, .095]) [see Table 1 for subgroup M(SD)].

Behavioral Level of Analysis (Positive Mood Induction).

S-GIMME subgroup moderated differences in affective variability across the positive mood induction, F(2,67)=3.295, p=.043, n2p=.091, which was maintained after controlling for patient’s depressive symptoms (p=.043). Posthocs revealed that Subgroup C exhibited greater variability in continuous affect ratings during the induction, compared to Group A (p=.013; 95% CI [7.384, 59.441]). Group B did not differ from either Group A or C (ps≥.168) [see Table 1 for subgroup M(SD)]. In contrast, S-GIMME subgroup did not moderate differences in average affect during the induction, F(2,67)=.319, p=.728, n2p=.010.

Behavioral Level of Analysis (Affective Go/NoGo Task).

To assess response inhibition, a 3 (S-GIMME Subgroup: A, B, C) × 2 (Emotional Face: Sad, Happy) × 2 (Stimulus Type: Emotional Face as “Go” or “NoGo”) mixed-model analysis of variance was performed on the dependent measure of d-prime (d’). Although there were no interactive effects of Emotion or Stimulus with S-GIMME subgroup (ps≥.199), there was a main effect of S-GIMME subgroup, F(2,63)=3.941, p=.024, n2p=.110, which was maintained after controlling for patient’s depressive symptoms (p=.027). Posthocs revealed that Subgroup C exhibited worse response inhibition across blocks compared to either Subgroup A (p=.017; 95% CI [−.846, −.087]) or B (p=.014; 95% CI [−.970, −.114]). In contrast, Subgroup A and B did not differ (p=.687) [see Table 1 for subgroup M(SD)].

Molecular Level of Analysis.

A 3 (S-GIMME Subgroup: A, B, C) × 4 (Region: Caudate, Pallidum, Putamen, Nucleus Accumbens) mixed-model analysis of variance was performed on the dependent measure of nT2*w values. Although there were no interactive effects of Region (p=.179), there was a main effect of S-GIMME subgroup, F(2,63)=3.432, p=.038, n2p=.093, which was maintained after controlling for patient’s depressive symptoms (p=.034). Posthocs revealed that, compared to Subgroups A (p=.050; 95% CI [.000, .066]) and B (p=.013; 95% CI [.010, .085]), Subgroup C exhibited higher nT2*w values (i.e., lower levels of tissue iron concentrations) across regions, whereas Subgroup A and B did not differ (p=.370).

Discussion

Given the significant role of heterogeneity in the etiology and treatment of depression (Akil et al., 2018; Lynch et al., 2020), the current study sought to parse a well-known neurobiological substrate, resting state functioning connectivity (RSFC), in order to separate homogeneous and heterogeneous patterns of functioning among patients with treatment-resistant depression (TRD). Using a well-validated method to characterize patients’ RSFC (Gates et al., 2017), we identified common (group-level) pathways as well as three brain-based subgroups derived from heterogeneous RSFC profiles (see Figure 1A–D). This revealed a highly interconnected network with the expected autoregressive (i.e., each node’s lagged influence on itself), bilateral, ipsilateral, and within-network connections, as well as multiple between network pathways, suggesting that the connectivity within the selected regions had high relevance to the resting state. Notably, three common negative lagged pathways were also found (encompassing dACC→pgACC and bilateral insula→VLPFC nodes) that were redundant with positive contemporaneous paths, suggesting function within the network was not constrained to positive, relatively contemporaneous influences and may also be manifested through negative feedback (see also Price et al., 2020). Considerable RSFC heterogeneity was also identified. Of the three identified RSFC subgroups, subgroup A was the largest, containing 49% of patients and the fewest additional pathways, perhaps representing a more typical presentation of TRD (i.e., containing the plurality of patients as well as RSFC pathways most similar to the group-level pattern). Subgroup B contained 27% of patients and included two additional negative lagged paths (left VLPFC→pgACC; right anterior insula→PCC), which may represent a subtype exhibiting additional negative feedback between VAN and DMN regions. Finally, subgroup C contained 24% of patients and the largest number of positive contemporaneous pathways, perhaps representing a subtype with a more diffuse pattern of connectivity across key regions.

The current study also sought to determine if RSFC heterogeneity among patients with TRD would generalize to discrete patterns of functioning across transdiagnostic positive and negative valence systems (Cuthbert & Insel, 2013). To test within-sample generalization, subgroups were compared across self-report, behavioral and molecular units of analysis that were chosen to capture transdiagnostic elements of cognitive-affective functioning relevant to depression (Price & Duman, 2020) (see Table 2 for a summary of findings across levels). Clinician ratings of depressive symptoms were not discriminated by RSFC-subgroups, suggesting that patients within each subgroup were experiencing similar levels of disorder-specific impairment. However, RSFC-based categorization predicted discrete aberrations in transdiagnostic cognitive-affective markers across multiple levels of analysis, even after controlling statistically for the influence of depressive symptoms, which suggests that RSFC subgroups discriminated patients’ levels of transdiagnostic functioning above and beyond any variability captured by a disorder-specific measure of functioning. Critically, these transdiagnostic patterns were largely unified, showing that subgroups exhibited consistent differences across levels of analysis. At the self-report level, positive affect was stratified across subgroup (Subgroup C>A>B). At the behavioral level, subgroup C was characterized by higher variability in affect during the positive mood induction and worse response inhibition during the affective Go/NoGo task, compared to other subgroups. At the molecular level, Subgroup C also exhibited reduced striatal tissue iron concentration, which is a dopaminergic substrate thought to be related to the ability to engage goal-directed responses and inhibit task-irrelevant ones (Adisetiyo et al., 2014; Larsen, Bourque, et al., 2020) and relevant to the pathophysiology of depression (Dunlop & Nemeroff, 2007; Grace, 2016).

Table 2.

Summary of RSFC subgroup in sample generalization across self-report, behavioral, and molecular levels of analysis.

Level of Analysis
Self-Report Behavioral Molecular
Subgroup A Intermediate Positive Affect >B*; <Cⴕ Lower Affective Variability <C*
Higher Response Inhibition >C* 		Higher Striatal Tissue Iron >C*
B Lower Positive Affect <A*; <C*** Higher Response Inhibition >C* Higher Striatal Tissue Iron >C*
C Higher Positive Affect >B***; >Aⴕ Higher Affective Variability >A*
Lower Response Inhibition <A*; <B* 		Lower Striatal Tissue Iron <A*; <B*
Open in a new tab

Note.

ⴕ

p = .060

*

p ≤ .050.

**

p ≤ .010.

***

p ≤ .001. Superscript letters A, B, C represent which RSFC subgroup was used in each posthoc comparison.

Notably, the majority of significant subgroup differences in functioning were specific to Subgroup C, which contained several positive DMN→CCN and VAN→CCN connections, with no unique paths originating in CCN regions. As the DMN has been widely linked to self-referential processing (Gusnard, Akbudak, Shulman, & Raichle, 2001) and, in the context of depression, negative rumination (Hamilton, Farmer, Fogelman, & Gotlib, 2015), these unique paths could reflect self-referential thinking and/or affective processing that is highly influential upon CCN regions. This baseline neural architecture might prime disruption during goal-oriented external tasks (Clark, Chamberlain, & Sahakian, 2009; Dwyer et al., 2014; Goulden et al., 2014), leading to the increased cognitive and emotion processing deficits observed in this subgroup. Specifically, downstream functioning in Subgroup C was marked by affective lability (i.e., positive affect that was elevated in daily life but also more variable in moment-to-moment fluctuations relative to the other patient subgroups) and response inhibition deficits, which were manifested during performance in all blocks of the affective Go/NoGo task and thought to be related to lower concentrations of striatal tissue iron (Adisetiyo et al., 2014; Larsen, Bourque, et al., 2020). Although increased positive affect in depressed patients is typically conceptualized as a strength to be capitalized upon in therapy, when flanked by the corresponding response inhibition deficits seen in Subgroup C, it may be a component of maladaptive functioning not well-captured by traditional diagnostic definitions of depression. Response inhibition encompasses both the ability to discriminate between salient stimuli that have varying relevance for subsequent goal-directed behaviors and the capability to generate the correct behavior once the relevant meaning has been extracted from said stimuli (Tottenham et al., 2011). Thus, increased difficulty first decoding and then inhibiting responses to emotional information (as reflected by the lower “d-prime” Go/NoGo values observed in Subgroup C) may contribute to patterns of inflexible emotion responding often observed among depression patients by making it more difficult to sensitively employ emotion regulation strategies that support periods of sustained positive affect (LeMoult & Gotlib, 2019), Similarly, diminished ability to engage in motivation and reward-driven behaviors, which is associated with the lower levels of striatal tissue iron concentration seen in Subgroup C (Adisetiyo et al., 2014; Larsen, Bourque, et al., 2020), reduces the probability of recruiting the adaptive emotion regulation skills and goal-directed behaviors that are required to overcome inflexible responding and sustain euthymic levels of positive affect. Given the specific RSFC profile found in Subgroup C, the current findings suggest that heightened DMN/VAN connection to the CCN may be associated with response inhibition deficits that could prevent patients in this subgroup from fully capitalizing upon an existing clinical strength (i.e., their elevated levels of positive affect, relative to other TRD peers). As such, using a synergistic combination of therapeutic techniques [e.g., emotion regulation skills selected from Dialectical Behavior Therapy (DBT) or Cognitive Therapy (CT) modules] and novel antidepressant interventions [e.g., enkephalinase inhibitors that impact striatal dopamine function (Peciña et al., 2019)] to address such deficits could be warranted for this subgroup.

The absence of group differences in depressive symptoms and self-reported negative affect (i.e., depression, anger, and anxiety) was inconsistent with prior research showing that data-driven parsing of RSFC can be used to predict depression severity (Beijers et al., 2019). Because of inclusion criteria, patients in the current study exhibited moderate-to-severe TRD, which may have led to homogeneity in patterns of RSFC associated with negative affect, thereby reducing power to detect subgroup differences in this domain. Indeed, this pattern was seen in the current study for both disorder-specific (depression severity) and various transdiagnostic measures of negative affect. Associations between RSFC and negative affect have been well described in S-GIMME research utilizing more heterogeneous samples (Chahal et al., 2020; Price, Gates, et al., 2017), and these RSFC patterns were not replicated in the current, more homogenous sample. Thus, it is possible that RSFC patterns most relevant to negative affect were best captured by the common (group-level) pathways in the current study. Yet, where homogeneity may have limited power for some hypotheses, it also introduced opportunities to parse heterogeneity in functioning not well-described by disorder-specific symptom measures, and revealed numerous subgroup distinctions in external markers, particularly those related to affective lability and response inhibition.

The results from the current study contribute to a growing body of research using S-GIMME, as well as the same ROIs, to examine the association between heterogeneous RSFC and depression across development (Chahal et al., 2020; Price, Gates, et al., 2017). For example, one study found that adolescents who belonged to a subgroup defined by hyperconnectivity between DMN and CCN exhibited greater increases in internalizing problems, such as depression, across a two-year follow-up, compared to peers in a more diffusely-connected subgroup (Chahal et al., 2020). In a sample of 80 depressed adults, who were more closely aligned with the current sample, two subgroups were found; one, containing 71% of patients, demonstrated the expected connectivity across DMN, whereas the remaining patients were characterized by DMN hypoconnectivity and CCN hyperconnectivity (Price, Gates, et al., 2017). The current study was not able to test out-of-sample generalization of these two previously published subgroups for several reasons. First, the sample characteristics of the current study differ considerably from those used in (Price, Gates, et al., 2017) (i.e., TRD vs. moderate-to-severe depression; psychotropic medications permitted vs. an unmedicated sample). Second, advances in imaging technology (i.e., multiband coil providing TR of 800ms) has allowed for much more precise quantification of temporal patterns, which would lead to inadequate comparisons. However, taken together, these studies highlight the clear importance of RSFC heterogeneity in depression, particularly for DMN connections to CCN regions.

Findings from the current study could have implications for novel TRD treatment targets. There is an urgent clinical need for such insights, given that approximately 2/3 of depressed patients fail to remit following the first standard treatment attempt and at least 1/3 become highly refractory (Rush, 2007). TRD is also associated with discrete neurobiological profiles, compared to treatment-responsive depression (Akil et al., 2018), highlighting the need for precision medicine protocols that match patients to interventions best suited to their baseline neurobiological profile. Novel, brain-based intervention approaches, including behavioral neurocognitive training [e.g., computer-based cognitive bias therapeutics which have exhibited effects on connectivity patterns in depression (Wiers & Wiers, 2017)], neuromodulation approaches (e.g., rTMS), and/or novel psychopharmacological interventions [e.g., enkephalinase inhibitors, which target δ-opioid receptors, may exert antidepressant effects in part via the striatal dopaminergic system (Peciña et al., 2019)] could ideally be tailored to RSFC subgroups, consistent with previously reported treatment prediction effects (Drysdale et al., 2017). This represents a promising avenue for further research. A critical next step to support these avenues is to establish the out-of-sample generalizability of S-GIMME-derived subgroups. Although we were well-powered for RSFC subgrouping via S-GIMME [i.e., as shown in previous S-GIMME validation work in simulated datasets (Gates et al., 2017; Gates & Molenaar, 2012; Lane et al., 2019)] and subgroups were largely unified across levels of analysis, suggesting a level of reliability in subgroup classification, replicating findings in a separate sample would lay critical groundwork for testing potential treatment moderation effects—ideally within a randomized design using multiple bona fide treatments (e.g., conventional and/or novel brain-based approaches).

The current study had several limitations. The practical implications of the current findings may be limited if subtypes can only be identified by and related to less clinically available measures (fMRI, behavioral research assessments). Although the average number of treatment failures in the present sample was 2.56, patients were eligible if they had ≥1 treatment failure within the current depressive episode, a more liberal definition of TRD than has been used in many other studies. Another limitation of the current work is that S-GIMME is best suited to examine ≤15 ROIs in a single analysis, and the results may have differed with the inclusion of different brain regions. However, the selected regions were indicated by meta-analytic and prior research and provided clinically meaningful subgroups in previous S-GIMME studies, suggesting they were well-suited to test our hypotheses. This study did not include explicit reward paradigms commonly used in depression research (Borsini, Wallis, Zunszain, Pariante, & Kempton, 2020). The inclusion of such tasks in future research may be critical to understanding further heterogeneity in transdiagnostic positive valence systems among depressed patients. Finally, the sample consisted of predominately non-Hispanic White patients, which was not fully representative of the local Pittsburgh region or national distributions. Future work is needed to replicate and extend current findings in more diverse samples, which would be a critical next step to improve health equity for minoritized populations, who have been historically underrepresented in psychiatric research.

In conclusion, our data-driven categorization of RSFC profiles in a sample of patients with TRD revealed three subgroups with divergent patterns in markers of cognitive-affective functioning, observed consistently across multiple levels of analysis (self-report, behavioral, molecular). Consistent with the growing movement to parse neurobiological substrates of depressed mood (Beijers et al., 2019; Lynch et al., 2020), the current findings suggest that RSFC-based subgrouping is well-suited to capitalize on well-known heterogeneity in depression, pointing the way to unique neurobiological etiologies and/or novel treatment targets. In the current study, RSFC subgroups generalized well to measures of functioning across transdiagnostic positive and negative valence systems, even within the context of a treatment-seeking patient sample with certain uniform characteristics placing them at heightened risk for poor clinical outcomes (TRD, moderate-to-severe depressive symptoms). In particular, the more diffuse RSFC profile found in Subgroup C, which included heightened DMN/VAN connection to the CCN, was related to increased affective lability and response inhibition deficits that may represent a pattern of cognitive-affective functioning that is not well-described by traditional characterizations of depression. Future efforts in this vein may propagate our understanding of transdiagnostic brain-behavior heterogeneity and inform precision medicine protocols that capitalize on, rather than ignore or suppress, the important role of heterogeneity in patient outcomes. Such future approaches may be especially warranted for TRD given the desperate need for novel biologically-based interventions for this population (Akil et al., 2018; Rush, 2007).

Supplementary Material

Supplement

Acknowledgments

Financial Support

This project was supported by National Institutes of Health grant MH113857 awarded to R.B.P. M.L.W. is supported by NIMH Grant K23 MH119225.

Footnotes

Ethical Standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Conflicts of Interest

All authors report no biomedical financial interests or potential conflicts of interest.

References

  1. Adisetiyo V, Jensen JH, Tabesh A, Deardorff RL, Fieremans E, Di Martino A, . . . Helpern JA (2014). Multimodal MR imaging of brain iron in attention deficit hyperactivity disorder: a noninvasive biomarker that responds to psychostimulant treatment? Radiology, 272(2), 524–532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Akil H, Gordon J, Hen R, Javitch J, Mayberg H, McEwen B, . . . Nestler EJ (2018). Treatment resistant depression: a multi-scale, systems biology approach. Neuroscience & Biobehavioral Reviews, 84, 272–288. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. American Psychiatric Association (2013). Diagnostic and Statistical Manual of Mental Disorders (DSM-5®): American Psychiatric Pub. [Google Scholar]
  4. Beard JL, & Connor JR (2003). Iron status and neural functioning. Annual Review of Nutrition, 23(1), 41–58. [DOI] [PubMed] [Google Scholar]
  5. Beijers L, Wardenaar KJ, van Loo HM, & Schoevers RA (2019). Data-driven biological subtypes of depression: Systematic review of biological approaches to depression subtyping. Molecular Psychiatry, 24(6), 888–900. [DOI] [PubMed] [Google Scholar]
  6. Beltz AM, Moser JS, Zhu DC, Burt SA, & Klump KL (2018). Using person-specific neural networks to characterize heterogeneity in eating disorders: Illustrative links between emotional eating and ovarian hormones. International Journal of Eating Disorders, 51(7), 730–740. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Borsini A, Wallis ASJ, Zunszain P, Pariante CM, & Kempton MJ (2020). Characterizing anhedonia: A systematic review of neuroimaging across the subtypes of reward processing deficits in depression. Cognitive, Affective & Behavioral Neuroscience. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Cella D, Choi SW, Condon DM, Schalet B, Hays RD, Rothrock NE, . . . Amtmann D. 2019). PROMIS® adult health profiles: Efficient short-form measures of seven health domains. Value in Health, 22(5), 537–544. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chahal R, Weissman DG, Hallquist MN, Robins RW, Hastings PD, & Guyer AE (2020). Neural connectivity biotypes: Associations with internalizing problems throughout adolescence, Psychological Medicine. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Clark L, Chamberlain SR, & Sahakian BJ (2009). Neurocognitive mechanisms in depression: Implications for treatment. Annual Review of Neuroscience, 32(1), 57–74. [DOI] [PubMed] [Google Scholar]
  11. Clementz BA, Sweeney JA, Hamm JP, Ivleva EI, Ethridge LE, Pearlson GD, . . . Tamminga CA (2016). Identification of distinct psychosis biotypes using brain-based biomarkers. American Journal of Psychiatry, 173(4), 373–384. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cuthbert BN, & Insel TR (2013). Toward the future of psychiatric diagnosis: The seven pillars of RDoC. BMC Medicine, 11(1), 126. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Drysdale AT, Grosenick L, Downar J, Dunlop K, Mansouri F, Meng Y, . . . Etkin A. 2017). Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nature Medicine, 23(1), 28–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dubol M, Trichard C, Leroy C, Sandu A-L, Rahim M, Granger B, . . . Artiges E(2018). Dopamine transporter and reward anticipation in a dimensional perspective: A multimodal brain imaging study. Neuropsychopharmacology, 43(4), 820–827. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Dunlop BW, & Nemeroff CB (2007). The role of dopamine in the pathophysiology of depression. Archives of General Psychiatry, 64(3), 327–337. [DOI] [PubMed] [Google Scholar]
  16. Dwyer DB, Harrison BJ, Yücel M, Whittle S, Zalesky A, Pantelis C, . . . Fornito A. 2014). Large-scale brain network dynamics supporting adolescent cognitive control. Journal of Neuroscience, 34(42), 14096–14107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Erikson KM, Jones BC, & Beard JL (2000). Iron deficiency alters dopamine transporter functioning in rat striatum. The Journal of Nutrition, 130(11), 2831–2837. [DOI] [PubMed] [Google Scholar]
  18. Finn ES, Shen X, Scheinost D, Rosenberg MD, Huang J, Chun MM, . . . Constable RT (2015). Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nature Neuroscience, 18(11), 1664–1671. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Gates KM, Lane ST, Varangas E, Giovanello K, & Guiskewicz K. 2017). Unsupervised classification during time series model building. Multivariate Behavioral Research, 52(2), 129–148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gates KM, & Molenaar PC (2012). Group search algorithm recovers effective connectivity maps for individuals in homogeneous and heterogeneous samples. Neuroimage, 63(1), 310–319. [DOI] [PubMed] [Google Scholar]
  21. Gates KM, Molenaar PC, Iyer SP, Nigg JT, & Fair DA (2014). Organizing heterogeneous samples using community detection of GIMME-derived resting state functional networks. PloS One, 9(3), e91322. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Goldstein-Piekarski AN, Ball TM, Samara Z, Staveland BR, Keller AS, Fleming SL, . . . Williams LM (2020). Mapping neural circuit biotypes to symptoms and behavioral dimensions of depression and anxiety. The Lancet Psychiatry. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Goulden N, Khusnulina A, Davis NJ, Bracewell RM, Bokde AL, McNulty JP, & Mullins PG (2014). The salience network is responsible for switching between the default mode network and the central executive network: replication from DCM. Neuroimage, 99, 180–190. [DOI] [PubMed] [Google Scholar]
  24. Grace AA (2016). Dysregulation of the dopamine system in the pathophysiology of schizophrenia and depression. Nature Reviews Neuroscience, 17(8), 524. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Grahek I, Everaert J, Krebs RM, & Koster EH (2018). Cognitive control in depression: Toward clinical models informed by cognitive neuroscience. Clinical Psychological Science, 6(4), 464–480. [Google Scholar]
  26. Gratton C, Laumann TO, Nielsen AN, Greene DJ, Gordon EM, Gilmore AW, . . . Schlaggar B L. (2018). Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation. Neuron, 98(2), 439–452. e435. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Greicius MD, Krasnow B, Reiss AL, & Menon V. 2003). Functional connectivity in the resting brain: A network analysis of the default mode hypothesis. Proceedings of the National Academy of Sciences, 100(1), 253–258. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Gusnard DA, Akbudak E, Shulman GL, & Raichle ME (2001). Medial prefrontal cortex and self-referential mental activity: Relation to a default mode of brain function. Proceedings of the National Academy of Sciences, 98(7), 4259–4264. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Haacke EM, Cheng NY, House MJ, Liu Q, Neelavalli J, Ogg RJ, . . . Obenaus A. 2005). Imaging iron stores in the brain using magnetic resonance imaging. Magnetic Resonance Imaging, 23(1), 1–25. [DOI] [PubMed] [Google Scholar]
  30. Hamilton JP, Farmer M, Fogelman P, & Gotlib IH (2015). Depressive rumination, the default-mode network, and the dark matter of clinical neuroscience. Biological Psychiatry, 78(4), 224–230. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Hamilton JP, Sacchet MD, Hjørnevik T, Chin FT, Shen B, Kämpe R, . . . Borg N. 2018). Striatal dopamine deficits predict reductions in striatal functional connectivity in major depression: A concurrent 11 C-raclopride positron emission tomography and functional magnetic resonance imaging investigation. Translational Psychiatry, 8(1), 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Hasler G, & Northoff G. 2011). Discovering imaging endophenotypes for major depression. Molecular Psychiatry, 16(6), 604–619. [DOI] [PubMed] [Google Scholar]
  33. Henry T, & Gates KM (2017). Causal search procedures for fMRI: Review and suggestions. Behaviormetrika, 44(1), 193–225. [Google Scholar]
  34. Horner MS, Siegle GJ, Schwartz RM, Price RB, Haggerty AE, Collier A, & Friedman ES (2014). C’mon get happy: Reduced magnitude and duration of response during a positive-affect induction in depression. Depression and Anxiety, 31(11), 952–960. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Kaiser RH, Andrews-Hanna JR, Wager TD, & Pizzagalli DA (2015). Large-scale network dysfunction in major depressive disorder: A meta-analysis of resting-state functional connectivity. JAMA Psychiatry, 72(6), 603–611. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Karalunas SL, Fair D, Musser ED, Aykes K, Iyer SP, & Nigg JT (2014). Subtyping attention-deficit/hyperactivity disorder using temperament dimensions: Toward biologically based nosologic criteria. JAMA Psychiatry, 71(9), 1015–1024. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  37. Kim J, Zhu W, Chang L, Bentler PM, & Ernst T. 2007). Unified structural equation modeling approach for the analysis of multisubject, multivariate functional MRI data. Human Brain Mapping, 28(2), 85–93. doi: 10.1002/hbm.20259 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Kozak MJ, & Cuthbert BN (2016). The NIMH research domain criteria initiative: background, issues, and pragmatics. Psychophysiology, 53(3), 286–297. [DOI] [PubMed] [Google Scholar]
  39. Lane ST, Gates KM, & Molenaar PCM (2015). gimme: https://cran.r-project.org/web/packages/gimme/index.html.
  40. Lane ST, Gates KM, Pike HK, Beltz AM, & Wright AG (2019). Uncovering general, shared, and unique temporal patterns in ambulatory assessment data. Psychological Methods, 24(1), 54. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Larsen B, Bourque J, Moore TM, Adebimpe A, Calkins ME, Elliott MA, . . . Roalf DR (2020). Longitudinal development of brain iron is linked to cognition in youth. Journal of Neuroscience, 40(9), 1810–1818. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Larsen B, & Luna B. 2015). In vivo evidence of neurophysiological maturation of the human adolescent striatum. Developmental Cognitive Neuroscience, 12, 74–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Larsen B, Olafsson V, Calabro F, Laymon C, Tervo-Clemmens B, Campbell E, . . . Luna B. 2020). Maturation of the human striatal dopamine system revealed by PET and quantitative MRI. Nature Communications, 11(1), 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. LeMoult J, & Gotlib IH (2019). Depression: A cognitive perspective. Clinical Psychology Review, 69, 51–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Litvina E, Adams A, Barth A, Bruchez M, Carson J, Chung JE, . . . Harris KM (2019). BRAIN Initiative: Cutting-edge tools and resources for the community. Journal of Neuroscience, 39(42), 8275–8284. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Lynch CJ, Gunning FM, & Liston C. 2020). Causes and consequences of diagnostic heterogeneity in depression: Paths to discovering novel biological depression subtypes. Biological Psychiatry. [DOI] [PubMed] [Google Scholar]
  47. Montgomery S, & Åsberg M. 1977). A new depression scale designed to be sensitive to change: Acad. Department of Psychiatry, Guy’s Hospital. [DOI] [PubMed] [Google Scholar]
  48. Mumford JA, & Ramsey JD (2014). Bayesian networks for fMRI: a primer. Neuroimage, 86, 573–582. [DOI] [PubMed] [Google Scholar]
  49. Nichols TT, Gates KM, Molenaar PC, & Wilson SJ (2014). Greater BOLD activity but more efficient connectivity is associated with better cognitive performance within a sample of nicotine-deprived smokers. Addiction Biology, 19(5), 931–940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Peciña M, Karp JF, Mathew S, Todtenkopf MS, Ehrich EW, & Zubieta J-K (2019). Endogenous opioid system dysregulation in depression: Implications for new therapeutic approaches. Molecular Psychiatry, 24(4), 576–587. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Peterson ET, Kwon D, Luna B, Larsen B, Prouty D, De Bellis MD, . . . Pohl KM (2019). Distribution of brain iron accrual in adolescence: Evidence from cross-sectional and longitudinal analysis. Human Brain Mapping, 40(5), 1480–1495. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Price RB, Beltz AM, Woody ML, Cummings L, Gilchrist D, & Siegle GJ (2020). Neural Connectivity Subtypes Predict Discrete Attentional-Bias Profiles Among Heterogeneous Anxiety Patients. Clinical Psychological Science, 8(3), 491–505. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Price RB, & Duman R. 2020). Neuroplasticity in cognitive and psychological mechanisms of depression: an integrative model. Molecular Psychiatry, 25(3), 530–543. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Price RB, Gates KM, Kraynak TE, Thase ME, & Siegle GJ (2017). Data-driven subgroups in depression derived from directed functional connectivity paths at rest. Neuropsychopharmacology, 42(13), 2623–2632. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Price RB, Lane ST, Gates KM, Kraynak T, Horner M, & Thase M. 2017). Parsing heterogeneity in directed brain connectivity during positive mood: A community detection analysis in depressed and healthy adults. Biological Psychiatry, 81, 347–357. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Price RB, Tervo-Clemmens BC, Panny B, Degutis M, Griffo A, & Woody ML (2021). Correlates of Striatal Tissue Iron in Depressed Patients. Submitted. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Rottenberg J. 2017). Emotions in depression: What do we really know? Annual Review of Clinical Psychology, 13, 241–263. [DOI] [PubMed] [Google Scholar]
  58. Rush AJ (2007). STAR* D: What have we learned? American Journal of Psychiatry, 164(2), 201–204. [DOI] [PubMed] [Google Scholar]
  59. Siegle GJ, Thompson W, Carter CS, Steinhauer SR, & Thase M. E. J. B. p. (2007). Increased amygdala and decreased dorsolateral prefrontal BOLD responses in unipolar depression: Related and independent features. Biological Psychiatry, 61(2), 198–209. [DOI] [PubMed] [Google Scholar]
  60. Smith SM, Miller KL, Salimi-Khorshidi G, Webster M, Beckmann CF, Nichols TE, . . . Woolrich MW (2011). Network modelling methods for FMRI. Neuroimage, 54(2), 875–891. [DOI] [PubMed] [Google Scholar]
  61. Tottenham N, Hare TA, & Casey B. 2011). Behavioral assessment of emotion discrimination, emotion regulation, and cognitive control in childhood, adolescence, and adulthood. Frontiers in Psychology, 2, 39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Vo LT, Walther DB, Kramer AF, Erickson KI, Boot WR, Voss MW, . . . Gratton G. 2011). Predicting individuals’ learning success from patterns of pre-learning MRI activity. PloS One, 6(1), e16093. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Wiers CE, & Wiers RW (2017). Imaging the neural effects of cognitive bias modification training. Neuroimage, 151, 81–91. [DOI] [PubMed] [Google Scholar]
  64. Williams LM (2016). Precision psychiatry: A neural circuit taxonomy for depression and anxiety. The Lancet Psychiatry, 3(5), 472–480. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Williams LM, & Goldstein-Piekarski AN (2020). Applying a neural circuit taxonomy in depression and anxiety for personalized psychiatry. In Personalized Psychiatry (pp. 499–519): Elsevier. [Google Scholar]
  66. Woody ML, & Gibb BE (2015). Integrating NIMH research domain criteria (RDoC) into depression research. Current Opinion in Psychology, 4, 6–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Yang Z, Xu Y, Xu T, Hoy CW, Handwerker DA, Chen G, . . . Bandettini PA (2014). Brain network informed subject community detection in early-onset schizophrenia. Scientific Reports, 4, 5549. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Zhang X, Wang F, Hamilton JP, Sacchet MD, Chen J, Khalighi M, . . . Glover GH (2019). Decoupling of Dopamine Release and Neural Activity in Major Depressive Disorder during Reward Processing Assessed by Simultaneous fPET-fMRI. bioRxiv, 861534. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement
NIHMS1760407-supplement-Supplement.docx (31.7KB, docx)

RESOURCES