Baseline Functional Connectivity in Resting State Networks Associated with Depression and Remission Status after 16 Weeks of Pharmacotherapy: A CAN-BIND Report

Gwen van der Wijk; Jacqueline K Harris; Stefanie Hassel; Andrew D Davis; Mojdeh Zamyadi; Stephen R Arnott; Roumen Milev; Raymond W Lam; Benicio N Frey; Geoffrey B Hall; Daniel J Müller; Susan Rotzinger; Sidney H Kennedy; Stephen C Strother; Glenda M MacQueen; Andrea B Protzner

doi:10.1093/cercor/bhab286

. 2021 Aug 20;32(6):1223–1243. doi: 10.1093/cercor/bhab286

Baseline Functional Connectivity in Resting State Networks Associated with Depression and Remission Status after 16 Weeks of Pharmacotherapy: A CAN-BIND Report

Gwen van der Wijk ^1,^✉, Jacqueline K Harris ^2,³, Stefanie Hassel ^4,⁵, Andrew D Davis ^6,⁷, Mojdeh Zamyadi ⁶, Stephen R Arnott ⁶, Roumen Milev ⁸, Raymond W Lam ⁹, Benicio N Frey ^10,¹¹, Geoffrey B Hall ^7,¹², Daniel J Müller ^13,^14,^15,¹⁶, Susan Rotzinger ^17,¹⁸, Sidney H Kennedy ^13,^16,^17,^19,²⁰, Stephen C Strother ^6,²¹, Glenda M MacQueen ^4,⁵, Andrea B Protzner ^1,^5,²²

¹ Department of Psychology, University of Calgary, Calgary AB T2N 1N4, Canada

² Department of Computing Science, University of Alberta, Edmonton AB T6G 2S4, Canada

³ Alberta Machine Intelligence Institute, Edmonton AB T5J 3B1, Canada

⁴ Cumming School of Medicine, Department of Psychiatry, University of Calgary, Calgary AB T2N 4N1, Canada

⁵ Mathison Centre for Mental Health Research and Education, University of Calgary, Calgary AB T2N 4Z6, Canada

⁶ Rotman Research Institute, Baycrest Health Sciences, Toronto ON M6A 2E1, Canada

⁷ Department of Psychology, Neuroscience & Behaviour, McMaster University, Hamilton ON L8S 4L6, Canada

⁸ Queen’s University, Departments of Psychiatry and Psychology, and Providence Care Hospital, Kingston ON K7L 3N6, Canada

⁹ Department of Psychiatry, University of British Columbia, Vancouver BC V6T 2A1, Canada

¹⁰ Department of Psychiatry and Behavioural Neurosciences, McMaster University, Hamilton ON L8S 4L8, Canada

¹¹ Mood Disorders Program and Women’s Health Concerns Clinic, St. Joseph’s Healthcare, Hamilton ON L8N 4N6, Canada

¹² Imaging Research Centre, St. Joseph's Healthcare Hamilton, Hamilton ON L8N 4N6, Canada

¹³ Department of Psychiatry, University of Toronto, Toronto ON M5T 1R8, Canada

¹⁴ Campbell Family Mental Health Research Institute, Centre for Addiction and Mental Health, Toronto ON M5T 1R8, Canada

¹⁵ Department of Pharmacology and Toxicology, University of Toronto, Toronto ON M5S 1A8, Canada

¹⁶ Institute of Medical Sciences, University of Toronto, Toronto ON M5S 1A8, Canada

¹⁷ Centre for Mental Health, University Health Network, Toronto ON M5G 1L7, Canada

¹⁸ Keenan Research Centre for Biomedical Science, Li Ka Shing Knowledge Institute, St Michael's Hospital, Toronto ON M5B 1T8, Canada

¹⁹ Centre for Depression and Suicide Studies, Unity Health Toronto, Toronto ON M5B 1W8, Canada

²⁰ Krembil Research Institute, Toronto Western Hospital, Toronto ON M5T 0S8, Canada

²¹ Department of Medical Biophysics, University of Toronto, Toronto ON M5G 1L7, Canada

²² Hotchkiss Brain Institute, University of Calgary, Calgary AB T2N 4N1, Canada

^✉

Address correspondence to Gwen van der Wijk, Department of Psychology, University of Calgary, 2500 University Drive NW, Calgary T2N 1N4, Alberta, Canada. Email: gwen.vanderwijk@ucalgary.ca

PMCID: PMC8924431 PMID: 34416758

Abstract

Understanding the neural underpinnings of major depressive disorder (MDD) and its treatment could improve treatment outcomes. So far, findings are variable and large sample replications scarce. We aimed to replicate and extend altered functional connectivity associated with MDD and pharmacotherapy outcomes in a large, multisite sample. Resting-state fMRI data were collected from 129 patients and 99 controls through the Canadian Biomarker Integration Network in Depression. Symptoms were assessed with the Montgomery-Åsberg Depression Rating Scale (MADRS). Connectivity was measured as correlations between four seeds (anterior and posterior cingulate cortex, insula and dorsolateral prefrontal cortex) and all other brain voxels. Partial least squares was used to compare connectivity prior to treatment between patients and controls, and between patients reaching remission (MADRS ≤ 10) early (within 8 weeks), late (within 16 weeks), or not at all. We replicated previous findings of altered connectivity in patients. In addition, baseline connectivity of the anterior/posterior cingulate and insula seeds differentiated patients with different treatment outcomes. The stability of these differences was established in the largest single-site subsample. Our replication and extension of altered connectivity highlighted previously reported and new differences between patients and controls, and revealed features that might predict remission prior to pharmacotherapy. Trial registration:ClinicalTrials.gov: NCT01655706.

Keywords: fMRI, functional connectivity, major depressive disorder, resting state networks

Introduction

Major depressive disorder (MDD) is a prevalent and debilitating disorder. Although several effective treatments are available, including psychotherapy, pharmacotherapy, and neurostimulation therapy, treatment outcomes vary greatly among patients (Rush et al. 2006; Cuijpers et al. 2020). Many neuroimaging studies have investigated brain changes related to MDD pathology and antidepressant treatment, but while the results are promising, inconsistent findings have led to growing concerns about reproducibility (Poldrack et al. 2017). To improve treatment outcomes for patients with MDD, a robust understanding of the characteristics associated with MDD and antidepressant treatment outcomes is essential.

Functional connectivity during rest, as captured by functional magnetic resonance imaging (fMRI) of the brain, has emerged as a potentially informative measure of the neural underpinnings of both the depressive state and its (successful) treatment (Dichter et al. 2015; Gudayol-Ferre et al. 2015; Kaiser et al. 2015; Mulders et al. 2015). Based on the more robust findings highlighted in meta-analyses and literature reviews, the three networks that have most consistently been linked to MDD pathology include the default mode network (DMN), the salience network (SN), and the cognitive control network (CCN) (Gudayol-Ferre et al. 2015; Mulders et al. 2015; Brakowski et al. 2017; Dunlop et al. 2019). Core regions of the DMN are the posterior cingulate cortex (PCC), precuneus, the bilateral angular gyrus, and the medial prefrontal cortex (mPFC; Mulders et al. 2015; Dunlop et al. 2019). In MDD, the anterior cingulate cortex (ACC) has also been found to be part of the DMN (Zhou et al. 2010). This network is generally thought to be related to internal processing in conscious participants (Raichle 2015; see Boly et al. 2008 for a discussion on DMN activity during unconscious states). The core regions of the SN include the bilateral insulae, the amygdalae, the temporal poles, and the dorsal anterior cingulate cortex (dACC; Seeley et al. 2007; Mulders et al. 2015). It is involved in emotion processing and bottom-up attentional processes (e.g., monitoring for salient stimuli; Kaiser et al. 2015). Core CCN regions include the bilateral dorsolateral prefrontal cortex (dlPFC), frontal eye fields (FEF), dorsomedial prefrontal cortex (dmPFC), and the posterior parietal cortex (PPC), which together exert top-down control over cognitive and emotional processes (Mulders et al. 2015; Dunlop et al. 2019). Disturbed connectivity within and between these networks is thought to lead to the abnormal emotion processing and mood regulation in MDD (Kaiser et al. 2015; Rayner et al. 2016). Given the frequency with which these networks are highlighted in previous work and their functional relevance to MDD pathology, we focused on these three resting state networks in our study.

A meta-analysis and literature review summarizing resting state fMRI connectivity alterations in patients with MDD compared with controls reported that MDD was characterized by stronger connectivity within the DMN, especially anteriorly, between the CCN and DMN, and between the anterior DMN and SN (Kaiser et al. 2015; Mulders et al. 2015). In addition, weaker connectivity within the CCN, between the SN and the posterior DMN, and between the posterior DMN and CNN was found in patients with MDD compared with controls across studies (Kaiser et al. 2015; Mulders et al. 2015). However, the studies included in these papers generally had small samples (typically < 30 patients) collected at a single site, and variable methodologies, limiting the generalizability of these findings. In addition, MDD itself is a heterogeneous disorder, which likely contributes to the variable findings (Goldberg 2011; Kessler et al. 2017). Several authors have called for large, multisite datasets to increase replicability (e.g., Button et al. 2013; Gong and He 2015).

Several studies have also investigated connectivity differences between patients with different treatment outcomes. Here, we focus on findings related to pharmacotherapy, which is the most common first-line treatment for MDD. Three reviews indicate that findings often include the same three networks as mentioned above, most commonly showing a decrease in connectivity within the DMN and an increase in connectivity within the CCN and between frontal (CCN) areas and limbic (SN) regions with treatment (Dichter et al. 2015; Gudayol-Ferre et al. 2015; Brakowski et al. 2017). However, the authors of these reviews emphasize that results vary greatly among studies and that similar issues to those raised in the previous paragraph prevent robust conclusions and limit translational relevance (Dichter et al. 2015; Gudayol-Ferre et al. 2015; Brakowski et al. 2017). Indeed, a recent meta-analysis was not able to detect any resting state connectivity patterns predictive of treatment response to pharmacotherapy (Long et al. 2020). They included only studies examining the association between symptom improvement with treatment and baseline brain-wide resting state connectivity at the voxel level, leaving six studies for their analysis of pharmacotherapy response predictors. In addition to underlining the variability in findings to date, this highlights the scarcity of studies looking specifically at baseline fMRI connectivity predictors of treatment outcomes, which could be especially informative for clinical practice.

In the current study, we made use of a relatively large, multisite resting state fMRI dataset to replicate and extend previous findings of connectivity alterations in patients with MDD compared with controls, and differences in connectivity prior to treatment among patients with differing outcomes. The dataset was collected as part of the Canadian Biomarker Integration Network in Depression 1 (CAN-BIND-1) study and included controls and patients diagnosed with MDD, who were treated for 16 weeks with escitalopram, and an add-on of aripiprazole after 8 weeks if symptoms decreased less than 50% (Lam et al. 2016; Kennedy et al. 2019). Depression symptoms were monitored using the Montgomery-Åsberg Depression Rating Scale (MADRS; Montgomery and Asberg 1979). Patients were divided into three groups: “early remitters” reached a score of ≤10 on the MADRS by week 8, “late remitters” reached this threshold after 16 weeks, and “non-remitters” did not reach remission by the end of the 16-week trial. Resting state fMRI data were recorded before the start of treatment. Connectivity between key nodes from the DMN, SN, and CCN, and the rest of the brain was compared between groups using partial least squares (PLS; McIntosh and Lobaugh 2004). We expected to replicate previous patient-control differences in connectivity within and between the DMN, SN, and CCN. Namely, we expected stronger connectivity within the DMN, especially anteriorly, between the CCN and DMN, and between the anterior DMN and SN, and weaker connectivity within the CCN, between the SN and the posterior DMN, and between the posterior DMN and CNN in patients with MDD compared with controls. In addition, we expected to find differences among patients with different treatment outcomes in these same networks.

Methods

Participants and Treatment

Study participants included 108 controls and 200 patients (70 and 126 females, 38 and 74 males, respectively) with a primary diagnosis of MDD, who were recruited from six academic health centres across Canada (University of British Columbia, UBC; University of Calgary, UCA; McMaster University, MCU; Centre for Addiction and Mental Health, CAM; Toronto General Hospital, TGH; Queen’s University, QNS) as part of the CAN-BIND-1 study (Lam et al. 2016; Kennedy et al. 2019). Briefly, patients were between 18 and 60 years of age, spoke sufficient English to complete the study, met the criteria for a major depressive episode according to the DSM-IV-TR, as assessed with the Mini International Neuropsychiatric Interview (MINI, Sheehan et al. 1998), had symptom scores of ≥24 on the MADRS (Montgomery and Asberg 1979), and their current episode lasted 3 months or longer. Patients were excluded if they met the diagnostic criteria of bipolar-I or -II disorder or any other primary psychiatric or personality disorder (except for generalized anxiety disorder and social anxiety disorder), experienced psychotic symptoms in the current episode, had a history of neurological disorders, head trauma, or other unstable medical conditions, had a high risk of suicide or a hypomanic switch, experienced substance dependence/abuse in the last 6 months, were currently pregnant or breastfeeding, showed previous nonresponse to four adequate pharmacotherapy interventions, had a previous unfavorable response to escitalopram or aripiprazole, or had any contraindications to MRI. Patients who had been taking antidepressant medication prior to participation went through a washout period (lasting at least 5 half-lives). Controls were between 18 and 60 years of age, had no psychiatric or unstable physical health diagnosis, no history of neurological disorders, head trauma, or other unstable medical conditions, and spoke sufficient English to complete the study.

Ethics approval was obtained from all participating centres. The ethics committees include: UBC Clinical Research Ethics Board (Vancouver); UCA Conjoint Health Research Ethics Board (Calgary); University Health Network Research Ethics Board (Toronto); CAM Research Ethics Board (Toronto); Hamilton Integrated Research Ethics Board (Hamilton); and QNS Health Sciences and Affiliated Teaching Hospitals Research Ethics Board (Kingston). The participants provided written, informed consent for all study procedures.

All patients were given escitalopram for the first 8 weeks of the study. The initial dose was 10 mg/d, which was increased to 20 mg/d if MADRS symptom scores did not drop by 20% after 2 weeks or 50% after 4 weeks. Patients who improved less than 50% over 8 weeks were given a flexibly dosed add-on of aripiprazole (2–10 mg/d) for an additional 8 weeks. Patients who improved 50% or more remained on their effective dose of escitalopram. Treating psychiatrists could prescribe lower doses for patients who did not tolerate higher doses, and the use of nonpsychotropic medications for stable conditions, nonprescription analgesics, supplementals, vitamins, and oral contraceptives were allowed at the discretion of the study psychiatrist.

Depression symptoms were assessed with the MADRS, using the structured interview guide to enhance reliability (Williams and Kobak 2008), every 2 weeks during treatment. For the group analyses, patients were assigned to one of three groups based on whether, and when, they reached remission, which was defined as having a MADRS score ≤ 10, roughly equivalent to the clinical consensus of a cut-off score of ≤7 on the Hamilton Rating Scale for Depression (Zimmerman et al. 2004). Remission status was chosen as the main outcome measure for this analysis because this is the ultimate treatment goal, as residual symptoms can lead to significant morbidity (Lecrubier 2002; Thase 2003; Dupuy et al. 2011). Patients who achieved remission at 8 weeks of treatment and maintained this at 16 weeks were considered early remitters (ER), patients who reached this threshold at 16 weeks were considered late remitters (LR), and patients whose MADRS scores were above 10 throughout the 16 weeks were considered non-remitters (NR). Patients who reached remission at 8 weeks, but then relapsed by week 16 (N = 6) were not included. An additional 64 participants were excluded due to: treatment not being initiated (N = 10), missing clinical data (N = 23), and missing fMRI data or poor fMRI data quality (N = 31), leaving a sample of 129 patients for analysis. From the control sample, 9 participants were excluded due to poor fMRI data quality. The characteristics of the patient and control groups, as well as ER, LR, and NR patient groups, are presented in Tables 1 and 2, respectively. Statistical differences were assessed in Excel, and any missing individual data points were replaced by the average of their allocated group (e.g., controls or ER). The characteristics for which there were missing data points are marked in Tables 1 and 2, and for each of these, there were no more than three individuals with missing data per group.

Table 1.

Characteristics of controls and patients with MDD (Mean ± SD) and their statistical differences

	Controls (N = 99)	Patients with MDD (N = 129)	Statistical difference
Sex	64 Female (65%), 35 male (35%)	81 Female (63%), 48 male (37%)	χ² (3) = 0.1, P = 0.77
Age (years)	33.0 ± 10.6 (range: 18–60)	35.1 ± 12.2 (range: 18–61)	t(223) = 1.4, P = 0.15
Site	5 CAM (5%), 16 MCU (16%), 12 UBC (12%), 18 TGH (18%), 12 QNS (12%), 34 UCA (34%)	5 CAM (4%), 25 MCU (19%), 46 UBC (36%), 20 TGH (16%), 9 QNS (7%), 24 UCA (19%)	χ² (5) = 20.0, P = 0.001^*
Education (years) ^a	18.5 ± 2.3	16.9 ± 2.1	t(203) = 5.4, P < 0.001^*
Race/Ethnicity ^b	1 Aboriginal (1%), 1 Arab (1%), 9 East Asian (9%), 3 Latin American/Hispanic (3%), 11 South Asian (11%), 3 Southeast Asian (3%), 69 White (70%), 4 Other (4%), 2 Prefer not to answer (2%)	2 Aboriginal (2%), 1 Arab (1%), 4 Black (3%), 8 East Asian (6%), 2 Jewish (2%), 9 Latin American/Hispanic (7%), 4 South Asian (3%), 5 Southeast Asian (4%), 103 White (80%), 4 Other (3%)	NA^c
Handedness	11 left (11%), 85 right (86%), 3 ambidextrous (3%)	12 left (9%), 113 right (88%), 4 ambidextrous (3%)	NA^c
Current marital status	55 never married (56%), 2 separated (2%), 27 married (27%), 5 divorced (5%), 9 domestic partnership (9%), 1 widowed (1%)	68 never married (53%), 10 separated (8%), 29 married (22%), 10 divorced (8%), 9 domestic partnership (7%), 3 widowed (2%)	NA^c

Open in a new tab

^*significant at P < 0.05. NA = not applicable.

^aOne missing data point from a patient was replaced by the group average.

^bParticipants were asked which of the presented race/ethnicity categories they most closely identified with. Categories were based on Canadian census questionnaires (Statistics Canada 2006).

^cStatistical comparison was not conducted due to one or more categories having insufficient samples to perform a reliable test.

Table 2.

Characteristics of ER, LR and NR to antidepressant pharmacotherapy (Mean ± SD), and their statistical differences

	ER (N = 36)	LR (N = 46)	NR (N = 47)	Statistical difference
Sex	23 female (64%), 13 male (36%)	30 female (65%), 16 male (35%)	28 female (60%), 19 male (40%)	χ² (3) = 0.3, P = 0.84
Age (years)	34.6 ± 11.7 (range: 21–59)	34.3 ± 12.1 (range: 18–59)	36.4 ± 12.9 (range: 18–61)	F(2, 126) = 0.4, P = 0.67
Site	1 CAM (3%), 8 MCU (22%), 16 UBC (44%), 3 TGH (8%), 2 QNS (6%), 6 UCA (2%)	1 CAM (2%), 11 MCU (24%), 12 UBC (26%), 8 TGH (17%), 4 QNS (9%), 10 UCA (22%)	3 CAM (6%), 6 MCU (13%), 18 UBC (38%), 9 TGH (19%), 3 QNS (6%), 8 UCA 17%)	NA^b
Education (years) ^a	17.1 ± 2.3	16.7 ± 2.1	16.9 ± 2.0	F(2, 126) = 0.2, P = 0.80
Race/Ethnicity ^b	1 Black (3%), 3 East Asian (8%), 2 Latin American/Hispanic (6%), 1 South Asian (3%), 1 Southeast Asian (3%), 28 White (78%), 2 Other (6%)	1 Aboriginal (2%), 1 Arab (2%), 2 East Asian (4%), 2 Jewish (4%), 4 Latin American/Hispanic (9%), 1 South Asian (2%), 1 Southeast Asian (2%), 36 White (78%), 1 Other (2%)	1 Aboriginal (2%), 3 Black (6%), 3 East Asian (6%), 3 Latin American/Hispanic (6%), 2 South Asian (4%), 3 Southeast Asian (6%), 39 White (83%), 1 Other (2%)	NA^c
Handedness	3 left (8%), 33 right (92%)	3 left (7%), 42 right (91%), 1 ambidextrous (2%)	6 left (13%), 38 right (81%), 3 ambidextrous (6%)	NA^c
Current marital status	18 never married (50%), 3 separated (8%), 9 married (25%), 3 divorced (8%), 2 domestic partnership (6%), 1 widowed (3%)	28 never married (61%), 4 separated (9%), 8 married (17%), 1 divorced (2%), 3 domestic partnership (7%), 2 widowed (4%)	22 never married (47%), 3 separated (6%), 12 married (26%), 6 divorced (13%), 4 domestic partnership (9%)	NA^c
Duration of current episode (months) ^a	25.5 ± 29.6 (range: 2–131)	23.0 ± 30.9 (range: 1–130)	37.2 ± 40.4 (range: 2–151)	F(2, 126) = 2.2, P = 0.11
Number of previous MDD episodes ^a	3.3 ± 3.0 (range: 0–13)	3.1 ± 3.1 (range: 0–15)	3.3 ± 4.1 (range: 0–20)	F(2, 126) = 0.04, P = 0.96
Baseline MADRS score	26.6 ± 4.3	30.9 ± 5.3	30.4 ± 5.6	F(2, 126) = 7.9, P = 0.001^*
MADRS after 8 weeks of treatment	4.8 ± 3.1	18.5 ± 5.6	23.4 ± 8.3	F(2, 126) = 94.4, P < 0.001^*
MADRS after 16 weeks of treatment	3.2 ± 3.0	5.5 ± 3.0	19.3 ± 6.3	F(2, 126) = 163.8, P < 0.001^*
Treatment regimen	36 ESC (100%)	13 ESC; 33 ESC + ARI	5 ESC; 42 ESC + ARI	χ² (2) = 72.3, P < 0.001^*

Open in a new tab

^*significant at P < 0.05. CAM = Centre for Mental Health and Addiction; ESC = escitalopram; ARI = aripiprazole; NA = not available.

^aMissing data points (max. 3 per group) from individual participants were replaced by the group average.

^bParticipants were asked which of the presented race/ethnicity categories they most closely identified with. Categories were based on Canadian census questionnaires (Statistics Canada 2006).

^cStatistical comparison was not conducted due to one or more categories having insufficient samples to perform a reliable test.

MRI Data Acquisition and Preprocessing

The structural and functional MRI scans used in this study were collected before the start of treatment. Resting-state fMRI scans were 10-min long, during which patients were asked to relax, have their eyes open, and focus on a fixation cross. A subgroup of 104 participants (64 patients and 40 controls) also performed an affective go/no-go task, lasting about 10 min. During the task, participants made go/no-go responses to stimuli presented on images containing affective content (see Albert et al. 2010 for a detailed description of the task). Task instructions, stimulus presentation, and task support materials were standardized across sites to ensure consistency. E-Prime version 2.0 (Psychology Software Tools) was used to record behavioral data. In the current study, go/no-go task data were only used for regions of interest (ROIs) or “seed” selection (see Seed Selection below).

Detailed information on the CAN-BIND MRI protocols can be found in MacQueen et al. (2019). Four different models of MRI scanners, all using 3.0 Tesla MRI systems with multicoil phased-array head coils, were used for data collection (Discovery MR750 3.0 T, GE Healthcare; Signa HDxt 3.0 T, GE Healthcare; MAGNETOM TrioTim, Siemens Healthcare; Intera 3.0 T, Philips Healthcare). To ensure data could be validly aggregated across sites, thorough quality control and standardization procedures were applied (see Lam et al. 2016 for details). The anatomical scans were obtained using a whole-brain T₁-weighted magnetization prepared gradient echo sequence with the following parameters: voxel dimensions (in mm) = 1 × 1 × 1, echo time (TE) = 2.2–2.9 ms, repetition time (TR) = 6.2–1900 ms, inversion time (TI) = 450–950 ms, flip angle = 8° or 15°, field of view (FOV) = 240–256 mm, matrix = 240 × 240 or 256 × 256, and number of slices = 176–192. The large range in TR was due to Siemens scanners reporting very different values for their proprietary sequence compared with GE and Philips; however, image parameters were visually optimized to produce similar contrast across scanners. To aid the confirmation of participant orientation, a small Vitamin E capsule was used as a stereotactic marker by placing it at the right temple during recording. Functional scans were acquired with a whole-brain T₂^*-sensitive blood oxygenation level-dependent (BOLD) echo planar imaging (EPI) series, with the following parameters: voxel dimensions (in mm) = 4 × 4 × 4, echo time (TE) = 25 or 30 ms, repetition time (TR) = 2 s, flip angle = 75° or 90°, FOV = 256 mm, matrix = 64 × 64, and number of slices = 34–40. All sites acquired slices in an interleaved order (even numbered slices before odd at QNS, odd before even numbered slices at MCU, CAM, TGH, UCA, and UBC).

The resting state fMRI data were preprocessed with the OPPNI pipeline (Churchill et al. 2015, 2017; software available at https://github.com/raamana/oppni) using the following steps: 1) the volume with the least amount of head displacement was determined using a principal component analysis and all volumes were registered to this volume with rigid-body motion correction (MOTCOR) via AFNI’s 3dvolreg; 2) significant outlier volumes were identified, removed, and replaced by interpolated values using neighboring volumes through censoring (CENSOR) as implemented in (Campbell et al. 2013); software available at: nitrc.org/projects/spikecor_fmri); 3) slice-timing correction (TIMECOR) was performed with Fourier interpolation via AFNI’s 3dTshift; 4) spatial smoothing across MRI scanners at different sites was matched using the 3dBlurToFWHM module in AFNI to smooth the fMRI images to the smoothness level of FWHM = 6 mm in three directions (x,y,z); 5) AFNI’s 3dAutomask algorithm was used to obtain a binary mask excluding nonbrain voxels using default parameter settings, and the resultant mask was applied to all EPI volumes prior to subsequent pipeline steps; 6) neuronal tissue masking was performed by estimating a probabilistic mask to reduce the variance contribution of nonneuronal tissues in the brain (macro-vasculature, ventricles) using the first part of the PHYCAA+ algorithm to estimate task-run and subject-specific neural tissue masks (Churchill and Strother 2013); software available at nitrc.org/projects/phycaa_plus); 7) several nuisance regressors (low-frequency temporal trends, head motion effects, and global signal modulations) were calculated and then regressed-out from the data concurrently via multiple linear regression (Carbonell et al. 2011; Churchill, Oder, et al. 2012; Churchill, Yourganov, et al. 2012); 8) physiological noise components were estimated and removed through data-driven physiological correction (PHYPLUS) using the second part of the data-driven PHYCAA+ algorithm (Churchill and Strother 2013); software at nitrc.org/projects/phycaa_plus); 9) low-pass filtering (LOWPASS) was carried out using a linear filter to remove BOLD frequencies above 0.10 Hz; and 10) spatial normalization to a structural template (sNORM) was carried out with all scans aligned to the MNI152 template (4-mm resolution) using two transformations (individual BOLD images to individual T₁-weighted image and individual T₁-weighted image to MNI template image, combined into one aggregated transform) via FSL’s FMRIB’s Linear Image Registration Tool module.

The same steps were performed for the task fMRI data, except that no corrections were applied for global signal modulations, physiological noise components, or higher frequency BOLD signals. After the OPPNI pipeline, the EPInorm strategy was used for additional spatial normalization on both the resting-state and task fMRI data, as this was shown to consistently reduce variability across participants, lower estimates for co-registration distances among participants and address effects of signal drop-out and distortion, including those stemming from phase encoding direction (Calhoun et al. 2017). Briefly, the data were registered directly to a cohort-specific EPI template using an affine followed by a nonlinear transformation. Only voxels in the gray matter (including cortical, subcortical, and cerebellar regions), as determined by a gray matter mask, were included in the analyses. To account for potential differences in signal quality and tissue coverage, voxels from the gray matter mask were excluded from analysis on the basis of signal-to-noise ratio (SNR, voxel-wise intensity mean over the imaging time-course, divided by the standard deviation) where voxels were excluded if the calculated SNR was less than 100 in at least 5% of participants (Drysdale et al. 2017). In our analysis, this resulted in the exclusion of 2227 of the 17 965 voxels originally in the gray matter mask. These excluded voxels were primarily located bilaterally at the outer edge of the brain, the orbitofrontal cortex, and the inferior temporal lobes (see Supplementary Fig. 1).

Seed Selection

Four seeds were selected for analysis from the ACC, the PCC, the insula, and the dlPFC, because of their prominent role in three important networks associated with MDD pathology and antidepressant treatment response, namely the anterior DMN, posterior DMN, SN, and CCN, respectively (Dichter et al. 2015; Gudayol-Ferre et al. 2015; Mulders et al. 2015; Brakowski et al. 2017). Two seeds were selected from the DMN, as the anterior and posterior sections of this network have been found to show distinct alterations in response to antidepressant treatment (Li et al. 2013).

The selection of specific seed coordinates was based on analyses contrasting task versus resting state in the subgroup of participants who performed the affective go/no-go task (McIntosh et al. 1997; Keightley et al. 2003; Grady et al. 2010; Bellana et al. 2016). This sample specific contrast was used to extract the optimal peak locations of the four seeds in the networks of interest for this sample. As the affective go/no-go task has been found to increase activation in both the SN and the CCN compared with rest (Shafritz et al. 2006; Chikazoe 2010), the voxels with the most stable increase in activity (as indicated by bootstrap ratios [BSRs], see Data Analysis with PLS) during the task within the insula and dlPFC were selected as seeds for the SN and CCN. In contrast, the voxels showing the most stable increase in signal during rest within the ACC and PCC were selected as the seeds for the anterior and posterior DMN. ROIs were then defined as a 12 × 12 × 12 mm cube (27 voxels in total) centered on the seed voxel for each region (ACC, PCC, insula, and dlPFC). Neighboring voxels that fell outside of the gray matter mask were not included. The signal averaged accross these voxels during rest was extracted for each individual and used to examine resting-state functional connectivity between the ROIs and the rest of the brain through correlation analyses.

This procedure was conducted twice to make the seed locations as representative as possible of the overall networks for the participants included in each of our two series of analyses: first for controls and patients together, to select the seeds for the patient-control comparisons, and then for patients only, to select the seeds for the comparisons between early, late, and NR. MDD and remission status were not considered in these seed selection analyses (i.e., participants were grouped together), to prevent seed selection from being influenced by group effects. Of note, the selected seeds were highly similar in both cases, except for the dlPFC seed (MNI coordinates patients and controls: ACC: [0, 48–4]; PCC: [0–60, 28]; Insula: [−40, 20, 0]; dlPFC: [−48, 16, 32]; MNI coordinates patients only: ACC: [0, 48–4]; PCC: [0–60, 28]; Insula: [−40, 20 4]; and dlPFC: [−52, 28, 28]).

Data Analysis with PLS

Statistical analysis of functional connectivity was performed using the PLS Graphic User Interface version 6.1311050 (Rotman Research Institute, http://www.rotman-baycrest.on.ca/pls) in MATLAB 2012a (The MathWorks, Inc.). PLS is a multivariate statistical technique that enables the detection of group and/or condition differences in the association between variables of interest and whole-brain signal patterns (McIntosh and Lobaugh 2004). By entering the BOLD signal from an ROI (averaged across ROI voxels over time within participant) as variables of interest, we were able to examine functional connectivity as quantified by the correlations between the average activity in a given network’s ROIs and all other voxels of the brain across participants. Thus, PLS examines the relationship between the signal in the ROI and the other voxels for the same participant in the larger context of group differences. PLS analyses identify latent variables (LVs), that is, connectivity patterns highlighting similarities and/or differences between groups (MDD and controls/ER, LR, and NR), that explain the largest amount of variance in the data. Each LV is made up of three components. The first is the singular value, which represents the strength of the effect expressed in the LV. The second contains the condition loadings, which indicate the contrast between groups as highlighted by the LV. The third component holds the element loadings, which describe where in the brain the connectivity differences identified by the LV are expressed.

The statistical assessment of LVs is done at two levels. First, the significance of the overall pattern is examined through permutation testing. Briefly, the data are randomly shuffled between groups and PLS analysis is then performed on the shuffled data for each permutation. LVs are considered significant when the singular value is more extreme than 95% of the singular values obtained from the shuffled data. We performed 1000 permutations per analysis. Second, bootstrap resampling is used to determine the consistency of the identified spatial pattern across participants. In this process, PLS analyses are repeated with different subsamples, building a distribution of loadings for each element depending on which participants are included. Bootstrap samples are created by randomly resampling the data with 50% replacement within groups, but maintaining the same number of data entries as in the main analysis (i.e., some participants are included more than once, while others are left out). These are then used to calculate BSRs by dividing the element loadings by the standard error of the distribution for each element. BSRs represent the stability of the spatial pattern showing the LV contrast and are similar to z-scores in that absolute BSR values ≥2 correspond to a confidence interval of ~95%. We performed bootstrap resampling 1000 times per analysis. As the element loadings are all calculated in one mathematical step, no correction for multiple comparisons is necessary (i.e., patterns of connectivity across the whole brain are tested at once instead of individual connections). In addition to P-values, we report the percent crossblock covariance explained (PCCE) by each LV, which is calculated by dividing the squared singular value of the LV by the sum of squared singular values of all LVs in the analysis. PCCE represents the amount of covariance in the data that are explained by the LV.

First, we ran four PLS analyses comparing controls and all patients (grouped together), one for each ROI, to replicate previously found connectivity alterations in MDD. Next, we performed four PLS analyses including only patients, now grouped according to remission status (ER, LR, and NR), one for each of the four ROIs. For each significant LV, the spatial pattern was examined both manually and using the automated anatomical labeling (AAL) atlas (Tzourio-Mazoyer et al. 2002) provided with the Fieldtrip toolbox (Oostenveld et al. 2011). Specifically, BSR thresholds were increased until the 10–15 clusters (involving 10 voxels or more) with the highest BSRs remained, and the areas in these clusters are reported in the results section. Manual inspection of these clusters was done using the MNI2TAL software developed by BioImage Suit and the Talairach atlas (Talairach and Tournoux 1988). In addition, the voxels in these clusters were automatically labeled using the AAL atlas. Voxels that did not receive a label and regions with fewer than 3 voxels were not included. As we were primarily interested in group differences, we only report significant LVs that distinguished groups here. Each analysis also identified an LV representing a network that was common among groups, which we report in the supplementary materials (Supplementary Figs. 2–5 and Supplementary Tables 1–8). These networks resemble the typical resting state networks of interest as described in previous reviews (e.g., Mulders et al. 2015; Dunlop et al. 2019), indicating that our seed selection procedure was successful in capturing the anterior/posterior DMN, SN, and CCN.

Results

Participants

There were no significant differences between patients and controls in the demographic characteristics presented in Table 1, except for controls having more years of education and a difference in the number of controls and patients tested at the different sites (Table 1). The three patient groups only differed in terms of their scores on the MADRS. By definition, ER had lower scores after 8 weeks of treatment, and both early and LR had lower scores after 16 weeks of treatment compared with NR (Table 2). However, ER also showed lower MADRS scores at the start of treatment (Table 2). This baseline difference was beyond our control because group assignment was based on MADRS scores after 8 or 16 weeks of treatment, but we did explore potential effects due to these differences (see subsection Correction for MADRS Baseline Differences).

Replication of Patient—Control Differences

Anterior Cingulate ROI

The PLS analysis comparing connectivity between the ACC ROI and the rest of the brain between patients with MDD and controls identified one significant LV (P = 0.010, PCCE = 22.8%) that highlighted differences between patients and controls. In line with our expectations, the pattern of differences included stronger connectivity between the ACC ROI and DMN regions (bilateral vmPFC, left PCC, and left angular gyrus), the SN (right insula and bilateral dorsal ACC), and parts of the CCN (bilateral dmPFC and dlPFC) in patients compared with controls (see Fig. 1A). In addition, this pattern included weaker connectivity between the ACC ROI and other parts of the CNN (bilateral PPC), the left extrastriate cortex and right premotor cortex, and stronger connectivity between the ACC ROI and the left orbitofrontal cortex, left middle temporal gyrus and bilateral middle cingulate cortex in patients compared with controls. The labels of all the regions included in the 10–15 most stable clusters in this contrast, as determined by bootstrapping (|BSR| ≥ 2.1), were identified using the AAL, and are presented in Supplementary Table 9.

Results from PLS analyses comparing patients with MDD and controls in terms of (A) connectivity between the ACC ROI (red sphere) and all other voxels, and (B) connectivity between the PCC ROI (red sphere) and all other voxels, projected onto a smoothed cortical surface using BrainNet Viewer (Xia et al. 2013). The ROI consisted of a cube of 27 voxels, centered around the seed voxel (ACC: [0, 48–4]; PCC: [0–60, 28]). The correlation bar graph shows group-dependent differences in the correlation between the ROI voxels and the areas identified in the brain image. The error bars indicate the 95% confidence intervals derived from bootstrap estimation. The brain image illustrates the areas that expressed this contrast most stably across participants, as determined by bootstrapping. Only the 10–15 clusters (>10 voxels) with the highest BSRs (ACC: |BSR| ≥ 2.1; PCC: |BSR| ≥ 2.0) are presented. Subcortical areas are projected onto the nearest surface, if significant. Cerebellar clusters are not illustrated, but they are listed in Supplementary Table 9 (ACC) and Supplementary Table 10 (PCC). Orange clusters indicate stronger, while purple clusters indicate weaker connectivity in patients as compared with controls. C = controls, P = patients with MDD, vmPFC = ventromedial prefrontal cortex.

Posterior Cingulate ROI

The PLS analysis of connectivity between the PCC ROI and the rest of the brain in patients with MDD and controls did not identify significant LVs differentiating the two groups. However, one LV showing group differences in connectivity approached significance (P = 0.058, PCCE = 21.7%). As expected, this pattern included stronger connectivity between the PCC ROI and the left precuneus (posterior DMN) and weaker connectivity between the PCC ROI and several CCN (bilateral dlPFC, right PPC) and SN (right dorsal ACC) areas in patients compared with controls (see Fig. 1B). In addition, this pattern included weaker connectivity between the PCC ROI and central motor areas (right SMA, bilateral premotor cortex), in patients compared with controls. The labels of all the regions included in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| ≥ 2.0), were identified using the AAL and are presented in Supplementary Table 10.

Insula ROI

The PLS analysis comparing patients with MDD and controls in connectivity between the ROI located in the left insula and the rest of the brain revealed one significant LV describing differences between patients and controls (P = 0.046, PCCE = 28.0%). As expected, the pattern of differences included weaker connectivity between the left insula ROI and bilateral PCC (posterior DMN) in patients compared with controls (see Fig. 2A). Contrary to our expectations, we did not identify increased connectivity between the left insula ROI and anterior DMN. In addition, the pattern included stronger connectivity between the left insula ROI and the left dlPFC (from the CCN) as well as bilateral occipital/posterior temporal regions in patients compared with controls. The labels of all the regions included in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| ≥ 2.0), were identified using the AAL and are presented in Supplementary Table 11.

Dorsolateral Prefrontal Cortex ROI

The PLS analysis of connectivity between the ROI located in the left dlPFC and the rest of the brain in patients and controls identified one significant LV revealing a pattern of group differences (P = 0.022, PCCE = 31.9%). In line with our expectations, this pattern included weaker connectivity between the left dlPFC ROI and other CCN areas (right dlPFC, right PPC) in patients (see Fig. 2B). At odds with our expectations, however, no connections between the left dlPFC ROI and DMN regions were part of the pattern of differences. In addition, this pattern included stronger connectivity between the left dlPFC ROI and SN regions (left dorsal ACC and bilateral insulae), and the left anterior PFC and weaker connectivity the left dlPFC ROI and visual processing areas (left extrastriate cortex, left fusiform gyrus) in patients compared with controls. The labels of all the regions included in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| ≥ 2.0), were identified using the AAL and are presented in Supplementary Table 12.

Early, Late and Nonremitter Comparison—Anterior Cingulate Seed

The PLS analysis examining connectivity between the ACC ROI and all voxels in the brain across the patient groups (ER, LR, and NR) identified one significant LV (P = 0.008, PCCE = 31.4%) that differentiated between ER and the other two groups (see bar graph in Fig. 3). This group specific contrast included stronger connectivity between the ACC ROI and bilateral parts of the cerebellum, the bilateral PCC, and the right occipital cortex (BA18), and weaker connectivity between the ACC ROI and the bilateral dlPFC, the left medial anterior PFC, right subgenual ACC, right caudate nucleus, right superior temporal gyrus, left insula, and the right extrastriate cortex (BA19) in ER compared with the other two groups (see Fig. 3). The labels of all the regions included in the 10–15 clusters showing the most stable group differences, as determined by bootstrapping (|BSR| ≥ 2.4), were determined using the AAL atlas and are presented in Table 3.

Results from PLS analysis comparing early, late and nonremitters in terms of connectivity between the ACC ROI (red sphere) and all other voxels, projected onto a smoothed cortical surface using BrainNet Viewer (Xia et al. 2013). The ROI consisted of a cube of 27 voxels around the seed voxel [0, 48–4]. The correlation bar graph shows group-dependent differences in the correlation between the ROI voxels and the areas identified in the brain image. The error bars indicate the 95% confidence intervals derived from bootstrap estimation. The brain image illustrates the areas that expressed this contrast most stably across participants, as determined by bootstrapping. Only the 10–15 clusters (>10 voxels) with the highest BSRs (|BSR| ≥ 2.4) are presented. Subcortical areas are projected onto the nearest surface, if significant. Cerebellar clusters are not illustrated, but they are listed in Table 3. Orange clusters indicate stronger, while purple clusters indicate weaker connectivity in early compared with late and NR.

Table 3.

Regions showing distinct connectivity with the ACC ROI in early, late and NR automatically labeled using the AAL atlas

Contrast	Cluster	Cluster size	Area	Number of voxels	Peak BSR	Peak MNI coordinates
						x	y	z
LV1: ER > LR & NR	1	18	L crus II of cerebellar hemisphere	16	3.94	−40	−72	−40
	2	10	R lingual gyrus	3	3.36	12	−48	0
			R lobule IV, V of cerebellar hemisphere	3	3.86	8	−48	0
			Lobule IV, V of vermis	3	3.37	0	−52	0
	3	10	R median cingulate/paracingulate gyri	3	3.05	4	−40	32
			L posterior cingulate gyrus	5	3.51	−4	−36	32
	4	16	R calcarine fissure	10	3.50	16	−64	16
			R Cuneus	3	2.63	16	−64	20
LV1: ER < LR & NR	5	153	L medial superior frontal gyrus	49	−5.10	−12	44	24
			R medial superior frontal gyrus	25	−6.24	4	48	0
			L medial superior frontal gyrus, orbital part	10	−5.68	0	48	−8
			R medial superior frontal gyrus, orbital part	12	−7.94	4	48	−4
			L anterior cingulate/paracingulate gyri	43	−4.41	0	48	0
			R anterior cingulate/paracingulate gyri	8	−4.15	4	52	8
	6	13	L inferior frontal gyrus, triangular part	13	−5.18	−48	28	16
	7	65	R dorsolateral superior frontal gyrus	12	−3.73	24	56	16
			R middle frontal gyrus	41	−5.14	40	44	12
			R inferior frontal gyrus, triangular part	10	−4.77	48	40	12
	8	33	L dorsolateral superior frontal gyrus	17	−4.94	−24	64	0
			L middle frontal gyrus	12	−4.40	−28	48	8
	10	22	R middle frontal gyrus	22	−4.77	36	44	28
	11	15	R superior temporal gyrus	13	−4.33	56	−4	−4
	12	11	L insula	9	−3.63	−36	16	−4
	13	11	R middle temporal gyrus	3	−3.08	40	−64	4
			R inferior temporal gyrus	3	−3.02	44	−64	−4
	14	17	L middle frontal gyrus	16	−3.31	−40	52	4

Open in a new tab

Note. Only the regions included in the 10–15 most stable clusters per LV are presented in this table (|BSRs| ≥ 2.4, Cluster Size ≥ 10). LR = late remitter, NR = non-remitter, R = right, L = left.

Early, Late, and Nonremitter Comparison—Posterior Cingulate Seed

The result of the PLS analysis investigating connectivity between the PCC ROI and all voxels in the brain identified one significant LV (P = 0.016, PCCE = 24.7%) which distinguished the three groups. Early and LR showed a pattern of connectivity that was most consistently expressed as stronger connectivity between the PCC ROI and the left precuneus/right PCC, parts of the right cerebellum, the bilateral superior temporal gyrus, bilateral orbital PFC, left dorsal and pregenual ACC, and the left fusiform gyrus, and weaker connectivity between the PCC ROI and the right superior temporal gyrus compared with NR. Importantly, ER exhibited this pattern of connectivity more strongly than LR (see bar graph in Fig. 4). All regions identified in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| > 2.3), were labeled using the AAL atlas and are listed in Table 4.

Results from PLS analysis comparing early, late and nonremitters in terms of connectivity between the PCC ROI (red sphere) and all other voxels, projected onto a smoothed cortical surface using BrainNet Viewer (Xia et al. 2013). The ROI consisted of a cube of 27 voxels, centered around the seed voxel [0–60, 28]. The correlation bar graph shows group-dependent differences in the correlation between the ROI voxels and the areas identified in the brain image. The error bars indicate the 95% confidence intervals derived from bootstrap estimation. The brain image illustrates the areas that expressed this contrast most stably across participants, as determined by bootstrapping. Only the 10–15 clusters (>10 voxels) with the highest BSRs (|BSR| ≥ 2.4) are presented. Subcortical areas are projected onto the nearest surface, if significant. Cerebellar clusters are not illustrated, but they are listed in Table 4. Orange clusters indicate stronger, while purple clusters indicate weaker connectivity in ER compared with late and nonremitters, as well as in late compared with nonremitters. pgACC = pregenual anterior cingulate cortex.

Table 4.

Regions showing distinct connectivity with the PCC ROI in early, late and NR automatically labeled using the AAL atlas

Contrast	Cluster	Cluster size	Area	Number of voxels	Peak BSR	Peak MNI coordinates
						x	y	z
LV1: ER < LR < NR	1	15	R superior temporal gyrus	12	3.49	52	−44	12
			R middle temporal gyrus	3	4.27	52	−44	8
LV1: ER > LR > NR	2	33	R fusiform gyrus	7	−4.92	32	−56	−20
			R crus I of cerebellar hemisphere	10	−3.17	40	−60	−28
			R lobule VI of cerebellar hemisphere	16	−5.25	32	−56	−24
	3	21	L dorsolateral superior frontal gyrus	3	−4.45	−32	56	0
			L middle frontal gyrus	12	−3.02	−36	60	8
			L middle frontal gyrus, orbital part	4	−3.31	−32	56	−8
	4	10	R superior temporal gyrus	7	−4.02	52	−4	−12
			R middle temporal gyrus	3	−3.35	48	−8	−16
	5	10	L posterior cingulate gyrus	6	−2.81	−8	−48	20
			L precuneus	3	−2.17	−8	−52	24
	6	16	R superior frontal gyrus, orbital part	8	−3.85	24	64	−8
			R middle frontal gyrus, orbital part	6	−3.09	24	60	−12
	7	19	L medial superior frontal gyrus, orbital part	6	−3.00	0	48	−12
			L anterior cingulate/paracingulate gyri	10	−3.71	0	44	−4
	8	14	L supplementary motor area	10	−3.50	0	16	48
			L medial superior frontal gyrus	3	−2.26	0	20	40
	9	10	L medial superior frontal gyrus	8	−3.39	−4	36	32
	10	11	L middle cingulate/paracingulate gyri	4	−2.52	0	−44	52
			L precuneus	7	−3.32	−4	−48	48
	11	12	L fusiform	12	−3.28	−32	−52	−20

Open in a new tab

Note. Only the regions included in the 10–15 most stable clusters per LV are presented in this table (|BSRs| ≥ 2.0, Cluster Size ≥ 10). LR = late remitter, NR = non-remitter, R = right, L = left.

Early, Late, and Nonremitter Comparison—Insula Seed

The PLS analysis of connectivity between the left insula ROI and the rest of the brain in early, late, and NR identified two significant LVs (LV1: P = 0.005; PCCE = 27.8%, LV2: P = 0.044, PCCE = 23.1%) differentiating between patient groups. The first differentiated late and ER, with ER showing stronger connectivity between the left insula ROI and the bilateral PCC/precuneus, bilateral fusiform gyrus, right PPC, bilateral anterior and medial PFC, left ACC, the left FEF, the left dlPFC, and the right angular gyrus, and weaker connectivity between the left insula ROI and the right insula, right inferior frontal gyrus, and parts of the right cerebellum, as compared with LR (see Fig. 5). All regions identified in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| ≥ 2.4), were labeled using the AAL atlas and are listed in Table 5.

Results from PLS analysis comparing early, late and nonremitters in terms of connectivity between the insula ROI (left—red sphere) and all other voxels, projected onto a smoothed cortical surface using BrainNet Viewer (Xia et al. 2013). The ROI consisted of a cube of 27 voxels, centered around the seed voxel [−40, 20, 4]. (A) The contrast and spatial pattern identified in the first LV, (B) contrast and spatial pattern identified for the second LV. The correlation bar graph shows group-dependent differences in the correlation between the ROI voxels and the areas identified in the brain image. The error bars indicate the 95% confidence intervals derived from bootstrap estimation. The brain image illustrates the areas that expressed this contrast most stably across participants, as determined by bootstrapping. Only the 10–15 clusters (>10 voxels) with the highest BSRs (|BSR| ≥ 2.4) are presented. Subcortical areas are projected onto the nearest surface, if significant. Cerebellar clusters are not illustrated, but they are listed in Table 5. In (A), orange clusters indicate stronger, while purple clusters indicate weaker connectivity in ER compared with LR. In (B), orange clusters indicate stronger, while purple clusters indicate weaker connectivity in nonremitters compared with early and LR.

Table 5.

Regions showing distinct connectivity with the left insula ROI in early, late and NR, automatically labeled using the AAL atlas

Contrast	Cluster	Cluster size	Area	Number of voxels	Peak BSR	Peak MNI coordinates
						x	y	z
LV1: ER > LR	1	22	L dorsolateral superior frontal gyrus	14	5.75	−24	56	24
			L middle frontal gyrus	8	4.45	−24	52	24
	2	14	L middle frontal gyrus	14	5.32	−44	20	48
	3	31	L dorsolateral superior frontal gyrus	12	5.04	−28	40	40
			L middle frontal gyrus	19	4.85	−28	32	40
	4	40	R middle occipital gyrus	9	4.01	44	−68	24
			R angular gyrus	13	4.94	44	−56	24
			R middle temporal gyrus	17	4.31	44	−56	20
	5	21	R postcentral gyrus	5	4.19	32	−44	60
			R superior parietal gyrus	15	4.72	32	−48	60
	6	13	L middle temporal gyrus	5	4.47	−56	−64	−4
			L inferior temporal gyrus	7	3.59	−56	−64	−8
	7	10	R inferior frontal gyrus, orbital part	4	3.03	40	24	−20
			R temporal pole: superior temporal gyrus	6	4.12	44	24	−20
	8	54	L posterior cingulate gyrus	7	3.04	−4	−52	28
			L superior parietal gyrus	11	3.96	−16	−64	40
			L precuneus	19	4.08	−8	−60	32
			R precuneus	15	3.92	8	−60	36
	9	12	R inferior temporal gyrus	12	4.05	48	−60	−12
	10	10	R superior parietal gyrus	10	3.96	20	−64	56
	11	25	L medial superior frontal gyrus	11	3.18	−8	52	16
			R medial superior frontal gyrus	4	3.61	8	56	12
			L anterior cingulate/paracingulate gyri	9	3.28	0	44	8
LV1: ER < LR	12	11	R inferior frontal gyrus, opercular part	3	−5.05	44	8	8
			R rolandic operculum	4	−5.58	48	0	8
			R insula	4	−3.19	48	8	4
	13	24	L insula	12	−3.48	−40	4	0
			L putamen	7	−5.09	−28	0	8
	14	15	R lobule IV, V of cerebellar hemisphere	5	−2.97	8	−56	−12
			Lobule IV, V of vermis	8	−3.97	0	−52	−12
LV2: ER & LR < NR	1	18	R lingual gyrus	4	2.99	20	−60	−4
			R fusiform gyrus	8	3.57	32	−56	−20
			R lobule VI of cerebellar hemisphere	6	3.98	32	−60	−24
	2	11	R rolandic operculum	6	3.94	40	−12	20
	3	10	R lingual gyrus	10	3.20	12	−88	−8
	4	14	R middle temporal gyrus	11	3.09	44	−68	4
LV2: ER & LR > NR	5	18	L medial superior frontal gyrus	8	−4.33	0	60	4
			L medial superior frontal gyrus, orbital part	7	−3.15	−4	64	−8
	6	19	R crus I of cerebellar hemisphere	8	−3.20	16	−76	−32
			R crus II of cerebellar hemisphere	8	−3.99	20	−80	−40
	7	10	R crus I of cerebellar hemisphere	7	−3.82	28	−64	−40
	8	11	L lobule VIII of cerebellar hemisphere	6	−3.45	−24	−60	−40
	9	13	R inferior frontal gyrus, orbital part	6	−2.92	40	24	−8
			R insula	5	−3.43	36	24	−4
	10	11	L putamen	7	−3.11	−28	8	−4
	11	14	L angular gyrus	9	−3.00	−52	−68	24
			L middle temporal gyrus	5	−2.88	−52	−68	20
	12	11	L hippocampus	3	−2.75	−32	−8	−28
			L parahippocampal gyrus	6	−2.86	−24	−8	−32

Open in a new tab

Note. Only the regions included in the 10–15 most stable clusters per LV are presented in this table (LV1: |BSR| ≥ 2.1, Cluster Size ≥ 10; LV2: |BSR| ≥ 2.4, Cluster Size ≥ 10). LR = late remitter, NR = non-remitter, R = right, L = left.

The second LV distinguished between remitters and NR. Specifically, NR showed stronger connectivity between the left insula ROI and parts of the right cerebellum, right extrastriate cortex, and right operculum, and weaker connectivity between the left insula ROI and bilateral parts of the cerebellum, right insula, left angular gyrus, left putamen, left mPFC, and left parahippocampal gyrus compared with both early and LR. All regions identified in the 10–15 most stable clusters, as determined by bootstrapping (|BSR| ≥ 2.1), were labeled using the AAL atlas and are listed in Table 5.

Early, Late, and Nonremitter Comparison—dlPFC ROI

The PLS analysis comparing early, late, and NR in connectivity between the left dlPFC ROI and the rest of the brain did not identify any significant LVs highlighting differences between patient groups.

Correction for MADRS Baseline Differences

As reported in the Participants section, ER had lower MADRS scores at baseline compared with late- and NR. To examine the influence of this baseline MADRS difference (i.e., depression severity) on our connectivity results, we regressed out the baseline MADRS scores from the dataset prior to analysis and reran the four PLS analyses comparing patient groups (one for each seed). Because permutation testing has been found to inflate Type I error rates when performed on residualized data (Winkler et al. 2020), we did not perform permutation testing on these PLS’s. Instead, we correlated the spatial connectivity patterns from these baseline MADRS-corrected analyses with those from the analyses reported above to assess their similarity. Based on guidelines put forth by (Akoglu 2018), we considered correlation coefficients above 0.8 to indicate strong to very strong similarity. All baseline MADRS-corrected analyses showed correlations above this threshold (r = 0.864–0.996, see Supplementary Table 13). This suggests that the differences in connectivity between early, late, and NR were not unduly influenced by the differences in baseline MADRS scores among groups.

Correction for Recording Site and Scanner Type

Another possible confound in our analyses was that data were collected at different sites and using different scanners. The CAN-BIND group utilizes rigorous standardization of MRI protocols across sites to minimize the influence of such potential confounds (MacQueen et al. 2019). Nonetheless, we examined the influence of recording site and scanner type effects in the manner described above for baseline MADRS scores. Here, we reran our patient—control as well as early, late, and nonremitter analyses for each seed.

Recording Site

The correlations of analyses corrected for recording site differences with our main analyses (both patient—control comparisons and early, late, and nonremitter comparisons) were high (r = 0.848–0.988), suggesting that recording site had little to no effect on our findings (see Supplementary Table 13).

Scanner Type

We found high correlations for the early, late, and nonremitter analyses corrected for scanner type (r = 0.850–0.962), but correlations were only medium for some of the patient—control analyses (ACC ROI: r = 0.674; INS ROI: r = 0.526; dlPFC ROI: r = 0.586; see Supplementary Table 13). Using visual inspection, we found that, while our findings were unaffected at the network level (i.e., we still found stronger connectivity within the [anterior] DMN, between the anterior DMN and the CCN and SN and between the CCN and SN, and weaker connectivity within the CCN in patients as compared with controls), there were some differences at the region level (see Supplementary Figs 6–8). Most notably, some areas only showed up unilaterally after scanner type was regressed from the data instead of bilaterally. In addition, connectivity between the ACC ROI and right insula was weaker in patients compared with controls, revealing stronger connectivity (ACC ROI—dACC) as well as weaker connectivity (ACC ROI—right insula) between the anterior DMN and SN in patients as compared with controls after the correction. Similarly, the connectivity between the left insula ROI and the PCC included both a cluster showing weaker connectivity, and a cluster showing stronger connectivity between these areas in patients.

Stability of Findings in Subsample

To assess the stability of our patient group comparisons, we ran additional PLS analyses on the largest single-site population of patients (16 ER, 12 LR, and 18 NR, from the UBC site). Specifically, we used nonrotated (i.e., hypothesis-driven) PLS to test whether the group differences we identified in the main analyses with the ACC, PCC, and insula ROI were present at the single-site subsample level (one each for the ACC and PCC, two for the insula ROI). In addition to inspecting the analysis results visually, we correlated the stable (|BSR| ≥ 2) element loadings of the main analyses with those of the subsample analyses to assess their similarity. As these were post-hoc analyses, we applied Bonferroni correction to the results. The results are summarized in Table 6. Two out of the four LVs tested were significant according to the Bonferroni corrected P-value (P < 0.013), one fell into the uncorrected window of significance (P = 0.043), and one showed a trend toward significance (P = 0.092). Correlations between stable element loadings ranged between 0.80 and 0.90.

Table 6.

Significance and correlation of subsample analyses for each seed ROI

Seed ROI	LV	Significance	Correlation with full sample analysis pattern
ACC ROI	1	P = 0.092	r = 0.80
PCC ROI	1	P = 0.007^**	r = 0.90
INS ROI	1	P = 0.010^**	r = 0.90
INS ROI	2	P = 0.043^*	r = 0.87

Open in a new tab

^**Significant at Bonferroni corrected P-value < 0.013.

^*Significant at uncorrected P-value < 0.05. ROI = region of interest, INS = insula.

Post-Hoc Analysis Examining Remission to Escitalopram Alone or Combined Escitalopram and Aripiprazole

To test whether the differences observed in insula connectivity between early and LR (see Early, Late and Nonremitter Comparison—Insula ROI) were related to differences in the medication regimen received (see Table 2), we ran a PLS analysis examining differences in (left) insula connectivity between remitters to escitalopram alone (N = 33) and remitters to combined escitalopram and aripiprazole (N = 49). This revealed one significant LV differentiating between the two groups (P < 0.001, PCCE = 37.4%), indicating that, before treatment, participants remitting to escitalopram alone most consistently showed stronger connectivity between the left insula ROI and bilateral medial and anterior PFC areas and the bilateral dorsal ACC (see Supplementary Table 14 and Supplementary Fig. 9) compared with participants who remitted to combined escitalopram and aripiprazole.

Discussion

We aimed to replicate and extend previous findings linking abnormal fMRI connectivity in three major resting state networks to the depressed state and antidepressant treatment outcomes using a large, well-characterized, multisite sample. Overall, we were able to replicate the more robust differences between patients with MDD and controls highlighted in previous reviews (Kaiser et al. 2015; Mulders et al. 2015). In addition, we identified differences in connectivity within and between these same networks within the patient population. ACC connectivity differentiated early from late and NR, PCC connectivity distinguished early, late, and NR, and left insula connectivity revealed both early/late remitter and remitter/nonremitter contrasts. Importantly, these differentiating connectivity patterns held in a single-site subsample, indicating a degree of stability of these group differences. As we looked at the relationship between connectivity before the start of treatment and remission status after 16 weeks of pharmacotherapy, these patterns might be useful in predicting antidepressant treatment outcomes.

In line with previous literature (Kaiser et al. 2015; Mulders et al. 2015), we found stronger connectivity within the DMN, most notably in the anterior subnetwork, and between the anterior DMN and the CCN and SN, and weaker connectivity within the CCN in patients as compared with controls. We also found weaker connectivity between the posterior DMN and the CCN and SN in patients, but this contrast only approached significance. Interestingly, altered connectivity between the anterior DMN and the SN, and the DMN and the CCN, was only apparent when using DMN regions as seeds (ACC and PCC ROIs). This could indicate that DMN abnormalities are stronger and thus most useful in distinguishing patients from controls. While the DMN certainly takes a prominent place in the literature on brain network alterations related to MDD, so do the two other networks investigated in this project. An alternative interpretation might be that DMN effects were stronger because we recorded data during resting state, when the DMN is typically active, while both the CCN and SN are typically more active during tasks (Nickerson 2018).

In addition to these expected effects, the connectivity patterns differentiating patients and controls included connections that have not been highlighted in previous reviews. Most interestingly, we identified stronger connectivity between CCN and SN regions in patients using both the CCN and SN seeds (dlPFC and insula ROIs, respectively). Few studies have found altered connectivity between the CCN and SN directly (e.g., Pang et al. 2018). However, these two networks are involved in emotion processing and mood regulation, which have been found to be abnormal in patients with MDD (e.g., Park et al. 2019); therefore, our finding of aberrant connectivity between them is not surprising. Mulders et al. (2015) mention that the SN is less well-defined during resting state (in comparison to the DMN and CCN), so effects might be harder to detect for this network, especially with small samples. Our finding of increased connectivity between the SN and CCN in a larger sample, together with the potential functional relevance of these connections, indicates that SN-CCN connectivity should be studied in more detail in future research. In addition, several regions outside of our networks of interest were highlighted by our analyses, including areas related to visual and motor processing. While generally given little attention, such regions have been observed in previous studies examining differences between patients with MDD and controls (Kaiser et al. 2015; Mulders et al. 2015), indicating that network alterations associated with MDD are more widespread than is generally highlighted.

Nonetheless, the three selected networks revealed important baseline connectivity differences between patients who reached remission early (within 8 weeks of treatment) or late (within 16 weeks), or who did not reach remission over the course of this study. ER displayed weaker connectivity within the anterior DMN, and between the ACC seed and regions from the CCN and SN compared with both late and NR. The ACC is often highlighted in studies on fMRI connectivity related to antidepressant treatment (e.g., see the reviews by Dunlop et al. 2019, and Brakowski et al. 2017 or the meta-analysis by Long et al. 2020), and has been proposed to play a central role in MDD pathology (Mayberg et al. 2005; Hamani et al. 2011; Pizzagalli 2011). While the subgenual part of the ACC is most often highlighted, the pregenual ACC, where our seed was located, has also been found to be part of the altered DMN in MDD and associated with treatment effects (Jing et al. 2020; Peng et al. 2020). Our finding that ACC connectivity with regions from both other networks differentiates patients with different treatment outcomes further supports this idea and highlights the potential of this feature for predicting remission in response to pharmacotherapy. Furthermore, the observation that LR showed the opposite effect indicates that ACC connectivity might be especially relevant for predicting early remission to pharmacotherapy.

The connectivity pattern of the posterior DMN seed differentiated all three groups, with ER showing the strongest connectivity within the posterior DMN and between the PCC seed and other cingulate areas, NR exhibiting the weakest connectivity between these regions, and LR exhibiting intermediate connectivity strengths. Previous studies have found both stronger (e.g., Korgaonkar et al. 2020) and weaker (e.g.Yan et al. 2019) connectivity within the DMN to be related to treatment outcomes. As we found stronger connectivity in the posterior, but weaker connectivity within the anterior DMN to be associated with (early) remission, our findings might be explained by the observation that the anterior and posterior DMN subnetworks respond differently to pharmacotherapy (Li et al. 2013). Li et al. (2013) found that DMN connectivity was increased in patients before treatment in both subnetworks, but only decreased with pharmacotherapy in the posterior DMN. A recent study by Goldstein-Piekarski et al. (2018) indicates that connectivity between the PCC and ACC/mPFC might be especially important for remission, with remitters showing higher connectivity prior to treatment compared with NR. This is in line with our findings of stronger baseline connectivity between the PCC and ACC in both early and LR compared with NR. Furthermore, previous studies have reported general treatment effects in DMN connectivity (Goldstein-Piekarski et al. 2018; Korgaonkar et al. 2020), and our separation of DMN connectivity into anterior and posterior sections further specifies that such general remission effects might be specific to posterior DMN connectivity, as our posterior DMN analysis revealed similar effects for early and LR (across remission timing and medication differences), while our anterior DMN analysis highlighted differences between early and LR.

With the left insula as seed (i.e., focusing on the SN), we identified two connectivity patterns: one distinguishing early and LR, and one distinguishing remitters (early and late) from NR. The first connectivity pattern included stronger connectivity in LR within the SN (between bilateral insulae) and weaker connectivity between the insula seed and regions from the other two networks (e.g., bilateral mPFC, left dlPFC). Connectivity between limbic SN and frontal CCN regions has often been highlighted as showing treatment-relevant effects (Dichter et al. 2015; Gudayol-Ferre et al. 2015; Brakowski et al. 2017). In fact, it has been proposed that the right anterior insula acts as a switch between the DMN and CCN and behaves aberrantly in the context of MDD (Manoliu et al. 2013). While our seed was located in the left anterior insula, our finding of weaker connectivity between the insula seed and the other two networks implies that, in our study as well, the interactions between the SN, DMN, and CCN were important for remission. This effect might be specifically related to remission to the aripiprazole add-on many LR received, as ER, who only received escitalopram for the duration of the study, showed the opposite effect. Our follow-up analysis comparing remitters to escitalopram alone and remitters to combined escitalopram and aripiprazole revealed overlap with the early/late remitter contrast. Most notably, both LR and remitters to combined escitalopram and aripiprazole showed weaker connectivity between the insula seed and the anterior DMN (i.e., bilateral mPFC), indicating that this connection is especially relevant for remission to combined escitalopram and aripiprazole. The second contrast further highlighted the importance of SN—DMN connectivity for both early and late remission, with stronger connectivity between the insula seed and the left mPFC and the left angular gyrus at baseline being associated with remission.

In addition, this second pattern included multiple cerebellar areas whose connectivity with the insula seed differentiated remitters and NR. Although the cerebellum was not the focus of our study, it showed up in most of our analyses as contributing to the differences between patients and controls, and different patient groups. While the cerebellum has traditionally been associated with motor function, this part of the brain is increasingly recognized as a structure playing an important role in a diverse set of processes and disorders, including the pathology and treatment of MDD (Konarski et al. 2005; Minichino et al. 2014; Depping et al. 2018). Indeed, most of our analyses revealed connections between our seeds and regions in the posterior lobes of the cerebellum (lobule VI-IX), which has more specifically been related to nonmotor functions (e.g., cognitive and affective; Depping et al. 2018). The remitter/nonremitter contrast in insula connectivity, which showed the most prominent involvement of the cerebellum in our analyses, revealed greater connectivity between the insula seed and right lobule VI, VIIA (crus I and II), and left lobule VIII of the cerebellum. Lobule VI has been associated with affective processing, and crus I and II of lobule VIIA have been related to the DMN and CCN, once again pointing toward the importance of networks involved with emotion processing and mood regulation for antidepressant treatment effects. Our analyses thus suggest that cerebellar connectivity could be important for treatment success prediction and should thus be studied in more detail.

Interestingly, the analysis with the dlPFC as the seed region (from the CCN) did not reveal any differences among the three patient groups. Previous findings of CCN connectivity have been inconsistent, with some proposing that MDD-related alterations in CCN connectivity might not be sensitive to treatment with antidepressant medication (Gudayol-Ferre et al. 2015). In contrast, Tozzi et al. (2020) found that the connectivity of the CCN during a response inhibition task was not only associated with treatment response in their relatively large sample but was also able to distinguish between responders to different types of pharmacotherapy, suggesting that there are treatment-related effects in CCN connectivity when measured during task conditions. Taken together, this body of work suggests that CCN connectivity might only be weakly related to pharmacotherapeutic effects when measured during resting state (e.g., through its connections with other networks).

To test the stability of our findings, we performed post-hoc analyses for the three seed regions (ACC, PCC, and insula) that revealed differences between early-, late-, and/or NR on the largest single-site dataset. Specifically, we wanted to exclude the possibility that our findings were unduly influenced by intersite variability (e.g., due to data being collected using different scanners). Half of the contrasts survived significance testing with correction for multiple comparisons, and all showed trends toward significance and revealed similar patterns of connectivity as observed in the main analyses. These findings suggest that the baseline differences we observed between our three patient groups have a degree of stability, are not dependent on a specific site/scanner setup, and are therefore more likely to generalize to other samples. The lack of significance of part of our post-hoc tests is likely related to the smaller sample size used in these analyses.

While our study addressed certain issues relating to reproducibility and generalizability, such as sample size and scanner differences, other factors fell outside the scope of this paper. For example, individual variation in the expression of MDD and response to antidepressant treatment among patients has also been proposed to play a role in the difficulty of finding robust neuroimaging biomarkers (Wise et al. 2014; Dichter et al. 2015). Although the bootstrapping procedure in PLS analyses ensures that the identified patterns of connectivity show a degree of stability across participants, this approach is not designed to determine if these findings might apply at an individual patient level. Thus, exploring whether these potential markers of treatment outcome translate to individual patients would be a worthwhile next step toward clinical applications. Similarly, many forms of antidepressant treatment are being investigated and it is currently largely unclear which effects are treatment specific or generally associated with better outcomes. Even when only considering pharmacotherapies, some studies have reported common effects across treatment with different medications (e.g., Goldstein-Piekarski et al. 2018; Korgaonkar et al. 2020), while others report medication-specific effects (e.g., Chin Fatt et al. 2020; Tozzi et al. 2020). Although our SN connectivity results and follow-up analysis comparing remitters to escitalopram alone and remitters to combined escitalopram and aripiprazole suggest that insula connectivity might distinguish remitters to these different treatment regimes, the design of our study did not allow for firm conclusions in this area. Other sources of variance, such as demographic and clinical characteristics and the use of different methodology, have also been related to variable findings (Dichter et al. 2015; Kaiser et al. 2015; Wise et al. 2014) and could be explored in more detail in future studies.

Limitations

Our study had several limitations that deserve consideration. First, the treatment regimen differed between ER (by definition recipients of escitalopram monotherapy), and among late and NR, who received either escitalopram alone or with adjunctive aripiprazole. While ER only received escitalopram, both late and NR included individuals who received either escitalopram alone or escitalopram and aripiprazole together, making it hard to disentangle specific medication effects. While suboptimal, this is common and difficult to avoid in clinical research on MDD, as the use of multiple medications and individually adjusted regimens are standard treatment practice. Second, the participants in our study, as is common in the field of neuroscience (Chiao and Cheon 2010), were drawn from what have been coined WEIRD (Western, Educated, Industrialized, Rich, and Democratic; Henrich et al. 2010) societies. More work involving participants and researchers from non-WEIRD populations should be conducted to extend our findings beyond WEIRD populations. Third, connectivity can be quantified in multiple ways, and using different approaches has been related to the variable findings reported in the literature. We chose seed-based correlation because it is the most commonly used method, and it has shown a greater degree of stability when compared with other metrics of connectivity (Fiecas et al. 2013), but it does have its own disadvantages. For example, by correlating a few seeds selected a priori with the rest of the voxels in the brain, our analyses focused on a limited set of connections, meaning that relevant connections could have been missed. In addition, even within the category of seed-based correlation, there are different approaches to calculate connectivity, and the one we used (across-subject cross-correlation) is different from the one employed by Fiecas et al. (Fiecas et al. 2013; within-subject cross-correlation, see Roberts et al. 2016 for details on the difference between these two approaches). Lastly, we selected the ACC as the anterior DMN seed, because this area appears to play a key role in the pathology and treatment of MDD and is most commonly selected as anterior DMN seed in seed-based connectivity studies of MDD (Pizzagalli 2011; Mulders et al. 2015). However, this region might only be included in the DMN in patients with MDD (Zhou et al. 2010) and might therefore have skewed our comparison of MDD and control participants for this seed.

Conclusion

Here, we replicated and extended abnormal patterns of fMRI connectivity in patients with MDD compared with controls and examined connectivity characteristics associated with remission status after 16 weeks of antidepressant treatment in patients. Our replication of connectivity differences between patients and controls confirmed the importance of the DMN, SN, and CCN for MDD pathology and revealed an additional finding of stronger connectivity between the SN and CCN in patients. Our patient group analyses further revealed multiple patterns of connectivity within and between the DMN, SN, and CCN that were associated with remission status. Interestingly, our work highlights previously identified features as well as new ones and further specifies their relationship to the timing of remission. Namely, in addition to confirming the importance of ACC connectivity for predicting treatment response, our analyses revealed that ACC connectivity could distinguish early from late remitters at baseline. Similarly, our analyses with the left insula as seed both supported the previously suggested importance of SN—DMN connectivity for treatment success and highlighted the novel finding of insula—mPFC connectivity potentially being related to dual pharmacotherapy. PCC connectivity showed a general remission effect, indicating that posterior DMN connectivity might be a treatment independent predictor of remission. Although reproducibility and generalizability are still major concerns in current neuroimaging research, the fact that we were able to replicate previous single-site findings in a larger, multisite sample (with MRI data collected from different Tesla scanners) and demonstrate stability in the patterns differentiating patients based on remission status in a single-site subsample of our data is encouraging and supports the potential of our findings for future clinical use.

Supplementary Material

CANBIND_Supplementary_materials_final_bhab286

Click here for additional data file.^{(7.9MB, pdf)}

Funding and Conflict of Interest

CAN-BIND is an Integrated Discovery Program carried out in partnership with, and financial support from, the Ontario Brain Institute, an independent non-profit corporation, funded partially by the Ontario government. The opinions, results and conclusions are those of the authors and no endorsement by the Ontario Brain Institute is intended or should be inferred. Additional funding is provided by the Canadian Institutes of Health Research (CIHR), Lundbeck, and Servier. Funding and/or in-kind support is also provided by the investigators’ universities and academic institutions. All study medications are independently purchased at wholesale market values. We acknowledge support to G.W. in the form of a Mamdani Family Foundation Graduate Scholarship and Alberta Graduate Excellence Scholarship (AGES: International). R.M. has received consulting and speaking honoraria from AbbVie, Allergan, Janssen, KYE, Lundbeck, Otsuka, and Sunovion, and research grants from CAN-BIND, CIHR, Janssen, Lallemand, Lundbeck, Nubiyota, OBI, and OMHF. R.W.L. has received honoraria or research funds from Allergan, Manuscript File 2Asia-Pacific Economic Cooperation, BC Leading Edge Foundation, CIHR, CANMAT, Canadian Psychiatric Association, Hansoh, Healthy Minds Canada, Janssen, Lundbeck, Lundbeck Institute, Michael Smith Foundation for Health Research, MITACS, Ontario Brain Institute, Otsuka, Pfizer, St. Jude Medical, University Health Network Foundation, and VGH-UBCH Foundation. B.N.F. received a research grant from Pfizer, outside of this work. S.R. has received grant funding from the Ontario Brain Institute and Canadian Institutes of Health Research and holds a patent Teneurin C-Terminal Associated Peptides (TCAP) and methods and uses thereof. Inventors: David Lovejoy, R.B. Chewpoy, Dalia Barsyte, Susan Rotzinger. S.H.K. has received research funding or honoraria from the following sources: Abbott, Alkermes, Allergan, BMS, Brain Canada, Canadian Institutes for Health Research (CIHR), Janssen, Lundbeck, Lundbeck Institute, Ontario Brain Institute, Ontario Research Fund (ORF), Otsuka, Pfizer, Servier, Sunovion, and Xian-Janssen. S.C.S. reports partial support from Canadian Biomarker Integration Network in Depression and CIHR (MOP 137097) grants during the conduct of the study, and grants from Ontario Brain Institute, Canadian Foundation for Innovation and Brain Canada, outside the submitted work. He is also the Chief Scientific Officer of the neuroimaging data analysis company ADMdx, Inc (www.admdx.com), which specializes in brain image analysis to enable diagnosis, prognosis, and drug effect detection for Alzheimer disease and various other forms of dementia. J.K.H., S.H., A.D.D., M.Z., G.B.H., D.J.M., G.M.M., and A.B.P. have no funding or potential conflicts of interest to disclose.

Notes

We would like to thank Keith Ho for his help with organizing the clinical data, and all the patients, controls, and health care workers involved in this project for the generous contribution of their time and their participation in this work.

Authors’ Contributions

G.W.: Study concept, data analysis and interpretation, writing (original draft), writing (review, editing, and approval); J.K.H.: Data collection, data curation, writing (review, editing, and approval); S.H.: Data collection, data curation, writing (review, editing, and approval); A.D.D.: Data collection, data curation, writing (review, editing, and approval); M.Z.: Data collection, data curation, writing (review, editing, and approval); S.R.A.: Data curation, writing (review, editing, and approval); R.M.: Study concept, data collection, writing (review, editing, and approval); R.W.L.: Study concept, data collection, writing (review, editing, and approval); B.N.F.: Study concept, data collection, data curation, writing (review, editing, and approval); G.B.H.: Study concept, data collection, data curation, writing (review, editing, and approval); D.J.M.: Study concept, data collection, data curation, writing (review, editing, and approval); S.R.: Study concept, data collection, data curation, writing (review, editing, and approval); S.H.K.: Study concept, data collection, writing (review, editing, and approval); S.C.S.: Data curation, writing (review, editing, and approval); G.M.M.: Study concept, data collection, data curation; A.B.P.: Study concept, data analysis, writing (original draft), writing (review, editing, and approval).

Data Availability Statement

The data analyzed in this paper are not publicly available but may be made available if permission is granted by the CAN-BIND Investigator’s Team and Ontario Brain Institute upon reasonable request.

References

Akoglu H. 2018. User's guide to correlation coefficients. Turk J Emerg Med. 18(3):91–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
Albert J, Lopez-Martin S, Carretie L. 2010. Emotional context modulates response inhibition: neural and behavioral data. Neuroimage. 49(1):914–921. [DOI] [PubMed] [Google Scholar]
Bellana B, Liu Z, Anderson JAE, Moscovitch M, Grady CL. 2016. Laterality effects in functional connectivity of the angular gyrus during rest and episodic retrieval. Neuropsychologia. 80:24–34. [DOI] [PubMed] [Google Scholar]
Boly M, Phillips C, Tshibanda L, Vanhaudenhuyse A, Schabus M, Dang-Vu TT, Moonen G, Hustinx R, Maquet P, Laureys S. 2008. Intrinsic brain activity in altered states of consciousness: how conscious is the default mode of brain function? Ann N Y Acad Sci. 1129:119–129. [DOI] [PMC free article] [PubMed] [Google Scholar]
Brakowski J, Spinelli S, Dorig N, Bosch OG, Manoliu A, Holtforth MG, Seifritz E. 2017. Resting state brain network function in major depression - depression symptomatology, antidepressant treatment effects, future research. J Psychiatr Res. 92:147–159. [DOI] [PubMed] [Google Scholar]
Button KS, Ioannidis JP, Mokrysz C, Nosek BA, Flint J, Robinson ES, Munafo MR. 2013. Power failure: why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci. 14(5):365–376. [DOI] [PubMed] [Google Scholar]
Calhoun VD, Wager TD, Krishnan A, Rosch KS, Seymour KE, Nebel MB, Mostofsky SH, Nyalakanai P, Kiehl K. 2017. The impact of T1 versus EPI spatial normalization templates for fMRI data analyses. Hum Brain Mapp. 38(11):5331–5342. [DOI] [PMC free article] [PubMed] [Google Scholar]
Campbell KL, Grigg O, Saverino C, Churchill N, Grady CL. 2013. Age differences in the intrinsic functional connectivity of default network subsystems. Front Aging Neurosci. 5:73. [DOI] [PMC free article] [PubMed] [Google Scholar]
Carbonell F, Bellec P, Shmuel A. 2011. Global and system-specific resting-state fMRI fluctuations are uncorrelated: principal component analysis reveals anti-correlated networks. Brain Connect. 1(6):496–510. [DOI] [PMC free article] [PubMed] [Google Scholar]
Chiao JY, Cheon BK. 2010. The weirdest brains in the world. Behav Brain Sci. 33(2–3):88–90. [DOI] [PubMed] [Google Scholar]
Chikazoe J. 2010. Localizing performance of go/no-go tasks to prefrontal cortical subregions. Curr Opin Psychiatry. 23(3):267–272. [DOI] [PubMed] [Google Scholar]
Chin Fatt CR, Jha MK, Cooper CM, Fonzo G, South C, Grannemann B, Carmody T, Greer TL, Kurian B, Fava Met al. 2020. Effect of intrinsic patterns of functional brain connectivity in moderating antidepressant treatment response in major depression. Am J Psychiatry. 177(2):143–154. [DOI] [PubMed] [Google Scholar]
Churchill NW, Oder A, Abdi H, Tam F, Lee W, Thomas C, Ween JE, Graham SJ, Strother SC. 2012. Optimizing preprocessing and analysis pipelines for single-subject fMRI. I. Standard temporal motion and physiological noise correction methods. Hum Brain Mapp. 33(3):609–627. [DOI] [PMC free article] [PubMed] [Google Scholar]
Churchill NW, Raamana P, Spring R, Strother SC. 2017. Optimizing fMRI preprocessing pipelines for block-design tasks as a function of age. Neuroimage. 154:240–254. [DOI] [PubMed] [Google Scholar]
Churchill NW, Spring R, Afshin-Pour B, Dong F, Strother SC. 2015. An automated, adaptive framework for optimizing preprocessing pipelines in task-based functional MRI. PLoS One. 10(7):e0131520. [DOI] [PMC free article] [PubMed] [Google Scholar]
Churchill NW, Strother SC. 2013. PHYCAA+: an optimized, adaptive procedure for measuring and controlling physiological noise in BOLD fMRI. Neuroimage. 82:306–325. [DOI] [PubMed] [Google Scholar]
Churchill NW, Yourganov G, Oder A, Tam F, Graham SJ, Strother SC. 2012. Optimizing preprocessing and analysis pipelines for single-subject fMRI: 2. Interactions with ICA, PCA, task contrast and inter-subject heterogeneity. PLoS One. 7(2):e31147. [DOI] [PMC free article] [PubMed] [Google Scholar]
Cuijpers P, Stringaris A, Wolpert M. 2020. Treatment outcomes for depression: challenges and opportunities. Lancet Psychiatry. 7(11):925–927. [DOI] [PubMed] [Google Scholar]
Depping MS, Schmitgen MM, Kubera KM, Wolf RC. 2018. Cerebellar contributions to major depression. Front Psych. 9:634. [DOI] [PMC free article] [PubMed] [Google Scholar]
Dichter GS, Gibbs D, Smoski MJ. 2015. A systematic review of relations between resting-state functional-MRI and treatment response in major depressive disorder. J Affect Disord. 172:8–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
Drysdale AT, Grosenick L, Downar J, Dunlop K, Mansouri F, Meng Y, Fetcho RN, Zebley B, Oathes DJ, Etkin A. 2017. Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat Med. 23(1):28. [DOI] [PMC free article] [PubMed] [Google Scholar]
Dunlop K, Talishinsky A, Liston C. 2019. Intrinsic brain network biomarkers of antidepressant response: a review. Curr Psychiatry Rep. 21(9):87. [DOI] [PMC free article] [PubMed] [Google Scholar]
Dupuy JM, Ostacher MJ, Huffman J, Perlis RH, Nierenberg AA. 2011. A critical review of pharmacotherapy for major depressive disorder. Int J Neuropsychopharmacol. 14(10):1417–1431. [DOI] [PubMed] [Google Scholar]
Fiecas M, Ombao H, Lunen D, Baumgartner R, Coimbra A, Feng D. 2013. Quantifying temporal correlations: a test-retest evaluation of functional connectivity in resting-state fMRI. Neuroimage. 65:231–241. [DOI] [PubMed] [Google Scholar]
Goldberg D. 2011. The heterogeneity of “major depression.”. World Psychiatry. 10(3):226. [DOI] [PMC free article] [PubMed] [Google Scholar]
Goldstein-Piekarski AN, Staveland BR, Ball TM, Yesavage J, Korgaonkar MS, Williams LM. 2018. Intrinsic functional connectivity predicts remission on antidepressants: a randomized controlled trial to identify clinically applicable imaging biomarkers. Transl Psychiatry. 8(1):57. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gong Q, He Y. 2015. Depression, neuroimaging and connectomics: a selective overview. Biol Psychiatry. 77(3):223–235. [DOI] [PubMed] [Google Scholar]
Grady CL, Protzner AB, Kovacevic N, Strother SC, Afshin-Pour B, Wojtowicz M, Anderson JA, Churchill N, McIntosh AR. 2010. A multivariate analysis of age-related differences in default mode and task-positive networks across multiple cognitive domains. Cereb Cortex. 20(6):1432–1447. [DOI] [PMC free article] [PubMed] [Google Scholar]
Gudayol-Ferre E, Pero-Cebollero M, Gonzalez-Garrido AA, Guardia-Olmos J. 2015. Changes in brain connectivity related to the treatment of depression measured through fMRI: a systematic review. Front Hum Neurosci. 9:582. [DOI] [PMC free article] [PubMed] [Google Scholar]
Hamani C, Mayberg H, Stone S, Laxton A, Haber S, Lozano AM. 2011. The subcallosal cingulate gyrus in the context of major depression. Biol Psychiatry. 69(4):301–308. [DOI] [PubMed] [Google Scholar]
Henrich J, Heine SJ, Norenzayan A. 2010. Most people are not WEIRD. Nature. 466(7302):29. [DOI] [PubMed] [Google Scholar]
Jing Y, Zhao N, Deng XP, Feng ZJ, Huang GF, Meng M, Zang YF, Wang J. 2020. Pregenual or subgenual anterior cingulate cortex as potential effective region for brain stimulation of depression. Brain Behav. 10(4):e01591. [DOI] [PMC free article] [PubMed] [Google Scholar]
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 Psychiat. 72(6):603–611. [DOI] [PMC free article] [PubMed] [Google Scholar]
Keightley ML, Seminowicz DA, Bagby RM, Costa PT, Fossati P, Mayberg HS. 2003. Personality influences limbic-cortical interactions during sad mood induction. Neuroimage. 20(4):2031–2039. [DOI] [PubMed] [Google Scholar]
Kennedy SH, Lam RW, Rotzinger S, Milev RV, Blier P, Downar J, Evans KR, Farzan F, Foster JA, Frey BNet al. 2019. Symptomatic and functional outcomes and early prediction of response to escitalopram monotherapy and sequential adjunctive aripiprazole therapy in patients with major depressive disorder: a CAN-BIND-1 report. J Clin Psychiatry. 80(2):18m12202. [DOI] [PubMed] [Google Scholar]
Kessler R, Van Loo H, Wardenaar K, Bossarte R, Brenner L, Ebert D, Jonge P, Nierenberg A, Rosellini A, Sampson N. 2017. Using patient self-reports to study heterogeneity of treatment effects in major depressive disorder. Epidemiol Psychiatr Sci. 26(1):22–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
Konarski JZ, McIntyre RS, Grupp LA, Kennedy SH. 2005. Is the cerebellum relevant in the circuitry of neuropsychiatric disorders? J Psychiatry Neurosci. 30(3):178–186. [PMC free article] [PubMed] [Google Scholar]
Korgaonkar MS, Goldstein-Piekarski AN, Fornito A, Williams LM. 2020. Intrinsic connectomes are a predictive biomarker of remission in major depressive disorder. Mol Psychiatry. 25(7):1537–1549. [DOI] [PMC free article] [PubMed] [Google Scholar]
Lam RW, Milev R, Rotzinger S, Andreazza AC, Blier P, Brenner C, Daskalakis ZJ, Dharsee M, Downar J, Evans KRet al. 2016. Discovering biomarkers for antidepressant response: protocol from the Canadian biomarker integration network in depression (CAN-BIND) and clinical characteristics of the first patient cohort. BMC Psychiatry. 16:105. [DOI] [PMC free article] [PubMed] [Google Scholar]
Lecrubier Y. 2002. How do you define remission. Acta Psychiatr Scand Suppl. 106:7–11. [DOI] [PubMed] [Google Scholar]
Li B, Liu L, Friston KJ, Shen H, Wang L, Zeng LL, Hu D. 2013. A treatment-resistant default mode subnetwork in major depression. Biol Psychiatry. 74(1):48–54. [DOI] [PubMed] [Google Scholar]
Long Z, Du L, Zhao J, Wu S, Zheng Q, Lei X. 2020. Prediction on treatment improvement in depression with resting state connectivity: a coordinate-based meta-analysis. J Affect Disord. 276:62–68. [DOI] [PubMed] [Google Scholar]
MacQueen GM, Hassel S, Arnott SR, Jean A, Bowie CR, Bray SL, Davis AD, Downar J, Foster JA, Frey BNet al. 2019. The Canadian biomarker integration network in depression (CAN-BIND): magnetic resonance imaging protocols. J Psychiatry Neurosci. 44(4):223–236. [DOI] [PMC free article] [PubMed] [Google Scholar]
Manoliu A, Meng C, Brandl F, Doll A, Tahmasian M, Scherr M, Schwerthoffer D, Zimmer C, Forstl H, Bauml Jet al. 2013. Insular dysfunction within the salience network is associated with severity of symptoms and aberrant inter-network connectivity in major depressive disorder. Front Hum Neurosci. 7:930. [DOI] [PMC free article] [PubMed] [Google Scholar]
Mayberg HS, Lozano AM, Voon V, McNeely HE, Seminowicz D, Hamani C, Schwalb JM, Kennedy SH. 2005. Deep brain stimulation for treatment-resistant depression. Neuron. 45(5):651–660. [DOI] [PubMed] [Google Scholar]
McIntosh AR, Lobaugh NJ. 2004. Partial least squares analysis of neuroimaging data: applications and advances. Neuroimage. 23:S250–S263. [DOI] [PubMed] [Google Scholar]
McIntosh AR, Nyberg L, Bookstein FL, Tulving E. 1997. Differential functional connectivity of prefrontal and medial temporal cortices during episodic memory retrieval. Hum Brain Mapp. 5(4):323–327. [DOI] [PubMed] [Google Scholar]
Minichino A, Bersani FS, Trabucchi G, Albano G, Primavera M, Delle Chiaie R, Biondi M. 2014. The role of cerebellum in unipolar and bipolar depression: a review of the main neurobiological findings. Riv Psichiatr. 49(3):124–131. [DOI] [PubMed] [Google Scholar]
Montgomery SA, Asberg M. 1979. A new depression scale designed to be sensitive to change. Br J Psychiatry. 134:382–389. [DOI] [PubMed] [Google Scholar]
Mulders PC, Eijndhoven PF, Schene AH, Beckmann CF, Tendolkar I. 2015. Resting-state functional connectivity in major depressive disorder: a review. Neurosci Biobehav Rev. 56:330–344. [DOI] [PubMed] [Google Scholar]
Nickerson LD. 2018. Replication of resting state-task network correspondence and novel findings on brain network activation during task fMRI in the human connectome project study. Sci Rep. 8(1):17543. [DOI] [PMC free article] [PubMed] [Google Scholar]
Oostenveld R, Fries P, Maris E, Schoffelen J-M. 2011. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci. 2011:156869. [DOI] [PMC free article] [PubMed] [Google Scholar]
Pang Y, Chen H, Wang Y, Long Z, He Z, Zhang H, Liao W, Cui Q, Chen H. 2018. Transdiagnostic and diagnosis-specific dynamic functional connectivity anchored in the right anterior insula in major depressive disorder and bipolar depression. Prog Neuropsychopharmacol Biol Psychiatry. 85:7–15. [DOI] [PubMed] [Google Scholar]
Park C, Rosenblat JD, Lee Y, Pan Z, Cao B, Iacobucci M, McIntyre RS. 2019. The neural systems of emotion regulation and abnormalities in major depressive disorder. Behav Brain Res. 367:181–188. [DOI] [PubMed] [Google Scholar]
Peng X, Wu X, Gong R, Yang R, Wang X, Zhu W, Lin P. 2020. Sub-regional anterior cingulate cortex functional connectivity revealed default network subsystem dysfunction in patients with major depressive disorder. Psychol Med. 51(10):1–9. [DOI] [PubMed] [Google Scholar]
Pizzagalli DA. 2011. Frontocingulate dysfunction in depression: toward biomarkers of treatment response. Neuropsychopharmacology. 36(1):183–206. [DOI] [PMC free article] [PubMed] [Google Scholar]
Poldrack RA, Baker CI, Durnez J, Gorgolewski KJ, Matthews PM, Munafo MR, Nichols TE, Poline JB, Vul E, Yarkoni T. 2017. Scanning the horizon: towards transparent and reproducible neuroimaging research. Nat Rev Neurosci. 18(2):115–126. [DOI] [PMC free article] [PubMed] [Google Scholar]
Raichle ME. 2015. The brain's default mode network. Annu Rev Neurosci. 38:433–447. [DOI] [PubMed] [Google Scholar]
Rayner G, Jackson G, Wilson S. 2016. Cognition-related brain networks underpin the symptoms of unipolar depression: evidence from a systematic review. Neurosci Biobehav Rev. 61:53–65. [DOI] [PubMed] [Google Scholar]
Roberts RP, Hach S, Tippett LJ, Addis DR. 2016. The Simpson's paradox and fMRI: similarities and differences between functional connectivity measures derived from within-subject and across-subject correlations. Neuroimage. 135:1–15. [DOI] [PubMed] [Google Scholar]
Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, Niederehe G, Thase ME, Lavori PW, Lebowitz BDet al. 2006. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR*D report. Am J Psychiatry. 163(11):1905–1917. [DOI] [PubMed] [Google Scholar]
Seeley WW, Menon V, Schatzberg AF, Keller J, Glover GH, Kenna H, Reiss AL, Greicius MD. 2007. Dissociable intrinsic connectivity networks for salience processing and executive control. J Neurosci. 27(9):2349–2356. [DOI] [PMC free article] [PubMed] [Google Scholar]
Shafritz KM, Collins SH, Blumberg HP. 2006. The interaction of emotional and cognitive neural systems in emotionally guided response inhibition. Neuroimage. 31(1):468–475. [DOI] [PubMed] [Google Scholar]
Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E, Hergueta T, Baker R, Dunbar GC. 1998. The Mini-international neuropsychiatric interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry. 59(Suppl 20):22–33quiz 34-57. [PubMed] [Google Scholar]
Statistics Canada . Ethnic Origin Reference Guide, 2006Census. Catalogue no. 97–562-GWE2006025, Ottawa, ON, Canada. http://www12.statcan.gc.ca/census-recensement/2006/ref/rp-guides/ethnic-ethnique-eng.cfm. [Google Scholar]
Talairach J, Tournoux P. 1988. Co-planar stereotaxic atlas of the human brain. New York: Thieme Medical Publishers, Inc. [Google Scholar]
Thase ME. 2003. Achieving remission and managing relapse in depression. J Clin Psychiatry. 64(Suppl 18):3–7. [PubMed] [Google Scholar]
Tozzi L, Goldstein-Piekarski AN, Korgaonkar MS, Williams LM. 2020. Connectivity of the cognitive control network during response inhibition as a predictive and response biomarker in major depression: evidence from a randomized clinical trial. Biol Psychiatry. 87(5):462–472. [DOI] [PMC free article] [PubMed] [Google Scholar]
Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, Mazoyer B, Joliot M. 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage. 15(1):273–289. [DOI] [PubMed] [Google Scholar]
Williams JB, Kobak KA. 2008. Development and reliability of a structured interview guide for the montgomery asberg depression rating scale (SIGMA). Br J Psychiatry. 192(1):52–58. [DOI] [PubMed] [Google Scholar]
Winkler AM, Renaud O, Smith SM, Nichols TE. 2020. Permutation inference for canonical correlation analysis. Neuroimage. 220:117065. [DOI] [PMC free article] [PubMed] [Google Scholar]
Wise T, Cleare AJ, Herane A, Young AH, Arnone D. 2014. Diagnostic and therapeutic utility of neuroimaging in depression: an overview. Neuropsychiatr Dis Treat. 10:1509–1522. [DOI] [PMC free article] [PubMed] [Google Scholar]
Xia M, Wang J, He Y. 2013. BrainNet viewer: a network visualization tool for human brain connectomics. PLoS One. 8(7):e68910. [DOI] [PMC free article] [PubMed] [Google Scholar]
Yan CG, Chen X, Li L, Castellanos FX, Bai TJ, Bo QJ, Cao J, Chen GM, Chen NX, Chen Wet al. 2019. Reduced default mode network functional connectivity in patients with recurrent major depressive disorder. Proc Natl Acad Sci U S A. 116(18):9078–9083. [DOI] [PMC free article] [PubMed] [Google Scholar]
Zhou Y, Yu C, Zheng H, Liu Y, Song M, Qin W, Li K, Jiang T. 2010. Increased neural resources recruitment in the intrinsic organization in major depression. J Affect Disord. 121(3):220–230. [DOI] [PubMed] [Google Scholar]
Zimmerman M, Posternak MA, Chelminski I. 2004. Derivation of a definition of remission on the Montgomery-Asberg depression rating scale corresponding to the definition of remission on the Hamilton rating scale for depression. J Psychiatr Res. 38(6):577–582. [DOI] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

CANBIND_Supplementary_materials_final_bhab286

Click here for additional data file.^{(7.9MB, pdf)}

Data Availability Statement

[ref1] Akoglu H. 2018. User's guide to correlation coefficients. Turk J Emerg Med. 18(3):91–93. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref2] Albert J, Lopez-Martin S, Carretie L. 2010. Emotional context modulates response inhibition: neural and behavioral data. Neuroimage. 49(1):914–921. [DOI] [PubMed] [Google Scholar]

[ref3] Bellana B, Liu Z, Anderson JAE, Moscovitch M, Grady CL. 2016. Laterality effects in functional connectivity of the angular gyrus during rest and episodic retrieval. Neuropsychologia. 80:24–34. [DOI] [PubMed] [Google Scholar]

[ref4] Boly M, Phillips C, Tshibanda L, Vanhaudenhuyse A, Schabus M, Dang-Vu TT, Moonen G, Hustinx R, Maquet P, Laureys S. 2008. Intrinsic brain activity in altered states of consciousness: how conscious is the default mode of brain function? Ann N Y Acad Sci. 1129:119–129. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref5] Brakowski J, Spinelli S, Dorig N, Bosch OG, Manoliu A, Holtforth MG, Seifritz E. 2017. Resting state brain network function in major depression - depression symptomatology, antidepressant treatment effects, future research. J Psychiatr Res. 92:147–159. [DOI] [PubMed] [Google Scholar]

[ref6] Button KS, Ioannidis JP, Mokrysz C, Nosek BA, Flint J, Robinson ES, Munafo MR. 2013. Power failure: why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci. 14(5):365–376. [DOI] [PubMed] [Google Scholar]

[ref7] Calhoun VD, Wager TD, Krishnan A, Rosch KS, Seymour KE, Nebel MB, Mostofsky SH, Nyalakanai P, Kiehl K. 2017. The impact of T1 versus EPI spatial normalization templates for fMRI data analyses. Hum Brain Mapp. 38(11):5331–5342. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref8] Campbell KL, Grigg O, Saverino C, Churchill N, Grady CL. 2013. Age differences in the intrinsic functional connectivity of default network subsystems. Front Aging Neurosci. 5:73. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref9] Carbonell F, Bellec P, Shmuel A. 2011. Global and system-specific resting-state fMRI fluctuations are uncorrelated: principal component analysis reveals anti-correlated networks. Brain Connect. 1(6):496–510. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref10] Chiao JY, Cheon BK. 2010. The weirdest brains in the world. Behav Brain Sci. 33(2–3):88–90. [DOI] [PubMed] [Google Scholar]

[ref11] Chikazoe J. 2010. Localizing performance of go/no-go tasks to prefrontal cortical subregions. Curr Opin Psychiatry. 23(3):267–272. [DOI] [PubMed] [Google Scholar]

[ref12] Chin Fatt CR, Jha MK, Cooper CM, Fonzo G, South C, Grannemann B, Carmody T, Greer TL, Kurian B, Fava Met al. 2020. Effect of intrinsic patterns of functional brain connectivity in moderating antidepressant treatment response in major depression. Am J Psychiatry. 177(2):143–154. [DOI] [PubMed] [Google Scholar]

[ref13] Churchill NW, Oder A, Abdi H, Tam F, Lee W, Thomas C, Ween JE, Graham SJ, Strother SC. 2012. Optimizing preprocessing and analysis pipelines for single-subject fMRI. I. Standard temporal motion and physiological noise correction methods. Hum Brain Mapp. 33(3):609–627. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref14] Churchill NW, Raamana P, Spring R, Strother SC. 2017. Optimizing fMRI preprocessing pipelines for block-design tasks as a function of age. Neuroimage. 154:240–254. [DOI] [PubMed] [Google Scholar]

[ref15] Churchill NW, Spring R, Afshin-Pour B, Dong F, Strother SC. 2015. An automated, adaptive framework for optimizing preprocessing pipelines in task-based functional MRI. PLoS One. 10(7):e0131520. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref16] Churchill NW, Strother SC. 2013. PHYCAA+: an optimized, adaptive procedure for measuring and controlling physiological noise in BOLD fMRI. Neuroimage. 82:306–325. [DOI] [PubMed] [Google Scholar]

[ref17] Churchill NW, Yourganov G, Oder A, Tam F, Graham SJ, Strother SC. 2012. Optimizing preprocessing and analysis pipelines for single-subject fMRI: 2. Interactions with ICA, PCA, task contrast and inter-subject heterogeneity. PLoS One. 7(2):e31147. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref19] Cuijpers P, Stringaris A, Wolpert M. 2020. Treatment outcomes for depression: challenges and opportunities. Lancet Psychiatry. 7(11):925–927. [DOI] [PubMed] [Google Scholar]

[ref20] Depping MS, Schmitgen MM, Kubera KM, Wolf RC. 2018. Cerebellar contributions to major depression. Front Psych. 9:634. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref21] Dichter GS, Gibbs D, Smoski MJ. 2015. A systematic review of relations between resting-state functional-MRI and treatment response in major depressive disorder. J Affect Disord. 172:8–17. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref22] Drysdale AT, Grosenick L, Downar J, Dunlop K, Mansouri F, Meng Y, Fetcho RN, Zebley B, Oathes DJ, Etkin A. 2017. Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat Med. 23(1):28. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref23] Dunlop K, Talishinsky A, Liston C. 2019. Intrinsic brain network biomarkers of antidepressant response: a review. Curr Psychiatry Rep. 21(9):87. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref24] Dupuy JM, Ostacher MJ, Huffman J, Perlis RH, Nierenberg AA. 2011. A critical review of pharmacotherapy for major depressive disorder. Int J Neuropsychopharmacol. 14(10):1417–1431. [DOI] [PubMed] [Google Scholar]

[ref25] Fiecas M, Ombao H, Lunen D, Baumgartner R, Coimbra A, Feng D. 2013. Quantifying temporal correlations: a test-retest evaluation of functional connectivity in resting-state fMRI. Neuroimage. 65:231–241. [DOI] [PubMed] [Google Scholar]

[ref26] Goldberg D. 2011. The heterogeneity of “major depression.”. World Psychiatry. 10(3):226. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref27] Goldstein-Piekarski AN, Staveland BR, Ball TM, Yesavage J, Korgaonkar MS, Williams LM. 2018. Intrinsic functional connectivity predicts remission on antidepressants: a randomized controlled trial to identify clinically applicable imaging biomarkers. Transl Psychiatry. 8(1):57. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref28] Gong Q, He Y. 2015. Depression, neuroimaging and connectomics: a selective overview. Biol Psychiatry. 77(3):223–235. [DOI] [PubMed] [Google Scholar]

[ref29] Grady CL, Protzner AB, Kovacevic N, Strother SC, Afshin-Pour B, Wojtowicz M, Anderson JA, Churchill N, McIntosh AR. 2010. A multivariate analysis of age-related differences in default mode and task-positive networks across multiple cognitive domains. Cereb Cortex. 20(6):1432–1447. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref30] Gudayol-Ferre E, Pero-Cebollero M, Gonzalez-Garrido AA, Guardia-Olmos J. 2015. Changes in brain connectivity related to the treatment of depression measured through fMRI: a systematic review. Front Hum Neurosci. 9:582. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref31] Hamani C, Mayberg H, Stone S, Laxton A, Haber S, Lozano AM. 2011. The subcallosal cingulate gyrus in the context of major depression. Biol Psychiatry. 69(4):301–308. [DOI] [PubMed] [Google Scholar]

[ref32] Henrich J, Heine SJ, Norenzayan A. 2010. Most people are not WEIRD. Nature. 466(7302):29. [DOI] [PubMed] [Google Scholar]

[ref33] Jing Y, Zhao N, Deng XP, Feng ZJ, Huang GF, Meng M, Zang YF, Wang J. 2020. Pregenual or subgenual anterior cingulate cortex as potential effective region for brain stimulation of depression. Brain Behav. 10(4):e01591. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref34] 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 Psychiat. 72(6):603–611. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref35] Keightley ML, Seminowicz DA, Bagby RM, Costa PT, Fossati P, Mayberg HS. 2003. Personality influences limbic-cortical interactions during sad mood induction. Neuroimage. 20(4):2031–2039. [DOI] [PubMed] [Google Scholar]

[ref36] Kennedy SH, Lam RW, Rotzinger S, Milev RV, Blier P, Downar J, Evans KR, Farzan F, Foster JA, Frey BNet al. 2019. Symptomatic and functional outcomes and early prediction of response to escitalopram monotherapy and sequential adjunctive aripiprazole therapy in patients with major depressive disorder: a CAN-BIND-1 report. J Clin Psychiatry. 80(2):18m12202. [DOI] [PubMed] [Google Scholar]

[ref37] Kessler R, Van Loo H, Wardenaar K, Bossarte R, Brenner L, Ebert D, Jonge P, Nierenberg A, Rosellini A, Sampson N. 2017. Using patient self-reports to study heterogeneity of treatment effects in major depressive disorder. Epidemiol Psychiatr Sci. 26(1):22–36. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref38] Konarski JZ, McIntyre RS, Grupp LA, Kennedy SH. 2005. Is the cerebellum relevant in the circuitry of neuropsychiatric disorders? J Psychiatry Neurosci. 30(3):178–186. [PMC free article] [PubMed] [Google Scholar]

[ref39] Korgaonkar MS, Goldstein-Piekarski AN, Fornito A, Williams LM. 2020. Intrinsic connectomes are a predictive biomarker of remission in major depressive disorder. Mol Psychiatry. 25(7):1537–1549. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref40] Lam RW, Milev R, Rotzinger S, Andreazza AC, Blier P, Brenner C, Daskalakis ZJ, Dharsee M, Downar J, Evans KRet al. 2016. Discovering biomarkers for antidepressant response: protocol from the Canadian biomarker integration network in depression (CAN-BIND) and clinical characteristics of the first patient cohort. BMC Psychiatry. 16:105. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref41] Lecrubier Y. 2002. How do you define remission. Acta Psychiatr Scand Suppl. 106:7–11. [DOI] [PubMed] [Google Scholar]

[ref42] Li B, Liu L, Friston KJ, Shen H, Wang L, Zeng LL, Hu D. 2013. A treatment-resistant default mode subnetwork in major depression. Biol Psychiatry. 74(1):48–54. [DOI] [PubMed] [Google Scholar]

[ref43] Long Z, Du L, Zhao J, Wu S, Zheng Q, Lei X. 2020. Prediction on treatment improvement in depression with resting state connectivity: a coordinate-based meta-analysis. J Affect Disord. 276:62–68. [DOI] [PubMed] [Google Scholar]

[ref44] MacQueen GM, Hassel S, Arnott SR, Jean A, Bowie CR, Bray SL, Davis AD, Downar J, Foster JA, Frey BNet al. 2019. The Canadian biomarker integration network in depression (CAN-BIND): magnetic resonance imaging protocols. J Psychiatry Neurosci. 44(4):223–236. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref45] Manoliu A, Meng C, Brandl F, Doll A, Tahmasian M, Scherr M, Schwerthoffer D, Zimmer C, Forstl H, Bauml Jet al. 2013. Insular dysfunction within the salience network is associated with severity of symptoms and aberrant inter-network connectivity in major depressive disorder. Front Hum Neurosci. 7:930. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref46] Mayberg HS, Lozano AM, Voon V, McNeely HE, Seminowicz D, Hamani C, Schwalb JM, Kennedy SH. 2005. Deep brain stimulation for treatment-resistant depression. Neuron. 45(5):651–660. [DOI] [PubMed] [Google Scholar]

[ref47] McIntosh AR, Lobaugh NJ. 2004. Partial least squares analysis of neuroimaging data: applications and advances. Neuroimage. 23:S250–S263. [DOI] [PubMed] [Google Scholar]

[ref48] McIntosh AR, Nyberg L, Bookstein FL, Tulving E. 1997. Differential functional connectivity of prefrontal and medial temporal cortices during episodic memory retrieval. Hum Brain Mapp. 5(4):323–327. [DOI] [PubMed] [Google Scholar]

[ref49] Minichino A, Bersani FS, Trabucchi G, Albano G, Primavera M, Delle Chiaie R, Biondi M. 2014. The role of cerebellum in unipolar and bipolar depression: a review of the main neurobiological findings. Riv Psichiatr. 49(3):124–131. [DOI] [PubMed] [Google Scholar]

[ref50] Montgomery SA, Asberg M. 1979. A new depression scale designed to be sensitive to change. Br J Psychiatry. 134:382–389. [DOI] [PubMed] [Google Scholar]

[ref51] Mulders PC, Eijndhoven PF, Schene AH, Beckmann CF, Tendolkar I. 2015. Resting-state functional connectivity in major depressive disorder: a review. Neurosci Biobehav Rev. 56:330–344. [DOI] [PubMed] [Google Scholar]

[ref52] Nickerson LD. 2018. Replication of resting state-task network correspondence and novel findings on brain network activation during task fMRI in the human connectome project study. Sci Rep. 8(1):17543. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref53] Oostenveld R, Fries P, Maris E, Schoffelen J-M. 2011. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci. 2011:156869. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref54] Pang Y, Chen H, Wang Y, Long Z, He Z, Zhang H, Liao W, Cui Q, Chen H. 2018. Transdiagnostic and diagnosis-specific dynamic functional connectivity anchored in the right anterior insula in major depressive disorder and bipolar depression. Prog Neuropsychopharmacol Biol Psychiatry. 85:7–15. [DOI] [PubMed] [Google Scholar]

[ref55] Park C, Rosenblat JD, Lee Y, Pan Z, Cao B, Iacobucci M, McIntyre RS. 2019. The neural systems of emotion regulation and abnormalities in major depressive disorder. Behav Brain Res. 367:181–188. [DOI] [PubMed] [Google Scholar]

[ref56] Peng X, Wu X, Gong R, Yang R, Wang X, Zhu W, Lin P. 2020. Sub-regional anterior cingulate cortex functional connectivity revealed default network subsystem dysfunction in patients with major depressive disorder. Psychol Med. 51(10):1–9. [DOI] [PubMed] [Google Scholar]

[ref57] Pizzagalli DA. 2011. Frontocingulate dysfunction in depression: toward biomarkers of treatment response. Neuropsychopharmacology. 36(1):183–206. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref58] Poldrack RA, Baker CI, Durnez J, Gorgolewski KJ, Matthews PM, Munafo MR, Nichols TE, Poline JB, Vul E, Yarkoni T. 2017. Scanning the horizon: towards transparent and reproducible neuroimaging research. Nat Rev Neurosci. 18(2):115–126. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref59] Raichle ME. 2015. The brain's default mode network. Annu Rev Neurosci. 38:433–447. [DOI] [PubMed] [Google Scholar]

[ref60] Rayner G, Jackson G, Wilson S. 2016. Cognition-related brain networks underpin the symptoms of unipolar depression: evidence from a systematic review. Neurosci Biobehav Rev. 61:53–65. [DOI] [PubMed] [Google Scholar]

[ref61] Roberts RP, Hach S, Tippett LJ, Addis DR. 2016. The Simpson's paradox and fMRI: similarities and differences between functional connectivity measures derived from within-subject and across-subject correlations. Neuroimage. 135:1–15. [DOI] [PubMed] [Google Scholar]

[ref62] Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, Niederehe G, Thase ME, Lavori PW, Lebowitz BDet al. 2006. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR*D report. Am J Psychiatry. 163(11):1905–1917. [DOI] [PubMed] [Google Scholar]

[ref63] Seeley WW, Menon V, Schatzberg AF, Keller J, Glover GH, Kenna H, Reiss AL, Greicius MD. 2007. Dissociable intrinsic connectivity networks for salience processing and executive control. J Neurosci. 27(9):2349–2356. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref64] Shafritz KM, Collins SH, Blumberg HP. 2006. The interaction of emotional and cognitive neural systems in emotionally guided response inhibition. Neuroimage. 31(1):468–475. [DOI] [PubMed] [Google Scholar]

[ref65] Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E, Hergueta T, Baker R, Dunbar GC. 1998. The Mini-international neuropsychiatric interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry. 59(Suppl 20):22–33quiz 34-57. [PubMed] [Google Scholar]

[ref66] Statistics Canada . Ethnic Origin Reference Guide, 2006Census. Catalogue no. 97–562-GWE2006025, Ottawa, ON, Canada. http://www12.statcan.gc.ca/census-recensement/2006/ref/rp-guides/ethnic-ethnique-eng.cfm. [Google Scholar]

[ref67] Talairach J, Tournoux P. 1988. Co-planar stereotaxic atlas of the human brain. New York: Thieme Medical Publishers, Inc. [Google Scholar]

[ref68] Thase ME. 2003. Achieving remission and managing relapse in depression. J Clin Psychiatry. 64(Suppl 18):3–7. [PubMed] [Google Scholar]

[ref69] Tozzi L, Goldstein-Piekarski AN, Korgaonkar MS, Williams LM. 2020. Connectivity of the cognitive control network during response inhibition as a predictive and response biomarker in major depression: evidence from a randomized clinical trial. Biol Psychiatry. 87(5):462–472. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref70] Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, Mazoyer B, Joliot M. 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage. 15(1):273–289. [DOI] [PubMed] [Google Scholar]

[ref71] Williams JB, Kobak KA. 2008. Development and reliability of a structured interview guide for the montgomery asberg depression rating scale (SIGMA). Br J Psychiatry. 192(1):52–58. [DOI] [PubMed] [Google Scholar]

[ref72] Winkler AM, Renaud O, Smith SM, Nichols TE. 2020. Permutation inference for canonical correlation analysis. Neuroimage. 220:117065. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref73] Wise T, Cleare AJ, Herane A, Young AH, Arnone D. 2014. Diagnostic and therapeutic utility of neuroimaging in depression: an overview. Neuropsychiatr Dis Treat. 10:1509–1522. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref74] Xia M, Wang J, He Y. 2013. BrainNet viewer: a network visualization tool for human brain connectomics. PLoS One. 8(7):e68910. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref75] Yan CG, Chen X, Li L, Castellanos FX, Bai TJ, Bo QJ, Cao J, Chen GM, Chen NX, Chen Wet al. 2019. Reduced default mode network functional connectivity in patients with recurrent major depressive disorder. Proc Natl Acad Sci U S A. 116(18):9078–9083. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ref76] Zhou Y, Yu C, Zheng H, Liu Y, Song M, Qin W, Li K, Jiang T. 2010. Increased neural resources recruitment in the intrinsic organization in major depression. J Affect Disord. 121(3):220–230. [DOI] [PubMed] [Google Scholar]

[ref77] Zimmerman M, Posternak MA, Chelminski I. 2004. Derivation of a definition of remission on the Montgomery-Asberg depression rating scale corresponding to the definition of remission on the Hamilton rating scale for depression. J Psychiatr Res. 38(6):577–582. [DOI] [PubMed] [Google Scholar]

PERMALINK

Baseline Functional Connectivity in Resting State Networks Associated with Depression and Remission Status after 16 Weeks of Pharmacotherapy: A CAN-BIND Report

Gwen van der Wijk

Jacqueline K Harris

Stefanie Hassel

Andrew D Davis

Mojdeh Zamyadi

Stephen R Arnott

Roumen Milev

Raymond W Lam

Benicio N Frey

Geoffrey B Hall

Daniel J Müller

Susan Rotzinger

Sidney H Kennedy

Stephen C Strother

Glenda M MacQueen

Andrea B Protzner

Abstract

Introduction

Methods

Participants and Treatment

Table 1.

Table 2.

MRI Data Acquisition and Preprocessing

Seed Selection

Data Analysis with PLS

Results

Participants

Replication of Patient—Control Differences

Anterior Cingulate ROI

Figure 1 .

Posterior Cingulate ROI

Insula ROI

Figure 2 .

Dorsolateral Prefrontal Cortex ROI

Early, Late and Nonremitter Comparison—Anterior Cingulate Seed

Figure 3 .

Table 3.

Early, Late, and Nonremitter Comparison—Posterior Cingulate Seed

Figure 4 .

Table 4.

Early, Late, and Nonremitter Comparison—Insula Seed

Figure 5 .

Table 5.

Early, Late, and Nonremitter Comparison—dlPFC ROI

Correction for MADRS Baseline Differences

Correction for Recording Site and Scanner Type

Recording Site

Scanner Type

Stability of Findings in Subsample

Table 6.

Post-Hoc Analysis Examining Remission to Escitalopram Alone or Combined Escitalopram and Aripiprazole

Discussion

Limitations

Conclusion

Supplementary Material

Funding and Conflict of Interest

Notes

Authors’ Contributions

Data Availability Statement

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases