Dorsolateral Prefrontal Cortex and Subcallosal Cingulate Connectivity Show Preferential Antidepressant Response in Major Depressive Disorder

Abstract

BACKGROUND:

Major depressive disorder is associated with abnormal connectivity across emotion and reward circuits as well as other established circuits that may negatively impact treatment response. The goal of this study was to perform an exploratory reanalysis of archival data from a clinical trial to identify moderators of treatment outcome of sertraline over placebo.

METHODS:

EMBARC (Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care) study participants completed magnetic resonance imaging before randomization to either sertraline or placebo for 8 weeks (n = 279). Seed-based functional connectivity was computed using 4 bilateral seeds (2 spheres defined bilaterally): amygdala, dorsolateral prefrontal cortex (DLPFC), subcallosal cingulate cortex, and ventral striatum. Functional connectivity maps were generated, principal component analysis was performed, linear mixed effects models were used to determine moderators of treatment outcome, and post hoc analyses were used to determine level of connectivity (low and high, −1 and +1 SD from the mean) that was most sensitive to improved depression severity (baseline to week 8) based on treatment.

RESULTS:

Greater mean reduction in the 17-item Hamilton Rating Scale for Depression score by 8 weeks occurred with sertraline relative to placebo when connectivity in the DLPFC was low (3-way interaction test, p = .05). Conditional on low connectivity in the DLPFC and subcallosal cingulate cortex and high connectivity in the ventral striatum and amygdala, there was on average a 4.8-point greater reduction in the 17-item Hamilton Rating Scale for Depression score with sertraline relative to placebo (p = .003).

CONCLUSIONS:

The level of functional connectivity seeded in both the DLPFC and the subcallosal cingulate cortex networks may play an important role in identifying a favorable response to sertraline over placebo.

Major depressive disorder (MDD) is characterized by disturbances in affect, mood, neurovegetative function, cognition, and psychomotor activity (1). In recent years, increasing attention has been directed toward disrupted emotional and reward processing associated with MDD (2,3) as well as disrupted communication between brain regions (4–6). Neuroimaging studies have shown that these symptoms are associated with abnormalities in many brain regions, especially the amygdala (AMY), dorsolateral prefrontal cortex (DLPFC), ventral striatum (VS), and subcallosal cingulate cortex (SCC).

The AMY, which has the core function of emotion processing, is an important hub of the emotion-processing network. Dysregulation of emotional processes is considered to be a core feature of MDD (7,8), and it coincides with dysfunction of the emotion-processing network (9,10). The DLPFC, the main hub of the central executive network (i.e., executive function), has abnormally low levels of brain function in patients with MDD (11,12). The VS, the core of the reward network, has also been shown to have abnormalities in patients with MDD. The striatal circuitry—including the VS—involves connections with multiple regions (PFC, AMY, and hippocampus). Each of these connections has some distinct involvement in behavioral responses regarding rewards and losses. A number of functional magnetic resonance imaging (fMRI) studies have found reduced brain activity in the VS and other basal ganglia structures among depressed patients during reward processing (13–15) as well as abnormal recruitment of these substrates during the processing of positive and negative events (16–18). These findings provide insight into the possible role of the VS in the negative affective biases that are associated with depression. In addition to these frontal and subcortical regions, the SCC has emerged as an important component of MDD pathophysiology (19,20). A study by Dunlop et al. (21) showed that functional connectivity seeded in the SCC with the left anterior ventrolateral PFC/insula, dorsal midbrain, and left ventromedial PFC was associated with outcomes of remission. In addition, a recent study by Liston et al. (22) showed hyperconnectivity of the SCC in MDD that was modulated after repetitive transcranial magnetic stimulation and may play a crucial role in reducing depression severity.

Functional connectivity is one method of assessing the communication between brain regions. The first functional connectivity study in MDD showed reduced connectivity between the anterior cingulate cortex and the medial thalamus, AMY, and pallidostriatum (23). Additional studies have shown the disruption of functional networks in MDD as well as how these disruptions are related to treatment outcome (4). Since then, there have been differing results primarily because of small sample sizes, lack of a placebo control, or data from only one site. Yet, a review of functional connectivity studies involving MDD showed that in general, MDD causes disruptions in brain networks (e.g., anterior portions of the default mode network) (6). While each of these studies expands our understanding of the network disruptions, to make tangible improvements in the selection for MDD treatment, we must determine whether relationships exist between functional networks in MDD and pharmacotherapy treatment outcome as opposed to simple placebo response.

Kozel et al. (24) showed connectivity measures in various brain regions—especially the SCC—to be highly correlated with treatment outcome. Additionally, Dunlop et al. (21) showed that positive summed resting-state functional connectivity of the SCC with the left anterior ventrolateral PFC, dorsal midbrain, and left ventromedial PFC was associated with remission with treatment with cognitive behavioral therapy and treatment failure with medication, whereas negative summed connectivity showed an inverse relationship with cognitive behavioral therapy and medication. However, these studies, as well as others that have used functional connectivity to understand treatment outcomes in MDD, have not helped us translate these findings to clinical practice because of one main reason—namely, determining whether functional connectivity in a brain region can differentially predict treatment outcome as opposed to general prediction inclusive of placebo response.

In this study, we sought to determine whether there is an association between treatment response and pretreatment resting-state functional connectivity. In this large multisite neuroimaging study, we used a data-driven approach that simultaneously considered the connectivity between key MDD-related brain regions (AMY, DLPFC, SCC, and VS) and all other regions in the brain (in a voxelwise approach). Participants were recruited as part of the EMBARC (Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care) study (25), a longitudinal multisite, randomized, double-blind, placebo-controlled trial of sertraline designed to evaluate moderators of treatment response. Given the size of EMBARC, rigorous placebo-controlled design, unbiased statistical analysis, and resting-state fMRI (rsfMRI) acquisition largely similar across the field, any moderators of response to sertraline versus placebo identified here have the potential to inform our understanding of the brain bases of antidepressant treatment, stratify patients based on predicted treatment outcome, and guide personalized medicine for depression.

METHODS AND MATERIALS

Participants

EMBARC study participants (N = 309) were 18 to 65 years of age and had MDD diagnosed using the Structured Clinical Interview for DSM-IV Axis I Disorders. Additional inclusion criteria included a 16-item Quick Inventory of Depressive Symptomatology–Self-Report score ≥14 at both the screening and the randomization visits, first major depressive episode beginning before age 30, and either a chronic current episode (duration ≥2 years) or recurrent MDD (at least 2 lifetime episodes). A complete list of eligibility criteria and their justification has been published previously (25). The institutional review board at each of the following clinical sites approved the study: University of Texas Southwestern Medical Center (TX), Massachusetts General Hospital (MGH), Columbia University (CU), and University of Michigan (UM). All participants provided written informed consent before participating in the study. Participants underwent neuroimaging before treatment.

Treatment

Stage 1 of EMBARC included an 8-week double-blind, placebo-controlled trial of sertraline. For details on randomization, see Trivedi et al. (25). Briefly, participants were assessed (via clinical measures and neuroimaging) before treatment and then randomly assigned to 1 of 2 treatment groups: placebo or sertraline. Treatment continued for 8 weeks.

Assessment

The 17-item Hamilton Rating Scale for Depression (HAMD₁₇) was used to assess depression severity of participants with individual items summed to yield total scores that indicated the following ranges of depression severity: no depression (score 0–7), mild depression (score 8–13), moderate depression (score 14–18), severe depression (score 19–22), and very severe depression (score ≥23) (26). HAMD₁₇ was administered at 7 time points (T) in the study: baseline (T0), week 1 (T1), week 2 (T2), week 3 (T3), week 4 (T4), week 6 (T6), and week 8 (T8). If a participant’s HAMD₁₇ score was ≤7 at week 8, the participant was considered to be in remission.

MRI Acquisition

All participants underwent neuroimaging to identify neurobiological moderators of treatment outcome. MRI scans were performed on 3T MRI systems at all EMBARC sites (CU: General Electric, Chicago, IL; MG: Siemens, Malvern, PA; TX and UM: Phillips, Andover, MA). We used rsfMRI to investigate regional interaction. Additionally, a high-resolution T1-weighted image was acquired as an anatomical reference during the same session. The rsfMRI parameters across the 4 sites were similar: single-shot echo-planar imaging, repetition time = 2000 ms, echo time = 28 ms, voxel size = 3.2× 3.2 × 3.1 mm³, 39 axial slices, 180 image volumes, and duration of 6 minutes. The high-resolution T1-weighted image parameters were also similar: 160 sagittal slices, voxel size = 1 × 1 × 1 mm³, and field of view = 256 × 256 × 160 mm³. The acquisition parameter details for each site are provided in Table S1.

MRI Data Processing

The rsfMRI data were preprocessed using CONN (27) and SPM8 (28). Briefly, the data were preprocessed using SPM8 with slice timing correction, motion corrected (realignment and unwarp), spatially normalized to the Montreal Neurological Institute template (matrix = 91 × 109 × 91, resolution = 2 × 2 × 2 mm³), and smoothed using a Gaussian kernel (full width at half maximum of 8 mm). Structural MRI scans were segmented into gray matter, white matter, and cerebrospinal fluid. Scrubbing was performed using ART (29), which is a component of CONN. Briefly, ART was used to detect outliers that met the following criteria: normalized global blood oxygen level–dependent signal z ≥ 3.0 and subject motion threshold ≥ 0.5. These outliers were used as movement covariates. Physiological and other spurious sources of noise were estimated using the ACompCor method (30) and used as first-level covariates. An additional 6 rigid body parameters (translational and rotational motion) characterized each participant’s motion and were used as covariates. The residual blood oxygen level–dependent time series was then bandpass filtered using 0.009 Hz < f < 0.08 Hz to keep only the appropriate frequency fluctuations. Using a seed-based approach, two 5-mm-radius spheres were defined bilaterally for each region based on previously published literature. The Montreal Neurological Institute coordinates were as follows: AMY [±22, 0, −22] (31), DLPFC [±46, +18, +44] (32), SCC [±6, +24, −11] (21), and VS [±8, +8, −8] (33). Pearson’s correlation coefficients were computed between the time course of each bilateral seed and all other voxels to generate a correlation map. These correlation maps were then transformed to a z score map using Fisher’s inverse hyperbolic tangent transformation. Last, a region of interest (ROI) analysis was conducted on known regions of each network (network defined as the seed as well as the functionally connected brain regions): bilateral DLPFC and inferior parietal cortex for the DLPFC network, bilateral nucleus accumbens and medial frontal cortex for the VS network, bilateral AMY for the AMY network, and bilateral SCC and posterior cingulate cortex for the SCC network. To define functional ROIs and standardize the ROI size, each region’s anatomical region was first defined based on Talairach Daemon database in AFNI (34). Next, a functional ROI was defined by choosing the top 500 z score voxels in each region of the network (as shown in Figure 1), and the functional ROIs were used as a mask to calculate a single z score for each region (35). Last, a principal component reduction was performed on all regions (11 brain regions) combined to obtain weights for network connectivity averages.

Figure 1.

Qualitative connectivity maps of the dorsolateral prefrontal cortex (DLPFC), ventral striatum (VS), amygdala (AMY), and subcallosal cingulate cortex (SCC) networks. Connectivity maps were arbitrarily thresholded at z score ≥ 0.15 and cluster size ≥ 250 voxels to illustrate the nodes in each network qaulitatively.

Statistical Analysis

Moderation Analyses.

We modeled HAMD₁₇ scores, ${\text{y}}_{\text{ijk}\ell }$, using the following general linear model:

$${\text{y}}_{\text{hijk}\ell }=\mu +{\phi }_{\text{h}}+\lambda {\text{x}}_{\ell }+{\beta }_{\text{i}}{\text{z}}_{\text{i}\ell }+{\gamma }_{\text{j}}+{\tau }_{\text{k}}+{\varsigma }_{\text{jk}}+{\alpha }_{\text{ij}}{\text{z}}_{\text{i}\ell }+{\delta }_{\text{ik}}{\text{z}}_{\text{i}\ell }+{\eta }_{\text{ijk}}{\text{z}}_{\text{i}\ell }+{\epsilon }_{\text{hijk}\ell },$$

where ${\text{y}}_{\text{hijk}\ell }$ are subject-specific HAMD₁₇ scores; ${\text{z}}_{\text{i}\ell }$ are subject-specific connectivity values in i = 1,.,4 networks (DLPFC, VS, AMY, SCC), calculated as weighted averages of within-network nodes; ${\gamma }_{\text{j}}{\text{Z}}_{\text{i}\ell }$ is treatment group, j = 1,2 (placebo, sertraline); ${\phi }_{\text{h}}$ is gender, h = 1,2; ${\tau }_{\text{k}}$ is time, $k=1,\dots ,7$ (T0, T1, T2, T3, T4, T6, T8); ${\text{x}}_{\ell }$ is age; and $\ell =1,\dots ,{\text{n}}_{\text{j}}$ participants.

In the model above, all pairwise and three-way interactions that involve treatment, time, and at most one network (denoted above by the appropriate combination of parameter subscripts i, j, and k) are included. The covariance matrix of the random vector $\epsilon$, whose elements are denoted ${\epsilon }_{\text{hijk}\ell }$ in the model above, includes separate variance components for participants (between-participant variance, ${\sigma }_b^2$), for site (between-site variance, ${\sigma }_{\text{s}}^2$), and for within-participant variance, ${\sigma }_{\text{w}}^2$.

All fixed-effects parameters of the linear model have been estimated by maximum likelihood under normal theory, and all variance components have been estimated by restricted maximum likelihood (conditional ML from model residuals) using SAS, version 9.3 software (SAS Institute Inc., Cary, NC). The parameter estimates and the covariance matrix of parameter estimates were imported into R (R Project for Statistical Computing; http://r-project.org) for all tests of contrasts pertaining to the specific hypotheses of the study.

Primary Tests of Three-way Interactions.

The primary hypothesis of interest was that the reduction from baseline HAMD₁₇ scores, specifically by the end of the measurement period T8, would be greatest in the sertraline treatment group relative to placebo, but that the reduction might depend on connectivity levels in at least 1 of the 4 networks. This hypothesis was tested by specific contrasts of the three-way interaction parameter estimates, most notably the following 1 degree of freedom tests of the null:

$${\eta }_{\text{i}17}-{\eta }_{\text{i}11}={\eta }_{\text{i}27}-{\eta }_{\text{i}21},$$

for each i. These are network-level tests of regression parameter differences between the treatment groups on their respective T8–T0 change scores.

Secondary Tests of Conditional Means.

We secondarily tested for treatment differences of T8–T0 change in mean HAMD₁₇ scores, conditional on several connectivity values across the 4 networks. These conditional null tests were the following:

$${\varsigma }_{17}-{\varsigma }_{11}+\sum _{i=1}^4\left({\eta }_{i17}-{\eta }_{i11}\right)\cdot z_i={\varsigma }_{27}-{\varsigma }_{21}+\sum _{i=1}^4\left({\eta }_{i27}-{\eta }_{i21}\right)\cdot z_i,$$

given specific values for $z_i$. The logic behind these secondary conditional tests was based on the insufficiency of a significant three-way interaction to support our primary hypothesis. For example, although we might discover a significant three-way interaction in a network, if the change in mean HAMD₁₇ scores from T0 to T8 is not significantly different between treatment groups at any connectivity value within which the three-way interaction was found, the interaction itself is not interpretable clinically. We arbitrarily chose to define low, average, and high values for each $z_i$ as $z_i=-1$ (1 SD below the mean of $z_i$); $z_i=0$ (mean of $z_i$); and $z_i=1$ (1 SD above the mean of $z_i$). These conditional null tests represent a set of tests, one for each combination of the $z_i$. There are, therefore, 3⁴ = 81 such tests. As these secondary tests serve only to explain the primary three-way interaction test, we set the false discovery rate control at 10%.

RESULTS

Participant Characteristics

Of the 309 EMBARC participants, 296 met the inclusion criteria and intent-to-treat, of which 279 participants had neuroimaging data (Table 1). No significant differences in age, gender, age at onset, race, education, employment status, number of major depression episodes, baseline HAMD₁₇ score, or current episode duration were noted between the treatment groups (p > .05). The number of participants at each site stratified by treatment were as follows: Columbia University, n = 79 participants (39 sertraline, 40 placebo); Massachusetts General Hospital, n = 46 participants (23 sertraline, 23 placebo); University of Texas Southwestern Medical Center, n = 95 (48 sertraline, 47 placebo); University of Michigan, n = 59 participants (29 sertraline, 30 placebo). Table S2 shows demographics and clinical summaries stratified by site and treatment.

Table 1.

Baseline Sociodemographic and Clinical Variables as a Function of Treatment Arm

	Sertraline		Placebo
Categorical Variables	n	%	n	%	χ2 (df)	p
Gender					1.84 (1)	.18
Male	41	29.5	52	37.1
Female	98	70.5	88	62.9
Race					2.66 (2)	.27
White	86	61.9	98	70.0
African American	32	23.0	22	15.7
Other	21	15.1	20	14.3
Employment Status					0.01 (2)	.99
Employed	76	54.7	76	54.3
Unemployed	58	41.7	59	42.1
Unknown	5	3.6	5	3.6
Continuous Variables	Mean	SD	Mean	SD	t (df)	p
Age, Years	38.0	14.0	37.0	13.0	0.47 (275)	.64
Age of Onset, Years	16.2	6.1	16.2	5.7	−0.02 (275)	.99
Education, Years	15.0	2.6	15.2	2.7	−0.66 (275)	.51
Number of MDEs	9.2	19.3	8.7	14.2	0.20 (226)	.84
Duration of Current Episode, Months	41.3	69.3	41.3	76.8	0.09 (275)	.93
HAMD17	18.6	4.6	18.7	4.1	−0.29 (274)	.77

There were 7 participants (4 with placebo and 3 with sertraline) with too many MDEs to count, 18 participants (8 with placebo and 10 with sertraline) with no number of MDEs to count, and 6 participants (4 placebo and 2 with sertraline) with no employment status.

HAMD₁₇, 17-item Hamilton Rating Scale for Depression; MDEs, major depression episodes.

Functional Connectivity

Figure 1 shows the average functional connectivity maps (i.e., z score maps) in the DLPFC, VS, AMY, and SCC networks where the z maps were arbitrarily thresholded (z score ≥ 0.15, k ≥ 250) to qualitatively visualize the nodes associated with each network. The results of the principal component reduction showed that 4 principal components were obtained, consistent with the network derivations, and within-network weights were comparable to yield network connectivity values that were essentially node averages for each network (Table 2).

Table 2.

Summary of Principal Component Values and Their Order as a Function of Seeded Region Connectivity

	DLPFC				VS			AMY		SCC
	L DLPFC	R DLPFC	L IPC	R IPC	L NA	R NA	L/R MFC	L AMY	R AMY	L/R SCC	L/R PCC
PC1	−0.44	−0.51	−0.50	−0.51	−0.02	−0.09	0.12	0.00	−0.00	−0.08	0.05
PC2	0.01	0.08	−0.07	−0.00	0.62	0.54	0.56	0.05	−0.04	−0.00	−0.02
PC3	0.01	0.00	−0.01	0.00	−0.03	−0.03	0.07	−0.68	−0.71	0.11	−0.12
PC4	−0.00	−0.02	0.04	0.01	−0.12	−0.05	0.21	0.02	−0.02	0.66	0.71

AMY, amygdala; DLPFC, dorsolateral prefrontal cortex; IPC, inferior parietal cortex; L, left; MFC, medial frontal cortex; NA, nucleus accumbens; PC, principal component; PCC, posterior cingulate cortex; R, right; SCC, subcallosal cingulate cortex; VS, ventral striatum.

Primary Tests of Three-way Interactions

Within the DLPFC network, we found that for participants with low connectivity values, treatment with sertraline yielded a significantly larger decrease in HAMD₁₇ scores from T0 to T8 relative to placebo, whereas for participants with high connectivity values, placebo was just as effective as sertraline (p = .05) (Table 3). Figure 2 shows separate regressions for the sertraline and placebo groups for HAMD₁₇ score change (i.e.,week 8–baseline) versus DLPFC connectivity level, whereas the value of the SCC, VS, and AMY components were fixed at [−1, +1, +1], respectively. For the sertraline treatment group, improvement in HAMD₁₇ scores was evident across all levels of DLPFC connectivity; however, comparable improvement for the placebo group occurred only at high levels of DLPFC connectivity. None of the other networks exhibited significant three-way interactions, although a similar pattern, as just described in the DLPFC, was noted in the SCC (p = .07). Table 3 shows all primary three-way interaction tests. Table S3 shows the generic analysis of variance table from the mixed model.

Table 3.

Summary of Primary Three-way Interaction Tests, $z_i$ × Treatment × Time (T8 – T0)

i	Network	t (df)	p
1	DLPFC	−1.62 (1367)	.052
2	VSN	0.42 (1375)	.662
3	AMY	0.80 (1370)	.788
4	SCC	−1.45 (1372)	.073

AMY, amygdala; DLPFC, dorsolateral prefrontal cortex; i, network of interest; SCC, subcallosal cingulate cortex; T, time point; VS, ventral striatum.

Figure 2.

Change in 17-item Hamilton Rating Scale for Depression (HAMD₁₇) score from baseline (T0) to exit (T8) is plotted as a function of dorsolateral prefrontal cortex (DLPFC) component score. The other components were kept constant at subcallosal cingulate cortex = −1, ventral striatum = +1, and amygdala = +1. The significant difference between treatments in HAMD₁₇ score change occurred primarily at low DLPFC connectivity levels. While improvement for participants given sertraline (Sert) occurred regardless of DLPFC connectivity level, the placebo (Plac) group improved only for the higher DLPFC connectivity level (no different than sertraline). The dotted lines represent pointwise 95% least significant intervals.

Secondary Tests of Conditional Means

As shown in Table 4, the highest mean difference of a 4.8-point improvement in HAMD₁₇ score was detected when the DLPFC and SCC component scores were low and the VS and AMY component scores were high (i.e., DLPFC = −1; SCC = −1; VS = +1; AMY = +1, p = .003) (Figure S1A). This difference is noted in Figure 2 by comparing the regression lines at a DLPFC connectivity value of −1. Conversely, the conditional mean differences in HAMD₁₇ reduction did not differ between the treatment groups when the connectivity levels within the DLPFC and SCC were high (i.e., +1) (Figure S1B). All other significant combinations are shown in Table 4 and Figure S2, primarily owing to low DLPFC connectivity and low SCC connectivity.

Table 4.

Difference of Change in HAMD₁₇ Scores From 0 to 8 Weeks Between Sertraline and Placebo (14 out of 81 Total Combinations; FDR = 10%)

DLPFC	VS	AMY	SCC	Mean Difference	t (df)	p
−1	1	1	−1	4.79	2.76 (1372.8)	.003
−1	0	1	−1	4.46	2.77 (1365.4)	.003
−1	1	0	−1	4.14	2.63 (1380.4)	.004
−1	−1	1	−1	4.12	2.22 (1362.6)	.013
−1	0	0	−1	3.80	2.92 (1372.0)	.002
−1	1	1	0	3.62	2.37 (1368.3)	.009
0	1	1	−1	3.53	2.30 (1375.7)	.011
−1	−1	0	−1	3.47	2.35 (1364.3)	.010
−1	0	1	0	3.28	2.33 (1366.5)	.010
0	0	1	−1	3.20	2.27 (1365.3)	.012
−1	0	0	0	2.63	2.42 (1370.4)	.008
0	0	0	−1	2.54	2.30 (1369.1)	.011

The values of −1, 0, and +1 denote 1 SD below, at mean, and 1 SD above, respectively.

AMY, amygdala; DLPFC, dorsolateral prefrontal cortex; FDR, false discovery rate; HAMD₁₇, 17-item Hamilton Rating Scale for Depression; SCC, subcallosal cingulate cortex; VS, ventral striatum.

In each of the single degree-of-freedom contrasts, sertraline had a larger mean reduction of HAMD₁₇ scores from baseline to week 8 for participants with low DLPFC and SCC connectivity values. That is, if a participant’s connectivity in all nodes of the DLPFC and SCC was low, the participant was more likely to have a large HAMD₁₇ reduction by T8 if given sertraline rather than placebo. However, at high DLPFC and SCC connectivity values, there was no difference between sertraline and placebo.

DISCUSSION

In this large study of outpatients with MDD, using a data-driven approach, we found that levels of functional connectivity seeded in the DLPFC and SCC play an important role in identifying a favorable response to sertraline over placebo. In general, the largest improvement in depression severity (i.e., lower HAMD₁₇ score across time) was detected when the functional connectivity seeded in the DLPFC and SCC was low and the functional connectivity seeded in the VS and AMY was high. Further, participants with low DLPFC and low SCC connectivity receiving sertraline treatment had a larger improvement in depression from baseline to week 8 compared with participants in the placebo group.

An interesting, clinically relevant finding of this report is that patients with both low DLPFC and low SCC connectivity receiving sertraline had a larger improvement (based on reduced HAMD₁₇ score from baseline to week 8) compared with the placebo group. Broadly, decreased functioning of the DLPFC—hub of the central executive network—has been shown to be a feature of MDD during resting and task states (36,37). In addition, decreased connectivity within this network has also been seen in response to negative stimuli compared with positive stimuli, which suggests underresponsiveness of the DLPFC to negative stimuli in patients with MDD (38). These findings have helped with targeted treatment. A number of studies have shown that functioning of the DLPFC normalizes with various treatments. For example, transcranial magnetic stimulation targeted at the DLPFC (39,40) and various antidepressants (41,42) have been shown to increase activation of the DLPFC, thereby reducing MDD symptoms.

Our study also showed that low connectivity (before treatment with a selective serotonin reuptake inhibitor) to the SCC—a core component of MDD pathophysiology—related to better outcomes with a selective serotonin reuptake inhibitor. This is in contrast to previous imaging studies that showed increased SCC activity before treatment and decreased activity after a variety of treatments, including antidepressants and electroconvulsive therapy (36,43,44). However, the key difference in this study is that by exploring key networks related to MDD, we found that a combination of both low SCC and low DLPFC connectivity predicted better treatment outcomes with sertraline.

While combined low connectivity to the SCC and DLPFC was consistently linked to improved MDD severity with selective serotonin reuptake inhibitor treatment, hyperconnectivity to the VS and AMY coincided with these findings. Hyperconnectivity of the AMY has been consistently reported in other MDD studies (38,45). However, the VS and its associated regions are usually hypoconnected and related to the underresponsiveness of the reward system in MDD, a possible neurological sign of increased anhedonic features (46). This finding may suggest that biological features that underlie pathophysiology of MDD may be distinct from biological markers that predict improvement with antidepressant treatment versus placebo.

Limitations in the current report should be noted. First, the eligibility restrictions and closer clinical tracking possible in research studies relative to routine clinical care may limit the generalizability of these findings to the broad population of depressed patients. Second, it is possible that some sertraline responders may have, in fact, had a placebo response rather than a pharmacological response to the drug. Therefore, the results reported here may underestimate the true effect, had the placebo response rate of the population been somewhat lower. Third, regions of interest were extracted after smoothing, and there is a potential that this step could contaminate the ROIs. Fourth, we set the false discovery rate control at 10% for our secondary conditional tests, which is higher than the typical 5%. However, these tests followed our primary three-way interaction tests and served simply to allow a clinical explanation for the results of the three-way interactions. Therefore, we chose to be a little more liberal with the false discovery rate threshold, a practice that is becoming more common in the genetics and imaging literature. Last, the scan length (6 min) for computing resting-state functional connectivity is relatively short. While longer scan durations may result in more reliable data, the duration of the scan was restricted in EMBARC to minimize participant burden and accommodate additional neuroimaging scans. In summary, the results in this report should be considered preliminary, and future replication of these findings in an independent sample is necessary for validation.

Based on the findings of this study, there are several steps to be taken next to mitigate the limitations of the current report, as follows: first, attempt to extend these findings to other treatment modalities and functional neuroimaging modalities; second, investigate whether there is a relationship between speed of response or adverse response, remission rates, and level of connectivity (e.g., a person with high connectivity in a brain region might respond faster, or more adversely, to a given treatment compared with someone with low connectivity); third, investigate whether these moderators change with treatment over time using posttreatment imaging. Finally, as this was an exploratory reanalysis of archival data, our next step is to replicate the findings in a larger and more heterogeneous population (47).

In conclusion, the present study identified the following nodes with disrupted functional connectivity in patients with MDD: bilateral DLPFC and inferior parietal cortex, bilateral nucleus accumbens and medial frontal cortex, bilateral AMY, and bilateral SCC and posterior cingulate cortex. Low connectivity within the DLPFC or the SCC predicted a significantly favorable treatment outcome for sertraline versus placebo. The highest mean difference of improvement in depression severity was detected when DLPFC and SCC connectivity were low and VS and AMY connectivity were high. Sertraline treatment had a larger mean reduction in depression severity from T0 to T8 for participants with low DLPFC and SCC connectivity values compared with placebo. The findings of this study demonstrate that the level of functional connectivity seeded in both the DLPFC and the SCC networks was associated with a favorable response to sertraline over placebo.