Introduction
The bed nucleus of the stria terminalis (BNST) is a small, heterogeneous structure located within the basal forebrain that plays a critical role in coordinating autonomic, neuroendocrine, and behavioral activity.1 The BNST is a component of the dorsal extended amygdala and part of the neural circuitry underlying the processing of sustained threat and chronic stress.2 The BNST also contributes to the emergence of a negative emotional state during drug withdrawal,3 tying it to dysfunctional reward states. These findings have led researchers to propose that the BNST may play a role in several psychiatric disorders4 and that the intrinsic connectivity of the BNST may serve as a biomarker of clinical dysfunction.5
Previous research sought to determine whether psychiatric diagnoses characterized by negative emotion might impact the intrinsic connectivity of the BNST. For instance, clinical anxiety disorders were previously found to be associated with reduced intrinsic connectivity between the BNST and the dorsolateral prefrontal cortex (dlPFC).6,7 Post-traumatic stress disorder (PTSD) and dissociative PTSD symptoms have also been associated with intrinsic connectivity of the BNST–dlPFC and BNST–striatum, respectively.8 Intrinsic connectivity of the BNST has also been shown to prospectively predict the efficacy of cognitive behavioral therapy in reducing alcohol consumption in individuals with alcohol use disorder.9
To date, however, no studies have investigated whether intrinsic connectivity of the BNST is impacted by major depressive disorder (MDD). Nevertheless, both preclinical and clinical research suggest that the BNST may play a role in the emergence of MDD. For instance, animal studies found that corticotropin releasing factor within the BNST was associated with anhedonia, a core symptom of MDD,10 following exposure to chronic stressors.
Pharmacological treatments for MDD, such as ketamine, have also been shown to impact the extracellular availability of catecholamines within the BNST in animals.11,12 In the human brain, MDD was associated with increased alpha power of local field potentials within the BNST,13 and deep brain stimulation (DBS) of the BNST reduced clinical symptoms in individuals with treatment-resistant depression.14,15 Interestingly, a previous meta-analysis found that MDD was associated with reduced intrinsic connectivity between the amygdala and striatum.16 Both the central nucleus of the amygdala and the BNST are components of the dorsal extended amygdala,17 and previous research suggests that these two structures have overlapping patterns of intrinsic connectivity.18 Another meta-analyses also suggested that MDD may be associated with reduced intrinsic connectivity within the prefrontal cognitive control network.19
One potential explanation for the absence of research investigating BNST intrinsic connectivity in patients with MDD is the difficulty of using functional magnetic resonance imaging (fMRI) to measure neural activity within small neural structures. The human BNST is approximately ∼190 mm2, and researchers have suggested that neuroimaging techniques with higher spatial resolution are needed to understand how the BNST contributes to human health and disease.2,20 However, increasing the spatial resolution of fMRI data has been shown to decrease the temporal signal-to-noise ratio (tSNR) of the BOLD signal, which can impair the reliability and reproducibility of intrinsic connectivity measures.21 Ultra-high field (7T) fMRI is one technique that can effectively reconcile these competing demands and, indeed, previous research found that the reduced tSNR associated with higher spatial resolution was mitigated in 7T fMRI compared to 3T and 1.5T fMRI.22 Consequently, ultra-high field (7T) fMRI confers unique advantages to studies investigating the relationship between intrinsic connectivity of the BNST and MDD.
Building on the observations that the BNST exhibits intrinsic connectivity with both the striatum and the PFC5,7,18 and that these neural networks are impacted by MDD, this study sought to determine the relationship between MDD and BNST intrinsic connectivity with the striatum and PFC. Ultra-high field (7 Tesla (7T)) fMRI was used to compare intrinsic connectivity of the BNST between individuals with MDD and healthy volunteers (HVs). Given prior evidence that MDD is associated with reduced intrinsic connectivity between the amygdala and striatum, the working hypothesis was that connectivity between the BNST and the striatum would be reduced in individuals with MDD.
Because MDD has also been associated with reduced connectivity within the prefrontal cognitive control network, the secondary hypothesis was that MDD would be associated with reduced connectivity between the BNST and PFC. Exploratory analyses were conducted to determine whether differences in BNST connectivity were associated with the severity of anxious or depressive symptoms in individuals with MDD.
Methods
Participants
Data were drawn from several different clinical trials (NCT02543983, NCT02484456, and NCT02122562, conducted under ZIAMH002857 and ZIAMH002927); full details, including inclusion and exclusion criteria for each study, can be found in the Supplement. Participants included 29 HVs (19 F/10 M, 32.44 ± 12.36 years) and 17 individuals with MDD (11 F/6 M, 38.12 ± 13.18 years). Clinical characteristics and demographic information for the current sample are available in Table 1.
Briefly, MDD diagnoses were confirmed using the Structured Clinical Interview for the Diagnostic (SCID) and Statistical Manual of Mental Disorders (DSM-IV, DSM-IV TR, or DSM-5, depending on the original study; see Supplement for additional information). HVs had no personal history of any mood disorder, as confirmed using structured clinical interviews (SCID), or family history (first-degree relative) of any mood disorder. All participants were medically healthy as assessed via a medical history review, physical examination, blood tests, and electrocardiogram.
All data included in this study were collected at baseline, prior to the administration of any placebo or active pharmacological manipulation administered as part of a larger clinical trial. Participants were excluded based on poor data quality and acceptable head motion (see Blood oxygen level dependent (BOLD) fMRI preprocessing and analysis, below). The NIH Combined Central Nervous System IRB approved each individual study, and written informed consent was provided by all participants prior to enrollment in their original study.
Self-report questionnaires
Clinical symptoms were assessed in individuals with MDD using the clinician-administered Hamilton Anxiety Rating Scale (HAM-A)23,24 and the Montgomery-Asberg Depression Rating Scale (MADRS)25 to assess anxiety and depressive symptoms. The HAM-A is a widely used outcome measure for clinical trials and exhibits good internal consistency (coefficient alpha = 0.85).26 MADRS total scores also exhibit reliable internal consistency (ICC = 0.86–0.97) and are sensitive to the effects of clinical treatment.27
MRI data acquisition
MRI data were acquired using a Siemens 7T Magnetom scanner and a 32-channel head coil. A T1-weighted MPRAGE was acquired using the following parameters: repetition time (TR) = 2200 ms, echo time (TE) = 3 ms, voxel size = 0.7 mm isotropic, flip angle 7°, acquisition matrix = 320 × 320, 256 sagittal slices. Resting-state fMRI data were acquired using single echo, echo planar imaging (EPI) sequence using the following parameters: TR = 2500 ms, TE = 25 ms, voxel size = 1.8 mm isotropic, flip angle 50°, acquisition matrix = 108 × 108, 60 oblique axial slices, 250 volumes over the course of 10 min and 25 s.
Two different sequence versions were used to collect the structural and functional MRI data. The first EPI sequence was used to collect resting-state fMRI data for three studies, two of which had a TR = 2500 with 250 volumes (ZIAMH002927; NCT02122562) and one of which had a TR = 2000 with 300 volumes (NCT02543983). A different EPI sequence was used to collect resting-state fMRI data for NCT02484456 (TR = 2500 ms, 250 volumes).
Full details for each study's sequence parameters can be found in Supplemental Table S1. Group-level fMRI analyses statistically controlled for the two different scan sequences.
Blood oxygen level dependent (BOLD) fMRI preprocessing and analysis
Our pipeline was designed to maximize our predictive power at each stage of data processing. Given the small size of the human BNST, preprocessing was performed without smoothing so that BOLD time series could be extracted from the BNST region of interest (ROI) while avoiding partial volume effects. After extracting the BNST time series, the preprocessed fMRI data were smoothed before calculating seed-to-voxel correlations, as prior work has shown that smoothing facilitates the ability to successfully make group-level inferences.28 This strategy of omitting smoothing until the seed time series is extracted is based on prior empirical work that successfully assessed the intrinsic connectivity of the periaqueductal gray matter.29
BOLD fMRI data were preprocessed using fmriprep version 20.2.3. T1 anatomical images were segmented into gray matter, white matter, and cerebrospinal fluid (CSF) using FreeSurfer.30 fMRI data were slice time corrected, realigned to correct for head motion, resampled to 2 mm isotropic resolution, and normalized into standard space using the MNI152NLin2009cAsym template. Independent component analysis (ICA-AROMA)31 was performed for artifact removal.
Framewise displacement (FD) was computed based on the absolute sum of relative motions.32 Principal components were extracted from CSF and white matter masks using CompCor.33 The following nuisance signals were regressed from the BOLD fMRI data: temporal derivatives and quadratic terms,34 frames that exceeded a threshold of 0.3 mm FD, components derived from CSF and white matter images, and a band-pass filter (0.009–0.08). Functional volumes were statistically censored on the basis of head motion (threshold = 0.5 mm mean FD or 75 differential variance (DVARS)). Participants without at least 225 uncensored time points (a minimum of 7.5 minutes of scan time) were excluded to ensure the reliability of our calculations.
The smoothing step was omitted during this stage of preprocessing to ensure that data from surrounding areas were not smoothed into the BNST. Average time series were extracted from the residualized functional images using a probabilistic bilateral BNST ROI (for details, see Regions of interest below). tSNR values from the BNST time series indicated that the temporal stability of these signals was acceptable for analysis (see Supplement). BOLD fMRI data were then smoothed using a 6 mm FWHM kernel, and Pearson's correlations were calculated between the BNST time series and all other voxels in the brain via seed-to-voxel analyses.
The resulting statistical maps were Fisher r-to-z transformed for the purposes of group-level analysis.
Regions of interest
BNST: Bilateral BNST ROIs were defined using the publicly available probabilistic atlas of Torrisi and colleagues35 and thresholded at 50%.
Striatum: Striatal ROIs were defined using the publicly available atlas of Xiao and colleagues36 encompassing bilateral areas of the putamen, caudate nucleus, nucleus accumbens, globus pallidus-internal, and globus pallidus-external.
Frontal lobe: A frontal lobe ROI was defined using the publicly available atlas of Glasser and colleagues37 encompassing all cortical areas of the frontal lobe rostral to Brodmann area 6 (a complete list of included areas can be found in Supplemental Table S2).
Statistical and analytic strategy
To determine whether the intrinsic connectivity of the BNST was associated with a clinical diagnosis of MDD, Fisher-transformed statistical images were entered into the AFNI program 3dMVM38 with MDD diagnosis as a predictor and age, sex, and scan sequence as covariates. Multiple comparisons were corrected based on cluster extent. An initial cluster forming threshold of p < 0.0005 (NN = 1) was selected.
A cluster size of 22 voxels was then calculated as required for significance (cluster level alpha = 0.05) using updated versions of 3dFWHMx and 3dClustSim. Updated versions of these programs incorporate a mixed autocorrelation function (acf) to better model non-Gaussian noise structure.39,40 Correction for multiple comparisons was conducted across an ROI encompassing the gray matter within the frontal lobe and striatum (for details, see Regions of interest above).
To test whether the magnitude of BNST intrinsic connectivity was associated with the severity of clinical symptoms, parameter estimates were extracted from areas where intrinsic connectivity exhibited a significant relationship with clinical diagnosis after surviving correction for multiple comparisons. Whether clinical symptoms predicted parameter estimates of intrinsic connectivity was assessed using multiple regression models while controlling for age, sex, and scan sequence. Importantly, these analyses were restricted to participants diagnosed with MDD to ensure that clinical diagnosis was not driving these relationships.
HAM-A scores and MADRS total scores were entered as predictors into separate models, and the alpha level was set at p < 0.025 to correct for multiple comparisons using the Bonferroni method.
Results
Effect of diagnosis on intrinsic connectivity
Group-level analyses found that, compared to HVs, a diagnosis of MDD was associated with significantly reduced connectivity between the bilateral BNST and areas of the ventral striatum (F = 18.18, cluster size = 27 voxels, x = 21.5, y = 9.5, z = −12.5) (Fig. 1). The cluster identified in the ventral striatum was located anterior and lateral to the BNST seed ROI (see Supplemental Fig.
S1). No other findings were significant after correcting for multiple comparisons. Exploratory analyses at more liberal statistical thresholds (see Supplement) confirmed that the effects of MDD diagnosis on the intrinsic connectivity of the BNST were limited to areas of the striatum.
Associations between clinical symptoms and intrinsic connectivity
Multiple regression models detected no significant effect for HAM-A anxiety scores (B = 0.154, p > 0.1) on parameter estimates of BNST intrinsic connectivity with the striatum in MDD participants. Similarly, the effect of MADRS total scores on extracted parameter estimates also did not reach significance (B = 0.157, p > 0.1).
Discussion
This study sought to determine whether the intrinsic connectivity of the BNST serves as a neural marker of MDD diagnosis and the severity of clinical symptoms. The results demonstrate that MDD was associated with reduced intrinsic connectivity between the BNST and ventral striatum. No association was observed between intrinsic connectivity of the BNST and the severity of anxious or depressive symptoms in individuals with MDD.
The BNST exhibits strong intrinsic connectivity with the striatum during resting-state scans,18,35 and experimentally induced anxiety has been found to decrease intrinsic connectivity between the BNST and ventral striatum.41 A previous meta-analysis also found that MDD was associated with reduced intrinsic connectivity between the amygdala and striatum.16 Both the central nucleus of the amygdala and the BNST are components of the dorsal extended amygdala,17 and the present findings suggest that reduced functional coupling between the dorsal extended amygdala and striatum may be associated with MDD.
MDD is characterized by depressed mood, anhedonia, heightened anxiety, sensitivity to stress, and several other symptoms.42–44 Previous research found that the striatum is a major hub for the neural processing of rewards and incentives.45 Importantly, one meta-analysis demonstrated that individuals with MDD exhibited reduced neural responses within the striatum during the anticipatory and consummatory processing of rewards.46 However, because the MDD participants included in the present analysis were recruited across different clinical trials that used differing scales to measure anhedonia, this study was unable to directly test this hypothesis. Future research should investigate whether reduced intrinsic connectivity between the BNST and striatum in individuals with MDD reflects altered motivational processing.
These findings contribute to the emerging literature investigating neural biomarkers of MDD. Previous research found that BNST BOLD response to stress predicted more rapid remission in patients with late-life depression,47 suggesting that BNST activity may function as a prognostic biomarker that could be used to predict clinical progression in MDD. In contrast, reduced phase transitions within the BNST-nucleus accumbens circuit following DBS were associated with improved clinical symptoms post-treatment,48 suggesting that circuit dynamics between the BNST and ventral striatum could function as a pharmacodynamic/response biomarker to assess response to clinical interventions.
In addition, BNST activity at baseline was found to function as a predictive biomarker, indicating how participants with treatment-resistant depression would respond to DBS of the BNST and nucleus accumbens.49 Finally, BNST activity was found to be elicited by pain and threat processing, and the magnitude of this activity correlated with the severity of clinical symptoms in patients with treatment-resistant depression.50 Future research is needed to determine whether reduced intrinsic connectivity between the BNST and ventral striatum functions as a risk biomarker in individuals who will eventually develop MDD or whether it functions as a diagnostic biomarker, only observed once individuals develop MDD.
Individuals with MDD are a heterogeneous population and differ across multiple factors, including lack of response to different types of clinical treatments and the presentation of comorbid psychiatric disorders.51 In addition, anhedonia symptoms associated with MDD can themselves be due to different factors, including endogenous anhedonia, stress-induced anhedonia, and anhedonia resulting from antidepressant medication.52 As such, it is difficult to determine exactly what is driving these findings that a diagnosis of MDD predicts connectivity between the BNST and striatum. Although the results tested for associations between intrinsic connectivity and symptom severity, the sample was underpowered for this type of analysis. Future studies with larger sample sizes are warranted to determine the degree to which connectivity between the BNST and striatum reflects MDD symptom severity or heterogeneity in clinical presentations.
Limitations
Despite notable strengths, including the use of ultra-high field fMRI to assess BOLD activity within the BNST, several limitations bear mention. First, the small sample size of the MDD group (N = 17) prevented investigation of how demographic factors such as sex, psychiatric comorbidity, treatment history, and age may have impacted the observed relationships. Second, it is unclear from our data whether reduced intrinsic connectivity between the BNST and striatum was a pre-existing risk factor for MDD or only emerged following MDD onset.
Future research should investigate whether clinical treatments for MDD impact connectivity between the BNST and striatum. Finally, although the BNST and striatum are consistently implicated in the processing of sustained threats and rewards, respectively, these data were collected during a resting-state scan. Consequently, any inferences about how reduced connectivity between the BNST and striatum relate to specific clinical manifestations are circumstantial.
Conclusion
These results provide evidence that connectivity between the BNST and striatum is impacted in individuals with MDD. Given that MDD is characterized by increased dysphoria and reduced responsiveness to reward, these findings may reflect one pathway through which clinical diagnosis impacts emotional processing. Previous research demonstrated that BNST activity can predict clinical progression in MDD and response to treatment.
The present results can guide future work investigating risk factors for MDD and provide potential targets for clinical intervention.