Resting-State Functional Connectivity in Treatment-Resistant Depression

This study was approved by the local ethical committee, and written informed consent was obtained from all participants. The patients in the present analysis were part of a large cohort study of major depression in the Chinese population of Han nationality. Patients were recruited consecutively, and the diagnosis of major depressive disorder was made with the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID) (23). Exclusion criteria were bipolar disorder, any history of major illness, previous psychiatric therapy, cardiovascular disease, age less than 18 or over 60 years, use of vasoactive medications, and alcohol or drug abuse. Originally 82 right-handed patients were recruited, of whom 22 were excluded by the above criteria. Finally, 60 patients were included in the study, none of whom had received antidepressant treatment before enrollment. Severity of depression was quantified using the 17-item Hamilton Depression Rating Scale (HAM-D) (24) and the Clinical Global Impressions (CGI) severity item (25). To be included in the study, patients had to have a HAM-D total score ≥18 and a CGI severity score ≥4 on the day of MR scanning.

After MR imaging, antidepressant treatment was started for all patients. Three classes of antidepressants were used: tricyclics, typical serotonin-norepinephrine reuptake inhibitors, and typical selective serotonin reuptake inhibitors. All antidepressants were empirically prescribed according to the clinical judgment of the treating psychiatrist. No patient was treated with ECT or had received ECT in the past. Refractory depression is defined as a poor response after at least two trials with antidepressants from different classes, with adequate dosages, duration (6 weeks for each trial), and compliance (26, 27). A poor response is defined as a reduction of <50% in HAM-D score with a minimum dosage of 150 mg/day of imipramine or the equivalent for 6 weeks. This outcome measure was chosen because it allows simple analyses that aid interpretation, particularly from a clinical perspective. Nonrefractory patients are those who had a reduction >50% in HAM-D score after treatment.

In addition, 48 right-handed healthy comparison subjects were recruited from the local area by poster advertisements. Comparison subjects were screened using the non-patient edition of the SCID to confirm the lifetime absence of a history of psychiatric or neurological illness and were interviewed to exclude any family history of psychiatric illness.

All participants were found by two experienced radiologists to have no abnormalities on conventional MRI.

Patients and comparison subjects underwent scanning using a GE Signa EXCITE 3-T MR system (GE Healthcare, Milwaukee) with an 8-channel phased array head coil. During scanning, participants were instructed to relax with their eyes closed without falling asleep; after the experiment, each participant confirmed not having fallen asleep during scanning. Participants were fitted with soft earplugs and positioned carefully in the coil with comfortable support. MR images sensitive to changes in BOLD signal (repetition time=2,000, echo time=30 msec, flip angle=90 degrees) were obtained with a gradient-echo echo-planar imaging (EPI) sequence. Five dummy scans were discarded to remove the impact of magnetization stabilization. The slice thickness was 5 mm (no slice gap) with a matrix size of 64×64 and a field of view of 240×240 mm², resulting in a voxel size of 3.75×3.75×5 mm³. Each brain volume comprised 30 axial slices, and each functional run contained 200 image volumes.

Preprocessing and statistical analysis of functional images were carried out using the SPM2 software package (www.fil.ion.ucl.ac.uk/spm/). For each participant, EPI images were slice-time corrected, realigned to the first image in the first series, and unwarped to correct for artifacts due to susceptibility-by-movement interaction. The resulting images were spatially normalized to the Montreal Neurological Institute (MNI) EPI template in SPM2, and each voxel was resampled to 3×3×3 mm³. The processed images were smoothed with an isotropic Gaussian kernel (full-width at half-maximum=8 mm). Functional connectivity was examined using the Resting-State fMRI Data Analysis Toolkit (REST) software package (http://resting-fmri.sourceforge.net) using a seed voxel correlation approach (28, 29). As structural and functional studies in patients with depression have revealed regional deficits in the limbic-thalamo-prefrontal and limbic-striatal-pallidal-thalamic systems (1–9, 11–15), we selected as seeds 13 areas that constitute these: the left and right hippocampus, insula, dorsal lateral prefrontal areas, amygdala, putamen, and thalamus and the anterior cingulate cortex. Using REST, after bandpass filtering (0.01–0.08 Hz) (21) and linear trend removal, a reference time series for each seed was extracted by averaging the fcMRI time series of voxels within each region of interest as defined in the WFU (Wake Forest University) PickAtlas (30, 31). The Ideal Filter was used for bandpass filtering by transforming the time series into the frequency domain by discrete Fourier transform, assigning a value of zero to the excluded frequency, and then transforming back to the time domain by inverse discrete Fourier transform. Thirteen correlation analyses were performed voxel-wise between each seed reference and the rest of the brain. Finally, the correlation coefficients in each voxel were transformed to z-value images using the Fisher r-to-z transformation to improve normality before averaging across subjects. Using SPM2, the impact of potential physiological artifacts such as cardiac or respiratory noise (32, 33) was minimized by regressing out components with high correlations with CSF or white matter or low correlations with gray matter.

For the patient data, individual z-value maps were analyzed with a random-effects one-sample t test to identify voxels showing a significant positive or negative correlation with the seed time series, the correlations being thresholded using a p<0.05 family-wise error correction for multiple comparisons. Voxel-based comparison of z-value maps among the three groups was performed using a design model of one-way analysis of variance with age and disease duration as covariates followed by post hoc two-sample t tests. The statistical significance of each region was estimated by distributional approximations from the theory of random Gaussian fields (34). In this method, clusters in smooth areas are shrunk while those in rough areas are expanded to account for differences in smoothness (34). Significance thresholds were set at 0.05 after family-wise error correction with an extent of more than five contiguous voxels. MNI coordinates were transformed to Talairach coordinates using mni2tal (http://imaging.mrc-cbu.cam.acuk/download/MNI2tal).

Relative to the comparison group, both the nonrefractory and refractory groups showed significantly reduced connectivity within prefrontal-limbic-thalamic areas bilaterally (Table 2). The nonrefractory group showed the more distributed decrease in connectivity, especially in the anterior cingulate cortex and the left and right prefrontal cortex, hippocampus, insula, and amygdala (Table 2, Figure 1A), while in the refractory group decreased connectivity was mainly in prefrontal areas and the thalamus areas bilaterally (Table 2, Figure 1B). Direct comparison between the groups showed decreased connectivity in the nonrefractory compared with the refractory group within the left amygdala-anterior cingulate cortex-right insula-precuneus region (Table 2, Figure 1C). These findings were not correlated with illness duration or age.