Participants
A total of 134 patients with MDD were recruited from the First Affiliated Hospital of Zhengzhou University. The inclusion criteria were as follows: (1) meeting the diagnostic criteria for MDD as specified by the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), (2) presenting a score ≥ 21 on the 24-item Hamilton Depression Rating Scale (HAMD), and (3) being of Han Chinese ethnicity. The following exclusion criteria were used: (1) presenting with comorbid psychotic illness, (2) having a family history of inherited diseases, (3) having organic mental disorders, (4) having psychoactive substance abuse, and (5) contraindications for MRI. MDD patients were divided into the FH group (FH group) and without FH group (NFH group) according to whether their relatives within three generations had MDD.
HCs were recruited from the community through poster advertisement at the same time. Inclusion criteria consisted of (1) having no family history of mental disorders, (2) lifetime absence of psychiatric illnesses and substance abuse, (3) having no contraindications for MRI, (4) having an HAMD score < 7, and (5) being of Han Chinese ethnicity. The study was approved by the Medical Research Ethics Committee of the First Affiliated Hospital of Zhengzhou University. All participants or guardians signed informed consent before the experiment.
Measurement of inflammation
A 5 ml of peripheral fasting blood samples were drawn from the patients at 8:00 AM. The blood samples were placed in an EP tube and stored in a -80 ℃ refrigerator for testing. The plasma levels of C-reactive protein (CRP), homocysteine (HCY), and interleukin-6 (IL-6) were measured by electrochemiluminescence using German cobase 411 kits.
Image acquisition
MRI data was obtained on a 3.0T GE DISCOVERY MR750 scanner (General Electric, Fairfield Connecticut, USA) equipped with a high-speed gradient and an 8-channel head coil. Foam pads and headphones were used to minimize head movement and scanner noise. The main parameters for structural MRI were set as follows: repetition time (TR)/echo time (TE) = 8.2/3.2 ms, flip angle=7°, slice thickness = 1 mm, slices = 188, field of view (FOV) = 256 × 256 mm2, matrix size = 256 × 256, and voxel size = 1 × 1× 1 mm3, thickness = 1 mm, no gap. Functional images were scanned using an echo-planar imaging sequence with the following parameters: TR/TE = 2000/40 ms, slices = 32, matrix size = 64 × 64, flip angle = 90°, FOV = 220 × 220 mm2, voxel size = 3.44 × 3.44 × 3.44 mm3, thickness = 4 mm, gap = 0.5 mm, and a total of 180 volumes.
VBM analysis
The VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm8) implemented in the SPM12 (http://www.fil.ion.ucl.ac.uk/spm) was employed to preprocess the structural T1-weighted images. First, the artifacts of the images were checked, and the origins of the images were adjusted to the anterior commissure. Subsequently, the images of each participant were normalized to the standard Montreal Neurological Institute (MNI) space by using an affine followed by nonlinear transformation and resampled to 1.5 1.5 1.5 mm3. The normalized images were then segmented into gray matter, white matter, and cerebral spinal fluid maps. The data quality of the segmented maps was checked. The probabilistic gray matter maps were further smoothed using an 8 mm full width at half maximum Gaussian kernel.
DFC analysis
Preprocessing of fMRI images was performed using the DPARSF toolbox (http://rfmri.org/dpabi). The pipeline included as following: (1) remove the first 5 volumes (remaining 175 volumes), (2) slice-timing, (3) head motion correction, all participants were retained under the head motion criteria of translation < 2 mm or rotation < 2 in any direction, (4) normalized to the standard MNI space with a voxel size of 3 3 3 mm3, (5) regression of nuisance covariates (i.e., white matter, cerebrospinal fluid, the Friston-24 parameters of head motion, and global signal), (6) spatial smoothed with full width at half maximum = 6 mm, (7) linear detren, (8) filtering (0.01–0.1 Hz). The framewise displacement (FD) across time points for each subject was calculated to assess head motion. Finally, as functional connectivity is sensitive to the confounding factor of head motion, scrubbing was performed for motion correction to reduce the negative influence [20]. If the FD exceeded 0.5 mm, the value of the signal at that point was interpolated using the cubic spline.
The sliding window correlation approach was performed to assess the time-varying dFC of the brain regions identified by VBM analysis. Based on our previous dFC studies [10, 21], the window length was set to 50 TRs (100 s) and the shift step size was 5 TRs (10 s). In each window, the Fisher's z-transformed Pearson's correlation coefficient between the average time series of each seed region and the remaining voxels in the whole brain was calculated. Thus, a set of sliding-window correlation maps for each participant was obtained. Finally, the dFC was estimated by calculating standard deviation at each voxel across sliding-windows.
Statistical analysis
One-way ANCOVA analysis was conducted to investigate the difference in gray matter volume (GMV) and dFC among the HCs, FH, and NFH groups, while controlling for age, gender, and years of education. In order to remove the influence of head movement, the mean FD was also included as a covariate while performing the one-way ANCOVA analysis on dFC. Multiple comparison correction was performed using Gaussian random field (GRF) theory with cluster corrected p < 0.05 and a voxel height of p < 0.005. Post-hoc comparisons were then conducted using two-sample t-tests, and the significance threshold set at p < 0.05/3 (Bonferroni corrected).
Correlation analysis
The identified changed GMV and dFC were then used to test the relationship with clinical and inflammatory variables in FH and NFH groups. Person correlation analysis was conducted between the VBM/dFC values and HAMD, and CRP, HCY, and IL-6 scores, with the significance threshold was set at p < 0.05.
Classification analysis
Classification analysis was implemented to test whether abnormal GMV and dFC of cerebellum can be used as a potential biomarker for distinguishing FH from NFH patients. Moreover, in order to investigate whether the combined structure and function achieve better classification performance than structure or function alone, the identified altered VBM, dFC, and the combination of the VBM and dFC of the cerebellum were used as features, respectively. We selected the radial basis function kernel for the support vector machine classifier as the classification model and employed a grid search algorithm to choose the optimal parameters for the classifier. Additionally, to address the issue of imbalanced data, the Random Under Sampler function was used to randomly remove some samples from the majority class [22], reducing the number of samples in the majority class to a level similar to that of the minority class. Subsequently, a five-fold cross-validation method was applied to obtain stable model performance. Finally, the receiver-operating characteristic curve was implemented to evaluate the classification performance, and the five-fold average area under the curve (AUC) values were reported.