In this study, patients were recruited from the Department of Psychiatry, the First Hospital of Shanxi Medical University, and met the following criteria: (1) Major depression disorder (MDD) diagnosis with an actual mild-to-moderate episode according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria; (2) Score 7 < 24-item Hamilton Depression Rating Scale (HAMD-24) scores <24; (3) 18 years ≤ age ≤ 50 years; (4) right-handedness; (5) no comorbid axis I diagnosis that is all psychological diagnostic categories; (6) no history of neurological diseases; and (7) no physical limitations prohibiting them from undergoing an MRI examination. Two patients were excluded because of incomplete clinical evaluation. In addition, all healthy controls were recruited from the community nearby. The criteria for healthy controls included: (1) HAMD-24 scores < 7; (2) no family history of psychosis; and (3) no psychological problems and symptoms of psychopathology.

Finally, 47 right-handed MMD patients (17 men; mean age 26.12 ± 5.06 years) and 43 age- and sex-matched right-handed healthy controls (13 men; mean age 26.27 ± 5.69 years) participated in the study. This study was conducted on the basis of the latest version of the Declaration of Helsinki Declaration of 1975, as revised in 2008. The Ethics Committee of Shanxi Medical University approved this study. Participants included in this study all signed informed consent forms.

Depressive and anxiety symptoms were assessed based on the HAMD-24 items and Hamilton Anxiety Rating Scale (HAMA)—14 items by two research psychiatrists independently (Bodescu et al., 2015). Both scales have been confirmed to have good reliability coefficients. In addition, patients with MMD and healthy controls also underwent neurocognitive assessment via Wechsler Memory Scale (WMS), which has presented good reliability and validity (Bodescu et al., 2015). The WMS includes 10 subtests, particularly questions that would determine the participants’ long-term memory, short-term memory, and transient memory. For MMD patients, Pearson’s correlation analyses between the amygdala volumes, Wechsler memory, and severity in depression and anxiety symptoms were performed at baseline while controlling for nuisance variables, including age, sex, and years of education.

After pretests, MMD patients included were assigned to take a Chinese herbal medicine named SG, which is mainly made up of Acanthopanax and H. perforatum (Yang, 2015), licensed and widely used in China since 2008 (Sun et al., 2009). According to the recommended medication dose, MMD patients took 1.44 g of SG per day, including 0.72 g in the morning and 0.72 g in the evening for 8 weeks. During the treatment, all participants were acquired to complete the HAMD assessment for two time points (4 and 8 weeks).

Clinical outcome was defined by the rating of the HAMD scales. Remission was scaled as a HAMD score < = 7, and treatment failure was scaled as HAMD score > 7 at both weeks 4 and 8.

All participants lay supine on the bed of a Siemens Skyra 3.0-T MRI scanner with a standard 32-channel head coil to collect the MRI data at the Shanxi Provincial People’s Hospital. T1-weighted anatomical data were acquired by covering the entire brain using the MPRAGE sequence (repetition time/echo time = 2,300/2.95 ms, FA = 9°, data matrix = 225 × 240, 160 slices, slice thickness = 1.2 mm). Functional image was obtained using an EPI sequence (repetition time/echo time = 2,500/3.0 ms, FA = 90°, field of view = 240 × 240 mm, data matrix = 64 × 64, slice thickness = 4 mm, and 32 slices, 212 volumes).

FreeSurfer was used to analyze the structural MRI data. The details are described elsewhere (Dale et al., 1999), and it has been shown to be robust in exploring cortical and subcortical features (Rosas et al., 2002; Salat et al., 2004). Briefly, the primary preprocessing included Talairach transformation, removal of the non-brain tissue, and tissues segmentation. Three experienced researchers checked tissue segmentation. Subsequently, we obtained the subcortical volumes from the automated procedure implemented in FreeSurfer (Liu et al., 2016). The subcortical volumes (i.e., left thalamus, left hippocampus, left nucleus accumbens, left amygdala, right thalamus, right hippocampus, right nucleus accumbens, and right amygdala) were extracted and compared between the two groups. One-way analysis of covariance was performed to explore the subcortical volume differences after adding the effects of age, sex, and total brain size as nuisance variables (Davis et al., 2017; Reddan et al., 2018). Furthermore, correlations between gray matter volumes and depression scores were calculated by using Pearson’s analysis.

Functional MRI data were processed and analyzed using FSL tools (FMRIB’s Software Library, version 6.00). Preprocessing steps included motion correction using MCFLIRT (Jenkinson et al., 2002), distortion correction with field map (FUGUE), and removal of non-brain structures using brain extraction tool (BET) (Smith et al., 2002), then the image was smoothed using a 5-mm FWHM Gaussian filter. The resulting functional MRI (fMRI) image was normalized to the structural image using a boundary-based registration (BBR) tool (Greve and Fischl, 2009). Then, the structural image was normalized to the MNI152-2mm template using linear and non-linear registrations (FLIRT and FNIRT) (Jenkinson et al., 2002).

The fMRI image was examined by researchers using a two-step quality checking procedure. First, data were excluded due to poor quality (movement >1 mm). Second, the registration quality is assessed by the amount of warp distortion according to the BBR-based function-to-structure realignment. Three subjects (two for MMD patients and one for healthy control groups) were not included in the analysis due to head movements during the fMRI scans.

Because structural MRI analysis revealed smaller right amygdala volume in MMD patients than in healthy controls, we decided to select the right amygdala as the seed for the further functional MRI analyses. The seed of the right amygdala was defined from the Harvard Oxford subcortical structural atlas (90% threshold).

Resting-state functional MRI (rs-fMRI) analysis was conducted using a general linear model according to the well-documented procedures (Fox et al., 2006). By using brushes, we draw nuisance regions of interest of cerebrospinal fluid (CSF) and white matter (WM) in the standard MNI152 template layer by layer and then save them as a mask. The nuisance masks in the standard space are then converted to individual functional space, and the time series of CSF and WM were generated for each participant. Time signal extraction from the seed and nuisance regions was then performed for each subject. The mean time signal of each individual subject’s seed was set as an explanatory variable (EV) with realignment parameters, and the time series of nuisance regions were set as controlled variables. Significant differences in functional connectivity between MMD patients and healthy controls were determined using an independent sample t-test. The group-level statistical images were corrected using the parametric family-wise error method at the cluster level, cluster-forming threshold by Z > 2.3, and a corrected cluster significance threshold of p < 0.05.

Logistic regression analyses were used to explore the relationship between a dichotomized measure [remission (HAMD score ≤ 7) and treatment failure (HAMD score > 7)] assessed at both weeks 4 and 8 and potential factors [age, sex, right amygdala-nucleus accumbens (NAc) functional connectivity, and right amygdala volume] for calculating sensitivity, specificity, and the area under the receiver operating characteristic curve.

A bootstrapped mediation analysis was performed to examine the relationships among the initial right amygdala volume/right amygdala-based functional connectivity, Wechsler memory, and treatment efficacy (HAMD_baseline—HAMD_{4–week/8–week}). The PROCESS macro in SPSS was used with 5,000-bootstrap samples, which identified 95% confidence intervals (CIs) for model components. With the Wechsler memory as the mediator, we tested two models: (1) right amygdala volume as the independent variable and treatment outcome as the dependent variable and (2) coupling between right amygdala and NAc as the independent variable and treatment outcome as the dependent variable. These results determined the indirect effects of the Wechsler memory on right amygdala features and treatment outcome, yielding the 95% CIs of the indirect effects. A significant mediation occurred when the 95% CIs did not include zero (Hayes and Preacher, 2014).

The flow chart of the participant enrollment and follow-up in this study is summarized in Figure 1. Demographic (i.e., age, sex, and year of education) and clinical variables (HAMD and HAMA), as well as WMS in MMD patients and healthy controls (HCs), are presented in Table 1. The clinical variables, as quantified by HAMD (t = 13.245, p < 0.001) and HAMA (t = 10.151, p < 0.001), were significantly larger in MMD patients than HCs. In addition, neurocognitive characteristics, as measured by WMS (t = 2.332, p = 0.022), were significantly weaker in MMD patients than HCs (Table 1 and Figure 2, top panel). These suggest that MMD patients are not only accompanied by negative emotions but also with memory deficits.

For MMD patients, Person’s correlation results demonstrated that both HAMA and WMS scores were not significantly related to HAMD score (HAMA vs. HAMD: r = 0.054, p = 0.772; WMS vs. HAMD: r = 0.030, p = 0.841; Figure 2, bottom panel); Additionally, only HAMA and WMS scores yielded a significant result in MMD patients (r = −0.292, p = 0.046; Figure 2, bottom panel).

Structural MRI results showed that the gray matter volume of the left thalamus, right thalamus, and right amygdala in MMD patients was significantly smaller than that of HCs after using variables of age, sex, education level, and total brain size as regressors (left thalamus: F = 8.000, p < 0.001; right thalamus: F = 8.912, p < 0.001; right amygdala: F = 5.728, p = 0.005; Table 2 and Figure 3; p < 0.006, corrected with Bonferroni correction).

The gray matter volume of the right amygdala was negatively correlated with HAMD score (r = −0.362, p = 0.012; Figure 3), whereas the gray matter volumes of the left and right thalamus were not correlated with HAMD score in MMD patients (r = −0.171, p = 0.273; r = −0.154, p = 0.318; Figure 3). No significant results were detected between the two groups in other subcortical and cortical volumes (left hippocampus: F = 2.785, P = 0.046; right hippocampus: F = 2.324, P = 0.081; left NAc: F = 3.262, p = 0.025; right NAc: F = 0.521, p = 0.669; left amygdala: F = 2.841, p = 0.064; Table 2; p < 0.006, corrected with Bonferroni correction).

Results from rs-fMRI indicated that the right amygdala showed significantly weaker functional connectivity with the NAc and stronger functional connectivity with the cerebral cortex in MMD patients than in HCs (Z > 2.3, p < 0.05 cluster-wise corrected; Table 3 and Figure 4, top panel). Interestingly, results identified a marginally significant positive relationship between HAMD score and right amygdala–NAc functional connectivity in MMD patients (Figure 4, bottom panel).

Forty-seven patients from the original sample returned for follow-up evaluation after SG treatment at two time points (4 and 8 weeks), and the depression severity (assessed by HAMD) are presented in Table 4.

We used logistic regression to combine potential measures into a signal signature of outcome groups after weeks 4 and 8 of SG treatment separately. At 4 weeks, the model generated by the logistic regression analysis is not significant (F = 0.593, p = 0.988). At 8 weeks, the logistic regression analysis resulted in a significant model (F = 14.100, p = 0.007), showing that treatment outcome was significantly predicted by age (t = 4.358, p = 0.037) and right amygdala volume (t = 9.708, p = 0.002), whereas sex (t = 1.165, p = 0.280) and right amygdala–NAc functional connectivity (t = 1.165, p = 0.280) were not significant. This model provided a good prediction of developing depression after 8 weeks of SG treatment [area under the receiver operating characteristic curve = 0.858 (95% CI 73.6–98.1%); p < 0.001, Figure 5].

Mediation analysis revealed the indirect effect of right amygdala volume on treatment efficacy via Wechsler memory in patients with MMD [direct effect = −0.001, p > 0.05; indirect effect = −0.003, 95% CI (−0.0067, −0.0001), Figure 6, left panel]. This result preliminary provides evidence of Wechsler memory in mediating the indirect role of right amygdala reduction on treatment efficacy. In contrast, the effect of right amygdala and NAc functional connectivity on treatment efficacy was not mediated by the Wechsler memory in patients with MMD [direct effect = 1.841, p > 0.05; indirect effect = −0.165, 95% CI (−3.1959, 2.5793), Figure 6, right panel].

It is worth noting that there are some limitations worth considering in this study. First, the relatively small sample size limited the exploration of the relationship between subcortical dysfunctions and symptoms in MMD patients. Second, further studies are needed to investigate the extent to which our findings are specific to MMD to further explore commonalities and differences between psychiatric disorders linked to amygdala morphology. Third, we only focus on the effect of SG treatment in this study; there is no placebo, which prevents us from exploring the effect and mechanism of placebo. Although this study has limitations, it also raises further suggestions of initially right amygdala volume served as a significant predictor to identify the treatment efficacy after 8 weeks of SG treatment for MMD patients, and especially, the effect of initially right amygdala volume on treatment efficacy was mediated by the Wechsler memory. In the future, these results could be replicated and summarized in psychiatric disorders by using longitudinal studies or large sample size studies. A more detailed understanding of the anatomy of the amygdala underlying depression in MMD will help to formulate more targeted treatment protocols.