Introduction

Major depressive disorder (MDD) is the leading cause of disease burden globally, ranking second in disability-adjusted life years according to the World Health Organization in 2020 [1, 2]. Repetitive transcranial magnetic stimulation (rTMS), which employs an excitatory pattern over the left dorsolateral prefrontal cortex (DLPFC), has been used as a non-invasive therapeutic approach in patients with treatment-resistant MDD [3] who have not responded to multiple antidepressants [4]. The canonical rTMS protocol for MDD treatment uses high-frequency stimulation (>5 Hz) over the left DLPFC [5,6,7,8]. A new rTMS protocol for MDD therapy has recently been approved by the Food and Drug Administration in the United States, namely the Stanford Intelligent Accelerated Neuromodulation Therapy (SAINT), which adopts intermittent theta burst stimulation over the left DLPFC [9] to exert excitatory effects on the left DLPFC. However, it has limited efficacy, with response rates ranging from 30–50% [10,11,12], primarily because its biological mechanisms are elusive. Therefore, a profound understanding of the biological foundations underlying the rTMS treatment of MDD is essential to improve its clinical efficacy.

Previous genetic, postmortem, and neuroimaging studies have shown that MDD can be due to the dysregulation of the DLPFC circuitry. These findings include alterations in the structure, markers of synaptic neurotransmission, and connectivity abnormalities with downstream structures [13,14,15,16]. Notably, Padmanabhan et al. demonstrated, through resting-state functional magnetic resonance imaging (rs-fMRI), that brain lesion locations that induce depressive symptoms could be mapped into a connected brain circuit centered on the left DLPFC [17]. Conversely, rTMS over the left DLPFC in patients with MDD has been theorized to improve MDD by remotely modulating the distributed depression-related brain regions that are functionally connected to the left DLPFC synergetically [18,19,20,21,22,23,24]. Furthermore, Siddiqi et al. revealed that brain lesions inducing depressive symptoms and the stimulation sites of neuromodulation techniques for MDD converge in the same brain circuitry [25]. These findings suggest a direct causal relationship: damage to the particular brain circuit triggers depression, whereas a targeted therapeutic stimulation of the same circuit alleviates depressive symptoms. This hypothesis was further substantiated by connectomics studies that demonstrated the functionally mutual connection of depression-related brain circuitry with the left DLPFC [26, 27]. These findings highlight the importance of understanding the interactions among the regions in the depression-related brain circuit connected to the left DLPFC to elucidate the disorder’s pathophysiological mechanisms.

Regarding depression-related brain circuit, numerous brain regions functionally connected to the left DLPFC have been identified as critical in the pathophysiology of MDD. These include amygdala (AMY), nucleus accumbens (NAC), anterior insula (AI), subgenual anterior cingulate cortex (sgACC), and ventromedial prefrontal cortex (VMPFC). Previous studies have demonstrated that the rTMS targeting the left DLPFC modulates functional connectivity (FC) between the sgACC and other regions implicated in MDD [28, 29]. Additionally, several brain regions that psychopathologically contribute to MDD symptoms–such as the NAC, AMY, VMPFC, and insula–, functionally converge in the left DLPFC [27, 30,31,32]. Notably, the SAINT stimulation over the left DLPFC have been shown to improve MDD symptoms by changing the directional temporal shift between the sgACC and AI [33, 34]. In addition, the suicide preventive effects of SAINT were associated with the connection related to the insula [35]. Several studies have further revealed that these regions are components of a neural circuit broadly associated with depression. The sgACC, VMPFC and left DLPFC converge within this circuitry, integrating lesion site connectivity and neuromodulation targets [25]. Furthermore, the left DLPFC, sgACC, VMPFC, insula, NAC, and AMY form a distinct depression-specific brain circuit closely linked to emotional dysregulation in patients with MDD, as identified through connectomics methods [27]. These regions exhibit spatial correlations with the depression-related circuit delineated by Siddiqi et al. [25]. MDD results from the disruption of orchestrated functional organizations among the distributed brain regions associated with MDD; therefore, identifying the functional relationships across the brain regions in the depression-related circuit associated with the left DLPFC could provide new insights into the pathophysiological mechanisms of MDD and the biological mechanisms of rTMS treatment in patients with MDD.

Previous research has revealed aberrant FC among brain regions using traditional methods, capturing static temporal characteristics without considering the dynamic aspects of brain activity [33, 36]. However, the dynamics of whole-brain activity are crucial for regulating emotional and cognitive processes [37,38,39,40,41,42]. Dynamic causal modeling (DCM) was recently developed to investigate the dynamic patterns of brain activity by estimating the causal relationships across brain regions of interest [43]. Conventional DCM was first applied to task-based fMRI studies [43, 44]; however, a new DCM method was recently developed to evaluate the causal relationships among distributed brain areas using rs-fMRI data [30, 45,46,47,48,49,50]. Notably, several previous studies have applied this method to the rs-fMRI datasets of individuals with MDD [51,52,53,54]. However, no study has investigated the aberrant dynamic patterns across brain areas focusing on depression-related brain circuitry in patients with MDD. Furthermore, most studies have examined aberrant FC in the brains of patients with MDD using datasets from a single site with small sample sizes [9, 36]. This limitation may have resulted in inconsistent findings across previous studies, hindering our understanding of the neural basis underlying MDD and rTMS treatment.

Here, we aimed to reveal the aberrant dynamic aspects across the regions associated with depression in patients with MDD by employing large-sample, multi-site rs-fMRI datasets comprising 270 healthy controls (HCs) and 175 patients with MDD across three imaging sites. Brain circuits functionally connected to the left DLPFC are crucial in the pathophysiological mechanisms of MDD. Consequently, we hypothesized that regions exhibiting abnormal FC with the left DLPFC might be crucial in patients with MDD. Initially, we identified areas within the whole brain that showed altered FC with the left DLPFC. We then applied DCM analysis to estimate the causal relationships among regions manifesting aberrant FC with the left DLPFC, in addition to the regions in the depression-related brain circuit such as the left DLPFC, AMY, NAC, AI, sgACC, and VMPFC. Determining the aberrant causal relationships across brain areas in the depression-related brain circuit and the regions aberrantly connected with left DLPFC could provide deeper insights into the potential mechanisms of rTMS treatment for MDD.

Materials and methods

Dataset and participants

In total, 445 rs-fMRI images were used from the data obtained from the Department of Psychiatry, Wakayama Medical University (WMU), and the publicly available database of the DecNef Project Brain Data Repository (https://bicr-resource.atr.jp/srpbsopen/) [55]. This open dataset was collected as part of the Japanese Strategic Research Program for the Promotion of Brain Science (SRPBS) supported by the Japanese Advanced Research and Development Programs for Medical Innovation (AMED). Participants with MDD and HCs were recruited for the study from three sites: the WMU protocol, held at WMU; the University of Tokyo (UTO) protocol, held at UTO; and the Center of Innovation at Hiroshima University (COI) protocol, performed at the Hiroshima University. The latter two protocols are included in the SRPBS dataset. Overall, 270 HCs and 175 patients with MDD were recruited using the three protocols. The sample size was determined based on the previous studies performing DCM analysis [51, 52]. Notably, some participants were excluded from the analyses due to their extra head motion and misregistration of their functional image into the standard space. Therefore, 177 HCs and 120 patients with MDD were included in the analysis. Table 1 shows the demographic characteristics of all the participants included in the analyses. The demographic characteristics of all participants and MRI parameters of all sites are summarized in Supplementary Tables 1 and 2. This study was conducted in accordance with the Declaration of Helsinki and approved by the ethics committees of Wakayama Medical University (2954). Written informed consent was obtained from all participants. Figure 1 illustrates the schematics of our study flowchart.

Table 1 Demographic characteristics of the participants included in the analyses.
Full size table
Fig. 1: Flowchart of the study protocol and schematic concept of the potential mechanism of the rTMS treatment for MDD.
Fig. 1: Flowchart of the study protocol and schematic concept of the potential mechanism of the rTMS treatment for MDD.
Full size image

a Flowchart of the entire study protocol is shown. We employed large-sample, multi-site rs-fMRI datasets, comprising 270 HCs and 175 patients with MDD across the three imaging sites. After removing the site differences by applying ComBat harmonization, we explored the regions showing aberrant FC with the left DLPFC in patients with MDD. Furthermore, we performed DCM analysis to explore aberrant causal connections among the brain regions within the depression-related brain circuit and those aberrantly connected with left DLPFC in MDD. b Schematic concept of the potential mechanism of rTMS treatment for MDD is shown. We hypothesized that rTMS treatment mitigated depressive symptoms by modifying the aberrant interactions among the regions in the depression-related brain circuit and those aberrantly connected with left DLPFC in MDD. HC healthy control, MDD major depressive disorder, FC functional connectivity, DCM dynamic causal modeling, rTMS repetitive transcranial magnetic stimulation.

Depressive symptoms were assessed using the 17-item version of the Hamilton Rating Scale for Depression (HAMD) in patients with MDD at WMU. However, for patients with MDD at UTO, the Beck Depression Inventory revised version (BDI-II) was used to assess subjective depression symptoms for 32 patients with MDD, and the Center for Epidemiologic Studies Depression Scale (CES-D) was used to determine the depression symptoms for the 30 patients with MDD. For patients with MDD at COI, BDI-II was used to assess depressive symptoms. HAMD scores were transformed into BDI-II scores based on a study by Furukawa et al. [56]. The BDI-II and CES-D scores were then transformed into the PROMIS depression metric based on the study by Seung et al. [57].

Pre-processing of rs-fMRI data

We used SPM12 (Wellcome Department of Cognitive Neurology, https://www.fil.ion.ucl.ac.uk/spm/) to pre-process the MRI data. We applied slice-timing correction, and the EPI images were realigned to the first image to correct for head movements in the scanner. The corrected EPI images were co-registered to the anatomical T1 image in the native space and normalized into the Montreal Neurological Institute (MNI) space using a unified segment and Diffeomorphic Anatomical Registration through Exponential Lie Algebra (DARTEL) in SPM12. Spatial smoothing was applied to the normalized images (2 mm isotropic voxel) with a Gaussian kernel of full-width half-maximum at 6 mm. We also normalized the segmented gray matter (GM) image to individually specify the voxels in the gray matter to extract the blood-oxygen-level-dependent (BOLD) time series. Furthermore, we normalized the segmented cerebrospinal fluid (CSF) and white matter (WM) images to extract non-neural signals of CSF and WM regions. One patient with MDD in WMU was excluded from the analysis due to the misregistration from the native space to MNI space. Patients with head motion of >2 mm were excluded from the study. Frame-to-frame head motion during the scan was evaluated using frame-wise displacement (FD) [58]. We eliminated participants whose head movement was >2 mm or >30% of the volumes having FD > 0.5 mm (“scrubbing”); therefore, 20 HCs and 16 patients with MDD in WMU, 10 HCs and 4 patients with MDD in UTO, and 63 HCs and 34 patients with MDD in COI were excluded from the analysis. Thus, 297 participants were included in the final analysis. The demographic characteristics of all participants included in the analyses are summarized in the Tables 1, 2.

Table 2 The severity of depressive symptoms.
Full size table

Furthermore, a temporal bandpass filter (0.009–0.08 Hz) was applied for seed-based FC analysis [29, 58]. Non-neural signal sources associated with nine parameters, including the six motion parameters and average signals over the CSF, WM, and whole brain, were removed from the BOLD signal data using linear regression. For the seed-based FC analysis, we performed the scrubbing procedure by removing those volumes with FD > 0.5 mm based on a previous study [59]. FD was determined as the head motion between two consecutive volumes (the summation of absolute displacements in translation and rotation). Regarding the six motion parameters, a two-sample t-test showed no significant difference in head motion between the HCs and patients with MDD. Details of the results of the rs-fMRI sessions are provided in the Supplementary Results. Therefore, we did not include head-motion parameters as nuisance covariates in subsequent analyses.

Region of interest specification for seed-based FC analysis and DCM

The seed mask for the left DLPFC was specified as a 10-mm radius of a spherical region of interest (ROI) centered at [x, y, z] = [−41.25, 42.25, 29.25] in the MNI coordinate (Fig. 2a), which was defined by taking the average of the four MNI coordinates of the left DLPFC adopted in previous studies ([−44, 38, 34] and [−38, 44, 26] [29], [−42, 44, 30] [60], and [−41, 43, 27] [61]).

Fig. 2: Regions which showed abnormal FC with the left DLPFC in MDD.
Fig. 2: Regions which showed abnormal FC with the left DLPFC in MDD.
Full size image

a The left DLPFC was used as a seed region for the FC analysis. b Patients with MDD had a significantly decreased FC between the bilateral Thal and left DLPFC compared with that in HCs (p < 0.05, corrected for multiple comparisons using the permutation method, with age and sex as nuisance covariates). c Patients with MDD had a significantly increased FC between the VIS and left DLPFC compared with that in HCs (p < 0.05, corrected for multiple comparisons using the permutation method, with age and sex as nuisance covariates). The color bar shows the (1-p) values. DLPFC dorsolateral prefrontal cortex, HC healthy control, MDD major depressive disorder, FC functional connectivity, Thal thalamus, VIS visual cortex.

The ROI mask for the bilateral AMY was created using the automated anatomical labeling atlas 3 (AAL3) [62]. Based on previous studies [63,64,65,66,67,68], the ROI for the bilateral NAC included an 8-mm radius of spherical ROIs centered at [x, y, z] = [14, 10, 0] and [−14, 10, 0] (mm in the MNI coordinate). The ROI for the bilateral AI was specified as a 6-mm radius of spherical ROIs centered at [x, y, z] = [–4,25,37] and [−32, 24, −6] [69]. For the sgACC, the ROI mask was created a 6-mm radius of a spherical ROI centered at [x, y, z] = [–10,6,16] based on a study by Fox et al. [29]. For the VMPFC, the ROI was specified as a 10-mm radius of a spherical ROI centered at [x, y, z] = [0, 46, −6], which was defined based on previous reports on the involvement of the VMPFC in decision making [65,66,67].

Furthermore, we incorporated regions demonstrating abnormal FC with the left DLPFC, specifically the bilateral thalamus (Thal) and visual cortex (VIS) areas, as ROIs for subsequent DCM analyses. Using the center of gravity for the clusters identified in the FC analysis, we created the right and left Thal masks as a 6-mm radius of a spherical ROI centered at [x, y, z] = [7.21, −12.8, 8.77] and [−6.78, −14.3, 14.4], respectively, and a VIS mask as a 6-mm radius of a spherical ROI centered at [x, y, z] = [−2.24, −68.9, 6.7]. Figure 3 shows all ROIs in the DCM analysis using BrainNet Viewer (http://www.nitrc.org/projects/bnv/) [70].

Fig. 3
Fig. 3
Full size image

Regions of interest for the DCM analysis. The ROI mask for the bilateral AMY was created using the automated anatomical labelling atlas 3 (AAL3). The ROI for the bilateral NAC comprised an 8-mm radius of spherical ROIs centered at [x, y, z] = [14, 10, 0] and [−14, 10, 0]. The ROI for the bilateral AI was specified as a 6-mm radius of spherical ROIs centered at [x, y, z] = [–4,25,37] and [−32, 24, −6]. The ROI for the sgACC was specified as a 6-mm radius of a spherical ROI centered at [x, y, z] = [–10,6,16]. The ROI for the VMPFC was specified as a 10-mm radius of a spherical ROI centered at [x, y, z] = [0, 46, −6]. The ROI for the bilateral Thal was created as a 6-mm radius of a spherical ROI centered at [x, y, z] = [7.21, −12.8, 8.77] and [−6.78, −14.3, 14.4]. The ROI for the VIS was specified as a 6-mm radius of a spherical ROI centered at [x, y, z] = [−2.24, −68.9, 6.7]. DLPFC dorsolateral prefrontal cortex, AMY amygdala, NAC nucleus accumbens, AI anterior insula, sgACC subgenual anterior cingulate cortex, VMPFC ventromedial prefrontal cortex, Thal thalamus, VIS visual cortex, ROI region of interest, DCM dynamic causal modeling.

Seed-based FC analysis and combat harmonization

We extracted the time course of all voxels included in the ROI for the left DLPFC and averaged them across all voxels in the ROI. FD was used in the subsequent scrubbing procedure [21]. FD was the head motion from one volume to the next and was calculated as the sum of the absolute values of the differentiated realignment estimates at every time point [71]. We removed volumes with FD > 0.5 mm to reduce spurious changes in FC due to head motion [21]. To create a whole-brain FC map, we calculated Pearson’s correlation coefficients between the time course of the ROI and that of all voxels in the GM. The correlation coefficient maps were transformed into z-score maps using Fisher’s r-to-z transformation.

Furthermore, we used the ComBat harmonization method to control for site differences in FC based on the adjusted general linear model harmonization method [72]. ComBat harmonization was examined using the MATLAB toolbox (https://github.com/Jfortin1/ComBatHarmonization/tree/master/Matlab). We applied the transformed two-dimensional data with the participants’ age, sex, and disease factors as covariates, along with protocol effects. The ComBat harmonization effect is shown in Supplementary Fig. 1.

After ComBat harmonization, group-level analysis for group comparison was conducted using a two-sample t-test. A threshold-free cluster enhancement (TFCE) approach was adopted with 10 000 permutations to correct for multiple comparisons, with age and sex as nuisance covariates [73]. The significance level was set at p < 0.05.

DCM

We employed spectral DCM implemented in SPM12 (version R7771, http://www.fil.ion.ucl.ac.uk/spm/software/) for the pre-processed rs-fMRI data. The DCM employs a neuronally plausible model for the observed BOLD signals and allows the estimation of causal relationships between the different nodes of the network. As described previously [30, 51], the details of DCM are provided in the Supplemental Information. Briefly, the BOLD time series of the ROIs were extracted, and non-neural signals of the WM and CSF and six head motion parameters were regressed. We used a fully connected model with bidirectional connections between any pair of ROIs for each participant. We estimated 64 free parameters because the fully connected model contained eight ROIs.

Parametric empirical bayes for group DCM

We used a standard Parametric Empirical Bayes (PEB) analysis process to conduct a group analysis and Bayes model averaging [74, 75]. The PEB took participant-specific connectivity parameters estimated from the spectral DCM to the group level, where they were modeled using a general linear model under the Bayesian hierarchical framework. Therefore, the estimated connection strengths and their uncertainties were considered from the participant to group level in the group analysis. As reported previously [30, 51], the details of these analyses are provided in the Supplementary Material.

First, we compared the causal connections across the networks between HCs and patients with MDD under the PEB framework. Age, sex, and site effects were added as nuisance covariates to the models. Furthermore, we tested the associations between causal connections and depression severity in patients with MDD under the PEB framework, with age, sex, and site effects as nuisance covariates. Depressive symptoms were assessed using the PROMIS Depression Metric [57]. We focused on connection parameters with abnormal connection values in patients with MDD. We adopted the definition of “strong evidence” by thresholding the effects at 95% posterior probability (Pp) [76]. The details of these analyses are provided in the Supplementary Material.

Results

Group comparison of the seed-based FC analysis between HCs and patients with MDD

According to the study flowchart shown in Fig. 1, we initially investigated the regions displaying the aberrant FC with the left DLPFC in patients with MDD. We found that patients with MDD exhibited reduced FC between the left DLPFC seed and bilateral Thal compared with that in HCs and increased FC between the left DLPFC seed and visual areas compared with that in HCs (Fig. 2a–c and Table 3).

Table 3 Regions which showed abnormal functional connectivity with left DLPFC.
Full size table

Group analysis of the DCM analysis between HCs and patients with MDD

Data for each group were confirmed for normal distribution using the Kolmogorov-Smirnov test and homogeneity of variances was verified using Levene’s test. The mean and standard deviations of 64 parameters for HC and MDD groups are shown in Supplementary Fig. 2. We then performed DCM analysis to delineate aberrant causal connections among the brain regions in the depression-related brain circuit and those aberrantly connected with left DLPFC (Fig. 1). The ROIs included in the current DCM analysis are detailed in Fig. 3. DCM revealed that patients with MDD had increased causal connections from the AI to AMY, sgACC to NAC, and Thal to VIS, and increased self-inhibitory connection of the AMY compared with that in HCs (Pp > 0.95; Fig. 4). Furthermore, MDD showed decreased causal connections from the left DLPFC to VIS, AMY to sgACC, AMY to VIS, NAC to sgACC, NAC to VIS, AI to NAC, and AI to VMPFC and decreased self-inhibitory connection of the Thal compared with that in HCs (Pp > 0.95; Fig. 4).

Fig. 4
Fig. 4
Full size image

Group comparison of the causal connection parameters between HCs and patients with MDD. The connection parameters whose posterior probability is more than 0.95 (strong evidence) are shown. Edges colored in yellow to red/blue indicate that the causal connection is stronger/weaker in patients with MDD than in HCs. HC healthy control, MDD major depressive disorder, DLPFC dorsolateral prefrontal cortex, AMY amygdala, NAC nucleus accumbens, AI anterior insula, sgACC subgenual anterior cingulate cortex, VMPFC ventromedial prefrontal cortex, Thal thalamus, VIS visual cortex.

Associations between the DCM connection parameters and depression severity

Finally, we investigated the relationships between causal connections and depression severity in MDD patients (Fig. 1). The causal connection from the AMY to sgACC was positively associated with depression severity. In contrast, connections from the NAC to VIS and from the AI to VMPFC was negatively associated with depression severity in patients with MDD (Pp > 0.95; Fig. 5). Notably, our focus was on those connections significantly correlated with depressive severity (Pp > 0.95) and displaying aberrant causal connections in MDD patients compared to HCs (Pp > 0.95).

Fig. 5
Fig. 5
Full size image

Associations between the causal connection parameters and depression scores. The connections having strong evidence (Pp>0.95) are depicted in color. The connections associated with depression scores and abnormal connections in patients with MDD were those from the AMY to sgACC, AI to VMPFC, and NAC to VIS. MDD, major depressive disorder, AMY amygdala, NAC nucleus accumbens, AI anterior insula, sgACC subgenual anterior cingulate cortex, VMPFC ventromedial prefrontal cortex, Thal thalamus, VIS visual cortex, Pp posterior probability.

Discussion

In this study, we used a large-sample multi-site rs-fMRI dataset of 270 HCs and 175 patients with MDD to investigate the aberrant causal connections between brain areas within the depression-related brain circuitry and brain regions aberrantly connected to the left DLPFC in MDD patients. We found that the Thal and VIS had altered FC with the left DLPFC in patients with MDD. Furthermore, we applied DCM to estimate the aberrant causal connections across brain areas within the depression-related brain circuit, including the left DLPFC, AMY, NAC, AI, sgACC, and VMPFC, as well as Thal and VIS, which were aberrantly connected to the left DLPFC in MDD. We found that patients with MDD had increased causal connections from the AI to AMY, sgACC to NAC, and Thal to VIS and increased self-inhibitory connection of the AMY compared with that in HCs. Additionally, patients with MDD showed decreased causal connections from the left DLPFC to VIS, AMY to sgACC, AMY to VIS, NAC to sgACC, NAC to VIS, AI to NAC, and AI to VMPFC and decreased self-inhibitory connection of the Thal compared with that in HCs. Finally, we found that the causal connection from the AMY to sgACC was positively correlated with depression severity, whereas that from the NAC to VIS and from the AI to VMPFC was negatively correlated with depression severity in patients with MDD. These data suggest that aberrant causal connections across regions in the depression-related brain circuit and those aberrantly connected to left DLPFC in MDD might underpin the neural mechanisms associated with depression severity in patients with MDD.

In this study, we obtained two notable findings. Most previous studies reported that the pathophysiological mechanisms of MDD were mainly due to disruption among the higher-order brain regions [36, 48]; however, one of our main findings was that the aberrant causal connections flowed from multiple depression-related regions (left DLPFC, AMY, NAC, and Thal) to the VIS in patients with MDD, suggesting that the VIS, a primary brain region, may be essential in the mechanisms of MDD. Furthermore, we revealed aberrant causal connections between subcortical brain regions and the VMPFC. Notably, the aberrant causal connections flowed from the AI to multiple regions (AMY, NAC, and VMPFC), sgACC to NAC, and AMY to sgACC in patients with MDD, which has not been reported in previous studies. Our novel findings have not been reported previously. This may be because, first, we explored the aberrant interactions among regions in the depression-related circuitry and the regions aberrantly connected with left DLPFC in MDD using a large-sample harmonized multi-site dataset. Second, we investigated the aberrant causal connections across the ROIs using DCM, which could not be captured using FC analysis.

In addition to the novel findings of altered causal connections to the VIS in patients with MDD, we found negative correlations between depression scores and causal connections from the NAC to VIS. Notably, some recent studies have reported abnormal functional integrity in the VIS in patients with MDD. Yang et al. [77] found that nodal degrees and efficiencies in the VIS were significantly decreased in patients with MDD. Lu et al. [78] showed that FC within the VIS was reduced in patients with MDD and that this reduced FC within the VIS was correlated with the clinical course. Other previous studies reported that the altered FC between the VIS and prefrontal cortex are crucial in the abnormal information updating processes in patients with MDD [79] and that the abnormal FC between the VIS and DLPFC could be used to predict the treatment response to electroconvulsive therapy or rTMS in patients with MDD [80]. Ding et al. [81] showed that FC between the NAC and VIS was increased in patients with MDD. Furthermore, Siddiqi et al. reported that the VIS is included in the depression-related brain circuitry [25], suggesting that damage to the occipital lobe could lead to post-stroke depression [82, 83]. These findings indicate that abnormal relationships among the VIS, subcortical regions, and prefrontal regions could greatly influence the pathophysiological mechanisms of MDD, aligning with our findings.

DCM revealed that patients with MDD had increased and decreased causal connections among certain depression-related regions, including subcortical regions and the VMPFC. Additionally, there was a reduction in self-inhibitory connections of Thal in patients with MDD compared with those in HCs. Individuals with depression show blunted responses to emotion regulation and rewarding stimuli within the corticostriatal circuit, including the AMY, NAC, AI, sgACC, VMPFC, and Thal, suggesting that disruptions in this corticostriatal circuit are fundamental in the pathophysiological mechanism of MDD [81, 84,85,86,87,88,89,90,91,92,93,94]. Furthermore, we found that the causal connection from the AMY to sgACC was positively correlated with depression scores. Previous studies have shown that FC between the AMY and sgACC could be used to predict the treatment response to rTMS [95], cognitive behavioral therapy, and antidepressants [96] in patients with MDD. These studies indicate that FC between the AMY and sgACC is significant for the treatment mechanisms of MDD. However, the positive association between the causal connection from the AMY to sgACC and clinical symptoms contradicted our finding that this connection was decreased compared with that in HCs. This discrepancy suggests that normalizing the abnormal causal connection from the AMY to sgACC may not improve clinical symptoms in patients with MDD, indicating that the aberrant causal connection pattern from the AMY to sgACC in patients with MDD is qualitatively different from that in HCs. We also found that the causal connection from the AI to VMPFC was negatively correlated with depression scores, which aligns with recent findings that the FC between the VMPFC and insula decreased during emotion regulation in patients with MDD [97, 98]. Decreased causal connections from the AI to VMPFC may be important in aberrant emotion regulation in patients with MDD.

To further interpret our findings, we categorized the aberrant causal connections into two groups: (1) interconnections between regions within the depression-related circuitry and those aberrantly connected to the left DLPFC in MDD, and (2) connections within the core brain circuitry intrinsically associated with depression. From this perspective, we observed that abnormal causal flows from multiple ROIs to the VIS predominantly fell into the first category, whereas aberrant connections between subcortical regions and the VMPFC aligned with the second category. This distinction suggests that rTMS might alleviate depression by modulating not only aberrant connections within the core depression-related circuitry but also the connections between the depression-related circuit and regions that are aberrantly connected to the left DLPFC in MDD.

Finally, our results suggest the potential mechanisms underlying rTMS treatment in patients with MDD. The finding that the left DLPFC had abnormal connections with the NAC, Thal, and VIS indicates that excitatory stimulation through rTMS of the left DLPFC may remotely modulate the aberrant activity of the NAC, Thal, and VIS. This mechanism may explain the visuospatial cognitive improvement following rTMS treatment over the left DLPFC in patients with MDD, as reported by Tsai et al. [99]. Consequently, these remote effects may extend to subcortical brain regions associated with depression, including the AMY, AI, and sgACC, potentially leading to the improvement of functional disruption among depression-related regions. Our inference regarding the potential mechanism of rTMS treatment for MDD is supported by previous studies demonstrating that neural circuits associated with MDD are functionally connected to the left DLPFC [25, 29]. Furthermore, a recent TMS/fMRI study directly demonstrated that stimulation over the left DLPFC increases activity in depression-related circuitry, including the left DLPFC, sgACC, AI, and Thal, with effects lasting at least 30 min after rTMS [100], confirming our hypothesis. However, further research is required to corroborate this hypothesis.

Limitations

First, we could not discuss the direct effects of rTMS treatment in patients with MDD because most participants with MDD in our study did not receive rTMS treatment. Second, we had limited medication data for the datasets and did not test for non-medicated patients; therefore, we could not eliminate the effect of medications on the results. Third, we did not have any data on illness duration or clinical states, such as acute or remitted, and we did not estimate their effects on the causal connections among the brain regions. Fourth, rTMS is predominantly administered to individuals with treatment-resistant depression, defined as those who have not responded to at least one antidepressant. Therefore, there may be differences between the MDD sample in this study and the typical treatment-resistant population.

Conclusion

DCM revealed aberrant causal connections among the regions in the depression-related circuits and those aberrantly connected with left DLPFC in MDD, including the left DLPFC, AMY, NAC, AI, sgACC, VMPFC, Thal, and VIS, and their associations with depressive symptoms. Our findings provide deeper insight into the pathophysiological mechanisms of MDD and elucidate the potential mechanisms of rTMS treatment in patients with MDD.