Shared but distinct functional connectome profiles underlying rumination in depressed and healthy individuals

Abstract

Background

Rumination—a passive, repetitive focus on negative thoughts and emotions—is a key risk factor for the onset and recurrence of depression. While prior studies have linked rumination to altered functional connectivity (FC) in brain networks, most relied on self-reported measures and resting-state fMRI, potentially missing key neural dynamics during active ruminative states.

Methods

We recruited 43 patients with major depressive disorder (MDD) and 42 matched healthy controls (HCs). Using a validated rumination state task (RST), we induced active rumination and distraction states during fMRI scanning. We constructed whole-brain functional connectomes and examined both FC patterns and a graph-theoretical feature (eigenvector centrality) across conditions. We also examined associations between the graph-theoretical feature and depressive symptom severity.

Results

Both MDD patients and HCs exhibited enhanced FC between the core and medial temporal lobe (MTL) subsystems of the default mode network (DMN) during rumination compared to distraction. However, edge-level analyses revealed group differences: HCs showed greater FC increases between the ventral attention network (VAN) and the DMN-MTL subsystem. We found a significant interaction effect regarding eigenvector centrality in a VAN node (Par_Med_4_R). During rumination, patients with MDD exhibited elevated eigenvector centrality compared to distraction, whereas HCs did not. This elevation correlated positively with depression severity.

Conclusions

This study provides the first comprehensive characterization of the active rumination connectome in MDD. Results highlight a shared DMN mechanism and a VAN-specific topological alteration in MDD, supporting the theory that excessive automatic constraints from attentional systems may drive maladaptive rumination.

Clinical trial number

Not applicable

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07724-0.

Introduction

Rumination is typically defined as passive and perseverative thoughts that focus on one’s depressive symptoms, as well as their possible causes and implications [1]. It is well established that rumination is associated with the onset, maintenance, and increased relapse risk of depressive episodes [2–4]. Recent research has implicated rumination as a viable and plausible target for potential novel clinical interventions [5]. Thus, understanding rumination’s neural underpinnings is crucial as it paves the way for the development of next-generation therapy that targets the key neural process underlying rumination.

The brain is intrinsically organized into different large-scale networks [6, 7], which consist of integrated regions (nodes) that are interconnected heavily (edges). Such an intrinsic functional connectome enables functional specialization and integration, and may reorganize to support different psychological processes [8]. Moreover, these nodes and edges can be viewed as a complex network that could be analyzed through the graph theoretical approach [9, 10]. A recent theoretical framework [11] posits that rumination, an automatic and constrained form of spontaneous thought, is associated with a certain interaction pattern among brain networks. Thus, understanding the association between the pattern of such a network, its graph theoretical features, and rumination could shed important light on rumination’s neural underpinnings. Indeed, accumulating evidence from recent research using resting-state functional MRI (rs-fMRI) and electroencephalogram (EEG) has found an association between self-reported ruminative traits and functional connectivities (FC) among a wide range of brain regions in both healthy adults and patients with depression [12–20].

Although these brain-wide association studies have brought important insight into rumination’s neural mechanisms, some noteworthy limitations remain. First, to achieve a replicable and robust association between self-reported, questionnaire-measured traits and phenotypes derived from fMRI, a large sample size may be required [21]. Second, most existing studies correlated the self-reported ruminative trait with the FCs during a typical resting state, which raised the possibility that individuals with high trait rumination may not be ruminating during the resting-state measurement [22]. In light of these caveats, it might be plausible to first induce individuals into an active, subject-driven “ruminative state” and characterize the functional connectome features during such a state. Milazzo and colleagues have found that FCs during an active rumination state could be leveraged to classify it from other mental states (e.g., a typical resting state) using a machine learning algorithm in a group of healthy individuals [23]. The ruminative state was also found to be associated with altered whole-brain FC profiles of the posterior cingulate cortex in patients with MDD [24]. However, the inducements used in these studies may not align with the methods commonly used in the literature [25, 26]. Besides, a typical resting state was used as the control condition, which could be problematic because participants might engage in spontaneous rumination [1]. Accordingly, we have developed a rumination state task (RST), which uses prompts modified from the rumination induction task (RIT) [27] and the ruminative response scale (RRS) [28]. A well-designed control condition, the distraction state, was also included. During the distraction state, individuals were asked to vividly think about an object, which differs from the rumination state in many dimensions and can help prevent subjects from engaging in ruminative thoughts. With this task, we have demonstrated that FCs between the core subsystem and the MTL subsystem of the default mode network were enhanced during rumination, and this effect could be replicated across 3 different MRI scanners in a group of healthy adults [29]. In a follow-up study, we found that the FC profiles of the superior frontal gyrus (SFG) were altered in patients with MDD compared to healthy controls (HCs) [30].

Existing evidence has implicated that depressive rumination is associated with a specific functional connectome pattern, which could be altered in patients with MDD. Comprehensively characterizing the whole-brain functional connectome and analyzing its graph-theoretical features during an active rumination state could provide essential insights into the neural processes underlying rumination and offer crucial empirical evidence for the recent theoretical frameworks [11, 31]. Here, the objectives of the present study were to characterize the functional connectome during the rumination/ distraction states and explore any abnormalities in patients with MDD, analyze the graph theoretical features of the nodes and identify the key node that is associated with MDD, and explore the relationship between the graph theoretical features and the depressive symptoms. We hypothesize that (1) both patients with MDD and healthy controls will exhibit significantly altered FCs among the subsystems of DMN, particularly between the core and MTL subsystems; (2) Patients with MDD will show altered whole-brain functional connectome organization during rumination, reflected in abnormal graph-theoretical properties in specific nodes, relative to HCs; (3) Individual differences in graph-theoretical features of these key nodes will be associated with depressive symptom severity.

Methods

Participants

The present study is a reanalysis of a previously reported cohort [30]. While our previous study examined degree centrality during the rumination state, here we conducted entirely new analyses addressing the functional connectome features underlying rumination in depression. Participants in this study were outpatients and inpatients with MDD recruited from Guangji Hospital in Suzhou, Jiangsu, China, from December 2021 to September 2022. We recruited HCs from the local community. MDD patients were included if they (1) met DSM-5 criteria for a current major depressive episode, (2) were aged between 18 and 45 years, and (3) scored 17 or greater on the 17-item Hamilton depression rating scale (HAMD). MDD patients and HCs were excluded if they had (1) acute physical illness, (2) MRI contraindications, (3) alcohol or substance abuse history, (4) pregnancy or breastfeeding, (5) chronic neurological disorders, (6) head trauma, or (7) left-handedness. Exclusion criteria that only applied to MDD patients include: (1) bipolar disorder, (2) psychotic disorder or current psychotic symptoms, (3) intellectual disability, (4) autism spectrum disorder, (5) delirium, and (6) neurocognitive disorder. HCs with any lifetime psychiatric disorder were also excluded. An initial diagnosis was provided by the attending clinician, followed by final diagnostic confirmation by F.-N. J., a clinically active psychiatrist and author of the study. A total of 139 participants were initially screened, and 13 patients with MDD and 6 HCs were excluded as they failed to meet the predefined inclusion criteria.

The final sample for analysis consisted of 43 MDD patients and 42 HCs (see Table 1). One patient did not complete the scan and was excluded. One patient and two HCs were removed due to missing demographic information. Excessive motion (mean framewise displacement > 0.2 mm) led to the exclusion of 13 patients and 9 HCs. We also excluded one left-handed patient. Finally, two patients were excluded due to technical issues. For further details, please refer to our previous publication [30]. No significant differences regarding the sex ratio, age, education, and head motion were revealed between MDDs and HCs.

Table 1.

Demographic and clinical information of participants

Characteristic	HC (N = 43)	MDD (N = 42)	P value*
Sex			0.8
Female	33 (77%)	31 (74%)
Age (years)	28.95 (10.12)	25.86 (7.60)	0.11
Education (years)	13.58 (1.91)	12.88 (2.25)	0.13
Drug naive	NA	14 (33%)
Missing	NA	3
HAMD	NA	23.24 (5.59)
Head motion (rum.)	0.09 (0.03)	0.10 (0.04)	0.3
Head motion (dis.)	0.08 (0.03)	0.08 (0.03)	0.6

All values are mean (SD) except Sex and Drug naive, which reported as n (%)

*Pearson’s χ² test; Two sample t-test

HC, healthy control; MDD, major depressive disorder; rum., rumination; dis., distraction

Clinical measures

One of the coauthors, F.-N. J., who is a practical psychiatrist, administered the 17-item HAMD [32] to patients with MDD. The HAMD used in the current study consisted of 9 items (depressed mood, guilt feelings, suicidal impulses, work and interests, retardation, agitation, anxiety psychic, anxiety somatic, hypochondriasis) rated on a 5-point scale (0–4; 0 = absent, 1 = doubtful or mild, 2 = mild to moderate, 3 = moderate to severe, 4 = very severe) and eight items (initial insomnia, middle insomnia, delayed insomnia, gastrointestinal, general somatic, sexual interests, loss of insight, weight loss) rated on a 3-point scale (0–2; 0 = absent, 1 = doubtful or mild, 2 = clearly present). The total score of the 17-item HAMD ranges from 0 to 52.

Image acquisition and preprocessing

All images were acquired on a 3T SIEMENS MAGNETOM Skyra MRI scanner using a 32-channel whole-brain coil (SIEMENS Medical, Erlangen, Germany). The preprocessing of fMRI data was conducted using the toolbox for Data Processing & Analysis for Brain Imaging on Surface (DPABISurf) [33], which is based on DPABI [34] and runs in a MATLAB R2020a platform (The MathWorks Inc., Natick, MA, US). For details, please refer to the Supplementary Information.

Rumination state task

All participants underwent a rumination state task, which has been validated in both healthy individuals [29, 35] and patients with MDD [30]. In general terms, this task consists of four sessions: a resting state, a negative autobiographical memory state, a rumination state, and a distraction state. Each session lasted for 8 min. Only the fMRI data of the rumination and distraction states were analyzed in the present study. The order of these two sessions was counter-balanced across participants. During the rumination condition, participants were instructed to reflect on themselves according to prompts such as “Think: Why can’t I handle things better in these events I just remembered?” or “Think: Why do I encounter these events other people don’t?”. On the other hand, they were instructed to imagine and focus on specific external topics according to prompts such as “Think about: A train stopped at a station” or “Think about: The layout of a typical classroom” during the distraction state. All prompts were adapted from the rumination inducement task [25, 36] and RRS [37]. A total of 4 prompts lasting 2 min were displayed consecutively without inter-stimulus intervals in each state. For details regarding this task, please refer to our prior publications [29, 30, 35]. Participants also reported their momentary emotional states at the end of each session. These manipulation-check data have been fully reported and validated in our prior publications using the same task [30]. Briefly, previous findings showed significantly increased negative affect after the rumination condition, whereas emotional states returned to baseline after the distraction condition.

Construction of the functional connectome

A functional connectome was constructed by computing Pearson’s correlation among the average time series of 384 cortical regions of interest (ROIs) covering the entire surface [38], resulting in a 384 × 384 matrix for each participant. The raw Pearson’s correlations then underwent Fisher’s r-to-z transformation and served as FCs. These ROIs were picked from the original 400 parcellation atlas and could be grouped into 17 large-scale networks. A total of 16 brain regions were not assigned to any networks and were excluded from the analyses. We further grouped these networks into 6 networks (visual network, VN; somatomotor network, SMN; dorsal attention network, DAN; ventral attention network, VAN; limbic network, LN; frontoparietal network, FPN) and 3 subsystems of the default mode network (DMN) [6, 39]. These three subsystems of DMN, the core, DMPFC, and MTL subsystems, have been well established in the literature [39–41] and are theoretically central to the mechanisms of rumination, allowing for a more meaningful and functionally grounded interpretation of DMN involvement [42]. In contrast, although subsystem frameworks have been tentatively proposed for other networks (e.g., the frontoparietal network [43, 44]), there is no comparable consensus regarding their functional subdivision, nor is their relationship to rumination well understood. To avoid introducing arbitrary distinctions, these networks were therefore analyzed at the whole-network level. To assign subsystem labels to the Schaefer atlas, we mapped the ROIs described in the Andrews-Hanna framework onto the Schaefer parcellation: each network was assigned to the subsystem whose spatial location and functional characterization it best matched. This approach leverages the natural three-part DMN organization inherent to the Schaefer-17 network atlas and ensures consistency with the canonical functional architecture proposed in the theoretical literature. Within-network (or within-subsystem) FC was computed by averaging the connectivity values across all ROI pairs belonging to the same network or subsystem. Conversely, between-network (or between-subsystem) FC was calculated by averaging the connectivity values across all ROI pairs belonging to two different networks or subsystems.

Graph theory metric

We used a graph theoretical metric, eigenvector centrality, to characterize each node’s topological feature. The eigenvector centrality characterizes a node’s importance in a given network according to its connection to other important (central) nodes [45, 46]. This makes eigenvector centrality particularly suitable for detecting nodes that participate in hierarchically organized, large-scale information flow, such as that proposed for rumination-related processes within the DMN and its interactions with other large-scaled brain networks. We calculated eigenvector centrality using the DPABINet [47] and the following formula. Let A = (a_{i, j}) denote the adjacency matrix of a graph, which is the functional connectome here. The eigenvector centrality x_i of node I was calculated by Eq. (1).

1

Where 𝜆 ≠ 0 is a constant. 𝜆 is calculated by Eq. (2).

2

𝜆 is the largest eigenvalue for the absolute values of A. Eigenvector centrality was calculated at sparsity thresholds ranging from 0.10 to 0.34, with an interval of 0.01. This ranging sparsity threshold was adopted to alleviate the thresholding [48] and head motion effects [49]. This specific thresholding range was picked to discern prominent topological properties in brain networks (high global and local efficiency) [49]. The area under the curve (AUC) across the sparsity range was calculated for further statistical tests. Here, AUC represents an integrated summary of eigenvector centrality across thresholds. The resulting AUC value for each node was then used as the dependent variable in subsequent statistical analyses (e.g., group comparisons and mixed-effects models).

Statistical analysis

We conducted a Pearson’s χ² test and a two-sample t-test to compare the demographic and clinical features between MDD patients and HCs. For illustration purposes, we first conducted a one-sample t-test, with head motion as a covariate, at the edge level. The t-values were used in Fig. 1 as a measure of connectivity strength. We used a general linear model (GLM) to decompose a two-way mixed effect ANOVA of a two-level fixed within-subject factor (rumination vs. distraction) and a two-level random between-subject factor (MDD patients vs. HCs). The mixed-effect analysis examined the Interaction effect, the Condition effect, and the Group effect simultaneously. The interaction effect specifically reflects whether the difference between the rumination and distraction conditions differs between patients with MDD and healthy controls (HCs). The condition effect reflects the overall difference between the rumination and distraction conditions across both groups, whereas the group effect captures the overall difference between patients with MDD and HCs across both conditions. This mixed-effect analysis was conducted at both network/subsystem and edge levels. Head motion was included as an additional covariate. Given a significant Condition effect was revealed, we then conducted paired t-tests in MDD patients and HCs separately using GLM with head motion as an additional covariate. Again, these paired t-tests were conducted at both network/subsystem and edge levels. Multiple comparison correction was conducted using FDR correction (q < 0.05 for the mixed effect analysis, q < 0.05/2 for the follow-up paired t-tests). At the edge level, we calculated the ratio of the number of significant edges within and between networks/subsystems to the total number of edges for better interpretation. To formally compare the ratio of edges showing significant rumination-related effects, we further conducted a two-proportion chi-square test. We conducted a mixed-effect analysis on each node’s AUC of eigenvector centrality, with head motion as a covariate. Multiple comparison correction was performed using the FDR correction. We also conducted a Pearson’s correlation analysis to examine the association between the arithmetic difference of the AUCs of eigenvector centrality between rumination and distraction, and the patients’ depressive symptom severity (HAMD score). Given the exploratory nature of the present study, we did not pre-register the analysis pipeline.

Fig. 1.

The whole-surface functional connectome profile was altered during rumination as compared to distraction states. The distances among nodes represent the strength of functional connectivity. Nodes showing stronger functional connectivities are closer to each other. Color denotes the networks nodes belong to. The node showing a significant interaction effect regarding the eigenvector centrality is displayed as a triangle and highlighted with an arrow. Notably, the spatial proximity between the DMN core (highlighted in yellow) and the DMN MTL subsystem (highlighted in green) was markedly closer during rumination compared to distraction, suggesting a tighter functional coupling between these subsystems under ruminative states

Results

Demographic and clinical features

A total of 42 MDDs and 43 HCs were included in the statistical analysis; no significant differences were found regarding sex ratio, age, education, or head motion (all P values > 0.05).

Functional connectome characteristics of different mental States

As depicted in Fig. 1, one-sample t-tests were used to characterize the whole-brain functional connectome patterns for each mental state, providing readers with an intuitive overview of the connectivity structure before proceeding to the subsequent statistical comparisons. We observed similar functional connectome patterns in the rumination and distraction states of MDD patients and HCs. Nodes from the identical network/subsystems are grouped together, while those belonging to different networks/subsystems are divided apart. Nevertheless, some nuanced alterations were notable: nodes from the core and the MTL subsystems were closer together during rumination than during the distraction state, highlighting connections later identified as statistically significant in our follow-up analyses. Furthermore, the VAN node Par_Med_4_R—later identified through the mixed-effects analysis as exhibiting a significant effect in eigenvector centrality—was also highlighted.

Functional connectome alterations during rumination

No significant interaction or group effect was revealed at either the network/subsystem or nodal levels in the mixed-effects analysis. We found a significant Condition effect between the core and the MTL subsystems (t(82) = 6.40, P < 0.001, Cohen’s f² = 0.50). Follow-up paired t-tests were conducted. As depicted in Fig. 2., we found significantly enhanced FCs between the core and the MTL subsystems during rumination compared with distraction in both MDD patients and HCs (MDD: t(40) = 4.12, P < 0.001, Cohen’s f² = 0.42; HC: t(41) = 4.62, P < 0.001, Cohen’s f² = 0.52, Fig. 2). As shown in the accompanying violin plots, considerable inter-individual variability was observed; however, the vast majority of participants exhibited increased core–MTL connectivity during rumination relative to distraction (MDD: 30 increased, 12 decreased; HC: 36 increased, 7 decreased). Again, we found no significant interaction or group effect when conducting the mixed-effect analysis at the edge level. Significant condition effects were revealed in a large number of edges. Similarly, we performed paired t-tests in MDD patients and HCs separately to interpret these condition effects (Fig. 3). As expected, a large proportion of FCs (~ 20% in MDD patients vs. ~30% in HCs) between the core and the MTL subsystems were significantly enhanced in the rumination condition compared to the distraction condition. Intriguingly, a relatively high percentage of FCs (~ 14%) between VAN and the MTL subsystem was significantly enhanced in HCs but not in MDD patients (~ 4%) during the rumination condition compared to the distraction condition (Fig. 3B). This difference was highly significant (χ²(1) = 37.41, P < 0.001). Only a small proportion of FCs (< 10%) were decreased during the rumination condition compared to the distraction condition. Of note, a large number of edges from the FPN, VAN, and DAN showed significant alterations during rumination compared to distraction in both MDDs and HCs.

Fig. 2.

Shared network alterations during the active rumination state at the network level in MDD patients and HCs. Both patients with MDD and HCs showed significantly enhanced FC between the DMN core and MTL subsystems during rumination as compared to distraction. ***: P < 0.001. Abbreviations: VN, visual network; SMN, somatomotor network; DAN, dorsal attention network; VAN, ventral attention network; LN, limbic network; FPN, frontoparietal network; DMN, default mode network; MTL, medial temporal lobe; MDD, major depressive disorder; HC, healthy control; Rum, rumination; Dis, distraction

Fig. 3.

Distinct alterations in functional connectivity at the edge level in patients with MDD and HCs during rumination as compared to distraction. Note the larger ratio of edges showing enhanced functional connectivity between VAN and DMN MTL subsystem in HCs during rumination. (A) Edges showing significantly different functional connectivties during rumination as compared to distraction in two groups. (B) The ratio of significantly increased/decreased edges in each network. Abbreviations: VN, visual network; SMN, somatomotor network; DAN, dorsal attention network; VAN, ventral attention network; LN, limbic network; FPN, frontoparietal network; DMN, default mode network; MTL, medial temporal lobe

Eigenvector centrality alteration during the rumination condition

We found a significant interaction effect in an ROI, Par_Med_4_R, that is located in the right medial parietal lobe (F(1,82) = 18.37, P_adj = 0.04, Cohen’s f² = 0.22). Post-hoc analysis showed that the eigenvector centrality’s AUCs were significantly increased in MDD patients (t(40) = 4.70, P < 0.001, Cohen’s f² = 0.38) but not in HCs (t(41) = -0.60, P = 0.55, Cohen’s f² = 0.01) during the rumination condition compared to the distraction condition. Again, as shown in the accompanying violin plots, the majority of participants showed the same direction of effects despite the notable inter-subject variations (MDD: 35 increased, 7 decreased; HC: 21 increased, 22 decreased). Besides, eigenvector centrality’s AUCs of MDD patients were significantly higher than HCs during rumination (t(82) = 4.02, P < 0.001, Cohen’s f² = 0.20) but not during distraction (t(82) = -0.38, P = 0.70, Cohen’s f² = 0.002). Furthermore, we found that the arithmetic difference regarding Par_Med_4_R’s eigenvector centrality between rumination and distraction conditions was significantly associated with the patients’ depressive symptom severity (HAMD scores, r(40) = 0.35, P = 0.02) (Fig. 4).

Fig. 4.

One node from VAN, Par_Med_4_R, showed a significant interaction effect regarding the eigenvector centrality. (A) The surface render of Par_Med_4_R. (B) The AUCs of Eigenvector centrality of Par_Med_4_R was significantly increased during rumination in patients with MDD but not in HCs. (C) Eigenvector centrality across a density range between 10% and 34%. Each point and error bar denote the mean and standard deviation at each density level, respectively. Asterisks indicate a significant difference at this density threshold. (D) The arithmetic differences of eigenvector centrality between rumination and distraction were related to the severity of depressive symptoms of patients with MDD. ***: P < 0.001. Abbreviations: MDD, major depressive disorder; HC, healthy control; Rum, rumination; Dis, distraction; AUC, area under the curve

Discussion

We systematically characterized the whole-brain functional connectome during an induced, active rumination state in MDD patients and HCs. We found that the FCs between the DMN core and DMN MTL subsystems were enhanced during rumination compared to distraction. This enhancement was shared between the MDD patients and HCs. In the follow-up, more nuanced edge-level analysis, we also found involvement of nodes from VAN, DAN, and FPN, implicating a complex association between these networks and depressive rumination. We further investigated the topological features of this “rumination connectome” using the eigenvector centrality as the metric, and found a significant mixed effect in a node from the VAN. The arithmetic change of this node’s eigenvector centrality is associated with the depressive symptom severity of MDD patients. Our results suggest a shared but distinct rumination connectome in MDD patients and HCs. They highlighted the pivotal role of the functional coupling between DMN core and DMN MTL subsystem in the neural basis of rumination. The present findings also showed that the alteration of the topological features of the nodes, as characterized by VAN, distinguished rumination in MDD patients.

Consistent with our previous findings [29], we found an enhanced FC between the DMN core subsystem and the MTL subsystem in both patients with MDD and HCs. This result is in line with the longstanding notion that depressive rumination is associated with an abnormally increased FC within the DMN [42]. Several recent large-scale rs-fMRI studies have failed to find increased DMN FC in patients with MDD compared to HCs [50, 51]. Our results implicated that rumination might not be associated with a widespread enhanced FC across the DMN but only involves certain subsystems of the DMN, and such enhancement might only be observed during an active ruminative state. The MTL subsystem facilitates more context-specific, imagery-driven cognitive processes, including episodic memory, episodic future thinking, and the generation of novel thoughts. In contrast, the core subsystem supports the self-referential processes, emotion, and social functions [41, 52]. In our recent study using an intracranial electroencephalogram (iEEG) [53], we found that the low-frequency power in the precuneus (a pivotal region of the core subsystem) was increased, and the high-gamma band power in the hippocampus (a subcortical region that is closely coupled with the MTL subsystem) was reduced during rumination as compared with distraction. To sum up, existing empirical evidence supports the notion that the constraint imposed by the core subsystem on the MTL subsystem may be the central mechanism of rumination. Additionally, this mechanism can be observed in both patients with MDD and healthy adults. This pattern suggests that the DMN–MTL coupling identified here may reflect a fundamental cognitive mechanism that is engaged during internally oriented, self-referential processing in both MDD patients and healthy adults. As rumination can also occur in individuals without psychopathology, the absence of a group effect may indicate that this neural mechanism is shared across populations, even if its frequency, persistence, or clinical impact differs in depression. The functional coupling between other brain networks and DMN could lead to the uncontrollability of rumination in depression. It is worth noting that our findings not only align with the previous results regarding the FC edge but also the direction of the effect. Such robustness highlights the reproducibility of the active induced rumination state and implicates a potential clinical usage, such as the identification of potential targets of brain stimulation [54, 55].

The recently proposed dynamic framework of thought (DFT) has conceptualized rumination as a specific type of spontaneous thought [11, 31]. According to the DFT, rumination occurs when attention and thought processes are excessively constrained by automatic influences, leading to a state in which the individual becomes stuck in repetitive thinking. Automatic constraints refer to a set of mechanisms that function independently of cognitive control, such as sensory or affective salience, which limit attention and shape the flow of thought. On the neural level, automatic constraint is hypothesized to be mediated by brain regions in the core subsystem, VAN, and DAN. Our findings support the theoretical framework by demonstrating strong VAN-related automatic constraints during rumination. Although the cross-sectional design does not allow causal inference, the observed VAN involvement may reflect a core neural mechanism that contributes to—or at least accompanies—maladaptive rumination in depression. On the contrary, the automatic constraint from the core subsystem might be the shared process in both patients with MDD and HCs. Indeed, even though we failed to observe any mixed effect in VAN on the network level, we did find a smaller portion of significantly increased VAN-MTL subsystem edges in MDD (~ 4%) as compared to HCs (~ 14%) during rumination, as compared to distraction. Furthermore, the significantly enhanced eigenvector centrality in MDD during rumination might reflect stronger FCs with other “rich club” nodes. Supportively, a recent longitudinal rs-fMRI study using static and dynamic FC analyses in the IMAGEN cohort (N = 595) revealed that rumination is differentially associated with large-scale network alterations, particularly in the VAN and DMN, and that these patterns partially mediate changes in internalizing and externalizing symptoms during the transition to adulthood [56]. Together, these results indicate an abnormal FC profile of VAN in patients with MDD during rumination.

We found a significantly positive association between the eigenvector centrality of a particular node in VAN and the HAMD score, which is in line with a substantial body of literature [1]. For example, Tang and colleagues [57] have recently reported a positive association between the rumination trait and HAMD scores in a relatively large sample of patients with MDD. Our results provide a possible network mechanism for this behavioral association. VAN controls the attention process alongside DAN [58]. The VAN comprises the temporoparietal junction (TPJ) and ventral frontal cortex, and is primarily anchored in the right hemisphere. The VAN is involved mainly when unexpected events occur and distract an individual’s attention from the ongoing task [59]. Dong and colleagues [60] recently analyzed the developmental changes of VAN. They reported that VAN might serve as an intermediary between unimodal (sensorimotor) and transmodal (default mode) networks, likely playing a key role in information flow and integration across brain systems. Specifically, the revealed VAN node is embedded in a circuit which is believed to be involved with the processing of physical and social information and encoding of potential plans accordingly [61]. This region has also been found to be selectively activated during memory tasks [62]. The more prominent role of this VAN node in the functional connectome of patients with MDD during active rumination might make them more vulnerable to the emotionally salient information from both sensorimotor and default mode network regions, leading to a more severe subjective sense of depressive symptoms. In addition, heightened involvement of this region may bias patients toward internally oriented, temporally extended forms of cognition—such as dwelling on past events or anticipating future threats—rather than attending to the present moment. Because present-moment awareness is a core component of mindfulness, such a shift away from the “here and now” may further undermine psychological well-being [63].

Our study highlighted two potential brain networks (DMN and VAN) that could serve as targets for brain stimulation techniques such as transcranial magnetic stimulation (TMS) [64] or transcranial direct-current stimulation (tDCS) [65, 66]. In two seminal studies [67, 68], Fox and colleagues found that rTMS studies positioning the stimulation coil nearest to the dorsolateral prefrontal cortex (DLPFC) site most negatively correlated with the subgenual anterior cingulate cortex (sgACC) tended to show the largest reductions in depressive symptoms. These findings indicated that fMRI could be leveraged to guide non-invasive brain stimulation [69]. Furthermore, Siddiqi et al. [70] demonstrated that different symptoms might correspond to different brain networks, and better treatment could be achieved by tailoring targets according to these brain networks. One challenge to using fMRI to get the targets for brain stimulation is the low replicability of fMRI [71, 72]. We noted that the FC between the core and MTL subsystems of DMN might be a particularly robust target since the elevated FC between these two networks could be reproduced across different sites and samples [29]. These findings, together with prior evidence such as Chou and colleagues [73], that fMRI-identified circuits—such as DMN and VAN components—may represent promising candidates for future neuromodulation studies. However, the present study does not test interventions, and further research is needed to determine their causal or clinical relevance.

The present study provides a detailed characterization of whole-brain functional connectivity during an active, well-controlled rumination state. To optimize the reproducibility and transparency of our findings and facilitate future research on the neural underpinnings of rumination, we have openly shared the raw MRI data, clinical data, and codes used for analysis. We hope this dataset, alongside our previously shared dataset of a group of healthy adults [29], could serve as a repertoire for future interested researchers. Despite the strengths of this study, several limitations should be acknowledged. The sample size of the present study is relatively modest, which may limit the replicability and generalizability of the connectivity effects identified. Despite this limitation, an important strength of the present work lies in the use of a validated, experimentally controlled rumination task. This paradigm allows us to isolate the neural mechanisms engaged explicitly during active rumination, providing a level of cognitive specificity that is not achievable with resting-state designs [74]. A subset of the included patients with MDD were receiving antidepressant medication during the study. Due to the limited sample size, we did not systematically investigate the effects of drug usage. Future studies could aggregate a larger sample of first-episode, drug-naïve patients to rule out the effect of drugs. Rumination is associated with the dynamic features of the brain’s functional connectome [18, 35], while the present study only investigated the static functional connectome during active rumination. Recent theoretical frameworks have proposed that rumination is a transdiagnostic psychological process [4, 75]. The current study only includes patients with MDD; future studies could investigate the differences and similarities regarding rumination’s functional network characteristics across a range of diagnoses, such as anxiety, bipolar disorder, and psychotic disorder. Notably, considerable inter-individual variability was observed in our primary results, consistent with the well-documented heterogeneity in both depressive symptoms and ruminative trait [1]. Although the group-level effect was directionally consistent across most participants, this variability underscores the need to interpret group-level findings with caution. Larger datasets and designs explicitly modeling individual differences (e.g., dimensional symptom measures or multilevel models) will be crucial for refining the generalizability of these results.

In summary, this study systematically characterized the functional connectome underlying active, continuous rumination in patients with MDD and healthy controls. We identified a shared network feature across groups—namely, increased functional connectivity between the core and MTL subsystems of DMN. Additionally, a significant interaction effect emerged for the eigenvector centrality of a region within the VAN; notably, the difference in centrality between rumination and control conditions was positively correlated with depression severity. These findings suggest that rumination may be driven by excessive top-down constraint from the DMN core to the MTL subsystem, while heightened VAN involvement may contribute to the uncontrollability of rumination in MDD. Our results also highlight the potential of using fMRI-guided non-invasive brain stimulation to target depressive rumination.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Concept and design: Xiao Chen, Chao-Gan Yan. Acquisition, analysis, or interpretation of data: Xiao Chen, Feng-Nan Jia. Drafting of the manuscript: Xiao Chen, Feng-Nan Jia. Critical revision of the manuscript for important intellectual content: Xiao Chen, Feng-Nan Jia, Fei-Yi Liu, Yan-Song Liu, Xun Liu, Chao-Gan Yan. Statistical analysis: Xiao Chen. Administrative, technical, or material support: Chao-Gan Yan. Supervision: Xun Liu, Chao-Gan Yan.

Funding

This work was funded by the National Natural Science Foundation of China (No. 32300933, No. 82572192, No. 82122035, No. 81671774, and No. 81630031), the Beijing Nova Program of Science and Technology (No. 20230484465), the Beijing Natural Science Foundation (No. J230040), the Sci-Tech Innovation 2030 - Major Project of Brain Science and Brain-inspired Intelligence Technology (No. 2021ZD0200600), the National Key R&D Program of China (No. 2017YFC1309902), the Key Research Program of the Chinese Academy of Sciences (No. ZDBS-SSW-JSC006), the Scientific Foundation of Institute of Psychology, Chinese Academy of Sciences (No. E2CX4425YZ, No. E3CX1315 and No. Y9CX422005), the Key Project of the Suzhou Municipal Applied Basic Research Program (Medical and Health Sciences, No. SYW2024008), the key diagnosis and treatment program for Suzhou Municipal Health Commission (No. ZDXM2024015).

Data availability

All the raw brain imaging data and clinical data have been made openly available (10.17605/OSF.IO/WTS7F). All codes related to the analysis and plotting of the present study are openly shared through https://github.com/XiaoChenPhD/rum_network_stable.

Declarations

Ethics approval and consent to participate

All participants have provided voluntary, informed consent at the screening visit stage. The study protocol was approved by the Institutional Review Board of the Institute of Psychology, Chinese Academy of Sciences, and conducted in accordance with the Declaration of Helsinki and relevant national regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.