Shared and distinctive brain networks underlying trait and state rumination

Although trait and state rumination play a central role in the exacerbation of negative affect, evidence suggests that they are weakly correlated and exert distinct influences on emotional reactivity to stressors. Whether trait and state rumination share a common or exhibit distinct neural substrate remains unclear. In this study


Introduction
Rumination, as described by Nolen-Hoeksema and colleagues, is a persistent, self-focused, and negatively toned pattern of thinking, often revolving around past events [1].Previous research highlights the association of rumination with various psychiatric disorders including non-suicidal self-injury, anxiety, substance abuse, and eating disorders [2], indicating its broad impact across different mental health issues.In particular, rumination is considered to be an important factor in the development and maintenance of major depression [3].Even in healthy individuals, ruminative tendencies increase the risk of depression [4].This insight shifts the focus from rumination as a mere characteristic of depression to a potential causal factor.
Rumination is a multifaceted construct that can be divided into stable trait rumination and momentary/state rumination [5].Trait rumination refers to an individual's habitual tendency to engage in passive and repetitive self-focus in response to distress over time [1].
State rumination refers to ruminating in the present moment or in response to an acute stressor [6].State rumination can gradually evolve into habitual trait rumination [5].Both state and trait rumination play a crucial role in exacerbating negative affect and cognition (e.g., depression and negative memory bias) [7,8].However, there is also evidence indicating that the two forms of rumination are weakly correlated and exert different influences on emotional reactivity [6,[9][10][11].For example, Zoccola and Dickerson showed that trait rumination may not predict actual ruminative responses to novel or distressing circumstances [11].Accordingly, distinguishing between trait and state rumination is crucial, as trait questionnaires have limited efficacy in predicting individual differences in stressor-specific rumination following challenging laboratory tasks [11].Furthermore, Hilt and Aldao et al. found that state rumination but not trait rumination was associated with increased negative emotional reactivity [9].Trait rumination may preclude reactivity to certain types of stressors but is more likely to be associated with chronic changes in reactivity over time, whereas state rumination may be a more heterogeneous and dynamic process that exhibits greater covariation with reactivity [9].These findings suggest that there are meaningful differences between state and trait rumination.
The above studies have shed light on both the similarities and differences between these two forms of rumination.However, the potential neural mechanisms underlying these similarities and differences are largely unknown.Previous neuroimaging studies conducted in healthy samples have shown that trait and state rumination are associated with activity or connectivity of brain regions in the Default Mode Network (DMN), Salience Network (SN), Control Network (CN) and Dorsal Attention Network (DAN) [12][13][14][15][16][17][18].For example, trait rumination was found to be positively correlated with activity in the dorsal anterior cingulate cortex (dACC) when successfully inhibiting a response to negative information [14].Healthy individuals with higher levels of trait rumination showed decreased functional connectivity (FC) within the anterior DMN (i.e.medial prefrontal cortex (MPFC) and ventral ACC), as well as increased FC from the anterior DMN to the posterior DMN (i.e.precuneus/posterior cingulate cortex (PCC) and bilateral angular gyrus) and the DAN [16].Furthermore, state rumination during resting state assessment was negatively associated with regional homogeneity (ReHo) in the right dorsolateral prefrontal cortex (DLPFC) [12].Activity in the dorsal MPFC increases when attention is specifically directed to self-referential or introspective mental activity [18].Seed-based FC analysis revealed higher connectivity of the striatum with the inferior frontal gyrus in participants reporting more unwanted thoughts [12].These findings have provided evidence for the neural bases of trait and state rumination, but the question of whether they share common or exhibit distinct neural substrates remains unresolved.
In this study, we aimed to identify the shared and/or distinct brain connectivity features of trait and state rumination in heathy subjects using fNIRS technique and connectome-based predictive modeling (CPM) [19].fNIRS is harmless, tolerant to body movements, highly portable, and suitable for all possible participant populations and experimental settings, both inside and outside the laboratory, in addition to achieving high temporal resolution and reasonable spatial resolution [20].The CPM algorithm, developed to predict brain-behavior relationship at an individual level [19,21], has been applied in recent fNIRS studies [22,23].Our approach involved constructing separate predictive models for trait and state rumination to investigate their unique functional connectivity (FC) characteristics.Moreover, to explore the possibility of shared connectivity profiles between trait and state rumination, we applied the FC model developed for one type of rumination (trait or state) to predict the other.Given that previous studies highlighted the DMN, SN, CN and DAN involved in rumination [12][13][14][15][16][17][18], we hypothesized that the FC patterns within these canonical networks might serve as predictive markers for both trait and state rumination.

Participants
Forty-three (15 men, 28 women) healthy subjects with no history of neurological or psychological disorders, diabetes, and acute or chronical coronary heart disease were obtained from a previous study [24].The study was approved by the Ethics Committee of the University Hospital and the University of Tu¨bingen.All participants gave informed consent in accordance with the Declaration of Helsinki.The demographic information was presented in Table 1.

Assessment of rumination
State rumination was assessed using a visual analogue scale ranging from 0 % to 100 %.Specifically, participants were asked to approximately rate the amount of time spent on ruminative activities following a resting state functional near-infrared spectroscopy (fNIRS) measurement.Trait rumination was assessed using the subscale rumination of the Ruminative Response Scale (RRS) [25].

Data acquisition and preprocessing
Seven-minute resting-state fNIRS data were acquired using an ETG-4000 Optical Topography System (Hitachi Medical Company, Tokyo, Japan) with a sample rate of 10 Hz.Subjects were instructed to sit still with their eyes closed and keep their head as still as possible, think of nothing and let their thoughts flow during the scan.As described in Goldbeck et al. (2019), the probe set (consisting of 52 channels) were placed in the form of a rectangle over parietal areas covering the precuneus, to measure parts of the DMN.The 52-channel probe set was positioned using reference points Pz (Channel 16), T3 (Channel 43) and T4 (Channel 52) of the international 10-20 electrode system [26].Cortical parcellation corresponding to each of the 52 channels was determined using the nfri_anatomlabel function in the NIRS-SPM toolbox, based on the Yeo 7 network and the Automated Anatomical Labeling (AAL) atlas [27,28].
High-and low frequency noise was removed using a bandpass filter (0.01-0.1 Hz).Correlation-based signal improvement [29] as well as component-based nuisance removal were performed to reduce motion artefacts.The preprocessed channel signals were visually inspected, and those noised channels were interpolated by the surrounding channels.Finally, the global components in fNIRS signals were removed using the principal component spatial filter algorithm [30].

Connectome-based predictive modeling
For each participant, the FC matrix (52*52) was computed by Pearson's correlation between all channel's signals using the rsHRF toolbox [31].The resultant FC values were converted to z-scores by Fisher's z-transform to improve the normality.
CPM was conducted to predict individual differences in trait and state rumination [19,32,33].The code is available at https://github.com/YaleMRRC/CPM.Briefly, CPM takes FC matrices and rumination scores as input to generate a predictive model of trait/state rumination from FC matrices.A cross-validated confound regression was performed prior to feature selection to control for confounds (i.e., age and sex) in the prediction analysis [34].The positive and negative features in the FC matrices were identified from the training dataset using Pearson correlation analyses with a significance threshold of p < 0.05.The sum of the FCs in positive features and sign-inverted FCs in negative features for each subject is entered into a robust linear regression model.The resulting regression model is then applied to the test data set to predict trait/state rumination.
The leave-one-out cross-validation (LOOCV) was used to test the model performance.Model performance was evaluated using Pearson correlation coefficient (r) between the actual and predicted rumination scores.To generate null distributions to assess the statistical significance of r, we randomly shuffled the observed rumination scores 5000 times and re-ran the prediction procedure with each shuffled data.Statistical significance was set at p < 0.05.

Results
The CPM algorithm reliably predicted trait (LOOCV: r = 0.408, p = 0.011) and state rumination (LOOCV: r = 0.390, p = 0.008).Fig. 1 and Fig. 2 shows the spatial distribution of FC patterns associated with trait and state rumination, respectively.Specifically, FC within the DMN and DAN as well as FC between the DMN, CN, DAN and SN were predictive for trait rumination.In contrast, a limited number of features contributed to the prediction of state rumination, primarily involving FC between the DMN and CN.To further explore unique FC patterns associated with trait and state rumination, we also reported results in terms of brain regions using the AAL atlas (see supplementary Figure S1 and Figure S2).FC networks associated with trait rumination showed bilateral superior parietal gyrus (SPG) (within the DAN) and precuneus (within the DMN) as connectivity hubs, whereas FC networks associated with state rumination showed the left inferior parietal lobule (IPL) (within the CN) as a connectivity hub.
To investigate the potential common neural substrates underlying trait and state rumination, we used FC networks related to trait (or state) rumination as fixed features for predicting state (or trait) rumination with the CPM procedure.The results showed that FC networks related to trait (or state) rumination significantly contributed to the prediction of state (or trait) rumination.However, compared to the data-driven CPM procedure mentioned above, both trait (LOOCV: r = 0.333, p = 0.016) and state rumination (LOOCV: r = 0.276, p = 0.042) showed relatively lower predictive efficacy.According to the results of the Steiger's Z-test, there was no significant difference between the two methods (p > 0.05, Steiger's Z-test).

Discussion
This is the first brain imaging study to apply the FC-based machine Fig. 1.Predictive network features for trait rumination based on Yeo-7 network atlas.A positive (orange color) or negative (blue color) predictive feature indicates that higher FC is associated with higher or lower trait rumination scores respectively.learning algorithm for predicting trait and state rumination in healthy controls at the single subject level.Examining rumination in a cohort of healthy individuals may attenuate potential confounding effects of concurrent comorbidities and pharmacological interventions, thereby promoting a more nuanced understanding of rumination as an independent cognitive process.CPM has identified FC within the DMN and DAN, as well as FC between the DMN, CN, DAN and SN, which contributed to the prediction of trait rumination.Conversely, a limited number of FC were predictive of state rumination, primarily involving FC between the DMN and CN.In addition, the predictive features of trait rumination can be robustly generalized to predict state rumination, and vice versa.These findings indicate that both the DMN and non-DMN systems are important for rumination.Furthermore, trait and state rumination share common but also exhibit distinct connectivity profiles.
Regions of the DMN, characterized by deactivation during goaldirected cognitive tasks and increased activity in self-referential processing [36,37], were proposed to work collectively and adaptively to facilitate different forms of self-generated thoughts [38].The DMN is widely recognized as the key brain network underlying rumination [39].Previous neuroimaging findings have suggested that FC within the DMN may predict individual differences in ruminative tendencies [16,40].In the present study, connectivity of the parietal and temporal regions of the DMN were predictive of trait rumination, indicating that the dorsal medial prefrontal cortex (DMPFC) subsystem, which mainly participates in theory of mind and mental simulation [17,39], is involved in trait rumination.In addition, FC within the DAN can be used to predict trait rumination.The DAN, which shows increased activation during cognitive task performance, has been implicated in the top-down control of attention [41,42].Reduced attentional control, such as attentional disengagement, plays a critical role in the emergence and maintenance of brooding responses [43].Trait rumination has been associated with difficulties in attentional disengagement from negative self-referential information [44][45][46].The current findings provide further evidence for the DAN connectivity associated with individual ruminative tendencies [47].
In addition to intra-network connectivity, FC between the DMN, CN, DAN and SN was also predictive of trait rumination.The CN and DAN underpin various cognitive functions such as cognitive control, goaldirected thinking, and planning [48,49].Previous research has shown rumination traits are associated with cognitive control deficits [50], thereby suggesting a plausible link between trait rumination and CN/DAN functioning [47,51,52].The SN, conceptualized as a bottom-up processor of both internal and external salient events, is responsible for initiating cognitive control by signaling the engagement of the CN/DAN while suppressing DMN activity [53,54].Specifically, the SN can suppress the DMN to promote disengagement from internally directed thoughts, thereby facilitating the activation of CN/DAN for goal-directed behavior [55].Furthermore, the triple network model of Fig. 2. Predictive network features for state rumination based on Yeo-7 network atlas.A positive (orange color) or negative (blue) predictive feature indicates that higher FC is associated with higher or lower state rumination scores respectively.psychopathology proposes that aberrant SN detection and mapping of external and internal stimuli can lead to abnormal engagement of the CN/DAN and DMN, which may underlie depressive rumination in adults [51,56,57].Consistent with previous studies, our results suggest that functional integration of the triple network (i.e., the SN coordinated switching between the DMN and CN/DAN) contributes to trait rumination [58].
Relative to trait rumination, state rumination has been predicted by a limited number of FC features, primarily involving connectivity between the DMN and CN.Using typical task-based designs, researchers observed increased activity in the DMN regions during rumination compared to the distraction state [59].For the CN, neuroimaging evidence suggests that state rumination is related to functional stability in the CN regions [60].Previous findings have demonstrated the involvement of the DMN and CN in state rumination.Of particular interest for the present study is the functional coupling between the DMN and CN.As already mentioned, functional interactions between the DMN and CN are crucial for cognitive control [49].Greater dominance of the DMN over the CN may influence the ability to exert cognitive control over ruminative thinking [61].Furthermore, inadequate regulation from the CN to the DMN has been associated with repetitive negative thinking about self-related topics during rumination [62,63].Therefore, state rumination may be characterized by a specific pattern of dynamically changing interactions between large-scale brain networks, particularly the DMN and CN [64,65].
Notably, FC associated with trait rumination was stronger and broader than state rumination.Based on former research, state rumination is a relatively narrow process and construct, whereas trait rumination is a broadly defined concept [3,40].Rumination traits and rumination states may be related to different cognitive processes and FC patterns [60,66].Thus, not all FC features that are involved in individual rumination tendency are also associated with rumination state.More importantly, we found that the bilateral SPG and precuneus serve as connectivity hubs in FC networks related to trait rumination, whereas the left IPL acts as a connectivity hub in FC networks related to state rumination.The precuneus is known to be involved in self-referential processing, visuo-spatial imagery and autobiographical memory [67].The connectivity strength of the precuneus has been negatively associated with trait rumination in depressed patients [68].The SPG, which is responsible for attentional control, plays an important role in trait rumination, as individuals with a high tendency to ruminate often show difficulties in disengaging attention from negative self-referential information [46].The IPL, as an important node of the CN, contributes to cognitive control [48,49].Rumination states are associated with deficits in cognitive control [62,63].These findings suggest that trait and state rumination may involve different functional network architectures.Our results also showed that the predictive features of trait rumination can be robustly generalized to predict state rumination, and vice versa.This finding suggests that trait and state rumination may share common connectivity fingerprints, consistent with the notion that the two psychological constructs are interrelated.Overall, the current results suggest a common, yet distinct neural network underlying trait and state rumination.
Several limitations should be noted.The first is that our findings were derived from a cohort of young, healthy adults and cannot be generalized to a clinically depressed sample.According to previous research, patients diagnosed with major depression experience more serious or chronic distress than healthy individuals, suggesting a higher level of maladaptive rumination in depressed patients [69].Moreover, the network bases of rumination have been found to show subtle differences between depressed patients and healthy controls [17].To gain a full understanding of the pathological role of rumination in major depression, future research should be conducted using clinical samples.The second concerns the assessment of state rumination.Because the induction of rumination (e.g., by recall of autobiographical information) may induce artificial or confounding neural activation unrelated to rumination per se, this study relied on spontaneous rumination during the resting state fNIRS measurement.However, state rumination was assessed using a visual analogue scale, which differed from the measurement of trait rumination.To improve our understanding of the common and unique neural substrates underlying rumination, FC features associated with state rumination can be further examined using a psychometrically validated questionnaire (e.g., the Brief State Rumination Inventory) [70].Finally, the sample size of this study is relatively small, which may lead to model overfitting, inflated prediction estimates, and poor generalization performance.The current findings need to be further validated with a larger sample size in the future.

Conclusion
In summary, this study has revealed distinct and overlapping neural correlates of trait and state rumination.Trait rumination is associated with FC within the DMN and DAN, as well as FC between the DMN, CN, DAN, and SN.On the other hand, state rumination is primarily linked with FC between the DMN and CN.A significant aspect of these findings is the generalizability of the predictive features: the neural markers identified for trait rumination can also robustly predict state rumination, and vice versa.This cross-predictability underscores the presence of shared neural mechanisms between the two forms of rumination, in addition to their unique characteristics.By delineating the common and distinct neural fingerprints of trait and state rumination, this research opens up possibilities for identifying biomarkers that could be pivotal in designing clinical trials, potentially leading to more targeted and effective therapeutic interventions for depression.

Declaration of Competing Interest
All authors declare no conflict of interest.

Table 1
Demographic characteristics of the study sample.