Rumination related activity in brain networks mediating attentional switching in euthymic bipolar patients

Participants and clinical data

Depression and mania levels were assessed by a trained clinical psychologist (PC, ALK) before the scanning session using the MADRS (Montgomery and Asberg 1979) and YMRS (Young et al. 1978), respectively. Self-report questionnaires were filled in by all participants, covering anxiety (STAI) and tendency to ruminate (Ruminative Response Scale—RRS). We used both versions of the STAI for state and trait anxiety (Bergeron et al. 1976) and a short 10-item version of the RRS (Treynor et al. 2003) comprising only the reflection and brooding subscales.

Introspection task

Each trial started with a 2 s instruction screen, announcing the type of task (internal or external), followed by a 4 s stimulus screen (single word) during which subjects were requested to indicate their response on a 3-point scale (Fig. 1a). Internal trials explicitly asked participants to evaluate how much they were currently feeling the state indicated by the word (an adjective) on a scale with 3 options: low (≤ 3), medium (Joormann et al. 2011; Drevets et al. 2008; Maalouf et al. 2010), high (≥ 7). External trials instead asked participants to indicate the number of letters present in the word on the same scale (< 3, 4–6, 7). The same words and the same three-level scale were used in both tasks, allowing us to equate visual stimulation and cognitive demands on response selection. There were 28 positive and 28 negative items in total, taken from the Profile of Mood State Questionnaire—POMS (McNair et al. 1971), plus 8 clinically relevant items (e.g., “stressed”, “ruminative”). To obtain counterbalanced stimuli for negative and positive conditions, some of the items of the POMS had to be reverted. Each item was presented once in both the internal and external conditions, resulting in 112 trials given in a single session (13 min). Orthogonally to this, words were presented in either the same task (repetition) or different tasks (switch) across successive trials. Valence was equally distributed among switch and repetition trials, and words in switch and repetition trials were randomized and counterbalanced across participants. Mean word length was similar between negative and positive terms. A fixation-cross was shown (jittered duration from 500 to 1500 ms) between successive trials. The task was implemented using E-Prime 2.0 software (Psychology Software Tools Inc., USA), and reaction times and responses were recorded using an MRI-compatible button box (HH—1 × 4—CR, Current Designs Inc., USA). Visual stimuli were displayed using an LCD projector (CP-SX1350, Hitachi, Japan) and projected on a screen at the rear of the scanner, which the participants could comfortably see through a mirror.

fMRI acquisition

Functional brain images were acquired with a 3T Magnetom TIM Trio scanner (Siemens, Germany) and a 32-channel head coil using a standard echo-planar imaging sequence [36 transverse slices with 20% gap, 64 × 64 base resolution, voxel size: 3.2 mm × 3.2 mm × 3.2 mm, repetition time (TR): 2100 ms, echo time (TE): 30 ms, flip angle (FA): 80°, field of view (FOV): 192 mm]. Image quality was inspected for each participant to ensure the absence of signal drop out in ventral prefrontal regions. Anatomical images were also acquired for precise localization and normalization to standard templates, using a T1-weighted 3D sequence (TR/TI/TE: 1900/900/2.32 ms, flip angle = 9°, field of view = 230 mm, PAT factor = 2, voxel dimensions: 1 mm, isotropic 256 × 256 × 192 voxel). One run of the behavioral task was acquired with 380 scans.

fMRI data analysis

In both models, movement corrections (realignment parameters) were incorporated as covariates (six nuisance regressors). Contrast images were generated for each condition of interest in each participant, and then entered in a second-level (group) analysis using a flexible ANOVA model and random-effects statistics (Penny and Holmes 2004), with depression scores (MADRS) incorporated as a covariate. Main effects are shown at a threshold corrected with the family-wise error fwe p = 0.05, whereas interaction effects are shown at an uncorrected threshold p = 0.001 with a number of voxel > 5 and T score > 3.1.

Multiple regression

To determine any parametric correlation of brain activity with individual tendency to ruminate, each condition (contrast image) generated for each participant in the first level of analysis was entered in a second level (group) analysis with rumination scores (RRS) as a covariate.

Psychophysiological interactions (PPI)

To examine task-related modulations of functional connectivity between brain regions according to attention focus and switching conditions, we conducted a PPI analysis using seed regions identified in the group level analysis (BD > HC main effect). The time-courses of selected regions were extracted from the individual (subject) level using the eigenvariate function in SPM. Volumes of interest (VOIs) were then defined as a sphere of 6 mm radius and PPIs computed for each condition. Two analyses were conducted: one using a VOI centred on the subgenual ACC (x = 3, y = 29, z = − 11) and one centred on the pregenual ACC (x = − 6, y = 44, z = 1), as it has been postulated that these regions may be involved in pathological self-referential processing but differentially so (Marusak et al. 2016). A first level analysis (GLM) was conducted using PPIs with psychological estimates for each task condition. Contrasts between conditions were then incorporated in a second level analysis in a flexible factorial statistical design and significant clusters were defined at p = 0.001.

Mediation analysis

Based on the fMRI results in BD patients, parameters of activity (beta values) from pgACC and entorhinal cortex as well as rumination (RRS) scores were introduced in a mediation analysis. Beta values were extracted from pgACC (x = − 6, y = 44, z = 1) during negative internal repetition trials and from entorhinal cortex (x = − 24, y = − 1, z = − 29) during switch from internal to external as compared to external repetition trials (contrast int.ext > ext.ext), as directed by our hypothesis and our fMRI results. Direct, indirect, and total effects of pgACC (mediator) and entorhinal cortex activities (independent variable) on rumination scores (dependent variable) were concurrently estimated with an implemented script for SPSS software [PROCESS, (Hayes 2017)]. A bootstrap test for the indirect effect (5000 samples, confidence intervals set at 95%) was performed according to Hayes (2017) recommendations.

Demographic data

As shown in Table 1, there was no statistical difference between BD patients and healthy controls for age, gender, educational level, and mania scores. Handedness laterality was also matched, with 2 left-handed in the BD group and 3 in the HC group. However, euthymic BD patients scored significantly higher on depression (MADRS, p < 0.05) and anxiety measures (STAI-S, p < 0.01; STAI-T, p < 0.001), and they showed higher tendency to ruminate (RRS, p < 0.05).

Behavioral data

Response times in the two word-judgment tasks are shown in Fig. 1b. A 2 × 2 × 2 ANOVA showed a marginal group effect (p = 0.07, F = 3.4) and interaction (focus*switch*group, p = 0.055, F = 3.9). Patients were generally slower than controls. Despite the borderline interaction, and given our specific hypothesis concerning the different task conditions, we performed a further analysis of RTs examining internal and external trials separately. A two-way ANOVA (group × switch) on trials from the external focus task testing for the impact of switching from the other (internal) task or repeating it showed no group effect (p = 0.18, F = 1.8) but a significant effect of switch (p < 0.05, F = 5.38), and no interaction (p = 0.28, F = 1.18). However, planned comparisons with Bonferroni-corrected t-tests revealed a significant difference between switch and repeat trials in BD patients only (p < 0.05), whereas this difference did not reach significance in HC (p > 0.05). On the other hand, a two-way ANOVA on trials from the internal focus task only showed a marginal group difference (p = 0.05, F = 3.9), with patients being slower than healthy controls, but no other effect. Bonferroni-corrected t test comparisons revealed a significant difference of switch vs repeat in HC (p < 0.05) but not in patients, while the difference between groups was significant for repetition trials only (p < 0.05) but not switch trials (p > 0.0.05). Thus, overall, BD patients tended to have more difficulties with the external focus task when it was preceded by internal focus, unlike controls who showed an opposite effect with slower switching from external to internal focus.

fMRI data

Main effect of patients vs healthy controls

A direct group comparison across all conditions (BD > HC, p = 0.05 FWE corrected) revealed highly significant increases in the patients for medial brain areas including sgACC (T = 8.08, x = 3, y = 29, z = − 11), vmPFC (T = 9.95, x = − 6, y = 41, z = − 23, T = 8.25, x = − 12, y = 41, z = − 8), and PCC (T = 6.95, x = − 6, y = − 34, z = 46), as well as in the inferior parietal cortex (L: T = 9.10, x = − 45, y = − 61, z = 37, R: T = 8.96, x = 60, y = − 55, z = 28), and superior occipital gyrus (R: T = 13.59, x = 24, y = − 88, z = 34), as shown in Fig. 2a (in red).

To further assess these group differences as a function of task conditions, we computed a group × attention interaction over the whole brain (BD > HC *internal > external, p = 0.001 uncorrected) that revealed higher activity for patients in regions associated with self-referential processing (Fig. 2a, in yellow). These included peaks in the PCC/precuneus (T = 5.19, x = − 3, y = − 61, z = 25), left vmPFC (T = 3.56, x = − 3, y = 44, z = − 11) and dmPFC (T = 3.88, x = − 6, y = 62, z = 16), plus the left middle frontal gyrus (T = 4.95, x = − 30, y = 26, z = 49) and left inferior parietal cortex (T = 4.99, x = − 45, y = − 67, z = 40). These data accord with our hypothesis of differential engagement of self-related processes in patients.

Switching from internal to external attention

The main effect of switching attentional focus (switch > repetition trials, p = 0.05 FWE, across both groups) showed activity in posterior medial parietal areas, including PCC (T = 8.04, x = 0, y = − 28, z = 28) and precuneus (T = 7.81, x = − 6, y = − 73, z = 40) (Additional file 1: Table S1C), in line with previous research on task switching (Piguet et al. 2016; Yin et al. 2015). Switching from internal to external focus revealed increases in several limbic structures such as amygdala, hippocampus, and striatum across groups, suggesting a delayed deactivation of these areas after a self-reference state (Additional file 1: Table S1D). The reverse contrast showed no effect, i.e., no such inertia of activity in external-focus network.

To test our second hypothesis that patients may have selective difficulties disengaging from internal focus on self-related information, we then compared brain activity between groups on trials that required switching from internal to external focus (int.ext condition) as compared to repeated external focus trials (ext.ext condition; Fig. 2b, in red). Remarkably, patients (vs controls) showed significantly higher activity in the left entorhinal cortex (T = 4.16, x = − 24, y = − 1, z = − 29), an area that has been previously related to rumination tendency in healthy subjects (Piguet et al. 2014). A further analysis separated trials according to the valence of the preceding word meaning (see “Methods”), in order to compare switching from a negative internal trial vs switching from a positive internal trial in patients vs controls (interaction of group × valence on int.ext trials). Results (Fig. 2b, in yellow) confirmed higher activity in the entorhinal cortex (T = 2.75, x = − 21, y = − 4, z = − 29) in this condition. No such effects were observed when switching from an external to internal attentional focus (i.e., ext.int trials) or when repeating the internal focus condition (i.e., int.int trials).

Rumination-related activity in patients

Next, we performed a whole brain multiple regression analysis using the RRS scores from each individual as a regressor in order to identify areas whose activity increased as a function of ruminative tendencies. Significant clusters (p = 0.001 unc., cluster size > 5) were found during repetition trials specifically for the internal focus condition with negative valence for the patients only. Higher rumination trait was associated with higher activity in this condition in the sgACC (x = 3, y = 29, z = − 11), vmPFC (x = 0, y = 35, z = − 20), pgACC (x = − 6, y = 44, z = 1), bilateral anterior insula (L: x = − 42, y = 17, z = − 2, R: x = 33, y = 17, z = 1), middle/posterior cingulate (x = 3, y = − 25, z = 34), and angular gyrus in parietal cortex (x = − 51, y = − 67, z = 40), as shown in Fig. 3. No significant correlation was found during the external focus repetition or the internal switch condition.

An additional correlation analysis of activity (betas) in clusters that were found more active in BD patients vs HC (i.e. Fig. 2a) was also performed (see Additional file 1: Table S2) and confirmed the specificity of this correlation during negative internal repetition trials. No correlation was observed in any condition in healthy controls. Trait rumination thus appears related to activity in brain areas classically implicated in self-referential processing and self-monitoring, in the condition with the most loading on internal focus.

Task-specific modulation of functional connectivity of ACC

Finally, we compared functional connectivity of sgACC (x = 3, y = 29, z = − 11) during the internal vs external focus condition in patients using a PPI analysis, since this region is among those hypothesized as central in rumination processes (Hamilton et al. 2015) and depressive state (Mayberg et al. 2005) and was found overactive in our patients as compared to controls. This analysis revealed significantly increased connectivity (p = 0.05 FWE) of sgACC with pgACC (x = − 3, y = 38, z = − 8), medial superior frontal gyrus (x = 0, y = 32, z = 43), and bilateral anterior insula (left: x = − 36, y = 8, z = − 11, right: x = 39, y = 17, z = − 11) (Fig. 4 in blue). A similar analysis also compared connectivity of pgACC when switching away from internal to external focus (vs repetition of external trials) and again found significant increases in patients (p = 0.001 unc, cluster size > 15) centered on the entorhinal cortex (x = − 27, y = − 1, z = − 26) (Fig. 4 in red). Using the same two seeds, no other significant connectivity pattern was found in patients or controls in other conditions. These connectivity results demonstrate increased engagement of self-processing networks in patients, specifically during internal focus, and lingering effects of brain activity related to internal focus when switching to the external focus task.

Entorhinal cortex activity predicts rumination scores via pgACC activity

A mediation analysis was next conducted, to assess the mediation effect via the pgACC activity of the ability to switch from an internal focus (entorhinal cortex hyperactivity) on rumination trait (RRS scores). In this analysis we defined as “path a” the link between entorhinal cortex and pgACC, as “path b” the link between pgACC and rumination and as “path c” the link between entorhinal cortex and rumination. Mediation analysis from the bootstrap analysis showed a significant indirect effect (M = 3.5, SE = 1.4), with a 95% bias corrected confidence interval excluding zero (1.04, 6.69) and a mediation effect at 69.4%, indicating that the association between the entorhinal cortex activity and rumination passes through pgACC activity. The direct effect and the other coefficient paths of the model were also significant (see Fig. 5 for paths coefficients and p values).

Internally focused attention and rumination-related activity

Behaviorally, BD patients showed an overall tendency to respond slower than controls, but particularly during internal trials. They were also slower in switching from an internal to an external focus as compared to repeating an external trial, and slower when repeating the internal focus condition. These data together suggest a difficulty to disengage from the internal focus or a transient decline in attention due to cognitive or affective overload during self-referential processing. These findings accord with psychopathological models suggesting that the tendency to ruminate reflects or results in executive control deficits in BD (Ghaznavi and Deckersbach 2012) and that a dysfunctional cognitive style with exaggerated internal-focus may impair transitions to a task-positive network (external focus) in remitted depressed patients (Marchetti et al. 2012).

In line with behavioral performance, brain activity during internal focus showed higher recruitment of midline cortical areas in patients, particularly in the ventro-medial PFC and PCC/precuneus, as well as in the parietal angular gyrus. These cortical midline structures overlap with the DMN and have been extensively related to self-referential functions in healthy humans (Northoff et al. 2006), including autobiographical memory (Summerfield et al. 2009) and mind-wandering (Mason et al. 2007). In a recent fMRI study directly comparing self-reference to resting state, Davey and colleagues showed that these areas (mPFC, PCC/precuneus, and left IPL) specifically activate during self-referential mental activity (Davey et al. 2016) and not only during resting state, with self-related content being possibly generated by PCC/precuneus activity and regulated by mPFC. However, self-focused attention during meditation (mindfulness) was recently shown to decrease activity in PCC and medial PFC as compared to mind-wandering in healthy subjects, supporting the hypothesis that exaggerated DMN-like activity may occur during unconstrained trains of thought (Scheibner et al. 2017). Therefore, the respective roles of mPFC and PCC in self-directed thinking (e.g., internally-focused tasks) versus unconstrained mind-wandering or resting state are still debated. Here we show that activity in this network is related to both an exaggerated internal focus and a tendency to ruminate in patients, not to unconstrained thought processes.

Indeed, activity in both anterior (ACC and mPFC) and posterior (PCC) midline regions was correlated with individual rumination scores, specifically during the repetition of negative internal trials. Both sgACC and mPFC have been previously associated to rumination in depressed (Cooney et al. 2010; Johnson et al. 2009) and healthy subjects (Kross et al. 2009). Taken together, these data suggest that ruminative thinking in mood disorder patients might reflect exaggerated and prolonged activity in these areas, triggered by negative self-reference.

In addition, we found that the tendency to ruminate also correlated with activity in bilateral anterior insula (AI). Although less frequently emphasized, a few previous studies suggested a role for AI in self-reflection (Herwig et al. 2012; Modinos et al. 2009), as well as in the integration of self-related information during decisions about mental effort investment (Otto et al. 2014), and in autobiographical self-relevant memories (Araujo et al. 2015). In a recent study investigating episodic counterfactual thoughts, a process similar to rumination where people imagine alternative ways in which past events could have happened, De Brigard et al. (2017) found increased insula activity, together with ACC, medial PFC, and inferior parietal cortex (De Brigard et al. 2017), consistent with our results. Insula has also been associated with heightened interoception in mood disorders (Paulus and Stein 2010). An increased focus on negative interoceptive information might therefore also promote automatic self-directed thoughts focused on past events, and/or contribute to negative affect and anxiety (Knutson et al. 2014) that are associated with self-reference and rumination in patients.

In summary, activity in self-referential brain networks is not only higher in patients as compared to controls, but further directly correlates with tendency to ruminate during negative internal focus, pointing to a plausible neural substrate of self-directed ruminative thoughts. Furthermore, rumination-related activity in regions associated with saliency and interoception, such as the subgenual ACC and bilateral AI, could possibly reflect the implication of these processes in triggering rumination and other affective symptoms of mood disorders, such as anxiety. We further tested how these regions interact as a network.

ACC connectivity in bipolar patients

Among areas found to be hyperactive and associated with rumination in our BD patients, the sgACC has been consistently implicated in the regulation of automatic emotional behavior (Ghaznavi and Deckersbach 2012; Phillips et al. 2008) and considered as a key node in the neurocircuitry of mood disorders (Drevets et al. 2008; Price and Drevets 2010). Activity in sgACC is increased in both MDD and BD (Piguet et al. 2016), even when corrected for structural differences (Price and Drevets 2010). However, despite these similarities, sgACC blood flow at rest has also been proposed as a specific target discriminating between unipolar and bipolar depression (Almeida et al. 2013).

In our study, activity in sgACC was globally higher in euthymic BD patients compared to healthy controls, correlated with rumination traits during the repetition of internally focused judgments, and showed enhanced connectivity with anterior brain structures (pgACC, mPFC, and insula) during internal focus. Increased functional connectivity of the sgACC with amygdala and PCC has been previously shown by our group in BD patients (relative to healthy controls) during resting state, regardless of their current mood (Rey et al. 2016). Moreover, sgACC-PCC and sgACC-amygdala, as well as sgACC-DMN connectivity has been linked to rumination in other non-euthymic populations (Hamilton et al. 2015; Rey et al. 2016). According to Hamilton et al. (2015), this increased connectivity may represent an integration of self-referential processes (implicating the DMN) with affective information (mediated by the sgACC), forming a circuit substrate for rumination in MDD. Our new data are partially consistent with this hypothesis and therefore point to shared mechanisms for both MDD and BD. Accordingly, as already noted, bipolar patients tend to ruminate equally to MDD patients (Ghaznavi and Deckersbach 2012). Here, when assessing task-related connectivity of the sgACC in BD patients, significant correlations of activity occurred with pgACC, bilateral anterior insula, and dorsomedial PFC during internal focus as compared to external focus, delineating an interconnected network for emotional self-reflection in BD that appears similar to the one expected in unipolar patients. Furthermore, activity in ACC and AI correlated both with the tendency to ruminate and between them, suggesting increased activity in the salience network (Menon and Uddin 2010; Medford and Critchley 2010), as also shown in response to negative-valenced stimuli in MDD (Hamilton et al. 2012). Taken together, our data therefore suggest that sgACC is not only hyperactive in BD patients but also more strongly coupled with pgACC, dorsomedial PFC, and insula during internal focus, with this activity pattern being correlated with the tendency to ruminate when internally focused judgments are repeated and negatively valenced. We might then conclude that such an activity pattern may constitute a dimension shared across mood disorders, in line with a recent meta-analytic study (Marusak et al. 2016) suggesting that both the sgACC and pgACC provide transdiagnostic neural markers reflecting common neurobiological substrates involved in developmental risk (sgACC) or in the broad expression of emotional psychopathology (pgACC) across disease boundaries.

Switching from internal states—pregenual ACC and entorhinal cortex

While both the sgACC and pgACC were more activated in patients and correlated with rumination traits, we found differential task-related connectivity for these two areas. Unlike sgACC (see above), the pgACC exhibited greater coupling with the left entorhinal cortex when switching from an internal to an external focus in BD patients. Interestingly, the same area was found hyperactive in BD patients relative to HC when comparing the two groups on switching trials. These findings support our hypothesis of higher limbic activity during a switch from negative self-reference states. Moreover, as the entorhinal is critically implicated in memory formation and retrieval (Piguet et al. 2014), our data point to an important role for enhanced communication of ventral parts of ACC with brain systems mediating autobiographical memory during internally focused attention.

Remarkably, the activity in the entorhinal cortex was previously found by our group to correlate with rumination scores in healthy subjects during both resting state and repetition of an easy visual attention task (Piguet et al. 2014), and in another group of bipolar patients (and their healthy controls) during a more difficult double switching task (unpublished results). In the current study, the entorhinal cortex activity during switching from internal to external attention as compared to external repetition trials was also found to significantly correlate with RRS scores in patients and to distinctively couple with pgACC, adding further support to the role of this region and its functional connectivity in relation with the hyperactive self-referential network in BD. As discussed above, activity in pgACC itself was both increased and correlated with rumination in our patients, underscoring a functional link of the entorhinal cortex with the ruminative circuitry. Our mediation analysis provides further evidence in support to this network, suggesting that entorhinal cortex hyperactivity during switch predicts tendency to ruminate only when mediated by hyperactivity of the pgACC. In accordance with our results, increased connectivity between the entorhinal cortex and sgACC has also been reported in bipolar patients during emotion labeling (Almeida et al. 2009). Moreover, entorhinal cortex together with ACC are parts of circuits implicated in automatic emotion regulation (Phillips et al. 2008; Rive et al. 2013) and were found hyperactive in depressed individuals during rumination (Cooney et al. 2010) and emotion regulation (Rive et al. 2013). Our results therefore suggest again a similar involvement in relation to negative self-reference in BD and MDD. The lack of deactivation of this region during switch from internally to externally attention might represent an excessive allocation of attentional resources to self-directed thoughts, at the expense of cognitive processes (Joormann et al. 2011). Finally, this allows refining the network implicated in rumination, often simplified with ROI analyses restricted to mPFC and PCC in current models (Williams 2016).

General conclusions and limitations

Although our novel paradigm provides important insights into the neural circuitry underlying focus of attention and activity in self-referential brain networks in euthymic bipolar patients, the current study is not without some limitations. First, all patients had medications, of different types and doses, the impact of which cannot be directly addressed here. Second, the clinical heterogeneity of this group (consisting of both bipolar type I and II), together with its modest size, may limit the generalization of our findings. Finally, we did not include another clinical control group, such as unipolar patients, to assess the specificity of brain activation patterns and their relation to different symptom dimensions of mood disorders. Based on our results, it would be necessary to assess if the mechanisms in action pertain to a dimension common to unipolar and bipolar type II disorders, for example, or are found throughout affective disorders in general.

In conclusion, we delineate a hyperactive functionally interconnected network of medial and anterior brain regions encompassing sgACC and pgACC, mPFC, insula, as well as PCC, whose activity is modulated by self-referential processing demands and underpins the tendency to ruminate in BD patients. This network is further connected to other cortico-limbic structures implicated in autobiographical memory, such as the entorhinal cortex, that shows lingering activity in BD when switching from an internal to an external focus. This persistent activation might not only account for heighted neural activity of the DMN at rest, but also contribute to cognitive difficulties in controlling self-referential thoughts and underlie the pronounced ruminative state often observed in mood disorder patients. These data speak in favor of excessive self-referential processing and rumination as central cognitive and neural features in both bipolar and unipolar patients, arguing for specific rumination-focused therapeutic interventions in mood disorders (Topper et al. 2017).