Abstract
Substantial individual differences exist in how acute stress affects large-scale neurocognitive networks, including Salience (SN), Default Mode (DMN) and Central Executive Networks (CEN). These network-level changes upon acute stress may predict vulnerability to long-term stress effects, which can only be tested in prospective longitudinal studies. Using a longitudinal design, we investigated whether the magnitude of acute-stress induced functional connectivity changes (delta-FC) predicts the development of posttraumatic stress disorder (PTSD) symptoms in a relatively resilient group of young police recruits that are known to be at high risk for trauma-exposure.
Using resting-state fMRI, we measured acute-stress induced delta-FC in 190 police recruits before (baseline) and after trauma exposure during repeated emergency aid services (16-month follow-up). Delta-FC was then linked to the changes in perceived stress levels (PSS) and post-traumatic stress symptoms (PCL and CAPS).
Weakened connectivity between the SN and DMN core regions upon acute stress induction at baseline predicted longitudinal increases in perceived stress level but not of post-traumatic stress symptoms, whereas increased coupling between the overall SN and anterior cerebellum was observed in participants with higher clinician-rated PTSD symptoms, particularly intrusion levels. All effects remained significant when controlling for trauma exposure-levels and cortisol stress-reactivity. Except these neural effects, neither hormonal nor subjective measures were relevant.
The reconfiguration of large-scale neural networks upon acute stress induction is relevant for assessing and detecting risk and resilience factors for PTSD. This study highlights the SN connectivity-changes as a potential marker for trauma-related symptom-development, which is sensitive even in a relatively resilient sample.
Introduction
Acute stress has been shown to induce a re-allocation of resources from the three large-scale neural networks: the salience (SN), default mode (DMN) and central executive network (CEN) 1,2. This neural network reconfiguration has been speculated to prioritize resources to facilitate processing of challenging situations in health1–3. However, frequent and chronic exposure to stressors likely leads to unfavorable consequences, such as stress-related disorders that have been associated with altered connectivity of SN and DMN4–6. To date it remains unclear whether the magnitude of brain network reorganization upon short-term challenges is predictive of one’s sensitivity to the negative consequences of long-term trauma exposure. Here, using a well-powered prospective longitudinal design, we tested this question in 190 Dutch police recruits who experienced a variety of potentially traumatic events in the line of duty.
So far, the vast majority of studies about the neural mechanisms underlying stress-related psychopathology has used cross-sectional designs. Abnormal hyper-connectivity between the core regions of the SN (e.g., amygdala-insula and amygdala-dACC), and hypo-connectivity between the core regions of the DMN (e.g., vmPFC-PCC), as well as between core regions of these two networks (e.g., amygdala-vmPFC) were observed in PTSD patients in contrast to healthy controls7–10. Due to the nature of cross-sectional designs with post-trauma assessments, however, these studies cannot tell whether the observed abnormalities existed prior to trauma exposure or were acquired along symptom development. The scarce pioneering longitudinal studies with small sample sizes (e.g., N between 15 to 50)11–13 mostly pointed to a hyperactive amygdala as predisposing factor for PTSD. These studies, however, were likely statistically underpowered according to recent guidelines14,15 and therefore require replications with large sample sizes. Besides, although these studies provided hints for a linkage between the altered functions of core regions of the SN DMN and stress-related disorders, only very few investigations directly deployed network-level approaches. Our recent study examining the network connectivity changes in response to acute stress induction, found that connectivity changes of the SN (including dorsal anterior cingulate cortex, anterior insula and amygdala) and DMN (including posterior cingulate cortex, precuneus and ventromedial prefrontal cortex) were correlated with cortisol increases after stress induction, respectively2. Given the implication of cortisol in inhibiting sympathetic stress responses and regaining physiological homeostasis following acute stressors16,17, results from our previous study suggested an adaptive reorganization of brain functional networks in the face of challenges. With these findings, as well as the evidence showing the involvement of the SN, DMN and CEN in stress-related symptoms6,10, we asked whether these short-term adaptive responses at the neural network level could predict long-term consequences after trauma exposure. Yet, stress-induced dynamical changes within and between these large-scale networks have not prospectively been tested in relation to long-term consequences after trauma exposure.
Using a longitudinal design, the current study aimed to elucidate whether the magnitude of acute-stress induced network reconfiguration could predict stress-related symptom development after trauma exposure. Specifically, we used acute stress-induced connectivity changes (i.e., delta-FC) of the SN, DMN, and CEN at baseline (i.e., Wave1 assessment) to predict the perceived stress and PTSD symptom levels after continuous exposure to police-operation related trauma in the training period of recruits. We further investigated acquired abnormalities in network responses to stress induction after trauma exposure, using neuroimaging data collected at the follow-up assessment (i.e., on average 16 months after the baseline assessment)
Based on the literature, we expected to observe predictive effects of acute stress-induced delta-FC in SN and DMN on trauma symptom development. More specifically, the adaptive responses to acute stress that were indicated by increasing SN and decreasing DMN connectivity at the baseline assessment2 were expected to be associated with lower levels of stress and PTSD symptoms after trauma exposure. For acquired effects, individuals with higher post-trauma stress levels were expected to show intensified SN connectivity and reduced DMN connectivity after stress induction as suggested in the PTSD literature7–10. Although CEN connectivity changes after acute stress induction were not associated with individual cortisol stress responses in our previous work2, decreased CEN connectivity after acute stress induction was observed under task conditions1 and CEN connectivity in general has been implicated in stress-related processing and psychopathology4,18–20. Hence, we also explored whether acute stress-induced delta-FC of CEN could predict long-term post-trauma stress levels.
Materials and methods
Participants
An initial sample of 321 police recruits from Dutch Police Academy participated in the current study in accordance with the principles of the Declaration of Helsinki and with the approval from the Independent Review Board Nijmegen (IRBN), the Netherlands. All participants gave their written informed consent before the study upon their first lab visit (Wave1). Exclusion criteria included any current psychiatric or neurological disorder, history of, or current endocrine or neurological treatment, current use of psychotropic medication, and current drug or alcohol abuse (full details in protocol article21). A number of participants (N=86) reported that their core trauma, the trauma most central to their symptoms, had occurred already before our baseline assessment. As these participants did not fit our prospective study design, data from them were excluded. Further exclusion included data from the participants who exhibited excessive motion artefacts (i.e., top 5% of participants with largest relative displacement from each fMRI session)2, resulting in a final sample of 190 participants (mean age=23.88) that had complete data from all measurements (see Figure 1 for detailed sample selection and Table S1 for demographic information).
Procedure
The baseline assessment (i.e., Wave1) took place in parallel with the early police curriculum of mostly in-class theoretical trainings. In this assessment, participants filled out questionnaires in the morning that investigate their baseline level of perceived stress (using Perceived Stress Scale, PSS) and stress-related symptoms (using PTSD Checklist for DSM-5, PCL-5). After a series of other experiments21, the acute stress induction was conducted in the late afternoon (i.e., between 4-7pm) to ensure stable salivary cortisol levels, which consisted of a SECPT (Socially Evaluated Cold Pressure Task) and MA (mental arithmetic) task (see details in the Supplemental Materials and Methods). Hormonal and subjective stress responses were assessed multiple times throughout the experiment (i.e., approximately at -10, 0, +10, +20 and +30 minutes with respect to the experiment onset), and two sessions of resting-state fMRI (rs-fMRI) data were acquired immediately before and after stress induction to assess stress-induced functional connectivity changes (Figure 2). After an average of 16 months (SD=1.9), participants were tested for the follow-up assessment (i.e., Wave2), when identical measurements were repeated to investigate consequences of exposure to trauma-like events. In the Wave2 assessment, the Police Life Event Scale (PLES)22 was additionally used to index the amount of trauma exposure. Participants also participated in a telephone interview comprising the Clinician-Administered PTSD Scale for DSM-5 (CAPS; see full details about all measurements in the protocol article21).
Data acquisition and analysis
Imaging data acquisition
Acquisition of the resting-state fMRI data occurred before and after the stress induction experiment in both Wave1 and Wave2 assessments, where participants were instructed to lie still and watch a small white cross at the screen center. Imaging data were acquired using a 3T Siemens Magnetom Prismafit MRI scanner (Erlangen, Germany) with a 32-channel head coil. A multi-band T2*-weighted EPI sequence with acceleration factor 8 (MB8), optimized from the recommended imaging protocols for the Human Connectome Project, was used to acquire a total number of 500 volumes of BOLD-fMRI images (TR=735ms, TE=39ms, flip angle=52°, voxel size=2.4×2.4×2.4mm3, slice gap=0mm, FOV=210mm). High-resolution structural images (1×1×1mm3) were also acquired, using a T1-weighted MP-RAGE sequence (TR=2300ms, TE=3.03ms, flip angle=8°, FOV=256×256×192mm3).
Assessment of stress-related measures
Perceived stress level and PTSD symptom levels were measured at both the baseline (Wave1) and follow-up (Wave2), prior to the implementation of the acute stress induction experiment (see Procedure). Additionally, CAPS interviews were conducted at Wave2 assessment. While sum scores of CAPs were used to indicate posttraumatic symptom levels, the change scores of PSS (i.e., Wave1 score subtracted from Wave2 score, delta-PSS) and PCL (delta-PCL) were calculated to indicate the development of posttraumatic stress levels.
Assessment of stress-induced hormonal and behavioral measures
To index acute endocrine and subjective stress responses, salivary samples and self-reported ratings of negative affect were measured throughout the stress induction experiment in Wave1 and Wave2 assessments. Following our previous practice2, increases in salivary cortisol and negative affect ratings were calculated for each participant to index the magnitude of acute stress responses. Specifically, cortisol increase was defined as the cortisol level 20 minutes after stress induction onset (i.e., at time +20 min. when responses peaked) subtracting baseline level immediately before stress induction (i.e., at time 0 min.). Negative affect increase was calculated as the difference in ratings between the baseline (time 0 min.) and 10 minutes after the onset of stress induction (time +10 min.; Figure 2).
fMRI preprocessing and analysis
Preprocessing
Preprocessing of rs-fMRI data included motion correction, 5mm spatial smoothing, ICA-AROMA based denoising23, and high-pass filtering with a cut-off of 100 seconds. Mean signal intensity of white matter and cerebrospinal fluid, as well as head motion parameters were regressed out to minimize psychophysiological confounds and motion artefacts24,25. The resulting residual images were subsequently registered to the MNI atlas and used for statistical analyses. Detailed preprocessing can be found in the Supplemental Materials and Methods.
Statistical analyses
Identifying delta-FC of RSNs
Resting-state networks (RSNs) of interest (i.e., SN, DMN and CEN) were identified from the components of a group-level independent component analysis (ICA) that showed the highest spatial correlation with pre-selected functional ROIs (i.e., anterior SN, left CEN, right CEN and ventral DMN) from the Stanford FIND atlas2. Changes in acute stress-induced functional connectivity (i.e., delta-FC) of these RSNs were defined as the connectivity difference between scans pre- and post-stress induction in each of three RSNs (i.e., SN, DMN and CEN).
Coefficients of delta-FC that indicated acute stress effects on RSNs at both local (i.e., delta-FC within each RSN) and more global levels (i.e., delta-FC of RSNs with the brain regions also outside the pre-defined network) at baseline (i.e., Wave1 assessment) were used to predict stress-related symptom development. In contrast, changes of delta-FC coefficients between two waves for each individual RSN were further calculated for testing acquired effects of trauma exposure.
Testing predictive and acquired effects
RSNs of interest (i.e., derived from group-level ICA) were thresholded at Z>3 to include only the core regions of each network and then applied to individual data for extracting local-level connectivity coefficients. In contrast, delta-FC images indicating connectivity changes of dual-regression derived individual RSNs before and after stress induction were averaged and thresholded at p<0.167, which were then used to extract more global-level connectivity coefficients as the associations between the RSNs and acute stress induction involved wide-spread brain regions (also see Supplemental Materials and Methods)2. Changes in these extracted connectivity coefficients at baseline (i.e., differences in connectivity coefficients before and after stress induction in Wave1 assessment) were examined for their predictive effects on posttraumatic stress levels after trauma exposure. Additionally, as cortisol stress reactivity was previously found predictive of subsequent PTSD symptom development in a limited number of studies26,27, we tested whether this could be replicated using acute stress-induced cortisol increases at baseline to predict stress-related symptomatology in our study.
In addition to predictive effects, changes between baseline and follow-up assessments in the extracted global and local-level connectivity coefficients, as well as in cortisol stress responses were further examined as acquired effects of stress-related symptom development.
We calculated delta-PSS, delta-PCL and CAPS sum scores to indicate each individual’s posttraumatic stress levels and linked them to the aforementioned connectivity coefficients at the global and local levels for each RNS. We further explored the development of specific symptom clusters, using the sum scores of each sub-cluster in delta-PCL and CAPS measures (Table S2).
Analytical approaches
Spearman rank correlation was used for all correlation analyses to mitigate the influences from extreme values and reduce the chance of false positives. Concerning the results for our a-priori hypotheses (i.e., regarding the delta-FC of SN and DMN), FDR corrections were applied to account for the number of analyses involving three outcome measurements (i.e., delta-PCL, delta-PSS and CAPS scores). For more exploratory analyses concerning the delta-FC of CEN, FDR corrections were conducted to account for two levels of network connectivity (i.e., local and more global levels) and three outcome measurements. Follow-up tests on sub-cluster symptom scores were carried out only if predictive or acquired effects were observed. Concerning these analyses, FDR corrections were conducted to account for the number of analyses involving all four sub-cluster symptom scores, for the local and more global level connectivity, separately. In case of significant results concerning delta-PCL or delta-PSS (either the overall changing score or the sub-cluster score), semi-partial Spearman correlation was further conducted to control for the baseline PCL or PSS level from Wave1 assessment. Finally, we used a Generalized Additive Model (GAM) to explore whether the predictive effects of hypothesized neural measures remain significant when accounting for influences of stress reactivity at hormonal and behavioral levels from baseline assessment, and the impact of trauma exposure (i.e., indicated by the number of experienced trauma types). Similar analyses were conducted for acquired effects to account for changes in effects of cortisol and negative affect as a function to acute stress induction between two assessments. Unlike multiple linear regression that estimates a single parameter for each predictor, GAM finds unspecified (non-parametric) functions that relate the predicted Y (dependent variable) values to the predictor values, and thus allows non-parametric fit28,29.
All statistical analyses were conducted using R version 3.6.130, with pcor function from RVAideMemoire package31 and gam function from mgcv package32 specifically for running semi-partial correlation and GAM analyses, respectively.
Results
Acute stress responses
Successful acute stress induction was observed in both baseline (Wave1) and follow-up (Wave2) assessments. Specifically, increases in salivary cortisol and reported negative affect were observed in Wave1 following stress induction, as reflected in main effects of sampling time (Fcortisol(4, 677.36)=76.82, p<0.0001; Faffect(4,719.87)=51.50, p<0.005). Similar significant effects were observed in Wave2 (Fcortisol(4,644.18)=123.4, Faffect(4,675.06)=51.03, p’s<0.0001). In short, our experimental manipulation successfully induced acute stress, indicated by increases in negative affect and cortisol for both waves (see Figure 2).
Traumatic experiences and posttraumatic stress measures
In between our two waves of data collection, the police recruits on average experienced 6.63 different types of potentially traumatic events (SD=3.78) with a range between 0 and 17. Most frequently experienced trauma were encountering suicide (including attempt; 31.4%), severe (traffic) accidents (23.6%) and physical assault (17.8%; Figure S1).
From Wave1 to Wave2, average stress symptom levels slightly increased with no statistically significant changes at the group level in either perceived stress (PSS: t(182)=1.48, p=0.14) or overall PTSD symptom levels (PCL: t(189)=0.77, p=0.44). Nevertheless, closer inspection revealed large variance in individual symptom trajectories (Figure 3; Table S2), together with a significant increase in intrusion symptom level (t(189)=2.22, p<0.05; other symptom clusters p>0.05). At Wave2, the average clinical interview (CAPS) score of overall PTSD symptoms was 1.79 (SD=4.01; sum score range: 0-27), with three participants having developed full-blown PTSD according to DSM-5 criteria (Figure S2). These results are in line with our expectation that the police recruits are overall resilient to the potentially traumatic experiences. At the same time, a substantial proportion of recruits exhibited clinically relevant increases in symptom levels (i.e., reported at least one symptom in each cluster in Wave2 PCL). We therefore proceeded to test whether the large variation in PTSD symptom trajectories could be explained by the stress-related neural network connectivity changes.
Predictive effects of baseline acute stress responses
In line with our expectation, a decreased coupling between SN and DMN core regions (i.e., posterior cingulate cortex/precuneus) following stress induction at baseline was predictive of larger increases in perceived stress level after trauma exposure (Rs=-0.19, p=0.0094; Figure 4). This effect remained significant after FDR correction for multiple comparisons (padjusted=0.0167) and when baseline level PSS was controlled for (Rs=-0.19, p=0.0039). This predictive effect of SN-DMN coupling remained as the only significant predictor in our follow-up analysis using Generalized Additive Model (GAM) when influences of baseline cortisol and subjective affect, as well as trauma exposure amount were considered (F=7.20, p=0.008). Against our hypotheses, we observed no predictive effects for self-reported PTSD symptoms (delta-PCL) or clinician rated PTSD levels (CAPS scores), nor with respect to hypothesized DMN connectivity changes after stress induction (all uncorrected p’s>0.08).
Subsequent exploratory analyses for CEN revealed that higher clinician-rated PTSD symptoms (i.e., CAPS total score) were predicted by increased delta-FC, both at the local (i.e., delta-FC within the CEN; Rs=0.21, p=0.0031) and more global levels (i.e., delta-FC of CEN with brain regions also outside the network; Rs=0.19, p=0.0089). After FDR corrections, only the effect of local CEN delta-FC remained significant (padjusted=0.019). Follow-up tests on the sub-cluster symptoms revealed that delta-FC within CEN predicted levels of alteration in mood and cognition (Rs=0.19, p=0.0085), as well as hyper-arousal symptoms (Rs=0.25, p=0.00058). These effects also remained significant after multiple comparison correction (padjusted<0.035; Figure S3).
In comparison to these baseline neural responses to acute stress induction, we did not find cortisol reactivity, nor negative affect, predictive of stress-related symptomology development (all p’s>0.05).
Acquired effects after trauma exposure
Changes in acute stress responses from Wave1 to Wave2 at hormonal, behavioral and neural levels were linked to increases in symptomology to test for acquired abnormalities. Although we did not observe any association for cortisol and negative affect levels, nor for connectivity changes of DMN or CEN (all p’s>0.05), increased coupling between the overall SN and anterior cerebellum (i.e., increased delta-FC as a function of acute stress induction) from Wave1 to Wave2 was associated with higher PTSD symptom levels (i.e., CAPS total score; Rs=0.-18, p=0.019). Yet, this effect just missed significance when correcting for multiple comparisons (pFDR=0.057). Follow-up tests examining sub-cluster symptoms suggested that this effect might have been driven by intrusion symptom (Rs=-0.22, p=0.0038; other symptoms p>0.05), with the participants showing higher intrusion symptom levels also exhibiting larger neural coupling in response to acute stress induction (Figure 5). The effect for sub-cluster symptom also remained significant after FDR correction (pFDR=0.015), and after accounting for the influence of changes in cortisol and negative affect between two assessments (F=6.05, p=0.015). No significant effects were observed for delta-PCL, nor delta-PSS scores.
Discussion
In this prospective longitudinal study, we investigated whether acute stress-induced neural network changes could function as a marker of stress vulnerability. To this end, we tested the predictive effects of such changes on long-term stress-related symptomatology after exposure to real-life trauma in police recruits. Reduced global connectivity of the SN in response to baseline acute stressors predicted increased post-trauma stress levels 16 months later. A different pattern emerged for neural network changes between assessments that followed symptom development and thus appear acquired rather than a pre-trauma risk factor: individuals with higher levels of PTSD intrusions symptoms at follow-up showed increased coupling between the SN and anterior cerebellum after acute stress induction in the follow-up versus baseline assessment. Interestingly, both acquired and predictive neural effects were found above and beyond potential associations of hormonal and subjective stress measures. Together, these findings suggest the SN synchronicity in response to acute stress as a factor for stress vulnerability that is predictive of long-term symptom development.
To our knowledge, this is the first longitudinal study that has used a network-based approach to link stress-induced connectivity changing patterns of large-scale brain networks at baseline to the subsequent symptom development after trauma exposure. Our findings therefore provide new empirical evidence that the magnitude of large-scale network reconfiguration upon acute stress exposure is relevant for investigating resilience and risk factors for stress-related symptomatology.
In line with our predictions, acute stress-induced decreases in overall SN connectivity with brain regions (i.e., PCC and precuneus) predicted higher perceived stress level after trauma exposure. Acute stress has been shown to immediately prompt SN engagement at the potential cost of neural resources that would otherwise have been allocated to other brain circuits1,2. This stress-induced reconfiguration of brain function is hypothesized to facilitate the coping with the challenging situations at hand by reallocating neural resource towards the SN for attention direction towards evolutionary relevant stimuli and integration of top-down appraisal and bottom-up visceral and sensory information (see review by Uddin33). Insufficient SN involvement in response to acute stress therefore may signal suboptimal processing and thus result in undesirable long-term consequences, such as the observed increases in stress levels after exposure to real life trauma.
Interestingly, SN reconfiguration upon acute stress induction seemed to not only signal a risk factor for later symptom development, but also indicate acquired abnormalities that were associated with increasing symptom levels. Our finding of SN-cerebellum coupling in participants with relatively high PTSD symptoms is in line with a growing number of studies that has linked cerebellum to emotional processing and regulation, particularly to negative emotional memories34–36, as well as to pathophysiology of PTSD37–41.
Furthermore, our exploratory analyses for CEN found that acute stress-induced connectivity changes in this network prior to trauma exposure could predict the overall post-trauma symptom level, and specific negative mood/cognition and hyperarousal symptom levels. These results suggest that the reconfiguration of CEN upon acute stressors may allow the tracking of symptom development after trauma exposure, and are in line with the role of dorsal lateral prefrontal cortex (dlPFC), a core region of the CEN in emotion regulation and with the idea that the modulation of dlPFC activity and connectivity could be beneficial in alleviating PTSD symptoms42,43.
Contrary to our prediction, however, we did not observe any associations between acute stress induced delta-FC of DMN at baseline and long-term consequences of trauma exposure. However, the finding of decreased connectivity between SN and posterior DMN suggests that the regions in DMN may function distinctively in response to stressful events over time, hence the mean coefficient indicating overall cohesion of all regions within the network at baseline could not capture individual variability in longitudinal symptom development. Additionally, we did not find evidence in support of cortisol reactivity predicting PTSD symptom development, which was reported in a few recent longitudinal studies27,44. Discrepancy may arise from differences in sample characteristics (i.e., combat solders vs. police recruits), analytical approaches (i.e., subtyping vs. continuous modelling), and the timing of assessment with regard to trauma exposure (i.e., once a year for four years vs. twice with 16 months in between) between the previous and our studies. Future investigations that study cortisol stress response in relation to symptom development should consider these differences.
In an attempt to investigate the predictive factors for PTSD symptom development and to disentangle predictive from acquired effects of the neural networks, we focused on a relatively healthy and resilient sample with limited variations in stress-related psychopathology in the current study. This may arguably be considered as a disadvantage as most relevant longitudinal (and cross-sectional) studies have focused on clinical populations with maximized variances pertinent to the symptomology7,45. However, our sample here is not affected by the typical confounds that surround more severe psychopathology either (e.g., medication intake). Our dimensional approach therefore potentially allows more direct interpretations of the prediction findings and our resilient sample sheds important light on how stress resilience versus vulnerability might be instantiated in the brain. Further, an important advantage of this study was the leverage of a well-established acute stress challenge in combination with a relatively large sample size.
In conclusion, the current study used connectivity changes of large-scale neural networks in response to an acute stress challenge to predict subsequent stress-related symptoms after trauma exposure in police recruits. Whereas SN-DMN connectivity prospectively predicted the longitudinal changes in perceived stress level, increased SN-cerebellum connectivity was acquired in participants with higher PTSD symptom levels. These findings suggest that acute stress-induced SN connectivity changes may serve as a potential marker of PTSD vulnerability.
Data Availability
Derived data supporting the findings of this study are available from the corresponding author [WZ] on request.
Conflict of Interest
The authors report no biomedical financial interests or potential conflicts of interest.
Acknowledgements
This work was supported by a VICI grant (#453-12-001) from the Netherlands Organization for Scientific Research (NWO) and a consolidator grant from the European Research Council (ERC_CoG-2017_772337) awarded to Karin Roelofs. We are particularly grateful to Dutch Police Academy (Politieacademie) for their cooperation. We also thank our former and current colleagues Ingrid Kersten, Naomi de Valk, Geoffrey Bertou, Leonore Bovy, Iris Hulzink, Tiele Dopp, Marijolein Hartgerink, Bart Becker, Madine Zoet, Delphin van Benthem, Job de Brouwer, Lisanne Nuijen, Pepijn van Houten, Klaas van Groesen and Nienke Flipsen for their help in setting up the study, recruiting participants and acquiring data, and to Paul Gaalman for his technical assistance in fMRI data acquisition.