18 June, 2020
Unravelling confidence abnormalities and incentivization effects in
obsessive-compulsive disorder and gambling disorder: a fMRI study
Channah Osinga1,2
1 University of Amsterdam, Faculty of Science, Amsterdam, The Netherlands
2 Department of psychiatry, Universitair Medische Centra – Location AMC, Amsterdam, The Neth-erlands
Abstract
Nowadays, compulsivity is used as a transdiagnostic construct in Psychiatric research as it is strongly associated with obsessive-compulsive disorder (OCD) and gambling disorder (GD). A rela-tively new construct, closely interrelated with compulsivity, is confidence. Several studies suggest underconfidence in OCD and overconfidence in GD leading to pathological decision-making. Whether these confidence abnormalities translate to abnormal brain activity in the ventromedial prefrontal cortex (vmPFC), a confidence coding area, remain poorly understood. Also, the effect of incentives can potentially be used in transdiagnostic research, since GD is associated with Sensi-tivity to Rewards (SR) and OCD with SensiSensi-tivity to punishments (SP). However the incentivization effect on confidence in both OCD and GD is understudied. This study will also investigate neurobi-ology of SR and SP. Finally this study will use BIS/BAS values as another measure of SR and SP to investigate its effect on confidence. We used a incentivized confidence elicitation task with func-tional magnetic resonance imaging (fMRI) to investigate confidence encoding and the incentiviza-tion effect on confidence. Also BIS/BAS scales were taken as an addiincentiviza-tional measure of SR and SP. We included 28 GD patients, 29 OCD patients and 45 healthy control subjects (HC) in the behav-ioral analyses and 24 GD patients, 28 OCD patients and 44 HC in the neuroimaging analyses. Re-sults showed significant underconfidence in OCD, however no confidence abnormalities were de-tected in confidence coding brain areas. Additionally, we found a significant higher confidence in gain contexts in GD patients compared to HCs, indicating higher reward sensitivity. This was con-firmed by finding a significant relationship between BAS values and confidence solely in GD pa-tients. Nonetheless, neuroimaging results did not reveal any significant group differences in incen-tive encoding. Together, these findings suggest that confidence and incenincen-tives can thus for not be reliably used as transdiagnostic constructs in OCD and GD research.
Keywords: compulsivity, confidence, incentives, OCD, GD, BIS/BAS, Reward Sensitivity, Punish-ment Sensitivity
Contents
1. Introduction ... 3 1.1 Confidence a transdiagnostic construct ... 3
2
1.2 How to measure confidence? ... 3
1.3 Neural Basis of confidence ... 3
1.4 Punishment and Reward Sensitivity ... 4
1.5 Behavioral Inhibition/Activation system ... 4
1.6 Research questions & Hypotheses ... 5
2. Materials & Methods ... 5
2.1 Subjects ... 5
2.2 Task procedure ... 6
2.3 Behavioral analyses ... 6
Linear-mixed effects model ... 6
Linking BIS/BAS with confidence ... 7
2.4 fMRI analyses ... 7
fMRI data acquisition and preprocessing ... 7
GLM model ... 8 Contrasts ... 8 2.5 Exclusions ... 8 3. Results ... 8 3.1 Behavioral results ... 8 Confidence judgements ... 8
Incentivization effect on confidence ... 9
Correlation BIS/BAS and confidence ... 9
3.2 fMRI results ... 10
Confidence encoding ... 10
Processing potential gain ... 12
Processing potential loss ... 13
4. Discussion ... 13
4.1 Confidence abnormalities ... 13
4.2 Confidence coding in the brain ... 14
4.3 Loss and Reward Sensitivity related to confidence ... 14
4.4 Loss and Reward processing in the brain ... 15
4.5 Conclusions and future directions ... 15
References ... 17
Appendix ... 19
A) Translated BIS/BAS scales ... 19
B) BAS subgroups analyses ... 21
C) Correlation analyses ... 21
3
1. Introduction
1.1 Confidence: a transdiagnostic construct
Nowadays, psychiatric research is increas-ingly using a transdiagnostic approach in-stead of research based on classically diag-nosed disorders. The main focus of this transdiagnostic research is on understanding underlying mechanisms and constructs that drive maladjusted behaviour across psychiat-ric disorders. Compulsivity is one of the con-structs that plays a key role in understanding and linking symptom variation across various disorders. It is defined by self-defeating, re-petitive behaviours and the sense of limited control over these behaviours (Figee et al., 2016). One of the psychiatric disorders most strongly associated with compulsivity is ob-sessive-compulsive disorder (OCD), since it represents one of its key symptoms. How-ever, it is also a characteristic of other psychi-atric disorders such as gambling disorder (GD) (Maccallum et al., 2007).
In addition, a relatively new construct that has received attention within psychiatry is confidence. Confidence is defined as a subjective feeling about the validity of one’s choices, thoughts or judgements (Luttrell et al., 2013). It is a form of metacognition, which refers to the ability to think about our own thinking (Martinez, 2006). The accuracy of meta-cognitive confidence is important since it affects guiding future decisions and behav-iours. When confidence judgements are sys-tematically inaccurate, it could contribute to persistent pathological decision-making ob-served in psychiatric disorders. Several stud-ies have shown that compulsivity and confi-dence are closely interrelated. In the case of OCD, there is evidence of decreases in confi-dence in several cognitive domains such as memory and perception (Hoven et al., 2019). In addition, greater distortions in confidence are related to an increase in compulsive
checking behaviour. There is much less con-sistent evidence for confidence abnormalities in GD. However, some studies do provide ev-idence for overconfev-idence in GD while per-forming perceptual decision-making tasks (Brevers et al., 2013 & Brevers et al., 2014). Overall, these studies suggest a link between compulsivity and confidence in the form of underconfidence for OCD patients and over-confidence for GD patients.
1.2 How to measure confidence?
Confidence can be measured retrospectively by using a confidence scale, after a choice has been made (Stankov, Kleitman & Jack-son, 2015). To assess whether someone is underconfident or overconfident, calibration (or confidence bias) is calculated, which is re-flecting the difference between mean confi-dence levels and mean performance. When confidence is higher than the actual perfor-mance, we speak of overconfidence. On the contrary, underconfidence is found when con-fidence is lower than the actual performance. When the calibration score approaches zero, it reflects a relatively good calibration (Hoven et al., 2019).
1.3 Neural basis of confidence
A number of studies have shown that there is sufficient evidence of confidence abnormali-ties in OCD and GD, however, the neurobio-logical basis of these confidence abnormali-ties is unknown. From studies assessing con-fidence encoding in healthy individuals it has become increasingly apparent that the ven-tromedial prefrontal cortex (vmPFC) is not only involved in the representation of value of choices (Smith et al., 2010), but also repre-sents confidence in these choices. For exam-ple, activity in the vmPFC was uniquely corre-lated to confidence levels during the moment of choice within a perceptual decision-making
4 task (Gherman & Philiastides, 2018).
Moreo-ver, Lebreton et al., (2015) found that judge-ments and their respective confidence levels were quadratically related and automatically reflected in the activity of the vmPFC. Lastly, Shapiro & Grafton (2020) confirmed the role of the vmPFC in encoding both value and confidence during a value-based decision-making task. These results suggest that the activity in the vmPFC can be seen as a readout of confidence during the moment of choice.
There are also some others regions found to be involved in confidence encoding. One study found evidence for the anterior cingulate cortex (pgACC), which also pre-dicted the degree of variation in confidence between subjects (Bang & Fleming, 2018). In addition to these prefrontal regions, striatal regions and mesolimbic structures such as the amygdala appeared to be associated with confidence encoding (Guggenmos et al., 2016; Hebart et al., 2016). While the evi-dence for the neural basis of confievi-dence cod-ing in the vmPFC is growcod-ing, it is unclear what the neurobiological basis of confidence abnormalities in OCD and GD entails.
1.4 Punishment and Reward Sensitivity
As the neural basis of confidence and value of the choice seems to overlap, this suggests an interaction of value and confidence at a behavioral level. Indeed, there is experi-mental evidence for a biasing effect of value on confidence. Lebreton et al., (2018) used an incentivized confidence elicitation task to examine the effect of incentives (gains vs. losses) on confidence in healthy individuals. They showed that higher and lower incen-tives significantly increased and decreased confidence judgements, respectively, while performance remained unchanged. It is, how-ever, unknown whether this incentive bias is apparent in our clinical context of OCD and
GD. Studies did investigate sensitivity to re-wards (SR) and sensitivity to punishments (SP) within OCD and GD populations. Wardell et al., (2015) showed that SR rather than SP was associated with different gam-bling motives among gamblers. Also, only motives related to winning, but not negative affect motives predicted gambling severity, thereby suggesting an increased SR. Another study confirmed this result by showing solely a positive relationship between SR and gam-bling problems (Gaher et al., 2015). Accord-ing to the literature, OCD patients are more sensitive to punishments. Morein-Zamir et al., (2013) showed that, compared to controls, OCD patients failed to reduce response inhi-bition after receiving punishments, indicating failure of cognitive control in punishment con-texts. Moreover, recently it was shown that OCD patients have an increased SP com-pared to healthy controls subjects (HC) (Raf-fard et al., 2020). In light of these earlier find-ings, it is hypothesized that in GD, confidence levels might be especially inflated in gain contexts. On the other hand, loss cues may amplify underconfidence in OCD.
1.5 Behavioural Inhibition/Activation System
The most frequently used method to measure SP and SR is the BIS/BAS scale. It is devel-oped by Carver and White (1994) and stands for the Behavioral Inhibition system (BIS) and the Behavioral Activation System (BAS). Re-search has shown high BIS values are corre-lated with OCD symptoms (Berger & Anaki, 2014; Fullana et al., 2004). On the other hand, multiple studies provide evidence for a correlation between high BAS values and gambling behavior (O’Connor, Stewart & Watt, 2009; Brunborg et al., 2011; Demaree et al., 2008; Mercer & Eastwood, 2010). Kim and Lee et al., (2011) even showed, within a non-clinical sample, that a high BAS group, compared to a low BAS group, shows higher
5 confidence levels. Additionally, they showed
that the high BIS group preferred safe deci-sions after losing experiences reflecting less confidence. This study aims to investigate the relationship between BIS/BAS and confi-dence in our clinical groups
1.6 Research questions & hypotheses
The first aim of the current study is to investi-gate whether the findings of underconfidence and overconfidence in decision-making of re-spectively, OCD and GD patients, can be replicated using a confidence elicitation task. Moreover, we will investigate how these con-fidence abnormalities are reflected in brain activity using fMRI, where we expect to find abnormal confidence related activity within the vmPFC in both patient groups. Also, this study will investigate the effect of incentives on confidence. We expect a significant inter-action effect, such that confidence will espe-cially increase within gain contexts in GD pa-tients, whereas confidence will decrease spe-cifically in loss contexts in OCD.
Moreover, in the light of dysfunctional sensitivity for rewards and punishments in these disorders, we will investigate if there is evidence for abnormal brain activity during processing of potential gain in GD patients and potential loss in OCD patients. Since multiple fMRI studies (Balodis et al., 2012; Van Holst et al., 2012) showed disturbed ac-tivity in the striatum in GD patients during ward-related tasks, it is expected that this re-sult will be replicated. On the contrary a study by Admon et al., (2012) showed that another value encoding area, the amygdala, plays a key role in loss processing in OCD patients. Therefore, we hypothesized that OCD pa-tients will show abnormal activity in the amyg-dala in loss context, relative to the HC. Lastly, this study will investigate the relationship be-tween BIS/BAS and confidence in both pa-tients groups, since BIS/BAS constructs are
related to SP and SR. First, we expect higher BAS scores in GD compared to HC, whereas we hypothesize a higher BIS score in OCD patients. Moreover, we expect a positive rela-tionship between BAS scores and confidence (i.e. subjects with higher BAS scores are gen-erally more confident), a negative relationship between BIS scores and confidence (i.e. sub-jects with higher BIS scores are generally less confident).
2. Materials & Methods 2.1 Subjects
31 patients with GD (mean±SD: 38.4±11,8 years, 26 males), 29 OCD patients (34.3±8.1 years, 11 males) and 49 HCs (39.5±14.0 years, 28 males) participated in the study. Recruitment of OCD patients took place via the department of psychiatry at the Amster-dam Universitair Medische Centra (UMC), lo-cation AMC. GD patients were recruited from the Jellinek Addiction Treatment Center in Amsterdam. The HC were recruited using ad-vertisement folders. Before inclusion, all par-ticipants were pre-screened using the inter-national psychiatric interview (MINI)
(Sheehan et al., 1998). Participants had nor-mal, or corrected to normal vision and an age above 18 to be able to sign the informed con-sent. Lastly, participants with MRI contraindi-cations, and/or an IQ below 80 were ex-cluded.
2.2 Task procedure
To investigate confidence abnormalities in GD and OCD patients and the effect of incen-tives on confidence, an incentivized confi-dence elicitation task was used. An overview of the task is shown in figure 1. Each trial of the task started with a fixation screen for 750 milliseconds (ms), followed by two Gabor Patches (150 ms). After that, participants had
6 to decide which one of the patches had the
highest contrast using the left or right index finger (self-paced). After the choice, a jitter was introduced (4500-6000 ms). Subse-quently, the incentive was shown (900 ms; gain, neutral or loss). This way, we isolated the effect of incentive on confidence without affecting performance. Then, participants gave a confidence judgement about their pre-viously made perceptual decision on a confi-dence scale, ranging from 50% to 100% in steps of 5% (self-paced). The initial position of the cursor on the confidence scale varied between 65% and 85%. Finally, after the con-fidence judgement, a feedback screen of 900 ms appeared which showed whether the per-ceptual decision was correct and the amount of points. In order for the test subjects to get used to the task, practice sessions containing 10 trials were held outside and inside the fMRI scanner.
In addition, calibration sessions were performed in order to adjust the evidence of the task in such a way to reach an average performance of 70%, to ensure an individu-ally matched difficulty level. The main task consisted of two sessions with 72 trials per session. The incentives conditions each con-tained 24 trials and were randomly intermixed across trials. The main measures of the task were: confidence ratings (varying between 50% and 100%), accuracy of the choice (which could be either correct or incorrect) and the amount of evidence (calculated by taking the difference in contrast between both Gabor patches). In addition to the confidence elicitation task, BIS/BAS sensitivity was measured using the Dutch translation of Carver and White’s BIS/BAS scales (1994) (appendix A).
2.3 Behavioral analyses
Linear-mixed-effects model
All behavioral analyses were performed using R (R core team 2018). To investigate confi-dence differences between the groups and the effect of incentives on confidence, a lin-ear mixed-effects model was fitted using the lme4 function (Bates et al., 2014). The right model was selected by comparing the Akaike information criterion (AIC), the Bayesian in-formation criterion (BIC) and by assessing model fit by using chi-square tests on the log-likelihood values. In this model, confidence was the dependent variable. Fixed effects in-cluded group, incentive, accuracy and evi-dence. Also, an interaction effect between group and incentive as well as a three-way interaction between group, evidence and ac-curacy was set. Moreover, a random inter-cept for participants was incorporated in the model to include differences in baseline lev-els of confidence between participants. By in-specting the model's residuals, it became clear that the assumptions of homoscedastic-ity and normalhomoscedastic-ity were met. P-values were obtained using the lmerTest function (Kuz-netsova, Brockhoff and Christensen, 2017). Linking BIS/BAS with confidence
The average BIS values and BAS values, in-cluding the three subcategories, were calcu-lated for each participant, and were com-pared with group as a factor in an ANOVA test. After that, BIS and BIS values were cor-related with the average confidence per group, using the Pearson correlation test.
7
2.4 fMRI analyses
FMRI data acquisition and preprocessing fMRI data were acquired using a 3.0 Tesla In-terna MRI scanner (Philips Medical Systems, Best, The Netherlands). After obtaining the T1-weighted structural anatomical image, T2*- weighted functional images, sensitive to blood oxygenation level dependent signals (BOLD), were acquired using a three-echo combined interleaved EPI sequence. The im-aging parameters were specified as: TR, 2.375 seconds; TE, 9.0ms, 24.0 ms, and 43.8ms; total echo train length, 75ms; voxel size, 37 transverse slices; 3 mm slice thick-ness; 0.03 mm slice-gap.
During preprocessing, functional scans were first weighed and combined which resulted in a total of 570 volumes per session. Thereby, the scans were realigned to the first volume. Subsequently, for all sub-jects the first 30 dummy scans and non-task related scans were discarded. The exact number of non-task related scans varied for each subject, because it depended on the
to-tal time needed to finish the experiment. Mo-tion parameters were also adjusted in this stage to set the new first volume to zero mm movement. Subsequently, the functional im-ages were co-registered with T1-weighted structural images, and were normalised to the Montreal Neurological Institute (MNI) space. Afterwards spatial smoothing was applied us-ing a 6 mm Gaussian Kernel at full-width at half-maximum. Despite the adjustment of the motor parameters, some subjects showed ar-tifacts in some scans. The Art-Repair toolbox (Mazaika et al., 2009) was used to repair these motion artifacts. The toolbox maintains a threshold for the mean deviation from the mean intensity of the BOLD signal. Volumes that had a deviation from above 1.5% were repaired by interpolating from the 12 adjacent volumes. All fMRI preprocessing steps were performed using MATLAB R2018a (Math-works Inc., Sherborn, MA, USA) with SPM12 software (Wellcome Department of Cognitive Neurology, London, UK).
Figure 1: Confidence elicitation task. Participants viewed two Gabor Patches, and made a perceptual decision as to which one had the highest contrast. After the decision, an incentive screen appeared representing either the re-ward, loss or neutral condition. Participants then rated their confidence in their perceptual decision using the confi-dence rating scale. Finally, the participants received feedback on whether their perceptual choice was correct. Hereby also showing the amount of points won.
8 GLM model
A general linear model (GLM) was con-structed, using two moments of interest: the moment of choice and the incentive/rating moment. We defined an onset regressor for the choice moment, which was parametrically modulated by confidence rating. Moreover, we defined three onset regressors at incen-tive/rating moment for every incentive sepa-rately.
Contrasts
The contrasts were defined per subject, after which they were used for second-level anal-yses (group analanal-yses). First, to study confi-dence encoding and group differences, we set a contrast comparing the moment of choice, modulated by the pmod confidence, to baseline. ROI analyses were performed with a mask consisting of the vmPFC. The mask was created with the WFU Pickatlas 2.5, wherein a spherical 8mm ROI was de-fined with MNI coordinates (x,y,z): -2,52,-2, based on the previous study from Lebreton et al., (2015).
Furthermore, to compare incentive processing between groups, we set a ‘reward processing’ contrast comparing the prospect of a possible ‘gain’ with a ‘neutral’ context, and a separate ‘loss processing’ contrast comparing the prospect of a possible ‘loss’ with a ‘neutral’ context. Reward (gain>neu-tral) and loss processing (loss>neu(gain>neu-tral) were compared between HC and GD, and HC and OCD, respectively, using independent two sample t-tests. For our ROI analyses, two masks consisting of the striatum and amyg-dala were created using the WFU pickatlas 2.5 and the automatic anatomical labelling (aal) atlas.
2.5 Exclusions
Two GD patients and three HCs, with an av-erage deviation of <5% of the initial cursor on
the confidence scale or that showed an aver-age performance of <50% were removed from all further analyses. Furthermore, due to a technical bug one GD patient and one con-trol participant were removed as well. Fur-thermore, four GD patients, one OCD patient and one session of the functional imaging of one GD patient and one HC were excluded from fMRI analyses, due to considerable head movement. However, the behavioral data of these participants were included in the analyses. In addition, functional imaging data for the second session was missing for two OCD patients, and first session data was missing for one HC. To conclude, 28 GD pa-tients, 29 OCD patients and 45 HCs were in-cluded in the behavioral analyses. For the fMRI analyses, 24 GD patients, 28 OCD pa-tients and 44 HCs were included.
3. Results
3.1 Behavioral results
Confidence judgements
The linear-mixed effect model revealed signif-icant lower confidence judgements in OCD patients (Mean±SD:72.59±16.61) compared to HC (76.13±15.07) (β = -6.16 ± 1.97, t =-3.13, p=.0022). However, GD patients (79.01±16.32) did not significantly differ from HCs (β = 2.52 ± 2.0, t =1.26, p=.21).
We also found an interaction effect between evidence and accuracy, which means that confidence increases in correct trials and de-creases with errors (β = 28.23 ± 3.43, t =8.24, p<.001). In addition, this interaction ef-fect was found to be group-dependent, which was significant in both OCD (β = -22.4 ± 4.22, t =-5.31, p<.001) and GD (β = -16.46 ± 4.70, t =-3.51, p<.001) compared to HC. Figure 2 shows this three-way interaction where in OCD, evidence appears to be less integrated
9 for both correct and incorrect answers,
whereas in GD, evidence appears to be less integrated only in incorrect answers.
Incentivization effect on confidence
As expected, the linear mixed-effect model revealed a significant interaction effect be-tween group and incentive on confidence, with a significant difference between HC and GD patients (β = 1.19 ± 0.34, t =4.49, p<.001) (Figure 3). This result shows that confidence increases in GD, especially when a reward is at stake. On the other hand, no differences between HCs and OCD patients was found regarding the effect of incentive on confi-dence.
Correlation BIS/BAS and confidence With regards to BIS/BAS scores, no signifi-cant differences in BAS scores were found between GD patients (43.45±4.44), HCs (41.80±6.06) and OCD patients (42.56±3.52)
(p=.50). BIS values did significantly differ be-tween the groups as determined by one-way ANOVA (F(2,90) = 19.61, p <.001). A Tukey post hoc test revealed that the mean BIS val-ues of OCD (22.19±1.88, p<.001) and GD (20.86±3.52, p<.001) patients were signifi-cantly higher compared to the HC
(17.41±3.50). There was no significant differ-ence between OCD and GD groups (p=.35).
Furthermore, a Pearson product-mo-ment correlation test was computed to as-sess the relation between BIS/BAS values and confidence. As expected there was a sig-nificant positive correlation between BAS val-ues and confidence in GD (r=-.43, n=20, p=.04). This means that increases in BAS values (i.e. sensitivity to reward) were corre-lated with increases in confidence in GD. (Figure 4). Contrary to the expectations, no significant negative correlation was found be-tween BIS values and confidence in OCD (r=-0.24, n=25, p=.22). A summary of the correla-tion analyses can be found in Appendix B.
Figure 2: Three-way interaction between evidence, accuracy and group. Shown is the interaction effect be-tween evidence and accuracy differing across the three groups. In HC, confidence increases with correct trials and increasing evidence, while it decreases with incorrect trials and decreasing evidence. However, in both patient the model revealed a different pattern. In OCD evidence seems to be less integrated in both right and wrong answers, while in GD, evidence appears to be less integrated in wrong answers (p<.001).
10
3.2 fMRI results
Confidence encoding
First of all, modulation of the BOLD re-sponses by confidence during choice mo-ment was investigated and compared both within and between groups. Table 1 shows a complete list of cluster activation from these analyses. At choice moment, a positive rela-tionship between subjective confidence levels of HC and activity in the vmPFC was found using a whole-brain analysis. sThis means that vmPFC activity is increasing with higher confidence. Moreover, whole-brain results in-dicated a positive relationship between confi-dence and activity in the bilateral striatum and left primary motor cortex. On the other hand, negative activity, indicating higher ac-tivity with lower confidence, was found in the primary somatosensory cortex, dorsal part of the ACC (dACC), dorsal lateral part of the prefrontal cortex (dlPFC) and bilateral insula.
When contrasting confidence at choice moment for OCD, using our ROI anal-ysis of the vmPFC, significant activity was found in the vmPFC, which was, however very small (see table 1). Moreover, whole-brain results showed positive effects in the bi-lateral striatum, left primary motor cortex, precuneus, and the left visual association cortex. Negative activity was found solely in the right primary motor cortex. Regarding confidence encoding in GD, we did not find any significant positive activity. However, at whole-brain level, subjective confidence sig-nificantly negatively correlated with activity in the right dorsomedial prefrontal cortex (dmPFC), right insula, as well as the left par-ahippocampal gyrus and left cerebellum. Per-forming group comparisons, no significant group differences were revealed both at a whole-brain level as well as in our ROI anal-yses. The neuroimaging results of the confi-dence contrasts of all three groups is shown in figure 5.
Figure 3: Effect of incentive on confidence. Shown are the mean confidence levels on the y-axis and the incentives on the x-axis which gradually goes from a loss condition (-1) to a win condition (1). Confidence levels of OCD patients were significantly lower than HCs (β = -6.16 ± 1.97, t =-3.13, p=.0022). Additionally, we found a significant interaction ef-fect between group and incentive, with a significant differ-ence between GD patients and HC, indicating higher confi-dence levels in GD compared to HCs when a reward is at stake (β = 1.19 ± 0.34, t =4.49, p<.001).
Figure 4: Correlation between BAS and confidence in GD. This figure shows the mean confidence levels relative to the mean BAS values in GD patients. Pearson’s correla-tion test revealed a significant positive correlacorrela-tion between BAS values and confidence in GD (r=-.43, n=20, p=.04).
11
Table 1: comprehensive list of cluster activation for confidence contrast (+/-)
Group Effect Brain Region KE T P
(FWE-cluster corrected Peak MNI coordi-nates HC Confidence at choice mo-ment + Striatum (L) Striatum (R) 188 134 6.48 5.74 <0.001 <0.001 -27 5 -7 24 2 -10 Primary motor cortex (L) 102 4.89 0.002 -36 -19 56 vmPFC (LR) 39 4.66 0.001 0 47 -4 Confidence at choice mo-ment - Primary soma-tosensory cortex (R) 615 6.45 <0.001 36 -25 47 dACC (LR) 368 6.36 <0.001 -9 20 38 dlPFC (R) 98 6.09 0.002 36 32 29 Insula (L) Insula (R) 94 243 5.31 6.01 0.003 <0.001 -33 20 5 33 23 5 OCD Confidence at choice mo-ment + Primary Motor cortex (L) 162 6.93 <0.001 -36 -22 53 Striatum (R) Striatum (L) 184 6.73 <0.001 9 20 -4 -18 14 -7 Precuneus (R) 54 4.31 0.021 21 – 79 20 Visual Associa-tion cortex (L) 75 4.59 0.004 -9 -91 20 vmPFC (LR) with mask 1 3.51 0.042 0 44 5 Confidence at choice mo-ment - Primary motor cortex (R) 153 5.60 <0.001 51 -16 50 GD Confidence at choice mo-ment - Cerebellum (L) 96 6.86 0.001 -30 -64 -25 dmPFC (R) 544 6.24 <0.001 3 23 35
Figure 5: Neuroimaging results of positive/negative confidence contrasts at choice moment. Red coloured regions showed a signifi-cant positive relationship between brain activity and confidence. Blue coloured regions showed a signifisignifi-cant negative relationship between brain activity and confidence. A) Results of OCD patients showed positive effects in the vmPFC, striatum, primary motor cortex, left visual association cortex, whereas negative activity was found solely in the left primary motor cortex. B) in HCs, positive activation was found in the
vmPFC, striatum and left primary motor cortex, whereas negative activation was found in the right primary somatosensory cortex, dACC, dlPFC and insula. C) in GD, no significant positive activation was found, however negative effects were found in the left cerebellum, right
12
Insula (R) 134 5.56 <0.001 30 23 5
Parietal cortex (L) -> BA40
58 4.61 0.011 -54 -43 47
Processing potential gain
The ROI analysis did not reveal any signifi-cant group differences within the striatum when comparing the BOLD responses during the moment of incentive/rating for reward tri-als versus neutral tritri-als. Also when using a whole-brain analysis no significant group dif-ferences were detected. Therefore only the within-group effects will be discussed here. Table 2 shows a complete list of cluster acti-vation from these analyses. In both groups,
we found significant activity in the bilateral caudate, using the striatum ROI (Figure 6). Moreover, significant activity was found in the right precuneus, left SMA and left dlPFC in HC, whereas in GD, activity was found in the bilateral dACC, right hippocampus and right insula.
Table 2: comprehensive list of cluster activation for gain > neutral contrast in HC and GD Group Brain Region KE T P (FWE-
clus-ter corrected
Peak MNI co-ordinates
HC Caudate Nucleus
(LR) 115 119 7.15 5.59 <0.001 <0.001 12 11 -4 -12 17 -7
Precuneus (R) 10072 8.79 <0.001 30 -67 41
SMA (L) 71 5.96 0.006 -45 8 32
Figure 6: Significant striatum activity during gain trials compared to neutral trials. The ROI analysis with a mask consisting of the striatum, showed significant activity in the bilateral caudate nucleus (red coloured) in both HC and GD. However no group differences were detected. The shown activations had a threshold of 0.001 (p < 0.05, FWE corrected).
13 dlPFC (L) 49 4.79 0.030 -48 35 23 GD Caudate nucleus (LR) 25 23 6.29 5.68 0.008 0.010 -9 8 -1 9 11 -1 dACC (LR) 184 223 6.00 5.63 <0.001 <0.001 3 -28 38 0 50 11 Hippocampus (R) 153 5.47 <0.001 30 -28 -7 Insula (R) 93 6.31 0.001 36 20 -1
Processing potential loss
The ROI analysis, comparing the BOLD re-sponses in the amygdala during the moment of incentive/rating for loss trials versus neu-tral trials, did not reveal any significant group differences. Also when using a whole-brain analysis no significant group differences were detected. Therefore only the within-group ef-fects will be discussed here. Firstly, a signifi-cant effect was found in the right supplemen-tary motor area (SMA) in
both HC and OCD. Furthermore, significant activity was detected in the left lingual gyrus, right caudate nucleus and right posterior cin-gulate cortex in HC, whereas in OCD, signifi-cant activity was found in the left calcarine fissure, right angular and left supramarginal gyrus. The neuroimaging results as well as the complete list of cluster activation from these analyses can be found in appendix D.
4. Discussion
4.1 Confidence abnormalities
The first aim of this study was to replicate the findings of underconfidence and over-confidence in decision-making, of respec-tively OCD and GD, using a confidence elic-itation task. We found significant lower con-fidence in OCD patients compared to HC, which is in line with previous findings (Hoven et al., 2019). On the contrary, no significant effect of group on confidence was found in the GD group. This finding is not in line with the hypothesis of increased confidence in GD compared to HC, as well as with earlier studies reporting higher confi-dence in GD patients (Brevers et al., 2013; Brevers et al., 2014).
One possible explanation is that ear-lier studies did not correct for performance differences between participants, possibly
biasing the results. The current study cor-rected for differences in baseline levels of confidence, by using a random intercept, and for performance differences. Another explanation is that GD patients might not be overconfident in perceptual tasks as used in the current study. As shown in figure 3, the confidence levels of GD patients were not significantly higher than HCs, however when inspecting the interaction effect be-tween group and incentive, GD patients had higher confidence than HC. This suggests that GD patients seem to especially over-confident in a gambling-relevant context
4.2 Confidence coding in the brain
The second aim of this study was to investi-gate group differences in neural signature of confidence encoding, and specifically in the vmPFC. Firstly, we replicated the finding from Lebreton et al., 2015, showing signifi-cant activity in the vmPFC in HCs at the mo-ment of choice. This confirms that the
14 vmPFC plays a role in automatic confidence
encoding. Also in OCD, activity was found in the vmPFC, but the activity was only visible in a single voxel. However, at a whole-brain level the vmPFC activity was not significant. In addition, no significant activity was found in the vmPFC in GD group. Group analyses revealed no significant differences in confi-dence encoding between OCD and HC as well as GD and HC. This is not in line with the hypothesis of finding abnormal vmPFC activity related to confidence in the patient groups. Finding little or no activity in the vmPFC in either patient group could be due to the low sample size of the patient groups compared to the HCs, causing the fMRI analyses to be underpowered.
On the other hand, in both patient groups there was a significant disturbed re-lationship between evidence and accuracy on confidence (see figure 2), meaning that evidence is less integrated for both correct and incorrect choices, which might disturb translation to proper confidence levels. In the current study, however, evidence was not integrated in the neuroimaging anal-yses. To assess the effect of evidence on confidence in the brain, the evidence level can be added as a parametric modulator at the moment of choice to see whether this would result in activity in the vmPFC in the patient groups and reveal group differ-ences.
Besides the vmPFC, the activity of the bilateral striatum and left primary motor cortex were found to be positively linearly correlated to confidence in HC and OCD. Whereas significant negative effects were found in the insula in both GD and HC. These areas were also mentioned by stud-ies of Guggenmos et al., 2016 and Hebart et al., 2016, confirming that a widespread brain network is involved in confidence en-coding.
4.3 Loss and Reward Sensitivity related to confidence
The third aim of the study was to investigate group differences in the effect of incentives on confidence. Firstly, our findings demon-strate a significant interaction effect be-tween group and incentive on confidence. As expected, GD patients showed signifi-cant higher confidence when a potential gain was at stake in comparison with HC. This indicates that GD patients have a higher reward sensitivity than HC. It also confirms the findings of Wardel et al., (2015) and Gaher et al., (2015) showing a link tween reward sensitivity and gambling be-havior. On the contrary, OCD patients did not show an exaggerated decrease in confi-dence with the prospect of loss, which does not correspond to our hypothesis. Earlier studies showing loss sensitivity in OCD pa-tients used other kinds of methods, includ-ing a go/no-go task and the Sensitivity to Punishment and Sensitivity to Reward Questionnaire. Here we operationalized SP by determining the confidence under the in-fluence of negative incentives.
The fourth aim of the study was to investigate whether the BIS/BAS constructs correlated to confidence in our clinical groups. First of all, OCD patients had signifi-cantly higher BIS values compared to healthy controls. However, the BIS values in OCD did not significantly correlate with lower confidence. On the other hand BAS were not significantly different between groups. However, the correlation between BAS and confidence only appeared to be significant in the GD group, showing that as BAS values increases, confidence in-creases as well. The results of the two measures of SP and SR do coincide with each other. They provide evidence for re-ward sensitivity in GD, which is related to higher confidence. However, we did not find
15 evidence for loss sensitivity in relation to
lower confidence in OCD.
4.4 Loss and Reward processing in the brain
The fifth and last aim of this study was to in-vestigate group differences in reward and loss processing in the brain. Despite finding significant activity in the striatum in both GD and HC during gain trials, no significant group differences were detected. This means that the behavioral finding of signifi-cant higher confidence with higher incen-tives in GD, did not result from differences in brain activity when processing potential gains, which is not in line with our hypothe-sis and earlier research of van Holst et al., 2012 and Balodis et al., 2020 that found ab-normal activity in the striatum during reward anticipation. One explanation is that in the current study, the incentive consisted of points instead of an actual monetary amount to be won or lost, which could have dampened the incentivization effect.
However, there is still a lot of contro-versy in the literature regarding neural re-ward processing in GD patients. On the one hand, studies as van Holst et al., 2012 show attenuated striatum activity during reward anticipation in GD patients, while on the other hand, studies such as Balodis et al., 2020 show diminished activity in the stria-tum when processing potential gains. In a review of Leyton and Vezina (2013) this dif-ference is explained by the presence or ab-sence of gambling-relevant cues, where presence causes overactivity in the striatum and absence causes diminished activity in the striatum. It would therefore be interest-ing to investigate the influence of different cues on processing potential rewards in GD patients.
Regarding OCD, abnormal activity was expected in the amygdala. However,
we did not find evidence for involvement of the amygdala in the processing of loss in both groups. This lack of findings in the loss condition can be explained by the fact that the design of the task is more gain focussed instead of loss focussed, as participants cannot actually lose money. This may also explain the lack of any behavioral effects of loss on confidence within OCD.
4.5 Conclusions and future directions
We identified underconfidence in OCD pa-tients, while overconfidence in GD patients was not identified. We also confirmed that the vmPFC is involved in automatic confi-dence encoding. Value encoding areas such as the striatum and insula were also related to confidence signaling, however, no abnormal activity was detected in both pa-tient groups. Since the small sample sizes of both patients groups probably caused the study to be underpowered, future research should include more participants. Also it would be interesting to look at connectivity patterns between the vmPFC and value en-coding areas since multiple studies con-firmed the involvement of this widespread network in confidence encoding. Moreover, since both patient groups showed a dis-turbed relationship between evidence and confidence, this effect should be assessed on a neural level.
Regarding incentives, we found that GD patients are reward sensitive, repre-sented by both higher confidence in gain contexts and a significant positive correla-tion between BAS values and confidence. Finally, this study does not provide evidence for abnormal incentive processing in the brain in both clinical groups. Future re-search can modify the incentivized confi-dence task to make sure that the task is bet-ter suited to pick up loss as well as gain ef-fects.
16 All in all, we can conclude that the
constructs of confidence abnormalities can thus far not be reliably used in transdiag-nostic psychiatric research or function as a biomarker for compulsive disorders. How-ever, before this can be said with confi-dence, more research is needed.
17
References
Admon, R., Bleich-Cohen, M., Weizmant, R., Poyurovsky, M., Faragian, S., & Hendler, T. (2012). Functional and structural neural indices of risk aver-sion in obsessive–compulsive disorder (OCD). Psy-chiatry Research: Neuroimaging, 203(2-3), 207-213. Bang, D., & Fleming, S. M. (2018). Distinct encoding of decision confidence in human medial prefrontal cor-tex. Proceedings of the National Academy of Sci-ences, 115(23), 6082-6087.
Balodis, I. M., Kober, H., Worhunsky, P. D., Stevens, M. C., Pearlson, G. D., & Potenza, M. N. (2012). Di-minished frontostriatal activity during processing of monetary rewards and losses in pathological gam-bling. Biological psychiatry, 71(8), 749-757.
Berger, U., & Anaki, D. (2014). The behavioral Inhibi-tion System (BIS) mediates major aspects of the rela-tionship between disgust and OCD symptomology. Journal of Obsessive-Compulsive and Related Disor-ders, 3(3), 249-256.
Brevers, D., Cleeremans, A., Bechara, A., Greisen, M., Kornreich, C., Verbanck, P., & Noël, X. (2013). Im-paired self-awareness in pathological gamblers. Jour-nal of Gambling Studies, 29(1), 119-129.
Brevers, D., Cleeremans, A., Bechara, A., Greisen, M., Kornreich, C., Verbanck, P., & Noël, X. (2014). Im-paired metacognitive capacities in individuals with problem gambling. Journal of gambling studies, 30(1), 141-152.
Brunborg, G. S., Johnsen, B. H., Mentzoni, R. A., Molde, H., & Pallesen, S. (2011). Individual differ-ences in evaluative conditioning and reinforcement sensitivity affect bet-sizes during gambling. Personal-ity and Individual Differences, 50(5), 729-734. Carver, C. S., & White, T. L. (1994). Behavioral inhibi-tion, behavioral activainhibi-tion, and affective responses to impending reward and punishment: the BIS/BAS scales. Journal of personality and social psychology, 67(2), 319.
Demaree, H. A., DeDonno, M. A., Burns, K. J., & Everhart, D. E. (2008). You bet: How personality dif-ferences affect risk-taking predif-ferences. Personality and Individual Differences, 44(7), 1484-1494. Bates, D., Mächler, M., Bolker, B., & Walker, S. (2014). Fitting linear mixed-effects models using lme4. arXiv preprint arXiv:1406.5823.
Figee, M., Pattij, T., Willuhn, I., Luigjes, J., van den Brink, W., Goudriaan, A., ... & Denys, D. (2016).
Com-pulsivity in obsessive–compulsive disorder and addic-tions. European Neuropsychopharmacology, 26(5), 856-868.
Fullana, M. A., Mataix-Cols, D., Caseras, X., Alonso, P., Menchón, J. M., Vallejo, J., & Torrubia, R. (2004). High sensitivity to punishment and low impulsivity in obsessive-compulsive patients with hoarding symp-toms. Psychiatry Research, 129(1), 21-27.
Gaher, R. M., Hahn, A. M., Shishido, H., Simons, J. S., & Gaster, S. (2015). Associations between sensi-tivity to punishment, sensisensi-tivity to reward, and gam-bling. Addictive behaviors, 42, 180-184.
Gherman, S., & Philiastides, M. G. (2018). Human VMPFC encodes early signatures of confidence in perceptual decisions. Elife, 7, e38293.
van Holst, R. J., Veltman, D. J., Büchel, C., van den Brink, W., & Goudriaan, A. E. (2012). Distorted expec-tancy coding in problem gambling: is the addictive in the anticipation?. Biological psychiatry, 71(8), 741-748.
Hoven, M., Lebreton, M., Engelmann, J. B., Denys, D., Luigjes, J., & van Holst, R. J. (2019). Abnormali-ties of confidence in psychiatry: an overview and fu-ture perspectives. Translational psychiatry, 9(1), 1-18. Kim, D. Y., & Lee, J. H. (2011). Effects of the BAS and BIS on decision-making in a gambling task. Per-sonality and Individual Differences, 50(7), 1131-1135. Kuznetsova A., Brockhoff P.B.,, & Christensen RHB. (2017). “lmerTest Package: Tests in Linear
Mixed Effects Models.” _Journal of Statistical Soft-ware_, *82*(13), 1-26. doi:
10.18637/jss.v082.i13 (URL: http://doi.org/10.18637/jss.v082.i13).
Lebreton, M., Abitbol, R., Daunizeau, J., & Pessig-lione, M. (2015). Automatic integration of confidence in the brain valuation signal. Nature neuroscience, 18(8), 1159.
Lebreton, M., Langdon, S., Slieker, M. J., Nooitge-dacht, J. S., Goudriaan, A. E., Denys, D., ... & Luigjes, J. (2018). Two sides of the same coin: Monetary in-centives concurrently improve and bias confidence judgments. Science advances, 4(5), eaaq0668. Leyton, M., & Vezina, P. (2013). Striatal ups and downs: their roles in vulnerability to addictions in hu-mans. Neuroscience & Biobehavioral Reviews, 37(9), 1999-2014.
18
Luttrell, A., Briñol, P., Petty, R. E., Cunningham, W., & Díaz, D. (2013). Metacognitive confidence: A neuro-science approach. Revista de Psicología Social, 28(3), 317-332.
Maccallum, F., Blaszczynski, A., Ladouceur, R., & Nower, L. (2007). Functional and dysfunctional impul-sivity in pathological gambling. Personality and Indi-vidual Differences, 43(7), 1829-1838.
Mazaika, P. K., Hoeft, F., Glover, G. H., & Reiss, A. L. (2009). Methods and software for fMRI analysis of clinical subjects. Neuroimage, 47(Suppl 1), S58. Mercer, K. B., & Eastwood, J. D. (2010). Is boredom associated with problem gambling behaviour? It de-pends on what you mean by ‘boredom’. International Gambling Studies, 10(1), 91-104.
Morein-Zamir, S., Papmeyer, M., Gillan, C. M., Crock-ett, M. J., Fineberg, N. A., Sahakian, B. J., & Robbins, T. W. (2013). Punishment promotes response control deficits in obsessive-compulsive disorder: evidence from a motivational go/no-go task. Psychological med-icine, 43(2), 391–400.
Martinez, M. E. (2006). What is metacognition?. Phi delta kappan, 87(9), 696-699.
O’Connor, R. M., Stewart, S. H., & Watt, M. C. (2009). Distinguishing BAS risk for university students’ drink-ing, smokdrink-ing, and gambling behaviors. Personality and Individual Differences, 46(4), 514-519.
Raffard, S., Capdevielle, D., Attal, J., Novara, C., & Bortolon, C. (2020). Apathy in Obsessive-Compulsive Disorder and Its Psychological Correlates: Compari-son With Individuals With Schizophrenia. The Journal of Neuropsychiatry and Clinical Neurosciences, 32(2), 168-174.
R Core Team (2018). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-pro-ject.org/.
Shapiro, A. D., & Grafton, S. T. (2020). Subjective value then confidence in human ventromedial prefron-tal cortex. Plos one, 15(2), e0225617.
Smith, D. V., Hayden, B. Y., Truong, T. K., Song, A. W., Platt, M. L., & Huettel, S. A. (2010). Distinct value signals in anterior and posterior ventromedial prefron-tal cortex. Journal of Neuroscience, 30(7), 2490-2495. Stankov, L., Kleitman, S., & Jackson, S. A. (2015). Measures of the trait of confidence. In Measures of personality and social psychological constructs (pp. 158-189). Academic Press.
Wardell, J. D., Quilty, L. C., Hendershot, C. S., & Bagby, R. M. (2015). Motivational pathways from re-ward sensitivity and punishment sensitivity to gam-bling frequency and gamgam-bling-related problems. Psy-chology of Addictive Behaviors, 29(4), 1022.
19
Appendix
A) Translated BIS/BAS scales
In de onderstaande vragenlijst ziet u stellingen staan waar u het mee eens of oneens kan zijn. Geef voor elke stelling aan in welke mate u het ermee eens of oneens bent. Dit kunt u doen door het hokje aan te vinken.
Beantwoord alle stellingen, sla er geen over. Per stelling is slechts één antwoord mogelijk. Probeer zo eerlijk mogelijk antwoord te geven, er zijn geen goede of foute antwoorden.
1. Als ik denk dat er iets onprettigs gaat gebeuren, raak ik
meestal behoorlijk "opgefokt". helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
2. Ik ben bezorgd om het maken van fouten. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
3. Als ik iets wil, ga ik er meestal helemaal voor. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
4. Vaak doe ik dingen om geen andere reden dan dat het wel eens
leuk zou kunnen zijn. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
5. Kritiek of een standje raken mij behoorlijk. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
6. Als ik iets krijg wat ik wil, voel ik me opgewonden en
op-geladen. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
7. Ik doe een hoop moeite om dingen die ik wil te krijgen. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
8. Ik verlang sterk naar spanning en nieuwe sensaties. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
20
9. Ik voel me behoorlijk overstuur als ik denk of weet dat
ie-mand boos op me is. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
10. Ik ben altijd bereid iets nieuws te proberen als ik denk dat het leuk
zal zijn. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
11. Als ik iets goed doe, wil ik er graag mee doorgaan. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
12. Zelfs als mij iets ergs staat te gebeuren, ervaar ik zelden
angst of nervositeit helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
13. Ik handel vaak zoals het moment me ingeeft. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
14. Als ik een kans zie iets te krijgen wat ik wil, ga ik er meteen op af. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
15. Als mij goede dingen overkomen, raakt dat me sterk. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
16. Ik voel me bezorgd als ik denk dat ik slecht heb gepresteerd op
iets. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
17. Ik zou het spannend vinden een wedstrijd te winnen. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
20
18. Vergeleken met mijn vrienden heb ik erg weinig angsten. helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
19. Als ik een mogelijkheid zie iets te krijgen wat ik leuk vind, word
ik direct opgewonden. Helemaal mee eens Beetje mee eens Beetje mee oneens Helemaal mee oneens
20. Als ik een wedstrijd zou winnen, zou ik erg enthousiast zijn helemaal mee eens beetje mee eens beetje mee oneens helemaal mee oneens
Score calculation Answering options:
Totally agree (4 points) Somewhat agree (3 points) Somewhat disagree (2 points) Totally disagree (1 point) BIS items: 1,2,5,9,12*,16,18* BAS items:
BAS Reward Responsivess: 6,11,15,17,19 BAS Drive: 3,7,14,20
BAS Fun Seeking: 4,8,10,13 *Reversed scored
21
B) BAS subgroups analyses
C) Correlation analyses
21
D) Neuroimaging results of loss>neutral contrast at incentive/rating moment
Figure 8: Neuroimaging results of loss>neu-tral contrast at incentive/rating moment in HC. ROI analysis with a mask of the bilateral amygdala did not show any significant activity. However, whole-brain analysis did show signifi-cant activity in the lingual gyrus (red), right SMA (blue), caudate nucleus (green) and the posterior cingulate gyrus (purple). The shown activations had a threshold of 0.001 (p < 0.05, FWE corrected).
Figure 7: Neuroimaging results of loss>neutral contrast at incentive/rating moment in OCD. ROI analysis with a mask of the bilateral amygdala did not show any significant activity. However, whole-brain analysis did show significant activity in the left calcarine fissure (red), SMA (blue), right angular gyrus (purple), left supramarginal gy-rus (lightblue) and in the premotor cortex (green). The shown activations had a thresh-old of 0.001 (p < 0.05, FWE corrected).
