Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Dealing with uncertainty: A high-density EEG investigation on how intolerance of uncertainty affects emotional predictions

  • Fiorella Del Popolo Cristaldi ,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing

    fiorella.delpopolocristaldi@phd.unipd.it

    Affiliation Department of General Psychology, University of Padua, Padova, Italy

    ⨯
  • Giovanni Mento,

    Roles Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Writing – review & editing

    Affiliations Department of General Psychology, University of Padua, Padova, Italy, Padua Neuroscience Center (PNC), University of Padua, Padova, Italy

    ⨯
  • Michela Sarlo,

    Roles Conceptualization, Methodology, Supervision, Writing – review & editing

    Affiliation Department of Communication Sciences, Humanities and International Studies, University of Urbino Carlo Bo, Urbino, Italy

    ⨯
  • Giulia Buodo

    Roles Conceptualization, Methodology, Project administration, Supervision, Writing – review & editing

    Affiliation Department of General Psychology, University of Padua, Padova, Italy

    ⨯

Abstract

Intolerance of uncertainty (IU) can influence emotional predictions, constructed by the brain (generation stage) to prearrange action (implementation stage), and update internal models according to incoming stimuli (updating stage). However, neurocomputational mechanisms by which IU affects emotional predictions are unclear. This high-density EEG study investigated if IU predicted event-related potentials (ERPs) and brain sources activity developing along the stages of emotional predictions, as a function of contextual uncertainty. Thirty-six undergraduates underwent a S1-S2 paradigm, with emotional faces and pictures as S1s and S2s, respectively. Contextual uncertainty was manipulated across three blocks, each with 100%, 75%, or 50% S1-S2 emotional congruency. ERPs, brain sources and their relationship with IU scores were analyzed for each stage. IU did not affect prediction generation. During prediction implementation, higher IU predicted larger Contingent Negative Variation in the 75% block, and lower left anterior cingulate cortex and supplementary motor area activations. During prediction updating, as IU increased P2 to positive S2s decreased, along with P2 and Late Positive Potential in the 75% block, and right orbito-frontal cortex activity to emotional S2s. IU was therefore associated with altered uncertainty assessment and heightened attention deployment during implementation, and to uncertainty avoidance, reduced attention to safety cues and disrupted access to emotion regulation strategies during prediction updating.

Citation: Del Popolo Cristaldi F, Mento G, Sarlo M, Buodo G (2021) Dealing with uncertainty: A high-density EEG investigation on how intolerance of uncertainty affects emotional predictions. PLoS ONE 16(7): e0254045. https://doi.org/10.1371/journal.pone.0254045

Editor: Maria Serena Panasiti, Sapienza University of Rome, ITALY

Received: February 6, 2021; Accepted: June 19, 2021; Published: July 1, 2021

Copyright: © 2021 Del Popolo Cristaldi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The datasets generated and analyzed during the current study, along with the R analysis script, are available on Figshare public repository, DOI: 10.6084/m9.figshare.13560569.

Funding: The study was supported by a grant from MIUR (Dipartimenti di Eccellenza DM 11/05/2017 n. 262) to the Department of General Psychology.

Competing interests: The authors have declared that no competing interests exist.

Introduction

The world is an uncertain place, and the uncertainty we experience in everyday life can have a dramatic influence on our emotional life. For example, the COVID-19 pandemic outbreak, and the associated risk of contracting the virus, carried a huge amount of uncertainty, substantially impacting over the emotional experience and mental health of people worldwide. Or else, not knowing how the plot is going to develop while watching a new TV series usually increases the rewarding value of the series itself. From both examples, it is clear how dealing with uncertainty can intensify affective experience.

The literature has long suggested that uncertainty pushes to a polarization of the impact of both positive and negative affect, leading to an increased attention and emotional engagement towards uncertain events [1]. Furthermore, individual differences in the subjective predisposition to tolerate uncertainty also play a non-negligible role in shaping emotional processing within uncertain environments. Intolerance of uncertainty (IU) is the dispositional characteristic that reflects individual differences in tolerating and adapting to uncertain situations [2]. It has been defined as the inability to tolerate the aversive reaction triggered by a perceived lack of sufficient or salient information, sustained by the correlated perception of uncertainty [3]. Individuals high in IU tend to appraise uncertainty as threatening irrespective of its probability and outcomes, to experience heightened physiological arousal in uncertain situations, and to consider the possibility of the occurrence of a negative event as unacceptable [2,4,5]. IU is characterized by cognitive, affective, and behavioral facets that have a broad influence on emotional experience, i.e., inflated estimates of threat probability, increased attention to threat and hypervigilance, deficient safety learning, behavioral and cognitive avoidance of exposure to uncertainty, and heightened physiological reactivity to threat uncertainty [6]. Moreover, the fear of the unknown (which is assumed to be the key feature at the basis of IU) has been proved to be a lower-order construct able to account for statistically significant variance in several higher-order constructs [2,7,8]. Indeed, it is positively associated with neuroticism [3,9], and with symptoms of different anxiety disorders (see [3] for a review); and it can act as a trigger for the activation of the behavioral inhibition system (BIS) [2,3], and the fight-flight-freeze defensive response [2,3]. For these reasons, IU has been proposed as a sound trans-diagnostic risk and maintaining factor, potentially crucial for vulnerability assessment and/or as a target for clinical treatment, for a wide range of emotional traits and disorders, such as neuroticism, trait anxiety and negative affectivity traits, as well as anxiety disorders, depressive disorders, obsessive-compulsive disorders, trauma- and stressor-related disorders, and eating disorders [1–4,10].

In the last years, the degree of uncertainty intrinsic to contextual information has been targeted as an influential feature in shaping emotional processing, as defined within the predictive coding framework [11–14]. According to this conceptualization, emotions are interoceptive predictive models, developing across three neurocomputational stages: prediction generation, in which the brain generates predictions according to prior experience and contextual information available in the environment; prediction implementation, in which bodily and behavioral changes are prearranged to efficiently deal with the predicted situation; prediction updating, in which predictions are compared with actual incoming stimuli, and mismatching ones are encoded as prediction errors and used to adjust future predictions [11–16]. Contextual differences in the relative balance of the amount of information between certain and uncertain situations are assumed to be crucial in determining whether uncertainty is appraised as threatening [3]. Moreover, it has been proposed that biases in the anticipation of emotional stimuli (e.g., inflated threat estimates) characterizing high-IU individuals could result from a disrupted prediction error signaling, which in turn results in a failure to update emotional predictions [6]. For these reasons and given the significance of IU as a trans-diagnostic risk factor, a better understanding of how individual differences in IU affect the neural correlates of emotional predictions as a function of contextual uncertainty could represent a valuable contribution to advancing knowledge in the field, leading to promising clinical and preventive implications. In fact, human organisms are embedded in constantly changing environments, conveying information with varying degrees of (un)certainty, and in which a continuous adaptation of emotional predictions is required. Assessing the neural correlates of individual differences in IU across the different neurocomputational stages of emotional predictions would allow to identify potential targets to treat, or to early intervene on, thus preventing high-IU from developing into clinically relevant conditions.

Nevertheless, experimental studies investigating at a neurocomputational level the effects of IU on emotional processing as a function of contextual uncertainty [17–23] are still scarce and provided conflicting results, with none of them embracing the predictive perspective on emotions [13,14]. Amongst these studies, uncertainty was manipulated mainly through two experimental paradigms. Some studies, measuring startle reflex and event-related potentials (ERPs), employed threat-of-shock paradigms [17,18,24], in which participants were presented with cues preceding predictable vs. unpredictable electric shocks. Other studies, recording skin conductance, electromyographic activity (EMG), ERPs, or functional magnetic resonance imaging (fMRI) [19–23], employed emotional S1-S2 paradigms, with 100% vs. 50% instructed emotional congruency between S1s (symbolic cues) and S2s (emotional pictures). Experimental tasks varied from passive viewing of stimuli [20,22], to tasks requiring shock/S2 ratings [17,18,21,23], or expectancy ratings about the upcoming S2 [19,21]. All these studies measured IU through the Intolerance of Uncertainty Scale (IUS), either in its original (27-item) or short (12-item) form [5,25], and divided the samples into two groups (high- vs. low-IU) according to the IUS scores distribution. Uncertainty was manipulated mostly through two extreme levels (fully predictive vs. random), and only negatively-valenced emotional stimuli were employed and compared with a neutral condition. Importantly, both S1-S2 and threat-of-shock paradigms can be reconsidered within the predictive framework [13,14] as a function of the three stages of the construction of emotional predictions, with S1/cue processing potentially reflecting the generation stage, the inter-stimulus interval (ISI) between S1/cue and S2/shock reflecting the implementation stage, and the S2/shock processing representing prediction updating.

However, the results collected are fragmentary and sometimes in conflict with the predictions formulated by the theoretical models of IU [1,2,4,6]. Regarding the generation stage, some studies [19,22] found no differences between high- and low-IU participants in S1/cue-locked skin conductance and corrugator EMG responses as a function of cue predictive meaning. Other studies [18,20] found that high-IU participants showed a larger P2 ERP component to S1s/cues irrespective of their predictive meaning, reflecting a heightened early automatic attention allocation towards cues and their potential threatening meaning. Focusing on prediction implementation, most of the studies found no significant effects of IU on ERPs amplitude, skin conductance, EMG responses, and expectancy ratings measured during the ISI between S1/cue and S2/shock [18,21,24]; while [19] found that high-IU participants showed higher threat expectancy ratings following cues conveying uncertain predictive information, consistently with the overestimation of threat predicted by the theoretical models of IU [6]. As for the prediction updating stage, some studies measuring skin conductance [21,22] failed to find significant effects of IU on S2-locked indices. Other studies found that in conditions of uncertainty IU negatively predicted neural activity within posterior frontomedian cortex (PFMC) [23], and the magnitude of the startle response [17], thus unexpectedly suggesting that as IU increases, the aversive reactions to uncertain threat attenuate, probably due to the perceived lack of controllability typical of uncertain contexts. Lastly, in high-IU participants uncertain negative S2s were found to elicit smaller Late Positive Potential (LPP) amplitudes, reflecting a dampening effect on the emotional processing of uncertain stimuli [20]. Altogether, the results related to the updating stage are in contrast with the assumption that uncertainty is associated with affect intensification and heightened physiological reactivity, as predicted by the theoretical models of IU [1,2,4,6].

The present study aimed to investigate through high-density EEG whether and how IU predicted the neural activity, both at the scalp and at the source level, which develops in contexts of different predictive value across the stages of generation, implementation and updating of emotional predictions. To this aim, we employed an emotional S1-S2 paradigm (cfr. [26]) in which contextual uncertainty was manipulated by means of valid (100%), moderately predictive (75%), and random (50%) emotional congruency between S1 and S2. Standardized emotional faces and pictures with positive, negative, or neutral valence were employed as S1s and S2s, respectively. Furthermore, we measured IU through the Italian adaptation of the IUS [27], and we used it in our analyses as a continuous predictor rather than dichotomizing its scores into two groups, thus avoiding flattening individual differences in IU and losing information and power [28]. Besides allowing to separately investigate the three stages of emotional predictions, the S1-S2 paradigm we employed contributed to fill at least three gaps in the extant literature. First, an intermediate level of contextual uncertainty (75%), which could better resemble a more realistic, moderately certain environmental context, was included. Second, a condition of positive valence was added, thus allowing to disentangle valence- vs. arousal-specific effects. Third, the optimal compromise between spatial and temporal resolution offered by high-density EEG [29] allowed to provide a compelling comprehensive neurocomputational evidence, thus contributing to narrow the gap between ERP and fMRI findings.

For each neurocomputational stage, we targeted specific ERP components, which have been shown to respond to the experimental manipulation (cfr. [26]). As for prediction generation, we focused on S1-N170, which is sensitive to predictive contexts, and reflects structural encoding and coarse emotional processing of facial expressions [30–32]. As to the implementation stage, we targeted the Contingent Negative Variation (CNV) [33] arising during the ISI, which reflects the orientation of attention towards motivationally relevant stimuli (early portion), and the motor preparation for a required or possible action (late portion), besides being sensitive to predictive contextual factors [34–37]. As regards prediction updating, we considered both early and late S2-ERPs, i.e., the P2, indexing early automatic attention allocation and the degree of updating of internal models, and the LPP, reflecting late motivated and sustained attention to emotional stimuli [37–41].

We hypothesized higher IU scores to predict (i) a larger N170 irrespective of the predictive context [18,20] during prediction generation; (ii) larger early and late CNV amplitudes following negative S1s (reflecting the anticipatory processes elicited by inflated threat estimates, see [19]) in uncertain predictive contexts (75% and 50%) during the implementation stage; (iii) either smaller (according to ERP findings, e.g., [20]) or larger (according to the IU theories’ predictions, e.g., [1]) P2 and LPP amplitudes to emotional S2s in uncertain predictive contexts (75% and 50%) during prediction updating.

Methods

Participants

Thirty-six right-handed undergraduates at University of Padova took part in the study as volunteers (16 males, age: M = 23.25, SD = 1.85, range = 20–29). The sample size was based on previous ERP research measuring IUS within S1-S2 paradigms [20]. A post-hoc power analysis (α = 0.05, η2 = 0.02) showed that the final sample size (N = 36) allowed to reach an average power of 0.541.

A total of 294 undergraduates were initially screened for participation through an online survey evaluating the inclusion criteria for the study: no history of neurological or psychiatric disorders; normal or corrected-to-normal vision; no medication; right-handedness, as assessed by the Edinburgh Handedness Inventory [42]; absence of extreme levels of blood-injection-injury fear, as assessed by the Fear Survey Schedule [43,44]. Since some negative pictures depicted gory scenes, highly fearful participants (N = 108), who scored 4 on items of the Fear Survey Schedule concerning blood, injuries and/or weapons (0–4 score), were discarded from participation for ethical reasons. Of the remaining participants, 23 were excluded because left-handed, and 127 because they did not meet the other inclusion criteria.

All participants signed an informed consent. All experimental procedures were approved by the Ethical Committee for the Psychological Research of the University of Padua (protocol no. 2859) and were conducted in accordance with the Declaration of Helsinki.

Stimulus material and procedure

Before arrival, all participants answered an online version of the IUS, administered in its Italian 12-item validated adaptation [27]. Participants were instructed to answer the survey alone, in a silent room, taking all the time needed.

Upon arrival, participants seated in a dimly lit room, at a 90-cm viewing distance from a 24-inch computer monitor (1280 x 1024 resolution). They received detailed information about EEG montage and the experimental task. Then, an elastic 128-ch EEG net was applied for EEG recording.

After a 10 min adaptation period, a computerized S1-S2 paradigm was presented on the screen. S1s were 24 colored emotional faces selected from the NimStim Set of Facial Expressions [45]: 4 male and 4 female Caucasian models, each posing a fearful, happy, and neutral facial expression. S2s were 120 colored emotional pictures selected from the International Affective Picture System (IAPS) [46]: 40 highly-arousing negative, 40 highly-arousing positive, and 40 neutral. Positive and negative emotional pictures did not differ for mean standardized arousal ratings (Ms = 6.43 and 6.44, respectively; t(93) = 0.12, p = .9). NimStim and IAPS picture numbers are listed in S1 File. S1s and S2s were presented pseudo-randomly (no more than 3 subsequent same-valence couplings) in their original size, in the center of the screen against a black background, through E-prime software [47].

At the beginning of the S1-S2 paradigm, participants read the on-screen the task instructions at their own pace. They were told that they would see a face followed by a picture and that they only had to look at the screen, trying to move as little as possible. Each trial started with the presentation of S1 for 500 ms, followed by a fixed inter-stimulus interval (ISI) of 2000 ms, in which the screen remained black. Then S2 was presented for 1500 ms. A white fixation cross against a black background was displayed during the inter-trial interval (ITI), whose length randomly varied between 800 and 1200 ms. The total number of trials was 360.

S1 and S2 valence was manipulated within-subjects through three levels: positive (POS), negative (NEG) and neutral (NEU). Contextual uncertainty was implicitly manipulated within-subjects through 3 continuous blocks of 120 trials each. In the 100% block, S1-S2 emotional congruency was complied in all the trials, that is, S1 valence was fully predictive of S2 valence. In the 75% block, S1-S2 pairs were congruent in 75% of the trials, so S1 valence was moderately predictive of S2 valence. In the 50% block, S1 valence was unpredictive of S2 valence, since each S1 could be randomly followed by a positive, negative, or neutral S2. The blocks seamlessly followed one another, with brief breaks randomly distributed during the whole procedure, and a between-subjects counterbalanced order.

Participants did not receive any explicit information about the different between-blocks contextual uncertainty levels. At the end of the S1-S2 paradigm, the experimenter ensured that the exact S1-S2 probabilistic ratio had remained implicit for each participant by asking if they had caught any relationship between S1 and S2. Participants only reported to have caught a general within-trials emotional congruency or incongruency, but none of them reported neither the precise probabilistic ratio nor to have caught the block manipulation.

Finally, participants completed a self-paced computerized Emotional Recognition Task. For each S1, participants were asked to select which emotion best represented the facial expression displayed, choosing between the following options: anger, disgust, fear, happiness, sadness, surprise, and no emotion. To avoid automaticity in responding, the order of choice options was randomized. The percentage of correct responses was computed for each participant. The Emotional Recognition Task was used to check whether all participants correctly recognized S1 valence. No participant scored lower than 60% accuracy in the task, therefore we could reliably exclude that their predictions could have been affected by a misled S1 valence recognition.

At the end of the experimental session each participant was thanked for the participation and was given detailed information about the research question and objectives of the study.

Electrophysiological recordings, brain source modelling and data analysis

During the S1-S2 paradigm, the EEG was continuously recorded at a 500 Hz sampling rate through a pre-cabled 128-channel HydroCel Geodesic Sensor Net (HCGSN-128), using a Geodesic high-density EEG System (EGI® GES-300). All electrodes were referenced online to the vertex, and scalp voltages were amplified through a 24-bit DC amplifier (Net Amps™ 300). The impedance was kept below 60 kΩ for each sensor. Signal preprocessing was accomplished offline through the MATLAB toolbox EEGLAB 14.1.2b [48]. The EEG signal was downsampled at 250 Hz and filtered with a digital band-pass filter (0.01–40 Hz, -6dB) using a Hamming windowed sinc finite impulse response filter (filter order = 82500). To compute the ERPs, the continuous EEG was segmented into 4500 ms epochs, from 500 ms before to 4000 ms after S1 onset. The signal was then baseline-corrected to two different baselines: from 500 ms before to the onset of S1 for the purpose of S1-locked ERP analysis; from 200 ms before to the onset of S2 for the purpose of S2-locked ERP analysis. Bad channels were reconstructed with the spherical spline interpolation method [49,50]. Epochs were then digitally inspected through the TBT EEGLAB plug-in, which was applied to electrodes from E40 to E100 (frontal electrode sites were excluded because no eye blinks/movements correction was yet applied). The TBT algorithm performed an automatic rejection of epochs and interpolation of channels on an epoch-by-epoch basis: channels that exceeded a differential average amplitude of ±150 μV were identified and marked for rejection. Channels marked as bad on more than 30% of all epochs were excluded and subsequently interpolated. Epochs having more than 10 bad channels were also excluded. The Infomax algorithm [51] implemented in EEGLAB was used to perform Independent Component Analysis [52]. Artifact-reduced data were decomposed into independent components, which were visually inspected to discard those related to eye blinks, eye movements, heartbeat, and muscular signals, according to their morphology and scalp distribution. The remaining components were then projected back to the electrode space to obtain cleaner EEG epochs. Epochs were further visually inspected and residual artifact-contaminated trials were rejected. Experimental conditions did not differ for final number of epochs accepted (see S1 Table). Data were finally re-referenced to the average of all electrodes. Individual average and grand average ERPs were computed for all experimental conditions, applying a weighted average in order to control for any potential unbalanced number of epochs per condition [53,54].

The ERPs statistical analysis was performed via Brainstorm software, using the Fieldtrip functions [55,56]. A whole-brain paired two-tailed t-test (α = .05) permutation approach [57,58] was used, performing 1000 Monte-Carlo cluster-based corrected permutations over all 128 channel locations. This approach allows to effectively control the familywise error rate, and it can be used also for complex factorial designs, since providing greater statistical power than traditional spatiotemporal averaging approaches when applied to a priori defined time windows [57–59]. The permutations were computed over 6 a priori time windows, corresponding to 4 distinguishable ERP components: S1-locked N170 (140–180 ms) for prediction generation; S1-locked early CNV (eCNV, 1500–2000 ms) and late CNV (lCNV, 2000–2500 ms) for prediction implementation; S2-locked P2 (200–300 ms), early LPP (eLPP, 400–600 ms) and late LPP (lLPP, 600–800 ms) for prediction updating. Interaction effects were tested through a difference-based strategy widely used in the literature, which allowed to further reduce the familywise error rate [59]. All the planned paired-wise comparisons performed are summarized in S1 File. Based on the results of the permutation analysis, ERPs were extracted as the mean voltage amplitude in the abovementioned time windows from the following electrode clusters: an occipital cluster (E70, E74, E75, E81, E82, E83) for the N170, a left-central cluster (E40, E41, E42, E46, E47) for eCNV and lCNV, a parietal cluster (E67, E71, E72, E75, E76, E77) for the P2, and a slightly different parietal cluster (E60, E61, E62, E67, E72, E77, E78, E85) for eLPP and lLPP.

The cortical sources of ERP activity were modeled via Brainstorm software [55]. ICBM152 anatomical template [60] was used to approximate the individual anatomies of each participant and warped to the EEG sensor positions, by employing rigid rotations and translations of digitized landmarks. The conductive head volume was modeled according to the realistic forward model OpenMEEG BEM [61]. The source space was constrained to the cortex and modeled as a grid of 15002 orthogonal current dipole triplets normally oriented to the cortical surface. The inverse modeling was based on sLORETA algorithm and implemented with default parameter settings. This algorithm has proved to be a good source estimator, leading to small localization errors even with low signal-to-noise ratio [62]. The noise covariance matrix was computed from the average of EEG baselines. For each participant, sources were projected to a standard anatomical template (MNI) and their activity was transformed in absolute Z scores relative to the baselines. Finally, a spatial smooth with a FWHM of 3 mm was applied to each source. Cortical activations were located according to the anatomical Destrieux atlas [63] adapted for cortical space solution. Based on each ERP source reconstruction, source map vertices were clustered in the following regions of interest (ROI): right superior temporal sulcus (r-STS) for the N170; left anterior cingulate cortex (l-ACC), left supplementary motor area (l-SMA), and left dorsal-posterior cingulate cortex (l-dPCC) for the eCNV and lCNV; bilateral temporo-parietal junction (TPJ) for the P2; right orbitofrontal cortex (r-OFC), and right temporal pole for the eLPP and lLPP. Absolute values of each ROI were time-averaged from the pertaining ERP time windows, extracted for each participant, and transformed using a natural logarithm.

To assess whether IUS score predicted the neural activity during the task both at the scalp and at the source level, separate linear mixed-effects models (R package: lme4 [64]) with individual random intercept were estimated, with each averaged ERP and ROI activity as dependent variables, IUS as a fixed continuous predictor, S1/S2 valence, block, and the interaction IUS × S1/S2 valence × block as fixed factors. The slopes of the IUS trend for each level of the factors (block, S1/S2 valence) were estimated, and their pairwise differences were tested by means of post-hoc Tukey test (R package: emmeans [65]).

Results

ERPs and cortical sources reconstruction

Prediction generation stage–N170 (140–180 ms) to S1 onset.

A significant negative occipital cluster, reflecting a larger N170, was found in 100% block when comparing emotional with neutral faces (POS vs. NEU p = .024, cluster statistic (c) = -476, cluster size (s) = 153; NEG vs. NEU p = .004, c = -871, s = 246), in 75% block when comparing fearful with positive and neutral faces (NEG vs. NEU p = .002, c = -1080, s = 306; NEG vs. POS p = .018, c = -455, s = 178), and in 50% block when comparing positive with neutral faces (POS vs. NEU p = .03, c = -413, s = 164) (Fig 1, panel A). No significant effects were found when comparing difference waves between blocks.

thumbnail
Download:
Fig 1. Modulation of ERPs and brain sources during prediction generation stage.

PANEL A: Grand average ERP waveforms following negative (red lines), positive (green lines) and neutral (blue lines) faces (S1) in the (a) 100%, (b) 75% and (c) 50% blocks. Waveforms are plotted from an occipital cluster of electrodes (E70, E74, E75, E81, E82, E83). Shaded areas denote standard error. The N170 was computed between 140 and 180 msec from S1 onset (time 0). PANEL B: On the top, cortical maps reconstruction of the (a) POS-NEU and (b) NEG-NEU differences in brain activations in the N170 temporal window (140–180 ms), regardless of blocks. On the bottom, time course of (c) r-STS activations to positive (green lines), negative (red lines), and neutral (blue lines) faces, regardless of blocks. r-STS = right superior temporal sulcus.

https://doi.org/10.1371/journal.pone.0254045.g001

N170 source reconstruction highlighted an emotional modulation on the right superior temporal sulcus (r-STS), showing a maximum pattern of activation after fearful faces (Fig 1, panel B). This pattern replicated consistently across blocks.

Prediction implementation stage–eCNV (1500–2000 ms) and lCNV (2000–2500 ms) to S1 onset.

Permutation analyses showed no significant effects of block, nor interaction effects with S1 valence on both early and late CNV. Nevertheless, visual inspection of grand average ERP waveforms from a central-left cluster of electrodes showed some amplitude differences, suggesting that a slightly larger CNV could have been elicited in the 50% block as compared to the others (see S1 Fig). However, since this result is not supported by statistics, it will not be further discussed.

CNV source analysis showed the involvement of a left network, extending over the l-ACC, the supplementary motor area (l-SMA), and the posterior-dorsal part of the cingulate gyrus (l-dPCC).

Prediction updating stage–P2 (200–300 ms), eLPP (400–600 ms), and lLPP (600–800 ms) to S2 onset.

A significant negative parietal cluster was found, signaling a reduced positivity to positive and negative, and so a larger P2 to neutral pictures in all the blocks. Furthermore, a larger P2 was found in 100% block comparing negative to positive pictures (Table 1; Fig 2, panel A). No significant effects were found when comparing difference waves between blocks.

thumbnail
Download:
Fig 2. Modulation of ERPs and brain sources during prediction updating stage.

PANEL A: Grand average ERP waveforms following negative (red lines), positive (green lines), and neutral (blue lines) pictures (S2) in the (a) 100%, (b) 75%, and (c) 50% blocks. Waveforms are plotted from a parietal cluster of electrodes (E60, E61, E62, E67, E71, E72, E75, E76, E77, E78, E85). Shaded areas denote standard error. P2 was computed between 200 and 300 msec, and LPP between 400 and 800 msec from S2 onset (time 0). PANEL B: On the top, cortical maps reconstruction of the (a) NEU-POS and (b) NEU-NEG differences in brain activations in the P2 temporal window (200–300 ms). On the bottom, time course of (c) l-TPJ, and (d) r-TPJ activations to positive (green lines), negative (red lines), and neutral (blue lines) pictures, regardless of blocks. l-TPJ = left temporoparietal junction, r-TPJ = right temporoparietal junction. PANEL C: On the top, cortical maps reconstruction of the (a) POS-NEU and (b) NEG-NEU differences in brain activations in the total LPP temporal window (400–800 ms). On the bottom, time course of (c) r- OFC and (d) r-temporal pole activations to positive (green lines), negative (red lines), and neutral (blue lines) pictures, regardless of blocks. r-OFC = right orbitofrontal cortex.

https://doi.org/10.1371/journal.pone.0254045.g002

thumbnail
Download:
Table 1. Results of planned comparisons in the prediction updating stage.

https://doi.org/10.1371/journal.pone.0254045.t001

P2 source reconstruction highlighted a larger activation of bilateral temporo-parietal junction (TPJ) to neutral than emotional pictures in all the blocks (Fig 2, panel B).

A significant positive parietal cluster was found, confirming the effect of S2 valence on LPP: positive and negative pictures elicited a larger early and late LPP than neutral pictures in all the blocks. Furthermore, a larger lLPP was found in 100% block comparing negative with positive pictures (Table 1; Fig 2, panel A). No significant effects were found when comparing difference waves between blocks.

LPP source analysis showed that, as compared with neutral, emotional stimuli elicited a larger activation of the right orbitofrontal cortex (r-OFC) and the right temporal pole in all the blocks (Fig 2, panel C).

Mixed-effects models

Prediction generation stage–N170 and r-STS activity (140–180 ms to S1 onset).

No significant main nor interaction effects of the predictors (IUS, S1/S2 valence, block, and their interaction) were found for N170 and r-STS activity (fitted models are summarized in S2 Table).

Prediction implementation stage–early (1500–2000 ms to S1 onset) and late (2000–2500 ms to S1 onset) CNV, l-ACC, l-SMA, and l-dPCC activity.

In the early time window (1500–2000 ms), a significant main effect of IUS was found for eCNV amplitude (F(1, 34) = 4.4, p = .044, η2 = .014, Fig 3A) and early l-ACC activity (F(1, 34) = 6.29, p = .017, η2 = .021, Fig 3B): higher IUS score predicted larger eCNV negativity and smaller early l-ACC activity regardless of block and S1 valence. No other significant main effects or interactions were found for eCNV, and early l-ACC, l-SMA and l-dPCC activity (fitted models are summarized in Tables 2 and S3).

thumbnail
Download:
Fig 3. Regression plots in the two time windows of the prediction implementation stage.

Relationships between IUS scores and (A) early CNV amplitude, (B) early l-ACC activity, (C) late CNV amplitude in 100% (in red), 75% (in green), and 50% (in blue) blocks, (D) late l-ACC and (E) late l-SMA activity. Shaded areas denote the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0254045.g003

thumbnail
Download:
Table 2. Summary of fitted models in the prediction implementation stage.

https://doi.org/10.1371/journal.pone.0254045.t002

As for the late time window (2000–2500 ms), analysis of lCNV showed significant main effects of IUS (F(1, 34) = 6.92, p = .013, η2 = .022) and block (F(2, 272) = 4.07, p = .018, η2 = .025), better explained by a significant IUS × block interaction (F(2, 272) = 3.48, p = .032, η2 = .022, Fig 3C): a statistically significant difference in the relationship between IUS and lCNV was found between blocks, regardless of S1 valence. The slope analysis revealed that the slope of the relationship between IUS score and lCNV amplitude was statistically different from 0 only in the 75% block (b = -0.061, bSE = 0.017, 95% CI = -0.0940, -0.0286), suggesting that higher IUS score predicted larger lCNV amplitude only in the 75% block, regardless of S1 valence. Post-hoc contrasts showed a significant difference between the slopes in the 75% vs 100% block (t(272) = -2.42, p = .043), with more negative values in the first. A significant main effect of IUS was found for late l-ACC (F(1, 34) = 4.69, p = .037, η2 = .016, Fig 3D) and l-SMA activity (F(1, 34) = 4.17, p = .049, η2 = .014, Fig 3E): higher IUS score predicted smaller late l-ACC and l-SMA activity regardless of block and S1 valence. No other significant main effects or interactions were found for lCNV, and late l-ACC, l-SMA and l-dPCC activity (fitted models are summarized in Tables 2 and S4).

Prediction updating stage–P2 and bilateral TPJ (200–300 ms to S2 onset), early (400–600 ms to S2 onset) and late (600–800 ms to S2 onset) LPP, r-OFC and r-temporal pole activity.

In the first time window (200–300 ms), analysis of P2 showed a significant main effect of block (F(2, 272) = 4.12, p = .017, η2 = .027), better explained by the significant IUS × block interaction (F(2, 272) = 3.98, p = .02, η2 = .026, Fig 4A), and a significant IUS × S2 valence interaction (F(2, 272) = 4.45, p = .013, η2 = .029, Fig 4B): a statistically significant difference in the relationship between IUS and P2 amplitude was found between blocks regardless of S2 valence, and between S2 valence levels regardless of block. Nevertheless, the slope analysis revealed that the slopes of the relationship between IUS score and P2 amplitude were not statistically different from 0 in any of the block and S2 valence levels. Post-hoc contrasts showed significant differences between the slopes in the 75% vs 100% block (t(272) = -2.72, p = .019), and between POS vs. NEU S2 valence levels (t(272) = -2.9, p = .011), with more negative slope values for the former. Furthermore, a significant main effect of block (F(2, 272) = 3.44, p = .034, η2 = .022) was found for l-TPJ activity. Nevertheless, since overall predictive effects have already been tested by means of permutation analyses, and no effects of block were found on P2 amplitude and its corresponding neural sources, this result will not be further commented. No other significant main effects or interactions were found for P2, and bilateral TPJ activity (fitted models are summarized in Tables 3 and S5).

thumbnail
Download:
Fig 4. Regression plots in the three time windows of the prediction updating stage.

Relationships between IUS scores and (A) P2 amplitude in 100% (in red), 75% (in green), and 50% (in blue) blocks, (B) P2 amplitude to neutral (in blue), positive (in green), and negative (in red) S2s, (C) early LPP amplitude in 100% (in red), 75% (in green), and 50% (in blue) blocks, (D) early and (E) late r-OFC activity to neutral (in blue), positive (in green), and negative (in red) S2s. Shaded areas denote the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0254045.g004

thumbnail
Download:
Table 3. Summary of fitted models in the prediction updating stage.

https://doi.org/10.1371/journal.pone.0254045.t003

In the second time window (400–600 ms), analysis of the eLPP showed a significant main effect of block (F(2, 272) = 3.82, p = .023, η2 = .023), better explained by the significant IUS × block interaction (F(2, 272) = 4.69, p = .01, η2 = .028, Fig 4C), and a main effect of S2 valence (F(2, 272) = 17.15, p < .001, η2 = .103): a statistically significant difference in the relationship between IUS and eLPP amplitude was found between blocks regardless of S2 valence, while the main effect of S2 valence replicates the results of the permutation analysis suggesting a larger eLPP for emotional as compared to neutral S2s. Nevertheless, the slope analysis revealed that the slopes of the relationship between IUS score and eLPP amplitude were not statistically different from 0 in any of the blocks. Post-hoc contrasts showed significant differences between the slopes in the 75% vs 100% block (t(272) = -2.92, p = .011), with more negative slope values for the former. Furthermore, a significant main effect of S2 valence (F(2, 272) = 23.15, p < .001, η2 = .133), better explained by the significant IUS × S2 valence interaction (F(2, 272) = 8.77, p < .001, η2 = .05, Fig 4D) was found for the early r-OFC activity: a statistically significant difference in the relationship between IUS and early r-OFC activity was found between S2 valence levels regardless of blocks. Nevertheless, the slope analysis revealed that the slopes of the relationship between IUS score and early r-OFC activity were not statistically different from 0 in any of the S2 valence levels. Post-hoc contrasts showed significant differences between the slopes in the POS vs. NEU and NEG vs. NEU S2 valence levels (t(272) = -3.89 and -3.29, p < .011 and = .003, respectively), with more negative slope values for the emotional as compared to the neutral S2 valence levels. Lastly, a significant main effect of S2 valence (F(2, 272) = 4.74, p = .01, η2 = .031) was found for r-temporal pole activity, replicating the larger activation to emotional stimuli already shown by LPP source reconstruction. No other significant main effects or interactions were found for eLPP, and early r-OFC and r-temporal pole activity (fitted models are summarized in Tables 3 and S6).

In the third time window (600–800 ms), analysis of the late r-OFC activity showed a significant main effect of S2 valence (F(2, 272) = 19.12, p < .001, η2 = .113), better explained by a significant IUS × S2 valence interaction (F(2, 272) = 6.45, p = .002, η2 = .038, Fig 4E): a statistically significant difference in the relationship between IUS and late r-OFC activity was found between S2 valence levels regardless of blocks. Nevertheless, the slope analysis revealed that the slopes of the relationship between IUS score and late r-OFC activity were not statistically different from 0 in any of the S2 valence levels. Post-hoc contrasts showed significant differences between the slopes in the POS vs. NEU and NEG vs. NEU S2 valence levels (t(272) = -3.14 and -3.08, p = .005 and .006, respectively), with more negative slope values for the emotional as compared to the neutral S2 valence levels. A significant main effect of S2 valence was found for lLPP (F(2, 272) = 13.11, p < .001, η2 = .084) and r-temporal pole activity (F(2, 272) = 9.47, p < .001, η2 = .059), replicating results already shown by permutation analyses and corresponding source reconstruction. No other significant main effects or interactions were found for lLPP, and late r-OFC and r-temporal pole activity (fitted models are summarized in Tables 3 and S7).

Discussion

To our knowledge, this study represents the first attempt to investigate the modulating role of IU on the neural activity elicited during the generation, implementation, and updating stages of emotional predictions, as a function of contexts with fully valid (100%), moderately valid (75%), or random (50%) predictive values.

Independent of IU, both contextual uncertainty and emotional valence exerted a detectable influence on the neural correlates of emotional predictions, especially during the generation and updating stages. Though a detailed discussion of these effects goes beyond the aims of this study, a brief summary of the main findings is still needed. During prediction generation, S1 valence modulated the N170 towards a processing advantage for emotional faces, with slight valence-specific differences among predictive contexts. This effect was subtended by a higher cortical activity in the r-STS, an area involved in the perceptual coding of changeable facial properties, among which emotional facial expression [66]. The N170/r-STS modulation is coherent with the extraction of emotional and contextual information from the environment, involving domain-specific brain circuits, and assumed to be crucial to continuously generate predictions [15,16]. During prediction updating, a strong emotional modulation emerged on the P2 and LPP components across all blocks: larger LPPs were found for emotional than neutral pictures, whereas the opposite was observed for P2s. The latter effect could index the infrequency of neutral S2s, and the consequent updating of the overall predictive model, subtended by a coherent modulation in the TPJ, an area involved in contextual updating [37,67]. The LPP modulation replicated a well-established effect in the literature, signaling a greater motivated attention towards emotional stimuli, subtended by domain-specific neural circuits (i.e., r-OFC, r-temporal pole) [38,39].

Concerning the main aim of this study, that is investigating the modulating role of IU on the neural activity elicited during the generation, implementation, and updating stages, the following results emerged. Contrary to what hypothesized, but in line with some previous evidence [19,22], individual differences in IU did not affect the prediction generation stage, either at the ERP or at the source level, suggesting that the extraction of emotional and contextual information conveyed by the faces develops in a similar manner along the different IU levels.

During the implementation stage, IU exerted a modulating role both on ERPs and sources activity, in partial consistency with our hypotheses, and with interesting differences as a function of time. In the early implementation stage, a higher IU was associated with a context- and valence-independent decrease of neural activity within l-ACC, an area assumed to subtend the behavioral and cognitive avoidance of exposure to uncertainty [6], and the continuous assessment of environmental uncertainty levels [68]. Net of the needed caution in interpreting brain sources activity as reconstructed from EEG, a reduced l-ACC activation to increasing IU levels could signal a disrupted environmental assessment of uncertainty. This could in turn translate into a heightened deployment of anticipatory resources, as reflected in larger eCNV amplitudes, in order to try to solve uncertainty [69]. In the late implementation stage, instead, higher IU levels were found to predict a context- and valence-independent decrease of neural activity within both l-ACC and l-SMA. So, as the implementation stage proceeds, the disrupted environmental uncertainty assessment, subtended by l-ACC decreased activity, seems to match an inhibition in action plans programming, as suggested by the reduced l-SMA activity, thus potentially contributing to the behavioral inhibition typical of high-IU individuals [4] (cfr. [2,3] for potential relationships with BIS activation). Furthermore, the heightened deployment of anticipatory resources, subtended by a larger CNV and reflecting an attempt to solve uncertainty, specialized into a context-specific effect: IU scores were found to negatively predict larger lCNVs only within the moderately predictive (75%) as compared to the fully predictive (100%) context. It can be suggested that, as IU increases more anticipatory resources are pre-allocated only in those contexts in which the amount of uncertainty is perceived as “solvable” (i.e., moderately predictive 75%), while in more certain contexts (i.e., fully predictive 100%) it could be metabolically effective to save resources, since there is no uncertainty to solve. This interpretation is consistent with IU models suggesting that neural, cognitive and behavioral peculiarities of high-IU individuals are aimed to increase perceived certainty and to reduce uncertainty exposure [4,10].

Within prediction updating, results were partially consistent with our hypotheses: higher IU levels were found to be associated both with valence-specific effects, such as smaller P2s to positive S2s and smaller r-OFC activity to emotional S2s, and with context-specific effects, such as smaller P2s and eLPPs to S2s in the moderately predictive (75%) context. Although slope analyses failed to find significant differences, and so these results should be considered with caution, some explanations could still be suggested. The valence-specific decrease of P2 amplitude to positive as compared to neutral S2s, reflecting a reduced automatic attention allocation, could represent an early neural correlate of the deficient safety learning which characterizes high-IU individuals [6]. This, in turn, could prevent emotional predictive models from being efficiently updated, by integrating environmental safety cues, because of the reduced attention allocation towards them. Furthermore, higher IU was found to be associated with a context-specific decrease in P2s and eLPPs when comparing the moderately predictive (75%) with the fully predictive (100%) contexts. This result, suggesting a dampening in S2 processing both at early and later stages, could reflect the uncertainty avoidance and a disrupted prediction error signaling typical of high-IU individuals [6]. Moreover, this result specifically arises in a context in which the updating of emotional predictive models is required (i.e., moderately predictive 75%), as compared to a context in which there is no need to update internal models since no expectancy violation occurs (i.e., fully predictive 100%). Lastly, both in the early and late stages of prediction updating, IU was found to be associated with a smaller r-OFC activity to emotional as compared to neutral S2s. Since OFC is involved in emotion regulation [70,71], its reduced activity could subtend a limited access to emotion regulation strategies, which is assumed to be a peculiar facet of high-IU traits [72], and which seems to spread across positive and negative valence and also across all predictive contexts.

Overall, our results suggest that IU differently modulated the neural activity developing across the three stages of emotional predictions construction. Prediction generation was not affected by individual differences in IU. During the implementation stage, instead, higher IU was associated with a disrupted environmental uncertainty assessment, and a consequent inhibition in anticipatory action plans programming, associated with a facilitation in the deployment of computational resources. These processes serve to solve uncertainty, specifically arising in moderately predictive contexts in which uncertainty is perceived as solvable. Finally, IU modulated the updating stage by predicting context-dependent uncertainty avoidance, and context-independent reduced attention to safety cues and disrupted access to emotion regulation strategies.

Implications of these results are promising. They demonstrate that individual differences in IU can actually affect how emotional predictions are constructed at a neural level. Given the conceptual relationships between IU, emotional traits and disorders (e.g., neuroticism, anxiety), and the activation of the associated behavioral patterns (e.g., BIS, fight-flight-freeze response), and being IU considered as the lower-order construct accounting for them [2,3], it can be hypothesized that such emotional traits might share similar neural correlates of the construction of emotional predictions. Thus, our results encourage broader considerations, useful for preventive and clinical applications within the domain of affective psychopathology. The CNV component and its neural sources, as well as the P2, the LPP and the r-OFC activity could in fact be considered as potential neural indicators to assess, or to intervene on (e.g., through biofeedback), when developing preventive and treatment measures for clinical and subclinical populations. Furthermore, our results suggest that, when assessing the relationships between IU and emotion, contextual uncertainty should be taken into more careful account, since its amount has proved to impact neural correlates of emotional predictions [26].

In conclusion, some limitations of our study must be addressed. First, the choice to screen out participants showing extreme levels of blood-injection-injury fear might have created a restricted range recruitment bias, and the sample was slightly sex-unbalanced (16 M vs. 20 F). Moreover, the final sample size was relatively small (N = 36), and it allowed to reach a power of about 0.5, lower than the desirable 0.8 level. Nevertheless, this issue was partially counterbalanced by the large number of trials we employed in our experimental paradigm, which enabled to heighten power by increasing the signal-to-noise ratio [73]. Second, a passive viewing task might have contributed to an overall decrease of ERP amplitudes, thus preventing some weaker modulations to reach statistical significance [74]. Furthermore, the lack of behavioral data makes it difficult to draw inferences on how the effects of IU on neural correlates of emotional predictions could translate into behavior and/or subjective affective experience. Lastly, the limited EEG spatial resolution implies caution when interpreting brain sources activations. We encourage future studies to overcome limitations by employing an active task and collecting subjective valence and expectancy ratings.

Net of these limitations, our paradigm still brought several elements of novelty to the literature. First, we used a totally uninstructed experimental task, in which participants were left free to implicitly construct their own predictive models as it occurs in everyday life. Second, IU was modeled as a continuous predictor, thus better approximating the construct, and preventing some slight individual differences to be lost in a dichotomization of the variable. Third, an intermediate level of contextual uncertainty (75%) was introduced to better resemble a more realistic, moderately certain environmental context. Fourth, highly arousing emotional stimuli were employed as S1s and S2s, including pleasant stimuli, thus allowing to isolate valence-specific effects. Fifth and last, high-density EEG allowed to investigate the neural correlates of emotional predictions both at the scalp and, with due caution, at the source level, thus conveying a comprehensive neurocomputational evidence.

Supporting information

S1 Fig. Modulation of ERPs and brain sources during prediction implementation stage.

Grand average ERP waveforms during prediction implementation in the 100% (continuous black line), 75% (continuous grey line), and 50% (dashed black line) blocks. Waveforms are plotted from a central cluster of electrodes (E40, E41, E42, E46, E47). Shaded areas denote standard error. CNV was computed between 1500 and 2500 msec from S1 onset (time 0). For visualization purposes, waveforms were low-pass re-filtered at 10 Hz.

https://doi.org/10.1371/journal.pone.0254045.s001

(TIF)

S1 Table. Descriptive statistics and ANOVA results on the final number of epochs accepted for each experimental condition.

Mean (M), standard deviation (SD) and F-test (F) of final epochs number for each level of the independent variables (block, S1/S2 valence).

https://doi.org/10.1371/journal.pone.0254045.s002

(DOCX)

S2 Table. Summary of the model fitted on the prediction generation stage.

Dependent variables: N170 and r-STS.

https://doi.org/10.1371/journal.pone.0254045.s003

(DOCX)

S3 Table. Summary of the model fitted on the early time window of the prediction implementation stage.

Dependent variables: eCNV and early l-ACC, l-SMA and l-dPCC.

https://doi.org/10.1371/journal.pone.0254045.s004

(DOCX)

S4 Table. Summary of the model fitted on the late time window of the prediction implementation stage.

Dependent variables: lCNV and late l-ACC, l-SMA and l-dPCC.

https://doi.org/10.1371/journal.pone.0254045.s005

(DOCX)

S5 Table. Summary of the model fitted on the first time window of the prediction updating stage.

Dependent variables: P2 and bilateral TPJ.

https://doi.org/10.1371/journal.pone.0254045.s006

(DOCX)

S6 Table. Summary of the model fitted on the second time window of the prediction updating stage.

Dependent variables: eLPP and early r-OFC and r-temporal pole.

https://doi.org/10.1371/journal.pone.0254045.s007

(DOCX)

S7 Table. Summary of the model fitted on the third time window of the prediction updating stage.

Dependent variables: lLPP and late r-OFC and r-temporal pole.

https://doi.org/10.1371/journal.pone.0254045.s008

(DOCX)

S1 File. Supplementary stimulus material and data analysis information.

Stimulus material: NimStim and IAPS picture numbers employed in the experimental paradigm. Data analysis: Planned planned paired-wise comparisons performed in the ERPs permutation analysis.

https://doi.org/10.1371/journal.pone.0254045.s009

(PDF)

Acknowledgments

We kindly thank Dr. Tania Moretta for her valuable help on data analysis.

References

  1. 1. Einstein DA. Extension of the Transdiagnostic Model to Focus on Intolerance of Uncertainty: A Review of the Literature and Implications for Treatment. Clin Psychol Sci Pract. 2014;21(3):280–300. pmid:25400336
  2. 2. Carleton RN. Fear of the unknown: One fear to rule them all? J Anxiety Disord. 2016;41:5–21. pmid:27067453
  3. 3. Carleton RN. Into the unknown: A review and synthesis of contemporary models involving uncertainty. J Anxiety Disord. 2016;39:30–43. pmid:26945765
  4. 4. Shihata S, McEvoy PM, Mullan BA, Carleton RN. Intolerance of uncertainty in emotional disorders: What uncertainties remain? J Anxiety Disord. 2016;41:115–24. pmid:27212227
  5. 5. Carleton RN, Norton MAPJ, Asmundson GJG. Fearing the unknown: A short version of the Intolerance of Uncertainty Scale. J Anxiety Disord. 2007;21(1):105–17. pmid:16647833
  6. 6. Grupe DW, Nitschke JB. Uncertainty and anticipation in anxiety: An integrated neurobiological and psychological perspective. Nat Rev Neurosci. 2013;14(7):488–501. pmid:23783199
  7. 7. Carleton RN. The intolerance of uncertainty construct in the context of anxiety disorders: theoretical and practical perspectives. Expert Rev Neurother. 1 agosto 2012;12(8):937–47. pmid:23002938
  8. 8. Hong RY, Cheung MW-L. The Structure of Cognitive Vulnerabilities to Depression and Anxiety: Evidence for a Common Core Etiologic Process Based on a Meta-Analytic Review. Clin Psychol Sci. 2015;3(6):892–912.
  9. 9. Mahoney AEJ, McEvoy PM. Trait Versus Situation-Specific Intolerance of Uncertainty in a Clinical Sample with Anxiety and Depressive Disorders. Cogn Behav Ther. 1 marzo 2012;41(1):26–39. pmid:22032804
  10. 10. Tanovic E, Gee DG, Joormann J. Intolerance of uncertainty: Neural and psychophysiological correlates of the perception of uncertainty as threatening. Clin Psychol Rev. 2018;60:87–99. pmid:29331446
  11. 11. Clark A. Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav Brain Sci. 2013;36(3):181–204. pmid:23663408
  12. 12. Friston K. The free-energy principle: A unified brain theory? Nat Rev Neurosci. 2010;11(2):127–38. pmid:20068583
  13. 13. Barrett LF. The theory of constructed emotion: an active inference account of interoception and categorization. Soc Cogn Affect Neurosci. 2017;12(1):1–23. pmid:27798257
  14. 14. Seth AK, Friston KJ. Active interoceptive inference and the emotional brain. Philos Trans R Soc B Biol Sci. 2016;371(1708):20160007. pmid:28080966
  15. 15. Bar M. The proactive brain: using analogies and associations to generate predictions. Trends Cogn Sci. 2007;11(7):280–9. pmid:17548232
  16. 16. Knill DC, Pouget A. The Bayesian brain: The role of uncertainty in neural coding and computation. Trends Neurosci. 2004;27(12):712–9. pmid:15541511
  17. 17. Nelson BD, Shankman SA. Does intolerance of uncertainty predict anticipatory startle responses to uncertain threat? Int J Psychophysiol. 2011;81(2):107–15. pmid:21619900
  18. 18. Tanovic E, Pruessner L, Joormann J. Attention and anticipation in response to varying levels of uncertain threat: An ERP study. Cogn Affect Behav Neurosci. 2018;18(6):1207–20. pmid:30112670
  19. 19. Chen JT-H, Lovibond PF. Intolerance of Uncertainty Is Associated With Increased Threat Appraisal and Negative Affect Under Ambiguity but Not Uncertainty. Behav Ther. 2016;47(1):42–53. pmid:26763496
  20. 20. Gole M, Schäfer A, Schienle A. Event-related potentials during exposure to aversion and its anticipation: The moderating effect of intolerance of uncertainty. Neurosci Lett. 2012;507(2):112–7. pmid:22172930
  21. 21. Grupe DW, Nitschke JB. Uncertainty is associated with biased expectancies and heightened responses to aversion. Emotion. 2011;11(2):413–24. pmid:21500909
  22. 22. Morriss J. What do I do now? Intolerance of uncertainty is associated with discrete patterns of anticipatory physiological responding to different contexts. Psychophysiology. 2019;56(9):e13396. pmid:31111965
  23. 23. Schienle A, Köchel A, Ebner F, Reishofer G, Schäfer A. Neural correlates of intolerance of uncertainty. Neurosci Lett. 2010;479(3):272–6. pmid:20570602
  24. 24. Morriss J, Biagi N, Dodd H. Your guess is as good as mine: A registered report assessing physiological markers of fear and anxiety to the unknown in individuals with varying levels of intolerance of uncertainty. Int J Psychophysiol. 2020;156:93–104. pmid:32763250
  25. 25. Freeston MH, Rhéaume J, Letarte H, Dugas MJ, Ladouceur R. Why do people worry? Personal Individ Differ. 1994;17(6):791–802.
  26. 26. Del Popolo Cristaldi F, Mento G, Buodo G, Sarlo M. What’s next? Neural correlates of emotional predictions: A high-density EEG investigation. Brain Cogn. 2021;150:105708. pmid:33714004
  27. 27. Bottesi G, Noventa S, Freeston MH, Ghisi M. Seeking certainty about Intolerance of Uncertainty: Addressing old and new issues through the Intolerance of Uncertainty Scale-Revised. PLoS ONE. 2019;14(2):e0211929. pmid:30742694
  28. 28. Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med. 2006;25(1):127–41. pmid:16217841
  29. 29. Michel CM, Murray MM. Towards the utilization of EEG as a brain imaging tool. NeuroImage. 2012;61(2):371–85. pmid:22227136
  30. 30. Blau VC, Maurer U, Tottenham N, McCandliss BD. The face-specific N170 component is modulated by emotional facial expression. Behav Brain Funct. 2007;3(1):7. pmid:17244356
  31. 31. Bentin S, Allison T, Puce A, Perez E, McCarthy G. Electrophysiological studies of face perception in humans. J Cogn Neurosci. 1996;8(6):551–65. pmid:20740065
  32. 32. Robinson JE, Breakspear M, Young AW, Johnston PJ. Dose-dependent modulation of the visually evoked N1/N170 by perceptual surprise: a clear demonstration of prediction-error signalling. Eur J Neurosci. 2018. pmid:29602233
  33. 33. Walter WG, Cooper R, Aldridge VJ, McCallum WC, Winter AL. Contingent negative variation: an electric sign of sensori-motor association and expectancy in the human brain. Nature. 1964;203(4943):380–4.
  34. 34. Chennu S, Noreika V, Gueorguiev D, Blenkmann A, Kochen S, Ibáñez A, et al. Expectation and attention in hierarchical auditory prediction. J Neurosci. 2013;33(27):11194–205. pmid:23825422
  35. 35. Mento G. The passive CNV: carving out the contribution of task-related processes to expectancy. Front Hum Neurosci. 2013;7:827. pmid:24376409
  36. 36. Mento G, Tarantino V, Vallesi A, Bisiacchi PS. Spatiotemporal neurodynamics underlying internally and externally driven temporal prediction: A high spatial resolution ERP study. J Cogn Neurosci. 2015;27(3):425–39. pmid:25203276
  37. 37. Gómez CM, Arjona A, Donnarumma F, Maisto D, Rodríguez-Martínez EI, Pezzulo G. Tracking the time course of Bayesian inference with event-related potentials: A study using the central Cue Posner Paradigm. Front Psychol. 2019;10:1424. pmid:31275215
  38. 38. Hajcak G, MacNamara A, Olvet DM. Event-related potentials, emotion, and emotion regulation: an integrative review. Dev Neuropsychol. 2010;35(2):129–55. pmid:20390599
  39. 39. Schupp HT, Cuthbert BN, Bradley MM, Cacioppo JT, Ito T, Lang PJ. Affective picture processing: The late positive potential is modulated by motivational relevance. Psychophysiology. 2000;37(2):257–61. pmid:10731776
  40. 40. Olofsson JK, Nordin S, Sequeira H, Polich J. Affective picture processing: An integrative review of ERP findings. Biol Psychol. 2008;77(3):247–65. pmid:18164800
  41. 41. Kimura M, Takeda Y. Automatic prediction regarding the next state of a visual object: Electrophysiological indicators of prediction match and mismatch. Brain Res. 2015;1626:31–44. pmid:25598206
  42. 42. Oldfield RC. The Assessment and Analysis of Handedness: the Edinburgh Inventory. Vol. 9, Neuropsychologia. Pergamon Press; 1971. pmid:5146491
  43. 43. Wolpe J, Lang PJ. A fear survey schedule for use in behavioral modification. Behav Res Ther. 1964;2:27–30. pmid:14170305
  44. 44. Bertolotti G, Zotti AM, Michielin P, Vidotto G, Sanavio E. A computerized approach to cognitive behavioural assessment: An introduction to CBA-2.0 primary scales. J Behav Ther Exp Psychiatry. 1990;21(1):21–7. pmid:2197296
  45. 45. Tottenham N, Tanaka JW, Leon AC, McCarry T, Nurse M, Hare TA, et al. The NimStim set of facial expressions: Judgments from untrained research participants. Psychiatry Res. 15 agosto 2009;168(3):242–9. pmid:19564050
  46. 46. Lang PJ, Bradley MM, Cuthbert BN. International affective picture system (IAPS): affective ratings of pictures and instruction manual. Tech Rep -8 Univ Fla Gainesv FL. 2008.
  47. 47. Schneider W, Eschman A, Zuccolotto A. E-prime. Pittsburgh, PA: Psychology Software Tools; 2010.
    • 48. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods. 15 marzo 2004;134(1):9–21. pmid:15102499
    • 49. Ferree TC. Spherical splines and average referencing in scalp electroencephalography. Brain Topogr. 2006;19(1–2):43–52. pmid:17019635
    • 50. Perrin F, Pernier J, Bertrand O, Echallier JF. Spherical splines for scalp potential and current density mapping. Electroencephalogr Clin Neurophysiol. 1989;72(2):184–7. pmid:2464490
    • 51. Bell AJ, Sejnowski TJ. An information-maximization approach to blind separation and blind deconvolution. Neural Comput. 1995;7(6):1129–59. pmid:7584893
    • 52. Stone J V. Independent component analysis: an introduction. Trends Cogn Sci. 1 febbraio 2002;6(2):59–64. pmid:15866182
    • 53. Kotowski K, Stapor K, Leski J. Improved robust weighted averaging for event-related potentials in EEG. Biocybern Biomed Eng. 2019;39(4):1036–46.
    • 54. Leski JM. Robust weighted averaging. IEEE Trans Biomed Eng. 2002;49(8):796–804. pmid:12148818
    • 55. Tadel F, Baillet S, Mosher JC, Pantazis D, Leahy RM. Brainstorm: A User-Friendly Application for MEG/EEG Analysis. Comput Intell Neurosci. 2011;2011:13. pmid:21584256
    • 56. Oostenveld R, Fries P, Maris E, Schoffelen J-M. FieldTrip: Open Source Software for Advanced Analysis of MEG, EEG, and Invasive Electrophysiological Data. Comput Intell Neurosci. 2011;2011.
    • 57. Groppe DM, Urbach TP, Kutas M. Mass univariate analysis of event-related brain potentials/fields I: A critical tutorial review. Psychophysiology. 2011;48(12):1711–25. pmid:21895683
    • 58. Fields EC, Kuperberg GR. Having your cake and eating it too: Flexibility and power with mass univariate statistics for ERP data. Psychophysiology. 2020;57(2):e13468. pmid:31456213
    • 59. Luck SJ, Gaspelin N. How to get statistically significant effects in any ERP experiment (and why you shouldn’t). Psychophysiology. 2017;54(1):146–57. pmid:28000253
    • 60. Evans AC, Janke AL, Collins DL, Baillet S. Brain templates and atlases. NeuroImage. 2012;62(2):911–22. pmid:22248580
    • 61. Gramfort A, Papadopoulo T, Olivi E, Clerc M. OpenMEEG: opensource software for quasistatic bioelectromagnetics. Biomed Eng Online. 2010;9(1):45. pmid:20819204
    • 62. Samuelsson JG, Peled N, Mamashli F, Ahveninen J, Hämäläinen MS. Spatial fidelity of MEG/EEG source estimates: A general evaluation approach. NeuroImage. 1 gennaio 2021;224:117430. pmid:33038537
    • 63. Destrieux C, Fischl B, Dale A, Halgren E. Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. NeuroImage. 2010;53(1):1–15. pmid:20547229
    • 64. Bates D, Mächler M, Bolker BM, Walker SC. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67(1).
    • 65. Lenth R V. Estimated Marginal Means, aka Least-Squares Means. 2020.
    • 66. Haxby J V., Hoffman EA, Gobbini MI. The distributed human neural system for face perception. Trends Cogn Sci. 2000;4(6):223–33. pmid:10827445
    • 67. Geng JJ, Vossel S. Re-evaluating the role of TPJ in attentional control: Contextual updating? Neurosci Biobehav Rev. 2013;37(10):2608–20. pmid:23999082
    • 68. Peters A, McEwen BS, Friston K. Uncertainty and stress: Why it causes diseases and how it is mastered by the brain. Vol. 156, Progress in Neurobiology. Elsevier Ltd; 2017. pag. 164–88. pmid:28576664
    • 69. Paulus MP, Stein MB. An Insular View of Anxiety. Biol Psychiatry. 2006;60(4):383–7. pmid:16780813
    • 70. Golkar A, Lonsdorf TB, Olsson A, Lindstrom KM, Berrebi J, Fransson P, et al. Distinct Contributions of the Dorsolateral Prefrontal and Orbitofrontal Cortex during Emotion Regulation. PLoS ONE. 2012;7(11):e48107. pmid:23144849
    • 71. Goldin PR, McRae K, Ramel W, Gross JJ. The Neural Bases of Emotion Regulation: Reappraisal and Suppression of Negative Emotion. Biol Psychiatry. 2008;63(6):577–86. pmid:17888411
    • 72. Ouellet C, Langlois F, Provencher MD, Gosselin P. Intolerance of uncertainty and difficulties in emotion regulation: Proposal for an integrative model of generalized anxiety disorder. Eur Rev Appl Psychol. 2019;69(1):9–18.
    • 73. Baker DH, Vilidaite G, Lygo FA, Smith AK, Flack TR, Gouws AD, et al. Power contours: Optimising sample size and precision in experimental psychology and human neuroscience. Psychol Methods [Internet]. 2020 [citato 11 maggio 2021]; Disponibile su: https://psycnet.apa.org/fulltext/2020-52357-001.pdf. pmid:32673043
    • 74. Gaillard AWK, Perdok J. Slow brain potentials in the CNV-paradigm. Acta Psychol (Amst). 1980;44(2):147–63. pmid:7368953