Clarifying the neural substrates of threat and safety reversal learning in humans

NeuroImage 207 (2020) 116427

Contents lists available at ScienceDirect

NeuroImage journal homepage: www.elsevier.com/locate/neuroimage

Clarifying the neural substrates of threat and safety reversal learning in humans Hannah S. Savage a, *, Christopher G. Davey b, c, Miquel A. Fullana d, e, Ben J. Harrison a, ** a

Melbourne Neuropsychiatry Centre, Department of Psychiatry, The University of Melbourne & Melbourne Health, Victoria, Australia Orygen, The National Centre of Excellence in Youth Mental Health, Parkville, Australia Centre for Youth Mental Health, The University of Melbourne, Parkville, Australia d Adult Psychiatry and Psychology Department, Institute of Neurosciences, Hospital Clinic, Barcelona, Spain e Institut d’Investigacions Biom
ediques August Pi i Sunyer (IDIBAPS), Centro de Investigacion Biomedica en Red de Salud Mental (CIBERSAM), Barcelona, Spain b c

A R T I C L E I N F O

A B S T R A C T

Keywords: Threat Safety Reversal learning Fear conditioning Ventromedial prefrontal cortex Anterior cingulate cortex

Responding flexibly to sources of threat and safety is critical to the adaptive regulation of emotions, including fear. At a neural systems level, such flexibility is thought to rely on an extended neural circuitry involving the dorsal anterior cingulate cortex (dACC) and ventromedial prefrontal cortices (vmPFC), although precisely how this occurs remains unclear. Using a novel fear reversal task and functional magnetic resonance imaging (fMRI), we examined the neural correlates of threat and safety reversal learning and their associations with individual differences in anxious responding in a large sample of healthy adolescents and young adults. Overall, participants demonstrated successful threat and safety reversal learning, as indexed by subjective ratings. At a whole-brain level, threat reversal was associated with significant activation of the bilateral anterior insular cortex and dACC, in particular its rostral subregion. Conversely, safety reversal led to significant activation of the anterior vmPFC, together with posterior mid-line regions. Further analyses of regional responses suggested a more selective role for the rostral dACC in threat signal updating, as well as a direct association of its activity with participants’ change in subjective anxious arousal to the reversed threat. Taken together, our findings complement existing neurocircuitry models of human fear regulation, particularly regarding the importance of midline cortical regions, and provide further insights into their specific contribution to flexible threat-safety signal processing. In particular, our results suggest that rostral dACC function may be more centrally involved in regulating levels of anxious arousal when flexibility is required. They also raise important questions regarding the vmPFC’s role in safety learning, particularly involving its hypothesized subregional contributions to response inhibitory versus stimulus value processing functions.

1. Introduction Responding flexibly to sources of threat and safety is critical to adaptive emotional behavior and well-being. People with anxiety disorders, for instance, appear to show basic deficits in the flexible processing of learned threat and safety signals, such that emotional responses to threats become entrenched and often generalized to ‘safe’ stimuli or situations (Duits et al., 2015; Grupe and Nitschke, 2013; Jovanovic et al., 2012; Lissek, 2012; Lissek et al., 2013). Neurobiologically, the capacity

to discriminate threat and safety signals, and accordingly to regulate fearful responses, appears to rely on the function of extended neural circuitry of medial prefrontal-subcortical brain regions, with key cortical areas including the dorsal anterior cingulate (dACC) and ventromedial prefrontal cortices (vmPFC; Fullana et al., 2015). In particular, evidence from rodent and human studies implicates dACC in the specific processing of threats, whereas the vmPFC appears more responsive to safety (Gilmartin and McEchron, 2005; Milad and Quirk, 2012; Milad et al., 2007). Nevertheless, precisely how these brain regions are involved in

* Corresponding author. Melbourne Neuropsychiatry Centre, Department of Psychiatry, The University of Melbourne, Level 3, 161 Barry St, Carlton, VIC, 3053, Australia. ** Corresponding author. Melbourne Neuropsychiatry Centre, Department of Psychiatry, The University of Melbourne, Level 3, 161 Barry St. Carlton, VIC, 3053, Australia. E-mail addresses: [email protected] (H.S. Savage), [email protected] (B.J. Harrison). https://doi.org/10.1016/j.neuroimage.2019.116427 Received 30 July 2019; Received in revised form 28 October 2019; Accepted 1 December 2019 Available online 1 December 2019 1053-8119/© 2019 Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

H.S. Savage et al.

NeuroImage 207 (2020) 116427

2. Materials and methods

threat and safety signal processing (Wallis et al., 2017), especially in situations when the learned value or meaning of these signals changes such as during reversal learning, remains to be understood. One established experimental test of flexible threat and safety signal processing is Pavlovian fear reversal, or reversal discrimination, which models the extent to which one is able to simultaneously update and inhibit fear responses when threat-safety outcome contingencies change. In practice, this reversal learning involves the conditioning of threat and safety signals during an initial training phase, followed by the switching of these contingencies in a reversal phase, whereby the former threat now signals safety, and vice versa. In studies of threat-safety reversal with brain functional magnetic resonance imaging (fMRI), the vmPFC has been implicated as having a selective role in safety signal updating. Most notably, Schiller et al. (2008) reported that the vmPFC’s response to a reversed safety signal was more prominent than during initial threat conditioning, suggesting that its activity may represent the increased value of a safety signal that was formerly a threat, as opposed to when it was a ‘naïve’ safety signal (see also Apergis-Schoute et al., 2017). An important issue that remains to be clarified from these studies is the extent to which vmPFC activity also directly discriminates a reversed safety signal from its status as a former threat, thus indexing both the value updating and response inhibitory processes that are thought to underlie successful reversal learning (Zhang et al., 2015). In a study of appetitive reversal learning, it was suggested that such value updating and inhibitory processes might be anatomically dissociable at the level of vmPFC function, with value updating being linked to the anterior (polar) vmPFC, and response inhibition to the posteriorsubgenual vmPFC (Zhang et al., 2015). However, these observations are somewhat at odds with Schiller et al. (2008) and Apergis-Schoute et al. (2017), who mapped safety signal updating to posterior-subgenual vmPFC activity. Similarly, while there is considerable evidence from classical fear conditioning literature linking the dACC to the processing of conditioned threats (Fullana et al., 2015), there are few studies to have directly examined its role in flexible threat processing, including reversal learning. These comparisons are likely to be informative, especially given recent evidence that suggests the dACC may have a more direct role in threat processing and fear regulation (Wallis et al., 2017), than the more generally held idea that it facilitates the expression of learned fear (Corcoran and Quirk, 2007; Laurent and Westbrook, 2009; Milad et al., 2007; Sierra-Mercado et al., 2011). Specifically, in Wallis et al. (2017), it was shown that inactivation of the marmoset area 32 (homologous to the human dACC) led to increased behavioral correlates of negative emotion and generalized fear responding. In the current study, we therefore sought to clarify the neural correlates of threat and safety reversal learning at a whole-brain level. We used a novel task, optimizing that used previously by Harrison et al. (2015) and Schiller et al. (2008), designed to map and identify specific learning-related changes in vmPFC and dACC activity. Regarding safety reversal learning, we anticipated that there would be broad involvement of anterior and posterior vmPFC subregions in tracking the reversal of a conditioned threat to safety signal, consistent with their respective hypothesized roles in value updating and response inhibition (Zhang et al., 2015). Regarding threat reversal, we predicted involvement of the dACC akin to past fear conditioning studies (Fullana et al., 2015), but also that it might respond distinctively to the reversed threat signal, consistent with Wallis and colleagues’ recent findings in a marmoset model, and their ‘fear regulation’ hypothesis (Wallis et al., 2017). We also sought to examine potential associations between the neural correlates of threat and safety reversal learning with autonomic and subjective markers of anxious arousal. We recruited a large cohort of adolescents and young adults to achieve our aim, and additionally examined the influence of age on the neural correlates of reversal learning. This is particularly relevant given evidence of neurodevelopmental influences on fear learning neural circuitry in human and animal studies (Casey et al., 2015; Zimmerman et al., 2019).

2.1. Participants One hundred and ten participants were recruited to the study. All participants met the following eligibility criteria: (i) they were aged between 16 and 25 years; (ii) had no current or past diagnosis of mental illness (iii) were competent English speakers, (iv) were not taking any psychoactive medication, and (v) had no contraindications to MRI, including pregnancy. All participants underwent the SCID-5 non-patient interview (First et al., 2015) conducted by experienced (psychology-graduate level) research assistants to exclude any mental health diagnoses. All participants had normal or corrected-to-normal vision and provided written informed consent, following a complete description of the study protocol, which was approved by The University of Melbourne Human Research Ethics Committee. Of the initial participant sample, 7 participants were excluded due to excessive head motion during scanning (see 2.5.2 Image preprocessing), 3 participants due to meeting diagnostic criteria for an anxiety disorder; and 6 participants due to technical difficulties during data acquisition. The final sample consisted of 94 participants (63 female) with a mean age of 21.31 years (S.D. 2.28 years). All participants completed the State-Trait Anxiety Inventory-Trait version (STAI-T; mean score 32.49
8.47, range 20–57; general Australian adult population: mean score 36.44
10.93, see Crawford et al., 2011) to provide a measure of general trait anxiety (Spielberger, 1983). 2.1.1. Experimental design Participants completed a differential threat-safety reversal task that was modelled closely on Schiller et al. (2008) and Harrison et al. (2017). Specifically, the reversal task used by Schiller et al. (2008) was modified to incorporate the stimuli and subjective ratings from the differential conditioning task used by Harrison et al. (2017); see Supplementary Fig. 1 in the Supplementary Material. A blue and a yellow sphere, presented for 2 s against a black background, were used as the conditioned stimuli (CS). The unconditioned stimulus (US) was an aversive auditory (white noise) burst (50 ms) presented at ~90 dB that co-terminated with the CSþ. The task had three phases (baseline, conditioning and reversal) that were acquired in a single experimental run. During baseline, each colored sphere was presented 5 times and the US did not occur. We used the baseline phase as a CS familiarization period, but also to provide an independent perceptually-matched reference state for mapping brain regional responses to the CS during conditioning and reversal (see Harrison et al., 2017 for detailed discussion). During conditioning, the US co-terminated with one of the CS (forming a CSþ) and not with the other (forming a CS-). The color of the CSþ was counterbalanced across subjects and the CS-US pairing occurred one third of the time, enabling the classification of CSþ unpaired trials and the subsequent analysis of CSþ responses without US confounding. During reversal, the pairing of the US and CS was switched (un-signaled), such that the conditioning phase CSþ became the ‘new CS-’, and the conditioning phase CS- became the ‘new CSþ’ (with US pairing). 10 presentations of the CSþ unpaired, 5 of the CSþ paired and 10 presentations of the CS- occurred during both conditioning and reversal with no more than two consecutive trials of the same stimuli. The first reinforced CSþ during conditioning was the second CSþ presentation, and upon reversal the first presentation of the new CSþ was reinforced. Across all phases, the inter-stimulus interval (ISI) between CS trials was 12s during which a white visual fixation cross was presented. At the conclusion of each phase, and as a natural progression of the task, participants were asked to rate each CS on five point Likert scales (Self-Assessment Manikins, SAM; Bradley and Lang, 1994) of valence and anxious arousal. To measure valence, participants responded to the question: “How unpleasant/pleasant did you find the [blue or yellow] sphere?“, with responses ranging from 1 ¼ ‘very unpleasant’ to 5 ¼ ‘very pleasant’. For anxious arousal, participants responded to the question: “How anxious did the [blue or yellow] sphere make you feel?“, with 2

H.S. Savage et al.

NeuroImage 207 (2020) 116427

responses ranging from 1 ¼ ‘not anxious’ to 5 ¼ ‘very anxious’. At the conclusion of the task participants were asked to rate how unpleasant they found the noise stimulus on a scale from 1 to 10; 1 ¼ ‘not unpleasant’ to ‘10 ¼ very unpleasant’ (mean rating 5.78
2.56; range 1–10). Subjective ratings were unavailable for one participant. Participants were then asked the following multiple-choice question to evaluate contingency awareness: “In part 2/3, did the noise stimulus usually occur in association with: a) the blue sphere, b) the yellow sphere, c) randomly, or d) you don’t know? (85.9% of participants were aware of the contingency change). Contingency awareness was unavailable for two participants. The task was programmed in Presentation (Neurobehavioral Systems, Inc.) and was delivered using MRI-compatible high-resolution goggles and headphones (VisuaStim Digital, Resonance Technology Inc.). Noise US were delivered via Resonance Technology EarBud headset, which also provided passive noise cancellation (~30 dB). Participants’ responses were registered with a fORP curved 4-button response box (Cambridge Research Systems Ltd.), which participants were familiarized with prior to scanning.

screening methods see Supplementary Information), in addition to those participants whose fMRI data was excluded, as noted above. For the SCR subgroup (n ¼ 50), participants’ filtered time-series were imported to the Psycho-Physiological modelling toolbox (PsPM; Bach and Friston, 2013) run in MATLAB R 2017b (The MathWorks, Inc.). All task trial onsets (baseline CSþ, baseline CS-, CSþ, CSþ Reinforced, CS-, new CSþ, new CSþ Reinforced, new CS-) were specified and convolved with a canonical skin conductance response function (Bach et al., 2009, 2013). Participants’ absolute modelled peak amplitude SCRs were exported to SPSS (v.24) for further analysis. As with the subjective ratings (see 2.3 Subjective ratings), we estimated differential threat and safety reversal scores for SCRs. 2.3.1. Image acquisition A 3T General Electric Discovery MR750 system equipped with an eight-channel phased-array head coil was used in combination with ASSET parallel imaging. The functional sequence consisted of a singleshot gradient-recalled EPI sequence in the steady state (repetition time, 2 s; echo time, 35 ms; and pulse angle, 90 ) in a 23 cm field-of-view, with a 64 x 64-pixel matrix and a slice thickness of 3.5 mm (no gap). Thirty-six interleaved slices were acquired parallel to the anterior–posterior commissure line with a 20 anterior tilt to better cover ventral prefrontal cortical brain regions. The total sequence time was 16 min and 10 s, corresponding to 481 whole-brain EPI volumes. A T1-weighted highresolution anatomical image (3D BRAVO) was acquired for each participant to assist with functional time series coregistration (140 contiguous slices; repetition time ¼ 7.9 s, echo time ¼ 3 s, flip angle ¼ 13 ; in a 25.6cm field of view, with a 256 x 256–pixel matrix and a slice thickness of 1 mm). To assist with head immobility, foam-padding inserts were placed around the participants’ head inside the coil.

2.1.2. Task training Prior to entering the scanner, participants were given brief instructions on the format and goals of the task. They were informed there were three parts to the task, referred to as ‘Part 1, Part 2 and Part 3’. They were told that during the task they would see blue and yellow spheres presented in different orders, and that during Part 1 (baseline) no noise stimuli would occur. For Parts 2 and 3 (conditioning and reversal), they were told that their job was to try to understand the relationship between the spheres and the noise stimulus. During training, they were exposed to one instance of the noise stimulus to mitigate novelty effects. They were also instructed on how to complete the SAM ratings of anxious arousal and valence when in the scanner. Immediately prior to commencing the scan, participants were reminded of the general task instructions.

2.3.2. Image preprocessing Imaging data was transferred to a Unix-based platform that ran MATLAB Version 9.3 (The MathWorks Inc., Natick, USA) and Statistical Parametric Mapping (SPM) Version 12 (Wellcome Trust Centre for Neuroimaging, UK). Motion correction was performed by aligning each participant’s time series to the first image using least-squares minimization and a six-parameter rigid-body spatial transformation. The SPM motion fingerprint toolbox (Wilke, 2012) was then used to quantify scan-to-scan head motion on the basis of the SPM motion parameters. Participants were excluded if movement exceeded 3 mm (~1 native voxel) mean total scan-to-scan displacement. These realigned functional images were co-registered to each participant’s respective T1 anatomical scans, which were segmented and spatially normalized to the International Consortium for Brain Mapping template using the unified segmentation approach. The functional images were interpolated to 2 mm isotropic resolution, and smoothed with a 5-mm full-width-at-halfmaximum (FWHM) gaussian filter.

2.2. Subjective ratings Subjective ratings were collected as above (see 2.2.1 Experimental design). Two-factor repeated measures ANOVAs of subjective ratings, as well as planned paired t-tests were performed to identify whether significant differential learning had occurred within and across experimental phases. To confirm the direction of updating in anxious arousal and valence ratings, differential scores of threat and safety reversal learning were calculated. The differential threat reversal score was calculated by subtracting participants’ ratings of the CS- from the new CSþ. Similarly, the differential safety reversal score was calculated by subtracting their ratings of the CSþ from the new CS-. Associations between subjective ratings, and trait anxiety scores were estimated by Pearson’s correlation in SPSS Statistics for Macintosh Version 24 (IBM Corp, USA).

2.3.3. Primary fMRI analysis Each participant’s preprocessed time-series was included in a firstlevel SPM general linear model (GLM) analysis, which specified the onsets of each CS event-type in each task phase to be convolved with canonical hemodynamic response function. The fixation-cross ISI periods throughout whole task, served as the implicit baseline. A high-pass filter (1/128 s) accounted for low-frequency noise, while temporal autocorrelations were estimated using a first-order autoregressive model. Primary contrast images were separately estimated for each CS response (CSþ, CSþ Reinforced, CS-, new CSþ, new CSþ Reinforced, new CS-) versus its corresponding perceptual (color-matched) baseline CS. The use of an independent baseline state (e.g., non-conditioned CS) allows one to better distinguish safety-signal activity (more selective to vmPFC) by removing the confounds of safety learning that may generalize to a fixation cross (for discussion see Harrison et al., 2017). The CSþ, CS-, new CSþ and new CS- contrast images were carried forward to the group-level using the summary statistics approach to random-effects analyses. We

2.3. Psychophysiology Skin conductance responses (SCR) were recorded using MRIcompatible finger electrodes (silver/silver chloride) fitted with conductance gel to the distal phalanges of the index and middle finger of participants’ non-dominant hand. Fingers were cleaned using alcohol wipes prior to the attachment of electrodes and participants were instructed to keep their hand as still as possible for the duration of the scan to decrease contamination of the trace by motion. The signal was amplified and sampled at 1000 Hz using PowerLab v8.0 (ADInstruments, Dunedin, NZ) and recording was scanner-triggered concurrently with the task presentation. Given the known challenges in recording high-quality SCRs during fMRI (Boucsein, 2012; Gray et al., 2009) we took a conservative approach to individual data screening, which resulted in only part of the sample being included in further analyses (for data pre-processing and 3

H.S. Savage et al.

NeuroImage 207 (2020) 116427

then applied a mask-corrected (small-volume) threshold of PFDR < 0.05, k ¼ 10 voxels to test significance.

specified a 2 x 2 factorial model that included phase (conditioning, reversal) and CS type (CSþ, CS-). We also specified a 2 x 2 factorial model, as above, with the CSþ, CS-, new CSþ and new CS- contrast images relative to the implicit baseline for consistency with other studies (see Supplementary Fig. 2). Our primary analyses concentrated on the differential reversal contrasts for threat- (new CSþ vs. CS-) and safety learning (new CS- vs CSþ). We also estimated differential conditioning (overall CSþ vs. CS-), as well as simple threat- (new CSþ vs. CSþ) and safety (new CS- vs. CS-) updating effects. All comparisons were thresholded with a whole-brain, false discovery rate (FDR) correction (PFDR < 0.05, cluster minimum (k) ¼ 10 contiguous voxels).

3. Results 3.1. Subjective ratings As expected, there were no significant differences at baseline in participants’ anxious arousal or valence ratings of the CS (arousal: t(92) ¼ 0.62, P ¼ 0.53; valence: t(92) ¼ 0.48, P ¼ 0.63). A two-factor repeated measures ANOVA of arousal ratings identified a significant main effect of ‘phase’ (F(1, 92) ¼ 5.38, P ¼ 0.023), ‘CS type’ (F(1, 92) ¼ 129.47, P < 0.001) and ‘phase x CS type’ interaction (F(1, 92) ¼ 5.04, P ¼ 0.03). Subsequent t-tests indicated significant differences during both conditioning (t(92) ¼ 10.12, P < 0.001) and reversal (t(92) ¼ 8.52, P < 0.001), confirming the differential aversiveness of the CSþ versus CS- (Fig. 1A). A two-factor repeated measures ANOVA of valence ratings identified a significant main effect of ‘CS type’ (F(1, 92) ¼ 103.43, P < 0.001) and ‘phase x CS type’ interaction (F(1, 92) ¼ 17.83, P < 0.001), but no main effect of ‘phase’ (F(1, 92) ¼ 2.62, P ¼ 0.11). Follow up t-tests indicated significant differences during both conditioning (t(92) ¼ 10.04, P <

2.3.4. Secondary fMRI analyses Associations between brain-level threat and safety reversal learning effects and corresponding subjective and autonomic indices, as well as trait anxiety scores (STAI-T), were examined by adding them as covariates of interest in separate one-sample t-test group models within SPM12. We also separately examined associations with participants’ age (years at time of scanning). All analyses were first inclusively masked to sample only those regions that were positively responsive to the threat and safety reversal contrasts, respectively (P < 0.05 uncorrected). We

Fig. 1. Subjective ratings of anxious arousal and valence during threat and safety reversal learning. Post conditioning and reversal phase ratings of the CS. For anxious arousal ratings (A), scores range from 1 ¼ ‘not anxious’ to 5 ¼ ‘very anxious’. (B) Differential threat and safety reversal learning scores for arousal. For valence ratings (C), scores range from 1 ¼ ‘very unpleasant’ to 5 ¼ ‘very pleasant’. (D) Differential threat and safety reversal learning scores for valence. CS ¼ conditioned stimuli. n ¼ 93. Bars show mean
SEM. *** ¼ P < 0.001. 4

H.S. Savage et al.

NeuroImage 207 (2020) 116427

0.001) and reversal (t(92) ¼ 7.99 P < 0.001), confirming the differential unpleasantness/pleasantness of the CSþ versus CS- within each phase (Fig. 1C). A significant difference was found between the differential threat and safety reversal scores for arousal (t(92) ¼ 11.37, P < 0.001; Fig. 1B) and valence (t(92) ¼ 10.17, P < 0.001; Fig. 1D), reflecting overall correct updating of these indices when the contingencies reversed. Furthermore, these change scores were also significantly correlated (arousal r ¼ 0.45, P < 0.001, valence; r ¼ -.52, P < 0.001), such that the more a participant updated to the new threat signal, they also updated to the new safety signal for each measure (Fig. 1B and D).

As illustrated in Fig. 4, safety reversal learning (new CS- vs. CSþ) was associated with significant activation of the anterior vmPFC (extending from medial frontopolar cortex to medial orbitofrontal cortex); ventral posterior cingulate extending to retrosplenial cortex; and the bilateral intraparietal lobule (angular gyrus). An additional area of activation was observed in the left primary somatosensory cortex. Fig. 4B presents regional signal changes estimated for the vmPFC and posterior cingulate cortex across the task phases in response to each CS. These results suggest that vmPFC responses were generally increased across conditioning and reversal, although its magnitude of deactivation to the threat signal was more apparent during conditioning than reversal. A similar pattern of responding was observed for the posterior cingulate cortex (Fig. 4B). Table 2 lists all regions of significant activation (and deactivation) during safety reversal learning. From the above analyses of both activation and deactivation responses to reversed threat and safety, certain regions appeared to equally track the reversal of both CS types. We performed two supplementary conjunction analyses to clarify the consistency of these effects, as follows: ‘new CSþ > former CS- and new CS- < former CSþ’; ‘new CS- > former CSþ and new CSþ < former CS-’. These analyses identified bilateral anterior insular cortex activity as significantly, and consistently increased in response to the reversed threat signal and decreased to the reversed safety signal (Supplementary Fig. 3 and Supplementary Table 1). To assist the comparison of our results with prior studies (Apergis-Schoute et al., 2017; Schiller et al., 2008) differential conditioning and simple threat and safety updating effects were examined. Results for differential threat and safety conditioning were highly consistent with past studies, including our own meta-analysis (Fullana et al., 2015) and are provided in Supplementary Fig. 4 (Supplementary Tables 2 and 3). No significant results were identified for simple threat and safety updating at our chosen threshold. Results for simple threat and safety updating at the lower threshold are provided in Supplementary Fig. 5 (Supplementary Tables 4 and 5).

3.2. SCR results Skin conductance responses (SCRs) were not significantly different between CS type (CSþ, CS-) at baseline (t(49) ¼ 0.47, P ¼ 0.64). A twofactor repeated measures ANOVA identified a significant main effect of ‘phase’ (F(1, 49) ¼ 4.28, P ¼ 0.04) and a ‘phase x CS type’ interaction (F(1, 49) ¼ 6.65, P ¼ 0.013), but no main effect of ‘CS type’ (F(1, 49) ¼ 1.55, P ¼ 0.22), suggesting that the differentiation between CS type was dependent on experimental phase. This was confirmed by follow up paired t-tests that indicated a significant difference between CS type during conditioning (t(49) ¼ 3.25, P ¼ 0.002) but not reversal (t(49) ¼ 0.88, P ¼ 0.37) reflecting overall, participants inability to reverse their responding when the contingency switched (see Fig. 2A). Differential threat and safety reversal scores were highly correlated but not significantly different (r ¼ 0.77, P < 0.001; t(49) ¼ 1.24, P ¼ 0.22; Fig. 2B) suggesting that the more a participant updated to the new threat signal, the less they were able to update to the new safety signal. 3.2.1. Primary fMRI results As illustrated in Fig. 3, threat reversal learning (new CSþ vs. CS-) was associated with significant activation of the bilateral anterior-to-mid insular cortex; rostral and caudal subregions of the dACC; as well as the left second somatosensory cortex (parietal operculum). Fig. 3B present regional signal changes estimated for the rostral dACC and anterior insular cortex regions across the task phases in response to each CS. These results suggest that responses in anterior-mid insular cortex were generally increased in response to threat across conditioning and reversal – a pattern not observed in the rostral region of the dACC. Table 1 lists all regions of significant activation (and deactivation) during threat reversal learning.

3.2.2. Secondary results: associations between anxious arousal, trait anxiety, and fMRI parameters Secondary analyses focused on potential associations between brain responses during threat and safety reversal and subjective, autonomic and trait measures of anxious arousal. No associations were identified between trait measures of anxious arousal and participants’ subjective ratings. At the brain level, differential anxious arousal ratings during threat

Fig. 2. Skin conductance responses (SCRs) during threat and safety reversal learning. (A) Peak amplitude (micro Siemens) of SCRs to each CS during conditioning and reversal phase. (B) Differential threat and safety reversal learning for SCRs. CS ¼ conditioned stimuli. n ¼ 50. Bars show mean
SEM.** ¼ P < 0.01. *** ¼ P < 0.001. 5

H.S. Savage et al.

NeuroImage 207 (2020) 116427

Fig. 3. Neural correlates of differential threat reversal learning. (A) Activation during differential threat reversal learning (new CSþ vs. CS-). Color bar is scaled as SPM T-scores. (B) Regional contrast estimates were extracted from cluster peaks as indicated by white circles in A. The magnitude of activation for each region was estimated against the corresponding perceptual baseline condition. CS ¼ conditioned stimuli. Bars show mean
SEM.

reversal learning were found to be significantly positively correlated with the magnitude of activation of the rostral and caudal dACC and bilateral anterior insular cortex (Fig. 5, Table 3). Participants’ differential threat reversal SCRs were also found to be significantly negatively correlated with the magnitude of left anterior insular cortex activation (Supplementary Fig. 6, Supplementary Table 6). In other words, participants who updated their SCR to the reversed threat signal demonstrated less corresponding activation within the left anterior insular region. No further significant associations were observed for subjective, autonomic or trait measures, nor did we observe any significant associations between participants’ age and the neural correlates of threat and safety reversal learning.

Table 1 Significant activation and deactivation associated with differential threat reversal learning (new CS þ vs. CS-). Anatomic Region

Activation: Anterior to mid-insular cortex Pregenual anterior cingulate cortex Second somatosensory cortex (parietal operculum) Dorsal midbrain (~substantia nigra) Rostral anterior cingulate cortex Caudal anterior cingulate cortex Dorsolateral prefrontal cortex Deactivation: Superior prefrontal cortex Primary motor cortex Inferior parietal cortex (angular gyrus) Ventral posterior cingulate cortex

Cluster Size (voxels)

Peak MNI Coordinates

Tmax

P-max

x

y

z

542 639 201

34 38 4

24 26 34

0 0 12

5.69 5.49 4.98

0.004 0.004 0.004

113 45

56 62

38 36

26 28

4.93 4.64

0.004 0.006

12

14

16

10

4.18

0.013

31

8

20

26

4.10

0.016

34

6

12

38

4.05

0.018

10

50

22

18

3.78

0.029

1796 1315 581 766 287 627

20 16 6 36 52 2

30 30 30 64 66 58

38 44 64 30 22 36

6.26 5.45 5.0 4.68 4.68 4.57

<0.001 0.001 0.002 0.004 0.004 0.005

4. Discussion While many previous studies have implicated an extended neural circuitry of medial prefrontal-subcortical regions in the learned discrimination of threat and safety signals, fewer have examined their involvement in flexible affective responding, such as during reversal learning. Our study is the first to clarify the neural substrates of threat and safety reversal learning at a whole-brain level, and to examine broader associations of this activity with individual differences in anxious arousal in a large sample of healthy adolescents and young adults. Whilst, overall, our results support the idea that the dACC is consistently involved in responding to learned threats, they also suggest it may contribute more dynamically to threat signal processing than what has been previously emphasised in fMRI studies (Corcoran and Quirk,

6

H.S. Savage et al.

NeuroImage 207 (2020) 116427

Fig. 4. Neural correlates of differential safety reversal learning. (A) Activation during differential safety reversal learning (new CS- vs. CSþ). Color bar is scaled as SPM T-scores. (B) Regional contrasts estimates were extracted from cluster peaks as indicated by white circles in A. The magnitude of activation for each region was estimated against the corresponding perceptual baseline condition. CS ¼ conditioned stimuli. PCC ¼ posterior cingulate cortex. Bars show mean
SEM.

2007; Laurent and Westbrook, 2009; Milad et al., 2007; Sierra-Mercado et al., 2011). Regarding the role of the vmPFC in safety reversal learning, our results seem to most strongly endorse a model that links vmPFC function, particularly its anterior subregion, to affective value-based computations in support of flexible emotional regulation. Recently, a ‘higher-order’ role for the dACC in facilitating threat appraisal and fear regulation was proposed. Specifically, in a marmoset model of fear conditioning and extinction, Wallis et al. (2017) showed that pharmacological inactivation of area 32 increased negative emotional responding (cardiovascular and behavioral correlates) and fear generalization, and impaired fear extinction. Potentially consistent with this, our study also provides evidence for the involvement of the dACC in the regulation of fear responses. We characterized a significant positive correlation between dACC activity and participants’ subjective anxiety changes during threat reversal, which is highly consistent with our recent work linking cognitive appraisals of bodily anxiety sensations during threat learning to dACC activity (Harrison et al., 2015). One interpretation is that the processing of the reversed threat signal requires increased appraisal efforts, due to the conflict between prior experiences and new learning, that influences the construction of the reported subjective experience (Finger et al., 2008). Interestingly, by comparison, we did not observe an association between dACC activity and conditioned or reversed SCRs to the threat signals, despite some prior evidence that such relationships may exist, at least for threat conditioning (Milad et al., 2007). Such findings have been important in linking human dACC activity directly to the expression of fear responses: an interpretation that is broadly consistent with anatomical descriptions of the dACC as

Table 2 Significant activation and deactivation associated with differential safety reversal learning (new CS- vs. CSþ). Anatomic Region

Activation: Ventromedial Prefrontal Cortex (extending from medial frontopolar cortex to medial orbitofrontal cortex) Intraparietal lobule (angular gyrus) Ventral posterior cingulate extending to retrosplenial cortex Primary somatosensory cortex Deactivation: Anterior to mid-insular cortex Ventral caudate nucleus Supplementary motor cortex Presupplementary motor cortex

Cluster Size (voxels)

Peak MNI Coordinates

Tmax

P-max

x

y

z

285

0

60

12

5.24

0.029

209 11 458

48 50 4

62 60 62

32 32 24

4.78 4.11 4.79

0.029 0.033 0.029

92

44

24

38

4.59

0.029

632 366 275 128 92 124 275

32 32 8 8 58 62 6

26 28 8 10 38 40 30

4 2 6 4 18 10 40

6.49 6.32 6.41 5.27 5.02 4.75 4.96

<0.001 <0.001 <0.001 <0.001 <0.001 0.003 0.001

7

H.S. Savage et al.

NeuroImage 207 (2020) 116427

Fig. 5. Neural responses during threat reversal predict subjective ratings of anxious arousal. Differential threat reversal learning scores were added as a covariate of interest in a one-sample t-test model. Color bar is scaled as SPM T-scores.

stimuli with representations of self, to infer subjective meaning and personal significance. This information thus likely directly influences the higher-order construction and regulation of emotional responses (Roy et al., 2012). Contrary to our initial hypothesis, the posterior region of the vmPFC (~BA 25), as identified by previous related studies (Apergis-Schoute et al., 2017; Schiller et al., 2008; Zhang et al., 2015), was not observed in our analysis of safety reversal learning, nor in our analysis of simple safety updating effect. This region of the vmPFC is thought to be important for attention direction and selective response inhibition, as well as potentially processing reward-like properties of safe stimuli (Apergis-Schoute et al., 2017; Schiller et al., 2008; Zhang et al., 2015). Differences in study design between the study by Schiller et al. (2008) and ours, namely the number of CS trials during conditioning and reversal, may have reduced our effect size and influenced these observations. It is important to remember that fear reversal and other fear regulation paradigms, can only infer - not prove - the engagement of inhibitory processes. To address this issue directly, it would be optimal to employ specific tests of inhibitory learning, such as conditioned inhibition paradigms (Jovanovic et al., 2005). The anterior insular cortex, often referred to as the ‘viscerosensory’ cortex, is hypothesized to be a hub for converging visceral afferent representations (Singer et al., 2009). This region has repeatedly been shown to track changes in sympathetic arousal (Alvarez et al., 2015; Critchley et al., 2000, 2004, 2005), and has been linked to all forms of emotional responding in fMRI studies (Craig, 2005, 2009; Gu et al., 2013). Our analyses, including the supplementary conjunction analyses, suggest that this region tracks changes in arousal-affective state, regardless of the direction of these changes – it both consistently increased and decreased its activity in response to threat and safety signals, respectively. Previous fear reversal studies have implicated the involvement of subcortical structures including the amygdala and striatum in the processing of aversive value and prediction errors (Apergis-Schoute et al., 2017; Li et al., 2011; Schiller et al., 2008) however, we did not replicate these findings. While our novel approach complicates comparisons, the involvement of these regions has recently been called into question as meta-analysis of fear learning paradigms have failed to substantiate their robust and consistent activation (Fullana et al., 2015; Fullana et al., 2018). Such results address ongoing debate regarding the fundamental processes examined in human fear learning experiments and the impact of fMRI modelling and analysis approaches (Fullana et al., 2015, 2018a, 2018b; Morriss et al., 2018). Overall, participants demonstrated successful threat and safety reversal learning, as measured by changes in anxious arousal and valence ratings of the CS, both within and across learning phases. We did not

Table 3 Significant positive correlation between threat reversal (new CS þ vs. CScontrast) and differential threat reversal learning score for arousal. Anatomic Region

Anterior to mid-insular cortex Pregenual anterior cingulate cortex Primary motor cortex Pregenual anterior cingulate cortex Anterior insular cortex Frontal operculum, inferior frontal cortex

Cluster Size (voxels)

Peak MNI Coordinates

Tmax

P-max

x

y

z

307 124 21

44 34 8

22 24 16

8 4 30

6.49 4.40 5.57

<0.001 <0.001 <0.001

31 66

6 6

20 22

36 28

5.51 5.30

<0.001 <0.001

124 20

34 54

24 6

4 0

4.65 4.40

<0.001 <0.001

‘visceromotor’ interoceptive cortex. Taken together, our findings suggest that the dACC, particularly its rostral area, may have a more direct role in integrating multisensory information about the internal state with external demands in order to focus, constrain and regulate fearful responding (Duncan and Barrett, 2007). Studies employing a variety of fear learning and regulation paradigms have consistently linked activity of the vmPFC to the processing of safety signals; specifically, the evaluation of positive affect, stimulus-value encoding, response inhibition, and conceptual processing more generally (Apergis-Schoute et al., 2017; Harrison et al., 2017; Morris and Dolan, 2004; Roy et al., 2012; Schiller and Delgado, 2010; Schiller et al., 2008). When examining these studies closely, there has also been notable variation in the location of vmPFC subregions implicated across them, which is relevant as these subregions are likely to make distinct functional contributions to safety signal processing on the basis of their unique cytoarchitecture and anatomical connectivity (Myers-Schulz and Koenigs, 2012; Mackey and Petrides, 2014; Wallis et al., 2017; Zhang et al., 2015). The vmPFC area identified in the current study, consistent with our past findings (Fullana et al., 2015; Harrison et al., 2017), has a most prominent anterior focus, located approximately within Brodmann area 10, or the frontopolar cortex. This anterior region has previously been implicated in the processing and encoding of positive affect and value (Harrison et al., 2017; Myers-Schulz and Koenigs, 2012; Ramnani and Owen, 2004; Zhang et al., 2015), and overlaps anatomically with the ‘core’ vmPFC component of the default mode network (Andrews-Hanna et al., 2010; Harrison et al., 2008; Davey et al., 2016). While we cannot rule out that this region may also play a role in response inhibition, the anterior vmPFC likely determines and integrates the affective value of

8

H.S. Savage et al.

NeuroImage 207 (2020) 116427

Council of Australia (NHMRC) Project Grant (1161897) to BJH, CGD and MAF. HSS was supported by an Australian Government Research Training Program (RTP) Scholarship. BJH was supported by a NHMRC Career Development Fellowship (1124472). CGD was supported by an NHMRC Career Development Fellowship (1061757). The authors thank Laura Finlayson-Short, Yara Toenders, Alec Jamieson and Lisa Incerti for their contributions to data collection, as well as staff from the Sunshine Hospital Medical Imaging Department (Western Health, Melbourne).

replicate previous studies employing different fear learning paradigms, that have found that higher trait anxiety (often studied in patients with anxiety disorders) is associated with decreased flexible or reversible learning (Barrett and Armony, 2009; Sehlmeyer et al., 2011; Apergis-Schoute et al., 2017; Wong and Lovibond, 2018). We can only speculate on the extent to which subjective ratings influenced learning dynamics (Lipp and Purkis, 2006; Lonsdorf et al., 2017), when compared to past studies where reversal learning was unsignaled and uninterrupted (Schiller et al., 2008). Interestingly, while participants’ subjective reports indicated successful reversal learning, their autonomic (SCR) responses only differentiated the threat and safety signals during the conditioning phase. While it is not clear why participants’ autonomic responses did not update during reversal, it is not unusual that subjective and autonomic measures are inconsistently correlated (Weinstein et al., 1968; Hodgson and Rachman, 1974; Rosebrock et al., 2017), and may be a result of subjective ratings and SCR indexing different components of emotional learning and responding (Szczepanowski and Pessoa, 2007). Although our study employed a common fMRI design approach (a greater number of CSþ trials relative to CS- trials), the unequal exposure to each stimulus type within conditioning and reversal and/or the inclusion of subjective ratings may have confounded participants learning and must be considered. Furthermore, our task design had fewer trials when compared to Schiller et al. (2008) which, when combined with the low reinforcement rate and our conservative approach to data screening, likely reduced the power of the SCR analysis compared to the analysis of subjective changes. The inclusion of adolescents and young adults in our study may have introduced variability in reversal learning, given noted maturational changes in fear-learning circuitry in animal and human studies (Kim and Richardson, 2010; Britton et al., 2013; Ganella and Kim, 2014; Hartley and Lee, 2015; Ganella et al., 2017). In saying this, however, we did not observe significant cross-sectional associations with participants age and the neural correlates of reversal learning. Fear reversal tasks investigate the flexible maneuvering of fearful responding, modelling real-world emotional learning where responses to threat and safety cues must be updated and controlled simultaneously. Future studies should investigate this processing in patient groups wherein this process may be aberrated (Duits et al., 2015; Grupe and Nitschke, 2013; Jovanovic et al., 2012; Lissek, 2012; Lissek et al., 2013) to establish the clinical validity and/or predictive value of this task.

Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.neuroimage.2019.116427. References Alvarez, R.P., Kirlic, N., Misaki, M., Bodurka, J., Rhudy, J.L., Paulus, M.P., Drevets, W.C., 2015. Increased anterior insula activity in anxious individuals is linked to diminished perceived control. Transl. Psychiatry 5, e591. Andrews-Hanna, J.R., Reidler, J.S., Sepulcre, J., Poulin, R., Buckner, R.L., 2010. Functional- anatomic fractionation of the brain’s default network. Neuron 65, 550–562. Apergis-Schoute, A.M., Gillan, C.M., Fineberg, N.A., Fernandez-Egea, E., Sahakian, B.J., Robbins, T.W., 2017. Neural basis of impaired safety signaling in Obsessive Compulsive Disorder. Proc. Natl. Acad. Sci. U. S. A. 114, 3216–3221. Bach, D.R., Flandin, G., Friston, K.J., Dolan, R.J., 2009. Time-series analysis for rapid event-related skin conductance responses. J. Neurosci. Methods 184, 224–234. Bach, D.R., Friston, K.J., 2013. Model-based analysis of skin conductance responses: towards causal models in psychophysiology. Psychophysiology 50, 15–22. Bach, D.R., Friston, K.J., Dolan, R.J., 2013. An improved algorithm for model-based analysis of evoked skin conductance responses. Biol. Psychol. 94, 490–497. Barrett, J., Armony, J., 2009. Influence of trait anxiety on brain activity during the acquisition and extinction of aversive conditioning. Psychol. Med. 39, 255–265. Boucsein, W., 2012. Electrodermal Activity. Chapter 2 - Methods of Electrodermal Recording. Springer Science & Business Media, pp. 133–136. Bradley, M.M., Lang, P.J., 1994. Measuring emotion: the self-assessment manikin and the semantic differential. J. Behav. Ther. Exp. Psychiatry 25, 49–59. Britton, J.C., Grillon, C., Lissek, S., Norcross, M.A., Szuhany, K.L., Chen, G., Ernst, M., Nelson, E.E., Leibenluft, E., Shechner, T., Pine, D.S., 2013. Response to learned threat: an FMRI study in adolescent and adult anxiety. Am. J. Psychiatry 170, 1195–1204. Casey, B.J., Glatt, C.E., Lee, F.S., 2015. Treating the developing versus developed brain: translating preclinical mouse and human studies. Neuron 86 (6), 1358–1368. https:// doi.org/10.1016/j.neuron.2015.05.020. Corcoran, K.A., Quirk, G.J., 2007. Activity in prelimbic cortex is necessary for the expression of learned, but not innate, fears. J. Neurosci. 27, 840–844. Craig, A., 2005. Forebrain emotional asymmetry: a neuroanatomical basis? Trends Cogn. Sci. 9, 566–571. Craig, A.D., 2009. How do you feel–now? The anterior insula and human awareness. Nat. Rev. Neurosci. 10, 59–70. Crawford, J., Cayley, C., Lovibond, P.F., Wilson, P.H., Hartley, C., 2011. Percentile norms and accompanying interval estimates from an Australian general adult population sample for self-report mood scales (Bai, BDI, CRSD, CES-D, DASS, DASS-21, STAI-X, STAI-Y, SRDS, and SRAS). Aust. Psychol. 46, 3–14. Critchley, H.D., Elliott, R., Mathias, C.J., Dolan, R.J., 2000. Neural activity relating to generation and representation of galvanic skin conductance responses: a functional magnetic resonance imaging study. J. Neurosci. 20, 3033–3040. € Critchley, H.D., Wiens, S., Rotshtein, P., Ohman, A., Dolan, R.J., 2004. Neural systems supporting interoceptive awareness. Nat. Neurosci. 7, 189–195. Critchley, H.D., Rotshtein, P., Nagai, Y., O’doherty, J., Mathias, C.J., Dolan, R.J., 2005. Activity in the human brain predicting differential heart rate responses to emotional facial expressions. Neuroimage 24, 751–762. Davey, C.G., Pujol, J., Harrison, B.J., 2016. Mapping the self in the brain’s default mode network. Neuroimage 132, 390–397. Duits, P., Cath, D.C., Lissek, S., Hox, J.J., Hamm, A.O., Engelhard, I.M., Hout, M.A., Baas, J.M.P., 2015. Updated meta-analysis of classical fear conditioning in the anxiety disorders. Depress. Anxiety 32, 239–253. Duncan, S., Barrett, L.F., 2007. Affect is a form of cognition: a neurobiological analysis. Cognit. Emot. 21, 1184–1211. Finger, E.C., Marsh, A.A., Mitchell, D.G., Reid, M.E., Sims, C., Budhani, S., Leibenluft, E., 2008. Abnormal ventromedial prefrontal cortex function in children with psychopathic traits during reversal learning. Arch. Gen. Psychiatr. 65, 586–594. First, M., Williams, J., Karg, R., Spitzer, R., 2015. Structured Clinical Interview for DSM5—Research Version (SCID-5 for DSM-5, Research Version; SCID-5-RV). American Psychiatric Association, Arlington, VA.
Fullana, M.A., Harrison, B.J., Soriano-Mas, C., Vervliet, B., Cardoner, N., Avila-Parcet, A., Radua, J., 2015. Neural signatures of human fear conditioning: an updated and extended meta-analysis of fMRI studies. Mol. Psychiatry 21, 500–508. Fullana, M.A., Albajes-Eizagirre, A., Soriano-Mas, C., Vervliet, B., Cardoner, N., Benet, O., Radua, J., Harrison, B.J., 2018a. Fear extinction in the human brain: a meta-analysis of fMRI studies in healthy participants. Neurosci. Biobehav. Rev. 88, 16–25.

4.1. Conclusion Consistent with the current neurocircuitry models of fear regulation, our findings highlight the involvement of key mid-line cortical regions in the processing of threat and safety, and provide an additional and novel insight into the specific contribution of these regions to the flexible processes related to updating threat and safety representation. Specifically, our results suggest that the rostral dACC is centrally involved in the appropriate regulation of threat-related responses. They also reinforce the vmPFCs role in safety learning, while raising important questions about the hypothesized subregional contributions to both inhibitory and value processing. The involvement of these regions in regulating human emotional learning processes deserves ongoing investigation. In particular, future studies using novel paradigms that evoke flexible responding to changing situations, contexts and demands, with an increased focus on real-world affective value and meaning, will be critical for understanding these brain functional dynamics in more detail. Declaration of competing interest The authors declare no conflict of interest. Acknowledgements This study was supported by a National Health and Medical Research 9

H.S. Savage et al.

NeuroImage 207 (2020) 116427 methodological considerations for the design and analysis of studies on human fear acquisition, extinction, and return of fear. Neurosci. Biobehav. Rev. Mackey, S., Petrides, M., 2014. Architecture and morphology of the human ventromedial prefrontal cortex. Eur. J. Neurosci. 40, 2777–2796. Milad, M.R., Quirk, G.J., 2012. Fear extinction as a model for translational neuroscience: ten years of progress. Annu. Rev. Psychol. 63, 129–151. Milad, M.R., Quirk, G.J., Pitman, R.K., Orr, S.P., Fischl, B., Rauch, S.L., 2007. A role for the human dorsal anterior cingulate cortex in fear expression. Biol. Psychiatry 62, 1191–1194. Morris, J.S., Dolan, R.J., 2004. Dissociable amygdala and orbitofrontal responses during reversal fear conditioning. Neuroimage 22, 372–380. Morriss, J., Hoare, S., van Reekum, C.M., 2018. It’s time: a commentary on fear extinction in the human brain using fMRI. Neurosci. Biobehav. Rev. 94, 321–322. Myers-Schulz, B., Koenigs, M., 2012. Functional anatomy of ventromedial prefrontal cortex: implications for mood and anxiety disorders. Mol. Psychiatry 17, 132–141. Roy, M., Shohamy, D., Wager, T.D., 2012. Ventromedial prefrontal-subcortical systems and the generation of affective meaning. Trends Cogn. Sci. 16, 147–156. Ramnani, N., Owen, A.M., 2004. Anterior prefrontal cortex: insights into function from anatomy and neuroimaging. Nat. Rev. Neurosci. 5, 184–194. Rosebrock, L., Hoxha, D., Norris, C., Cacioppo, J., Gollan, J., 2017. Skin conductance and subjective arousal in anxiety, depression, and comorbidity. J. Psychophysiol. 31, 145–157. Schiller, D., Delgado, M.R., 2010. Overlapping neural systems mediating extinction, reversal and regulation of fear. Trends Cogn. Sci. 14, 268–276. Schiller, D., Levy, I., Niv, Y., LeDoux, J.E., Phelps, E.A., 2008. From fear to safety and back: reversal of fear in the human brain. J. Neurosci. 28, 11517–11525. Sehlmeyer, C., Dannlowski, U., Schoning, S., Kugel, H., Pyka, M., Pfleiderer, B., Zwitserlood, P., Schiffbauer, H., Heindel, W., Arolt, V., Konrad, C., 2011. Neural correlates of trait anxiety in fear extinction. Psychol. Med. 41, 789–798. Sierra-Mercado, D., Padilla-Coreano, N., Quirk, G.J., 2011. Dissociable roles of prelimbic and infralimbic cortices, ventral hippocampus, and basolateral amygdala in the expression and extinction of conditioned fear. Neuropsychopharmacology 36, 529. Singer, T., Critchley, H.D., Preuschoff, K., 2009. A common role of insula in feelings, empathy and uncertainty. Trends Cogn. Sci. 13, 334–340. Spielberger, C.D., 1983. Manual for the State-Trait Anxiety Inventory STAI (Form Y)(" Self-Evaluation Questionnaire"). Szczepanowski, R., Pessoa, L., 2007. Fear perception: can objective and subjective awareness measures be dissociated? J. Vis. 7, 1–17. Wallis, C.U., Cardinal, R.N., Alexander, L., Roberts, A.C., Clarke, H.F., 2017. Opposing roles of primate areas 25 and 32 and their putative rodent homologs in the regulation of negative emotion. Proc. Natl. Acad. Sci. 114, E4075–E4084. Weinstein, J., Averill, J.R., Opton Jr., E.M., Lazarus, R.S., 1968. Defensive style and discrepancy between self-report and physiological indexes of stress. J. Personal. Soc. Psychol. 10, 406–413. Wilke, M., 2012. An alternative approach towards assessing and accounting for individual motion in fMRI timeseries. Neuroimage 59, 2062–2072. Wong, A.H.K., Lovibond, P.F., 2018. Excessive generalisation of conditioned fear in trait anxious individuals under ambiguity. Behav. Res. Ther. 107, 53–63. Zhang, Z., Mendelsohn, A., Manson, K.F., Schiller, D., Levy, I., 2015. Dissociating value representation and inhibition of inappropriate affective response during reversal learning in the ventromedial prefrontal cortex. eNeuro 2, 1–16. Zimmerman, K.S., Richardson, R., Baker, K.D., 2019. Maturational Changes in Prefrontal and Amygdala Circuits in Adolescence: Implications for Understanding Fear Inhibition during a Vulnerable Period of Development. Brain Sci. 9 (3) https:// doi.org/10.3390/brainsci9030065.

Fullana, M.A., Albajes-Eizagirre, A., Soriano-Mas, C., Vervliet, B., Cardoner, N., Benet, O., Radua, J., Harrison, B.J., 2018b. Amygdala where art thou? Neurosci. Biobehav. Rev. 102, 430–431. Gilmartin, M.R., McEchron, M.D., 2005. Single neurons in the medial prefrontal cortex of the rat exhibit tonic and phasic coding during trace fear conditioning. Behav. Neurosci. 119, 1496–1510. Ganella, D.E., Kim, J.H., 2014. Developmental rodent models of fear and anxiety: from neurobiology to pharmacology. Br. J. Pharmacol. 171, 4556–4574. Ganella, D.E., Drummond, K.D., Ganella, E.P., Whittle, S., Kim, J.H., 2017. Extinction of conditioned fear in adolescents and adults: a human fMRI study. Front. Hum. Neurosci. 11 (647), 1–13. Gray, M.A., Minati, L., Harrison, N.A., Gianaros, P.J., Napadow, V., Critchley, H.D., 2009. Physiological recordings: basic concepts and implementation during functional magnetic resonance imaging. Neuroimage 47, 1105–1115. Grupe, D.W., Nitschke, J.B., 2013. Uncertainty and anticipation in anxiety: an integrated neurobiological and psychological perspective. Nat. Rev. Neurosci. 14 (7), 488–501. Gu, X., Hof, P.R., Friston, K.J., Fan, J., 2013. Anterior insular cortex and emotional awareness. J. Comp. Neurol. 521, 3371–3388. Harrison, B.J., Pujol, J., Lopez-Sola, M., Hernandez-Ribas, R., Deus, J., Ortiz, H., SorianoMas, C., Yucel, M., Pantelis, C., Cardoner, N., 2008. Consistency and functional specialization in the default mode brain network. Proc. Natl. Acad. Sci. U. S. A. 105, 9781–9786. Harrison, B.J., Fullana, M.A., Soriano-Mas, C., Via, E., Pujol, J., Martínez-Zalacaín, I., Tinoco-Gonzalez, D., Davey, C.G., L opez-Sol
a, M., Perez Sola, V., Mench on, J.M., Cardoner, N., 2015. A neural mediator of human anxiety sensitivity. Hum. Brain Mapp. 36, 3950–3958. Harrison, B.J., Fullana, M.A., Via, E., Soriano-Mas, C., Vervliet, B., Martinez-Zalacain, I., Pujol, J., Davey, C.G., Kircher, T., Straube, B., Cardoner, N., 2017. Human ventromedial prefrontal cortex and the positive affective processing of safety signals. Neuroimage 152, 12–18. Hartley, C.A., Lee, F.S., 2015. Sensitive periods in affective development: nonlinear maturation of fear learning. Neuropsychopharmacology 40, 50–60. Hodgson, R., Rachman, S., 1974. II. Desynchrony in measures of fear. Behav. Res. Ther. 12, 319–326. Jovanovic, T., Keyes, M., Fiallos, A., Myers, K.M., Davis, M., Duncan, E.J., 2005. Fear potentiation and fear inhibition in a human fear-potentiated startle paradigm. Biol. Psychiatry 57, 1559–1564. Jovanovic, T., Kazama, A., Bachevalier, J., Davis, M., 2012. Impaired safety signal learning may be a biomarker of PTSD. Neuropharmacology 62 (2), 695–704. Kim, J.H., Richardson, R., 2010. New findings on extinction of conditioned fear early in development: theoretical and clinical implications. Biol. Psychiatry 67, 297–303. Laurent, V., Westbrook, R.F., 2009. Inactivation of the infralimbic but not the prelimbic cortex impairs consolidation and retrieval of fear extinction. Learn. Mem. 16, 520–529. Li, J., Schiller, D., Schoenbaum, G., Phelps, E.A., Daw, N.D., 2011. Differential roles of human striatum and amygdala in associative learning. Nat. Neurosci. 14, 1250–1252. Lipp, O.V., Purkis, H.M., 2006. The effects of assessment type on verbal ratings of conditional stimulus valence and contingency judgments: implications for the extinction of evaluative learning. J. Exp. Psychol. Anim. Behav. Process. 32, 431–440. Lissek, S., 2012. Toward an account of clinical anxiety predicated on basic, neurally mapped mechanisms of Pavlovian fear-learning: the case for conditioned overgeneralization. Depress. Anxiety 29, 257–263. Lissek, S., Bradford, D.E., Alvarez, R.P., Burton, P., Espensen-Sturges, T., Reynolds, R.C., Grillon, C., 2013. Neural substrates of classically conditioned fear-generalization in humans: a parametric fMRI study. Soc. Cogn. Affect. Neurosci. 9 (8), 1134–1142. Lonsdorf, T.B., Menz, M.M., Andreatta, M., Fullana, M.A., Golkar, A., Haaker, J., Heitland, I., Hermann, A., Kuhn, M., Kruse, O., 2017. Don’t fear ‘fear conditioning’:

10