Abstract

Highly influential models have proposed that responses to different types of threat are mediated by partially segregated neural systems, with the amygdala underlying phasic responses to explicit threat (fear) and the bed nucleus of the stria terminalis (BNST) mediating sustained responses to ambiguous threat (anxiety). However, newer models have suggested similar recruitment of both regions across a wide spectrum of threat. Therefore, to empirically test these models and further elucidate the activation profiles and connectivity patterns of the amygdala and the BNST during threat processing, 20 participants were scanned using high-resolution fMRI (1.5 mm3). Using fearful faces and human screams as aversive stimuli, two threat conditions were created: Explicit Threat in which threats were certain and predictable (fear) and Ambiguous Threat in which threats were uncertain and unpredictable (anxiety). Results indicated that, although the amygdala and the BNST both showed heightened engagement across both threat conditions, the amygdala showed preferential engagement during Explicit Threat and displayed functional connectivity with regions involved in stimulus processing and motor response. By contrast, the BNST preferentially responded during Ambiguous Threat and exhibited functional connectivity with prefrontal regions underlying interoception and rumination. Furthermore, correlations with questionnaires measuring trait anxiety, worry, and rumination suggested that individual differences in affective style play a modulatory role in regional recruitment and network connectivity during threat processing.

INTRODUCTION

Anxiety disorders constitute some of the most common and debilitating mental disorders in the general population, with nearly a one in three lifetime incidence (Kessler, Petukhova, Sampson, Zaslavsky, & Wittchen, 2012). Moreover, these disorders are frequently chronic, economically costly, highly comorbid, and often difficult to treat. Although a great deal of progress has been made in basic neuroscience research on the neural mechanisms underlying threat detection, current therapies and clinical outcomes have not paralleled this progress, as evidenced by the fact that the most commonly prescribed and efficacious therapies were developed decades ago, and many individuals remain symptomatic despite existing treatments (LeDoux & Pine, 2016; Griebel & Holmes, 2013; Hyman, 2013). Together, this knowledge underscores the need for continued bidirectional translational research to elucidate the underlying neural circuitry of fear and anxiety to better understand pathological disorders, identify individuals at elevated risk, and ultimately develop new and targeted therapies.

One path toward these goals of increased specificity and more targeted interventions is research aimed at differentiating between anxiety and fear to better understand the specific circuits that support these responses. Psychologically, anxiety can be defined as a prolonged state of apprehension elicited by an uncertain or unpredictable prospective threat. Although the term “anxiety” is often used interchangeably with “fear,” more precisely, fear describes the phasic response to an immediate and identifiable threat. Correspondingly, converging evidence has suggested that this subtle psychological distinction between fear and anxiety is paralleled by partially segregated neural circuits (Avery, Clauss, & Blackford, 2016). Spearheaded by Davis and Walker, this highly influential model theorizes that responses to phasic and sustained threats are mediated, respectively, by the amygdala and the bed nucleus of the stria terminalis (BNST)—a basal forebrain region considered part of the “extended amygdala” (Davis, Walker, Miles, & Grillon, 2010; Walker, Toufexis, & Davis, 2003; Davis, 1998). In earlier versions of this hypothesis, a strict double dissociation was proposed, suggesting that the amygdala mediates phasic responses to Explicit Threat (fear), whereas the BNST responds gradually and displays more sustained responses to unpredictable, ambiguous, or diffuse threat (anxiety). This hypothesis has since been revised to suggest a more subtle functional segregation, proposing that the amygdala contributes to both phasic and sustained fear, with the medial division of the central nuclei mediating phasic fear whereas the lateral nuclei and its projections to the BNST underlie sustained anxious responses (Davis et al., 2010).

Several neuroimaging investigations in humans have further supported a functional dissociation between the amygdala and the BNST during threat processing. In one study, three distinct virtual reality contexts were used to indicate safety, predictable threat of shock or unpredictable threat of shock, respectively. In line with previous animal literature, transient activity in the amygdala was found to be greatest during predictable threat, whereas the BNST showed a positive linear trend in both transient and sustained activity from safety to predictable threat to peak responsivity in unpredictable threat contexts. These results were interpreted to suggest that phasic fear responses are mediated by transient activity in the amygdala, but that, in situations of prolonged exposure to threat, this transient amygdala response may give way to activation of the BNST to maintain anxiety (Alvarez, Chen, Bodurka, Kaplan, & Grillon, 2011). Years later, Klumpers and colleagues presented complementary results using a shock paradigm with cues signaling safety or potential threat. Comparing the anticipatory period to waiting period following a threat cue to the moment of shock confrontation, no evidence for amygdala involvement was found during shock anticipation, but robust amygdala activation was observed during the actual aversive outcome (shock). In comparison, the BNST was found to be significantly elevated during shock anticipation. Although the findings generally support a similar regional dissociation, due to the nature of the study design, these results indicate that the BNST may instead give way to the amygdala, with the BNST playing a role in helping to predict potential outcomes, whereas the amygdala mediates instantaneous responses during acute danger (Klumpers, Kroes, Baas, & Fernández, 2017).

However, still others advocate a different view. Contrary to the notion that the amygdala is primarily involved in phasic fear, sustained changes in amygdala activation and connectivity have been observed during extended periods of anticipatory threat (McMenamin, Langeslag, Sirbu, Padmala, & Pessoa, 2014). Furthermore, in a recent review, Shackman and Fox (2016) amalgamated work that suggests that both the amygdala and the BNST exhibit similar functional profiles in response to a variety of aversive threats. Many of the studies reviewed demonstrate that both the amygdala and the BNST display phasic responses to immediate and short-lived threat, both regions are engaged by uncertain or ambiguous threat, and both show heightened activity during sustained exposure to threat (Shackman & Fox, 2016). This would suggest that the prominent view of a strict functional dissociation warrants reevaluation, and additional thorough investigation examining the specific nature of the differential contributions of the amygdala and the BNST is needed.

The lack of consensus in the field regarding the roles of the amygdala and the BNST in threat processing may in part stem from differences in how paradigms separate aspects of threat to psychologically elicit both fear and anxiety. Furthermore, much of the work of Davis and colleagues was drawn from animal studies, which typically evaluate defense behaviors, whereas human studies and the human experience incorporate subjective feelings. Finally, the combination of the very small size of the BNST and the relatively low spatial resolution of standard fMRI presents an obstacle, one that may cause a false assumption of BNST activation or misattribution of activity to another region and thus discrepancies in reported results.

Therefore, in this study we aimed to further empirically test and delineate the neurobiological mechanisms underlying these theoretical models using high-resolution fMRI (1.5 mm3), as well as employing careful delineation of amygdala nuclei groups (basolateral amygdala [BLA]) and the BNST using ultrahigh resolution anatomical masks (Leal, Noche, Murray, & Yassa, 2017; Avery et al., 2014; Leal, Tighe, Jones, & Yassa, 2014). To investigate the functional activation and connectivity profiles of the amygdala and the BNST during threat processing, the Threat Anticipation Task was designed to vary threat on two key dimensions: certainty/uncertainty of threat occurrence and immediacy/temporal unpredictability of an aversive outcome. From this, two threat conditions were created, one in which threat was certain and predictable (Explicit Threat), intended to elicit fear, and another in which threat was uncertain and unpredictable (Ambiguous Threat) to elicit anxious anticipation. We hypothesized that, in line with the newer proposed models, both the amygdala and the BNST would show heightened responses to Explicit and Ambiguous Threats but would display functional dissociations in their degree of activation, with the amygdala responding more to Explicit Threat and the BNST to Ambiguous Threat, in the manner proposed by Davis and Walker. Similarly, although we anticipated some degree of overlap in the connectivity profiles of the BLA and the BNST, we hypothesized that, relative to the BNST, the BLA would show increased connectivity with stimulus processing and motor response regions (Klumpers et al., 2017), supporting the notion that the amygdala is more closely tied to phasic responses to immediate and identifiable threats (fear), whereas the BNST would show relatively increased connectivity to medial prefrontal regions (Klumpers et al., 2017), supporting its role in more prolonged states of apprehension (anxiety) through worry and rumination. Finally, using self-report questionnaires to measure state and trait anxiety, worry, and rumination, we hypothesized that higher scores in anxiety-related traits would mirror our group analyses during threat, relating to increased activity in the BLA and the BNST, increased connectivity between the BLA and sensorimotor processing regions, and increased connectivity between the BNST and higher order medial prefrontal regions. Finally, if differences were to emerge between anxiety-like traits, we would hypothesize that worry and rumination, the more cognitive aspects of anxiety, would be most closely linked to connectivity between the BNST and the medial pFC (mPFC; Paulesu et al., 2010).

METHODS

Participants

Twenty healthy young adults (16 women) participated in the study (mean age = 20.2 years, SD = 1.88 years). All participants were right-handed, with normal or corrected-to-normal vision and hearing, and had no disclosed history of neurological or psychiatric disorders. Participants were recruited through on-campus flyers and an online research participation system (SONA Systems) and were paid for their participation. Written informed consent was obtained before experimental sessions, and experimental protocols were approved by the University of Louisville institutional review board before data collection. No participants were excluded from any analyses.

Procedure

The study was divided into two consecutive days. On the first day, participants visited the laboratory to sign consent forms, read through task instructions, and complete self-report questionnaires measuring personality traits: State–Trait Anxiety Inventory (STAI; Spielberger, Gorsuch, & Lushene, 1970), Penn State Worry Questionnaire (PSWQ; Meyer, Miller, Metzger, & Borkovec, 1990), and Rumination Response Scale (RRS; Treynor, Gonzalez, & Nolen-Hoeksema, 2003). Importantly, participants were also instructed to complete a short practice round of the Threat Anticipation Task to become familiar with the paradigm and to reduce any startle response that would not be amenable to scanning. The practice round of the Threat Anticipation Task was composed of 10 trials, organized in miniblocks of two successive trials of the same condition to simulate the actual task, with the order of condition blocks pseudorandomized. Numbers of practice trials per condition were as follows: two Explicit Threat, two Explicit Neutral, four Ambiguous Threat (two with an aversive outcome and two with a neutral outcome), and two Ambiguous Neutral. On the second day, participants completed the fMRI portion of the study at the University of Louisville School of Medicine, followed by postscan ratings of visual stimuli.

Stimuli

Images of fearful and neutral faces (white and black, male and female faces) were acquired from the Chicago Face Database (Ma, Correll, & Wittenbrink, 2015). Audio clips of aversive human screams (seven male, seven female) were used for threat conditions. In addition, multitalker babble in a coffee shop (neutral human sound) and nature sounds of a flowing river and chirping birds (neutral nature sound) were used for control conditions. All audio clips were edited to 2 sec in length and normalized using Audacity (DC offset removed and maximum amplitude set to −1dB, the highest amplitude recommended for distortion-free playback on all equipment). During scanning, visual stimuli were displayed through E-Prime (Psychology Tools) onto an Invivo Esys LCD TV monitor at the back of the scanner bore, which was viewed by participants through a mirror on the head-coil. Auditory stimuli were present binaurally through headphones at a predetermined constant level. Red triangles and blue squares were used as cues for different conditions.

Scanning Paradigm

In the Threat Anticipation Task (Figure 1), participants were presented with human faces paired with different sounds (human screams, multitalker babble, or nature sounds). The task contained four conditions: Explicit Threat, Ambiguous Threat, Explicit Neutral, and Ambiguous Neutral. A cue (500 msec) was presented at the beginning of each trial to indicate condition. To evoke threat processing that elicited imminent fear (Explicit Threat) or anticipatory anxiety (Ambiguous Threat), the likelihood of aversive outcome (fearful face + human scream) and onset time of aversive stimulus presentation were manipulated.

Figure 1. 

Example trials for the Explicit Threat, Ambiguous Threat, Explicit Neutral, and Ambiguous Neutral conditions. In trials where cued stimuli did not occur (Ambiguous Threat: fearful face and human scream; Ambiguous Neutral: neutral face and multitalker babble), a neutral face and nature noises were instead presented.

Figure 1. 

Example trials for the Explicit Threat, Ambiguous Threat, Explicit Neutral, and Ambiguous Neutral conditions. In trials where cued stimuli did not occur (Ambiguous Threat: fearful face and human scream; Ambiguous Neutral: neutral face and multitalker babble), a neutral face and nature noises were instead presented.

Explicit Threat trials were cued by a “100%” circumscribed by a red triangle (500 msec). Participants were informed that a red triangle indicated a potential threat, and the probability within the red triangle signaled the likelihood that the aversive outcome would occur (i.e., for the Explicit Threat condition, there was a 100% certainty that the fearful face + human scream would be presented). Immediately following the cue in Explicit Threat trials, a fearful face and human scream were presented (2000 msec). Aversive stimuli were followed by an intertrial interval (ITI) of 1500 msec (total trial length = 4000 msec, 24 trials). Thus, in the Explicit Threat condition, threats were both certain and predictable (fear; Figure 1).

The Ambiguous Threat condition simulated uncertain and unpredictable threats by varying the likelihood of aversive outcome and onset time of aversive stimuli presentation. Ambiguous Threat trials were cued with a red triangle containing probabilities of 80%, 60%, 40%, or 20% that a fearful face and scream would occur, creating event uncertainty. In addition, cues were followed by a variable delay period during which a black screen was shown. Participants were informed that a fearful face and scream could occur at any time, creating temporal unpredictability (in actuality: range 500–5000 msec). On trials when aversive outcomes did not occur, a neutral face and nature sounds were instead presented (2000 msec). Thus, in the Ambiguous Threat condition, threats were both uncertain and unpredictable (anxiety). Ambiguous Threat trials were formulated such that, within each cue probability, aversive stimuli did occur that percentage of the time (e.g., the 60% cue was used 12 times, and aversive stimuli were presented following 7 of those 12 trials or 58.33% of the time). Across all probability conditions of Ambiguous Threat trials, aversive stimuli occurred 50% of the time (24 aversive trials and 24 neutral trials). Stimulus presentation was followed by a variable ITI (500–5000 msec, total trial length = 8000 msec, 48 trials; Figure 1).

Each of the threat conditions was matched with a control condition, cued by blue squares. Participants were instructed that a blue square signaled safety. Explicit Neutral trials were cued with a blue square containing 100% probability, which was immediately followed by a neutral face paired with multitalker babble (total trial length = 4000 msec, 24 trials). Ambiguous Neutral trials were cued with a blue square containing either a 60% or 20% and were followed by the same variable waiting period as Ambiguous Threat trials. However, event outcomes were either a neutral face and multitalker babble, or a neutral face and nature sounds (total trial length = 8000 msec, 24 trials). Participants were informed that aversive stimuli (fearful face + scream) would never occur during Explicit Neutral or Ambiguous Neutral conditions (Figure 1).

The Threat Anticipation Task employed a hybrid event-related design that contained miniblocks of 16 sec (four Explicit Threat/Explicit Neutral trials or two Ambiguous Threat/Ambiguous Neutral trials per miniblock). Miniblocks were presented in a pseudorandom order. This design was chosen to balance considerations for the psychological state of the participant with statistical power. Each condition consisted of 24 trials, with the exception of the Ambiguous Threat condition where 48 trials were presented (24 aversive outcome and 24 neutral outcome), to ensure the estimation of activation was equal across threat conditions when analyzing trials in which aversive stimuli were presented (Explicit Threat [100% aversive occurrence × 24 trials] and Ambiguous Threat [50% overall aversive occurrence × 48 trials]). After each miniblock, an additional pseudorandom variable ITI (jitter) was incorporated to increase design efficiency for hemodynamic response estimation (0–14,000 msec). Finally, a fixation cross was presented for 30 sec in the beginning and end of the task, which was utilized for additional low-level baseline estimation for the fMRI analysis.

After scanning, participants rated all faces, presented in a pseudorandom order, using a 7-point Likert scale to assess valence (1 = extremely pleasant, 4 = neutral, 7 = extremely unpleasant). Unique faces were presented for every trial (total = 120). Behavioral data were analyzed using SPSS (Version 25.0.0, SPSS, Inc.). A probability level of p < .05 was considered statistically significant.

Neuroimaging Methods

Imaging Data Acquisition

Structural.

All structural MRI images were acquired using a Siemens 3T Skyra MR scanner located at the University of Louisville School of Medicine. A 20-channel head coil was used for radiofrequency reception. Participants were given earplugs to reduce scanner noise and were additionally given headphones to receive instructions and auditory stimuli. Foam padding was added to limit motion if additional room remained within the head coil, and a piece of folded tape was placed over the participant's forehead as a reminder to remain still throughout the scan. Structural images were obtained via a T1-weighted magnetization-prepared rapid gradient-echo sequence (MPRAGE) in 208 sagittal slices. Imaging parameters were as follows: echo time (TE) = 2.26 msec, repetition time (TR) = 1700 msec, flip angle = 9.0°, field of view (FoV) = 204 mm, and voxel size = 0.8 × 0.8 × 0.8 mm. Scan parameters were consistent for all imaging sessions associated with this study.

Functional.

Functional BOLD images were collected using gradient-echo T2*-weighted echoplanar imaging (TR = 3000 msec, TE = 30 msec, multiband accelerated Factor 2, FoV = 192 mm, 78 transverse slices with whole-brain coverage, 1.5 mm3 voxels, flip angle = 90°). Slices were oriented obliquely along the AC–PC line. An additional high-contrast full-head BOLD image was obtained to facilitate three-stage registration (TR = 7390 msec, TE = 30 msec, FoV = 192 mm, 100 transverse slices, 1.5 mm3 voxels, flip angle = 90°).

Imaging Analyses

Functional Analyses

Image preprocessing and data analysis were implemented using the FSL package (Version 5.0.9, analysis group, FMRIB, Oxford, United Kingdom, www.fmrib.ox.ac.uk/fsl/). FSL preprocessing pipeline constituted optiBET—brain extraction (Lutkenhoff et al., 2014), time-series prewhitening, and high-pass filtering (0.01 Hz). Functional images were smoothed with a Gaussian kernel of 6 mm FWHM for whole-brain analyses and 3 mm FWHM for ROI analyses. Smoothing of 1.5 mm was additionally evaluated for ROIs (BLA, BNST) and was found to produce nearly equivalent results, suggesting that 3 mm did not oversmooth these regions but produced less variable responses (Supplementary Figure 1).1 Three-stage registration was performed. Individual's functional task images were first registered to a high-contrast full-head functional volume to facilitate registration of the multislice acquisition (multiband) images. Functional images were then registered to high-resolution MPRAGE scans via six-parameter linear registration, and the MPRAGE images were in turn registered to the Montreal Neurological Institute (MNI) 152 T1-1 mm template via a 12-parameter nonlinear registration (Andersson, Jenkinson, & Smith, 2007). These registrations were combined to align the functional images to the 1 mm standard template. Following preprocessing, lower-level statistics were implemented in FEAT. Using multiple regression analysis, statistical maps representing the association between the observed time series (e.g., BOLD signal) and one or a linear combination of regressors for each subject were constructed. Regressors for the main effects were constructed by modeling each of the conditions versus low-level fMRI baseline (ITI, jitter, and fixation): Explicit Threat, Ambiguous Threat (only trials with an aversive outcome), Explicit Neutral, Ambiguous Neutral, and a dummy variable modeling the Ambiguous Threat trials in which neutral stimuli were presented. The contrasts of interest were created by comparing threat conditions against one another: Explicit Threat > Ambiguous Threat and Ambiguous Threat > Explicit Threat. For each regressor, a double-gamma hemodynamic response function was convolved with an event vector starting at the cue onset through stimulus presentation (duration of 2500 msec for Explicit Threat and Explicit Neutral; duration of 3000–7500 msec for Ambiguous Threat and Ambiguous Neutral). In addition to modeling the whole trial of the Explicit Threat and Ambiguous Threat conditions, individual trials epochs were evaluated. For Explicit Threat, one model contained the Cue epoch (500 msec) and another model examined the Stimulus epoch (2000 msec). For Ambiguous Threat, the first model assessed the Cue + Delay epoch (1000–5500 msec), and a separate model contained the Stimulus epoch (2000 msec). Higher level analyses were conducted using FLAME 1 + 2 to combine and spatially normalize all subjects. The higher level models employed nonparametric permutation methods through FSL's randomize function (Nichols & Holmes, 2002). Paired-sample t tests for each contrast of interest were performed using the threshold-free cluster enhancement (TFCE) method, which detects clusters of contiguous voxels without first setting an arbitrary statistical cutoff (e.g., Z > 2.58), and controls the family-wise error rate at p < .05 (Smith & Nichols, 2009). Each contrast underwent 5000 permutations. randomize produces corrected 1 − p maps, which were used for all figures and tables. A conjunction analysis was additionally conducted by thresholding TFCE corrected maps (p < .05) for Explicit Threat and Ambiguous Threat main effects and then combining these maps to visualize commonalities between Explicit and Ambiguous Threat processing. Figures of statistical brain maps were created using FSLview.

Ultra high-resolution anatomical masks (normalized to MNI space) were acquired to accurately delineate the BLA (0.65 mm3; Leal et al., 2014, 2017) and the BNST (0.60 mm3; Avery et al., 2014) for ROI analyses, kindly shared by the authors. As stated in the introduction, Davis has shown that the medial division of the central neucleus of the amygdala (CeA) may mediate fear whereas the lateral portion of the CeA mediates anxiety. Because this level of resolution could not be achieved, only the BLA was selected in an attempt to cleanly dissociate between the roles of the amygdala and the BNST. To ensure optimal regional alignment for veritable signal extraction, two types of registration were explored: the Advanced Normalization Tool (Avants, Tustison, & Song, 2009) and FSL's three-stage registration. Two individuals viewed and compared each mask on participants' EPI images relative to a standard brain and independently confirmed the use of FSL's registration. Figure 2 shows these registered masks on the MNI 152 T1-1 mm standard brain and a representative participant. Following masking of these regions, feat-query was used to extract percent signal change (PSC) from each ROI. Only main effects were modeled for this analysis to associate discrete hemodynamic response function responses for the conditions of interest versus low-level baseline. A 2 × 2 factorial ANOVA and follow-up t tests were then performed to assess differences in functional activation across regions by condition.

Figure 2. 

BLA and BNST masks. Row A: Overlaid on the MNI 152 T1-1 mm standard brain. Row B: Overlaid on a representative participant's structural image. Row C: Overlaid on a representative participant's EPI image. The BLA is shown in red, and the BNST is shown in yellow.

Figure 2. 

BLA and BNST masks. Row A: Overlaid on the MNI 152 T1-1 mm standard brain. Row B: Overlaid on a representative participant's structural image. Row C: Overlaid on a representative participant's EPI image. The BLA is shown in red, and the BNST is shown in yellow.

Functional Connectivity

A priori seed regions were selected for this analysis: BLA and BNST. Whole-brain seed-based functional connectivity was performed by using the following steps: (1) Lower-level participant-specific models (FSL's FEAT) were run by applying high-pass filtering (100 sec), subsequently the residuals and mean functional output were added together (FSL's res4d and mean_func); (2) the average time course was also extracted over three brain masks: ventricles, white matter, and participant space whole brain (FSL's meants); (3) a second lower-level participant-specific model combined the two previous outputs to remove the signal from the ventricles and white matter and to globally normalize the functional signal; (4) subsequently, the residuals and mean functional output were again added together (FSL's res4d and mean_func) to produce a preprocessed participant-specific time series that was high-pass filtered, was controlled for white matter and ventricle signals, and was globally normalized; (5) this participant-specific time series was then used with regressors for the conditions of interest and masked for specific seed ROIs; (6) finally, higher level group models combining all participants were run for each seed (FSL's FEAT). Regions displaying significant functional connectivity were then masked using a 10-mm radius sphere centered around the peak voxel, and PSC was extracted from each ROI across conditions. t tests were performed to compare differentiation in degree of connectivity from each seed. Reported brain regions were required to meet two criteria to be considered functionally connected: (1) display connectivity significantly different from zero in either threat condition and (2) reveal a significant differentiation between the BLA and the BNST as determined by t tests between parameter estimates.

Questionnaires

Self-report questionnaires measuring anxiety-related personality traits (STAI, PSWQ, and RRS) were incorporated into functional analyses. Both the State and Trait scales were utilized within the STAI. First, parameter estimates of average BOLD activation were extracted from each ROI (BLA and BNST) for Explicit and Ambiguous Threat conditions, and questionnaires were correlated by condition. In addition, regions displaying functional connectivity with ROIs were masked using a 10-mm radius centered around the peak voxel, and mean functional connectivity parameter estimates between regions were extracted for Explicit and Ambiguous Threats and correlated with scores from the questionnaires of interest. To correct for multiple comparisons, the Benjamini–Hochberg procedure was implemented to control the false discovery rate at a level <0.05 (Benjamini & Hochberg, 1995).

RESULTS

Behavioral Results

Analysis of postscan face ratings were initially conducted using all fearful faces associated with Explicit Threat/Ambiguous Threat (and excluding the 24 neutral faces in the Ambiguous Threat condition, corresponding to fMRI analyses) and all neutral faces associated with Explicit Neutral/Ambiguous Neutral (collapsed across multitalker babble and nature sounds in Ambiguous Neutral). Results revealed a significant main effect of Threat (Threat, Neutral; F(1, 19) = 160.64, p < .001), with fearful faces being rated as significantly more negative than neutral faces, but no main effect was found for Certainty (Explicit, Ambiguous; F(1, 19) = 3.58, p = .07). In addition, there was no significant interaction of Threat × Certainty, F(1, 19) = 3.83, p = .07, as fearful faces associated with Explicit and Ambiguous Threats were rated as equally negative (Explicit Threat: M = 5.65, SD = 0.57; Ambiguous Threat: M = 5.66, SD = 0.59), t(19) = −0.51, p = .61, despite faces associated with Ambiguous Neutral being rated as less pleasant than faces associated with Explicit Neutral (Ambiguous Neutral: M = 3.71, SD = 0.43; Explicit Neutral: M = 3.61, SD = 0.43), t(19) = −2.23, p = .04.

Following inclusion of neutral faces associated with the Ambiguous Threat condition (from trials when aversive stimuli did not occur and a neutral face and nature noises were instead presented), importantly, these were rated as significantly less pleasant than faces associated with Ambiguous Neutral trials (Ambiguous Threat [neutral faces]: M = 3.82, SD = 0.37; Ambiguous Threat [neutral faces] vs. Ambiguous Neutral: t(19) = 3.08, p < .001). Thus, a clear grading was present in the behavioral data: Neutral faces associated with the Explicit Neutral condition were rated as the most pleasant, whereas faces associated with the Ambiguous Neutral condition were rated as less pleasant, and neutral faces associated with the Ambiguous Threat condition were rated as least pleasant of all.

Neuroimaging Results

Task-Related Whole-brain Activity

Group-level general linear model analysis was performed to examine the neural circuits recruited for Explicit and Ambiguous Threats. Because our aim was to investigate both the neural similarities and differences between fear and anxiety, we first assessed the commonalities between Explicit and Ambiguous Threat processing by conducting a conjunction analysis. Significant clusters were observed in bilateral amygdala, bilateral primary and secondary visual areas (inferior to the calcarine fissure/BA 17/BA 18/BA 19 and fusiform gyrus/BA 37), bilateral auditory processing (superior and middle temporal gyri/BA 22/BA 21, respectively), and bilateral sensory input relay centers in visual and auditory pathways (LGN, medial geniculate nucleus), as well as the right inferior frontal gyrus (rIFG) extending the anterior portion of pars opercularis and pars triangularis (BA 44/BA 45, respectively; Figure 3). Although these TFCE-corrected conjunction results did not reveal involvement of the BNST, the voxelwise (uncorrected) conjunction showed bilateral BNST (p = .001). Furthermore, a conjunction analysis was conducted comparing all threat (Explicit Threat, Ambiguous Threat) versus all neutral (Explicit Neutral, Ambiguous Neutral) to assess regions involved in threat processing after contrasting against conditions that elicit similar levels of visual and auditory processing. This analysis showed very similar results (amygdala, rIFG, enhanced auditory processing), with the exception that significant differences in visual cortical activation no longer emerged (Supplementary Figure 2). The voxelwise (uncorrected) conjunction for all threat compared with all neutral additionally revealed activity in the right BNST (p = .01).

Figure 3. 

Functional activation: Conjunction analysis displaying commonalities between Explicit and Ambiguous (Amb) Threats (green), Explicit Threat > Ambiguous Threat (red), and Ambiguous Threat > Explicit Threat (blue). Aud = auditory cortex; Vis = visual cortex; Amy = amygdala.

Figure 3. 

Functional activation: Conjunction analysis displaying commonalities between Explicit and Ambiguous (Amb) Threats (green), Explicit Threat > Ambiguous Threat (red), and Ambiguous Threat > Explicit Threat (blue). Aud = auditory cortex; Vis = visual cortex; Amy = amygdala.

Conversely, to investigate the differences in neural regions recruited by Explicit Threat and Ambiguous Threat, these conditions were contrasted directly. The contrast Explicit > Ambiguous Threat revealed greater activation in bilateral primary visual areas and bilateral auditory regions. In comparison, the contrast Ambiguous > Explicit Threat was associated with greater activation in mPFC, extending from the pre-SMA (medial BA 6/BA 8) rostrally toward the dorsal ACC (dACC/BA 32; Figure 3; Table 1). An exploratory voxelwise (uncorrected) analysis additionally revealed bilateral amygdala activation for Explicit > Ambiguous Threat (p = .005). Finally, the left BNST was observed in the whole-brain voxelwise analysis for Ambiguous > Explicit Threat (p = .03), but no amygdala activation was found for this contrast.

Table 1. 
Cluster and Peak Report for Whole-brain TFCE-Corrected Analyses for Explicit > Ambiguous Threat, Ambiguous > Explicit Threat, and the Conjunction Analysis between Threat Conditions, as Shown in Figure 3 
RegionXYZCluster Size
Explicit > Ambiguous Threat 
VVPS −35 −87 −21 45666 
Left auditory cortex −44 −22 5225 
Right auditory cortex 55 −18 1797 
  
Ambiguous > Explicit 
dACC −11 33 20 5837 
  
Conjunction 
VVPS 38 −35 −32 86665 
Left auditory cortex/LGN −40 −24 64608 
Right auditory cortex/LGN 33 −26 29841 
rIFG 46 32 7824 
rIFG 39 26 −12 1926 
Right temporal pole −29 12 −33 1080 
Right amygdala −18 −6 −28 515 
Left temporal pole −31 −1 −44 370 
Left amygdala 18 −5 −20 267 
RegionXYZCluster Size
Explicit > Ambiguous Threat 
VVPS −35 −87 −21 45666 
Left auditory cortex −44 −22 5225 
Right auditory cortex 55 −18 1797 
  
Ambiguous > Explicit 
dACC −11 33 20 5837 
  
Conjunction 
VVPS 38 −35 −32 86665 
Left auditory cortex/LGN −40 −24 64608 
Right auditory cortex/LGN 33 −26 29841 
rIFG 46 32 7824 
rIFG 39 26 −12 1926 
Right temporal pole −29 12 −33 1080 
Right amygdala −18 −6 −28 515 
Left temporal pole −31 −1 −44 370 
Left amygdala 18 −5 −20 267 

Coordinates in MNI space.

ROI Analyses

Based on previous literature and recent theoretical models, our primary impetus for this study was to compare and contrast the relative contributions of two key regions underlying threat processing: the amygdala and the BNST. Our ROIs were masked, and PSC was extracted across the time series by condition. When assessing Region (BLA, BNST) × Condition (Explicit Threat, Ambiguous Threat) effects, no significant main effects were noted for Region, F(1, 76) = 3.68, p = .06, or Condition, F(1, 76) = 1.37, p = .25, but a significant Region × Condition interaction was found in a pattern consistent with our hypothesis, F(1, 76) = 141.54, p < .0001. One-sample t tests indicated that both the amygdala and the BNST were significantly elevated above baseline across threat conditions (BLA Explicit Threat: t(19) = 6.68, p < .0001; BLA Ambiguous Threat: t(19) = 4.97, p < .0001; BNST Explicit Threat: t(19) = 3.85, p = .001; BNST Ambiguous Threat: t(19) = 9.53, p < .0001). Follow-up pairwise comparisons revealed that, although the BLA showed a more elevated response in Explicit Threat relative to Ambiguous Threat, this difference was not statistically significant, t(38) = 1.48, p = .15, Cohen's d = .47. However, the BNST showed increased activity during the Ambiguous Threat condition compared with Explicit Threat, t(38) = 3.95, p < .001, d = 1.25 (Figure 4). Analyses were additionally conducted in two control ROIs directly above and below the BNST in the head of the caudate and ventral striatum, respectively, to support that the extracted signal was reliably related to the BNST and that reported results were not contaminated by signal from nearby structures (Supplementary Figure 3).

Figure 4. 

PSC extracted from ROIs across the explicit threat and ambiguous threat conditions. The figure on the left depicts PSC by region for both threat conditions, across all trial epochs. The right figure displays regional PSC during each trial epoch: Cue (Explicit Threat), Cue + Delay (ambiguous threat), and stimulus (explicit threat, ambiguous threat). Across the whole trial period and in discrete trial epochs, significant Region × Condition interactions were found. Error bars represent SEM.

Figure 4. 

PSC extracted from ROIs across the explicit threat and ambiguous threat conditions. The figure on the left depicts PSC by region for both threat conditions, across all trial epochs. The right figure displays regional PSC during each trial epoch: Cue (Explicit Threat), Cue + Delay (ambiguous threat), and stimulus (explicit threat, ambiguous threat). Across the whole trial period and in discrete trial epochs, significant Region × Condition interactions were found. Error bars represent SEM.

Next, to further understand the roles of Explicit Threat and Ambiguous Threat, we separated trials into Cue, Delay, and Stimulus presentation epochs. This approach allowed us to examine what occurs during the Delay epoch of Ambiguous Threat trials, the epoch that specifically differentiates Explicit and Ambiguous Threats. Of note, only Ambiguous Threat trials in which the aversive stimulus occurred were analyzed. Analysis of Region (BLA, BNST) × Condition (Explicit Threat, Ambiguous Threat) effects during the Cue (+Delay) epoch revealed no significant main effect of Region, F(1, 76) = 0.61, p = .44, and no significant main effect of Condition, F(1, 76) = 0.05, p = .82; however, there was a significant Region × Condition interaction, F(1, 76) = 20.94, p < .0001. Similarly, during the aversive Stimulus epoch, no main effects were found for Region, F(1, 76) = 1.02, p = .35, or Condition, F(1, 76) = 3.67, p = .06, but again a significant Region × Condition interaction emerged, F(1, 76) = 66.77, p < .0001. Pairwise comparisons revealed that the BLA showed increased activity to the certain Cue in the Explicit Threat condition (100%) relative to the uncertain Cue + Delay in the Ambiguous Threat condition (80–20%), t(38) = 2.30, p < .05, d = .72, but interestingly, no difference was observed in BLA activity between the Explicit and Ambiguous Threat conditions during the aversive Stimulus epoch, t(38) = 0.43, p = .67, d = .14. The BNST, by comparison, demonstrated an elevated response to the uncertain Cue and anticipatory Delay in the Ambiguous Threat condition, t(38) = 3.17, p < .01, d = 1.00, and showed a further potentiated response during the aversive stimulus presentation, t(38) = 5.23, p < .0001, d = 1.66 (Figure 4). Furthermore, a repeated-measures analysis was conducted to reinforce that these trial epochs could be reliably separated (Supplementary Figure 4).

Functional Connectivity

To assess brain regions that represent possible coherence across the time course of activation during Explicit and Ambiguous Threats, seed-based functional connectivity from the BLA and the BNST was analyzed. Brain regions were required to meet two criteria to be considered functionally connected: (1) display connectivity significantly different from zero in either threat condition (as determined by a one-sample t test on the extracted parameter estimates) and (2) reveal a significant differentiation between the BLA and the BNST (as determined by a t test between parameter estimates extracted for each region).

When comparing connectivity with the BLA relative to the BNST, significant results were found in regions supporting stimulus perception (ventral visual processing stream [VVPS]), emotion detection and processing (ventral insula, ventromedial pFC [vmPFC], dorsomedial pFC [dmPFC]), and motor preparation and execution (medial and lateral primary motor cortex [PMC]). Connectivity between the BLA and VVPS was significantly elevated across both threat conditions (including all trial epochs when the Ambiguous Threat condition was subdivided). Nevertheless, there was a significant difference in the degree of connectivity between the BLA and VVPS relative to the BNST (Explicit: d = 6.86, Ambiguous: d = 8.81). These same findings were also observed between the BLA and emotional detection and processing regions (ventral insula, vmPFC, dmPFC), with significantly increased connectivity being found across both threat conditions and all trial epochs (BLA > BNST; ventral insula—Explicit: d = 2.92, Ambiguous: d = 5.84; vmPFC—Explicit: d = 4.94, Ambiguous: d = 14.65; dmPFC—Explicit: d = 6.73, Ambiguous: d = 5.84). Finally, enhanced functional connectivity was observed between the BLA and the PMC in the Explicit Threat condition and during the Stimulus epoch of Ambiguous Threat trials (Explicit: d = 8.44, Ambiguous: d = 3.20; Figure 5).

Figure 5. 

Seed-based functional connectivity from the BLA and the BNST. Brain regions were required to meet two criteria to be considered functionally connected: (1) displayed connectivity significantly different from zero in either threat condition and (2) revealed a significant differentiation between the BLA and the BNST. Yellow arrows depict significantly greater connectivity from the BLA compared with the BNST. Green arrows show significantly greater functional connectivity with the BNST versus the BLA. Solid lines represent significant connectivity across both threat conditions and all trial epochs, and dashed lines signify significant connectivity only during the Stimulus epoch. DAI = dorsal anterior insula, VI = ventral insula.

Figure 5. 

Seed-based functional connectivity from the BLA and the BNST. Brain regions were required to meet two criteria to be considered functionally connected: (1) displayed connectivity significantly different from zero in either threat condition and (2) revealed a significant differentiation between the BLA and the BNST. Yellow arrows depict significantly greater connectivity from the BLA compared with the BNST. Green arrows show significantly greater functional connectivity with the BNST versus the BLA. Solid lines represent significant connectivity across both threat conditions and all trial epochs, and dashed lines signify significant connectivity only during the Stimulus epoch. DAI = dorsal anterior insula, VI = ventral insula.

In contrast, the BNST revealed increased functional connectivity and significantly greater connectivity relative to the BLA with the dorsal anterior insula during both threat conditions and across all trial epochs (Explicit: d = 6.69, Ambiguous: d = 5.57). In addition, a significant functional relationship was found between the BNST and subgenual ACC (sgACC) during Explicit Threat. Interestingly, both the BLA and the BNST exhibited elevated connectivity with the sgACC during the Ambiguous Threat condition across the whole trial, but a significant difference emerged during the Stimulus epoch (BNST > BLA: d = 3.96; Figure 5).

Questionnaire Correlations

To examine whether individual differences in anxiety-related personality traits were associated with brain activation and connectivity, questionnaires (STAI, PSWQ, RRS) were correlated with subject-level parameter estimates derived from the functional activation ROI analyses (see ROI Analyses section) and from parameter estimates of functional connectivity between seeds and ROIs that resulted from functional connectivity analyses (see Functional Connectivity section). After correcting for multiple comparisons (false discovery rate < .05), results emerged exclusively within the Stimulus epoch of Ambiguous Threat trials. Increased connectivity between the BLA and PMC was negatively correlated with state anxiety (r = −.756, p < .001) and trait anxiety (r = −.599, p = .005). Furthermore, results revealed that increased functional connectivity between the BNST and the sgACC was negatively related to worry (r = −.620, p = .004) and total rumination (r = −.630, p = .003; Figure 6).

Figure 6. 

Correlations between questionnaires and parameter estimates extracted from functional connectivity analysis. All results shown are within the Ambiguous Threat condition, during the Stimulus epoch. (A) Negative correlation between state anxiety (STAI) and functional connectivity parameter estimates between the BLA and PMC. (B) Negative correlation between trait anxiety (STAI) and functional connectivity parameter estimates between the BLA and the PMC. (C) Negative correlation between worry (PSWQ) and functional connectivity between the BNST and the sgACC. (D) Negative correlation between total rumination (RRS) and functional connectivity between the BNST and the sgACC. bilat. PMC = bilateral PMC.

Figure 6. 

Correlations between questionnaires and parameter estimates extracted from functional connectivity analysis. All results shown are within the Ambiguous Threat condition, during the Stimulus epoch. (A) Negative correlation between state anxiety (STAI) and functional connectivity parameter estimates between the BLA and PMC. (B) Negative correlation between trait anxiety (STAI) and functional connectivity parameter estimates between the BLA and the PMC. (C) Negative correlation between worry (PSWQ) and functional connectivity between the BNST and the sgACC. (D) Negative correlation between total rumination (RRS) and functional connectivity between the BNST and the sgACC. bilat. PMC = bilateral PMC.

DISCUSSION

To the best of our knowledge, this study is the first to investigate the differential contributions of the amygdala and the BNST in Explicit and Ambiguous Threat processing, using task-based high-resolution fMRI (1.5 mm3) and precise delineation of these brain structures via ultrahigh resolution anatomical masks. Our paradigm employed a multimodal stimulus task intended to psychologically elicit feelings of fear or anxiety through cues signaling certain and predictable threats or uncertain and unpredictable threats, respectively. Although the BLA and the BNST both displayed heightened activity to Explicit and Ambiguous Threats, important distinctions were noted in degree of recruitment, temporal activation profiles, and functional connectivity. Specifically, the BLA showed preferential involvement in Explicit Threat processing, responding to the certain cue and to the presence of the threatening stimulus across both conditions. The BNST, by contrast, indicated biased engagement during Ambiguous Threat, showing significantly increased activity at the uncertain cue and exhibited distinct patterns of functional connectivity relative to the BLA. Notably, the current findings additionally present valuable insight into how alterations in this network activity and connectivity may relate to individual differences in anxiety-related personality traits.

Behavioral Findings

Immediately after scanning, participants rated all fearful and neutral face stimuli on a 7-point Likert scale to assess perceived valence. As anticipated, all fearful faces associated with threat conditions were rated as significantly more negative than neutral faces when compared with their respective control conditions. Furthermore, there were no differences when ratings of fearful faces were compared between the Explicit Threat and Ambiguous Threat conditions. However, ratings of neutral faces associated with the Ambiguous Neutral condition were rated as significantly less pleasant than faces associated with the Explicit Neutral condition, suggesting that simply waiting for the arrival of an unpredictable stimulus, despite knowing that the stimulus would be neutral, may put individuals in a mildly anxious state and consequently alter processing of the stimulus itself (Somerville et al., 2012). Importantly, this effect was seen to a greater degree with the neutral faces associated with Ambiguous Threat trials. Neutral faces that were presented after anticipation of a potential threat were rated as significantly less pleasant than faces in Ambiguous Neutral trials, indicating that the anxious state induced by a cued threat led to a more negative association of neutral faces in Ambiguous Threat. These findings suggest our manipulation of Explicit and Ambiguous Threats induced negative affect equally, but more importantly, that our manipulation of Ambiguous Threat vis-à-vis the unpredictable probability of threat occurrence and temporal nature of the threat during the delay period induced negative affect before the stimulus was presented (i.e., Ambiguous Threat–neutral faces vs. Ambiguous Neutral).

Commonalities and Functional Dissociations

Whole-brain Functional Activation

To characterize the neural mechanisms associated with Explicit and Ambiguous Threats, the results will be discussed in terms of commonalities and differences observed in threat processing. The initial conjunction analysis assessing the similarities across both threat conditions revealed activation in the amygdala, primary visual and auditory cortices, sensory thalamic relay centers (medial geniculate nucleus, LGN), and rIFG. These finding provide an initial overview of common neural correlates recruited across a broad spectrum of threat. Given the current paradigm and stimuli presented, it is unsurprising that an upregulation in visual and auditory cortices was seen across both threat conditions. Furthermore, it is well known that the amygdala plays a key functional role in detecting salient and novel cues in the environment that predict affective or threatening events (Blackford, Buckholtz, Avery, & Zald, 2010; Adolphs, 2008). The addition of increased activation of the rIFG suggests enhanced negative context monitoring and rapid surveillance of the environment for potential danger, alongside a general withdrawal response (Banich & Depue, 2015; Depue, Orr, Smolker, Naaz, & Banich, 2015; Hampshire, Chamberlain, Monti, Duncan, & Owen, 2010; Corbetta & Shulman, 2002). Taken together with the behavioral results, the observed upregulation of regions related to sensory modalities, in combination with increased amygdala and rIFG response, highlights the common neural mechanisms for general threat detection and supports the validity of the current paradigm.

Directly contrasting the Explicit and Ambiguous Threat conditions uncovered differences associated with certain and predictable versus uncertain and unpredictable threat processing. Greater activation was observed in visual and auditory cortices for the Explicit Threat versus Ambiguous Threat contrast. One interpretation of this suggests that, although both types of threat recruited these sensory regions, Explicit Threat is more stimulus bound and exemplifies a stronger representation of the stimulus. However, it is also possible that, because of the nature of the study design, an initially strong response in these regions during the Ambiguous Threat condition may be diminished due to the inclusion of the Delay epoch or that the aversive Stimulus epoch in the Explicit Threat condition may be exhibiting a stronger relative influence due to the lack of a Delay epoch between the Cue and aversive outcome. In the opposing contrast, Ambiguous Threat versus Explicit Threat revealed greater activation in the dACC extending to the dmPFC, suggesting higher-level detection of emotion and conflict monitoring (Egner, Etkin, Gale, & Hirsch, 2008; Eisenberger & Lieberman, 2004), likely in anticipation of threat. Recent research highlights the ACC as a central locus for signaling outcome uncertainty in a valence-specific manner. Through a Pavlovian procedure in monkeys investigating the certainty versus uncertainty of punishments and rewards, one study identified a novel punishment uncertainty signal in the ACC, demonstrating that some neurons are selectively excited by the prospect of uncertain punishment and are most strongly activated during greatest uncertainty (50% probability of an aversive outcome; Monosov, 2017).

ROI Analysis

Focusing on our a priori ROIs, we next assessed the differential contributions of these regions to Explicit and Ambiguous Threats. Across the whole trial period, we found that both regions were significantly activated across threat conditions but importantly found a significant region by condition interaction in a pattern consistent with our hypothesis. Globally, the BLA displayed preferential responsivity to certain and predictable threats relative to the BNST, as shown by qualitatively increased activity in Explicit Threat relative to Ambiguous Threat, although this finding was not statistically significant. An opposing pattern of activity was observed in the BNST, with significantly more activation being found in Ambiguous Threat compared with Explicit Threat. These findings indicate that, although a partial functional dissociation was observed, in the manner proposed by Davis and colleagues, both regions displayed elevated activity across conditions, lending support to perspectives outlined by Shackman and Fox (2016).

Additional insight was uncovered following division of trials into epochs. For Explicit Threat, the Cue epoch and aversive Stimulus epoch were modeled separately, and for Ambiguous Threat, trials were separated into the Cue + Delay epoch and the Stimulus epoch. Although a significant region by condition interaction was still present in both the Cue (+Delay) and aversive Stimulus epochs, subsequent pairwise comparison revealed additional insights into the functional roles of each region. First, a significant difference was noted when comparing the Explicit Threat Cue to the Ambiguous Threat Cue + Delay epoch, with the BLA exhibiting increased responsivity to the concretely paired certain cue in Explicit Threat, at a magnitude comparable to BLA's response to the aversive Stimulus. However, when comparing the two threat conditions during the Stimulus epoch, no difference was observed in BLA activity (due to the elevated BLA response in Ambiguous Threat during the Stimulus epoch). Together, this explains why a significant difference did not emerge between conditions across all epochs of the trial. However, more importantly, this demonstrates that the BLA preferentially responds to the threatening stimuli's overt display, regardless of whether the onset of that stimulus is immediate (Explicit Threat) or temporally delayed (Ambiguous Threat).

In contrast, the BNST showed increased activity to the uncertain cue and unpredictable anticipatory delay (Cue + Delay) in the Ambiguous Threat condition and continued to display an elevated response throughout the stimulus presentation. Although both the BLA and the BNST displayed heightened activity at all threat cues, plausibly serving as an alerting system to potential danger, the magnified response of the BNST in the Ambiguous Threat condition suggests that the BNST may underlie increased vigilance when the specifics of a threat are unknown.

These results compliment previous work by Somerville, Whalen, and Kelley (2010), which showed that the BNST continuously tracks threat proximity (low, medium, or high risk of receiving a shock) and that this threat monitoring was exaggerated in individuals with high trait anxiety. In this paradigm, participants never actually received a shock while being scanned, and notably, the study reported that the amygdala showed minimal task-modulated activity even at exploratory thresholds. Klumpers et al. (2017) reports a similar dissociation in the roles of the amygdala and the BNST using a shock paradigm. During the presentation of a cue signaling potential shock (16% or 33% reinforcement rate), significant activation was noted in the BNST in two independent samples. In contrast, no evidence was found for amygdala involvement during uncertain shock anticipation, but robust amygdala activation was exhibited during the actual aversive outcome (with high probability for localization in the BLA; Klumpers et al., 2017). Thus, taken all together, these results help clarify the functional roles of the amygdala and the BNST: A regional dissociation can be attributed to the BLA preferentially responding the “actual presence” of an aversive stimulus or a concretely paired cue, whereas the BNST exhibits a functional specialization for the detection of a potential threat and maintains hypervigilance until threat arrival or situational resolve.

Functional Connectivity

Analysis of seed-based functional connectivity was assessed from these two ROIs: the BLA and the BNST. Results revealed increased functional connectivity between the BLA and bilateral VVPS across Explicit and Ambiguous Threats, with the strongest connectivity being observed during the Stimulus epoch of the Ambiguous Threat condition (Klumpers et al., 2017). These findings build on previous results, which demonstrated that the amygdala responded to the aversive stimulus itself across both types of threat and thus more strongly in the Explicit Threat condition on average, as it contained only a certain cue and stimulus presentation (no delay epoch). Moreover, whole-brain results showed that Explicit Threat was more stimulus bound when contrasted directly with Ambiguous Threat. Added evidence from functional connectivity then suggests that the stimulus bound nature of Explicit Threat may be mediated through the BLA and its back projections to upregulate visual processing (Pessoa & Adolphs, 2010; Amaral, Behniea, & Kelly, 2003). In addition, the BLA displayed increased functional connectivity with cortical motor areas indicating a role in preparation for and executing a motor response (Avendaño, Price, & Amaral, 1983; Llamas, Avendaño, & Reinoso-Suárez, 1977). A similar temporal pattern of results was observed, with enhanced connectivity being observed in the Explicit Threat condition and during the Stimulus epoch of the Ambiguous Threat condition. Together, these results suggest that, in the face of threat, the amygdala may facilitate coordinated activity between sensory processing areas and motor control, so as to afford quick and adaptive behavioral changes.

Finally, the BLA showed increased functional connectivity with the emotional detection and processing regions (vmPFC, dmPFC, and ventral insula) across the whole trial for both threat conditions. It is well known that the amygdala shares extensive connections with the mPFC (Phan, Wager, Taylor, & Liberzon, 2002), whose activity is thought to underlie many facets of cognitive and emotional processing including emotional detection, appraisal, self-monitoring, and emotion regulation (Etkin, Egner, & Kalisch, 2011), whereas the insula has additionally been implicated in general affective processing and integration of body state representations (Craig, 2002, 2009; Critchley, Wiens, Rotshtein, Öhman, & Dolan, 2004). This indicates that, in addition to facilitating gross motor movement planning for defensive behaviors or escape, the BLA may contribute to specific motor selection in concert with changes in emotional and body state, especially when that state is representative of discomfort, thought to be represented in the ventral insula (Jezzini, Caruana, Stoianov, Gallese, & Rizzolatti, 2012).

In both Explicit and Ambiguous Threats, the BNST was found to exhibit extensive functional connectivity with the insula, specifically in the most dorsal anterior portions. In addition to underlying integration of body states, the anterior insula in particular has been hypothesized to play a role in the “perception” of subjective interoceptive states (Grupe & Nitschke, 2013). Thus, this metacognitive aspect of interoception may in part underlie feelings of anticipatory anxiety when a potential threat is detected through increased awareness and interpretation of physiological arousal (Herrmann et al., 2016; Damasio & Carvalho, 2013). The BNST additionally displayed increased functional connectivity with the sgACC in Explicit Threat and during the Stimulus epoch of Ambiguous Threat, a prefrontal region putatively involved in internal mentation. Together, these results suggest greater connectivity of the BNST to regions supporting higher level perception of interoceptive state, as well as prefrontal regions that may modulate these responses through reflection and rumination, suggesting a role of the BNST in the more psychological aspects of anxiety (Torrisi et al., 2018; Klumpers et al., 2017; Andrews-Hanna, 2012; Mobbs et al., 2007). Finally, the sgACC is additionally known to be highly involved in communication with the amygdala for downregulation of negative affect (Connolly et al., 2013; Banks, Eddy, Angstadt, Nathan, & Phan, 2007), bringing up the intriguing question of whether this region has the same top–down control over the BNST.

Relationships with Questionnaires

To evaluate the generalizability of the neural findings beyond the utilized paradigm, personality questionnaires measuring anxiety, worry, and rumination were collected, and relationships between individual differences in personality and alterations in ROI recruitment and network connectivity were assessed during threat processing. However, given that our sample size was modest (N = 20), the results should be interpreted with care as we do not want to overspeculate on relationships with individual differences. As such, the relationships with questionnaire scores are primarily presented as broad support for the task results, rather than claims on how individual traits specifically modulate threat processing.

After correcting for multiple comparisons, no significant results were found when questionnaires were correlated with parameter estimates in ROI activity across threat conditions. However, correlations with functional connectivity parameter estimates revealed two functional pathways that exhibited relationships with anxiety-like personality traits, suggesting that functional connectivity may be a better predictor of behavior than regional activation alone. First, connectivity between the BLA and bilateral PMC was found to be negatively related to state and trait anxiety (STAI state, STAI trait) during the Stimulus epoch of Ambiguous Threat. Initially, this may seem counterintuitive. However, these results likely indicate better integration of emotional and motor responses for individuals with lower trait anxiety. Because these findings were specific to the stimulus period of Ambiguous Threat, this implies that the anticipatory delay influenced responses to the stimulus (otherwise, the same results would be seen in Explicit Threat). Therefore, these results suggest that individuals with lower anxiety have increased connectivity between the amygdala and cortical motor systems, which may reflect enhanced motor planning during the anticipatory delay in preparation for the arrival of an aversive stimulus. Thus, this increased communication and better preparedness for protective or defensive motor behaviors may reduce anxiety, or conversely, individuals who are less anxious may have better ability to prepare an appropriate motor response in the face of a potential threat. In addition, increased functional connectivity between the BNST and sgACC, likely an index for communication between emotional and regulatory systems, was found to be related to reduced worry and total rumination. These findings are supported by the functional connectivity results, which demonstrated a dissociation between the connectivity profiles of the BLA and the BNST, with the BLA showing increased connectivity to stimulus processing and motor response regions whereas the BNST showed enhanced communication with medial prefrontal regions putatively involved in internal mentation. Worry and rumination are processes more closely tied to anticipation and as opposed to reactivity, and the BNST was likewise preferentially involved in Ambiguous Threat processing, suggesting that this functional pathway may underlie some of the more cognitive aspects of anxiety (Muris, Roelofs, Rassin, Franken, & Mayer, 2005).

Limitations and Future Directions

Limitations of this study should be acknowledged. First, our sample size was modest, using only 20 healthy participants. As previously stated, results should be interpreted with care given the limited sample size, and future studies will be needed to investigate how individual differences in personality specifically modulate components of threat processing. Second, all participants were considered psychologically healthy, and although our study indicates that differences in personality may alter regional activity and connectivity, investigation of clinical populations is needed to specifically elucidate the neural underpinnings of anxiety disorders. Our sample was also predominantly female, and given that the BNST is known to be a sexually dimorphic region (Hines, Davis, Coquelin, Goy, & Gorski, 1985), it is unknown how these mechanisms may vary by gender. In addition, our analyses only focused on the functional role of the BLA relative to the BNST, and therefore, it is possible that another picture may have emerged for other amygdala regions, such as the CeA. Finally, to maximize statistical power, trial lengths were brief (max = 5500 msec until stimulus presentation), and we therefore could not confidently address the phasic versus sustained response profile debate for the amygdala and the BNST.

Therefore, future studies would benefit from larger sample sizes with roughly equal numbers of men and women to assess gender differences and longer trial lengths to investigate phasic and sustained responses of these ROIs under Explicit and Ambiguous Threat conditions. In addition, although our study used 80–20% probabilities of an aversive outcome in the Ambiguous Threat condition, individual probability conditions had too few trials to be able to assess how different levels of threat likelihood affected processing. Therefore, future studies should expand on the current design, including more trials to evaluate parametric modulation of threat likelihood. This could provide insight into whether the BNST tracks these different probabilities of occurrence and how these mechanisms differ in individuals with high trait anxiety. Finally, although our study only included a threat condition that was both certain and predictable, additional studies could explore conditions in which threat is certain but delayed or where threat is cued as certain but never arrives. In the first scenario, we would hypothesize that the BNST would respond at the cue and show a sustained response until stimulus arrival, whereas the latter may serve as a model for generalized anxiety through a simulated state of perpetual anticipation for a fear that may never occur.

Summary and Conclusions

In summary, results from functional activation contrasts revealed that Explicit Threat engages more stimulus bound processing, as evidenced by increased activation in visual and auditory cortices. By contrast, Ambiguous Threat processing involves the dmPFC and the dACC, suggesting higher level emotional detection. These results were further supported by analysis of ROIs, which showed that the BLA exhibited preferential involvement in Explicit Threat, as measured by PSC, and displayed heightened responsivity to the presence of aversive stimuli presentation across conditions. These findings demonstrate that activity of the amygdala is more concretely tied to the threatening stimulus itself,or a concretely paired cue, putatively mediating feelings of fear. The BNST, by comparison, showed preferential involvement during Ambiguous Threat processing and exhibited significantly elevated activity at the uncertain cue and showed a potentiated response to the aversive stimulus presentation. This further supports that BNST activity may predominantly serve as an alerting system, responding as soon as a prospective threat is detected, and putatively mediating feelings of anticipatory anxiety. However, as these analyses did not include mediation model with subjective feelings, the precise relationships between these regional dissociations and the feelings of fear and anxiety will have to be explored in future work. In addition, functional connectivity results demonstrated that, on a whole, the BLA display increased connectivity with regions supporting stimulus processing and gross motor response, whereas the BNST was found to be more functionally related to anterior prefrontal regions that underlie interoception, internal mentation, and rumination. Importantly, these current findings were strengthened by relationships with individual differences in personality trait and mood state, which further emphasized these partial functional dissociations, and suggested that differences in individual affective state may play a modulatory role in these key networks during threat processing.

Based on the current results, we believe that both proposed models on threat detection, as they relate to fear and anxiety, have validity. Our results support that the BLA is more involved in Explicit Threat (fear) processing, whereas the BNST shows preferential engagement to Ambiguous Threat (anxiety), as proposed by Davis and colleagues. However, contrary to this model, our results indicated that all regions respond to both threat conditions, lending support to the perspectives of Shackman and Fox. Therefore, we instead propose an alternative idea that amends these disparities. Over and above the type of threat being processed, the BNST appears to exhibit a functional specialization for the detection of a potential threat, putatively serving as an alerting system to maintain hypervigilance and, thus, worry and rumination until the arrival of a threat or resolution of the threatening situation. In complement to the BNST, the BLA preferentially responds to the certainty of threat occurrence or the actual presence of a threatening stimulus, regardless of whether that threat is immediate or occurs after an anticipatory delay. Together, these results and this altered view of threat processing may help explain the inconsistencies that currently exist in the literature and inform future research.

Acknowledgments

The authors gratefully acknowledge Brooke Siers for assistance with study design and data acquisition and Teodora Stoica, Leonard Faul, and Brooke Siers for help with the manuscript.

Reprint requests should be sent to Brendan E. Depue, Department of Psychological and Brain Sciences and Department of Anatomical Sciences and Neurobiology, University of Louisville, Louisville, KY 40292 or via e-mail: brendan.depue@louisville.edu.

Note

1. 

Supplementary material for this paper can be retrieved from www.nilcamp.net.

REFERENCES

REFERENCES
Adolphs
,
R.
(
2008
).
Fear, faces, and the human amygdala
.
Current Opinion in Neurobiology
,
18
,
166
172
.
Alvarez
,
R. P.
,
Chen
,
G.
,
Bodurka
,
J.
,
Kaplan
,
R.
, &
Grillon
,
C.
(
2011
).
Phasic and sustained fear in humans elicits distinct patterns of brain activity
.
Neuroimage
,
55
,
389
400
.
Amaral
,
D. G.
,
Behniea
,
H.
, &
Kelly
,
J. L.
(
2003
).
Topographic organization of projections from the amygdala to the visual cortex in the macaque monkey
.
Neuroscience
,
118
,
1099
1120
.
Andersson
,
J. L.
,
Jenkinson
,
M.
, &
Smith
,
S.
(
2007
).
Non-linear optimisation
.
FMRIB (Technical Report TR07JA1)
.
Oxford, UK
:
University of Oxford FMRIB Centre
.
Andrews-Hanna
,
J. R.
(
2012
).
The brain's default network and its adaptive role in internal mentation
.
Neuroscientist
,
18
,
251
270
.
Avants
,
B. B.
,
Tustison
,
N.
, &
Song
,
G.
(
2009
).
Advanced normalization tools (ANTS)
.
Insight Journal
,
2
,
1
35
.
Avendaño
,
C.
,
Price
,
J. L.
, &
Amaral
,
D. G.
(
1983
).
Evidence for an amygdaloid projection to premotor cortex but not to motor cortex in the monkey
.
Brain Research
,
264
,
111
117
.
Avery
,
S. N.
,
Clauss
,
J. A.
, &
Blackford
,
J. U.
(
2016
).
The human BNST: Functional role in anxiety and addiction
.
Neuropsychopharmacology
,
41
,
126
141
.
Avery
,
S. N.
,
Clauss
,
J. A.
,
Winder
,
D. G.
,
Woodward
,
N.
,
Heckers
,
S.
, &
Blackford
,
J. U.
(
2014
).
BNST neurocircuitry in humans
.
Neuroimage
,
91
,
311
323
.
Banich
,
M. T.
, &
Depue
,
B. E.
(
2015
).
Recent advances in understanding neural systems that support inhibitory control
.
Current Opinion in Behavioral Sciences
,
1
,
17
22
.
Banks
,
S. J.
,
Eddy
,
K. T.
,
Angstadt
,
M.
,
Nathan
,
P. J.
, &
Phan
,
K. L.
(
2007
).
Amygdala–Frontal connectivity during emotion regulation
.
Social Cognitive and Affective Neuroscience
,
2
,
303
312
.
Benjamini
,
Y.
, &
Hochberg
,
Y.
(
1995
).
Controlling the false discovery rate: A practical and powerful approach to multiple testing
.
Journal of the Royal Statistical Society. Series B (Methodological)
,
57
,
289
300
.
Blackford
,
J. U.
,
Buckholtz
,
J. W.
,
Avery
,
S. N.
, &
Zald
,
D. H.
(
2010
).
A unique role for the human amygdala in novelty detection
.
Neuroimage
,
50
,
1188
1193
.
Connolly
,
C. G.
,
Wu
,
J.
,
Ho
,
T. C.
,
Hoeft
,
F.
,
Wolkowitz
,
O.
,
Eisendrath
,
S.
, et al
(
2013
).
Resting-state functional connectivity of subgenual anterior cingulate cortex in depressed adolescents
.
Biological Psychiatry
,
74
,
898
907
.
Corbetta
,
M.
, &
Shulman
,
G. L.
(
2002
).
Control of goal-directed and stimulus-driven attention in the brain
.
Nature Reviews Neuroscience
,
3
,
201
215
.
Craig
,
A. D.
(
2002
).
How do you feel? Interoception: The sense of the physiological condition of the body
.
Nature Reviews Neuroscience
,
3
,
655
666
.
Craig
,
A. D.
(
2009
).
How do you feel–now? The anterior insula and human awareness
.
Nature Reviews Neuroscience
,
10
,
59
70
.
Critchley
,
H. D.
,
Wiens
,
S.
,
Rotshtein
,
P.
,
Öhman
,
A.
, &
Dolan
,
R. J.
(
2004
).
Neural systems supporting interoceptive awareness
.
Nature Neuroscience
,
7
,
189
195
.
Damasio
,
A.
, &
Carvalho
,
G. B.
(
2013
).
The nature of feelings: Evolutionary and neurobiological origins
.
Nature Reviews Neuroscience
,
14
,
143
152
.
Davis
,
M.
(
1998
).
Are different parts of the extended amygdala involved in fear versus anxiety?
Biological Psychiatry
,
44
,
1239
1247
.
Davis
,
M.
,
Walker
,
D. L.
,
Miles
,
L.
, &
Grillon
,
C.
(
2010
).
Phasic vs sustained fear in rats and humans: Role of the extended amygdala in fear vs anxiety
.
Neuropsychopharmacology
,
35
,
105
135
.
Depue
,
B. E.
,
Orr
,
J. M.
,
Smolker
,
H. R.
,
Naaz
,
F.
, &
Banich
,
M. T.
(
2015
).
The organization of right prefrontal networks reveals common mechanisms of inhibitory regulation across cognitive, emotional, and motor processes
.
Cerebral Cortex
,
26
,
1634
1646
.
Egner
,
T.
,
Etkin
,
A.
,
Gale
,
S.
, &
Hirsch
,
J.
(
2008
).
Dissociable neural systems resolve conflict from emotional versus nonemotional distracters
.
Cerebral Cortex
,
18
,
1475
1484
.
Eisenberger
,
N. I.
, &
Lieberman
,
M. D.
(
2004
).
Why rejection hurts: A common neural alarm system for physical and social pain
.
Trends in Cognitive Sciences
,
8
,
294
300
.
Etkin
,
A.
,
Egner
,
T.
, &
Kalisch
,
R.
(
2011
).
Emotional processing in anterior cingulate and medial prefrontal cortex
.
Trends in Cognitive Sciences
,
15
,
85
93
.
Griebel
,
G.
, &
Holmes
,
A.
(
2013
).
50 years of hurdles and hope in anxiolytic drug discovery
.
Nature Reviews Drug Discovery
,
12
,
667
687
.
Grupe
,
D. W.
, &
Nitschke
,
J. B.
(
2013
).
Uncertainty and anticipation in anxiety: An integrated neurobiological and psychological perspective
.
Nature Reviews Neuroscience
,
14
,
488
501
.
Hampshire
,
A.
,
Chamberlain
,
S. R.
,
Monti
,
M. M.
,
Duncan
,
J.
, &
Owen
,
A. M.
(
2010
).
The role of the right inferior frontal gyrus: Inhibition and attentional control
.
Neuroimage
,
50
,
1313
1319
.
Herrmann
,
M. J.
,
Boehme
,
S.
,
Becker
,
M. P.
,
Tupak
,
S. V.
,
Guhn
,
A.
,
Schmidt
,
B.
, et al
(
2016
).
Phasic and sustained brain responses in the amygdala and the bed nucleus of the stria terminalis during threat anticipation
.
Human Brain Mapping
,
37
,
1091
1102
.
Hines
,
M.
,
Davis
,
F. C.
,
Coquelin
,
A.
,
Goy
,
R. W.
, &
Gorski
,
R. A.
(
1985
).
Sexually dimorphic regions in the medial preoptic area and the bed nucleus of the stria terminalis of the guinea pig brain: a description and an investigation of their relationship to gonadal steroids in adulthood
.
Journal of Neuroscience
,
5
,
40
47
.
Hyman
,
S. E.
(
2013
).
Psychiatric drug development: Diagnosing a crisis
.
Cerebrum
,
2013
,
5
.
Jezzini
,
A.
,
Caruana
,
F.
,
Stoianov
,
I.
,
Gallese
,
V.
, &
Rizzolatti
,
G.
(
2012
).
Functional organization of the insula and inner perisylvian regions
.
Proceedings of the National Academy of Sciences, U.S.A.
,
109
,
10077
10082
.
Kessler
,
R. C.
,
Petukhova
,
M.
,
Sampson
,
N. A.
,
Zaslavsky
,
A. M.
, &
Wittchen
,
H. U.
(
2012
).
Twelve-month and lifetime prevalence and lifetime morbid risk of anxiety and mood disorders in the United States
.
International Journal of Methods in Psychiatric Research
,
21
,
169
184
.
Klumpers
,
F.
,
Kroes
,
M. C. W.
,
Baas
,
J.
, &
Fernández
,
G.
(
2017
).
How human amygdala and bed nucleus of the stria terminalis may drive distinct defensive responses
.
Journal of Neuroscience
,
37
,
3830
3916
.
Leal
,
S. L.
,
Noche
,
J. A.
,
Murray
,
E. A.
, &
Yassa
,
M. A.
(
2017
).
Disruption of amygdala–entorhinal–hippocampal network in late-life depression
.
Hippocampus
,
27
,
464
476
.
Leal
,
S. L.
,
Tighe
,
S. K.
,
Jones
,
C. K.
, &
Yassa
,
M. A.
(
2014
).
Pattern separation of emotional information in hippocampal dentate and CA3
.
Hippocampus
,
24
,
1146
1155
.
LeDoux
,
J. E.
, &
Pine
,
D. S.
(
2016
).
Using neuroscience to help understand fear and anxiety: A two-system framework
.
American Journal of Psychiatry
,
173
,
1083
1093
.
Llamas
,
A.
,
Avendaño
,
C.
, &
Reinoso-Suárez
,
F.
(
1977
).
Amygdaloid projections to prefrontal and motor cortex
.
Science
,
195
,
794
796
.
Lutkenhoff
,
E. S.
,
Rosenberg
,
M.
,
Chiang
,
J.
,
Zhang
,
K.
,
Pickard
,
J. D.
,
Owen
,
A. M.
, et al
(
2014
).
Optimized brain extraction for pathological brains (optiBET)
.
PLoS One
,
9
,
e115551
.
Ma
,
D. S.
,
Correll
,
J.
, &
Wittenbrink
,
B.
(
2015
).
The Chicago face database: A free stimulus set of faces and norming data
.
Behavior Research Methods
,
47
,
1122
1135
.
McMenamin
,
B. W.
,
Langeslag
,
S. J.
,
Sirbu
,
M.
,
Padmala
,
S.
, &
Pessoa
,
L.
(
2014
).
Network organization unfolds over time during periods of anxious anticipation
.
Journal of Neuroscience
,
34
,
11261
11273
.
Meyer
,
T. J.
,
Miller
,
M. L.
,
Metzger
,
R. L.
, &
Borkovec
,
T. D.
(
1990
).
Development and validation of the Penn State Worry Questionnaire
.
Behaviour Research and Therapy
,
28
,
487
495
.
Mobbs
,
D.
,
Petrovic
,
P.
,
Marchant
,
J. L.
,
Hassabis
,
D.
,
Weiskopf
,
N.
,
Seymour
,
B.
, et al
(
2007
).
When fear is near: Threat imminence elicits prefrontal-periaqueductal gray shifts in humans
.
Science
,
317
,
1079
1083
.
Monosov
,
I. E.
(
2017
).
Anterior cingulate is a source of valence-specific information about value and uncertainty
.
Nature Communications
,
8
,
134
.
Muris
,
P.
,
Roelofs
,
J.
,
Rassin
,
E.
,
Franken
,
I.
, &
Mayer
,
B.
(
2005
).
Mediating effects of rumination and worry on the links between neuroticism, anxiety and depression
.
Personality and Individual Differences
,
39
,
1105
1111
.
Nichols
,
T. E.
, &
Holmes
,
A. P.
(
2002
).
Nonparametric permutation tests for functional neuroimaging: A primer with examples
.
Human Brain Mapping
,
15
,
1
25
.
Paulesu
,
E.
,
Sambugaro
,
E.
,
Torti
,
T.
,
Danelli
,
L.
,
Ferri
,
F.
,
Scialfa
,
G.
, et al
(
2010
).
Neural correlates of worry in generalized anxiety disorder and in normal controls: A functional MRI study
.
Psychological Medicine
,
40
,
117
124
.
Pessoa
,
L.
, &
Adolphs
,
R.
(
2010
).
Emotion processing and the amygdala: From a ‘low road’ to ‘many roads’ of evaluating biological significance
.
Nature Reviews Neuroscience
,
11
,
773
783
.
Phan
,
K. L.
,
Wager
,
T.
,
Taylor
,
S. F.
, &
Liberzon
,
I.
(
2002
).
Functional neuroanatomy of emotion: A meta-analysis of emotion activation studies in PET and fMRI
.
Neuroimage
,
16
,
331
348
.
Shackman
,
A. J.
, &
Fox
,
A. S.
(
2016
).
Contributions of the central extended amygdala to fear and anxiety
.
Journal of Neuroscience
,
36
,
8050
8063
.
Smith
,
S. M.
, &
Nichols
,
T. E.
(
2009
).
Threshold-free cluster enhancement: Addressing problems of smoothing, threshold dependence and localisation in cluster inference
.
Neuroimage
,
44
,
83
98
.
Somerville
,
L. H.
,
Wagner
,
D. D.
,
Wig
,
G. S.
,
Moran
,
J. M.
,
Whalen
,
P. J.
, &
Kelley
,
W. M.
(
2012
).
Interactions between transient and sustained neural signals support the generation and regulation of anxious emotion
.
Cerebral Cortex
,
23
,
49
60
.
Somerville
,
L. H.
,
Whalen
,
P. J.
, &
Kelley
,
W. M.
(
2010
).
Human bed nucleus of the stria terminalis indexes hypervigilant threat monitoring
.
Biological Psychiatry
,
68
,
416
424
.
Spielberger
,
C. D.
,
Gorsuch
,
R. L.
, &
Lushene
,
R. E.
(
1970
).
The State Trait Anxiety Inventory manual
.
Palo Alto, CA
:
Consulting Psychologists
.
Torrisi
,
S.
,
Gorka
,
A. X.
,
Gonzalez-Castillo
,
J.
,
O'Connell
,
K.
,
Balderston
,
N.
,
Grillon
,
C.
, et al
(
2018
).
Extended amygdala connectivity changes during sustained shock anticipation
.
Translational Psychiatry
,
8
,
33
.
Treynor
,
W.
,
Gonzalez
,
R.
, &
Nolen-Hoeksema
,
S.
(
2003
).
Rumination reconsidered: A psychometric analysis
.
Cognitive Therapy and Research
,
27
,
247
259
.
Walker
,
D. L.
,
Toufexis
,
D. J.
, &
Davis
,
M.
(
2003
).
Role of the bed nucleus of the stria terminalis versus the amygdala in fear, stress, and anxiety
.
European Journal of Pharmacology
,
463
,
199
216
.

Author notes

*

These authors contributed equally to this work.