One of the principal goals of affective neuroscience is to delineate neurobiological responses to emotional challenges. Typically, this goal has been pursued in laboratory contexts by inducing emotion and then examining changes in subjective experience (Dimberg, 1987
; Hubert and de Jong-Meyer, 1991
; Magai et al., 2006
), peripheral physiology (Sequeira et al., 2009
; Vrana et al., 1988
) and neural responses (Carretie et al., 2009
; Hagemann et al., 2003
). For instance, conditioning studies have explored manifestations of fear responses to aversive stimuli, and dozens of studies have demonstrated increased skin conductance and brain activation in response to electric shock, loud noise and other aversive stimuli (e.g. (Cheng et al., 2003
; Cheng et al., 2007
; Dunsmoor et al., 2007
; Fischer et al., 2002
; Knight et al., 2004b
; LaBar et al., 1998
; Phelps et al., 2004
); for review see (Sehlmeyer et al., 2009
Although this work is of fundamental importance, one limitation has been the frequent confounding of emotional responses inherent in the psychological representation of the event with the cascade of responses that take place once the body is actually enduring a challenge. For example, the anticipatory anxiety that happens while one awaits a visit to the dentist is importantly distinguishable from the responses that occur once the drilling has begun. It is crucial to parse these two components and carefully examine the anticipatory period before the aversive assault because 1) much of our emotional life is spent in the anticipation of future events, and 2) this window allows a purer examination of the impact of the emotional representation of a stressor in the absence of the responses to the stressor itself. Unfortunately, less is known about the anticipatory component of emotional responding than is known about the responses that occur once the demanding situation is actually unfolding. Still less is known about the way that individual difference variables moderate these anticipatory responses.
While the vast majority of fear conditioning studies have employed a strategy in which a cue is presented and then nearly immediately (e.g. less than 500 ms) followed by an aversive stimulus, a handful of neuroimaging studies in humans have employed a “trace” conditioning strategy in which the cue is followed by a waiting period (e.g. from 500 ms to 8 s) before the aversive stimulus is deployed (Buchel et al., 1999
; Carter et al., 2006
; Cheng et al., 2008
; Cheng et al., 2006
; Knight et al., 2004a
). Unfortunately, most of these paradigms have not analyzed the waiting period separately from the receipt of the shock, making it difficult to interpret which element (anticipation or receipt of shock) is driving the findings. The literature on pain anticipation has addressed anticipatory emotional responses more directly, often using longer anticipatory delay periods, and a variety of aversive stimuli to induce anticipatory responses. Early studies of anticipatory anxiety have typically examined anticipatory effects on autonomic responding. For example, anticipation of electric shock (Chua et al., 1999
; Kopacz and Smith, 1971
) and venipuncture (Geddes et al., 1993
) have been shown to lead to increased skin conductance responses (SCRs), and potentiate protective reflexes such as the eyeblink startle reflex (Grillon et al., 1993
; Grillon et al., 1991
; Vrana et al., 1988
A seminal study by Ploghaus and colleagues was the first neuroimaging study to disentangle the anticipation and receipt of pain, and they found a separable network of regions invoked purely by emotional distress in the absence of external stimulation (Ploghaus et al., 1999
). Subsequent studies have begun to elucidate the neural correlates of anticipatory anxiety using standard threat of shock tasks in which shocks were delivered (Chua et al., 1999
; Schunck et al., 2008
; Straube et al., 2009
), or not (Kumari et al., 2007
), as well as threat of heat stimuli (Wager et al., 2004
), or upsetting images (Nitschke et al., 2006
; Waugh et al., 2008
). These studies typically show increased activation in bilateral insula, anterior cingulate cortex (ACC), dorsomedial prefrontal cortex (dmPFC) and ventrolateral prefrontal cortex (vlPFC) (Mee et al., 2006
). These brain regions have been associated with representing visceral body states and conscious feelings related to interoceptive processes (insula), integrating contextual cues and sensory information (ACC), self-referential processing (dmPFC), and representing the affective stimulus value and expectation of negative outcomes (vlPFC). Only one relatively small study (N = 16) has examined the relationship between varying levels of shock intensity and anticipatory responses in the brain (Straube et al., 2009
), with findings that suggest that anticipatory processes are not “all or nothing”, but are instead modulated by the intensity of the anticipated threat.
There is considerable individual-related variation in anticipatory anxiety. One particularly important individual-difference dimension for understanding variation in negative emotional responses is neuroticism. Compared to individuals low in neuroticism, individuals high in neuroticism are more prone to anxiety and negative affect, respond to environmental stressors more negatively (Costa and McCrae, 1980
), and are particularly concerned with averting possible threats (Zelenski and Larsen, 1999
). Moreover, neuroticism has been associated with an avoidant coping style (Bolger, 1990
; McCrae and Costa, 1986
; Parkes, 1986
) and increased levels of suppression (Gross and John, 2003
), Neuroimaging studies have focused on the impact of neuroticism on responses during exposure to aversive images and faces, and have found exaggerated neural responses (Canli, 2004
; Canli et al., 2001
; Carroll et al., 2007
; Haas et al., 2007
) in the left middle temporal gyrus, middle frontal gyrus, and insula. The autonomic literature, however, has been mixed, with some studies showing increased electrodermal responding to aversive images (Norris et al., 2007
) and others showing decreased responding (De Pascalis et al., 2007
). It is less clear whether neuroticism affects anticipatory responding before the stimulus actually occurs. Only one study to date has used fMRI to examine neuroticism in response to threat of noxious stimuli (Kumari et al., 2007
), finding that neuroticism was negatively correlated with a number of prefrontal and parietal regions during threat greater than safe conditions in a brief task in which shocks were not delivered. Given that neuroticism has been identified as a risk factor for anxiety disorders, many of which are characterized by chronic worry and anxiety about the future (Barlow, 2002
; Bienvenu and Stein, 2003
), neuroticism may modulate anticipatory anxiety even in healthy individuals.
The goal of the present study was to investigate how anticipation of varying levels of shock intensity impacts anxiety experience, autonomic responding, and neural activity. To address this goal, we employed a task in which a long anticipatory cue period (7-11 s) allowed for robust responses to be generated. To create maximal levels of anxiety with minimal levels of habituation, three layers of unpredictability were embedded into the shock trials: event (whether the shock will occur or not), temporal (when will it occur), and intensity (how strong will it be) unpredictability. These levels of unpredictability have been shown to potentiate emotional reactivity (D'Amato and Gumenik, 1960
; Monat et al., 1972
), autonomic (Geer and Maisel, 1972
) and neural responding (Carlsson et al., 2006
). Multiple levels of shock intensity permitted investigation of a dynamic range of responses. In order to disentangle anticipatory anxiety from shock response, the anticipatory period was analyzed separately from the receipt of shock. A relatively large number of trials were employed to enhance power to detect effects, and a large sample of healthy women was employed to permit investigation of individual differences in neuroticism without confounding effects of gender or previous history of psychopathology. Multiple output channels were investigated to better elucidate the richness of anticipatory emotional responses.
Based on the extant literature on anticipatory anxiety, we hypothesized that as shock intensity increased, anticipatory anxiety would result in stepwise increases in 1) anxiety experience, 2) electrodermal responding, and 3) neural activity in brain regions including the insula, anterior cingulate, thalamus, and prefrontal cortex (but not in the amygdala). Based on studies showing exaggerated reactivity during negative stimulus presentation due to neuroticism, we further hypothesized that individuals high in neuroticism would show increases in anticipatory responding before stimulus presentation. Specifically, we expected greater increases in anxiety, electrodermal responding and brain activity in the middle temporal gyrus, frontal gyrus and insula relative to those low in neuroticism.
Study 1: Experiential and Autonomic Responses to Shock Anticipation
In Study 1, we devised a shock anticipation task that involved anticipating electric shocks to the wrist in three different intensity levels: safe (no shock), medium shock, and strong shock. To increase anxiety and prevent habituation, three layers of uncertainty were embedded into the shock trials: event, temporal, and intensity uncertainty. The task was administered in a sample large enough to examine the potential effect of neuroticism. We sought to examine the effects of shock anticipation on anxiety experience and electrodermal responses, and to test whether neuroticism moderated these responses.
All potential participants were screened using an interview based on the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders (SCID for DSM-IV) (First et al., 1995
). Eligible participants were healthy females that did not meet criteria for any psychiatric disorder within the past year, or for lifetime posttraumatic stress, bipolar, obsessive–compulsive, or psychotic disorders, and were not currently taking psychotropic medications. Ninety-five women (mean age 21.8 ± 2.2) participated in the anticipatory anxiety task. The study was approved by the Stanford University Institutional Review Board, and all subjects gave informed consent and were paid for participation.
Participants underwent a shock calibration procedure and training phase before beginning the task. In preparation for the task, participants received a series of brief (50 ms) electric shocks delivered to the left wrist above the median nerve with a Grass SD-9 stimulator (West Warick, RI) and were instructed to set a maximum voltage (MV) that was “painful and difficult to tolerate but not unbearable.” A brief training phase followed which mirrored the task, however without actual shock presentation.
The shock anticipation task consisted of three conditions presented in a pseudorandom order with no more than two sequential trials of the same condition: safe (8 trials), medium shock (13 trials), strong shock (13 trials). Conditions were indicated by the words “safe” (on a green background), “medium” (on an orange background), and “strong” (on a red background). During safe trials, participants were instructed that no shocks would be administered. Instructions for the medium and strong trials featured three levels of uncertainty which were explained during the training phase. Event uncertainty was implemented by indicating that shocks would only be delivered during 85% of the trials. Temporal uncertainty by was implemented by indicating that trials would last between 0-20 seconds and shocks could occur at any time; in actuality, trials ranged between 7-11 s (avg 9 s) and terminated with shock in 85% of the trials; both medium and strong conditions contained a single “quick” trial terminated with a shock at 3 s in order to maintain the claim that shocks could occur at any time. Intensity uncertainty was implemented by leaving the exact strength of the shock unspecified within a 20% window; for medium trials the range was indicated as 40-60% of MV and for strong trials the range was indicated as 70-90% of MV (in actuality, medium trials were always at 55% of MV and strong trials were always at 85% of MV).
Measures Continuous anxiety ratings
Continuous ratings of emotional experience were obtained using a rating dial similar to that used previously (Mauss et al., 2005
; Ruef and Levenson, 2007
). The dial allows participants to move a pointer along a 180° scale, with the anchors “no anxiety ‘0’” at 0° and “very high anxiety ‚10’” at 180°. Participants were asked to adjust the dial position as often as necessary to reflect the current amount of anxiety they felt.
Participants’ skin conductance (SC) was recorded from the middle phalange of the index and middle finger of the left hand using 11-mm inner diameter Ag/AgCl electrodes filled with isotonic electrode paste (TD-246, Med Associates, Inc., St. Albans, VT, USA)
to provide a continuous index of sympathetic activity (Boucsein, 1992
The neuroticism scale (N) of the NEO-PI-R (Costa and McCrae, 1992
) was administered on a day prior to the anticipatory anxiety task. The neuroticism scale assesses an individual's tendency to experience psychological distress and sensitivity to negative cues in the environment. In keeping with previous literature showing the utility of dissociating the anxious and depressive aspects of neuroticism, the score for the N1 “anxiety” scale was utilized in subsequent analyses (mean score = 21.4 +/- 5.7 SD) (Haas et al., 2007
Data Reduction and Analysis
All analyses only examined responses during the anticipatory cue period, up to the point but not including the shock. The two “quick trials” where not analyzed since the 3 second window may not be long enough to obtain meaningful responses, leaving 12 medium and strong trials for analysis respectively. Using ANSLAB software (Wilhelm and Peyk, 2005
), we calculated skin conductance responses (SCRs) by determining the maximum SC during each trial (prior to the shocks) from which the average SC during a 1 s baseline before each cue onset was subtracted. SCRs were then averaged across trials within each condition. Negative SCRs were recoded to zero. Rating dial values were averaged across the whole cue phasei
We first analyzed main effects of the task on continuous anxiety ratings and SCRs in one way ANOVAS (safe, medium, strong). Next, we compared participants with high and low neuroticism (as determined by median split) in their responses to the three conditions in a Condition (safe, medium, strong) X Group (high vs. low neuroticism) ANOVA for anxiety ratings and SCRs.
Results and Discussion
Effects of Task Continuous anxiety ratings
A one-way ANOVA for anxiety ratings yielded an effect of Condition, F(2,178)=228.16, p<.001, partial eta squared = .72 (see ). Bonferroni-corrected paired t-tests indicated that anxiety ratings displayed the expected dose-response differences between the three conditions (strong > medium > safe) (all ps<.001).
Figure 1 Study 1: Effects of shock anticipation on (A) continuous ratings of anxiety and (B) skin conductance response across three levels of shock intensity. Paired t-tests indicated significant differences between the conditions for both (A) and (B) (all ps (more ...) Electrodermal responses
A one-way ANOVA for SCRs resulted in an effect of Condition, F(2, 188)=81.57, p<.001, partial eta squared=.47 (see ). Bonferroni-corrected paired t-tests indicated that SCRs displayed the expected dose-response differences between the three conditions (strong > medium > safe) (all ps<.001).
Effects of Neuroticism Continuous anxiety ratings
In addition to the task main effects reported above, the Condition (safe, medium, high) X Group (high neuroticism, low neuroticism) ANOVA for continuous anxiety ratings revealed significant main effect for Group F(1, 88)=8.58, p=.004, partial eta squared=.09 (see ), but not their interaction F(2, 176)=0.466, p=.54, partial eta squared=.005. High neuroticism participants showed elevated levels of subjective anxiety compared to their low neuroticism counterparts across conditions.
Study 1: The effect of neuroticism (median split) on continuous ratings of anxiety. Repeated-measures ANOVAs showed a significant effect of neuroticism on anxiety ratings (p=.004).
The Condition (safe, medium, high) X Group (high neuroticism, low neuroticism) ANOVA for SCRs was not significant for Group, F(1,93)=.247, p=.621, partial eta squared=.003 or their interaction, F(2, 186)=.025, p=.9, partial eta squared=0.
As expected, the results from Study 1 revealed a stepwise increase in both anxiety reports and electrodermal responding from safe to medium to strong trials. This suggests that the shock anticipation task is sufficiently potent to induce anticipatory anxiety and engage preparatory responses while awaiting an aversive event. It is also clear that individuals higher in neuroticism showed elevated anxiety reports in all three task conditions, but there were no differences in electrodermal responses.
Study 2: fMRI BOLD Responses During Threat Anticipation
In Study 2, we sought to extend the findings from Study 1 by examining the effects of shock anticipation on neural responses during BOLD neuroimaging. We hypothesized that increasing levels of threat would be associated with increasing neural responses in brain regions implicated in visceral perception (insula and thalamus), motor preparation (caudate), assigning affective stimulus value, and self-referential processing (vlPFC and dmPFC). In this independent sample, we further sought to investigate whether neuroticism moderated neural responses to the shock anticipation task. We hypothesized that individuals higher in neuroticism would show enhanced neural responses during shock anticipation.
In Study 2, 51 right-handed healthy female individuals (mean age 22.2 +/- 2.4) participated in the shock anticipation task during fMRI scanning. Participants met the same eligibility criteria described in Study 1, as well as additional criteria to ensure magnet compatibility. The study was approved by the Stanford University Institutional Review Board, and all subjects gave informed consent and were paid for participation.
The procedure for the fMRI task was the same as described in Study 1. The task varied slightly in that it contained 12 safe, 13 medium, and 13 strong trials, and the inter stimulus interval (ISI) varied from 3 to 6s. As with Study 1, the “quick” trials were dropped from analysis, leaving 12 trials per each of the three conditions. Rather than providing continuous ratings of anxiety, at the end of each medium and strong shock trial, participants provided ratings of the anxiety that they experienced during the anticipatory period prior to the delivery of the shock, using a scale of 1=not at all to 5=very much. The ratings for 3 subjects were not obtained due to software malfunction. The trial structure was as follows: condition cue (7-11 s), shock (50 ms, only on a subset of trials), rating (6 s, only after medium and strong trials), ISI (3 to 6 s).
Imaging was performed on a GE 3 Tesla Signa magnet (General Electric Medical System, Milwaukee, Wisconsin). Blood oxygenation-level dependent (BOLD) signal was acquired with a T2*-weighted gradient echo spiral-in/out pulse sequence (Glover and Law, 2001
) and a custom-built quadrature “dome” elliptical bird cage head coil. Head movement was minimized using a bite-bar. Four hundred and forty-six functional volumes were obtained during the functional run from 22 sequential axial slices (TR/TE = 1500/30 msec, flip angle = 70°, FOV = 22 cm, matrix = 64 × 64, single-shot, in-plane resolution = 3.438 mm2
, and slice thickness =4.5 mm). Three-dimensional high-resolution anatomical scans were acquired using fast spin-echo spoiled gradient recalled (SPGR) (.859 x 1.2 mm; FOV = 22 cm, frequency encoding = 256).
Neuroticism scores were obtained using the same method described in Study 1. In Study 2, the mean neuroticism N1 score = 20.71 +/- 4.75 SD. Neuroticism scores did not differ significantly between participants in Study 1 and Study 2.
Data Reduction and Analysis
Functional data were analyzed with Analysis of Functional NeuroImages (AFNI) software (Cox, 1996
). Preprocessing included coregistration, motion correction, 4 mm3
isotropic Gaussian spatial smoothing, high-pass filtering (.011 Hz), and linear detrending. Only volumes which demonstrated less than 1 mm of motion in the x, y, and z directions were included. Three participants exhibited volumes with motion above this threshold; these subjects were removed from all subsequent analyses, leaving 48 participants for these analyses. There was no evidence of stimulus-correlated motion when conducting correlations of condition-specific reference functions and x, y, or z motion correction parameters (all ps > .05).
A multiple regression model implemented with AFNI 3dDeconvolve included baseline parameters to remove polynomial trends to the fifth order, as well as individual motion-related variance in the BOLD signal in six orientations. The model also included the following regressors: safe period, medium anticipation period, strong anticipation period, medium shock, strong shock, medium quick trial, strong quick trial, ratings after medium trials, and ratings after strong trials, all of which were convolved with the gamma variate model of the hemodynamic response function. Contrasts were computed by weighting the appropriate columns in the design matrix accordingly.
As with Study 1, brain analysis only examined the anticipatory period, up to but not including the shock (the separate shock regressors accounted for variation due to the shock itself). The first contrast compared safe trials vs. shock anticipation (medium and strong anticipation periods grouped together). The second contrast compared medium vs. strong anticipation trials. In order to determine which regions had a true linear effect across all three conditions, a conjunction analysis was performed on the second contrast such that the main effects from the first contrast (thresholded at corrected p < 0.05) were used to mask the results from the second contrast. The resultant map showed brain regions where responses increased on average from safe to medium to strong trials. Statistical maps were resampled to 3.438 mm3
and converted to Talairach atlas space (Talairach and Tournoux, 1988
), and second-level statistical parametric maps were produced according to a random-effects analysis to enhance the generalizability of the results.
To correct for multiple comparisons, AlphaSim, a Monte Carlo simulation bootstrapping program in the AFNI library (Cox, 1996
), was employed to identify a joint probability consisting of a voxel-wise threshold and a minimum cluster volume threshold to establish a cluster-wise p-value that protects against false-positive detection of activation clusters (Forman et al., 1995
). A voxel-wise threshold of p = .001 (t = 3.50) resulted in a minimum cluster volume threshold of 343 mm3
(8 voxels) to protect against false-positive detection of clusters of activation at p < .05. All clusters cited in this paper survived this correction.
The relationship between neuroticism and BOLD responses was investigated using a whole-brain regression analysis across participants in which individual neuroticism scores were entered as the regressor of interest and examined for the contrast of (strong shock anticipation > medium shock anticipation) using 3dRegAna from the AFNI library (Cox, 1996
). The same multiple comparisons threshold was employed as above.
Results and Discussion
Effects of Task Anxiety ratings
A paired samples t-test was used to compare anxiety ratings made in the medium and strong conditions, and results indicated that ratings in strong trials were significantly higher than medium trials (medium M=2.3, SD =0.65, strong M=3.5, SD =0.89) t(45)= -15.69, p < .001. There were no ratings provided after safe trials.
The contrast (shock anticipation > safe) revealed brain regions including bilateral anterior insula, mid cingulate cortex, anterior cingulate cortex, bilateral dorsomedial PFC, bilateral ventrolateral PFC, bilateral thalamus, bilateral caudate, bilateral precentral gyrus, hypothalamus, and periaqueductal gray. The contrast (strong shock anticipation > medium shock anticipation) masked by the contrast (safe vs. shock anticipation) revealed a subset of regions including bilateral anterior insula, mid cingulate cortex, anterior cingulate cortex, inferior frontal gyrus and culmen (see and ).
Figure 3 Study 2: Effects of shock anticipation on neural responses for (A) the contrast shock anticipation > safe (starting top left and moving clockwise, slices are displayed at x=36, x=0, z=40 , z=11) and (B) the conjunction analysis with strong shock (more ...)
Table 1 Study 2: Neural regions exhibiting an effect of shock anticipation for the contrast shock anticipation > safe (top), and the conjunction analysis with strong shock anticipation > medium shock anticipation (bottom) at the threshold p < (more ...) Effects of Neuroticism Anxiety ratings
A one-way ANOVA for anxiety ratings indicated that anxiety ratings did not differ based on neuroticism (median split) for medium or strong trials (ps > .55).
We examined the association between neuroticism and neural responses during the shock anticipation task using a regression with the contrast (strong shock anticipation > medium shock anticipation). We found that neuroticism was negatively correlated with activation in left BA 46 and left insula (see and ). For display purposes, individual subjects’ mean contrast value in BA 46 for each condition are plotted based on neuroticism, defined by a median split ().
Table 2 Study 2: Neural regions exhibiting a correlation with neuroticism at corrected p<.05, T=3.5. A whole-brain regression analysis was performed on the contrast strong shock anticipation > medium shock anticipation. Coordinates in Talairach (more ...)
Figure 4 Study 2: Neural response to threat varies as a function of neuroticism. (A) the relationship between BOLD response to each condition and neuroticism (median split) in right BA 46 (brain slice depicted in (B), the contrast strong shock anticipation > (more ...) Summary
As predicted, the results from Study 2 parallel those from Study 1, indicating an increase in anxiety experience from medium to strong shock trials. Study 2 extends these findings by showing that as threat level increases, there are concomitant increases in brain activation in a network of regions involved in visceral perception and integration, motor preparation, context interpretation, and assigning affective value. A subset of these regions show a linear response from safe to medium to strong trials. This study did not show an effect of neuroticism on anxiety ratings, likely due to the reduced sensitivity of the ratings used in Study 2 (continuous ratings vs. punctuate ratings on a five point scale). The results do indicate, however, that relative to individuals who are low in neuroticism, those high in neuroticism show a reduction in brain activation based on threat level in key brain regions.