Impact of induced anxiety on neural responses to monetary incentives

Abstract

Previous research demonstrates that aversive stimuli can interrupt appetitive processing and that brain regions involved with the processing of potential rewards, such as the ventral striatum (VS), also respond to threatening information. Potential losses can likewise activate the VS and, thus, the full extent to which threat can impact neural responses during incentive processing remains unclear. Here, unpredictable threat of shock was used to induce anxiety while participants performed the monetary incentive delay (MID) task during functional magnetic resonance imaging (fMRI). During anticipation, anxiety impacted neural responses within the bilateral VS and distributed regions of the occipital cortex. Anxiety enhanced activity within the VS to both gain and loss trials. Furthermore, anxiety enhanced activity to both gain and loss trials within dorsal areas of BA19. However, anxiety only enhanced activity during gain, but not loss trials, within ventral areas of BA19. These results suggest that during anticipation, induced anxiety enhanced VS activity to incentives generally, which might reflect changes in the subjective salience of gains and losses. Collectively, these results suggest that the impact of induced anxiety on responses to monetary incentives depend on the neural region, type of incentive, and stage of processing.

Introduction

Evidence from associative learning models demonstrates that potentially harmful stimuli inhibit behavioral responses to rewards (Dickinson and Pearce, 1977), which has led to the suggestion that appetitive and aversive stimuli compete during neural processing (Choi et al., 2014). However, although appetitive and aversive responses are in direct competition on a behavioral level, neural responses to motivational stimuli depend on the function of the corresponding neural structure. While abundant translational research has demonstrated that regions of the ventral striatum (VS), including the nucleus accumbens, are responsible for the motivational processing of both positive and negative stimuli (Knutson and Greer, 2008; Berridge et al., 2009). VS Blood oxygen level-dependent (BOLD) activity can reflect different neural processes (Dreher et al., 2006; Preuschoff et al., 2006), which has led to the suggestion that single-factor theories may be unable to fully explain patterns of VS activity (Cooper and Knutson, 2008). Valence refers to the relative positive vs negative quality of a stimulus (Lewis et al., 2007; Knutson and Greer, 2008), while the term salience refers to the psychological prominence of a stimulus. VS BOLD responses can reflect either the valence or salience of motivational stimuli, depending on the context in which the stimuli are presented (Cooper and Knutson, 2008). Interactions between stimulus type and psychological context (e.g. anxiety) may be important for understanding exaggerated neural activity within the striatum in persons diagnosed with Social Anxiety Disorder (SAD) and Post-traumatic Stress Disorder (PTSD; Guyer et al., 2012; Patel et al., 2012; Hattingh et al., 2013). A deeper understanding of the neural mechanisms underlying the interactions between appetitive and aversive processing may also shed light on the role of the striatum in processing motivational stimuli.

Although previous research has sought to address the neural basis of interactions between threat and reward processing, several factors have complicated the search for broader insights. First, responding to motivational stimuli involves multiple stages of information processing (Knutson et al., 2001; Ernst and Paulus, 2005; Richards et al., 2013). Neural responses preceding the delivery of a reward (i.e. anticipation stage) differ from neural responses to the delivery of the reward itself (i.e. outcome). Second, both the anticipation of potential gains and losses elicit activity within the VS (Knutson and Greer, 2008). As such, it is important to understand whether threatening information impacts the neural processing of rewards specifically or the neural processing of incentives generally. Finally, the same monetary outcome can be the result of success or failure in attaining the desired result, depending on the type of gamble. This formulation refers to psychological framing (Tversky and Kahneman, 1981), and research has demonstrated that the coding of outcomes differs as a function of framing (Hardin et al., 2009).

Here, we set out to determine how induced anxiety impacts neural processing during the monetary incentive delay (MID) task. Although, all types of aversive stimuli elicit defensive responses, important distinctions exist between responses to predictable vs unpredictable threats. In keeping with previous research (Davis et al., 2010; Fox and Shackman, 2017), we operationally define anxiety as a response elicited by an unpredictable aversive stimulus. Previous research utilizing a modified MID task showed that threatening stimuli reduce striatal responses during reward anticipation (Choi et al., 2014). However, important factors distinguish Choi et al.’s design from the present one. First, Choi et al. manipulated threat by delivering aversive electrical stimuli during the target stimulus on a trial by trial basis. The present design utilizes unpredictable electrical stimuli to induce a prolonged anxious state across blocks of trials. Second, Choi et al. did not employ loss trials. The inclusion of loss trials in our design allows us to determine if induced anxiety impacts reward processing specifically or incentive processing generally. Lastly, Choi et al. only assessed neural responses during the anticipation period and did not analyze neural responses to the outcomes of incentivized trials. Thus, the present design allows us to determine how induced anxiety impacts the anticipation and outcome of both positive and negative incentives.

During an anxious state, potential gains may be perceived as less positive and potential losses may be perceived as more negative. However, the motivation to achieve rewards and avoid losses may be enhanced during a negative anxious state, leading to enhanced subjective salience of motivational stimuli. VS activity reflects the valence of stimuli when outcomes are certain but reflects the salience of stimuli during uncertain outcomes (Cooper and Knutson, 2008). In the present design, the outcome is determined by task performance, which is uncertain, since the task is individually tailored to yield 66% successful responses. Thus, we predict that induced anxiety will enhance neural activity to both gains and losses within the VS during the anticipation period, reflecting enhanced salience of positive and negative monetary incentives.

For the outcome period, we hypothesize that induced anxiety will differentially impact neural responses to positive and negative outcomes. The medial prefrontal cortex (mPFC) responds to the delivery of gains and losses during the MID task (Dillon et al., 2008), and previous research suggests that the mPFC tracks the subjective value of both food-based and monetary rewards (Levy and Glimcher, 2011; Bartra et al., 2013). Anxious states may decrease participants’ subjective valuation of monetary gains, and we predict that mPFC responses to gains will be attenuated in the anxious condition. In contrast, induced anxiety may enhance the subjective negative value of losses, and in turn, enhance neural responses to losses within the mPFC. Furthermore, based on previous research demonstrating differential roles of the ventral and dorsal striatum in processing gains and losses, respectively (Seymour et al., 2007), we tested for interactions between anxiety and trial type in predicting neural responses during successful and unsuccessful outcomes from a priori regions of interest within the striatum. Finally, we tested whether state anxiety at baseline and gender influenced the impact of anxiety on neural responses to incentives within the striatum.

Methods

Participants

Sixty-five right-handed volunteers were recruited via flyers, print advertisements and internet listservs for participation in this study. Written informed consent was obtained from all participants and approved by the National Institute of Mental Health (NIMH) Combined Neuroscience Institutional Review Board. All participants were free from the following exclusion criteria: (i) current or past Axis I psychiatric disorder as assessed through a clinician administered SCID-I/NP (First et al., 2001), (ii) first-degree relative with a psychotic disorder, (iii) medical condition conflicting with safety or design of the study, (iv) brain abnormality on magnetic resonance imaging (MRI) as assessed by a radiologist, (v) positive toxicology screen, or (vi) MRI contraindication. Fifteen participants in total were excluded from functional magnetic resonance imaging (fMRI) analysis (see Supplementary Material: Methods: Exclusion Criteria, for details) and the final sample consisted of 50 participants (M = 26.4 ± 5.08 s.d. years; 25 women). Participants who made key responses outside of the possible range (i.e. keys: 1–7) for self-reported valence and arousal were excluded from those analyses, respectively (valence N = 49; arousal N = 48).

Experimental procedure

Following informed consent, participants completed a state anxiety inventory [Spielberger State–Trait Anxiety Inventory–State version (STAI-S)] and were instructed how to perform the MID task. Participants were told that they would be endowed with $20.00 and that it was possible to increase or decrease this amount based on their performance during the experiment. Participants were told that they would perform the MID task twice and that they would receive the average of their final endowment from the two runs of the MID paradigm, in addition to their compensation of $175.00 for participation in the experiment. Following the completion of the MID paradigm, participants were asked to report which geometric shapes served as cues for which trials. Participants were then debriefed and thanked for their participation.

Shock procedure

To account for individual differences in subjective experience of the shock, stimulation intensity was calibrated for each participant prior to the task. Participants used a 1–10 scale to identify a level of electrical stimulation that was `highly annoying but not painful’. Participants were told that, during anxiety conditions of the task, they could receive an electrical stimulus at any time. Electrical stimuli were delivered for 100 ms to the subject’s left wrist using a Digitimer constant current stimulator (DS7A; Digitimer, UK) via 2 Ag/AgCl 6 mm electrodes.

MID paradigm

Trials consisted of three components (Figure 1): (i) a geometric shape (e.g. a circle, triangle, or square) serving as a cue to denote gain, loss, and neutral trials, respectively; (ii) a white square serving as a target; and (iii) an outcome screen informing participants as to whether they responded appropriately (i.e. `Success!’ or `Failure!’), the monetary consequences of their performance (i.e. gain: `You won money!’ or `You’re total is unchanged!’; loss: `You did not lose money!’ or `You lost money!’; even: `You’re total is unchanged!’) and their current endowment. Participants were endowed with $20.00 at the beginning of each run. Success during gain trials resulted in $5.00 being added to the current endowment, while failure during loss trials resulted in $5.00 being subtracted from the current endowment.

The cue was presented for 250 ms, followed by a jittered interval (2500 ± 500 ms) before the presentation of the target. The target was initially displayed for 250 ms, and following the third presentation of that trial type the duration of the target was titrated. Specifically, 20 ms was added or subtracted from the response window to achieve 66% accuracy across each trial type in each condition. After the presentation of the target, a jittered delay (1500 ± 500 ms) preceded the presentation of the outcome screen (2000 ms duration). A jittered interval (3000 ± 1000 ms) was used between trials. The average duration of each trial, including the inter-trial interval, was 9500 ms.

At the beginning of each block, participants were presented with a screen which stated `This block will be a Safe/Shock block’ for 2000 ms. A yellow or a blue border surrounded the screen during either the safe or threat blocks (counterbalance across participants). Participants were informed about which cues signaled which trial types and which border colors signaled which condition.

Participants completed two runs of the MID paradigm. Each run contained six blocks, alternating between the two conditions (three safe and three anxiety), and each block included nine trials for analysis. The order of safe and anxiety blocks was counterbalanced across runs and across participants. Electrical stimuli occurred during a single trial within one of the three anxiety blocks during each run. Electrical stimuli were presented during either the anticipation period or during the outcome period, counterbalanced across runs. Blocks, during which a shock was administered, consisted of 11 trials, of which two trials, one containing the shock and the subsequent trial were excluded from analysis (i.e. censored trials). The average length of each block was 85.5 s (while the block containing the `censored trials’ was 104.5 s). The total run of the task lasted for 632.5 s.

During one run of the experiment, one electrical stimulus was delivered immediately following the incentive cue for a gain trial during the third anxiety block of the experiment. During the other run of the experiment, participants received either one (N = 24) or two (N = 26) electrical stimuli, due to a missing inline script within the Eprime program. Participants who did not respond successfully to the target received one electrical stimulus immediately before the presentation of the feedback screen for a neutral trial during the second anxiety block. Participants who did successfully respond to the target received one electrical stimulus immediately preceding the presentation of successful feedback, in addition to a second electrical stimulus preceding the presentation of an errant unsuccessful feedback screen for the same trial. Importantly, the exclusion of the censored trials from analysis ensured that all trials included in the analysis occurred at least 10 s after the delivery of the electrical shock. To ensure that between subject differences in the number of electrical stimuli delivered did not impact our observed pattern of results, we directly tested whether self-reported subjective state, behavioral responses or neural activity were impacted by the total number of shocks delivered during the task (i.e. two vs three electrical stimuli). Although the group that received three shocks exhibited nominally slower reaction times during gain trials, self-reported subjective state and patterns of neural activity were not impacted by the number of electrical stimuli administered (see Supplementary Material: Results: Between Subject Differences in Shock Number and Supplementary Figures S1–S5 for full details). Therefore, we chose to combine both groups of participants within our main results section.

Self-report of subjective state

Participants completed the STAI-S (Spielberger et al., 1999) prior to entering the scanner and before receiving any electrical shocks and again following the completion of the MID paradigm. Self-reported valence and arousal during the MID task was assessed retrospectively using seven-point Likert scales (see Supplementary Methods for details).

BOLD fMRI data acquisition

MRI data were acquired using a GE 3 T scanner and 8-channel head coil. BOLD signal was acquired using T2*-weighted echo-planar imaging (EPI) across 45 ascending interleaved axial slices using the following parameters: TR = 2500 ms, TE = 30 ms, flip angle = 65°, acquisition matrix = 72 × 72, 3×3×2.5 mm voxels, 253 volumes. A 1.0 mm isotropic T1 weighted anatomical image was also collected using a gradient-echo sequence (T = 7796 ms; TE = 2.984; flip angle, 7°; acquisition matrix = 256 × 256).

BOLD fMRI pre-processing and analysis

FreeSurfer (Fischl et al., 2002) was used to generate white matter and ventricular images from the T1 weighted anatomical scan. Ventricular and white matter images were then resampled to 2.5 mm isotropic images and were eroded by one voxel in each direction to ensure they did not overlap with gray matter. The rest of pre-processing and analysis was performed using Analysis of Functional NeuroImages (AFNI; Cox, 1996). The first four functional volumes were discarded to allow for steady-state equilibrium. Using the afni_proc.py python script, functional volumes were slice time corrected, re-aligned to the last function volume, co-registered to the corresponding T1 weighted anatomical image and non-linearly warped to the MNIa_caez_colin27_T1_18 template. Functional volumes were resampled to 2.5 mm isotropic voxels and were smoothed using a 6 mm full width at half maximum (FWHM) kernel.

Predictors of interest for the first-level general linear models entailed 23 regressors of interest including six regressors (safe: gain, loss, and neutral; anxiety: gain, loss, and neutral) for three types of events: the anticipation period, successful outcomes and unsuccessful outcomes. The anticipation period was modeled starting with the presentation of the incentive cue and included the entirety of delay preceding the target. Outcomes were modeled from the onset to offset of the feedback screen (2000 ms) Regressors were also included for the onset and offset of safe and anxiety blocks (2000 ms) and for the censored trials (i.e. the trial on which the shock occurred and the subsequent trial). The following nuisance signals were then regressed from the functional volumes: (i) six head motion parameters and their first-order derivatives, (ii) the first three principal components from the time series extracted from the lateral ventricles, (iii) the first three principal components from the time series extracted from the third and fourth ventricles and (iv) local white matter signal from a 15 mm sphere surrounding each gray matter voxel using the ANATICOR approach (Jo et al., 2010). Adjacent functional volumes with a Euclidean norm motion derivative > 0.3 mm were censored from analyses, and participants with greater than 5% censored volumes were excluded from analysis. Beta-images were generated at the single subject level using linear mixed-effects modeling with restricted maximum likelihood estimates (Chen et al., 2013).

Statistical and analytic strategy

Reaction times to the target and self-reported arousal and valence ratings were analyzed using 3 (trial type: gain, loss, and neutral) × 2 (condition: safe and anxiety) repeated measure analysis of variance (rANOVA) models using SPSS v24. Repeated measure analysis of covariance (ANCOVA) models were used to assess how between subject differences in state anxiety at baseline and gender impact BOLD responses. Post hoc paired t-tests were used to probe trial type and interactions effects and are Bonferroni corrected for multiple comparisons (0.05/3 = 0.017).

Beta-images corresponding to (i) the anticipation period, (ii) successful feedback and (iii) unsuccessful feedback were entered into 3 × 2 rANOVA models (3dANOVA3) to test for effects of trial type, condition and their interaction. Parameter estimates were then extracted from resulting clusters and subjected to post hoc paired t-tests using SPSS v24.

All group level statistical maps were thresholded at P < 0.05, corrected across the whole brain within gray matter, based on cluster extent, using a cluster-forming threshold of (P < 0.0005, k = 22). Correction for multiple comparisons was performed via updated versions of 3dFWHMx and 3dClustSim which incorporate a mixed autocorrelation function to better model non-Gaussian noise structure (Cox et al., 2016; Eklund et al., 2016). A priori analyses of interactions between anxiety and trial type during successful and unsuccessful outcomes were performed on parameter estimates extracted from 5 mm spheres centered on ventral (± 12, 10, −1) and dorsal (±13, 10, 14) areas of the striatum (see Supplementary Figure S10 for details).

Results

Behavior

RANOVAs demonstrated that trial type impacted reaction times to the target stimulus [F(2,98) = 17.07, P < 0.0001] (Figure 2). Reaction times during gain trials [T(49) = 5.00, P < 0.001] and loss trials [T(49) = 3.62, P < 0.001] were faster than reaction times during neutral trials. Reaction times during gain trials were faster than during loss trials [T(49) = 2.67, P < 0.01]. Condition impacted reaction times, such that responses were slower during the anxiety compared to the safe condition [F(1,49) = 6.37, P < 0.05]. The interaction between condition and trial type was not statistically significant.

Self-report

State anxiety

Paired t-tests demonstrated that self-reported state anxiety (STAI-S) was enhanced [T(49) = 3.97, P < 0.001] immediately following the experiment (M = 28.12 ± 7.97 s.d.) compared to baseline (M = 23.86 ± 5.40 s.d.).

Valence

RANOVAs demonstrated that condition interacted with trial type to predict the self-reported liking of the target cues [F(2,96) = 3.27, P < 0.05] (Figure 3). Participants reported liking cues for gain trials [T(48) = 3.36, P < 0.005] and loss trials [T(48) = 2.77, P < 0.01] less during the anxiety condition. Participants self-reported liking of cues for neutral trials did not differ between the anxiety and safe condition [T(48) = 0.46, P = 0.65].

Arousal

RANOVAs demonstrated that trial type predicted self-reported arousal to the incentive cues [F(2,92) = 35.09, P < 0.0001], which was characterized by higher self-reported arousal to gain trials [T(47) = 9.08, P < 0.0001] and loss trials [T(47) = 3.74, P < 0.001] compared to neutral trials and higher arousal scores for gain trials compared to loss trials [T(47) = 4.76, P < 0.0001] (Figure 3). Furthermore, the effect of trial type was qualified by an interaction with the anxiety condition [F(2,92) = 3.18, P < 0.05], which was driven by a trend toward higher self-reported arousal for neutral trials during the anxiety condition [T(47) = 1.97, P = 0.055] but not gain or loss trials (all P’s > 0.15).

BOLD fMRI

Interactions between trial type and anxiety

Anticipation

RANOVAs demonstrated that during the anticipation of the target, an interaction effect was found within the bilateral striatum (Figure 4, Table 1) [left striatum: x = −11.2, y = 11.0, z = −5.2, F(2,98) = 15.34, kE = 32; right striatum: x = 13.8, y = 13.5, z = −0.2, F(2,98) = 17.71, kE = 69] which extended from areas of the VS into the head of the caudate. The interaction was driven by enhanced BOLD responses during the anxiety condition during the anticipation of gain [left VS: T(49) = 4.38, P < 0.001; right VS: T(49) = 4.62, P < 0.001] and loss trials [left VS: T(49) = 3.75, P < 0.001; right VS: T(49) = 3.93, P < 0.001] and a trend toward reduced BOLD responses during the anxiety condition during neutral trials [left: T(49) = 2.30, P = 0.026; right: T(49) = 2.31, P = 0.025].

The anxiety condition also interacted with trial type to predict responses within visual association cortex, including distributed areas of BA18 [x = 11.2, y = −76.5, z = 22.2, F(2,98) = 26.61, kE = 1445] as well as dorsal [right hemisphere: x = 26.2, y = −89.0, z = 22.2, F(2,98) = 17.12, kE = 137; left hemisphere: x = −26.2, y = −86.5, z = 24.8, F(2,98) = 23.17, kE = 188] (Figure 5) and ventral [x = 41.2, y = −74.0, z = 2.2, F(2,98) = 17.86, kE = 43] regions of BA19 (Figure 6). The interactions within areas BA18 and dorsal areas of BA19 were driven by reduced BOLD responses during the anxiety condition during the anticipation of neutral trials [BA18: T(49) = 3.10, P < 0.005; left dorsal BA19: T(49) = 2.49, P = 0.0163; right dorsal BA19: T(49) = 2.54, P = 0.014] and enhanced BOLD responses during the anticipation of gain [BA18: T(49) = 5.22, P < 0.001; left dorsal BA19: T(49) = 3.98, P < 0.001; right dorsal BA19: T(49) = 4.53, P < 0.001] and loss trials [BA18: T(49) = 5.04, P < 0.001; left dorsal BA19: T(49) = 5.10, P < 0.001; right dorsal BA19: T(49) = 3.86, P < 0.001]. Interactions within ventral areas of BA18 were characterized by a trend toward reduced BOLD responses during the anxiety condition during the anticipation of neutral trials [T(49) = 2.18, P = 0.034]. However, enhanced BOLD responses during the anxiety condition were only observed during gain [T(49) = 5.26, P < 0.001], but not loss trials [T(49) = −0.91, P = 0.37].

Outcome

Whole Brain. No interactions between the anxiety condition and trial type during the presentation of successful or unsuccessful feedback were observed at whole brain corrected statistical thresholds.

Striatal regions of interest (ROIs)

In order to determine if anxiety interacted with trial type to predict neural responses to successful feedback within areas of the VS, parameter estimates were extracted from a priori ROIs (Supplementary Figure 10). RANOVA models demonstrate a main effect of trial type on BOLD responses within areas of the left [F(2,98) = 9.86, P < 0.001] and right [F(2,98) = 9.11, P < 0.001] VS. Interactions between condition and trial type did not reach significance (all P’s > .2).

In order to determine if anxiety interacted with trial type to predict neural responses to unsuccessful feedback within areas of the dorsal striatum, parameter estimates were extracted from a priori ROIs (Supplementary Figure 10). RANOVA models demonstrate a main effect of trial type on BOLD responses within areas of the left [F(2,98) = 13.89, P < 0.001] and right [F(2,98) = 4.50, P = 0.014] dorsal striatum. Furthermore, areas of the right dorsal striatum exhibited larger BOLD responses during the anxiety condition [F(1,49) = 6.09, P = .017]. However, interactions between condition and trial type did not reach significance (all P’s > .2).

Main effects of trial type and threat

RANOVAs demonstrated that there were main effects of trial type and the anxiety condition during both the anticipation and outcome period of the MID paradigm. For completeness we report the main effects of trial type and anxiety condition within the supplemental materials (for full details, see Supplementary Materials: Results, Tables S1–S5, Figures S5–S9).

Interindividual variability in BOLD response

Parameter estimates were extracted from clusters within the VS exhibiting an interaction between anxiety condition and trial type and were entered into rANCOVA models, with gender and self-reported anxiety at baseline, to assess between-subject variability in BOLD responses. Self-reported anxiety at baseline interacted with the anxiety condition to predict BOLD responses within the left [F(1,48) = 6.67, P = .013] and right [F(1,48) = 5.15, P = .028] VS, such that the anxiety condition increased BOLD responses to a greater degree in participants who reported high state anxiety at baseline. Self-reported anxiety did not impact effects of trial type or interactions between trial type and condition (all p’s > .15). Interactions between gender and other effects of interest did not reach significance (all P’s > .15).

Discussion

Our results demonstrate that anxiety enhanced VS responses during the anticipation of both positive and negative incentives. We hypothesized that the valence of gain trials would be less positive during the anxiety condition, while the valence of loss trials would be more negative during anxiety. Furthermore, we hypothesized that anxiety would enhance the salience of both positive and negative monetary incentives. Previous research has demonstrated that VS responses reflect the valence of certain outcomes but reflect the salience of uncertain outcomes (Cooper and Knutson, 2008). Our results suggest that during the anticipation of uncertain outcomes, anxiety enhances VS activity to incentives generally, which supports the interpretation that induced anxiety impacts the salience of monetary incentives. Notably, we did not observe any interaction between trial type and the anxiety condition during the outcome period, even when employing a priori ROI analyses and a more liberal statistical threshold. Although it is possible that these patterns reflect a more prominent effect of anxiety on the neural processing of salient monetary stimuli during the anticipation period, future research is needed to formally assess differences in how anxiety impacts the neural processing of monetary incentives during different stages of processing.

In contrast to our results, Choi et al. (2014) showed that gain trials that are paired with the threat of an aversive electrical stimulus result in decreased VS responses during anticipation. Additionally, while we show slower reaction times during induced anxiety, Choi et al. report faster reaction times during threat trials. Importantly, we manipulated anxiety via threat of unpredictable electrical shock across blocks of trials, whereas Choi et al. manipulated threat on a trial-by-trial basis. As such, our pattern of findings may be specific to sustained aversive states and may not generalize to experimental manipulations involving transient responses to threat. However, other important distinctions between Choi et al. and the present study are notable. Most of the anticipation periods (75%) reported in Choi et al. were 12 s long, while most experiments using the MID task, including the design reported here, have anticipation periods of less than 5 s. Consequently, it is possible that the relatively longer anticipation period used in Choi et al. may impact the effect of threatening information on VS activity during anticipation.

In addition to advancing the basic neuroscience question of how anxiety impacts incentive processing circuits, our results may also shed light on divergent findings regarding striatal activity observed in clinical populations. Both PTSD and SAD patients exhibit heightened sensitivity to threat and reduced positive affect (Brown et al., 1998; Etkin and Wager, 2007; Nawijn et al., 2015). However, while PTSD participants do not exhibit differential VS activity during incentive anticipation (Nawijn et al., 2016), participants diagnosed with SAD exhibit increased VS responses during the anticipation of gains (Guyer et al., 2012). Although prior evidence suggests that anhedonia is associated with blunted VS responses during reward anticipation (Wacker et al., 2009; Stoy et al., 2012), the literature at large demonstrates that VS activity can reflect different properties of motivational stimuli in different contexts (Cooper and Knutson, 2008). Our results underscore the complex nature of patterns of VS function and highlight the importance of future research to unravel the interactions among stimulus type, situational context and emotional state that determine dysregulated neural activity in individuals with mental disorders.

In addition to impacting neural responses within the VS, anxiety interacted with trial type to predict occipital responses during the anticipation period. Previous research has demonstrated that neural responses to visual stimuli within the occipital cortex reflect the learning of both appetitive and aversive associations (Cheng et al., 2003; León et al., 2018). Further, threat of shock can impact neural responses within occipital cortex during visual discrimination behavior linked to monetary rewards (Hu et al., 2013). These data suggest that not only do affective stimuli impact sensory activity but that interactions between appetitive and aversive stimuli are reflected in neural responses within the visual system. The visual system is hierarchically organized, and it is possible that neural responses within different areas of the visual system reflect different properties of motivational stimuli being processed. Dorsal areas of BA19 fall within the dorsal `where’ visual stream, which is dedicated to processing the location of stimuli within the environment, while ventral areas of BA19 fall within the ventral `what’ visual stream dedicated to the processing of object recognition. Spatial processing is necessary for adaptive responses to both positive and negative stimuli and it is possible that enhanced activity to gains and losses within dorsal areas of occipital cortex during anxiety reflects increased spatial processing during monetary incentives. In contrast, anxiety only enhanced activity to gains, but not losses, within ventral areas of occipital cortex which might reflect competition between positive and negative valence within areas of the ventral visual stream. Future research will be needed to directly address how sensory and striatal circuits interact during motivational processing and the role of occipital cortex in mediating responses to aversive and appetitive stimuli.

This study is not without limitations. First, our results cannot directly address which psychological processes are impacted by induced anxiety. Although participants exhibit increased VS responses during the anticipation of gains and losses, which may reflect increased salience of motivational stimuli during anxiety, we did not collect any measure of self-reported salience and are unable to directly address this possibility. Further, uncertain monetary outcomes likely elicit multifaceted psychological and neural processing, which may involve factors unrelated to the salience of stimuli. Second, it is surprising that our results did not identify an impact of induced anxiety on mPFC responses during the delivery of monetary gains or losses. Third, we did not employ a practice MID paradigm before scanning, as is typically done, and the absence of the practice task may result in enhanced learning during early trials of the experiment. Fourth, the identification of interactions between anxiety and trial type via mass univariate analysis, as done here, likely results in inflated effect sizes. As such, post hoc tests and figures can determine directionality but cannot provide an unbiased estimate of effect size. Finally, persons suffering from anxiety disorders experience anxiety which is much more chronic than the transient induction of anxiety utilized here. Consequently, our results may not reflect the same type of neural process observed in those diagnosed with anxiety disorders.

Collectively, our results suggest that induced anxiety can impact neural processing of monetary incentives within both sensory and motivational circuits. Our results demonstrate that induced anxiety increases VS responses during the anticipation of gains and losses. Additionally, our results demonstrate that while anxiety enhances occipital activity to incentives generally within areas of the dorsal visual stream, occipital responses within areas of the ventral visual stream are only enhanced by anxiety during gain trials. These results shed light on the complex nature of interactions between appetitive and aversive information within the human brain and may serve to inform future research investigating the neural mechanisms underlying disturbances in positive affect associated with clinical diagnoses.

Acknowledgments

This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).

Conflict of interest. None declared.

Funding

This work was supported by the Intramural Research Program of the National Institutes of Mental Health, project no. ZIAMH002798 (clinical protocol 02-M-0321, NCT00047853) to C.G.

References