Coping in the Clinic: Effects of Clinically Elevated Anxiety on Dynamic Neurophysiological Mechanisms of Escape/Avoidance Preparation

Abstract

BACKGROUND:

Treatments for anxiety and related disorders target exaggerated escape/avoidance as a core feature, but current methods fail to improve escape/avoidance habits for many treatment-seeking individuals. To support developing tools that increase treatment efficacy by targeting mechanisms more directly, the current work examined potential distinctions in the neurophysiologies of escape and avoidance and tested how clinical anxiety affects these neurophysiologies.

METHODS:

Twenty-five treatment-seeking individuals with varied principal diagnoses (e.g., generalized anxiety disorder, posttraumatic stress disorder) and 20 non-treatment-seeking control subjects participated. In the study task, approximately 5.25-second cues predicted aversive images that could be avoided (blocked by a button press before image onset), escaped (ended by a button press after image onset), or not controlled. To examine neural processing and defensive response modulation, anticipatory event-related potentials were derived, and startle reflexes were probed throughout each cue.

RESULTS:

Multidimensional profiles were observed such that 1) anticipatory event-related potential enhancement was only reliable during avoidance preparation, and event-related potentials potentially reflected perceived/instrumental control; and 2) startle reflexes were inhibited during avoidance preparation, relatively enhanced during escape preparation, and further enhanced during uncontrollable anticipation, thus potentially reflecting fear-related activation. Treatment-seeking status, then, did not affect cortical processing, but it did moderate context-dependent fear (if individuals with severe depression were excluded) such that treatment-seeking individuals without depression showed exaggerated startle during escape, but not avoidance, preparation.

CONCLUSIONS:

Data suggest a specific effect of anxiety on fear system activation during preparation to escape aversion. This effect warrants further investigation as a precision target for interventions that directly modulate the specific underlying neural circuitry.

Because situational avoidance plays a key role in maintaining anxiety and related disorders by preventing corrective relearning of situational threat value (1), evidence-based treatments consistently identify avoidance as a critical intervention target. Despite the focus on avoidance, building evidence suggests that for many treatment-seeking individuals, current techniques still have limited effectiveness in ameliorating avoidant habits (2,3); for many, avoidant habits either impede treatment engagement (4) or reappear/never fully remit after treatment that otherwise reduces reported symptoms (5). In light of these issues, a need for novel tools that increase the durability and generalizability of treatment by targeting avoidance mechanisms early in treatment is increasingly apparent (2,6).

Nonhuman animal work suggests that one crucial step toward improving treatment could be to more precisely target mechanisms of distinct avoidant-type behaviors—especially reactive escape from present threat versus proactive avoidance of upcoming threat (7). Regarding specific mechanisms, research suggests that fear-related neural activity (especially in the critical central amygdala–ventral periaqueductal gray pathway) (8) and resulting downstream physiological/hormonal expression are critical in driving reactive escape (7) but may not be critical to proactive avoidance (which instead depends on habit-related striatal circuitry) (9). Relevant to clinical practice, animal work also suggests that extinction methods that reduce learned fear at the same time can fail to reduce avoidance of previously feared stimuli (10)—perhaps providing an analog for cases in which intervention reduces situational fear but does not ameliorate avoidant habits.

Extending reactive escape to proactive avoidance comparison to humans, our laboratory recently also observed evidence for different fear-related concomitants of these behaviors, while at the same time suggesting that this may co-occur with motor-related processing that does not vary across escape/avoidance contexts (11,12). The study measured reflex and autonomic physiology during cues to demonstrate whether upcoming aversive exposure could be escaped (shortened after onset) or avoided (prevented before onset), and it was found that startle reflex reactivity (a marker of context-dependent amygdala activation) (13) and skin conductance were each enhanced during escape compared with avoidance preparation, while cardiac deceleration was similarly enhanced in both escape and avoidance contexts. Given the potential for multiple dimensions of coping preparation and differential modulation of these dimensions, a prime aim of the current work was to examine how neural physiology is modulated across escape/avoidance contexts while replicating findings for measures (e.g., startle) that do differentiate escape and avoidance.

To accomplish the first aim, this study incorporated electroencephalography [scalp-level recording of cortical pyramidal cell–generated electric field potentials (14)] and derivation of event-related potentials (ERPs) that arise during escape/avoidance preparation. The ERP of interest was a slow-wave negativity—the contingent negative variation (CNV)—that arises in tasks in which a warning cue predicts a subsequent stimulus signaling time to act (15). The CNV comprises early- and late-interval slow-wave negativities, and the late component, in particular, reflects motor preparation processing carried out through frontal cortex–basal ganglia interactions (16). In line with its sensitivity to motor processing, late CNV has been shown to be enhanced by various features that increase response motivation, including when responding terminates painful stimulation (17). At the same time, there is also evidence that slow-wave negativities can scale with perceived, more than actual, control—such that amplitudes are enhanced if participants believe that they control a desired outcome relative to when they actively respond but do not believe that it influences outcome (18). Because no study has compared CNVs across escape and avoidance preparation contexts, it is unknown whether CNV amplitude is similar for each behavior because motor demand is equal, or if it instead is enhanced during avoidance preparation inasmuch as perceived control is higher than that during escape preparation.

In addition to examining a neural escape/avoidance processing dimension, another principal aim of this study was to test how clinically elevated anxiety affects escape/avoidance processing. Prior work began such examination by relating trait-level anxiety (measured using the State-Trait Anxiety Inventory [STAI]) (19) to physiology in a non–treatment-seeking student sample. Results indicated a positive correlation between anxiety and startle magnitude enhancement during escape compared with avoidance preparation but no impact on other indices (e.g., heart rate), suggesting an effect of anxiety on escape-related fear system activation without an impact on motor processing (12). In the current work, we tested how clinical anxiety (anxiety associated with functional impairment) affects escape/avoidance processing by recruiting treatment-seeking individuals. In doing so, one goal was to examine whether anxiety affects neural physiology as it does the startle reflex, or if anxiety does not alter neural physiology just as it does not alter autonomic indices of motor preparation (a critical aim because prior work finds some evidence of anxiety effects in volitional effort [e.g., (20)] or multiple demand [e.g., (21)] tasks but has not examined effects on simple escape/avoidance preparation). In addition, this work sought to determine whether anxiety that is clinically impairing specifically affects escape-related physiology as does anxiety that is heightened but not necessarily functionally impactful, or if, instead, clinical anxiety has broader effects across escape/avoidance contexts as might be predicted based on the suggestions that anxiety becomes impairing when it impedes adaptive neurobehavioral tuning to context-specific demands (22,23).

By characterizing neural along with physiological dimensions of escape/avoidance processing and examining clinical changes in that processing, an ultimate objective of this research was to establish precision targets for interventions that could modify escape/avoidance behaviors more directly and efficiently. To accomplish the study aims, participants completed a task in which cues indicated if subsequent aversive content could be prevented, shortened, or not controlled at all. Throughout the task, neural physiology of coping preparation was assessed by deriving cue-related ERPs, and startle reflex modulation was also assessed as in prior research. Using this approach, we hypothesized that cue-related CNVs might be more sensitive to the task demand of preparing action than to the aversive feature of exposure certainty, and if so, it would be similarly enhanced in escape and avoidance contexts (relatedly, we also expected to replicate motor patterns for heart rate and skin conductance) (see the Supplement). Alternatively, we hypothesized that if CNV in particular were sensitive to perceived control, then its amplitude would be further enhanced when exposure could be completely controlled (avoided) relative to when exposure could be shortened (escaped) but not fully prevented. To examine clinical effects, our sample in this study included community-dwelling adults seeking treatment for anxiety or related disorders in addition to a comparison group not seeking such treatment. Based on our analog findings, we predicted that clinically elevated anxiety would specifically affect startle (fear expression) in escape, but not avoidance, contexts. At the same time, we also could not a priori rule out an alternative prediction that clinically elevated (as opposed to heightened but not necessarily impairing) anxiety would more broadly reduce granularity across escape and avoid contexts and/or would affect neural and reflex dimensions of escape/avoidance processing.

METHODS AND MATERIALS

Methods are overviewed here, and more specific replicationrelated details are provided in the Supplement.

Participants

Participants were recruited from outpatient mental health clinics in a Southeastern medical center and surrounding community. The study was approved by the institutional review board, and participants provided written consent. Participants included 25 adults seeking treatment for anxiety or related disorders and 20 community-dwelling non–treatment-seeking adults (age 18–65 years). In the treatment-seeking group, 5 individuals reported severe depression [Beck Depression Inventory (BDI) score > 28] (24) in addition to anxiety. To allow comparison to our prior work in which the sample did not include individuals with severe depression (12), analyses including all participants were then repeated in individuals without severe depression.¹

Sample Characterization

Questionnaires included a demographics survey, the STAI–Trait Anxiety (STAI-T) form (19); BDI, Second Edition (BDI-II) (24); and Illness Intrusiveness Rating Scale (25), a survey of mental health–related functional interference. Psychiatric diagnosis was assessed by doctoral-level staff via a Mini-International Neuropsychiatric Interview (26).

Study Task

The experimental task included visual cues, a visual go signal, and pictures as response-motivating outcomes (Figure 1). The initial 5- to 5.5-second presentation of the predictive cue was immediately followed by a 500-ms go signal (in avoidance trials) or aversive (disgusting/violent) picture (in escape/no-control trials). In avoidance and escape trials, if participants pressed a button within 500 ms after anticipated stimulus onset, the stimulus was replaced by a neutral (everyday scene/object) image for another 2.5 seconds. If the button was not pressed within 500 ms then the go signal was replaced by an aversive picture for 2.5 seconds (avoidance trials) or aversive exposure continued for another 2.5 seconds (escape trials). On no-control trials, the initial 500-ms aversive presentation was followed by a 2.5-second additional exposure regardless of any button presses (no-control trials with button press removed were <1% of all trials).

Figure 1.

Task design including actual cues, go signal, and timings. Resp, response; RT, response time.

Task orders were counterbalanced across participants such that each image occurred in each context (no image repetition). In addition to visual stimuli, a 50-ms, 105-dB white noise burst was presented during each cue (2.5 seconds or 1 second before anticipated stimulus onset) to probe startle reactivity.

Procedure

After interviews or surveys, participants were seated in front of the task computer and affixed with sensors, headphones, and a mouse to respond. Before the task, participants were told each cue’s meaning and were told 1) not to press the button on uncontrollable trials, 2) that they would lose control if they pressed the button before cue offset, and 3) to ignore noises played over headphones. Participants then completed 5 practice trials to verify understanding and present startle habituation probes.

After task completion but before sensor removal, participants rated the pleasantness of anticipating avoidable, escapable, and uncontrollable aversion and of viewing disgusting/violent and everyday images using a Likert-type scale that ranged from 1 (very unpleasant) to 9 (very pleasant; 5 indicates neither pleasant nor unpleasant). Sensors were then removed, and subjects were debriefed and paid.

Data Processing and Analysis

Startle Reflex.

Startle reactivity was indexed using an MP150 data acquisition and EMG100C amplifier (BIOPAC Systems, Inc) and 2 Ag/AgCl sensors measuring left orbicularis oculi electromyography. Responses to each probe were scored and standardized as T scores for each subject.

Cue-Related ERPs.

Electroencephalography data were collected using actiCHamp (Brain Products GmbH) and 32 active sensors arrayed in International 10–20 system. Following standard preprocessing procedures including time-locking to cue and trial averaging, resulting ERPs were quantified with complementary strategies: 1) in a traditional approach, the signal was averaged in time windows and sensor clusters of maximal voltage for each component; and 2) to separate cue-related and startle probe–related time-series components, ERPs were submitted to temporospatial principal component analysis (PCA) (27). Because the aim of PCA was to separate cue- and probe-related components rather than interrogate spatial distinctions, primary analyses were run on factors from the initial temporal PCA, while subsequent spatial PCA was used to identify factors accounting for >0.5% variance per convention (28). The sample point and sensor of maximal voltage was identified using PCA software and exported factor scores for analysis. Analyses of temporospatial subfactors are presented in the Supplement.

Other Measures.

Along with cue ERPs, startle probe ERPs were quantified as in prior research (29). Heart rate and skin conductance changes were also collected as in our prior work (12). These data are presented in the Supplement.

Data Analytic Plan.

Main analyses involved mixed-effects analyses of variance with experimental factors as repeated measures (30). Analysis of cue-related ERPs centered around 3 (response context: avoidance, escape, and no-control) × 2 (group) analyses of variance. Startle analysis centered around 3 (context) × 2 (probe time: 2.5- or 1-second pre-picture onset) × 2 (group) analyses of variance (F values used Greenhouse-Geisser–corrected dfs) (31). For all analyses, significant omnibus effects were followed up with paired-samples (for experimental conditions) or independent-samples (for group) two-tailed t tests (α = 0.05). A post hoc power analysis indicated that the 20 treatment-seeking individuals and 20 control subjects resulted in power 1 − β = 0.80 to detect group × context interactions of small/medium size (η² = 0.04, consistent with 12) at α = 0.05 (repeated-measure correlation set to 0.5 as observed in CNV scores).

RESULTS

Sample Characteristics and Manipulation Checks

Sample and manipulation check analyses are presented in the Supplement. To summarize, regarding sample characteristics, analyses indicated no demographic (biological sex, race/ethnicity, age) group differences but expected higher levels of anxiety, depression, and mental health–related functional impairment in treatment-seeking individuals. Regarding manipulation checks, findings paralleled prior work: 1) rated aversiveness decreased with increasing control, 2) participants were similarly fast and accurate across escape and avoidance conditions when poorly performing individuals were excluded (see Methods and Materials), and 3) group did not moderate ratings or behavior.

Task Physiology

Contingent Negative Variation.

Figure 2 depicts raw ERPs along with factors that emerged from temporal PCA. For raw CNVs, early (time window: 800–1200 ms; sensor cluster: FC1, FC2, Cz, CP1, CP2) and late (time window: 4500–5000 ms; sensor cluster: Fz, FC1, FC2, Cz) components appeared over frontal sites. Both early and late CNVs were modulated by context (early: F_2,37 = 4.2, p = .02, η² = 0.03; late: F_2,37 = 4.1, p = .02, η² = 0.03) but not by group (early: F_2,37 = 0.2, p = .69, η² < 0.01; late: F_2,37 = 0.28, p = .60, η² = 0.01) or by the group × context interaction (early: F_2,37 = 1.0, p = .37, η² = 0.01; late: F_2,37 = 1.9, p = .16, η² = 0.02). Effects did not change for early (context: F_2,32 = 3.7, p = .03, η² = 0.03; group: F_1,33 = 0.1, p = .74, η² < 0.01; interaction: F_2,32 = 0.3, p = .72, η² < 0.01) or late (context: F_2,32 = 3.5, p = .04, η² = 0.03; group: F_1,33 = 0.5, p = .50, η² = 0.02; interaction: F_2,32 = 1.4, p = .26, η² = 0.01) CNVs with depressed individuals removed. Follow-up for early CNVs indicated enhancement during avoidance preparation (t₃₉ = 2.6, p = .01, d = 0.42) and a tendency toward enhancement during escape preparation (t₃₉ = 1.8, p = .09, d = 0.28) compared with uncontrollable anticipation. For late CNVs, amplitude was enhanced in avoidance compared with uncontrollable contexts (t₃₉ = 2.8, p = .009, d = 0.44) and did not differ between escape and avoidance (t₃₉ = 1.6, p = .11, d = 0.26) or escape and uncontrollable (t₃₉ = 1.4, p = .18, d = 0.22) contexts.

Figure 2.

Cue-related event-related potentials. Panel (A) depicts raw event-related potentials while panels (B) and (C) depict early (factor 2) and late (factor 1) contingent negative variations from temporal principal component analysis. In panel (A), topographies depict distribution across the scalp for early (left) and late (right) contingent negative variation, and waveform depicts electroencephalography (EEG) signal averaged in centroparietal sensors of maximal voltage (Fz, FC1, FC2, Cz). In panels (B) and (C), signal is depicted at sensor FC1, a sensor of maximal voltage for both components.

After PCA to separate cue-related and probe-related ERP components, 21 temporal factors and 3 spatial subfactors emerged. Early and late CNVs corresponded with temporal factors 2 (maximal negativity at sensor FC1, sample point 1252–1254 ms) and 1 (maximal negativity at sensor FC1, sample point 4970–4972 ms); each spatial subcomponent of these factors explained >0.5% variance (see the Supplement). Similar to raw CNVs, each factor was modulated by context (early: F_2,37 = 4.7, p = .01, η² = 0.04; late: F_2,37 = 6.0, p = .004, η² = 0.06) but not by group (early: F_1,38 = 0.1, p = .73, η² < 0.01; late: F_1,38 = 0.07, p = .79, η² < 0.01) or by the interaction (early: F_2,37 = 1.7, p = .19, η² = 0.02; late: F_2,37 = 1.1, p = .35, η² = 0.01). Effects did not change for early (context: F_2,32 = 4.2, p = .02, η² = 0.04; group: F_1,33 = 0.1, p = .78, η² < 0.01; interaction: F_2,32 = 1.2, p = .29, η² = 0.01) or late (context: F_2,32 = 5.5, p = .006, η² = 0.06; group: F_1,33 < 0.1, p = .91, η² < 0.01; interaction: F_2,32 = 0.4, p = .67, η² < 0.01) CNVs with depressed individuals removed. Follow-up indicated factor 2 negativity was enhanced for avoidance (t₃₉ = 2.6, p = .01, d = 0.40) and escape (t₃₉ = 2.4, p = .02, d = 0.38) compared with an uncontrollable context, and there was no difference between avoidance and escape (t₃₉ = 0.4, p = .72, d = 0.06). For factor 1, negativity was enhanced in avoidance compared with uncontrollable contexts (t₃₉ = 3.4, p = .002, d = 0.53) and tended toward enhancement in escape compared with uncontrollable contexts (t₃₉ = 2.0, p = .06, d = 0.31), with no difference between avoidance and escape (t₃₉ = 1.4, p = .16, d = 0.23) contexts.

Startle Reactivity.

Figure 3 shows startle reactivity across contexts in the entire sample. In the entire sample, blinks were modulated by a context × probe time interaction (F_2,43 = 6.2, p = .005, η² = 0.03) but not by group or any group moderation of experimental factors. Follow-up indicated no context differences early in the cue, whereas later in the cue, blinks were enhanced during escape compared with avoidance preparation (t₄₄ = 2.3, p = .02, d = 0.35) and during uncontrollable anticipation compared with escape preparation (t₄₄ = 2.5, p = .02, d = 0.37). In addition, blink magnitudes decreased across avoidance preparation (t₄₄ = 2.9, p = .005, d = 0.44), tended to increase across uncontrollable anticipation (t₄₄ = 1.9, p = .07, d = 0.28), and did not change throughout escape preparation (t₄₄ = 0.2, p = .85, d = 0.03).

Figure 3.

Blink reactivity across the cuing interval in avoidance, escape, and no-control contexts.

When individuals with severe depression were removed, group did moderate the effect of context on blink reactivity (F_2,38 = 3.2, p = .05, η² = 0.03), with this effect being reliable late (F_2,38 = 5.1, p = .01, η² = 0.04), but not early (F_2,38 = 0.5, p = .62, η² = 0.01), in the cue. Follow-up specified that whereas healthy control subjects showed no difference in reactivity across avoidance and escape contexts (t₁₉ = 0.1, p = .90, d = 0.03) and also inhibited reactivity in escape relative to uncontrollable contexts (t₁₈ = 2.2, p = .04, d = 0.54), treatment-seeking individuals conversely showed enhancement in escape relative to avoidance contexts (t₁₉ = 4.2, p < .001, d = 0.95) and no difference across escape and uncontrollable contexts (t₁₉ = 0.5, p = .63, d = 0.11).² Furthermore, between-group comparison showed that blink reactivity differed between treatment-seeking individuals and control subjects in escape (t₃₈ = 2.9, p = .006, d = 0.91) but not in avoidance (t₃₈ = 0.4, p = .69, d = 0.12) or uncontrollable (t₃₈ = 0.7, p = .47, d = 0.23) contexts.

To follow up the change in results owing to severe depression, linear regression was used to test STAI-T and BDI-II scores and their interaction as predictors of an escape-avoidance difference score in the entire sample. In this analysis (STAI-T: b = 0.34, SE = 0.10, t₄₄ = 3.3, p = .002 and BDI-II: b = −0.31, SE = 0.11, t₄₄ = 2.8, p = .007), each predicted escape-related potentiation (total: R²_3,41 = 0.31, p = .002). Following this, STAI-T scores were then correlated with the escape-avoidance modulation score within minimally depressed (BDI < 14) and mild/moderately depressed (13 < BDI < 29) groups, and this relationship was visually inspected in participants with severe depression (BDI > 28). Results indicated a positive relationship in the minimally depressed group (r₂₁ = 0.48, p = .02) and a suggestive relationship in the mild/moderately depressed group (r₁₅ = 0.44, p = .08). In the severely depressed group, all 5 participants showed a small or inverted escape-avoidance difference (Figure 4 summarizes clinical effects on startle).

Figure 4.

Effects of anxiety and depression on context modulation of blink reactivity. Panel (A) depicts blink reactivity in avoidance, escape, and no-control contexts for control subjects and treatment-seeking individuals (Tx Seekers). Inset depicts blink magnitudes in the treatment-seeking group combined across individuals with and without depression. Scale of inset is the same as for the main graph. Panel (B) depicts the relationship between trait anxiety (State-Trait Anxiety Inventory–Trait Anxiety [STAI-T]) and the escape-avoidance difference score in minimally depressed (red), mild/moderately depressed (blue), and severely depressed (purple) groups. Red and blue lines depict the STAI-T–difference score relationship in the minimally depressed and mildly/moderately depressed groups, respectively. BDI, Beck Depression Inventory.

DISCUSSION

Current data extend elucidation of multidimensional escape/avoidance processing to neural physiology and of context-varying anxiety effects to treatment-seeking individuals. Regarding neural physiology of escape/avoidance, complementary analyses each indicated that enhancement of a preparatory CNV was only conventionally reliable when action completely avoided aversive exposure, suggesting that CNVs might track control in coping contexts. At the same time, results replicated findings of increasing startle reflex priming from avoidance to escape and then to uncontrollable contexts, replicating that such priming tracks operation of an affective (fear) modulation system rather than motor processing (11,12). Regarding anxiety effects, current data found that treatment-seeking anxious (but not severely depressed) individuals and non–treatment-seeking individuals differed specifically in startle reactivity during escape preparation; clinical impairment did not moderate exaggeration of reactivity specifically in an escape context and there was a lack of effects on motor processing. At the same time, results suggested that severe depression may have blunted modulation of startle across task contexts—a finding that could indicate important clinical-biobehavioral subgroups but also demands replication with larger samples of highly depressed individuals.

As movement toward precision targeting of psychiatric disorder–relevant processes continues, it will be critical to understand the multiple neurobehavioral systems that mediate each process and which systems do or do not require clinical intervention. As one advance in such work, current data characterize a neural dimension of motor processing and suggest that the CNV and other measurements could reflect multiple motor systems, such that CNV enhancement in this study, was particularly reliable in avoidance preparation, whereas cardiac deceleration was equivalent for escape and avoidance and skin conductance was particularly increased for escape preparation (see the Supplement). The CNV modulation observed here is consistent with an interpretation that it may uniquely reflect a perception-of-control system or perhaps a system for internally cuing action when it precedes salient outcomes (18). While late CNVs are often enhanced in emotional compared with nonemotional contexts, this likely relates more to enhanced motor demand inherent in such contexts than to emotional tone itself. Relatedly, late CNV amplitude does not change across contexts in which the aversiveness of anticipated stimulation varies but active demand is equivalent and functions similar to end stimulation in each context (17). In the current work, meanwhile, reliable enhancement of late CNV in avoidance, but not in escape, contexts was consistent with a difference in the function of action even as outcome aversiveness was similar; that is, action provided complete control in an avoidance context but only partial control in an escape one.

Regarding clinical effects, a lack of group differences for the CNV or other motor measurements suggests that anxiety did not impede any aspect of motor processing, and therefore anxious individuals were not in less control over the task than their non–clinically anxious counterparts. Other studies examining anxiety effects on CNVs have done so in more dynamic contexts (i.e., ones that involve voluntary gradations of motor force in return for varying reward or competing/shifting task demands), and some published studies show anxiety-related differences in such contexts (20,21,32–36). In this study, meanwhile, motor demand was consistent and relatively simple across contexts, and motivation was also presumably similar inasmuch as all participants rated disgusting/violent images as similarly unpleasant. Given unvarying motor demand in this task, it is perhaps not surprising that individual differences in motor processing also did not arise. At the same time, more work is needed to determine if anxiety-related differences in other dynamic contexts are reliable and could thus represent a viable intervention target, which is especially important because methods of prior dynamic motor processing studies vary widely.

With no group differences on motor processing, the presence of effects for startle suggests that anxiety did affect how a fear modulation system operated specifically during escape preparation. Animal and human work supports startle as an index of how fear is expressed in discrete threat contexts by 1) observing reflex priming across contexts including fear conditioning (13), picture viewing (37), and imagery (38); and 2) showing critical relationships between reflex priming and activity in a central amygdala–ventral periaqueductal gray pathway in discrete threat contexts (39,40). Importantly, central amygdala activity and downstream changes including startle priming represent a context-dependent frozen fear profile that also arises while animals prepare to escape threat and then disappears as they learn to avoid threat entirely (41). In humans, exaggeration of frozen fear activation during preparation to escape aversive exposure could in turn help motivate adopting broader avoidant strategies (inasmuch as activation in discrete contexts motivates individuals to consistently avoid even controllable exposure). This possibility warrants future testing by relating escape-related startle potentiation to real-world coping measures (e.g., Brief Coping Orientation to Problems Experienced) (42).

To determine whether the potential driver of habitual avoidance identified here can be targeted in direct treatment, subsequent studies need to test if it 1) is individually reliable and 2) can be altered by modulating its specific neural mediators (e.g., using brain stimulation technology such as transcranial magnetic stimulation). As alluded to above, a critical target to modify escape-related fear modulation could be the central amygdala or its inhibitory inputs (e.g., the ventromedial prefrontal cortex) (43). Relatedly, human work with a similar task as in the current work found that natural lesions of inhibitory inputs to the amygdala also lead to exaggerated startle in escape, but not avoidance or uncontrollable, contexts (44), supporting an idea that intervention that enhances rather than impedes amygdalar inhibitory inputs may effectively reduce escape-specific fear. Importantly, context specificity of effects also supports a possibility of targeting fearful freezing specifically in controllable but certain exposure situations, which is important given that current findings also suggest that anxiety might not alter how fear is expressed in other [i.e., complete control or strong (45) uncontrollable] contexts. If further work confirms that escape-related fearful freezing can be individually measured and modulated in anxious individuals, then this could have strong clinical implications if such work also shows that modulation can be implemented early in treatment in a way that enhances outcome.

Testing viability of modulating escape-related circuitry in a clinically impactful way represents a fruitful further research program, and this research must also determine if different tasks or measurements should be targeted in different treatment-seeking subgroups—e.g., in individuals with severe depression as tentatively suggested here. In addition to this work, various other steps are also needed to address the limitations of the current research, with the first step being to expand the sample to investigate patterns within specific diagnostic groups (including individuals with severe depression) and also further examine demographic moderators. Next, systematically altering the task to understand boundary conditions will be important, with potential alterations including incorporating a simple response preparation control condition, incorporating dynamic response thresholds and investigating effects with different (e.g., tactile) aversive outcomes. Finally, examining pleasant coping preparation could be critical to determine if similar reactive/proactive distinctions also arise in the approach-based coping domain.

Even prior to further research, though, the current work demonstrates clinical anxiety effects on fear expression during escape preparation that could be a target for novel intervention tools. To the degree that exaggerated fearful freezing activation in controllable exposure contexts is one key driver of persistent real-world avoidance, objectively assessing this change and targeting it with precision—perhaps using brain stimulation that modulates specific neural mediators—could become part of treating real-world avoidance in a way that has more durable, generalized effects. Moreover, inasmuch as such assessment and treatment can be accomplished rapidly using new technologies, this and complementary strategies could go a long way toward improving issues of treatment nonadherence and nonresponse—issues that continue to limit effectiveness of treatments for anxiety and related disorders for all who seek them (2,3).