Anticipatory Threat Responding: Associations with Anxiety, Development, and Brain Structure

Abstract

Background

While translational theories link neurodevelopmental changes in threat learning to pathological anxiety, findings from studies in patients inconsistently support these theories. This may reflect difficulties in studying large patient samples with wide age-ranges using consistent methods. A dearth of imaging data in patients further limits translational advances. We address these gaps through a psychophysiology and structural brain imaging study in a large sample of patients across the lifespan.

Methods

A total of 351 participants (age 8–50 years; 209 females; 195 healthy participants and 156 medication-free, treatment-seeking anxious patients) completed a differential threat conditioning and extinction paradigm that has been validated in pediatric and adult populations. Skin conductance response indexed psychophysiological responding to conditioned (CS−, CS+) and unconditioned (UCS) threat stimuli. Structural magnetic resonance imaging data were available for 250 participants. Analyses tested anxiety and age associations with psychophysiological responding, as well as associations between psychophysiology and brain structure.

Results

Regardless of age, patients and healthy comparisons demonstrated comparable differential threat conditioning and extinction. Magnitude of skin-conductance responding to both CS types differentiated patients from comparisons and covaried with dorsal prefrontal cortical thickness; structure-response associations were moderated by anxiety and age in several regions. Unconditioned responding was unrelated to anxiety and brain structure.

Conclusions

Rather than impaired threat learning, pathological anxiety involves heightened skin-conductance response to potential but not immediately-present threats; this anxiety-related potentiation of anticipatory responding also relates to variation in brain structure. These findings inform theoretical considerations by highlighting anticipatory responding to potential threat in anxiety.

Introduction

Translational theories link neurodevelopmental changes in threat learning to pathological anxiety (1–5). However, findings in patients inconsistently support these theories (3,6,7). Disconnect between theory and data may reflect difficulties in recruiting large patient samples; inconsistent methods across studies; and failure to examine wide age ranges within studies. Furthermore, few studies in patients relate physiology to brain measures, limiting translational advances across developmental stages. Here, we address these gaps by integrating psychophysiology with structural brain imaging to study threat learning in anxious and healthy individuals spanning childhood, adolescence, and adulthood (N=351; ages 8–50 years).

Threat learning encompasses conditioning and extinction. Conditioning is a highly-conserved process through which a neutral stimulus becomes associated with a threat, such that subsequent encounters with the stimulus elicit anticipatory responding to the danger that might follow; extinction reflects attenuation of conditioned threat responding when the stimulus no longer predicts the occurrence of threat (2,8–11). A core feature of pathological anxiety is exaggerated fear of anticipated threats (12,13), and contemporary theories attribute anxiety to aberrant threat learning, conceptualized as rapid or exaggerated conditioning or attenuated extinction (1,3,4,14). Other data suggest that developmental changes in threat learning contribute to the emergence of anxiety disorders in late childhood and early adolescence (2,5,15–18). Research guided by these theories aims to inform anxiety treatment (1,2,5,19–23).

While studies in non-human animals and healthy individuals provide support for these theories (2,19,24,25), meta-analyses of studies comparing threat learning between healthy and anxious pediatric (6) as well as adult (3,7) samples yield mixed findings. Thus, all meta-analyses find evidence of perturbed threat learning, but differ in the specific affected processes they identify. Such inconsistency generates a need for more data on associations among pathological anxiety, age, and threat learning (6,25,26). The current report addresses this need.

Paradigms model threat learning by pairing a neutral conditioned stimulus (CS) with an unconditioned threat stimulus (UCS)(11,27); with learning, the CS comes to elicit a conditioned response in anticipation of danger (9,13,14,28,29). Threat learning may therefore rely on both conditioned and unconditioned threat responding (30). While ample research focuses on anxiety-related differences in conditioned responding (3,6,7,13,31), fewer studies address aspects of unconditioned threat responding (13,31), particularly as such responding changes with development (25,32). Thus, comparing conditioned and unconditioned threat response among pediatric and adult anxiety patients and healthy volunteers addresses important gaps.

Finally, we extend neuroanatomical data. Substantial research, particularly in healthy subjects, links conditioned threat responding to structure and function in prefrontal cortex, amygdala, and hippocampus (2,8,33–40). Moreover, other data suggest that age moderates the neural architecture of threat learning (2,25,41). However, no studies relate individual differences in brain structure to psychophysiological threat-response measures that relate most strongly to anxiety disorders across age. We first identify such skin-conductance measures, and then identify their structural correlates. Finally, we evaluate the moderation of these skin-conductance-response/brain-structure associations by age and anxiety diagnosis.

To achieve these goals, we studied a large sample of anxious and healthy children, adolescents, and adults (N=351). All participants completed a differential (i.e., involving both threat and safety learning) conditioning and extinction paradigm previously validated in pediatric and adult populations (32,42); a subset (n=250) completed structural imaging. Analyses proceeded in three stages, testing specific hypotheses arising from prior research. First, we tested anxiety and age effects on skin-conductance response (SCR) indices of conditioning and extinction. Based on prior findings (3,6,7), we hypothesized comparable differential conditioning and extinction effects in patients and comparisons but enhanced anxiety-related responding to both conditioned threat and safety cues during the task, both in youth and adults. Second, we examined response to unconditioned threat; given no prior reports of anxiety-related differences in UCS-responding and the prominence of anticipatory fears in anxiety, we hypothesized anxiety effects to manifest more strongly in responses to conditioned than unconditioned threats. Finally, we examined correlations between brain structure (cortical thickness and gray-matter volume) and conditioned responding, as well as their moderation by anxiety and age. Given prior research on structure-SCR associations (19,36–38,40,43), we hypothesized effects to emerge in prefrontal regions as well as amygdala and hippocampus. Primary hypothesis tests in all three areas considered SCR, given data on reliability (44), the ease with which SCR responses to the CS and UCS can be compared, and the availability of prior data on brain-structure correlates of SCR. Secondary analyses examined anxiety and age effects on startle-probe-related electromyography (EMG) and self-reported fear.

Methods and Materials

Participants

387 individuals underwent conditioning and extinction; analyses include N=351 (see Table 1, supplement), with 195 healthy (108 females; ages 8–46 years) and 156 anxious (101 females; 8–50 years) participants not differing in age, sex, or IQ, all ps>.08, all studied at the National Institute of Mental Health. Written informed-consent was acquired from adult participants as well as parents, and written assent was acquired from youth for an Institutional-Review-Board approved protocol. Previously-reported psychophysiology data for 162 participants (72 anxious, 90 healthy) (32,44) were combined with unpublished data for 189 participants to generate the sample (N=351).

Table 1.

Sample demographic and clinical characteristics.

Characteristics	Healthy (N=195)	Anxious (N=156)
	Mean (SD)	Mean (SD)
Age (years)	21.46 (9.14)	19.78 (9.99)
IQ (WASI)	114.30 (11.55)	114.31 (13.33)
Characteristic	No. (%)	No. (%)
Female	108 (55.4%)	101 (64.7%)
Diagnosis
Generalized anxiety disorder	-	121 (77.6%)
Social anxiety disorder	-	94 (60.3%)
Separation anxiety disorder	-	29 (18.6%)
Specific phobia	-	28 (17.9%)
Panic disorder	-	11 (7.1%)
Attention-deficit/hyperactivity disorder	-	5 (3.2%)
Major depressive disorder	-	4 (2.5%)
Selective mutism	-	2 (1.3%)
Oppositional defiant disorder	-	1 (0.6%)

Note: N=351; WASI = Wechsler Abbreviated Scale of Intelligence.

Anxiety diagnosis.

Psychiatric status was determined using structured interviews by trained clinicians. Pediatric patients met criteria for generalized anxiety, social anxiety, and/or separation anxiety disorders as primary diagnosis and presenting complaint for treatment. Adult patients were additionally eligible for panic disorder. Healthy participants were diagnosis-free. See supplement.

Auxiliary analyses utilized standard anxiety symptom questionnaires. Youth and their parents completed the Screen for Child Anxiety Related Emotional Disorders (45) and adults completed the trait subscale of the State-Trait Anxiety Inventory (46); data were combined by Z-scoring (see supplement).

Threat conditioning and extinction task

We used an uninstructed, differential threat-learning task previously shown to produce conditioning with acceptable dropout rates in youth and adults (18,26,32,42,44,47,48). In the task (Fig. 1), photographs of two women displaying neutral expressions (49) served as CS+ and CS−. The UCS (presented at CS+ offset) was a 1-second presentation of the CS+ woman displaying fear co-occurring with a 95dB female scream for all participants. The task involved three phases. During pre-conditioning, each CS appears four times. During conditioning, each CS appears 10 times; the CS+ was followed by the UCS with an 80% reinforcement schedule. During extinction, CSs each appear eight times. See supplement for additional details.

Figure 1. Schematic representation of the threat conditioning and extinction paradigm.

During conditioning, one face (CS+) was repeatedly paired with a fearful face co-terminating with a scream (UCS); the other face (CS−) was never paired with the UCS. During extinction, both faces were presented in the absence of the UCS.

Note: CS = conditioned stimulus; UCS = unconditioned stimulus.

Psychophysiology.

SCR was determined by the square-root-transformed difference in base-to-peak amplitude within 5s after stimulus onset, in line with previous studies (32,35,44,47,50). Additionally, startle probes were delivered 5–6s after stimulus onset, and response was measured using eye-blink startle EMG. Primary analyses utilize SCR; supplemental material provide EMG methods, results, and discussion on combining psychophysiology measures.

Subjective fear ratings.

Before and following conditioning, and following extinction, participants rated their fear of the CSs using a 10-point Likert scale (1=no fear, 10=extreme fear) (32,47). These complemented psychophysiological responses to the CSs.

Brain imaging

MRI images (1mm³) were collected for 250 of the participants (71%; 145 healthy [82 females, Mage=21.3 years]; 105 anxious [71 females, Mage=19.0 years]) in a separate visit. Data were processed using FreeSurfer. Analyses tested associations between structural imaging measures and psychophysiological indices, and their moderation by anxiety and age using permutation tests (51). We considered whole-brain cortical thickness, using the TFCE (threshold-free cluster enhancement) statistic (52), and subcortical gray-matter volume (GMV). MRI data from 115 participants appear in previous reports that use different analyses (41,53). Analyses applied familywise-error (FWE) rate correction. See supplement for additional details.

Data analysis

First, we examined anxiety and age effects on differential threat conditioning and extinction through omnibus Anxiety×Age×Phase×CS interactions on SCR to conditioned cues; trial-by-trial analyses complemented analyses on averaged SCR (11). In auxiliary analyses, EMG and self-report data were analyzed in a similar manner. Second, effects on unconditioned responding were tested through the Anxiety×Age interaction on SCR to the UCS. Finally, we examined relationships between brain structure and SCR responses, with the primary analysis using the SCR measure that best differentiated healthy comparisons from patients, as well as moderation of structure-response relations by anxiety and age. To do so, we regressed SCR on cortical thickness and GMV measures. All analyses used general linear models, whereby anxiety group (anxious, healthy) was a between-subjects factor; age was a continuous covariate. Effect-sizes are reported as partial-eta-squared. All tests were two-sided and set significance at α=0.05.

Results

Response to conditioned cues

SCR.

Averaged psychophysiological responding to conditioned cues by task phase is summarized in Figure 2A. Repeated-measures analysis of covariance (ANCOVA) testing the Anxiety×Age×Phase×CS effect on averaged SCR yielded a significant Phase×CS interaction, F(2,696)=21.62, p<0.001, ${\eta }_p^2=0.06$, with follow-up paired-samples t-tests indicating greater response to CS+ relative to CS− during conditioning, t(350)=8.12, p<0.001, but not during pre-conditioning or extinction, ps>0.16. This pattern indicates successful conditioning followed by extinction.

Figure 2. Conditioned and unconditioned psychophysiological threat response.

A) Conditioned skin conductance responses (SCR) by stimulus type (CS−, CS+) averaged across each phase of the task (pre-conditioning, conditioning, extinction), by anxiety group (healthy, anxious). B) Averaged SCR to the unconditioned stimulus by anxiety group (healthy, anxious).

Note: SCR data were square-root-transformed microsiemens units. Pre-cond. = Pre-conditioning, Cond. = Conditioning, n.s. = not significant. Error bars represent one standard error of the mean.

We also observed a main effect of Anxiety on SCR, F(1,348)=10.46, p=0.001, ${\eta }_p^2=0.03$, whereby patients exhibited greater mean response to the conditioned cues across the task relative to healthy controls. Further group comparisons indicated that patients generated stronger responses relative to controls to both CS− and CS+ in each task phase, all ps<0.032.

Additionally, we noted a main effect of Age, F(1,348)=83.90, p<0.001, ${\eta }_p^2=0.19$, indicating decreasing response with increasing age. This effect was qualified by an Age×Phase×CS interaction, F(2,696)=7.77, p<0.001, ${\eta }_p^2=0.02$. Follow-up analyses yielded a significant Age×CS interaction during conditioning, F(1,348)=27.91, p<0.001, ${\eta }_p^2=0.07$, indicating decreased differential conditioning with age, but no age differences during pre-conditioning or extinction, ps>0.86.

No significant Anxiety interaction effects emerged, either across the task or separately during conditioning or extinction, indicating no difference in differential threat learning processes. Additional trial-by-trial SCR analyses are reported in supplement, indicating similar findings. A complementary dimensional analysis of anxiety symptom severity indicated a significant positive association between symptoms and averaged SCR to the conditioned cues, r=0.14, p=0.010.

EMG analyses appear in the supplement. Briefly, these indicate two notable non-significant effects: Phase×CS interaction (p=0.061, CS+>CS− only during conditioning and extinction) and Anxiety main effect (p=0.090, anxious>healthy).

Self-reported fear.

Analyses of subjective fear reports are reported in supplement. Briefly, subjective fear paralleled SCR, demonstrating increased fear of CS+ relative to CS− following conditioning that was diminished following extinction. Moreover, fear of CS+ following conditioning correlated positively with the magnitude of SCR to CS+, supporting convergence of subjective and psychophysiological measures. Finally, as with SCR, we noted a main effect of Anxiety on fear reports, indicating greater fear of conditioned cues throughout the task, but no Anxiety interaction effects.

Brain structure correlates.

Since anxiety group differences emerged in SCR responding across CSs and task phases, averaged SCR across CSs and phases was used to index conditioned responding. Analyses tested the main effect of cortical thickness, as well as moderation by anxiety and age using the three-way Thickness×Anxiety×Age interaction, in predicting the magnitude of conditioned response. This indicated a significant association between cortical thickness and conditioned response (controlling for age and anxiety) in a left-hemisphere cluster extending from dorsomedial to dorsolateral PFC (dmPFC, dlPFC; Fig. 3A, Table 2), whereby less thickness predicted greater conditioned response. Another cluster in left retro-splenial cortex demonstrated a positive association between thickness and conditioned response. Furthermore, anxiety moderated the association between cortical thickness and conditioned responding in bilateral ventral occipital cortex (Fig. 3B), such that patients exhibited a more positive thickness-SCR association in this region. Finally, age moderated association thickness-response associations in several clusters (Fig. 3C), including bilateral posterior-insula and temporal-occipital cortex, right mid-cingulate cortex, and left middle-frontal gyrus. Analysis of GMV revealed age moderation in bilateral hippocampus. The effect of age was consistent across all regions, such that among younger participants, thicker cortex or greater GMV was positively associated with conditioned responding, but with age this association became negative.

Figure 3. Brain structure correlates of psychophysiological response to conditioned cues.

Result of analysis predicting individual averaged skin conductance responses to conditioned cues (CS− and CS+) across the task by cortical thickness, anxiety status (healthy, anxious), and age (in years). A) Association between cortical thickness and conditioned response, controlling for anxiety and age. B) Modulation of association between cortical thickness and conditioned response by anxiety. C) Moderation of association between cortical thickness and conditioned response by age. Each surface’s color reflects −log(p-value) of the threshold-free cluster enhancement statistic; brighter colors represent stronger effects (threshold: p_FWE<0.05).

Note: L = left; R = right.

Table 2.

Location, peak significance level (p_FWE<0.05), and size (number of vertices) of clusters showing significant associations between cortical thickness or gray matter volume (GMV) and magnitude of conditioned psychophysiological response. For GMV, p-values refer to the entire structure. For cortical thickness, p-values refer to the TFCE statistic.

Effect	Location	Peak p-value (FWE-corrected)	Cluster size (vertices)
Cortical thickness
	L dmPFC-dlPFC	0.009	115
	L retrosplenial cortex	0.021	27
Cortical thickness × Anxiety
	L visual association cortex	0.029	16
	R visual association cortex	0.039	15
Cortical thickness × Age
	R midcingulate cortex	0.019	173
	L temporo-occipital cortex	0.025	156
	L posterior insula	0.025	146
	R posterior insula	0.035	92
	R temporo-occipital cortex	0.042	40
	R parieto-occipital cortex	0.046	22
	L visual association cortex	0.046	9
	L ventral medial frontal gyrus	0.047	8
Gray matter volume × Age
	R hippocampus	0.017	-
	L hippocampus	0.033	-

Note: L = left; R = right; dmPFC = dorsomedial prefrontal cortex; dlPFC = dorsolateral prefrontal cortex; TFCE = threshold-free cluster enhancement.

In summary, analyses of psychophysiological responses indicate comparable threat conditioning and extinction between patients and healthy controls. Compared to controls, patients demonstrated increased conditioned SCR responding to both CS− and CS+. Magnitude of this responding was inversely related to dorsal-PFC thickness; anxiety and age moderation effects emerged in other cortical regions as well as bilateral hippocampus.

Response to unconditioned stimulus

SCR.

Analysis of SCR to the UCS indicated comparable response to the unconditioned threat stimulus in the patient and healthy groups, F(1,348)=0.24, p=0.62, ${\eta }_p^2<0.01$ (Fig. 2B). We noted a significant main effect of Age, F(1,348)=34.29, p<0.001, ${\eta }_p^2=0.02$, indicating decreasing response with increasing age. Additional trial-by-trial analyses indicated diminishing unconditioned response across trials and with age (supplement; Fig. S1B), but no Anxiety effects. Of note, these analyses indicate that absence of Anxiety effects on unconditioned responding is not due to ceiling effects.

An auxiliary analysis directly comparing averaged SCR to the conditioned cues with averaged SCR to the UCS within a single model yielded a significant Stimulus×Anxiety interaction, F(1,348)=7.86, p=0.005, ${\eta }_p^2=0.02$, further indicating that the anxiety effect on response was specific to increased conditioned, but not unconditioned, responding. Dimensional analysis indicated that symptom severity and averaged SCR to the UCS were not correlated, r(319)=0.03, p=0.65.

Self-reported fear.

Full statistics are provided in supplement. Briefly, subjective fear of the conditioned cues did not depend on the magnitude of unconditioned response to threat.

Brain structure correlates.

Analyses of cortical thickness and GMV indicated no significant association, either direct or moderated by anxiety or age, between variation in brain structure measures and magnitude of unconditioned response.

In summary, the magnitude of unconditioned threat responses diminished with age. However, unlike anticipatory responses to the conditioned cues, response to the unconditioned stimulus did not differ as a function of anxiety, did not relate to conditioned subjective fear, and did not relate to variation in brain structure.

Discussion

This study examined the associations that anxiety exhibits with conditioned and unconditioned threat responding across age. Three key findings emerged. First, as hypothesized, across age, anxiety patients and healthy comparisons demonstrated comparable differential threat conditioning and extinction. Second, despite intact threat learning, the magnitude of conditioned SCR responding was greater to both CS− and CS+ in patients relative to healthy comparisons. The magnitude of such responding also covaried with subjective fear of the conditioned cues and brain structure in several hypothesized regions. Third, the magnitude of unconditioned psychophysiological responding did not relate to anxiety status, subjective fear of conditioned cues, or variation in brain structure. Together, these findings suggest that differential threat learning remains intact in pathological anxiety. Instead, anxiety involves heightened SCR to both CS+ and CS− but not UCS; the magnitude of such diagnosis-related SCRs also correlates with variation in brain structure.

This study is the largest single report comparing threat conditioning and extinction between anxiety patients and healthy comparisons across development. The findings of comparable differential threat conditioning and extinction in patients and comparison subjects is consistent with prior meta-analyses (3,6,7). Thus, findings do not unequivocally support theories that relate anxiety to aberrant threat conditioning or extinction (3,4). Moreover, the use of a single, established paradigm informs theories on development and anxiety (2,6,25). Importantly, age did not moderate anxiety effects on these processes.

Instead, our findings highlight greater responding to both conditioned cues as differentiating anxious from healthy participants. This is consistent with previous findings generated in separate studies among youth and adults demonstrating anxiety-related enhancement of responding to CS−, CS+, or both (3,6,7). Greater response to CS− has been hypothesized to reflect anxiety-related aberrations in safety learning (54–56); greater response to CS+ has been hypothesized to reflect enhanced threat learning or reduced fear extinction (3,4,57). Both patterns emphasize some form of perturbed learning, and neither hypothesis is fully supported by our data. This calls for alternative explanations of the observed patterns.

A notable finding from the current study stems from comparing responses to the unconditioned stimulus. This finding indicates that enhanced SCRs in patients do not occur to the UCS, suggesting perturbed anticipatory responding, as opposed to acute-threat responding. This distinction could arise from differences in the function of conditioned and unconditioned responses to threat. Unconditioned threat stimuli signal unambiguous, immediate danger; as such, they elicit reflexive defensive responses that require minimal computation to execute (9,29). In contrast, conditioned stimuli only predict the potential as opposed to immediate occurrence of danger; such prediction may be influenced by multiple processes that jointly estimate the probability, magnitude, or proximity of danger and accordingly influence adaptive defensive responding (55,58,59). Excessive conditioned responding may reflect perturbations in any of these processes, each involving biased threat estimates in response to any stimulus that predicts danger.

The uninstructed and probabilistic nature of the conditioning schedule used here may have led anxious relative to healthy participants to view both CS+ and CS− as conveying relatively high levels of danger, leading to greater anticipatory responses in patients to both cues (55,58). Such an effect was also observed prior to the presentation of the first UCS, whereby patients demonstrated greater responding to the initial face presentations during pre-conditioning (supplemental data). As participants were aware of the aversive nature of the paradigm, patients may have shown increased anticipatory responding to the first stimuli presented in the task; this group difference diminished during pre-conditioning as the stimuli were continually non-reinforced.

Several brain-structure correlates of anticipatory threat response were identified. Analyses specifically examined correlations with SCR measures that differentiated patients from healthy comparisons, to inform understanding of clinical psychophysiological correlates. Less left-dmPFC and -dlPFC thickness was associated with greater anticipatory psychophysiological responding. Considerable functional imaging literature implicates these regions, particularly left-sided, in threat learning (33,34). Furthermore, other work suggests that altered function or structure in these regions contributes to maladaptive anticipation (55,60–62) and emotion regulation (63,64) processes. Our results bridge these findings, offering the possibility that dmPFC and dlPFC support effective regulation of anticipatory responses to potential danger. Indeed, preliminary findings from lesioned patients also support this possibility (65).

A positive association emerged between cortical thickness in left retro-splenial cortex (RSC) and conditioned anticipatory response. RSC is one of several regions implicated in threat conditioning (33), and is suggested to mediate the encoding of episodic or contextual memory of CS-UCS associations (66,67). Our results extend these findings by showing that structural variation in this region relates directly to the expression of conditioned psychophysiological responses indicative of diagnostic differences in physiology.

Additionally, the association between thickness in bilateral clusters in ventral visual association areas and anticipatory response varied with anxiety status. Prior research links structure and function in occipital cortex to anxiety disorder clinical features and treatment response (41,68–71). Reciprocal connections between amygdala and visual cortex may account for such findings, as these connections are thought to facilitate processing of biologically-relevant stimuli in the context of threat conditioning (72–76). Our findings add to this literature, potentially linking patients’ increased psychophysiological responding to visual threat stimuli to perturbations in cortical regions mediating visual processing. Additional research using conditioned stimuli of other modalities is needed to explore the specificity of this effect.

Age-dependent associations emerged between several structures and individual differences in conditioned responding. The consistent pattern of age-moderation suggests that a group of regions may constitute a network supporting threat anticipation processes in ways that change with development. Broadly, other work finds these regions to show relatively protracted maturation with age (77,78). Some data suggest that the mid-cingulate cortex, particularly its anterior extent, acts as a key hub in networks mediating threat conditioning, modulation of negative affect, and anticipation (33,55,79). Consistent with these prior findings, our data may suggest that mid-cingulate cortex supports anticipatory responding to threat. Similarly, prior functional and structural imaging work relates posterior insula to threat conditioning (33,37); this region has also been linked to the integration of interoceptive information (80). Given such prior work, our data also implicate the posterior insula in conditioned anticipatory preparation for harm. Finally, associations between GMV and anticipatory responding were also observed in bilateral hippocampus, a structure implicated in threat learning processes, potentially via context representation (33,39,81). As thinning in these regions is associated with greater anticipatory response, it is possible that some of their functions are regulatory, involving integration of somatic, affective, and contextual information. Nevertheless, it is important to emphasize that brain structure might not directly map onto function (82); thus, inferences on functional roles for these regions are limited. Longitudinal studies focusing on both the structure and function of these regions are necessary to more completely understand age moderation of associations among brain structure, function, and conditioned response to threat.

Of note, our findings suggest that neither aberrant differential conditioning nor extinction exhibit strong, direct associations with pathological anxiety across development. Nevertheless, it remains possible that more nuanced anxiety differences in differential learning of anticipatory threat responses exist. One possibility is that analytical challenges in capturing dynamic learning processes mask such subtle differences. Methods that directly model associative learning processes (14,83,84) may be more powerful in identifying such differences. Alternatively, it has been suggested that other effects that derive from threat learning, such as tests of extinction recall or generalization of learned threat, may better capture anxiety deficits (19,26,56,85,86). Such effects may also reflect the elicitation of anticipatory responses (e.g., to generalized stimuli), and could prove valuable avenues for research linking anxiety, anticipation, and response to learned threat.

Exaggerated fear of potential danger is a core feature in the presentation of anxiety symptoms. Here, we identify a potential psychophysiological correlate of this maladaptive anticipatory fear response. Importantly, the magnitude of the anticipatory response differentiated between patients and healthy comparisons, but also correlated with reported fear of the conditioned cues, thereby linking psychophysiological and subjective-fear responses to potential threat. As such, this paradigm provides an experimental setting primed for uncovering the nature of associations among anticipatory psychophysiological responding, subjective-fear, and anxiety symptoms. Follow-up studies could use repeated assessments of conditioned fear alongside anxiety ratings embedded in the threat learning paradigm to examine how anticipatory psychophysiological responses and subjective fear might interact to contribute to the experience of anxiety symptoms (10,26,87–89).

Along these lines, identifying a psychophysiological correlate of a pathological process in anxiety could potentially inform treatment development (5,90). For example, increased anticipatory psychophysiological response could serve as a specific target for interventions that aim to directly reduce physiological arousal, such as particular forms of cognitive-behavioral therapy (CBT) and biofeedback techniques. Future research could explore whether neuroscience-guided interventions, such as brain stimulation methods and neurofeedback (91,92), could potentially downregulate neural processes mediating anticipatory responses or upregulate regulatory processes. Given the absence of anxiety differences in response to the unconditioned acute-threat stimulus, psychotherapy and CBT might focus on addressing anticipation-focused cognitions and somatic responses. Additional research could further explore whether the magnitude of anticipatory psychophysiological response could serve as a biomarker for anxiety treatment outcome.

Several limitations should be acknowledged. First, this was a cross-sectional study, limiting the extent of inference about causality; a longitudinal design would allow stronger inferences about developmental and causal processes (93). Second, this study was not designed to directly link individual differences in threat learning and treatment outcome, thus limiting the scope of therapeutically-relevant inference. Third, establishing baseline (11) for SCR to UCS is inherently challenging due to anticipation once associations have been learned. Here, CS+ and UCS events were separated by an adequate duration as recommended (11); nevertheless, future research should consider this issue. Fourth, since brain structure variably maps onto function (82), inference on the functional role of identified brain regions is limited. Fifth, we measured both SCR and EMG; the use of multiple psychophysiological indicator variables could interfere with their indexing of the target processes (94). Sixth, we used a paradigm that is well-suited for developmental research but uses a pre-set volume level for all participants; this limits comparisons with prior studies in adults in which UCS aversiveness was set individually. Several strengths mitigate these limitations and address general shortcomings in threat learning research (11,27). First, the large sample size increases precision in the estimates of associations (95). Second, participants were carefully assessed and free of medications known to impact threat learning and psychophysiology (11). Third, a wide age range generates inferences on age differences with reasonable statistical power. Finally, task and setting were identical for all participants, reducing measurement confounds and noise.

In summary, the current study examined associations among conditioned and unconditioned responses to threat, anxiety, and age. Our findings highlight anticipatory threat responding as differentiating between patients and healthy controls and identify brain structure correlates of this response. These findings may bear implications for our conceptualization of anxiety and its treatment and study.

KEY RESOURCES TABLE

Resource Type	Specific Reagent or Resource	Source or Reference	Identifiers	Additional Information
Add additional rows as needed for each resource type	Include species and sex when applicable.	Include name of manufacturer, company, repository, individual, or research lab. Include PMID or DOI for references; use “this paper” if new.	Include catalog numbers, stock numbers, database IDs or accession numbers, and/or RRIDs. RRIDs are highly encouraged; search for RRIDs at https://scicrunch.org/resources.	Include any additional information or notes if necessary.
Software; Algorithm	PALM — Permutation Analysis of Linear Models version alpha116	Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutation inference for the general linear model. NeuroImage, 2014;92:381–397 (Open Access)	RRID:SCR_017029
Software; Algorithm	FreeSurfer version 5.3.0	Laboratory for Computational Neuroimaging at the Athinoula A. Martinos Center for Biomedical Imaging; http://surfer.nmr.mgh.harvard.edu	RRID:SCR_001847
Software; Algorithm	SPSS 25	IBM Corp. Released 2017. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.	RRID:SCR_002865
Software; Algorithm	FSL FAST 4	Zhang, Y. and Brady, M. and Smith, S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans Med Imag, 20(1):45–57, 2001.	RRID:SCR_009472
Software; Algorithm	Surf Ice 2	https://www.nitrc.org/projects/surfice/	NA