Safety learning during development: implications for development of psychopathology

Abstract

Fear and safety learning are necessary adaptive behaviors that develop over the course of maturation. While there is a large body of literature regarding the neurobiology of fear and safety learning in adults, less is known regarding safety learning during development. Given developmental changes in the brain, there are corresponding changes in safety learning that are quantifiable; these may serve to predict risk and point to treatment targets for fear and anxiety-related disorders in children and adolescents. For healthy, typically developing youth, the main developmental variation observed is reduced discrimination between threat and safety cues in children compared to adolescents and adults, while lower expression of extinction learning is exhibited in adolescents compared to adults. Such distinctions may be related to faster maturation of the amygdala relative to the prefrontal cortex, as well as incompletely developed functional circuits between the two. Fear and anxiety-related disorders, childhood maltreatment, and behavioral problems are all associated with alterations in safety learning for youth, and this dysfunction may proceed into adulthood with corresponding abnormalities in brain structure and function—including amygdala hypertrophy and hyperreactivity. As impaired inhibition of fear to safety may reflect abnormalities in the developing brain and subsequent psychopathology, impaired safety learning may be considered as both a predictor of risk and a treatment target. Longitudinal neuroimaging studies over the course of development, and studies that query change with interventions are needed in order to improve outcomes for individuals and reduce long-term impact of developmental psychopathology.

1. Introduction

Fear and safety learning are critical features required for survival [1, 2]. These adaptive behaviors are modulated by different brain regions and circuits [1, 2], and there are multiple ways in which an individual may learn safety [3–6]. An individual’s ability to learn safety, or to retain learned information, may be altered in the context of fear and anxiety-related psychopathology—either increasing the individual’s risk for such a mental health disorder or resulting from the disorder itself [7]. Safety learning may also be differentially affected by exposure to stress, trauma, and adversity, and indeed aspects of safety learning may act as biomarkers of psychopathology [8]. Animal models of learning threat and safety have shown robust age-related effects, with altered fear regulation circuitry in the juvenile period compared to the early postnatal period; additionally, puberty marks a significant change in fear expression [9]. Given the developmental changes in the brain that have also been shown in human studies, there may be corresponding changes in fear and safety learning that are quantifiable in clinical settings and may provide predictors of risk and potential treatment targets for fear and anxiety-related psychopathology in children and adolescents [10, 11]. Preclinical and clinical studies have provided insight on the neurobiology of safety learning in both youth (juvenile, in preclinical studies) and adult populations. Prior reviews of this literature have mostly focused on adults (e.g. [8, [13–14]); the reviews of the literature on development of safety learning have looked specifically at infancy in preclinical models [12], fear conditioning, and specific forms of safety learning—extinction learning [13] and conditioned inhibition [14]. The present review aims to describe the broad field of work regarding safety learning over the course of development from childhood to adolescence including a translational perspective from animal to clinical research, with implications for biomarkers and clinical interventions.

In this review, we first provide an abbreviated overview of fear learning (section 2) followed by descriptions of multiple experimental methods to assess safety learning (section 3). Within each subsection describing a particular method we explore neurobiological mechanisms associated with these forms of safety learning from childhood and adolescence, informed by the more extensive adult literature, and include extant relevant research on psychopathology. In section 4, we review studies of safety learning across developmental windows, focusing on childhood and adolescence. We conclude in section 5 by exploring the potential of safety learning as both a biomarker and a treatment target in pediatric cohorts.

2. Fear Conditioning/Fear Learning

In most studies of safety learning, threat or fear is established first, in order to provide contrast for safety learning [15]. This fear is typically acquired using Pavlovian conditioning (depicted in Figure 1), in which a neutral cue (conditioned stimulus, CS+) is repeatedly paired with an aversive stimulus (unconditioned stimulus, US) such as a shock, a loud noise, or a fear-eliciting facial cue [16]. Repeated presentation of the CS+/US pairing leads to the development of a conditioned response (CR) to the CS+ even in the absence of the US [16]. The fear CR may be increased psychophysiological responses, such as startle response, elevated electrodermal activity (for example, skin conductance response—SCR), increased heart rate, or increased self-reported fear [17]. Such CRs may be indicative of behavioral traits—for example, SCR may be indicative of emotional reactivity, and responses to emotional stimuli have been shown to be associated with levels of psychopathology [18]. In the context of trauma-related psychopathology, such as posttraumatic stress disorder (PTSD), re-experiencing symptoms may be thought of as persistent CRs—behaviors and emotions which may be evoked by trauma-associated stimuli, even in the absence of danger [19]. Additionally, increased physiological reactivity is observed in individuals with PTSD during fear conditioning [20], however this finding varies based on the type of trauma exposure, age at exposure, and chronicity of exposure, among other factors [21]. The knowledge an individual has regarding the CS+ predicting the US is termed contingency awareness [22], and while not necessary for generating a CR [23, 24], contingency awareness is a strong correlate of the CR [25]. Fear learning is predominantly a subcortical process, engaging features of the limbic system such as the amygdala, the insula, and the anterior cingulate cortex (ACC) [26]. The amygdala is a key region implicated in fear conditioning, and in individuals with a number of fear and anxiety-related disorders, including phobic disorders, hyperactivity of the amygdala and insula is observed during fear learning [27, 28]. While animal and human studies have largely attributed the role of the ventral medial prefrontal cortex (vmPFC) to extinction of learned fear and recall of previously extinguished learned fear, new evidence from a recent lesion study also indicates a causal role of the vmPFC in fear learning [29].

Figure 1.

Schematic of traditional fear conditioning (top, red) and extinction learning (bottom, green) paradigms. The CS− signals the absence of danger during fear learning, as the CS− is never paired with the US. During the extinction learning phase, individuals gradually learn that the CS+ no longer signals threat and is now safe. The different backgrounds for fear and extinction learning indicate that these two processes usually occur in different contexts—for example, fear learning may occur in the context of a trauma, and that context may include cognitive, experiential, and social information that contribute to the context. Extinction learning, then, typically occurs in a different context from the traumatic experience.

Fear-potentiated startle (FPS) combines Pavlovian fear conditioning and the acoustic startle response (ASR) to measure not only fear but also safety learning [30] and a heightened FPS response to safety cues is considered a candidate biomarker for PTSD [8, 31, 32]. The startle response is a ubiquitous, cross-species response to strong exteroceptive stimuli, recorded in humans using electromyogram (EMG) of the orbicularis oculi (eye blink muscle) contraction [33, 34]. Exaggerated startle is a hallmark symptom of patients with PTSD [35], and an auditory stimulus is most commonly used to generate the response in experimental settings [36]. Brown and colleagues (1951) first studied how fear could potentiate the acoustic startle response (ASR) using a fear conditioning paradigm [36]. The FPS response provides a good candidate for a biomarker because it is related to PTSD symptoms [37, 38]. Kolb (1984) suggested that exaggerated startle in PTSD is a conditioned response to trauma-related stimuli [39]. The fear response component is measured as the increased magnitude of the startle response to the auditory stimulus that elicits the startle response (acoustic startle probe) when measured during presentation of the CS+ pairing, compared to presentation of the startle probe alone [40, 41]. A unique feature of FPS is that it can be measured in animals and humans using analogous methods and therefore is ideally suited for translational research; however, as listed above, other fear responses have been used as CR’s in fear conditioning studies.

3. Safety Learning

There are multiple experimental paradigms for testing safety learning used in animal and human research [15]. In the following subsections, we describe four such paradigms (see Box 1 for a visual representation of each paradigm). Safety signals are learned cues that predict the non-occurrence of an aversive event, therefore acting as inhibitors of fear and stress responses [15]. Unlike external inhibitors which attenuate fear through increased attention or orientation to novel stimuli, safety signals are conditioned inhibitors which act to reduce fear responses as a result of learning [15]. Safety learning usually does not erase learned fear responses, but rather represents new learning to supersede the prior fear learning under safety conditions [42–45]. Learning safety is more cognitively demanding than fear acquisition, therefore safety learning involves cortical components of the brain in addition to limbic regions [26]. Prior reviews highlight diminished safety learning and weakened ability to modulate fear responses using safety cues in adults with PTSD [46, 47]. A similar effect is observed for individuals with anxiety disorders, who show less inhibition of fear responses to the non-reinforced safety cue compared to controls [48–52]. Startle responses to safe conditions are specifically predictive of onset of anxiety disorders but not unipolar depressive disorders [48, 49]. This predictive observation is similar to that of earlier work indicating that while individuals with anxiety do not show significantly different responses to threat cues compared to controls, they do show significantly elevated CRs to safety cues [40, 49, 53]. Two meta-analyses confirm that this effect of increased fear responses to the CS− is robust across anxiety disorders (including generalized anxiety disorder, panic disorder, social phobia, other specific phobias, and PTSD) [54, 55]. Overgeneralization of threat cues is marked by diminished discrimination between threat and safety signals [46, 51, 56, 57]. In individuals with PTSD, such weakened ability to differentiate can lead to hypervigilance and exaggerated physiological responses to trauma-related stimuli and situations in the context of safety [46].

Box 1: Visual Definitions of Safety Learning.

Box 1. provides visual summaries of the safety learning paradigms described below.

In differential conditioning (A), the CS+ (red circle) is presented with an aversive cue (US; orange warning triangle). Another conditioned stimulus, the CS− (green circle), is not presented with the US. Individuals over time learn that the CS+ signals danger while the CS− signals safety. In extinction learning (B), the CS+ that was previously paired with the US is repeatedly presented without the US, such that individuals learn that the CS+ no longer signals threat. In conditioned inhibition (C), a stimulus A (yellow apple) is paired with the US, however when the stimulus X is present (green X) with stimulus A, the US is not presented. Individuals learn that stimulus X signals safety, because A is only paired with the US when presented without X. In conditional discrimination (D), stimulus A when paired with X (yellow X) is presented with the US. Stimulus B (green book) is never paired with the US. When X is presented with B, the safety signal, the US is not present, as it was when X was paired with A. Stimulus B signals safety, and individuals learn to differentiate the meaning of X depending on whether it is paired with stimulus A or stimulus B. As a transfer test, stimulus A may be presented with stimulus B, resulting in a reduced fear response if transference of safety information has successfully occurred. A novel stimulus Y may be presented with stimulus A following training. The purpose of the AY compound is to test whether the fear in inhibited due to the novelty of the compound cue acting as an external inhibitor. If the fear response elicited by AY is larger than that elicited by AB, then B is considered to be functioning as a true safety signal and not an external inhibitor.

3.1. Differential Conditioning

During fear conditioning, another neutral cue may also be presented without co-occurrence of the US, and this is termed the non-reinforced CS− [58]; see Figure 1. While individuals are learning the association between the CS+ and the US, they concurrently learn that the CS− predicts the absence of the US, therefore the CS-does not elicit fear responses [13]. In such differential conditioning paradigms, the CS− represents a safety signal to inhibit fear, resulting in differential responses between the threat (CS+) and safety (CS−) cues [55, 59] (see Box 1, “A. Differential Conditioning”). Differential responding to threat and safety cues is likely partially mediated by the prefrontal cortex (PFC) and hippocampus [60] in healthy individuals.

One common method for determining whether individuals are able to cognitively discriminate between stimuli is self-reported contingency awareness [61]—that is, how much individuals expect to experience the US when presented with the CS+, the CS−, or another stimulus [23]. Evidence to date supports an effect of age on the ability to discriminate between threat and safety cues. For children, awareness of the CS+−US contingency can be enhanced by increasing the number of CS+−US pairings [62]. However, for the safety cue CS−, children seem to remain more uncertain about the absence of the US than adolescents and adults regardless of the number of training trials [62]. It is important to note that some studies have not found such age-related differences—one study of 8–13 year old children did not find an age-related difference in self-reported contingency awareness, and 80% of youth were able to correctly distinguish between CS+ and CS− at the end of the acquisition phase [63]. This discrepancy may be due to differences in stimuli—the latter study was an aversive conditioning paradigm in which CSs were neutral faces and the US was a fearful face paired with a scream [50], instead of differently colored shapes as in the former studies. These emotionally salient social stimuli may facilitate better discrimination than paradigms which use shapes as CSs and a loud noise or air blast as the US.

Individuals may also rate the valence of stimuli—that is, how threatening, negative, or positive they may evaluate the stimulus to be. Children are less likely than adolescents and adults to have stronger negative and positive valence evaluations of CSs, potentially reflecting a lesser degree of differentiation between threat and safety cues [62]. When looking at youth with psychopathology, however, clinically anxious children and young adolescents have exhibited diminished discrimination between threat and safety cues compared to non-anxious youth following conditioning, with emotional and neutral faces used as stimuli [64]. Lau and colleagues (2008) replicated and expanded on these data, finding that adolescents with anxiety disorders showed greater self-reported fear ratings to both the CS+ and the CS− relative to healthy counterparts [50]. Fear ratings to both threat and safety predicted avoidance of a second study visit [50]. Such avoidant behavior is important to observe, given that avoidance may interfere with safety learning [65], as well as with seeking or remaining in treatment [66, 67].

Psychophysiological data, including fear-potentiated startle (FPS) and skin conductance response (SCR), have also been used to determine whether individuals are able to learn threat and safety signals [61]. In studies of youth ages 8–13, children under the age of 10 have shown diminished discrimination between threat and safety cues based on contingency awareness [63, 68]. Older children (10 and up) and adolescents have demonstrated more adult-like discrimination patterns [63, 68]. In one differential conditioning study utilizing faces as CSs, healthy children exhibited greater startle magnitude when presented with the CS+ compared to the CS; this difference increased with age, indicating that older children showed greater discrimination between threat and safety stimuli [63]. Differential conditioning may emerge earlier in the lifespan when measured by psychophysiological data than conscious discrimination as represented by contingency awareness/expectancy data. Ingram and Fitzgerald (1974) investigated SCR to threat (CS+) and safety (CS−) cues and found that as early as 3 months of age, full-term infants showed greater magnitude of SCR to CS+ compared to CS− [69]. Children in middle childhood also showed increased SCR to CS+ compared to CS− in the acquisition phase, with older children reporting more fear to stimuli more closely resembling the CS+ in a paradigm that included morphed stimuli [70]. Finally, in the same study, children ages 9–10 showed better fear discrimination and memory about the CS+−US contingency compared to children ages 5–8 [70]. 13–17 year olds show more adult-like discrimination than children; larger SCRs and startle blink reflexes to CS+ compared to CS− have observed in 13–17 year olds [71]. Comparing adolescents to adults, adolescents overall may present with higher skin conductance level than adults, however unlike children, both adolescents and adults have shown greater SCRs to the CS+ compared to the CS−[13], replicating findings from an earlier study [72]. A third study supported this finding, in that both adolescents and adults showed greater magnitude of SCR to the stressful sound (a woman’s scream) compared to a neutral tone [73]. However, adolescents showed less cognitive discrimination between CS+ and CS− compared to adults based on verbal fear ratings of stimuli [74]. This cognitive finding may be mediated by dlPFC activity—here, dlPFC activation was associated with safety cues more strongly than threat cues in adults, which the authors posit to represent greater ability to distinguish between stimuli, resulting in more proficient verbal report of explicit threat and safety [74].

Differences in discrimination also emerge across the age range in relation to psychopathology. SCR to threat cues, for example, has been found to be greater in youth under the age of 10 who had higher anxiety compared to youth under 10 with lower anxiety [68]. Unlike the younger children, FPS to threat and safety cues was greater in youth above the age of 10 who had higher anxiety compared to those with lower anxiety [68]. After controlling for age and trauma exposure, only FPS to the safety cue was predictive of anxiety levels [68]. Research to date regarding startle responses in youth with PTSD is limited; adults with PTSD show greater FPS responses not only to the CS+ but also to the CS− compared to controls during fear conditioning in a differential conditioning paradigm, representative of enhanced fear learning [75] which may be interpreted as overgeneralization or hypervigilance behaviors that in part define PTSD. Greater trait anxiety in adolescents has similarly been associated with impairments in discrimination, compared to adolescents with lower trait anxiety, as evidenced by startle responses [76]. Behavioral inhibition is a temperament trait that presents during toddlerhood and persists during childhood characterized by withdrawal in the presence of novelty, general fearfulness to unfamiliar situations, and is a risk factor for development of anxiety disorders—especially social anxiety [77]. Adolescents with high levels of behavioral inhibition and adolescents with lifetime anxiety have shown increased startle response to safety cues [77]. Longitudinal studies have indicated that with increasing age, anxiety symptoms become more negatively associated with amygdala-dlPFC connectivity in behaviorally inhibited youth during tasks that require maintaining attention to a threat [78]. In contrast, those with low behavioral inhibition showed a more positive association between anxiety and amygdala-dlFPC connectivity during the same task [77]. Relatedly, Fairchild and colleagues (2008) defined conduct disorder as a persistent pattern of serious aggressive and antisocial behaviors with disregard for the rights of others, and conduct disorder may present early in life with a persistent life course, or may be limited to adolescence [79]. Previous research in adolescent males with early-onset and adolescent-onset conduct disorder has shown reduced startle responses when viewing affective pictures compared to healthy controls [79]. These males with conduct disorder also had lower skin conductance levels at baseline, and less SCR discrimination between CS+ and CS− in the fear conditioning task, compared to healthy controls [79]. There was no effect of group on SCR to the US, suggesting that group differences were not due to blunted aversiveness of the US in the conduct disorder group [79].

Developmental sex differences in discrimination may be associated with hormonal activational effects during puberty, as estradiol levels and genotype have been linked with sex differences in impaired safety learning in adults [80]. Pituitary adenylate cyclase-activating polypeptide (PACAP) is a protein that plays a role in regulating the cellular stress response [81]. Adult women with high levels of PACAP and the CC variant of a gene coding for the PACAP receptor (PAC1R) have shown greater startle responses to the safety cue CS− compared to women with low levels of PACAP of carriers of the G allele for PAC1R [81]. This effect of PACAP levels or PAC1R genotype was not observed for adult men [81].The physiological changes during puberty may in part explain some of these sex differences, given the influence of sex hormones on fear and safety learning [82]. In a longitudinal sample of youth ages 8–16, greater pubertal development was associated with increased FPS to the threat cue but not the safety cue [83]. Similarly, in another study using differently colored shapes as the CS+ and the CS−, healthy children ages 10–13 showed larger baseline startle responses associated with older age, but not pubertal status [84]. However, FPS was significantly greater for those in the mid to late pubertal stage group compared to those in the pre to early pubertal stage group [84]. Girls from this sample reported more negative valence (aversiveness) of the US compared to boys, and, longitudinally, girls also showed greater fear potentiation compared to boys [84], yet this effect was not replicated [83]. As has been observed in adults, girls who carry the CC risk genotype of the PAC1R gene, which is associated with stress response and PTSD[81], have increased FPS to safety cues compared to girls with the G allele or boys with either genotype [85]. These data suggest that some sex differences in risk may emerge early.

Sex differences in fear conditioning may be associated with risk for negative clinical outcomes. One study of 8–13 year olds exposed to trauma found that girls showed less discrimination between threat and safety cues compared to age-matched boys, and conditioned fear to danger cues was specifically predictive of self-blame and fear of repeated trauma in girls [86]. For boys, higher intrusive symptoms of PTSD predicted fear responses to the CS+ even after controlling for trauma exposure [86]. In addition to genetic risk, families may confer risk through parental behavior. Individuals whose parents had major depressive disorder showed increased magnitude of startle response, and in turn, their children showed a sex-specific effect: girls show increased baseline magnitude of startle response throughout the experiment but not boys [87]. This sex-specific finding was replicated in adolescent youth of parents with a history of anxiety disorder—the larger startle response was greater in magnitude for children with a family history of anxiety, and the response was largest for high risk females [88]. Parental anxiety may inhibit children from properly learning how to identify safe and non-safe cues [89], and the effects of parental psychopathology may be more apparent in girls. The mechanism through which this may be occurring remains unclear.

Another way that family dynamics can impact development of safety learning is via childhood maltreatment, which occurs in an estimated one out of every four children[90]. Maltreatment is defined by abuse and/or neglect leading to injury, death, emotional harm, or serious risk of harm due to a caregiver’s actions or their failure to provide adequate care [91]. Hein and Monk (2017 [92]) conducted a meta-analysis and described a brain network particularly susceptible to the effects of childhood maltreatment—the social information processing network (SIPN) [93]—due to its composition of structures which undergo significant postnatal development [94]. Maltreated children had blunted SCR to CS+ cues, and lack of CS+/CS− discrimination during fear conditioning [95]. Maltreated children also had reduced amygdala and hippocampal volumes; reduced volumes were negatively correlated with SCR to CS+ during the early phase of conditioning [95]. These data showcase how different early life experiences may lead to differential effects on fear learning, and likely safety learning as well. Maltreated children had responses similar to those observed in individuals who experience chronic, interpersonal traumas yielding passive responses—which may be linked with inability to escape—and blunted sympathetic nervous system responses [21].

Taken together, the literature to date appears to indicate diminished ability to discriminate between threat and safety cues in childhood compared to adolescence and adulthood, and youth with higher anxiety symptoms also show such an impairment across childhood and adolescence. These findings may be reflective of overgeneralization of fear which interferes with safety learning in youth with anxiety disorders and high trait anxiety—fear generalization is in fact a hallmark of anxiety disorders [96]. Measuring responses to threat and safety cues within differential conditioning paradigms may be an appropriate way to assess fear generalization. Additionally, as with adults, there is sex-specific variation in differential safety learning in adolescents. Caregivers and the early environment may impact development of safety learning, neural networks related to threat and safety processing, and be associated with psychopathology.

3.2. Extinction Learning

During extinction learning, a conditioned stimulus (CS+) that was previously paired with an aversive cue (US) to elicit a fear response (CR) is then repeatedly presented without the US, so that the individual learns a once-threatening stimulus is now safe [45]. Extinction learning does not result in erasure of the original fear memory and related responses; rather, extinction learning creates a new memory based on the CS+ no longer predicting the US [97]. When the CS+ is encountered again in the future, the new memory competes with the original memory [97]. This process is considered central to extinction learning [98, 99], with inhibitory pathways via the vmPFC recognized as being part of the neurobiology of extinction [22, 100, 101]. In healthy human subjects, the vmPFC acts as a top-down regulator of amygdala activity in the context of multiple forms of safety learning [102–104], in that greater activity in the vmPFC is associated with lower activity in the amygdala [105, 106]. In this sense, the vmPFC may be thought of as the neural substrate for inhibitory learning [22]. The effects of inhibitory learning are often specific to the context in which extinction has occurred [22, 107]. This means that fear may return when the previously extinguished CS+ is encountered in a different context [22]. Context can be manipulated between fear conditioning and extinction learning (see the two backgrounds representing context in Figure 1). The context in which fear is learned is the “threat context”; the context in which safety is learned (and threat extinguished) is the “safety context” [58]. The hippocampus encodes such information [108] and may be involved in the contextual components of extinction learning and extinction recall [22].

Adolescents with anxiety disorders have demonstrated deficient safety learning in a fear conditioning/extinction paradigm—compared to healthy counterparts, adolescents with anxiety disorders showed heightened fear ratings to both the threat and safety cue following conditioning, which remained elevated even following extinction [50]. Even when no differences in self-reported arousal were found, anxious children compared to controls maintained greater difficulty in extinguishing conditioned psychophysiological responses as measured by SCR [109]. Clinically anxious children and adolescents, compared to healthy controls, showed both larger FPS response and greater SCRs following extinction learning, indicative of a failure to learn safety in those with anxiety disorders compared to those without [64]. Some studies have found that differences in SCR and FPS do not always emerge in anxious compared to non-anxious adolescents even when they reported more fear to the CS+ and the CS− during both conditioning and extinction [110, 111]. As in differential conditioning, such discrepancies between studies of extinction may be due to differences in stimuli or differences in experimental cohorts. Determining the root of these variations will be vital for safety learning to be considered a candidate biomarker for fear and anxiety related psychopathology.

Although there are no published studies of extinction in children and youth with PTSD, adults with PTSD compared to trauma exposed controls have shown elevated FPS responses to the CS+ during early and middle stages of extinction, representing a higher ‘fear load’ to be extinguished [75]. Additionally, individuals with more severe re-experiencing PTSD symptoms had greater FPS to CS+ compared to those with less severe re-experiencing symptoms [75]. Re-experiencing symptoms may reflect individuals’ weakened ability to inhibit fear [19, 112, 113]. During extinction learning, higher vmPFC activity is associated with higher levels of emotional regulation and lower levels of psychopathology [114–117]. Conversely, hypoactivation of the vmPFC during extinction is associated with impairments in safety learning, and this observation has been repeatedly made in individuals with PTSD [27, 28]. Adults with anxiety disorders show impaired extinction learning, such that those with an anxiety disorder have greater CRs to the CS+ during the extinction learning phase when the cue is no longer paired with the US compared to healthy controls [54, 118, 119].

Extinction recall is the ability to retain the learned safety information regarding the extinguished CS+ given a passage of time since the initial extinction training. Following completion of an extinction learning paradigm, children with anxiety rated the extinguished CS+ as more fear provoking than the CS− compared to healthy controls, suggesting a failure to encode new inhibitory memories [64]. When youth were repeatedly assessed on extinction learning paradigms over multiple test sessions, level of fear at the first visit predicted greater avoidant behaviors at the second visit [50]—indicating that greater fear during a conditioning task may be predictive of later avoidance, a specific symptom of PTSD. Evidence from one study indicated that trauma-exposed youth (6–11 years) were more avoidant of both previously extinguished threat cues as well as the safety cue based on distance metrics of approach to cues, and fail to re-extinguish fear to the threat cue during extinction recall, compared to non-trauma exposed control youth [120]. The distance from the threat cue in trauma-exposed youth was associated with greater activation of the dACC and anterior insula [120]. The inability to retain learned safety information and to re-extinguish fear accompanied by greater reactivity of salience-related regions may indicate mechanisms through which trauma exposure confers increased risk for psychopathology [120]. This may also affect developmental maturation, as the child becomes more oriented towards scanning the environment for threat and preparing for possible fight or flight, and less oriented towards social, educational, and emotionally relevant stimuli in the environment [121, 122]. In another neuroimaging study, adolescents and adults were presented with the CS+ and morphed images of the CS+ and CS− three weeks after fear conditioning and extinction. Response patterns specific to both adolescents and adults with anxiety were revealed [123]. These patterns indicated that anxious adolescents and adults had lower subgenual ACC activation when appraising threat compared to healthy individuals, and overall more self-reported fear to the CS+ compared to healthy individuals [123]. Diminished fear extinction and extinction recall in youth, particularly when context is manipulated, may have neural bases similar to that of adults: reduced hippocampal and vmPFC volumes confer a weakened ability to extinguish fear and recall safety information about a previously conditioned CS+ [16, 60, 118, 119, 124–126]. Studies of adolescents with PTSD show reduced gray matter volume (GMV) in both regions [127], and those without PTSD have higher hippocampal volumes and age-normative vmPFC volumes [128]. This volumetric reduction in the vmPFC, as well as reduced GMV in the ventrolateral PFC (vlPFC) and dlPFC, is associated with greater severity of PTSD in adolescents [129]. 10–17 year old offspring of parents with past or current anxiety disorders had greater activation of the amygdala and dlPFC to safety cues (CS− > CS+), but did not have impairments in extinction learning [130]. On the other hand, 10–17 year old youth with a current anxiety disorder showed differential fronto-amygdala circuitry compared to healthy controls for CS+ > CS− and impaired extinction [130].

Sex differences may also play a role in development of extinction learning. Low estrogen levels in naturally cycling women with PTSD have been associated with slower extinction learning in a FPS paradigm compared to women with PTSD and higher estrogen levels [131]. Functional neuroimaging studies reveal that adult women with high levels of estradiol have increased vmPFC activation during extinction learning, compared to women with low levels [132]. Women with higher estradiol were also better able to recall extinction learning a day later compared to the women with lower estradiol [133]. Based on such findings regarding the effects of circulating gonadal hormones on safety learning, it is likely that puberty may influence extinction learning during this key developmental period.

Evidence to date indicates that extinction learning is a process which involves integration of higher order cognitive structures, such as the PFC, and limbic regions, including hippocampus and amygdala. Differential connectivity between these regions may underlie developmental and pathological variation in extinction learning, defining possible biomarkers of risk to psychopathology and a treatment target. While some studies have investigated differences in extinction learning among youth exposed to trauma, youth with anxiety disorders, and healthy youth, none have specifically looked at extinction learning in youth with PTSD. Finally, as with differential conditioning, sex differences and hormonal effects may be associated with variation in extinction learning and may emerge during the adolescent period.

3.3. Conditioned Inhibition (A+/AX−)

Conditioned inhibitors are safety signals which predict the absence of the aversive US [22] and attenuate the fear CR [134]. In this laboratory model, individuals learn that while the stimulus A when presented alone signals danger, presentation of stimulus A in compound with a second, neutral stimulus X (AX−) signals the absence of the US, so that individuals learn to inhibit fear responses in the presence of X [59]. This paradigm has been used less frequently in human research [135] than in animal studies [136, 137]. This method of safety learning works by capitalizing on lateral and basolateral amygdala (BLA) GABAergic neurons, GABAergic neurons in the intercalated neurons, neurons projecting from the infralimbic region (IL), and other inputs to the central amygdala inhibiting its outputs [15]. A recent study which conducted a parallel human and rodent conditioned inhibition paradigm identified an additional role of the ventral hippocampus, elucidating a subpopulation of ventral hippocampal neurons which project to the prelimbic cortex (PL, the rodent analog of the human dACC) but not IL (the rodent analog of the human anterior vmPFC) or the BLA [138]. This neuronal subpopulation was more responsive to safety and compound cues rather than threat cues, and activity in this population was associated with freezing behavior in the rodents [138].

The first published study of conditioned inhibition in youth with anxiety disorders compared to age-matched controls indicated no behavioral or physiological differences in those with an anxiety disorder, compared to those without, during fear acquisition [139]. However, during testing of conditioned inhibition (in which the safety compound and novel compound cues were presented), children with anxiety disorders had greater SCRs to all cues and greater expectation of the US when presented with the novel compound cue compared to controls [139]. Additionally, children with anxiety disorders showed greater vmPFC activity during presentation of the safety vs. novel compound, while children without anxiety disorders showed the opposite response pattern [139]. Given the small sample size (n=17 and 18 per group), it is not possible to make conclusions given the preliminary nature of the study, however this study does show feasibility for use of this paradigm in developmental cohorts, and presents foundational evidence for differences in this form of safety learning for youth with and without anxiety disorders. Recently presented data at a national conference suggested that for children under 10 years old with anxiety, discrimination was impaired to an even greater extent than their age-matched controls who already showed less discrimination than older children [140]. This is still an emerging field of research in children and adolescents, and there are very limited findings reported in the literature [14].

3.4. Conditional Discrimination (AX+/BX−)

The conditional discrimination paradigm is a modification of the conditioned inhibition paradigm [6]. Stimulus A is paired with the aversive US in the presence of X (AX+) [6]. X is also combined with stimulus B to signal absence of the US (BX−) [6]. Following training, presenting A and B together yields a reduced response compared to the response elicited by AX+ if an individual is successfully able to transfer learned safety information from B to A [6]. To confirm that the safety signal is not acting as an external inhibitor, a novel stimulus Y can be presented with stimulus A following training [15]. If AY does not elicit as large of a reduction in conditioned fear response as seen for AB the effect has not been due to external inhibition [6, 30].

As is the case for conditioned inhibition, there are only a few studies that have begun to investigate conditional discrimination during development [14]. A recent conference presentation on 71 healthy participants ages 12–30 showed greater hippocampal activation to the safety compound cue (AB) in individuals with low vs high trauma [141]. Additionally, hippocampal connectivity associated with the safety compound (AB) increased across age [141]. In a recently published dissertation, children ages 8–11 were trained on an AX+/BX− FPS paradigm; children whose parents had an overprotective parenting style had higher FPS to the safety transfer “AB” compound trials than the learned safety cue BX−, indicative of diminished transfer of safety information [89]. Such an impairment may cause individuals to perceive the world as more threatening than it actually is, and this may be associated with a hyperactive amygdala and/or weakened inhibitory prefrontal control of the amygdala. For children whose parents had greater severity of depression, lower FPS to the threat cue AX+ and the safety cue BX− was observed [89]. This may be indicative of a blunting effect of parental depression where defensive mechanisms are employed to reduce the aversiveness of unpleasant stimuli leading to attenuated fear responses. Although there are limited studies in pediatric cohorts, in adult veterans with high symptoms of PTSD showed impaired transfer of safety information on the AB trials in a FPS paradigm [8, 38]. Such overgeneralization and impaired inhibition are key features of PTSD and were not observed in individuals with depression [8, 32].

There are multiple paradigms and mechanisms in which safety learning is studied in adults and in youth. All represent inhibitory learning, yet may have different neural underpinnings [142] and utilize different ways of transfer of safety information—conditioned inhibitors use safety cues to transfer inhibition, while extinction utilizes contexts [46]. Preclinical work in rodents provides an opportunity to develop this understanding from an neuroanatomical perspective, and studies such as that of Meyer and colleagues (2019) are exemplary of how to translate these models and concepts from preclinical to clinical research [138]. The evidence in adults indicates sex-specific differences in safety learning and a relation between altered safety learning and psychopathology, prompting a need to understand functional safety learning in youth, and how safety learning may become dysfunctional during development in the context of adversity that may lead to psychopathology following exposure. Notably, children with anxiety show less discrimination between threat and safety cues [140, 143]. Safety learning behaviors of children are also impacted by parental, especially maternal, trauma exposure and psychopathology, above and beyond that of the child’s own trauma exposure and psychopathology [144]. Evidence from both adults and youth indicate that the alterations in safety learning may be associated with structural and functional changes in key frontolimbic regions—namely amygdala, prefrontal cortex, and hippocampus—implicated in fear and safety learning. These brain regions may be especially sensitive to the effects of environment at different times during development, which may lead to diminished fear and safety learning as well as psychopathology (see Figure 2.).

Figure 2. Brain and behavioral development from infancy to adulthood.

While developmental trajectories are not fixed and may vary based on sex, environmental factors, region/network, and even by individual, here we provide indications of potential developmental trajectories given what has been measured and reported in the literature to date. While arrows are used for simplicity to reflect time, it should be additionally noted that development is not necessarily linear.

4. Developmental Periods

Experience and genetics both influence how the brain and behaviors such as safety learning develop and change during maturation [145]. Throughout development, adverse experiences may impact biological and psychiatric phenotypes of the individual. The impact of such experiences may differ depending on the type, duration, frequency and timing of adversity. The broader context in which the adversity occurs also matters: comorbid psychiatric conditions, food insecurities, and environmental toxins may increase risk of negative outcomes, while supportive family environments may be protective [146]. Exposure to adverse events during different developmental periods may affect the ways in which individuals learn safety and potentially alter this behavior, as well as the underlying neural correlates of safety learning described above. During childhood, subcortico-subcortical connections are refined such that children are able to engage sensorimotor skills to perceive and respond to their environments, detect threat signals, and respond to reward cues [147]. Early life stress and trauma is associated with regional abnormalities in core components of the frontolimbic circuitry [148]. Multiple studies indicate amygdala hypertrophy [129, 149–151] and prefrontal and hippocampal atrophy [152, 153]. These results align with regional functional alterations [117, 154–157]. Adolescence marks a significant period of increased reactivity to emotional and social cues as subcortico-cortical networks are refined [147]. Later in adolescence and into adulthood, cortico-cortical networks reach maturation [147]. More specifically, the amygdala, hippocampus, and prefrontal cortex are considered key regions for fear and safety learning, which are also implicated in the pathophysiology associated with fear and anxiety related disorders, including PTSD. The development of the amygdala is characterized by rapid initial growth in the postnatal period [158–160], followed by sustained growth marked by volumetric peaks between ages 9–11 [158–160], and subsequent volumetric decline; the overall trajectory is typically understood to be an inverted U-shaped growth curve. It should be noted that while some consider this decline to be mediated by synaptic pruning [158–160], it may also be mediated by expanding white matter tracts. The 9–11 age period has been defined by some as preadolescence, and it is also within this period that hippocampal volumes are believed to reach volumetric peaks [160, 161]. Unlike the amygdala, some studies support continued volumetric increases in the hippocampus in males 13–14 and 18–21 years of age [162], with postmortem studies indicating myelination of the hippocampus continuing into adulthood [163], and animal studies indicate the hippocampus as a region of continued neurogenesis [164]. Evidence for structural sex differences has been observed in the amygdala and hippocampus [165, 166]; both regions express receptors for sex hormones [167–169], and the female amygdala reaches its volumetric peak ~1.5 years sooner than the male amygdala [160]. However, another study has shown larger hippocampal and amygdala volumes in more sexually mature adolescents, with greater volumes in males compared to females [170]. However, there is great variation in brain development, with environment and experience playing important roles [159].

Resting state functional connectivity (rsFC) is a measure of the brain activity in the absence of a directed task or evoking stimuli. Using this measure can aid in understanding spontaneous co-activation of intrinsically connected networks. From the ages 7 to 31 there are distinct increases in segregation and integration of networks [171]. Regions relevant to fear and safety learning, such as hippocampus and mPFC, are part of the default mode network, which as a whole is typically more active during rest than during task; the network which specifically responds to both danger and reward is termed the salience network and includes the amygdala [171]. These functional specialization patterns that are part of maturation likely arise from genetic and molecular mechanisms interacting with the environment [172].

One significant environmental factor impacting brain development is exposure to childhood adversity. The way in which different types of adversities shape neurodevelopment is defined by McLaughlin and colleagues as the dimensional model of adversity and psychopathology, with threat experiences more likely to be associated with alterations in amygdala activity and other salience network features, while deprivation experiences are considered more likely to be associated with alterations in executive functioning and reward processing networks [173, 174]. There are two distinct hypotheses for adverse experiences altering maturation trajectories [148]. The first, the cumulative risk approach, proposes that with an increasing number of adverse exposures, individuals are at greater risk for negative physical and mental health outcomes and nonspecific alterations in maturation [175]. The second proposes a dimensional approach, such that there are differential effects based on the type of adversity: threat may be associated with changes in fear learning and emotional processing (frontolimbic circuitry), whereas deprivation may be associated with alterations in cognition and executive functioning [95, 173, 176]. Accelerated maturation may be adaptive—in a harsh or unpredictable environment, reaching adult-like phenotypes at an earlier age would be favorable to maximize reproductive success prior to potential death [177]. The interaction between the hypothalamic pituitary adrenal (HPA) axis and the hypothalamic pituitary gonadal (HPG) axis is hypothesized to underlie this accelerated maturation process. Stress responses to adversity activate the HPA axis, which can in turn accelerate onset of puberty by activation of pubertal hormones via the HPG axis [178–180]. This accelerated maturation pattern may be more commonly observed in youth exposed to violence and other types of threat-related adversity, while an attenuated maturation pattern may be more likely in youth exposed to neglect and other types of deprivation-related adversity. Data from the Fragile Families and Child Wellbeing Study support the hypothesis of differential effects of different early life experiences, including childhood violence exposure and social deprivation, finding greater amygdala activation and sustained amygdala activity in response to angry faces during adolescence in children exposed to violence, compared to children who experienced social deprivation [181]. For children who experienced social deprivation, decreased activation of the ventral striatum in response to happy faces during adolescence was observed [181].

As development proceeds, there may be important periods of time at which environment and experience can have unique influences on the maturation of the nervous system and behavior [182]. While the human brain is adaptable and dynamic throughout the lifespan [182], there are periods of greater plasticity, and as such, Knudsen (2004) defines sensitive periods as “when the effects of experience are particularly strong on a limited period in development” [183]. Others define sensitive periods as “a time period (or life stage) in which experience shapes a trait to a larger extent than the same experience does in other time periods.” [184, 185] For example, the “early experience hypothesis” posits that the first three years of life are an important period given the expanse of change in brain development that occur during this period [182, 186]. Not only does developmental timing of experiences matter in the context of sensitive periods, but also does the amount, or “dose”, or the experience [187, 188]. There is also not one fixed key developmental period for one system, but rather there is the potential for multiple important periods which may differ by system. Atypical maturation of the brain regions described in the former two paragraphs involved in threat and safety learning during distinct sensitive periods may lead to abnormal learning and put individuals at risk for fear and anxiety related disorders [189]. In this section, we focus on how experience and environment during important developmental periods are associated with changes in brain structure and function, as well as behavior. Understanding these key developmental periods will be important for identifying windows of opportunity in which systems are most amenable to intervention in order to reduce severity and long-term incidence/effects of psychopathology.

4.1. Childhood

As mentioned above, peak age for amygdala volumes has been reported as preadolescence (ages 9–11) in healthy individuals, with the most rapid growth rates occurring in the postnatal period [158–160, 190, 191]. Within these windows, development of the amygdala may be particularly susceptible to environmental effects, including early adversity. Broadly, the U.S. Department of Health and Human services states that early adversity includes child abuse and neglect, exposure to violence, and family economic hardship—all of which can lead to both short and long term health consequences [192]. One such consequence may be diminished safety learning, mediated by altered brain development. For example, evidence from an international adoption study of institutionalized children showed that children adopted at after 15 months of age had larger amygdale volumes than children adopted at earlier ages, and age of adoption significantly correlated with amygdala volume when controlling for age, IQ, and presence of an anxiety disorder[151]. At age two, institutionalized youth were allocated to either further institutionalization or foster care; those who were placed in foster care showed normative white matter development when assessed at ages 8–11 [193]. Those who continued to be institutionalized showed reduced integrity of white matter tracts involved in limbic circuitry, fronto-striatal circuitry, and sensory processing—further emphasizing the fact that early environments confer a strong influence on neural development [193]. These structural findings have been corroborated by functional imaging studies. Previously institutionalized children showed greater amygdala reactivity to emotional faces and reduced eye contact in social situations compared to control children who were never institutionalized [154]. Notably, the patterns of amygdala reactivity in institutionalized children was more similar to adults than to healthy children, suggesting that increased vigilance for threat may be an adaptation leading to accelerated development [154]. A recent retrospective study found that exposure to adverse childhood experiences (ACEs) at ages 3–4 (as well as 16–17) explained greater variance in right amygdala reactivity when viewing threatening versus neutral faces, more than ACE exposure at other points of development [194]. Together, these studies highlight that amygdala structure and function are susceptible to early life experiences [160].

Greater amygdala reactivity may be associated with diminished safety learning, and conversely, less hyperactivity associated with increased PFC regulation may be a resiliency trait facilitating safety learning [195, 196]. When processing threat and safety-related stimuli in an emotional faces viewing task, children showed positive connectivity (which suggests that the PFC and amygdala may be activated at the same time) whereas adolescents and adults exhibited negative connectivity (meaning that the higher PFC activity is associated with lower amygdala activity) [116, 117]. Children and adolescents who experienced maternal deprivation early in life showed amygdala hyperreactivity and altered development of amygdala-prefrontal connectivity; while comparison children showed the typical immature pattern (amygdala-mPFC positive coupling) the previously institutionalized children displayed a more mature pattern (amygdala-mPFC negative coupling) [117]. Children who exhibited the negative connectivity pattern earlier in life following childhood adversity had lower levels of psychopathology, indicating that this “accelerated maturation” may confer resilience [117]. While amygdala connectivity with other subcortical/limbic structures seems largely stable between the ages of 4 and 23, the transition from childhood to adolescence (around age 10) is marked by this distinct shift in amygdala-cortical functional connectivity [197]. A decrease in functional connectivity between the amygdala and parahippocampal gyrus/posterior cingulate occurs, and a decrease in functional connectivity between the insula and superior temporal sulcus is also simultaneously observed [197]. A timeline delineating what is currently hypothesized regarding development of behavioral and neural processes can be found in Figure 2.

Important periods of plasticity in brain development may differ by brain region. For example, one study indicated volumetric differences in gray matter when measured in adulthood showed important developmental periods based on when childhood sexual abuse (CSA) occurred [198]. CSA between the ages of 3–5 was associated with reduced hippocampal volume and CSA between ages 14–16 was associated with reduced frontal volumes [198]. In males specifically, neglect, but not abuse, prior to age 7 predicted hippocampal volume in adulthood [199]. Conversely in females, abuse, but not neglect, at ages 10, 11, 15, and 16 predicted hippocampal volume [199]. In terms of contextual memory, which is critical for extinction learning [200], fMRI studies of encoding in healthy children show developmental changes in the medial temporal lobe and PFC, as well as increased connectivity between the two regions [200]. Literature to date suggests that contextual memories and neural correlates of context encoding do not vary with age in 8 to 19 year olds [201]. Larger hippocampal volumes were found in those with better contextual memory, which was not found in children exposed to violence [201]. Timing of key developmental periods may also differ between types of learning mechanisms [202]. As reviewed in section 4.1, several studies using psychophysiological measures have indicated that safety learning, and discrimination between threat and safety, becomes more refined after age 10 [63, 68]. In addition, one study of FPS in trauma-exposed children and adolescents showed that CS+/CS− discrimination was affected by maternal availability. Children <10 years of age whose mothers were near them during fear conditioning showed improved safety learning compared to those whose mothers were not available [203]. Adolescents were not affected by maternal availability, in that they showed good discrimination regardless of the mother’s status, overall indicating that childhood is more likely to be a sensitive period for maternal buffering to modulate learning of danger and safety signals [203].

Recent research from Teicher, Ohashi, & Khan indicates network susceptibility to effects of childhood adversity through age 21 [204]. Thus, both childhood and adolescence are important periods to investigate the relation between experience, brain development, and behavioral phenotypes. As is clear from the present review, there is a limited body of research regarding key developmental periods in safety learning. The evidence to date is not conclusive, but rather identifies areas of ongoing investigation worth further exploration, particularly given the potential for use of safety learning as a biomarker or even a treatment target.

4.2. Adolescence

As the average age of onset for anxiety disorders falls around age 14 [205], with prevalence increasing during late childhood and through adolescence, this developmental period may be considered an important time for identifying those at risk for adult anxiety [206]. Indeed, as Casey and colleagues (2016) emphasize [207], risk for psychopathology and criminally relevant behaviors is elevated at this stage [208, 209] while significant changes to the brain are taking place [210]. Changes to the underlying neurobiology may precede the behavioral expression of anxiety disorders, and such changes could also confer increased risk to trauma-related disorders such as PTSD, which does not have a clear age of onset given the dependence on exposure to a traumatic event [211]. After age 12 there is between 3:1 and 2:1 odds ratio for anxiety in girls relative to boys; given that estrogen plays a significant role in the neurobiology of fear in women [81, 131–133], it is likely that neuroendocrine pubertal changes are in part at play for the manifestation of symptoms [212]. Timing of exposure to adversity in the context of puberty is also relevant to subsequent amygdala function. In adults with a history of physical maltreatment between ages 3 and 6 (pre-pubertal exposure), blunted amygdala response to emotional faces versus shapes was observed [213]. On the other hand, adults with a history of peer emotional abuse between ages 13 and 16 (post-pubertal exposure), showed increased amygdala responsivity to emotional faces versus shapes [213].

A repeatedly replicated finding across animal and human studies is attenuated extinction learning during adolescence [208, 214–217]. In humans, this finding remains significant even when controlling for sex and trait anxiety [214]. Whether or not this finding is replicated in safety signal learning is unclear, however for extinction learning, this may be due to developmental maturation being reached for the amygdala, while the PFC is still developing [214]. This is termed the “imbalance model” reflecting a hyperactive subcortical amygdala and hippocampus, and an underdeveloped, hypoactive PFC [218, 219]. From a more nuanced understanding, the refinement of limbic circuits linking subcortical (including amygdala and hippocampus) features with cortical (including the vmPFC and other prefrontal regions) features is coincident with adolescence [208, 219] and may underlie the behavioral perturbations observed not only as pathology, but also more broadly as risk-taking behaviors. Rodent studies corroborate this hypothesis—in mice, lack of plasticity in prefrontal regions during adolescence is associated with blunted regulation of fear extinction [214]. A similar mechanism may be involved in discrimination between threat and safety signals. As stated above, children have greater difficulty differentiating threat and safety, and underdeveloped functional circuits which involve the PFC may lead to a weakened ability to form distinct threat categories. Indeed, when actively labeling threats, adolescents reported less discrimination based on self-reported fear ratings of CSs between threat and safety compared to adults, and adolescents were more likely than adults to utilize subcortical structures which reach maturation earlier during differential conditioning compared to adults [74]. Those adolescents who were better able to differentiate between cues showed greater engagement of later-maturing prefrontal regions [74]. Amygdala reactivity to negative cues in healthy youth decreases with age in tandem with greater structural and functional connectivity with the vmPFC [116, 129, 220, 221]. The changes in these specific regions are likely indicative of similar patterns across respective neural systems—association, limbic, subcortical, and executive. Functional connectivity studies have confirmed reduced coupling between amygdala, hippocampus, and vmPFC at rest in individuals with a history of childhood trauma [222, 223]. Resilient youth who were able to properly regulate emotions showed effective coupling between the amygdala and prefrontal cortex [224, 225].

In summary, important periods are evident for development of gray matter, white matter circuits, and brain function. Therefore, developmental timing of experiences is important, and can result in differential psychopathology and behavioral performance on safety learning paradigms. For a summary of the literature to date regarding key developmental periods, see Figure 2.

5. Clinical Significance of Safety Learning

Unfortunately, children may be exposed to a variety of heterogenous adverse experiences—including directly experiencing physical or sexual abuse, witnessing war trauma, being neglected or maltreated, and living through a global pandemic. Because of this heterogeneity, it is clear that a “one size fits all” model cannot be suitable for establishing clinically viable biomarkers of risk and resilience, nor can such a model be suitable for treatment. Differences in safety signal learning, and its underlying neurobiology, may provide candidate biomarkers and/or treatments for conditions such as PTSD and anxiety, and can be considered for clinical application and further investigation.

5.1. Safety Learning as a Predictor of Risk

Controlling for age and trauma exposure, FPS to safety cue predicts child anxiety levels in youth ages 11–13, indicating that diminished safety signal learning may be a risk factor for anxiety disorders [77, 226]. Children with anxiety and at risk for anxiety compared to healthy controls show larger anticipatory and unconditioned SCRs during fear learning across stimuli—including to safety cues—as well as larger orienting and anticipating SCRs during extinction learning [48]. Because the responses in those with and those at risk for anxiety do not differ, it appears as though these behaviors may be prodromes for anxiety disorders in youth. Similarly, adolescents with high levels of neuroticism—a risk factor for anxiety disorders—show elevated startle response to safe conditions [227]. Elevated fear responses to safety cues may be behavioral representations of overgeneralization, which may be a symptom of anxiety disorders that presents early on in the course of pathology such that it may be an indicator of risk. Such alterations may be the result of diminished emotion regulation, and may both increase risk for fear and anxiety-related disorders as well as be a result of the disorder. For example, youth with history of abuse have poorer emotion regulation capacity, putting them at increased risk for PTSD and other psychopathology in adulthood; resilient counterparts are able to reappraise negative emotions and lower reactivity [228–232].

Parental psychopathology may indicate risk for diminished safety learning in their offspring [144], which may then confer additional risk for fear and anxiety-related disorders, given that fear conditioning and extinction learning mechanisms are moderately heritable [233]. Indeed, we have presented above how intergenerational factors are associated with differences in safety learning in offspring [144]. Simultaneously behavioral testing of parents and offspring may enhance the predictive validity of safety signals as heritable predictors of risk, and two studies of parent-child dyads have shown evidence for vicarious fear learning in children acquired from parents, which may be indicative of a potential mechanism contributing to the transmission of fear and anxiety disorders in families [234, 235]. Dyadic research approaches that assess parents/caregivers and youth when investigating safety learning as a biomarker would advance the field with a comprehensive understanding of the familial influences on offspring phenotypes [88, 144, 203, 234–237].

Safety learning may posit a viable biomarker for trauma-related psychopathology and other outcomes following early life stress and adversity in developing youth. Recommendations for future research to utilize learning paradigms to both better understand such disorders in youth and to direct treatment include moving towards multi-day, more nuanced, and more ecologically valid paradigms as well as conducting treatment prediction and mediational studies with data collected at multiple timepoints [238].

5.2. Enhancing Safety Learning as a Therapeutic Approach

Anxiety and fear-related disorders, including PTSD, are of great concern in youth given that diagnoses are associated with lower academic achievements and high rates of comorbidities which may continue into adulthood [239]. Therefore, enhancing individuals’ ability to learn safety may be of therapeutic benefit, and targeting brain structures that underlie safety learning may facilitate threat/safety discrimination and improve extinction learning. Examples would include neurostimulation of the vmPFC and hippocampus, as regions that are implicated in safety learning and inhibit amygdala-mediated fear responses. Therapeutic interventions which may indirectly affect these regions may also be of benefit. Current treatments, primarily cognitive behavioral therapy (CBT), which includes prolonged exposure (PE) therapy, have been identified based on findings from adult populations. There are no evidence-based pharmacotherapies available for adolescent PTSD in particular, altogether highlighting a need for treatments that are developed with an understanding of the neurobiology unique to youth.

Therapies that enhance emotion regulation and target underlying neural circuits may confer greater ability to dampen reactivity to negative stimuli and allow youth to better interpret and respond to their environments without hypervigilance, avoidance, or hyperarousal [240]. CBT is the most common evidence-based behavioral treatment for anxiety disorders [113]. Such treatments have been primarily developed and studied in adults, yet an estimated 31.9% of adolescents have an anxiety disorder and 5% have PTSD [205, 241]. The Child Mind Institute Children’s Mental Health Report states that up to 80% of children with anxiety disorders do not receive treatment. Medications including serotonin-selective reuptake inhibitors (SSRIs) are commonly used in child and adolescent anxiety. However, some parents prefer not to use medications for their children. Psychotherapies for youth have mostly been confined to CBT. While some studies show large effect, others are not as optimistic on improved outcomes over treatment as usual (TAU). Yet a review of CBT for adolescents with anxiety indicates that two thirds of youth respond favorably to CBT, with moderate to large effects compared to control conditions [242]. For one group of anxious youth who underwent CBT, treatment responders had SCRs similar to non-anxious youth, in that SCRs were elevated at the beginning of extinction learning and progressively declined across extinction [243]. For those anxious youth who did not respond to treatment, no change in SCR over extinction was observed [243]. CBT may be a beneficial way to improve safety learning—more specifically extinction learning—and psychophysiological measures may be indicative of treatment response. Trauma-focused cognitive behavioral therapy (TF-CBT) has shown benefit to improving emotion regulation and reducing PTSD symptoms in adolescents [244]. Youth receiving TF-CBT with high levels of comorbid depression have less positive treatment outcomes, and studies examining the treatment effects in comorbid anxiety are lacking [245]. For those youth who do not respond to CBT, diminished returns in some adolescents could be hypothesized to be due to reduced extinction learning during the adolescent phase [246]; this could potentially be recovered via increased number of exposure trials during extinction [247]—translating to an increased number of exposure-based CBT sessions. Finally, reduced responsivity to CBT in approximately one third of adolescents may also be due to underdeveloped fronto-amygdala circuitry which likely contributes to both individual differences in anxiety phenotypes and their treatment [248].

Extinction learning serves as the laboratory model of exposure therapy, and therefore understanding the underlying mechanisms is critical for enhancing treatments and their outcomes [22]. Exposure therapy is a specific form of CBT, the goal of which is to reduce maladaptive behaviors related to situations or cues associated with a traumatic event, a specific phobia, or feelings of panic/anxiety [249]. Here, context becomes increasingly important such that exposure therapy should occur across multiple contexts in order for inhibitory learning to be effective not just in the treatment setting, but in the real-world setting where the feared stimulus is most likely to be encountered. Thus, virtual and augmented reality-based exposure therapy may offer new opportunities for enhanced therapeutic outcomes. It is important to note that while safety learning processes may improve therapeutic outcomes, paradoxically, safety signals present during therapy may hinder outcomes. For example, explicit safety signals and behaviors themselves may actually need to be removed [250] so that they do not interfere with successful extinction learning [251, 252]. These safety signals could include other persons, objects, settings, or behaviors which effectively distract or reduce the intensity of the exposure experienced [250], such that while distress is acutely relieved, when such safety signals are not present, fear and distress returns [253]. The impact such safety signals and behaviors depends on their quantity and strength [254], and therefore in some cases safety signals and behaviors may not always diminish returns of exposure therapy [255–257]. Craske and colleagues (2014) recommend gradually phasing out safety signals when necessary to reduce likelihood of dropout; however others suggest immediately removing safety signals whenever possible for best outcomes [250, 258]. The present review has aimed to focus on mechanisms of safety learning and how these mechanisms may be leveraged and enhanced in the treatment setting, rather than advocate for the use of safety signals in the context of exposure therapy.

Pharmacological agents can also be applied to enhance extinction learning and extinction recall. A randomized controlled trial of children ages 6–14 with dog or spider phobia treated with a single session of PE or PE enhanced with D-cycloserine (DCS) found that at one week follow-up, the DCS group showed less avoidance and less fear when exposed to the stimulus in a different context, compared to the PE only group [259]. Rodent studies of adolescents and adults show DCS to specifically enhance extinction recall [247, 260] and chronic administration of SSRIs paired with extinction training to prevent return of fear [261, 262]. Whether pharmacologically or behaviorally, treatments aiming to enhance generalization of safety learning are likely to have broader positive effects. Considerations for exploring efficacy and underlying mechanisms of other interventions should be prioritized, given that a “one size fits all” model is not sufficient for treatment of diverse populations, and 50% of both adults and children with anxiety do not adequately respond to the current gold standard treatments such as CBT and SSRIs [263]. Combination of these treatments does yield better results; therefore, combination treatments and adjunct therapeutic options should also be considered. For those who experience deficits in safety learning and extinguishing learned fear, it may be appropriate to consider combination treatment or alternative therapies that are non-contingent on the ability to learn safety and extinguish fear [264].

An advantage of understanding brain structures and circuits involved in safety learning is that these circuits can serve as targets for neurostimulation. Having a brain target would allow the use of methods like repeated transcranial magnetic stimulation (rTMS), transcranial direct current stimulation (tDCS), and other stimulation methods. This can be done in conjunction with psychotherapy and pharmacotherapy. rTMS has been FDA approved for treatment-resistant depression (the primary target being the dlPFC) and obsessive compulsive disorder; research is ongoing to evaluate possibility of use for treatment of fear- and anxiety-related disorders as well [265]. In preclinical animal studies, brain stimulation and pharmaceutical studies have shown that enhancing PFC function prior to extinction training leads to more efficacious extinction learning[266, 267]. Because this research is still emerging, the work to date has only been done in adults. Multiple rTMS studies have used the dlPFC as a stimulation target for reducing symptoms of PTSD [268–271], panic disorder [272, 273], and generalized anxiety disorder with success [274]; one study also used tDCS with clinically significant improvement for generalized anxiety disorder [275]. One study used rTMS of the vmPFC paired with one of two conditioned cues during extinction learning and found that SCRs during extinction recall were reduced for the CS paired with TMS during extinction learning [276]. When pairing active rTMS of the dlPFC compared to sham with exposure therapy in a small pilot sample of n=9, a large effect for improvement of hyperarousal symptoms was observed [277]. Non-invasive stimulation methods hold promise for reducing posttraumatic stress and anxiety symptoms, and may also enhance exposure therapy. As more studies highlight such efficacy, as well as confirm safety, future research should explore the potential use of these methods in youth, with a focus on the specific brain regions and circuits differentially implicated in safety learning in developing youth.

Mind-body interventions may also show promise for youth as a form of preventative community intervention, early intervention following trauma exposure or in disease prodromal phase, treatment for those endorsing less severe psychopathology or refusing other treatments [278, 279], and adjunctive treatment in more severe cases. If individuals are unable to properly inhibit fear and learn safety, mind-body interventions like mindfulness, as well as dance/movement therapy, art therapy, and yoga, may provide skills to reduce hyperarousal, hypervigilance, and help cope with re-experiencing [280–290]. In young adults ages 22–29, mindfulness-based interventions have shown improvements in mindful attention, attachment, rejection sensitivity, and posttraumatic stress symptoms [291]. One small (n=38) randomized controlled trial of mindfulness-based stress reduction (MBSR) to address consequences of early life stress in adolescents (anxiety, depression, substance use) found that MBSR reduced depressive symptoms, cannabis use, and stress reactivity as assessed by self-report and cortisol response [292]. As the improvements may be relevant to emotional regulation, mindfulness-based approaches may also enhance safety learning and reduce overall trauma-related psychopathology. In fact, one study of adults who participated in a 4-week mindfulness intervention delivered via phone showed that training in mindfulness reduced spontaneous recovery of fear responses to a previously extinguished threat cue compared to waitlist control—potentially indicative of an improved ability of those who trained in mindfulness to retain learned safety information [293]. By becoming more rooted in the present, individuals may better be able to integrate contextual cues to properly learn safety as well as maintain learned safety. By engaging in moderate aerobic exercise through movement-based modalities, individuals may also be able to recover deficits in hippocampal volumes and function related to stress and depression [294, 295], as well reverse negative structural and functional effects of inflammation on the PFC and amygdala [296–301]. In doing so, individuals may be better equipped to extinguish fear and learn safety [302], since the hippocampus, PFC, and amygdala are all critical structures involved in these processes [302]. Moderate aerobic exercise and deep diaphragmatic breathing can also aid in improving vagal tone, resulting in increased heart rate variability (HRV) [303] and improved top-down inhibition of sympathetic arousal to non-threat (i.e. safe) stimuli or threat-related stimuli in safe contexts [304]. As safety signal learning is associated with PTSD phenotypes, and new evidence supports a role of HRV in moderating these associations [304–306], therapeutic modalities that target HRV may confer an indirect benefit to safety learning. Vagal nerve stimulation also positively affects stress physiology as measured via brain imaging, blood-based biomarkers of inflammation, and wearable devices, pointing to new approaches for preventing and treating stress-related conditions like PTSD [307]. Vagal nerve stimulation can be manipulated via surgical implantation, non-invasive external stimulation, and deep diaphragmatic breathing [307]; again this may in turn result in indirect benefits to safety learning behaviors.

The evidence base supporting multiple therapeutic modalities to both extend and enhance safety learning behaviors and reduce psychopathology in developmental cohorts continues to grow. The use of biomarkers to identify at-risk youth and inform treatment, and a wealth of diverse of methods available for both current use and potential future use holds promise for personalized medicine which considers development.

6. Conclusions and Future Directions

In this review, we comprehensively described some of the most commonly used methods of safety learning in humans that have been translated from rodent models. We then explored the neural underpinnings of these processes, and the variation that may exist in behavior and the neural bases of that behavior, in adults with and without psychopathology. This laid the foundation for understanding variation in safety learning processes in youth, and how they may be associated with a number of common psychiatric concerns, which may peak in prevalence around adolescence. It is within this adolescent phase that we see a shift in fronto-amygdala connectivity, which is key to safety learning and may explain differences in extinction learning and recall in adolescent populations. This may also underlie diminished returns for one third of adolescents who are less responsive to CBT-based treatments. The majority of youth, however, respond well to CBT. With these clinical implications in mind, we explored key developmental periods for safety learning, and how early environments and experiences may result in differences in safety learning and incidence of psychopathology. We suggest that while threat-related experiences such as trauma exposure may be associated with exaggerate fear responses, impaired safety learning, and lead to more hyperreactive phenotypes, maltreatment and neglect may lead to blunted responses. Finally, we discussed the potential role of safety learning as a predictor of risk and target for treatment. Diminished learning of safety signals appears to be a predictor of risk for fear and anxiety-related disorders in older children and adolescents, and facilitating safety learning could be of therapeutic benefit for both children and adults. For adolescents, and for children and adults as well, therapies that heavily emphasize safety learning may be enhanced with additional sessions, pharmacotherapy, and/or mind-body interventions that center patients for treatment and increase efficacy, as well as provide life-long coping skills. While the literature describing the neurobiology of safety learning is large and continually growing in adults, studies of safety learning in developmental cohorts (especially longitudinal studies) are just starting to emerge. Studies that probe associations between safety learning, psychopathology, and changes over the course of psychiatric treatment could be enhanced and expanded, in order to improve outcomes for individuals and reduce long-term impact of developmental psychopathology.