Introduction

Exaggerated and inappropriate responses to threats are characteristic of anxiety disorders and often lead to avoidance behaviors and functional impairment. Common behavioral treatments for anxiety disorders aim to reduce these inappropriate fear responses through gradual exposure, which is based on extinction principles. Fear conditioning studies in both animals and humans provide valuable insight into how fears are learned and extinguished [see recent reviews, 1•, 2•]. Through knowledge gained from basic research, treatments for individuals with anxiety disorders can be refined.

In this review, we will summarize the most recent findings in the literature concerning fear and extinction learning, concentrating on empirical manuscripts of human research published in the past year. First, we will briefly review key concepts and the neurobiological basis of fear conditioning, extinction and extinction retention. Next, we will highlight potential therapeutic targets, namely cognitive processes that may be altered in anxiety and/or leveraged to enhance treatment response. These potential targets include fear generalization, memory consolidation, and reconsolidation. We will also discuss factors related to effectiveness, efficiency and durability of extinction-based treatments. Then, we will consider adolescence as a key developmental stage when 1) these perturbations may emerge and 2) interventions may impact the course of anxiety disorders.

Fear Conditioning, Extinction and Extinction Retention

In differential fear conditioning studies, the unconditioned stimulus (UCS; e.g., shock) is paired with a neutral conditioned stimulus, CS+ (e.g., large ring). Another stimulus, the CS- (e.g., small ring), is not paired with the UCS (Fig. 1–1). After pairing, the CS+ alone can elicit the fear response (e.g., skin conductance response, startle response, subjective fear ratings). Through conditioning, fear responses to the CS+ become elevated relative to the CS-. During extinction, fear responses decrease with repeated CS+ presentations in absence of the UCS (Fig. 1–2). When re-exposed to the CS+ at a later point, the CS-threat association formed during fear acquisition and the CS-no threat association formed during extinction compete to determine whether a fear response is elicited or inhibited (i.e., extinction retention) (Fig. 1–3). Fear may return in different contexts (i.e., renewal), with UCS exposure (i.e., reinstatement) or with time (i.e., spontaneous recovery) [3].

Fig. 1
figure 1

Standard fear conditioning paradigms highlight three phases: 1) Fear acquisition, 2) Extinction and 3) Extinction Retention. During fear acquisition, fear responses are learned when a conditioned stimulus (CS+) is paired with an unconditioned stimulus (e.g., a shock; UCS). Meanwhile, a neutral stimulus (CS-) is never paired with the UCS, resulting in little to no fear response. During fear extinction, the CS+ and CS- are repeatedly presented without the UCS; thus, safety is learned, and fear responses diminish. When re-exposed to the CS+ following extinction training (i.e., extinction retention), fear and safety memories compete when re-exposed to the CS+. Depending on the circumstance, fear response may return or be inhibited. In this review, we expand the focus to processes beyond the standard fear conditioning paradigm: A) Fear Generalization. Following acquisition, fear generalization may occur, where stimuli that appear similar to the CS+ elicit fear responses in proportion to their similarity with the CS+. B) Fear Memory Consolidation. Fear memories are later stabilized in a process known as consolidation. C) Extinction Memory Consolidation. Extinction memories then undergo consolidation, where the safety associations are stabilized. D) Reconsolidation. When extinction learning is conducted within the reconsolidation window, learned fear responses may be persistently inhibited such that safety associations are more easily recalled. In cases of extinction training outside the reconsolidation window, fear response mimics standard extinction retention, where learned fear memories are often recalled

Full size image

Consistent with animal work [see reviews, 4, 5], functional neuroimaging studies of fear conditioning and extinction in humans have implicated several key brain regions, e.g., the amygdala, hippocampus, and medial prefrontal cortex (mPFC) [e.g., 6]. Simply, the amygdala is involved in both fear conditioning and extinction [e.g., 7]. The hippocampus modulates fear acquisition and extinction by discriminating related stimuli [8], providing contextual information [9, 10], and governing explicit knowledge of the CS-UCS association [11]. PFC regions also modulate the fear response, but in different ways. While the dorsal anterior cingulate (ACC) is positively associated with fear expression during acquisition [12], the ventromedial PFC inhibits the amygdala and mediates fear reduction during extinction and extinction retention [13, 10].

Past and More Recent Work

Individuals with anxiety disorders exhibit difficulties distinguishing threat-safety boundaries. Most notably, individuals with anxiety demonstrate exaggerated fear responses to both conditioned stimuli relative to healthy controls [see review, 14], possibly due to increased worry [e.g., 15], ambiguity [e.g., 16], or contextual threat [e.g., 17]. Although anxious and healthy groups show similar levels of differential fear conditioning (i.e., similar CS+ > CS-) [see review, 14], anxious groups exhibit extinction-related deficits relative to healthy groups. For example, fear responses to the CS+ may extinguish at a slower rate. Additionally, fear of the CS-, the safety cue, may be elevated during extinction in anxious individuals [see review, 14].

Extinction

Given these extinction deficits, numerous attempts have aimed to enhance extinction learning using a variety of pharmacological agents [see review, 18•]; however, few studies have detected direct effects on extinction learning. A recent study found that, relative to placebo controls, two-week administration of escitalopram to healthy adults prior to fear conditioning facilitated faster extinction [19]. New evidence also suggests that D-cylcoserine (DCS) given prior to exposure therapy accelerates the efficiency, not efficacy or durability, of treatment response in social anxiety disorder patients [20]. While trying to enhance extinction learning seems to be a sensible approach to reduce anxiety, fear and extinction learning activate a cascade of cognitive processes that extend beyond learning itself. Targeting fear generalization, memory consolidation, and reconsolidation may yield novel avenues for addressing inappropriate fear responses and anxiety.

Fear Generalization

Following fear learning, fear may broaden or generalize to stimuli sharing similar characteristics. For example, if someone is bitten by a particular dog, fear may generalize to other dogs. In fear generalization paradigms in a research setting, the CS+ and CS- may be two different sized rings, and the generalization stimuli may be morphed images varying in size between the CS+ and CS- [21]. As shown in Fig. 1-A, fear generalization is expressed as a quadratic pattern of fear response, with higher fear to stimuli most similar to the CS+ and fear levels decreasing as similarity declines [21]. Generalization effects of learned fear have been reported using both self-report and psychophysiological measures [22, 21]. Initial evidence using these indices suggests fear generalization is stable over 8 months [23•]; therefore, treatment manipulations targeting generalization can be attributed to the intervention rather than changes over time.

More recently, the neural underpinnings of fear generalization have received attention. Several brain regions are sensitive to subtle differences in threat and safety using fear generalization paradigms. In neuroimaging studies in healthy adults, insula responses showed generalization gradients with maximal response to the CS+, suggesting sensitivities to threat. The ventromedial prefrontal cortex (vmPFC), on the other hand, has shown the opposite pattern with highest response to the CS- [24, 25•], suggesting sensitivities to safety. In Lissek et al. 2013, the hippocampus is also implicated in discriminating safety, with an activation pattern similar to the vmPFC. Additionally, hippocampal-vmPFC functional connectivity increased with greater dissimilarity to the CS+ [25•].

Anxious adults have exhibited evidence of overgeneralization (i.e., fear generalizes to more stimuli) compared to healthy controls. Adults with panic disorder and generalized anxiety disorder (GAD) showed reduced quadratic patterns of subjective and startle response along the stimulus continuum compared to healthy controls [26, 27], indicating more fear generalization. Consistent with these behavioral and psychophysiological indices, women with GAD also showed a less steep gradient in vmPFC activation than healthy controls, which predicted diagnostic status [28•, 29]. Additionally, less vmPFC cortical thickness and greater functional vmPFC connectivity with the amgydala and parahippocampal gyrus also predict lower levels of vmPFC threat-safety discrimination [29]. Relative to healthy adults, adults with anxiety disorders may have a lower threshold to detect CS+ features, or alternatively, deficient recruitment of vmPFC to inhibit fear responses.

Stimulus complexity may also influence fear generalization. Some CS stimuli used in fear conditioning paradigms are simple (e.g., colored squares, sized rings) [13, 21], while other studies use complex stimuli (e.g., different women) [30, 22]. Particular stimulus properties may acquire threat-value, and the CS- may dictate such features [31]. For example, in Fig. 1–1, size differentiates threat (e.g., large ring) and safety (e.g., small ring), rather than any other feature (e.g., color). Evidence suggests that extinguishing the most salient feature of a complex stimulus first may be optimal in preventing fear at retrieval [32]. However, even though separate features of the CS+ may be extinguished through generalization stimuli, fear returns when these features are presented simultaneously [31], suggesting that the CS+ may be viewed holistically. Future studies should investigate how best to extinguish generalizable features to maximize fear reduction.

Complexity of fear learning also extends to the environmental cues associated with fear (i.e., context). Relative to extinguishing a single context, extinguishing multiple contexts attenuates renewal and reinstatement effects [33–35]. However, it is unclear whether the reduced fear after extinguishing multiple contexts is due to increased generalization of extinction or additional knowledge gained by alternating contexts. Regardless, treatments conducted in the clinic must be generalized to personal surroundings; therefore, understanding the complexity associated with different contexts is necessary for successful treatment.

Consolidation

Through consolidation, the different UCS associations learned during fear acquisition (Fig. 1-B) and extinction (Fig. 1-C) are converted into long-term memories. Rapid eye movement (REM) sleep is involved in emotional memory consolidation [36] and activates the fear-related neural circuitry, including the amygdala and hippocampus [37]. Sleep may enhance and refine negative memories by strengthening connectivity between fear circuit structures such as the amygdala, hippocampus, and vmPFC during memory retrieval [38]. As such, REM sleep plays an important role in the consolidation of both fear learning and extinction learning.

Acute sleep deprivation following fear acquisition affects the later expression of learned fear. In animal studies, depriving REM sleep following fear learning impairs consolidation of fear memories, as indexed by lower levels of behavioral freezing to a previously feared tone [39]. Additionally, longer amounts of acute sleep deprivation produce larger associated impairments in fear memory consolidation in rats [40]. Although replication is needed, more recent work with a translational focus has shown that acute sleep deprivation impairs fear memory consolidation in humans. For example, less REM sleep following fear acquisition resulted in lower shock expectancy ratings, smaller skin conductance responses, and less basolateral amygdala activation to the conditioned stimulus [41•]. This result suggests weaker consolidation of learned fear, consistent with animal studies [42].

Although acute sleep deprivation following fear learning may result in reduced fear expression, chronic sleep deprivation in humans may lead to increased threat sensitivity. Chronically disrupting sleep results in greater amygdala activation to threatening stimuli, possibly resulting from decreased ventral ACC inhibition of the amygdala [43]. Therefore, acute sleep deprivation may lower short-term fear responses, at the expense of higher long-term fear responses, producing a maladaptive cycle that leads to the development and maintenance of fear-related disorders such as post-traumatic stress disorder (PTSD) [44].

Similar to its role in fear learning, sleep also plays a key role in consolidating extinction learning. The consolidation of extinction learning can be enhanced by increasing the amount of sleep following extinction learning. For example, participants who slept for longer intervals following extinction learning, compared to a sleep deprivation group, demonstrated higher levels of extinction retention [45]. Consolidation of extinction learning can also be enhanced by reducing the time interval between extinction learning and sleep. For instance, if extinction learning is closely followed by sleep, fear reductions generalize from an extinguished to a non-extinguished conditioned stimulus [46].

Clinical applications of this work may include increasing sleep following extinction-based treatments (e.g., exposure therapy). Increasing sleep following exposure therapy produces larger reductions in anxiety symptoms. One study demonstrated that afternoon naps as short as 1.5 hours immediately following successful exposure therapy for spider phobia resulted in significantly lower reported levels of distress and catastrophic cognitions at one-week follow-up [47]. Additionally, more restful sleep following CBT sessions for social anxiety disorder was associated with larger symptom reductions at subsequent sessions [48]. Mirroring findings in basic extinction learning paradigms [46], reducing the time interval between exposure therapy and sleep also enhances treatment outcomes. For example, sleep more closely following a single session of exposure therapy produced greater reductions in physiological reactivity to a feared stimulus, which also generalized to new stimuli [49•]. In short, sleep may reduce both threat-related cognitions and physiological reactivity to feared stimuli, as well as increase the rate, retention and generalization of these fear reductions [49•].

Despite these promising findings, however, the mechanisms that produce enhanced consolidation of extinction learning require elucidation. Several explanations for the effects of sleep on extinction learning have been proposed using sleep deprivation studies. First, neural data suggest that sleep deprivation may impair the new CS+ association (i.e., the CS+ no longer predicts the UCS). Compared to a control group, participants who were sleep deprived following extinction learning demonstrated more left middle temporal gyrus activation in response to a CS+ during an extinction retention task [45], possibly reflecting greater UCS expectancy violation [50]. Alternatively, sleep deprivation may impair learning that the context does not predict the UCS. In other words, rather than impairing the consolidation of amygdala-dependent cued memory, sleep deprivation may impair the consolidation of hippocampus-dependent contextual memory [51, 52]

In summary, modifying sleep patterns may serve as a tool to reduce fear. However, the mechanisms for the implementation of sleep modifications should be considered. Reducing sleep following fear learning impairs the consolidation of fear memories. However, utilizing this approach in anxiety disorders may have long-term contraindications, as a more chronic course of sleep deprivation increases threat sensitivity and may impair extinction learning. In contrast, improving consolidation by either increasing the amount or proximity of sleep following extinction learning has demonstrated more clinical utility. This approach has already demonstrated direct clinical relevance, given that sleep has been shown to augment exposure therapy in a number of ways. As previously noted, however, further research that examines a longer time course of these processes is required to understand the stability of these processes over time and to guide the development of interventions targeting these processes.

In addition to manipulating sleep, fear and extinction memory consolidation may be targeted in other ways. For example, transcranial direct stimulation of the dorsolateral PFC (dlPFC) following fear acquisition reduces subsequent fear responses measured during extinction learning, which suggests that neural stimulation of this region disrupts fear memory consolidation [53]. This stimulation may also alter connectivity between structures involved in the fear circuit. For example, resting-state connectivity between the amygdala and dlPFC increases following conditioning and is correlated with explicit memory of the CS−UCS association learned during fear acquisition [54•].

Pharmacological interventions may also be employed to target the consolidation of extinction memories. Although some pharmacological interventions are administered prior to extinction learning, group differences are often delayed, suggesting that the consolidation of the extinction memory may underlie the pharmacological changes, rather than extinction per se. For example, intranasal oxytocin administered to healthy adults prior to extinction learning improved extinction retention [55]. Similarly, fear reduction after reinstatement was enhanced in a sample of healthy adults who were given either DCS or valproic acid prior to extinction [56]. Relative to placebo controls, cannabinoids, namely synthetic Delta-9-tetrahydrocannabinol (THC), given prior to extinction learning enhanced extinction retention and elicited greater ventromedial PFC and hippocampus activation in healthy adults [57•, 58]. Additional evidence suggests that cannabinoids, e.g., synthetic cannabidiol, given after extinction enhances consolidation of extinction learning relative to placebo [59].

Other studies conducted in clinical contexts may provide additional support that enhancing consolidation following extinction is clinically viable. Relative to placebo controls, administering cortisol before psychotherapy sessions improves symptom reduction in individuals with a specific phobia [60, 61], though one study found symptom reduction only at long-term follow-up [61]. Additionally, some research suggests that DCS enhances extinction learning, although more recent evidence indicates that symptom reduction with DCS augmentation was dependent on low fear levels following exposure sessions (i.e., successful exposure therapy) [62, 63]. In contrast, symptoms worsen when DCS is administered prior to treatment sessions that do not lower fear levels (i.e., unsuccessful exposure therapy) [62]. Similarly, although yombinine, a noradrenergic antagonist, enhances self-reported reductions of social anxiety following exposure therapy, the effects were moderated by session success [64]. Therefore, DCS and yombinine may target consolidation mechanisms indiscriminately. Future research should examine the viability of utilizing neural-stimulation and pharmacological interventions to enhance consolidation of extinction memory specifically in the context of extinction-based learning treatments such as exposure therapy.

Reconsolidation

When a consolidated memory is recalled it becomes temporarily unstable, rendering it vulnerable to disruption. Updates to the memory that occur during this period of instability are referred to as reconsolidation [65]. While reconsolidation is not a new field of research, it has recently regained interest [see review, see 66], likely due to its potential for clinical utility in the treatment of anxiety disorders.

Reconsolidation studies follow a general format. First, participants are conditioned to fear two or more stimuli. The next day, one of the conditioned stimuli is presented without an unconditioned stimulus pairing (e.g., electrical shock, loud noise) to reactivate the fear memory, rendering it unstable. Once the fear memory is labile, pharmacological or behavioral manipulations can disrupt this memory during the reconsolidation window (e.g., within 6+ hours post activation). In behavioral studies that condition two CS+, the unreminded CS+ serves as a within-subject control measure. A successful reconsolidation update effect would show a fear reduction in skin-conductance, startle response, or subjective ratings during subsequent retrieval [66] (see Fig. 1-D).

Human reconsolidation studies often administer propranolol, a beta-adrenergic receptor blocker that mimics the effects of protein synthesis inhibitors used in animal studies [67, 68]. Propranolol paradigms have had relative success in reducing fear following reconsolidation in non-clinical samples [69, 70], though some findings are limited to self-reported fear reductions [71]. Instructed paradigms may be a more clinically relevant approach to reduce fear, because the CS−UCS contingencies are based on explicit cognitions, rather than implicit classical conditioning. Using an instructed fear paradigm, participants receiving propranolol prior to extinction learning experienced successful reduction in both the behavioral expression of fear (as measured by startle eye blink) and subjective feelings of anxiety, while the placebo group exhibited return of the fear memory [72•]. This study suggests that a reconsolidation paradigm can alter cognitively based fear memories. Some question the success of propranolol to influence older memories [73]; however, propranolol reduced physiological reactivity following reactivation of past traumatic events relative to placebo control [74]. Additionally, compared to controls, significant symptom improvement was detected after individuals with PTSD activated long-term traumatic memories using script-driven imagery and received deep transcranial magnetic stimulation of the mPFC [75].

Like pharmacological manipulations, behavioral manipulations presented within the reconsolidation window can also interfere with reactivated memories. Studies employing these paradigms are largely variants of Schiller et al., 2010 [76], who showed that conducting extinction training within the reconsolidation window effectively reduced skin conductance responses to conditioned stimuli during retention. Some studies have replicated these findings using a different UCS [e.g., a loud, aversive noise; 77•] and different reinforcement schedules [78]; however, others have failed to see reductions in fear expression when conducting extinction during the reconsolidation window [79, 80]. The reason these studies fail to observe fear reduction is unclear. The use of fear-relevant stimuli [e.g., fearful faces; 80] may explain some of the discrepancy, though previous studies using fear-relevant stimuli in a propranolol paradigm have successfully reduced fear following extinction during the reconsolidation window [69, 70].

Two recent studies have investigated neural correlates of fear reduction following extinction training conducted during the reconsolidation window [78, 81•]. Individuals that underwent extinction training during the reconsolidation window recruited the vmPFC and amygdala-vmPFC coupling less in response to the reminded CS+ during re-extinction than standard extinction procedures [78, 81•]. This finding is clinically relevant, as extinction training during reconsolidation may recruit regions related to fear (i.e., amygdala) less.

Taken together, reconsolidation research is moving in a promising direction towards clinical applications. For instance, behavioral paradigms may target fear memory processes more directly than pharmacological approaches. Yet, researchers need to improve ecological validity by using fear-relevant stimuli, imagined stimuli, or different UCSs. Upcoming study designs should also continue to test the durability of using reconsolidation update mechanisms in clinical applications. For example, some researchers have demonstrated continued effects in healthy samples at one-month [70] and one-year follow-up [76], but further research is needed in clinical samples, including pediatric age ranges, and individuals at-risk for clinical anxiety.

Development and Puberty

The cognitive processes such as those described above may play a role in the etiology and maintenance of anxiety. Adolescence may be a key developmental stage when these perturbations emerge. Adolescence is a transitional period that confers risk for development of multiple anxiety disorders including social anxiety disorder and GAD [82]. Targeting interventions during this period when these processes are most malleable may prove beneficial.

Adolescence is a period of both physical and psychological changes. Across development, the brain shows evidence of increased synaptic pruning and myelination [83]. In adolescence, an imbalance between cortical and subcortical regions activation may lead to heightened emotional responses [84]. As such, several reviews have highlighted the need to study fear conditioning and extinction across development [85, 86, 87•, 88].

As in adults, differential fear conditioning has been demonstrated in children and adolescents [89•]. Some studies indicate that older children show more discrimination compared to younger children. In addition, fear generalization patterns appear more analogous to adults than younger children [90]. Consistent with animal work investigating the effects of development [91], healthy adolescents may exhibit deficits in extinction. For example, healthy adolescents reported more fear to the CS+ relative to the CS- after extinction [92, 30], suggesting extinction resistance. In addition, adolescents had deficient extinction learning compared to adults and younger children [87]. Although youth have deficits in extinction, to our knowledge, only one study on youth with obsessive-compulsive disorder has attempted to enhance extinction learning using DCS [93].

Identifying youth at risk for chronic anxiety in adulthood during these vulnerable periods may alter the course of anxiety disorders [94]. Consistent with prior work in at-risk youth [95], 7–14 year old children with a maternal history of an anxiety disorder exhibited impaired extinction compared to low-risk youth, as evidenced by greater physiological reactivity to CS+ than to CS-. In addition, SCR to both CSs increased with continued extinction training, rather than decreasing [96•]. Longitudinal studies that examine extended time courses are needed to identify developmental trajectories in both at-risk and clinical populations.

Recent neuroimaging studies have detected developmental differences during conditioning [97] and extinction retention [89•]. During fear acquisition, healthy adolescents, compared to healthy adults, showed less self-reported discrimination and greater amygdala activation in response to the CS+ relative to the CS- [97]. Consistent with previous work in anxious adults [98], both adults and adolescents with anxiety disorders exhibited subgenual ACC hypoactivation during extinction retention relative to healthy groups, highlighting regions that are anxiety-related, independent of development. Anxious youth showed different perturbations in vmPFC activation than anxious adults [89•], highlighting regions that are anxiety-related and developmentally sensitive. Together, these findings suggest that adolescence may be a prime target for interventions that disrupt fear memories and/or enhance extinction-related processes.

Puberty may also impact fear and safety learning. For instance, physiological reactivity is increased during puberty [e.g., 99]; however, few studies have investigated the effects pubertal status on fear and safety learning in adolescence. Puberty is also accompanied by hormonal changes and the onset of menstrual cycle. Studies in adult women suggest that estrogen impacts fear extinction, with endogenous and exogenous estradiol modulating vmPFC activation and enhancing extinction retention [100, 101]. Future studies in development should investigate the impact of puberty and hormonal changes during and following fear conditioning.

Conclusions

Here, we have reviewed the recent studies over the past year in fear conditioning, concentrating on several processes following fear learning. After a fear is learned, individuals with anxiety disorders may overgeneralize and have difficulties with discriminating threat and safety. In addition, sleep disturbances may impact the consolidation of fear and safety learning and may lead to maladaptive cycles in anxiety disorders. Studying these different time periods following fear conditioning and extinction may facilitate or enhance treatment development. For example, reconsolidation update mechanisms appear promising, as targeting these mechanisms reduces fear; however, the work conducted in healthy subjects needs to be extended to clinical populations and long-term outcomes needs to be delineated. Additionally, studies outlined herein review opportunities to alter fear expression in adults. Understanding how fear generalization, consolidation, extinction and extinction retention operate independently and interactively in anxiety disorders may provide increased clinical utility. Although promising research has suggested that these processes can be targeted in adults, these processes may be particularly malleable during vulnerable developmental periods, e.g., adolescence. Intervening during these vulnerable periods could be a more fruitful approach, considering the potential to alter the developmental trajectory of anxiety disorders.