The Study of Fear Extinction: Implications for Anxiety Disorders

In the laboratory, fear is acquired when a neutral conditioned stimulus (e.g., a light or tone) is paired with an aversive unconditioned stimulus (e.g., a mild shock). After several such pairings, the subject learns that the conditioned stimulus predicts the unconditioned stimulus, and subsequent presentation of the conditioned stimulus elicits a variety of fear responses, including freezing in the rodent and skin conductance responses in the human. Once acquired, fear to the conditioned stimulus can be extinguished by repeatedly exposing the subject to the conditioned stimulus in the absence of any aversive outcome. During extinction training, the subject's fear responses gradually decline, and when tested the following day, the subject typically exhibits long-term extinction recall.

The fear conditioning and extinction model has been used extensively to examine the neurobiology underlying fear processes in animals and more recently in humans. It is important to note that the fear conditioning model does not necessarily model the etiology of anxiety disorders, as most individuals with anxiety disorders cannot recall a specific conditioning episode that precipitated the disorder (except in posttraumatic stress disorder [PTSD], in which the occurrence of a traumatic event is specified in the diagnostic criteria [10]). The advantage of the fear conditioning model, however, is that it produces the behavioral symptoms commonly exhibited in anxiety disorders (e.g., avoidance) and thus can be used to test ways of reducing these symptoms, such as via extinction. We argue that the fear extinction model is suited to examination of potential biomarkers of anxiety for several reasons. First, a key feature of clinical anxiety disorders is a failure to appropriately inhibit, or extinguish, fear (11). Individuals with anxiety disorders avoid fear-provoking situations and stimuli or endure them by employing a range of “safety behaviors” designed to protect the individuals from harm. Avoidance and the use of safety behaviors prevent the individual from challenging his or her unrealistic beliefs and thus prevent fear extinction. Hence, the fear extinction model provides a direct measure of what is widely accepted to be a central underlying dysfunction in anxiety disorders. As such, the measurement of neural activity during extinction may provide a sensitive measure of the neural circuitry involved in the maintenance of anxiety disorders.

Second, as noted above, exposure therapy is a dominant component of CBT. Even though anxiety disorders are characterized by dysfunctional cognitive processes (e.g., overestimation of the probability and cost of negative outcomes), research has shown that exposure therapy without explicit cognitive intervention is just as effective and invokes just as much cognitive change as comprehensive CBT that combines behavioral and cognitive interventions (12). Exposure therapy was based on the extinction procedure used in animal studies of fear inhibition. Thus, in addition to potentially detecting the neural basis for the underlying dysfunction in anxiety disorders, extinction can also be used as a valid model of the most effective psychological treatment for anxiety disorders.

A third advantage of the fear extinction model is that comparison of animal studies and human neuroimaging studies suggests considerable similarity between the neural structures involved in extinction in the rodent and in the human, as reviewed in more detail below. The cross-species validity of the extinction model permits rodents to be used to address questions that are not possible to address using human subjects, such as trialing the effects of novel drugs on extinction and subsequent relapse, with the assurance that these findings are readily translatable to the human population.

The neurobiology of fear acquisition is well characterized in rodents and humans (13). Briefly, it is widely accepted that the basolateral complex of the amygdala is the main neural structure in which information about conditioned and unconditioned stimuli converge. This finding has been supported by studies in rodents using lesions, pharmacological inactivation, electrophysiology, and drug antagonists, which together have demonstrated that interfering with normal functioning of the basolateral amygdala disrupts the acquisition and expression of fear conditioning (14). There is also evidence from rodent studies that the prelimbic division of the medial prefrontal cortex is involved in regulating the expression of learned fear. Inactivation of the prelimbic cortex reduces the expression of cued and contextual fear (15), while microstimulation increases conditioned freezing and reduces extinction (16), prompting the assertion that the prelimbic cortex regulates fear expression by activating the amygdala. This view is supported by findings that prelimbic neurons potentiate their response to a tone-conditioned stimulus and that extinction failure is associated with a persistence of prelimbic neuronal response after extinction training (17). Functional MRI (fMRI) studies have demonstrated that humans show robust increases in activity in the amygdala and the dorsal anterior cingulate (which appears to be functionally analogous to the prelimbic cortex) during fear acquisition and expression (18, 19). Other research (20) has demonstrated that in healthy humans, the cortical thickness of the dorsal anterior cingulate is positively correlated with skin conductance responses during fear conditioning acquisition and that activation of this structure during acquisition of conditioning increases in response to a stimulus paired with shock relative to a stimulus not paired with shock. It should be noted, however, that a recent study failed to replicate the correlation between the cortical thickness of the dorsal anterior cingulate and fear acquisition levels in healthy subjects (21). In addition, other regions have also been implicated in fear expression in humans, including the insula, the thalamus, and brainstem regions such as the periaqueductal gray.

Fear extinction, on the other hand, involves interactions between the infralimbic region of the medial prefrontal cortex, the basolateral complex of the amygdala, and the hippocampus. It is proposed that when an extinguished cue is presented in the extinction training context, the hippocampus activates the infralimbic cortex, which in turn activates inhibitory interneurons in the basolateral amygdala that inhibit the output neurons in the central amygdala, thus preventing conditioned responding. In contrast, when the extinguished cue is presented in a context other than extinction training, the hippocampus does not activate the infralimbic cortex and central amygdala activity is not inhibited, and thus conditioned responding returns (22).

Subsequent analysis of the neurobiology underlying fear extinction in humans using fMRI has revealed remarkable preservation of this circuitry across species. Earlier fMRI studies demonstrated that the amygdala exhibits increased activation to the conditioned stimulus during early extinction training, and this activation decreases across extinction training (23). Other studies demonstrated that the amygdala and the orbitofrontal cortex (part of the ventromedial prefrontal cortex) exhibit increased activity during extinction training for an olfactory cue (24) and that there are differences in amygdala and hippocampal activity during extinction training in comparison to a nonextinguished comparison group (18). Later studies focusing on the neurobiology of extinction recall consistently demonstrated that extinction recall is associated with increased activity in the ventromedial prefrontal cortex (19, 25, 26), a structure that has been proposed to be the human homologue of the rat infralimbic cortex. Furthermore, it has been shown using structural MRI that extinction recall is positively correlated with the thickness of the ventromedial prefrontal cortex (21, 27, 28).

Several studies have also provided evidence for increased hippocampal activity during extinction recall (18, 26). Furthermore, one study reported increased hippocampal and ventromedial prefrontal cortex activity during recall in the extinction context but not in the original conditioning context (25). These findings support the notion that the hippocampus modulates when and where extinction is expressed, depending on the contextual information. Hence, there is much evidence to support the notion that a distinct neural circuitry involving interactions among the amygdala, the ventromedial prefrontal cortex, and the hippocampus underlies the ability to extinguish fear and that this circuitry has been preserved across evolution.

In addition, a burgeoning literature is investigating the molecular signaling pathways within this neural circuitry that support the formation of extinction memories, such as brain-derived neurotrophic factor (BDNF). In rodents, BDNF mRNA increases in the basolateral amygdala after extinction (29), and BDNF infusions in the infralimbic cortex and the hippocampus enhance extinction (30). Moreover, a BDNF genetic polymorphism alters fear extinction retention in both rodents and humans (31, 32). Other signals that are involved in extinction and appear to have been conserved across evolution include neuropeptide Y (33, 34) and cannabinoids (35, 36). A full review of the molecular signals implicated in extinction is beyond the scope of this article, but several excellent reviews have been published in recent years (37, 38).

It is widely accepted that anxiety disorders are maintained as a result of a failure to appropriately extinguish fear. In this section, we address two questions: 1) Do clinically anxious populations exhibit alterations in the neural circuitry that mediates normal extinction? and 2) Does the extinction model provide a sensitive measure of this dysfunction in clinical populations?

To date, there are very few published studies in this domain to answer this question. One study (61) reported that genetic heritability accounted for 35%–45% of the variance associated with conditioning and extinction rates. Consistent with that finding, another study (62) assessed the potential use of the extinction model to predict vulnerability to anxiety disorders. The study examined the extinction of skin conductance responses and corrugator muscle electromyogram (EMG) responses to an aversively conditioned stimulus (colored circles) in firefighters during cadet training. Participants were reassessed for PTSD symptoms within 24 months after training, by which time all had been exposed to a work-related trauma. Reduced extinction of EMG responses during extinction training at the time of cadet training accounted for 31% of the variance associated with subsequent PTSD symptoms 24 months later. Thus, these initial studies support the premise that impaired extinction may be a precursor to anxiety and that early detection of this impairment may be used to predict vulnerability to anxiety. Nonetheless, in a small sample of monozygotic twins, we reported impaired extinction retention in participants with PTSD, which was not present in their co-twins (54), suggesting the lack of any preexisting presence of impaired extinction retention in participants with PTSD. One possible explanation for this discrepancy is that the sample size may have been too small. Another is that extinction retention deficits may be acquired, whereas the ability to learn to fear and extinguish fear within a session may be associated with genetic predisposition to developing PTSD. Animal studies appear to support distinct mechanisms for extinction learning and extinction recall, and recent imaging studies in twins suggest that while some of the psychopathology of PTSD appears to precede trauma exposure, other neural deficiencies appear to be acquired (see below).

No studies have investigated whether alterations in the functional activity of the neural circuitry involved in extinction can predict subsequent development of PTSD. However, studies that have investigated differences in brain morphology using structural MRI have consistently demonstrated that PTSD is associated with decreased hippocampal volume (7), although it remains controversial whether or not alterations in brain morphology are a precursor to or a consequence of PTSD. At least one study supports the notion that a smaller hippocampal volume is predictive of PTSD development (63). The study demonstrated that monozygotic twins discordant for trauma exposure in whom the trauma-exposed twin developed PTSD had smaller hippocampal volumes than monozygotic twins discordant for trauma exposure in whom the trauma-exposed twin did not develop PTSD. Furthermore, symptom severity in the participants with PTSD was negatively correlated with their own hippocampal volume as well as that of their co-twin. On the other hand, a later study using the same participants (64) found reduced gray matter density in the anterior cingulate of those with PTSD relative to their combat-unexposed co-twins, as well as to combat-exposed twins without PTSD and their co-twins, suggesting that neural abnormalities in this region may be a consequence of PTSD. A more recent study (65) examined resting state activity using PET in dizygotic twin pairs in which one co-twin had PTSD compared with twin pairs in which one co-twin had been exposed to trauma but did not develop PTSD. The study reported a higher resting state in the dorsal anterior cingulate and midcingulate cortex in participants with PTSD and their co-twins relative to trauma-exposed participants and their co-twins, suggesting that alterations in the neural circuitry underlying conditioning may be a risk factor for subsequent development of PTSD. Clearly, more work is needed to elucidate the extent to which preexisting dysfunctions in extinction ability and the neural circuitry underlying extinction contribute to subsequent development of PTSD and other anxiety disorders.

As noted previously, even the most successful treatments for anxiety disorders are associated to some extent with relapse, and some patients fail to respond at all. At this point, the factors that predict responsiveness to treatment remain largely elusive. One possibility is that the extinction model may be employed to predict the likelihood of responding to CBT-based treatments that primarily use extinction procedures. To our knowledge, no studies have investigated extinction ability prior to treatment and correlated the magnitude of extinction retention with the success of treatment response. However, two studies have examined neural activity in the circuitry mediating extinction prior to treatment. One examined resting metabolic activity in participants with obsessive-compulsive disorder (OCD) before they received behavioral therapy or fluoxetine treatment (66). In the behavioral therapy group, positive treatment response was correlated with higher pretreatment metabolism in the left frontal orbital cortex, while the reverse was the case for positive response to the medication treatment (that is, treatment response was negatively correlated with pretreatment left frontal orbital cortex metabolism). The second study examined the pattern of neural activation in response to fearful and neutral faces in individuals with PTSD prior to CBT (67). They found that poor treatment response, as measured 6 months after treatment, was associated with greater activation in the amygdala and the ventral anterior cingulate region. Although these findings are preliminary, they suggest that pretreatment measurement of neural activity in the extinction circuitry could provide important information regarding the intensity, duration, and type of therapy required to prevent or reduce subsequent relapse.

As exposure therapy is based on extinction, laboratory investigations of the neurobiology of extinction in the rodent and the human have proved fruitful in providing ways of enhancing CBT for anxiety disorders. Without a doubt, the most successful of these investigations has been that of the effect of d-cycloserine on the extinction of conditioned fear, which was initially demonstrated to enhance extinction of conditioned fear in rats and to reduce stress-precipitated relapse (68, 69). Since then, d-cycloserine has been shown to enhance exposure therapy in humans with a range of anxiety disorders (70). Investigations of d-cycloserine and extinction have influenced a new wave of thinking in pharmacotherapy for anxiety disorders. Rather than developing drugs that merely mask the symptoms of anxiety (and often interfere with the effectiveness of psychological therapies), researchers are investigating the potential of drugs that can be used to augment the underlying therapeutic mechanisms of CBT. Numerous novel pharmacological enhancers of extinction are currently being investigated in preclinical and clinical research (71).

Other studies have focused on directly stimulating the neural structures that have been proposed to be dysfunctional in anxiety. In repetitive transcranial magnetic stimulation (rTMS), cortical neurons are stimulated noninvasively and without the side effects often associated with ECT. In one study (72), rTMS of the lateral prefrontal cortex was shown to reduce compulsive urges in people with OCD as measured 8 hours after stimulation, an effect that was not found when the midoccipital region was stimulated. Another study (73) reported that 10 sessions of high-frequency stimulation to the dorsolateral prefrontal cortex significantly reduced PTSD symptoms up to 3 months after treatment (73). These findings suggest that mere stimulation of neural regions underlying extinction can be therapeutic.

Results from rTMS studies raise the possibility that extinction procedures may be combined with brain stimulation to enhance the effectiveness of such treatments. Indeed, this has already been investigated at a preclinical level in rodents. One study (74) found that combining conditioned stimulus presentations with infralimbic stimulation reduced conditioned freezing during extinction training in rats. Furthermore, this reduction in freezing remained evident during extinction recall, an effect that was absent when the stimulation was not paired with the conditioned stimulus presentation. As rTMS is already being examined in clinical trials, the potential exists for translating this finding from the rodent to clinical populations. Notably, the recent advances made in enhancing current treatments for anxiety reviewed here all stemmed from preclinical investigations of extinction in nonhuman animals, thus illustrating the validity of extinction as a model of treatment.

Several limitations of the fear extinction model and its proposed underlying neural circuitry have been noted. For example, it was recently reported (75) that PTSD was less prevalent in Vietnam War veterans who had experienced damage to the ventromedial prefrontal cortex or the amygdala, a finding that appears contradictory to the model suggesting that anxiety results from a failure of prefrontal inhibition over amygdala activity. However, the average lesion area in that study was not limited to the ventromedial prefrontal cortex; large portions of the surrounding anterior, lateral, and dorsal prefrontal cortices were included. Future studies with focal lesions that are more confined to the ventromedial region should specifically examine the relationship between ventromedial lesions and PTSD prevalence.

Another limitation to the extinction model is that neuroimaging studies of anxiety have not always produced results that are consistent with the current neural model of fear extinction. For example, some studies have reported no difference in amygdala activation during the presentation of trauma-related reminders in PTSD patients (76, 77). In addition, some studies have reported hyperactivation in the ventromedial prefrontal cortex during symptom provocation in PTSD patients compared with healthy (78) and trauma-exposed (79) comparison subjects, or no difference in ventromedial prefrontal activation between groups (80). These inconsistencies are concerning, and the underlying reason for them is unclear, but they may stem from the fact that symptom provocation and fear extinction tasks measure slightly different functions (fear processing as opposed to fear inhibition).

Finally, it is clear that the extinction model does not capture all aspects of clinical anxiety, particularly cognitive components such as anticipatory anxiety (9). Similarly, it does not capture all aspects of any given anxiety disorder. For example, the underlying pathogenesis of OCD, which is characterized by intrusive thoughts and rituals, is not well modeled by extinction and may be regulated by an entirely different neural circuit from that of extinction (9). Furthermore, while extinction is a dominant component of CBT for anxiety disorders, it is not the only mechanism of therapeutic change. For example, CBT also involves exposure to the feared outcome itself, an aspect that is perhaps better modeled by habituation paradigms (81). Nevertheless, despite these limitations, the data reviewed here indicate that the fear extinction model can be used to understand the psychopathology of anxiety disorders and to determine similarities and differences among the various diagnostic categories.

In this review, we have presented four arguments. First, the extinction model is advantageous because it models the commonly accepted underlying dysfunction in anxiety disorders (CBT); because it models the most widely used treatment of anxiety disorders; and because there is much evidence for the cross-species validity of the model. Second, a basic neural circuitry that regulates fear extinction has been identified in rodent and human research, and there is moderate evidence to suggest that this neural network is dysfunctional in anxiety disorders. Third, there is preliminary evidence to suggest that the extinction model may potentially be used to detect vulnerability to anxiety disorders and to predict the likelihood of treatment response. Fourth, there is considerable evidence to suggest that the extinction model can be used as a means to investigate ways of enhancing existing treatments for anxiety. We argue that these factors make the extinction model—and its underlying neural circuitry—an excellent candidate for a biomarker of anxiety disorders, as well as a useful tool for understanding the psychopathology of anxiety. However, there is further work to be done to help validate the extinction model as a biomarker of anxiety disorders across diagnostic categories and to ensure that the model continues to lead to developments in treatment. Below, we summarize what we believe to be the most important future directions.

The advent and continuous development of neuroimaging tools have allowed us to form a solid base of knowledge about the neural dysfunctions across anxiety disorders. The recent explosion of interest in understanding the neural substrates of fear extinction in rodents reflects the importance of the neurobiology of fear inhibition and its relevance to anxiety disorders. Research using healthy human subjects has confirmed the cross-species validity of this model and has demonstrated that the behavioral and neural mechanisms underlying fear extinction have been strongly preserved across evolution. Our current knowledge base from these two fields should allow us to move forward and merge the two together using a multimodal across-species approach. While this experimental model is not perfect, as it does not capture all aspects of any given anxiety disorder, we argue that it is a good model for understanding the neural circuits underlying learning not to fear per se, which can then be directly translated to understanding anxiety disorders and improving treatment outcomes for patients.

Received April 7, 2011; revision received June 15, 2011; accepted June 20, 2011.