View
227
Download
0
Category
Preview:
Citation preview
8/11/2019 Review Relapse, 2013
1/37
Fear Extinction and Relapse:State of the Art
Bram Vervliet,1 Michelle G. Craske,2
and Dirk Hermans1
1Department of Psychology, University of Leuven, Leuven 3000, Belgium;email: bram.vervliet@ppw.kuleuven.be, dirk.hermans@ppw.kuleuven.be
2Department of Psychology, University of California, Los Angeles, California 90095-15email: mgcraske@ucla.edu
Annu. Rev. Clin. Psychol. 2013. 9:21548
TheAnnual Review of Clinical Psychologyis online at
http://clinpsy.annualreviews.org
This articles doi:10.1146/annurev-clinpsy-050212-185542
Copyright c2013 by Annual Reviews.All rights reserved
Keywords
Pavlovian conditioning, learning theory, memory retrieval, anxiety
treatment, translational research, relapse prevention
Abstract
Exposure-based treatments for clinical anxiety generally are very eff
but relapse is not uncommon. Likewise, laboratory studies have show
conditioned fears are easy to extinguish, but they recover easily. This a
is striking, and numerous fear extinction studies have been publishe
highlight the processes responsible for the extinction and return of ac
fears. This review examines and integrates the most important result
animal and human work. Overall, the results suggest that fear extinc
relatively easy to learn but difficult to remember. It follows that
ments will benefit from an enhanced focus on the long-term retrieval
extinction. We review the available studies on the prevention of ret
fear and the prospects of weakening fear memories forever. We sho
the behavioral principles outlined in learning theory provide a cont
inspiration for preclinical (neurobiological) and clinical research on t
tinction and return of fear.
215
Click here for quick links to
Annual Reviews content online,
including:
Other articles in this volume
Top cited articles
Top downloaded articles
Our comprehensive search
FurtherANNUAL
REVIEWS
8/11/2019 Review Relapse, 2013
2/37
Contents
I N T R O D U C T I O N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1 6
RETURN OF FEAR IN CLINICAL PRACTICE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
RETURN OF FEAR IN CLINICAL ANALOGUE STUDIES:
ROLE OF CONTEXT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Extinguished Fears Can Return with a Change of External Context . . . . . . . . . . . . . . . 220
Extinguished Fears Can Return with a Change of Internal State. . . . . . . . . . . . . . . . . . . 220RETURN OF FEAR IN THE LABORATORY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
R e n e w a l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 1
Spontaneous Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
R e i n s t a t e m e n t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 2
THE LEARNING THEORY PERSPECTIVE ON FEAR CONDITIONING
AND EXTINCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Fear Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Deviant Fear Conditioning in Anxious Individuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Mechanisms of Extinction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
THE DEVELOPMENT OF THE RETRIEVAL MODEL
OF EXTINCTION IN RATS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224What We Learned from Reinstatement Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
What We Learned from Renewal Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
The Retrieval Model of Extinction and Return of Fear . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
The Neurological Model of Extinction and Return of Fear . . . . . . . . . . . . . . . . . . . . . . . 227
Multiple Pathways to Return of Fear? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
PRECLINICAL HUMAN STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Reinstatement in Human Fear Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Renewal in Human Fear Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Spontaneous Recovery in Human Fear Conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
ROADS TO RELAPSE PREVENTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Preventing Relapse by Strengthening the Extinction Memory. . . . . . . . . . . . . . . . . . . . . 235Preventing Relapse By Weakening the Fear Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
FUTURE RESEARCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
What Constitutes a Context? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Individual Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Extinction Versus Reconsolidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Time Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Return Versus Relapse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Retrieval Versus New Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
INTRODUCTION
Behavior therapy originated with Wolpes (1958) systematic application of basic learning pri
ciples to the treatment of fear (systematic desensitization; see Eelen & Vervliet 2006). Over t
years, behavior therapy has maintained a preferential relationship with fundamental learning th
ory. However, the cognitive revolution in psychology in the 1970s and 1980s sparked stro
216 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
3/37
Extinction of de
novo conditioned
fears in healthy
humans
Extinction of de
novo conditionedfears in animals
Extinction of de
novo conditioned
fears in anxious
individuals
Extinction of pre-
existing fears inanxious individuals
Clinical trials of
exposure-based
treatments of
anxiety
Figure 1
The different steps in the translational research program on fear extinction. Fear extinction in animals represents the most funda
preclinical research line (extreme left); clinical trials of exposure-based treatments in anxious patients represent the most appliedresearch line (extreme right). Preclinical research in healthy humans verifies animal findings in the human species. Preclinical rein anxious individuals verifies the relevance of the studied processes for clinical anxiety. Clinical analogue studies that target extinof preexisting fears in anxious individuals can test the application of fundamental principles in real-world fears with selected treacomponents. Clinical trials evaluate the added efficacy of adjustments to treatment packages. The arrows indicate a continuousinfluence between the different steps, in both directions.
Clinical researapplied researchevaluates interv(and moderatinfactors) in the c
settingReturn of fearincrease of the findex from the the last treatmesession to a folltest
criticisms against the traditional learning theory perspective on fear and anxiety disorders. Ac-
cordingly, fundamental learning research contributed less to (cognitive-) behavioral therapy that
emerged in the 1980s. The backlash against traditional learning theory failed to take account
of important theoretical developments within the field of learning theory (Mineka & Zinbarg
2006, Rescorla 1986). The recent rise of (affective) neuroscience research has reinstalled interest
in learning theory as a framework for preclinical animal models of psychopathology. The fear
domain has once again pioneered this new evolution. The domain is an excellent example of
successful translational research, in which preclinical and clinical research programs mutually re-
inforce each other (see Figure 1 and sidebar Translating Research Findings From the Laboratory
to the Clinical Setting and Back). The focus on fear also bolsters connections among learning
theory, clinical psychology, psychiatry, affective neuroscience, pharmacology, and genetics. As we
argue in this review, this cross-fertilization is in large part due to the elegance and simplicity of
the Pavlovian fear-conditioning procedure that has been adopted widely for the study of fear and
anxiety.
Prevention of relapse is a common theme in the renewed research program on fear. Effective
pharmacological and psychotherapeutic treatments for anxiety have existed for quite some time
(see, e.g., meta-analyses by Eddy et al. 2004, Hofmann & Smits 2008, Mitte 2005). The most
effective psychotherapies share exposure techniques as a key ingredient. This involves repeatedly
exposing the anxious client to the situations that elicit fear. Typically, this leads to reductions of
fear over repeated exposures. Exposure-based therapy is very successful in reducing fear levels
in anxiety patients, as is pharmacotherapy. Despite these successes, however, these therapies also
share a common weakness: poor retention of fear reduction in the long-term. Treatment efficacy
studies have shown that successful fear reductions are sometimes short-lived and followed by a
return of fear symptoms that can culminate in a full-blown relapse (see, e.g., Eddy et al. 2004).
www.annualreviews.org Fear Extinction and Relapse 217
8/11/2019 Review Relapse, 2013
4/37
TRANSLATING RESEARCH FINDINGS FROM THE LABORATORY TO THECLINICAL SETTING AND BACK
Translational research encompasses all research that combines laboratory with more applied research methodolo-
gies. Clinical research feeds into basic laboratory research by formulating current questions and problems in clinical
practice; basic laboratory research feeds into clinical research by revealing underlying mechanisms and suggesting
novel intervention techniques. The success of translational research depends in large part on the validity of the
experimental model used to mimic the disorder in the laboratory. In the fear domain, the translation can be verified
at different steps. The Pavlovian conditioning model can be applied in nonhuman animals, healthy individuals, and
anxious individuals. Extinction (exposure) can be investigated with de novo conditioned fears or with preexisting
fears in anxious individuals. Each step has added values and limitations; the combination of the different steps
constitutes the strength of the translational research program.
Extinction: thegradual decay ofconditioned responsesdue to repeatedpresentations of theCS without theassociated US
Pavlovianconditioning: a
learning phenomenonthat occurs whenpairings of two stimulicause a change in thebehavioral reaction toeither stimulus
Conditional stimulus(CS): a stimulus thatelicits a reactionconditional upon itsprevious pairings withanother stimulus
Conditionalresponse (CR): thereaction elicited by aconditional stimulusthat has been paired
with an unconditionalstimulus
The current challenge is not to achieve fear reduction, but rather to maintain it over time. Relap
prevention has become the major focus of fear extinction research, as it is for addiction an
depression research.
Return of fear is a widespread phenomenon. It occurs with clinical anxiety that has been treate
but it also occurs with experimentally induced fears in the laboratory. Fears can be easily creat
through a Pavlovian conditioning procedure, in which an innocuous stimulus (e.g., a green ligh
is systematically followed by an aversive stimulus (e.g., an electrical shock). This typically results
conditional fear reactions to the light (e.g., freezing in rats). In conditioning terms, the innocuo
stimulus is called the conditional stimulus (CS), the aversive stimulus is the unconditional stimul
(US), and the fear reaction the conditional response (CR).
As in exposure-based treatments, repeated presentations of the light CS in the absence
the aversive shock US typically leads to a gradual reduction of the fear CR. This is called t
extinction effect. One of the most basic findings on extinction is that extinguished responses c
reoccur (Pavlov 1927). In the case of fear conditioning, this provides a preclinical laboratory mod
for the return of fear. Clinical and preclinical research complement each other. Clinical studies c
record conditions under which return of fear occurs, examine individual differences in the return
fear, and test treatment adjustments to prevent the return of fear. The advantage of this research
its direct applicability (no translational gap). The disadvantages are the constraints upon studyi
the underlying mechanisms (due to time limitations), ethical considerations (e.g., brain lesionin
lack of control over etiology and potentially confounding factors, and so on. Preclinical laborato
research can recreate the etiological factors believed to underlie the development of anxiety an
examine the phenomenon of interest in a highly controlled way. The main disadvantage is t
gap with clinical practice and the uncertainty about whether the revealed mechanisms are actua
relevant to clinical phenomenology. A combination of clinical and preclinical research is need
to advance our understanding of disorders and to improve existing treatments. Return of fe
provides a textbook example of how this combination can lead to scientific and practical progre
Of particular interest in the return of fear domain is the significant contribution from Pavlovi
fear-conditioning research in healthy humans. Fears can be created in healthy individuals with t
same basic procedure as in animals: repeated pairings of an innocuous with an aversive stimulu
Most measures of fear differ from rat studies (subjective ratings, skin conductance), but som
are highly similar (the fear-potentiated startle reflex). The preclinical human model can transla
animal findings to the human species, and it permits the study of processes that are clinica
very relevant but difficult to research in rat models (e.g., cognitive processes). The disadvanta
218 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
5/37
Unconditionalstimulus (US):a stimulus thatelicits a reactionunconditionally(independent of
pairings with otstimuli)
Fear conditionPavlovian condiprocedure with aversive stimuluUS and fear meas dependent va(CR)
Preclinical resfundamentallaboratory reseathat investigatesprocesses with crelevance, usingexperimental mthe disorder
is the ethical constraint on the types of experimental manipulations that can be used (e.g., brain
lesioning, traumatic experiences). Human fear conditioning bridges the gap between preclinical
animal research and clinical patient research.
The main theme of this review is that extinction and return of fear are intimately connected.
Return of fear is at the core of extinction; any theory of extinction will have to explain return
of fear and vice versa. We give a selective overview of the preclinical rat studies, preclinical
human studies, and clinical studies that have shaped our understanding of fear extinction and
the conditions under which fears return. The main guide throughout this review is learning
theory. This approach provides a continuous source of inspiration for new developments in the
field of fear extinction and return. Learning theories of extinction have inspired the search for
conditions of return of fear, and observations of return of fear have prompted new theoretical
developments. These developments have had a major influence on affective neuroscience research
into fear extinction and on clinical research into exposure treatment. The integration of learning
theory, affective neuroscience, and clinical psychology has also sparked the development of an
innovative treatment paradigm that relies on behavioral-pharmacological interactions to produce
sustainable fear reduction. We discuss this paradigm shift when reviewing the empirical status of
prevention-of-return strategies.
A major challenge in translational research is to connect fundamental research findings to
clinically relevant phenomena. Sometimes theconnection canbe found on theprocedural level, and
at other times the connection is primarily on a conceptual level. We aim to show that most theory-
inspired research on return of fear can be connected to clinical practice, even if the connection
seems remote at first sight. For that purpose, we refer to a relatively simple example of phobic fear
(although the same principles apply to panic disorder, social anxiety disorder, and posttraumatic
stress disorder):
Bob experienced a panic attack inside an elevator. As a consequence, he became fearful of elevators
and avoided elevators whenever possible. Bobs therapist proposed a cognitive-behavioral treatment
strategy based on repeated exposures to elevators. After a few sessions, Bob was able to use elevators
with minimal discomfort and no longer avoided them. Both the therapist and Bob were very satisfied
with these results and concluded that further treatment was not necessary. A few months later, Bob
experienced some moments of fear inside elevators again, and his tendency to avoid them was renewed.
Subsequently, Bob experienced a full-blown relapse of his anxiety disorder.
This very return of fear is the primary target of this review.
RETURN OF FEAR IN CLINICAL PRACTICE
The return of fear was first documented as the reemergence of fear following systematic desen-
sitization (Rachman 1966). Typically, the level of fear is assessed during behavioral avoidance
tests at pre- and posttreatment and at a follow-up assessment weeks to months later. The client is
asked to gradually approach his/her feared object as closely as possible and to report the level of
fear at each step. Concurrently, psychophysiological indices of anxiety can be measured (e.g., skin
conductance, heart rate). The routine observation is fear reduction from pre- to posttreatment as-
sessment. Return of fear is defined by an increase of fear from posttreatment to follow-up. Return
of fear has been specifically studied following exposure therapy for specific phobias, obsessive-
compulsive disorder, agoraphobia, and performance anxieties. In the existing literature, estimates
of return of fear range from 19% to 62% (see review by Craske & Mystkowski 2006). Other
studies report rates of relapse more broadly by indexing recurrence of panic attacks or overall
www.annualreviews.org Fear Extinction and Relapse 219
8/11/2019 Review Relapse, 2013
6/37
ratings of clinical severity. For example 23% to 27% of patients with panic disorder reported
relapse (Brown & Barlow 1995, Fava et al. 2001) following exposure-based therapies. It is cle
that return of fear after successful treatment is not uncommon and that it poses a serious challen
to the long-term outcome of existing (exposure-based) treatments.
RETURN OF FEAR IN CLINICAL ANALOGUE STUDIES:ROLE OF CONTEXT
Return of fear puts serious limitations on the long-term outcome of (exposure-based) anxie
treatments. An important research target is to reveal the conditions that can trigger return
fear. Clinical observations provide little information because of the lack of experimental contro
Clinical analogue studies that take an experimental approach are more informative. These stu
ies often use a simplified version of cognitive behavioral therapy interventions (focusing on t
exposure ingredient) and experimentally examine specific manipulations between posttreatme
and follow-up. Within this line of research, the role of context in the return of fear is the mo
systematically investigated feature.
Extinguished Fears Can Return with a Change of External Context
Five exposure studies (total of 285 fearful participants) have shown that posttreatment fear
higher when assessments occur in a context different from the treatment context. Rodrigu
et al. (1999) exposed spider-fearful college students to two spiders during a one-session exposu
protocol. Assessments at two-week follow-up revealed a significant return of fear in a differe
context (presentation of the same spiders in a different room with an experimenter who differ
from the therapist), as indicated by number of steps completed on a behavioral avoidance test a
associated heart rate (although subjective fear ratings did not indicate return of fear). Using ve
similar designs and spider-fearful samples, Mineka et al. (1999) and Mystkowski et al. (2002, 200
found significant return of self-reported fear, but not heart rate, during a behavioral avoidance te
In a public speakinganxious sample, Culver et al. (2011) found return of self-reported fear aheart rate at a two-week follow-up test in a context different from treatment. All studies contain
control conditions with fear assessments in the treatment context, which elicited no return of fea
Extinguished Fears Can Return with a Change of Internal State
Contexts can also include the internal state (i.e., arousal) of an individual. A study with 43 spide
fearful individuals showed that incongruent internal states during treatment and follow-up c
produce a return of fear (Mystkowski et al. 2003). Internal states were manipulated by the ingesti
of caffeine or placebo. Groups that had received the same ingestion before treatment and te
showed sustained fear reduction; groups that had received different ingestions showed a significa
return of self-reported fear.
These clinical analogue studies demonstrate the important role of internal and external conte
in the return of fear after exposure treatment. But what exactly is the role of the context her
What does it tell us about the mechanism of exposure and return of fear? How can we neutrali
the detrimental effect of context changes and improve long-term extinction? Such questions a
difficult to answer using clinical studies alone, but preclinical laboratory research on the extincti
and return of fear has been extremely instrumental.
220 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
7/37
a Contextual renewal of extinguished fears
Trials [context A]
Fear conditioning(tone-shock)
Fear
Trials [context B]
Fear extinction(tone-nothing)
Fear
[Context B] [Context A/C]
Fear test(tone-alone)
Fear
b Spontaneous recovery of extinguished fears
Fear conditioning(tone-shock)
Fear
Fear extinction(tone-nothing)
Fear
Immediate Delayed
Fear test(tone-alone)
Fear
c Reinstatement of extinguished fears
Fear conditioning(tone-shock)
Fear
Fear extinction(tone-nothing)
Fear
Shocks
Fear test(tone-alone)
Fear
Trials Trials
Trials Trials
Figure 2
The different procedures used to trigger the return of fear in the laboratory.
RETURN OF FEAR IN THE LABORATORY
It has long been known that extinction is fragile and that extinguished conditional responses
return easily (Pavlov 1927). This is now also documented in the domain of fear conditioning.
Three procedures have been used to trigger the return of fear in the laboratory (seeFigure 2).
One includes explicit context changes, whereas the other two do not.
Renewal
Figure 2arefers to the observation of return of fear when tests occur in a context different from
extinction (in rat research, the context is usually the experiment cage). In ABA renewal, fear
conditioning occurs in context A, extinction in context B, followed by test of the extinguished CS
in context A again. In ABC renewal, the stimulus is tested in a novel context C. In AAB renewal,
www.annualreviews.org Fear Extinction and Relapse 221
8/11/2019 Review Relapse, 2013
8/37
conditioning and extinction both occur in context A, followed by a test of the stimulus in a nov
context B (Bouton 2002).
Spontaneous Recovery
Figure 2brefers to the observation that a time interval after extinction can produce fear recove
to the extinguished CS. The amount of fear recovery in rats is proportional to the length of t
extinction-test time interval, with a complete recovery after 14 days (Quirk 2002).
Reinstatement
Figure 2crefers to the observation that unsignaled presentations of the aversive event can produ
fear recovery to the extinguished CS (Rescorla & Heth 1975).
Bobs fears may have extinguished through exposures to the elevator in the therapists office building,
but they reemerge when he enters an elevator in a different building (renewal). Bobs fears may also
reemerge when he has not been inside an elevator for some time (spontaneous recovery). Or Bobs fear
of elevators may reemerge after he experiences an out-of-the-blue panic attack unrelated to elevators
(reinstatement).
Observations of return of fear show that extinction/exposure does not simply erase the fe
Any theory of extinction should explain (a) how fears can extinguish without erasure and
how posttreatment manipulations can elicit the return of extinguished fears. The learning theo
perspective has proven to be a very successful framework in the behavioral analysis of extincti
and the development of extinction theories.
THE LEARNING THEORY PERSPECTIVE ON FEAR CONDITIONINGAND EXTINCTION
A theory of return of fear implies a theory of extinction, which is itself based on a theory of ho
fears are acquired in the first place.
Fear Conditioning
A light that systematically precedes a shock elicits fear. Standard learning theory states that t
light-shock experiences leave a memory trace that consists of three components: a mental repr
sentation of the light CS, a mental representation of the shock US, and an association betwe
these two representations. Future perceptions of the light will activate its representation and,
virtue of the CS-US association, also the US representation. This activation of the US repr
sentation produces the conditional fear reactions and can be understood as the light signals th
shock or the light retrieves the memory of the shock. The CS-US association is a hypothetic
construct that is held responsible for the conditional fear response.
The fear response is a reaction pattern unique to the CS. It is not a lingering reaction to
past traumatic event (shock US) but rather the emotional component of the anticipation o
reoccurrence of that event. Fear is always future oriented. The adaptive nature of fear is such th
it urges organisms to take action and prepare for upcoming danger (Lang et al. 2000).
The strength of the conditional fear reaction is determined by two variables: the strength
the CS-US association and the intensity of the US memory. From a cognitive perspective, th
corresponds to a probability intensity estimation. A stimulus that signals high probability
an intense threat will elicit strong fear. Low probability and/or weak threat will produce less fea
222 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
9/37
ANXIOUS INDIVIDUALS SHOW DEVIANT PATTERNS OF FEAR CONDITIONING
Examining the behavior of anxious individuals in the Pavlovian fear-conditioning model has generated interest
hypotheses about the deviant processes in anxiety. First, anxious individuals may suffer from heightened fear
activity to dangerous stimuli. Second, anxious individuals may suffer from impaired extinction learning and, m
broadly, safety learning. Third, anxious individuals may suffer from overgeneralization of acquired fears to n
dangerous stimuli. Finally, anxious individuals may suffer from a failure to associate specific stimuli to the aver
event (unconditional stimulus), resulting in chronic threat perception and anxiety. Results consistent with al
these hypotheses have been found; an important question is whether there is an underlying unifying mechanism
Bobs level of fear is determined by (a) an explicit/implicit estimation of the chance of another panic
attack inside an elevator (with higher estimations relating to a stronger elevator-panic association) and
(b) an explicit/implicit estimation of the hazard of panic attacks (with higher hazard estimations relating
to more intense representations of panic and its consequences).
Deviant Fear Conditioning in Anxious Individuals
Pavlovian fear conditioning in itself is an adaptive mechanism. Learning to fear certain signalscan be crucial for survival. Examining the behaviors of anxious individuals in a standard fear-
conditioning procedure can shed light on the deviant processes in irrational anxiety: Where does
the mechanism go wrong (see sidebar Anxious Individuals Show Deviant Patterns of Fear Con-
ditioning)? A meta-analysis on fear-conditioning studies showed that individuals with an anxiety
disorder continue to react fearfully to stimuli that no longer signal danger (i.e., weakened fear
extinction; Lissek et al. 2005). This has been further supported by more recent studies with panic
disorder patients (Michael et al. 2007), posttraumatic stress disorder (PTSD) patients (Blechert
et al. 2007; Milad et al. 2007, 2009), and children with anxiety disorders (Craske et al. 2008). Fear
extinction deficits also correlate with symptom severity in PTSD (Norrholm et al. 2011).
Impaired fear extinction is also present in individuals at risk for anxiety disorders, suggesting
that it is an important etiological factor. Children from parents with anxiety disorders showedslower fear extinction than healthy control children exhibited (Craske et al. 2008). Impairments
in conditioned fear extinction predicted 31% of the variance of PTSD development in firefighter
recruits (Guthrie & Bryant 2006; for comparable results in soldiers deployed in Afghanistan, see
Lommen et al. 2013). Individuals with subclinical levels of anxiety (high scorers on a trait-anxiety
questionnaire; another risk factor for clinically severe anxiety) show a deviant pattern of neural
correlates during fear extinction, consistent with the pattern seen in clinically anxious samples
(Sehlmeyer et al. 2011).
It is of crucial importance to understand the mechanisms that underlie fear extinction and how
extinction can be improved. Standard learning theory provides different possibilities concerning
the mechanisms of extinction.
Mechanisms of Extinction
Reduction of the fear response may simply reflect the destruction of the underlying CS-US as-
sociation (e.g., Rescorla & Wagner 1972). Without that association, confrontations with the CS
will no longer activate the memory of the aversive US and hence elicit no fear. The CS returns
to a neutral state, as if no conditioning experience has taken place. In our example, to Bob the
elevator is just an elevator again. Second, the association may remain intact, but the memory of
www.annualreviews.org Fear Extinction and Relapse 223
8/11/2019 Review Relapse, 2013
10/37
the US may become devalued instead. Repeated CS presentations activate the US memory ov
and over again, without experiencing its aversive impact any more. If this induces a devaluation
the US memory, CS-based retrieval of this less aversive US memory will elicit less fear (Resco
& Heth 1975). In our example, elevators may still activate Bobs memory of the panic attack, b
this memory is not so frightening any more. Third, the extinction trials may leave the CS-U
association and the US memory intact and induce a (temporary) deactivation of the US memo
instead. The temporary deactivation inhibits the previously conditioned fear response (extinctio
This deactivating mechanism is commonly conceptualized as the learning of an inhibitory asso
ation (as opposed to the normal excitatory CS-US association). According to this view, extincti
can be seen as the development of an inhibitory CS-US association that counteracts the prev
ously formed excitatory CS-US association (Bouton 1993). The excitatory association activa
the US memory, but the inhibitory association serves to deactivate it (e.g., by augmenting t
activation threshold of that memory). The net effect will be the absence of fear to the CS (i.
extinction). In our example, Bob may have associated elevators with the absence of panic attack
which deactivates the fear arousing memory of the panic attack.
The destruction of the CS-US association provides the most favorable outcome, with no thre
of return of fear (erasure). Devaluation of the US memory reduces fear, but the intact CS-U
association leaves the possibility for the return of fear should the US memory regain its origin
intensity. The suppression of the CS-US association runs the risk of return of fear in the eve
that the suppressing source (inhibitory CS-US association) weakens.
THE DEVELOPMENT OF THE RETRIEVAL MODELOF EXTINCTION IN RATS
The research of Mark Bouton and others on renewal and reinstatement has greatly shaped o
understanding of extinction and relapse and has culminated in the retrieval model of extinctio
In order to fully explicate the specifics of this model, we discuss the rat conditioning studies th
stimulated its development.
What We Learned from Reinstatement StudiesReinstatement through US revaluation. Extinction may rely on the progressive devaluation
theUS memory(the devaluation hypothesis, describedabove). In that case, US-alone presentatio
duringreinstatement testing would reintensify theUS representation andthereby produce a retu
of fear to the associated CS (Rescorla & Heth 1975). However, Bouton & Bolles (1979a) show
that reinstatement of fear is a context-dependent phenomenon, a finding that is difficult to ali
with the US revaluation hypothesis. Rats that had received unsignaled shocks in a context differe
from extinction (a separate cage with distinctive features) failed to show the reinstatement effe
when they were exposed to the CS in the extinction context again. Together with the renew
phenomenon, this finding points to the important role of context in the return of fear.
Reinstatement exceeds contextual fear, but depends on it. It is conceivable (and often demo
strated) that unsignaled shocks induce a general fearof the entire context that surroundsthe subje
(Grillon 2002). Nevertheless, the amount of reinstated fear to the CS typically exceeds the amou
of fear elicited by the test context. This shows that the fear increase is a genuine return of th
fear-eliciting properties of the CS rather than a reflection of newly conditioned contextual fe
On the other hand, an additional experiment showed that contextual fear is the primary trigger f
the return of CS fear. Rats that received lengthy exposure to the test context after the unsignal
224 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
11/37
shocks displayed less reinstatement of CS fear as compared to rats that were exposed to an unre-
lated context (Bouton & Bolles 1979a). Arguably, the lengthy context exposure extinguished fear
of the context before the CS was presented. In sum, reinstated fear to the CS goes above and
beyond contextual fear, but also depends on it. As we discuss below, Bouton (2002) proposed that
the newly formed context-US association retrieves the CS-US association from memory.
Reinstatement depends on a context-CS association. Various learning models hypothesize
that presenting a CS in a specific context establishes a context-CS association in memory (e.g.,
sometimes-opponent process theory; Wagner 1981). Conditioning to the context can then produce
conditioned responding to the CS as well, mediated by the context-CS association. Accordingly,
Westbrook et al. (2002) also found a return of fear of the CS when the unsignaled shocks were
administered in the extinction context and tests occurred in a different context. Arguably, the
CS becomes associated with the context in extinction; later context-US conditioning will then
produce fear of the CS (mediated conditioning). This explanation of reinstatement emphasizes
the importance of new learning rather than retrieval.
The set of basic reinstatement findings suggests that the experience of an out-of-the-blue panic attack
will reevoke Bobs fears of elevators (a) if he is still in a state of heightened anxiety/arousal when having
to enter an elevator or (b) if he experienced the attack in a context that he associates with elevators
(e.g., an office building). In general, one could say that the unsignaled panic attack evokes a general
uncertainty about when and where attacks will recur. Because elevators have been associated with panic
in the past, they are odds-on predictors of future attacks for Bob.
What We Learned from Renewal Studies
Renewal through contextual fear. The contextual fear mechanism of reinstatement could also
underlie the ABA-renewal phenomenon. A return to the conditioning context A may reelicit some
contextual fear, which would then reinstate fear of the CS. However, closer inspection of the
renewal phenomenon rules out this explanation. First, researchers have observed ABA renewal in
the absence of any detectable fear of conditioning context A (Bouton & Bolles 1979b, Bouton &
King 1983). Second, renewal is also observed when the CS is tested in a novel context C that is
presumably neutral (ABC renewal).
Renewal through stimulus generalization. Stimuli may elicit different percepts in distinct
contexts. For instance, a yellow line may appear more orange against a red background. Also, con-
texts and stimuli may merge into a unitary representation, forming a Gestalt, before entering the
conditioning process (configural learning theories; e.g., Pearce 1987). In a renewal experiment,
the extinction CS may be perceived as different from the acquisition CS. It is well known that
conditional responses generalize on the basis of perceptual similarity (Pavlov 1927). On the first
extinction trial in a different context, the CS will evoke the conditional response partly (general-
ization decrement). As a consequence, only a part of the conditional response (and US association)
is submitted to extinction. When the original acquisition CS is presented at test (in ABA renewal),
thenonextinguished part of the conditional response (and US association) will be activated,leading
to return of fear (see Vervliet et al. 2004, 2005). This will also happen in ABC renewal, because
the nonextinguished part can generalize to CS in context C. A simple test of this mechanism,
however, is to observe whether there is indeed a response decrement on the first trial of ex-
tinction. Bouton and his colleagues have repeatedly reported that there is no detectable decre-
ment in the conditional response from the last CS acquisition trial in context A to the first CS
www.annualreviews.org Fear Extinction and Relapse 225
8/11/2019 Review Relapse, 2013
12/37
extinction trial in context B (Bouton 1994). This stands in sharp contrast to the large increase of t
conditional response when the CS is presented in the renewal test context. Hence, a simple gene
alization explanation is not sufficient. The data seem to point to the more complex and surprisi
hypothesis of an intrinsic asymmetry in the generalization of acquisition and the generalizati
of extinction. This was not anticipated by standard learning theory.
Renewal through contextual inhibition. CS-alone extinction trials may promote learning th
the entire extinction context is safe (conceptualized as the formation of an inhibitory contextU
association). This novel contextual meaning prevents fear reactions to the CS, although the CS-Uassociation remains intact. By implication, presenting the CS in another contextshould recover t
fear reaction (renewal). This contextual inhibition hypothesis was tested by assessing whether t
extinction context of the CS would reduce fear reactions to other separately conditioned stim
as well. Thus, rats were conditioned to a light and a tone in context A, followed by extincti
of the light in context B. During tests of the tone in context B, rats showed no decrease in fe
responding (Bouton & King 1983). Hence, in contrast to the contextual inhibition hypothes
there was no evidence of general inhibitory/safety learning about the extinction context. Th
shows that extinction is specific to both context and stimulus.
The set of renewal results suggests that a fear of elevators may reemerge in Bob when he has to enter
an elevator in a context different from exposure treatment, even when this novel context is itself notanxiety provoking. In addition, the fragility of the fear-extinction effectis not necessarily due to a failure
of eliciting the full fear in treatment or because Bob feels safe in the therapy context(e.g., because of the
presence of the therapist). Even in the absence of these extinction-preventing factors, the long-term
fear-extinction effect remains delicate.
The Retrieval Model of Extinction and Return of Fear
Bouton and colleagues proposed that extinction learning is based on the formation of a secon
inhibitory CS-US association that coexists with the earlier formed excitatory association. Wh
the inhibitory association is active, the expression of the CS-US association is suppressed (the C
evokes no fear). When the inhibitory association is inactive, the CS-US association is express
(the CS evokes fear again). The data show that the activity of inhibitory associations depen
on the context in which the CS is presented. Bouton and colleagues proposed that activation
the inhibitory association requires the concurrent presence of both the CS and the extinctio
context. Graphically, this is represented by an extra representation node that receives input fro
both the context and the CS, and which is connected (through an inhibitory association) wi
the US representation (see Figure 3). The extra node functions as an AND gate so that bo
the extinction context and the CS need to be present before the inhibition is active. The mo
important aspect of this model is that the excitatory CS-US association does not involve su
an AND gate and is therefore independent from the context (in accordance with the observ
asymmetry between the generalization of acquisition and the generalization of extinction).
This model can also be applied to spontaneous recovery if time is considered as a context
well. External and internal contexts dynamically develop over time so that the context at the tim
of extinction is most likely to differ from the context at a long-term test. The extinction memory
more difficult to retrieve in this newcontext than theconditioningmemory, leading to spontaneo
recovery of fear. Reinstatement is also accounted for from a retrieval viewpoint. The unsignal
US presentations in a test for reinstatement produce conditioning with the context (context-U
association). This novel association retrieves the earlier CS-US association, producing recover
fear responses to the CS.
226 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
13/37
CS US Memory after conditioning
Memory after extinction
CX
USCS
Figure 3
A graphical representation of the retrieval model of extinction. The memory after extinction is morecomplex than the memory after conditioning. Circles indicate memory representations of stimuli (CS,conditional stimulus; US, unconditional stimulus; CX, context); arrows indicate an excitatory association; theline with a bar at the end represents an inhibitory association.
The representation of the retrieval model of extinction is rather abstract, but Mark Bouton
has proposed several ways to interpret this more meaningfully. First, by connecting it to the
occasion-setting literature, it becomes clear that the extinction context actually signals when the
CS-US inhibitory association is valid. The AND gate can be seen as a hierarchical node through
which the context connects to the (inhibitory) association between the CS and the US. Thecontext sets the occasion for the inhibitory association. Second, extinction turns a CS into an
ambiguous stimulus with two opposing meanings. The excitatory association provides a danger
meaning, whereas the inhibitory association provides a safety meaning. From this perspective,
the extinction context disambiguates the meaning of the CS by delineating the occasions when
the CS is safe. Finally, the model can be viewed from a memory-retrieval perspective. Some
memories are weak and easily forgotten in the sense that more retrieval cues are needed in
order to remember them fully. Often, contexts provide potent retrieval cues in this regard. The
data seem to suggest that extinction memories (CS-US inhibition) are weak and easily forgotten
outside of the extinction context. In other words, the extinction context functions as a potent
retrieval cue for the extinction memory that is otherwise easily forgotten (hence, return of fear
outside of the extinction context). Original excitatory fear memories, on the other hand, are easilyactivated and do not depend on contextual retrieval cues. From this perspective, the link between
the context representation and the AND gate node is conceptualized as a retrieval link of the
inhibitory CS-US association (the extinction memory). The extinction context acts to retrieve the
extinction memory, thereby signaling when the CS-US inhibitory association is valid and hence
disambiguating the dangerous-safe CS. In our example, the treatment context may act to retrieve
the fear extinction memory, thereby signaling when the elevatorno panic association is valid and
hence disambiguating dangerous-safe elevators to Bob. An important implication is that the long-
term success of extinction will depend on the retrievability of the extinction memory (Bouton
2002, Craske et al. 2012). This concurs with the clinical perspective that the achievement of fear
extinction (i.e., fear reduction over the course of exposure therapy) is not the major challenge, but
rather the sustainability of obtained extinction effects (i.e., prevention of return of fear).
The Neurological Model of Extinction and Return of Fear
The neurobiological basis of (conditioned) fear extinction has been well mapped (Quirk & Milad
2012). The amygdala plays a central role in the conditioning of fear reactions. Sensory information
about the CS and the US is transmitted from the thalamus to the basolateral nucleus of the
amygdala. The plasticity in that center forms the basis of the CS-US association, and connections
www.annualreviews.org Fear Extinction and Relapse 227
8/11/2019 Review Relapse, 2013
14/37
to the central nucleus of the amygdala, with its own connections to brainstem regions, serve t
expression of fear. Other regions that are implicated in fear are the dorsal anterior cingula
cortex (presumably related to threat anticipation) and the insular cortex (presumably related
interoception, awareness, and sensitivity to visceral activity). Fear extinction relies on activ
of the amygdala, hippocampus, and ventromedial prefrontal cortex (vmPFC) in concert. T
hippocampus is known for its role in computing a unitary representation of complex stimuli, su
as contexts. The vmPFC has connections to the intercalated cells of the amygdala, which ser
an inhibitory (gamma-aminobutyric acidmediated) function. It is believed that the hippocamp
provides the contextual information and signals when the vmPFC should activate the inhibito
influence in the amygdala, resulting in inhibition of the fear reaction. If the CS is encounter
within the extinction context, this extinction network is activated and the fear reaction is inhibite
If the CS is encountered outside of the extinction context, this extinction network is not activat
and fear is expressed. Interestingly, the available research supports a much more consistent ro
for the hippocampal-vmPFC network in extinction recall (e.g., 24 hours later) than during init
extinction learning (Milad & Quirk 2002, Quirk et al. 2000). This suggests that the network
particularly important for the retrieval of the extinction memory. It is clear that the neurologic
model of extinction maps well onto the behavioral retrieval model of extinction (Quirk & Mil
2012).
Multiple Pathways to Return of Fear?
Rodent studies on the extinction and return of fear are mostly in line with the retrieval model
extinction. Alternative explanations based on standard learning theory have been systematica
discarded. However, phenomena and their underlying mechanisms may vary depending on proc
dural parameters, species, subject samples, and individual variables. Some studies have replicat
critical results in appetitive and instrumental conditioning in rats, showing generalizability of t
retrieval model of extinction (Bouton 2002). Nevertheless, there is no a priori reason to expect th
extinction and return of fear are subsumed by the same mechanism in humans (or other anim
species and procedures). Even within the human species, there may be differences between healt
and clinical samples, or other individual variables may exist. Moreover, return of fear may re
on a combination of different mechanisms that operate at the same time (e.g., part of the retu
of fear reflects contextual inhibition, the other part reflects context-specific CS inhibition). Th
may have implications for the implementation of strategies to prevent return of fear, but little
known in this regard. The extinction and the return of fear remain a complex issue, even thou
the retrieval model is the major guideline for experimental research.
PRECLINICAL HUMAN STUDIES
The memory retrieval model of extinction is exclusively based on rat conditioning research. A
though the model provides an interesting perspective on clinical fear extinction and relapse, the
is obviously a large translational gap between conditioned rats and anxious patients. Testing t
translation on several intermediate steps can bridge that gap. One important step consists of fea
conditioning studies in healthy human individuals.1 The advantage of this type of paradigm is t
1A major advantage of the Pavlovian fear-conditioning procedure is that it can be similarly applied in animals and humans. Ba number of procedural differences exist that may be important. First, due to ethical constraints, the participant individuaselects an intensity of the aversive US regarded as uncomfortable, but not painful at the start of the experiment. Sumildly unpleasant USs may produce only weak fear reactions. Second, the fear measures often differ from those in anim
228 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
15/37
high level of experimental control (fear conditioning) and the absence of a cross-species transla-
tion problem. Hence, the establishment of return of fear and testing the underlying mechanism in
nonclinical human participants provides crucial information about the applicability of the memory
retrieval model of extinction to clinical anxiety and treatment.
Reinstatement in Human Fear Conditioning
Reinstatement has long been established in animal fear-conditioning research (Rescorla & Heth
1975), but demonstrations in humans are more recent. There are now 15 published studies on fearreinstatement with a total of 742 participants, comprising a variety of USs (loud noise, electrical
shock, air blast to the throat), CSs (neutral pictures, lights, fear-relevant pictures), and fear mea-
sures (skin conductance, fear-potentiated startle, subjective ratings of fear and threat expectancy).
Together, these studies establish the robustness of the reinstatement procedure to elicit return of
fear in preclinical research in humans. They also reveal important aspects of the fear-reinstatement
mechanism in humans.
Reinstatement is context dependent in humans. LaBar & Phelps (2005) tested whether hu-
man fear reinstatement is influenced by contextual changes, as in rats (Bouton & Bolles 1979a).
Following conditioning and extinction in one context, participants received unsignaled USs ei-
ther in the same or in a novel context (a different room). Final tests of the extinguished CS alwaysoccurred in the extinction context. As expected, reinstatement was observed only in the group
that received no context changes. This suggests a common mechanism between human and rat
conditioning studies. In addition, two amnesic patients with hippocampal damage failed to show
reinstatement in the same context (LaBar & Phelps 2005). This supports the hypothesis that the
hippocampus is implicated in the processing of contexts and the productionof return of fear (Quirk
& Milad 2012).
Reinstatement can occur with various timings after acquisition and extinction. In rodent
studies, the phases of acquisition, extinction, and recovery test are typically distributed over sep-
arate days. In contrast, in human studies these phases are often combined in a single session
(mostly for practical reasons). Due to these time differences, the rodent studies may rely more
on long-term memory processes with possible implications for the reinstatement mechanism.
Nevertheless, human fear reinstatement has been shown with a 24-hour or 48-hour time interval
between conditioning and extinction reinstatement, or with the three phases distributed over three
separate days (Kindt & Soeter 2013, Kindt et al. 2009, Norrholm et al. 2006, Schiller et al. 2009,
Soeter & Kindt 2011). In addition, Kindt et al. (2009) and Schiller et al. (2009) showed successful
reinstatement when participants were retested one month and one year later, respectively. This
supports the generalizability of the reinstatement findings between rodent and human preclinical
research.
Reinstatement produces a return of attentional biases. Anxiety patients often show an at-
tentional bias for threat-related material (Bar-Haim et al. 2007). Attention to threat stimuli is a
first step in the production of a fear reaction. This bias may be a causal or maintaining factor
of anxiety symptomatology (MacLeod et al. 2002). Such cognitive biases are difficult to model
conditioning experiments. The only exceptionis the fear-potentiated startle reflex, although effortsare being madeto translatea measure of freezing to humans. Third, human studies mostly use visual stimuli, whereas animal studies mostly use auditivestimuli as CSs. Finally, the timing of CS, US, intertrial intervals, and intersession intervals differs greatly. These differencesshould be taken into account when comparing results from animal and human experiments.
www.annualreviews.org Fear Extinction and Relapse 229
8/11/2019 Review Relapse, 2013
16/37
in rodents. Preclinical human research has shown that fear-conditioned stimuli elicit an atte
tional bias similar to the bias seen in anxious patients. Furthermore, this attentional bias decreas
through extinction and returns at a reinstatement test (Dirikx et al. 2004, Hermans et al. 200
Van Damme et al. 2006). Together, these data suggest that return of fear can be accompanied
a return of attentional biases, which may further exacerbate fear reactivity.
Reinstatement is larger to negatively valenced stimuli. Human participants typically evalu
fear-conditioned stimuli as negative (rated on a positive-negative scale). Conditioned negati
valence is more difficult to extinguish than conditioned fear reactions (Vansteenwegen et al. 200
Unextinguished negative valence may provide a vulnerability factor for return of fear. Diri
et al. (2004, 2007) and Hermans et al. (2005) found a significant correlation between the rat
negativity of the CS after extinction and the amount of return of fear in a subsequent test f
reinstatement. Theauthors proposed a mechanism for reinstatement, based on theassumption th
fear represents a combination of negative valence and arousal (Lang et al. 1990). They speculat
that the reinstating USs induce general arousal that can recombine with the residual negati
valence of the CS and produce return of fear. This hypothesis awaits further experimentation, b
it has the advantage that it can predict individual differences in reinstatement based on end-o
extinction performance (unextinguished negative valence).
Reinstatement is not always stimulus specific in human fear conditioning. Human fea
conditioning studies routinely include a nonconditioned control stimulus. This is a stimul
(termed the CS-) that is presented equally as often as the CS+ but is never paired with t
US. The conditional fear response is defined as the differential reaction to the CS+ and CS-,
their only difference is the association with the US. It follows that the reinstatement of fear shou
be specific to the CS+. However, a number of studies have observed equally strong increases
fear responding to the CS- as well (Dirikx et al. 2009, Kull et al. 2012, Sokol & Lovibond 2012
This seems to contradict the view of reinstatement as a return of fear phenomenon, as there w
no prior fear to the CS-. Moreover, it is unclear why some studies find differential reinstatement
the CS+ and not to the CS- whereas others do not. One interesting observation is that participan
with higher levels of self-reported trait anxiety showed more fear of the CS- at reinstatement t
(Kindt et al. 2009, Soeter & Kindt 2010). This suggests that return of fear is a widespread, norm
phenomenon, but anxious individuals may show nonspecific levels of return of fear that augme
the risk for relapse. Some studies may have unknowingly recruited more anxious participants th
others, leading to more or less differential reinstatement.
There are two candidate mechanisms for the relationship between trait anxiety and nondiffe
ential reinstatement. First, anxious individuals have a propensity for fear generalization (Liss
et al. 2010). Consequently, fear of the CS- may reflect generalization from the returned fear
the CS+. Second, anxious individuals have a propensity to fear the entire experimental conte
(Grillon 2002). Consequently, the reinstating USs may induce a higher level of contextual fear
that any stimulus presented in that context will induce fear (Sokol & Lovibond 2012).
Reinstatement can be triggered by other aversive stimuli. Most studies on reinstateme
have used the exact same US as used during acquisition to elicit return of fear. However, litt
is known about the boundary conditions of the reinstating US. Sokol & Lovibond (2012) test
the degree to which other aversive stimuli produce return of fear after extinction. They fou
increased fear to the CS+ regardless of whether the same US or a different aversive stimulus w
used (for similar findings in rats, see Rescorla & Heth 1975). Importantly, participants who h
received administrations of the different aversive stimulus expected this stimulus to follow the C
230 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
17/37
at test (rather than the original US). This has significant implications for the retrieval model of
extinction, as it suggests that not all aspects of the US are crucial for retrieval. Rather, the newly
learned contextual fear (context fear) retrieves the old CS fear, regardless of the fear-inducing US.
New fears can trigger old fears.
Reinstatement in preclinical human research: Conclusion. Human studies on fear reinstate-
ment have shown that the effect is context dependent, occurs primarily in response to CSs with
unextinguished negative valence, and is triggered by other aversive stimuli as well. Together,
these data suggest that any stimulus that induces arousal in the test context can trigger return of
fear, particularly to CSs with unextinguished negative valence. The clinical implication would be
that any stressful life event that induces fear or arousal is able to revive old fears that have been
extinguished through exposures.
The results on attentional biases show that reinstatement may elicit a return not only of the
fear symptoms themselves but also of their maintaining factors. Finally, the return of fear may
easily spill over to other stimuli that are not related to the threat, particularly in individuals with
high trait anxiety. The clinical implication is that, if left untreated, the return of fear through
reinstatement is very likely to spread out and trigger a full-blown relapse. Accordingly, a clinical
study found that unextinguished negative valence at posttreatment predicted individual levels of
return of fear (Vasey et al. 2012).
Bobs fear of elevators may return after stressful life events other than panic attacks (e.g., marital
problems). In addition, Bobs vulnerability for relapse may increase if he continues to dislike elevators
after treatment. The return of fear can be accompanied by an increased attention toward escape routes,
phone boxes, alarm buttons, elevator movements, etc., which probably hinders reextinction of fear.
The return of fear may also extend to other situations that previously were not associated with the
panic attacks (riding in cars, trains, buses). This suggests that each relapse episode contains the risk of
exacerbating the symptomatology.
Renewal in Human Fear Conditioning
Around the same time that the fear reinstatement effect was established in human fear condition-
ing, researchers also demonstrated the contextual renewal effect. To recapitulate, renewal occurs
when shifting the surrounding context between extinction and test produces a return of fear to the
extinguished CS. Renewal hasattracted more interest than reinstatement, resulting in 24 published
studies with a total of 1,248 participants. The phenomenon has now been demonstrated with a va-
riety of USs (loud noise, electrical shock, negative International Affective Picture System images),
a variety of CSs (neutral pictures, lights, fear-relevant pictures), fear measures (skin conductance,
fear-potentiated startle, subjective ratings of threat expectancy, fMRI brain scanning), a variety of
contexts (different rooms, different central lightings in the same room, pictures of rooms, virtual
reality contexts, background sounds), and different participant samples (healthy individuals, anx-
ious patients). Together, these studies establish the robustness of the renewal procedure to elicit
return of fear in preclinical research in humans. In addition, the studies reveal important aspects
of the fear renewal mechanism in humans.
ABA renewal is independent of contextual fear. One potential mechanism for ABA renewal is
based on the assumption that the context A may acquire a certain level of conditional fear during
the conditioning phase (see section What We Learned from Renewal Studies). Returning to that
context will reactivate a general fear response. Tests of the CS in that context will then result
www.annualreviews.org Fear Extinction and Relapse 231
8/11/2019 Review Relapse, 2013
18/37
in the recording of fear, not because fear of the CS has returned, but because the entire conte
elicits fear. Alvarez et al. (2007) observed that the conditioning context A elicited more fear th
the extinction context B, which is in line with the hypothesis above. However, the level of fear
context A did not correlate with the level of fear of the extinguished CS tested in that contex
Although this is a statistical null finding, it suggests that the renewal of fear to the extinguish
CS is independent of the amount of fear elicited by the test context (cf. rodent findings; Bout
& King 1983).
ABC renewal occurs but is weaker than ABA renewal. A more stringent test of the influen
of fear of the test context on renewal is to assess the extinguished CS in a neutral context C (AB
renewal). Effting & Kindt (2007) found an increase of fear to the extinguished CS in a nov
context C, but less so when compared to ABA renewal. Also, the ABC renewal displayed an equ
increase to the CS- (nondifferential return of fear), whereas the ABA renewal was specific to t
CS+ (differential return of fear). Neumann & Kitlertsirivatana (2010) also observed weaker AB
renewal compared to ABA renewal, but this time the return of fear was differential in both renew
procedures. Many procedural differences exist between the two studies, and it is not possible
identify the crucial parameter for differential versus nondifferential ABC renewal.
Milad et al. (2005a) and Neumann & Kitlertsirivatana (2010) observed complete ABA renew
as compared with another CS in the test context that had never been extinguished. These studisuggest that ABA renewal may in general lead to a maximal return of fear, whereas ABC renew
leads to a submaximal return of fear (cf. rodent studies; Harris et al. 2000). Based on results
rodents, one would expect even smaller renewal effects within an AAB renewal procedure (Thom
et al. 2003). Unfortunately, there are no reports of AAB renewal in human fear conditioning.
ABA renewal does not depend (entirely) on inhibition of the extinction context. Renew
can result from general inhibition (safety) learning during extinction. Instead of learning that t
CS is no longer followed by the US (CS-US inhibition), participants can learn that the US nev
occurs in the extinction context B (context-US inhibition). Testing the CS in a different conte
will remove this contextual inhibition and produce renewal of fear to the CS (see section Wh
We Learned from Renewal Studies). Milad et al. (2007) fear conditioned two stimuli during tacquisition phase (CS1 and CS2). Next, CS1 was extinguished in context B, whereas CS2 w
not. On the next day, CS2 elicited more fear in context B than CS1. This shows that the fea
extinction effect to CS1 is not entirely due to inhibition by context B. Arguably, the extinctio
context B is not a general signal of safety (conditioned inhibitor) but rather acts specifically on t
conditioned-and-extinguished CS1.
ABA renewal can occur with various timings after acquisition and extinction. Where
most human studies combine the phases of acquisition, extinction, and renewal test in one sessio
some have separated phases or days. Studies from the Milad laboratory routinely investigate AB
renewal effects with acquisition and extinction occurring in one session, followed by renewal tes
in both contexts 24 hours later (Milad et al. 2005a; see also Kalisch et al. 2006). The ABA resu
in this procedure confirm that the context specificity of fear extinction is present with long-ter
memory mechanisms as well. There is at present no study that has distributed the three phas
over separate days (cf. rodent studies). Importantly, Zeidan et al. (2011) have shown that the AB
renewal effect is stable over time when the procedure is conducted three times with 12 wee
between each testing. This shows the reliability of the ABA renewal procedure and supports
potential for diagnostic purposes and the screening of novel intervention methods.
232 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
19/37
ABA renewal is stronger with fear-relevant stimuli and contexts. Fear extinction is delayed
when fear-relevant pictures (e.g., spiders and snakes) are used as conditional stimuli instead of
geometric figures (for a review, see Ohman & Mineka 2001). This may reflect a different extinction
mechanism and hence a different renewal mechanism. In two studies, Neumann & Longbottom
(2008) observed contextual renewal with fear-relevant stimuli. Together with the reinstatement
results by Kindt et al. (2009), Soeter & Kindt (2010, 2011), and Kindt & Soeter (2013), these
findings show that return of fear occurs with fear-relevant stimuli as well, despite their slower
extinction rate.
Interestingly, the size of the renewal effect depended on an interaction between the fear-
relevance of the CSs and the naturalistic context in which it was tested (picture of indoor office or
outdoor bush). The mechanism of this interaction is not yet clear and merits further research.
Fear conditioning is not entirely context free in humans. A major tenet of the retrieval theory
of extinction is that conditioning memories are much easier to retrieve than extinction memories,
with the latter requiring input from contextual retrieval cues. Indeed, Bouton and colleagues
routinely observed no decrement in fear responding to the CS when it is first presented in the
extinction context B. This contrast with the clear effect of the context change after extinction
provided the impetus for much of the return of fear research. Most human studies have not
tested the fear decrement after acquisition, and from those that tested it, the majority reported a
statistically significant fear decrement (reviewed by Vervliet et al. 2013). This observation casts
doubts on the exact mechanism underlying renewal in human fear conditioning.
ABC renewal elicits more ambiguity than ABA renewal. Bouton formulated his retrieval
model of extinction in terms of ambiguity. That is, extinction turns a CS into an ambiguous
stimulus (with both an excitatory and inhibitory association); the context serves to disambiguate
the current meaning of the CS (activating the inhibitory association or not). In humans, the
latency to provide a rating of the current status of the CS has been interpreted as a measure
of CS ambiguity (Lissek et al. 2008). Using an ABA- and ABC-renewal procedure, Neumann
& Kitlertsirivatana (2010) found that the latency increased for any stimulus whenever a context
changed (after acquisition as well as extinction). However, the latency was highest for the CS+
when presented in the novel test context C. This suggests that the ambiguity is low in a test for
ABA renewal but higher in a test for ABC renewal.
Renewal can be triggered by changes in perceptual features of the CS. Vervliet et al. (2004,
2005) found that conditional fear generalizes easily to a stimulus that shares common features
with the CS, whereas fear extinction with this stimulus does not generalize back to the CS. This
return of fear demonstrates ABA renewal with stimulus changes rather than context changes. Fear
extinction with the CS itself did generalize to the other stimulus (i.e., no AAB renewal).
Extinction recruits activity of the hippocampus and vmPFC in concert. In a study by Milad
et al. (2007; see above), participants were tested in an MRI scanner during all phases of the ex-
periment. Day 2 testing in the extinction context (B) revealed augmented activity in two loci of
the vmPFC and the hippocampus during the extinguished stimulus. Moreover, a positive corre-
lation was found between activity in these areas and the strength of extinction recall as measured
by skin conductance. This pattern of brain activity is similar to what has been found in rodents
and supports the neurological theory that fear acquisition is based on amygdala activity and that
(the recall of) extinction additionally recruits activity of the vmPFC and the hippocampus in con-
cert to inhibit amygdala activity (and hence inhibit fear). Arguably, the hippocampus forms a
www.annualreviews.org Fear Extinction and Relapse 233
8/11/2019 Review Relapse, 2013
20/37
unitary representation of the extinction context (which is often a diffuse set of background cue
and the vmPFC is the inhibitory agent that activates the inhibitory parts of the amygdala, t
intercalated cells (Quirk & Milad 2012). Further in line with this account, Milad et al. (2005
reported that thickness of the vmPFC correlated with the level of extinction (skin conductanc
when participants viewed the CS+ in the extinction context.
Kalisch et al. (2006) found very similar results in a similar design. Acquisition and extincti
learning phases occurred in separate contexts on Day 1; tests were conducted in both contex
on Day 2 (the acquisition context test phases started with an unsignaled shock, which is akin to
reinstatement procedure). Again, the extinguished CS elicited augmented activity in the vmPF
and the (anterior) hippocampus in the extinction context. In line with the neurological theory
extinction recall, the vmPFC and the hippocampus correlated positively in the extinction conte
but not in the acquisition context.
Renewal in preclinical human research: Conclusion. Preclinical research in humans has no
widely demonstrated that conditioned fears can return with a change of context after extinctio
ABA renewal can elicit a complete return of fear and is generally larger than ABC renewal. In li
with the retrieval model of extinction, ABA renewal is not related to the level of fear elicited by t
context A, and the extinction effect is not (or at least not entirely) due to general inhibition/safe
learning to context B. This suggests that extinction is based on the development of context-speci
CS inhibition (Bouton 2002). However, stimulus generalization is not ruled out as an alternatiexplanation, and some additional critical tests are lacking to fully validate the retrieval model
extinction in human fear conditioning (see Vervliet et al. 2013).
Brain scanning studies with fMRI techniques have revealed a remarkable similarity with t
neurocircuitry revealed in rodent brain lesion research. The fear extinction network requir
concerted activity in the hippocampus, vmPFC and the amygdala. This supports the translati
between preclinical rodent and human studies.
Human studies have also revealed additional information relevant to the renewal mechanism
First, the type of CSs and contexts used is not trivial. Certain matches between CSs and contex
produce stronger renewal effects than others. There is no current theory that explains these effec
Second, the ambiguity of the conditioned-and-extinguished CS is relatively low when presented
the conditioning context A. The prior learning history with context A may directly disambiguathe meaning of the CS. The ambiguity is higher when the conditioned-and-extinguished CS
tested in a novel context C. We have seen that the renewal effect is lower in context C as compar
to A. This may suggest that the novel context C is not as good a disambiguating agent as th
conditioning context A. Possibly, the renewal effect in C is more malleable than in A, which wou
provide more opportunity for renewal prevention in ABC renewal as compared to ABA renew
The results suggest that return of Bobs fear may be stronger to the original elevator where he experi-
enced the panic attack as compared to a similar elevator in another building. The latter may elicit more
ambiguity (comprising both danger and safety properties). Interestingly, anxious individuals tend to
interpret ambiguous situations as dangerous, which is possibly a marker for anxiety pathology (Lissek
et al. 2006). In addition, some buildings (contexts) may trigger more fear of elevators than others (e.g.,high buildings, old buildings, hotels, places far away from home). These factors may modulate the
occurrence and intensity of return of fear.
Spontaneous Recovery in Human Fear Conditioning
Spontaneous recovery occurs when a time interval between the last CS extinction trial and the fi
CS test trial produces a return of conditioned responding (Pavlov 1927). This phenomenon h
234 Vervliet Craske Hermans
8/11/2019 Review Relapse, 2013
21/37
received less attention in human research, but at least five studies with a total of 471 healthy human
participants have established the basic effect (Guastella et al. 2007; Huff et al. 2009; Norrholm et al.
2008; Schiller et al. 2008, 2009). These studies included different fear measures (fear-potentiated
startle reflex, skin conductance, subjective ratings of threat expectancy), different types of CSs
(neutral lights and pictures, fear-relevant pictures), and different types of USs (electrical shock,
airblast to the throat). The time interval between extinction and test varied between 24 hours and
94 hours. No study hassystematically tested a dose-dependent relationship between extinction-test
interval and the amount of spontaneous recovery (cf. rodent research; Quirk 2002).
ROADS TO RELAPSE PREVENTION
Return of fear is now widely established in rodent and human preclinical research. In the clinical
setting, return of fear jeopardizes the long-term efficacy of exposure-based treatments of anxiety. A
return of fear can trigger a full-blown relapse of the anxiety disorder. Preventing the return of fear
seems to be a major route toward the prevention of relapse of anxiety disorders. The behavioral
and neurological analysis of the extinction and return of fear have provided the impetus for a
variety of prevention strategies that have been put to test.
Preventing Relapse by Strengthening the Extinction Memory
Arguably, return of fear results from a failure to activate/retrieve the extinction memory. Strategies
to enhance the strength and retrievability of extinction memories may prevent return of fear.
Massive extinction prevents relapse in rats. Augmenting the number of extinction trials may
strengthen the extinction memory and weaken the return of fear. Denniston et al. (2003) found
almost no ABA and ABC renewal when the extinction trials exceeded the acquisition trials with a
multiplication factor of 100. However, no human studies have evaluated the number of exposure
trials with respect to offsetting context renewal, spontaneous recovery, or reinstatement.
Compound extinction prevents relapse in rats but produces unfavorable results in humans.
Associative learning models hypothesize that extinction learning is driven by error reduction. Early
in extinction, theCS produces a prediction error by signalinga US that no longeroccurs. Inhibition
learning reduces this error. Based on this hypothesis, augmenting the prediction error should
lead to stronger inhibition learning and maybe stronger memory formation. Using rats, Thomas
et al. (2003) combined three fear-conditioned stimuli in extinction as a means of augmenting the
prediction error. As expected, extinction proceeded faster to this compound stimulus, and ABA-
renewal tests revealed weaker return of fear to the individual stimuli (as compared to individually
extinguished stimuli). Along the same lines, Rescorla (2006) found that combining individual
and compound trials in extinction produces less spontaneous recovery and reinstatement of the
conditioned fear responses.
To date, there have been no prevention studies with compound extinction trials in humans.
However, Lovibond et al. (2000) and Vervliet et al. (2007) found deleterious effects of a compound
extinction procedure on the extinction effect itself. Tests with the individual stimuli revealed intact
levels of fear. The mixed individual-compound extinction procedure awaits further experimenta-
tion in humans.
Extinction retrieval cues may prevent relapse in humans, but the exact conditions are un-
clear. Theretrieval view on extinction implies that facilitating theretrievalof extinctionmemories
will prevent the return of fear. Brooks & Bouton (1994) proposed that random presentations of
www.annualreviews.org Fear Extinction and Relapse 235
8/11/2019 Review Relapse, 2013
22/37
an extra stimulus during the entire extinction phase would provide this stimulus the ability
retrieve the extinction memory (possibly through a stimulus-context association). They provid
the proof of principle of this method in an appetitive conditioning preparation in rats. Three h
manfear-conditioning studies have now shown beneficial effects of retrieval cues on the preventi
of ABA renewal (Dibbets et al. 2008, Dibbets & Maes 2011, Vansteenwegen et al. 2006). Ext
tests in the Dibbets studies suggested that the retrieval cue turned into a safety signal (in contra
to Brooks & Bouton 1994). That is, the retrieval cue also reduced fear of separately condition
stimuli. In addition, fear returned when the conditioned-and-extinguished CS was presented
its own extinction context but in the absence of the retrieval cue. This is an unfavorable outcom
as it is counterproductive in clinical practice to create dependence on safety signals (breaki
such dependence is often a primary goal of treatment). More research is needed to outline th
procedural conditions to produce a benign retrieval cue versus a malignant safety cue.
A clinical study found only a very weak effect of retrieval cues on the return of fear after exposu
treatment (of public speaking anxiety; Culver et al. 2011). It is clear that even though retriev
cues may be beneficial for the prevention of return of fear, we need much more information abo
the conditions for success.
Mental reinstatement prevents relapse in humans. Rather than presenting a cue to retrie
the extinction memory, one can just ask human participants to mentally reinstate the treatme
episode (Mystkowski et al. 2006). This strategy produced a clear reduction of contextual renew
of fear in an exposure treatment study with spider-anxious individuals. This provides a proof
principle that memory retrieval can prevent return of fear. It remains to be seen how nonverb
retrieval cues can assume this function.
Multiple contexts extinction can prevent relapse in rats and humans, under some cond
tions. Extinction is typically specific to the context in which learning took place. Conducti
extinction in multiple contexts would obviously expand the number of contexts capable of pr
ducing the extinction effect. This method has proven successful to reduce and even abolish AB
renewal in rodent and human fear conditioning (Bandarian Balooch et al. 2012, Gunther et
1998). An analogous clinical study showed the same effect in spider-fearful individuals who we
exposed to videotaped spiders against different background contexts (Vansteenwegen et al. 200
However, a rat conditioning study comprising three experiments and a human conditioni
study comprising two experiments failed to find a reducing effect on ABA or ABC renewal (Bout
et al. 2006, Neumann et al. 2007). The exact reasons for this discrepancy are unclear, but tw
additional studies have revealed specific conditions that modulate the effects of multiple contex
extinction. One rat conditioning study found a beneficial effect of multiple contexts extinction
ABA renewal only with extensive extinction training (the number of acquisition trials multipli
by 15; Thomas et al. 2009). One human conditioning study found that multiple contexts extincti
abolished ABC renewal when the extinction contexts were similar to the test context (Bandari
Balooch & Neumann 2011). Finally, the rat study by Gunther et al. (1998) also showed th
multiple contexts extinction failed to reduce renewal when fear conditioning had occurred in t
same number of contexts.
Internal states can also function as context. Kircanski et al. (2012) and Culver et al. (201
found that greater variability in terms of fear levels throughout exposure (i.e., repeated increas
followingdecreases in minuteto minuteanxietylevels) is a positive predictor of follow-upoutcom
in contaminant anxietyand public speaking anxiety. Arguably,the variability of fear levels produc
multiple internal contexts during extinction, thereby enhancing the retrievability of extinction
the long-term. These correlational studies need further experimental examinat
Recommended