|Home | About | Journals | Submit | Contact Us | Français|
Conditioned drug craving and withdrawal elicited by cues paired with drug use or acute withdrawal are among the many factors contributing to compulsive drug taking. Understanding how to stop these cues from having these effects is a major goal of addiction research. Extinction is a form of learning in which associations between cues and the events they predict are weakened by exposure to the cues in the absence of those events. Evidence from animal models suggests that conditioned responses to drug cues can be extinguished, although the degree to which this occurs in humans is controversial. Investigations into the neurobiological substrates of extinction of conditioned drug craving and withdrawal may facilitate the successful use of drug cue extinction within clinical contexts. While this work is still in the early stages, there are indications that extinction of drug- and withdrawal-paired cues shares neural mechanisms with extinction of conditioned fear. Using the fear extinction literature as a template, it is possible to organize the observations on drug cue extinction into a cohesive framework.
Addiction is characterized as a compulsively relapsing pattern of drug use that persists in the face of negative consequences and a desire to stop using. It is a complex and multifaceted disorder that has resisted simple solutions, and current therapies are disappointing in that many addicts have difficulty maintaining abstinence.
Conditioned drug craving and withdrawal elicited by drug-paired cues are among the many factors contributing to compulsive drug use. It is known that addicts respond to cues such as drugs, drug paraphernalia, people using drugs, and environments where drugs were consumed, with subjective reports of craving and withdrawal as well as physiological indices of arousal, anxiety, and discomfort (Childress et al., 1986; Ehrman et al., 1992; Foltin and Haney, 2000; Herman, 1974; Hugdahl and Ternes, 1981; O’Brien et al., 1977; Pomerleau et al., 1983; Sideroff and Jarvik, 1980; Wikler, 1948). When asked to consider factors contributing to relapse, addicts identify drug-paired cues as being equally or more powerful than other influences including moods or impulsive choices (Heather et al., 1991).
Understanding how to stop these cues from having these effects is a major goal of addiction research. In this review we focus on one way in which this might be accomplished: extinction, a form of learning in which associations between cues and the events they predict are weakened by exposure to the same cues in the absence of those events. Accumulating evidence indicates that conditioned drug craving and withdrawal in animals can be extinguished through exposure to drug- or withdrawal-paired cues in the absence of drug administration or acute withdrawal. The degree to which this occurs in human addicts is less clear, but numerous laboratories (including our own) are engaged in studies designed to identify the behavioral and neurobiological mechanisms of drug- or withdrawal-paired cue extinction in the hope that increased understanding of those mechanisms will facilitate successful application of extinction procedures in therapeutic settings.
The organization of this review is as follows: we describe what has been learned about the behavioral features of extinction of drug- and withdrawal-paired cues in animals, the outcome of clinical studies of extinction of drug cues in people, and the emerging literature on the neurobiological substrates of extinction of conditioned drug craving and withdrawal. In our discussion of extinction we draw substantially from the fear extinction literature to provide a framework within which to consider the relatively modest literature on extinction of drug- and withdrawal-paired cues. To provide background before turning to these issues, we begin by discussing the concept of drug administration as an associative learning episode, describing paradigms for studying the acquisition and expression of conditioned responses to drug- and withdrawal-paired cues in laboratory animals, and presenting some of the major findings on the neurobiological mechanisms of conditioned drug craving and withdrawal.
Craving and withdrawal elicited by drug-paired cues can be conceptualized as conditioned responses (CRs) that develop through a Pavlovian conditioning process, such that over the course of an addict’s drug use history, these cues become signals for impending drug use. Pavlov (1927) observed that dogs injected repeatedly with morphine or apomorphine exhibited salivation, restlessness, and vomiting upon exposure to what he called “the preliminaries of injection” (e.g., the appearance of the experimenter or being brought into the laboratory). He proposed that the situation was analogous to the conditioning of salivation to the sound of a metronome paired with food, such that the responses were preparatory responses to a pre-injection ritual that preceded and predicted the onset of drug effects.
That drug exposure engages an associative learning process is corroborated by decades of research on conditioned tolerance to the analgesic effects of morphine in rats. Rats administered morphine for the first time show analgesia as assessed by tail flick or hot plate tests, but this analgesia undergoes tolerance such that it diminishes with repeated morphine administrations. Siegel and colleagues (1976) discovered that this tolerance was context-specific, such that rats injected repeatedly with morphine in one context exhibited renewed analgesia when injected with morphine in a different context. This situational specificity suggests that tolerance is not just a physiological response to repeated drug administration, but that it also involves a learning component. Siegel proposed that the rats learned to associate morphine with the context in which it is administered, and as a function of this association, the context acquired the ability to elicit a conditioned physiological response to oppose the effects of the drug. Consistent with this, rats administered morphine repeatedly and then tested in the same context following an injection of saline exhibit hyperalgesia (Siegel, 1975). The hyperalgesic response is a type of CR called a conditioned compensatory response (CCR), defined as a conditioned homeostatic mechanism triggered by drug-paired cues that serves to maintain the system at or close to its normal set point in the presence of drugs that otherwise would perturb this balance (Siegel et al., 2000). CCRs are elicited by cues paired with a variety of drugs in addition to morphine, and show all of the features and effects associated with other types of Pavlovian CRs including external inhibition, latent inhibition, extinction, conditioned inhibition, overshadowing, blocking, and sensory preconditioning (for review see Siegel et al., 2000). CCRs are observed in humans as well and may contribute to incidences of overdose, particularly in people who are found to have what should be non-lethal quantities of drug in their system (Siegel et al., 1982). Theoretically, an experienced user who takes drugs in a new environment or otherwise in the absence of cues that normally accompanied drug administration may be vulnerable because no CCR is mounted to counteract the effects of the drug.
CCRs also are relevant to the maintenance of drug use in addicts because they can be unpleasant and involve physical and psychological signs of withdrawal (Childress et al., 1986; Kenny et al., 2006; Siegel and Ramos, 2002; Wheeler et al., 2008). These responses can motivate progressively more aggressive drug seeking because drug use is negatively reinforced by alleviating withdrawal (Kenny et al., 2006; Wikler, 1948) and because the drug is more rewarding when consumed in the presence of a withdrawal state (Hellemans et al., 2006b; Hutcheson et al., 2001; Robinson and Berridge, 1993).
The neurobiological mechanisms underlying the acquisition and expression of conditioned drug craving and withdrawal have been the subject of intensive study. Much of this work has been done using animal models, where the experimental conditions can be controlled with precision. Two paradigms are used most often in this type of research: the drug self-administration paradigm and the place conditioning paradigm. In both, initially neutral cues are paired with drug administration or acute withdrawal, but they differ in how conditioned responses to those cues are assessed.
In the drug self-administration paradigm, animals (nonhuman primates, rats, mice) learn to emit an operant response (typically a lever press) that is reinforced by an intravenous infusion of a drug of abuse (Figure 1). Drug cues are introduced into this paradigm in one of two ways. In the more common method (e.g., Grimm and See, 2000), a cue such as a multi-modal stimulus consisting of visual (light) and auditory (tone) components occurs simultaneously with drug infusions. That is, when an animal completes a required schedule of responding, the cue is triggered at the same time that the drug is administered. In the second method (e.g., Kruzich and See, 2001), Pavlovian cue-drug pairing sessions are conducted separately from drug self-administration sessions. Animals first acquire lever pressing, then in a separate session in which the lever is unavailable, they are presented several times with cues followed by drug administered by the experimenter. In both methods, following this initial training phase, lever pressing is extinguished over the course of one or more sessions in which lever presses produce neither drugs nor cues. The purpose of lever press extinction is to reduce the rate of lever pressing so as to be able to detect modulation of drug-seeking behavior during subsequent test sessions. During these tests, the cues are reintroduced, either by being presented periodically independent of the animal’s responding or by being made contingent on lever pressing. The typical outcome is an increase in the rate of lever pressing after exposure to the cues as compared to before. It is presumed that the cues enhance drug-seeking behavior by eliciting drug craving, and the renewed lever pressing is considered a model of relapse.
In humans, drug-paired cues elicit not only drug craving but also conditioned withdrawal, likely through different neural mechanisms (Robbins et al., 1997). To isolate the withdrawal component of the conditioned response, it is possible to use a modified version of the drug self-administration paradigm in which cues are paired not with drug administration but with drug abstinence. This method is not often used, but it may be valuable as a complement to studies of conditioned withdrawal in the place conditioning paradigm (discussed below in section II.A.2.). Establishing withdrawal-paired cues is accomplished most easily by using an opiate receptor antagonist such as naloxone to precipitate withdrawal in opiate-dependent animals. This has been done in two different ways, which, paradoxically, produce opposite results. In one approach (Goldberg et al., 1969; Kenny et al., 2006), animals are trained to self-administer an opiate over multiple sessions and in the process become dependent on the drug. In the case of opiates, dependence is reflected by well-characterized physical and motivational signs during periods of drug abstinence or, alternatively, antagonist-precipitated withdrawal (see Gracy et al., 2001; Koob and Le Moal, 2008). Withdrawal then is precipitated prior to self-administration sessions into which an extended cue (such as a continuous tone) is introduced. Importantly, the drug is available to the animal during these sessions under the same contingency that was in place during the initial training phase. In a test session conducted after the completion of training, the cue is present but the animal is not administered the opiate receptor antagonist. Similar to drug-paired cues, withdrawal-paired cues trained in this way increase drug-seeking behavior (Goldberg et al., 1969; Kenny et al., 2006). The second approach involves pairing a cue with precipitated withdrawal in Pavlovian conditioning sessions conducted separately from self-administration sessions (Hellemans et al., 2006b). When trained in this way, a withdrawal-paired cue actually decreases drug-seeking behavior in test (Hellemans et al., 2006b). One possible explanation for the discrepancy is that the opiate is more rewarding in the first approach because it is self-administered during withdrawal (Robinson and Berridge, 1993), giving it both positive- and negative-reinforcing effects that can overcome a general behavioral suppressive effect of conditioned withdrawal (Hellemans et al., 2006b).
The self-administration paradigm is advantageous in that it models human drug-taking and drug-seeking behavior very closely. However, it is difficult and time-consuming because of the need to surgically implant intravenous catheters, maintain the patency of those catheters over time, and train animals to self-administer drug over multiple training sessions. Because of the time and degree of expertise required to do these kinds of experiments, many labs opt for a simpler alternative: place conditioning.
Contrary to the drug self-administration paradigm, in the place conditioning paradigm administration of the drug is under the control of the experimenter rather than the subject, and measurement of the conditioned motivational effects of drug- or withdrawal-paired cues is indirect. Place conditioning most commonly involves a 2- or 3-chambered apparatus in which the chambers are distinguished by wall color, floor texture, or both. Training proceeds in several stages. First, animals (generally rats or mice) are given a pre-test in which they are exposed to the apparatus for a short period of time (10–30 min) and their exploration of the chambers is recorded. Typically the apparatus is configured such that a population of subjects show no inherent preference for one or the other chamber; individual animals that show a strong bias can be eliminated from the experiment on the basis of the pre-test, before time- or labor-intensive aspects of the studies begin. Training involves administering a drug or precipitating drug withdrawal and then confining an animal in one of the chambers for a period of time (usually 30–60 min). Animals also are confined in the other chamber for the same amount of time following an injection of saline to equate exposure to the two contexts. The number and spacing of confinements varies; for example, with precipitated withdrawal, a single context-withdrawal pairing generally is sufficient, whereas with cocaine, multiple context-drug pairings typically are required. At least 24 hrs after the completion of training, animals once again are given the opportunity to explore the apparatus freely for the same 10–30 min period as was done in the pre-test. The post-test almost always is conducted while the animals are in a drug-free state. Animals trained in this way come to prefer a drug-paired compartment or avoid a withdrawal-paired compartment, where preference or avoidance is measured in terms of time spent in each compartment in the post-test. Animals showing conditioned place preference (CPP) spend more time in the drug-paired compartment than in the saline-paired compartment, whereas animals showing conditioned place avoidance (CPA) spend less time in the withdrawal-paired compartment than in the saline-paired compartment. The change in time spent in the paired compartment from pre-test to post-test is taken as a measure of the motivational effects of cues (environments) paired with drugs or withdrawal – motivational effects that in humans lead to drug-seeking behavior.
Although it might be proposed that the place conditioning paradigm and the drug self-administration paradigm are different ways of modeling the same process, several caveats are important. First, in place conditioning, the drug- or withdrawal-paired cues are contexts, whereas in the drug self-administration paradigm, discrete cues such as lights and tones are employed. This is significant because contextual cues and discrete cues engage somewhat divergent neural mechanisms (Kim and Fanselow, 1992). Second, place conditioning involves passive drug administration, which engages different neural mechanisms than does self-administration (Jacobs et al., 2003). Finally, indirect measures of the motivational effects of drug cues as they relate to relapse (as in the place conditioning paradigm) may not map perfectly onto direct measures of drug-seeking behavior (as in the drug self-administration paradigm). Nevertheless, despite their fundamental differences, the two paradigms produce concordant results the large majority of the time (Bardo and Bevins, 2000).
It is important to note that drug cues can elicit conditioned responses or modulate behavior through more than one mechanism. As described above, drug- or withdrawal-paired cues can be Pavlovian conditioned stimuli (CSs), defined as elements of a contingency in which CS presentation predicts either drug administration or acute withdrawal (Figure 2). However, drug-paired cues also can serve as discriminative stimuli or occasion setters, which signify that a contingency between other events is in effect. A discriminative stimulus (SD) signals that a response (operant) will be followed by an outcome (reinforcer), and an occasion setter signals that a different cue (a CS) will be followed by an outcome (an unconditioned stimulus or US). Sometimes it can be difficult to make these distinctions, particularly with people in real-life situations, and it is possible under some circumstances for a cue to serve more than one role; however, in animal models the roles played by drug- or withdrawal-paired cues generally are clear-cut. CSs can be discrete cues (e.g., a light, a tone, or the sound of a syringe pump) triggered at the same time as drug self-administration, or environments paired with passive drug administration or precipitated withdrawal. An example of an SD would be a signal such as houselight illumination, the presence of which indicates that lever pressing will lead to drug administration, and the absence of which indicates that lever pressing will have no programmed consequence. An example of an occasion setter would be a context in which Pavlovian pairings are conducted; if an animal receives cue-drug pairings in context A but cue alone (without drug) in context B, the significance of the cue will depend on the context in which it is presented, and the ability of that cue to elicit conditioned responses likewise will be modulated by the context in which a test occurs.
It is important to maintain these distinctions because CSs, SDs, and occasion setters have different behavioral features and almost certainly rely on different neurobiological mechanisms. For example, CSs extinguish readily when the CS is no longer followed by the US, whereas occasion setters are very resistant to extinction even following extensive exposure to the occasion setter in the absence of the US (Bouton and Swartzentruber, 1986; Rescorla, 1986). Because our major interest is in extinction of Pavlovian conditioned responses, in this review we focus on drug- and withdrawal-paired cues as CSs.
There has been considerable interest in determining the neurobiological underpinnings of conditioned drug craving and withdrawal. A full review of this literature is beyond the scope of this paper, but because some familiarity with it is essential for understanding the evolving literature on the neurobiology of drug cue extinction, in this section we describe some of the most significant findings.
Several studies have used brain imaging techniques including positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) to examine brain activation in addicts in response to drug-paired cues. Even though the studies vary in terms of the participants’ drug of choice (cocaine, heroin, crack cocaine, alcohol, or nicotine) and the manner of presenting drug-paired cues (viewing images or videos, handling paraphernalia, engaging in script-guided imagery, or some combination of these), the results have been remarkably consistent. Most report activation of prefrontal cortical areas, particularly anterior cingulate (Brody et al., 2002; Childress et al., 1999; Grusser et al., 2004; Kilts et al., 2004; Kilts et al., 2001; Maas et al., 1998; McBride et al., 2006; McClernon et al., 2007; Sell et al., 1999; Wexler et al., 2001; Wilson et al., 2005), but also dorsolateral prefrontal cortex (Brody et al., 2002; Grant et al., 1996; Maas et al., 1998; Sell et al., 1999; Wilson et al., 2005; Yang et al., 2009) and orbitofrontal cortex (Bonson et al., 2002; Childress et al., 1999; Grant et al., 1996; Kilts et al., 2001; McClernon et al., 2007; Yang et al., 2009). The prefrontal cortex is involved in response inhibition, decision making, and emotional regulation, among other functions, and is highly interconnected with limbic and reward circuitries (Miller and Cohen, 2001). Somewhat less consistent are reports of activation of the amygdala (Bonson et al., 2002; Childress et al., 1999; Grant et al., 1996; Kilts et al., 2001; McClernon et al., 2007; Yang et al., 2009) and the ventral striatum (Braus et al., 2001; Kareken et al., 2004; Sell et al., 1999), as might be expected from the emotionally charged and motivationally significant nature of drug cues. In several studies, activity in one or more of these areas correlated significantly with subjective reports of drug craving (Bonson et al., 2002; Brody et al., 2002; Grant et al., 1996; Kilts et al., 2004; Kilts et al., 2001; Maas et al., 1998; Sell et al., 2000; Smolka et al., 2006).
Studies in laboratory animals have examined both acquisition and expression of conditioned responses to drug- and withdrawal-paired cues. Here we focus on expression rather than acquisition, because in many cases acquisition studies are complicated by the possibility that an experimental manipulation could modulate the intensity of acute drug effects or acute withdrawal as opposed to modulating learning about the relationship between cues and the drug or withdrawal state. In other words, it can be difficult to determine whether an apparent impairment or facilitation of learning about drug- or withdrawal-paired cues is a secondary effect of increasing or decreasing the salience of the unconditioned stimulus (US) rather than an effect on the associative learning process per se.
Consistent with the human imaging literature, studies using Fos or other immediate-early genes as markers of neuronal activity have reported activation of cortical, limbic, and reward circuitries following exposure to drug-paired cues. In most of these studies, rats were exposed to pairings of an environment with a drug administered by the experimenter, then some later time were re-exposed to the environment in the absence of the drug. Activated prefrontal areas include anterior cingulate (Brown et al., 1992; Franklin and Druhan, 2000; Honsberger and Leri, 2008; Neisewander et al., 2000; Schmidt et al., 2005; Schroeder et al., 2000; Schroeder and Kelley, 2002; Zavala et al., 2008), prelimbic cortex (Franklin and Druhan, 2000; Miller and Marshall, 2004, 2005b; Schmidt et al., 2005; Schroeder et al., 2000; Schroeder and Kelley, 2002; Zavala et al., 2008), infralimbic cortex (Franklin and Druhan, 2000; Schroeder et al., 2000; Schroeder and Kelley, 2002) (but see Miller and Marshall, 2004, 2005b), claustrum (Brown et al., 1992; Franklin and Druhan, 2000), and orbitofrontal cortex (Koya et al., 2006; Zavala et al., 2008). Activation of the basolateral amygdala (BLA) has been reported by several groups (Baker et al., 1999; Brown et al., 1992; Franklin and Druhan, 2000; Miller and Marshall, 2004, 2005b; Neisewander et al., 2000) but not all (Schroeder et al., 2000; Schroeder and Kelley, 2002; Zavala et al., 2008), whereas conditioned activation of the central amygdala (CeA) has not been observed (Baker et al., 1999; Franklin and Druhan, 2000; Miller and Marshall, 2004, 2005b; Neisewander et al., 2000; Schroeder et al., 2000; Schroeder and Kelley, 2002; Zavala et al., 2008). Finally, activation of the nucleus accumbens (NAcc), including core, shell, or both, is a common observation (Franklin and Druhan, 2000; Koya et al., 2006; Miller and Marshall, 2004; Neisewander et al., 2000; Schmidt et al., 2005; Schroeder et al., 2000; Schroeder and Kelley, 2002) (but see Brown et al., 1992).
The contribution of each of these areas to the expression of conditioned responses to drug cues has been investigated further using lesion, inactivation, and pharmacological techniques. Because of the central role it plays in responding to drugs of abuse, NAcc has been the focus of many of these studies. Single units in NAcc respond selectively with increased firing rates to cocaine-paired cues (Carelli, 2000; Carelli and Deadwyler, 1996). Inactivation of NAcc core (but not shell) blocks cue-induced reinstatement of cocaine-seeking (Di Ciano et al., 2008; Fuchs et al., 2004; Ito et al., 2004) and heroin-seeking (Rogers et al., 2008; Zhou et al., 2007); similarly, infusion of a mitogen-activated protein kinase (MAPK) inhibitor into NAcc core blocks expression of cocaine CPP (Miller and Marshall, 2005a) (but see Gremel and Cunningham, 2008). These effects are mediated at least in part by glutamate release within NAcc, as cue-induced reinstatement of heroin seeking is associated with increased glutamate release in NAcc core (Hotsenpiller et al., 2001; LaLumiere and Kalivas, 2008), and targeted infusion of AMPA receptor antagonists into NAcc core blunts cue-induced reinstatement of cocaine- and heroin-seeking (Backstrom and Hyytia, 2007; Di Ciano and Everitt, 2001, 2004; LaLumiere and Kalivas, 2008) and expression of cocaine and ethanol CPP (Gremel and Cunningham, 2009; Kaddis et al., 1995). By contrast, the role of dopamine release within NAcc is controversial. Increased dopamine release following exposure to drug-paired cues has been reported by some (Di Ciano et al., 1998; Fontana et al., 1993; Gerasimov et al., 2001; Ito et al., 2000; Phillips et al., 2003) but not all investigators (Brown and Fibiger, 1992). Likewise, impairment of cue-induced reinstatement of drug-seeking or expression of CPP following localized infusions of dopamine receptor antagonists has been shown in some studies (Bossert et al., 2007; Hiroi and White, 1990, 1991b; Kaddis et al., 1995; LaLumiere and Kalivas, 2008) but not others (Di Ciano and Everitt, 2004; Gremel and Cunningham, 2009).
The dorsomedial prefrontal cortex (dmPFC; anterior cingulate and prelimbic cortices) is a major source of glutamatergic input to NAcc (Christie et al., 1985) and has been implicated in relapse to drug-seeking by priming drug administrations and stress (McFarland et al., 2003, 2004; McFarland and Kalivas, 2001). This area seems to play a critical role in responding to drug cues as well, as dmPFC lesions or inactivation diminish cue-induced reinstatement of cocaine-seeking (Di Ciano et al., 2007; Di Pietro et al., 2006; McLaughlin and See, 2003; Weissenborn et al., 1997) and heroin-seeking (Rogers et al., 2008; Van den Oever et al. 2008; but see Schmidt et al. 2005). Dopamine is released within dmPFC following exposure to a previously methamphetamine-paired context (Lin et al., 2007), and targeted administration of a D1 dopamine receptor antagonist into prelimbic cortex blocks cue-induced reinstatement of heroin-seeking (See, 2009).
The amygdala is another major source of glutamatergic input to NAcc (Christie et al., 1987) and is reciprocally connected with the dorsomedial PFC (Llamas et al., 1977; Room et al., 1985). There is a great deal of evidence that the amygdala, and particularly the basolateral complex of the amygdala (BLA; lateral and basal nuclei), is required for expression of conditioned responses to drug-paired cues. BLA lesions or inactivation attenuate cue-induced reinstatement of cocaine-seeking (Kruzich and See, 2001; McLaughlin and See, 2003; Meil and See, 1997; Whitelaw et al., 1996) (but see Fuchs et al., 2002) and heroin-seeking (Fuchs and See, 2002; Grimm and See, 2000; Rogers et al., 2008) and can impair expression of amphetamine and ethanol CPP (Gremel and Cunningham, 2008; Hiroi and White, 1991a). Conversely, administration of D-amphetamine into BLA potentiates cue-induced reinstatement of cocaine-seeking (Ledford et al., 2003). Dopamine, but not glutamate, release within BLA seems to be critical, as targeted infusions of dopamine receptor antagonists into BLA block cue-induced reinstatement of cocaine-seeking (Di Ciano and Everitt, 2004; See et al., 2001) and expression of ethanol CPP (Gremel and Cunningham, 2009), whereas intra-BLA infusions of AMPA and/or NMDA glutamate receptor antagonists have no effect on cue-induced reinstatement of cocaine-seeking (Di Ciano and Everitt, 2004; See et al., 2001). BLA single units respond with increased firing rates to cocaine-paired cues (Carelli et al., 2003), suggesting that amygdalar input may contribute to the similar cocaine cue-responsivity of single units in NAcc. Serial processing of drug cue-related information from BLA to NAcc is supported by the finding that “disconnection” of these two structures via unilateral infusion of a dopamine receptor antagonist into BLA and contralateral infusion of an AMPA receptor antagonist into NAcc core impairs the ability of a cocaine-paired cue to maintain cocaine-seeking behavior under an extended chain schedule of reinforcement (Di Ciano and Everitt, 2004).
The literature on the neurobiological mechanisms underlying the expression of conditioned withdrawal is relatively modest, but some important information is available. Frenois and colleagues (2005; Lucas et al., 2008) used Fos expression to map brain activation patterns in rats undergoing acute (precipitated) opiate withdrawal or conditioned withdrawal elicited by re-exposure to a previously withdrawal-paired context. Interestingly, largely overlapping circuitries were engaged: acute withdrawal was associated with activation of the extended amygdala (CeA, bed nucleus of the stria terminalis or BNST, and NAcc shell) and some of its major afferents (VTA and hippocampus), as well as a number of brainstem structures and nuclei (periaqueductal gray, nucleus of the solitary tract, and locus coeruleus or LC). Likewise, conditioned withdrawal was associated with activation of elements of extended amygdala (BNST and NAcc shell, but not CeA), as well as BLA, regions of hippocampus, VTA, and LC. BLA and the extended amygdala have been implicated in fear, anxiety, stress, and aversive motivation (Cardinal et al., 2002; Walker et al., 2003), and NAcc is a site of neuroplastic changes occurring during the development of drug dependence that are associated with dysphoria and anhedonia (Carlezon et al., 1998). Recruitment of this “anti-reward” system (Koob and Le Moal, 2005) is a hallmark of the progression of addiction (Koob and Le Moal, 1997) and may be responsible for the experience of withdrawal (acute or conditioned) as a stressful, anxiogenic, dysphorogenic state that triggers an intense desire to take drugs.
Despite activating similar circuitries, acute withdrawal and conditioned withdrawal are not identical mechanistically. The adrenergic a2 receptor antagonist clonidine blocks both acute withdrawal symptoms and withdrawal-associated activation of LC (Aghajanian, 1978; Tseng et al., 1975), but clonidine does not block conditioned withdrawal, as it has no effect on suppression of bar pressing by a previously withdrawal-paired CS and actually enhances the expression of CPA to a previously withdrawal-paired environment (Schulteis et al., 1998).
In two studies the role of the amygdala in mediating conditioned withdrawal has been examined, and consistent with the literature on conditioned craving, this region seems to play an important role. In one study (Hellemans et al. 2006a), intra-BLA infusions of antisense oligonucleotides to Zif268 (an immediate early gene) prior to a brief re-exposure to a previously withdrawal-paired CS impaired later expression of conditioned withdrawal (measured as suppression of bar pressing) to that same CS, presumably by blocking memory reconsolidation. In the other study (Heinrichs et al. 1995), destruction of CeA corticotropin-releasing factor (CRF) – containing neurons by localized infusion of immunotoxins attenuated the expression of conditioned withdrawal as assessed via response suppression in the presence of a previously withdrawal-paired CS.
Conditioned responses (CRs) to drug- and withdrawal-paired cues, like all other types of Pavlovian conditioned responses, can be reduced through extinction. Extinction involves repeated presentations of a CS in the absence of the unconditioned stimulus (US) with which it was paired previously. The outcome of exposure to this type of training (“extinction training”) is a decline in the magnitude and/or frequency of the CR (“extinction”). Because there is considerable evidence that they are mediated by different mechanisms, a distinction is made between within-session extinction, defined as extinction occurring during extinction training, and extinction retention, defined as extinction memory assessed at some point after the completion of extinction training (typically at least 24 hrs).
In the drug self-administration paradigm, cue extinction training involves conducting several test sessions (after the completion of lever press extinction) in which no drug is administered but the cues are present (Figure 1). Extinction of the response-reinstating ability of the cues occurs over the course of the test sessions and is measured as a lessening of their ability to modulate drug-seeking behavior. It is important to maintain the distinction between operant (i.e., lever press) extinction and cue extinction, which are different processes that likely engage different neural mechanisms. In this review we are concerned with cue extinction and do not discuss lever press extinction in any detail.
In the place conditioning paradigm, extinction training can occur in either of two ways: animals can be given free access to the place conditioning apparatus in repeated test sessions, or animals can be confined in the formerly drug- or withdrawal-paired context in the absence of drug administration or precipitated withdrawal (i.e., following an injection of saline) and subsequently given free access tests to assess extinction. When extinction has occurred, there no longer is a preference for or aversion to the previously drug- or withdrawal-paired context; that is, animals spend approximately equal amounts of time in each.
Extinction is not due to forgetting or “unlearning” of the significance of the CS. In most paradigms Pavlovian conditioned memories are long-lasting and resistant to forgetting, such that without exposure to the CS in the absence of the US, the conditioned response does not disappear (Quirk, 2002; Stinus et al. 2000). On the other hand, extinction memory is fragile in that extinguished CRs are subject to recovery under several circumstances. For example, extinction typically is expressed selectively within the context in which extinction training occurred (the renewal effect) (Bouton and Bolles, 1979). Likewise, extinguished CRs re-emerge over time following extinction training, an effect called spontaneous recovery (Pavlov, 1927). These effects indicate that extinction does not undo original learning, but is itself a form of new learning in which the CR comes to be suppressed in a context- and time-dependent manner (Bouton, 1993).
Various types of CRs to drug-paired cues undergo renewal and spontaneous recovery. In animals, spontaneous recovery of tolerance to the analgesic effects of morphine has been observed (Millin and Riccio, 2002). Renewal of cocaine and morphine CPP in a variant of the CPP paradigm in which a distinctive floor texture is the drug-paired cue (Parker et al., 2006), and of nose poking for nicotine- and ethanol- paired cues (Chaudhri et al., 2008; Diergaarde et al., 2008), has also been reported. Interestingly, and similar to other types of CRs (Chelonis et al., 1999; Gunther et al., 1998; but see Bouton et al., 2006), renewal of CRs to drug-paired cues is attenuated when extinction training is conducted in multiple contexts (Chaudhri et al., 2008). Human studies have been less consistent, with two groups reporting renewal of craving induced by nicotine-and alcohol-paired cues in smokers and moderate to heavy social drinkers, respectively (Collins and Brandon, 2002; Thewissen et al., 2006), and two groups finding no evidence of renewal of craving or salivation elicited by alcohol-paired cues in heavy social drinkers and alcoholics (MacKillop and Lisman, 2008; Stasiewicz et al., 2007). Both renewal (Collins and Brandon, 2002) and spontaneous recovery (Brooks et al., 2004) of CRs to drug-paired cues are attenuated when an extinction cue, defined as an incidental cue that is present during extinction training, is present during testing as well, similar to other forms of extinction (Brooks and Bouton, 1993, 1994; Brooks et al., 1999).
Extinction can be employed in a clinical context as part of treatment programs for psychiatric conditions that involve a significant learned component. For example, post-traumatic stress disorder (PTSD) is characterized by extreme fearfulness, intrusive thoughts, nightmares, and increased arousal, which develop following exposure to a severe stressor. Many of these responses are triggered by cues that were present at the time of the stressor, suggesting that they acquired significance through a Pavlovian conditioning mechanism in which the stressor acted as a US. One form of PTSD treatment is exposure therapy, an extinction-based protocol in which the patient is exposed to fear-triggering cues in a safe environment until fear responses decline (Rothbaum and Davis, 2003). A challenge inherent in using extinction clinically is overcoming its context dependence and general fragility, but techniques are being developed to make extinction memory stronger and more permanent (Walker et al., 2002). Some of these techniques have been applied successfully in clinical studies (Ressler et al., 2004).
As described above, addiction also involves a learning component, and as such it may be amenable to extinction-based interventions. Typically, however, addiction treatment programs emphasize avoidance of drug cues rather than nonreinforced exposure to them, as recovering addicts are advised not to return to drug-paired environments or associate with people with whom they consumed drugs. This is an effective strategy to the extent that avoidance is possible, but it is not optimal because there often are instances of unintended exposure to cues that once signaled drug procurement or use (e.g., money). A better approach would be to reduce the ability of drug cues to provoke craving and withdrawal so that if they are encountered they will be less powerful.
Several groups have examined the utility of exposing drug users to drug cues in the absence of drug effects. These studies differ in a number of respects, including the characteristics of the participants, their drug of choice (opiates, cocaine, nicotine, or alcohol), extent of drug use (addict, heavy user, or user looking to moderate intake), and inpatient or outpatient status. Different types of cues and methods of presenting them have been employed, including images of drugs or people using drugs; videos of people preparing or using drugs; audio tapes of drug deals; and “in vivo” exposure involving handling of drugs or drug paraphernalia, preparing to self-administer drugs, and in some cases, self-administering either saline or a drug itself with a drug antagonist “on board” to block physiological drug effects. A few studies have involved virtual reality technology to simulate environments associated with drug use, such as “crack houses” (Saladin et al., 2006) or bars (Lee et al., 2007). The details of the exposure protocol, including the number and duration of exposures and number and spacing of exposure sessions, vary widely; moreover, in some studies cue exposure was conducted in conjunction with other types of therapy, such as cognitive-behavioral or relaxation therapy, whereas in others cue exposure was used alone. Response measures include subjective reports of craving, withdrawal, and emotional states; physiological or autonomic responses such as salivation or skin conductance; and outcome measures such as relapse rates. Finally, some but not all studies include untreated or alternatively treated control groups and/or follow-up assessments after the completion of therapy.
Overall, the results have been disappointing. There have been some reports of reductions in craving and withdrawal following drug cue exposure, but the effects tend to be modest and time-limited (Childress et al., 1986; Drummond and Glautier, 1994; Lee et al., 2004; Lee et al., 2007; Loeber et al., 2006; Monti et al., 1993; Monti et al., 2001; O’Brien et al., 1990; Powell et al., 1993; Rohsenow et al., 2001; Sitharthan et al., 1997). Several studies observed no benefit (Corty and McFall, 1984; Dawe et al., 1993; Kasvikis et al., 1991; Kavanagh et al., 2006; McCusker and Brown, 1995; Niaura et al., 1999; Raw and Russell, 1980) and one reported detrimental effects in terms of increased dropout and relapse among participants receiving drug cue exposure (Marissen et al., 2007). A meta-analysis of the literature through 2001 revealed no statistically significant benefit of drug cue exposure therapy (Conklin and Tiffany, 2002). There are some indications that cue exposure may be more efficacious with certain kinds of drugs than with others; for example, alcohol users seem to respond particularly well, as many of the studies reporting positive effects involved exposure to alcohol cues in alcoholics or heavy drinkers looking to moderate intake (Drummond and Glautier, 1994; Lee et al., 2007; Loeber et al., 2006; Monti et al., 1993; Monti et al., 2001; Rohsenow et al., 2001; Sitharthan et al., 1997). In this regard it is potentially interesting that the two failures to observe renewal in cue exposure studies both involved alcohol-paired cues (MacKillop and Lisman, 2008; Stasiewicz et al., 2007), suggesting that extinction of CRs to alcohol cues may generalize across contexts particularly well. Nevertheless, collectively these studies fail to substantiate the efficacy of drug cue exposure therapy.
It is unclear why the results of these studies have been so discouraging thus far. Several authors speculate that incorporating advances in the understanding of extinction in laboratory animals would improve the outcome (Conklin, 2006; Conklin and Tiffany, 2002; Havermans and Jansen, 2003; Powell, 1995; Siegel and Ramos, 2002). For example, since renewal is attenuated following extinction in multiple contexts, exposure therapy conducted in multiple environments might promote generalization of extinction (Conklin and Tiffany, 2002; Havermans and Jansen, 2003; Siegel and Ramos, 2002). Likewise, since both renewal and spontaneous recovery are lessened in the presence of extinction cues, incorporating a portable token that can be taken home (such as a distinctive coin or rock) into cue exposure sessions might be beneficial (Conklin and Tiffany, 2002; Havermans and Jansen, 2003; Siegel and Ramos, 2002). It also has been proposed that discrete, tangible drug cues such as drug paraphernalia represent only a subset of cues associated with drug use and may not be the most salient ones. Interoceptive cues related to drug self-administration and initial drug effects are powerful, but with few exceptions (e.g., O’Brien et al., 1984) they have not been incorporated into exposure therapy sessions (Siegel and Ramos, 2002). Similarly, the perception of drug availability is itself a drug cue, and its absence during cue exposure sessions may affect both the degree to which other drug-paired cues elicit craving and withdrawal responses (Droungas et al., 1995; McBride et al., 2006; Thewissen et al., 2005; Wilson et al., 2005) and the generalization of extinction to environments in which drugs are accessible (Powell, 1995). Finally, since drug users are often exposed to hundreds if not thousands of pairings of cues with drug effects over the course of their addiction, successful cue exposure therapy may require many exposures to extinguish entrenched associations (Conklin and Tiffany, 2002).
Another possibility is that if the neural underpinnings of drug cue extinction were clearer then it might be more evident why current drug cue exposure protocols have had limited success. In other contexts, basic neuroscience research has led to more efficacious clinical interventions for psychiatric conditions; analogous studies of the neurobiology of drug cue extinction might have a similar impact.
As yet little is known about the neurobiological mechanisms of extinction of conditioned drug craving and withdrawal. There have been few systematic investigations; in fact, some of the relevant findings were incidental observations in studies designed to characterize other processes. Different groups have employed cues paired with different drugs or classes of drugs, which may engage different mechanisms, and the vast majority of the data comes from models of conditioned drug craving, such that almost nothing is known about extinction of conditioned withdrawal. Clearly, there is much work to be done in this area, although the topic is attracting increased interest.
Without any theoretical framework within which to organize this fragmentary literature, it can be difficult to piece together a cohesive picture or apply what has been learned clinically. However, there is relevant information available from another line of research. Extinction of a different type of Pavlovian conditioned response – conditioned fear – has been the subject of intensive study for a number of years and a great deal has been learned about its underlying neural circuitry, neuropharmacology, and molecular mechanisms. At the core of the fear extinction process is a recruitment of circuitry in the ventromedial prefrontal cortex that inhibits conditioned fear responses arising from limbic structures, particularly the amygdala. Accumulating evidence indicates that this same basic process is engaged by extinction of other kinds of Pavlovian CRs (Akirav et al., 2006) and operant responses (LaLumiere et al., 2008a) as well.
In many cases parallels can be drawn between observations in the fear and drug cue extinction literatures, suggesting that extinction of conditioned drug craving and withdrawal may be mediated by a similar cortico-limbic mechanism. For this reason, It is helpful to describe some of the major features of fear extinction before turning to a discussion of the neurobiology of extinction of drug- and withdrawal-paired cues. This will facilitate use of the fear extinction literature as a template to organize the observations on drug cue extinction into a cohesive framework.
The fear extinction literature is too large to be reviewed in its entirety, and interested readers are referred to other sources (Myers and Davis, 2007; Quirk and Mueller, 2008). Here we emphasize major substrates and mechanisms, focusing in particular on those that may be shared with extinction of conditioned drug craving and withdrawal.
In the fear conditioning paradigm, an organism (often a rat or mouse) is exposed to pairings of an initially neutral cue, such as a light, tone, or context, with an aversive stimulus such as foot shock, and comes to exhibit conditioned fear responses in the presence of that cue. In rodents fear is defined operationally in any of several ways; the most common measures are freezing (cessation of all movements except those required for respiration) and fear-potentiated startle (an increase in the amplitude of the acoustic startle response in the presence of a fear-eliciting cue vs. in its absence). Extinction of conditioned fear occurs readily when the CS is presented repeatedly in the absence of the aversive US.
The neurobiology of conditioned fear acquisition has been studied extensively. The amygdala in particular is critical: sensory information about the CS and US is transmitted via cortical and subcortical pathways to BLA, where essential plasticity underlying the formation of the CS-US association occurs. BLA projects to CeA, which in turn innervates hypothalamic and brainstem regions responsible for triggering autonomic and behavioral fear responses (for reviews see Davis and Shi, 2000; Rodrigues et al., 2004). Fear extinction involves this circuitry as well, but also recruits additional structures, particularly the medial prefrontal cortex (mPFC). Consistent with its general role in emotional regulation and behavioral inhibition, in extinction mPFC plays an active role in inhibiting conditioned responses by suppressing neural activity within the amygdala. Both the amygdala and mPFC are sites of essential synaptic plasticity underlying extinction, including NMDA receptor-mediated forms of plasticity, although each structure plays a somewhat different role in the encoding, consolidation, and expression of extinction memory. In what follows we describe what has been learned about each of these components in more detail.
Evidence for a role of the amygdala in fear extinction is extensive. The amygdala is activated during fear extinction training (Herry and Mons, 2004; LaBar et al., 1998; Phelps et al., 2004). Individual BLA neurons fire selectively to extinguished cues both as extinction training progresses (Herry et al., 2008) and after extinction training is complete, in a context-dependent manner (Herry et al., 2008; Hobin et al., 2003). Pre-extinction training inactivation of the basal nucleus (Herry et al., 2008) or post-extinction training lesions of the intercalated cell masses (ICMs) lying between BLA and CeA (Likhtik et al., 2008) impair within-session extinction and extinction retention, respectively. Both encoding and consolidation of extinction memory can be modulated by intra-BLA infusions of a wide variety of drugs including neurotransmitter receptor agonists and antagonists (Berlau and McGaugh, 2006; Falls et al., 1992; Harris and Westbrook, 1998; Hikind and Maroun, 2008; Kim et al., 2007a; Laurent and Westbrook, 2008; Lee and Kim, 1998; Roche et al., 2007; Yang et al., 2006) as well as modulators of downstream second messengers, transcription, and translation (Lin et al., 2003c; Lin et al., 2003b; Lu et al., 2001). The phosphorylation state of second messengers and gene expression patterns within BLA are altered following extinction training (Cannich et al., 2004; Chhatwal et al., 2005b; Chhatwal et al., 2006; Heldt and Ressler, 2007; Herry and Mons, 2004; Lin et al., 2003c; Lin et al., 2003b), and manipulation of gene expression within BLA prior to extinction training modulates extinction memory consolidation (Chhatwal et al., 2006).
The contribution of GABAergic neurotransmission within the amygdala has been a topic of particular interest. Based on findings of recovery of fear CRs following extinction (e.g., the renewal effect), it has been proposed that amygdalar plasticity underlying fear memory remains at least partially intact following extinction but that the expression of that memory is suppressed. One way that this might occur is via recruitment of GABAergic neurotransmission within the amygdala to dampen amygdalar throughput (Davis and Myers, 2002) (for a different view, see Dalton et al., 2008; Kim et al., 2007b; Lin et al., 2003a; Mao et al., 2008). There is an extensive network of GABAergic interneurons within BLA (Muller et al., 2003; Rainnie et al., 1991), as well as GABAergic ICMs lying between BLA and CeA that gate impulse traffic from one to the other (Royer et al., 1999), and either or both of these could be critical (Jungling et al., 2008; Rosenkranz and Grace, 2001; Rosenkranz et al., 2003; Royer and Pare, 2002). Straightforward pharmacological manipulations of GABAergic tone in the amygdala are problematic due to the potential for either sedation or epileptogenesis; however, it has been shown that extinction induces changes in GABA-related genes and modulates GABAA receptor binding in BLA (Chhatwal et al., 2005b; Heldt and Ressler, 2007), and that selective lesions of ICMs impair extinction retention (Likhtik et al., 2008). Hence, modulation of the activity of GABAergic systems within BLA likely plays a key role in extinction of conditioned fear.
The medial prefrontal cortex (mPFC) is a major source of input to amygdalar GABAergic neurons. Stimulation of mPFC induces firing in BLA interneurons (Rosenkranz and Grace, 2001) and inhibits conditioned responses of BLA principal neurons to a previously fear conditioned cue (Rosenkranz et al., 2003); it also induces Fos expression in ICMs (Berretta et al., 2005) and blunts responses of CeA neurons triggered by stimulation of BLA, presumably via activation of ICMs (Quirk et al., 2003). Hence, mPFC is in a position to exert inhibitory control over the amygdala. Consistent with this, there is a great deal of evidence indicating that it plays a central role in extinction.
The first studies implicating mPFC in fear extinction demonstrated that pre-training, electrolytic lesions of this region produce an impairment (Morgan and LeDoux, 1995; Morgan et al., 1993). Later studies employing more focal lesions (Quirk et al., 2000) or localized inactivation (Sierra-Mercado et al., 2006) identified a prominent role for the ventral medial or infralimbic (IL) region of mPFC, specifically. Medial PFC is activated during fear extinction training (Barrett et al., 2003; Burgos-Robles et al., 2007; Herry and Mons, 2004; Kalisch et al., 2006; Knapska and Maren, 2009; Milad et al., 2007; Phelps et al., 2004; Santini et al., 2004; Santini et al., 2008), and subnormal mPFC activation is associated with poor extinction retention (Burgos-Robles et al., 2007) and is seen in disorders such as PTSD that are thought to involve an extinction deficit (Bremner et al., 2005). Medial PFC thickness is associated positively with retention of extinction memory (Milad et al., 2005), whereas manipulations that adversely affect mPFC circuitry, such as acute or chronic stress and associated dendritic retraction, impair extinction retention (Izquierdo et al., 2006; Miracle et al., 2006).
Medial PFC seems to be involved specifically in consolidation and expression of extinction memory. Medial PFC lesions impair extinction retention while sparing within-session extinction (Quirk et al., 2000). Single units within IL fire selectively to presentations of a previously fear-conditioned cue during an extinction retention test 24 hrs after extinction training but not during the extinction training session itself (Milad and Quirk, 2002). When IL microstimulation is paired with presentations of a previously fear conditioned cue in nonextinguished animals, freezing to those cues is attenuated (Milad and Quirk, 2002; Milad et al., 2004), consistent with the putative role of this region in the expression of extinction. Pre-extinction training, intra-mPFC or intra-IL infusions of neurotransmitter receptor agonists or antagonists (Burgos-Robles et al., 2007; Hikind and Maroun, 2008; Laurent and Westbrook, 2008; Mueller et al., 2008; Pfeiffer and Fendt, 2006; Sotres-Bayon et al., 2009; Zushida et al., 2007), as well as modulators of downstream second messengers, transcription, and translation (Hugues et al., 2006; Hugues et al., 2004; Mueller et al., 2008; Santini et al., 2004), modulate extinction retention without affecting within-session extinction. Immediate post-extinction training administration of many of these agents has a similar effect.
Like other forms of learning, extinction involves NMDA receptor-mediated synaptic plasticity. Studies involving systemic administration of NMDA receptor antagonists prior to extinction training report dose-dependent impairments of both within-session extinction and extinction retention (Baker and Azorlosa, 1996; Dalton et al., 2008; Santini et al., 2001; Sotres-Bayon et al., 2007; Sotres-Bayon et al., 2009). Systemic NMDA receptor antagonists also impair extinction retention when administered immediately after extinction training (Burgos-Robles et al., 2007; Laurent and Westbrook, 2008; Santini et al., 2004), indicating that NMDA receptors are involved in consolidation as well as encoding of extinction memory.
NMDA receptors within BLA and IL contribute to different aspects or phases of extinction learning and memory. Microinfusions of NMDA receptor antagonists into BLA prior to fear extinction training impair both within-session extinction and extinction retention (Falls et al., 1992; Laurent et al., 2008; Laurent and Westbrook, 2008; Lee and Kim, 1998; Lin et al., 2003b), whereas immediate post-extinction training intra-BLA infusions of the NR2B-specific antagonist ifenprodil have no effect on subsequent extinction retention (Laurent and Westbrook, 2008; Sotres-Bayon et al., 2009). This suggests that NMDA receptor-dependent synaptic plasticity within BLA is involved in encoding of extinction memory, but that NR2B-containing NMDA receptors within BLA are not required for consolidation. Conversely, infusions of NMDA receptor antagonists into IL prior to extinction training have no effect on within-session extinction but impair later extinction retention (Burgos-Robles et al., 2007; Laurent and Westbrook, 2008; Sotres-Bayon et al., 2009). Intra-IL infusions immediately after extinction training impair extinction retention as well (Burgos-Robles et al., 2007; Laurent and Westbrook, 2008). This suggests that NMDA receptor-dependent synaptic plasticity within IL is involved primarily in consolidation of extinction memory.
Whereas antagonism of NMDA receptors impairs extinction, enhancement of NMDA receptor function by the NMDA receptor partial agonist D-cycloserine (DCS) facilitates extinction. DCS acts at the glycine modulatory site on the NR1 NMDA receptor subunit to increase calcium influx without causing damage due to neurotoxicity (Sheinin et al., 2001). Systemic administration of DCS either before (Bouton et al., 2008; Ledgerwood et al., 2003, 2005; Lee et al., 2006; Mao et al., 2008; Walker et al., 2002; Weber et al., 2007; Woods and Bouton, 2006; Yang and Lu, 2005) or immediately after (Ledgerwood et al., 2003) extinction training facilitates extinction, suggesting that DCS modulates consolidation of extinction memory. To date no studies have examined the effect of infusions of DCS directly into IL, but local infusion of DCS into BLA prior to (Mao et al., 2006; Walker et al., 2002) or immediately after (Ledgerwood et al., 2003) fear extinction training mimics the effects of systemic administration.
The findings on amygdala and mPFC involvement in fear extinction suggest a model in which the two structures interact dynamically and contribute to different aspects of extinction learning and memory (Figure 3) (Quirk and Mueller, 2008). When a CS is presented in the absence of the US with which it had been paired previously, the contingency change is detected through a mechanism that is not well understood (McNally et al., 2004). NMDA receptor-dependent synaptic plasticity occurs in BLA and contributes to within-session extinction; IL also is recruited and NMDA receptor-dependent synaptic plasticity occurring here is involved in consolidation of extinction memory. Later, at the time of an extinction retention test, IL is activated and inhibits the expression of fear by inhibiting the amygdala via innervation of GABAergic interneurons in BLA and/or ICMs. The nature of the synaptic plasticity occurring at each site is as yet unclear, but may include long-term potentiation (LTP), long-term depression (LTD), synaptic depotentiation, or a combination of these (Herry and Garcia, 2002; Kim et al., 2007b; Li et al., 2009; Royer and Pare, 2002; Santini et al., 2001).
Many of the findings on extinction of drug- or withdrawal-paired cues also implicate these three major substrates: amygdala, mPFC, and NMDA receptor-mediated synaptic plasticity. In light of this, it can reasonably be anticipated that some of the mechanisms of fear extinction outlined above may overlap with the mechanisms of extinction of drug-associated cues. There also are some findings involving additional mechanisms whose significance is as yet unclear.
Studies using the immediate early gene Fos as a marker of neuronal activity suggest that a broad network of structures engaged by drug cue acquisition is re-engaged by extinction. Among these structures is BLA, where Fos expression is increased following exposure to a nonextinguished drug-paired cue or context, but is unchanged relative to that seen in drug-naïve controls following exposure to an extinguished drug-paired cue or context (Baker et al., 1999; Neisewander et al., 2000). Perhaps similarly, in smokers, amygdala activation by smoking-related cues (as assessed by fMRI) diminished over the course of an extinction-like smoking cessation program in which participants smoked reduced nicotine content cigarettes in order to weaken the association between the physical act of smoking and the delivery of a bolus of nicotine (McClernon et al., 2007).
Studies involving BLA manipulations indicate that the plasticity reflected in altered amygdala reactivity following extinction is critical. In the drug self-administration paradigm, rats with excitotoxic lesions of BLA induced after acquisition of cocaine-seeking take longer to cease lever pressing during test sessions in which cocaine delivery is discontinued but previously response-contingent, drug-paired cues are presented noncontingently (Fuchs et al., 2002). In the same paradigm, inactivation of BLA via localized infusions of the voltage-gated sodium channel blocker tetrodotoxin (TTX) immediately following each of several test sessions has a similar effect (Fuchs et al., 2006). One interpretation of these findings is that extinction of the response-reinstating value of drug-paired cues is impaired by disruption of BLA function. Consistent with this interpretation, excitotoxic BLA lesions induced after acquisition of cocaine conditioned place preference (CPP) impair CPP extinction over repeated test sessions (Fuchs et al., 2002). Finally, enhancement of BLA function via localized infusions of glucose (Schroeder and Packard, 2003) or the cholinergic agonist oxotremorine (Schroeder and Packard, 2004) prior to or immediately after nonreinforced confinements in a previously amphetamine-paired environment facilitates extinction of amphetamine CPP, as assessed off-drug in a separate test session.
Only two behavioral studies have examined the role of mPFC in drug cue extinction, but the data suggest that intact function of this region is important as well. In one study, temporary inactivation of mPFC (prelimbic and infralimbic cortices) via localized infusion of the anesthetic bupivacaine prior to nonreinforced confinements in a previously amphetamine-paired context impaired extinction of amphetamine CPP (Hsu and Packard, 2008). Localized infusion of the NMDA receptor antagonist AP5 prior to extinction training had similar effects (Hsu and Packard, 2008). By contrast, in the second study, excitotoxic mPFC lesions had no effect on either the acquisition or extinction of cocaine CPP (Zavala et al., 2003). However, in this study the lesions were targeted to the prelimbic region of mPFC, which does not seem to be required for extinction of conditioned fear (Burgos-Robles et al., 2009; Vidal-Gonzalez et al., 2006). Moreover, because these permanent lesions were made before all phases of training, the data may also reflect some aspects of recovery of function or other compensatory neuroadaptations.
Zavala et al. (2007), using Fos as a marker of neuronal activity, found that the prelimbic and infralimbic regions of mPFC were activated significantly by nonextinguished cocaine-paired cues but were not activated at all by extinguished cocaine-paired cues. Significant behavioral differences between the extinguished and nonextinguished groups may, however, have contributed to the differences in Fos expression, which complicates interpretation of these findings. A previous study from the same group (Neisewander et al. 2000) included conditions that controlled for these differences in behavior, but Fos expression was not examined in prelimbic or infralimbic cortex.
NMDA receptor antagonists impair extinction of drug-paired cues. As described above, infusions of AP5 into mPFC impair extinction of amphetamine CPP (Hsu and Packard, 2008). In the drug self-administration paradigm, infusions of AP5 into BLA immediately after multiple test sessions are associated with more persistent cue-maintained drug-seeking behavior, perhaps because extinction of the response-reinstating value of the cues was impaired (Feltenstein and See, 2007). Finally, pre-extinction training, systemic administration of the NMDA receptor antagonist MK-801 blocks facilitation of cocaine CPP extinction by CDPPB, a modulator of mGluR5 receptors, which are functionally linked to NMDA receptors (Gass and Olive, 2009).
Conversely, extinction of drug- and withdrawal-paired cues is facilitated by DCS. Extinction of cocaine CPP is facilitated when DCS is administered systemically either prior to or immediately after repeated place preference tests, or when infused into BLA immediately after repeated tests (Botreau et al., 2006; Paolone et al., 2009). Likewise, extinction of naloxone-induced conditioned place aversion (CPA) in morphine-dependent rats is facilitated by systemic administration of DCS immediately prior to confinement in the formerly naloxone-paired environment in the absence of acute withdrawal (Myers and Carlezon, 2010). DCS does not enhance the rate of extinction of ethanol CPP but does retard subsequent reconditioning, suggesting a facilitation of extinction that may have been obscured by floor or performance effects (Groblewski et al., 2009). Finally, in an ethanol self-administration paradigm, rats receiving DCS prior to sessions in which ethanol delivery was discontinued but response-contingent cues continued to occur stopped responding more rapidly than did rats receiving vehicle, presumably because DCS facilitated extinction of the secondary reinforcing value of the cues (Vengeliene et al., 2008).
There are a few studies examining the contribution of neuromodulator systems and molecular mechanisms to extinction of drug-paired cues.
Using a place conditioning paradigm in mice in which ethanol conditions either CPP (when administered before confinement in the paired environment) or CPA (when administered after confinement in the paired environment), Cunningham et al. (1998) found that systemic administration of the opiate receptor antagonist naloxone prior to each of several test sessions facilitates extinction of CPP but retards extinction of CPA. Naloxone had no effect on extinction recall when administered prior to an extinction retention test conducted after extinction training was complete. Similar findings have been reported with fear extinction: administration of naloxone prior to extinction training sessions impairs both within-session extinction and extinction retention, whereas administration immediately after extinction training sessions or immediately prior to an extinction retention test has no effect (McNally and Westbrook, 2003). The fear extinction observations are part of a larger literature on the role of endogenous opioids in conditioned fear acquisition, extinction, and inhibition, which when taken as a whole suggests that endogenous opioids might serve as an “error signal” driving associative learning (for discussion see McNally et al., 2004; Myers and Davis, 2007). It is possible that naloxone modulates this error signal similarly for drug cues and fear cues, albeit in opposite directions for cues with appetitive and aversive motivational valences. In this regard endogenous opioid signaling could be viewed as a general associative learning mechanism.
A similar bidirectional, valence-specific effect on extinction of place conditioning has been observed in mice lacking the gene for Narp, an AMPA receptor aggregating protein. Narp knockout mice show delayed extinction of morphine CPP (Crombag et al., 2009) and enhanced extinction of morphine withdrawal-induced CPA (Reti et al., 2008). It is not immediately obvious how to interpret these findings, particularly since the Narp knockout was constitutive, but one possibility is that Narp contributes to error signaling, similarly (perhaps) to endogenous opioids.
Endogenous cannabinoids also have been implicated in drug cue extinction. Parker et al. (2004) found that systemic administration of Δ9-tetrahydrocannabinol (THC) or cannabidiol (CBD) – cannabinoids found in marijuana – prior to confinement in a previously cocaine- or amphetamine-paired environment facilitated extinction of cocaine and amphetamine CPP, at doses where they themselves conditioned neither CPP nor CPA. These effects were not blocked by co-administration of a CB1 cannabinoid receptor antagonist, suggesting that they are mediated through other mechanisms. Perhaps similarly, systemic administration of a cannabinoid reuptake inhibitor prior to fear extinction training facilitates extinction of conditioned fear (Chhatwal et al., 2005a), although it is thought that modulation of fear extinction by endogenous cannabinoids occurs via a CB1 receptor-dependent mechanism (Marsicano et al., 2002).
Finally, in an investigation of the molecular mechanisms underlying acquisition and extinction of conditioned responses to cocaine-paired cues, Zhang et al. (2006) found that transgenic mice lacking the c-fos gene selectively in D1 dopamine receptor-expressing neurons show delayed extinction of cocaine CPP. This impairment may be secondary to other gene expression changes seen in these mice, most notably a lack of increased expression of the NR1 NMDA receptor subunit and the GluR2 AMPA receptor subunit within the nucleus accumbens and caudate putamen in response to repeated cocaine administrations.
There is much work to be done before the mechanisms underlying extinction of conditioned drug craving and withdrawal will be understood, but progress is being made. Immediate, major goals of this work include developing a map of the underlying circuitries, characterizing the interactions among the structures that are involved, and identifying the relevant synaptic plasticity and molecular mechanisms. Given the evidence outlined above, it seems likely that extinction of drug- and withdrawal-paired cues involves similar cortico-limbic mechanisms as have been implicated in fear extinction, but more evidence is needed to solidify that view.
For example, the contribution of mPFC to extinction of conditioned craving or conditioned withdrawal remains unclear. There is little in the way of direct evidence for a role of mPFC, with the exception of the Hsu and Packard (2008) study mentioned above. In addition, the evidence for an involvement of mPFC in fear extinction is suggestive but not necessarily indicative of an involvement in extinction of drug- and withdrawal-paired cues. However, when considered in the context of emerging evidence that prefrontal regions contribute to conditioned fear and drug-seeking behavior in a parallel manner (for review see Peters et al., 2009), a stronger case can be made. Prelimbic (PL) and infralimbic (IL) cortices play complimentary roles with regard to conditioned fear, such that PL contributes to the expression of learned fear whereas IL is critical for its suppression (Corcoran and Quirk, 2007; Knapska and Maren, 2009; Milad and Quirk, 2002; Vidal-Gonzalez et al., 2006). There seems to be a similar regionally-specific contribution of mPFC to the expression and suppression of drug-seeking behavior: dmPFC (anterior cingulate and prelimbic cortices) mediates reinstatement of drug-seeking by drug cues, drug primes, and stress (Capriles et al., 2003; Di Pietro et al., 2006; McFarland et al., 2004; McFarland and Kalivas, 2001; Park et al., 2002), whereas IL is involved in inhibiting drug-seeking following drug primes or extinction of operant responding (McFarland et al., 2004; McLaughlin and See, 2003; Ovari and Leri, 2008; Peters et al., 2008a; Peters et al., 2008b). While no studies have looked at the role of IL specifically in extinction of conditioned responses to drug- and withdrawal-paired cues, it would fit well within this overall framework for IL to be a contributor. Future studies that explicitly examine regional specificity and target prefrontal subregions selectively are a priority.
The amygdala is a potential downsteam target of mPFC actions in extinction, considering that BLA is required for expression of conditioned craving and withdrawal responses and involved importantly in their extinction. Similar to the scenario that has been proposed to occur with fear extinction, mPFC might exert top-down inhibition of the amygdala following extinction of drug- and withdrawal-paired cues via projections to BLA interneurons, ICMs lying between BLA and CeA, or both, and thereby suppress conditioned drug craving and withdrawal. Future work could explore this possibility by (for example) inactivating IL with localized infusions of tetrodotoxin during extinction of drug-paired cues, and determining if this would block the reduction in cue-induced Fos expression observed in BLA following extinction training.
The amygdala is not the only candidate downstream target of mPFC, however. Drug-and withdrawal-paired cues activate a broad network of structures including extended amygdala and reward circuitries, and Fos studies indicate that many of these areas, like BLA, are activated less by extinguished than by nonextinguished drug-paired cues (Neisewander et al., 2000; Zavala et al., 2007). This coordinated response is suggestive of top-down inhibition occurring in a global fashion over a distributed network of plasticity. It could be that this inhibition is orchestrated by IL, whose projections are widespread (Takagishi and Chiba, 1991), and that IL interaction with different “nodes” in the network independently modulates different kinds of CRs to drug- and withdrawal-paired cues. It will be important in future work to determine the contribution of structures other than the amygdala and mPFC to drug cue extinction and to build a case for interactions among them, should they exist.
Finally, the nature of the synaptic plasticity underlying extinction of drug- and withdrawal-paired cues remains to be elucidated. Identifying the sites and types of extinction-related NMDA receptor-dependent synaptic plasticity, and determining which phases of extinction learning and memory it mediates within each site, will be an important step. Based on the evidence reviewed above, it appears that NMDA receptors within BLA are involved in the encoding and consolidation of drug cue extinction memory, but it is not clear which phase(s) of extinction learning are mediated by NMDA receptors in mPFC because of the limited data available.
As alluded to earlier, basic research on fear extinction is beginning to have an impact clinically. For example, neuroimaging studies have found consistently that PTSD is associated with amygdala hyperactivity and medial prefrontal cortex hypoactivity during exposure to trauma-related cues (Etkin and Wager, 2007), which, together with the animal data, suggests that dysfunction of inhibitory cortical control over the amygdala contributes to inappropriate triggering and diminished extinction of fear responses in PTSD patients (Milad et al., 2008; Shin et al., 2004; Shin et al., 2005). To restore functionality in cases of intractable and debilitating PTSD, it has been proposed that techniques such as deep brain stimulation could be used to correct the balance of amygdalar and prefrontal cortical activity (Koenigs and Grafman, 2009). Another clinical strategy for patients with less incapacitating symptoms is to couple standard cue exposure therapy with treatments that enhance NMDA receptor functionality. DCS, which is FDA-approved for the treatment of tuberculosis, has been used in several clinical studies showing that it can facilitate cue exposure therapy for phobias, social anxiety disorder, and obsessive-compulsive disorder in people in much the same way that it facilitates fear extinction in animal models (Hofman et al., 2006; Kushner et al., 2007; Ressler et al., 2004). While there have been some failures to observe DCS effects, particularly with subclinical populations (Guastella et al., 2006), the use of this drug represents a major advance in this area and is considered a prototypical example of bench-to-bedside translational research.
Similar clinical insights and advances may emerge from an advancing understanding of drug cue extinction. For example, addicts perform poorly on neuropsychological tasks that reflect prefrontal function — in some cases exhibiting deficits comparable to those seen in patients with prefrontal cortex lesions (Bechara et al., 2001) — and show reduced (Adams et al., 1993; Fishbein et al., 2005; Forman et al., 2004; Fu et al., 2008; Hester and Garavan, 2004; Kaufman et al., 2003; Kubler et al., 2005; Lee et al., 2005; Li et al., 2008) or altered (Ersche et al., 2005; Goldstein et al., 2001; Yucel et al., 2007) prefrontal responsivity while engaging in them. Moreover, addiction is associated with reduced prefrontal volume and gray matter density (Franklin et al., 2002; Jernigan et al., 1991; Kim et al., 2006; Liu et al., 1998). These findings might explain why drug cue exposure therapy has been relatively unsuccessful; perhaps more intensive therapy than has been tried to date will be required to compensate for limited prefrontal function in addicts. Alternatively, combining drug cue exposure therapy with “talk” therapies designed to maximize skills based on prefrontal functionality, such as behavioral inhibition and gratification delay, may be helpful.
Coupling drug cue exposure therapy with D-cycloserine (DCS) is another potentially useful approach. The demonstrated efficacy of DCS in facilitating both extinction of drug- and withdrawal-paired cues in animals and cue exposure therapy for fear and anxiety disorders in people argues strongly for investigating its utility as an adjunct to drug cue exposure therapy in addicts. It could be that DCS-induced facilitation of extinction in addicts will compensate for extinction deficits and improve the course of therapy. Consistent with this idea, it very recently was reported that DCS facilitates extinction of conditioned craving and skin conductance responses to smoking-related cues in nicotine-dependent smokers (Santa Ana et al., 2009). This important finding provides a strong rationale for further clinical research. At the present time there are several ongoing clinical studies involving pairing DCS with exposure to drug-paired cues (National Clinical Trial number NCT00827281, NCT00430573, NCT00926900, NCT00759473); their outcome may substantiate the utility of using pharmacological interventions to weaken conditioned responses to cues that trigger addictive behavior.
As basic research on extinction of drug- and withdrawal-paired cues progresses, additional clinical insights will emerge. While far in the future, it is not difficult to envision a scenario in which pharmacotherapeutics or other tools will be developed to target specific brain regions or synaptic plasticity mechanisms in order to facilitate and maximize the generalizability of drug cue extinction. The exact direction that this will take remains to be seen. Regardless, it is important to remember that extinction likely is a process that is adaptive within a narrow range of function, and treatments designed to non-specifically boost or retard its pace may have unintended consequences on cognitive function and the quality of life.
This work was supported by the National Institute on Drug Abuse (DA012736 to W.A.C. and DA027752 to K.M.M.). We thank Dr. Danielle Graham and Dr. John Muschamp for their careful reading of the manuscript and many helpful comments.