Using c-fos to study neuronal ensembles in corticostriatal circuitry of addiction

Abstract

Learned associations between drugs and environment play an important role in addiction and are thought to be encoded within specific patterns of sparsely distributed neurons called neuronal ensembles. This hypothesis is supported by correlational data from in vivo electrophysiology and cellular imaging studies in relapse models in rodents. In particular, cellular imaging with the immediate early gene c-fos and its protein product Fos has been used to identify sparsely distributed neurons that were strongly activated during conditioned drug behaviors such as drug self-administration and context- and cue-induced reinstatement of drug seeking. Here we review how Fos and the c-fos promoter have been employed to demonstrate causal roles for Fos-expressing neuronal ensembles in prefrontal cortex and nucleus accumbens in conditioned drug behaviors. This work has allowed identification of unique molecular and electrophysiological alterations within Fos-expressing neuronal ensembles that may contribute to the development and expression of learned associations in addiction.

1. Neuronal ensembles in addiction

Learned associations play an important role in addiction. With repeated drug use, addicts learn to associate drug effects with stimuli or cues in the drug environment, such as drug paraphernalia, context, and people (Goldberg, 1976; O'Brien et al., 1986; Stewart et al., 1984; Wikler, 1973). Long after cessation of drug use, these cues can promote drug craving and relapse (O'Brien et al., 1986; Siegel, 1999; Wikler, 1973). These associations are complex combinations of many different stimuli or cues that are used to recognize people, places, and events with a high degree of resolution. Thus any neural mechanism capable of encoding these associations must have a comparably high degree of resolution. However, nearly all investigations of the molecular and cellular mechanisms of learning and behavior in addiction research manipulate neural activity or assess molecular and cellular alterations in all neurons, or all neurons of a given cell type, at the resolution of discrete brain sites (Bowers et al., 2010; Hyman et al., 2006; Kalivas et al., 2005; Kalivas, 2009; Koob, 2006; Lu et al., 2006; Mameli and Luscher, 2011; Nestler et al., 1993; Nestler, 2001; Pickens et al., 2011; Russo et al., 2010; White and Kalivas, 1998; Wolf and Ferrario, 2010b). While having important effects on drug-related behaviors, alterations at the level of brain sites do not have sufficient resolution to encode complex highly specific learned associations and cannot distinguish between the many distinct learned associations that are encoded within the same brain site.

Instead, learned associations are hypothesized to be encoded within sparsely distributed patterns of neurons called neuronal ensembles (Buzsaki and Moser, 2013; Guzowski et al., 2004; Knierim and Zhang, 2012; Nicolelis et al., 1997; Pennartz et al., 1994; Pennartz et al., 2009; Penner and Mizumori, 2012; Schwindel and McNaughton, 2011), an idea based on the ‘cell assemblies’ hypothesis of Donald Hebb (1949). The particular ensemble activated during behavior is determined by the specific activity pattern of afferent inputs that convey information about interoceptive stimuli (from the body), exteroceptive stimuli (from the environment), and previous experience (from association cortex) (Figure 1) (Best et al., 2001; Deadwyler and Hampson, 1995; Deadwyler and Hampson, 1997; Doetsch, 2000; Guzowski et al., 2004; Johnson et al., 2009; Leutgeb et al., 2005; Mittler et al., 1994; Nicolelis et al., 1997). Those neurons that receive the most integrated stimulatory input will be part of the activated ensemble. Considering the small proportion of neurons estimated to compose a neuronal ensemble (approximately 1–5%) (Bossert et al., 2011; Chawla et al., 2005; Cruz et al., 2013; Fanous et al., 2012; Guzowski et al., 1999; Guzowski et al., 2004; Koya et al., 2009) and the thousands to millions of neurons available in a given brain area, then the number of possible ensemble patterns is enormous and capable of encoding learned associations with sufficient resolution. This number is increased further because neuronal ensembles encoding different learned associations can intermingle in the same brain area and individual neurons can participate in different neuronal ensembles (Schwindel and McNaughton, 2011).

Figure 1.

Schematic drawing to illustrate how different activity patterns of stimulatory afferent input to a brain area can select different patterns of neurons called ‘neuronal ensembles’ that mediate distinct learned associations. The longer red arrows symbolize different patterns of afferent input activated by Drug and Drug-related cues while the longer blue arrows symbolize different patterns of afferent input activated by Non-drug rewards and Non-drug-related cues. Those neurons that receive the highest and most persistent levels of integrated stimulatory input will be part of the neuronal ensemble. The red ovals symbolize neurons that receive the highest levels of excitatory drug-related afferent inputs and encode drug-related learned associations that mediate drug-related behaviors while the blue ovals symbolize neurons that receive the highest levels of excitatory non-drug-related afferent inputs and encode non-drug-related learned associations that mediate non-drug-related behaviors. In the real brain, each of these neuronal ensembles is comprised of thousands to millions of neurons in many different brain areas.

In this review, we discuss neuronal ensembles in conditioned drug behaviors such as drug self-administration and context- and cue-induced reinstatement of drug seeking (Crombag et al., 2008b; Pickens et al., 2011). We focus on the prefrontal cortex and nucleus accumbens within the corticostriatal circuitry that play critical roles in conditioned drug behaviors. To support a role for neuronal ensembles in these behaviors, we begin with evidence accumulated from correlations between lever pressing and neuronal activity that was assessed using in vivo electrophysiology or cellular imaging with the immediate early gene (IEG) c-fos and its protein product Fos in activated neurons. The subsequent description of neurobiological mechanisms provides a more detailed understanding of the relationship between neural activity and c-fos gene activation and provides the basis for recently developed c-fos promoter-based techniques that (1) allow causal role evidence for Fos-expressing neuronal ensembles mediating these conditioned drug behaviors and (2) allow us to identify unique molecular and electrophysiological alterations within these ensembles. The overall goal of this review is to provide support for the hypothesis that sparsely distributed Fos-expressing neurons in the corticostriatal circuitry act together as a unit to form neuronal ensembles that encode and mediate conditioned drug behaviors. A more detailed review of the techniques themselves can be found in (Cruz et al., 2013).

2. Correlates of Fos-expressing neuronal ensembles in drug self-administration

In vivo electrophysiology in drug self-administration models

Since the early 1990’s, nearly all studies of neuronal ensembles in addiction examined correlations between single unit recordings of neurons and lever-pressing for intravenous cocaine, heroin, or drug-related cues in self-administration models in rats (Carelli et al., 1993; Carelli and Deadwyler, 1997; Chang et al., 1994; Chang et al., 1996; Deadwyler et al., 2004; Kiyatkin and Brown, 2007; Woodward et al., 2000). In the nucleus accumbens, cocaine or heroin self-administration has both tonic and phasic effects on single unit activity (Carelli and Deadwyler, 1996; Carelli et al., 1999; Carelli, 2002a; Chang et al., 1997b; Chang et al., 1998; Kiyatkin and Rebec, 1996; Kiyatkin and Rebec, 1999; Kiyatkin et al., 2000; Peoples et al., 1998a; Peoples and Cavanaugh, 2003; Peoples et al., 2004). Drug infusions throughout a self-administration session increase dopamine levels that tonically inhibit most neurons but carry low-resolution information (Haracz et al., 1998; Kiyatkin and Rebec, 1996; Kiyatkin, 2002; Peoples et al., 1998b; Peoples et al., 1999; Peoples and Cavanaugh, 2003; Peoples et al., 2004). At the same time, lever-pressing for drugs correlates with increases or decreases of glutamate-mediated phasic firing in response to either drug reward or to cues associated with drug reward (Carelli et al., 1993; Chang et al., 2000; Guillem et al., 2014; Haracz et al., 1998; Kiyatkin and Rebec, 1996; Kiyatkin and Rebec, 1999; Kiyatkin, 2002; Peoples et al., 1997; Wakabayashi and Kiyatkin, 2014). A key finding has been that different rewards such as cocaine, heroin, water, or sucrose induce phasic firing in largely different sets of neurons, which suggests that the stimulus properties of each reward are encoded in distinct ensembles (Cameron and Carelli, 2012; Carelli and Deadwyler, 1994; Carelli, 2002a; Carelli, 2002b; Carelli and Wondolowski, 2003; Chang et al., 1998; Deadwyler et al., 2004; Opris et al., 2009; Roop et al., 2002). Environmental cues associated with drug reward can also induce phasic firing either prior to drug reward presentation or during drug-free conditions (Carelli, 2002a; Chang et al., 1994; Chang et al., 1997b; Chang et al., 1998; Chang et al., 2000; Wheeler et al., 2011). Many of these neurons are activated only by cues that were previously associated with one reward (cocaine) but not by another reward (food) (Carelli and Ijames, 2001; Carelli, 2002b) which suggests that reward-associated cues or contexts are also encoded in distinct ensembles. Similar types of drug- and cue-induced phasic firing have been observed in different areas of the prefrontal cortex (Chang et al., 1997a; Chang et al., 1997b; Chang et al., 1998; Chang et al., 2000; Guillem et al., 2010; West et al., 2014). Overall, these studies support the hypothesis that distinct neuronal ensembles encode different learned associations underlying drug- and non-drug related behaviors.

c-fos and Fos expression in drug self-administration models

Cellular imaging of IEGs, such as c-fos, zif268, and arc, has also been used to provide correlative evidence for neuronal ensembles in addiction. These IEGs are rapidly induced within activated neurons (Cohen and Greenberg, 2008; Herdegen and Leah, 1998; Morgan and Curran, 1991) and their mRNA or protein products have long been used as markers of behaviorally activated neurons. C-fos mRNA and its protein product Fos have been the most commonly used IEG markers of neuronal activity in addiction research (Brenhouse and Stellar, 2006; Crombag et al., 2002; Graybiel et al., 1990; Hope et al., 1992; Konradi et al., 1994; Moratalla et al., 1993; Persico et al., 1993; Steiner and Gerfen, 1993; Young et al., 1991), with the first paper published in 1989 (Robertson et al., 1989). Immunohistochemical assays indicated that Fos expression is increased during cocaine or heroin self-administration throughout the corticostriatal circuitry, including medial prefrontal and orbitofrontal cortex, nucleus accumbens core and shell subregions, and the dorsal striatum (Ben-Shahar et al., 2004; Larson et al., 2010; Madsen et al., 2012; Martin-Garcia et al., 2014; Pich et al., 1997; Thiel et al., 2010; Zahm et al., 2010). Even when drug is not onboard, contextual cues and discrete cues previously paired with drug self-administration activate many of the same prefrontal cortex and striatal areas during reinstatement of lever pressing (Bastle et al., 2012; Bossert et al., 2011; Cruz et al., 2014; Fanous et al., 2012; Fanous et al., 2013; Hamlin et al., 2008; Kufahl et al., 2009; Mahler and Aston-Jones, 2012; Neisewander et al., 2000; Shalev et al., 2003; Zavala et al., 2007; Zhou et al., 2013) or exposure to the paired context alone without lever responding (Doherty et al., 2013). Perhaps due to higher variability, in situ hybridization and quantitative PCR assays for c-fos mRNA identify activation of only a subset of these same brain areas (Celentano et al., 2009; Daunais et al., 1993; Daunais et al., 1995; Hearing et al., 2008a; Hearing et al., 2008b; Koya et al., 2006; Kuntz et al., 2008; Kuzmin and Johansson, 1999). Overall, cellular imaging studies indicate that c-fos mRNA and Fos protein are expressed within sparsely distributed neurons in many of the same prefrontal cortex and nucleus accumbens regions shown with in vivo electrophysiology studies using similar drug and cue conditions in drug self-administration and relapse models.

Neural activity and expression of c-fos and Fos

Although Fos expression and in vivo electrophysiology studies identify similar brain areas during lever responding for cocaine and heroin and related cues, the exact correspondence between Fos expression and electrophysiological activity at the cell level in intact brain of behaving animals is not known. Previous attempts to directly examine the relationship between Fos expression and electrophysiological activity in the striatum and hippocampal dentate gyrus of intact rats suggest that Fos expression correlates with synaptic activity levels and not with action potentials (Labiner et al., 1993; Sgambato et al., 1997). However, we do not know if the electrophysiological recordings in these studies were from Fos-positive or Fos-negative neurons. Indeed, the majority of recorded neurons were likely Fos-negative and thus would have had lower synaptic activity levels that were below the threshold for generating action potentials. Future in vivo electrophysiology experiments need to determine whether Fos expression or c-fos promoter activation is induced in the same neurons being recorded.

Regardless of whether action potentials are necessary for Fos expression or not, strong and persistent synaptic activity leads to both Fos expression and action potentials (Cohen and Greenberg, 2008; Curran and Morgan, 1986; Morgan and Curran, 1986; Morgan and Curran, 1988; Rajadhyaksha et al., 1999; Thomas and Huganir, 2004; Xing et al., 1996). The only difference may be that the synaptic activity threshold required for Fos expression is lower than that required for action potentials. Under these conditions, Fos expression still functions as a marker of the most electrophysiologically active neurons in a brain area, and this set of neurons likely contains most if not all of the neurons that generate action potentials responsible for conditioned behaviors.

3. Neurochemistry of c-fos and Fos expression in addiction research models

In addiction research, most c-fos and Fos expression studies have focused on the striatum, including the nucleus accumbens. The neurotransmitter systems that regulate drug-induced c-fos and Fos expression in striatal medium spiny neurons (MSNs) are similar to those that regulate in vivo electrophysiological activity. We focus here on glutamate and dopamine, the two most highly studied neurotransmitters in addiction research. The excitatory neurotransmitter glutamate mediates drug-induced c-fos induction in striatum of behaving animals; reviewed in (Cohen and Greenberg, 2008; Vanhoutte et al., 1999). NMDA glutamate receptor antagonists block drug-induced Fos expression (Berretta et al., 1992; Ghosh et al., 1994; Konradi et al., 1996; Ohno et al., 1994; Rajadhyaksha et al., 1999; Snyder-Keller, 1991; Torres and Rivier, 1993) while electrical or chemical stimulation of cortical glutamatergic inputs induces Fos expression in striatal neurons (Berretta et al., 1997; Berretta et al., 1999; Canales et al., 2002; Fu and Beckstead, 1992; Gerfen et al., 2002; Liste et al., 1995; Parthasarathy and Graybiel, 1997; Tschanz et al., 1991; Tschanz et al., 1994). Thus glutamate alone can induce c-fos and Fos expression.

Dopamine also plays a role in drug-induced c-fos induction in the striatum (Badiani et al., 1998; Labandeira-Garcia et al., 1996; Paul et al., 1995). Psychostimulant-induced Fos expression requires activation of both D1-type (Berretta et al., 1992; Young et al., 1991) and D2-type (LaHoste et al., 1993; Ruskin and Marshall, 1994) dopamine receptors. However, dopamine alone does not induce Fos expression. Anesthetics such as chloral hydrate or urethane have been used to attenuate glutamatergic input to the striatum during cocaine and amphetamine administration without altering drug-induced dopamine (Kreuter et al., 2004; Ryabinin et al., 2000; Torres and Rivier, 1993). Under these conditions, cocaine-induced dopamine could not induce Fos expression in the absence of glutamate.

The interaction between glutamate and dopamine on Fos expression can be explained, at least for D1-type MSNs, by the upstate-downstate hypothesis that proposes drug-induced dopamine enhances ongoing glutamate-mediated synaptic activity instead of dopamine inducing synaptic activity directly (Nicola et al., 2000; O'Donnell, 2003; Onn et al., 2000; Pennartz et al., 1994; Surmeier et al., 2007). When synapses are sufficiently activated by excitatory glutamatergic inputs on the dendritic spine head, the synapses transform from the downstate to the upstate. Concurrent drug-induced activation of D1-type dopamine receptors on the neck of the spine further enhances activity of the upstate synapses while attenuating activity in downstate synapses (Surmeier et al., 2007). This produces an effect at the systems level where drug-induced dopamine enhances activity of the set of neurons that receive the highest levels of ongoing glutamatergic excitatory input from cue-induced afferents, while attenuating activity in the majority of neurons that are receiving lower levels of excitatory afferent input; the net effect is that drug-induced dopamine enhances signal to noise during information processing in the striatum. Similar evidence for the upstate-downstate hypothesis has been found within the prefrontal cortex using in vivo electrophysiology (O'Donnell, 2003; Surmeier et al., 2007). It should be noted that this upstate-downstate explanation does not work for psychostimulant-induced Fos expression in D2-type MSNs in rats injected outside of their home cage; although dopamine enhancement of glutamatergic input in general still appears to be involved in Fos expression in these neurons (described below in the section entitled ‘cAMP/PKA pathway in c-fos promoter activation in D1-type MSNs’).

In summary, glutamate and dopamine act synergistically to mediate Fos expression. While glutamate alone can activate neural activity to induce Fos, drug-induced dopamine enhances ongoing glutamatergic excitatory input to induce Fos expression. We hypothesize that Fos is expressed in the small number of neurons that received the highest levels of cue-induced glutamatergic excitatory input during conditioned drug behavior.

4. Molecular and cellular mechanisms of drug-induced c-fos promoter activation

We know more about the molecular and cellular mechanisms of c-fos promoter activation in the brain than for almost any other mammalian promoter. Ever since the first demonstrations of c-fos induction in brain (Dragunow et al., 1987; Morgan et al., 1987), thousands of papers and many excellent reviews (Brami-Cherrier et al., 2009; Cohen and Greenberg, 2008; Farivar et al., 2004; Herdegen and Leah, 1998; Hughes et al., 1999; Lyons and West, 2011; Morgan and Curran, 1991; Nestler, 2001) have been written on the subject. Many of these studies used cell or slice cultures that were critical for determining candidate mechanisms, but not all these mechanisms are necessarily operating in the intact brains of behaving animals. Thus we attempt, whenever possible, to emphasize data confirmed using intact brains in behaving animals or slices from adult brains. In addition, we have not included data from knockout mice because these data are difficult to interpret due to the likely occurrence of many compensatory alterations induced within these mice that are unknown and may or may not be directly related to the deleted target gene. Lastly we refer to ‘c-fos promoter activation’ in this section because drug-induced c-fos and Fos expression levels are determined primarily by molecular and cellular activation of the c-fos promoter to induce transcription (Morgan and Curran, 1991).

The c-fos promoter

The two most critical components of the c-fos promoter for neural activity-induced expression are the serum response element (SRE) and the calcium response element (CaRE) (Figure 2); reviewed in (Chen et al., 1992; Cohen and Greenberg, 2008; Herdegen and Leah, 1998; Morgan and Curran, 1991). The transcription factor proteins serum response factor (SRF) and Elk-1 bind constitutively as a complex to the SRE while CREB binds constitutively to the CaRE. Ultimately both Elk-1 and CREB need to be phosphorylated to activate the c-fos promoter. The CaRE site is sometimes called the cyclic AMP response element (CRE), but since the cAMP/PKA pathway appears to play only an indirect role in activating CREB in the intact brain in addiction research models (see cAMP/PKA section below), we refer to this site only as CaRE.

Figure 2.

Schematic drawing to illustrate the neurochemistry and molecular mechanisms of c-fos and Fos expression in strongly activated neurons in prefrontal cortex and nucleus accumbens of awake behaving animals. Glutamate is the main excitatory neurotransmitter that increases neural activity. Dopamine enhances (red arrow and + sign) glutamate-mediated neural activation of the small proportion of neurons that have the highest levels of neural activity while inhibiting neural activation of the majority of neurons that have lower levels of neural activity. The detailed electrophysiological mechanisms underlying this glutamate-dopamine interaction are described in (Surmeier et al., 2007). Strong persistent neural activity induces calcium (Ca²⁺) influx through NMDA-type glutamate receptors and voltage-sensitive calcium channels (VSCCs) to levels that are sufficient for phosphorylating and activating ERK/MAPK via the Ras-Raf-MEKK pathway. ERK/MAPK activation leads to phosphorylation of Elk-1 that is associated with serum response factor (SRF) as well as phosphorylation of CREB via ribosomal S6 kinase (RSK). Elk-1/SRF and CREB are transcription factors that, when phosphorylated, can induce transcription of the coding sequence for c-fos. Transcribed c-fos mRNA and the translated protein product Fos can be used as markers of strongly activated neurons.

A critical role for calcium-dependent activation of the ERK/MAPK pathway

Drug-induced c-fos promoter activation in the brains of awake behaving rats is mediated primarily by calcium-dependent activation of the ERK/MAPK pathway (Bading et al., 1997; Bertran-Gonzalez et al., 2008; Brami-Cherrier et al., 2005; Chen et al., 1992; Cohen and Greenberg, 2008; Ferguson and Robinson, 2004; Mattson et al., 2005; Miller and Marshall, 2005; Radwanska et al., 2005; Sgambato et al., 1998a; Sgambato et al., 1998b; Valjent et al., 2000). Recent excellent reviews of the role of this pathway in corticostriatal circuitry can be found in (Brami-Cherrier et al., 2009; Cahill et al., 2014). In brief, neural activity initiates calcium influx through NMDA receptors and L-type voltage-sensitive calcium channels (VSCC) to activate a variety of calcium-dependent molecules, including Ras-GRP, that mediate calcium-dependent activation of the Ras/Raf kinase pathway that phosphorylates and activates ERK (Agell et al., 2002; Cahill et al., 2014) (Figure 2). Phosphorylated ERK then translocates to the nucleus where it can phosphorylate the transcription factors Elk-1 and CREB (via RSK: ribosomal S6 kinase) on the c-fos promoter (Besnard et al., 2011; Cahill et al., 2014; Chen et al., 1992; Cohen and Greenberg, 2008; Morgan and Curran, 1991).

As described in the previous neurochemistry section, glutamate and dopamine can act synergistically to depolarize neurons and induce calcium influx and ERK activation. Similarly, any neurochemical system that regulates electrophysiological activity and intracellular calcium levels, such as inhibitory receptors (e.g., GABAergic), excitatory receptors (e.g., acetylcholinergic), or modulatory receptors (e.g., presynaptic mGluR2/3 receptors), can affect ERK-mediated c-fos promoter activation (Herdegen and Leah, 1998; Lyons and West, 2011).

One of the key aspects of neuronal activation of the ERK pathway is that it requires consistent (not sporadic) high levels of calcium influx (Blank et al., 2004; Morgan and Curran, 1988; Xing et al., 1996). Thus, ERK-dependent activation of c-fos reflects a summation of activity-dependent calcium influx over seconds to minutes. Only strong consistent activity over this timeframe increases calcium levels past the threshold required to activate the c-fos promoter (Deisseroth and Tsien, 2002; Deisseroth et al., 2003; Ghosh et al., 1994; Konradi et al., 1996; Mermelstein et al., 2000; Morgan and Curran, 1989; Murphy et al., 2002; Wu et al., 2001). This mechanism agrees with the hypothesis that the c-fos promoter is induced in only the most consistently and strongly activated neurons observed with in vivo electrophysiology studies.

Possible alternative calcium-dependent signaling pathways

Other calcium-dependent intracellular signaling pathways (distinct from the ERK/MAPK pathway) may regulate c-fos promoter activation. Numerous cell and slice culture studies indicate that calcium-dependent CaM kinase IV can rapidly phosphorylate CREB within the first few minutes of neural activity before ERK activation takes over the dominant role; reviewed in (Deisseroth and Tsien, 2002). Although group I mGluR-mediated release of calcium from intracellular stores appears not to play a significant role in ERK activation or c-fos promoter activation (Choe et al., 2002; Wang et al., 2007), group I mGluR receptors are able to activate the ERK pathway using a novel calcium-independent Homer protein-dependent pathway (Mao et al., 2005; Yang et al., 2004). Another calcium-dependent molecule called DREAM can bind to and constitutively inhibit the c-fos promoter (Carrion et al., 1999). Increased levels of calcium during neural activity can bind to DREAM and attenuate DREAM-mediated inhibition of the c-fos promoter. While these alternative calcium-dependent pathways may play a role in c-fos promoter activation, they await confirmation in the intact brains of behaving (non-knockout) animals.

cAMP/PKA pathway in c-fos promoter activation in D1-type MSNs

Psychostimulant induction of c-fos and Fos in D1-type MSNs (Berretta et al., 1993; Bertran-Gonzalez et al., 2008) suggests a role for D1 dopamine receptor stimulation of the cAMP/PKA pathway in c-fos promoter activation. The D1/cAMP/PKA pathway can act synergistically with the glutamate/calcium/ERK pathway by a combination of electrophysiological and biochemical mechanisms. The electrophysiological mechanism is based on the upstate-downstate hypothesis described above. Dopamine-induced cAMP/PKA pathway enhances the electrophysiological effects of ongoing glutamatergic input onto MSN dendritic spines by modulating ion channels within synapses (Surmeier et al., 2007) that increases calcium-dependent ERK activation leading to c-fos promoter activation. One biochemical mechanism involves D1/cAMP/PKA phosphorylation of NMDA receptors in the synapse to enhance activation of the NMDA/calcium/ERK pathway (David et al., 2014; Ron, 2004). Another biochemical mechanism proposes that D1/cAMP/PKA activates a DARPP32-mediated mechanism that indirectly enhances NMDA/calcium activation of the ERK pathway (Valjent et al., 2005). However this mechanism is dependent largely on experiments with knockout mice that need to be validated in wild-type mice.

Enhanced glutamate-induced ERK in c-fos promoter activation in D2-type MSNs

When psychostimulants are administered to rats (but not mice) outside their home cage, as done in most behavioral assays, then c-fos and Fos are induced equally in D1-and D2-type MSNs (Badiani et al., 1999; Jaber et al., 1995; Mattson et al., 2005; Mattson et al., 2007; Uslaner et al., 2001); reviewed in (Badiani and Robinson, 2004). Since D2 receptor activation reduces cAMP levels (Stoof and Kebabian, 1981), it is unlikely that cAMP is necessary for c-fos promoter activation in these neurons. Badiani and Robinson have shown that excitatory glutamatergic inputs from the cortex to the striatum are much more active in rats outside their home cages than in rats kept in their highly habituated home cage environment (Badiani et al., 1998; Ferguson and Robinson, 2004). Since activation of glutamatergic inputs from the cortex to the striatum preferentially induce Fos in D2-type MSNs (Berretta et al., 1992; Berretta et al., 1997; Parthasarathy and Graybiel, 1997), psychostimulant-induced Fos expression in D2-type MSNs in rats injected outside their home cage is likely due to increased excitatory glutamatergic input and activation of calcium-dependent ERK in these neurons (Ferguson and Robinson, 2004).

Signaling from the synapse to the nucleus

One of the more important questions is how information about synaptic activity in dendritic spines is conducted to the nucleus for c-fos promoter activation (reviewed in (Bengtson and Bading, 2012; Bito et al., 1997; Matamales, 2012; Wiegert and Bading, 2011). Locally within dendritic spines, glutamatergic input can depolarize the synapse and activate NMDA and voltage-gated calcium channels that increase calcium influx and activate ERK and CaM kinase signaling to produce significant local effects within the activated spine (Bowers et al., 2010; Citri and Malenka, 2008; Luscher et al., 2000). Distinct from local regulation of synaptic function, some of the activated molecules (e.g., phospho-ERK) may be transported from the dendritic spines to the nucleus to activate the c-fos promoter (Deisseroth and Tsien, 2002; Hardingham et al., 2001; Zhai et al., 2013). However, the relatively long distances from these dendritic spines to the nucleus and the rapid activation of the c-fos promoter within minutes suggests that other mechanisms may be involved (Matamales, 2012). One possibility involves calcium-induced calcium release from endoplasmic reticulum (calcium waves) (Matamales, 2012). A simpler electrophysiological explanation is that synaptic depolarization events in dendritic spines are conducted rapidly by well-established cable properties of dendrites to the cell soma that integrates these depolarization events (Carter et al., 2007; Hille, 1992; Matamales, 2012). Once the depolarization threshold for activation of voltage-gated calcium channels is reached in the cell soma, calcium influx into the soma can activate the ERK (and CaM kinase IV) pathway and activate the c-fos promoter in the nucleus (Bengtson and Bading, 2012; Bito et al., 1997; Cahill et al., 2014).

Summary for c-fos promoter activation in addiction research models

The described neurochemical and molecular biological mechanisms all support the idea that c-fos promoter activation is an indicator of strong and persistent calcium influx into synapses of MSNs that received the most excitatory glutamatergic input. Drug-induced dopamine can synergistically enhance c-fos promoter activation in these strongly activated MSNs. It should be noted that a lack of Fos expression in a neuron does not imply a complete lack of neural activity, only that it is not depolarized strongly or persistently enough to produce enough intracellular calcium to activate the ERK signalling pathway. The earlier section describing in vivo electrophysiology and cellular imaging studies indicate that the neurons receiving the most excitatory glutamatergic input are determined by the context and cues present during drug administration. Altogether these data support the hypothesis that Fos-expressing neurons in corticostriatal circuitry can act together as a unit to form neuronal ensembles that encode and mediate conditioned drug behaviors. It is important to note that we treat c-fos and Fos only as markers of activated neurons in this hypothesis and do not imply that these molecules are directly involved in neuronal ensemble function. To test our hypothesis, we developed several techniques that exploit c-fos promoter activation in behaviorally activated neurons.

4. Causal roles for neuronal ensembles in behavior

In one sense, we have been able to identify sparsely distributed neuronal ensembles using c-fos mRNA and Fos protein expression for a long time. However, we have not been able to test the hypothesis until very recently. To examine causal roles for neuronal ensembles in learned behavior, one requires a method for manipulating one set of sparsely distributed neurons that were activated specifically by one set of stimuli without affecting surrounding neurons activated by distinct sets of stimuli. Existing inactivation methods, such as lesions, reversible inactivation with GABAergic agonists, receptor antagonists, or optogenetic and DREADD technologies (Rogan and Roth, 2011) either inactivate most neurons or selected neurons based on cell type, regardless of their activation state during behavior or stimuli presentation. To selectively manipulate behaviorally activated neuronal ensembles, we exploited the selective activation of the c-fos promoter in neurons strongly activated during behavior.

Daun02 inactivation was the first procedure to demonstrate a causal role for endogenously produced neuronal ensembles in a learned behavior (Bossert et al., 2011; Cruz et al., 2014; Fanous et al., 2012; Koya et al., 2009). This procedure, reviewed in (Cruz et al., 2013), inactivates only neurons that were previously activated during a specific behavior or during reactivation of a specific memory (Koya et al., 2009). We used c-fos-lacZ transgenic rats with a transgene that contains a c-fos promoter that induces transcription of the lacZ coding sequence, leading to translation of the protein product β-galactosidase (βgal) in strongly activated neurons, but not in surrounding non-activated neurons (Kasof et al., 1995a; Kasof et al., 1995b; Kasof et al., 1996). Once β-gal is the induced in strongly activated neurons, the inactive prodrug Daun02 is injected into the brain area of interest where it is catalyzed by β-gal to the active product daunorubicin that appears to induce apoptotic cell death of the previously activated neurons (Bakina and Farquhar, 1999; Farquhar et al., 2002; Ghosh et al., 2000). In Daun02 inactivation experiments, c-fos-lacZ rats are exposed on ‘Induction day’ to the previously drug-associated training- or extinction-related context and cues, or a novel environment as a control condition, for 15–30 min to induce β-gal protein in the activated neurons. Daun02 is injected into the brain 90 min later when β-gal levels are maximal (Koya et al., 2009). On ‘Test day’, three days later, the rats are exposed to training- or extinction-related context and cues to assess whether inactivation of the previously activated βgal-expressing neurons altered the ability of the cues and contexts to reactivate the same neurons and induce context-specific behavior.

The first application of the Daun02 inactivation procedure was context-specific sensitization of cocaine-induced locomotor activity where the sensitized response is expressed in only the drug-paired environment (A) and not in a non-drug-paired environment (B) (Anagnostaras and Robinson, 1996; Badiani and Robinson, 2004; Crombag et al., 2002; Hope et al., 2006; Koya et al., 2009; Marin et al., 2009; Mattson et al., 2008; Robinson et al., 1998; Stewart et al., 1984; Stewart and Vezina, 1991; Uslaner et al., 2003; Vezina and Stewart, 1984; Vezina et al., 1989). We first found that context-specific cocaine-induced locomotion induced Fos in ~3% of nucleus accumbens neurons. In the next experiment, intra-accumbens Daun02 injections 90 min after cocaine-induced reactivation of the drug-paired neuronal ensemble in the drug-paired environment (A) on Induction day attenuated subsequent cocaine-induced reactivation of the drug-paired ensemble and the sensitized locomotor response in the drug-paired environment (A) on Test day (Figure 3). In control animals, however, intra-accumbens Daun02 injections 90 min after cocaine-induced activation of a non-drug-paired neuronal ensemble in a non-drug-paired environment (B) on Induction day did not attenuate subsequent cocaine-induced reactivation of the drug-paired ensemble or the sensitized locomotor response in the drug-paired context (A) on Test day. This indicates that activation of a context-specific pattern of accumbens neurons is necessary for context-specific sensitization of cocaine-induced locomotion in rats.

Figure 3.

Daun02 inactivation of Fos-expressing neuronal ensembles in rat nucleus accumbens attenuated sensitized cocaine-induced locomotion and β-galactosidase-labeled neural activity (shown on the y-axes) on Test day only when Daun02 was injected into the nucleus accumbens following cocaine injections to rats in the Cocaine-paired locomotor activity chamber on Induction day. The x-axes indicate conditions used on Induction day three days prior to Test day. On induction day, Vehicle or Daun02 was injected into the nucleus accumbens 90 min following systemic injections of Saline or Cocaine to rats. Daun02 did not attenuate cocaine-induced locomotion or neural activity when non-drug-associated neurons were activated in the Saline-injected rats. The column on the left indicates cocaine-induced locomotion on the first day of repeated injections during sensitization training, and is provided only to show that Daun02 attenuated cocaine-induced locomotion on test day in sensitized rats to approximately the same level as that seen in non-sensitized rats. * indicates a significant difference (p<0.05) between Vehicle and Daun02 groups following cocaine injections three days earlier on induction day. Modified figure from (Koya et al., 2009).

We then used the Daun02 inactivation procedure in two studies that assessed functional roles for neuronal ensembles in a rat model of drug relapse, called context-induced reinstatement of drug seeking (Crombag et al., 2008a), which is based on the ABA renewal model (Bouton and Bolles, 1979). Rats were first trained to self-administer drugs in one context (A); each drug infusion was paired with discrete cues (light or light+tone). Then drug-reinforced lever responding in the presence of the discrete drug cue was extinguished in a non-drug context (B). Subsequently, context-induced reinstatement of drug seeking was assessed by re-exposing rats to the discrete cues in the drug-associated (A) context (Crombag and Shaham, 2002). We specifically assessed roles for ventral mPFC ensembles in context-induced reinstatement of heroin seeking (Bossert et al., 2011) and for accumbens shell ensembles in context-induced reinstatement of cocaine seeking (Cruz et al., 2014). We found that context-induced reinstatement of heroin seeking correlated with Fos induction in ~6% of ventral mPFC neurons (Bossert et al., 2011) while context-induced reinstatement of cocaine seeking correlated with Fos induction in ~3.4% of accumbens shell neurons (Cruz et al., 2014). In the next set of experiments, we found that prior Daun02 injections into ventral mPFC or accumbens shell after exposure to the drug-associated context (A) decreased subsequent context-induced β-gal expression and reinstatement of drug seeking in context A (Figure 4A and 4B). Daun02 injections following exposure to the control extinction-associated context (B) or to a novel context (C) did not alter β-gal expression or reinstatement of drug seeking in context A on test day. These data indicate that the ability of specific cues and contexts to induce reinstatement of drug seeking is mediated by specific patterns of Fos-expressing neuronal ensembles that are selected by these cues and contexts.

Figure 4.

Daun02 inactivation of Fos-expressing neuronal ensembles in different drug relapse models in rats. Lever-pressing behavior on test day is indicated on the y-axes of graphs in the left-hand column. Neural activity, indicated by number of β-galactosidase (β-gal)-labeled neurons, on test day is indicated on the y-axes of graphs in the right-hand column. The x-axes indicate conditions used on Induction day three days prior to Test day. Panel A: Daun02 (but not Vehicle) injections into ventromedial prefrontal cortex (vmPFC) following exposure to the Heroin context (but not the Extinction context) on Induction day attenuated context-induced reinstatement of heroin seeking (lever pressing) and neural activity three days later on Test day. Modified figure from (Bossert et al., 2011). Panel B: Daun02 (but not Vehicle) injections into nucleus accumbens shell following exposure to the Cocaine context (but not to a Novel context) on Induction day attenuated context-induced reinstatement of cocaine seeking (lever pressing) and neural activity three days later on Test day. Modified figure from (Cruz et al., 2014). Panel C: Daun02 (but not Vehicle) injections into orbitofrontal cortex (OFC) following exposure to the Heroin context (but not to a Novel context) on Induction day attenuated cue-induced heroin seeking (lever pressing) and neural activity three days later on Test day. Modified figure from (Fanous et al., 2012). * indicates a significant difference (p<0.05) between Vehicle and Daun02 groups following exposure to the drug context three days earlier on induction day.

We also used Daun02 to demonstrate a causal role for orbitofrontal cortex (OFC) neuronal ensembles in cue-induced heroin seeking (Fanous et al., 2012). Rats were first trained to lever press for heroin paired with discrete cue (light+tone) presentation and then kept in their home cage for 14 d. On test day, re-exposure to the discrete cues in the training context under extinction conditions increased lever responding and induced Fos expression in ~12% of OFC neurons. In the next experiment, we used the Daun02 inactivation procedure to assess a causal role for context-dependent Fos-expressing neuronal ensembles in the lateral OFC in cue-induced heroin seeking. We found that prior injections of Daun02 into OFC after exposure to the heroin-associated environment attenuated subsequent cue-induced β-gal expression and heroin seeking on test day (Figure 4C). In contrast, prior Daun02 injections after exposure to a novel control environment (without levers or the discrete cues to activate a different pattern of OFC neurons) had no effect on subsequent cue-induced β-gal expression or heroin seeking on test day.

In all four studies, the drug-associated contexts and cues increased Fos expression in prefrontal cortex and nucleus accumbens. Selective inactivation of these Fos-expressing neurons in the test groups attenuated subsequent reactivation of these neurons and drug seeking behavior when exposed again to the drug-associated contexts and cues on test day. Inactivation of Fos-expressing neurons that were activated by alternate contexts or cues in the control groups did not attenuate subsequent reactivation of the drug-associated neurons or behaviors on test day. These studies provide evidence that drug-associated contexts and cues activate specific patterns of Fos-expressing neuronal ensembles that mediate the conditioned drug behaviors. They also provide the rationale for identifying unique molecular, structural, and electrophysiological alterations within Fos-expressing neurons.

5. FACS sorting of selectively activated neuronal ensembles

Based on the above studies, molecular neuroadaptations within context- and cue-selected Fos-expressing neurons are likely to play unique and important roles in conditioned drug behaviors. We recently developed an antibody-based fluorescence-activated cell sorting (FACS) method to purify activated Fos-expressing neurons and assess gene expression (Guez-Barber et al., 2011; Guez-Barber et al., 2012), similar to the previously described GFP expression-based FACS method (Lobo et al., 2006; Lobo, 2009). In our method, reviewed in (Cruz et al., 2013), cells from adult brain tissue are enzymatically dissociated and then neurons are labeled with fluorescent NeuN antibodies while strongly activated neurons are labeled with fluorescent Fos or β-gal antibodies. The labeled cells are forced to pass single file through a narrow flow cell in a flow cytometer and sorted according to their light-scattering and immunofluorescent characteristics. Gene expression can then be assessed using qPCR and microarrays.

Until recently, nearly all molecular neuroadaptations had been found in brain homogenates that did not distinguish between neurons activated or not activated during behavior (Kalivas, 2009; Koob et al., 2004; Nestler et al., 1993; Nestler, 2001; Nestler, 2005; Olive et al., 2012). It had long been speculated that gene expression in the small proportion of neurons activated during behavior may be diluted or masked using these brain homogenates. To test this idea, acute methamphetamine was injected rats and IEG expression in brain homogenates was compared with IEG expression in FACS-separated Fos-positive and Fos-negative neurons in the dorsal striatum (Liu et al., 2014). Methamphetamine induced 3–20-fold increases of the IEGs arc, homer-2, c-fos, fosB and its isoforms (ΔfosB and a novel isoform ΔfosB-2) in Fos-positive, but not Fos-negative neurons, while IEG mRNA induction was 10-fold lower or absent when assessed in unsorted samples from single dorsal striatum homogenates. Thus FACS is useful for identifying unique alterations in behaviorally activated neurons.

FACS was used to assess unique gene alterations in β-gal and Fos-expressing striatal neurons in context-specific sensitization of cocaine-induced locomotion (Guez-Barber et al., 2011). Fos-expressing neuronal ensembles in nucleus accumbens in ventral striatum had already been shown to mediate context-specific sensitization of cocaine-induced locomotion (Koya et al., 2009). Cocaine-induced gene expression in activated striatal neurons following FACS purification of β-gal neurons from c-fos-lacZ transgenic rats 90 min after test injections of cocaine to naïve rats or cocaine-sensitized rats and compared this to gene expression in control rats that received test injections of saline (Guez-Barber et al., 2011). Microarray and qPCR analyses indicate that expression of the IEG activity markers arc, fosB, and nr4a3 were increased in activated neurons from cocaine-injected rats but not in non-activated neurons from the same cocaine-injected rats or in all neurons from saline-injected rats. Notably, gene expression was not significantly different between the two control groups: non-activated neurons from cocaine-injected rats and all neurons from saline-injected rats.

FACS was also used to assess gene expression in Fos-expressing neurons from the prefrontal cortex, including the OFC, following cue-induced heroin seeking in rats (Fanous et al., 2013). OFC neuronal ensembles were previously shown to play a causal role in mediating cue-induced heroin seeking after prolonged withdrawal (Fanous et al., 2012). Rats were first trained to self-administer heroin and then remained in their home cages for 14–30 days. On test day, half of the rats were re-exposed to the training context for 90 min under extinction conditions, while control rats were kept in their home cage. After FACS separation of Fos-positive and negative neurons, qPCR analyses indicated that cue-induced heroin seeking increased expression of the IEG activity markers arc, fosB, egr1, and egr2 in Fos-positive neurons but not in Fos-negative neurons from the same ‘test’ rats or inall neurons from the ‘no test’ rats that were left in the home cage after drug self-administration training. Again, gene expression was not significantly different between the two control groups: non-activated neurons from rats exposed to the heroin cue and all neurons from rats kept in their home cages.

In both studies, IEGs were induced only within the behaviorally activated Fos-expressing neurons and not at all in the Fos-negative neurons. This all-or-nothing activation of IEGs in Fos-expressing neurons does not fit with the more graduated range of neuronal activity levels observed from in vivo electrophysiology experiments. As suggested above in our section on neurobiology of c-fos promoter activation, it is likely that only a small proportion of neurons experience enough strong and consistent activation so that calcium influx crosses the threshold for induction of c-fos and other IEGs. It is notable that the IEGs induced selectively in Fos-expressing neurons are either transcription factors (e.g., fosB) capable of inducing further waves of gene expression or involved in synaptic adaptations (e.g., arc and homer). It is possible that only this small proportion of Fos-expressing neurons receive sufficiently high levels of integrated stimulatory input during learning and memory retrieval to undergo unique adaptations that may contribute to the development and consolidation of learned associations.

6. Electrophysiological alterations in activated neuronal ensembles

Alterations in synaptic strength are thought to be the main cellular mechanism underlying learning (Bowers et al., 2010; Bredt and Nicoll, 2003; Luscher and Malenka, 2011; Martin et al., 2000). However, these previous studies examined drug-induced global alterations in randomly selected neurons regardless of their activation state during drug-related behavior. As discussed above, alterations found in the general population do not have the resolution to encode specific learned behaviors. We had already shown that Fos-expressing accumbens neurons are necessary for context-specific sensitization of cocaine-induced locomotion while the majority of accumbens neurons are less directly involved (Koya et al., 2009). Thus, we assessed whether unique synaptic alterations were induced in Fos-expressing accumbens neurons that were strongly activated during cocaine-induced locomotion in sensitized animals and compared them to alterations in the surrounding non-activated accumbens neurons (Koya et al., 2012).

We used c-fos-GFP transgenic mice that were previously developed by Alison Barth (Barth et al., 2004; Barth, 2007; Benedetti et al., 2012; Clem and Barth, 2006; Yassin et al., 2010). The transgene contains a c-fos promoter that induces green fluorescent protein (GFP) in strongly activated Fos-expressing neurons in electrophysiological slice preparations. Cocaine injections on test day induced silent synapses only in GFP-expressing accumbens neurons in mice that were previously sensitized with 5 once daily injections of cocaine (Koya et al., 2012). Silent synapses contain functional NMDA receptors but do not have functional AMPA receptors (Brown et al., 2011; Huang et al., 2009; Lee and Dong, 2011; Lee et al., 2013). In contrast, cocaine injections on test day did not induce silent synapses in GFP-expressing neurons in mice that were not sensitized with repeated cocaine injections (Koya et al., 2012). Thus repeated cocaine injections during sensitization are necessary for silent synapse formation in behaviorally activated neuronal ensembles.

The silent synapses induced in our activated GFP-expressing neurons have different characteristics (Koya et al., 2012) than those induced in randomly selected accumbens neurons following repeated cocaine injections (Brown et al., 2011; Huang et al., 2009; Lee and Dong, 2011; Lee et al., 2013). NMDARs in silent synapses from these randomly selected neurons contain increased levels of GluN2B subunits that are thought to indicate the formation of new synapses with NMDA receptors but no functional AMPA receptors. In contrast silent synapses in our GFP-expressing neurons do not have GluN2B-containing NMDA receptors (Koya et al., 2012). We speculate that silent synapses on GFP-expressing neurons are due instead to AMPA receptor endocytosis, a commonly observed compensatory response to strong synaptic activation (Loweth et al., 2014).

Silent synapses may be just one stage in a dynamic regulation of nucleus accumbens synapses following psychostimulant injections, similar to that observed in the general population of accumbens neurons (Kourrich et al., 2007; Wolf and Ferrario, 2010b). Repeated cocaine injections to rats during sensitization enhance responsiveness of context-specific nucleus accumbens neurons to subsequent cocaine test injections resulting in enhanced Fos expression on test day (Crombag et al., 2002; Hope et al., 2006; Koya et al., 2009; Mattson et al., 2007). We hypothesize that repeated cocaine injections to mice during sensitization similarly enhance responsiveness of selectively activated synapses on accumbens neurons. Strong activation of these synapses on test day initially produces strong activation of these neurons and GFP expression while synapses on the surrounding neurons are less activated and do not induce GFP. AMPA receptors then undergo endocytosis as a compensatory response to stronger activation of the synapses on these GFP-expressing neurons (Bowers et al., 2010; Carroll et al., 2001; Huganir and Nicoll, 2013) that we detect as silent synapses. Previous work suggests that this compensatory response is temporary and that silent synapse will return to a heightened state of responsiveness following drug abstinence (Kourrich et al., 2007), perhaps due to incorporation of more highly conducting calcium-permeable GluR2-lacking AMPA receptors into these synapses (Boudreau and Wolf, 2005; Wolf and Ferrario, 2010a), so that they can induce Fos and GFP expression again. Unfortunately we cannot observe the entire time course of this process in selectively activated neuronal ensembles because they are not labeled with GFP prior to the cocaine injection on test day or for more than 4–6 hours after test day injections (Koya et al., 2012). Nevertheless enhanced neurotransmission at these synapses may play an important role in long-term enhanced responsiveness of neuronal ensembles to drug-paired contexts and cues.

7. Summary

We provided correlational and causal role evidence for the hypothesis that Fos-expressing neuronal ensembles in the prefrontal cortex and nucleus accumbens mediate conditioned drug behaviors in rats. The specific pattern of Fos-expressing neurons is determined by the context and cues associated with the behavior. A critical question is whether these neurons are actually encoding the learned association or merely involved in retrieval of the memory. We do not have the answer to this question; however our working definition of ‘encoding’ neurons are neurons that have undergone long-lasting cue-specific adaptations that alter their activity response, as well as that of the neural network, to cue-related stimuli during learning. Neurons that do not undergo these long-lasting adaptations are not ‘encoding’ neurons; they can be involved in retrieval of the memory or have activity that merely correlates with behavior without playing a significant role in encoding the memory.

We hypothesize that these long-lasting cue-specific adaptations that encode memories can be found in Fos-expressing neuronal ensembles activated during conditioned drug behaviors. We have identified some candidate molecular and electrophysiological adaptations. However we cannot yet manipulate these adaptations selectively in the Fos-expressing neuronal ensembles to demonstrate that they play causal roles in altering the activity response of neurons in these ensembles, or altering the behavioral response, to cue-related stimuli during learning. Future experiments could use our recently developed c-fos-tet_O-Cre transgenic rat system, or the Tet-tag mice from Mark Mayford’s laboratory (Garner et al., 2012; Matsuo et al., 2008; Reijmers et al., 2007), that can induce transgenes in Fos-expressing neurons that were activated during behavior. To understand addiction-related learning, future studies will also have to focus on Fos-expressing neuronal ensembles activated during the development of learning in addition to the current work that examined the expression of already learned behaviors.

In future experiments, we will use our c-fos-tet_O-Cre transgenic rat system to investigate some of the hypotheses generated by in vivo electrophysiology and cellular imaging studies that suggest distinct ensembles encode drug reward versus non-drug rewards such as food (see above). It will be important to directly assess the degree of overlap between these ensembles in these rats as a way to test whether drugs use the same neuronal ensembles that are used in natural behaviors involving non-drug rewards. If these ensembles do not overlap, then it may be possible to selectively target the ensembles (or unique molecular or cellular alterations within them) that encode addiction-related memories in human addicts. We could then use these targets to ablate or attenuate the response of these neuronal ensembles to relapse-inducing cues. Our preclinical data indicate that this is at least theoretically possible with minimal effects on other memories.

Highlights.

• Correlative evidence supports a role for neuronal ensembles in addiction.

• Fos-expressing neuronal ensembles mediate conditioned drug behaviors.

• Unique neuroadaptations are induced within these Fos-expressing ensembles.