Going and stopping: Dichotomies in behavioral control by the prefrontal cortex

Abstract

The rodent dorsal medial prefrontal cortex (PFC), specifically the prelimbic cortex (PL), regulates the expression of conditioned fear and behaviors interpreted as reward-seeking. Meanwhile, the ventral medial PFC, namely the infralimbic cortex (IL), is essential to extinction conditioning in both appetitive and aversive domains. Here we review evidence that supports, or refutes, this “PL-go/IL-stop” dichotomy. We focus on the extinction of conditioned fear and the extinction and reinstatement of cocaine- or heroin-reinforced responding. We then synthesize evidence that the PL is essential for developing goal-directed response strategies, while the IL supports habit behavior. Finally, we propose that some functions of the orbital PFC parallel those of the medial PFC in the regulation of response selection. Integration of these discoveries may provide points of intervention for inhibiting untethered drug seeking in drug use disorders, failures in extinction in Post-traumatic Stress Disorder, or co-morbidities between the two.

The inhibition of aberrant fear in Post-traumatic Stress Disorder (PTSD) and drug-seeking behavior in addiction represent major hurdles in treating these conditions. Furthermore, co-morbidities are commonly reported: For example, cocaine use is associated with anxiety, anxiety attacks, and PTSD¹, suggestive of common or interactive biological etiologies. A better understanding of the overlapping (and non-overlapping) behavioral, cellular, and molecular mechanisms underlying the successful suppression of reward- and fear-related behaviors may result in novel intervention strategies.

This review begins with a brief overview of the neuroanatomy of the mPFC. We then review evidence that the PL prefrontal cortex serves as a “go” structure, energizing the expression of fear- and reward-related behaviors, while the IL compartment guides “stopping”². We next attempt to reconcile the “PL-go/IL-stop” model with evidence that the PL is essential to goal-directed decision making, which can include response inhibition, while the IL supports stimulus-response habits. In the interest of scope, we emphasize studies using Pavlovian fear conditioning and cocaine- and food-reinforced conditioning, in particular those using lesions, stimulation, and inactivation (pharmacological, optogenetic, and chemogenetic) techniques in rodents. We also address the reinstatement of heroin-reinforced behaviors, but we note that a large number of significant reports have been neglected due to length restrictions. Accordingly, we aim to complement, rather than replace, excellent recent reviews on these topics^3–10.

Part 1. Structures and functions of the mPFC

The mPFC has long been considered part of a mesocorticolimbic system involved in both reward- and threat-related conditioning. For example, the dorsal mPFC is essential for maintaining instrumental responding for food when reinforcement availability is uncertain^11,12, while the ventral mPFC is associated with response inhibition following extinction conditioning in both aversive and appetitive domains, as discussed below. The mPFC receives input from the amygdala and hippocampus, particularly the ventral subregion, as well as other limbic structures, positioning it to integrate information regarding salience, value, and contextual cues associated with appetitive and aversive outcomes^13–15. The mPFC in turn innervates amygdalar and striatal structures to regulate, for example, motor output.

The mPFC can be separated into multiple different subregions including the ACC, PL, IL, and the medial oPFC¹⁰ (fig.1), which are differentiated based on efferent and afferent projection patterns. For example, within the ventral striatum, the PL largely innervates the NAC core, while the IL preferentially innervates the shell compartment¹⁰. The PL also projects to both the basolateral and lateral nuclei of the amygdala, while the IL targets the basal, central, and medial compartments, as well as the GABAergic intercalated cell masses that separate these regions^2,16,17.

Figure 1. Subregions of the rodent prefrontal cortex.

The rodent prefrontal cortex includes the PL (yellow), medial oPFC (green), oPFC (blue; highlighting areas commonly referred to as the ventral and lateral oPFC), anterior cingulate cortex (purple), and IL (orange). Outlines of these regions are transposed onto coronal sections from the Mouse Brain Library¹²⁶.

These medial wall structures can also be grouped according to their functional outputs and their positioning along the dorsoventral axis, with the dorsomedial PFC containing the ACC and dorsal PL, and a ventral compartment containing the ventral portion of the PL, the IL, and the medial oPFC². Behaviorally, these dorsal/ventral mPFC networks have been termed “go” and “stop” systems, respectively, that guide behavior².

PL/IL dichotomies in conditioned fear: Going and stopping

In several domains, the PL and IL exert distinct, sometimes opposing, influences over behavior. Perhaps the most commonly cited example of this phenomenon pertains to the expression and extinction of conditioned fear. In fear conditioning experiments, an innocuous stimulus, such as an auditory tone, is paired with an unconditioned stimulus, such as a foot shock. This once-innocuous stimulus, the conditioned stimulus (CS), takes on fear-eliciting properties, which can be inferred in the rodent by conditioned startle or freezing. The term “conditioned fear” most often refers to conditioned freezing, a defensive response to threat. This term is widely used in the field, but complicated by the implication that the experimenter can infer the subjective state of the animal (fear)¹⁸. We will nonetheless use this imperfect phraseology in the present review, in line with current practice.

While the tone-shock association is thought to be stored within the amygdala, the retrieval of this memory and consequent expression of conditioned fear is PL-dependent. For example, PL lesions and inactivation interfere with conditioned fear expression^19–24. Additionally, reducing the activity of PL inhibitory interneurons disinhibits PL output to the BLA and accordingly, enhances the expression of conditioned fear²⁵. Optogenetic studies further indicate that the PL is required for conditioned fear expression and have interestingly revealed that the retrieval of “new” fear memory requires PL-BLA interactions, but “old” fear memory requires PL-thalamic interactions²⁶. Thus, fear retrieval circuits shift with time, but the PL remains a key cortical node for conditioned fear expression (fig.2a).

Figure 2. Connections and functions of the PL and IL.

(a) Connections and functions of the PL and IL are discussed in this review. For example, interactions between the PL and BLA support the expression of conditioned fear, as well as the reinstatement of cocaine-reinforced behaviors following extinction. PL innervation of the NAC additionally promotes the reinstatement of cocaine-reinforced responding, and interactions with the dorsal striatum are associated with goal-directed response selection. The dorsal striatum may also be a site of integration of oPFC- and mPFC-dependent learning and memory. Meanwhile, the IL is essential to fear extinction, particularly the retention of extinction memory in both appetitive and aversive domains, as well as habit formation. PL-dependent fear expression can be blunted by vHC innervation of inhibitory interneurons, sustaining low levels of conditioned fear following extinction conditioning. The authors note that this figure highlights the connections and models discussed in this manuscript and do not represent all anatomical connections between these regions. (b) These interactions are essential to several types of conditioning – response-outcome, stimulus-outcome, and stimulus-response. Examples are provided. (R = response; O = outcome; S = stimulus.)

In fear extinction conditioning, repeated presentation of the CS in the absence of foot shock leads to a reduction, or decay, in startle or freezing. This process is thought to reflect new learning rather than memory erasure^8,27. Early studies indicated that lesions of the mPFC that included the IL do not impair the initial acquisition of extinction conditioning, but rather, extinction retention, resulting in aberrantly high levels of conditioned fear despite extinction training^28,29. This phenomenon has since been replicated using pharmacological and optogenetic inactivation – i.e., IL inactivation during extinction conditioning blocks subsequent extinction memory retrieval^{e.g.,22,30,31} (and see^21,23 in which IL inactivation impaired both extinction conditioning and retrieval). Conversely, optogenetic stimulation of the IL in conjunction with extinction conditioning reduces conditioned freezing both during training and also in subsequent retrieval tests, mirroring the effects of electrical stimulation of the IL^31;cf.,32,33. Additionally, the retention of extinction memory is dependent on new gene transcription and protein synthesis in the IL^34,35.

These and other findings indicate that extinction-induced IL neuroplasticity is necessary for the retention of fear extinction³⁶. This plasticity is rapid, given that IL inactivation immediately following extinction training has failed in some reports to modify subsequent retrieval^22,23. In counter-conditioning procedures, conditioned fear can be blunted by the co-presentation of a separate CS that is not paired with an aversive outcome; for example, this CS can be paired with positive reinforcement. In this case, inactivation of the IL, but not PL, blocks the ability of reward-related cues to mitigate conditioned freezing following extinction conditioning²⁴. Early models suggested that the IL facilitates fear extinction by stimulating inhibitory intercalated interneurons in the amygdala to suppress amygdala output; however, this model is evolving with evidence that a key functional target of the IL is instead the BLA (reviewed⁹) (fig.2a).

Interestingly, whether these general principles translate to avoidance behaviors is unclear. The expression of active avoidance (escaping a foot shock signaled by a CS) can be blocked by reversible inactivation of either the IL or PL, and surprisingly, PL inactivation in this setting can leave conditioned freezing intact^37,38. By contrast, IL inactivation increases conditioned freezing even in the absence of extinction training and also blocks extinction retention^37,38. In other reports, lesions containing both the PL and ACC interfered with the expression of avoidance, but lesions selective to either structure alone had no effects^38,39. In ref.³⁹, lesions selective to the PL were, in contrast, sufficient to reduce conditioned lick suppression, in general accordance with the conditioned fear studies discussed above, and suggesting that the expression of avoidance, specifically, uniquely recruits multiple structures, complicating a simple “PL-go/IL-stop” model.

From fear to reward: Convergence in the PL

As in the context of aversive conditioning, the acquisition, expression, and extinction of “reward-seeking” behaviors can be dissociated in rodents^see,6. For example, in conditioned place preference (CPP) testing, cocaine, for example, is repeatedly paired with a context. Following several pairings, the experimental animal prefers the cocaine-paired context, evidence of knowledge of the context-cocaine association. This stimulus-outcome conditioning parallels classical fear conditioning in that an initially innocuous stimulus, the context, is paired with an experimenter-delivered outcome, cocaine (fig.2b).

In cocaine self-administration studies, mice or rats are placed in conditioning chambers in which they can generate an operant response (e.g., nose poke, lever press, chain pull) reinforced with cocaine. Cocaine is delivered most commonly via intravenous catheter. Often, stimuli such as lights or tones signal delivery. Thus, multiple associations may be formed – that a stimulus predicts an outcome (as in CPP and classical fear conditioning), and that a response produces an outcome (unlike in CPP and classical fear conditioning) (fig.2b). From a translational perspective, an appealing aspect of self-administration paradigms is that laboratory animals will, like humans, volitionally self-administer drugs of abuse. Moreover, relapse-like behavior can be assessed – in this case, the cocaine-reinforced response is first extinguished, or alternatively, the animal may simply undergo forced abstinence. Then, the animal is returned to the drug-associated chamber. The degree to which responding is energized by re-exposure to the drug-associated context, cues, a drug prime, or other stimuli such as stressors models the relapse of drug seeking following abstinence in humans and is termed “reinstatement.”

The mPFC regulates behavioral sensitivity to positive reinforcement. Laboratory animals will self-administer electrical stimulation to the mPFC; cocaine energizes this behavior; and further, stimulation of the mPFC induces CPP, while lesions or inactivation of the mPFC can attenuate cocaine-CPP (reviewed⁶). A large body of research has also been devoted to identifying factors regulating reinstatement, since understanding this process could elucidate the mechanisms driving relapse in clinical populations. Much of this work has implicated the PL as a critical node in driving reinstatement behavior. For example, lesions or inactivation of the PL generally decrease responding in cocaine (though possibly not mephamphetamine⁴⁰) reinstatement tests. These include reinstatement elicited by stressors, which can also be blocked by PL-targeted infusions of a dopamine D1 receptor antagonist^41,42. PL-targeted infusions of a D1 antagonist also interfere with the reinstatement of heroin-reinforced responding⁴³. PL inactivation reduces behavioral sensitivity to a cocaine prime^41,44–47 and cocaine-associated cues^48–51, occluding the reinstatement of cocaine-reinforced behaviors. Notably, PL inactivation failed in at least two reports to impact responding when rats were exposed to drug abstinence but not extinction conditioning^40,52, suggesting a more nuanced role for PL in the absence of extinction conditioning.

Cocaine, amphetamine, and dopamine infusion into the PL reinstates drug-related responding following extinction conditioning^44,53. Moreover, following prolonged withdrawal, cocaine-associated cue presentation preferentially activates PL, and not IL, neurons⁵⁴. Blockade of new protein synthesis in the PL, but not the dorsally-situated ACC, interferes with cocaine-induced reinstatement following the reconsolidation of a cocaine-related memory^55,56. Notably, rats subjected to extinction conditioning or abstinence, and not reconsolidation training (in which a previously consolidated memory is recalled and again consolidated), were unaffected by protein synthesis blockade in one report⁵⁶. These findings could suggest that the mechanisms driving the reinstatement of cocaine seeking following the reconsolidation of a cocaine-associated memory distinctively involve de novo protein synthesis within the PL.

Cocaine-induced reinstatement augments synaptic glutamate release from PL terminals in the NAC core^57,58. Inactivation of PL terminals in the NAC decreases reinstatement and interferes with reinstatement-induced modifications in dendritic spines in the NAC^46,59. Meanwhile, inhibition of BLA-PL interactions or BLA projections to the PL also reduces reinstatement^50,60, and in fear conditioning contexts, BLA silencing occludes the electrophysiological responsiveness of PL neurons to shock-associated tones⁶¹. Thus, PL projections to, and innervation by, the BLA appear to facilitate both conditioned fear expression and cocaine-reinforced responding (fig.2a). Notably, blockade of neuroplasticity in the PL mitigates cocaine-induced reinstatement even in rats considered cocaine-vulnerable, i.e., rats that self-administer at high frequencies⁶². This may be because high rates of cocaine self-administration are associated with immediate-early gene activation in the mPFC-NAC pathway. Meanwhile, resilience is instead associated with immediate-early gene expression in BLA-NAC pathways⁶².

Does the IL impact drug seeking?

Relapse is a major challenge in the successful treatment of cocaine abuse, hence a strong focus in the field on the regulatory mechanisms associated with the reinstatement of cocaine-reinforced behaviors. This is as opposed to their extinction. Nonetheless, some investigations suggest that the IL is involved in the extinction of cocaine seeking. Following cocaine-CPP, optogenetic stimulation of the IL enhances, while inhibition occludes, the extinction of place preference⁶³. Stimulation of AMPA or β2-adrenergic receptor systems immediately following extinction training in cocaine self-administering rats enhances the subsequent retrieval of extinction memory⁶⁴. Meanwhile, inactivation of the IL or β2-adrenergic inhibition interferes with extinction retrieval⁶⁴, paralleling effects in conditioned fear extinction experiments³⁵. Manipulations delayed by 3 hours have no consequences, suggesting that IL plasticity is essential to the consolidation of extinction conditioning⁶⁴. Additionally, these manipulations in the PL have no effects, supporting a “PL-go/IL-stop” model.

Importantly, this “go/stop” model does not negate influences of other regulatory structures. The partial NMDA receptor agonist D-Cycloserine enhances the extinction of drug-seeking behavior in response to cocaine-associated cues, but these actions are attributable to activity in the NAC, rather than IL⁶⁵. Further, the blockade of new protein synthesis in the IL failed in one report to deter extinction conditioning, while inhibition in the subiculum and BLA instead disrupted extinction⁶⁶. This report utilized very few training sessions, and similarly, IL inactivation did not impact the extinction of a cocaine-reinforced response when delivered in conjunction with a single extinction training session in another report⁶⁷. These findings are parsimonious with experiments in which single-unit recordings in the IL of fear-conditioned rats revealed sensitivity to “extinguished” CSs only following successful extinction⁶⁸, and a general model in which extinction-induced IL plasticity is essential for the retention of extinction memory³⁶.

Thus, unlike the PL, the IL does not appear to facilitate responding in reinstatement tests^41,69,44,42. Instead, IL stimulation following extinction conditioning enhances extinction retention, mitigating reinstatement^67,70. Conversely, IL inactivation following extinction conditioning exaggerates the subsequent reinstatement, as well as the spontaneous recovery, of cocaine-reinforced behaviors^67,71. This can be blocked by simultaneous inactivation of the PL, re-establishing response inhibition⁶⁷. Interestingly, the timing of infusions in this report⁶⁷ implicates the IL in the retrieval of the extinction memory, whereas the IL may not be necessary for the retrieval of Pavlovian fear extinction memory, only retention³¹. This report also suggests that PL and IL systems are competitive in some contexts. In another example, mice with selective reduction of the pro-plasticity protein Brain-derived Neurotrophic Factor (BDNF) in the PL have blunted cocaine-CPP and conditioned fear expression, and in the absence of reinforcement, they more rapidly inhibit responding for food than control mice, favoring a presumably IL-dependent extinction strategy^72,73.

Interactions between the IL and NAC shell are thought to be necessary for extinction training to inhibit subsequent cocaine reinstatement behavior⁶⁷ (fig.2a). During a drug-free period following cocaine exposure, silent (immature) synapses linking IL projection neurons to medium spiny neurons in the NAC mature. Interference with this process exaggerates the subsequent reinstatement of cocaine-directed responding, suggesting that synapse maturation in this network facilitates the inhibition of drug-seeking behaviors⁷⁴.

Despite the studies discussed above, several lines of evidence challenge an “IL-stop” model as it pertains to the reinstatement of drug seeking in general. For example, inactivating a subset of IL neurons stimulated by heroin-associated cues, the IL in general (via muscimol/baclofen infusions), or local cannabinoid receptors reduces, rather than disinhibits, the reinstatement of heroin-reinforced responding^75–78. One interpretation may be that the IL promotes the retention of cocaine-extinction memory, but inhibits the retention of heroin-extinction memory. However, blocking GluR2 endocytosis in the IL (but not PL) decreases cue-induced reinstatement of heroin responding⁷⁹, and IL inactivation induces the expression of heroin-CPP⁸⁰, suggesting that IL stimulation, by contrast, can mitigate heroin seeking under some circumstances.

In other reports, IL inactivation reduced, rather than exaggerated, the reinstatement of cocaine-⁴⁵, methamphetamine-⁸¹, and nicotine-directed⁸² responding, and also cocaine-directed responding after abstinence⁵². The reinstatement of alcohol-directed responding can be either disinhibited or delayed by IL inactivation^83,84. Willcocks and McNally⁸⁴ propose, as an alternative to the “IL-stop” model, that the IL gates sensitivity to extinction-associated contextual cues. In another report, selective inactivation of alcohol-cue-stimulated IL neurons exaggerated reinstatement behaviors, while global inactivation did not⁸³. Thus, a focus on context-extinction associations and cell-type-specific influences may lead to a more nuanced understanding of the regulation of reinstatement by the IL. Novel information could reinforce the convergences between fear- and reward-related regulatory systems highlighted in fig.2a, or alternatively, give rise to models that cannot generalize across appetitive and aversive domains, or across distinct drugs of abuse.

Part 2. Actions, habits, and mPFC structures

Reward-related behaviors can be characterized by the associations that support them. For example, responding directed toward obtaining reinforcement can be maintained by the predictive relationship between the response and expected outcome⁸⁵ – a process referred to as goal-directed action selection – while a habit system is instead supported by stimulus-response associations (fig.2b). The actual reinforcer plays a reinforcing function in habitual behaviors, serving to strengthen the stimulus-response association, but it is not encoded as a goal. The incentive salience of drugs of abuse on the one hand, and drug-induced biases towards habit-based response strategies on the other, are both considered factors in the development and maintenance of addiction^86–90. Given that the PL and IL are key regulators in toggling between goal-directed actions and habits^3,7,91, these structures likely regulate both goal-directed and habit-based drug seeking.

The most common way to test whether instrumental responding occurs according to outcome-based (goal-directed) vs. stimulus-response (habit) contingencies is by assessing responding following some alteration in reinforcer value. This can be accomplished by pairing a reinforcer with, for example, lithium chloride-induced malaise. If responding persists despite this devaluation, then responding is independent of outcome value and interpreted as being under the control of a stimulus-response habit. Response inhibition by contrast reflects goal-directed action selection.

Instrumental contingency degradation can also be used to classify response strategies. Here, organisms are typically trained to generate two distinct reinforced responses, such as a nose poke and a lever press. Then, the likelihood that one response will be reinforced is reduced, or “degraded.” Rodents that are sensitive to the predictive relationship between actions and their outcomes – that are goal-directed – will selectively inhibit responding, evidence of knowledge of the response-outcome relationship. By contrast, equivalent engagement of both responses is considered habitual.

The PL regulates reward-related decision making

PL lesions interfere with response-outcome conditioning, resulting in insensitivity to reinforcer value and instrumental contingency degradation^92–94,11. The PL appears necessary for encoding the value of reinforcement, but not necessarily expressing this knowledge, given that inactivation following training or reinforcer devaluation leaves behavioral sensitivity to devaluation intact^95,96,12.

In recent studies aimed at recapitulating the neurobiological effects of early-life stressor exposure, PFC GABA_Aα1 expression was chronically reduced via viral-mediated silencing of Gabra1. Gene knockdown decreased synaptic marker expression, and knockdown in the PL interfered with the acquisition, but not extinction, of a cocaine-reinforced response^97,98, evidence that, as with food, the healthy PL may encode the incentive value of cocaine. Other studies have reported that large lesions of the mPFC that include the PL increase cocaine-reinforced responding during self-administration acquisition, an apparently contradictory effect⁹⁹. The ability to selectively manipulate particular genes in discrete brain regions may help to elucidate disease mechanisms in future investigations.

As implied in the prior paragraph, cocaine seeking can be considered goal-directed in nature, for example, sensitive to disruptions in seeking-taking response chains^100,101, as well as absence or punishment. Inactivation studies indicate that the PL is essential to the (goal-directed) inhibition of cocaine seeking under these circumstances^102–104. Accordingly, PL stimulation can inhibit cocaine-reinforced responding when responding also results in foot shock¹⁰⁴. Interestingly, another report indicated that PL-targeted lesions – rather than stimulations – also inhibited cocaine-reinforced responding in response to foot shock¹⁰⁵. In this report, lesions extended to the medial oPFC, however. This is relevant because inactivation of the medial oPFC reduces cocaine-reinforced behaviors¹⁰⁶, and PL and medial oPFC lesions can have opposite effects in food-reinforced tasks¹². Thus, we argue that the PL can be considered a “go” structure, but because PL function is responsive to outcome value, “going” may include the inhibition of seeking behaviors, i.e., when reinforcers lose value.

The IL supports habit behavior – an adaptive function?

Unlike the PL, lesions of the IL in rats maintain goal-directed behavior despite extensive response training that otherwise causes habits⁹³. Furthermore, IL inactivation following extensive response training reinstates goal-directed behavior after habits have formed¹⁰⁷. In other words, the IL is essential to the acquisition and expression of habit behavior, and this could presumably include cocaine-reinforced habits. Meanwhile, the IL is also essential to extinction memory retention, as discussed above, resulting in the inhibition of conditioned fear and cocaine-reinforced behaviors.

How might we reconcile these findings? It may be that the IL suppresses behavioral sensitivity to previously-learned associations when this behavioral sensitivity is no longer advantageous^cf.,3. For example, in the case of fear conditioning, freezing deprives the animal of valuable opportunities to instead seek food, mates, etc. Thus, when a CS no longer predicts threat, the IL promotes the retention of extinction, allowing for exploratory behavior. In appetitive contexts, habits free attentional resources to attend to other events when a familiar behavior is highly likely to be reinforced. Investigations using T-maze tasks indicate that IL activity is essential for both the acquisition and expression of habit-based response strategies, and even orchestrates toggling back and forth between old and new habits^108,109. These findings suggest that the IL actively promotes behaviorally-advantageous response strategies, rather than simply driving habit- or extinction-based behaviors irrespective of context.

Synergies with oPFC function and final considerations

Within the PFC, the oPFC is positioned both ventrally and laterally to the PL and IL (fig.1). It is inter-connected with the mPFC, as well as amygdalar and hippocampal structures, and across species, it receives information from all sensory modalities, a unique property within the PFC^{10,16,14,110–114}. The connectional networks of the oPFC and mPFC are considered distinct, but notably, none of the focal projections of the various PFC subregions are fully segregated within the striatum¹¹⁵. For instance, PL innervation occupies a large territory of dorsal striatum that overlaps significantly with oPFC innervation¹¹⁵. This provides possible points of convergence in PL- and oPFC-dependent decision making.

The oPFC plays a role in determining reinforcer value and integrating available information regarding outcome features, magnitude, and other characteristics in the service of reward-related decision making^90,113,116. Although the oPFC is largely associated with stimulus-outcome conditioning (again, fig.2b), oPFC neurons encode both stimulus-outcome and response-outcome associative contingencies in non-human primates¹¹⁷. In rodents, several manipulations of the oPFC can occlude response-outcome conditioning – these include lesions¹¹⁴; CaMKII-driven Gi-coupled Designer Receptors Exclusively Activated by Designer Drugs (DREADDs)¹¹⁸; site-selective knockdown of the plasticity-associated neurotrophin Bdnf^118,119, the GABA_Aα1 receptor subunit⁹⁷, and Fragile X Mental Retardation Protein (FMRP)¹²⁰; asymmetric lesions disconnecting the oPFC from the dorsal striatum¹¹⁹; and asymmetric oPFC Bdnf knockdown and BLA lesions¹¹⁸. These findings suggest that some of the functions of the oPFC parallel those of the PL, specifically to support response-outcome learning and memory.

Prolonged exposure to stress hormones causes dendritic spine elimination in the oPFC¹²¹. Spine deficiencies are durable, detectable after a time point when spine modifications have recovered in the mPFC. If the oPFC influences mPFC-dependent behavior, the long-term loss of synaptic contacts in this region could conceivably contribute to durable stress-related failures in, for example, response-outcome goal-directed action selection^122,123. Further characterization of how (and where) mPFC- and oPFC-dependent learning and memory systems intersect could increase knowledge regarding the long-term consequences of stress, drugs, and trauma on aberrant decision-making processes and mental health.

An additional consideration is that the PFC matures considerably throughout postnatal life, with substantial developmental modifications occurring well into adolescence. For example, vHC innervation of the mPFC develops during adolescence, and disruptions in this process impair the ability of these projections to gate competing excitatory projections from the BLA¹²⁴. This would be expected to promote fear expression via disinhibition of the PL⁶¹, and also cue-induced cocaine seeking, given that stimulation of BLA-PL interactions energizes drug-reinforced behaviors⁶⁰ (fig.2a). The protracted developmental trajectory of the PFC may also open a window of opportunity for insults, such as cocaine and traumatic stressors, to cross-sensitize¹²⁵.

Better understanding the circuit-level sequelae, as well as the extracellular and intracellular signaling factors and structural and neurodevelopmental dynamics that determine vulnerability to cocaine, stressors, and trauma may lead to novel approaches to mental illness. Identification of common etiologies between drug- and fear-related disorders may elucidate novel strategies that could treat these illnesses or symptom comorbidities, or that could serve as adjuncts to behavioral intervention therapies. We argue that the “Pl-go/IL-stop” model, while imperfect (for example, unable to fully account for the role of the IL in the reinstatement of drug-seeking or avoidance behaviors), provides a conceptual framework for these future investigations, in which new evidence favoring the model, as well as challenging it, will be instructive.

Box 1. Anatomical abbreviations.

ACC, anterior cingulate cortex; BLA, basolateral amygdala; DLS, dorsolateral striatum; DMS, dorsomedial striatum; IL, infralimbic prefrontal cortex; mPFC, medial prefrontal cortex (a term referring to the structures positioned along the medial wall of the prefrontal cortex including the prelimbic and infralimbic cortices); NAC, nucleus accumbens; oPFC, orbitofrontal prefrontal cortex; PL, prelimbic prefrontal cortex; vHC, ventral hippocampus

Box 2. Does neural remodeling in the PFC correspond with changes in fear-related behaviors?

Dendritic spines are the primary sites of excitatory plasticity in the brain. Do changes in spine density or structure parallel modifications in IL-dependent behaviors? Following fear extinction conditioning, axospinous synapses and dendritic spines in the IL indeed proliferate¹²⁸. The heads of spines on excitatory IL neurons are thin, suggestive of the proliferation of new, immature spines, which could subsequently stabilize and form synapses.

Failures in fear extinction following stressor exposure are conversely associated with dendrite retraction in the IL and dendritic spine loss in a mouse model of Amyotrophic Lateral Sclerosis^129,130, see also^131,121 for stress-related failures in extinction and IL neural structure. Further, impairments in fear extinction following early-life antipsychotic treatment are associated with abnormally high PL dendritic spine densities¹³². In this case, aberrant fear expression may reflect a structural imbalance between PL/IL systems. Accordingly, failures in typical fear expression have been associated with reduced PL dendritic spine density¹³³.

Box 3. Acute cocaine impacts dendritic spines and PL function.

Abundant evidence indicates that cocaine induces dendritic spine proliferation in the mPFC^134,135. Although these studies historically focused on the ACC, experiments using in vivo multiphoton imaging suggest that cocaine induces spine proliferation in the PL, and it is most robust following the first exposure to the drug¹³⁶. Could these rapid changes serve as a neuroanatomical substrate for the encoding of response-outcome associations (e.g., subserving fruitful drug acquisition strategies), or simply disorganize goal-directed response choice? We have previously paired a single cocaine injection with instrumental contingency degradation in mice. Cocaine disrupted new learning regarding the predictive relationship between actions and their consequences, resulting in a deferral to familiar habit-based strategies when mice were subsequently tested drug-free¹²⁷ (fig.3a–c). Injections delayed by 4 hours had no effects, suggesting that cocaine interfered with the consolidation of new response-outcome memory (fig.3c). In a conceptually similar experiment, pairing an instrumental behavior with cocaine accelerated the development of habitual control over that behavior¹³⁷. Thus, cocaine seeking can be goal-directed and PL-dependent, but at the same time, cocaine exposure weakens goal-directed response selection strategies, resulting in a bias towards habit-based behavior.

Under typical circumstances, PL interactions with the DMS are thought to support response-outcome conditioning⁷. By contrast, the DLS regulates habit-based response strategies via interactions with sensorimotor cortex^138,139. Methamphetamine and cocaine decrease dendritic spine density, synaptic marker expression, and activity of the cytoskeletal regulatory element cortactin in the DMS, while spines in the DLS proliferate^140,127 (fig.3d). These patterns of synaptic marker and dendritic spine changes could contribute to drug-induced biases in response strategies that favor stimulus-dependent habit-based systems [see also^141–147 (fig.3e)].

Box 4. IL Dopamine D2: Diverging roles in motor habits and conditioned fear extinction.

Multiple research groups, using multiple techniques, report that a history of repeated exposure to cocaine or amphetamine induces habit-based responding for food when rats or mice are tested drug-free (reviewed box 3 and¹³⁵). Systemic treatment with dopamine D1 receptor antagonists blocks amphetamine-induced habits, while D2 blockade accelerates habit formation¹⁴³. These effects might be attributable to IL dopamine receptor systems, given that IL-specific infusion of D1 inhibitors, D2 agonists, or dopamine itself also occlude habit formation^148,149, potentially by decreasing the excitability of IL pyramidal neurons³. Interestingly, Mueller et al. reported that the D2 antagonist raclopride blocks, rather than enhances, the retention of fear extinction memory¹⁵⁰. With the caveat the limited pharmacological dose ranges have been applied (presumably for feasibility), IL dopamine D2 thus appears to be necessary for extinction-induced IL plasticity on the one hand, but stimulation interferes with IL-dependent habits on the other hand.

Summary.

The prefrontal cortex supports the expression and inhibition of fear- and reward-related behaviors. These dualities are attributable to discrete functional domains making up this brain region, which allow it to stimulate or inhibit behavior, depending on an organism’s experiences. The authors review evidence that supports, or refutes, this “go/stop” function.