Contemporary approaches to neural circuit manipulation and mapping: focus on reward and addiction

Abstract

Tying complex psychological processes to precisely defined neural circuits is a major goal of systems and behavioural neuroscience. This is critical for understanding adaptive behaviour, and also how neural systems are altered in states of psychopathology, such as addiction. Efforts to relate psychological processes relevant to addiction to activity within defined neural circuits have been complicated by neural heterogeneity. Recent advances in technology allow for manipulation and mapping of genetically and anatomically defined neurons, which when used in concert with sophisticated behavioural models, have the potential to provide great insight into neural circuit bases of behaviour. Here we discuss contemporary approaches for understanding reward and addiction, with a focus on midbrain dopamine and cortico-striato-pallidal circuits.

1. Introduction

A major goal of systems and behavioural neuroscience is to map complex behavioural states to ever-more well-defined neural systems. In addition to understanding how adaptive behaviour is orchestrated, it is also critical to understand how these systems are altered in states of psychopathology, such as addiction. Over the past decade, we have seen major advances in new techniques for the dissection of neural circuits that can be used in concert with sophisticated behavioural models to elucidate neural bases of behaviour. In this review, we discuss contemporary approaches to understanding the brain systems implicated in reward and addiction, including limitations of these methods, with a focus on insights gleaned about midbrain dopamine and cortico-striato-pallidal systems.

The neural circuitry and psychological processes implicated in addiction are complex and, in some cases, controversial. Over the years, addiction has been hypothesized to involve negative reinforcement and aversive states [1], exaggerated positive reinforcement and hedonic states [2,3], aberrant reward learning or habit formation [4–6], dysfunction in top–down cognitive control [7] and enhanced incentive motivation [8]. These psychological processes rely on neural circuitry that emerging research has increasingly revealed to be heterogeneous [9], with discrete functions or characteristics being applied to increasingly specific sets of neurons based on neuroanatomical subregion [10,11], connectivity [12,13], activity pattern [14] and neurotransmitter content [15,16]. We suggest that progress in understanding the neural circuitry relevant to addiction and related disorders will require the use of new technical approaches that offer a chance to tie complex psychological processes to precisely identified circuits, as well as to manipulate and probe those systems to investigate potential therapeutic avenues.

2. Methods for circuit dissection and function mapping

(a). Targeting and control with optogenetics and chemogenetics

Over the past decade a diverse set of optogenetic tools were developed that allow for rapid and temporally precise manipulation of genetically and/or anatomically defined cells in intact, behaving animals [17]. This approach relies on the transduction of neurons with light-sensitive microbial opsins that, when bathed in light, depolarize or hyperpolarize the cell. Channelrhodopsin-2 (ChR2), a blue-light-activated cation channel, is the most commonly used opsin for rapid neuronal excitation [18–20]. Conversely, halorhodopsin (eNpHR3.0), a chloride pump, and archaerhodopsin (eArch3.0), a proton pump, are commonly used to silence neurons using yellow or green light [21,22]. Opsins can be conditionally targeted via the introduction of an opsin-expressing virus, transduction of which can be limited to a specific anatomical region by introduction of a general-promoter virus that infects any nearby neuron [23,24]. Alternatively, opsin expression can be limited to genetically distinct cells by the introduction of a recombinase-dependent (e.g. Cre) opsin-encoding virus in Cre-driver transgenic mouse or rat lines [25,26]. Retrogradely transported viral vectors delivered into the terminal fields of neurons of interest, or conditional transduction approaches involving the delivery of a recombinase-encoding virus into a terminal field and a recombinase-dependent, opsin-encoding virus near the cell bodies, allow for projection-specific neuron targeting [27,28]. Viral constructs can be used alongside optogenetic and electrophysiological manipulations to visualize anatomical connections and probe functional connectivity [29,30]. Additional methods allow for visualization and manipulation of distinct neuronal ensembles activated during specific behavioural states [14,31].

Combining these approaches can provide detailed mapping and precise control over the signalling of targeted neurons. A growing toolkit of transgenic rodent lines, microbial opsins and methods for light introduction are now available [32–35]. These manipulations are powerful for investigating circuit function on rapid timescales, such as in acute locomotion [36] or anxiety-like behaviour [37], or during discrete environmental events, like reward-predictive cue presentation or reward delivery [38].

While optogenetic approaches are ideal for temporally precise, transient manipulation of neuronal signalling, they are less useful for certain chronic experimental manipulations, because of tissue damage caused by laser-induced heating and phototoxicity, and the necessity to tether animals to a light source (wireless optogenetic tools, e.g. [32], solve this problem but remain uncommon). Chemogenetic approaches provide an alternative method for prolonged neuronal control through the introduction of engineered G protein-coupled receptors, which can be targeted to defined neuronal populations as described above [39–41]. These Designer Receptors Exclusively Activated by Designer Drugs (DREADDs) are muscarinic family receptors modified to respond to an otherwise pharmacologically inert (at least in rodents) ligand, clozapine-N-oxide (CNO) [42]. Administration of CNO systemically, or into a region of interest, excites neural activity, through the G_q-coupled DREADD (hM3Dq), or inhibits neural activity, through the G_i-coupled DREADD (hM4Di) [43–45]. More recently, a new G_i-coupled DREADD based on the kappa-opioid receptor was developed, which is activated by the otherwise inert ligand salvinorin B, thus allowing for bidirectional control of activity with DREADDs sensitive to different ligands [46]. As with optogenetics, the use of transgenic animal lines, and/or combinatorial virus approaches allows for cell-type and projection-specific targeting of DREADD receptors [45]. Chemogenetic approaches are particularly useful for studies requiring sustained tonic control of neurons, or for longitudinal studies, such as those examining weight gain/obesity [47].

Like any method, optogenetic and chemogenetic approaches come with important caveats (discussed in detail elsewhere, see [48–50]). For example, viral vector transduction may alter the baseline physiology and/or morphology of neurons [51,52], opsin or DREADD expression in cells is often toxic, and expression strength changes over time. These issues also complicate comparison of results across experimental cohorts and research laboratories. Perhaps a more fundamental problem, however, relates to the physiological (and hence, behavioural) relevance of optogenetic and chemogenetic manipulations. Chemogenetic approaches induce activity changes over relatively long timescales (in minutes), which also cannot be assumed to mimic natural circuit function. Likewise, current optogenetic approaches activate neurons in bulk synchrony, and at high frequencies, which is not representative of endogenous system or local circuit dynamics. Moving forward, it will be critical to implement activity-guided ‘closed-loop’ circuit manipulations in ways that better approximate, and are more relevant to, natural activity patterns in vivo [49,53].

(b). Optogenetics and chemogenetics in conjunction with electrophysiology

Optogenetic and chemogenetic tools can be used in combination with both in vivo and ex vivo electrophysiology to accomplish a variety of aims. For instance, ex vivo electrophysiology combined with optogenetics can be used to more clearly define circuit connectivity and to investigate synaptic characteristics, including pharmacological effects and plasticity mechanisms, based on defined afferents [54–62]. When combined with in vivo single unit electrophysiology, optogenetics can be useful for identifying neurons based on genetic or hodological characteristics [63] in order to determine how these populations encode task events [64], such as reward-predictive cues [61,65]. This approach could also be used to investigate how defined neuronal populations encode drug-seeking and other drug-related events, including triggers of relapse to drug seeking. Combined in vivo physiology and optogenetics or chemogenetics can also be used to determine the contribution of specific afferents to task-related firing, or to examine how optogenetically induced behavioural effects are reflected in downstream changes in activity [64,66,67].

Combining in vivo electrophysiology with optogenetics and chemogenetics, especially when light is targeted to the same neurons being recorded, is not without potential technical difficulties that warrant consideration. Fibre optic implants used to generate behavioural effects are large in diameter, requiring placement away from recording sites (approx. 200 µm) to avoid damaging neurons [68]. The higher light power required to then achieve sufficient optical stimulation at recording sites may work well for behavioural manipulations, but it can produce problems for recording, such as photoelectric artefacts [69,70], retinal stimulation [63], as well as distortion of waveforms or synchronous firing that prevents accurate unit sorting [68]. Another major limitation of this approach is that, in comparison to imaging techniques (see below), the yield of optogenetically tagged neurons is low. Some of these problems can be minimized. For example, photoelectric effects tend to be a bigger problem when measuring low-frequency data, such as local field potentials [70], and can be removed in some cases with high-pass filtering [69]. Synchronous activity can be reduced by employing longer, structured low-intensity stimulus waveforms (such as sinusoids [71]). Some researchers have reduced problems associated with high light power usage by etching small-core optic fibres to a point and mounting them very close to recording sites so that lower light power can be used [71]. Finally, there are engineering hurdles for the construction of inexpensive, reliable multichannel recording implants that incorporate optical control, and a dearth of field-standard designs limits progress.

(c). In vivo imaging with genetically encoded indicators of neural activity

The combination of optical imaging approaches with genetically encoded indicators of neural activity, such as calcium indicators (e.g. GCaMPs), sensors for synaptic activity, and voltage indicators [72–74], avoids some problems associated with optogenetic identification of electrically recorded cells. Like light-sensitive opsins and DREADDs, activity sensors can be selectively expressed in defined neuronal populations. A major advantage of this approach is the potential to image a large number of identified cells. Two-photon calcium imaging with thinned-skull cranial windows, or with implantable lenses and prisms, can be used to simultaneously record the activity of hundreds (or even more than 10 000 [75]) of neurons for long periods of time, which makes it possible to measure the same neurons across multiple behavioural conditions or during learning [76,77]. Yet, use of two-photon imaging is mostly limited to experiments in which the animal is head-fixed, restricting the range of behavioural tests, and so far is of limited utility for examining activity in deep-brain regions [48,74]. An alternative method, single-photon counting through implanted fibre optics, is suitable for deep-brain imaging, and was recently used to measure activity of direct and indirect pathway striatal neurons selectively expressing the calcium indicator GCAaMP3 [78]. A related fibre optic method, fibre photometry, is sensitive enough to measure changes in calcium signalling from axons, allowing for projection-specific imaging [79]. Unfortunately, the activity of individual neurons cannot be isolated with either of these fibre optic approaches. Miniaturized single-photon epifluorescence microscopes [80–84] that interface with chronic implantable lenses circumvent that limitation, allowing for chronic in vivo calcium imaging in deep-brain structures in freely moving animals, with improved resolution. A recent study [85] used this microendoscopic approach to image calcium transients via GCAMP6m fluorescence, which allowed activity measures from hundreds of genetically defined neurons in lateral hypothalamus (LH) of freely moving mice across multiple days of behavioural testing. This approach revealed that, even within this identified subpopulation, different neurons showed preferential activity during different behaviours. In another recent paper using this approach [86], calcium activity in AGRP neurons in the hypothalamus was measured during feeding in freely moving, food-restricted mice. The authors found nearly all AGRP neurons showed reduced activity upon seeing a food pellet or the initiation of food consumption. Thus, current microendoscopic approaches for in vivo calcium imaging offers a new way to examine the heterogeneity or homogeneity of neural responding in deep-brain structures during behaviour. Two limitations, however, are that unlike two-photon approaches, one-photon imaging does not allow for optical sectioning through tissue, which limits fine neuronal structure analysis, and light scattering or the presence of fluorescence outside the focal plane can impede resolution of single neurons.

Together, optogenetics, chemogenetics and in vivo imaging can be used to activate, silence, measure and map specific neuronal populations on various timescales and under a diverse set of experimental conditions to probe adaptive and pathological behavioural mechanisms [48,87–90]. The ability to define behaviour-relevant neural circuits with this degree of precision represents a major advance in behavioural and systems research. These approaches have recently been applied to studies of reward and addiction in animal models, which we will now discuss.

3. Mapping circuits of adaptive and pathological reward: midbrain dopamine

The neuromodulator dopamine is central to adaptive reward processes, as well as disorders of dysfunctional learning and exaggerated motivation, such as addiction [91–96]. Given that addictive drugs share the ability to elevate dopamine transmission [3,8], researchers have historically assumed that understanding what dopamine ‘does’ will offer some insight into addiction and related psychiatric disorders. This remains a matter of active debate [94,95,97–99]. Prominent theories hinge on whether dopamine signalling serves as a reward-prediction error (RPE) that causes Pavlovian stimulus–reward learning [5] or, rather, if it controls the incentive motivational value (the degree to which things are ‘wanted’) of rewards and associated Pavlovian cues [97]. By extension, in the context of addiction, these theories suggest that drug-evoked dopamine contributes to addiction primarily by producing aberrant learning, such that drugs and drug cues cause an overestimation of reward value that promotes drug-related behaviours. Conversely, drug-evoked dopamine may exaggerate the motivational value of drugs and associated stimuli to spur seeking. Of course, it is possible that some intermediate version of these ideas (and/or others; [6,95,100]) is true, and in behaving animals dopamine has diverse functions in the brain. Nevertheless, exactly how and what dopamine does, and how its function may vary depending on projection or other factors is an important question, as dopamine is integral to drug seeking, consumption and relapse.

Investigating the role of dopamine in these processes, however, is complicated by the anatomical and cellular heterogeneity of dopamine neurons and the circuitry in which they are embedded. While the primary efferent projections from the ventral tegmental area (VTA) and substantia nigra (SN) release dopamine [93], these regions also contain glutamatergic projections, and GABAergic interneurons and projections [15,93,101,102]. Furthermore, dopaminergic neurons, which for a long time were treated as a uniform class, exhibit different functional and physiological characteristics related to projection target [12,13,103,104], and subtypes exist that co-release glutamate and GABA [105–109]. Owing to this complexity, midbrain dopamine systems received attention from optogenetics researchers studying reward processes early on, as this method allows for precise targeting of dopamine neurons within the surrounding heterogeneity. Additionally, the temporal precision conferred by optogenetics is ideal for investigating functional roles for dopamine neurons in temporally discrete events and behavioural states. There are now reliable tools for optogenetic and chemogenetic manipulation of dopamine neuron firing and dopamine release [110–112] in transgenic mice and rats [25,111].

In an early such study, Tsai et al. [111] found that phasic (50 Hz) optogenetic stimulation of dopamine neurons in the VTA of tyrosine hydroxylase (TH)-cre mice using ChR2 was sufficient to produce a conditioned place preference (figure 1a). Tonic (1 Hz) optogenetic stimulation failed to condition behaviour. These data confirmed an extensive literature of electrical stimulation, lesion and pharmacology studies [95]. In another study from Witten et al. [25], TH-cre rats were used to demonstrate that optogenetic activation of dopamine neurons in the VTA (paired with an external cue) supports robust intracranial self-stimulation (ICSS). This result confirmed and extended classic data from electrical stimulation studies suggesting that positive reinforcement is mediated, at least in part, by dopamine signalling [131,132]. In a follow-up study, Steinberg et al. [115] refined this notion further, showing that optogenetic activation of dopaminergic terminals specifically within the nucleus accumbens (NAc) is sufficient for ICSS. Additional studies found similarly robust optogenetic ICSS of substantia nigra dopamine neurons in TH-cre mice [116,117], suggesting that positive reinforcement is broadly represented across dopamine subregions.

Figure 1.

Select reward circuit functions as defined by optogenetic and chemogenetic manipulations. (a) A variety of cortico-striato-pallidal, amygdalar, thalamic and midbrain circuits, and specific cell types within them, contribute to distinct motivated behaviours, such as place preference [85,111,113,114], intracranial self-stimulation [25,62,115–121] and place aversion [79,113,114,120,122–125]. (b) Neural projections that contribute to relapse produced by cocaine-prime, cues or prime + cue [10,126–128], footshock-resistant alcohol seeking [129] and cocaine place preference [130] include prefrontal cortical, striato-pallidal and amygdalar circuits. BLA, basolateral amygdala; BNST, bed nucleus of the stria terminalis; CeA, central nucleus of the amygdala; core, nucleus accumbens core; IL, infralimbic cortex; LHb, lateral habenula; LH, lateral hypothalamus; LDTg, laterodorsal tegmentum; PL, prelimbic cortex; shell, nucleus accumbens shell; SNR, substantia nigra; VP, ventral pallidum; VTA, ventral tegmental area. (Online version in colour.)

More recently, these findings were extended through probing of dopamine neuron function in more complex learning and decision-making tasks using optogenetics. The ability to activate dopamine neurons in a behaviourally and temporally specific manner was key to these experimental approaches. Steinberg et al. [38], for example, found that transient activation (with ChR2) of VTA dopamine neurons in TH-cre rats at the time of reward delivery during an associative blocking paradigm [133] was sufficient to establish a conditioned response to a blocked cue, circumstances in which a response would normally not form. These findings suggested that the artificial dopamine signal can mimic an RPE to support learning, consistent with certain learning theories [5] as well as with results from another recent study reporting RPE-like signals in optogenetically identified dopamine neurons during a cue–reward conditioning task [65]. Future work remains necessary to resolve the role of VTA dopamine signalling in establishing Pavlovian associations, and/or facilitating learning indirectly by modifying the motivational value of Pavlovian cues and rewards [97,134]. In another recent study, Saddoris et al. [135] manipulated NAc dopamine terminals during a delay discounting task wherein the delay to reward and reward magnitude were independently varied. They found that optogenetic stimulation of dopamine terminals in the NAc specifically during the cue was sufficient to increase preference for a delayed reward option over an immediate one, but did not change preference for a large over small reward, suggesting a specific role for accumbens dopamine in delay-based decisions, and that at least in some cases increasing dopamine neuron activity is insufficient to enhance the learned value of lesser rewards.

While these studies are consistent with prior conceptions of dopamine contributions to conditioned reward, the manipulation of non-dopamine neurons within the VTA suggests a contribution of these cell populations to both reward and aversion (figure 1a). Mice will avoid a location in which they receive optogenetic activation of VTA GABA neuron cell bodies (local and projection neurons were not differentiated), suggesting that activity of those neurons promotes aversion, presumably via inhibition of dopamine neurons [122]. However, Stamatakis et al. [118] found that optogenetic activation of VTA neurons projecting to the lateral habenula that release GABA (but not detectable levels of dopamine) promotes place preference and supports ICSS. In a complementary study, Root et al. [123] determined that optogenetic activation of VTA glutamatergic neurons projecting to the lateral habenula is both acutely aversive and also produces conditioned place aversion. Finally, a study by Brown and colleagues showed that GABA projection neurons within the VTA selectively innervate the NAc, and synapse onto cholinergic interneurons, which have been shown to regulate both activity of local medium spiny neurons, as well as neurotransmitter release from nearby terminals [108,136]. Optogenetic activation of these VTA GABA-NAc neurons produced a pause in the firing of cholinergic interneurons within the NAc, which facilitated cue–outcome learning in a fear conditioning task [137]. Together with results indicating that optogenetically defined VTA GABA neurons exhibit opposing firing patterns to dopamine neurons during Pavlovian conditioning [65], these data suggest a complex interaction of dopamine and non-dopamine midbrain neurons in learning and motivational processes.

Moving forward, it will be important to use modern technology to make conceptual advances on the psychology of reward, and in particular move beyond largely confirmatory studies focused on the role of dopamine in reward learning. Missing from currently published work using these techniques, for example, are experiments probing the role of dopamine neurons in regulating the incentive motivational properties of reward-associated cues [91,92], in effort-based responding [100], or the development or expression of habitual responding [138,139].

Notably, there are scant published data from studies using the circuit dissection techniques discussed above to investigate dopamine systems in drug seeking, consumption and models of drug relapse or craving. In one of the only studies to date, Bass et al. [140] found that optogenetic stimulation of VTA dopamine neurons acutely attenuated alcohol consumption in rats, but only when tonic (5 Hz) stimulation was delivered (phasic activation at 50 Hz had no effect). These results illustrate the utility of using approaches that allow for precise control of neural activity patterns. Given the anatomical heterogeneity of these systems, and studies suggesting that dopamine neurons do not have uniform physiological responses to drugs (i.e. [141,142]), future research will likely reveal highly complex dopaminergic involvement in drug-related behaviours.

4. Mapping a core cortico-striato-pallidal circuit that mediates relapse and compulsive drug use

While there is little published research using optogenetics or chemogenetics to study the role of dopamine neurons in behavioural models of addiction, there is more work investigating ‘downstream’ cortico-striato-pallidal circuitry, especially within animal models of relapse. A central problem in the treatment of addiction is the long-term risk of relapse of drug use [2,143] and a variety of animal models have been used in an attempt to capture its many potential causes, such as contextual and discrete drug-paired cues, stress and drugs themselves [2,92,144–146]. Work using these procedures delineated a set of brain structures necessary for multiple forms of drug-seeking response reinstatement/relapse, including cue-induced, stress-induced and drug-primed [144,147–150], leading to the suggestion that a core circuit consisting of dopamine projections to dorsomedial prefrontal cortex (dmPFC) and NAc, dmPFC projections to NAc and NAc projections to ventral pallidum (VP), is critical for reinstatement of drug seeking for most reinstatement triggers. Several recent studies have exploited optogenetic and chemogenetic approaches to map drug-associated behavioural function to this cortico-striato-pallidal circuitry (figure 1b).

(a). Prefrontal cortical inputs to the nucleus accumbens

While the prelimbic cortex, or dmPFC, is often proposed to drive drug seeking and reinstatement [151], it undoubtedly plays a complicated role in cue-reactivity and cognitive control [152], and may contribute to distinct aspects of behaviour depending on projection target. Therefore, the ability to manipulate distinct cortical projections provides a level of specificity that may be critical in delineating this circuit's function. A handful of recent studies have applied optogenetic techniques to investigate the necessity of projections from dmPFC to the NAc core in reinstatement and measures of compulsive drug seeking (figure 1b). Stefanik et al. [126] examined the role of dmPFC projections to the NAc core in cocaine-primed and combined cocaine-primed plus cue-induced reinstatement of cocaine seeking. They found that tonic optogenetic inhibition of dmPFC axonal fibres in the NAc core (using ArchT) suppressed reinstatement of cocaine seeking induced by cocaine-prime alone, or reinstatement induced by cocaine-prime plus cues. Optogenetic inhibition of dmPFC cell bodies themselves (using eNpHR3.0) also reduces reinstatement of food seeking induced by yohimbine stress [153], though it remains unknown whether projections to NAc core are required for stress-induced reinstatement. Optogenetic inhibition of mPFC, as well as insula, inputs to NAc core also reduces measures of compulsive alcohol seeking, including aversion-resistant intake of quinine-adulterated alcohol, and footshock-resistant operant responding for alcohol [129]. The role of medial PFC (mPFC) inputs to NAc core, at least in supporting aversion-resistant alcohol intake, depends on NMDA receptors active at hyperpolarized potentials, which are found selectively at mPFC, but not BLA, synapses onto NAc core neurons in drinking, but not naive, rats [129]. Because these studies inhibited dmPFC or its projections throughout the entire test session, it remains unknown whether dmPFC neuron activity during discrete events (e.g. during cue presentations or reward delivery) is of particular importance for the generation of drug-seeking behaviour. Future studies confining optogenetic inhibition or activation to unique behavioural epochs are needed to provide novel information on what specific aspects of drug-seeking behaviour are mediated by mPFC–NAc projections.

The importance of dmPFC projections to NAc core in reinstatement and compulsive drug seeking is also related to the maturation of silent synapses by the insertion of non-calcium-permeable α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors at dmPFC-NAc core synapses during drug withdrawal, at least for cocaine, as Ma et al. found [127]. When the maturation of silent synapses in the dmPFC to NAc core projection is reversed using an optogenetic long-term depression (LTD) induction protocol, cocaine seeking supported by response contingent cue presentations is reduced. By contrast, the same optogenetic LTD-induction protocol aimed at ventromedial prefrontal cortex (vmPFC) projections to NAc shell results in increased cocaine seeking [127], which is consistent with the hypothesis that vmPFC projections to NAc shell may be important for the suppression of actions owing to extinction or satiation [154–156]. Of note, this study highlights an additional means by which optogenetic and chemogenetic approaches can be used to demonstrate causal roles of specific projections—in this case, by using optogenetic stimulation to change the strength of a particular set of synapses and subsequently querying the effect on later behaviour. This approach is especially powerful when used to reverse changes in synaptic strength that mediate the behaviour in question [127].

This type of approach was first used by Pascoli et al. [54]. Reversal of cocaine-induced synaptic plasticity via optogenetic stimulation at vmPFC-to-NAc shell synapses reversed cocaine-induced behavioural sensitization in mice, demonstrating the feasibility of using optogenetic stimulation to alter select drug-related synaptic changes with concomitant behavioural changes. More recent work from Pascoli et al. [157] focused on the NAc shell showed that cocaine self-administration selectively triggers insertion of calcium-permeable AMPA receptors at mPFC synapses, but not basolateral amygdala or ventral hippocampal synapses, at least at synapses onto D1-dominant NAc shell neurons. Reversal of this change at mPFC–NAc shell synapses, using a 13 Hz optogenetic LTD induction method targeted at ventral hippocampus inputs (which produces a heterosynaptic effect on mPFC inputs), prior to a cued relapse test did not affect the vigour of cue-supported cocaine seeking as indicated by a change in active lever presses, but did impair response discrimination by increasing inactive lever responses. This selective effect on inactive lever presses is not surprising given that Pascoli et al. [157] targeted the NAc shell, which receives greater vmPFC inputs than the NAc core, and that their virus injections may preferentially infect vmPFC neurons. However, simultaneous reversal of synaptic changes at both the vmPFC–NAc shell and ventral hippocampal–NAc shell inputs abolished relapse selectively [157]. This study emphasizes the utility of projection-specific manipulation afforded by optogenetic approaches in providing understanding of the complex control of drug-seeking behaviour, in this case, of the coordinated role of multiple inputs to a given circuit. Overall, the current optogenetic evidence for the role of prefrontal cortex (PFC) projections to NAc in reinstatement is promising, although these studies have so far used relatively simple relapse models.

(b). Nucleus accumbens projections: D1 versus D2

Investigating the role of NAc in craving and relapse to drug seeking is complicated by the presence of two genetically defined populations of neurons that do not entirely coincide with their projection target. In the dorsal striatum, D1-dominant neurons, which contain mainly D1 dopamine receptors and co-transmit substance P and dynorphin, largely project along the direct pathway to the midbrain. Dorsal striatal D2-dominant neurons, which contain mainly D2 dopamine receptors and co-transmit enkephalin and neurotensin, largely project along the separate indirect pathway [158–161]. In the NAc, by contrast, while D2-dominant neurons also largely project to the VP, D1-dominant neurons project to both the midbrain (the ‘direct’ pathway) and the VP (the ‘indirect’ pathway) [162–164]. Therefore, it is important to consider potential heterogeneity of function based both on neurotransmitter content and projection target.

Indeed, the control of relapse by vmPFC to NAc shell projections, described above, is selective to inputs on D1-dominant neurons [157]. Direct manipulations of D1- and D2-dominant neurons themselves, largely in transgenic mice, have also revealed distinct, roughly opposing roles for these neuronal types in drug reward and self-administration. Optogenetic activation of D1-dominant neurons enhances cocaine-conditioned place preference and cocaine-induced locomotion in animals previously exposed to cocaine [130]. Chemogenetic activation of D2 neurons, conversely, suppresses cocaine self-administration, and chemogenetic inhibition of D2-dominant neurons increases the breakpoint of progressive ratio responding for cocaine [165]. Additionally, optogenetic activation of D2-dominant neurons reduces cocaine-conditioned place preference [130]. Repeated optical stimulation of NAc D2 neurons during withdrawal, but not immediately before cocaine administration, attenuates the expression of behavioural sensitization to cocaine [166].

A direct manipulation of D1 versus D2 neurons themselves within a relapse model, however, awaits testing. The lack of progress on this front may be owing at least in part to the challenges associated with studying reinstatement of drug self-administration in mice [167] and the absence of widely available transgenic rat models for studying D1- and D2-dominant neurons. Viral vectors using enkephalin or dynorphin gene promoters have been used to selectively target these neurons in the dorsal striatum of the rat [43], demonstrating opposing effects on behavioural sensitization, but this viral approach has not been used in the NAc in models of relapse or in other tests of more specific reward processes.

(c). Nucleus accumbens projections: ventral pallidum and midbrain

Optogenetic excitation of NAc efferents has been used to dissect the microstructure of NAc core inputs to the midbrain, which may shed some light on potential mechanisms by which this projection could influence reinstatement. By measuring the effects of optogenetic activation of ChR2-expressing NAc projections onto VTA neurons in a slice preparation, Xia and co-workers [55,56] were able to demonstrate that NAc shell inputs to the VTA primarily target non-dopaminergic neurons. They found that while optical stimulation of NAc inputs generated fast inhibitory post synaptic currents (IPSCs) in many VTA neurons, none of these neurons was TH-positive. Additionally, application of the mu-opioid agonist, DAMGO, inhibited the amplitude of IPSCs generated by optical stimulation of NAc inputs to VTA, suggesting that the actions of mu-opioid agonists in the VTA may be mediated in part by disinhibition of VTA GABAergic neurons. More recent work in mice has confirmed that NAc inputs to VTA largely synapse onto VTA GABAergic neurons [168]. Cocaine exposure potentiates these synapses, resulting in reduced activity of VTA GABA neurons. Bocklisch et al. [168] also found that while optogenetic stimulation of NAc inputs to VTA induced large IPSCs in most GAD-positive neurons, smaller IPSCs were induced in 67% of dopaminergic neurons. While a direct inhibitory projection from NAc onto VTA dopamine neurons appears to exist, bulk optogenetic inhibition of NAc inputs to VTA had the net effect of generating an increase in dopamine neuron activity.

Recent work suggests that NAc projections to the VP play an important role in reinstatement. Stefanik et al. [128] demonstrated that ArchT-mediated inhibition of NAc core inputs to the lateral VP throughout the reinstatement session reduced cocaine-plus-cue-primed reinstatement of cocaine seeking. In comparison, optogenetic inhibition of NAc core inputs to substantia nigra had no effect [128]. These results, in combination with other studies discussed above, support a common reinstatement circuitry model that includes dmPFC projections to NAc core and NAc core projections to the VP (figure 1b). Yet, NAc core neurons projecting to VP may not be required for context-induced reinstatement of drug seeking. Recently, Khoo et al. [169] found that excitation (with ChR2) or inhibition (with eNpHR3.0) of NAc core projections to VP had no effect on contextual renewal of beer-seeking responses. Whether the projection from NAc core to VP is differentially involved in reinstatement based on the particular antecedent, or the particular drug, remains unknown. Of note, the necessity of NAc core neurons projecting to the VP in reinstatement, which are primarily GABAergic, is difficult to reconcile with the demonstration that normal neuronal activity in VP is required for reinstatement of drug seeking elicited by cues, stress and drug prime (see below). Promotion of reinstatement by NAc core inputs to VP may alternatively be mediated by substance P, the major co-transmitter present in D1 medium spiny neurons. Careful investigation of specific projections from the NAc to the VP, especially of D1-dominant NAc neurons projecting to the VP, is required to further understand the interactions among NAc and VP in relapse behaviour.

(d). Ventral pallidum projections to ventral tegmental area

A recent paper using chemogenetic-mediated inhibition of VP inputs to VTA suggests that VP subregions and/or microcircuits may differentially contribute to specific types of drug-related behaviours. Mahler et al. [10] expressed the inhibitory DREADD hM4Di either in rostral or caudal VP. Chemogenetic inhibition of rostral VP neurons suppressed cue-induced but not cocaine-primed reinstatement of cocaine seeking. Conversely, similar inhibition of caudal VP suppressed cocaine-primed but not cue-induced reinstatement. Additionally, whereas rostral VP projections to VTA were critical for cue-induced reinstatement, caudal VP neurons projecting to VTA were not necessary for reinstatement, as chemogenetic inhibition of caudal VP fibres in VTA had no effect on either cocaine-primed or cue-induced reinstatement [10]. They [10] also found that cue-induced reinstatement of cocaine seeking requires both rostral VP neurons and VTA dopamine neurons, as chemogenetic ‘disconnection’ of these neurons reduced cue-induced cocaine seeking.

How might VP neurons, the majority of which are GABAergic [170–172], promote reinstatement via their projections to VTA, given the importance of activation of VTA dopamine neurons for reinstatement [10]? Optogenetic activation of VP fibres in VTA produces monosynaptic IPSCs in VTA neurons; approximately half of VTA neurons that show that IPSCs are TH-positive [55]. Mahler et al. [10] reported that application of CNO in the VTA of rats expressing the inhibitory DREADD hM4Di in VP resulted in excitation of ‘type 1’ neurons (putative dopaminergic, but see [12,103,173,174]), and inhibition of ‘type 2’ neurons. If VP inputs to VTA contribute to reinstatement of drug seeking via inhibition of dopaminergic neurons, it suggests that these dopamine neurons may be functionally distinct from the VTA dopamine neurons whose activity is required for reinstatement. Interestingly, while glutamatergic inputs to VTA in particular are implicated in reinstatement of drug seeking, Mahler et al. found that unilateral VTA glutamate blockade and contralateral DREADD-mediated inhibition of rostral VP did not affect reinstatement, suggesting that rostral VP and VTA glutamate influence reinstatement through convergent, rather than serial, circuits. The role of VP in reinstatement, for both rostral and caudal subregions, is likely mediated by projections to other structures as well. The VP sends its densest projection to LH [175], a structure that is of great interest in emerging circuit work focused on reward-seeking behaviour [85], particularly for its projections to the midbrain [61].

(e). Emerging circuits implicated in adaptive and maladaptive non-drug reward seeking

Circuit investigations with optogenetics and chemogenetics have revealed important functions for other VTA/SN afferents, including the LH, bed nucleus of the stria terminalis (BNST), lateral habenula and laterodorsal tegmentum. The LH has long been implicated in reinforcement processes and consummatory behaviour [176,177] and recent papers have begun to dissect LH–VTA projections in this context. Using optogenetics-assisted electrophysiology, Nieh et al. [61] found that distinct neuronal populations within the LH responded to different behavioural epochs within a sucrose-seeking task. A subset of LH–VTA neurons became active when mice were presented with a sucrose-predictive cue, while a separate population responded selectively during a sucrose-seeking behavioural response. Optogenetic stimulation of LH–VTA neurons increased both free access feeding and sucrose seeking in the presence of negative consequences. When the authors selectively targeted either LH^GABA or LH^glut VTA inputs, however, only optogenetic stimulation of the LH^GABA–VTA neurons increased feeding. Complementary to this study, Jennings et al. [85] found that optogenetic or DREADD-mediated activation of LH^GABA neurons increased feeding behaviour and produced place preference. Furthermore, in vivo Ca²⁺ imaging of LH^GABA neurons during a reward consumption task revealed diverse activity patterns. Some LH^GABA neurons were excited as mice engaged with food, while others were inhibited, and yet others showed no response. Optogenetic activation of GABAergic inputs to the LH from the BNST also generates feeding and supports ICSS [119]. Within the BNST to VTA projection, optogenetic stimulation of glutamatergic fibres from BNST to the VTA promotes aversion and anxiety-like behaviours, but selective stimulation of BNST^GABA–VTA projections produces place preference and anxiolytic behavioural effects [113], even though both projections selectively synapse onto non-dopamine VTA neurons. Similar bidirectional control of motivational states has been shown for lateral habenula and laterodorsal tegmentum VTA inputs, which preferentially synapse onto PFC and NAc shell projecting dopamine neurons, to promote aversion and place preference, respectively [114]. Together, recent optogenetic and chemogenetic studies point to complex rules for how midbrain circuits orchestrate motivated behaviour. The details of how these systems promote pathological behaviours, such as drug seeking and relapse, remain a largely unexplored avenue for future research.

5. Questions, limitations and future directions

The use of genetic tools for cell- and projection-specific mapping of anatomy and function within reward circuitry is becoming more commonplace in neuroscience and experimental psychology research, so it is important to take stock of emerging issues and limitations regarding their implementation, and consider goals for the future.

(a). Model system complexity and limitations

Neurons signal using more than one chemical transmission mechanism, including co-transmission of different classes of neurotransmitter from distinct vesicles or co-release from the same vesicle [178–183]. This has long been recognized [181,184], but modern techniques provide excellent tools to detect co-transmission/release while mapping it onto defined neurons. As described above, we now know that some dopamine neurons also transmit glutamate [105,185], as well as GABA [108], even through non-canonical mechanisms [106,186], including co-release [106]. Some neurons in the VTA that project to NAc transmit both dopamine and glutamate, but from spatially distinct synapses on the axon terminals [109]. Furthermore, some neurons projecting from the VTA to the LHb transmit both glutamate and GABA (but not dopamine), establishing an exception to the classic excitatory/inhibitory dichotomy of neuron classification [187]. These studies emphasize the importance of a fundamental question in neuroscience: what defines a neuron's phenotype? This issue is critical for the implementation of genetically encoded tools, the selectivity of which relies on the degree of genetic diversity inherent to the system being studied, because they commonly use a single-gene classification scheme for targeting neurons [188]. Intersectional genetic strategies that target expression based on multiple cell features (e.g. projection and genetic identity) and new transgenic lines that target different promoter systems offer ways forward [188–192].

Genetic specificity and neuronal phenotyping are central to recent studies that highlight important considerations for implementing circuit dissection tools. Lammel et al. [193] found that in the medial aspects of the ventral midbrain of TH-cre and TH-GFP mice, only one-half to two-thirds of yellow fluorescent protein+ (YFP+) neurons actually coexpressed TH, as measured by immunofluorescence. Non-coexpression was highest in medial VTA and midline structures like the interpeduncular nucleus. Dopamine transporter (DAT)-cre mice, conversely, had high specificity in labelling TH+ neurons. Lateral VTA and SN coexpression in these lines was also high, in line with previous reports [194]. Such non-specific or ectopic transgene expression in transgenic lines has been reported elsewhere [195]. These data are significant because they suggest that studies using TH-cre mice may unintentionally label and manipulate non-dopaminergic neurons, clouding our understanding of ‘dopamine-specific’ effects.

While it is important to consider ectopic Cre expression as a limitation of Cre-driver lines and related genetic tools [196,197], recent data comparing neuronal phenotypes in mice and rats suggest a broader issue. Studying outbred mice, Yamaguchi et al. [102] found that a substantial fraction of mouse neurons in the ventral midbrain that were positive for TH mRNA did not have detectable levels of TH protein. TH mRNA+/TH protein-lacking neurons were most prevalent in midline nuclei, the interfascicular nucleus, interpeduncular nucleus and rostral portion of the linear nucleus of the raphe (figure 2). Critically, this mismatch is not found in outbred rats, for whom TH protein is detected in 100% of TH mRNA-containing neurons across ventral midbrain nuclei [101]. Thus, the lack of TH+ cell specificity in TH-cre and TH-GFP lines described in Lammel et al. [193] may be at least partly explained by the (i) unique features inherent to the neurobiology of mice that are recapitulated in transgenic mouse lines and (ii) limitations of relying on protein-staining immunofluorescence to identify the genetic phenotype of neurons (see also [188]). This has significant implications for all promoter-based genetic tools, as it is possible that other neuron types exhibit a mismatch between gene expression and protein synthesis, resulting in false positive labelling and manipulation. More generally, mice are used in the vast majority of genetic targeting studies, and mapping of their neuronal characteristics and function is typically assumed to apply to other species, such as rats. It is important to point out, however, that relying on TH protein levels as a marker of ‘dopamine’ neurons, in both mice and rats, can be also problematic, as while all TH-protein-positive neurons appear to have the ability to synthesize dopamine, not all package it into vesicles for release [101,118]. These results not only highlight general considerations for implementing genetic tools across different species [198], but also suggest that investigators should be careful to avoid over-interpreting the results from such studies.

Figure 2.

Coexpression of tyrosine hydroxylase (TH) mRNA and protein in the ventral midbrain of the mouse. The cytochemical makeup of neurons within the ventral midbrain differs dramatically between mice and rats. In mice, but not rats, many neurons within midline nuclei that express TH mRNA do not produce detectable levels of the protein. IF, interfasicular nucleus; IPN, interpeduncular nucleus; RLi, rostral linear nucleus of the raphe; SNc, substantia nigra compacta; SNr, substantia nigra reticulata; VTA, ventral tegmental area. Data adapted from [102] with permission. (Online version in colour.)

(b). Conceptual limitations

Advanced circuit manipulation methods offer the opportunity to gain unprecedented insight into the neural mechanisms of behaviour, but as yet, at least within the context of learning and reward circuitry, much of the current literature using these techniques consists of studies that largely confirm, and in some notable cases, refine, published data from studies using older methods.

One issue that hinders conceptual innovation is the limited behavioural repertoire used in many studies. This is partly by necessity, given the technical difficulty of adapting some behavioural models to mice, the required model subject for most studies using genetic tools [167]. For instance, mice do not display reliable individual differences in sign-tracking versus goal-tracking [199], behaviours that have been used to dissociate the predictive and incentive value of conditioned cues in rat Pavlovian-conditioned approach paradigms [94,200,201]. These behaviours are known to differentially rely on dopamine signalling [134,202,203] and predict different forms of relapse [150,204,205]. Drug self-administration studies using intravenous catheters [206] and certain impulsivity tasks [207,208] are also much more challenging to conduct in mice. Additionally, some behaviours believed to serve as particularly useful models of human relapse and addiction, including compulsive cocaine seeking in the face of punishment [209], are unexplored in mice.

Increasingly, new tools are being developed to allow manipulation of cell-type specific populations in rats, including transgenic rats, such as the TH-cre and Chat-cre lines [25], though the published behavioural paradigms in which these rats are used are few. Transgenic rat constructs remain commercially unavailable for most major cell types, such as D1 versus D2 neurons, GABAergic and glutamatergic neurons, studies of which continue to require mouse models. We hope that studies of these populations in rats will be improved by new transgenic lines and through further development of cell-type specific viral promoters [43,210]. Finally, researchers continue to expand the range of relevant behavioural tests being conducted in conjunction with optogenetics or chemogenetics in mice, including some models of relapse [157], goal-directed versus habitual behaviour [211] and impulsivity [212].

(c). Translational limitations

Translating the functional circuit mapping information gained from optogenetic and chemogenetic approaches into clinically relevant applications remains a major obstacle. While steps have been made to introduce these methods for use in monkey and even human experiments [69,213], their invasive nature, and the toxicity of viral vectors, will prevent large-scale implementation. A recent study [214] used optogenetics experiments to inform deep-brain stimulation (DBS) parameters within the NAc shell of mice. Creed et al. [214] found that low-frequency electrical DBS in NAc shell, combined with blockade of D1 dopamine receptors, mimicked the effects of low-frequency optogenetic DBS, which had normalized neurotransmission via generation of LTD at mPFC synapses onto D1 neurons. While electrical DBS on its own only had transient effects on sensitization, optogenetic or electrical stimulation combined with D1 antagonism reduced locomotor sensitization to drug up to 24 h after stimulation, similar to optogenetic stimulation alone. Thus, while genetically encoded tools may currently be of limited direct utility as a clinical tool, they may be useful in the development of alternative methods that recapitulate optogenetic or chemogenetic results with therapeutic benefits, and/or to refine the application of currently available neurological manipulation methods like DBS, transcranial magnetic stimulation (TMS) and focused ultrasound. These clinical approaches have shown some promise in producing behavioural changes in human treatment populations, including addicts [215–220], but they lack anatomical and cell-type specificity, and the durability of effects is unclear. Thus, as yet unknown technological innovations are needed to accomplish safe and precise circuit manipulations suitable for human intervention.

6. Conclusion

Contemporary neuroscientific methods of circuit dissection are redefining research on the neural bases of reward and addiction. Critically, these approaches are beginning to provide new insights into the causal contributions of cortico-striato-pallidal and mesolimbic circuitry that were previously impossible. In particular, we are gaining an appreciation for how neural heterogeneity (according to cell type, projection, input or as yet unknown factors) within brain nuclei orchestrates distinct, or even opposing reward-related behaviours.

Given our growing ability to manipulate, measure and map activity in ever-more defined neuronal populations, it has become more important and valuable to be equally diligent in our dissection of the behavioural processes that are associated with activity within circuits. In part, this will involve greater attention to the temporal dynamics of neural activity that is correlated with, or necessary for, precise behavioural events. It will also require the design of behavioural experiments that are not only intended to confirm but also to potentially disprove long-standing psychological hypotheses regarding brain function. Accordingly, it is worthwhile to renew our commitment to improving upon behavioural models such that they better relate to mental health problems. Going forward, the ability to manipulate circuit elements in concert with behaviour will provide a better understanding of the way natural and drug rewards act upon the nervous system, and the critical differences between neural control of adaptive and maladaptive reward seeking.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by NIH grant nos DA036996 (B.T.S.), AA022290 (J.M.R.) and DA035943 and AA014925 (P.H.J.).