Central amygdala circuits in valence and salience processing

Abstract

Behavioral responses to environmental stimuli are dictated by the affective valence of the stimulus, good (positive valence) or bad (negative valence). These stimuli can innately elicit an affective response that promotes approach or avoidance behavior. In addition to innately valenced stimuli, valence can also be assigned to initially neutral stimuli through associative learning. A stimulus of a given valence can vary in salience depending on the strength of the stimulus, the underlying state of the animal, and the context of the stimulus presentation. Salience endows the stimulus with the ability to direct attention and elicit preparatory responses to mount an incentive-based motivated behavior. The central nucleus of the amygdala (CeA) has emerged as an early integration point for valence and salience detection to engage preparatory autonomic responses and behavioral posturing in response to both aversive and appetitive stimuli. There are numerous cell types in the CeA that are involved in valence and salience processing through a variety of connections, and we will review the recent progress that has been made in identifying these circuit elements and their roles in these processes.

Introduction

Surrounded by environmental stimuli, an organism must decide to either approach or avoid a stimulus based on its valence and perceived value. Valence is a defining feature of a stimulus that endows it with the capacity to elicit a positive or negative affective emotional state or response. Stimuli holding positive valence (e.g., food) are rewarding and promote approach behaviors. In contrast, stimuli of negative valence (e.g., painful shock) are aversive and promote avoidance behaviors. Stimuli that predict an appetitive or aversive outcome can also be assigned valence and themselves elicit an affective response. Salience is the ability of a stimulus to direct attention that engages decision-making processes and elicits an incentive motivational response. Salience reflects the strength of the stimulus if it is inherently valenced or the magnitude of the signaled outcome. Thus, salience is not predicated on the positive or negative features of the stimulus but rather its importance.

The amygdala has emerged as a brain region in which neural representations of opposite valences are distinguishable [1] and a site of incentive processing [2] through the integration of sensory inputs to drive behavioral outputs [3, 4]. The amygdala can be divided into multiple subdivisions that are extensions of both cortical and striatopallidal complexes [1, 5–8]. Among these subdivisions, the central amygdala (CeA) is emerging as a potentially critical site for valence and salience processing. The CeA is comprised of three subdivisions, the central or capsular (CeAc), the lateral (CeAl), and medial (CeAm). The CeA is composed of almost exclusively GABAergic medium spiny neurons [9] that are genetically and functionally heterogeneous within the given subdivisions. Many of these cells are highly interconnected within the CeA forming microcircuits [5, 10], but also send long-range projections to other brain regions [11, 12]. Canonically, the CeAc is viewed as the nociceptive amygdala as it receives nociceptive inputs from the spinal parabrachial tract [13]. The CeAl is connected to multiple brain regions, including the lateral (LA) and basolateral amygdala (BLA), and has emerged as a key node for associative memory [14, 15]. The CeAm is predominantly viewed as the major output of the CeA to drive behavioral responses and is implicated in threat memory consolidation [14, 16]. There is considerable evidence to support the distinct functions of the CeA subdivisions; however, these delineations are not always clear or consistent, which may reflect further functional delineations that occur along the rostral-to-caudal extent of the CeA [17]. Thus, the CeA may perform distinct functions based on the differential connectivity of cell types within specific subdivisions and the anatomical locations of these specific cell types.

The CeA is most prominently viewed as a structure critical for the processing of threat stimuli to elicit defensive behavioral responses [18]; however, there is ample evidence to suggest that the CeA also plays a role in appetitive motivations and reward processing [19–22]. The functional duality of the CeA in regulating appetitive and aversive behavior supports a potential role for the CeA in both valence and salience encoding through highly interconnected microcircuits and long-range projections. We will review the evidence that supports the role of the CeA in appetitive and aversive motivational processes and outline neural circuit connections that may be critical for these CeA functions.

The role of the CeA in learned and innate fear

CeA and Threat Conditioning

The most widely used behavioral paradigm for understanding the neural mechanisms of fear learning is Pavlovian threat conditioning, where a neutral stimulus is paired with a noxious unconditioned stimulus (US) to facilitate a conditioned stimulus (CS) association [23]. Once the CS/US association is learned, the CS elicits a conditioned threat response that scales with the intensity of the US [24]. In conventional Pavlovian threat conditioning, the US is typically a footshock, and the CS is either auditory, olfactory, or visual. The early effort to understand the role of the CeA in threat conditioning revealed that the CeA is critical for the expression of conditioned threat responses [25–29] through direct projections to various brain regions mediating both autonomic and motor defensive responses [30]. This original perspective that placed the CeA as a passive output station has been revised based on the observation that functional inactivation at various stages of threat conditioning blocked the acquisition of learning and consolidation of the threat memory, as well as expression [31]. This extended view of the CeA in threat conditioning has been supported by subsequent studies [14, 27, 32, 33] that have further delineated the CeAl as critical for the fear acquisition and the CeAm for fear expression [14].

Gene expression analysis has revealed that the CeA is comprised of heterogeneous cell types both within and across the CeA subdivisions [19, 34]. The development of genetic and optogenetic tools that allow for precise mapping and manipulation of these heterogeneous cell types has helped to refine the role of the CeA in threat conditioning and behavioral regulation. Two of the earliest cell types isolated and characterized for their roles in threat conditioning are somatostatin (SOM) and protein kinase C delta (PKCδ)-expressing neurons (Figure 1). These two cell types are largely non-overlapping and have been estimated to constitute approximately 80–90% of the neurons in the CeAl [10, 34–37]. Studies employing pharmacological, chemogenetic, optogenetic, or molecular methods showed that selective silencing of CeA-SOM neurons led to impaired fear learning showing reduced fear responses to the CS, whereas activation of these neurons sufficiently induced unconditioned and conditioned defensive behaviors [36, 38–40]. Like suppression of CeA-SOM neurons, inhibition of CeA-PKCδ neurons suppress unconditioned and conditioned fear responses [39]; however, the roles of these two cell types in threat conditioning are likely to be at least partially distinct. A significant portion of CeA-PKCδ neurons are inhibited by the CS [35] and activated by the US [39]. CeA-SOM neurons are activated by the CS following conditioning [40, 41]; though it should be noted that a subset of CeA-SOM neurons projecting to the globus pallidus are activated by the US [42], and a significant proportion of CeA-PKCδ neurons have been shown to be activated by the CS [39]. These response profiles are consistent with partially distinct populations that encode negative valence and salience, as will be discussed further below.

Figure 1. Cell types and connections of the central amygdala involved in valence and salience.

Numerous cell types defined by differential gene expression have been isolated and shown to contribute to valence and salience processing in the central amygdala (CeA). There are numerous reciprocal connections to other brain regions that have been identified as playing key roles in these processes. While there are known projections from the CeA to these regions, in many cases the cell types that project to these areas from the CeA or the cell types within these areas that receive CeA input remains to be resolved. The exception to this is projections from CeA-serotonin 2A receptor (Htr2a) cells which synapse onto the parabrachial nucleus (PBN) neurons in turn synapse on CeA-protein kinase C delta (PKCδ neurons [17]. SOM, somatostatin; CRF, corticotrophin releasing factor; Pnoc, prepronociceptin; Nts, neurotensin; Tac2, tachykinin 2; CeAl, lateral CeA; CeAm, medial CeA; CeAc, the central or capsular CeA; PVT, paraventricular thalamus; SNc, substantia nigra pars compacta; VTA, ventral tegmental area.

Since the initial characterizations of CeA-SOM and CEA-PKCδ neurons, numerous other studies using genetic markers to isolate subpopulations within the CeA have been completed; with these markers showing varying degrees of overlap with SOM and PKCδ. These markers include the genes encoding for corticotrophin-releasing factor (Crh), tachykinin 2 (Tac2) or neurokinin B (NkB), prepronociceptin (Pnoc), neurotensin (Nts), calcitonin gene-related peptide receptor 1 (Calcrl), and serotonin receptor 2a (Htr2a) (Figure 1). Of these, neurons expressing Crh gene have been the most extensively studied and demonstrated to play a unique role in conditioned threat processing. The neuropeptide corticotropin-releasing factor (CRF) released from neurons in the CeA facilitates learning in a conditioned threat paradigm in both mice [43] and rats [44], which is anxiogenic [44, 45] and antagonizes immediate fear extinction [46]. It should be noted, however, that chemogenetic activation of these neurons was shown to impair fear learning and promote extinction [47]. The exact nature of this difference is unclear but may reflect anatomical differences in the rostral versus caudal location of the neurons manipulated. Indeed, recent evidence demonstrates that parabrachial inputs to the caudal, but not rostral CeA is aversive and anxiogenic [17].

In addition to facilitating learning, CRF neurons have been shown to scale defensive responses towards flight in response to a serial compound stimulus [40]. In contrast to CeA-CRF neurons, CeA-SOM neurons bias behavior towards passive defensive responses [41]. These findings are consistent with the reciprocal inhibitory connectivity of CeA-CRF and CeA-SOM neurons [40] that would facilitate defensive action selection based on the relative strength of activation of these two populations. In contrast to CeA-CRF and CeA-SOM neurons that are predominantly localized to the CeAl and regulate various aspects of acquisition and expression of conditioned threat responses, CeA-NkB neurons are principally located in the CeAm and through NkB/Nk3R signaling regulate the consolidation of threat memory [16, 48]. How these specific cell types might encode valence and salience to influence threat learning and defensive selection will be discussed below.

CeA in Innate Versus Conditioned Threat

The CeA is involved in the expression of fear and defensive responses that are not gained through associative learning [26, 49, 50]. There appear to be distinct pathways processing innate vs. learned threat [50–53] and a hierarchical relationship between innate and conditioned threat responses [50]. This relationship has been shown to be mediated by serotonin 2A receptor (Htr2a)-expressing neurons in the CeA neurons that are influenced by Htr2a receptor signaling [50]. Calcium imaging in CeA-Htr2a neurons in response to innate versus learned threat odorants revealed that activity in these neurons is inhibited preferentially by the innate threat odorant. In a feeding suppression assay, the presence of either innate or learned threat odorants actively suppresses food consumption in hungry mice. When mice are presented with access to food in the presence of the innate fear odorant in one arm of a Y-maze and in the presence of a learned fear odorant in the other arm, the innate odorant suppresses consumption more than the learned fear odorant; thus, resulting in an increased food consumption in the presence of the learned fear stimulus [50]. Consistent with this observation, inhibition of these neurons, which magnifies the responses to the innate threat odorants, enhances innate fear responses but suppresses learned fear responses [50]. CeA-Htr2a neurons are preferentially biased towards innate threat odorant; however, it is interesting to note that the CeA-Htr2a neurons are inhibited by the innate threat but not the learned threat. It remains to be established whether these neurons are inhibited by the US used to condition the learned threat (e.g., electric footshock). If so, this would suggest that the differences observed related to innate versus learned odorant threats may be due to differences in the representations of the stimuli as a CS (learned threat) or a US (innate threat).

The role of the CeA in appetitive behavior

CeA and Incentive Motivation

Numerous lesion studies have demonstrated that the BLA and CeA regulate appetitive behavior [2, 22, 54]. Evidence suggests that the BLA and CeA work in parallel [55], with the BLA regulating associations between the CS and features of the US to link the emotional significance of the outcome to the predictive stimulus and promote consummatory conditioned responses (e.g., licking, chewing, and flinching) while the CeA links the affective component of the US to the CS to assign a general incentive motivational property to the CS to promote preparatory conditioned responses (e.g., changes in heart rate, blood pressure, approach and avoidance behaviors) [2]. A slight variation of this view has also been proposed with evidence supporting the role of the BLA in acquired value based its requirement for second order conditioning [56] and the CeA regulating orienting towards relevant stimuli and CS-reward association [22, 54].

In support of the role of the CeA in regulating both consummatory and preparatory responses, infusion of the mu-opioid receptor agonist DAMGO induces consummatory behavior towards reward-predicting CS (e.g., nibbles/sniffs of the lever) and enhances food US consumption that is reversed by muscimol-induced inactivation of the CeA [57]. Optogenetic stimulation of all neurons in the CeA has also been shown to enhance reward reinforcement [58–60] and drive preparatory and consummatory predatory behavior [61]. When CeA stimulation is paired with one of two equally preferred rewards, behavior is biased towards consumption of the CeA-paired reward [58]. Non-selective CeA stimulation also promotes consummatory responses directed towards a metal shock rod that is dependent upon shock delivery [58]. This same stimulation drives motivated responses to acquire the aversive US and assigns incentive motivation to a CS that predicts shock delivery [58].

Non-selective optogenetic stimulation of the CeA alone is not reinforcing and does not promote approach to a neutral object indicating that the incentive motivational aspects of this activation require concomitant sensory information from the appetitive or aversive US to elicit the conditioned response [58–60]. Interestingly, if the CeA optical stimulation was paired with CS-US delivery in a passive Pavlovian threat conditioning paradigm, the defensive response to the CS was enhanced [58]. This indicates that the CeA as a whole does not promote behavioral responses that are inherently appetitive (positive) or aversive (negative) but rather is dependent on the active engagement of behavior that results in the US delivery versus passive delivery of the US that is not contingent upon the animal’s behavior. However, it should be noted that isolation of specific cell types within the CeA revealed that activation some populations are reinforcing, and others are aversive [19]; thus, as we have already grown to appreciate, the CeA is not monolithic and there are likely dedicated cell types and pathways for regulating appetitive and aversive motivations that are likely interconnected.

CeA Cell Types and Appetitive Behavior

Consumption of food, including high-calorie palatable food, activates subsets of neurons in the amygdala that can preferentially respond to food-predicting cues or food consumption [62–64]. Several cell types have been identified in the CeA that regulate appetitive responses. Cell types that promote or suppress appetitive behaviors have also been identified in the CeA. As mentioned above, CeA-Htr2a neurons are inhibited by innate threat odorants but not conditioned threat odorants [50]. CeA-Htr2a neurons are partially overlapping with CeA-SOM neurons [20, 50] and are activated during food consumption [20]. In contrast, CeA-PKCδ neurons are activated by satiety signals, visceral malaise, and bitter tastants [65]. Activation of CeA-Htr2a neurons promotes feeding and activation of CeA-PKCδ neurons suppress feeding [20, 65]. Inhibition of these populations has the reciprocal effect [20, 65]. Although these two cell types receive inputs from similar brain regions [20, 65], the largest distinction appears to be in their connectivity to the parabrachial nucleus (PBN). CeA-Htr2a project to the PBN where they form inhibitory synapses with neurons that project to the CeA and synapse onto PKCδ neurons [20] (Figure 1).

Another cell type in the CeA that facilitates palatable food consumption is neurons expressing prepronociceptin (Pnoc), the precursor to the orexigenic opioid-like neuropeptide nociceptin. Pnoc expression in the CeA is largely non-overlapping with other makers, including Prkcd, Crh, Nts, Tac2, and Htr2a but partially overlapping with Sst [66]. Like CeA-Htr2a neurons, CeA-Pnoc neurons are activated during feeding, and this activation occurs in a calorically scalar manner that is more pronounced in response to highly palatable food. Additionally, this activation is partially insensitive to satiety and bitter tastants [66], suggesting that these cells may not be influenced by the activity of CeA-PKCδ neurons. Genetic ablation and chemogenetic inhibition of these neurons selectively reduce high-fat food consumption, which results in weight loss [66]. Stimulation of these neurons is sufficient to promote reinforcement, as measured by the real-time place preference (RTPP) assay and nose poking for self-stimulation of these neurons [66]. These findings suggest that CeA-Pnoc neurons may be a subpopulation that narrows incentive motivation [59] or assigns positive valence to palatable foods [66].

Similar to the orexigenic and rewarding properties of nociceptin and Pnoc-expressing neurons, the neuropeptide neurotensin (NTS) has also been broadly linked to feeding and reward [67]. Neurons in the CeA express numerous other markers to varying degrees [67] but appear to have differential requirements for the consumption of and preference for reinforcing stimuli. Ablation of CeA-NTS neurons decreases ethanol consumption but not sucrose, saccharin, or quinine liquid intake. However, direct activation of these neurons [19], as well as activation of CeA-NTS projections to the PBN, is reinforcing [67]. Interestingly, although CeA-NTS do not appear to be required for consumption of highly palatable liquids, stimulation of CeA-NTS terminals in the PBN is sufficient to enhance consumption of sweet fluids [67]; thus, these neurons may have a yet undetermined role in incentive motivation.

Valence versus salience encoding in the CeA

The CeA is important for the regulation of both appetitive and aversive behavior. The question is whether these functions are fully separable such that positive valence (reward) and negative valence (threat/aversion) are processed through independent pathways and cell types that are mutually antagonistic through direct or indirect inhibitory connections. Alternatively, there may be independent valence systems that do not interact and function to generate either preparatory approach or avoidance “states” to promote consummatory or defensive responses to stimuli that are innately valence or have acquired valence. Lastly, aversive, and appetitive motivation is likely processed through a shared salience encoding system in the CeA. To better understand valence and salience, it is critical to establish a basic framework for distinguishing between these two signals; this is not always unambiguous. In innate valence encoding, the signal will be opposite for oppositely valenced unconditioned stimuli, e.g., activated by appetitive stimuli and inhibited by aversive stimuli (solid line in Figure 2A), or vice versa (dotted line in Figure 2A) but do not respond to conditioned stimuli (Type 1 valence, Figure 2A). CeA-Htr2a neurons a good example as they are inhibited by innate threat stimuli and activated by appetitive stimuli [20, 50] and do not appear to be activated by a CS [17]. Neurons that encode valence will typically maintain responsiveness to the US as long as the US holds affective information but may acquire responsiveness to the CS (acquired valence), such that the CS can elicit an affective response (Type 2 valence, Figure 2B). Such responding was demonstrated for at least a subset of CeA-PKCδ neurons [39]. Salience encoding neurons should respond in the same direction regardless of the valence of the stimulus [68–71]. These cells can acquire responding to the CS [60–62] (Type 1 salience, Figure 2C) but not respond to the US [63] (Type 2 salience, Figure 2D). Alternatively, salience encoding neurons may initially respond to the US but rapidly diminish this responding as salience is transferred to the predictive CS [46] (Type 3 salience, Figure 2E). There are numerous additional experiments that are required to fully resolve the role of the CeA in valence and salience processing; however, considerable evidence exists to suggest that both processes are occurring within the structure.

Figure 2. Types of valence and salience encoding in the CeA.

(A) Type I valence neurons show opposite responding to appetitive and aversive stimuli and do not respond to a CS. (B) Type 2 valence neurons respond positively to appetitive and aversive stimuli and acquire responses to the CS to encode the valence of the associated stimulus. (C) Type 1salience neurons respond in the same direction to either a positive or negatively valenced appetitive or aversive stimulus and acquire responses to encode the salience of the associated stimulus. (D) Type 2 salience encoding neurons do not respond to the US but respond in the same direction to the CS regardless of the valence of the US. (E) Type 3 salience neurons initially respond to the US but fully transition to responding to the CS following conditioning. Possible two different response types are represented with either solid lines or dotted lines.

Valence Encoding in the CeA

Valence is the ability of a stimulus to elicit a positive or negative affective state response either through the inherent nature of the stimulus or the acquired association of the stimulus with a positive or negative affective experience. CeA-PKCδ neurons are strongly activated by aversive stimuli such as footshock [39], anorexigenic signals (cholecystokinin, lithium chloride, and lipopolysaccharide), bitter tastant (quinine), and even satiety [65]. CeA-Htr2a neurons are activated by food [20] and inhibited by an innate threat stimulus [50]. The activations of the cells by stimuli of opposing valence suggest that these neurons indeed reflect the valence of the stimulus (Figure 1, Table 1, and Table 2). This is further supported by the observation that virtually all CeA-PKCδ neurons maintain responsiveness to the aversive US across all conditioning trials and that a subset of CeA-PKCδ neurons develop a response to the CS following consolidation, indicating that negative valence has been assigned to the CS by these neurons [39]. It is currently not known whether CeA-Htr2a neurons develop a response to a stimulus paired with delivery of an appetitive US; however, it is interesting to note that a learned aversive CS does not elicit an inhibitory response in these cells [50], whether this reflects a lack of responsiveness or the sensitivity of the fiber photometry to detect these signals in a subset of neurons is not clear. The monosynaptic reciprocal inhibitory connections between CeA-PKCδ and CeA-Htr2a neurons within the CeA and the inhibitory synaptic input from CeA-Htr2a to PBN neurons that send excitatory projections to the CeA-PKCδ neurons [20, 65] are consistent with valence processing and the ability of cells with opposing valence to influence each other’s activity (Figure 1). The activation of CeA-Pnoc neurons by appetitive stimuli [66] supports a potential role for these cells in valence processing (Figure 1 and Table 2), but it is not clear how these cells respond to negatively valenced stimuli. It is worth noting that a subset of CeA-Pnoc neurons have overlapping expression with Sst [66]; however, we do not know how these Pnoc/Sst co-expressing neurons respond to the appetitive or aversive stimuli. The projections of CeA-Pnoc [66] and CeA-NTS [67] neurons to the PBN, like CeA-Htr2a cells, and the ability of these cells to elicit reinforcement upon activation [66, 67] further supports a role for these neurons in valence processing.

Table 1. Negative valence:

CeA cell types and connections implicated in negative valence encoding.

Region (Cell-type)	Projection	Reference
CeA	PBN→CeA	Han et al., 2015
		Sato et al., 2015
		Bowen et al., 2020
		Cai et al., 2014
	PVT→CeA	Chen et al., 2019
		Do-Monte et al., 2015
		Liang et al., 2020
	VTA→CeA	Zhou et al., 2019
		Jo et al., 2018
CeAl (PKCδ)		Kim et al., 2017
		Yu et al., 2017
		Botta et al, 2015
		Haubensak et al., 2010
	PBN→PKCδ	Campos et al., 2016
CeAl (PKCδ,SOM)		Hunt et al., 2017
CeAl (SOM)		Li et al., 2013
		Yu et al., 2016
	PVT→SOM	Penzo et al., 2015
	SOM→PVT	Penzo et al., 2014
	SOM→Globus pallidus	Giovanniello et al., 2020
CeAl (SOM,CRF)		Fadok et al., 2017
CeAl (CRF)		Sanford et al., 2017
		Jo et al., 2020
		Gafford et al., 2012
		Pomrenze et al., 2019
		Hartley et al., 2019
	PVT→CRF	Li and Kirouac, 2008
CeA (NTS)		Pomrenze et al., 2019
CeAl (Htr2a)		Isosaka et al., 2015
CeAm (Tac2)		Andreo et al., 2016
CeA (GABA)		Pomrenze et al., 2019

Table 2. Positive valence:

CeA cell types and connections implicated in positive valence encoding.

Region (Cell-type)	Projection	Reference
CeA		Mahler and Berridge, 2009
		Warlow et al., 2020
		Robinson et al., 2014
		Tom et al., 2019
	PBN→CeA	Chen et al., 2018
	PVT→CeA	Do-Monte et al., 2017
CeAc	PBN→CeA	Carter et al., 2013
CeAl, CeAm (SOM,CRF,NTS)		Kim et al., 2017
CeAl (CRF)	CRF→VTA	Heymann et al., 2020
CeAl (Htr2a)	Htr2a→PBN	Douglass et al., 2017
CeA (Pnoc)	Pnoc→PBN	Hardaway et al., 2019
CeA (NTS)	NTS→PBN	Torruella-Suarez et al., 2020
CeA (Drd2)		Kim et al., 2018,
CeAm (GABA)		Han et al., 2017

CeA-SOM neurons are activated by aversive CSs [40, 41], and at least a subset of CeA-SOM projections neurons is activated by an aversive US [42]. However, a longitudinal analysis of CeA-SOM neurons during threat conditioning and how they respond to appetitive stimuli has not been completed, making it difficult to assign these neurons to either valence encoding, salience encoding, or both. Given the breadth of Sst expression in the CeA and its overlap with numerous other markers, it is highly likely that these neurons encompass both.

Salience Encoding in the CeA

Salience refers to the ability of the stimulus to capture attention and promote readiness for associative learning [22, 54, 72] by eliciting arousal and engage a motivated pursuit or avoidance of the associated outcome. Of the neurons characterized in the CeA, the CeA-CRF neurons appear to be the most likely to encode salience (Figure 1, Table 1, and Table 2). CeA-CRF neurons are activated by a CS paired with an aversive US [43], like CeA-SOM neurons [40, 41]. Longitudinal analysis of CeA-CRF neuron activity across habituation, conditioning, extinction, and recall provides insight into the potential role of these neurons in salience processing [46]. CeA-CRF neurons are weakly activated by an unexpected neutral stimulus that rapidly habituates. During conditioning a significant number of these neurons are activated by the aversive US but this response rapidly habituates and becomes nearly undetectable as their responses to the CS emerges. During extinction, the activity of CeA-CRF neurons to the CS gradually decline that tracks the extinction of behavioral responses [46]. This is remarkably similar to the responses observed in basal forebrain neurons that respond to motivationally salient cues irrespective of the valence of the US [71]. Consistent with CRF in the CeA playing an important role in the preparatory response, direct microinfusion of CRF into the CeA elevates heartrate [73]. This likely reflects a response that would be elicited by locally derived CRF as these neurons are connected to other cell types within the CeA and other sources of CRF inputs to the CeA are sparse [43].

Additional support for CeA-CRF neurons as providing a salience signal is the observation that silencing these neurons during the presentation of a serial compound stimulus, which bolsters the salience of the stimulus to elicit a flight response [40, 74, 75], can block the flight behavior and bias it towards passive freezing [40]. Optogenetic stimulation of CeA-CRF neurons during CS presentation in a conditioned threat extinction paradigm attenuates extinction learning and is sufficient to reinstate extinguished conditioned response [46]. Intriguingly, using optogenetic stimulation of CeA-CRF neurons that is not aversive or reinforcing in a neutral context, we have observed that CeA-CRF neuron activation in place of a food reward in an instrumental appetitive task is sufficient to prevent the extinction of the instrumental response (Kong, Jo, and Zweifel unpublished observation), similar to what is observed with activation of these neurons preventing extinction of a threat response [39]. Future experiments to determine the response profile of CeA-CRF neurons during appetitive behavior will be key to further unraveling the role of these cells in salience processing. Of note, it has been shown with microdialysis that CRF levels in the CeA increase during feeding [76], further supporting the idea that these neurons also respond to appetitive stimuli.

CeA circuits for valence and salience processing

There are numerous afferent and efferent connections of the CeA that are likely critical for its role in valence and salience processing [1, 11]. It is interesting to note that the CeA shares similar connectivity to other subdivisions of the extended amygdala complex, including the bed nucleus of the stria terminalis (BNST) and nucleus accumbens (NAc) shell, and the shared processing of valence and salience signals by these structures is likely to be essential for affective state regulation [77, 78]. Indeed, the BNST receives input from the CeA, as well and the ventral midbrain, PBN, and paraventricular thalamus (PVT) [79], and activation of specific outputs of the BNST can either be appetitive or aversive [80]. Similarly, the NAc receives inputs from the PVT that are heavily collateralized to the CeA and BNST [78, 81]. NAc activity is also linked to the negative affect of pain [82], and distinct subdivisions of the NAc can either promote approach or avoidance behavior [83]. While there are likely numerous inputs to the CeA and outputs of the CeA that are critical for its role in valence and salience processing, here we will limit our discussion to three brain regions with reciprocal connections to the CeA that have known functions in valence and salience processing.

CeA-PBN

The CeA has reciprocal projections with the PBN (Figure 1), and their connection in positive valence has largely related to projections from the CeA to the PBN [20, 66, 67]. In contrast, the role of this connection in negative valence has most prominently been linked to PBN projections to the CeA. We have already discussed several cell types from the CeA that project to the PBN [20, 66, 67] and will focus our discussion here on PBN projections to the CeA. Within the PBN, there are multiple cell types that have been identified that play important roles in pain processing and negative affect [84–86]. Calcitonin gene-related peptide (CGRP) expressing neurons in the PBN that project to the CeA have been the most extensively studied for their role in appetite suppression and pain [87–92].

PBN-CGRP neurons encode painful stimuli such as footshock [84, 92]. Given the direct connections with spinal nociceptive afferents [93], this indicates that the PBN is a key pathway for relaying negative valence to the CeA for driving negative affect [17, 84, 87, 94, 95]. Indeed, pairing between the tone CS and selective activation of PBN→CeA pathway as the US elicits freezing behavior to the CS [84]. This indicates that the activation of PBN inputs to the CeA is sufficient to induce the acquisition of fear learning [95]. Specifically, when the PBN-CGRP terminals in the CeA are activated, it works as the US in auditory and contextual fear conditioning [17, 84] and generates anxiety-like behaviors [17]. Genetic silencing of the PBN-CGRP terminals in the CeA blocks threat learning indicating that the signal from the PBN to CeA is essential for relaying US information [84, 92]. PBN-CGRP neurons are also activated by a CS paired with a negative US [92]. The exact nature of how the PBN receives CS information is not clear, but this may be an important reflection of the assignment of negative valence to the CS that can subsequently elicit a negative affective response through the PBN.

PBN-CGRP neurons show a robust decrease in activity before and during feeding [92], which may be mediated in part by inhibitory projections from the CeA-NTS, -Htr2a, and -Pnoc expressing neurons. Indeed, selective activation of PBN-CGRP neurons causes anorexia; reduces food intake and leads to starvation [88, 89]. Optogenetic stimulation of retrogradely targeted PBN neurons from the CeA decreases food intake, but chemogenetic inhibition of them increases food intake and even after injection of nausea-inducing lithium chloride (LiCl) or visceral malaise-causing lipopolysaccharide (LPS) [88]. PBN-CGRP neurons form monosynaptic connections onto most PKCδ neurons in the CeA, and CeA-PKCδ neurons express the CGRP receptor [65, 84, 89]. The anorexic role of CeA-PKCδ neurons has been mentioned in the earlier section, and their role is strongly related to PBN-CGRP inputs as 60% of PBN neurons that project to the CeA-PKCδ neurons express CGRP [65].

In addition to suppression of feeding and mediating the affective component of pain through connections to the CeA, PBN-CGRP neurons also mediate the negative affect of visceral malaise that is critical for conditioned taste aversion (CTA). In CTA, aversion to a novel taste develops when the consumption is followed by transient gastrointestinal malaise [90, 91]. Stimulation of PBN instead of LiCl or LPS injection successfully induces CTA to novel food, and genetic inhibition/silencing of PBN-CGRP neurons attenuates CTA [90, 91]. These effects are partially driven from the PBN→CeA circuit as pairing stimulation of PBN-CGRP terminals in the CeA with 5% sucrose induced CTA and reduced food intake [91].

CeA-PVT

The PVT is considered one of the main hubs for emotional processing [96], and neurons in this area appear to encode both valence and salience of stimuli. Imaging of PVT neurons has revealed complex effects. In one study, imaging of the PVT as a whole revealed that the structure largely reflects salience irrespective of valence and responds overall with excitation to both appetitive and aversive stimuli, including to both the CS and US [68]. Single unit electrophysiology recordings in this study showed approximately 2/3 of neurons are salience encoding, and 1/3 are valence encoding [68]. Isolation of specific cell types supports the presence of two classes of cells [69]; however, the directionality of responding was slightly different than reported previously [68]. Gao and colleagues isolated type I (posterior PVT) and type II (anterior PVT) neurons and found that type 1 neurons are activated by aversive stimuli and inhibited by appetitive stimuli, whereas type II neurons are inhibited by both [69]. It should be noted that the stimuli used in these studies were different. Gao et al. used footshock and tail suspension as negative stimuli and social interaction and thermoneutrality as positive stimuli, and these stimuli were not paired with a CS [69]. Zhu and colleagues used water as the positive stimulus in water-restricted mice and air puff as the negative stimulus; both USs were paired with an odorant [68].

Projection-specific analysis of PVT neuron activation also supports a role for these neurons in valence/salience processing. A number of PVT neurons innervate the NAc shell, BNST, and CeA that are strongly collateralized, with a single neuron innervating all three structures [78, 81]. Otis et al. imaged PVT projections to the NAc, which are likely to reflect mostly collateralized neurons, and found that the majority of responsive cells were inhibited by a reward-predicting stimulus, and a smaller number were activated by the CS [81]. The authors did not report the responses of these neurons to the US. The reason for the discrepancies in the response profiles of the PVT in responding to positive and negative stimuli in the above studies is not clear but may reflect the anatomical location of the recordings, or the nature of the stimuli used. Nonetheless, it is interesting to note that neurons within the posterior PVT predominantly innervate the ventromedial NAc shell and more strongly collateralize to the BNST and CeA [78]. Posterior PVT neurons are activated by aversive stimuli and inhibited by appetitive stimuli [69]. Anterior PVT neurons are inhibited by both appetitive and aversive stimuli [69] and innervate the dorsomedial NAc shell [78]. How this relates to the previous observations that the dorsomedial NAc shell promotes approach the ventromedial NAc shell promotes avoidance [83] is not yet clear but is likely to be critical to the role of the PVT in integrating extended amygdala processing of valence and salience.

The duality of the PVT in valence and salience processing is consistent with inputs to this region arising from hippocampal, cortical, hypothalamic, and hindbrain structures [81, 97]. Of further note, it has also been shown that the PBN sends collateral projections to the CeA and PVT [98] and PVT projections to the CeA have been shown to modulate pain [99]; thus, the CeA is likely to receive salience-related information from cortical-PVT pathways and negative valence related information directly from the PBN and through the PBN-PVT pathway.

The PVT forms monosynaptic connections to CeA-SOM [38] and CeA-CRF neurons [77], and CeA-SOM neurons form the reciprocal connections with the PVT [12, 38]. PVT inputs potently modulate CeA activity [12, 38, 69, 100] and are critical for both acquisition and expression of conditioned threat [38, 100, 101]. The exact temporal requirements of the PVT in threat conditioning remain to unresolved [38, 101–103]. For example, Do-Monte and colleagues showed that inactivation of the PVT immediately following threat conditioning did not alter fear expression; however, inactivation at 24 hrs, 7 days, and 28 days did impair expression [101]. Interestingly, inactivation at 7- and 28-days impaired reconsolidation [101]. These authors also found that PVT neurons were more strongly activated by the CS at 24 hrs compared to 2 hrs post-conditioning [101]. Interestingly, Penzo and colleagues found that inhibition of PVT projections to the CeA prior to conditioning impaired fear retrieval 24 hrs later, similar to inhibiting this projection just prior to retrieval [38]. These data suggest that the PVT is critical for fear acquisition and delayed but not early fear expression. Whether these distinct temporal requirements reflect the dual functionality of the PVT in valence and salience processing will be a key area for future investigation.

Like the role of the PVT in threat conditioning, the role of this structure in reward processing also remains to be further established. Do-Monte and colleagues found that silencing the anterior PVT during reward omission increases the extinction burst observed in instrumental conditioning [104]. In contrast, stimulating the PVT attenuates instrumental responding in the presence and absence of reward [104]. They further observed that one population is inhibited by reward presentation but does not respond to reward omission, and a second is activated by reward omission but does not respond to reward presentation. Somewhat perplexing is the observation that inhibition of anterior PVT inputs to the NAc enhanced the extinction burst and inhibition of the anterior PVT inputs to the CeA attenuated the extinction burst [104], given that nearly 75% of PVT neurons were shown to send collateral projections to the NAc and CeA [81]. Whether there are differentially functioning outputs that project to both structures but have differentially weighted inputs to the NAc and CeA, or whether there is some other explanation for this observation remains to be determined.

CeA-VTA/SNc

The ventral midbrain, comprised of the ventral tegmental area (VTA) and substantia nigra pars compacta (SNc), has a crucial role in value-based decision-making [105, 106]. These midbrain nuclei are comprised of neurons that release glutamate, GABA, dopamine, and in some cases, combinations of these neurotransmitters [107] and have reciprocal connections with the CeA [108–111]. The ventral midbrain has long been implicated in valence and salience processing [112, 113] (Figure 1), positioning these nuclei as key regulators of these processes in the CeA through the potential use of multiple neurotransmitter systems.

Midbrain dopamine neurons are potently activated by rewards and reward-predictive stimuli [114], and both dopamine and GABA neurons within the VTA have been shown to be activated by aversive stimuli [49, 115–117] and cues that predict aversive outcomes [118, 119]. Dopamine is essential for Pavlovian threat conditioning through projections to the amygdala [120] and plays a key role in cued threat discrimination [121, 122]. Within the CeA, excitatory and inhibitory dopamine receptors, D1R and D2R, are differentially expressed on a variety of cell types [19, 33], consistent with the potential for dopaminergic modulation of multiple CeA processes. Indeed, similar to their key role in regulating Pavlovian associative fear learning, positive prediction errors by dopamine neurons during cued fear extinction have been shown to play a key role in the fear extinction process [123, 124] by signaling across different timescales [125].

In addition to dopamine-producing neurons, VTA-GABA neurons also project to the CeA [126]. Although glutamate-only neurons in the VTA do not appear to project to the amygdala, neurons that co-release dopamine and glutamate do [126], indicating that dopamine may regulate the CeA through co-neurotransmitter release. It was recently demonstrated that VTA-GABA neurons regulate innate defensive behaviors and that flight responses elicited by activation of these neurons are dependent upon GABA release in the CeA [49]. Thus, the VTA regulates both innate and learned defensive behavior through what could be distinct circuitry.

The exact role of dopamine in the CeA for regulating appetitive behavior is still largely unknown. During feeding, dopamine is increased in the CeA, and antagonism of D2R signaling in the CeA increases food consumption [127]. Consistent with this observation, inactivation of D2R expression in the CeA increases impulsivity in a food-seeking task [128]. Direct activation of VTA dopamine terminals in the CeA does not promote a real-time place preference [121]; however, optogenetic stimulation of subpopulations of CeA neurons that express D1R is sufficient to promote an operant response [19]. Collectively, this suggests that unlike dopamine projections to the NAc, dopamine in the CeA is not inherently reinforcing but does modulate CeA cell types critical for reward reinforcement.

As mentioned above, the CeA forms reciprocal connections with the ventral midbrain [109, 110]. CRF from the CeA to the VTA is anxiolytic and is likely mediated by activation of the CRF receptor CRHR1 [129]. Neurons within the CeA that express Crhr1 facilitate Pavlovian reward association and instrumental reward learning [130]; however, whether CeA projections to the VTA regulate salience processing in a CRF-dependent manner has yet to be established.

In contrast to the VTA, CeA projections to the SNc have been shown to regulate both appetitive and aversive behaviors through what is likely salience processing [70]. Steinberg et al. identified a strong GABAergic projection from the CeA to the lateral SNc [70]. They demonstrate that these GABAergic projections synapse predominantly onto GABA neurons in the SNc, that these neurons are activated by both appetitive and aversive stimuli, as well as cues that predict appetitive and aversive outcomes, and are essential for both appetitive and aversive learning [70]. These findings indicate that the CeA relays salience information from the CeA to the midbrain; however, whether reciprocal connections from the midbrain to the CeA form a similar function is not known. Future experiments designed to determine the temporal dynamics of dopamine release in the CeA during appetitive and aversive behaviors, and how inputs from the VTA and SNc influence valence and salience processing will significantly advance our understanding of this circuitry.

Summary

The CeA has clearly emerged as an important brain region for the convergence of valence and salience signals. There are numerous cell types that respond to valenced and salient stimuli, which through their local and long-range projections, coordinate autonomic and motor responses to mediate affective behavioral responses. In this view, the CeA performs functions as both a critical relay for generating defensive avoidance and consummatory approach responses and as a site of associative learning that permits stimuli with acquired valence and salience to engage preparatory autonomic and motor responses. Currently, we are merely at the tip of the iceberg in terms of understanding the circuitry of the CeA that performs these processes. To begin to fully understand the complexities of CeA function in this area, it will be essential to perform longitudinal recording or imaging from genetically defined CeA populations during novelty habituation, CS/US conditioning, recall, and extinction for both appetitive and aversive stimuli within both the rostral and caudal aspects of the structure. These studies will no doubt reveal distinct response profiles within these populations that may or may not segregate anatomically. It will then be essential to establish the input and output connectivity of these subpopulations to effectively establish their function within the context of the larger valence and salience circuits of the brain.