Norepinephrine ignites local hotspots of neuronal excitation: How arousal amplifies selectivity in perception and memory

Abstract

Emotional arousal enhances perception and memory of high-priority information but impairs processing of other information. Here, we propose that, under arousal, local glutamate levels signal the current strength of a representation and interact with norepinephrine (NE) to enhance high priority representations and out-compete or suppress lower priority representations. In our "glutamate amplifies noradrenergic effects" (GANE) model, high glutamate at the site of prioritized representations increases local NE release from the locus coeruleus (LC) to generate “NE hotspots.” At these NE hotspots, local glutamate and NE release are mutually enhancing and amplify activation of prioritized representations. In contrast, arousal-induced LC activity inhibits less active representations via two mechanisms: 1) Where there are hotspots, lateral inhibition is amplified; 2) Where no hotspots emerge, NE levels are only high enough to activate low-threshold inhibitory adrenoreceptors. Thus, LC activation promotes a few hotspots of excitation in the context of widespread suppression, enhancing high priority representations while suppressing the rest. Hotspots also help synchronize oscillations across neural ensembles transmitting high-priority information. Furthermore, brain structures that detect stimulus priority interact with phasic NE release to preferentially route such information through large-scale functional brain networks. A surge of NE before, during, or after encoding enhances synaptic plasticity at NE hotspots, triggering local protein synthesis processes that enhance selective memory consolidation. Together, these noradrenergic mechanisms promote selective attention and memory under arousal. GANE not only reconciles apparently contradictory findings in the emotion-cognition literature but also extends previous influential theories of LC neuromodulation by proposing specific mechanisms for how LC-NE activity increases neural gain.

1. Introduction

When jolted by a rough skydiving landing, psychologist James Easterbrook observed that his sense of space and time shrank and slowly re-expanded (Easterbrook 1982). This sparked his curiosity about how arousal influences attention. Later he published a review article in which he argued that under arousal, people rely more on central or immediately relevant information and less on peripheral information (Easterbrook 1959). Since his seminal paper, researchers have accumulated many more observations that arousal evoked by emotional events enhances some aspects of perception and memory but impairs others (for reviews, see Mather & Sutherland 2011; Reisberg & Heuer 2004). For example, victims of a crime tend to remember the weapon vividly but forget the perpetrator's face (Steblay 1992). People also pay attention to emotional information at the expense of neutral information (Dolcos & McCarthy 2006; Knight et al. 2007). These examples fit with Easterbrook's formulation that arousal impairs attention to peripheral information. But arousing stimuli can sometimes enhance memory of peripheral neutral information (Kensinger et al. 2007; Knight & Mather 2009). Thus, although it is clear that arousal shapes attention and memory, knowing that something is neutral or spatially peripheral is not enough to predict how it will fare under emotional conditions.

So, then, how does arousal influence the brain's selection of features to highlight versus suppress? An initial answer to this puzzle was provided by the arousal-biased competition (ABC) model, which posits that arousal does not have fixed rules about which types of stimuli to enhance or suppress. Instead, arousal amplifies the stakes of ongoing selection processes, leading to “winner-take-more” and “loser-take-less” effects in perception and memory (Mather & Sutherland 2011). The ABC model builds on biased competition models proposing that stimuli must compete for limited mental resources (Beck & Kastner 2009; Desimone & Duncan 1995; Duncan 2006). As conceptualized by Desimone and Duncan (1995), both bottom-up and top-down neural mechanisms help resolve competition.

Bottom-up processes are largely automatic, determined by the perceptual properties of a stimulus, and do not depend on top-down attention or task demands. For example, stimuli that contrast with their surroundings, such as a bright light in a dark room, engage attention automatically even if they are currently goal irrelevant (Itti & Koch 2000; Parkhurst et al. 2002; Reynolds & Desimone 2003). Top-down goals can also bias competition in favor of particular stimuli that otherwise would not stand out. Although not included in the original biased competition models, past history with particular stimuli is also a source of selection bias (Awh et al. 2012; Hutchinson & Turk-Browne 2012). For example, one's name or a novel stimulus tends to engage attention (Moray 1959; Reicher et al. 1976). In addition, faces, text, and emotionally salient stimuli all grab attention (e.g., Cerf et al. 2009; Knight et al. 2007; MacKay et al. 2004; Niu et al. 2012).

A core aspect of most current theories of visual attention is that these different signals are integrated into maps of the environment that indicate the priority or salience of stimuli across different locations (Itti & Koch 2000; Soltani & Koch 2010; Treisman 1998). Regions in frontoparietal cortex integrating sensory and top-down signals help represent such priority maps (Ptak 2012). Moreover, having both feedforward and feedback connections between sensory regions and cortical priority maps enables distributed representations of prioritized information to modulate their own processing (e.g., lower-level visual features) even further (Klink et al. 2014; Ptak 2012; Serences & Yantis 2007; Soltani & Koch 2010). Thus, priority signals are self-biasing to enhance efficient information processing in the brain.

In the ABC model, arousal further biases mental processing to favor high- over low-priority representations, regardless of whether initial priority is determined by bottom-up salience, emotional salience, or top-down goals. Thus, because spatially peripheral information is usually lower priority than central information, arousal usually impairs memory for it (Steblay 1992; Waring & Kensinger 2011). Yet, when peripheral information is perceptually salient or goal relevant, arousal instead enhances memory for it (e.g., Kensinger et al. 2007, Experiment 4). But the ABC model does not tackle how this works in the brain. Previous brain-based models of emotion and cognition also do not account for the dual role of arousal. Most models posit that the amygdala enhances perception and memory consolidation of emotionally salient stimuli, but fail to address how arousal sometimes enhances and sometimes impairs information processing.

In this article we propose the glutamate amplifies noradrenergic effects (GANE) model, in which arousal amplifies the activation difference between high- and low-priority representations via local synaptic self-regulation of the locus coeruleus–norepinephrine (LC–NE) system. According to the GANE model, hearing an alarming sound or seeing something exciting leads to a surge in NE release, which, in turn, enhances activity of neurons transmitting high-priority mental representations and suppresses activity of neurons transmitting lower-priority mental representations. As already outlined, priority is determined by top-down goals, bottom-up factors, and high-level stimulus features (Beck & Kastner 2009; Desimone & Duncan 1995; Fecteau & Munoz 2006).

According to the GANE model, the brain's primary excitatory neurotransmitter, glutamate, signals priority. Under arousal, elevated glutamate associated with highly active neural representations stimulates greater NE release, which then further increases glutamate via positive feedback loops. Thus, in these local “NE hotspots,” glutamate signals are amplified. At the same time, wherever NE is released and fails to ignite a local hotspot, inhibitory adrenoreceptors with lower thresholds of activation suppress activity. Higher NE concentration at hotspots also enhances delivery of energy resources to the site of active cognition, synchronizes brain oscillations, and modulates activity in large-scale functional networks. Thus, under arousal, local NE hotspots contrast with widespread NE suppression to amplify priority effects in perception and memory, regardless of how priority was instantiated.

2. Arousal-biased competition in perception and memory

We start by reviewing recent findings supporting Mather and Sutherland's (2011) ABC model and its novel predictions. Next, we turn to the question of how these arousal effects operate in the brain. A fundamental challenge in understanding how arousal influences cognition is that it sometimes enhances and sometimes impairs information processing. Although most emotion research focuses on how processing of emotional stimuli is enhanced compared with neutral stimuli, emotional arousal can also influence processing of neutral stimuli, and across studies, opposing effects are often seen. How can emotionally salient stimuli sometimes enhance memory for what just happened, but other times impair it? When do arousing stimuli enhance perception and when do they impair perception of subsequent stimuli? Many studies report that emotion increases selectivity (for reviews, see Levine & Edelstein 2009; Mather & Sutherland 2011; Murray et al. 2013), but how do we predict what gets selected?

2.1. Arousal enhances perception of salient stimuli, but impairs perception of inconspicuous stimuli

In previous research on how arousal influences subsequent perception, two types of findings were hard to reconcile. First, arousing stimuli impair perception of subsequent stimuli. For example, people preferentially perceive arousing stimuli (e.g., Anderson 2005; Keil & Ihssen 2004) but fail to perceive or encode neutral stimuli close to arousing stimuli either in time (e.g., embedded in a rapid series of images after an arousing image) (Smith et al. 2006) or in space (Kensinger et al. 2007; Tooley et al. 1987). Second, hearing or seeing an arousing stimulus enhances visual perception of a subsequent Gabor patch (Lee et al. 2014a; Padmala & Pessoa 2008; Phelps et al. 2006).

How can we explain both the enhancing and impairing effects of arousing stimuli on perception of stimuli that appear close in time or space? Initial evidence supports the ABC hypothesis that inducing arousal should have two opposing effects on perception: Arousal should enhance processing of high-priority (more salient) stimuli but impair processing of lower-priority (less salient) stimuli. When asked to report as many letters as they could from a briefly flashed array (Fig. 1), participants reported more of the high-salience letters and fewer of the low-salience letters after hearing an arousing emotionally negative sound than after hearing a neutral sound (Sutherland & Mather 2012). Similar results were obtained when arousal was induced by emotionally positive sounds (Sutherland & Mather, under review). These results indicate that arousal makes salient stimuli stand out more than they would otherwise.

Figure 1. Participants heard an arousing or neutral sound before a letter array was flashed briefly. They then reported as many of the letters as they could. Some of the letters were shown in dark gray (high contrast and, therefore, salient) and some in light gray (low contrast and less salient). Participants reported a greater proportion of the salient letters than the nonsalient letters, but this advantage for salient letters was significantly greater on arousing trials than on neutral trials, and the disadvantage for the nonsalient letters was significantly greater on arousing than on neutral trials (Sutherland & Mather 2012).

The ABC model also explains the enhanced processing of emotional stimuli, the focus of most previous theoretical accounts (e.g., Kensinger 2004; LaBar & Cabeza 2006; Mather 2007; Murty et al. 2010; Phelps 2004). People tend to prioritize emotional stimuli due to top-down goals (e.g., increasing pleasure and avoiding pain), their emotional saliency (e.g., associations with reward/punishment), and/or bottom-up salience (e.g., a gunshot is loud as well as a threat to safety [Markovic et al. 2014]). Thus, arousing stimuli should dominate competition for representation at their particular spatiotemporal position (Wang et al. 2012).

If the arousing stimulus appears in the exact same location as a neutral stimulus presented less than a second later, it will impair perception of that neutral stimulus, an effect known as emotion-induced blindness (Kennedy & Most 2012; Most et al. 2005). On the other hand, arousing stimuli tend to enhance the dominance of high-priority stimuli that are nearby but not competing for the same spatiotemporal spot. An emotionally salient word that impairs perception of a subsequent target word flashed in the same location 50 or 500 ms later can instead enhance perception of a target word flashed 1,000 ms later (Bocanegra & Zeelenberg 2009), because after the longer interval, the priority of the target word is no longer overshadowed by the emotionally salient word.

2.2. Arousal enhances perceptual learning about salient stimuli but impairs learning about nonsalient stimuli

Interspersing emotional or neutral pictures with a visual search task had opposite effects on perceptual learning of salient and nonsalient targets (Lee et al. 2012). In this study, the targets were always the same, but in one condition they were salient because they differed from the distractors, and in the other condition they were not salient because they were quite similar to distractors. Emotional images enhanced perceptual learning about the salient target lines but impaired learning of nonsalient targets (Fig. 2). Thus, whether arousal enhanced or impaired learning depended on the target's salience.

Figure 2. Estimated tuning curves for averaged “target” responses as a function of emotion in the high-salience condition (A) and low-salience condition (B). In the high-salience condition, having interspersed emotional pictures enhanced perceptual learning of the exact tilt of the target (55°), whereas in the low-salience condition, emotion impaired learning of the exact tilt of the same target. Figure adapted from Lee et al. (2012).

2.3. How arousal modulates neural representations depends on salience

A recent study took advantage of the fact that faces and scenes activate distinct representational regions in the brain to test the ABC hypothesis that arousal increases brain activation associated with processing of salient stimuli, whereas it decreases brain activation associated with processing of less salient stimuli (Lee et al. 2014b). On each trial, one yellow-framed face and one scene image appeared briefly side-by-side and then a dot appeared in the former location of one of the images (Fig. 3A). The participants' task was to indicate the side on which the dot appeared. Participants responded fastest to dots that appeared behind the salient faces on trials preceded by a tone conditioned to predict shock and thereby induce arousal. In a follow-up functional magnetic resonance imaging (fMRI) study, there was an arousal×saliency interaction in visual category-specific brain regions, such that arousal enhanced brain activation in the region processing the salient stimulus (i.e., fusiform face area) but suppressed brain activation in the region processing the nonsalient stimulus (i.e., parahippocampal place area) (Fig. 3B) (Lee et al. 2014b).

Figure 3. In the functional magnetic resonance imaging study by Lee et al. (2014b), tones conditioned to predict shock (CS+ tones) played before the display of a salient face, and a less salient scene (A) increased activity in the left fusiform face area (FFA) associated with face processing, while decreasing activity in the left parahippocampal place area (PPA) associated with scene processing, compared with tones conditioned not to predict shock (CS– tones) (B). *p < 0.05, **p < 0.005. CS = conditioned stimulus; ISI = interstimulus interval.

2.4. Arousal enhances or impairs memory consolidation of representations depending on their priority

So far, we have focused on how arousal enhances processing of subsequent inputs; however, arousal should have similar effects on mental representations currently active at the moment arousal is induced. Previous research has indicated that arousal induced after initial encoding sometimes impairs and sometimes enhances memory of preceding information (Knight & Mather 2009). The critical ABC hypothesis is that experimental manipulation of priority of information should alter the effect of subsequent arousal on memory consolidation.

In the first study testing this hypothesis, participants viewed lists of objects one object at a time, with one perceptual oddball in each list (Fig. 4) (Sakaki et al. 2014a). The oddball was either emotionally salient or neutral. Some participants were asked to recall the name of the oddball picture as soon as the list presentation ended. In this condition, the object shown just before the oddball (e.g., the cabbage in Fig. 4) was low priority. Other participants were asked to recall the name of the object shown just before the oddball (oddball-minus-1 object). Thus, in this condition, the oddball-minus-1 object (e.g., the cabbage) was high priority. After a series of lists, memory for details of all oddball-minus-1 objects was tested. As predicted, positively or negatively emotionally salient oddball pictures enhanced memory for prioritized oddball-minus-1 objects and impaired memory for nonprioritized oddball-minus-1 objects.

Figure 4. Schematic representations of a neutral trial in the prioritize-oddball condition (A) and a negative trial in the prioritize-oddball-minus-1 condition (B). Memory performance for oddball-minus-1 objects differed as a function of their priority and the valence of oddball pictures (C). Oddball pictures depicted here were obtained from iStockPhoto for illustration purposes and differ from those used in the experiments. Figures from Sakaki et al. (2014a).

Although the brain mechanisms underlying this priority×arousal interaction in memory have yet to be tested, fMRI evidence indicates that arousal enhances activity in regions processing a high-priority stimulus. For example, pairing shock with certain high-priority (i.e., standalone) neutral scenes enhances successful encoding-related activity in the parahippocampal place area (PPA), the brain region specialized to process scene information (Schwarze et al. 2012). Thus, arousal-induced enhancement of brain activity processing prioritized information not only occurs during perception (e.g., Lee et al. 2014b), but also predicts memory for such items.

2.5. Summary

Mather and Sutherland's (2011) ABC model accounts for both the enhancement and impairment effects of arousal on neutral stimuli across a wide variety of experimental contexts. It makes novel predictions: (1) Arousal before exposure to stimuli should amplify the effects of salience on perception and memory encoding; and (2) Arousal shortly after encoding information should amplify the effects of its goal relevance on memory consolidation. Both effects result from arousal's differential modulation of representations depending on priority. Other models also highlight the importance of interactions between arousal, attention, and goals (Kaplan et al. 2012; Levine & Edelstein 2009; Montagrin et al. 2013; Talmi 2013). However, so far there has been no account of how arousal amplifies the effects of priority in the brain.

3. Current brain-based models of arousal's modulatory effects

Before we present our account of how arousal can modulate neural representations differently depending on their priority, we outline how existing brain-based models of arousal and cognition fail to adequately address how arousal has opposite effects depending on representational priority (see Table 1 for an overview).

Table 1. Brain-based emotion-cognition theories.

3.1. Modular vs. “multiple waves” of emotion enhancement in perception

Noticing things like snakes and guns can increase the odds of survival. Consistent with this adaptive importance, emotionally salient stimuli are often detected more rapidly than neutral stimuli (Leclerc & Kensinger 2008; Mather & Knight 2006; Öhman et al. 2001). Explaining the privileged status of emotional stimuli has been the focus of brain models of emotion perception. One common assumption is that the evolutionary value of noticing emotional stimuli led to a specialized emotion module or pathway to evaluate emotional salience (Tamietto & de Gelder 2010). For example, in their multiple attention gain control (MAGiC) model, Pourtois et al. (2013) argue that emotional salience shapes perception via amplification mechanisms independent of other attentional processes. In the MAGiC model, the amygdala and other modulatory brain regions amplify neural responses to emotional relative to neutral stimuli along sensory pathways. The model also posits that these modulations occur parallel to and sometimes in competition with signals from bottom-up (exogenous) and top-down (endogenous) attentional control systems (see also Vuilleumier 2005b).

In contrast, Pessoa and Adolphs (2010) argue against a modular approach to emotion enhancement in perception. In their multiple waves model, affectively and motivationally significant visual stimuli rapidly engage multiple brain sites, including the amygdala, orbitofrontal cortex, anterior insula, and anterior cingulate cortex, that then bias processing to favor these stimuli. From their perspective, the amygdala helps prioritize emotional aspects of information processing by coordinating activity in other regions involved in selective attention. Thus, in the multiple waves model, emotion influences general-purpose perceptual and attention systems rather than harnessing independent brain mechanisms to enhance perception of emotional items.

The latter perspective is more compatible with our findings than are separate-system models; if emotional stimuli were processed via a system separate from that processing neutral stimuli, it is not clear how emotional arousal could have both enhancing and impairing effects on neutral stimuli depending on their priority. However, even this modulatory multiple waves approach to emotion–cognition interactions fails to explain the full picture of how emotional arousal influences cognitive processing, as it focuses only on the enhanced perception of arousing stimuli and ignores how arousal affects perceptual selectivity more generally.

3.2. Canonical amygdala modulation model of emotional memory enhancement

The act of noticing something creates initial trace representations that require additional resources over the next few minutes, hours, and days to consolidate into a longer-lasting memory. Much research indicates that emotional arousal experienced before, during, or after an event can enhance these memory consolidation processes (Hermans et al. 2014). The prevailing view of how emotion affects memory consolidation is that the amygdala enhances processes in the hippocampus and other memory-related brain regions in the medial temporal lobes, such that memory for emotional events is enhanced compared with memory for neutral events (e.g., McGaugh 2004). Consistent with this idea, activity in the amygdala during encoding predicts later memory for emotional items, but not memory for neutral items, as does greater amygdala functional connectivity with medial temporal brain regions (Dolcos et al. 2004; Kilpatrick & Cahill 2003; Richardson et al. 2004; Ritchey et al. 2008).

Converging rodent and human research indicates that NE facilitates the amygdala-mediated enhancement of emotional information. For example, NE released in the amygdala during arousal is associated with enhanced memory for the emotionally arousing event (McIntyre et al. 2002). Infusion of noradrenergic agonists into the basolateral amygdala after training also enhances memory for emotionally arousing events (Hatfield & McGaugh 1999; LaLumiere et al. 2003). In humans, administration of the β-adrenergic antagonist propranolol impairs emotional memories, whereas pharmacological agents that increase NE levels, such as a selective NE reuptake inhibitor, tend to enhance them (Chamberlain & Robbins 2013), and enhanced amygdala activity during encoding emotional stimuli is reduced by propranolol (Strange & Dolan 2004). Thus, NE–amygdala interactions enhance memory for emotional events.

Activation of the amygdala by NE can also impair memory for neutral information encountered near something emotional. For example, as already described earlier in the context of the Sakaki et al. (2014a) study, people often have worse memory for neutral-low priority information shown immediately before an emotional compared with a neutral “oddball” stimulus. Patients with amygdalar damage do not exhibit decrements in memory for neutral words preceding emotional oddball words, and in normal individuals, a β-adrenergic antagonist prevents this retrograde memory impairment (Strange et al. 2003).

Although not usually articulated, the amygdala modulation hypothesis presumably explains these impairment effects for neutral stimuli in terms of a trade-off in which the amygdala focuses resources on emotional stimuli, leaving fewer resources available to process and consolidate the neutral stimuli. However, this trade-off explanation fails to explain how NE–amygdala interactions can also enhance memory for nonarousing information (e.g., Barsegyan et al. 2014; Roozendaal et al. 2008).

3.3. Biased attention via norepinephrine model

In the biased attention via norepinephrine (BANE) model, Markovic et al. (2014) propose that affectively salient stimuli activate the LC–NE system to optimize their own processing. Like the ABC model (Mather & Sutherland 2011), the BANE model builds on biased competition models of attention (Markovic et al. 2014). The BANE model proposes that affect-biased attention “is distinct from both ‘classic’ executive top-down and bottom-up visual attention and is at least in part circumscribed by a different set of neural mechanisms” (Markovic et al. 2014, p. 230). In the BANE model, emotional salience is detected by an “anterior affective system,” including the amygdala and the orbitofrontal cortex, based on the recent history of reward and punishment. In turn, the amygdala's recruitment of the LC–NE system serves as an additional specialized pathway that further biases attention and memory in favor of the affectively relevant information that triggered NE release. However, like other models of emotion and cognition, the BANE model focuses exclusively on how affectively salient stimuli outcompete less salient stimuli and does not address how arousal induced by these stimuli sometimes enhances and sometimes impairs processing of proximal neutral information.

3.4. Emotional attention competes with executive attention for limited mental resources

Another line of work focuses on how emotional stimuli compete for executive resources (Bishop 2007; Choi et al. 2012; Eysenck et al. 2007), with some researchers positing that a ventral affective system competes with a dorsal executive system (Bush et al. 2000; Dolcos et al. 2011). For example, when task-irrelevant emotional stimuli capture attention, they diminish dorsal executive brain region function and therefore disrupt working memory for neutral faces that were just seen (Dolcos & McCarthy 2006; Dolcos et al. 2008). However, meta-analyses indicate that emotional responses are associated with both the ventral and dorsal prefrontal cortical regions (Phan et al. 2002; Shackman et al. 2011), and so the notion that emotional distractors lead the ventral prefrontal cortical region to inhibit the dorsolateral prefrontal cortical region (Dolcos et al. 2008) is unlikely to be universal across different contexts.

Instead of a ventral/dorsal antagonism model, the dual competition model posits that emotional stimuli compete for resources at both perceptual and executive levels of processing (Pessoa 2009; 2013). For example, when participants heard tones predicting shock, regions within the frontoparietal network were activated (Lim et al. 2009). Recruitment of these regions during intense emotional arousal should make them less available for concurrent neutral task-related processing and lead to behavioral impairments. At the perceptual level of the dual competition model, both cortical and subcortical structures help amplify visual cortex responses to emotional stimuli, again leading to the impaired perception of other concurrent stimuli.

As in the ABC framework, competition is a core feature of these models. These models, however, consider only one type of competition: that between arousing and neutral stimuli/tasks. Critically, our empirical results indicate that arousal also influences competition between two neutral stimuli, such that processing of high-priority stimuli is enhanced, whereas processing of lower-priority stimuli is impaired. It is not clear how, in competition models that focus on competition between arousing and neutral stimuli, arousal would interact differently with low- and high-priority neutral information. For example, such models cannot account for the differential effects of arousing sounds on subsequent perceptually salient versus nonsalient letters (Fig. 1).

3.5. Competition between items for memory consolidation

In a different type of competition account, Diamond et al. (2005) propose that there is “ruthless competition” between novel and existing memory representations, such that encoding a new emotional experience suppresses recently potentiated synapses, creating memory for emotional events at the cost of memory for information learned just before the emotional event (Diamond et al. 2005).

This ruthless competition hypothesis argues that the acquisition of new information via the hippocampus depotentiates the most recently activated synapses and that this suppression of recently formed memories is greater when the new information induces emotion or stress. Thus, inducing arousal should impair memory for a preceding sequence of items, regardless of whether those preceding items were themselves emotional or not. That is not the case, however. Inducing arousal via emotional or cold-pressor stress immediately after participants study a mixed list of emotional and neutral pictures selectively enhances memory for preceding emotional, but not neutral, pictures (Cahill et al. 2003; Liu et al. 2008).

3.6. An arousing stimulus sometimes impairs and sometimes enhances memory of what just happened

How can inducing arousal enhance memory for preceding emotional items but not neutral items? Investigators proposed that emotional arousal “tags” synapses associated with representations of emotional items, making these synapses the selective target of protein synthesis-dependent long-term potentiation (Bergado et al. 2011; Richter-Levin & Akirav 2003; Segal & Cahill 2009; Tully & Bolshakov 2010). The emotional tagging hypothesis predicts that emotionally salient stimuli are remembered better than neutral stimuli because emotional tags allow those particular synapses to capture the plasticity-related proteins released with subsequent inductions of arousal.

A problem for the emotional tagging model is that inducing emotional arousal sometimes enhances memory for preceding neutral stimuli (Anderson et al. 2006; Dunsmoor et al. 2015; Knight & Mather 2009; Nielson & Powless 2007; Sakaki et al. 2014a). Neither the emotional tagging hypothesis nor any of the other hypotheses outlined earlier can account for this retrograde enhancement of something neutral. In contrast to the emotional tagging hypothesis, behavioral studies demonstrate that whether something arousing will yield retrograde enhancement or impairment depends on the priority of the preceding information (sect. 2.5) (Ponzio & Mather 2014; Sakaki et al. 2014a).

3.7. Summary

Although there are many models describing how emotion enhances perception, attention, and memory in the brain, these theories fail to account for both the enhancing and impairing effects of emotional arousal (see Table 1 for a summary). In the following sections, we make the case for GANE, a model of how NE released under arousal can impact high- and low-priority representations differently despite its diffuse release across the brain.

4. Locus coeruleus, NE, and arousal

Like the GANE model, other theories also argue that the LC–NE system is important for emotion–cognition interactions (Markovic et al. 2014; McGaugh 2000; 2004; McIntyre et al. 2012). However, they have focused mostly on how NE interacts with the amygdala to enhance processing and consolidation of emotional stimuli at the expense of processing neutral stimuli (e.g., Strange & Dolan 2004; Strange et al. 2003). In contrast, we argue that the LC–NE system promotes selectivity for any prioritized stimuli, irrespective of whether they are emotional or nonemotional.

In this section, we review the functional anatomy of the LC–NE system. A small nucleus in the brainstem known as the locus coeruleus (LC) releases NE when people are aroused – whether by a reward or punishment, a loud noise, or a disturbing image. LC axons are distributed throughout most of the brain (Gaspar et al. 1989; Javoy-Agid et al. 1989; Levitt et al. 1984; Swanson & Hartman 1975), enabling NE to modify neural processing both locally and more globally in large-scale functional brain networks. How does the LC influence information processing in most cortical and subcortical regions? One might think that a hormone released under conditions of arousal would amp up brain activity. But instead, NE quiets most neuronal activity. In turn, this quiet backdrop makes those select few representations that NE amplifies stand out even more.

4.1. Functional neuroanatomy of the LC–NE system

The LC is the primary source of cortical NE and helps determine arousal levels (Berridge & Waterhouse 2003; Berridge et al. 2012; Samuels & Szabadi 2008a; 2008b). Tonic, or background, levels of LC activity help regulate levels of wakefulness (Carter et al. 2010). Phasic, or transient, bursts of LC activity occur in response to novel, stressful, or salient stimuli (Aston-Jones & Bloom 1981; Foote et al.1980; Grant et al. 1988; Sara & Bouret 2012; Sara & Segal 1991; Vankov et al. 1995) or to top-down signals associated with decision outcomes or goal relevance (Aston-Jones & Cohen 2005; Aston-Jones et al. 1999). Emotionally salient stimuli also induce LC phasic activity irrespective of whether stimuli are positive (Bouret & Richmond 2015; Grant et al. 1988) or aversive (Chen & Sara 2007; Grant et al. 1988).

With highly divergent branching axons, the LC projects to every major region of cortex, despite its relatively small number of neurons (13,000 per hemisphere in humans) (Foote & Morrison 1987). Subcortical regions that underlie memory, attention, and emotional processing, including the hippocampus, frontoparietal cortex, and amygdala, are also innervated by the LC (Berridge & Waterhouse 2003). LC axon varicosities release NE into extracellular space, allowing it to activate a broad swath of receptors within a diffusion zone (Beaudet & Descarries 1978; Descarries et al. 1977; O'Donnell et al. 2012).

In target brain sites, NE binds to multiple receptor subtypes (i.e., α₁, α₂, and β receptors) that are located both pre- and postsynaptically on neurons and astrocytes (Berridge & Waterhouse 2003; O'Donnell et al. 2012; Terakado 2014; Tully & Bolshakov 2010). Whereas α₂-adrenoreceptors limit global and local NE release by acting as autoreceptors and decrease cell excitability, β-adrenoreceptor activation generally increases cell excitability, network activity, and synaptic plasticity (Berridge & Waterhouse 2003; Marzo et al. 2009; Nomura et al. 2014; Starke 2001). α₁-Adrenoreceptors recruit phospholipase activation and typically increase cell excitability via the inhibition of potassium channels (Wang & McCormick 1993). Thus, the relative density and localization of adrenoreceptor subtypes help determine how arousal-induced NE release will affect neural processing in different brain regions.

4.2. NE decreases neuronal noise in sensory regions during arousal

In the 1970s, researchers proposed that LC–NE activity enhances signal-to-noise ratios in target neurons in sensory regions (Foote et al. 1975; Freedman et al. 1977; Segal & Bloom 1976; Waterhouse & Woodward 1980). For example, recording from individual neurons in awake squirrel monkeys revealed that NE application reduced spontaneous activity more than it reduced activity evoked by species-specific vocalizations (Foote et al. 1975). Noradrenergic regulation of signal-to-noise ratios is characterized by two simultaneous effects: (1) most neurons in a population decrease spontaneous firing, and (2) the few neurons that typically respond strongly to the specific current sensory stimuli either show no decrease or an increase in firing, unlike the majority of neurons for which the stimuli typically evoke weak responses (Foote et al. 1975; Freedman et al. 1977; Hasselmo et al. 1997; Kuo & Trussell 2011; Livingstone & Hubel 1981; O'Donnell et al. 2012; Oades 1985; Waterhouse & Woodward 1980).

Intracellular recording data in awake animals support and extend these early observations. Both inhibitory and excitatory neurons are depolarized in aroused cortex when mice run (Polack et al. 2013). Yet, consistent with earlier reports of a quieter cortex under arousal, inhibitory neurons are more depolarized than excitatory neurons (Polack et al. 2013). Moreover, surround inhibition dominates sensory responses during wakefulness compared with anesthesia, increasing the speed and selectivity of responses to stimuli in the center of the receptive field (Haider et al. 2013). NE mediates the increase in widespread depolarization and the increase in inhibitory activity in visual cortex that together increase the signal-to-noise ratio (Polack et al. 2013). The effect of NE has also been characterized as increasing the gain on the activation function of neural networks (Fig. 5) (Aston-Jones & Cohen 2005).

Figure 5. Norepinephrine gain modulation makes the non-linear input–output function more extreme, increasing the activity of units receiving excitatory input and decreasing the activity of units receiving inhibitory input. Adapted from Aston-Jones and Cohen (2005).

Arousal is also characterized by cortical desynchronization, both globally when comparing wakefulness with anesthesia (Constantinople & Bruno 2011) or locomotion with being stationary (Polack et al. 2013) and locally among neurons corresponding to attended representations (Fries et al. 2001). Such decreases in cortical slow wave synchrony under arousal are likely mediated by LC activity (Berridge & Foote 1991; Berridge et al. 1993). Synchronous slow wave neural activity may gate sensory inputs, whereas desynchronized activity permits communication of cortical representations of stimuli across the brain (Luczak et al. 2013). Cortical cell depolarization, desynchronization, and increased responsiveness to external input also occur with pupil dilation (Reimer et al. 2014; Vinck et al. 2014), and pupil dilation tracks LC activity (Murphy et al. 2014).

4.3 Summary

Years of research indicate that NE suppresses weak or random neuronal activity, but not strong activity. This is consistent with the increased selectivity seen under arousal (sect. 2). In the next section, we outline a model of how NE has such different outcomes depending on activity level.

5. Glutamate amplifies noradrenergic effects: The core noradrenergic selectivity mechanism under arousal

Now we turn to our GANE model, a novel brain-based account of how arousal amplifies priority effects in perception and memory. We propose that local glutamate–NE interactions increase gain under arousal. Glutamate is the most prevalent excitatory neurotransmitter in the brain (Meldrum 2000). Glutamate receptors such as α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) and N-methyl-D-aspartate (NMDA) receptors mediate rapid excitatory synaptic transmission, neural network connectivity, and long-term memory (Bliss & Collingridge 1993; Lynch 2004; Traynelis et al. 2010).

In addition to point-to-point transmission across a synapse, some glutamate escapes the synaptic cleft, resulting in “glutamate spillover” (Okubo et al. 2010). In this section, we outline evidence that glutamate spillover attracts and amplifies local NE release via positive feedback loops. These self-regulating NE hotspots generate even greater excitatory activity in the vicinity of synapses transmitting high-priority representations, in contrast with NE's suppressive effects in the more widespread non–hotspot regions.

5.1. The NE hotspot: How local NE–glutamate positive feedback loops amplify processing of high-priority information

5.1.1 High glutamate activity stimulates adjacent NE varicosities to release more NE

The first demonstrations of glutamate-evoked effects on NE found that glutamate increased NE release via NMDA and non-NMDA glutamate receptors on LC axons (Fink et al. 1989; Göthert & Fink 1991; Lalies et al. 1988; Nelson et al. 1980; Pittaluga & Raiteri 1990; 1992; Vezzani et al. 1987; Wang et al. 1992; see also Jones et al. 1987). In these studies, glutamate-evoked NE release occurred for NE varicosities in all cortical structures investigated in vitro: olfactory bulb, hippocampus, and throughout neocortex. In vivo experiments replicated the effect with targeted glutamate in rodent prefrontal cortex (Lehmann et al. 1992). Other neurotransmitters associated with arousal, such as histamine (Burban et al. 2010) and orexin (Tose et al. 2009), enhance glutamate-evoked NE release. Central to our hypothesis, glutamate-evoked NE release occurs in human neocortex (Fink et al. 1992; Luccini et al. 2007; Pittaluga et al. 1999).

How do these glutamate–NE interactions occur? LC axon varicosities rarely make direct synaptic contacts (e.g., only ~5% in rat cortex) (Vizi et al. 2010), but the distribution of these varicosities suggests they should often be found near glutamate terminals at excitatory synapses in neocortex (Benavides-Piccione et al. 2005; Gaspar et al. 1989). Another critical point is that LC neurons produce the NMDA receptor subunits needed for glutamate to modulate the release of NE from LC axon varicosities (Chandler et al. 2014; Grilli et al. 2009; Petralia et al. 1994; Zhu et al. 2003).

New technologies enable the visualization of glutamate spillover in cerebellum, neocortex, and hippocampus (Okubo et al. 2010; Okubo & Iino 2011). Multiple action potentials in a row yield sufficient spillover glutamate to activate nonsynaptic NMDA and group I metabotropic glutamate receptors (mGluRs) (which are co-expressed on NE varicosities and enhance glutamate-evoked NE release in rodent and human cortices [Luccini et al. 2007]), but probably yield insufficient glutamate to recruit lower-affinity AMPA receptors (Okubo et al. 2010). Extracellular concentrations of the spillover rapidly decrease as distance from the synaptic cleft increases (Vizi et al. 2010), and the upper limit of glutamate spillover effects is estimated to be no greater than a few micrometers (Okubo & Iino 2011).

That spillover glutamate is sufficient to activate NMDA, but not AMPA receptors is another key factor. Unlike AMPA receptors, NMDA receptors require synchronized glutamate stimulation and neuron depolarization to activate (Lüscher & Malenka 2012). Thus, local glutamate spillover must co-occur with phasic depolarizing bursts of activity in LC neurons to recruit additional local NE release. Furthermore, a unique feature of NMDA receptors is that they require a co-agonist, which could be either glycine or D-serine (Wolosker 2007). Glutamate stimulates astrocytes to release these co-agonists (Harsing & Matyus 2013; Van Horn et al. 2013), and both glutamate and NE stimulate astrocytes to release glutamate (Parpura & Haydon 2000). These additional glutamate interactions should further enhance NMDA receptor-mediated NE release (Fig. 6) (Paukert et al. 2014). Together, these local glutamate–NE interactions support the emergence and sustainment of hotspots in the vicinity of the most activated synapses when arousal is induced.

Figure 6. Norepinephrine (NE) “hotspot” mechanism. (1A) Spillover glutamate (green dots) from highly active neurons interacts with nearby depolarized NE varicosities in a positive feedback loop involving N-methyl-D-aspartate (NMDA) and other glutamate receptors that leads to greater local NE release (maroon dots). The glutamatergic NMDA receptors require concomitant depolarization of noradrenergic axons (lightning symbol). Thus, hotspots amplify prioritized inputs most effectively under phasic arousal. (1B) Glutamate also recruits nearby astrocytes to release serine, glycine (orange dots), and additional glutamate. (2) Greater NE release creates concentration levels sufficient to activate low-affinity β-adrenoreceptors, which enhances neuron excitability. (3) Via activation of β- and α_2A-auto-receptors, NE can stimulate and inhibit additional NE release, respectively. (4) Within hotspots, NE engages β-adrenoreceptors on pre-synaptic glutamate terminals to increase glutamate release. (5) Finally, NE binding to postsynaptic β-adrenoreceptors also inhibits the slow after-hyperpolarization, enabling the neuron to fire even longer. AMPA=α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; mGluR=metabotropic glutamate receptor.

Consistent with the existence of glutamate–NE interactions, local NE release in the region of an activated novel representation depends on the coincident timing of the novel event and an arousing event (Rangel & Leon 1995). For example, when footshock was administered to a rat while it explored a novel environment, NE levels rose substantially higher and remained elevated longer than when footshock was administered to the rat in its holding cage (Fig. 7) (McIntyre et al. 2002). The amygdala presumably activated in response to the novelty of the new environment (Weierich et al. 2010), and glutamate associated with that representational network amplified the NE release initiated by the shock.

Figure 7. A rat receiving a foot shock (FS) in its home cage exhibits a brief increase in norepinephrine (NE) levels (gray triangles). A novel training environment does not increase NE on its own (black squares), but NE levels increase dramatically when shock is combined with that novel training environment (black diamonds). Figure reprinted with permission from McIntyre et al. (2002).

Hotspot effects have also been observed in the bed nucleus of the stria terminalis immediately after training rats on an inhibitory avoidance task (Liu et al. 2009). When infused separately at low doses, glutamate and NE each had no effect. But when infused together at the same low doses, they produced marked memory enhancements. Infusion of a higher dose of glutamate led to memory enhancements that were blocked by propranolol, indicating that the glutamate effect required β-adrenergic activity, which, as we describe next, is another key feature of our hotspot model.

5.1.2. α- and β-adrenoreceptors exert different effects on neuronal excitability and require different NE concentrations to be activated

To be engaged, β-adrenoreceptors require relatively high NE concentrations, α₁-adrenoreceptors more moderate levels, and α₂-adrenoreceptors the lowest NE concentrations (Ramos & Arnsten 2007). Thus, under arousal, α₂-adrenoreceptor effects should be widespread, whereas β-adrenoreceptors should be activated only at hotspot regions because the local glutamate-evoked NE release there results in higher NE levels. Next, we describe the importance of this distinction for adrenergic autoreceptors.

5.1.3 Adrenergic autoreceptors inhibit or amplify their own NE release

Autoreceptors at NE varicosities serve as neural gain amplifiers by taking opposing action at low and high local levels of NE. The predominant presynaptic noradrenergic autoreceptor in humans is the α_2A-adrenoreceptor (Starke 2001), which inhibits NE release when it detects low or moderate levels of NE (Delaney et al. 2007; Gilsbach & Hein 2008; Langer 2008; Starke 2001). In contrast, presynaptic β-adrenoreceptors amplify NE release when activated by high levels of NE (Chang et al. 1986; Misu & Kubo 1986; Murugaiah & O'Donnell 1995a; 1995b; Ueda et al. 1985). In addition, α_2A-adrenoreceptors may lose affinity for NE when neurons are depolarized (Rinne et al. 2013), which would remove their inhibitory influence as a region becomes highly active. However, this loss of affinity recovers at saturating levels of NE (Rinne et al. 2013), which should help prevent the runaway excitation that could otherwise emerge because of the NE–glutamate feedback loop. Together with glutamate-evoked NE release (see sect. 5.1.1), the opposing effects of these different autoreceptors at low and high levels of NE provide an elegant way for the LC to modulate signal gain depending on the degree of local excitation.

5.1.4 Elevated local NE at hotspots engages β-adrenoreceptors on the glutamate terminals transmitting the prioritized representation

This stimulates an even greater release of glutamate, thereby amplifying the high-priority excitatory signal (Ferrero et al. 2013; Gereau & Conn 1994; Herrero & Sánchez-Prieto 1996; Ji et al. 2008; Kobayashi et al. 2009; Mobley & Greengard 1985). That β-adrenoreceptors require relatively high NE concentrations to be engaged further biases this form of cortical autoregulation towards the most active synapses. Through these feedback processes, high-priority representations are “self-selected” to produce a stronger glutamate message and excite their connections more effectively under arousal. This stronger glutamate message should also promote selective memory of such stimuli (see sect. 6.1). In contrast, activation of lower threshold α₂-adrenoreceptors inhibits glutamate release (Bickler & Hansen 1996; Egli et al. 2005), providing a mechanism for inhibiting lower-priority neural activity under arousal.

5.1.5 Higher NE levels at hotspots help prolong the period of neuronal excitation by temporarily inhibiting processes that normalize neuron activity

Under normal conditions, the slow after-hyperpolarization current habituates a postsynaptic neuron's responses following prolonged depolarization (Alger & Nicoll 1980). However, even here, NE seems to benefit prioritized inputs by prolonging neuronal excitation via β-adrenoreceptors inhibiting the slow after-hyperpolarization (Madison & Nicoll 1982; Nicoll 1988).

In summary, different receptor subtypes enable NE to ignite hotspots in regions with high glutamate levels while inhibiting activity elsewhere. As we outline later, this diversity in NE receptor subtypes also plays an important role in shaping synaptic plasticity to favor prioritized representations under phasic arousal.

5.2. NE hotspots modulate interneurons and GABAergic transmission to increase lateral inhibition of competing representations

Increases in glutamate and NE at hotspots should also enhance inhibitory activity that mediates competition among neurons. γ-Aminobutyric acid (GABA) is the most widespread inhibitory transmitter from neurons that suppress the responses of other neurons or neuronal circuits (Petroff 2002). Strong glutamate activity in cortical circuits stimulates local GABAergic activity, which increases the inhibitory effects of highly active regions on neighboring, competing neural circuits (Xue et al. 2014). Increases in NE also activate inhibition directly, with intermediate concentrations engaging maximal suppression (Nai et al. 2009).

Subtypes of interneurons respond differently to NE in ways that should further increase neural gain. Although LC–NE activity activates interneurons that mediate lateral inhibition (Salgado et al. 2012a), it can also suppress interneurons with feedforward connections (Brown et al. 2005), such that a strong signal will inhibit competing representations while enhancing activity in other neurons within its processing pathway.

5.3. NE directs metabolic resources to where they are most needed

To optimize processing of salient events, NE also helps coordinate the delivery of the brain's energy supplies, allowing it to mobilize resources quickly when needed (e.g., Toussay et al. 2013). The brain's most essential energy supplies, oxygen and glucose, are delivered via the bloodstream. One key way that NE coordinates energy delivery is by increasing the spatial and temporal synchronization of blood delivery to oxygen demand within the brain. For example, in mice, as NE levels increase, overall blood vessel diameter in the brain decreases, but the spatial and temporal selectivity of blood distribution to active task-relevant regions increases (Bekar et al. 2012).

In addition to distributing blood flow, NE also interacts with astrocytes locally to mobilize energy resources throughout the cortex. When a particular area of the brain needs more energy, it can obtain fuel not only from glucose, but also from glycogen in astrocytes (Pellerin & Magistretti 2012). NE speeds up the process of obtaining energy from glycogen (Magistretti et al. 1981; Sorg & Magistretti 1991; Walls et al. 2009). While α₁- and α₂-adrenoreceptors mediate glutamate uptake and glycogen production in astrocytes, β-adrenoreceptors stimulate the breakdown of glycogen to provide rapid energy support in highly active local regions (O'Donnell et al. 2012), further amplifying NE hotspot activity.

5.4. Summary

At the local neuronal level, NE suppresses most activity, but amplifies the strongest activity as a result of the differential effects of NE on different adrenoreceptor subtypes. The amplification of strong activity occurs via “NE hotspots,” where positive feedback loops between local NE and glutamate release increase the strength of activated representations. To sustain higher levels of activity, hotspots also recruit limited metabolic resources. At the circuit level, the increased glutamate and NE produced at hotspots recruit nearby astrocytes that supply additional energy to active neurons. On a broader scale, NE facilitates the redistribution of blood flow towards hotspots and away from areas of lower activity. Thus, by influencing multiple levels of brain function, NE selectively amplifies self-regulating processes that bias processing in favor of prioritized information.

6. Roles of the LC–NE system in memory

So far we have focused on how arousal increases the gain on prioritization processes in perception, attention, and initial memory encoding. Now we turn to memory consolidation processes. Experiencing an emotionally intense event influences the vividness and longevity of recent memory traces, enhancing or impairing them based on their priority (e.g., Fig. 4) (Knight & Mather 2009; Sakaki et al. 2014a). Much research has indicated that NE is involved in memory consolidation effects (for a review, see McGaugh [2013]), but there has been little focus on the interplay between NE's enhancing and impairing effects during memory consolidation.

The durability of memories depends on adjustments in the strength of communication across synapses via processes known as long-term potentiation (LTP) and long-term depression (LTD). Whether neural activity triggers LTP or LTD depends on the relative timing of spikes in pre- and postsynaptic neurons (Nabavi et al. 2014), and whether LTP and LTD are maintained depends on protein synthesis processes (Abraham & Williams 2008). We propose that two main NE mechanisms modulate LTP and LTD, leading to “winner-take-more” and “loser-take-less” outcomes in long-term memory: (1) hotspot modulation of the probability of LTP (higher NE levels engaging LTP) and LTD (relatively lower NE levels promoting LTD), and (2) NE-enhanced protein synthesis supporting long-term maintenance of LTP and LTD.

6.1. NE gates spike-timing-dependent LTD and LTP

Long-term potentiation and long-term depression are often studied in brain slices in a petri dish using high-frequency electric stimulation to induce LTP and repeated slow stimulation to induce LTD. But in the brain's natural context involving constant barrages of pre-synaptic activity generating postsynaptic spikes, the relative timing of pre- and postsynaptic activity helps determine whether LTP or LTD occurs. Furthermore, to avoid constant up-and-down adjustment of synapses based on random firing patterns, neuromodulators such as NE and dopamine signal when the relationship between pre-synaptic and postsynaptic activity is likely to be meaningful (Pawlak et al. 2010). In vivo studies indicate that spike-timing-dependent LTP or LTD requires these neuromodulators (Huang et al. 2014; Johansen et al. 2014). In particular, by binding to G-coupled receptors, NE modulates kinases and phosphatases that determine whether LTP or LTD induction occurs (Treviño et al. 2012b; Tully & Bolshakov 2010).

Different adrenoreceptor subtypes appear to mediate NE's regulation of spike-timing-dependent LTP and LTD. Spike-timing-dependent LTP is initiated primarily by β-adrenoreceptor activation, whereas α₁-adrenoreceptors promote spike-timing-dependent LTD (Salgado et al. 2012b). Critically, Salgado and colleagues reported that the LTP promoting activation of β-adrenoreceptors requires concentrations of NE ~25-fold higher (8.75 µM) than the NE concentration that promotes α₁-adrenoreceptor-mediated spike-timing-dependent LTD (0.3 µM) in vitro. This agrees with an in vivo estimate of a 30-fold increase in NE associated with LTP in dentate gyrus (Harley et al. 1996). The increase in NE required to support spike-timing-dependent LTP is substantially higher than the increases in NE levels seen when experimenters stimulate LC and measure NE in cortex or hippocampus using microdialysis (e.g., approximately twice baseline [Florin-Lechner et al. 1996], ~0.5 µM [Palamarchouk et al. 2000]). Thus, there is a discrepancy between the NE levels needed for spike-timing-dependent LTP to occur and the levels measured in laboratory studies. Our GANE model accounts for this difference, as it posits that LC activation interacts with prioritized representations to elicit much higher NE release in a select few local hotspots than elsewhere, such that the average cortical sampling location would not detect the NE levels needed to support LTP.

The NE hotspot model supports a range of simultaneous NE modulatory actions. At high-priority hotspots, NE levels should be sufficiently high to engage β-adrenoreceptors and initiate spike-timing-dependent LTP (Salgado et al. 2012b; Treviño et al. 2012b). Conversely, areas with lower glutamate activity, where NE levels are by comparison modestly increased, would undergo LTD as a result of the engagement of relatively higher affinity α₁-adrenergic receptors (Huang et al. 2014; Salgado et al. 2012b; Treviño et al. 2012a). Variations in NE levels in the alert brain thereby support bidirectional plasticity (Salgado et al. 2012b; Treviño et al. 2012b).

6.2. NE increases protein synthesis processes that promote memory consolidation: Critical role of β-adrenoreceptors

Arousal levels in the minutes and hours before or after an event also influence later memory for it. Here we review evidence that these wider time window effects of arousal depend on NE's enhancement of protein synthesis processes that determine the long-term durability of salient memories. Critically, such regulation of memory processes by NE appears to be mediated by β-adrenoreceptors, which we propose are selectively activated in high-priority representational networks.

The role of NE in gating the synthesis of plasticity-related proteins has been recognized for more than a decade (Cirelli et al. 1996; Cirelli & Tononi 2000). For example, plasticity-related proteins promoted by an LC–NE novelty signal can enhance long-term memory consolidation of another salient, but otherwise poorly consolidated event (i.e., learning that stepping off of a platform leads to a weak shock) that occurs 1 hour later or even 1 hour prior to the novelty experience (Moncada & Viola 2007; Moncada et al. 2011).

Blocking β-adrenoreceptors or protein synthesis prior to novelty exposure prevents novelty facilitation of LTP (Straube et al. 2003). What is particularly striking is that β-adrenoreceptor activation at time 1 primes synapses to induce LTP at time 2 an hour later, even when β-adrenoreceptors are blocked by propranolol during time 2 (Tenorio et al. 2010). However, if protein synthesis processes are blocked during time 2, the time 1 priming event does not lead to enhancement. The plasticity marker, Arc protein, is recruited by β-adrenoreceptor activation in the presence of NMDA receptor activation (Bloomer et al. 2008). Hotspots are characterized by high levels of glutamate release and β-adrenoreceptor activation; thus, emotional arousal should elevate Arc selectively in NE hotspots.

β-Adrenergic activation after learning or weak LTP induction can also convert short-term LTP to more lasting protein synthesis-dependent late LTP (Gelinas & Nguyen 2005; Gelinas et al. 2008). Likewise, stimulating the basolateral amygdala either before or after tetanization of the hippocampus converts early LTP to late LTP via a β-adrenoreceptor mechanism (Frey et al. 2001). Activation of β-adrenoreceptors also shields late LTP from subsequent depotentiation (Gelinas & Nguyen 2005; Katsuki et al. 1997).

Creation of long-lasting memories depends on the protein synthesis cyclic adenosine monophosphate (cAMP)/protein kinase A (PKA)/cAMP response element-binding protein (CREB) pro-signaling cascade (Kandel 2012; O'Dell et al. 2010). Neuronal ensembles in which the cAMP/PKA/CREB cascade has been activated, as happens with the engagement of β-adrenoreceptors, have been found to be selectively allocated to the engram representing a memory (Han et al. 2007). Furthermore, increasing excitability via different methods mimics the effects of CREB overexpression, suggesting that neurons are recruited to an engram based on their neural excitability (Frankland & Josselyn 2015; Zhou et al. 2009). Thus, by modulating CREB and other aspects of neural excitability, NE hotspots should help determine which neurons are allocated to an engram and stabilized in long-term memory.

6.3. Summary

Local NE concentration is the key to understanding how NE mediates arousal's dichotomous effects on memory. Previous research has indicated that different NE levels regulate different forms of spike-timing-dependent plasticity by engaging distinct adrenoreceptors. Whereas NE binding to moderate-affinity α₁-adrenergic receptors leads to LTD and memory suppression, NE binding to lower-affinity β-adrenoreceptors leads to LTP and memory enhancement. We propose that local discrepancies in NE levels arise from self-regulating NE–glutamate interactions. Where NE concentrations become high enough to engage low-affinity β-adrenoreceptors, a cascade of intracellular events triggers protein synthesis processes that enable long-term memory consolidation of the high-priority trace. In contrast, more modest increases in NE levels at less active regions lead to LTD, ensuring less important events are forgotten. Before or after encoding, the confluence of protein synthesis and β-adrenoreceptor activation selectively strengthens memory consolidation when these mechanisms are recruited close in time.

7. Beyond local GANE: Broader noradrenergic circuitry involved in increased selectivity under arousal

Beyond local effects, NE increases biased competition processes by altering how different brain structures interact. With its widely distributed afferents, the LC–NE system influences neural processing in many brain regions when an arousing event occurs. NE release can translate local hotspot effects to more global winner-take-more effects by modulating neuronal oscillations. Furthermore, cortical and subcortical priority signals modulate glutamate release in sensory regions and the hippocampus as mental representations are formed and sustained. As previously reviewed (see sect. 5.1), glutamate is essential for NE release to selectively amplify the processing of significant information. Thus, by stimulating local glutamate release and recruiting LC firing, key brain structures can optimize synaptic conditions for arousal to ignite hotspots.

7.1. Activation of inhibitory networks by NE primes neuronal synchronization among high-priority neural ensembles

So far, we have reviewed evidence that NE hotspots amplify the effects of priority, enhancing salient features while suppressing noisy background activity. In this section, we discuss the possibility that neuronal oscillations communicate activity in local hotspots more globally (Singer 1993).

The first candidate is gamma synchrony (30–80 Hz). Conceptual frameworks of neural oscillations posit that gamma synchrony supports gain modulation in local networks (Fries 2009), such that a target area can oscillate in phase with only one of two competing inputs. As a result, the synaptic input that more successfully synchronizes its activity with the target region is amplified, whereas the less synchronized input is suppressed. Gamma synchrony is likely a key component of selective attention (Baluch & Itti 2011; Fries 2009; Fries et al. 2001).

Gamma oscillations are generated by a feedback loop between excitatory pyramidal cells and fast-spiking parvalbumin-positive inhibitory interneurons (Buzsáki & Wang 2012; Cardin et al. 2009; Carlen et al. 2012; Sohal et al. 2009). Noradrenergic release activates these interneurons (Cox et al. 2008; Huang et al. 2013; Toussay et al. 2013) and increases gamma synchrony in these target regions (Gire & Schoppa 2008; Haggerty et al. 2013; Marzo et al. 2014). Emotional arousal also modulates gamma oscillations in regions that process motivational significance, such as the amygdala, sensory cortex, and prefrontal cortex (Headley & Weinberger 2013). These results suggest that arousal-induced NE release selectively biases gamma oscillations in favor of the most activated representations in local neuronal ensembles.

Consistent with the hotspot model, increases in local gamma power during cognitive processing in humans are associated with increases in glutamate levels (Lally et al. 2014). Increases in local gamma power are also associated with successful memory encoding in humans (Burke et al. 2013). Likewise, in rats, fear conditioning increases gamma synchronization in sensory cortex (Headley & Pare 2013). Increased gamma power predicts retention of tone–shock associations and enhanced representations of the tone associated with shock in the primary auditory cortex (Headley & Weinberger 2011).

Recent research indicates that β-adrenoreceptors recruit in-phase oscillations with gamma activity, whereas α₁-adrenoreceptors recruit out-of-phase oscillations (Haggerty et al. 2013). Given the higher threshold for activating β-adrenergic than α₁-adrenergic receptors (see sect. 5.1), these results suggest that high NE levels at hotspots engage β-adrenoreceptors, recruit in-phase oscillations, and increase local network connectivity for prioritized representations. Elsewhere, lower NE levels should only be sufficient to engage α₁-adrenoreceptors and thereby reduce local gamma power and diminish local synchronization.

In addition to modulating oscillations in local neuronal ensembles, NE also facilitates oscillatory coupling across regions. Current frameworks of neural synchrony posit that long-range/interregional communication between areas is modulated by oscillation in low-frequency bands, such as theta (4–8 Hz), whereas communication within local networks is modulated by high frequencies, including gamma synchrony (Canolty & Knight 2010; Von Stein & Sarnthein 2000). New research further suggests that optimal network function occurs when gamma is embedded in, and phasically facilitated by, slower theta (or even delta [Lakatos et al. 2008]) oscillations (Canolty & Knight 2010; but see Burke et al. 2013). This theta–gamma coupling seems to provide a mechanism for interregional communication and cross-location phase coupling across regions to help translate local NE hotspots into global effects.

Activation of the LC–NE system promotes hippocampal theta (e.g., Berridge & Foote 1991; Walling et al. 2011) and is linked to enhancement of novelty-related hippocampal theta (Kocsis et al. 2007). In humans, the phase coupling of gamma with slower oscillations has been described primarily for neocortex (Canolty et al. 2006), where the role of LC–NE in slower rhythms is less well studied. However, hippocampal theta entrains prefrontal cortical theta (Paz et al. 2008). Recently, selective LC–NE activation was found to increase neocortical theta in anesthetized animals (Vazey & Aston-Jones 2014). The parvalbumin neurons modulated by NE participate in setting not only gamma, but also theta rhythms (Varga et al. 2014; Wulff et al. 2009); thus, parvalbumin interneurons provide a mechanism for LC–NE support of phase-coupled rhythms. Indeed, lesions of NMDA receptors in the parvalbumin neurons result in decreased power of theta oscillations and reduced modulations of gamma oscillation by theta (Korotkova et al. 2010). NE modulation of the hyperpolarization-associated Ih current has also been proposed to support thalamocortical driving of slower neocortical oscillations (Yue & Huguenard 2001). Thus, by modulating gamma and theta, the LC–NE system can amplify the winner-take-more effects of hotspots.

7.2. Key brain regions help evaluate priority and modulate NE hotspots

Here we review how several key brain regions help enhance GANE selectivity mechanisms under arousal. These regions help detect saliency and interact with the LC to fine-tune priority signals via their own hotspot-like effects (e.g., amygdala) and/or other NE mechanisms (e.g., prefrontal cortex and thalamus).

The amygdala plays a central role in enhancing selectivity under arousal. It helps notice and track salient information (Sander et al. 2003) and recruits the LC when activated (e.g., Bouret et al. 2003; Fallon et al. 1978; Jones & Moore 1977; Price & Amaral 1981; Van Bockstaele et al. 1998). The LC, in turn, modulates amygdala activity via NE to further enhance the saliency signal (Sears et al. 2013). Through its strong anatomical projections to sensory cortices (Amaral et al. 2003), the amygdala amplifies cortical processing of behaviorally relevant events (Chau & Galvez 2012; Pessoa & Adolphs 2010). Such modulation of other regions may be mediated by amplification of saliency signals by glutamate–NE interactions within the amygdala (Fig. 7) (see Liu et al. 2009), thereby enhancing the amygdala's selective modulatory influence on other regions. In addition, as reviewed previously (see sect. 3.2), β-adrenoreceptors in the amygdala mediate the selective effects of arousal on memory.

The thalamus helps control the communication of sensory information across the brain (Sherman 2005). Within the thalamus, there are dense NE fibers and high levels of NE in the pulvinar posteriolateral/posteriomedial complex, but very few in the lateral geniculate nucleus (Morrison & Foote 1986; Oke et al. 1978). Through its widespread reciprocal connections with cortical and subcortical structures (Shipp 2003), the pulvinar helps filter inputs based on behavioral relevance (Fischer & Whitney 2012), promotes communication across brain regions (Saalmann & Kastner 2009; Saalmann et al. 2012), modulates gamma oscillations (Shumikhina & Molotchnikoff 1999), and controls the gain of sensory processing (Purushothaman et al. 2012). In addition, the pulvinar is sensitive to emotional saliency (Liddell et al. 2005; Padmala et al. 2010; Troiani & Schultz 2013). Thus, anatomically, NE is set up to modulate thalamic signals of priority.

Furthermore, in rats, NE increases signal-to-noise processing within the thalamus. When directly infused with NE, rat ventral posteriomedial thalamus exhibits reduced spontaneous firing, but enhanced firing in response to whisker stimulation (Hirata et al. 2006). When stimulated by phasic or tonic LC activation, ventral posteriomedial thalamus also exhibited increased firing in response to whisker stimulation (Devilbiss & Waterhouse 2011). However, an intriguing observation was that in sensory barrel field cortex, phasic stimulation of LC enhanced firing to strong whisker stimulation, but slightly impaired firing to weak whisker stimulation, an outcome consistent with the NE hotspot model. This differential response based on stimulus intensity did not, however, occur within the ventral posteriomedial thalamus, where both strong and weak sensory inputs increased firing (Devilbiss & Waterhouse 2011). This initial finding suggests that NE influences in sensory thalamus may occur through mechanisms other than NE hotspots. Thus, further work is needed to examine NE's modulatory role in the thalamus. In any case, the thalamus plays a key role in amplifying selectivity under arousal by coordinating responses to salient stimuli across the brain. Such local representations of salient stimuli are then subject to NE modulatory influences.

The prefrontal cortex (PFC), including the orbital frontal cortex (OFC) and anterior cingulate cortex (ACC), has reciprocal connections with the LC (Arnsten & Goldman-Rakic 1984; Jodo et al. 1998) and is an important regulator of LC output. PFC regions help appraise sensory information and recruit the LC based on goal relevance (Aston-Jones & Cohen 2005), motivational relevance (Mohanty et al. 2008), reward (for the OFC; Schoenbaum & Roesch 2005), conflict (Botvinick et al. 1999; Sheth et al. 2012), monetary loss (Gehring & Willoughby 2002), and pain (Rainville et al. 1997). The ACC is also a key site for integrating task-relevant and arousal inputs (Pessoa 2009; Shackman et al. 2011). In humans, LC innervation of the PFC is relatively sparse, especially in anterior regions (Gaspar et al. 1989; Javoy-Agid et al. 1989), but NE modulates working memory processes in PFC (Arnsten 2011; Wang et al. 2007).

These PFC noradrenergic influences on working memory have different mechanisms than the NE hotspot. First, in our model, β-adrenoreceptors support positive feedback loops at NE–glutamate hotspots, but α₂-adrenoreceptors suppress those feedback loops (see sect. 5.1). However, the facilitatory versus inhibitory role of these adrenoreceptors reverses in the context of working memory. β-Adrenoreceptors stimulate cAMP, whereas α₂-adrenoreceptors inhibit it (Duman & Enna 1986; Nomura et al. 2014; Robinson & Siegelbaum 2003). Inhibition of cAMP via stimulation of postsynaptic α₂-adrenoreceptors increases input resistance and enhances recurrent network activity and working memory performance (Wang et al. 2007). Thus, by activating via α₂-adrenoreceptors, moderate levels of arousal should enhance working memory processes that maintain goal-relevant information in mind, whereas by activating β-adrenoreceptors, high levels of arousal should impair these processes (Arnsten 2011; Kuhbandner & Zehetleitner 2011). Such impairments may, in turn, disrupt initiation of top-down prioritization goals after exposure to emotionally salient stimuli (Sutherland et al., in press).

One interesting question is what might occur when top-down priority and bottom-up priority conflict. The insula plays a key role in this aspect and integrates salience signals from internal and external stimuli (Craig 2009; Uddin 2015). The insula is involved in various types of saliency processing, including error detection (Ullsperger et al. 2010), interoception (Craig 2009), oddball detection (Harsay et al. 2012), aversive memory (Miranda & McGaugh 2004), and detection of events that require cognitive resources (Cai et al. 2015). Although not much is known about LC–insula interactions, the LC and other NE brainstem sites project to the insula (at least in rats) Robertson et al. 2013). Neuroimaging studies also suggest that elevated LC–NE activity is associated with encoding-related activity in the insula in response to aversive stimuli (Clewett et al. 2014; Rasch et al. 2009). Consistent with GANE, motivated (higher-priority) versus passive viewing of emotional faces enhances functional connectivity within face processing networks, including the insula and LC (Skelly & Decety 2012; but see Astafiev et al. 2010 for caution when interpreting results from LC fMRI).

7.3. NE amplifies activity in behaviorally relevant functional brain networks

Along with the dorsal ACC, the insula is a key node in a broader “salience network” (Eckert et al. 2009; Hermans et al. 2011) that helps integrate different sources of saliency (Seeley et al. 2007), guide adaptive behavior (Bressler & Menon 2010; Cocchi et al. 2013), and regulate shifts from rest to task-oriented behavior (Sidlauskaite et al. 2014). On the basis of these findings, recent models of the salience network propose that it mediates competitive interactions between antagonistic attention networks that prioritize internal versus external stimuli (Bressler & Menon 2010; Menon & Uddin 2010). Current data suggest that the LC–NE system modulates salience network activity. For example, β-adrenoreceptor blockade during stress reduces salience network activity (Hermans et al. 2011), and salience network activity is associated with pupil and autonomic responses to errors (Critchley et al. 2005) and overall arousal (Sadaghiani & D'Esposito 2014). In neuroimaging studies, the LC co-activates with the dorsal anterior cingulate during the detection of novel stimuli (Krebs et al. 2013) and during task switching (von der Gablentz et al. 2015), a proposed function of the salience network.

Anatomically, activation of the LC–NE system is well positioned to modulate activity based on priority, as some of the most dense NE innervation is to frontoparietal regions (Gaspar et al. 1989; Javoy-Agid et al. 1989; Morrison & Foote 1986) that coordinate attention to salient stimuli via priority maps (Ptak 2012). Indeed, phasic LC responses, as indexed by pupil dilation, correlate with activity in a dorsal frontoparietal network during focused attention (Alnæs et al. 2014). However, more generally, according to the GANE model, activation of the LC–NE system should amplify activity in whichever functional network is currently dominant. Consistent with a role for NE in mediating this process, while subjects rest, pupil dilation increases as activity in the functional network associated with resting state activity increases and activity in a competing motor network activity is suppressed (Yellin et al. 2015). In addition, NE preferentially enhances ventral frontoparietal attention network activity during the detection of salient events that trigger re-orienting (Corbetta et al. 2008; Strange & Dolan 2007). Thus, NE's influence on gain modulation also manifests at the whole-brain level.

7.4. Summary

Arousal's dual effects on cognition pervade multilevel brain systems to amplify the priority of important information. By modulating theta and gamma oscillations, NE preferentially synchronizes activity between high-glutamate regions, leading to “winner-take-more” effects in perception and memory. Like some earlier emotion–cognition theories (e.g., Pessoa & Adolphs 2010), the GANE model favors the perspective that the amygdala coordinates information transfer within broader networks that influence salience processing and is not the only route by which NE enhances processing of prioritized stimuli. Brain regions that evaluate saliency modulate LC activity either directly via afferent inputs or indirectly via broader networks. Without contextual signals from these central structures and the periphery, the LC would be blind to salient events that demand attention (Sara & Bouret 2012). In turn, the resulting increase in NE release activates these modulatory structures to further bias neural processing in favor of high-priority stimuli. On a larger scale, NE modulates activity in a salience network that mediates competitive interactions between frontoparietal attention networks supporting higher-level representations of priority. Thus, according to the GANE model, reciprocal interactions between the LC and hierarchical brain networks help strengthen and reinforce priority-biasing signals under phasic arousal (see Fig. 8).

Figure 8. Summary of the glutamate amplifies noradrenergic effects (GANE) model. (A) An example of how arousal biases perception and memory to favor prioritized information. High perceptual contrast (bottom-up) and top-down attention prioritize processing of the cow stimulus in the brain over a less salient hay bale. The sound of booming thunder induces arousal and triggers phasic norepinephrine (NE) release. (B) Salience-evaluating structures, such as the amygdala and prefrontal cortex, recruit locus coeruleus (LC) firing to enable NE to modulate ongoing processing at multiple levels of brain function. In the high-priority processing pathway, NE interacts with high local glutamate levels to create “hotspots” that increase the “cow” representational activity even further. These local hotspots recruit energetic resources, synchronize oscillations, lead to enhanced activity in high priority large-scale networks, and increase synaptic plasticity. Local glutamate–NE effects occur parallel to more broad-scale suppression, as NE recruits lateral and auto-inhibitory processes that suppress weaker glutamate signals in lower-priority processing pathways. Together, these noradrenergic mechanisms lead to “winner-take-more” and “loser-take-less” outcomes in perception and memory under arousal, such that the cow is even more likely to be remembered, whereas the hay bale is even more likely to be forgotten. ACC = anterior cingulate cortex; INS = insular cortex.

8. Existing models of LC modulation of cognition

In this section, we discuss how the GANE model relates to existing theories of LC neuromodulation of cognition that we have not already discussed.

8.1. Adaptive gain theory

The adaptive gain theory (Aston-Jones & Cohen 2005) posits that two different modes of LC activity (phasic vs. tonic) adaptively adjust the gain of cortical information processing to optimize behavioral performance. Phasic LC activity serves as a temporal attentional filter to selectively process task-relevant stimuli and filter out task-irrelevant stimuli, whereas tonic LC activity regulates overall arousal level in the brain. Phasic LC responses to target detection are constrained by background LC activity and occur most frequently during moderate levels of tonic activity (Usher et al. 1999). Adaptive gain theory provides predictions similar to those of the GANE model in terms of the role of the phasic LC mode: phasic LC activity should increase the gain of task-relevant inputs over noisy or task-irrelevant activity. Our GANE model provides a neuromechanism for these effects by proposing that low to moderate NE levels create ideal conditions to ignite and sustain local NE hotspots via greater phasic LC responses. In support of this notion, a recent fMRI study used baseline pupil dilation before trials of a reward-learning task as a measure of tonic LC–NE activity (Eldar et al. 2013). Both low baseline pupil diameter before the trial and high pupil dilation response during the trial were associated with stronger brain activation in response to task-relevant, but not task-irrelevant stimuli.

8.2. Network reset theory

The LC–NE system activates in response to various salient stimuli, including novel, uncertain, or emotionally salient stimuli (Sara 2009; Yu & Dayan 2005). The network reset theory proposes that when these stimuli are detected, the LC issues a phasic “reset” signal that reorganizes neural networks to facilitate behavioral and cognitive shifts accordingly (Bouret & Sara 2005; Sara & Bouret 2012). This theory explains why emotionally salient stimuli and the sudden onset of goal-relevant or perceptually salient stimuli are preferentially perceived and remembered: these events activate the LC, which then reconfigures functional brain networks to process new sources of priority while impairing ongoing processing of other stimuli. This model, however, does not offer a clear explanation of why phasic arousal induced when encountering emotional stimuli can enhance processing of preceding stimuli when they have high priority.

To explain both the facilitative and impairing effects of emotional arousal on preceding stimuli, the GANE model posits that the incidental release of NE by something emotional can instead maintain – or even enhance – ongoing functional network connectivity when those networks are highly activated. Stimulating the LC can inhibit feedforward inhibition by interneurons, thereby increasing the throughput of coincident sensory (glutamatergic) inputs (Brown et al. 2005). Although this “loosening” of neurotransmission enables network flexibility and the building of new representations, the GANE model's prediction that strong glutamatergic signals transmitting a prioritized representation will benefit from sudden LC activation explains how the “reset” signal triggered by phasic LC activity can still enhance processing of preceding high-priority stimuli.

8.3. Summary

The GANE model both complements and extends previous models of how cognition is influenced by the LC–NE system. According to adaptive gain theory, high phasic LC activity promotes exploitation of the current focus of attention over exploration of other options. In contrast, the network reset theory proposes that phasic LC activity promotes a global reset of attention. The GANE model reconciles these two theories by highlighting the role of priority. According to the GANE model, if the current focus of attention has sufficient priority to yield high glutamate release in synapses transmitting those stimuli, then a phasic increase in LC activity should enhance processing of those representations. Otherwise, increases in LC activity should shift attention and neural resource allocation towards new sources of priority.

The GANE model extends current models of LC function by positing that under arousal, local glutamate–NE interactions will amplify activity of high-priority representations regardless of how those representations initially became highly active. Thus, although the GANE model provides neural mechanisms that account for arousal increasing biased competition outcomes, it can also accommodate other models or modes of information prioritization (Friston 2010; Keitel et al. 2013; Reynolds & Heeger 2009; Wieser et al. 2011).

9. Potential boundary conditions and questions for future research

In this article, we have argued that arousal leads to winner-take-more and loser-take-less effects in perception and memory via local and global noradrenergic mechanisms in the brain. Yet, although the GANE model explains many findings observed in the emotion–cognition literature, there are a number of important questions for future research.

Arousal may not increase selectivity as effectively among older adults because of age-related changes in the LC–NE system, including loss of LC neurons (Manaye et al. 1995; Mather & Harley 2016; Sladek & Sladek 1978; Vijayashankar & Brody 1979). Recent autopsy evidence indicates that lower LC neuron density is related to the rate of cognitive decline prior to death, even after controlling for decline in other aminergic nuclei (e.g., dorsal raphe, ventral tegmental area) (Wilson et al. 2013). β- and α₂-adrenoreceptors may also be affected in aging (e.g., Bigham & Lidow 1995; Kalaria et al. 1989). Decreases in α₂-adrenoreceptor activity may contribute to age-related cognitive declines because agonists that engage α_2A-adrenoreceptors can improve age-related deficits in working memory (Arnsten & Cai 1993; Arnsten & Goldman-Rakic 1985; Ramos et al. 2006), potentially via α_2A-induced improvements in the ability to maintain focused attention (Decamp et al. 2011). Aging also affects how effectively glutamate triggers additional NE release (Gonzales et al. 1991; Pittaluga et al. 1993), which would disrupt the emergence and/or efficacy of NE hotspots in older adults.

Another question involves sleep, which plays a crucial role in selectively consolidating salient memory traces (Diekelmann & Born 2010), including emotional stimuli (Hu et al. 2006; Payne et al. 2008; 2012) and top-down prioritized information (Rauchs et al. 2011; Saletin et al. 2011). Emerging research suggests that the LC–NE system may enhance memory consolidation during slow wave sleep (non–rapid eye movement [NREM]), a period when high-priority neural ensembles reactivate (for a review, see Dang-Vu et al. 2008; Eschenko et al. 2012; Sara 2010). For example, a learning-dependent increase in LC activity occurs during slow wave sleep (Eschenko & Sara 2008), and depleting NE prior to encoding reduces slow wave sleep that night (Cirelli et al. 2005). Pharmacologically enhancing LC–NE system activity during slow wave sleep improves recognition of odors learned within the previous 3 hours, whereas blocking LC–NE activity impairs odor recognition (Gais et al. 2011). Blocking NE during sleep also leads to greater memory impairment for emotional than for neutral stimuli (Groch et al. 2011). The timing of transient LC activity coincides with the slow wave grouping of hippocampal sharp wave ripple complexes and sleep spindles that promote NMDA-mediated cellular plasticity (Diekelmann & Born 2010; Rosanova & Ulrich 2005). NE may interact with these processes, given evidence that pharmacological activation of β-adrenoreceptors facilitates the emergence of sharp waves and the induction of LTP (Ul Haq et al. 2012). Together these findings raise the intriguing possibility that the precise timing of NE release interacts with the reactivation of high-priority memory networks to facilitate GANE effects during slow wave sleep.

In this article, we focused on perception, encoding, and consolidation processes, but another important question for future research is how NE modulates memory retrieval (e.g., Sterpenich et al. 2006). For example, when encountering a new experience, our memory system can either store this novel information as a distinct memory (i.e., requiring pattern separation) or use it to reactivate existing memories (i.e., requiring pattern completion) (Bakker et al. 2008). Previous research indicated that arousal facilitates pattern separation (Segal et al. 2012) and that NE facilitates retrieval or pattern completion (Devauges & Sara 1991). But it has been unclear how NE/arousal modulates competition between these two hippocampal processing modes. Glutamate amplification of noradrenergic effects might also affect the stability of a salient memory after it is retrieved, or reconsolidated, because this process involves β-adrenoreceptor and NMDA receptor activation (Lee et al. 2006; Przybyslawski et al. 1999).

Another open question concerns the timing of these effects. Behavioral data indicate that presenting an emotionally salient item influences memory of items appearing in the past few seconds (e.g., Sakaki et al. 2014a) and memory of items appearing in the next few seconds, as well (e.g., Sutherland & Mather 2012). It is plausible that the phasic release of NE would have effects on this time scale, but research examining NE–glutamate interactions is needed to address this question.

On the tonic side of the equation, events that induce stress activate both the LC–NE system and the hypothalamic pituitary adrenal (HPA) axis (Pacak & Palkovits 2001; Sved et al. 2002), and these two systems interact in many ways, especially via the actions of corticotropin releasing factor (CRF). Released by the hypothalamus under stress, CRF helps to initiate the HPA axis response while also targeting the LC (Carrasco & Van de Kar 2003; Valentino & Van Bockstaele 2001; Van Bockstaele et al. 2001). CRF influences both tonic LC activity and sensory-evoked phasic discharge, either enhancing or impairing sensory-evoked phasic responses depending on waking state and CRF levels administered (Bangasser & Valentino 2012; Devilbiss et al. 2012; Zitnik et al. 2014). One possibility is that by modulating tonic levels of LC activity, stress also enhances or constrains the impact of phasic arousal responses (see sect. 8.1).

Human genetic studies suggest that different NE polymorphisms moderate the strength of arousal's influence on memory and perceptual processing. To date, much of this research has focused on the ADRA2B deletion variant in which there is reduced NE inhibitory signaling. In human ADRA2B deletion carriers, there is greater activity in the amygdala and insula during the viewing or encoding of emotional versus neutral images (Cousijn et al. 2010; Rasch et al. 2009). Such patterns of NE-related activity are believed to underlie the larger advantage of emotionally salient over neutral stimuli in memory (de Quervain et al. 2007) and perception (Todd et al. 2011; 2013) observed in these individuals.

It is, however, unclear how these genetic effects relate to the NE hotspot mechanisms outlined in the GANE model. Whereas α_2A-adrenoreceptors are found throughout much of the brain and have been clearly identified as autoreceptors regulating NE release, the α_2B-adrenoreceptors associated with this genetic polymorphism have a different profile (Brede et al. 2004). They are most dense in striatum, globus pallidus, and thalamus (De Vos et al. 1992; Saunders & Limbird 1999) and are essential for regulating the fetal blood supply (Brede et al. 2004). Thus, although it is possible that these genetic effects alter the feedback cycle in NE hotspots, the genetic differences could also be mediated by different developmental pathways, thalamic modulation of emotional input, or some other factor.

Related to this point about the differential brain localization of α_2B-adrenoreceptors is the more general question of how regional variation in receptor density (e.g., Zilles & Amunts 2009) modulates hotspot effects. Modeling and direct comparisons of NE–glutamate interactions across regions could help address this question. In addition, although we have focused on how the LC–NE system influences cognition, other neuromodulators such as serotonin, dopamine, and acetylcholine share many mechanisms of action with NE (Hurley et al. 2004) and interact with NE to regulate attention, memory, and arousal (Arnsten 2011; Briand et al. 2007; Sara 2009). Such interactions are likely to modulate the NE–glutamate interactions highlighted here (some examples already described in sect. 5.1 are interactions with orexin, histamine, glycine, and serine). These interactions may allow for more nuanced effects and some redundancy within the arousal system. However, given NE's core role in arousal and broad innervation of much of the brain, including source nuclei of other neuromodulators (e.g., ventral tegmental area and basal forebrain) (Jones 2004; Sara 2009), we expect that it plays the lead role in modulating cognitive selectivity as arousal levels fluctuate.

10. Conclusion

Selection is at the core of what allows our cognitive systems to function effectively, enabling us to process the constant influx of information and retrieve the experiences most relevant for adaptive behavior and maintenance of well-being. The ability to focus on salient information is especially important during situations that induce arousal, such as during exposure to threatening or exciting sounds or objects and the pressure to perform a challenging task. For more than 50 years, there has been robust behavioral evidence that arousal often simultaneously enhances and impairs processing of different types of neutral information (Easterbrook 1959). Yet brain-based accounts of how arousal influences cognition failed to address how such dual effects could arise.

Our GANE model fills this critical gap. In this framework, we propose that increases in NE levels under arousal enhance the selectivity of information processing. GANE builds on the previous ABC model (Mather & Sutherland 2011) to provide neural mechanisms of how NE leads to winner-take-more and loser-take-less effects in perception, attention, and memory. Unlike the ABC model, however, the GANE model does not require competition to be a fundamental mechanism. Instead, the GANE model selectively amplifies the activity of whatever priority mechanisms are operating.

Under phasic arousal, local glutamate signals corresponding to a highly activated percept interact with NE to create a hotspot of even higher levels of activity, whereas lower-priority representations are either neglected or further suppressed. These self-regulating hotspots are further aided by NE's recruitment of brain structures and large-scale functional networks that determine which stimuli deserve attention. NE directs blood flow and energetic resources to brain regions transmitting prioritized information. It supports selective memory consolidation via initiation of LTP and LTD. Through all of these processes, NE increases the gain of prioritized information in the brain, such that things that matter stand out even more and are remembered even better, while the mundane and irrelevant recede even farther into the background and are ignored or forgotten.