Locus Coeruleus Norepinephrine in Learned Behavior: Anatomical Modularity and Spatiotemporal Integration in Targets

1Norepinephrine (NE) is one of the four main neuromodulators in the brain, exerting widespread influence over almost all cortical and subcortical brain regions. Neurons in the locus coeruleus (LC) release NE to regulate baseline arousal and to facilitate a variety of sensory-motor and behavioral functions. Dysfunction in the LC-NE system has been implicated in the etiology of ADHD, schizophrenia, anxiety or stress, and depression, as well as in the cognitive decline observed in aging and Alzheimer’s disease.

2Despite its brain-wide effects and established involvement in CNS disorders, much about even the normal function of the LC-NE system in the brain remains unknown. For example, the conditions under which LC-NE neurons are transiently activated and the modes of NE action during learned behaviors are poorly understood, especially in comparison to other neuromodulatory systems, such as dopamine. There are at least three reasons for this gap. First, the activation of phasic NE has been traditionally examined in the context of sensory, motor, or goal-oriented events with little consideration of its role in signaling temporal stimulus-action-reward sequences or task parameters such as uncertainty. Second, the functions of phasic NE have been interpreted in the context of global or brain-wide influences, without considering selective spatial targeting. Third, the LC is a small nucleus, with a small number of neurons projecting throughout the brain, thus exhibiting relatively low innervation density as compared to other neuromodulatory systems, which typically have a larger number of neurons with a distinct projection network (e.g., substantia nigra and ventral tegmental area for dopamine). For this reason, methods for recording and manipulating LC-NE activity during behavior have been technically challenging. Recent studies have begun to overcome many of these challenges, enabling a more thorough understanding of the role of LC-NE in different aspects of cognition and behavior. Here, we review the current state of the field and present new ideas on the organization, activation, and function of LC-NE neurons.

3Despite its small number of neurons [∼3,000 neurons in the rodent brain and ∼50,000 in the human brain], the LC serves as the primary source of NE in the forebrain with a highly divergent set of projections. Indeed, most cortical and subcortical areas receive dense LC-NE axonal innervation with the marked exception of the striatum, which receives limited innervations from the LC (; Figure 1A). LC is reported to receive inputs from up to 111 distinct brain regions, including most brain-stem and forebrain regions (;; Figure 1B). Brainstem inputs, most notably from the gigantocellular reticular nucleus – which responds to tactile, visual, vestibular, and olfactory stimuli – result in LC activation following salient sensory stimuli. Meanwhile, top-down inputs from the prefrontal cortex (PFC) and central amygdala can modulate the intensity of LC activation. These convergent inputs suggest its recruitment in both bottom-up, sensory induced, as well as top-down, goal-directed, regulation of behavior. This input-output organization leaves the LC poised to modulate brain function in response to external stimuli and internal states, and modify it through learning. Determining the contexts that trigger LC activity and how, in turn, this activity affects brain circuits is critical to understanding the role of LC-NE in cognition.

4Norepinephrine neuronal bodies make up a dense LC “core,” with LC-NE dendrites extending into a pericoeruleus “shell” region. The pericoerulear region also contains GABAergic (LC-GABA), glutamatergic, neuropeptide-S-expressing, and cholinergic neurons. Whether all these neuronal subtypes make direct connections with LC-NE neurons, as well as their function in LC modulated behaviors, remains unresolved. Nonetheless, the neuronal subtype that has received the most attention is the LC-GABA population, which forms inhibitory synapses on LC-NE neurons (;;;; Figure 2A). Tracing experiments indicate that these LC-GABA neurons receive inputs from largely the same brain regions as LC-NE neurons, with only several regions exhibiting preferential projections to either LC-GABA or LC-NE neurons (; Figures 2B,C). Direct recordings from LC-GABA neurons show that they are highly active during wakefulness. A subpopulation of LC-GABA neurons are active simultaneously with LC-NE neurons, reflecting coincident inputs to both LC-NE and LC-GABA populations. Coincident inputs to LC-NE and LC-GABA neurons play a critical role in controlling the phasic activity of LC-NE neurons and resultant transient increases in arousal in response to salient sensory stimuli. Based on retrograde labeling, LC-GABA neurons would receive preferential, or non-coincident, inputs from several brain regions, suggesting that these regions can exert inhibitory control of LC-NE activity (; Figure 2C). Further studies using input mapping with electrophysiology are required to confirm the existence of preferential inputs to LC-GABA neurons. Yet, preferential inputs to LC-GABA neurons have been reported in previous experiments studying PFC inputs. Inactivation of the PFC in rats shows an increase in firing rate in the LC, and axonal photo-activation of PFC inputs in the LC decreases baseline LC-mediated arousal levels. Based on tracing experiments (Figure 2C), there are several other regions that likely exert preferential inputs to LC-GABA neurons, such as the superior colliculus or the intermediate reticular nucleus, but the function and nature of these input regions on the LC have been largely unexplored.

5Whether the other neuronal subtypes in the LC shell – e.g., glutamatergic, neuropeptide-S-expressing, or cholinergic – form direct connections with LC-NE neurons, or play a role in shaping LC-NE activity, remains unknown. However, these neuronal subtypes are predominantly active during wakefulness, indicating that they could be involved in spontaneous or learned behaviors. Future research using physiological recordings and manipulations targeting these different neuronal cell-types will enable a more comprehensive understanding of LC local circuits.

6One of the strongest behavioral correlates of LC-NE activity is its link with vigilance states. LC-NE neurons are silent during REM sleep, display low levels of activity during non-REM sleep, and are most active during wakefulness. Manipulating LC activity modulates vigilance state. High levels of photo-activation of LC-NE neurons, mimicking LC phasic activity, promotes sleep-to-wake transitions, while lower photo-activation levels, resembling increased tonic activity, impair the power of theta and delta frequencies measured with EEG during sleep. Similarly, minimal photo-activation of tonic LC-NE increases the probability of sensory-evoked awakenings. On the other hand, reducing LC-NE activity with photo-inactivation decreases wakefulness durations. Together, these results suggest that LC-NE plays an important role in controlling vigilance states, likely as part of a complex network involving neurons from the brainstem, mid-brain, thalamus, and hypothalamus.

7The link between LC-NE activity and arousal is likely relevant for explaining the role of LC-NE in learned behaviors during wakefulness. LC activation tightly correlates with pupil size and arousal level in mice, monkeys, and humans. This activity of LC-NE neurons alone is sufficient to promote global arousal in mice, as photo-activation or -inhibition of LC-NE neurons during wakefulness increases or reduces arousal levels, respectively. The mode of activation of LC-NE neurons seems to play a key role in arousal. Indeed, phasic activation over long periods (>1 h), but not tonic activation, of LC-NE activity increases arousal. Change in arousal can occur following phasic LC-NE activation by salient or novel sensory stimuli. Often, this phasic activation scales with the strength of behavioral action associated with the stimulus. Optimal cognitive performance correlates with moderate levels of baseline LC-NE activity, while uncontrolled LC-NE activity can lead to hyper-arousal and anxiety behaviors, exemplifying the inverted-U relationship between brain functions and NE activation.

8Arousal alters attention by suppressing low-salience stimuli while enhancing responses to highly salient or goal-directed information; accordingly, LC-NE is also thought to be critically involved in regulating attention through its brain-wide release. Older adults typically exhibit a decline in LC functional connectivity that correlates with an inability to suppress non-salient information during arousal. Global LC-mediated changes in arousal can alter attention via gain control, though the mechanism by which this gain control occurs remains unresolved. It has been shown that pharmacological blockade of NE activity impairs the membrane depolarization of cortical neurons that would naturally occur during high levels of arousal. Arousal levels have also been shown to modulate microglial response to brain injury via β2 adrenergic receptors. However, a detailed analysis of the cell types and neuronal subtypes being affected by NE during brain state changes is lacking. Moreover, we have a poor understanding of the degree of brain-wide modulation by LC-NE activity in vivo in awake behaving animals. Finally, defining the function of LC-NE activity as a global broadcast system of arousal might be too simplistic for understanding its role in learned behavior, as we discuss later with the emergence of modular functional data in the LC-NE system.

9The role of LC-NE neurons extends beyond arousal, attention, and anxiety. Recordings or pharmacological manipulations of phasic NE activity suggest a dual involvement in behavior and cognition. First, LC activation is critical for sustained attention, as its phasic activity correlates with behavioral responses in learned behaviors. Non-targeted recordings in the LC show a transient increase in firing rate coinciding with behavioral response in tasks involving operant or instrumental conditioning. This increase prior to motor execution scales with the amount of effort or vigor exerted, but also with expected reward size. During periods of poor behavioral performance, LC activity is decreased. Thus, it has been hypothesized that transient LC activity during a learned behavior is critical for maintaining focused attention on a task. Supporting this hypothesis, long-lasting blockade of LC activity slows reward seeking behaviors, and pairing LC activity with a sensory stimulus improves this response. Though there is abundant evidence for LC-NE activity facilitating behavioral execution, how this is achieved, especially in the context of other LC-NE functions, is an outstanding question in the field.

10In addition to its role in facilitating behavioral execution, LC activation is critically involved in learning and memory. LC projections to the hippocampus, amygdala, and PFC influence spatial learning and memory formation. For example, during passive reinforcement learning, LC axonal activity in the hippocampus is associated with the timing of reward in spatial memory encoding. In tasks requiring active learning, LC activity correlates with an animal’s ability to form new stimulus/response associations: attentional set-shifting paradigms show that LC activity is involved in the learning of a new discrimination and in making extra-dimensional shifts. In rats performing a linear maze navigation task, when animals undergo an extra-dimensional shift requiring them to use visual cues instead of the previously indicative spatial cues, boosting NE pharmacologically accelerates the detection of the cue-shift and learning of a new cue. It has also been shown that in rats performing a task requiring a shift in association to a new sensory cue within the same modality (intra-dimensional shift), LC spiking activity precedes prelimbic cortex spiking activity, suggesting that NE plays a critical role in the updating of task rules. Further, decreasing LC-NE activity in the PFC over long durations with chemogenetics impairs a rat’s ability to disengage from a cognitive task.

11One mechanism by which LC-NE activity could enable learning is through encoding of environmental novelty. In naïve animals, electrophysiological measurements show that LC activity increases in response to novel or salient stimuli, and this activity quickly habituates. In trained animals, non-targeted measurements of LC spiking activity during learning of an operant conditioning task shows transient LC activity linked with unexpected reward. Indeed, transient increase in LC activity may represent “unexpected uncertainty” and serve as a neural interrupt signal for unexpected events. However, it remains unclear whether increased LC activity in response to unexpected stimuli is causal for learning mechanisms, and how this might co-exist with its role in behavioral execution.

12It has been widely observed that LC dynamics correlate with fluctuations in pupil diameter, both in human and animal studies. Many studies have taken advantage of this correlation to study the role of the LC in arousal, learning, and uncertainty in human subjects. One such study suggests that LC-NE supports learning through global modulations of neural gain. High neural gain would enhance attention for selective stimulus features, whereas low gain would broaden attention, promoting processing of multiple stimulus features. fMRI data show that periods of high gain and focused learning are associated with tightly clustered functional connectivity, suggesting that increased gain enhances selective interactions in brain networks. In addition to the evidence for LC-NE activity modulating feature processing through changes in the gain of global networks, other experiments have implicated LC-NE in learning through driving behavioral responses to uncertainty and volatility. Indirect measurements of NE activity in humans using pupillometry report a correlation between unexpected outcomes and increases in transient LC activity. Experiments using human subjects performing a probabilistic learning task show that blocking NE binding to β-adrenergic receptors globally causes subjects to rely more on their prior expectations when uncertainty is high. Similarly, in a study using an auditory localization task where the predictability of the location varied over time, arousal was found to adjust the strength of perceptual biases in a changing environment. These studies again highlight the potential roles of LC-NE in human cognition: driving attention and behavioral execution, and facilitating learning through updating of priors. Although pupil dilations strongly correlate with LC-NE activity, other neuromodulators such as acetylcholine and serotonin also show some degree of correlation with pupil size. Further tools to investigate neuromodulator dynamics will be required to disentangle the role of LC-NE from that of other neuromodulators in human behaviors.

13Two predominating theories attempt to explain the role of the LC in sensory-motor behavior: the adaptive gain theory and the network reset theory. The adaptive gain theory seeks to explain the phasic and tonic modes of LC-NE activity. Phasic activity prevails during optimal behavioral performance, where transient increases in LC-NE activity facilitate task-specific decision processes. In contrast, tonic activity prevails during periods of poor performance, where a general increase in LC-NE activity increases the gain of a network indiscriminately, making targeted circuits more responsive to any stimulus. Thus, through adaptive gain, LC-NE activity optimizes the tradeoff between exploitation and exploration behaviors by switching between phasic and tonic activity, respectively. However, this theory does not explain which environmental stimuli would cue the LC to switch between these two modes. Further, this theory does not offer an explanation for whether different phasic activities exist temporally within tasks, and what the roles of these temporally distinct phasic activations might be. The network reset theory, on the other hand, suggests that contexts requiring a change in behavior transiently activate LC-NE neurons. These activating contexts lead LC-NE neurons to induce widespread cortical arousal and reset network activity in the brain. Similarly, it has been suggested that LC-NE activity signals “unexpected uncertainty,” causing a reset in network activity to enable an updating of priors. By signaling the need to update priors, LC-NE would suppress top-down, expectation driven information in favor of bottom-up sensory-induced signals to enable learning and behavioral optimization. However, it is not clear how LC responses during execution would not lead to a network reset, and how phasically induced arousal relates to the commonly described tonic regulation by NE of arousal and internal state.

14Though these theories of gain-modulation and network reset are not mutually exclusive, neither alone can fully explain the role of LC-NE in the brain, and little progress has been made in either refining or unifying them in the past 15 years. Recent technological advances and increased tool availability enabling precise measurements and manipulations of LC spiking activity in awake behaving mice present a means of evaluating these theories or advancing alternative proposals. Future studies on the role of LC-NE in learned behavior will require well-designed behavioral experiments to address these dual functions, while also considering the heterogeneous nature of LC activity and potential for spatial modularity of outputs.

15One way to reconcile the divergent roles for LC-NE activity is suggested by recent evidence of spatial modularity within the LC-NE neuronal population. Historically, the LC has been considered a homogenous nucleus, such that NE is uniformly released to forebrain, brainstem, cerebellum, and spinal cord. Yet, retrograde labeling shows that morphological characteristics and location within the LC nucleus is predictive of the particular region a LC neuron innervates. Recent findings using projection-based viral genetic labeling has confirmed this view. Retrograde labeling of LC-NE neurons from the infralimbic, prelimbic, motor cortices or the amygdala, olfactory bulb, and medulla yields very few axons in other regions; these data suggest modularity in NE neuron outputs. However, other reports using the same approach demonstrate broad outputs from the LC, with broad collateralization of LC-NE axons in most brain regions excluding the olfactory bulb and medulla. Retrograde tracings also demonstrate limited modularity of LC-NE axons projecting to primary sensory cortices. Moreover, a high throughput method for mapping projections from the LC shows that LC neurons can preferentially target one cortical area, but also innervate large areas of cortex at reduced density. LC-NE neurons may also display functional modularity: simultaneous recording of LC neuron activity in anesthetized rats shows ensembles where the activity of subgroups of neurons correlates in time. So far, no data exist on whether functional modularity exists in awake behaving animals and whether this modularity originates from a combination of distant and local (e.g., LC-GABA) inputs.

16Our current understanding of the LC-NE circuitry suggests at least partially distinct populations of LC-NE neurons with respect to their projection targets and functional organization (Figure 3A). These distinct LC-NE neurons would be able to confer target specific NE release through spatial modularity. In this spatial model of modularity, subpopulations of LC-NE neurons could receive different types of inputs. Distinct inputs to LC-NE neuron with highly convergent axons would enable local NE release (Figure 3B). In other behavioral contexts, inputs to LC-NE neurons with highly divergent axons, or global inputs to most LC-NE neurons, would enable global NE release (Figure 3B). Distinct outputs from LC neurons to cortical or subcortical targets would then encode distinct behavioral correlates. These proposals thus represent testable hypotheses of LC function in learned behaviors derived from spatial modularity of its inputs and outputs.

17The suggestion of a modular input-output organization in the LC raises the question of whether LC-NE subpopulations exhibit different molecular signatures that could confer different functional properties. Again, though the LC has often been considered a homogenous nucleus, there is evidence to suggest that LC-NE neurons are a molecularly heterogeneous population. For example, though all LC neurons contain NE, separable subpopulations also contain other neuropeptides, such as Neuropeptide Y, galanin, and cocaine-and-amphetamine-regulated transcripts. These subpopulations are found throughout the LC but have biased distributions; however, the anatomy of the projections from these neurons as well as the functional relevance of this co-release is uncharacterized. Galanin release from LC has been suggested to mediate active-coping behaviors, but whether this release occurs in projection-specific LC neurons is not known. It has also been shown that LC projections release the neuromodulator dopamine in the hippocampus and midline thalamus, where it plays a significant role in memory consolidation and stress responses, respectively. The extent to which dopamine is co-released from LC terminals in other areas of the brain, and how this co-release might affect other behaviors, remains unknown. In addition to having subsets of neurons capable of co-release of neuropeptides and neurotransmitters, LC neurons themselves appear to exhibit variability in neurotransmitter receptor expression, potentially enabling subpopulations to be activated by different inputs. LC neurons express receptors for many different neurotransmitters, such as GABA, orexin/hypocretin, and acetylcholine. In addition to these receptors, LC-NE neurons express several adrenoceptor subtypes, making LC-NE neurons themselves responsive to NE release. Indeed, it has been suggested that these adrenoceptor subtypes are differentially expressed throughout the LC, making subpopulations of LC-NE neurons differentially responsive to LC-NE activation. Like many other aspects of LC organization and function, the role of these different molecularly defined subpopulations remains unexplored, but provides a means by which LC-NE subpopulations could be differentially activated to facilitate distinct aspects of learned behavior.

18Local integration of NE release by targets is an important way by which the activity of single LC neurons exhibiting multiple task components, with projections to multiple areas, can nonetheless be utilized for separate functions. Due to different NE receptor expression throughout the cortex, heterogenous expression of NE receptors on different types of brain cells, or the underlying activity of a specific brain region, it is possible that the same NE signal can have different effects in output regions. Indeed, specific functions have been attributed to NE from region-specific pharmacological or optogenetic manipulations of LC activity. In subcortical regions, LC-NE activity in the hippocampus is critical for contextual memory formation and place map plasticity. In the basolateral amygdala, photostimulating LC axons increases anxiety-like behavior. In the context of memory formation, inactivating LC projections to the basolateral amygdala impairs memory acquisition, while inactivating LC-NE projections in the infralimbic cortex impairs memory extinction. In cortical regions, pharmacological blockade of local NE receptors in visual, somatosensory, or motor cortex reduces membrane depolarization of neurons that is typically associated with state changes. This depolarization in the presence of NE normally leads to an increase in firing rate of neurons and ultimately, increased sensory sensitivity. In prefrontal cortex regions such as the anterior cingulate cortex and limbic cortex, local manipulation of NE activity influences sustained attention, memory, decision making, and task acquisition. NE release in the anterior cingulate cortex promotes exploration and behavioral variation. In a set-shifting task, manipulating NE activity in the prelimbic or infralimbic cortex improves the ability of animals to learn new associations. Importantly, most of these functions related to LC-NE activation are linked in time to specific components of internal state or behavior, so that NE release within a target enables the subsequent contribution of that target to brain state or function. Thus the different effects of LC-NE signals on processing and behavior not only reflect differential targeting of LC-NE outputs (spatial modularity) but also the differential processing of NE signals in time based on the functional role of the target as well as its regional and cellular NE receptor expression and availability (temporal modularity). In combination with spatial modularity, local and global temporal integration of NE signals in diverse targets is therefore an important mechanism of LC-NE function (Figure 3C).

19Understanding the complex role of LC-NE signals in facilitating distinct aspects of behavior will require understanding the timescales on which LC-NE neurons are activated, how this activity affects performance in a learned behavior, and how these LC-NE signals are used by target regions to facilitate behavioral responses and learning. As in human studies manipulating the discriminability of a target, a learned behavior for studying LC-NE in model systems should incorporate different levels of sensory uncertainty. Thus, a task dependent on trial-history that incorporates sensory uncertainty and necessitates a goal-directed behavioral choice would enable a more unified understanding of the role of LC-NE in learned behaviors. Another important goal for the field is developing an understanding of how NE interacts with other neuromodulators to modulate brain states and behaviors. An established task that can be used across neuromodulatory systems would be well-suited for this goal. For example, in the cholinergic system, an auditory go/no-go task was used to show that cholinergic neurons encode reinforcement surprise, and that these responses scale with uncertainty. Using varying tone intensities, the difficulty was modulated on a trial-by-trial basis to reflect different degrees of uncertainty. A similar task might be used to study how LC-NE neurons encode task-relevant responses and uncertainty, all within the same framework. Additionally, a novel two-alternative forced choice task in head fixed mice has been used to assay perceptual and value-based decision making. This task, which has been standardized across laboratories, could be applied toward studying the role of LC in learned behaviors. Behaviors such as these, incorporating both behavioral execution and learning, combined with cutting-edge tools such as optogenetics, targeted labeling of neuronal populations, deep-brain imaging, and imaging of NE receptor dynamics, combined with specific hypotheses of LC-NE function as proposed here, will help elucidate the role of LC-NE in cognition.

20Despite the diverse roles of the LC in regulating arousal, attention, and facilitating more complex behaviors, our understanding of this nucleus is quite limited. Recent studies, however, are changing our understanding of the LC. What was formerly considered a homogenous nucleus exerting global, uniform influence over its many diverse target regions, is now suggested to be a heterogenous population of NE releasing cells, potentially exhibiting both spatial and temporal modularity that govern its function. These observations, combined with a rapidly expanding neuroscience toolkit, enable updating of existing theories, or potentially forming new ones, to explain the roles of LC-NE in learned behaviors.

21VB-P and GD wrote the initial draft. VB-P made all figures. All authors discussed the content and commented on the text and figures.

22The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.