Monoaminergic Neuromodulation of Sensory Processing

1Even at the earliest stages of sensory processing, the neuronal representation of external stimuli is modulated by internal brain states. While the evidence for such modulation is long-standing, the analysis of sensory representations has typically focused on the feed-forward stimulus-driven component while regarding the modulation by internal states as noise. However, recent results based on population recordings that highlight the extent of brain-state dependent modulation of sensory processing, the discovery of substantial modulation of sensory activity with locomotion, as well as novel tools to more selectively target modulatory circuit elements genetically have contributed to reviving the interest in the neuromodulation of sensory processing. Since the role of acetylcholine has received substantial attention and has been the subject of excellent recent reviews, we will focus here on the modulation by the monoamines dopamine (DA), noradrenaline (NA), and serotonin (5HT). For each of these modulatory systems, we will summarize anatomical data and electrophysiological findings to provide insights into their role in the modulation of sensory processing, placing an emphasis on studies in non-human primates and rodents, and highlighting the complementarity of these neuromodulatory systems (Figure 1). We will cover key classical studies and then shift our focus toward more recent work.

2Serotonin-synthesizing neurons are located in the brainstem in a small group of clusters [named B1–B9 after ]. These were first identified in the rat brain but their anatomical localizations in mouse and primate species, e.g.,, have been found to be largely consistent with that in the rat. These clusters are typically divided into a caudal and a rostral group. The caudal group, which consists of the raphe pallidus (B1), the raphe obscurus (B2), and raphe magnus (B3), projects mainly to the spinal cord and brain stem. The rostral group includes the dorsal raphe nucleus (B6, B7) and the median raphe nucleus (B5, B8, and B9) and projects to the cortex. Within this group, there is some topographic organization of the serotonergic innervation that reflects a rostral (frontal cortex) to caudal (occipital cortex) gradient. Such topographical organization suggests that rather than reflecting a signal that is broadcasted uniformly throughout the brain, serotonergic projections to different areas are more specific and may serve different roles. Indeed, retrograde studies in rats support topographically specific projections, for example to the prefrontal cortex (PFC). Although serotonergic neurons represent only a very small number of the neurons in the brain [approx. 28,000 in the mouse to several 100,000 in humans], they give rise to diverging projections to virtually all regions of the mammalian brain. Here, we will focus on the serotonergic projections to primary sensory cortical areas and subcortical structures involved in sensory processing.

3Anatomical studies in different rodent and monkey species have identified substantial serotonergic projections from the raphe nuclei to early sensory areas including the primary auditory, visual, and somatosensory areas, the olfactory bulb, and subcortical structures involved in sensory processing (Table 1). [Note that while in the auditory system, serotonergic innervation of the cochlear nucleus is established, serotonergic projections to the retina have been controversial].

4Comparisons of monoaminergic innervation found that the serotonergic innervation was substantially more pronounced than that for NA or DA in the macaque primary auditory cortex and in the macaque primary visual cortex. This pattern mirrors early reports of regional differences in cortical monoaminergic distribution, with a pronounced decrease in DA from frontal to occipital cortex, a weaker gradient (interrupted by a peak in the somatosensory cortex) for NA, and a roughly uniform distribution for serotonin (Figure 1B).

5Within primary sensory cortical areas, the distribution of serotonergic fibers in primates shows a characteristic laminar profile. In the primary visual cortex, anatomical data for different primate species agree that the distribution of serotonergic axons is highest in layer 4. This suggests that the serotonergic modulation in the primary visual cortex is biased toward targeting the visual input stage. Serotonergic fibers are also consistently found in cortical layer 4 in the primary auditory cortex and somatosensory cortex, although not preferentially.

6This contrasts with the monoaminergic innervation of primary sensory areas by NA and DA (see below) that is typically sparse if not absent in layer 4 (Figure 1B). (Note that while a transient dominance of serotonergic input to layer 4 occurs during the early post-natal development in the rodent primary visual cortex and barrel cortex, regional and laminar differences are less pronounced in rodents than in the primate species.)

7Consistent serotonergic innervation is also found in subcortical structures involved in sensory processing such as the sensory thalamus, the superior and inferior colliculi, and the cochlear nucleus. In the olfactory bulb, serotonergic fibers reach the glomerular layer, the primary input stage. Together, the extent of serotonergic innervation at the earliest stages of sensory processing makes this system well suited to directly modulate the incoming sensory information.

8The serotonergic projections to the primary sensory areas consist of small varicose axons that are widely distributed. Only a very small proportion of synaptic specializations, typically asymmetric, is found, suggesting that serotonin predominantly acts by volume transmission from varicosities. Note, however, that the degree to which neuromodulatory systems rely on “wired” transmission, i.e., highly localized and typically synaptic, or “volume” transmission, i.e., more spatially diffuse, is subject to debate (e.g.,;;; ). In the mammalian brain, seven serotonin receptor families, most with several subtypes, have been identified to date and contribute to the functional diversity of serotonin. A detailed overview is outside the scope of this review, but a few receptors should be highlighted. 5HT1A is expressed on cortical pyramidal neurons. In the macaque primary visual cortex, the most densely expressed receptors are 5HT1B and 5HT2A, predominantly in layer 4. 5HT1B is also strongly expressed in the LGN, but only weakly in other cortical areas including the auditory and somatosensory cortex. In the mouse, GABAergic neurons that express the 5HT3A do not express the calcium binding protein parvalbumin or somatostatin and may form a third non-overlapping class of inhibitory interneurons.

9Given the complexity of the input to serotonergic neurons, the wide distribution of serotonergic projections in the brain and the diversity of serotonin receptors, it is unsurprising that serotonin has been implicated in a spectrum of brain functions. These include the sleep–wake cycle, hormonal regulation, regulation during development as well as affective, cognitive, and sensorimotor functions. Serotonergic neurons show tonic and phasic modes of discharge that are thought to signal different information. For example, in the mouse transient responses of putative serotonergic neurons in the dorsal raphe nucleus reflect a variety of behavioral, sensorimotor, and reward-linked information. Phasic and tonic response patterns of optogenetically identified serotonergic neurons in the mouse dorsal raphe have been proposed to reflect the contextual valence on different time-scales. Serotonergic neurons have also been linked to signaling patience. But, even for the intensely studied links to reward signaling a simple computational account has proved challenging. Although it is unclear whether its modulatory role can be conceptualized by a simple overarching computational function, serotonin is thought to modulate sensory processing according to behavioral–motivational context.

10In reviewing studies of serotonergic modulation of early sensory processing across modalities, consistencies are notable, which will be our focus here (Table 2). This focus contrasts with previous perspectives highlighting the diversity of findings. (Note that we restricted this summary to studies of short-term sensory modulation and did not consider reports on adaptation or plasticity.) Some variability in the findings likely results from comparing results across anesthetized animals using different anesthetics or awake animals and different experimental approaches.

11Despite this variability, most findings are consistent with an overall serotonin-mediated decrease of the sensory response. In the auditory system, a decrease of the extracellular response was observed for the majority of neurons both in the cochlear nucleus of the rat and the inferior colliculus (IC) of bats, although some variability of the effects between cells was observed. Behaviorally, a reduced startle response to auditory tones was observed for intraventricular injection of serotonin, which is consistent with reduced auditory response (but could also reflect downstream processing). Similarly, a reduced mechanosensory response was observed in behaving mice during optogenetic stimulation of serotonergic raphe neurons, which could reflect downstream processing but would also be expected for a reduced sensory response in the somatosensory cortex as previously reported.

12For early visual processing, predominantly decreased responses were observed for the iontophoretic application of serotonin in the cat lateral geniculate nucleus (LGN) and rat V1. In the anesthetized macaque, when attempting to dissect the role of the two most strongly expressed receptors in V1, 5HT1B and 5HT2A, using receptor specific ligands, a diverse pattern and bi-directional modulation were observed for both. However, these effects were not compared to spontaneous variability of the responses resulting from the anesthesia or the iontophoretic application itself. For example, slow fluctuations in the neuronal responses have been shown to contribute to stimulus-independent co-variability (“noise correlations”) between neurons. When instead iontophoretically applying the endogenous ligand serotonin in awake macaques and comparing the effects against those of pH matched saline application, a recent study found an overall decrease of the sensory responses with serotonin. While there was some variability across cells, consistent with the results of, the inhibitory effect of serotonin was the dominant pattern across the sizeable neuronal population. This decrease was predominantly explained by a multiplicative change (gain change) of the neuronal tuning curves. Behaviorally, a recent study that systemically administered a serotonin-reuptake inhibitor to enhance the effect of serotonin while macaques performed a color discrimination task observed slowed reaction times as well as deteriorated perceptual performance, as expected for such reduced visual responses. Conversely, a gain reduction of the spontaneous response, consistent with an increased signal-to-noise ratio (SNR), or a gain reduction of the tuning curves for serotonin was observed in the mouse olfactory bulb.

13Taken together, a surprisingly consistent pattern of the serotonergic modulation of early sensory processing across modalities emerges. The decreased sensory response by serotonin – effectively lowering the salience of the sensory input – may reflect a sensory signature of how serotonin shapes behavior in downstream circuits, such as its proposed role as a behavioral inhibitor, to promote waiting or persistence.

14In rodents and primates, neurons that produce NA are located in the brainstem in clusters named A1–A7 after. Of these, the locus coeruleus (LC, A6) projects to most areas in the brain, except for to the basal ganglia [reviewed in ], and is the sole source of the noradrenergic innervation of the cerebral cortex. Immunohistochemical evidence indicates that within the LC the vast majority of neurons are noradrenergic. Similar to the serotonergic and dopaminergic systems, the absolute number of noradrenergic neurons in the LC is small [estimated between approx. 1500 per hemisphere in rodents and approx. 15,000 per hemisphere in humans, reviewed in ], but the projections of these neurons are divergent and wide-spread throughout the brain. Despite these wide-spread projections recent findings indicate a modular organization of the LC in rats and substantial anatomical specificity of the connections, for example to the prefrontal versus the motor cortex, or different modules projecting to the amygdala compared to the medial PFC (mPFC). It therefore seems likely that at least some degree of modularity is also characteristic of projections to sensory areas. In the following, we will again focus on the projections to the primary sensory areas and subcortical structures involved in early sensory processing.

15The LC sends divergent projections to the cortex and subcortical structures. Within the LC, these projections are roughly topographically organized [reviewed in ]. In the rat, cortex-projecting neurons are more prominent in the caudal portion of the LC and show some ventral (frontal cortex) to dorsal (occipital cortex) organization. Despite the rough topography, a recent study combining retrograde labeling and optogenetic stimulation in the mouse found that the LC projections to primary sensory cortices are not modality-specific (contrasting with these authors’ findings for the cholinergic system). Nonetheless, projections from LC show clear regional differences (Table 3). Early studies in rats and macaques observed an overall frontal (higher concentration) to occipital (lower concentration) gradient of NA throughout cortex, with the exception of somatosensory cortex where the concentration was highest. This gradient was weaker than that for DA and differed markedly from the absence of a gradient for serotonin (reviewed above and see Figure 1).

16Within the primary sensory cortex, the noradrenergic innervation shows some laminar specialization, particularly in primates. In the primate somatosensory cortex, noradrenergic fibers are fairly uniform and dense throughout layers. The innervation of the primary auditory cortex and primary visual cortex, in contrast, is sparser overall, and a “striking absence” of layer 4 innervation, particularly in V1, has been observed. Combined with the near absent innervation of the LGN, which provides the dominant feed-forward visual input to V1, this suggests that – in contrast to the serotonergic system – noradrenergic modulation does not target the visual input stage. Interestingly, such complementarity between the serotonergic and noradrenergic innervation of the sensory input stage is also present in the rat olfactory bulb. While “virtually no label” was observed in the glomerular layer (the site of the first synapse of the olfactory input) using anterograde tracer injections in the LC, this layer showed the densest serotonergic innervation (see above). Instead, pronounced noradrenergic innervation in the rat olfactory bulb was observed for the consecutive processing stages, the internal plexiform, granule, and external plexiform layers.

17Similar to the other monoaminergic systems, noradrenergic neurons in the LC show both tonic and phasic modes of activation [reviewed in ]. Their axons have characteristic NA-containing varicosities that can form synapses (e.g., ) and likely have non-synaptic release sites. Three adrenergic receptor families are expressed in the brain (alpha 1, alpha 2, and beta 1–3), which have characteristic pre- and postsynaptic sites and laminar expression patterns across cortex. Noradrenaline has the highest affinity for the α2 receptors, intermediate for α1 receptors, and lowest affinity for the β adrenergic receptors [reviewed in ]. In the primate PFC, these differences in affinity have been implicated in differentially modulating cognitive processes as a function of the level of noradrenergic tone. They are likely also an important factor in the substantial diversity in modulation of sensory processing observed for NA (see below).

18The brain’s noradrenergic system has long been linked to arousal (e.g.,; reviewed in, e.g.,; ) and to adapting network activity for optimal, flexible behavior. Consistent with such a general function, NA has been shown to modulate sensory processing in complex ways (Table 4).

19Earlier findings have been reviewed in detail. In brief, activating the LC or local application of NA has been found to result in a variety of effects, including inhibitory and/or facilitating modulation, selective gating, changes to a neuron’s receptive field, and changes to its SNR. For example, iontophoretic application of NA was found to predominantly decrease responses in the dLGN, dorsal, and ventral thalamus of cats, in A1 of squirrel monkeys and of rats, and in rat S1. Conversely, a predominantly facilitating effect was observed for iontophoretic application of NA or LC stimulation in the rat dLGN or V1, and for phasic LC stimulation in rat S1 or rat piriform cortex. In cat V1 Sato and colleagues, observed a substantial layer-dependence of the response modulation with LC stimulation. It was predominantly inhibitory in layers 2–4, mostly α receptor mediated facilitation in layer 5 and approximately balanced in both directions in layer 6. These laminar differences likely reflect different receptor expression profiles across layers. The results suggest that some of the variability between studies may also reflect laminar differences between studies within the targeted structures in addition to, e.g., dose-dependent effects. Mirroring this variability, noradrenergic effects on the neuronal SNR differed between studies. While some reported an enhanced SNR (e.g.,;; ) and others reported no net effect on SNR across the population. In line with the notion of optimizing the neuronal gain for behavior, this variability may reflect the noradrenergic role in different states of a dynamical system.

20Indeed, simultaneous recordings in the LC and barrel cortex of anesthetized rats combined with dynamical systems modeling showed that activity in the LC was highly predictive of dynamic changes in cortical excitability. Moreover, phasic responses in the LC or urethane anesthetized rats were elicited by aversive somatosensory stimuli and modulated the stimulus-induced gamma oscillations in the mPFC, suggesting a corresponding modulation of somatosensory processing. In awake mice, visual responses have been found to be substantially enhanced during phases when the animals were running compared to no locomotion. Intriguingly, intracellular recordings showed that this effect was accompanied by a depolarization of the membrane potential in layer 2/3 neurons, which in turn was blocked by noradrenergic antagonists. Indeed, both noradrenergic and cholinergic mechanisms were linked to this locomotion-dependent modulation, with partially complementary functions. These findings align with the notion of a noradrenergic role in adapting sensory circuits for optimal behavior. Additionally, they highlight the importance of interactions between neuromodulatory systems that go beyond the scope of this review.

21Dopamine synthesizing neurons represent a comparatively small population of neurons in the mammalian brain. The majority of DA neurons are found in three cell groups in the ventral midbrain (mesencephalon). Based on cytoarchitectonic features, most dopaminergic cells reside in the substantia nigra (SN) pars compacta (SNc, A9), the ventral tegmental area (VTA, A10) medial to the SN and in the retrorubral area (RRA; A8), which lies caudal and dorsal to the SN. Other, smaller dopaminergic cell groups are present in the periaqueductal gray matter (PAG; A11) and in the hypothalamus (A12–A15), the lateral parabrachial nucleus, the olfactory bulb (A16), and in the retina (A17).

22The most frequently used marker for identifying dopaminergic neurons is tyrosine hydroxylase (TH), the rate-limiting enzyme of DA synthesis. For TH-immunopositive neurons, the A8 group comprises approx. 5% of cells, while A9 and A10 account for the remaining 95% with a slight dominance of A9 over A10. The DA system undergoes considerable expansion from rodents to primates. By immunostaining for TH, the number of mesencephalic DA neurons has been estimated at approx. 20,000–30,000 in total in mice, 45,000 in rats, 110,000–220,000 in the SN in rhesus monkeys, and between 230,000 and 430,000 in the human SN. Thus, a large dopaminergic A9 group is a distinctive feature of the primate brain.

23Tyrosine hydroxylase labeling to identify dopaminergic neurons has been called into question because of a lack of specificity. Staining for the dopamine transporter (DAT), which removes DA from the extracellular space and is largely responsible for the termination of dopaminergic neurotransmission, is more specific and is not detected, e.g., in noradrenergic cells. However, DAT is not expressed in all DA neurons. For example, it cannot be used to label the diencephalic (hypothalamic) cell groups. The subgroup of VTA DA neurons that project to the PFC also contains very little DAT. Distinct expression profiles of TH and DAT can therefore be exploited to trace independent populations among dopaminergic neurons.

24There is mounting evidence that DA is also released from terminals of LC neurons as a co-transmitter of NA. These important findings underscore the potential caveats of using cellular protein markers instead of the transmitter receptor or the transmitter itself to investigate monoaminergic neurotransmission. Considerable complexity within the monoaminergic system is therefore likely to shape sensory and higher-order cognitive processing.

25A variety of different techniques has been used to investigate dopaminergic innervation of target structures, including direct (e.g., photometric or autoradiographic) neurotransmitter measurements, immunostaining for TH in comparison to staining for DA-β hydroxylase and DAT, DA receptor autoradiography, and DA receptor mRNA assays. It is important to keep in mind that these methods differ not just with regard to their sensitivity and specificity, but that they also target distinct stages of DA production and neurotransmission. This is a likely source of variability and even discrepancy in the literature. Ultimately, functional evidence is required that local DA modulates neuronal processing in target areas. Physiological studies are therefore the gold-standard for demonstrating effective dopaminergic innervation (see below).

26The DA system projects extensively to many subcortical and cortical structures of the brain, albeit with a clear concentration. noted, summarizing their work on the cortical distribution of dopaminergic afferents in the non-human primate, that “dopaminergic fibers preferentially innervate motor over sensory regions, sensory association over primary sensory regions, and auditory association over visual association regions.”

27The dopaminergic projections from the midbrain are typically subdivided into the mesostriatal pathway and the mesocorticolimbic pathway. The striatum is the major target of mesencephalic DA neurons. Ipsilateral projections to the dorsal striatum mainly arise from the SNc (nigrostriatal pathway). Limbic efferents including projections to the ventral striatum (nucleus accumbens) emanate largely from the VTA.

28The mesocortical pathways target in particular the motor cortex, the prefrontal, the anterior cingulate, and the rhinal cortices in rodents and in the non-human primate. Similar to the number of dopaminergic midbrain neurons, however, the density of cortical DA terminals varies considerably from species to species, reaching a maximum in primates. Unlike NA and, especially, serotonin, cortical levels of DA in the macaque brain exhibit a prominent gradient with a strong decrease along the fronto-occipital axis and only minimal amounts detectable in visual cortex (; see Figure 1B).

29Mesocortical projections originate from all major cell groups in the ventral mesencephalon, with a looser organization compared to the mesostriatal system. A clear topographical relationship, however, has been described for the most densely innervated frontal lobe. Dopaminergic projections to the dorsolateral PFC in the macaque brain arise from the dorsal and lateral regions of three midbrain cell groups A8, A9, and A10, whereas the ventromedial regions of the PFC receive projections from the medial parabrachial pigmented nucleus and the midline nuclei of the VTA. A similar medial–lateral arrangement has been found in rats. Here, most of the dopaminergic innervation of the medial frontal cortex comes from VTA neurons that project to the deep layers 5 and 6 of the mPFC and in particular to the orbitofrontal cortex, with substantially fewer cells from the SNc innervating predominately the superficial mPFC layers. SNc DA neurons preferentially target the lateral frontal cortex, e.g., the premotor areas.

30Overall, VTA dopaminergic neurons show little collateralization to extensive terminal fields, somewhat in contrast to SNc cells. This suggests that, in light of the topographical arrangement described above, discrete, anatomically circumscribed dopaminergic subsystems exist that could modulate selected target regions.

31Dopamine-containing varicosities can form multiple synaptic contacts along the course of an axonal fiber. In frontal cortex, these terminals establish conventional, symmetric synapses with postsynaptic dendritic shafts and spines, preferentially on pyramidal neurons. At the ultrastructural level, dopaminergic afferents form synaptic triads with postsynaptic spines that receive a second, presumably glutamatergic input. This configuration would allow DA to modulate ongoing neuronal transmission both pre- and postsynaptically.

32Dopamine receptors are expressed by pyramidal neurons and GABAergic interneurons alike, indicating that DA modulates excitatory and inhibitory synaptic transmission. Dopamine receptors are categorized into two major families: the D1 family comprising the D1 and D5 receptors, and the D2 family comprising the D2, D3, and D4 receptors. Both D1 and D2 families of receptors are G protein-coupled receptors, which initiate intracellular signaling cascades rather than directly inducing postsynaptic currents. D1 and D2 receptors are largely expressed in different cell populations. In most rat brain structures except for the VTA and some midbrain cell groups, D1 receptors outnumber D2 receptors. In the human brain, D1 receptor mRNA clearly dominates over D2 receptor mRNA in the cortical mantle, whereas D2 receptor mRNA is more abundant in the hippocampal formation, brainstem, and in subcortical structures such as the thalamus. Of note, a high proportion of DA (D1) receptors are found at extrasynaptic sites, suggesting that DA might also exert its effects via diffusion in the neuropil (volume transmission).

33Cortical DA receptors show a laminar-specific distribution profile. D1 receptors are expressed in all cortical layers in the primate frontal brain, often with a bilaminar infra-supragranular pattern with a relative paucity in layer 4 and a predilection for deeper layers. D2 receptors, in contrast, are typically confined to cortical layer 5.

34Compared to the densely innervated frontal cortex, the sensory cortices receive very sparse dopaminergic afferents (Table 5). In both rat and primate studies, the primary visual cortex shows the lowest DA fiber density of all investigated cortical structures. TH immunoreactivity in primate primary visual cortex V1 is restricted to layer 1, and few fibers reach layer 6 as well in V2. The same layers are targeted by dopaminergic fibers in the rat primary visual cortex.

35A slight increase in fiber density is seen in primary auditory cortex. In the gerbil, D1 receptors are found in infragranular layers mainly associated with pyramidal neurons. TH positive fibers loosely innervate auditory cortex layers 1 and 6 in primates. In primary somatosensory cortex, labeled fibers are also found in layer 5. A further increase in dopaminergic innervation is then observed in the association cortices of the temporal, parietal, and frontal lobe, where the aforementioned bilaminar profile emerges.

36Dopaminergic fibers have also been reported to innervate subcortical structures that are involved in processing sensory stimuli. Dopamine axons reach the LGN of the thalamus (LGN, visual nucleus). Here, D1 and D2 receptors are found in excitatory relay neurons and in inhibitory local interneurons. D2 receptors are also present in the thalamic medial geniculate nucleus (MGN, auditory nucleus) and in the ventrobasal complex (VB, somatosensory nucleus).

37While these first-order thalamic nuclei mainly relay sensory input to the cortex, the second-order thalamic nuclei are part of cortico–thalamo–cortical loops and could support higher brain functions operating on sensory information. The second-order thalamus is more strongly innervated by DA neurons than its first-order counterpart, reaching the levels of highest cortical density in some nuclei. Thalamic DA fibers are much denser in primates than in rodents. Dopaminergic afferents in the thalamus stem from multiple sources including the hypothalamus, ventral mesencephalon, the PAG, and the lateral parabrachial nucleus, possibly hinting at the existence of a distinct “thalamic dopaminergic system”. Interestingly, the densest dopaminergic projections to the thalamus are found in those second-order nuclei that are linked to frontal and limbic cortex, e.g., the mediodorsal (MD) nucleus and the midline nuclei. In fact, these DA projections share many anatomical features with the strong meso-prefrontal DA system. These findings suggest that DA could modulate higher-order frontal lobe functions both at the cortical level and by controlling its associated thalamic nuclei. Finally, it has been noted that VTA axon terminals and terminals from thalamic MD neurons converge in mPFC layer 5, forming a synaptic triad as described above. Thus, dopaminergic modulation of thalamic function might also have extra-thalamic components.

38The main auditory midbrain nucleus, the IC, is targeted by TH-immunoreactive nerve terminals arising from the subparafascicular thalamic nucleus, and the IC has been shown to express D2 receptors. Dopaminergic afferents to the superior colliculus, a crucial midbrain structure for oculomotor and visual processing, are sparser in comparison.

39Dopamine neuron activity is classically described as being time locked to the presentation of rewarding stimuli. Phasic firing of action potentials by DA neurons especially in the medial mesencephalon communicates a reward prediction error that scales with the difference between predicted and actual reward. These neurophysiological findings form the basis for the large number of studies that have investigated DA’s role in motivation, appetence, and reward-related learning (“motivational salience”). More recent experiments have revealed the existence of functionally distinct subgroups of DA neurons in particular in the lateral midbrain that are activated both by rewarding and aversive events. This suggests that these neurons are responsive to a more general group of sensory stimuli that are behaviorally relevant and should trigger an appropriate coordinated response (“cognitive salience”). The physiological effects of DA on sensory information processing, however, are not yet well understood. Given the sparse innervation of sensory structures, comparatively few studies have addressed the role of DA in directly modulating sensory inputs (Table 6).

40In rat and cat dorsal LGN, in vivo iontophoretic activation of D1 receptors produced inhibition, whereas engagement of D2 receptors resulted in excitation of extracellularly recorded relay neurons. Intracellular recordings in rat and mouse LGN slices, however, found that D1 receptors lead to an excitatory membrane depolarization in relay neurons, whereas D2 receptors on local GABAergic interneurons produced inhibition in postsynaptic neurons, in complete contrast to the in vivo results. It is currently unclear whether these diverging findings reflect dose-dependent effects, where application of small and large drug concentrations often yields opposite effects, or rather result from network mechanisms that differ in the in vivo and in vitro preparation.

41In primate primary visual cortex (V1), oscillatory activity (local field potentials) containing information about animated visual stimuli (movie clips) was enhanced in particular in supragranular layers following systemic administration of the DA precursor L-DOPA. However, focal application of DA failed to induce changes in neuronal activity, possibly reflecting the very sparse dopaminergic innervation of the occipital pole. In accordance with the notion that dopaminergic effects on visual processing are not mediated by early primary visual regions but possibly by higher-order brain areas, visually induced neuronal responses were directly modulated by DA in the monkey lateral PFC. Here, iontophoretically applied DA affected two distinct neuronal populations involved in encoding behaviorally relevant visual stimuli. In putative interneurons, DA inhibited activity in form of a subtractive shift in response levels with unchanged SNR. In putative pyramidal neurons, DA increased excitability in form of a multiplication in gain and enhanced SNR through a reduction of response variability across trials. By increasing the coding strength of sensory signals, DA could play a crucial role in how sensory information is represented, memorized, and interpreted in PFC.

42Acting on cognitive brain centers, DA might influence visual cortical processing by long-range interactions originating, e.g., in frontal cortex. Support for this hypothesis comes from exploring DA’s modulatory influence on PFC-guided allocation of visual attention in the macaque monkey. D1R were blocked by local antagonist injections into sites of the frontal eye fields (FEFs) that represented the same region of visual space (the “response field”) as simultaneously recorded neurons in visual cortex area V4. Prefrontal D1R antagonism caused the animals to saccade more frequently toward FEF response field targets, meaning this part of the visual field had grasped their attention. The response properties of V4 neurons were changed in a way that is consistent with a multiplicative increase in gain: first, there was an enhancement in the magnitude of responses to visual stimulation; second, the visual responses became more selective to stimulus orientation; third, the visual responses became less variable across trials. Thus, prefrontal top-down control over visual cortical neurons during visual attention is under the influence of D1 receptors.

43In the auditory system, both inhibitory and – less frequently – excitatory responses in IC neurons were observed after local iontophoretic application of DA in awake mice. The importance of subcortical DA for auditory information processing was underscored by showing that overexpressed MGN D2 receptors in a schizophrenia mouse model reduce the efficiency of synaptic transmission of excitatory thalamocortical afferents in auditory cortex. These changes were accompanied by behavioral deficits in auditory perception (acoustic startle response). Because thalamocortical afferents in visual or somatosensory cortex were not affected, studying this mouse model might help to provide a neuronal mechanism that links dopaminergic dysfunction to the generation of auditory hallucinations in schizophrenia. Auditory processing is modulated by cortical DA receptors as well. Systemic administration of D1 receptor antagonists or local injections directly into auditory cortex of gerbils impaired discrimination learning of acoustic stimuli, while D1 receptor agonists improved the animals’ performance.

44Finally, the representation of somatosensory information is also strengthened by DA: in vivo depletion of DA in mouse striatum worsened the laterality coding of whisker deflections in medium spiny neurons, i.e., the projection neurons of this structure, indicating that tactile acuity requires dopaminergic input. Intracellular recordings in rat VB thalamic slices showed that DA increases neuronal excitability in two different ways: activation of D1 receptors leads to membrane depolarization, whereas D2 receptors facilitate action potential discharge. In contrast, in rat S1, iontophoretic DA application produced inhibitory effects on neuronal firing.

45The discussed studies provide compelling evidence that the representation and processing of sensory information is heavily regulated by the brain’s monoamine transmitters. Despite their widespread cortical and subcortical targeting, the monoaminergic systems are by no means blurry, brain-wide modulators of neuronal activity. They send regionally and laminar-specific projections, which may allow them to control both feed-forward and feed-back influences on sensory signaling by distinct mechanisms. These regional and laminar differences are particularly pronounced in the macaque brain (Figure 1).

46The indolamine serotonin has prominent projections to the primary sensory areas and innervates all cortical layers including the thalamo-cortical input layer 4. Serotonin is well suited to regulate the sensory input stages by acting in these primary sensory areas and in the first-order thalamic relay nuclei. Despite some degree of variability between studies on the serotonergic modulation of sensory processing, the emerging pattern is that it dampens neuronal responses and reduces gain. The reduction in gain may reduce the salience of a sensory stimulus. Alternatively, if it only affects the spontaneous but not the stimulus-driven response, it may increase a neuron’s SNR and hence effectively increase stimulus salience. Probing the serotonergic modulation of sensory processing during perceptually driven behavior may resolve this seeming discrepancy. In general, however, the experimental data show that serotonin directly modulates sensory processing as early as the feed-forward sensory input stage.

47In contrast to serotonin the catecholamine DA has a pronounced fronto-occipital gradient (Figure 1), is less abundant in sensory cortices, particularly sparse in the primary visual cortex, and only weakly active in granular layers. This suggests that dopaminergic effects on sensory processing are not mediated primarily by local modulation of early sensory input stages (e.g., ) but instead by modulating long-range recurrent cortico-cortical and cortico-bulbar interactions originating in the strongly innervated supragranular and infragranular layers, respectively, e.g., of the frontal lobe cognitive control centers. Its modulatory effects stem from a variable and complex combination of inhibition and excitation in the targeted circuits. The net result is likely an increase in SNR, which is based both on additive operations, e.g., predominant suppression of responses to non-preferred stimuli (“sculpting inhibition”), and on multiplicative operations, e.g., increase in gain and response reliability (reduction of trial-to-trial variability). Thus, the role of DA could primarily be to adapt the read-out of sensory circuits to best serve task demands and behavioral goals.

48The catecholamine NA, conversely, may play a role in between these two extremes. It has a less pronounced fronto-occipital gradient than DA and, like serotonin, substantial projections to the primary sensory cortices, including the primary visual cortex. But in contrast to serotonin, the innervation of the granular layers is relatively sparse (Figure 1). The modulatory effects of NA on sensory processing are diverse. This variability may reflect adaptable modulation depending on the behavioral state of the animal, in line with the notion of optimizing the neuronal gain for behavior.

49Taken together, the current literature argues for a prominent complementarity in sensory neuromodulation by monoamines. This complementarity is prominent anatomically (Figure 1B), resulting in both direct bottom-up and indirect top-down control over sensory signaling by the indolamine serotonin and the catecholamines NA and DA, respectively. One of the main challenges will now be to dissect the individual contributions of the anatomically and functionally separable monoamine subsystems in shaping how sensory information is represented, processed, and evaluated by the brain’s sensory, cognitive, and motivational networks (Figure 1A). We believe that key to such insights will be the combination of increased specificity and precision when targeting these neuromodulatory systems with well characterized behavior.

50All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

51The 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.