A neuropeptide code for itch

Abstract

Itch is one of the most primal sensations, being both ubiquitous and important for the well-being of animals. For more than a century, a desire to understand how itch is encoded by the nervous system has prompted the advancement of many theories. Within the past 15 years, our understanding of the molecular and neural mechanisms of itch has undergone a major transformation, and this remarkable progress continues today without any sign of abating. Here I describe accumulating evidence that indicates that itch is distinguished from pain through the actions of itch-specific neuropeptides that relay itch information to the spinal cord. According to this model, classical neurotransmitters transmit, inhibit and modulate itch information in a context-, space- and time-dependent manner but do not encode itch specificity. Gastrin-releasing peptide (GRP) is proposed to be a key itch-specific neuropeptide, with spinal neurons expressing GRP receptor (GRPR) functioning as a key part of a convergent circuit for the conveyance of peripheral itch information to the brain.

Interest in itch and scratch-reflex mechanisms is more than a century old^1–7, and it has long been appreciated that itch and pain appear to be intricately entangled. Pain inhibits itch, and opioids correspondingly inhibit pain but induce itch^8,9. In addition, many pain-sensing sensory neurons (nociceptors, also known as C-nociceptive fibres) can also transmit itch^2,7. Historically, therefore, the focus of investigation of itch has been its relationship with pain^{1,5,7,10–13}.

Different types of itch can be classified in various ways. Depending on whether histamine release is involved, itch can be broadly divided into histaminergic itch (such as acute allergic itch) and non-histaminergic itch, both of which are mediated by a wide variety of itch-related receptors¹⁴. Itch can also be elicited by various external physical and non-physical stimuli, which can be used as a method of classification. For example, chemical itch is evoked by pruritogens that penetrate the epidermis of the skin, whereas mechanical itch is induced by innocuous light touch (such as that of a wool fibre) on the skin^1,5,15. Seeing others scratch or merely viewing an image of a bug crawling on the skin may elicit the urge to scratch^16,17, referred to as ‘contagious itch’. As a form of interoception¹⁸, itch can also occur spontaneously and can be modulated by internal states or homeostatic demands¹⁹. Such spontaneous itch can be physiological (many animals scratch spontaneously to clean their fur or maintain their body temperature^1,20), psychological (triggered by stress or anxiety, for example)²¹ or pathological (such as neuropathic or systemic itch)^1,2,22,23. Finally, itch can be classified as acute or chronic according to the duration of itch sensation: acute itch is transient, whereas chronic itch lasts for more than 6 weeks in humans^14,24, or more than 1 hour (and up to months) in mice (in which chronic itch is induced via periodic chemical sensitization of the skin for several days^25,26 or neural tissue injury^26–28). The lack of effective therapeutics for chronic itch, most types of which are refractory to antihistamines, poses a significant burden to patients and may compromise their quality of life to a degree often comparable to that for chronic pain²⁹.

A fundamental question in the somatosensory field is how different types of sensory modalities are encoded and represented in the brain. With respect to itch, most research has focused on how primary sensory neurons and/or the spinal cord differentiate between itch and pain; however, the answer to this century-old puzzle continues to elude researchers. In this Perspective, I describe a neuropeptide code hypothesis of itch, with an emphasis on the neuropeptides, signalling mechanisms and spinal neural circuits that encode itch in mice. I critically appraise the current state of knowledge of the spinal circuitry of itch and propose a model in which classical neurotransmitters and neuropeptides play distinct roles in conveying itch information to the CNS. Along the way, I discuss technical and conceptual challenges. Specifically, I explore alternative explanations for some conflicting results and revisit several commonly held views. The peripheral neuroimmune mechanisms of itch are not the focus of this Perspective, which are well covered in recent reviews^14,30,31. Brain mechanisms of itch and chronic itch are tangentially touched upon whenever relevant.

Coding theories of itch

Somatosensory information — including that important for itch, pain, temperature sensing and touch — is relayed from the skin to various categories of primary sensory neurons with cell bodies residing in the dorsal root ganglia (DRG) and trigeminal ganglia. These sensory neurons project to second-order interneurons or directly synapse with projection neurons in the dorsal horn of the spinal cord (FiG. 1). The spinal dorsal horn is divided into several morphologically distinguishable laminae. Laminae I and II receive direct inputs from slowly conducting unmyelinated C fibres and faster-conducting myelinated Aδ fibres, together encoding pain, itch, pleasant touch and temperature, whereas laminae III and IV are innervated mainly by Aβ low-threshold mechanoreceptors (LTMRs) that convey touch and proprioceptive information^32,33 (FIG. 1a).

Fig. 1 |. Dorsal horn architecture and coding theories for itch.

a | The dorsal horn, illustrating its division into several distinct laminae (I–IV). Laminae I and II are innervated by unmyelinated C fibres and fast-conducting myelinated Aδ fibres, which together convey information relevant to pain, itch, pleasant touch and temperature sensing. Interneurons in laminae I and II additionally receive mechanical itch information from Aβ low-threshold mechanoreceptors (LTMRs) via an indirect route (not shown)^58,188. Lamina II can be subdivided into two layers: unmyelinated peptidergic C fibres terminate mainly in lamina I and the outer layer of lamina II and convey information that requires slow conduction velocities, such as that important for chemical itch and chemical pain^33,247. The inner layer of lamina II is innervated mainly by unmyelinated non-peptidergic C fibres that convey pain, itch and other somatosensory information. Laminae III and IV are typically innervated by myelinated Aβ LTMR fibres that convey mechanical touch, mechanical itch and proprioceptive information^{32,33,58,188,193}. Lamina III can be further divided into two layers. The outer layer contains TAC2 neurons, which mediate mechanical itch⁵⁸. Dorsal horn projection neurons in lamina I and possibly laminae III and IV, integrate and relay itch information to the somatosensory cortex (not shown)^61,75,248. b | The main current coding theories of itch. According to the specificity (or labelled line) theory, information about itch-inducing and painful stimuli is conveyed via distinct neural pathways from the skin to the spinal cord³. There are itch-specific neurons in the periphery, which are activated by itch stimuli exclusively.The selectivity (or population coding) theory proposes that pruriceptors (represented by a circle filled half in red and half in blue) are a small subset of nociceptive neurons. Pruriceptors respond to both itch-inducing and painful stimuli, whereas a much larger population of nociceptors (red) convey pain information exclusively^3,19. This theory suggests that distinct labelled lines for pain and itch exist in the spinal cord and that pain transmission can activate spinal inhibitory neurons to inhibit or occlude itch by crosstalk between the two labelled lines^34,35. According to the intensity theory, weak activation of nociceptors provokes itch, whereas strong activation of the same population of neurons evokes pain, via different modes of firing pattern (tonic for weak and burst for strong stimuli)^3,10,12,13. Like the selectivity theory, the pattern theory^3,34 postulates that within the larger population of sensory nociceptive neurons (red) there is a subset of pruriceptors (half red, half blue). However, it suggests that the distribution of these pruriceptors may be different in the epidermis and dermis (not shown). Therefore, depending on the nature and location of a given stimulus, a mixture of pruriceptors and nociceptors will be differentially activated, leading to different population firing patterns (tonic versus burst) and thereby encoding itch or pain specificity^34,106. This theory does not explain how itch and pain information is encoded in the spinal cord. However, it can be assumed that itch and pain are conveyed separately in the spinal cord after the encoding by primary sensory neurons.

Many theories have been proposed to explain how itch is encoded and distinguished from pain in the sensory neurons and/or spinal cord^3,34–37 (FIG. 1b). For many, the most influential and attractive theory (perhaps for its simplicity) is the labelled line (or specificity) theory, which is deeply rooted in Johannes Müller’s ‘law of specific nerve energies’, first published in 1835 (REF.³⁸). It postulates that itch and pain are conveyed by dedicated and distinct neural pathways that begin in specialized peripheral detectors and continue within the spinal cord³. In the past decade, genetic and chemical ablation approaches, combined with behavioural assessment, have been used to try to identify itch-specific neurons (TABLE 1). However, as discussed in more detail later, extensive searching has yet to conclusively identify the components of a labelled line for itch in peripheral sensory neurons^39,40. It could be argued that this theory is incompatible with the intrinsic properties of the primary sensory neurons, which are typically polymodal^41,42.

Table 1 |.

Approaches used to define itch-specific neuropeptides and neurons

Manipulation	Criteria	Advantages	Disadvantages
Assessment of behavioural correlates of itch
Loss of neuropeptide function in mice (via genetic KO or siRNA-mediated knockdown)	Impaired itch behaviour and normal pain behaviour	Direct evidence Sensory neuron-specific KO is feasible	Compensation: multiple related neuropeptides and receptors Time consuming to generate double-KO or triple-KO mice and CKO mice Spinally restricted CKO may not be feasible
Chemogenetic or optogenetic inhibition or viral, chemical or intersectional genetic ablation of neurons	Impaired itch behaviour and normal pain behaviour	Spinally restricted inhibition/ablation possible Sensory neuron-specific inhibition/ablation possible Ablation can be fast and efficient	Low level of fidelity of a Cre line Neurotoxin approach may have off-target effects Most acute pain behaviour cannot be appropriately evaluated after cervical inhibition or ablation Genetic compensation Indirect effects
Loss of neuropeptide function via pharmacological approaches	Impaired itch behaviour and normal pain behaviour	Acute and convenient Agonists may be available Dose-dependent scratching behaviour can be assessed	Off-target effects Potential presynaptic effects originating in primary afferents Pharmacological artefacts
Gain of neuropeptide function via pharmacological approaches	Induced itch behaviour and normal pain behaviour	Acute and convenient Agonists may be available Dose-dependent scratching behaviour can be assessed	Off-target effects Potential presynaptic effects originating in primary afferents Pharmacological artefacts Ceiling or floor effects
Chemogenetic or optogenetic activation of neurons	Induced itch behaviour and normal pain behaviour	Spinal cervical cord-restricted activation Validation using BB-sap possible Sensory neuron-specific activation of central or peripheral fibres possible	Low level of fidelity of a Cre line Chemogenetic or optogenetic artefacts Artificial behaviour can only suggest the capacity of fibres and is not necessarily an indicator of an endogenous function of manipulated neurons
Assessment of neural correlates of itch
Electrophysiological recording of neuronal responses to chemical itch, pain, cooling or scratching	Inhibited by pain stimuli	Single-cell resolution Cell-type recording	Low level of fidelity of Cre or eGFP mouse lines Lack of specificity of some chemicals (e.g. capsaicin) May not simulate neuropeptide release in freely behaving animals

BB-sap, bombesin–saporin; CKO, conditional knockout; eGFP, enhanced GFP; KO, knockout; siRNA, small interfering RNA.

A competing theory is the selectivity (or population coding) theory, which assumes that most nociceptors are insensitive to pruritogenic stimuli and that the activity of this large population of nociceptors can inhibit that of the small subset of nociceptors that are also pruriceptors (itch-sensing neurons)^3,35,43. In other words, a painful stimulus will simultaneously activate both itch-insensitive nociceptors and pruriceptors (but will mask the sensation of itch through cross-inhibition within the spinal cord)³⁵: itch will be transmitted only when pain is absent. However, in light of the polymodality of most sensory neurons^42,44, it remains unclear whether itch-insensitive nociceptors exist. Moreover, cooling-sensitive neurons, which account for fewer than 10% of the DRG neurons in rodents^45–47, are efficacious in inhibiting itch^48,49 and even pain⁵⁰. This suggests that one sensory modality may inhibit another on non-quantitative grounds and that other factors, such as synaptic strength and wiring, may also play a part.

Several other theories have been proposed to explain itch coding (FIG. 1). The intensity coding theory posits that itch and pain are transmitted by the same fibres, but are distinguished by the different firing intensities that they induce^3,10,12,13. This theory was first proposed by von Frey, who used a plant bristle to stimulate the skin and found that a weak mechanical touch evoked itch, whereas a strong one led to pain⁴. However, this theory is not supported by more recent findings that, for some cutaneous fibres, either electrical or optogenetic stimulation can evoke itch in an intensity-dependent manner but does not evoke pain at any intensity tested^51–53.

The pattern theory hypothesizes that distinct patterns of firing evoked in sensory neuron populations (nociceptors and pruriceptors) that display partially overlapping innervation patterns in the skin by itch-inducing or painful stimuli may encode these two sensations^3,34 (FIG. 1b). A variation of this theory is the spatial contrast theory³⁷, which postulates that itch arises from selective activation of pruriceptors, but that surrounding nociceptors must also be quiescent, or their inhibition of pruriceptors reduced, for itch to be conveyed³⁷.

These coding theories have focused primarily on the coding of itch versus pain by the primary sensory neurons. However, as I will outline herein, it is now becoming clear that how itch and pain information is distinctly relayed to the spinal cord may be more important for our understanding of itch coding.

Chemical itch coding

Chemical itch (a common cause of acute itch) occurs as a result of a cascade of inflammatory and immune responses that involve receptors in skin, immune and nerve cells^14,30. A major breakthrough in our understanding of the coding of chemical itch over the past decade was the identification of several itch-specific neuropeptides and receptors.

The GRP–GRPR neural pathway.

Studies in which genome-wide mapping and behavioural screening were performed in search of spinal dorsal horn lamina-specific nociceptive genes⁵⁴ led, in 2007, to the serendipitous discovery of gastrin-releasing peptide (GRP) receptor (GRPR), a G-protein-coupled receptor (GPCR), as the first itch-specific receptor in the mouse spinal cord⁵⁵. GRPR-expressing neurons (hereafter referred to as GRPR neurons) make up approximately 20% of the excitatory interneurons in laminae I and II and receive inputs from C fibres, Aδ fibres, Aβ LTMRs and other spinal itch neurons^56–58. They convey itch information to the parabrachial nuclei via projection neurons^59–62. The neurotoxin bombesin–saporin (BB-sap) binds to and is internalized by GRPR neurons; however, it is not internalized by those expressing the related receptor neuromedin B (NMB) receptor (NMBR)⁶³, to which it also binds⁶⁴. This thereby affords a highly selective method to chemically ablate GRPR neurons in mice⁶⁵. By this method, ablation of spinal GRPR neurons abolished itch transmission without impairing pain behaviours⁶⁵. It could be argued that the normal pain behaviour observed in mice treated with BB-sap could be due to genetic compensation: this might occur if the population of neurons that were ablated contained cells sensitive to a mixture of sensory modalities, each of which accounted for only a small proportion of the total number of neurons sensitive to a given modality (as is the case for spinal Grp-expressing neurons (see later))⁵³. Additionally, the potential for rewiring of the dorsal horn circuitry as a result of developmental and genetic ablation of a given population of neurons may make it difficult to assess their functional requirement in the adult spinal cord⁶⁶. However, these confounding factors should be considered together with the results of gain-of-function studies. Pharmacological or optogenetic activation of GRPR neurons evokes itch-related but not pain-related scratching behaviour^55,67, suggesting that these neurons are dedicated to the transmission of itch alone. In my view, these combined findings rule out the possibility of genetic compensation or developmental rewiring, and effectively separate itch from pain at the spinal level (TABLE 1).

The search for itch-specific neurons or receptors is complicated by the fact that, in humans, some commonly used experimental pruritogens, such as histamine and cowhage (a non-histaminergic pruritogen), can also evoke pain and some algogens (such as capsaicin) can elicit itch^68,69. Likewise, in mice, intraplantar injection of non-histaminergic pruritogens (such as chloroquine, an antimalaria drug that induces pruritus in mice and in some African patients⁷⁰, bovine adrenal medulla 8-22 peptide (BAM8-22), PAR2 and SLIGRL) induces mechanical and thermal pain that is dependent on the cation channel TRPA1 (REFS^71,72). Thus, what might be thought to be a ‘pure’ pruritogen may activate a putative itch-specific neuron to transmit nociceptive information rather than inducing itch-related scratching behaviour. To overcome this limitation, an alternative approach takes advantage of the fact that itch can be inhibited by pain and cooling. This idea was first postulated by Sherrington⁶, who suggested that this inhibition was mediated through central mechanisms^57,73–75. However, it is only recently that it has been adopted as a criterion to define itch-specific neurons; electrophysiological recording confirmed that both chemical pain (induced using mustard oil) and a cooling agent (menthol) inhibit the activity of GRPR neurons⁵⁷. Thus, converging evidence has demonstrated that GRPR neurons fulfil three key criteria for defining itch-specific neurons: they transmit itch information but not pain information (as demonstrated by a loss-of-function approach), their activation evokes itch-related but not pain-related behaviour and they are inhibited by pain and cooling (as shown by electrophysiological recording) (TABLE 1).

GRP, a 27 amino acid peptide, is the ligand for GRPR and was for a long time considered to be a mammalian homologue of the amphibian neuropeptide bombesin^64,76–78. Amphibians, however, were later discovered to also possess GRP^79,80, indicating that this peptide is evolutionarily conserved in vertebrates. Grp mRNA has been shown to be abundantly expressed in lamina II of the mammalian dorsal horn, and GRP-immunoreactive fibres have been detected in the dorsal horn of the mammalian spinal cord^{53,54,59,81–83}. The origin of these fibres has recently attracted renewed attention, and several studies have investigated potential sources of GRP and, in particular, whether the neuropeptide is produced by spinal or primary sensory (DRG) neurons.

GRP and/or its mRNA have been detected via in situ hybridization, reverse transcription–PCR, western blot and immunohistochemical staining in DRG neurons across various animal species, including rodents, dogs, monkeys and humans^{53,55,59,81,82,84–93}. By contrast, several large-scale RNA-sequencing studies have reported that Grp mRNA was not detected in these neurons^94,95, which has been considered by some researchers to provide evidence for an absence of Grp expression in DRGs. However, absence of evidence is not evidence of absence, and it is possible that the low transcript levels of Grp mRNA could simply fall below the empirical detection threshold of these methods, a recognized limitation of such high-throughput assays^96,97. Indeed, an earlier study suggested that the level of Grp mRNA transcripts in DRGs is at least ten times lower than that of MAS-related G-protein-coupled receptor A3 gene (Mrgpra3) mRNA transcripts⁵⁹. It is also important to note that the relative level of mRNA detected for a given gene should not be equated with that of the corresponding protein nor with its functional importance⁹⁸. For instance, the mRNA for a given neuropeptide might be barely detectable because there is already a large quantity of its bioactive form stored in large dense-core vesicles sufficient for activity-dependent release⁹⁹. It has long been recognized that some neuropeptides, whose expression is below a detectable level, become detectable only when prolonged or repeated environmental stimuli are encountered⁹⁹. Taking these factors into account, I consider that there is strong evidence for the presence of Grp mRNA in DRGs.

In line with this suggestion, two research groups that performed dorsal root rhizotomy followed by immunostaining with a GRP antibody concluded that the GRP-positive fibres present in the mouse dorsal horn are predominantly of primary afferent origin^26,81, corroborating several studies conducted independently almost 40 years ago^93,100. However, two other groups using the same approach pointed to a spinal cord origin for these fibres^82,83. The reason for these discrepancies remains unclear. However, several methodological issues need to be considered regarding the detection of Grp mRNA and GRP, as discussed previously⁵⁹. In particular, there is a widespread notion that the GRP antibody used in these experiments is not specific owing to some background immunostaining problems associated with immunohistochemical staining^59,83. However, it is worth emphasizing that — using a variety of approaches that included western blotting⁵⁹, the creation of Grp-knockout (KO) mice²⁶, peptide preabsorption testing^81,82 and appropriate dilution of the antibody⁸³ — all four of the teams examining the source of GRP using the GRP antibody confirmed its specificity^{26,53,59,81–83}. Moreover, the GRP antibody is known to be one of the most reliable antibodies for cell type classification in the suprachiasmatic nucleus (SCN)^101–104. Although the level of Grp mRNA in the SCN is very low¹⁰⁴, immunohistochemical staining indicates that GRP is abundant in the central axonal terminals of GRP neurons¹⁰⁴, similar to the findings in DRG^26,53. I hypothesize that the low transcriptional activity of Grp and the rapid onset of transcription, translation and axonal transportation of GRP in response to itchy stimuli could all be important mechanisms by which the level of GRP is maintained. From the findings taken together, I believe that there is compelling evidence that some DRG neurons express GRP and that the amount of GRP present in the central terminals of primary afferents is adequate for an animal’s daily use.

To examine the functional role of GRP, the phenotype of mice in which there is a conditional KO (CKO) of Grp in sensory neurons has been analysed⁵³. Grp-CKO mice displayed deficits in non-histaminergic itch but not in histamine-evoked itch or pain behaviours⁵³, which phenocopies observations in mice with global KO of either Grp^53,56,105 or Grpr^55,65 and is further corroborated by studies that induced a pharmacological blockade of spinal GRPR⁸⁷. A knock-in Grp^iCre mouse line that recapitulates endogenous Grp expression has also been generated and used to show that optogenetic stimulation of cutaneous GRP-expressing fibres at frequencies up to 20 Hz evokes itch-related scratching behaviour, a finding that is consistent with the results of chemogenetic studies⁵³. Of note, the broad arborization of GRP fibres in the epidermis is in line with the finding that cutaneous itch arises from the epidermis or the dermoepidermal junction but not the dermis^5,106.

One could interpret these results as evidence that GRP neurons are part of a labelled line for itch, in support of the specificity theory. However, for GRP neurons to qualify as part of an itch-specific circuit, they should not respond to painful stimuli. It has long been recognized that some GRP neurons also express substance P (SP)^59,84,92, a nociceptive neuropeptide that activates several neurokinin receptors with different affinities^107,108. NK3R is expressed in lamina II excitatory and inhibitory neurons¹⁰⁹, and during nociceptive transmission, SP is released to activate spinal NK3R neurons^110–113. It is notable that SP has also been implicated in itch transmission at the spinal level^87,114; however, determining its site of action is complicated by its crosstalk with other neurokinin receptors and its peripheral action in itch¹¹⁵ (TABLE 1). Conceivably, a subset of neurons expressing both GRP and SP may therefore participate in nociceptive transmission via NK3R, in addition to transmission of itch via GRPR, meaning that GRP neurons cannot be considered part of a labelled line for itch. In the same vein, a recent study showed that a subset of primary sensory neurons expressing Mrgpra3 — previously thought to be itch-specific neurons³⁹ — can also transmit nociceptive information⁴⁰, further arguing against the existence of an itch-specific circuit in primary sensory neurons.

A neuropeptide code for chemical itch.

As noted earlier herein, the key question of how itch information is conveyed to the spinal cord has attracted relatively little attention compared with the coding by primary sensory neurons. Cutaneous sensory neurons possess two cardinal features. First, they have a certain level of modality specificity, manifesting itself in their topographic projection patterns in the spinal cord^{32,33,42,116,117} (FIG. 1a). Second, they possess a high degree of plasticity and diversity and can respond to different types of environmental stimulus^33,41,42. By contrast, electrophysiological studies indicate that the modular organization of the dorsal horn circuitry discriminates between modality-specific inputs and conveys this information to the brain^118,119. This prompts the proposition that the connectivity between the primary afferent fibres and their central targets dictates modality-specific transmission.

How does the coding of itch fit into this general schema? The identification of spinal GRPR neurons provides one of the first pieces of evidence supporting the existence of itch-specific circuitry in the spinal cord. Another clue comes from an important observation that there is a significant reduction of GRP immunoreactivity in neuronal fibres within the mouse spinal cord after robust scratching elicited by chloroquine⁵³. Chloroquine-evoked itch is mediated by primary sensory neurons that innervate spinal GRPR neurons^53,59,87. The reduction in the intensity of GRP immunoreactive fibres in the superficial dorsal horn after robust scratching suggests a depletion of GRP from the axonal terminals as a result of its release (and is presumably accompanied by activation of GRPR⁵³). By contrast, central SP immunoreactivity remains unaffected by scratching⁵³, implying that there is a lack of, or negligible, SP release during the itch–scratch cycle. Immuno-electron microscopy has revealed that, in some cases, both GRP and SP are stored in the axonal terminals that directly synapse with GRPR neurons, consistent with their partial colocalization in DRG neurons^59,84,87,92, and evidence indicates that SP is released in the spinal cord in response to a variety of noxious thermal, mechanical and chemical stimuli¹⁰⁸. This suggests that some peripheral nociceptors expressing both SP and GRP will be activated by pruritogenic stimuli, but that the nociceptive information encoded by SP is occluded from accessing the dorsal horn pain circuits because only GRP is released into the spinal cord.

In light of recent advances, I propose here a neuropeptide code model that incorporates the merits of various coding theories (FIG. 2). It posits that it is the neuropeptides that encode the specificity of itch.

Fig. 2 |. Hypothetical model for the neuropeptide coding of itch and pain.

The schematics illustrate the neuropeptide code model, which suggests that itch and pain are encoded and transmitted separately through discrete neuropeptides. Substance P (SP) is shown as an example of a pain-specific neuropeptide transmitter, whereas gastrin-releasing peptide (GRP) is shown as an example of an itch-specific neuropeptide transmitter. Other types of itch-specific or pain-specific neuropeptides may exist but are not shown for illustrative purposes (see also FiG. 3). a | Pruriceptors and nociceptors partially overlap, which is manifested in co-expression of GRP and SP. In the absence of itch or pain stimuli, neither GRP nor SP is released from its terminal, resulting in a lack of itch or pain transmission. For simplistic purposes, neurotransmitters are shown only in the cell bodies of nociceptors and pruriceptors. b | Under itchy conditions, pruriceptors are activated either spontaneously (resulting in tonic firing) or in response to itch stimuli (resulting in bursting firing). This activation is proposed to release GRP and glutamate into the spinal cord, activating GRP receptor (GRPR)-expressing neurons (GRPR neurons), which transmit itch information exclusively to the brain. A smaller number of nociceptors that co-express GRP and SP (indicated as the overlap of the ovals representing these two cell populations) are also activated but fail to release SP, owing to tight control of presynaptic neurotransmitter release via channels such as sodium or calcium channels. Therefore, no pain information is transmitted. This subset of neurons may also release a small amount of GRP at synapses with spinal nociceptive neurons expressing NK receptors (NKRs), but will fail to activate them owing to the lack of GRPR in these neurons. The GABAergic inhibitory interneurons that innervate itch-specific spinal neurons remain inactive. c | Under painful conditions, a population of nociceptors that contain some pruriceptors are concomitantly activated via a bursting firing pattern that provokes the release of SP and glutamate from these neurons^132,249,250, which activate spinal nociceptive neurons expressing NKRs. A small amount of GRP may be co-released but fails to activate these neurons owing to the lack of GRPR expression. Meanwhile, a smaller subset of pruriceptors expressing both GRP and SP or GRP alone (not shown) (indicated as the overlap of the ovals representing these two cell populations) is also activated and releases GRP and glutamate, which marginally activate GRPR neurons. A small amount of SP may also be co-released from pruriceptors/nociceptors via the central terminals that innervate GRPR neurons, but may fail to activate them owing to the lack of NKRs in these neurons. A concurrent activation of spinal GABAergic inhibitory neurons by SP and/or glutamate, however, inhibits the activity of GRPR neurons and thereby blocks itch transmission (see also Supplementary Fig. 1). Depending on the properties and location of the stimuli encountered, therefore, a balance between the activation of pruriceptors, nociceptors and spinal inhibitory neurons dictates whether the output of itch is completely or partially blocked. In the latter scenario, both itch information and pain information can be transmitted to the brain (not shown).

A signature of chemical itch is the relatively slow onset of scratching behaviour after exposure to the stimulus^5,55, perhaps due to the time required for the pruritogens to initiate a cascade of allergic and inflammatory events^7,14,120. For example, there is a latency of several minutes between chloroquine injection and the onset of scratching behaviour in mice⁵⁵. Chloroquine triggers the release of a plethora of proinflammatory mediators^70,121,122, many of which have pruritogenic properties and may, in turn, activate their respective receptors (including, but not limited to, MRGPRA3, Toll-like receptor 3 (TLR3), TLR7 and arginine vasopressin receptor 1A (AVPR1A))^{90,91,122,123}, to increase the excitability of pruriceptors. In line with this, the percentage of dissociated DRG neurons that responded to chloroquine, as measured by calcium imaging, is approximately two to three times that of those expressing what is proposed to be the major itch receptor for chloroquine, MRGPRA3 (approximately 4–5%)^{87,122,124–126}, indicating that additional receptors mediate chloroquine-evoked itch. It remains unclear to what extent MRGPRA3 contributes to itch elicited by chloroquine and other pruritogens, since only mice in which the genes encoding a cluster of MAS-related G-protein coupled receptors (a family that includes other itch-related or pain-related receptors) are knocked out has been examined for changes in chloroquine-evoked itch¹²². Furthermore, chloroquine is a promiscuous ligand that binds to a number of receptors¹²⁷. Likewise, no single GPCR can account for formalin-induced or mustard oil-induced chemical pain. These examples suggest that a repertoire of receptors and/or channels in primary sensory neurons (instead of one receptor for one pruritogen as widely presumed) synergistically orchestrate sequential inflammatory responses to a pruritogenic stimulus⁷². Logically, this makes sense: if one receptor (or one subset of pruriceptors) were dedicated to the transmission of only one pruritogenic stimulus, sensory neurons would quickly exhaust their arsenal, given the potentially vast number of natural pruritogens that exist. Instead, multiple skin, immune and neural pathways are thought to be involved in the transmission of itch evoked by one pruritogen^30,31,128.

Although this argument suggests that diversity in peripheral signalling mechanisms is required to coordinate the detection of varied and numerous pruritogenic stimuli, I suggest that it is not necessary or economical for dorsal horn interneurons to match this with multiple microcircuits to process the itch inputs. Itch is a subjective perception that (in humans, at least) may have a variety of qualities; however, all types of itch give rise to an unpleasant sensation that provokes an urge to scratch to counteract itch. To maximize the efficiency and accuracy of the coding that manifests the same motor output, I therefore argue that a process of integration and convergence of diverse peripheral itch inputs must occur and that this happens in the spinal dorsal horn.

The neuropeptide code model, therefore, postulates that a prerequisite for a subset of sensory neurons to participate in transmitting itch information is that the neurons contain itch-specific neuropeptides, such as GRP (FIG. 2). This hypothesis suggests that the activation of these neurons by pruriceptive stimuli would provoke them to release itch (but not pain) neuropeptides, which would bind to their receptors through monosynaptic connections^59,122. Thereby, itch information will be transmitted, while irrelevant, non-specific information will be filtered out or blocked (FIG. 2). Any leaky release of pain neuropeptides (under certain burst firing conditions; see later) would be rendered non-functional by the fact that the spinal itch-specific neurons lack the receptors for nociceptive neuropeptides. This is known as a mismatch phenomenon: the peptide release site and corresponding receptor expression are anatomically distinct¹²⁹, likely reflecting changes that have occurred in their evolutionary or ontogenetic past. For instance, GRPR neurons cannot be activated by SP to evoke pain-related scratching and/or biting behaviour¹³⁰, or by chemical pain or cooling⁵⁷, even though fibres containing both GRP and SP or those containing SP alone may directly innervate GRPR neurons⁵⁹.

A central question for the neuropeptide code model is how an itch neuropeptide is released independently of a pain neuropeptide. While recent studies suggest that SP is not released during the itch–scratch cycle, when some of these afferents co-expressing GRP and SP are presumably active^53,59, some confusion arose when researchers attempted to extrapolate the results obtained from electrophysiological recordings of spinal cord slice preparations to freely behaving mice. For example, it has been shown that burst firing or high-frequency stimulation of primary afferents, or the dorsal horn neurons they innervate, can evoke GRP or SP release under in vitro recording conditions^131,132. An earlier study found that the response of the dorsal horn GRP-sensitive neurons to monosynaptic C fibre stimulation could be completely blocked by 6-cyano-7-nitroquinoxaline-2,3-dione, a glutamatergic AMPA and kainate antagonist, under in vitro recording conditions, and thus claimed that it is glutamate instead of GRP that functions as an itch transmitter in C fibres¹³³. However, recent optogenetic stimulation of cutaneous GRP fibres of freely behaving mice even at a low frequency (1–5 Hz) could evoke itch-related scratching behaviour⁵³. Even contagious itch induced by a rather weak visual stimulus in the absence of physical stimulation corresponds to GRP release in the SCN¹⁰⁴. Moreover, rodents are known to engage in frequent self-scratching and/or grooming behaviours that enable them to maintain various physiological functions²⁰, which may imply that there is a tonic release of itch-related and grooming-related neuropeptides at the baseline level in the absence of external stimuli in these animals. Therefore, repeated stimulation of primary afferents at high frequencies, which is normally required for neuropeptide release under in vitro electrophysiological recording conditions¹³⁴, is not a prerequisite for GRP release from primary afferents in freely behaving mice.

Although it is difficult to evoke neuropeptide release from primary afferents using an electrophysiological recording approach, presynaptic glutamate release could be readily evoked in spinal slice preparations^{57,133,135,136}. However, pain is not frequently experienced unless a stimulus reaches certain noxious thresholds that enable both glutamate release and SP release from primary nociceptive afferents^137–139. These inconsistencies reflect limitations of electrophysiological recording of spinal slice preparations in detecting the slow activation of GPCR responses evoked by neuropeptide binding, contrasting with that of fast glutamatergic transmission at postsynaptic sites¹⁴⁰. It can be envisioned that the brake mechanism that is in place in the central terminals of primary afferents to gate the release of pain transmitters must be much tighter than the machinery that gates the release of itch neurotransmitters. For example, discrete channels and/or Ca²⁺ sensors expressed in the central nerve terminals of primary afferents may subserve as gatekeepers for these two types of stimulus^141,142 that distinguish between resting and stimulated states¹⁴³.

Since itch (GRP) and pain (SP) neurotransmitters partially overlap in sensory neurons (FIG. 2) and since the gating of itch transmitter release is supposed to be weaker than that of pain transmitter release, it could be predicted that the release of pain transmitters in response to noxious stimuli will be accompanied by that of itch transmitters (FIG. 2). To override itch information that may coincidentally travel from the spinal cord to the brain, the neuropeptide code model suggests that spinal interneurons that inhibit itch transmission are activated by pain transmission (FIG. 2). This model postulates that SP, GRP and glutamate simultaneously co-activate spinal pain neurons, itch neurons and inhibitory circuits for itch, respectively.

The neuropeptide code may shed new light on the development of vicious itch–scratch cycles under certain pathological itch conditions. In these conditions, primary nociceptors that are usually involved in the detection and transmission of non-itchy stimuli would be recruited to express itch neuropeptides^26,59. This could make the nociceptive apparatus that otherwise could be recruited for efficacious inhibition of itch ineffective. Furthermore, the spinal circuitry responsible for inhibiting itch might be disabled^144,145 (see the discussion later), in conjunction with an augmented release of GRP that sensitizes and recruits some ‘silent’ GRPR neurons into action^26,59. All these mechanisms could underlie the observations that painful mechanical, thermal, chemical and electrical stimuli induce itch but not pain in patients with chronic itch^146,147.

In summary, the neuropeptide code model proposes that, in the somatosensory system, the specificity of itch is encoded by the selective release of slow-acting itch neuropeptides rather than by the activation of itch stimulus-specific receptors or a specific subset of sensory neurons. Itch stimuli predominantly activate pruriceptors but also marginally activate nociceptors in the skin; however, the central terminals of nociceptors are unable to release pain neurotransmitters, and the spinal pain circuits thereby remain quiescent (FIG. 2).

Neuropeptide coding of histaminergic versus non-histaminergic itch.

Human and animal studies have shown that histamine-dependent and histamine-independent itch are conveyed through parallel peripheral pathways^{37,56,126,148,149}, indicating that distinct neuropeptides may be used to convey information about these types of itch. However, the question of how histaminergic and non-histaminergic itch information is relayed to the spinal cord — and, more specifically, which neuropeptides are involved — has been insufficiently considered. As mentioned earlier, evidence suggests that GRP encodes only non-histaminergic itch⁵³. An important question is, therefore, whether there is an equivalent neuropeptide that encodes histamine-evoked itch.

NMB, another mammalian bombesin-like neuropeptide, binds to NMBR with high affinity but also cross-activates GRPR^64,150. NMB has several biological and pharmacological properties that are analogous to those of GRP, including its capacity to induce immediate dose-related scratching behaviours when injected into the spinal cord at picomolar concentrations^56,64,76,151, an indicator that it is directly activating receptors involved in itch transmission. Comprehensive genetic studies in mice have demonstrated that NMB and NMBR are required for histaminergic itch but not for chloroquine-evoked itch^56,105. Consistent with this notion, a recent study found that sneezing reflex, a histamine-dependent acute allergic response¹⁵², is mediated by NMB¹⁵³. Interestingly, normal histaminergic itch is observed in Nmbr-KO mice; however, it was shown that this is not due to compensatory effects but rather is due to a cross-activation of GRPR neurons by NMB in the absence of NMBR¹⁰⁵. By contrast, Nmb-KO mice exhibited deficits in histaminergic itch, even in the presence of NMBR, indicating that NMB is required for itch evoked by histamine^53,56. In support of these findings, Grpr-KO mice exhibit normal histaminergic itch, owing to the presence of normal NMB-NMBR signalling^65,105. Because NMBR neurons can be weakly activated by GRP, they additionally relay some non-histaminergic itch information⁵⁶, with this information converging in GRPR neurons through glutamatergic transmission^56,105 (FIG. 3a). These studies suggest that GRP–GRPR and NMB–NMBR signalling are parallel peripheral pathways that encode itch information in a non-complementary manner^53,56,105. The partial overlap of NMB expression and GRP expression that has been observed by immunohistochemical staining¹⁰⁵ suggests that NMB and GRP may also be co-released, leading to competition between histaminergic and non-histaminergic itch transmission, resulting from the binding of NMB to GRPR neurons, where it acts as a functional antagonist, and the binding of GRP to NMBR neurons, where it acts as a weak agonist^56,105. This may explain why chloroquine-mediated itch manifests itself in a histamine-independent manner at the behavioural level, even though chloroquine induces the degranulation of mast cells (releasing histamine), and neurons expressing the chloroquine receptor MRGPRA3 are required for histamine-mediated itch^39,122.

Fig. 3 |. Neuropeptide coding in peripheral and central itch pathways.

a | Chemical itch pathways. A schematic illustration of the encoding of histaminergic and non-histaminergic itch according to the neuropeptide code model. According to this model, itch induced by histaminergic and non-histaminergic pruritogens is encoded by either gastrin-releasing peptide (GRP) or neuromedin B (NMB), but not both. GRP and NMB work not only in a complementary manner but also in synergy, with GRP acting as a partial agonist for NMB receptor (NMBR)¹⁰⁵. GRP and NMB neurons release GRP and NMB, respectively, to activate GPR receptor (GRPR)-expressing neurons (GRPR neurons) in laminae I and II of the spinal dorsal horn or NMBR neurons in lamina II (REFS^55,56,105. NMBR neurons send predominantly histaminergic and marginally non-histaminergic itch information to GRPR neurons via glutamate^56,105. B-type, or brain, natriuretic peptide (BNP) is co-released with NMB, binds to natriuretic peptide receptor C (NPRC) in the spinal cord and facilitates NMB-mediated histaminergic itch via NPRC-NMBR crosstalk¹⁵⁵. BNP and NMB are co-expressed in some neurons¹⁵⁵. Since NMB and GRP expression partially overlaps⁵⁶, some NMB may be co-released with GRP; however, this NMB will act as a functional antagonist to compete with GRP for binding to GRPR neurons. Conversely, a small amount of GRP is co-released with NMB and binds to NMBR, where it acts as a weak agonist to convey non-histaminergic itch information. b | The mechanical itch pathway as described by the neuropeptide code model. Neurons in the outer layer of lamina III (IIIo) of the dorsal horn that express TAC2 or originate from progenitors expressing the urocortin 3 gene (Ucn3) are proposed to be activated by innocuous tactile information (such as that arising from the touch of a von Frey fibre) via Aβ low-threshold mechanoreceptors (LTMRs) and to relay this information to GRPR neurons via glutamatergic transmission^58,188. The convergence of inputs from these IIIo neurons and from neurons arising from the same lineage but located in the inner layer of lamina II (IIi) may be a prerequisite for converting this innocuous touch into itch⁵⁸. The TAC2 neurons in IIi receive direct inputs from C fibres and/or Aδ fibres but are hypothesized to be silent under naive conditions owing to tonic inhibition by local GABAergic neurons. Under chronic itch conditions, they may be activated to send itch information to GRPR neurons (dashed arrow). The neuropeptides expressed by the GABA/glycine inhibitory neurons may differ and are not shown to avoid confusion. c | The contagious itch pathway according to the neuropeptide code model. In a contagious itch test, the observer watches the scratching of conspecifics displayed on a screen. The retina receives the visual itch information and conveys it to the suprachiasmatic nucleus (SCN)¹⁰⁴. It is hypothesized that GRP neurons in the retinorecipient zone or the core of the SCN release GRP to activate GRPR neurons, which send the information to downstream neural circuits to trigger a scratching behaviour¹⁰⁴. BAM8-22, bovine adrenal medulla 8-22 peptide; DRG, dorsal root ganglion; IIo, outer layer of lamina II.

B-type, or brain, natriuretic peptide (BNP), an inhibitory neuropeptide expressed in sensory neurons and encoded by the gene Nppb, has also been implicated in itch^25,154–156; however, its contribution to itch remains unclear and controversial. Mice with a global KO of Nppb exhibited deficits in histaminergic and non-histaminergic itch¹⁵⁴. However, it is possible that these deficits may be attributable to the loss of BNP expression in skin and immune cells¹⁵⁶ in these mice. The neurotoxin BNP–saporin was initially reported to exclusively ablate neurons expressing the BNP receptor natriuretic peptide receptor A (NPRA), encoded by Npr1 (REF.¹⁵⁴), and to abolish itch responses in mice¹⁵³. However, older evidence suggests that BNP binds to both NRPA and NPRC, encoded by Npr3, with equal affinity¹⁵⁷. This was recently supported by the findings reported in a preprint in which it was reported that both Npr1 and Npr3 spinal neurons were ablated with use of BNP–saporin¹⁵⁵. Most natriuretic peptide receptors that act as guanylyl cyclases (including NRPA and NRPC) inhibit the activity of the neurons in which they are expressed^158,159. Indeed, intrathecal injection of BNP inhibits nociceptive transmission¹⁶⁰, while exogenous application of BNP inhibits the excitability of primary nociceptors^161–163, suggesting that BNP may be a negative modulator of nociceptive transmission under certain inflammatory pain conditions. Furthermore, intrathecal or intradermal injection of BNP at picomolar or nanomolar concentration fails to induce immediate scratching behaviour in mice, indicating that administration of BNP alone at a physiologically relevant concentration does not stimulate pruriceptive transmission^154,155,160. The findings reported in the recent preprint mentioned above suggest that BNP is instead involved in facilitating histaminergic itch that is mediated by NMB through NPRC–NMBR crosstalk¹⁵⁵. This finding is in agreement with the observations that Nppb is markedly upregulated upon histamine release from mast cells in mice with allergic contact dermatitis^25,158,164. By contrast, the preprint suggests that Nppb is downregulated in DRGs of mice with a dry skin itch condition¹⁵⁵ that is devoid of a histaminergic component¹⁶⁵. Consistent with these findings, a study that used a behavioural screening approach to identify anti-Npr1 compounds used an antihistamine effect as a validation tool¹⁵⁸. These preliminary findings indicate that the role of BNP in DRGs may be limited to the facilitation of histamine-mediated itch and that NPRC–NMBR crosstalk may switch the otherwise inhibitory action of BNP to an excitatory one. Although BNP-expressing neurons were proposed to be a peripheral itch-specific neural pathway^154,164, they also express genes important for mediating mechanical pain, such as those encoding PAR2 (REFS^72,166) and neuropeptide Y (NPY) receptor type 2 (NPY2R)^167,168. A receptor may function differently in different tissues. For example, in contrast to DRGs, PAR2 acts as an itch receptor in the skin^169,170 Notably, CKO of Sst, the gene encoding somatostatin (SST), an inhibitory neuropeptide that is largely co-expressed in BNP neurons, in TRPV1-expressing sensory neurons in mice (Trpv1^Cre;Sst^f/f mice) increased mechanical pain but not itch¹⁷¹, providing another example that disputes the idea that the specificity of itch is encoded by a subpopulation of sensory neurons.

Together, these findings suggest that GRP and NMB encode histaminergic itch and non-histaminergic itch, respectively. By contrast, preliminary findings suggest that BNP acts as a neuromodulator that facilitates histaminergic itch transmission. Thus, the neuropeptide code model hypothesizes that distinct modes of neuropeptide action confer the specificity of different types of itch (FIG. 3a).

Spinal Grp neurons in chemical itch.

In recent years, there has been considerable interest in and debate on the possible role of spinal Grp neurons, which express Grp mRNA in itch^{59,83,154,172,173}. It was proposed that spinal Grp neurons exclusively express NPRA and release GRP to activate GRPR neurons in response to BNP release by primary sensory neurons¹⁵⁴. According to this model, BNP transmits itch, irrespective of whether the itch is histaminergic or non-histaminergic¹⁵⁴. From an evolutionary perspective, however, I believe that it is difficult to envisage why the dorsal horn would expend the metabolic costs required to support an additional layer of intralaminar transmission or a detour in the wiring to perform a function that is identical to that of GRPR neurons. In principle, spinal GPCRs — which are predominantly activated by neuropeptides released from sensory neurons — are responsible for mediating relatively slow transmission, such as that involved in the transmission of chemical itch and inflammatory pain, typically conducted by slowly conducting C fibres¹⁷⁴. However, once the information that encodes a warning signal has arrived at and been processed by neural circuits defined by GPCRs in the spinal cord, it is highly desirable to ensure its timely transfer to the brain. Local intralaminar communication between tightly apposed interneurons of laminae I and II, using a neuropeptide rather than fast-acting glutamate, would delay the body’s defensive response. Thus, I argue that neuropeptides synthesized within the dorsal horn are dispensable for modality-specific excitation during normal reflexive responses. Instead, a more sustained slow release of neuropeptides and their slower activation of GPCRs (within seconds to minutes) conducted by C fibres is optimally suited for mediating chemical itch and for tonic modulation of itch^99,134,140.

Several research groups have attempted to examine the possible role of spinal Grp neurons and have obtained contradictory results. They all used bacterial artificial chromosome (BAC)-based Cre and/or enhanced GFP (eGFP) mouse lines produced by the Gene Expression Nervous System Atlas (GENSAT) project¹⁷⁵. One group used the global ablation approach and found that mice lacking Grp^Cre neurons showed reduced itch but increased thermal pain behaviour. In addition, it was shown that stronger activation of these neurons recruits opioid signalling to inhibit pain, while weaker activation induces itch¹⁷². On the assumption that abnormal itch and pain originated from the spinal ablation of Grp^Cre neurons, it was proposed that spinal Grp^Cre neurons constitute a ‘leaky gate’ for itch and pain¹⁷². However, these mice also exhibited an ectopic ablation of lamina II nociceptive neurons (including nearly half of the population of neurons expressing protein kinase Cγ (PKCγ) neurons, which do not overlap with spinal Grp neurons⁵³). Therefore, the most parsimonious explanation for these findings, in my view, is that the reduced itch and enhanced thermal pain deficits stem from the brain, considering that there are numerous Grp neurons in the brain regions potentially involved in itch and pain transmission and modulation^53,61,173. For example, the decreased itch behaviour can be explained by the ablation of Grp neurons in the parabrachial nuclei implicated in itch transmission⁶¹. By contrast, a subsequent study that used biting behaviour as a behavioural surrogate of itch observed a rather modest reduction in the biting behaviour induced by chloroquine, histamine and serotonin (also known as 5-hydroxytryptamine (5-HT)), but not SLIGRL, nor noxious stimuli, after ablation of Grp^Cre neurons at the lumbar level, and concluded that spinal Grp neurons are required for itch but not pain transmission¹⁷³. It is now known that these BAC-based Cre and/or eGFP lines not only fail to express Cre and/or eGFP in sensory neurons but also capture no more than a quarter of Grp neurons in the spinal cord^{83,154,172,176}. Thus, the existing BAC Grp lines preclude functional analysis of a vast majority of spinal Grp neurons.

To circumvent this problem, a knock-in Grp^iCre mouse line has been generated that recapitulates the endogenous Grp expression in both sensory neurons and the spinal cord⁵³. An intersectional genetic approach that involves the use of two genetic lines expressing different recombinases (one of which is expressed in a region-specific manner¹⁷⁷) was used to ablate all spinal Grp neurons of Grp^iCre mice so that intradermal injection in the nape — a gold standard approach for investigating itch behaviour in mice — could be used. No pain and itch deficits were found after ablation of spinal Grp neurons⁵³. Molecular analysis, however, revealed that spinal Grp neurons express a myriad of itch-related and pain-related genes, such as those encoding NMBR, NPY receptor type 1 (NPY1R), NPR3 and NK3R^53,178. Interestingly, the transgenic spinal Grp neurons that are directly innervated by MRGPRA3 fibres are potently activated in the BAC Cre and/or eGFP lines — rather than inhibited, as would be the case if they were itch-specific neurons⁵⁷ — by nociceptive stimuli¹⁷² but not by chloroquine¹⁷⁹, in line with the suggestion that MRGPRA3 fibres also convey nociceptive information⁴⁰. Furthermore, they can be inhibited by a μ-opioid receptor (MOR) agonist¹⁷⁶ or endogenous opioids¹⁷². Collectively, these findings strongly indicate that the BAC Grp^Cre transgenic lines might have captured a subpopulation of spinal nociceptive neurons expressing MOR that can be directly activated by TRPV1/MRGPRA3 nociceptive fibres^40,109,172.

These controversies raise the important question of how to define an operational microcircuit that is physiologically meaningful for chemical itch. It may be difficult to believe that the Grp mRNA found in the dorsal horn has no role in itch. However, it has been proposed that many neuropeptides present in the dorsal horn may be vestigial remnants of ontogenetic development or past evolution¹⁸⁰. It is not surprising that none of the neuropeptides contained by dorsal horn neurons has been convincingly demonstrated to be required for acute somatosensory transmission or inhibition. Furthermore, numerous genes share overlapping expression with GRPR or NMBR⁹⁵, suggesting that the ablation of neurons expressing these genes may result in partial deficits in itch and pain. For instance, neuronal nitric oxide synthase (nNOS) is expressed not only in GABAergic inhibitory neurons implicated in pain inhibition¹⁸¹ but also in some GRPR neurons^67,181,182. Conceivably, therefore, activation of nNOS neurons in a BAC Cre line might result in more itch and fewer pain behaviours⁶⁷. In this case, if the expression of nNOS in GRPR neurons were unknown, these results might be interpreted as arguing against the intensity or differential coding theory.

In spite of the invaluable role of GENSAT’s Cre lines in advancing neuroscience, their limitations have become increasingly evident (TABLE 1). In the spinal cord, most BAC-based Cre lines probably label a subset of neurons with a mixture of sensory modalities and often lack fidelity. Several well-known factors, such as partial gene regulatory elements contained in the BACs, irregular genomic integration sites and variable copy number of transgenes, have similarly hampered the functional classification of other types of sensory neuron (such as retina ganglion cells¹⁸³). Perhaps linked to these difficulties and despite their fundamental roles as sensors and their therapeutic value, spinal GPCRs and their circuits remain understudied. This negligence similarly exists for circuit analysis in the brain¹³⁴. For instance, functional neural circuits in the extended amygdala and the SCN are typically classified according to neuropeptide expression rather than GPCR expression^102,184. That GPCRs have largely escaped attention can be attributed to technical difficulties in detecting these receptors in the CNS owing to their very low expression levels⁹⁸, as well as a paucity of specific antibodies. Maintaining the basal level expression of GPCRs, however, could be a mechanism whereby an accidental switch of the itch and pain circuits from a quiescent state to a pathological one is prevented. A real challenge, therefore, will be to identify a few truly modality-specific Cre lines defined by GPCRs.

Mechanical and contagious itch

Mechanical and contagious itch are idiosyncratic sensations (occurring only under special circumstances), with certain traits possibly unique to humans¹⁸⁵. For example, only a few facial regions are sensitive to mechanical itch¹⁵, whereas certain populations are prone to contagious itch^17,185. Mechanical itch is modulated by PIEZO2, an ion channel present in Merkel cells in the skin, which are involved in discriminating light touch¹⁸⁶. Contagious itch is a type of automatic motor mimicry, a primitive and pervasive phenomenon present in most vertebrates that enables social animals to act in synchrony to promote survival and social bonding^185,187. There are a number of differences and similarities between the coding of mechanical, chemical and contagious itch.

Mechanical itch.

In the mouse spinal cord, neurons of the Ucn3 lineage (that is, neurons derived from progenitors transiently expressing Ucn3 during development) and Tac2-expressing neurons (labelled with use of a Cre reporter) have been independently demonstrated to constitute the first relay station for Aβ-LTMR inputs that carry information about mechanical itch but not about pain or chemical itch^58,188. Of note, while lineage neurons — tracked through permanent genetic reporters such as tdTomato fluorescent protein — are commonly used as a surrogate for circuit analysis, they may not necessarily represent a true functional circuit in their entirety. These two excitatory interneuron populations, which are found in the outer layer of lamina III of the dorsal horn, are likely to overlap since they share several remarkable features^58,188. It was recently shown that Tac2-expressing neurons convey light touch information to GRPR neurons (FIG. 3), and that ablation of either Tac2-expressing neurons or GRPR neurons abolishes mechanical itch⁵⁸, indicating that mechanical itch transmission requires GRPR neurons. Several studies, however, indicated that mechanical itch is transmitted through NPY1R neurons (a group of excitatory interneurons in laminae I and II that were previously implicated in chronic pain^189,190 and were recently implicated in chemical itch^191,192) and that this transmission is not affected by the ablation of GRPR neurons using BB-sap^188,193,194. However, there was still a certain amount of chloroquine-induced scratching behaviours (about 50 scratches) in mice after BB-sap treatment. I therefore hypothesize that there was an incomplete ablation of GRPR neurons in these studies and that the remaining GRPR neurons are sufficient for mediating mechanical itch behaviour^188,193,194. Although one study reported that NPY1R and GRPR are not colocalized in the spinal cord, the specificity of the GRPR antibody used remains unclear¹⁹⁴. In contrast, other studies have reported substantial overlapping expression between Npy1r and Grpr in the dorsal horn^58,195. It can, therefore, be concluded that NPY1R neurons are a mixture of excitatory interneurons (including GRPR neurons) involved in both itch and pain transmission. Furthermore, this emerging evidence supports the notion that GRPR neurons convert innocuous light touch into the sensation of itch.

Is the coding of mechanical itch exclusively mediated by fast-conducting myelinated Aβ-LTMR inputs and/or through glutamatergic transmission or does neuropeptide coding play a role in this type of itch? To answer this question, it is worth considering that not all scratches evoked by innocuous light touch reflect mechanical itch. Rodents are well equipped with a repertoire of stereotyped reflexive behaviours mediated via Aβ-LMTR fibres³² — such as avoidance, body shivering, wiping and scratching — that allow them to distance themselves from or remove both non-itchy stimuli and irritants. This begs the question of how we may reliably distinguish between scratching behaviours evoked by non-itchy touch versus those evoked by itchy touch, an important but overlooked confounding factor that may lead to some controversial interpretations of results. It has long been appreciated that a certain degree of skin sensitization (manifesting itself in tactile insensitivity and tonic activation of C tactile fibres) is necessary for converting touch into itch^1,4,5,7. Consistent with this, mechanical itch can be elicited by application of von Frey hairs to the skin surrounding an area sensitized by histamine or bordering an itchy region^186,196, or to nape skin that has been sensitized by shaving in naive mice^{58,186,193,194}. This suggests that unmyelinated or histamine-sensitive C fibres are involved in mechanical itch^5,15,197, in a manner reminiscent of their role in mechanical allodynia^197,198, a painful mechanical hypersensitivity to innocuous touch mediated by synergistic inputs from Aβ and C fibres¹⁹⁹. Researchers may easily mistake non-itchy light touch-evoked scratching behaviour, which is mediated, independently of GRPR neurons, exclusively through Aβ-LTMR input⁵⁸, for mechanical itch. Neuropeptide signalling could be an integral part of the code that distinguishes non-itchy touch from itchy-touch input^5,7,15 (FIG. 3b; Supplementary Table 1).

Contagious itch.

Contagious itch, first reported in humans and non-human primates^17,187,200, was serendipitously discovered in mice in 2017, when it was found that wild-type mice mistakenly housed in the same cage as Braf^V600E mice (a genetic model of neuropathic itch²⁶) exhibited robust spontaneous scratching behaviour¹⁰⁴. Modelling contagious itch in naive mice is, however, challenging owing to the poor visual acuity of mice, the presence of spontaneous scratching, the fact that the visual itch stimulus is subtle and the randomness of the ‘looking’ behaviour of a freely moving observer towards a scratching conspecific. These technical nuances may explain why some researchers failed to observe contagious itch behaviour in mice using a paradigm that involved a weak pruritogen, such as histamine^201,202. In a study in which contagious itch was induced by exposure of an observer mouse to a visual recording of a demonstrator mouse scratching, it was shown that the probability of the observer mouse noticing a scratching demonstrator is maximized when the speed of scratching motion displayed on the computer monitor is 1 Hz: that is, one scratching bout (or four scratching strokes) per second^104,202. The discovery of contagious itch in mice suggests that recognizing and copying a scratching motion has an adaptive value and probably evolved millions of years ago. A recent study meticulously replicated the contagious itch protocol and discovered that a mouse model of autism spectrum disorders failed to imitate scratching behaviour without impacting spontaneous scratching behaviour²⁰³. The study is important, in light of a degree of scepticism about the existence of contagious itch in mice. Socially contagious pain and empathic behaviours, however, have been known to exist between familiar rodents for decades^204,205. Since contagious itch in mice does not require familiarity or prior training, it could be a primitive form of emotional contagion and could have arisen earlier than pain contagion in evolution²⁰⁶.

The development of a contagious itch test in mice has enabled the elucidation of the mechanisms involved in coding contagious itch. Mapping expression of FOS after contagious itch had been induced revealed that the most robustly active area is the SCN, where GRP is also expressed¹⁰⁴. GRP neurons in the ventrolateral ‘core’ of the SCN project to GRPR neurons in the dorsomedial ‘shell’ of the SCN¹⁰⁴. Importantly, the scratching behaviour evoked by observing the scratching motion of a conspecific was abolished through SCN-specific deletion of GRPR¹⁰⁴. Analogous to the reduced spinal GRP immunohistochemical staining observed after chloroquine-evoked scratching⁵³, GRP immunostaining was reduced in the SCN after the contagious itch test, suggesting a release of GRP and the activation of GRPR¹⁰⁴. These findings were surprising because the SCN receives direct inputs from the retina through intrinsic photosensitive retina ganglion cells (ipRGCs), which were thought to be involved only in non-image-forming responses²⁰⁷. Nevertheless, this demonstrated that SCN GRP encodes contagious itch information and suggested that the retina–ipRGC–SCN pathway may be an evolutionarily ancient visual pathway that has been repurposed to encode itch (FIG. 3c). Duplicating and repurposing older circuits and molecules for a novel function is a fundamental mechanism underlying the expansion of the complexity of organisms throughout evolution²⁰⁸.

It is clear that, despite profound differences in the properties of the stimuli (visual versus physical) and the principal neural circuits involved (the SCN versus the spinal cord), the coding of contagious, chemical and mechanical itch is remarkably similar. Their slow kinetics, the skin sensitization required for mechanical itch and the dependence on GRPR neurons all support the importance of the neuropeptide code model in these types of itch.

Modulation of itch

Itch evokes the urge to scratch. While scratching inhibits the initial cause of the itch, it may (by driving the release of inflammatory mediators) provoke a further urge to scratch¹²⁸. The neuropeptide code model posits that this rhythmic itch–scratch cycle is driven by the sequential release of itch neuropeptides and fast-acting neurotransmitters. As described earlier herein, the model suggests that, upon exposure to a pruritogenic stimulus, itch neuropeptides and glutamate activate GRPR neurons, directly or indirectly. It further proposes that subsequent scratching behaviour drives the release of glutamate from primary afferents, which activates spinal inhibitory neurons, which, in turn, release GABA and glycine to inhibit GRPR neurons. The inhibition of GRPR neurons must be both rapid and transient so that the brain, while being warned, can continue monitoring the environment and skin condition from moment to moment. How this extraordinarily intricate balance between promoting and inhibiting itch transmission is seamlessly executed in a state-specific and spatio-temporal-specific manner is a fascinating question for scientific enquiry.

Inhibitory neural circuits.

Approximately 25–30% of neurons of dorsal horn laminae I–III are functionally heterogeneous GABAergic and/or glycinergic interneurons¹⁸². These neurons are broadly classified according to their expression of neuropeptides such as galanin (GAL), dynorphin (DYN; encoded by Pdyn), NPY and parvalbumin (PV). DYN neurons are additionally composed of some excitatory neurons implicated in chronic pain²⁰⁹. An overriding question in the field is whether these spinal inhibitory neurons gate itch and pain separately. Since most transgenic Cre mouse lines used for spinal circuit research also express Cre in the brain, it is necessary to use an intersectional genetic strategy to achieve spinally restricted ablation of particular cell populations. Alternatively, virus-mediated manipulation of neural circuits should be performed within the cervical spinal cord (rather than the lumbar cord) to allow scratching behaviour evoked at the nape to be used to assess itch using BB-sap, as scratching behaviour may also reflect pain or non-itchy touch as discussed earlier^67,210. Intersectional ablation of spinal Pdyn^Cre lineage neurons in mice revealed that these neurons gate mechanical pain but not chemical itch^211,212. In contrast, chemogenetic inhibition and caspase-based ablation of Gal^Cre lineage neurons that also express DYN²¹³ in the cervical cord increase chemical itch but not spontaneous itch or pain behaviour⁶⁷. These seemingly conflicting results suggest that the Pdyn^Cre mouse line does not faithfully capture all adult spinal DYN inhibitory neurons, complicating the interpretation of results derived from the use of this line. Together, these studies indicate that DYN neurons are functionally heterogeneous and that subpopulations of these neurons exert anti-itch (Gal^Cre inhibitory neurons), anti-nociceptive (DYN inhibitory neurons) and pro-nociceptive (DYN excitatory neurons) functions^214,215. More recently, a group of Cre lineage neurons that transiently express Ptf1a (which encodes a transcription factor that specifies the fate of GABAergic neurons in the dorsal horn^216,217) were shown to partially gate spontaneous itch through the inhibition of GRPR neurons; however, they did not inhibit chemical itch²⁸. Since only 40% of Ptf1a^Cre lineage neurons were ablated in that study²⁸, it is unclear whether Ptf1a^Cre lineage neurons may additionally gate spontaneous pain. Nonetheless, these studies suggest that chemical and spontaneous itch may be differentially gated and that there may be distinct types of GRPR neurons that mediate chemical and spontaneous itch, respectively^28,57,67.

Gal^Cre lineage neurons have been found to inhibit less than one-third of GRPR neurons via GABA-mediated monosynaptic connections⁶⁷. This is less than the proportion of GRPR neurons (~60%) that are inhibited by GABAergic and/or glycinergic inputs in response to primary afferent stimulation⁵⁷ but similar to the proportion of GRPR neurons that are inhibited by cooling and chemical pain⁵⁷. Interestingly, PV^Cre lineage neurons in the inner layer of lamina II and in lamina III of the dorsal horn, which gate certain types of mechanical pain transmission²¹⁸, do not gate chemical itch and rarely synapse with GRPR neurons⁶⁷. This implies that the synaptic connectivity between local inhibitory neurons dedicated to gating nociceptive transmission and those gating GRPR neurons is distinct, stringent and hard-wired. By paired recordings, it was shown that there is a low connection probability between inhibitory neurons and other neurons in lamina II of the dorsal horn (11%)¹¹⁸. Importantly, monosynaptic connections between excitatory neurons within this region are also rather sparse¹¹⁹. Thus, synaptic connections between excitatory neurons of laminae I and II of different sensory modalities could also be stringent. This suggests that rapid transmission and convergence of submodality inputs, such as that occurring at the synaptic connections between NMBR neurons and GRPR neurons that convey histaminergic itch, are hard-wired. Such a strategy could allow efficient inhibition of itch through the exclusive inhibition of GRPR neurons, a convergent node for transmitting discrete types of itch. Together with the observation that chemical pain and cooling inhibit rather than excite GRPR neurons⁵⁷, the evidence of low connection probabilities within the spinal cord is in favour of the notion that the excitatory neural pathways that transmit pain and cooling do not undergo crosstalk with GRPR neurons directly in the spinal cord. Correspondingly, inhibitory modulation of itch and pain is modality specific and hard-wired.

Npy^Cre lineage neurons, widespread in the dorsal horn, receive direct monosynaptic inputs from C fibres and Aβ fibres and can be activated by mechanical, thermal and chemical stimuli^193,219. They were initially proposed to gate mechanical itch but not chemical itch on the basis of observations that mice with ablation of Npy^Cre lineage neurons developed spontaneous scratching behaviour and skin lesions and were sensitive to innocuous touch, which characteristics have been interpreted as a manifestation of mechanical itch^193,194. However, given that mechanical itch (like chemical itch) can be gated at the level of GRPR neurons (and possibly by inner layer of lamina II TAC2 and/or Ucn3 lineage neurons^58,188) (FIG. 3), it seems unlikely that the scratching evoked by chemical itch versus mechanical itch would activate two distinct populations of spinal inhibitory neurons (Gal^Cre lineage neurons versus Npy^Cre lineage neurons) if the intention of mechanical itch-induced scratching is also to inhibit itch. Instead, it is possible that scratching evoked by mechanical itch may mainly help an animal to remove irritating light touch on the skin, but not to inhibit the sensation of itch, thus resembling the scratching evoked by non-itchy touch (see earlier); if so, it would not be necessary to entail the major spinal inhibitory circuits to inhibit this type of mechanical touch unless such irritating light touch becomes persistent, possibly to the degree that might lead to an allergic skin response. In the latter scenario, the spontaneous scratching/biting behaviour that resulted from ablation of Npy^Cre lineage neurons^193,194 may be an indicator of spontaneous or neuropathic pain due to a loss of tonic gating, akin to that resulting from the spinal perturbation of NPY2R signalling²¹⁰. Moreover, the increased scratching behaviour evoked by von Frey fibre stimulation applied to the nape of mice lacking NPY neurons may be an indicator of mechanical hypersensitivity due to ongoing neuropathic pain. These rodents showed normal basal pain responses^189,193,194, supporting the notion that acute and fast withdrawal nocifensive behaviour does not require spinal inhibitory peptidergic gating⁵⁸. Consistently, there is ample evidence demonstrating the primary role of spinal NPY signalling in tonic gating of chronic and spontaneous pain via NPY1R and/or NPY2R^{189,190,210,220}. Thus, some of the spontaneous scratching, licking or biting behaviours and skin lesions that arise following the ablation of various populations of inhibitory neurons, which are typically considered as manifestations of itch rather than pain, could actually be due to spontaneous pain²¹⁰. Moreover, because inhibitory neuropeptides are not involved in fast inhibition of itch, the observations that exogenous NPY or NPY1R agonists inhibit chemical itch are likely to be pharmacological artefacts^191,192 (TABLE 1) that are attributable to NPY1R expression in GRPR neurons⁵⁸. A pharmacological artefact refers to the behavioural or neuronal response elicited by exogenously administered substances that otherwise would not occur under natural conditions (TABLE 1). For example, intrathecal injection of SST could induce itch-related scratching behaviour¹⁵⁵. However, neither spinal cord-specific (Lbx1^Cre;Sst^f/f mice) nor DRG-specific (Trpv1^Cre;Sst^f/f mice) deletion of Sst alters itch transmission¹⁷¹, indicating that endogenous SST in these two areas is not involved in itch modulation. Although GRPR neurons are subject to both local tonic and phasic GABAergic inhibition²²¹, tonic inhibition of these neurons appears to be the principal mechanism for gating the flow of itch information.

In summary, while spontaneous pain has been mostly ignored owing to overemphasis on itch, recent studies on functional characterization of various populations of spinal inhibitory neurons lend considerable support to anatomically segregated inhibitory circuits for itch and pain, respectively. Although the components of these inhibitory circuits are usually identified through their neuropeptide or transcription factor expression, they invariably use fast-acting GABA and/or glycine, rather than neuropeptides, to rapidly inhibit acute itch transmission^{57,58,67,74,188,222}. Given that some of the inhibitory circuits examined to date may overlap, further delineation of inhibitory microcircuits defined by physiologically relevant ligand–receptor interactions will be required to understand their activation mechanisms.

The dichotomy of glutamate: excitation and inhibition.

The role of glutamate in itch is also controversial and confusing because of conflicting claims in large part resulting from methodological differences and limitations^{57,60,87,133,223,224}. Two comprehensive genetic studies showed that mice lacking vesicular glutamate transporter 2 (VGLUT2), which regulates the probability of glutamate release from the central terminals²²⁵ in Nav1.8^Cre lineage and TRPV1 DRG neurons (Vglut2-CKO mice; Nav1.8 is also known as Scn10a and Vglut2 is also known as Slc17a6), exhibited enhanced itch but reduced pain behaviour^223,224. These studies cast doubt on the role of glutamate as a co-transmitter of itch in primary C fibres; however, it remains possible that glutamate released by a subset of C fibres in Vglut2-CKO mice may have been sufficient to transmit itch. Indeed, Vglut2-CKO mice exhibited only a 50% reduction of fast glutamatergic transmission between primary afferents and the spinal cord, as well as normal or partial attenuation of certain types of acute thermal and mechanical pain^223,224.

Within the spinal cord, however, it is clear that glutamate plays an essential role in local intralaminar itch transmission. It relays chemical itch from NMBR neurons⁵⁶ and mechanical itch from TAC2 neurons⁵⁸ to GRPR neurons, and relays chemical itch and mechanical itch from GRPR neurons to projection neurons^56,60,61. The finding that co-application of glutamate receptor, GRPR and NK1R antagonists into the spinal cord dramatically attenuates histaminergic and non-histaminergic itch may result from the combinatory effect of these antagonists acting at discrete sites within the itch transmission pathways¹¹⁴.

At first sight, the impaired pain behaviour and enhanced itch observed in Vglut2-CKO mice^223,224 appear to indicate a loss of the inhibition of itch mediated by pain. However, the deficits in itch and pain are actually fully dissociable; when pruritogenic responses were tested, the nociceptive transmission pathway was quiescent in both transgenic mice and control mice because of the endogenous tonic gating even without peripheral noxious input (as shown previously). Conversely, the pain deficits observed in Vglut2-CKO mice were shown to result from glutamate deficiency and to be independent of itch transmission^57,65. Given this evidence of distinct pain and itch circuits in the spinal cord, how could the loss of glutamate in C fibres of Vglut2-CKO mice give rise to augmented itch?

A long-held belief is that scratching activates mild noxious mechanical pain to inhibit itch^2,10,73. However, scratching — or rubbing or stroking in animals that cannot scratch — commonly induces a pleasant and hedonic sensation^4,10,226,227, which is likely to be relayed through some interneurons of laminae I and II innervated by C LTMRs (Supplementary Fig. 1) and neurons of laminae III and IV targeted by Aβ LTMRs to the brain via spinal projection neurons^32,228,229. Under normal circumstances, if scratching induces mild pain to inhibit itch, this would imply that pain neurotransmitters can be easily released and that spinal Gal^Cre lineage neurons are activated indirectly by a spinal nociceptive circuit instead of directly by primary afferents, which is inconsistent with the finding that GAL neurons also receive monosynaptic inputs from both C fibres and Aβ fibres¹⁸². Further arguing against the view is that scratching an itch evokes pleasant and relieves aversive sensation; in accordance with this, activation of C LTMRs inhibits, rather than activates, nociceptive C fibres^32,118. Accordingly, the enhanced scratching behaviour of Vglut2-CKO mice might be explained by the disinhibition of GRPR neurons due to a loss of the feedforward inhibition that is mediated by the spinal inhibitory neurons (which in turn results from a lack of glutamate release from C fibres). In this context, it can be hypothesized that glutamate may have a much less appreciated role in the inhibition of itch transmission. Scratching may release glutamate to instantaneously activate Gal^Cre neurons. Because diminished glutamate release from C fibres could weaken the activation of these spinal neurons, Vglut2-CKO mice might have to compensate by scratching harder. Furthermore, GRP expression or release might be enhanced to compensate for a deficiency of glutamate that otherwise is tonically co-released with GRP, leading to increased acute itch or the development of chronic itch. Thus, glutamate released from C fibres could be essential for facilitating GRP-GRPR signalling.

In short, disentangling peripheral and central neural pathways involved in the itch–scratch cycle is important for understanding the neuropeptide code for itch in context. Scratching an itch can simultaneously activate multiple spinal excitatory and inhibitory pathways in parallel. The genetic studies described above have uncovered an unexpected functional dichotomy in the action of glutamate released by C fibres. First, glutamate from primary afferents excites spinal itch neurons as a co-transmitter of itch neuropeptides. Subsequently, it inhibits itch through a different subset of primary afferents that activate spinal inhibitory neurons. A rapid alternation of glutamate-mediated excitation and inhibition of itch manifests itself in a temporally distinct site-dependent and state-dependent manner (Supplementary Fig. 1). Because of its opposing role in itch, glutamate release into the spinal cord must be made through discrete types of primary afferents shielded from each other at their central synaptic targets to safeguard modality-specific excitation and inhibition (Supplementary Fig. 1).

Descending modulation.

In addition to spinal inhibitory circuits, spinal itch transmission is under descending modulation via several parallel pathways originating in the periaqueductal grey and travelling to the spinal cord indirectly via the rostral ventromedial medulla (RVM) pathway^{67,105,230–232}. Serotonin (5-HT) derived from the dorsal raphe nucleus of the RVM has opposing functions in itch and pain modulation: it inhibits nociceptive neurons expressing 5-HT receptor 1A (5-HT_1A), an inhibitory GPCR, in lamina II of the spinal dorsal horn, while concomitantly facilitating itch via crosstalk between 5-HT_1A and GRPR, which form a heteromeric complex²³³ (FIG. 4). Although the effect of 5-HT itself on GRPR neurons is inhibitory because 5-HT_1A is an inhibitory GPCR under experimental conditions²³³, it becomes facilitatory when it is released in conjunction with GRP^230,233. TAC1 glutamatergic neurons in the periaqueductal grey, which synapse directly onto 5-HT neurons in the RVM, modulate the function of GRPR neurons indirectly via several descending pathways, including those involving 5-HT and GABA⁶⁷. Recently, a population of non-serotonergic GABAergic neurons that co-express G-protein-coupled oestrogen receptor (GPER) and MOR in the RVM was shown to inhibit spinal itch via regulation of opioid signalling by GPER²³².

Fig. 4 |. GRPR neural and signalling pathways.

a | The neural pathways mediating morphine-induced itch and analgesia. The rectangles represent the itch-specific and pain-specific neural pathways. The coloured circles represent μ-opioid receptor 1 (MOR1)-expressing neurons (MOR1 neurons; red) and neurons expressing the MOR1 isoform MOR1D and gastrin-releasing peptide (GRP) receptor (GRPR) (blue). According to the traditional view, there are more nociceptors (expressing MOR1) than pruriceptors (expressing MOR1D and GRPR) in the spinal cord. Therefore, pain transmission masks or occludes itch^2,22, and morphine inhibits pain by activating MOR1, thereby unmasking itch transmission. The contemporary view, however, states that two non-overlapping populations of excitatory interneurons in the spinal cord mediate morphine-induced itch and analgesia, respectively. MOR1 neurons convey nociceptive information, whereas MOR1D and GRPR neurons transmit itch. Morphine simultaneously activates MOR1 neurons to induce analgesia and MOR1D/GRPR neurons to induce itch⁸. b | Diverse GRPR signalling pathways. GRPR can be activated directly by GRP released from primary afferents to open the endoplasmic reticulum (ER) Ca²⁺ store and induce itch signalling^8,57,144,233. In addition, it can be indirectly activated by opioids through a unidirectional MOR1D–GRPR cross-signalling in a GRP-independent manner. GRPR can also be inhibited directly by neuropeptide Y (NPY) or by NPY receptor type 1 (NPY1R) agonists^58,191 and indirectly by κ-opioid receptor (KOR) agonists through KOR-activated protein kinase Cδ (PKCδ)-mediated phosphorylation (P) of GRPR¹⁴⁴. Lastly, GRPR is subject to descending modulation via 5-hydroxytryptamine (5-HT) receptor 1A (5-HT_1A)-GRPR crosstalk upon concurrent activation by their respective agonists, GRP and 5-HT. 5-HT_1A-GRPR crosstalk converts the inhibitory 5-HT_1A-mediated cAMP pathway into an excitatory pathway, thereby facilitating GRPR signalling²³³. These GPCR cross-signalling events may occur in distinct microdomains within the same GRPR neurons or in different subpopulations of GRPR neurons. Ins(1,4,5)P₃, inositol 1,4,5-trisphosphate.

In addition to the descending input from the RVM and the periaqueductal grey, the release of noradrenaline from the locus coeruleus into the spinal cord also exerts a potent inhibitory effect on spinal itch transmission^234,235. Spinal dorsal horn-projecting noradrenergic neurons inhibit acute and chronic itch, and noradrenaline may facilitate inhibitory input into GRPR neurons via the α_1A-adrenergic receptor²³⁵. Therefore, multiple descending systems may have a powerful role in the modulation of spinal GRPR neurons via long-range neurotransmitter-mediated signalling.

Opposing opioid signalling.

Besides endogenous modulation of itch, exogenous drugs and chemicals can also modulate itch transmission. One example often cited in support of the notion that pain and itch are inherently related is opioid-induced pruritus, one of the most prevalent side effects of epidural analgesia⁹. Research on the molecular and neural mechanisms by which opioids induce itch in the spinal cord of mice has further demonstrated that spinal itch and pain circuits are distinct. Most strikingly, morphine-induced itch is completely lost in Grpr-KO mice⁸, indicating an indispensable role of GRPR in opioid-induced itch. How could morphine activate GRPR in the spinal cord?

Pioneering work identified dozens of MOR spliced isoforms, all encoded by the gene Oprm1, in rats, mice and humans²³⁶. Of these, MOR1D — a spliced isoform and an inhibitory GPCR first identified in the mouse and rat spinal cord²³⁷ — was later identified in GRPR neurons as an important receptor for mediating morphine-induced itch (but not analgesia) in mice⁸. To examine MOR1D in situ, a rabbit antibody was generated with use of the same MOR1D amino acid sequence that had been used to generate the original rat antibody²³⁷, and it was found that MOR1D is expressed in some mouse GRPR neurons⁸. This replication of the original findings made in rats and mice²³⁷ should help dispel the doubt regarding the presence of MOR1D in mouse spinal cord, which, akin to Grp in DRGs, could be difficult to detect if the technique lacks sensitivity. A spinal isoform-specific knockdown indicated that a MOR1D isoform is required for morphine-induced itch but not analgesia⁸. Cross-activation of GRPR by MOR1D (but not MOR1) through heterodimerization involving seven amino acids of the MOR1D carboxy terminus was demonstrated in both the mouse dorsal horn and HEK 293 cells expressing these receptors, showing that GRPR can be internalized by morphine only in the presence of MOR1D but not MOR1 (REF.⁸). By contrast, MOR1, which is not expressed in GRPR neurons in lamina II, was shown to mediate morphine analgesia but not itch⁸. This study thus revealed a cell type-specific double dissociation between opioid-induced analgesia and itch (FIG. 4a). Importantly, MOR1Y, an orthologue of MOR1D, was recently isolated from human spinal cords and shown to similarly cross-activate GRPR²³⁸, indicating a conserved mechanism between rodents and humans. Thus, the transmission of itch and pain can be simultaneously and independently modulated by opioids, further demonstrating the dissociation between itch and pain circuits in the spinal cord.

In contrast to the findings that MOR1D/MOR1Y–GRPR crosstalk is important for mediating opioid-induced itch, recent studies have shown that deletion of Oprm1 using Oprm1^f/f mice in Vgat^Cre (Vgat is also known as Scl32a1) or Pdyn^Cre neurons in the CNS abolishes opioid-induced itch^192,239, indicating the involvement of disinhibition in this type of itch. Although the deficit is ascribed to the spinal Oprm1 activated by morphine, the deletion is not limited to the spinal cord. Importantly, this disinhibition mechanism, which occurs at the circuit level, fails to explain observations made at the molecular level, including a loss of morphine-induced itch in Grpr-KO mice, after spinal knockdown of GRPR or its downstream Ca²⁺ signalling molecules such as phospholipase Cβ (PLCβ), and disruption of GRPR-MOR1D interactions⁸. It can be predicted that a disruption of any component in the canonical G_q-protein-mediated Ca²⁺ signalling pathway would attenuate opioid-induced itch. Several possibilities may explain these discrepancies. First, Vgat^Cre or Pdyn^Cre could be transiently expressed in developing GRPR neurons. It was shown that some Grpr neurons express Vgat¹⁷⁸, while nNOS neurons partially express Pdyn, Oprm1 (REF.²³⁹) and GRPR⁶⁷. Conceivably, CKO of Oprm1 in these mice could have deleted MOR1D in Grpr neurons. Second, Oprm1 is broadly expressed in GABAergic neurons in discrete brain regions, and its deletion in these neurons is known to abolish morphine analgesia^240,241. By analogy, the brain Oprm1 in GABAergic neurons could be involved in opioid-induced itch. For instance, a loss of Oprm1 in RVM GABAergic neurons might attenuate descending inhibition of itch²³². However, given the crucial role of spinal GRPR, neither the spinal disinhibition mechanism nor the brain disinhibition mechanism is likely to contribute to spinal opioid-induced itch. Nonetheless, evidence suggests that the transmission of itch and pain can be simultaneously and independently modulated by opioids, further demonstrating the dissociation between itch and pain circuits in the spinal cord.

Unlike MOR agonists, κ-opioid receptor (KOR) agonists such as nalfurafine have strong anti-itch properties and have been used in clinical trials to treat uraemic pruritus²⁴². Exogenous KOR agonists initiate a non-canonical PKC signalling pathway within GRPR neurons, leading to inhibition of itch via PKC-mediated phosphorylation of GRPR for up to 3 days¹⁴⁴. Such a potent and long-lasting inhibitory effect indicates that the type of inhibitory signalling induced by exogenous KOR agonists (like that induced by DYN or by NPY1R agonists) is unlikely to occur under natural conditions, such as during scratching, as it would impede the warning function of itch. Indeed, the onset of inhibition of the spinal neuronal firing activity induced by scratching is rapid and transient, and ends instantaneously with the cessation of scratching^74,75, which is typical of fast synaptic inhibition of GRPR neurons mediated by GABA and glycine (on the order of milliseconds)^{57,67,74,188,221,243,244}. Consistent with this, mice lacking inhibitory neuropeptides such as DYN and GAL showed normal acute itch behaviours^144,245, whereas spinal transplantation of the precursors of cortical GABAergic inhibitory neurons could attenuate neuropathic itch behaviours in mice lacking Bhlhb5 (also known as Bhlhe22) (REFS^27,243).

Summary.

GRPR has emerged as an extremely versatile receptor that interacts with diverse GPCRs to modulate itch transmission under several well-known physiological, pharmacological, and pathological conditions (FIG. 4b). The presence of perhaps dozens of distinct GPCRs on GRPR neurons makes them amenable to pharmacological perturbation of itch signalling independently of GRPR, which conceivably could lead to a plethora of behavioural outcomes.

Concluding remarks

While itch has rapidly emerged as a popular topic that attracts researchers from several different disciplines, the field is — perhaps unavoidably — replete with competing claims and conflicting results. Here, I have attempted to show that many of these issues can be reconciled and made sense of through the consideration of alternative hypotheses. This discussion has underscored the importance of experiments that involve functional perturbation of neural circuits/molecules at the cervical spinal cord level in adult animals as well as the necessity of rigorous validation of evoked or spontaneous scratching behaviours, whose complexity is inadequately appreciated. Despite the challenges, powerful mouse genetics, ethologically relevant animal models and unprecedented genetic and pharmacological access to pivotal itch circuits have allowed the basic architecture of the spinal itch circuitry and the organizing principles that govern spinal itch transmission to be uncovered. Under the disguise of redundancy and the intermingling of numerous molecularly diverse interneurons of varied modalities lies a constellation of intricate networks, which are wired through highly efficient and unidirectional transmission programmes. These microcircuits — shaped over hundreds of millions of years of evolution — are endowed with precision and specificity and can be defined functionally according to the ligand–receptor interactions through which they operate under different physiological conditions.

The emerging properties of these microcircuits and the underlying coding logic of itch described here provide a coherent framework whereby neuropeptides and classic neurotransmitters play differing yet concerted roles in a highly spatio-temporal-specific pattern during the itch–scratch cycle. Moving forward, a more focused approach to substantiate these ideas — some remain speculative — is necessary. For example, how primary afferents control the differential release of itch neuropeptides versus pain neuropeptides is an enigma. Whether GRPR neurons are composed of subtypes that mediate histaminergic itch versus non-histaminergic itch or mechanical itch versus chemical itch is also unclear. After separating itch from pain in the spinal cord, we do not understand how itch information is relayed through projection neurons to the brain. Finally, it is logical to presume that the neuropeptide code outlined here can be extended to the coding of other sensory modalities with slow response kinetics, such as chemical pain and pleasant touch. With the advent of technologies such as monosynaptic rabies virus tracing²⁴⁶ and an emphasis on circuits defined by GPCR expression, future endeavours to decipher modality-specific connectivity between primary afferents and spinal microcircuits may prove fruitful. Crucially, the advances in illuminating the neuropeptide code for itch support the idea that how sensory information flows from primary afferents to the brain via the spinal cord can be approached and understood, a quest once thought beyond our reach.