1. Introduction

The ability to feel the warmth of the sun or the coolness of a breeze on a summer's day; a pleasant caress of skin; to explore the texture, size, and shape of an object; and to feel pain as a mechanism to protect us from what is dangerous around us has fascinated mankind for centuries. Perhaps this fascination is rooted in the fact that the somatic senses of temperature, pain, touch, itch, and proprioception endow us with the ability to continuously stay in contact with the world within and around us. Attempting to explain how we react to heat, the 17th century philosopher René Descartes depicted a thread connecting the skin with the brain and particles of fire pulling on the thread.20 In the 1880s, distinct sensory spots on the skin were recognized and shown to respond to specific stimuli, such as touch, heat, or cold. This finding was taken as evidence for different sensations relying on the existence of distinct nerves tuned to specific types of stimuli.83,94 In the beginning of the 20th century, the discovery of different nerve fiber types with distinct conduction velocities, activation thresholds, and refractory periods made it possible to link specific fiber types to unique qualities of sensation, such as proprioception and different kinds of touch. Unlike innocuous touch and pressure, which are communicated through nerve fibers responding only to mechanical stimuli, noxious stimuli were found to activate a class of polymodal nociceptive nerve fibers responding to several kinds of unpleasant stimuli, including high-threshold mechanical force and intense heat.7,12 This work led to the general idea that the capacity to discriminate perceptual qualities arises from unique functions of a variety of sensory neurons involved in somatic sensation. This is possible because different neuron types are tuned to respond and transduce certain, but not all, kinds of stimuli, such as warm and cool temperatures, noxious heat, noxious cold, light punctate sensation, painful pressure, sharp objects, hair pull, chemical irritants, inflammatory substances, chemical and mechanical itch, and various kinds of touch and vibration.29,38

Apart from providing sensory awareness about our body and its surroundings, the somatosensory system is also essential for other functions we perform effortlessly and without much thought. For example, a continuous flow of information in a sensory–motor feedback loop involving several dozen muscles is necessary to coordinate movements when taking a sip from a glass of water or for the seemingly simple task of walking. Somatosensory neurons are also organized in discreet pathways with autonomic sympathetic and parasympathetic motor neurons driving reflexes involved in physiological homeostasis, such as thermoregulation through reflex control of sweat glands and selective vasoconstrictor systems in muscle and skin active when standing up and during hypothermia, respectively.16,49,74 Furthermore, pleasant and painful sensations involve not only discriminative perception and coordination of efferent reflex functions but also emotional and motivational components important for behavior. For example, social touch such as hugging and caressing induce sensations of pleasantness to facilitate emotional bonding, affiliative behavior, and well-being,82,88 while noxious stimuli not only convey sensory discriminative and protective reflexes but also include an affective-motivational dimension of unpleasantness contributing to long-term behavior avoiding harm.112,125

The existence of molecularly unique neuron types coding for each quality and dimension of somatosensation is unlikely. Instead, summation of activity and inactivity from different types of nerve fibers is likely to contribute to the sensory qualities and dimensions involved. Several recent studies support this conjecture, eg. the discovery of warmth sensation relying on the simultaneous activation of warmth-sensing and inhibition of cold-sensing nerves97; the requirement of different sensory neurons for protective reflexes and affective coping behavior elicited by the same stimuli45; the suppression of noxious cold (−10 to 10°C) by Calca+ neurons18,81; the modulation of touch-evoked itch by Merkel cells and Slc17a8-lineage (low-threshold mechanoreceptor [LTMR]) sensory neurons32,33,108; and alleviation of Npy2r+ neuron-dependent acute pain behavior by the simultaneous activity of the A-LTMR touch neurons.3 With a thorough molecular categorization of sensory neuron types, experiments can be designed with even greater precision, opening possibilities for gaining new, deeper insights into the cellular basis for nociception and pruriception.

2. Categorizing sensory neurons

Sensory neurons have historically been classified based on physiological properties, morphology, target innervation, spinal termination, developmental hierarchy, and functional and neurochemical properties (reviewed by Emery and Ernfors, 202029). This work has been instrumental for deciphering how the peripheral sensory nervous system is organized. However, these strategies have measured a dozen or so quantitative features, such as response to cold, heat, and mechanical stimuli; conduction velocity; neuron size; or a handful of marker genes, which might not be sufficient for reliable identification of functional units. By contrast, single-cell RNA sequencing (scRNA-seq) can measure thousands of features, is quantitative, and allows for an unbiased classification of cell types. Nevertheless, the accuracy of cell-type assignment is affected by data quality (library complexity, sequencing depth, technical errors, and technical noise) and the fact that only a proportion of the messenger RNA (mRNA) molecules within a cell is successfully converted for sequencing. In addition, different cell types contain different quantities of mRNAs resulting in differences in sequencing quality, affecting cell-type assignment. This is particularly relevant for sensory neurons which, unlike most neurons of the central nervous system, are of highly variable size. Cell-type identification is also affected by the resolution used for clustering. The aim of clustering in scRNA-seq is to identify the diverse neuron types which represent unique functional units. However, all molecular cell-type taxonomies are hampered by the questions “what is the relation between variable molecular features and function?” and “what is a cell type?”152

Sensory neurons are intimately associated with a surrounding sheath of satellite glial cells which is often insufficiently removed during sample preparation. Single-nucleus RNA sequencing (snRNA-seq) can be used to mitigate this. Although most of a cell's genomic information can be accessed in the nucleus alone, there is less starting material when the cytosol is not included leading to a lower median number of unique genes detected per cell and the presence of ambient RNA confounding classification. Gene capture is a vital metric, and although the ease of clustering and identification of cell types might be improved in snRNA-seq through increasing the number of analyzed cells, much of biological insights into cell-type function can be lost due to the detection of a smaller proportion of the genes expressed by the cell. Because of this, we have always favored data generated by scRNA-seq and will in this review focus on results obtained by this method. Although there are several scRNA-seq studies published on the heterogeneity of mouse somatosensory dorsal root ganglion (DRG),14,60-62,114,137,151 trigeminal,65,91,145 and jugular60 neurons, there has been a general lack of efforts to integrate the data to find recurrent cell populations across the different studies using unbiased computational strategies. For this purpose, we have used intersample mapping to propagate information between studies and machine and deep-learning strategies to find neural correlates between our studies60,61,131,151 as well as the study of mouse DRG by Sharma et al.114 Because we know the relation between sequenced cells in these studies, our taxonomy of sensory neurons is heavily influenced by the molecular features as revealed in these studies.

3. Strategy of somatosensation along the body axis

General somatosensory neurons are present in the trigeminal ganglia, the superior and inferior (jugular) ganglia of the glossopharyngeal and vagal nerves, and the DRG, with a very small number in the geniculate ganglion.19,133 In addition, some accessory general somatosensory neurons may be present along the XIth cranial nerve. Sensory neurons located in the trigeminal ganglia convey sensory information from the face and oral cavity (including the dura mater and teeth, respectively), lips and mouth having the greatest representation. Neurons from the geniculate and jugular ganglia innervate parts of the head, the ear, and throat while DRG neurons convey cutaneous and deep sensation from the rest of the body. Unlike somatosensory neurons which detect and relay conscious sensations, visceral sensory neurons are largely involved in reflex functions regulating body homeostasis related to organ function and transmitting gustatory information from taste buds.60,79,136,140 These neurons are organized in cranial ganglia, including the petrosal and nodose ganglia which are dedicated visceral ganglia. Comparing neuron types in the general somatosensory trigeminal, jugular, and DRG identified by scRNA-seq suggests that the neuronal basis for somatosensation is conserved along the body axis, while visceral sensory neurons of the nodose ganglion are represented by completely different kinds of neurons.60 However, scRNA-seq of colonic DRG neurons43 has led to the identification of neuron types not found in other scRNA-seq studies of the DRG, suggesting the existence of somatosensory neuron types dedicated to visceral innervation. The data from this study have not yet been compared with other DRG scRNA-seq data sets using unbiased computational strategies, and thus, the similarities and differences to neurons identified in other studies remain to be explored.

4. Overview of sensory neuron types and their putative functions

Based on the taxonomy presented here, 19 molecular types of DRG sensory neurons are predicted. An overview of the neuron types, expression of sensory transducers, percent contribution to neurons of the dorsal root ganglion, peripheral innervation, key markers, and function is outlined in Figure 1. This figure also shows the hierarchical relation between the different adult neuron types as well as key transcription factors and growth factor receptors known to drive cell-type diversification during development. Interestingly, the molecular relation in adult largely reflects ontogeny.

Figure 1.:

Molecular and functional taxonomy of mouse dorsal root ganglion sensory neuron types. Molecular types and hierarchical relationship are based on highly variable genes in mouse scRNA-seq data. The classification is based on Usoskin et al. (2015), Zeisel et al. (2018), and Sharma et al. (2020), and the nomenclature which follows Emery and Ernfors (2020) is indicated at the end branch of the hierarchical tree. The first or top box represents a conflation of hierarchically related neuron types with information summarizing fiber type (C, Aδ, and Aβ), modalities of sensation evidenced by functional experimentation or predicted based on expression (M, mechanical; H, heat; HTMR, high-threshold mechanoreceptor; LTMR, low-threshold mechanoreceptor). Boxes from top to bottom: neuron type, major sensory transduction genes expressed, percentage of representation among all DRG neurons, innervation targets, examples of marker genes, and finally function of the neurons. Question mark indicates prediction based on expression. Note that the hierarchical relationship largely reflect ontogeny which is outlined in the hierarchical tree as examples of transcription factors and receptors. The neurogenic transcription factor Neurogenin-2 (Ngn2) drives early neurogenesis generating Aδ nociceptor, Aβ LTMR, and proprioceptor neuron types which diversify through transcriptional activation, repression, and signaling through Trk and Ret receptors. Neurogenin-1 is responsible for neurogenesis of later born neurons defined by Runx1, Prdm12, and TrkA signaling which diversify into mostly unmyelinated nociceptive and pruriceptive neurons. Nppb, natriuretic peptide precursor B; scRNA-seq, single-cell RNA sequencing; Sst, somatostatin.

Neurons of the atlas include the innocuous mechanoreceptors represented by 1 type of Aδ-LTMR, Aβ rapidly adapting (RA) LTMR, Aβ slowly adapting (SA) LTMR, proprioceptor, and 1 C-LTMR, showing overall consistency with functional units explored in the mouse and human.2,38,154 Although this review is not focusing on LTMRs, it is interesting to note that 2 functional features seem to separate LTMRs into distinct molecular types: rate of adaptation and conduction velocity. The molecular properties underlying these features could be mutually exclusive and because of this, separate as unique entities defining LTMR types. By contrast, the molecular classification of A-LTMRs does not reflect peripheral termination morphology. For example, only 1 type of Aβ RA-LTMR neuron type has been identified by scRNA-sequencing,114,131,151 which terminates as longitudinal lanceolate endings in hair follicles and in the Meissner and Pacinian corpuscles.38 Furthermore, the fibers of 1 type of Aβ SA-LTMR terminate either in Merkel cells or putatively also as circumferential endings in hair follicles.

Machine learning–based analyses have revealed that Usoskin et al.,131 Zeisel et al.,151 and Sharma et al., 2021114,131,151 (from now we refer to these studies as only “Usoskin,” “Zeisel,” and “Sharma” for brevity) independently identified the same nociceptor and pruriceptor neuron types in mouse DRG, although with varying resolution. The low-resolution view reveals the major divisions of neuron types, whereas at high resolution the individual subtypes within the major divisions can be identified. At high resolution, there are 14 types of nociceptors and pruriceptors with predicted C-fiber or Aδ-fiber properties. These can be conflated into 7 major types with subtypes based on the hierarchical relationships. We believe the subtypes represent functionally similar units because many of the known transducers for somatic sensations such as Trpm8, Trpv1, Trpm2, Trpm3, Piezo2, and a multitude of itch receptors are consistent with both the conflated and high-resolution taxonomy. The conflated taxonomy of nociceptors and pruriceptors include the following neuron types using the Usoskin nomenclature: TrpM8high, PEP1-3, and NP1-3. The full taxonomy includes 3 types of cold-sensing “TrpM8” neurons, 4 types of mechano-heat polymodal nociceptors (peptidergic “PEP1” neurons), 2 Aδ-nociceptor neuron types (putative heat or inflammatory peptidergic “PEP2” and putative a high-threshold mechanoreceptor type (A-HTMR) peptidergic “PEP3” neurons), 2 Mrgprd+ neuron types involved in light punctate reflex withdrawal and β-alanine itch (nociceptor and pruriceptor 1 “NP1” neurons), 2 Mrgpra3 neuron types involved in itch (nociceptor and pruriceptor 2 “NP2” neurons), and 1 somatostatin (Sst) and natriuretic peptide precursor B (Nppb) expressing neuron type involved in itch, allergic itch, and possibly perception of warmth (nociceptor and pruriceptor 3 “NP3” neurons) (Fig. 1). Although most of these major types have been shown to represent distinct functional units, unique functions for some of the subtypes remain to be shown. For example, there are 4 PEP1 subtypes, yet unique functions as compared to the other 3 have only been demonstrated for one of these 4. Conversely, Usoskin identified 2 and both Zeisel and Sharma one molecular type of proprioceptive neurons,114,131,151 while recent studies focusing on proprioceptive neurons found greater molecular heterogeneity, corresponding to subtypes for Ia, Ib, and type II proprioceptive nerve endings.95,144 Furthermore, a unique neuron type expressing Mrgprb4 has been functionally described135 but is yet to be identified in scRNA-seq studies. It is likely to be embedded among the NP2 neurons72 and fails to separate as a distinct cell type due to its scarcity. Better resolution and discovery of additional molecular neuron types could be expected by increasing the quality and number of sequenced cells.

5. Cross-species strategies of somatosensation

DRG neurons from nonhuman primate (NHP, rhesus macaque) have been analyzed by scRNA-seq.61 Integration and co-clustering as well as machine learning comparisons between high-quality mouse and NHP scRNA-seq data sets have revealed the existence of corresponding cell types between species, suggesting that the basal layout of sensory neuron types is largely conserved between these species.61 Nevertheless, when looking at the level of individual genes, major differences in expression between the corresponding neuron types of the mouse and NHP are observed, some of which directly relate to transduction of somatosensory stimuli. Transcriptomic analysis of human DRG has been performed using spatial transcriptomics128 and snRNA-seq.90 There is a lack of validated computational strategies for comparisons between data obtained from scRNA-seq and spatial transcriptomics. Therefore, the molecular correlates between human neurons as identified by spatial transcriptomics and mouse or NHP neurons have not been extensively analyzed.128 Nguyen et al.90 identified 15 clusters based on snRNA-seq of human neurons (predicted neuron types, H1-H15) and identified mouse–human cell type counterparts based on some variably expressed genes and co-clustering. Although some clusters could not be mapped back to any mouse counterpart (H4 and H12 clusters), others showed resemblance to mouse (H1, H2, H5 clusters corresponded to mouse PEP1 C-polymodal subtypes; H8 to mouse TrpM8high cold-sensing neurons; H9 to mouse PEP3; H10 to mouse NP1; H11 to mouse NP3; H13 to mouse Aδ-LTMRs; H14 to mouse Aβ-LTMRs; H15 to mouse proprioceptors; and C-LTMRs were not found), although some markers and inconsistencies were observed in the mouse–human comparison. These results give a first indication of the neural basis for human somatosensation; however, further studies are needed to understand the exact relationship and molecular makeup.

In mouse and NHP, the current data quality and molecular taxonomy of sensory neurons open for the assignment of function to defined neuron types with insights into the molecular features shaping their physiology. Below we describe the molecular heterogeneity of mouse and NHP sensory neuron types based on scRNA-seq studies where cell type correlates between different laboratories and studies have been shown using unbiased computational strategies120,138,156 and further summarize known functions of the different neuron types with a focus on nociceptive and pruriceptive neurons. The cell-type correlates and nomenclature used in these studies are presented in Figure 2. For those interested in interactive reading, we generated a single web site for the mouse Zeisel and Sharma data sets with an option to select the different cell-type nomenclatures (“Usoskin,” “Zeisel,” or “Sharma”) when querying gene expression in each of the different data sets (https://ernforsgroup.shinyapps.io/MouseDRGNeurons/) and a single web site containing both the STRT2 and SmartSeq2 macaque data (https://ernforsgroup.shinyapps.io/macaquedrg/). Expression of genes makes sense only when placed in known cell types. Thus, a molecular taxonomy of sensory neuron types is essential for insights into the neural basis of somatosensation and for understanding what goes wrong in chronic pain. Because of this, the review is centered around cell types rather than individual genes or molecules and there is a disproportionate impact by studies examining cell-type function by genetic strategies for ablation or by synthetic activation and inactivation of different neuron types using optogenetics or chemogenetics in mice. Furthermore, for those interested in LTMRs, we refer to other reviews on this topic.2,3,45,70,87,94,112

Figure 2.:

Schematic overview of the corresponding dorsal root ganglion neuron types between mouse and macaque data sets and “Usoskin,” “Zeisel,” “Sharma,” and “Kupari and Usoskin” nomenclatures. LTMR, low-threshold mechanoreceptor. NA, not available.

6. TrpM8high, C-cold-sensing neurons

Mouse TrpM8high neuron type is named “high” because in macaque (PEP3, see below) and possibly also in human another neuron type also expresses Trpm8, although at much lower level. The TrpM8high neuron type includes small-sized neurons critical for cool sensation that also importantly contribute to sensing painful cold. They innervate the skin and likely also mouth and esophagus,56,150 representing about 6% to 8% of all DRG neurons.21

This neuron type expresses unique features including Trpm8 and Foxp2 but lacks the mechanosensitive channel Piezo2 and heat-sensitive TRP channels (ie, Trpv1, Trpa1, Trpm2, and Trpm3). Usoskin and Sharma identified only one type of TrpM8high neuron, while Zeisel identified 3 subtypes. This difference is due to the strategy and resolution used for clustering and identification of cell types.61

Based on marker expression associated with myelination, Zeisel predicted 2 subtypes representing putative C fibers and 1 putative Aδ fiber.151 Whether or not the 3 TrpM8high subtypes distinguished by Zeisel represent functionally different neuronal types remains to be shown. Intriguingly, 3 functionally different subtypes of Trpm8+ neurons have been identified in the mouse trigeminal ganglion using Ca2+ imaging: 1 type that is active only during cool temperatures (28-20°C), another active only at noxious cold temperatures (<18°C), and a third with a hybrid response.148

Consistent with the unimodality of TrpM8high cold-sensing neurons, genetic knockout studies have revealed the TRPM8 channel as a critical transducer for cool temperatures in the mouse.4,17,22 Mice lacking TRPM8 also display a clearly reduced aversive behavior toward noxious cold, indicating that the TRPM8 ion channel contributes to, but is not essential for, noxious cold sensation. By contrast, ablating the entire population of Trpm8+ sensory neurons leads to a complete loss of cold sensitivity and a greater loss of noxious cold sensitivity compared with mice where only the Trpm8 gene has been knocked out.57,87,101 These studies highlight the significant role of TrpM8high neurons in innocuous and noxious cold sensing, while simultaneously indicating that other neuron types may also participate in the detection of noxious cold through a mechanism independent of the TrpM8high neuron types. This conclusion is also supported by another set of experiments. Low-threshold cold-sensitive fibers are largely unimodal consistent with the TrpM8high neurons, while high-threshold cold nociceptors also respond to noxious heat and mechanical stimuli.76 Among C-fiber neurons, the TrpM8high neurons are the only nociceptors not expressing Scn10a (Nav1.8). Consistently, deleting Nav1.8 in mice has no effect on cold behavior down to 5°C while responding to extreme cold stimuli (below freezing) is completely absent.73 Moreover, Calca+ (CGRP) and Scn10a+ C-fiber neurons of unknown molecular types were found to contribute to noxious cold sensing through a TRPM8-independent mechanism representing “silent” cold-sensing neurons sensitized during neuropathic cold allodynia.75 The identification of several channels proposed as alternative cold transducers11,76 has not helped in identifying neuron types contributing to the remaining noxious cold sensitivity in Trpm8-null mice, and furthermore, extreme cold below freezing seems to activate sensory neuron types regardless of type, likely because of tissue damage.118 Finally, TrpM8high neuron types have been shown to also take part in cold-induced analgesia, indicating that activity in these neuron types can suppress pain.103

7. C-mechano-pruriceptors NP1, NP2, and NP3

The NP1, NP2, and NP3 classes include neurons involved in protective mechanosensory reflexes, itch, and potentially also nociception. In addition, these neurons express various thermosensitive TRP channels and, according to physiological recordings of cells or nerves, at least some are heat sensitive, although the in vivo role for noxious heat transduction in mice and humans seems negligible.13,37,45,55,66,101 Among these neuron types, NP3 is the most different in the mouse and the NHP, while the NP1 and NP2 types are hierarchically closer to each other.61,151 The thermosensitive and chemosensitive channels Trpv1 and Trpa1 are expressed in the NP-class neurons. This expression might reflect their role for signal transduction through receptors for histamine and nonhistamine itch-inducing compounds rather than contributing as heat sensors.53,143

Forced activation of the MrgprdCre lineage neurons (including NP1, NP2, and NP3) by optogenetics fails to generate any place aversive behavior in naive mice, but does so in neuropathic animals.139 Thus, it has been proposed that none of the 3 neuronal types in the NP class are bona fide nociceptors, ie, high-threshold receptors with a function to transduce and encode noxious stimuli.139 Chemical itch is mediated through specific kinds of receptors whose expression patterns place chemical itch in the C-fiber NP class of neurons in the mouse and NHP. Exogenous and endogenous mediators released from damaged or infected tissues increase the extravasation of vessels attracting immune cells to the injured site for the inflammatory response.77 The released “inflammatory soup” is rich in purines, amines, cytokines, chemokines, lipid mediators, and growth factors. These include histamine, serotonin, leukotrienes, prostaglandins, bradykinin, and others of which some could directly activate or sensitize nociceptors and thereby mediate itch or pain. As is outlined below, NP2 and NP3 class neurons in mice are predicted to differentially respond to such mediators generating itch sensation while NP1 neurons are responsible for mechanical threshold detection and some types of itch. Consistent with this mouse taxonomy, itch in humans is conducted by nerves different from polymodal C-fiber nociceptors.52,110

8. NP1, C-mechanopruriceptor neurons

The nonpeptidergic NP1-type neurons are of small diameter, express Mrgprd in both mouse and NHP, and mostly innervate the superficial layer of the epidermis in the glabrous and hairy skin of the mouse.156 They are polymodal C fibers representing about 25% to 32% of DRG neurons and are involved in pruriception and threshold detection to light punctate mechanical stimuli (von Frey). This neuron type expresses unique molecular features, eg, Mrgprd, Mrgprx receptors, Lpar3, and Mdfic2. Zeisel defines 2 NP1 subtypes, while Usoskin and Sharma define only 1. This difference is due to the resolution used for clustering and identification of unique clusters.61 Whether or not the 2 subtypes of NP1 neurons represent functionally distinct units remains to be determined.

Based on physiological recordings, NP1-type neurons can be activated by mechanical, chemical, and heat stimuli in rodents26,66,71,105; however, although these neurons are critical for the withdrawal reflex evoked by punctate stimuli from von Frey filaments, their ablation does not affect heat or cold sensation.13,45 Furthermore, they are dispensable for the mechanical pain evoked by pinching.44 Consistent with the lack aversive behavior when activating all NP-class neurons (NP1-3),139 specific activation of NP1 fibers fails to generate any place aversive behavior in the naive animal,6,139 while it seems to do so after spared nerve injury and chronic inflammation (using complete Freund adjuvant).1,139 Thus, current studies support the idea that while NP1 neurons are not nociceptors in the naive state, they perform this role during chronic inflammatory pain. Still, the integrated contribution of NP1 neurons to mechanical allodynia during chronic pain, especially in relation to other neuron types, needs to be systematically evaluated. In the mouse, NP1 neurons are also involved in pruriception. Although NP1 neurons do not respond to histamine,66 β-alanine, a natural pruritogenic compound, directly binds and activates the MRGPRD receptor,117 which then engages TRPA1 and induces itch.141 Furthermore, the NP1 neuron type is the only one expressing the pruritogenic lysophosphatidic acid receptor Lpar3 in the mouse. Thus, in the naive situation, mouse NP1 neurons seem to primarily function as mechanopruriceptive neurons in reflex withdrawal to punctate stimuli and as sensors to some selected pruritogens. In NHPs, based on gene expression, NP1 neurons are predicted to have broader functions for itch, expressing receptors for histamine and different nonhistamine pruritogenic compounds, and often share expression with NP2 neurons, as will be described below.

9. NP2, C-pruriceptor neurons (chemo)

NP2-type neurons are of small diameter, express Mrgpra3 in the mouse, and mostly innervate the superficial layer of the epidermis while occasionally wrapping around hair follicles.25,37,156 They are polymodal high-threshold C fibers responding to both noxious mechanical and heat stimuli, but not cold,37 and represent 4% to 7% of sensory neurons.25,155 In vivo in the mouse, NP2 neurons have been proposed to be a neuron type dedicated to itch,37 although recent results suggest broader functions (see below). Zeisel defines 2 NP2 subtypes, while Usoskin and Sharma show only one.114,131,151 This difference is due to the resolution used for clustering and identification of unique clusters.61 Whether or not the 2 subtypes of NP2 neurons represent functionally distinct units remains to be determined.

Although the cells are heat sensitive at the cellular level, ablation of the Mrgpra3+ neurons in the mouse has no effect on pain behavior induced by heat and cold noxious stimuli, nor does it induce any shift in light punctate sensation evoked by von Frey hairs. Furthermore, no behavioral difference was observed in these mice in response to heat or mechanical stimuli in the inflamed animal.37 By contrast, mouse NP2 neurons are critical for chloroquine-induced itch mediated through the ligand activation of MRGPRA3 receptors that are uniquely expressed in these cells. Consistently, in mice where the Mrgpra3+ neurons are ablated, chloroquine itch is absent and histamine, serotonin, and several other itch agents display attenuated effects on induced scratching behavior.37,104 As in the case of MRGPRD, also MRGPRA3-dependent itch relies on TRP channels.141 Although NP1 and NP2 neuron types seem to be distinct functional units in the mouse, they seem to share many pruriceptive functions in NHPs. Contrary to the mouse, where chloroquine itch is mediated through the activation of MRGPRA3,37,67,106 this receptor is not conserved in humans, and chloroquine is instead engaging MRGPRX1.67 Thus, while chloroquine itch is conveyed by NP2 in the mouse, it is predicted to act in NHPs through both NP1 and NP2 neurons through the ortholog receptor MRGPRX1. A shared function of NP1 and NP2 neuron types in NHPs is predicted also for β-defensin-induced itch. β-defensin expression is elevated in the skin lesions of inflammatory dermal diseases such as psoriasis and atopic dermatitis, inducing itch through the activation of human MRGPRX1 and MRGPRX2 or mouse MRGPRA3 and MRGPRC11.124,130,153 While in the mouse β-defensins are predicted to act exclusively through the NP2 type of neurons based on receptor expression, in NHPs the ortholog receptors are expressed in both NP1 and NP2 neurons,61 again indicating species differences in the cell-type responses to external stimuli.

Itch can also be initiated through the protease-activated receptor F2RL1 (PAR2) by high levels of proteases observed in the skin of patients with chronic itch conditions, such as atopic dermatitis. In mice, PAR2 agonist-induced itch is absent in Par2-deficient mice.5,116,123,142 This receptor is expressed exclusively in NP3 neurons in mice where it is not only critical for itch but also evokes mechanical hyperalgesia.40 This functionality is not conserved in NHPs where PAR2 is not expressed in sensory neurons at all, indicating any contribution of PAR2 to itch in humans to be indirect, perhaps through skin cell types and/or immune cells.50 Although much focus has been on PAR2 as a contributor to itch in humans,10,122,123 the PAR family consists of 4 genes: F2R (PAR1), F2RL1 (PAR2), F2RL2 (PAR3), and F2RL3 (PAR4), which can be activated by endogenous and exogenous sources, including proteases. Nonhuman primate sensory neurons express PAR1 and PAR3 exclusively in the NP1, NP2, and the PEP2 type (an Aδ nociceptor). Thus, NP1 and NP2 neurons seem to be functionally similar in NHPs also in this respect. A shared function of NHP NP1 and NP2 neuron types is also predicted for cholestatic itch. Lysophosphatidic acid increases in sera of patients with cholestatic liver diseases and pruritus and can cause both pain and itch in humans.27 In mice, lysophosphatidic acid elicits both itch-related39,59 and acute pain-like behaviors, the latter being dependent on TRPV1.48,93 This is consistent with the effects of lysophosphatidic acid acting through LPAR3 expressed exclusively in mouse pruriceptive NP1 neurons and LPAR1 expressed in both pruriceptive NP3 and nociceptive PEP1 neurons (which also co-express TRPV1). By contrast, NHPs express LPAR3 in NP1 and NP2 neurons and LPAR1 in PEP2 (which also expresses TRPV1). Among the few pruritogens identified that mediate cholestatic itch apart from lysophosphatidic acid are bilirubin and bile acids8 which activate the MRGPRX4 receptor in human.149 MRGPRX4 is expressed in NP1 and NP2 neurons of NHPs. Thus, it seems that NP1 and NP2 neuron types are essential in NHP for chronic itch during cholestasis. Furthermore, a differential engagement of NP1, NP2, and PEP types of neurons by proteases and lysophosphatidic acid might represent itch and nociception, respectively, by the same compounds.

The proallergic cytokine thymic stromal lymphopoietin (TSLP) is a pruritogen142 acting through a heterodimeric receptor of IL7R and CRLF2 likely using an unknown indirect mechanism because sensory neurons do not express the receptor pair in either the mouse or NHPs. Another mast cell and keratinocyte released cytokine is IL33, which has been proposed to act directly on sensory neurons64 through the IL1RL1 (ST2) and IL1RAP receptors78; however, based on scRNA-seq, the effects are indirect as these receptors are absent in mouse and NHP sensory neurons.

Contrary to previous studies where activation of all NP class neurons (NP1-3) did not lead to aversive behavior,139 recent results indicate mouse NP2 neurons to be intrinsically multimodal based on their activation mode. Metabotropic activation evokes mostly pruriceptive behavior while ionotropic stimulation of the same NP2 cell type evokes pain-like behavior.113 This study corroborates NP2 neurons as pruriceptors but suggests that they can under some circumstances also function as nociceptors. Consistent with this, a putative contribution of NP2 neuron types to heritable musculoskeletal pain was inferred by partitioning genomic loci associated with chronic pain in human onto primate sensory neuron types. Genome-wide associations for pain converged on 2 different neuronal types, one of which was NP2.61 Thus, although a critical role of NP2 neurons for itch is firmly established, further studies on their contribution to chronic pain states are needed.

10. NP3, C-pruriceptor neurons (mechano-chemo)

The NP3 type neurons contribute to both mechanical and chemical pruriception and are predicted in mouse to also take part in warmth sensation. They are very small-diameter skin-innervating neurons44; express Sst, Nppb, and neurotensin (Nts); and represent about 5% to 9% of mouse dorsal root ganglion neurons in the adult.131

NP3 neurons are polymodal high-threshold C fibers responding to both noxious mechanical and heat but not cold stimuli,44 although the in vivo functional role for noxious mechanical and heat stimuli is likely negligible. Single-cell and bulk RNA-seq of mouse DRG have revealed expression of a rich repertoire of pruritogenic receptors in NP3 neurons, many of which are associated with inflammatory conditions and persistent itch.120,131 Thus, the unique expression of many of these receptors only in NP3 neurons predicts a central role for these neurons in responding to many pruritogens. Consistently, specific activation of NP3 neurons produces scratching behavior.44 It has also been shown that mice deficient in the neuropeptide NPPB, which is uniquely expressed in NP3 neurons, display a loss of behavioral responses to a range of pruritogenic agents (histamine, serotonin, PAR2 ligand, and chloroquine).37,87 Nonhistamine mast cell pruritogens, such as cysteinyl leukotrienes, interleukin-31, and sphingosine-1-phosphate,5,42,120 all produce itch and their cognate receptors are uniquely expressed in NP3 neurons (Cysltr2, Il31ra and Osmr, and S1pr1, respectively).120,131 These neurons also express TRPA1, which is required as a signaling component for itch.42 Serotonin5,120 can engage several receptors many of which are implicated in rodents.7,65,97,100,125,133,133 However, only Htr1a and Htr1f are robustly expressed in mouse sensory neurons and are specific for mouse NP3 neurons.151 Consistently, application of a serotonin receptor HTR1F agonist to cultured sensory neurons leads to calcium flux, but this effect is gone in cultures from mice in which the NP3 neurons were genetically ablated; furthermore, HTR1F agonist-induced itch behavior was reduced in such mice.121 Therefore, it seems that NP3 neurons are uniquely responsible for the direct itching effect of serotonin, cysteinyl leukotrienes, interleukin-31, and sphingosine 1-phosphate. Most of these functions are predicted to be conserved also in NHPs based on receptor expression. For example, S1pr1 is exclusively expressed in NHP NP3 neurons which also express the co-receptor Trpa1 and beyond this receptor, there are no other sphingosine 1-phosphate receptors in NHP sensory neurons. Nevertheless, some receptors for other pruritogens are expressed at lower levels also in NHP NP1 and NP2 neurons. For example, interleukin-31 receptors are expressed in NHP NP1 and NP2 in addition to NP3 neurons.61

As mentioned above, NP2 neurons contribute only partly to histamine-induced scratching in mice. It is possible that the remaining histamine itch phenotype is mediated through NP3 neurons because the histamine receptor Hrh1 is expressed in both NP2 and NP3 neurons in the mouse. By contrast, NHP NP3 neurons alone can be predicted to be critical for histamine-induced itch because HRH1 expression is abundantly present only in NP3 neurons. Itch induced by histamine is believed to rely on the downstream activation of TRPV1 by phospholipase-b347,115 and both are expressed in the NP3 neurons of mouse and NHP. Thus, it seems histamine itch in NHPs relies on NP3 neurons while in the mouse, both NP2 and NP3 neurons could contribute.

Toll-like receptors (TLR2-5 and TLR7) have been implicated in itch by coupling to TRPA1 or TRPV1.69,70,86,138 Based on mouse scRNA-seq data, TLR4 expression is somewhat enriched in the PEP classes of neurons while TLR5 is present in LTMRs.114,131,151 However, NHP sensory neurons express robust levels of TLR2 in NP3 and A-LTMRs, TLR4 and TLR7 in PEP2 neurons, and consistent with mouse, TLR5 is expressed in A-LTMRs (MYD88 adaptor in all cell types). Thus, although a putative TLR5-dependent itch mechanism could be conserved across the species, it seems that NHPs have gained functions related to TLR2, TLR4, and TLR7 in sensory neurons.

Acute mechanical itch remains intact in animals after ablation of all sensory neurons expressing Scn10a (Nav1.8), which include all nociceptors and pruriceptors except the TrpM8high neurons. Furthermore, activating Aβ-LTMRs through TLR5 with the bacterial protein flagellin—which in humans causes itch92—leads to predominantly TLR5-dependent scratching responses in mice.96 Both these results po