A subpopulation of itch‐sensing neurons marked by Ret and somatostatin expression

Itch, the unpleasant sensation that elicits a desire to scratch, is mediated by specific subtypes of cutaneous sensory neuron. Here, we identify a subpopulation of itch‐sensing neurons based on their expression of the receptor tyrosine kinase Ret. We apply flow cytometry to isolate Ret‐positive neurons from dorsal root ganglia and detected a distinct population marked by low levels of Ret and absence of isolectin B4 binding. We determine the transcriptional profile of these neurons and demonstrate that they express neuropeptides such as somatostatin (Sst), the NGF receptor TrkA, and multiple transcripts associated with itch. We validate the selective expression of Sst using an Sst‐Cre driver line and ablated these neurons by generating mice in which the diphtheria toxin receptor is conditionally expressed from the sensory neuron‐specific Avil locus. Sst‐Cre::AviliDTR mice display normal nociceptive responses to thermal and mechanical stimuli. However, scratching behavior evoked by interleukin‐31 (IL‐31) or agonist at the 5HT1F receptor is significantly reduced. Our data provide a molecular signature for a subpopulation of neurons activated by multiple pruritogens.


Introduction
The perception of physical and chemical stimuli through the skin is initiated by peripheral sensory neurons that have their cell body in the DRG. The complexity of somatosensation is reflected by the fact that a myriad of sensations including touch, pain, itch, and temperature are recognized by the peripheral nervous system. It has long been debated whether this functional complexity arises from activation of specific subtypes of sensory neuron for each stimulus modality, or from encoding and summation of neuronal activity generated by neurons that can detect a broad range of stimuli [1]. Recently, it has been proposed that the sensation of itch is a discrete sensory modality that utilizes a dedicated neuronal pathway tuned to perceive only this sensation [2][3][4].
Itch, the sensation that elicits a desire to scratch, serves a protective function against potentially harmful environmental irritants [5]. Although unpleasant, itch is inherently different from pain, both in its sensory quality and its behavioral outcome (scratching compared to withdrawal). The neuronal pathways that mediate itch versus pain also appear to be distinct. In the spinal cord, ablation of neurons expressing receptors for the neuropeptides, gastrinreleasing peptide (Grp) [4] or natriuretic polypeptide b (Nppb) [3], reduces itch responses to multiple pruritogens but does not affect nociceptive behavior. Similarly, subpopulations of C-fiber primary afferents are activated by pruritogens [6][7][8] and ablation or selective activation of DRG neurons positive for the Mas-related G proteincoupled receptor A3 (MrgprA3) impacts upon scratching behavior but not pain [2].
Itch is further categorized by its dependence upon histaminergic or non-histaminergic mechanisms. Histamine-dependent itch is elicited through activation of the H1 receptor (HRH1) and signaling through phospholipase-b3 (PLCb3) and the ion channel TRPV1 [9,10]. The existence of further histamine-independent pathways is supported by observations that many chronic pruritic syndromes such as atopic dermatitis are resistant to antihistamine therapy [11]. Mechanistically, histamine-independent itch is likely to be mediated by activation of the ion channel TRPA1. For example, injection of the antimalarial agent chloroquine induces itch [12] via MrgprA3 receptors [13] functionally coupled to TRPA1 [14], and TRPA1 is also required for itch produced by oxidative stress and leukocyte accumulation [15,16]. Other antihistamine-resistant itch responses include those elicited by cytokines such as interleukin-31 (IL-31) released from T cells during allergic itch [17,18]. IL-31 induces severe pruritus and may be a key mediator in atopic dermatitis [19].
Further information of the molecular profile of itch-sensing neurons and identification of molecular markers for different subtypes of itch neuron would be valuable for understanding how different pruritogens activate itch pathways. Intriguingly, a recent study which took an unbiased approach to classify sensory neuron subtypes identified three populations of putative itch receptors [20]. Each of these populations was enriched for transcripts associated with itch. However, from their gene expression profile, it was not clear whether they all signal only itch or can also contribute to other sensations such as pain.
To distinguish different populations of sensory neuron and ultimately define their function, we examined the expression pattern of the glial-derived neurotrophic factor (GDNF) receptor Ret in mouse DRG. Almost every DRG neuron expresses at least one neurotrophic factor [21], and approximately 60 percent of cells are marked by Ret [22]. The Ret tyrosine kinase is the signaling receptor for GDNF family ligands GDNF, neurturin, artemin, and persephin which bind via GPI-anchored co-receptors termed GFRa1-4 to initiate signaling through Ret [23]. Two distinct waves of Ret expression arise during development with the first occurring prior to E11.5 and the second emerging subsequently [22,24]. Early Ret-positive neurons develop into rapidly adapting mechanoreceptors and express the co-receptor GFRa2 and high levels of Ret [25,26]. Late Ret-positive neurons form a large heterogeneous group of non-peptidergic nociceptors that can be distinguished by their binding of the plant lectin IB4 [22,24]. A further population of Ret-positive neuron co-expresses the enzyme tyrosine hydroxylase (TH) [27,28] and forms C-fiber low threshold mechanoreceptors which have been implicated in the affective component of touch [29]. Finally, a fourth population of Ret-positive neurons that expresses the neurotrophin receptor TrkA in adult mice has also been described [22,30]. This population is rare, corresponding to around 10% of all Ret cells, and its role in vivo is unknown.
We sought to determine the function of this unique population of Ret-positive sensory neuron. To this end, we took a genetic approach and generated mice where eGFP expression was driven from the Ret locus exclusively in peripheral sensory neurons [31,32]. We identified multiple subpopulations of Ret-positive neurons in DRG which were quantified using flow cytometry. Microarray analysis of Ret-expressing neurons that were negative for IB4 uncovered a sparse population of cells enriched in transcripts for TrkA, neuropeptides such as somatostatin (Sst), and pruritogen receptors. We validated the expression of Sst in this population using an Sst-Cre driver line and generated a new mouse line to selectively ablate these neurons in vivo. Mice displayed normal nociceptive responses to thermal and mechanical stimuli. However, scratching behavior evoked by several classes of pruritogen was significantly reduced. Thus, Ret marks a population of itch receptors characterized by their co-expression of Sst and multiple pruritogen receptors.

Ret-eGFP expression in primary sensory neurons
To examine Ret expression in the adult peripheral nervous system, Ret eGFP/+ mice [31] were crossed with Avil cre/+ [32] mice to obtain heterozygote Avil-Cre::Ret eGFP/+ mice. This approach allowed us to target all DRG neurons and avoid extraneous GFP expression in surrounding tissues. Heterozygous mice were viable, exhibited no overt behavioral phenotype, and displayed robust eGFP fluorescence in peripheral sensory ganglia.
We investigated Ret-eGFP distribution in DRG by co-staining sections from Avil-Cre::Ret eGFP/+ mice with a selection of markers for different subtypes of sensory neuron. Ret-eGFP was present in 55 percent of neurons ( Fig 1M) and displayed a broad range of fluorescence intensities across different cells. We examined expression with IB4 and NF200, markers of non-peptidergic nociceptors and myelinated neurons, respectively, and observed overlap with the majority of Ret-eGFP-positive neurons (Fig 1A-D), reflecting the early and late Ret neurons described previously [22,24]. We further investigated Ret-eGFP co-expression with TH, a marker of C-fiber low threshold mechanoreceptors [27,28]. eGFP fluorescence was evident in many TH-positive neurons and these cells were not colabeled with NF200 or IB4 (Fig 1E-L). Our analysis also indicated that a small proportion of Ret-eGFP-positive neurons were not marked by either IB4, NF200, or TH, suggesting the existence of a novel subtype of Ret-expressing neuron. We quantified the overlap between Ret-eGFP and each marker and determined that 14% of Ret-eGFP-positive neurons were negative for all markers (Fig 1N), supporting the idea that this population may reflect a functionally uncharacterized subset of primary afferent neuron.
We utilized Avil-Cre-driven Ret-eGFP expression to examine the peripheral and central projections of Ret-positive sensory neurons. In the skin, Ret-eGFP fluorescence was broadly distributed and present in free nerve endings terminating in the dermis and epidermis (Appendix Fig S1), and in lanceolate endings encircling hairs (Appendix Fig S1). Similarly in spinal cord sections, Ret-eGFP was widely expressed across the dorsal horn. This was evident as a dense plexus of expression in lamina II o corresponding to IB4positive non-peptidergic nociceptors (Appendix Fig S2) and more diffusely through laminae III to V overlapping with NF200-labeled mechanoreceptor inputs (Appendix Fig S2). Notably, we also detected Ret-eGFP expression immediately ventral to IB4-positive terminals that coincided with PKCc, a marker for lamina II i /III interneurons (Appendix Fig S2), and in lamina I co-expressed with CGRP. Thus, RET-eGFP expression distinguishes multiple populations of peripheral sensory neuron that are likely to be functionally distinct.
To obtain quantitative data on the distribution of Ret-positive neuronal populations in DRG, we applied flow cytometric analysis to acutely dissociated neurons. We focused on levels of IB4 binding and native Ret-eGFP fluorescence as this would allow for quantitative measurements in live cells. In line with histological data, we observed both IB4-positive and IB4-negative populations of Ret-eGFP neuron (Figs 2A and EV1). Importantly, however, flow cytometric analysis revealed multiple well-defined subpopulations delineated by their levels of eGFP fluorescence and IB4 binding (termed Ret-eGFP Lo :IB4 Neg (A), Ret-eGFP Hi :IB4 Neg (B), Ret-eGFP Lo :IB4 Lo (C), Ret-eGFP Hi :IB4 Lo (D), and Ret-eGFP Hi :IB4 Hi (E)). There was a broad distribution in cell size across all populations as determined by forward scatter values, and no correlation between size and levels of IB4 binding or fluorescence (Fig 2B). We further validated flow cytometric analysis using fluorescent microscopy and observed that in both sensory neuron cultures ( Fig 2C) and sections of DRG ( Fig 2D) from Avil-Cre::Ret eGFP/+ mice, native eGFP fluorescence varied by an order of magnitude across IB4-positive and IB4-negative cells.

Transcription profiling of Ret-eGFP-and IB4-negative neurons
We reasoned that by defining the molecular composition of Ret-eGFP populations, we may be able to gain clues as to their function. We focused on IB4-negative neurons because flow cytometric data indicated that these cells formed two well-defined and potentially homogeneous populations. Moreover, while one of these populations presumably corresponds to RA mechanoreceptors [25,26], the other may reflect an as yet uncharacterized population of Ret-positive neurons. We performed differential microarray screening on sorted Ret-eGFP Hi :IB4 Neg and Ret-eGFP Lo :IB4 Neg cells and determined that these two populations do indeed cluster into distinct, homogeneous subsets. We further identified several functional markers in each population that gave a first indication as to their identity. For example, Ret-eGFP Lo :IB4 Neg cells were enriched in transcripts for TRPV1, CGRP (Calca), and the Ret co-receptors GFRa1 and GFRa3 (Fig 3A), while Ret-eGFP Hi :IB4 Neg neurons expressed significantly higher levels of Ret, GFRa2, and the mechanosensitive ion channel Piezo 2 (Fam38b) (Fig 3A). Intriguingly, neither population was enriched for TH. We thus speculate that Ret-eGFP Hi :IB4 Neg neurons may correspond to RA mechanoreceptors, while Ret-eGFP Lo :IB4 Neg cells represent a novel, functionally distinct subtype of nociceptor. We sought to define the function of the novel Ret-eGFP Lo :IB4 Neg population and observed that this subset was highly enriched in molecules previously implicated in itch ( Fig 3A, red dots). For example, transcripts for histamine-dependent itch mediators HRH1, PLCb3, and TRPV1 were all differentially expressed in Ret-eGFP Lo : IB4 Neg neurons, as were itch-associated neurotransmitters such as Nppb [3] and neuromedin B (Nmb) [33], and the co-receptors for IL-31, Il31ra, and Osmr. However, other molecules implicated in itch such as GRP, MrgprC11, and endothelin receptors were not over-expressed in Ret-eGFP Lo :IB4 Neg neurons, suggesting that these cells may contribute to a subtype of itch receptor.
To further investigate whether itch-associated receptor transcripts were indeed specifically expressed in Ret-eGFP Lo :IB4 Neg neuron, we performed a second microarray screen where we assessed differential expression in this population with respect to all DRG neurons. We sorted DRG neurons from Avil-Cre::R26 tdRFP mice in which the majority of peripheral sensory neurons are marked by RFP fluorescence (Appendix Fig S3) and subjected them to microarray analysis. Similar to differential screening between Ret-eGFP Lo :IB4 Neg and Ret-eGFP Hi :IB4 Neg populations, this dataset was also enriched in itch-associated molecules (Fig 3B). Fold change levels were lower than for comparisons with Ret-eGFP Hi : IB4 Neg neurons (Fig 3C), presumably reflecting the mixed molecular profile of all DRG neurons, and the fact that some transcripts such as Calca (CGRP) mark large populations of Ret-negative neurons. A Flow cytometric analysis of dissociated sensory neurons plotted according to their level of endogenous Ret-eGFP expression and IB4 binding (also indicated in Fig EV1). There are 5 well-defined subsets, 2 of which do not bind to IB4 but display a differential level of eGFP intensity (termed Ret-eGFP Lo :IB4 Neg , A, and Ret-eGFP Hi :IB4 Neg , B, respectively). The other 3 subsets bind IB4 and display a range of eGFP intensities, termed Ret-eGFP Lo  To validate the microarray analysis, we performed parallel quantitative RT-PCR analysis of 38 transcripts in Ret-eGFP Lo :IB4 Neg and Ret-eGFP Hi :IB4 Neg populations using a microfluidic platform (Fluidigm). We selected genes that represented not only itch-associated transcripts but also ion channels, signaling molecules, and transcription factors with a demonstrated role in the peripheral nervous system. In agreement with microarray data, high differential expression between Ret-eGFP Lo :IB4 Neg and Ret-eGFP Hi :IB4 Neg neurons was evident for itch-related genes such as the neuropeptide Nppb, the membrane receptors Il31ra, Osmr, Cysltr2, MrgprA3, and Hrh1, and the signaling molecule Plcb3. We also observed higher expression in this population for other markers such as Ntrk1 (TrkA), Calca (CGRP), and Trpv1 transcripts, marking it as a novel population of Ret-positive neurons, as well as the ion channels Trpm6, Trpc6, Trpm2, and P2rx3 and the serotonin receptors Htr1f and Htr1a ( Fig 3D and Appendix Fig S4). In the Ret-eGFP Hi :IB4 Neg population, we detected almost 10-fold higher expression of Ret and eGFP, validating the flow cytometry analysis. In addition, Gfra2 and the mechanosensitive ion channel Piezo 2 (Fam38b) (but not Piezo 1 (Fam38a)) were upregulated, supporting the assumption that these neurons function as RA mechanoreceptors. Intriguingly, we also observed that this population was enriched in transcripts involved in chloride transport including the GABA channel subunits a1 and c2 (Gabra1, Gabrg2) and the putative chloride channel anoctamin 6 (Ano6) (Fig 3D and Appendix Fig S4).

Sst Cre as a surrogate marker for Ret-eGFP Lo :IB4 Neg neurons
In order to investigate the function of Ret-eGFP Lo :IB4 Neg neurons, we required a molecular tool with which to selectively manipulate this population. From the microarray and quantitative RT-PCR analysis, the neuropeptide somatostatin (Sst) was among the highest enriched transcripts in Ret-eGFP Lo :IB4 Neg cells, displaying a 1,126-fold higher expression compared to the Ret-eGFP Hi :IB4 Neg population. Moreover, an Sst Cre driver line is available which has been well characterized in the CNS [34]. We thus examined the expression pattern of Sst Cre -mediated recombination in the PNS to determine whether it coincides with the Ret-eGFP Lo :IB4 Neg population and could be used to genetically target these cells.
We first assessed the anatomical properties of Sst Cre -positive neurons by crossing the Sst Cre driver line with Ret eGFP mice to generate heterozygote Sst-Cre::Ret eGFP/+ mice. In DRG sections from these animals, we detected weak eGFP fluorescence in a sparse population of cells corresponding to 1.3% of all neurons (Fig 4I). We explored this expression pattern in more detail by co-staining sections with IB4, NF200, and TH (Fig 4A-H). Sst Cre -driven Ret-eGFP expression was most evident in neurons not expressing any of these markers (85% of all Ret-eGFP-positive cells, Fig 4J), suggesting that Sst Cre may indeed mediate recombination in the Ret-eGFP Lo : IB4 Neg population.
To investigate whether Sst Cre is selective for Ret-eGFP Lo :IB4 Neg neurons, we crossed mice with a ubiquitous RFP reporter driven from the Rosa26 locus (Sst-Cre::Rosa26 RFP mice) and performed immunohistochemistry for neuronal markers on DRG sections. We again observed a low number of RFP-positive cells (1.8% of total neurons, Fig EV2) that were mostly negative for IB4 and NF200 (87% of all Rosa26 RFP -positive cells, Fig EV2). These values were not significantly different from the number of Sst-Cre::Ret eGFPpositive neurons (P = 0.36 for total cells and P = 0.7 for marker negative cells) implying that Sst Cre does not drive substantial recombination beyond the Ret-eGFP Lo :IB4 Neg population.
We further examined the peripheral projections of Sst Cre -positive sensory neurons using a Rosa26 SNAPCaaX reporter mouse line that allows for highly sensitive detection of Cre-positive cells [35]. We labeled skin from Sst-Cre::Rosa26 SNAPCaax mice with fluorescent TMR-Star SNAP substrate and observed prominent fluorescence mainly confined to neurons that formed free nerve endings and ran parallel to the dermal/epidermal border of the skin (Fig 4K-M). In occasional sections, we also detected a rare population of hair follicles that were innervated by Sst Cre -positive neurons (1.9 AE 1% of hair follicles in 3 out of 13 sections, Fig EV2). Importantly, this staining pattern was absent from skin taken from control mice not expressing Sst Cre .
To obtain a more quantitative assessment of the degree of intersection between Sst Cre -positive neurons and the Ret-eGFP Lo :IB4 Neg population, we applied flow cytometric analysis to acutely dissociated neurons from Sst-Cre::Ret-eGFP mice. We followed an identical preparation protocol to that performed previously for Avil-Cre:: Ret eGFP mice and used the same gates for analysis. In line with histological data, we observed very few IB4-positive cells among the Sst-Cre::Ret eGFP population, and furthermore, native GFP fluorescence was low (Fig 5A). Indeed, when comparing this data directly to flow cytometric analysis of Avil-Cre::Ret eGFP neurons (Fig 5B), it was apparent that four out of the five subpopulations defined in earlier experiments were absent in Sst Cre mice and only the Ret-eGFP Lo : IB4 Neg was present.
Finally, we sought to validate the identity of Sst Cre -positive sensory neurons by assessing their molecular composition compared to Ret-eGFP Lo :IB4 Neg cells. We sorted Sst-Cre::Ret eGFP cells using the same gating paradigm as described previously and IB4 Neg against probability. Ret has higher expression in the Ret-eGFP Hi :IB4 Neg subset, which also shows an upregulation of the Ret co-receptor Gfra2 and Fam38b, encoding for the mechanosensitive ion channel Piezo 2. The Ret-eGFP Lo :IB4 Neg subset displays an array of molecules previously associated with itch (marked in red). B Itch-associated transcripts are enriched in the Ret-eGFP Lo :IB4 Neg population compared to all DRG neurons. Volcano plot of fold change expression in Ret-eGFP Lo : IB4 Neg neurons versus sorted neurons from Avil-Cre::R26 tdRFP mice against probability. Molecules linked to itch perception are significantly upregulated in the Ret-eGFP Lo :IB4 Neg subset. C Gene expression within the Ret-eGFP Lo :IB4 Neg population is confirmed by triple comparison between Ret-eGFP Lo :IB4 Neg , Ret-eGFP Hi :IB4 Neg , and Avil-Cre::R26 tdRFP datasets. D Validation of differential microarray screening by quantitative RT-PCR. Transcripts encoding ion channels, receptors, neuropeptides, signaling molecules, and transcription factors were selected for quantitative RT-PCR. Differential expression between Ret-eGFP Lo :IB4 Neg and Ret-eGFP Hi :IB4 Neg populations correlates with microarray analysis (n = 3).    Fig S5). Furthermore, levels of enrichment for both populations were remarkably similar, especially for the markers eGFP and Ret, and for itch-associated transcripts such as Hrh1, Il31ra, Osmr, Cysltr2, Plcb3, and Nppb (Fig 5C and Appendix Fig S5). Collectively, these data imply that Sst Cre does indeed mark the Ret-eGFP Lo :IB4 Neg population and that this population is largely homogenous.

Functional characterization of Sst Cre neurons
To examine the function of Sst Cre -positive neurons, we reasoned that selective ablation of this population could uncover their role in driving distinct somatosensory behaviors. A well-described and effective method for selectively ablating neurons in vivo involves Cre-dependent expression of the diphtheria toxin receptor from the Rosa26 locus and subsequent treatment of animals with diphtheria toxin [36]. However, because Sst Cre is expressed widely in the central nervous system [34], this approach would not be suitable for deleting only Sst Cre -positive neurons in the peripheral nervous system. We thus generated a new mouse line where diphtheria toxin receptor is integrated into the sensory neuron-specific Avil locus preceded by a loxP-flanked stop cassette. Crossing these mice with a Cre driver line should therefore induce expression of the diphtheria toxin receptor only in peripheral sensory neurons, regardless of whether Cre is expressed in other tissues. Avil iDTR mice were generated using standard techniques for knock-in to the Avil locus (Fig EV3). Mice were healthy, displayed normal fertility, and did not exhibit any obvious defects. To assess Sst Cre -mediated recombination of the Avil iDTR transgene, Sst Cre mice were crossed with Avil iDTR animals to produce heterozygote Sst-Cre::Avil iDTR/+ mice, and expression of the diphtheria toxin receptor determined using immunohistochemistry. In control Avil iDTR mice (without the Cre), we detected no diphtheria toxin receptor expression. However, in mice with the Sst Cre allele, a small number of diphtheria toxin receptor-positive cells were evident in DRG that was also negative for IB4 (Fig 6A-C and G). We next investigated the efficiency of ablation of these cells by applying diphtheria toxin systemically in mice and evaluating the number of diphtheria toxin receptor cells using immunohistochemistry. We observed an almost complete loss of diphtheria toxin receptor immunoreactivity after toxin application (Fig 6D-F). To determine whether ablation also impacts upon endogenous Sst expression and does not affect other populations, we performed immunocytochemistry on dissociated DRG neurons plated on glass coverslips and labeled with Sst and NF200 antibodies, and IB4. Sst-Cre::Avil iDTR/+ mice treated with diphtheria toxin displayed a complete loss of Sst immunoreactivity with no change in the number of NF200-or IB4-positive neurons ( Fig EV4). Finally, we carried out quantitative RT-PCR on Sst and other transcripts associated with the Ret-eGFP Lo :IB4 Neg population (Hrh1, MrgprA3, Il31ra, and Htr1f). All transcripts were strongly downregulated upon diphtheria toxin-mediated ablation (Fig EV4), indicating that this strategy is an effective means of eliminating the Sst Cre population in vivo.
To explore the function of Sst Cre -positive cells, we initially performed calcium imaging on acutely dissociated DRG and investigated their activation by pruritogens. Based upon transcriptome analysis, we selected four receptor classes highly enriched in this population encompassing histamine-dependent (histamine) and histamine-independent signaling (chloroquine and IL-31), and a class of receptor, 5HT 1F not previously implicated in itch, but which couples to the inhibition of adenylate cyclase [37] and may therefore inhibit neuronal activity. Experiments were carried out on DRG from control Avil iDTR mice and Sst-Cre::Avil iDTR mice, both treated systemically with diphtheria toxin. Application of histamine to cells evoked a characteristic phasic calcium response in approximately 10% of neurons from control mice, which upon ablation was reduced to less than 1% (Fig 6H and L). Chloroquine and IL-31 activated a similar number of neurons in control mice with a corresponding decrease in the number of responsive cells in DRG from Sst-Cre::Avil iDTR mice (Fig 6I, J and L). Finally, we examined the function of the 5HT 1F receptor by applying the 5HT 1F agonist LY344864 to neurons. Unexpectedly, we found that application of low concentrations of LY344864 evoked calcium flux in a small population of neurons that was also reduced upon ablation of Sst Cre neurons (Fig 6K and L). In all cases, responses to pruritogens which remained after ablation were quantitatively similar to those from control cultures (Fig EV4).
The Avil iDTR strategy opens up the possibility to determine how ablation of the Sst Cre population impacts upon mouse behavior, and thus translate in vitro observations in DRG to events occurring in sensory neuron endings in the skin. To explore this, we first examined nociceptive thresholds in control Avil iDTR and Sst-Cre::Avil iDTR mice treated with diphtheria toxin. We observed no significant difference in thermal withdrawal thresholds in the hot plate test (P = 0.17, Fig 7A) or mechanical thresholds to calibrated von Frey filaments (P = 0.32, Fig 7B). We next considered responses to pruritogens by quantifying scratching behavior after intradermal injection of agonists tested in isolated DRG. We first assayed responses in control (Avil iDTR ) and Sst-Cre::Avil iDTR mice in the absence of diphtheria toxin. All compounds including the 5HT 1F agonist LY344864 evoked robust scratching behavior compared to vehicle (Figs 7C-F and EV5). We then applied diphtheria toxin systemically to these mice and reexamined scratching responses. Strikingly, we observed a significant reduction in scratching behavior in Sst-Cre:: Avil iDTR mice evoked by IL-31 and LY344864, but not by histamine or chloroquine (Figs 7C-F and EV5). Responses in control Avil iDTR mice were not influenced by diphtheria toxin application, indicating that Sst Cre neurons (and thus the Ret-eGFP Lo :IB4 Neg population) mediate itch sensation in vivo.

Discussion
Here, we show that a previously uncharacterized population of sensory neurons that express the receptor tyrosine kinase Ret function as itch receptors. Using flow cytometry of dissociated DRG neurons, we isolated this population based upon its weak expression of Ret and absence of IB4 binding. We demonstrate that these neurons express markers such as TrkA and Sst, and importantly are highly enriched for transcripts associated with itch sensation. Anatomical and functional analysis of these cells indicates that they form a rare subtype of DRG neuron and function to detect itch inducing stimuli.
Three populations of Ret-positive sensory neuron have been previously identified and characterized. Early Ret neurons develop

Heat withdrawal
Mechanical withdrawal (9) (6) (9) (6) No DTX DTX  into RA mechanoreceptors [25,26], while late Ret neurons become C-fiber low threshold mechanoreceptors and non-peptidergic nociceptors [22,28]. Intriguingly, a fourth population of Ret-positive neurons has also been described that expresses TrkA in adult mice but has unknown function [22,30]. This population is scarce, corresponding to around 10% of all Ret-positive cells [22,30], and presumably this rarity has made further analysis difficult. By taking advantage of the sensitivity and analytic power of flow cytometry, we have isolated these cells and performed gene expression profiling to determine their function. We initially used histochemistry to identify populations of Ret-eGFP-positive neuron expressing markers of mechanoreceptors (NF200), non-peptidergic nociceptors (IB4), and C-fiber low threshold mechanoreceptors (TH). However, limited by the number of markers that can be applied simultaneously to the same section, we turned to flow cytometry to analyze Ret expression in more detail. We based our analysis on the observation that Ret expression levels differ considerably across the DRG and that low Ret levels correlate with GFRa1/3 expression [24], while high Ret-expressing neurons express GFRa2 and form RA mechanoreceptors [25,26]. By co-labeling cells with IB4, we were able to distinguish multiple subtypes of Ret-positive neurons of which a Ret-low, IB4-negative population was clearly evident. That this population corresponds to the fourth, uncharacterized type of Ret neuron is supported by several lines of evidence. Firstly, it is IB4 negative, indicating that it is not part of the non-peptidergic nociceptor population. Secondly, it expresses almost 10-fold less Ret transcript compared to Ret-eGFP Hi : IB4 Neg neurons. Thirdly, Gfra3 is upregulated 29-fold in these cells, while Gfra2 has 21-fold higher expression in the Ret-eGFP Hi :IB4 Neg population. Finally, we detected many differentially expressed genes such as TrkA, CGRP, and TRPV1, in the Ret-eGFP Lo :IB4 Neg population which are known to be absent from C-LTMRs or RA mechanoreceptors [28]. While we did not perform a similar analysis on IB4-positive neurons, flow cytometry data indicate that further three populations can also be differentiated here which may also form distinct functional populations.
Given the powerful regulatory actions of neurotrophic factors in the peripheral nervous system, a major emphasis of previous research has been to investigate how deletion of these factors impacts upon the survival, differentiation, and maintenance of distinct populations. Here, we have not considered the effects of Ret ablation on Ret-eGFP Lo :IB4 Neg -negative neurons, concentrating instead upon the initial functional characterization of this population. Of note, the number of Ret-positive/TrkA-positive neurons is not reduced in Ret nociceptor-specific conditional knockout mice [30], suggesting that this population does not require Ret for its survival. It is, however, possible that the termination pattern of these neurons in the skin and spinal cord, and the expression of ion channels and signaling molecules are regulated by Ret signaling as has been shown for other populations [22,[24][25][26][27][28]30,38]. The identification of markers and molecular tools with which to target Ret-eGFP Lo :IB4 Neg -negative cells described here would now allow for further investigation of these phenotypes.
Among the differentially expressed genes present in Ret-eGFP Lo : IB4 Neg cells, we observed a strong enrichment of the neuropeptides CGRP, Nppb, neurotensin (Nts), and somatostatin (Sst). Of these, we focused our attention on Sst because of the availability of an Sst-Cre driver line [34] which we reasoned could be used to target the Ret-eGFP Lo :IB4 Neg population. Indeed, recombination of the Ret or Rosa26 locus by Sst-Cre occurred in very few neurons of the DRG, resulting in fluorescence in mainly IB4-and NF200-negative cells, analogous to the Ret-eGFP Lo :IB4 Neg population. Moreover, flow cytometry analysis and transcriptional profiling revealed that these neurons overlap. Of note, Sst-Cre-mediated recombination of eGFP and RFP reporters resulted in a lower number of positive cells when compared to the Ret-eGFP Lo :IB4 Neg population, or to antibody staining of Sst. We suggest that this may be due to the lower sensitivity of reporter fluorophores [35]. Indeed, in ablation experiments which require minimal expression of the diphtheria toxin receptor to induce cell death, we observed around 10% reduction in the number of cells responding to pruritogens such as histamine.
To define the function of the Sst-Cre/Ret-eGFP Lo :IB4 Neg population, we generated a new mouse line which allows for selective ablation of Cre-positive sensory neurons regardless of whether Cre recombinase is expressed in other tissues beyond the DRG. Using this mouse, we were able to selectively ablate Sst-positive neurons and found that both cellular and behavioral responses to pruritogens but not nociceptive behavior were reduced upon ablation. Intriguingly, we also observed several differences between in vitro studies on isolated DRG and in vivo studies examining scratching behavior. For example, histamine-and chloroquine-evoked responses were sensitive to diphtheria toxin ablation in isolated DRG, but were not affected at the whole animal level. One possible explanation for this discrepancy is that ablation of Sst-Cre-positive neurons does indeed eliminate those neurons which respond directly to the tested pruritogens. However, in vivo, alternative pathways involving other cell types may be evoked (especially by histamine and chloroquine) which in turn activate a broader spectrum of neurons to trigger scratching. It was also unexpected that scratching behavior was only partially reduced in vivo. Indeed, knockout mice for Nbbp show an almost complete loss of itch-evoked behavior [3], and Nbbp was highly upregulated in Ret-eGFP Lo :IB4 Neg neurons and is co-expressed with Sst [29,30]. Since our data indicate that Sst neurons are fully ablated upon treatment with diphtheria toxin (Fig EV4), this would suggest that Nbbp is also expressed outside of the Sst population. Thus, the Avil iDTR mouse line may be a useful tool for establishing functional overlap between different populations of sensory neuron.
We were intrigued by the high expression levels of the serotonin receptors Htr1f and Htr1a in Ret-eGFP Lo :IB4 Neg neurons. Serotonin has a well-established pruritogenic action when applied subcutaneously and this effect is mediated by 5-HT 2 receptors [40,41]. Transcripts for 5-HT 2 receptors were not upregulated in Ret-eGFP Lo : IB4 Neg cells and we did not investigate further whether serotonin directly activates these cells. Instead, we focused on the 5HT 1F receptor because it is coupled to G i signaling and should therefore inhibit neuronal activation. Unexpectedly, we observed an apparent activation of Ret-eGFP/IB4-negative sensory neurons by the 5HT 1F agonist LY344864 and robust scratching behavior when LY344864 was injected intradermally in control animals. A potential basis for this effect may come from the observation that many G i -coupled receptors are able to interact with G 14 subunits promoting the stimulation of phospholipase C [42]. Indeed, the transcript for G 14 (Gna14) was among the most differentially expressed genes in Ret-eGFP Lo :IB4 Neg neurons (Fig 5A-C), and thus, its presence could expand the signaling repertoire of these cells allowing for G i -coupled The Authors receptors such as 5HT 1F to generate neuronal activity. Of note, 5HT 1F receptor agonists have also been proposed as clinical targets for the treatment of migraine [43]. Our data indicate that these compounds could have unwanted side effects via activation of Sst-Cre/Ret-eGFP Lo :IB4 Neg neurons and generation of itch.
In addition to the four pruritogen receptors examined here, many molecules previously implicated in itch sensation such as Cysltr2, Plcb3, and Nppb were enriched in the Ret-eGFP Lo :IB4 Neg population. However, other itch-associated transcripts were absent from the transcriptome data including GRP, MrgprC11, endothelin receptors, and TSLP receptor transcripts (Il7r and Crlf2) which play a critical role in the development of atopic dermatitis [44]. These data support the recent finding that several populations of putative itch receptors exist in the mouse which may be tuned to selectively detect the multitude of irritants which can evoke scratching behavior [20]. Our data on the expression profile of Ret-eGFP Lo :IB4 Neg neurons provide a molecular inventory for one of these populations, and as such many more molecules with an as yet unexplored role in itch may be contained within this list. Moreover, this complexity suggests that therapeutic strategies for treating itch should target common signaling molecules or disrupt neuronal function, rather than inhibit specific receptors. The transcriptional profiling approach described here is a first step toward identifying these molecules.
To ablate Sst-positive sensory neurons, we crossed the Sst Cre line with an Avil iDTR line, thus obtaining Sst-Cre::Avil iDTR/+ mice that express the diphtheria toxin receptor specifically in Sst-positive DRG neurons.
The Avil iDTR line was obtained by classical knock-in of the iDTR cassette Lox-STOP-Lox-DTR (kindly provided by the Waisman group in Mainz) into the Avil locus, to replace exon 2 and 66 bp upstream thereof (Fig EV3). The targeting construct was transfected into A9 ES cells. Individual ES cell clones underwent Southern blot screening to identify homologous recombinants. DNA was digested with PstI and HindIII and hybridized with 5 0 or 3 0 probe, respectively, obtaining 9,600 (wild-type)-and 6,300 (targeted)-bp DNA fragments by using the 5 0 probe and 7,100 (wild-type)-and 5,600 (targeted)-bp DNA fragments by using the 3 0 probe. Positive clones were injected into 8-cell stage embryos to generate mice heterozygous for the targeted allele. To assess mouse genotype, PCR was performed with the following primers: aggagcgagggctcagttgggctgt (forward) and acaccaggttagcctttaagcctgc (reverse) for the wild-type and ctgccacccaggttaccatggagagagg (forward) and cattctagttgtggtttgtc caaactca (reverse) for the mutant allele, giving PCR products of 313 and 287 bp, respectively. Heterozygous mice were all viable and exhibited no overt behavioral phenotype.
All mice were bred and maintained at the EMBL Mouse Biology Unit, Monterotondo, in accordance with Italian legislation (Art. 9, 27. Jan 1992, no 116) under license from the Italian Ministry of Health, and in compliance with the ARRIVE guidelines. The Avil iDTR conditional allele has been deposited in the European Mutant Mouse Archive (www.emmanet.org).
Immunofluorescence DRG and spinal cord from adult mice (6 weeks or older) were postfixed in 4% PFA for 30 min (DRG) or 2 h (SC), embedded in 12% bovine gelatin, and sectioned using a vibratome at 50-100 lm. Sections were incubated for 30 min in 50% ethanol and then overnight in 0.3% Triton X-100, and 5% goat or donkey serum in PBS at 4°C containing one or more antibodies diluted as shown below. Secondary antibodies and streptavidin-647 were diluted (1:1,000 and 1:600, respectively) in 0.3% Triton X-100, 5% goat or donkey serum in PBS and left for 1-2 h at 4°C. Slides were mounted with Prolong gold antifade (Invitrogen, P36930).
For immunofluorescence analysis of the Avil-Cre::RET eGFP/+ mouse skin, the hair was removed and skin collected and postfixed in 4% PFA for 8-12 h, incubated in 30% sucrose overnight, and frozen in OCT. 35-lm sections were cut with a cryostat. Antibody staining was performed as described above.
For SNAP-tag analysis of the Sst-Cre::Rosa26 SnapCaaX/+ mouse peripheral nerve endings, the back skin of the animals was injected intradermally with 10 lM of Snap-Cell TMR-Star (New England Biolabs, S9105S). After 6 h, the skin was collected and treated as described above. Once sectioned, the skin was additionally stained with DAPI (Invitrogen, D1306).
To visualize the cells expressing the diphtheria toxin receptor, we used goat anti-human HB-EGF antibody (R&D Systems, AF-259-NA).
All images were visualized with a Leica SP5 confocal microscope and analyzed with ImageJ.
To analyze the distribution of Ret-and Sst-positive sensory neurons, we used flow cytometry on Sst-Cre::Ret eGFP/+ animals. Sensory neurons from 10 mice were collected, isolated, and stained following the same protocol used for the analysis of Avil-Cre:: RET eGFP/+ mice.
Flow cytometric online and offline analyses were performed in a FACS Aria III SORP (BD Bioscience) using FACS Diva software (BD Bioscience) and FlowJo (Tree Star, Inc.), respectively. For gene expression analysis, DRG cells were sorted (see Fig EV1) using an 85-lm nozzle (40 PSI) directly into RLT buffer (RNeasy kit, Qiagen) and RNA was purified according to manufacturer instructions. Two rounds of RNA amplification, labeling, and hybridization to Affymetrix â MOE430 2.0 GeneChip â expression arrays were performed by the EMBL Gene Core Facility. Microarray data were analyzed as detailed in the Appendix. All data have been submitted to ArrayExpress with the accession code E-MTAB-1836.

Fluidigm
To validate the microarray data, we used the microfluidic platform Fluidigm to perform quantitative RT-PCR. DRG from Avil-Cre:: RET eGFP/+ mice were collected and processed as described above. Cells were sorted directly into 5 ll of CellsDirect (Invitrogen, PN 11753-100 and 11753-500) 2× Buffer, and combined with 0.2 ll of RT/Taq mix, 1 ll of assay pool (formed by 38 individual assay pairs that were previously pooled and diluted to a final concentration of 500 nM), and 2.8 ll of water. The samples went through targetspecific RT and pre-amplification (16 cycles) following manufacturer instructions. Pre-amplified cDNA was then diluted 1:10 and combined together with 2.5 ll 2× SsoFast Evagreen Supermix (Bio-Rad, PN 172-5211) and 0.25 ll 20× Sample Loading Reagent (Fluidigm, PN 100-0388) and loaded onto the sample inlets of a 48.48 Fluidigm Dynamic Array. All 38 individual assays pairs (100 lM) were diluted by combining 0.25 ll of each assay with 2.25 ll of TE Buffer (TEKnova, PN T0224) and 2.5 ll of Assay Loading Reagent (Fluidigm, PN 85000736). The 48.48 chip was then loaded on the IFC Controller MX and 30 cycles of qPCR were performed on the Fluidigm BioMark HD system. Data were analyzed using Fluidigm Real-Time Analysis Software.
To compare the level of expression of the 38 genes between Ret-eGFP Lo :IB4 Neg -and Sst-Cre::Ret eGFP -positive sensory neurons, we used Fluidigm platform to perform qRT-PCR on DRG neurons sorted from Sst-Cre::Ret eGFP/+ animals. We followed the same procedure described above.
qRT-PCR DRG were dissected from adult Sst-Cre::Avil iDTR/+ and control Avil iDTR/+ (without the Cre) mice treated with 40 lg/kg of diphtheria toxin (2 injections, the second injection occurring 72 h after the first one). Mice were sacrificed 8 days after the second DTX injection. RNA was isolated using Trizol and reverse-transcribed using Superscript II (Invitrogen). Quantitative RT-PCR was then performed with LightCycler 480 SYBR Green I Master kit (Roche, 04707516001) following manufacturer instructions and using Roche LightCycler 480 instrument. The expression of each gene was calculated relatively to ubiquitin using the 2 ÀDDCt method.

Behavior
To evaluate the scratching response of animals prior and after ablation of the Sst-positive sensory neurons, we removed the hair of the neck of 6-to 8-week-old Sst-Cre::Avil iDTR/+ and control Avil iDTR/+ mice with an electric shaver. The animals were acclimatized for 2-3 days and the following day they were injected intradermally with 50 ll solution in the nape of the neck with one of the following reagents: 10 mM histamine (Sigma, H7250) in saline, 1 mM LY344864 (Abcam, ab120592) in saline, 5.6 lM IL-31 (Peprotech, 210-31) in 0.1% BSA/PBS, 12.5 mM chloroquine (Sigma, C 6628) in saline. Control injections were performed with appropriate vehicle. Animals were observed for 60 min and bouts of scratching counted. One bout was defined as an event of scratching lasting from when the animal lifted the hind paw to scratch until it returned it to the floor or started licking it. The actual time of scratching was also assessed, considered as the time from when the animal started scratching until it put its hind paw down.
The same animals were then injected i.p. with 40 lg/kg of DTX (Sigma D0564). After 72 h, a second injection of DTX was performed. After 5 days, mice were again acclimatized for 2-3 days, and at the 8 th day after the second DTX injection, scratching behavior was assessed again, using the same reagent of the first test, at the same concentration.
For the hot plate and the von Frey tests, Sst-Cre::Avil iDTR/+ and Avil iDTR/+ (without the Cre) mice were injected twice with 40 lg/kg of DTX, with the second injection occurring 72 h after the first one. The tests were performed 8 days after the second DTX injection, after 2-3 days of acclimatization.
For the hot plate test, each animal was placed on the hot plate (Ugo Basile) and left to acclimatize for 30 min. A gradient from 42 to 49°C was applied and stopped as soon as a paw withdrawal or licking was observed. This was repeated for 2-3 days. On the day of the actual test, animals were placed again on the hot plate and the gradient was applied: The temperature at which a paw withdrawal or licking was observed was registered as individual threshold for thermal nociception.
For the von Frey test, animals were acclimatized by placing them in cages on a mesh platform and by applying von Frey filaments (North Coast Medical) on the plantar surface of the hind paw. The actual test was performed following the simplified up-down method proposed by Bonin et al [47].
Sample sizes were determined from the power of the statistical test performed. No animals were excluded and all experiments were performed blinded with order of testing randomized.
Expanded View for this article is available online.