c-Maf-positive spinal cord neurons are critical elements of a dorsal horn circuit for mechanical hypersensitivity in neuropathy

Summary Corticospinal tract (CST) neurons innervate the deep spinal dorsal horn to sustain chronic neuropathic pain. The majority of neurons targeted by the CST are interneurons expressing the transcription factor c-Maf. Here, we used intersectional genetics to decipher the function of these neurons in dorsal horn sensory circuits. We find that excitatory c-Maf (c-MafEX) neurons receive sensory input mainly from myelinated fibers and target deep dorsal horn parabrachial projection neurons and superficial dorsal horn neurons, thereby connecting non-nociceptive input to nociceptive output structures. Silencing c-MafEX neurons has little effect in healthy mice but alleviates mechanical hypersensitivity in neuropathic mice. c-MafEX neurons also receive input from inhibitory c-Maf and parvalbumin neurons, and compromising inhibition by these neurons caused mechanical hypersensitivity and spontaneous aversive behaviors reminiscent of c-MafEX neuron activation. Our study identifies c-MafEX neurons as normally silent second-order nociceptors that become engaged in pathological pain signaling upon loss of inhibitory control.

In brief Frezel et al. use intersectional genetics to identify a population of deep dorsal horn excitatory neurons that act as silent second-order nociceptors that are only recruited into pain pathways after nerve injury. They can engage nociceptive pathways via the superficial dorsal horn or via deep dorsal horn projection neurons.

INTRODUCTION
Patients suffering from chronic pain not only display increased sensitivity to noxious stimuli but often also perceive innocuous stimuli (e.g., touch) as painful. This phenomenon is known as allodynia. A wide variety of alterations potentially contributing to allodynia have been proposed, ranging from changes in peripheral neurons and spinal neurons to changes at supraspinal sites. [1][2][3][4][5] Recently, it has been described that the corticospinal tract (CST), which in naive mice is regarded as an important element for top-down control of voluntary movement, 6,7 critically contributes to mechanical allodynia in nerve-injury-induced chronic pain states. 8 In these conditions innocuous low-threshold afferent input is thought to gain access to superficial dorsal horn nociceptive specific circuits via polysynaptic pathways, thus producing touch-evoked allodynia. [9][10][11][12] Liu et al. 8 suggested that CST neurons located in the somatosensory cortex (S1) synapse onto neurons in the deep dorsal spinal horn, an area also termed ''low-threshold mechanoreceptor (LTMR) recipient zone,'' as it receives input predominantly from myelinated sensory afferents conveying innocuous information about touch and proprioception. 7,13 Taken together, these findings suggest that deep dorsal horn neurons that integrate low-threshold primary afferent input and descending information from S1 are critical components of the neural circuitry that controls mechanical pain perception after nerve injury. While several deep dorsal horn interneuron populations have been identified that receive corticospinal and low-threshold peripheral input, 13 it is only incompletely understood which of them are required for mechanically evoked pain after nerve injury.
We have previously identified dorsal horn neurons expressing the transcription factor c-Maf as a main target population of CST neurons in S1. 14 Here, we have employed intersectional virusbased strategies for circuit tracing and functional manipulation to identify excitatory c-Maf neurons as critical elements of a spinal circuit involved in the generation of nerve-injury-induced mechanical allodynia.

c-Maf is expressed in subsets of deep dorsal horn excitatory and inhibitory interneurons
To address the role of c-Maf neurons in dorsal horn neural circuits, we generated a c-Maf Cre knockin mouse line ( Figure 1A). Eutrophic expression of Cre was verified using multiplex in situ hybridization on spinal cord sections of adult c-Maf Cre mice (Figure 1B). We found that 87.4% ± 3.9% of c-Maf + neurons co-expressed Cre mRNA and detected c-Maf mRNA in 78.8% ± 4.2% of Cre + neurons ( Figure 1C). We next analyzed the localization and molecular identity of spinal c-Maf/Cre neurons using immunohistochemistry and multiplex in situ hybridization. Neurons expressing c-Maf were present in laminae III and IV (Figure 1A), ventral to the protein kinase C g (PKCg) plexus, which delineates the border between the superficial and deep dorsal horn. 15 Only few c-Maf neurons (0.18% ± 0.1%) also expressed PKCg ( Figures 1A and 1F). To further characterize c-Maf neurons, we used c-Maf or Cre probes together with probes for a variety of previously reported marker genes for dorsal horn neurons [16][17][18][19][20] (Figures 1D-1H and S1). Consistent with a previous report, 21 we found that more than half of the c-Maf neurons were excitatory (52.9% ± 1.6% expressed VGLUT2, Figures 1D and 1G) and one-third were inhibitory (32.1% ± 0.4% expressed Pax2, 31.3% ± 1.6% expressed VGAT, and 29.1% ± 4.2% expressed Glyt2, Figures 1A, 1F, 1G, and S1). Single-cell profiling experiments 16,22 have suggested c-Maf as a marker for a molecular defined family of excitatory dorsal horn neurons present in 2 out of 15 excitatory (Glut1 and Glut2) and 3 out of 15 inhibitory (GABA11, GABA12, and GABA13) neuronal subpopulations. Consistent with the single-cell data, we found that 84% of the excitatory c-Maf neurons (44.5% ± 2.5% of all c-Maf neurons) expressed cholecystokinin (CCK) ( Figures 1E  and 1G), 80% (51.3% ± 3.8% of all c-Maf neurons) expressed RORa ( Figures 1E, 1G, and S1C), and 24.7% ± 2.8% of all c-Maf neurons expressed parvalbumin (PV) (Figures 1D, 1G, and S1A). Again, in line with single-cell data, PV was found to be expressed in both excitatory and inhibitory c-Maf neurons (7.6% ± 1.3% of the excitatory and 12.7% ± 0.6% of the inhibitory c-Maf neurons). Only few c-Maf neurons were calretinin (CR) positive (10.2% ± 2.0% of all c-Maf cells) (Figures 1G, S1B, and S1D). Vice versa, we also determined the percentages of c-Maf neurons among the CCK, RORa, CR, and PV populations and found that 27.7% ± 2.1% of CCK-positive neurons expressed c-Maf, 36.4% ± 3.2% of RORa-positive neurons, and 30.7% ± 4.6% of PV + neurons, and again only a small overlap with CR neurons was observed (4.23% ± 0.9% of CR + neurons, Figures 1H and S1D). Our data thus indicate that excitatory c-Maf neurons constitute a subset of the larger CCK population. To further verify this finding, we labeled spinal CCK cells by crossing CCK Cre mice to a reporter line (NuTRAP) that expresses GFP in a Credependent manner (labeled cells are termed CCK GFP ). In subsequent co-labeling experiments, we co-stained spinal cord sections with antibodies against GFP, c-Maf, and Pax2 ( Figure 1I). In line with our previous results, we found that 88.4% ± 2.2% of all excitatory c-Maf neurons (c-Maf + ; Pax2 À ) co-expressed GFP. Vice versa, about one-third of CCK GFP cells (31% ± 2.1%) were c-Maf positive, while almost no inhibitory c-Maf neurons (c-Maf + ; Pax2 + ) co-expressed GFP ( Figures 1I and 1J). Taken together, our findings are consistent with single-cell RNA-sequencing data 16 (http://linnarssonlab.org/dorsalhorn/) indicating c-Maf expression in a PKCg-negative subfamily of CCK neurons that represents 2 out of 15 excitatory populations of spinal dorsal horn neurons (Glut1 and Glut2).
Intersectional targeting strategies enable selective targeting of either excitatory or inhibitory c-Maf neurons As outlined above, c-Maf neurons constitute a mixed population of excitatory and inhibitory neurons, and c-Maf is also expressed in dorsal root ganglion (DRG) neurons. 23 We therefore chose intersectional strategies to selectively target the excitatory or inhibitory c-Maf neuron family in the spinal cord. To this end, we crossed either a Lmx1b Dre allele ( Figure S2) or a GlyT2:Dre transgene 24 into c-Maf Cre mice, and employed Cre and Dre double-dependent reporter transgenes delivered via recombinant adeno-associated viruses (rAAVs) (Figures 2A and 2J). Lmx1b is expressed in the vast majority of dorsal horn excitatory neurons, whereas GlyT2 is a marker gene for inhibitory neurons of the deep dorsal horn. Neither gene is expressed in DRG neurons ( Figure S2E). 25,26 To validate our intersectional strategies, we injected the left lumbar spinal cord of c-Maf Cre ;Lmx1b Dre double transgenic mice (hereafter referred to as c-Maf EX mice) and c-Maf Cre ;GlyT2:Dre mice (c-Maf IN mice) with an rAAV carrying a Cre/Dre double-dependent eGFP expression cassette (rAAV9.hEF1a.C on /D on -eGFP) (Figure 2A).
In c-Maf EX mice, the vast majority of eGFP + neurons were found in laminae III and IV of the dorsal horn and co-expressed Lmx1b, while the inhibitory marker Pax2 was virtually absent from eGFP + neurons ( Figures 2B-2F). As expected, DRG neurons were devoid of eGFP ( Figure 2D). Comparable results were obtained when reporter mice were used instead of reporter viruses ( Figure S3). Most c-Maf EX (eGFP + ) neurons were located ventral to the PKCg plexus ( Figure 2E). Only few eGFP + cells also expressed PKCg (1.47% ± 0.59% of eGFP + neurons, Figures 2E and S3F). We used calcitonin gene-related peptide (CGRP) immunostaining and isolectin B4 (IB4) binding to label laminae I/II o and II i , respectively, 27 which together comprise the termination area of most nociceptive fibers. 1,28 VGlut1 staining was used to label the LTMR recipient zone. 13,29 Most eGFP + neurons and their neuropil were located ventral to the IB4 and CGRP layers within the area of the VGlut1 axon terminals ( Figure 2F) consistent with the distribution of c-Maf immunoreactivity ( Figure 1A). In c-Maf IN mice, the vast majority of eGFP + neurons (88.7% ± 4.3%) were Pax2 + ( Figures 2H and 2I). Again, no eGFP expression was detected in DRG neurons ( Figure 2M). The localization of eGFP neurons ( Figures 2N and 2O) was comparable with that found in c-Maf EX neurons. Both c-Maf EX and c-Maf IN neurons were in the LTMR recipient zone, the area receiving low-threshold cutaneous and proprioceptive information, 13,29,30 indicating that c-Maf neurons likely receive non-nociceptive LTMR sensory input. 7,8,13,17,31,32 To further characterize the neurons targeted by our intersectional strategies, we analyzed their morphology and recorded basic biophysical parameters. Sparse labeling with an eGFP encoding replication-deficient rabies virus (see Albisetti et al. 33 ) revealed that many c-Maf EX neurons, especially in upper lamina III, displayed a vertical cell-like morphology with an apical dendrite extending toward the superficial laminae ( Figure 2G) while the morphology of many c-Maf EX neurons in deeper L III and L IV were less polarized ( Figure S4A), more resembling central cells. These observations are consistent with morphologies that have been reported previously for deep dorsal horn RORa cells or CCK cells. 10,17 Many c-Maf IN neurons could be identified as islet cells ( Figure 2P) or radial cells ( Fig S4B). Next, we used c-Maf Cre ;Lmx1b Dre /(GlyT2:Dre); Ai66 mice to identify and characterize the biophysical properties of c-Maf EX (c-Maf IN ) neurons. The majority of Maf EX neurons displayed an initial burst firing pattern (50%). About one-third (29%) showed tonic firing, 13% delayed firing, and 8% gap firing ( Figures 2H-2I). Maf IN neurons predominantly presented a tonic firing pattern (64%) while some showed initial burst (22%) or delayed (14%) firing ( Figures 2Q and 2R). Maf EX and Maf IN neurons had similar thresholds, resting membrane potentials, rheobase, action potential width, and input resistance but differed significantly in their average capacitance and after hyperpolarization ( Figure S4C-S4I).
Since our previous data suggested monosynaptic connections between CST neurons of S1 and spinal c-Maf neurons, 14 we examined eGFP expression also in supraspinal CNS areas. As expected, eGFP + cells were present in the primary somatosensory cortex (S1, Figures 3I and 3J, n = 4) but also in the red nucleus ( Figures S5E and S5G) and the rostroventral medulla (RVM, Figures S5F and S5H, n = 4), verifying that c-Maf EX and c-Maf IN neurons integrate descending supraspinal input with sensory input from different types of LTMRs and proprioceptors.
CST neurons innervating deep dorsal horn excitatory neurons have been reported to affect nociception in neuropathic mice by activating spinal excitatory CCK neurons. 8 To verify functional connections between layer V S1 neurons and spinal c-Maf neu-rons as well as to investigate potential differences between CST innervation of c-Maf EX or c-Maf IN neurons, we performed slice recordings from the respective subtype. To this end, we overexpressed ChR2-YFP in CST neurons by injecting AAV.ChR2-YFP into S1 and recorded from labeled c-Maf EX or c-Maf IN neurons in the spinal cord sections of c-Maf Cre ;Lmx1b Dre /(GlyT2:Dre);Ai66 mice after optogenetic stimulation of CST terminals. Latency and jitter recorded from c-Maf EX neurons were significantly higher than those recorded from c-Maf IN neurons (12.7 ± 1.1 ms vs. 6.8 ± 0.9 ms, respectively, p = 0.002) ( Figures 3K-3N), suggesting that CST input is processed differentially by spinal c-Maf EX or c-Maf IN neurons.

Behavioral effects of silencing of c-Maf IN and c-Maf EX neurons in healthy mice
To examine potential roles of c-Maf deep dorsal horn neurons in the processing of somatosensory stimuli, we examined the consequences of transient silencing of c-Maf EX or c-Maf IN neurons in a battery of sensory tests. To this end, we injected an rAAV encoding the Cre/Dre double-dependent inhibitory chemogenetic receptor hM4Di (rAAV.hSyn1.C on /D on hM4Di-mCherry) 36,37 into the lumbar spinal cord of c-Maf EX , c-Maf IN , or control mice (lacking either one or both recombinases) ( Figure 4A). To determine the specificity and efficacy of the C on /D on hM4Di-mCherry construct, we reacted spinal slices from c-Maf EX and c-Maf IN mice injected with the AAV.C on /D on hM4Di-mCherry with anti-mCherry, c-Maf ,and Pax2 antibodies. We found 75% ± 2.6% of the excitatory mCherry + (mCherry + ;Pax2 À ) and 77.4% ± 3.1% of the inhibitory mCherry + (mCherry + ;Pax2 + ) cells to contain detectable levels of c-Maf ( Figure 4A). Vice versa, 54.5% ± 4.2% of the c-Maf EX cells (c-Maf + , Pax2 À ) and 56.3% ± 3.3% of the c-Maf IN cells expressed detectable levels of mCherry. None of the mCherry + neurons detected after injection into c-Maf EX mice co-expressed Pax2. Starting 14 days after the intraspinal injection, mice were treated intraperitoneally (i.p.) with the chemogenetic agonist clozapine N-oxide (CNO) followed by sensory testing. Silencing of c-Maf EX neurons did not change the responses of the mice to innocuous (Figures 4B  and 4F; Table S1) or noxious mechanical stimuli ( Figure 4E and Table S1), or noxious heat ( Figure 4C and Table S1) or cold (Figure 4D) stimuli, nor did it alter performance in the rotarod test ( Figure 4G and Table S1). Thus, silencing c-Maf EX interneurons did not alter somatosensory thresholds in naive mice and did not impair gross motor coordination. Silencing of c-Maf IN neurons reduced sensory thresholds to punctate mechanical stimulation ( Figure S6B, von Frey; Table S1) but had no impact on thresholds of noxious thermal (heat and cold) or noxious mechanical (pin prick) stimuli (Figures S6C-S6F and Table S1). Cre/Dre double-dependent rAAV encoding the excitatory chemogenetic receptor hM3Dq 36,37 (rAAV.hsyn1.C on /D on hM3Dq) ( Figure S6A). Transient activation of c-Maf EX neurons led to a strong reduction in mechanical thresholds in the von Frey test ( Figure 4H), but no differences were observed in response to pin prick, brush, or heat stimulation (Figures 4I, 4K, and 4L; Table S1). Animals displayed a slight hypersensitivity to cold ( Figure 4J and Table S1). Consistent with the observed mechanical hypersensitivity, activation of c-Maf EX neurons also led to profound spontaneous aversive behaviors, including biting and flinching of the ipsilateral hindpaw ( Figure 4M and Table S1). After repeated CNO injections, c-Maf EX mice also developed skin lesions ( Figure S7), which together with the increased biting behavior may indicate the presence of itch-like sensations after chemogenetic activation of c-Maf EX neurons. Taken together, our data suggest that c-Maf EX neurons are dispensable for noxious stimulus-evoked responses in naive mice, but their chemogenetic activation induces strong mechanical hypersensitivity and spontaneous aversive behaviors. Activation of c-Maf IN neurons ( Figures S6G-S6K) conversely reduced responses to noxious mechanical and dynamic mechanical stimuli ( Figures S6J and S6K; Table S1).

Mechanical nociception in neuropathic mice requires dorsal horn c-Maf EX neurons
The experiments described above suggest that c-Maf EX neurons receive input from non-nociceptive touch-sensitive sensory fibers and connect to dorsal horn structures transmitting noxious mechanical stimuli but are silenced during acute nociceptive stimulation in naive mice. Previous work has suggested that under pathological conditions, such as after nerve injury and potentially also in response to peripheral inflammation, touch-sensitive sensory fibers gain access to dorsal horn nociceptive circuits giving rise to mechanical allodynia. 3,5,20 We therefore asked whether c-Maf EX neurons might be part of such allodynia circuits. To test this hypothesis, we examined the consequences of c-Maf EX neuron silencing on nociception in neuropathic or inflammatory pain models. Neuropathic pain was induced in c-Maf EX and control mice by a chronic constriction injury (CCI) of the left sciatic nerve. 39 To inhibit c-Maf EX neurons, animals were injected with the rAAV encoding the Cre/Dre doubledependent inhibitory hM4Di receptor ( Figure 5A). CCI surgery was performed 1 week after injection of the rAAV ( Figure 5B). Seven days after the CCI surgery, all mice displayed strong hypersensitivity to von Frey filament stimulation ( Figure 5C and Table S1). Transient silencing of c-Maf EX neurons significantly reduced this hypersensitivity ( Figure 5C and Table S1) and, even more, the responses to pin prick stimulation ( Figure 5D and Table S1). Responses to innocuous brush stimulation were not significantly affected ( Figure 5E and Table S1).
Inflammatory hyperalgesia was evoked by subcutaneous plantar injection of zymosan A into the left hindpaw. 40 Zymosan A was injected 10 days after the AAV injection, and mice were tested during the following 2 days (Figures S8A and S8B). Mice developed mechanical hypersensitivity within 24 h but, in contrast to what had been observed in mice after nerve injury, neither responses to von Frey filament stimulation nor responses to noxious pin prick stimulation were reduced by c-Maf EX neuron silencing ( Figures S8C-S8E and Table S1). Our data therefore indicate that c-Maf EX neurons are elements of circuits of mechanical nociception and act as second-order mechano-nociceptors after nerve injury but do not contribute to mechanical sensitization induced by inflammation.
Silencing of c-Maf IN neurons in naive mice produced mechanical hypersensitivity. We therefore investigated whether their activation after nerve injury would conversely reduce mechanical hypersensitivity in neuropathic mice. We indeed observed a profound reduction in pin-prick-evoked responses ( Figure 5G and Table S1) and a slight, statistically non-significant reduction in mechanical (von Frey) hypersensitivity ( Figure 5F and Table S1). These changes were strikingly similar to what was observed after the silencing of c-Maf EX neurons.

Inhibitory PV and c-Maf neurons control activity of c-Maf EX neurons
The observation that chemogenetic activation of c-Maf EX neurons produced allodynia and spontaneous pain-like behaviors in healthy mice, while their inhibition was without obvious effects, suggests that c-Maf EX neurons are silenced under physiological conditions. To search for such inhibitory input, we used monosynaptic rabies tracing with c-Maf EX neurons as the starter population. Primary infected neurons (c-Maf EX , eGFP + , TVA + neurons) and neurons presynaptic to the starter population (eGFP + but TVA À ) could be distinguished by the respective presence or absence of TVA immunoreactivity ( Figure 6A). About half of the eGFP + TVA À neurons were positive for Pax2 (52.3% ± 4.0%, Figures 6A and 6C). This was confirmed by in situ hybridization showing that 46.2% ± 1.5% of eGFP + neurons also expressed vGAT (Figures 6B and 6D), suggesting a large inhibitory input onto c-Maf EX neurons from local interneurons. We further investigated the identity of these inhibitory neurons using multiplex in situ hybridization ( Figure 6D). We found that 21.0% ± 2.6% of the presynaptic inhibitory neurons also expressed c-Maf ( Figure 6D) and 41.9% ± 6.3% expressed PV (Figures 6B and  6D). Other previously established markers for inhibitory dorsal  Table S1. In brief: *p < 0.05, **p < 0.01 (B-L: ANOVA, followed by pairwise comparisons; M: unpaired Student's t test). Scale bars, 100 mm (overview image) and 10 mm (higher-magnification images). 2.1% ± 1.1%; nNOS: 4.5% ± 3.5%; Gal: 5.6% ± 2.6%; NPY: 0.0% ± 0.0%, respectively; Figure 6D). c-Maf EX neurons therefore appear to be predominantly controlled by inhibitory c-Maf and PV neurons ( Figure 6E However, silencing c-Maf IN neurons in naive mice only partially recapitulated the behavioral effect of c-Maf EX neuron activation suggesting additional inhibitory input to c-Maf EX neurons, for example by inhibitory PV neurons. Consistent with this concept, a previous study reported the development of mechanical allodynia after ablation of all (excitatory and inhibitory) PV interneurons, yet no spontaneous pain or itch-like behavior was observed. 19 We found that about 20% of all dorsal horn PV neurons are excitatory (Figures S9A-S9C) and that some c-Maf EX neurons also express PV ( Figure 1G). The remaining $80% of spinal dorsal PV neurons co-express GlyT2 (Figures S9B and  Table S1). PWL, paw withdrawal latency; BL, baseline before injury; CCI, BL 7 days after chronic constriction injury and before CNO injection. 1 h to 24 h refers to time post CNO injection. Error bars denote ±SEM. Number of mice and statistics are shown in Table S1. In brief: *p < 0.05, **p < 0.01, ***p < 0.001 (ANOVA, followed by pairwise comparisons). Article ll OPEN ACCESS S9C). We therefore decided to specifically ablate inhibitory PV (PV IN ) neurons in the spinal cord, again using intersectional genetics. We generated PV Dre ; GlyT2:Cre (PV IN mice) and ablated PV IN neurons by injecting AAV encoding a Cre/Dre doubledependent version of an inducible diphtheria toxin receptor (iDTR) cassette (AAV.hSyn.flex.roxed-iDTR) followed by i.p. injection of diphtheria toxin (DTX) 10 days later ( Figure S9D). Seven to ten days after ablation, the number of PV neurons was reduced at the injected side ( Figure S9E (Figure S9E). After PV IN neuron ablation, mice showed a strong reduction in withdrawal thresholds upon von Frey filament stimulation ( Figure 6F and Table S1) and developed strong spontaneous aversive behaviors such as biting and flinching of the affected hindpaw ( Figure 6K and Table S1). However, responses to heat, cold, pin prick, and brush stimulation remained unchanged ( Figures 6G-6J and Table S1). This phenotype hence recapitulated the behavioral changes observed after activation of c-Maf EX neurons.
Taken together, we have shown that c-Maf EX neurons receive most of their inhibitory input from local PV and c-Maf interneurons. Inhibiting either of these inhibitory populations produced phenotypes consistent with the idea that they gate c-Maf EX neuron activity and thus prevent c-Maf EX neurons from engaging nociceptive circuits in naive mice ( Figure 6E).

Output of c-Maf EX neurons
Our data suggest that c-Maf EX neurons act as second-order mechano-nociceptors after forced chemogenetic activation or after release from local inhibition. We next addressed the nature of the output of c-Maf EX neurons to determine whether their target neurons are known components of circuits transmitting noxious information. To this end, we used orthogonal approaches. We quantified c-fos induction after chemogenetic activation of c-Maf EX neurons and performed anterograde tracing employing wheat germ agglutinin (WGA) expression. 44 Chemogenetic activation of c-Maf EX neurons strongly increased the number of c-fos immunoreactive cells in the lumbar dorsal spinal cord (55.0 ± 16.6 vs. 11.5 ± 1.8 c-fos + cells in the superficial laminae and 305.0 ± 62.8 vs. 73.2 ± 10.7 c-fos + cells in the deep dorsal horn, Figures 7A-7D). Chemogenetically stimulated c-Maf EX neurons thus provide excitatory input to nearby deep dorsal horn neurons and also relay excitation from the deep to the superficial dorsal horn. To further characterize these downstream neurons, we examined the expression of CR and PKCg, marker genes of known separate excitatory interneuron populations that link deep dorsal horn neurons to the more superficially located nociceptive circuits. 12,13,45 Only 1.8% ± 0.6% expressed PKCg, but 8.9% ± 0.4% of c-fos + neurons were positive for CR.
For anterograde tracing experiments with WGA, we injected a rAAV encoding a Cre/Dre double-dependent WGA transgene into the spinal cord of c-Maf EX mice crossed to the Cre and Dre double-dependent reporter line Ai66 (Rosa26 dstdTom/wt ). The tdTomato reporter line was used to distinguish trans-synaptically labeled (tdTom À ) from directly infected neurons (tdTom + ). We found that about half of the targeted neurons (trans-synaptically labeled) expressed Pax2 (48.6% ± 1.8% of postsynaptic WGA + , Figures 7F and 7I). CR was found in 11.8% ± 1.4% of postsynaptic WGA + neurons ( Figures 7G and 7I). In contrast, none of the postsynaptic WGA + neurons (tdTomato À ) expressed PKCg ( Figures 7H and 7I). This is in line with our observation that some CR + , but almost no PKCg + , neurons, were activated (c-fos + ) following activation of c-Maf EX neurons ( Figure 7E).
Finally, we asked whether c-Maf EX neurons could directly target ascending projection neurons of the dorsal horn. To this end, we labeled the synaptic terminals of c-Maf EX neurons with the fusion protein synaptophysin-mCherry while spinoparabrachial neurons (and potentially fibers of passage 46 ) were labeled from the (contra-)lateral parabrachial nucleus (LPb) using a rAAV2-retro-CAG-eGFP, a serotype that was specifically developed for improved axon terminal infection and retrograde transduction 47 ( Figure 7J). Many of the eGFP-labeled neurons were in the deep dorsal horn and a few scattered neurons were found in lamina I ( Figure 7K). Of all eGFP + LPb projection neurons in the deep dorsal horn, 62.7% ± 15.1% had at least one mCherry + VGlut2 + terminal in close apposition to a homer + puncta on the cell body or a dendrite ( Figures 7L-7O). No direct input of c-Maf EX neurons onto lamina I projection neurons was detected. To demonstrate the functionality of the connections between c-Maf EX neurons and ascending deep dorsal horn projection neurons, we overexpressed ChR2-YFP (AAV.Flex.ChR2-YFP) in spinal c-Maf neurons ( Figure 7P) and recorded light-evoked potentials in spinal projection neurons retrogradely labeled from the LPb with AAV2retro.tdTomato. Recorded cells responded with an average latency of 5.9 ± 0.9 ms (jitter: 1.2 ± 0.2 ms), suggesting monosynaptic connections  (Figures 7O-7S). Taken together, our results suggest that c-Maf EX neurons can engage nociceptive circuits via two pathways: activation of CR neurons in superficial laminae and direct activation of scattered deep dorsal LPb projection neurons.

DISCUSSION
The present study has focused on the function of dorsal horn c-Maf neurons in spinal nociceptive signaling. These neurons are of particular interest, as they constitute a main target population of the corticospinal tract whose activity is required for nerve-injury-induced neuropathic mechanical sensitization. 8 We demonstrate that c-Maf neurons fall into two subpopulations, excitatory and inhibitory interneurons, which exert opposing effects on pain. c-Maf EX neurons are part of a normally silent circuit required for nerve-injury-induced mechanical allodynia, and relays signals from non-nociceptive sensory fibers to dorsal horn nociceptive output structures. Under healthy conditions, these c-Maf EX neurons are silenced via feedforward inhibition by c-Maf IN and inhibitory PV interneurons.
Disinhibition as a source for functionalization of c-Maf EX neurons is in good agreement with previously proposed mechanisms for neuropathic sensitization. 48,49 However, disinhibition also occurs in inflammation. 50,51 Yet silencing of c-Maf EX neurons failed to reduce allodynia in mice with inflamed paws, supporting the concept that distinct allodynia circuits become activated depending on the nature of the underlying pathology. 10 Integration of c-Maf EX and c-Maf IN neurons in sensory circuits of the dorsal horn c-Maf EX neurons receive input from myelinated primary afferent fibers including TrkB-positive LTMRs. Input from these fibers has previously been shown to be essential for nerve-injuryinduced mechanical allodynia. 52 The prevalence of this non-nociceptive input to c-Maf neurons is consistent with the location of c-Maf neurons in the deep dorsal horn. Analyses of changes in c-fos expression following chemogenetic c-Maf EX neuron activation and anterograde tracing with WGA revealed that CR neurons but not PKCg neurons, another subpopulation previously reported to affect nerve-injury-evoked pain sensitivity, 53 are postsynaptic to c-Maf EX neurons. CR neurons have previously been proposed to connect VGlut3 lineage neurons to nociceptive circuits of the superficial dorsal horn. 11 CR neurons may therefore act as third-order interneurons in a pathway that relays LTMR input to dorsal horn nociceptive output neurons via c-Maf EX neurons. This is consistent with recent studies indicating that CR neurons receive polysynaptic input from Ab fibers (LTMRs) and in turn directly target spinoparabrachial projection neurons in lamina I. 12,45,54 In addition to the superficial dorsal horn output system, our study identified a second output pathway that links c-Maf EX neurons to higher-order nociceptive centers. c-Maf EX neurons frequently contacted scattered deep dorsal horn projection neurons that were retrogradely labeled from the LPb. While most recent work has focused on lamina I projection neurons, projection neurons in the deep dorsal horn act as an additional nociceptive output system. 55 In addition, work conducted by Browne et al. 56 suggests that laminae (III-V) projection neurons form an alternative route to provide qualitatively different sensory information to the LPb. Our data therefore suggest that excitatory deep dorsal horn c-Maf interneurons may engage pain circuits by directly activating deep dorsal horn projection neurons, independently of the superficial dorsal horn.
Our finding that ablation or silencing of c-Maf EX neurons had no major impact on nociceptive behavior in healthy mice suggests the presence of strong inhibitory control of these neurons. Retrograde tracing experiments performed to reveal the origin of such inhibition identified inhibitory PV and c-Maf IN neurons presynaptic to c-Maf EX neurons. Compromising the activity of these neurons led to mechanical hypersensitivity and spontaneous  19 However, in this study, no spontaneous pain-like or itch-like behavior was reported. This might be because in this previous study all PV neurons were ablated, which included about 15% excitatory neurons that may be necessary for the observed spontaneous aversive behaviors.
Our retrograde tracing experiments revealed additional sources of input to dorsal horn c-Maf EX neurons originating from supraspinal centers, including the primary somatosensory cortex S1, the red nucleus, and the RVM. In the context of the present study, the primary somatosensory cortex S1 is of particular interest because transection of the corticospinal tract specifically abolishes mechanical hyperalgesia in neuropathic mice. 8 This specificity fits with the innervation of dorsal horn c-Maf EX neurons by low-threshold mechanoreceptors and the virtual absence of input from unmyelinated nociceptors.
Our data indicate that c-Maf EX and c-Maf IN neurons are innervated by a similar set of peripheral as well as supraspinal neurons. This leads us to propose that c-Maf IN neurons provide feedforward inhibition to c-Maf EX neurons. However, this raises the conundrum of how a disynaptic connection can inhibit monosynaptic input from the same source. We provided evidence that the optogenetic innervation of CST terminals produced lightevoked responses in c-Maf EX neurons with almost twice the latency as compared with c-Maf IN neurons. Differential dendritic filtering of peripheral and supraspinal input to c-Maf EX and c-Maf IN neurons as suggested by Zhang et al. 57 might result in effective feedforward inhibition. Clearly, additional studies are required to address interconnectivity, for example between supraspinal sites and spinal neurons, in greater detail.
c-Maf neurons and neuropathic itch c-Maf EX neuron activation experiments revealed phenotypes that were reminiscent of neuropathic itch in human patients. In humans, nerve injury not only gives rise to neuropathic hyperalgesia and allodynia but can also lead to neuropathic itch (allokinesis). 58 It has been suggested that similar mechanisms underlie both pathologies. This concept is backed by our finding that activation of c-Maf EX neurons produces both neuropathic pain-like behaviors (i.e., spontaneous pain and mechanical allodynia) and itch-like behaviors (biting leading to self-inflicted skin lesions). Furthermore, both phenotypes are recapitulated by the ablation of inhibitory PV neurons. Together with our retrograde tracing experiments, these data suggest that, under disinhibitory conditions, LTMR input can give rise to pain-like or itch-like behavior, both depending on the recruitment of dorsal horn c-Maf EX neurons.
In summary, our study has identified c-Maf EX neurons as critical elements of a spinal allodynia and allokinesis circuit that connects innocuous input from touch-sensitive sensory fibers to dorsal horn nociceptive output structures. c-Maf EX neurons appear to serve a unique function in this circuit by integrating peripheral sensory input with both local inhibition and descending excitation from the corticospinal tract. They hence link this circuit not only to well-established disinhibitory processes in neuropathic pain but also to more recent concepts pointing to the importance of top-down modulation in neuropathic pain. 8 Limitations of the study In this study, we demonstrate that deep dorsal horn c-Maf EX neurons connect non-nociceptive somatosensory input to spinal nociceptive output pathways. We propose that the activity of these c-Maf EX neurons is normally silenced by inhibitory PV To ultimately prove that loss of PV IN and c-Maf IN function leads to c-Maf EX activation, combinatorial gain-and loss-offunction experiments would be required. However, this requires the identification of non-overlapping marker genes for the respective populations. We also demonstrate that c-Maf EX neurons are critical for mechanical hypersensitivity and allodynia after nerve injury. Compared with silencing other excitatory dorsal horn populations, e.g., CCK + or VGlut3 + neurons, 8,10 silencing c-Maf EX neurons had a relatively small impact on mechanical allodynia, although we and others reported an extensive overlap between c-Maf EX and CCK + or VGlut3 + neurons. 10 Methodological differences (chemogenetic silencing vs. toxin-mediated ablation or silencing) may account for these differences.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following: