Control of somatic membrane potential in nociceptive neurons and its implications for peripheral nociceptive transmission

&NA; We identified major ion channels influencing the resting membrane potential of nociceptive sensory neurons and demonstrated that changes of somatic/perisomatic membrane potential of these neurons can strongly influence peripheral nociceptive transmission. &NA; Peripheral sensory ganglia contain somata of afferent fibres conveying somatosensory inputs to the central nervous system. Growing evidence suggests that the somatic/perisomatic region of sensory neurons can influence peripheral sensory transmission. Control of resting membrane potential (Erest) is an important mechanism regulating excitability, but surprisingly little is known about how Erest is regulated in sensory neuron somata or how changes in somatic/perisomatic Erest affect peripheral sensory transmission. We first evaluated the influence of several major ion channels on Erest in cultured small‐diameter, mostly capsaicin‐sensitive (presumed nociceptive) dorsal root ganglion (DRG) neurons. The strongest and most prevalent effect on Erest was achieved by modulating M channels, K2P and 4‐aminopiridine‐sensitive KV channels, while hyperpolarization‐activated cyclic nucleotide‐gated, voltage‐gated Na+, and T‐type Ca2+ channels to a lesser extent also contributed to Erest. Second, we investigated how varying somatic/perisomatic membrane potential, by manipulating ion channels of sensory neurons within the DRG, affected peripheral nociceptive transmission in vivo. Acute focal application of M or KATP channel enhancers or a hyperpolarization‐activated cyclic nucleotide‐gated channel blocker to L5 DRG in vivo significantly alleviated pain induced by hind paw injection of bradykinin. Finally, we show with computational modelling how somatic/perisomatic hyperpolarization, in concert with the low‐pass filtering properties of the t‐junction within the DRG, can interfere with action potential propagation. Our study deciphers a complement of ion channels that sets the somatic Erest of nociceptive neurons and provides strong evidence for a robust filtering role of the somatic and perisomatic compartments of peripheral nociceptive neuron.


Introduction
In contrast to the majority of central nervous system neurons, peripheral somatosensory neurons normally generate action potentials (APs) at peripheral nerve endings, not at the axon hillock [2,3]. While somatic APs and electrogenesis are not required for AP propagation from the periphery to the spinal cord [4], sensory neuron somata are electrically excitable [3,12,111,122], and ectopic somatic activity [3,12,77,111,122], along with ectopic peripheral fibre activity [22,23,125], is thought to contribute to many chronic pain conditions. Moreover, measurements [28,33,110,113] and simulations [79] suggest that the axonal bifurcation (t-junction) within dorsal root ganglia (DRG) influences the transmission of spikes on their way to the spinal cord. Hitherto unexplained recent clinical studies have established that direct electrical stimulation (''neuromodulation'') of the DRG provides efficacious pain relief in neuropathic pain patients [20,95]. Taken together, these findings suggest that sensory ganglia may play a much stronger role in peripheral nociceptive transmission than is generally accepted. Moreover, sensory ganglia may represent a novel target for pain therapeutics [95]. Yet been done so far to directly test how electrophysiological properties of somatic/perisomatic compartment of sensory neuron affect peripheral somatosensory transmission.
The aims of this study were 1) to identify major ion channels influencing the resting membrane potential (E rest ) of nociceptive DRG neurons and 2) to investigate if (and how) manipulation with the activity of these channels within the somatic/perisomatic compartments of DRG would affect peripheral nociceptive transmission. In the first part we focused on the ion channels that are known to be expressed in nociceptive DRG neurons and would be expected to be active at, and possibly contribute to, the E rest of these neurons. These channels included 4-aminopiridine (4-AP)-sensitive voltage-gated K + (K V ) channels [26], slow-activating M channels (Kv7, KCNQ) [26,58,71,74,92,93,100,101], 2-pore K + ''leak'' channels (K2P) [1,80,115], sodium-activated K + channels (Slo2.x, K Na ) [32,91,116]; hyperpolarization-activated cyclic nucleotide-gated channels (HCN) [13,30,31,128], low voltage-activated T-type Ca 2+ channels (Ca v 3.x) [50,90,107,114], and voltage-gated Na + channels (VGNC) [6,60]. These analyses identified M channels, 4-AP-sensitive K V and K2P ''leak'' channels as those having most significant influence over the E rest . We then investigated how the hyperpolarization of somatic/perisomatic E rest in the DRG achieved by modulating somatic/perisomatic ion channel activities can influence nociceptive transmission in vivo. Finally, a computational model of a nociceptive DRG neuron was used to better understand how the relationship between morphology, membrane potential, and ion channels active at E rest influence nociceptive signal propagation. Our study identifies major ion channels that set somatic E rest of nociceptive neurons and provides firm evidence for a much stronger role of sensory ganglia in the peripheral nociceptive transmission than is generally thought.

Neuronal cultures and slice preparation
DRG neurons were cultured as described previously ( [74,76]; see [59] for step-by-step protocol). Briefly, adult male Sprague-Dawley rats (180-200 g) were humanely euthanized by cervical dislocation under the isoflurane anaesthesia. DRG from all spinal levels were removed and treated at 37°C in Hank's Balanced Salt Solution supplemented with collagenase (1 mg/mL; Sigma-Aldrich, St. Louis, MO, USA) and dispase (10 mg/mL; Invitrogen, Life Technologies, Grand Island, NY, USA) for $30 minutes. Ganglia were then gently triturated, washed twice, and resuspended in 600 lL culturing media (approx. 500,000 cells per isolation); this suspension was then plated as dense cultures onto glass coverslips coated with poly-D-lysine and laminin. Neurons were cultured for 2 to 5 days. No nerve growth factor was added to the culture to avoid inflammatory insult; we found that densely plated cultures survive well without trophic factors added. It is of note that our dissociation protocol provides cultures that are enriched with smalldiameter, high-threshold (presumed nociceptor) neurons because large-diameter, low-threshold mechanoreceptors mainly die during trituration due to the mechanical overstimulation, unless specifically protected [36,59].

Electrophysiology
Whole-cell and perforated patch recordings in current clamp configuration were performed at room temperature (unless indicated otherwise). Patch pipettes (resistance 2-4 MX) were fabricated from borosilicate glass capillaries using a DMZ-universal horizontal puller (Zeitz, Martinsried, Germany) or a Sutter P-97 puller (Sutter, Novato, CA, USA). Currents were amplified and recorded using an EPC-10 patch amplifier and Patchmaster 2.2 software (HEKA Electronik, Lambrecht, Germany) or an Axon patch 700B amplifier and pCLAMP 10.0 software (Axon Instruments, Union City, CA, USA), and were sampled at a frequency of 5 kHz. Liquid junction potentials were calculated with the algorithm developed by P.H. Burry [7] using pCLAMP software and subtracted post acquisition. Continuous current-clamp recording with no current injection was used for E m monitoring. Linear ramps of currents from 0 to 1 nA (1-second duration) were injected for measuring rheobase and other AP parameters. The extracellular solution contained (in mM): 160 NaCl, 2.5 KCl, 5 CaCl 2 , 1 MgCl 2 , 10 HEPES, and 8 glucose, pH 7.4. The intracellular solution for perforated patch experiments [70] [101].
Sharp electrode recordings were performed from DRG slices held in a submerged-type chamber and perfused with carbogenated extracellular solution (4-5 mL/min) at room temperature. Electrodes were pulled using a DMZ-universal horizontal puller to resistances of 70-120 MX when filled with a solution containing 1 M K-acetate (plus 1 mM KCl; pH 7.2 adjusted with acetic acid). Some recordings were performed with electrodes filled with 1 M KCl (plus 10 mM HEPES titrated to 7.2 with potassium hydroxide). Recordings were made using an SEC-05L amplifier (npi electrotonic, Tamm, Germany) and digitized (10 kHz) with a PC-based system (Digidata 1200 and Clampex 9.3, Molecular Devices, Sunnyvale, CA, USA) and analysed off-line (Clampfit 10.1). To measure the rheobase and to analyse AP properties, a family of 600-ms current injections (between À0.35 and +1 nA with 0.05 nA increment) was used. Because liquid junction potential should be <1 mV [89], no correction was applied.
To identify neurons as nociceptive, capsaicin (1 lM) has been applied at the end of the recording in all recording paradigms. Due to the small number of capsaicin-insensitive neurons and due to the fact that it was not always possible to apply capsaicin (eg, due to the premature loss of the recording), data from capsaicin-sensitive and capsaicin-insensitive neurons were not analysed separately.

Experiments with recombinant channels
In experiments testing specificity of K Na -modulating drugs, plasmids encoding human Kv7.2 and Kv7.3 (GenBank accession no. NM000218 and AF091247) were transfected into Chinese hamster ovary (CHO) cells using Lipofectamine 2000 (Invitrogen). In experiments testing effect of ST101 on Cav3.2, the plasmid encoding human Cav3.2 (GenBank accession no. AF051946; kind gift from Prof. Chris Peers, University of Leeds, UK) was transfected into human embryonic kidney (HEK293) cells and whole-cell recordings were performed. The recordings were made using an Axon 700B patch-clamp amplifier (Axon Instruments); signals were filtered at 2 kHz and analysed using pCLAMP 10 (Axon Instruments) and Origin 7.5 (OriginLab Corporation, Northampton, MA, USA).

Atomic absorption spectroscopy Rb + efflux assay
Rb + efflux assay to study the modulation of M channels has been described in detail previously [97]. Briefly, CHO cells stably transfected with Kv7. 2 where y is the response; A 1 and A 2 are the maximum and minimum response, respectively; x is the drug concentration, and p is the Hill coefficient.

Acute focal application of ion channel modulators to DRG in vivo
All surgical procedures were performed under deep anaesthesia with an intraperitoneal injection of pentobarbital sodium (10-20 mg/kg) in accordance with the Animal Care and Ethical Committee of Hebei Medical University (Shijiazhuang, China) under the International Association for the Study of Pain guidelines for animal use. Focal application of compounds to the DRG in vivo was performed as described before [96], with modifications. Briefly, a midline incision was made at the L4-L6 spinal level of an adult male rat (Sprague-Dawley; 180-200 g), and the L5 was identified at the midpoint of a link between both sides of iliac crest. A 0.8-mm hole (approximately 2 mm off the inferior edge of the transverse process) was drilled through the transverse process over the L5 DRG. Approaching of ganglion was verified by the twitch of the hind paw, at which point the drilling was stopped immediately. A hooked stainless steel blunt-tip cannula (inner diameter 0.64 mm, length 4 mm) was forced into the hole and connected to a polypropylene tube (inner diameter 0.41 mm, length 4.5 mm). The incision was closed with sutures and the cannula was firmly fixed in place with dental cement. Intramuscular injection of benzylpenicillin (19 mg/0.1 mL) was given immediately after surgery. Postoperatively, rats were housed individually in plastic cages with sawdust flooring and supplied with water and food ad libitum. Animals were left to recover for at least 24 hours before the experiments were carried out. Animals developing signs of distress were humanely euthanized by cervical dislocation under the isoflurane anaesthesia.
To evaluate the effect of focal application of ion channel modulators to DRG on the nociceptive processing, 5 lL of retigabine, pinacidil, or ZD7288 solution (each at 200 lM) or saline/vehicle control were injected via the DRG cannula immediately prior to the hind paw plantar injection of 50 lL of bradykinin (200 lM).
The animal was returned to the cage and video-recorded for 30 minutes. Time spent licking, flinching, and biting the injected paw over the period of 30 minutes was analysed by the operator blind to the composition of the injected solution.
In order to verify that drug exposure was limited to the DRG, a fluorescent dye, 5(6)-Carboxyfluorescein diacetate N-succinimidyl ester (Sigma; 20 lM in 5 lL), was injected via the cannula implanted as described above. Dye injection was performed on animals that received no other injections before; approximately 30 minutes after injection, the animal was sacrificed, both the L5 DRG and proximal inferior part of the lumbar spinal cord were excised, submerged in Tissue-Tek O.C.T. (Sakura, Alphen aan den Rijn, The Netherlands), frozen, and sectioned (15 lm) using a freezing microtome (CM1950, Leica Microsystems). Slices were then analysed for the presence of dye using confocal microscopy (TCS SP5 II, Leica Microsystems).

Computer modelling
A computational model of small-diameter nonmyelinated DRG neuron was constructed and simulated using NEURON (http:// www.neuron.yale.edu) [42,43] on an Intel-based Macintosh computer (Apple Inc, Cupertino, CA, USA). Simulations were analysed using IgorPro analysis software (Wavemetrics, Lake Oswego, OR, USA). Our model neuron had a morphology based on available literature: the soma was 25 lm in diameter [38,129], with a capacitance of 20 pF, while the diameters of the peripheral and central axons were 0.8 and 0.4 lm, respectively [35,45,79,112]. The stem axon arising from the peripheral axon had a diameter of 1.4 lm and was 75 lm in length, except where noted. Axonal compartments within the DRG were subdivided into 100 sections for computational accuracy [104]. For all compartments, E rest = À60 mV, R m = 10,000 Xcm 2 , C m = 1 lF/cm 2 , and R a = 100 Xcm [79]. These parameters resulted in a model with a somatic input resistance of 274 MX [38,129] and an apparent cell capacitance (ratio of membrane time constant and input resistance: s m /R N ) of 29.6 pF.
VGNC and delayed rectifier K + channels were expressed in all compartments with a density of 0.04 S/cm 2 , except at the soma, where VGNC was 0.02 S/cm 2 [84]. The voltage-dependence of the VGNC was adjusted to be approximately mid-way between values reported for Na V 1.7 and Na V 1.8 channels in DRG [18,106]. M and HCN channels were inserted in the soma, stem axon, and in most simulations, extended 100 lm into the peripheral and central axons. Conductance densities for these channels are reported in units of pA/pF at potentials of À30 mV (M channels V 1/2 ) and À100 mV (HCN channels 100% activation). E leak in all compartments was calculated from resting Na + , K + , and M channel or HCN currents to achieve a E rest of À60 mV [33,38]. APs were initiated in the peripheral axon distal to the t-junction by depolarizing current steps (0.2 nA, 1-ms duration). Where noted, constant current was injected into the soma (1-100 pA).

Compounds
List of all ion channel modulators used in this study, as well as their abbreviations and concentrations and sources are listed in Table 1. All compounds, except of XE991 (XE), were used at concentrations sufficient to produce maximal effect. XE was used at 3 lM (near IC 80 ), as at saturating concentrations it may affect other channels such as eag1 and Kv4.3 [123].

Statistics
All data are given as mean ± SEM. In the experiments where normal distribution of data cannot be expected, the following approach has been applied. 1) Differences between groups of paired values were analysed using paired Wilcoxon test. 2) Kruskal-Wallis analysis of variance (ANOVA) was used to compare among multiple groups. 3) Pairs within multiple groups were analysed by Mann-Whitney test with Bonferroni correction. 4) Differences in proportions of responsive cells were analysed using Fisher's exact test. In the case where normal distribution was confirmed, t-test (paired or unpaired, as appropriate) was used; where indicated, multiple groups were compared using one-way ANOVA with Bonferroni post hoc test. Differences were considered significant at P 6 0.05. Statistical analyses were performed using Origin 9.0 (OriginLab Corporation, Northampton, CA, USA), Minitab 16 (Minitab Inc, State College, PA, USA), and Prism 6.01 (GraphPad Software Inc, La Jolla, CA, USA).

Results
Membrane potential (E m ) of a neuron during resting state (resting membrane potential, E rest ) results from the steady-state interaction of a number of membrane conductances, mostly represented by ion channels [41,51], with small contribution by electrogenic pumps [61,63]. In order to identify ionic conductances that contribute to the somatic E rest of nociceptive neurons, we measured changes in E m (DE m ) in response to pharmacological inhibition or activation of ion channels expected to be active at voltages near the E rest . Our main experimental model was cultured small-diameter ($20 lm) DRG neurons with whole-cell capacitance of 26.3 ± 1.3 pF (n = 32); these neurons were predominantly capsaicin sensitive (71% or 174/245 of such cells responded to 1 lM capsaicin). Responsiveness to capsaicin indicates expression of nociceptive neuron marker TRPV1 [65], and thus, we describe the population of neurons under investigation as predominantly small-diameter nociceptors, although contribution of a small number of neurons of other modalities to this population cannot be excluded.
Accurate measurement of E rest can be influenced by the recording configuration, as well as the experimental preparation. Therefore, we used a combination of approaches to measure E rest of small DRG neurons. Specifically, 1) perforated-patch and 2) whole-cell recordings were made from cultured sensory neurons; 3) sharp electrode recordings and 4) a limited number of wholecell recordings were made from sensory neurons in acute DRG slices. Remarkably, similar values for E rest , near À60 mV, were observed across all experiments ( Table 2). For whole-cell experiments there was no significant drift of E m value after the breaking into the neuron, indicating little influence of the intracellular solution exchange during the recording period. Likewise, in sharp-electrode recording experiments, replacing the K-acetatebased pipette solution with the KCl-based solution resulted in no significant difference. Our E rest values were very close to those reported previously [3,5,38,71,72,103,129] and show that E rest of both cultured and acute DRG neuron somata are maintained within the same voltage range.

Channels contributing to E rest in cultured nociceptive DRG neuron somata
To identify ion channels contributing to the E rest of nociceptive neurons, we used perforated patch current clamp recordings to measure DE m in cultured, small-diameter DRG neurons in response to a set of well-characterized ion channel blockers or enhancers for M channels, K V , K ATP , K Na , HCN, Ca v 3x, and Na V channels. It has to be noted that if a given ion channel is not tonically active at E rest , then its pharmacological activation or enhancement will not inform about the channel contribution to E rest . Responses of varying amplitudes were observed for most compounds in various proportions of neurons ( Fig. 1). Fig. 1A and B summarizes these experiments, with panel A depicting responses to compounds that produced hyperpolarization and panel B depicting these produced depolarization. Number and percentage of neurons responsive to each compound (those where the absolute value of DE m was >1 mV) is listed above each dataset. Fig. 1C compares the mean DE m values for each compound determined from the subset of responsive neurons only. Exemplary E m responses are given in Fig. 2 and statistical analysis of the data is summarized in Table 3.
The most robust and most abundant responses were to the modulators of M-type K + channels. M channel subunits Kv7. 2 [99,121], Kv2s [10,119], and Kv3.4 [15,57] (reviewed in [26]). 4-AP (2 mM), which has little effect on M channels, produced a sizable depolarization of 7.4 ± 0.8 mV in 8/11 neurons (73%; Figs. 1 and 2I, J). This suggests that some K V channels expressed at nociceptor neuron cell bodies are partially open near the E rest . The effect of 4-AP was not significantly different from that of XE (Table 3).
We then turned our attention to channels that generally produce inward currents. Hyperpolarization-activated nonselective cation channels HCN underlie neuronal I h ; HCN1 and HCN2 are expressed in small-diameter nociceptors [30,31]. The HCN blocker ZD7288 (ZD; 10 lM) produced moderate hyperpolarization by À3.5 ± 0.5 mV in 11/19 (58%) neurons, suggesting that there is detectable background HCN activity in many cultured DRG somata ( Figs. 1 and 2B). The effect of ZD was significantly less pronounced as compared to RTG ( Table 3). The effect of ZD was not increased when measurements were repeated at 37°C (not shown).
Most voltage-gated Ca 2+ channels activate at voltages more depolarized than À60 mV, with the exception of T-type Ca 2+ channels (Ca v 3), which have an activation threshold near or even below      Table 1 for complete list of compounds). Values from capsaicin-sensitive (red) and -insensitive (green) neurons, as well as from neurons untested for capsaicin sensitivity (dark grey), are identified. Horizontal black bars in every group depict mean values of all neurons tested. Number of responsive neurons out of total neurons tested for each compound are indicated above each group as X/Y, where X is a number of responsive neurons and Y is a total number of neurons tested; percentage of responsive neurons for each compound is also indicated. (C) Bar chart summarizes effects of each of the compounds tested taking into account responsive neurons only (DE m changes below 1 mV, an average noise amplitude, were considered as no effect). Dotted lines indicate level of depolarization (upper line) or hyperpolarization (lower line) upon M channel inhibition or enhancement, respectively (as indicated).  Table 1 for complete list of compounds). Statistical analysis of responses is presented in Fig. 1 and Table 3. small neurons tested (Figs. 1 and 2L). The fact that ST101 did not affect Ca v 3.2 voltage dependence significantly (Fig. 5B 1 ), but strongly increased current amplitudes at voltages between À60 and À30 mV, suggests that at E rest of $À60 mV native T-type Ca 2+ channels in DRG neurons are just on the margin of their activation threshold, and thus, may be activated by relatively small depolarizations. Finally, we tested the contribution of VGNC; small, nociceptive DRG neurons most abundantly express tetrodotoxin (TTX)-sensitive Na v 1.7, Na v 1.6, and TTX-resistant Na v 1.8 and Na v 1.9 VGNC subunits; reviewed in [21]. In a small fraction of neurons (4/43, 9%), the blocker of TTX-resistant VGNC Na v 1.8 and Na v 1.9, A803467 (A803; 10 lM) induced sizable hyperpolarization of À10.3 ± 0.5 mV (Figs. 1 and 2J). These few neurons that did respond to A803 had a tendency to be more depolarized as compared to an average small DRG neuron (À46.2 ± 4.6 mV, n = 4 vs. À61.4 ± 3.1 mV, n = 43). It is possible that heterogeneous expression of these channels contributes to a more depolarized E rest in these particular neurons. TTX-sensitive VGNC may also contribute to E rest [24]. Accordingly, in most neurons (9/10; 90%), TTX induced a small hyperpolarization of À3.3 ± 0.3 mV (Figs. 1 and 2K). Effects of both A803 and TTX were significantly less strong as compared to RTG (Table 3).

Acute DRG slices
Dissociated DRG neurons in culture provide for a convenient experimental model. However, these cells are axotomized and maintained in vitro, and thus, may not necessarily maintain their native phenotype; therefore, data obtained from such neurons may not reflect the physiology of a neuron in situ. We therefore   performed additional experiments on neurons in acutely prepared DRG slices. As can be seen from Table 2, the basic electrophysiological characteristics of DRG somata as recorded with sharp intracellular electrodes are quite similar to these recorded with whole-cell patch electrodes (somewhat higher rheobase seen in sharp electrode recordings is likely to reflect larger leak currents introduced by sharp microelectrodes). Particularly relevant to this study is the fact that the values of E rest of cultured and acute DRG neuron somata are within the same voltage range; this ensures that the ion channels that remain active near À60 mV (which we have focused on in this study) are likely to be relevant to the maintenance of E rest of DRG somata both in situ and in vivo.
In acute DRG slices, we confirmed the key results obtained with cultured DRG neurons (Fig. 6). Since the E rest baselines recorded with sharp electrodes from DRG slices were on average noisier as compared with these recorded using the whole-cell current clamp from cultured DRG neurons, it was difficult to unambiguously classify slice recordings as ''responding'' or ''nonresponding'' to a compound (with our threshold parameter for response being set at DE m P 1 mV). Therefore, presented in Fig. 6A is a scatter plot showing all the individual responses recorded in such experiments; mean DE m data recorded from all neurons in each group are also given (horizontal black bars). These mean values are likely to underestimate true effects due to the contribution of nonresponding cells. Again, the strongest effect on E rest was found in response to enhancing or blocking M channels; thus, XE depolarized and RTG hyperpolarized E rest by 7.8 ± 1.2 mV (n = 20) and À4.4 ± 1.2 mV (n = 17), respectively. Blocking of HCN channels with ZD (50 lM) resulted in more modest hyperpolarization (À2.0 ± 1.3 mV; n = 11), while inhibition of K2P channels with BUP and K V channels with 4-AP induced comparably more moderate depolarization: 3.96 ± 0.89 mV; n = 8 and 2.87 ± 1.08 mV, n = 7, respectively).
It also has to be noted that in these sharp electrode recordings we were unable to select neurons by size, and therefore, these recordings were blind from a randomized neuron population. In some neurons we were able to test sensitivity to capsaicin (Fig. 6A), but it was not always possible to apply multiple drugs due to the recording stability issues. However, since the majority (up to 70%) of cell bodies in DRG are small-diameter nociceptors [37,64,86], it is reasonable to suggest that the majority of our sharp-electrode recordings were indeed performed on this type of neurons. This is consistent with the observation that among the neurons that were tested for the capsaicin sensitivity in this recording paradigm, 23/38 (60.5%) were capsaicin sensitive. These considerations suggest that our sharp electrode recordings are largely representative of nociceptive neurons.

Effect of somatic/perisomatic hyperpolarization on pain signalling in vivo
In order to evaluate the importance of somatic/perisomatic E rest on nociceptive transmission in vivo, we adapted a method of focal DRG drug injection developed by Puljak and colleagues [96] (with modifications; see Materials and Methods). In this approach, a cannula is inserted into a hole drilled through the transverse process of L5 vertebra; the cannula allows delivery of small volumes of drugs directly to the DRG (Fig. 7A, top panel). In order to verify that this injection technique delivers drugs specifically localized to DRG, and there is no ''spill over'' to the spinal cord, we injected a fluorescent dye 5(6)-Carboxyfluorescein diacetate N-succinimidyl ester (5 lL; 20 lM) through the cannula and tested the extent to which dye spread from the DRG to the proximal spinal cord. Confocal fluorescent imaging revealed abundant fluorescence in the DRG but a complete lack of staining in the spinal cord (Fig. 7B, see Materials and Methods for the experimental details).
Next, we tested the effects of focal DRG application of ion channel modulators on the pain induced by the hind paw injection of bradykinin (BK). BK is a potent endogenous proinflammatory and pain-inducing peptide (algogene) [25,56]; it produces obvious protective or ''nocifensive'' behaviour when injected into the hind paw of rats (flinching, biting, and shaking of the injected paw) [76]. We then tested compounds that significantly hyperpolarized E rest based on our in vitro experiments. Specifically, drugs that enhanced the activity of M and K ATP channels or blocked HCN channels were injected into the DRG prior to BK administration to the ipsilateral hind paw. Focal preapplication of RTG (5 lL, 200 lM), PIN (5 lL; 200 lM), or ZD (5 lL; 200 lM) significantly attenuated BK-induced nocifensive behaviour (P 6 0.01; one-way ANOVA with Bonferroni post hoc test, Fig. 7C). RTG produced the largest attenuation, although difference with PIN and ZD did not reach significance (one-way ANOVA with Bonferroni post hoc test). Importantly, focal DRG application of RTG to the L5 DRG contralateral to the side of BK injection did not produce any reduction in nocifensive behaviour (Fig. 7C; one-way ANOVA with Bonferroni post hoc test). This complete lack of effect of the contralateral focal RTG application is another strong piece of evidence against any spinal effects of a drug applied to DRG via cannula: dorsal roots that ascend from DRG to the spinal cord in rats are about 3 cm long In the column labels, red font denotes plantar paw injections and black font denotes focal application to DRG through the cannula. ⁄ , ⁄⁄ , and ⁄⁄⁄ denote difference from baseline at P 6 0.05, P 6 0.01, and P 6 0.001, respectively (one-way analysis of variance with Bonferroni posttest). ZD, ZD7288; PIN, pinacidil; RTG, retigabine; LOX, loxapine. [112], thus, a substance that diffused that distance and reached the spinal cord in a concentration sufficient to produce an effect would definitely produce a bilateral action. K Na enhancer LOX, which only produced marginal hyperpolarization in cultured DRG neurons, was without an effect (Fig. 7C). It has to be acknowledged that while focal application of drugs to DRG via cannula resulted in no significant spill-over to the spinal cord, the drugs applied in such a manner would affect not only DRG somata but also a perisomatic compartment: stem, t-junction, and adjacent segments of peripheral and central axon. In order to investigate how somatic/perisomatic hyperpolarization may affect transmission of nociceptive signals from the periphery to the spinal cord via the t-junction, we developed a computational model of a small-diameter unmyelinated DRG neuron.

Influence of DRG hyperpolarization on axonal AP propagation: a computational model
We have shown that the injection of compounds into the DRG that hyperpolarize sensory neurons in vitro (eg, RTG and ZD) reduced nocifensive behaviour in vivo. Our working hypothesis is that exogenously induced hyperpolarization enhances the likelihood of AP failure across the t-junction, the point of lowest safety factor for AP propagation [19,33,79,109], and in turn reduces the nocifensive response. Successful AP propagation through a branch point depends on the relative impedance loads between the parent axon and its daughters, as well as their active properties, the length of the stem axon, and any contribution by the soma (depending on its electrotonic proximity). Using the dimensions for peripheral, central, and stem axons from the literature ( [35,45,112]; Fig. 8A), we developed a reduced biophysical model of the DRG portion of an adult, mammalian small-diameter unmyelinated neuron to determine how variations in membrane potential produced by ion channel enhancement or blockade affects AP propagation. Active conductances were limited to a fast VGNC conductance (G Na ) between the activation ranges of Na V 1.7 and Na V 1.8 channels, a delayed rectifier K + conductance (G KDR ), KCNQ current (M channels), and HCN channels. Channel densities (G Na and G KDR ) were adjusted to achieve reliable propagation up to a stimulus frequency of approximately 100 Hz [33]. The conduction velocity of the model neuron was 0.33 m/s, as expected for thin, unmyelinated fibres [33,62]. Somatic APs were 76 mV, somewhat higher than in our recordings of 60-70 mV, but within values reported in the literature [33,79,129]. The goal of the model was to identify those factors limiting AP propagation from the periphery to the spinal cord, rather than to build a highly detailed and complete reconstitution of nociceptive neuron electrical response.
Simulated 30-Hz trains of APs initiated in the peripheral axon, a firing frequency within the upper range for unmyelinated nociceptive fibres [14,108], reliably propagated through the DRG (Fig. 8B 1 ). Potentials at the t-junction had a characteristic waveform; the spike invading the bifurcation was reduced in amplitude and was immediately followed by a larger spike generated in the stem axon. As the AP progressed into the central axon, its amplitude increased with distance from the t-junction. At this stimulus frequency, spike propagation was reliable with M channel current (I M ) densities from 0 and up to 240 pA/pF, while densities of 2.5 to over 20 pA/pF are reported in the literature [17,74,93,100].
We then simulated the effects of enhancing M channels with RTG by increasing I M 1.5-fold and shifting V 1/2 to À60 mV [72,118]. The minimal initial I M density required to produce a failure of spike propagation through the DRG, determined as the loss of at least one action potential during a 30-Hz train of 10 spikes, was 0.9 pA/pF (Fig. 8B 2 ), which is on the lower end of what was reported in the literature [17,74,93,100]. Under RTG conditions (I M density = 1.35 pA/pF and V 1/2 = À60 mV), somatic membrane was hyperpolarized by À3.2 mV. All APs reaching the t-junction were reduced in amplitude, but only 4 of the 10 spikes triggered a regenerative and propagating AP in the central axon. As reported previously, AP generation at the soma was not required for AP propagation from the peripheral axon to the central axon [3]. When I M density was increased to 1.575 pA/pF (initial I M density = 1.05 pA/pF) and V 1/2 was set to À60 mV, somatic E m hyperpolarized to À63.6 mV and AP propagation through the DRG was completely abolished, not only for high-frequency stimulation, but for lower frequencies (<30 Hz) as well (Fig. 8B 2 ). The transition between reliable conduction and failure occurred through a small voltage window, with the hyperpolarization produced by M channel enhancement.
Was AP failure due to hyperpolarization, the increased membrane conductance, or both? To answer this question we tested whether comparable hyperpolarization produced by somatic current injection interfered with AP propagation. Hyperpolarizing the soma by À3.7 mV with constant somatic current (À14 pA) had no effect on the reliability of AP propagation (Fig. 8B 3 ). Only when somatic hyperpolarization was further increased to À64.4 mV (À4.4 mV below E rest ) was spike propagation blocked ( Fig. 8B 3 ). These results suggest that enhancing M channels reduced AP transmission across the model's t-junction by both hyperpolarization and increased membrane conductance and, moreover, support the hypothesis that the physiological effects of RTG observed in vitro (see Fig. 6) are sufficient to account for its effects on AP propagation in vivo.
To further characterize how RTG conditions affect spike propagation across the t-junction, in the model we plotted E m as a function of distance along the peripheral and central axonal segments flanking the t-junction at regular time intervals (0.15 ms) under basal and RTG conditions. Under basal conditions (I M density = 0.9 pA/pF, V 1/2 = À30 mV), as the AP invaded the t-junction, its amplitude decreased to a minimum at the t-junction (Fig. 8C 1 ). A regenerative spike in the central axon was subsequently generated and increased in amplitude with distance from the t-junction. In contrast, under RTG conditions (Fig. 8C 2 ), the spike leaving the t-junction decreased in amplitude and widened with distance in a manner more consistent with the passive spread of potential.
As stated above, the low safety factor for AP propagation at the t-junction depends on the combined contributions of axonal bifurcation and the active properties of the membrane. The importance of the bifurcation is illustrated in Figure 8D 1 , where a spike invading the t-junction segment is plotted with and without a connected stem axon and soma. The presence of bifurcation results in a significant reduction in spike amplitude localized to the t-junction. The second later peak, arising from the regenerative spike in the stem axon, was also reduced. As a result, enhancing M channels in the central axon alone, distal from the t-junction, failed to block AP transmission under retigabine conditions using I M density values of 45 pA/pF and V 1/2 = À60 mV (not shown). Membrane excitability at the t-junction and surrounding axons also influenced the amplitude of the invading spike. For example, varying G Na density affected the spike waveform at the t-junction (Fig. 8D 2 ) and, in turn, the reliability of AP propagation; in general, raising G Na density elevated the safety factor for spike propagation across the t-junction. Na + channel inactivation had little effect on spike propagation. At the resting potential, 17% of the conductance was inactivated. With hyperpolarization, reactivation of the channels and an increased effective G Na was not sufficient to counter the effects of the lower impedance at the t-junction and the effect of hyperpolarization.
Stem axon length and diameter strongly affected the impedance relationship between parent and daughter axons, and as a result, affected AP propagation in the model. With longer stem axons of 200 lm or greater, basal I M density had to be increased to . However, when M channels were restricted to only the soma and stem axon, spike propagation was reliable, as in (E 1 ). Spikes failed when initial I M density was raised to 1.2 pA/pF (E 3 ). Likewise, when RTG enhancement was limited to only the axon segments proximal to the t-junction, spikes reliably propagate into the central axon; initial I M density had to be raised to 5.1 pA/pF to limit spike propagation (E 4 ).
13.5 pA/pF (V 1/2 = À60 mV) in order for M channel enhancement to interfere with AP transmission. When the stem axon had an equivalent diameter to the peripheral axon (0.8 lm), thereby decreasing the conductance ratio [(diam daughter1 ) 3/2 + (diam daughter2 ) 3/2 ]/ (diam parent ) 3/2 [19] from 2.67 to 1.35, the minimal I M density needed to block spikes was 78.3 pA/pF (V 1/2 = À60 mV). M channel subunit Kv7.2 expression has been observed not only in the soma, but also in the stem axon and peripheral fibres [100]. Thus, we examined how the spatial distribution of RTG-enhanced M channels at the soma, stem, and flanking axons affected spike propagation. Starting with an initial I M density of 0.9 pA/pF, as above, RTG modulation of M channels only in the soma and stem axon had no effect on spike propagation. Only when the initial I M density was increased to 1.2 pA/pF did RTG enhancement affect spike propagation (Fig. 8E 1-3 ). Likewise, RTG enhancement of M channels limited to only the peripheral and central axons flanking the t-junction had no effect on spike propagation until the initial density was increased to 5.1 pA/pF (Fig. 8E 4 ), consistent with the differences in surface area. It is important to note that the range of M channel initial densities used in these simulations ($1 to $5 pA/pF) is comparable to, if not less than, the ranges reported in the literature (2.5 to over 20 pA/pF [17,74,93,100]). If M channel densities were indeed higher, RTG enhancement would be even more potent. Lastly, the densities of M channels that blocked propagation were highly dependent on R N and G Na ; with higher R N , fewer M channels were required to achieve comparable hyperpolarization (and vice versa). Likewise, with greater excitability (ie, higher G Na ), larger M channel densities were needed to block spike propagation (not shown).
HCN channels are active at the E rest of DRG neurons in vitro, and pharmacological blockade of these conductances hyperpolarizes the membrane potential. Moreover, when HCN blocker is focally applied to the DRG via cannula, it also reduces the nociceptive response in vivo. Blocking HCN channels in our model DRG neuron also hyperpolarized membrane potential, as expected (Fig. 9A). However, hyperpolarization of À3.98 mV below E rest failed to affect spike propagation. In comparison, hyperpolarization of À3.6 mV produced by RTG enhancement of M channels completely blocked AP signalling. Only when membrane potential was hyperpolarized by at least 4 mV was there interference with AP propagation (Fig. 9B). As with the effect of M channel enhancement, the transition between reliable conduction and failure took place within a small voltage range of hyperpolarization produced by HCN channel blockade.
Blocking HCN channels required a greater magnitude of hyperpolarization than observed with M channel enhancement because it not only hyperpolarizes the t-junction (reducing the safety factor for spike propagation), but also reduces total membrane conductance, elevating the safety factor (but to a lesser degree). The effect on spike waveform at the t-junction and the proximal central axon was comparable to that observed for both M channel enhancement and somatic hyperpolarizing current injection. The model predicts that enhancing M channels might be a more potent inhibitor of nociceptive responses than blocking HCN channels; the possible explanation for this discrepancy is discussed below.

Discussion
Peripheral nociceptive transmission is generally conceptualised as an uninterrupted conduction of peripherally generated APs from their respective sites of origin (eg, skin terminals) to the spinal cord along sensory nerve axons. It has long been recognised that sensory neuron somata are electrically excitable [3,12,[46][47][48]77,111,122], but the cell body and stem axon residing in the sensory ganglia are generally not considered to be important for conduction (eg, [4]). In this study we addressed the following questions regarding somatic excitability of nociceptors and its role in the peripheral nociceptive transmission. 1) what major families of ion channels contribute to the somatic E rest ? 2) How does manipulating somatic/perisomatic E rest impact transmission of peripherally born nociceptive signals in vivo and in silico?
Firstly, we screened for classes of ion channels that influence E rest of nociceptors. We focused on ion channels that are 1) expressed in small-diameter nociceptive neurons and 2) active at or near À60 mV (ie, M channels, 4-AP sensitive K V , K2P ''leak,'' HCN, K ATP , low-threshold Ca V , and Na V channels). We found that manipulation of all these channels affected E rest in varying proportions of small-diameter DRG neurons. However, different channels had different efficacies and prevalence within the neuronal population tested. Comparison of effects presented in Figs. 1 and 2 and Table 3 shows the following ranked sequence for hyperpolarizing manipulations: "M channels ) ;TTX-sensitive VGNC = "K ATP P "K2P P ;HCN > "K Na > ;TTX-resistant VGNC P ;Ca v 3. For compounds producing depolarization, the effects were less graded, with inhibition of M channels, K2P, and 4-AP-sensitive Kv channels having stronger and more prevalent effects. Our current clamp recordings were designed to mimic the effects on the E rest of the acute modulation of ion channel activity by endogenous regulatory molecules (eg, inflammatory mediators, cytokines, hormones) released, for example, during acute inflammation or cancer, and also by peripherally active analgesics (for review, see [26,73,85]). Therefore, our screen has identified the complement of ion channels that would have the largest effect over the nociceptor's somatic E rest during such conditions. M channels were found to have strong influence over E rest , as both inhibition and enhancement of M channels caused $10 mV de-or hyperpolarization, respectively. M channels are physiologically inhibited by G q/11 -coupled protein-coupled receptors (GPCR) such as M 1 mAChR, bradykinin B 2 , protease activated receptor-2, and angiotensin II AT 1 , and such inhibition results in depolarization and increased firing (reviewed in [26,40,105]). In contrast, some G i/o -coupled GPCR, such as somatostatin receptors, increase M current [87,98]. In nociceptors, M channels can be augmented by neurokinin receptors [69,71], an action that results in reduced excitability [71]. Thus, M channels may represent a major endogenous mechanism for tuning the excitability of nociceptive neurons.  in small DRG neurons [26] as well as K2P channels, also strongly influenced E rest ; the K2P channel contribution was significantly higher at 37°C (suggesting large contribution of temperaturesensitive TREK-2 channel, which has low activity at room temperature [52] and is highly expressed in IB4-positive nociceptors [1]). K ATP channels, while present in $70% of nociceptors (as evidenced by the hyperpolarization induced by K ATP enhancer, PIN), were not active at rest, as the K ATP inhibitor GLIB failed to depolarize the E rest . In contrast, K Na channel inhibition produced moderate depolarization ($5 mV) in 55% of the neurons; the K Na enhancer LOX produced only a marginal effect.
Among depolarizing currents, we tested the contribution of HCN channels, T-type Ca 2+ currents, and VGNC. In $60% of the neurons, a modest hyperpolarization (relative to that produced by RTG) was observed with the blockade of HCN channels. Inhibition of TTX-resistant Na + channels produced large ($À10 mV) hyperpolarization of the membrane, but in a very small proportion (9%) of neurons. In the majority of DRG cells, TTX hyperpolarized E rest , but only by À2 to À3 mV. Other cationic or anionic background conductances that may potentially also contribute to E rest (eg, persistent currents through !-aminobutyric acid [GABA] A or Nmethyl-D-aspartate receptors expressed in DRG [66,67,124]) were not addressed in the current study.
Our next question was to establish how modulation of ''resting'' conductances of the somatic and perisomatic compartments of nociceptors affects transmission of sensory information from periphery to the spinal cord. We tested whether compounds that hyperpolarize E m of small DRG neurons interfere with the relaying of APs from the periphery in vivo. Indeed, focal application of 2 K + channel enhancers, RTG and PIN, as well as the I h blocker ZD strikingly attenuated nocifensive behaviour induced by the hind paw injection of BK (Fig. 7). The K NA enhancer LOX, which only produced nominal hyperpolarization in vitro, was without effect.
In order to better understand how these compounds were limiting pain information from reaching the spinal cord, we constructed a computational model of a small-diameter unmyelinated DRG neuron and reached the following conclusions: 1) The morphology of the DRG axon bifurcation based on anatomical measurements [35,45,112] fosters an intrinsically low safety factor for AP propagation that has been observed experimentally in amphibian and embryonic DRG neurons [79,109], as well as adult mammalian C-fibre neurons [33]. An electronically short stem axon, achieved by a short length and/or larger diameter, was essential to the low safety factor. The diameter provides for a larger conductance ratio and both the length and diameter allow potential at the soma to affect potential at the t-junction. Hyperpolarization, produced by enhancing M channels, blocking HCN channels, or somatic current injection, further lowered the safety factor and, as a result, interfered with AP propagation. Increased membrane conductance (induced by somatic/perisomatic K + channel enhancement) additionally contributed to lowering the safety factor. Without a t-junction, the safety factor in the axon is relatively high, and comparable ion channel modulation, for example, in the central axon distal to the t-junction, had no effect on spike transmission. Accordingly, injection of M channel enhancer flupirtine (close analogue of RTG) into sciatic nerve of control rats did not affect nociceptive transmission from the periphery (while similar injection of lidocaine expectedly did) [100]. Recent evidence suggests that the site of analgesic activity of systemically administered RTG is almost exclusively peripheral because, in contrast to its anticonvulsant activity, it was not antagonized by central application of XE [39]. Thus, since the t-junction most likely has the lowest safety factor for AP propagation within the peripheral nociceptive pathway, it is logical to hypothesize that AP failure at nociceptive neuron t-junctions may contribute to the analgesic effect of systemic RTG. 2) The soma of a small-diameter unmyelinated DRG neuron is electrotonically close enough to the t-junction to influence AP transmission. Although excitability (in this case, electrogenesis) at the soma does not normally affect spike propagation [4], hyperpolarization of the soma substantially influences membrane potential at the t-junction. As a result, manipulations that sufficiently hyperpolarize the t-junction interfere with the transmission of APs from the periphery to the spinal cord, again assuming a low safety factor (discussed above).
3) The combination of hyperpolarization and increased membrane conductance should be more potent at blocking spike propagation than either alone. Thus, the efficacy of ZD to attenuate BK-induced pain observed in vivo is not entirely consistent with modelling results. It has been reported recently that I h density was very low in C-fibre nociceptors, but much higher in nociceptive sub-populations of Ad and Ab fibres (5, 13, and 21 pA/pF at À100 mV, respectively) [31]. Notably, in our simulations ( Fig. 9), HCN channel block interfered with AP propagation only at high HCN channel densities ($25 pA/pF), suggesting a possible explanation for the weak effect of ZD on E rest in small nociceptors in vitro and relatively strong effect in reducing peripherally induced pain in vivo: the in vivo effect of ZD is likely to be mediated by Ad and Ab nociceptors. The model demonstrates proof-of-principle that interference with AP propagation, even at low stimulus frequencies, could be achieved by manipulating somatic/perisomatic conductances in DRG [44,75,78].
Importantly, both our in vivo and in silico experiments strongly suggest that somatic/perisomatic compartment of nociceptive neurons has indeed a strong filtering role and may impede incoming APs. Potentially in support of this finding is the recent clinical discovery that electrical stimulation (neuromodulation) of DRG in humans via the implanted electrodes provides efficacious pain relief in patients with various ''untreatable'' neuropathic pain syndromes [20,68,95]. The exact action of such DRG stimulation has yet to be discovered. However, the facts that a) DRG stimulation itself does not cause pain, and b) cessation of stimulation allows pain to return [68], suggests that this type of analgesia arises from the induced failure of peripherally generated APs to pass through the ganglion.
To our knowledge, the experiments presented here are the first to demonstrate that somatic/perisomatic E m can regulate sensory transmission from the periphery to the spinal cord. Thus, the DRG may play a much stronger role in controlling peripheral transmission than generally accepted, representing a hitherto underappreciated additional ''gate'' within the peripheral nociceptive system.

Conflicts of interest
The authors declare that they have no conflicting interests.