Using an engineered glutamate-gated chloride channel to silence sensory neurons and treat neuropathic pain at the source

See Basbaum (doi:10.1093/brain/awx227) for a scientific commentary on this article. Following nerve injury, sensory neurons become hyperexcitable and this is a key driver of neuropathic pain. Weir et al. develop a gene therapy approach that silences sensory neurons and reversibly normalizes pain thresholds in animal models of neuropathic pain. This new approach has significant translational potential.


Introduction
Neuropathic pain arises as a consequence of a lesion or disease of the somatosensory nervous system and affects 7-10% of the general population (Colloca et al., 2017). Many of the most common causes of neuropathic pain such as post-herpetic neuralgia, diabetic polyneuropathy and traumatic nerve injury, relate to injury to sensory neurons within the peripheral nervous system. A key pathophysiological driver for the induction and maintenance of neuropathic pain in such conditions is the development of hyperexcitability of sensory neurons, which develop ectopic activity (Wall and Gutnick, 1974;Wall and Devor, 1983;Kajander and Bennett, 1992;Boucher et al., 2000;Wu et al., 2001) and an abnormal response to natural stimuli (Shim et al., 2005;Smith et al., 2013). Such abnormalities have been demonstrated in multiple animal models of peripheral nerve injury (Kajander and Bennett, 1992;Boucher et al., 2000;Wu et al., 2001) as well as in patients suffering from peripheral neuropathic pain through the use of microneurography (Serra et al., 2012). Ectopic activity develops in small diameter unmyelinated and large diameter myelinated sensory axons, and arises at the site of the injured axon and at the level of the cell body (Wall and Devor, 1983). This hyperexcitability occurs due to profound changes in the expression as well as altered trafficking and post-translational modifications of ion channels within sensory neurons (Waxman and Zamponi, 2014).
Studies on patients using local anaesthetic blocks suggest that abnormal afferent input from sensory neurons is required for the maintenance of peripheral neuropathic pain (Haroutounian et al., 2014;Vaso et al., 2014). Systemic lidocaine reduces ectopic activity in animal models of nerve injury (Devor et al., 1992); however, dose escalation in humans is limited by motor, cardiac and CNS side effects. Similarly, the oral analgesics currently available that target aberrant afferent input have limited efficacy and poor tolerability (Finnerup et al., 2015).
Chemogenetics provides opportunities for a novel treatment approach to neuropathic pain by silencing aberrant primary afferent input. This technique enables precise control of neuronal activity through expression of receptors/ ion channels that have been engineered to be solely activated by a non-toxic drug (Sternson and Roth, 2014). This allows for temporal and spatial control of activity, determined by the pattern of expression of the receptor and timing of drug administration. From a translational perspective, a key attribute of the chemogenetic approach is that it enables non-invasive and chronic modulation of neuronal activity. A heteromeric glutamate-gated chloride channel (GluCl v1.0) modified to respond to low doses of ivermectin but not glutamate, has previously been shown to silence activity of mammalian CNS neurons in vivo for several days following a single dose of ivermectin (Lerchner et al., 2007;Lin et al., 2011). Effectiveness was limited by difficulties of ivermectin crossing the blood-brain barrier and suboptimal b-subunit trafficking (Lerchner et al., 2007). We therefore used a recently optimized version of GluCl (GluCl v2.0), wherein ivermectin sensitivity and effective trafficking of the b-subunit have been increased (Frazier et al., 2013) with the aim of silencing peripheral sensory neurons.

Materials and methods
Animals  wild-type C57BL/6 mice (University of Oxford Breeding Unit) of both genders were used for in vitro culture experiments including electrophysiology, somal Ca 2 + imaging and immunocytochemistry. Adult male (6-8 weeks old at time of intrathecal surgery) C57BL/6 mice were used for all behavioural experiments. For microfluidic experiments, embryonic Day 16 Sprague-Dawley rat embryos (Charles River) were used. Animals were group-housed in specific pathogen-free conditions on a 12:12 h light-dark cycle at 23 AE 1 C with free access to food and water. All behavioural testing was performed during the light phase. Animals were assigned to treatment groups by randomly drawing them from their housing cages and rehoused post-treatment in mixed cages. All procedures complied with the UK Animals (Scientific Procedures) Act (1986) and were performed under a UK Home Office Project Licence in accordance with University of Oxford Policy on the Use of Animals in Scientific Research. This study conforms to the ARRIVE guidelines (Kilkenny et al., 2013).

Surgical procedures
Intrathecal delivery of AAV was performed on 6-8-week-old mice (18-25 g) under 2% inhaled isoflurane anaesthesia. The dura was exposed by careful dissection of soft tissue between T10 and T11 vertebrae and punctured with a 30 G needle, before a small cannula (polyetrafluoroethylene tubing, 0.008" outer diameter Â 0.004" internal diameter, Braintree Scientific) was inserted 1 cm caudally into the subarachnoid space. Eight microlitres of AAV [4 ml of neat GluClb-AAV with 4 ml of neat GluCla-AAV, or phosphate-buffered saline (PBS)] was then infused at a rate of 1 ml/min. Following a 1-min rest period to limit backflow, the cannula was slowly removed and the animal recovered with the administration of local (2 mg/kg Marcain, AstraZeneca) and systemic (5 mg/kg Rimadyl, Pfizer) postoperative analgesia. The time between intrathecal infusion and animal sacrifice to assess viral transduction or the commencement of behavioural analysis was at least 3 weeks.
For the spared nerve injury (SNI) model we ligated and transected the tibial and common peroneal branches of the sciatic nerve, leaving the sural nerve intact (Decosterd and Woolf, 2000). Animals received postoperative analgesia as detailed above and were assessed behaviourally from 3 days post-SNI in addition to being monitored for self-mutilation (autotomy) daily. Autotomy of two or more nails/toes led us to sacrifice the animal. Any animals that did not show a 50% reduction in mechanical thresholds by 7 days post-SNI were excluded from further assessment and were not included in the final analysis. The presence of autotomy or the lack of mechanical hypersensitivity led us to exclude 2/30 animals. For whole-cell patch clamp experiments L3, L4 and L5 dorsal root ganglia (DRG) were pooled from animals that had undergone SNI surgery 3-5 weeks previously to generate DRG cell cultures (see below).

Behavioural testing
All behavioural studies were performed by an experimenter blind to treatment group and at a consistent time of day. Sample sizes were chosen based on similar experiments in the literature (thermal place preference experiments; Bautista et al., 2007), or calculated using standard power calculations based on our own previous studies (statistical power of 80% and avalue 5 0.05). For assessment of mechanical and thermal thresholds the minimum group size required to show a 25% change was 9 and the group size required to detect a 25% rescue in mechanical sensitivity following SNI was 8. Application of von Frey hair filaments (Stoelting) to the plantar surface of the hind paw using the 'up-down' method was used to assess mechanical sensitivity. Prior to testing, mice were acclimatized for 1 h on an elevated wire mesh floor and housed in 8 Â 5 Â 10-cm partitions. The 50% paw-withdrawal threshold was calculated as per Dixon (1980) on three different days and the average taken as the baseline value. Thermal thresholds were assessed using an infrared light source applied to the plantar surface of each hind paw (Hargreaves test). Latency to withdraw was tested three times per paw and averaged. All baseline data represent the average score of three separate days of testing. Except in the case of nerve injury, all sensory threshold data represent the average of right and left hind paws. Noxious mechanosensation was assessed by the pinprick test described previously (Arcourt et al., 2017). Mice were housed and acclimatized similarly to the von Frey test. A pinprick stimulus was applied to the plantar surface of the hind paw by a dissecting pin attached to a 1 g von Frey filament. Latency to withdraw was recorded by a Sony Xperia phone at 120 fps (8.33 ms per frame) and analysed by an investigator blind to treatment groups. Three measurements were taken for each hind paw per trial and averaged. Plotted latency represents the average of both paws. Baseline data represent the average of two trials performed on separate days.
Motor performance was assessed by the rotarod performance test (Ugo Basile, 47600). Mice were scored for their latency to fall from a horizontal cylinder rotating at 32 rpm. Animals were acclimatized to the equipment several days prior to testing by placing them on the cylinder; first while it was stationary and then while rotating. Following acclimatization, two baseline measurements were taken on separate days and averaged to generate baseline data. Each data point represents one trial performed on a given day. The Beam test (Carter et al., 2001) apparatus consisted of a 1-m long horizontal beam of 12-mm width suspended from the ground. Mice were placed at one end of the beam next to a light source and walked across the beam towards a darkened box. Trials were recorded once per day and video footage was used by an investigator blind to treatment groups, to assess number of mistakes (defined as missed hind paw placements or a hop). Mice were acclimatized to the apparatus prior to testing for 10 min. Baseline data were generated from the average of two trials on separate days.
Thermal place preference was performed with the use of two chambers with aluminium plates [in cm, 25 (w) Â 37 (d) Â 47 (h); Ugo Basile] as the base, connected by a short passage. One plate was maintained at room temperature (22 C) and the other was set to 16 C. Assigning temperatures to each plate was performed at random to prevent animal learning. The percentage time spent at each temperature was live-scored over a 10-min period. Animals were acclimatized to the equipment several days prior to testing by placing them on the plates at room temperature for 10 min. Following acclimatization, two baseline measurements were performed on separate days and averaged to generate baseline data.

Immunohistochemistry
For tissue collection, animals were deeply anaesthetized with pentobarbital and perfused through the vascular system with ice-cold 0.9% NaCl solution until tissues were sufficiently cleared of blood. L4 DRGs, nodose ganglia, superior cervical ganglia, lumbar sympathetic chain, hind paw glabrous skin and peripheral nerves were collected and post-fixed for 1-2 h in 4% paraformaldehyde (PFA). Lumbar spinal cord and brain tissue were collected simultaneously and post-fixed in 4% PFA overnight. Post-fixation, all tissue was stored in 30% sucrose for a minimum of 48 h before being embedded and sectioned by a cryostat. DRG, skin and nerve were sectioned at 15 mM, while brain and spinal cord were sectioned at 30 mM. Cell cultures were fixed at room temperature for 10 min in 4% PFA and from there on treated similarly to tissue sections.
Dorsal root ganglia neuron culture and electroporation DRG were rapidly excised (Sleigh et al., 2016) and enzymatically digested at 37 C for 80 min in dispase type II (4.7 mg/ml) plus collagenase type II (4 mg/ml) (Worthington Biochemical). DRG from all levels were used except for experiments examining biophysical properties following SNI wherein only L3, L4 and L5 tissues were harvested. Mechanically dissociated cells were plated onto laminin/poly-D-lysine (BD Biosciences) treated coverslips in complete Neurobasal Õ medium [Neurobasal Õ media supplemented with 2% (v/v) B27 and 1% (v/v) penicillin streptomycin (ThermoFisher Scientific)]. Mouse nerve growth factor (50 ng/ml; NGF, PeproTech) and 10 ng/ml glial-derived neurotrophic factor (GDNF, PeproTech) were added to the media and cultures used for experiments 24-72 h later. For cultures generated from GluCl AAV animals, neurons were plated and maintained in DMEM-F12 with 10% foetal calf serum (ThermoFisher Scientific) instead of complete Neurobasal Õ medium and used for experiments 4-24 h after plating. Electroporation was performed prior to plating using the Neon Õ transfection system (ThermoFisher Scientific). Dissociated neurons were resuspended at 5 Â 10 6 cells/ml in 10 ml Buffer R with 1 mg of total plasmid DNA. The electrical protocol applied was three 1500-V pulses of 10-ms duration. After electroporation, cells were immediately plated as described above and used 24-72 h later, when expression of the plasmid DNA was maximal.
Human embryonic kidney cell culture HEK 293 cells were routinely subcultured in Dulbecco's modified Eagle medium (DMEM; ThermoFisher Scientific) with 10% foetal calf serum. Cells were plated in six-well dishes 24 h prior to transfection with Lipofectamine TM 3000 (ThermoFisher Scientific) and plasmid DNA (total of 1 mg DNA). Twenty-four hours post-transfection, cells were replated onto coverslips at low density, before being used for patch clamp experiments 24-48 h later.

Derivation of sensory neurons from induced pluripotent stem cells
The control induced pluripotent stem cell (iPSC) line AD2-1 was derived from commercial fibroblasts (Lonza, CC-2511) and reprogrammed using the CytoTune Õ iPS Reprogramming Kit (ThermoFisher Scientific). Quality control checks of this line included: tests for Sendai virus clearance, fluorescence-activated cell sorting (FACS) for pluripotency markers, genomic integrity checks and embryoid body tri-lineage differentiation experiments. For differentiation, iPSCs were passaged onto Matrigel Õ -coated six-well plates using TrypLE express (ThermoFisher Scientific) and maintained in mTeSR1 supplemented with 10 mM ROCK inhibitor (ScienCell). Twenty-four hours after plating, the medium was exchanged to mouse embryonic fibroblast (MEF) conditioned medium (ScienCell) supplemented with 10 ng/ml human recombinant FGF2. Cells were allowed to expand on MEF-conditioned medium until $50% confluent, at which time differentiation was started according to Chambers et al. (2012). Briefly, medium was exchanged to knockout serum replacement (KSR) medium containing; knockout-DMEM, 15% knockout-serum replacement, 1% GlutaMAX TM , 1% non-essential amino acids, 100 mM bmercaptoethanol, 1% antibiotic/antimycotic (ThermoFisher Scientific), supplemented with the SMAD inhibitors SB431542 (Sigma, 10 mM) and LDN-193189 (Stratech, 100 nM). The medium was gradually transitioned from KSR medium to N2 medium (Neurobasal Õ medium, 2% B27 supplement, 1% N2 supplement, 1% GlutaMAX TM , 1% antibiotic/antimycotic) (ThermoFisher Scientific) over an 11-day period. On Day 2, the small molecules CHIR99021 (Apollo Scientific, 3 mM), SU5402 (R&D Systems, 10 mM) and DAPT (Sigma, 10 mM) were also added. SMAD inhibitors were removed from the media from Day 6 onwards. On Day 11, the now immature neurons were replated onto Matrigel Õ -coated coverslips (25 000 cells per 13 mm coverslip) in 100% N2 medium containing human recombinant NGF, GDNF, BDNF, NT3 (all at 25 ng/ml, PeproTech) and 10 mM ROCK inhibitor. CHIR99021 (3 mM) was included in the medium until Day 14, and laminin (1 mg/ml, ThermoFisher Scientific) was supplemented into the medium from Day 20 onwards. Medium changes were performed twice weekly after replating. Cytosine b-D-arabinofuranoside (AraC, 2 mM, Sigma) was included in the medium for 24 h following replating to remove the few non-neuronal dividing cells remaining in the culture. This differentiation resulted in a pure neuronal culture with extensive arborized neurites by 2-3 weeks after the end of the small inhibitor stage. Neurons were matured for 4 weeks before addition of AAV (GluClb-YFP-AAV with or without GluCla-mCerulean-AAV) at a multiplicity of infection (MOI) of 10 5 . All patch clamp and immunocytochemistry experiments were performed 1-3 weeks post-virus treatment.

Electrophysiology
Whole-cell patch clamp recordings were performed at room temperature (22 C) using an Axopatch 200B amplifier and Digidata 1550 acquisition system (Molecular Devices). GluCl + DRGs were defined by the presence of YFP and detected with an Olympus microscope with an inbuilt GFP filter set (470/40Â excitation filter, dichroic LP 495 mirror and 525/50 emission filter). Data were low-pass filtered at 2 kHz and sampled at 10 kHz. Series resistance was compensated 60-80% to reduce voltage errors. Patch pipettes of 2-6 M tip resistance were pulled from filamental borosilicate glass capillaries (1.5 mm outer diameter, 0.84 mm inner diameter; World Precision Instruments) and filled with internal solution containing (mM): 100 K-gluconate, 28 KCl, 1 MgCl 2 , 5 MgATP, 10 HEPES, and 0.5 EGTA; pH was adjusted to 7.3 with KOH and osmolarity set to 305 mOsm. For experiments of high intracellular Cl À , KCl was increased to 78 mM while K-gluconate was reduced to 50 mM. Extracellular solution contained (mM): 140 NaCl, 4.7 KCl, 1.2 MgCl 2 , 2.5 CaCl 2, 10 HEPES and 10 glucose; pH was adjusted to 7.3 with NaOH and osmolarity was set to 315 mOsm. Extracellular solution was perfused at a continuous rate of 1-2 ml/min. To assess voltage gated Na + currents in iPSC-sensory neurons, the internal solution contained (mM): 140 CsF, 10 NaCl, 1 EGTA and 10 HEPES. The extracellular solution contained (mM): 70 NaCl, 50 N-methyl-D-glucamine, 20 tetraethylammonium chloride, 1 CaCl 2 , 3 KCl, 1 MgCl 2 , 1 CaCl 2 , 10 HEPES, 10 glucose and 0.1 CdCl 2 . Ivermectin (Sigma-Aldrich) was diluted to a 10 000Â stock in dimethyl sulphoxide (DMSO) fresh daily and added via the perfusion system. Unless otherwise stated, post-ivermectin recordings were always made 15 min after addition of the drug. In voltage clamp, ramps from À90 mV to À40 mV were used to measure membrane conductance. Ramps of 100 ms duration were delivered in 10 s intervals and conductance was derived from the linear portion of the resultant current slope. To assess voltagegated Na + currents potential was stepped to À85 mV for 8 s before a 20 ms test step to 0 mV, from a holding potential of À100 mV. Resting membrane potential was assessed in bridge mode, while firing properties were assessed in current clamp mode. Input resistance (R Input ) was derived from the membrane potential deflection caused by a 20 pA hyperpolarizing current pulse at À60 mV. To determine rheobase, cells were depolarized from a holding potential of À60 mV by depolarizing current steps (50 ms) of increasing magnitude (Á25 pA) until an action potential was generated. Spontaneous activity was defined as the presence of action potential firing during a 1 min bridge-mode recording sweep made immediately upon establishing whole cell access. Response to 10 s application of 1 mM capsaicin was assessed in bridge mode while recording from small diameter (525 mM) neurons.

Ca 2 + imaging
For ratiometric Ca 2 + imaging, cells were incubated with media containing 2 mM Fura-2 AM and 80 mM pluronic acid (ThermoFisher Scientific) for 1 h at 37 C, before being washed with extracellular fluid (in mM, 145 NaCl, 5 KCl, 10 HEPES, 10 D-glucose, 2 CaCl 2 and 1 MgCl 2 , pH 7.4). For somal imaging, drug (vehicle or ivermectin) was added during the Fura-2 AM and pluronic acid incubation step. A 10 Â objective, dichroic LP 409 mirror and BP 510/90 emitter filter were used for Ca 2 + imaging. Pairs of images using excitation filters BP 340/30 and BP 387/15, respectively, were captured every 2 s. Ratiometric 340/380 calculation was performed with a background subtraction. GluCl + DRG were identified using BP 480/30 excitation filter, dichroic LP 505 mirror and BP 535/40 emission filter and labelled as such prior to recording protocols. Extracellular fluid was perfused with the addition of drugs by a gravity-driven application system with remotely controlled pinch valves. Therefore, the entire stimulation protocol and ivermectin application was performed in a time-locked and hands-free manner. Cells exhibiting a time-locked Ca 2 + transient 40.2 ÁF/F were defined as having responded.
Compartmentalized microfluidic chambers were composed of a polydimethylsiloxane (PDMS) piece in which a pattern of channels and microgrooves were moulded that was bonded to a glass bottom culture dish. The fabrication of the master template in which the PDMS pieces were moulded from was performed as previously described in detail (Park et al., 2006). Embryonic Day 16 rat DRG were processed for culture similarly to adult mouse tissue but with a reduced enzymatic digestion step of 45 min. We used rat DRG for this experiment as in our hands they grow through the microgrooves better and more consistently generate somal responses to terminal stimulation. Once dissociated embryonic rat DRG were resuspended in complete Neurobasal Õ medium and 10 000 cells were plated in a 4 ml volume in only one compartment using capillary action to draw the cell suspension along the entire compartment length. Ten microlitres of complete Neurobasal Õ medium was placed in the adjacent compartment. Cells were left to adhere for a minimum of 2 h before flooding the channels with complete Neurobasal Õ medium. To encourage axonal growth through the microgrooves we exploited the fluidic resistance to create a unidirectional NGF/GDNF gradient across the array. This was established according to Table 1. DRG neurons were treated with AAV (GluClb-AAV with or without GluCla-AAV) 3 days post-plating at an MOI of 100. Imaging experiments were performed 7 days post-plating. To identify neurons that had grown a neurite through the 5 ng/ml NGF, 1 ng/ml GDNF in 160 ml 50 ng/ml NGF, 10 ng/ml GDNF in 200 ml Day 2 0 ng/ml NGF, 0 ng/ml GDNF in 160 ml 50 ng/ml NGF, 10 ng/ml GDNF in 200 ml Days 3 and 5 0 ng/ml NGF, 0 ng/ml GDNF in 160 ml 50 ng/ml NGF, 10 ng/ml GDNF in 200 ml microgroove array, 100 ml of 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (DiI, Invitrogen, 40 mg/ml in complete Neurobasal Õ medium) was applied to the axonal compartment 2 h prior to Ca 2 + imaging, with 200 ml of medium in the somal compartment to maintain fluidic isolation. Neurons that were positive for DiI and GluCl subunits (GluCla and GluClb, or GluClb only) were identified and subsequently used to assess Ca 2 + response. Averages from one litter were used to compute group means and measures of group variance. Statistical comparisons were always made on the basis of litter number.

Statistical analysis
Data are expressed throughout as mean AE standard error of the mean (SEM

Evoked firing of sensory neurons is silenced by GluCl activation
To assess the efficacy of GluCl for silencing sensory neurons, we first performed whole-cell patch clamp recordings of dissociated mouse DRG neurons electroporated with GluCl plasmid DNA. At rest, neurons expressing both aand b-subunits of GluCl (Fig. 1A) had similar membrane and firing properties to control (neurons electroporated with GFP-only vector) (Supplementary Table 1), suggestive of no GluCl activity. However, low levels of ivermectin induced a slow, stable and dose-dependent conductance ( Fig. 1B and C). Neurons expressing only the b-subunit of GluCl did not respond to ivermectin, consistent with previous observations that this subunit does not form functional ivermectin-sensitive homomers (Frazier et al., 2013). This validated our use of GluCl b-only as a suitable control in future experiments. Unlike CNS neurons, sensory neurons maintain a high intracellular Cl À concentration ([Cl À ] i ) (Gilbert et al., 2007) and as such, initiating a Cl À conductance would not be expected to inhibit activity by the classical fashion of hyperpolarizing the membrane potential (V m ). We reasoned, however, that the large drop in membrane resistance associated with GluCl activation (Fig. 1D) would be sufficient to inhibit neuronal activity. Indeed under our recording conditions ([Cl À ] i = 30 mM) GluCl activation potently reduced evoked firing. After 15 min treatment with 20 nM ivermectin, 95.83% (23/24) of GluCl + neurons did not generate an action potential in response to a current stimulus 10-fold the pre-ivermectin rheobase value, while control neurons were unaffected ( Fig. 1E and F). The transient opening of Cl À channels can depolarize and activate sensory neurons (Cho et al., 2012). However, we noted no action potential firing in the 15 min prelude to assessment of evoked firing in GluCl + neurons (5 nM ivermectin n = 13 cells, 20 nM ivermectin n = 13). Cells recorded from ranged in diameter from 15 to 40 mm and silencing was effective in all sizes of neuron ( Supplementary Fig. 1). DRG [Cl À ] i levels are heterogeneous (Gilbert et al., 2007) and can be increased after nerve injury (Pieraut et al., 2007), altering the polarity and drive of Cl À flux. Under recording conditions of high intracellular Cl À ([Cl À ] i = 80 mM, E Cl À = À16 mV) ivermectin depolarized V m , but was still effective at silencing neuronal activity ( Fig. 1G and Supplementary Fig. 2). These results indicate that even when providing a depolarizing drive, GluCl activity can silence evoked electrical activity of sensory neurons.
GluCl silencing is effective at the axon as well as the soma To circumvent confounds of whole-cell patching and artificial modulation of intracellular ionic concentrations; we performed Ca 2 + -imaging experiments. The proportion of GluCl + neurons responding to KCl-induced depolarization and naturally occurring TRP channel agonists [menthol (McKemy et al., 2002) and capsaicin (Caterina et al., 1997)] following ivermectin was markedly reduced compared to control, indicative of effective silencing ( Fig. 2A). The fact that the Ca 2 + response to capsaicin was not completely abolished likely represents the direct entry of Ca 2 + through TRPV1 (a non-selective cation channel). This notion is supported by our finding that GluCl + neurons did not fire action potentials in response to capsaicin, following 20 nM ivermectin ( Supplementary Fig. 3). Whether GluCl activation silences neurons solely at the level of the soma or also blocks axonal action potential propagation has not previously been determined; but the question is highly relevant to sensory neurons given their long axonal projections. We used microfluidic compartments to separate the soma and neurites of GluCl + neurons ( Fig.  2B) (Park et al., 2006). Local depolarization of neurites leads to Ca 2 + transients in DRG cell bodies dependent on axonal transmission (Fig. 2C). Ivermectin applied only to neurite endings was able to almost entirely block soma responses to neurite depolarization ( Fig. 2D and E), suggesting that GluCl activation can interfere with primary afferent transmission at the level of the axon and not just the soma. The in vitro data illustrate that GluCl activation can potently silence sensory neurons and compelled us to assess whether the system was applicable to the in vivo setting.

Intrathecal AAV effectively delivers GluCl to sensory neurons in vivo
Following nerve injury, evidence exists that both myelinated (Wall and Gutnick, 1974;Wall and Devor, 1983;Kajander and Bennett, 1992) and unmyelinated afferents (Wu et al., 2001;Serra et al., 2012) are involved in the generation of neuropathic pain (Ossipov et al., 1999;Tarpley et al., 2004;Xu et al., 2015). We therefore aimed to deliver GluCl to a broad range of sensory neurons in vivo. We engineered AAV9 to express GluCla and GluClb separately; under the control of the ubiquitous CAG promoter and modified GluCla to contain a monomeric cerulean fluorophore in the place of YFP to better visualize subunit co-localization ( Supplementary Fig. 4). Fluorophore switch had no effect on channel function ( Supplementary Fig. 4). We used intrathecal delivery to selectively transduce neurons of the sensory ganglia while avoiding transduction of spinal cord neurons (Mason et al., 2010;Jacques et al., 2012). Four weeks post-surgery, GluCl was expressed in 66.1 AE 9.6% of L4 DRG neurons (n = 4 animals with one excluded from analysis due to zero expression) (Fig. 3A). Size frequency analysis of GluCl + neurons revealed equally distributed transduction of neurons across all diameters (Fig. 3B). Consistent with this result, co-labelling experiments showed that transduction rates were similar across different molecularly defined sub-populations ( Fig. 3C and D) and was not selective for a particular population. Homomeric GluCl channels provide little or no conductance in response to ivermectin (Frazier et al., 2013); however, we found that the vast majority (78.9%) of GluCl + neurons expressed both subunits ( Supplementary Fig. 5). GluCl + afferent endings were seen in superficial and deep dorsal horn laminae but crucially no expression was seen in ventral horn (G) Excitability of GluCl + DRG assessed while using an intracellular solution containing 80 mM Cl À . (Left) Example firing of a GluCl + neuron in response to supra-threshold current stimuli pre-and post-ivermectin (20 nM). (Right) Following ivermectin, 5/5 neurons were unable to fire an action potential in response to a current 10-fold their pre-drug rheobase value (full silencing). All grouped data are mean AE SEM, *P 5 0.05, **P 5 0.01, ***P 5 0.001; P 5 0.0001.
motor neurons (Fig. 3E), brain or autonomic structures ( Supplementary Fig. 6). In the periphery we observed GluCl expression in neurite endings of the skin and deep dermal nerve fibres ( Supplementary Fig. 7). Channel activity remains for several days post-ivermectin dosing due to slow dissociation from the channel (Slimko et al., 2002) and tissue retention of ivermectin (Lerchner et al., 2007). This allowed us to test silencing efficacy following AAV delivery and agonist treatment in vivo. GluCl-AAV animals were dosed with 5 mg/kg ivermectin i.p. 24 h prior to dissociation of lumbar DRG. Ex vivo GluCl + DRG neurons exhibited analogous features to those observed in vitro; a substantive basal conductance and an inability to generate action potentials in response to current injection ( Fig. 3F and G). Neurons did not exhibit a significantly different resting membrane potential (b-only naïve, À59.39 AE 2.23 mV versus ab ivermectin, À56.01 AE 1.9, P = 0.18 Student's t-test.) AAV is considered a suitable vehicle to achieve long-term expression of exogenous transgenes in target tissue (Hadaczek et al., 2010), which makes it a highly attractive approach from a translational perspective.
However, there are previous reports of partial shutdown of transgene expression over time when AAV is used to target DRG (Mason et al., 2010;Iyer et al., 2014). To assess expression longevity, we maintained a small cohort of GluCl-AAV animals for 7 months and found robust expression still evident ( Supplementary Fig. 8). We noted no reduction in cutaneous afferent innervation or upregulation of the injury marker ATF3 (Supplementary Fig. 9).
Motor performance and proprioception (as assessed by the rotarod and beam walk tests, respectively) were not affected in either experimental or control groups following ivermectin ( Fig. 4D and E). We investigated in detail the time course of GluCl-mediated behavioural responses. After a single dose, hypo-responsiveness was apparent by 24 h and was maintained for 3 days before thresholds returned to baseline by 6 days (Fig. 4F). Once GluCl channels have closed following ivermectin washout they should be amenable to reactivation by a subsequent dose of the drug (Lerchner et al., 2007). Indeed, we found that a second ivermectin administration had similar efficacy to the first after an interval of 3 weeks (Fig. 4G). These results highlight GluCl silencing of DRG neurons as a means to non-invasively manipulate sensory thresholds in a chronic and reproducible fashion.

GluCl activation reverses traumatic nerve injury-induced hypersensitivity
Chronic pain following nerve injury is driven by hyperexcitability (Smith et al., 2013) and ectopic activity of primary afferents (Haroutounian et al., 2014;Vaso et al., 2014). We asked whether our approach of long-term electrical silencing of afferents would be sufficient to counter this phenomenon in animals that had undergone SNI (Decosterd and Woolf, 2000). We found that GluCl activation could repetitively reverse established mechanical hypersensitivity following SNI ( Fig. 5A and Supplementary Fig. 10). Post hoc tissue analysis revealed that the degree of rescue correlated well with the proportion of L4 DRG cells expressing GluCl (Fig. 5B), supporting the view that primary afferent drive is critical for the maintenance of the neuropathic pain state. We used a two-chamber cold place preference as an operant measure of thermal hypersensitivity following nerve injury (Balayssac et al., 2014). One week after SNI, animals spend significantly less time in the cold (16 C chamber versus 22 C) than pre-injury levels. Ivermectin reversed the observed cold allodynia in the experimental group but had no effect on controls (Fig. 5C). As a cellular correlate, we studied DRG neurons taken from chronic SNI animals (25-33 days post-SNI) for their electrophysiological properties in vitro. Ex vivo neurons derived from injured DRG had a higher incidence of ectopic activity than control [ipsilateral (8/24 cells) versus contralateral (1/20 cells), P = 0.027 Fischer's exact t-test]. However, injured neurons from ivermectin-treated GluCl + animals exhibited no ectopic activity (Fig. 5D) and largely silenced evoked activity (Table 2), indicative of robust silencing. GluCl activation is therefore sufficient to silence nerve-injury induced ectopic activity of primary afferents and reverse pain related behaviour.

GluCl silences human iPSC-derived sensory neurons
The use of human model systems can potentially improve translation of preclinical findings in animal models into clinical therapeutics (Woolf, 2010;Percie Du Sert and Rice, 2014). To address whether the GluCl system may have efficacy in the context of human sensory neurons, we generated cultures of human iPSC-derived sensory neurons. We followed a differentiation protocol known to produce neurons with an expression profile and functional characteristics that are very similar to rodent sensory neurons (Chambers et al., 2012;Young et al., 2014;Clark et al., 2017). Resultant neurons uniformly express the sensory neuron marker Brn3a, project extensively arborized neurites and exhibit mature electrophysiological characteristics ( Fig. 6A-C). Analogous to mouse DRG, human iPSC-sensory neurons expressing GluCl and treated with ivermectin show a substantial conductance (Fig. 6D-F) and limited capacity to generate action potentials to otherwise suprathreshold current stimuli ( Fig. 6G and H).
These results indicate the generality of the GluCl system and its potential relevance for use in human sensory neurons.

Discussion
To summarize, we show that activation of GluCl channels results in robust silencing of rodent and human iPSCderived sensory neurons. AAV-mediated delivery is capable of highly specific, efficient and long-lasting expression of GluCl in sensory neurons in vivo without deficits in cutaneous afferent innervation or upregulation of the injury marker ATF3 and can normalize pain-related behaviour following peripheral nerve injury in a reproducible and reversible fashion. The GluCl channel that we used in this study is a heteromeric glutamate-gated chloride channel modified so be activated by low doses of ivermectin, but is insensitive to its natural ligand glutamate (Slimko et al., 2002). The initial version (GluCl v1.0) has previously been used to silence neurons in the CNS (Lerchner et al., 2007;Lin et al., 2011). In this study we used a recently modified version (GluCl v2.0), which shows higher sensitivity to ivermectin and improved trafficking of the b-subunit (Frazier et al.,  (H) Neuronal excitability as defined based on a rheobase value; 53 (excitable), 3-10 (partial silencing) or 410 (full silencing) standard deviations above the mean of control (vehicle treated b-only, mean = 103 AE 28.2 pA). P 5 0.001, chi-squared test ( 2 = 32.45 with 6 degrees of freedom); vehicle b-only, n = 9; vehicle ab, n = 11; ivermectin b-only, n = 11; and ivermectin ab n = 13. All grouped data are mean AE SEM, ***P 5 0.001.

2013
), but has (up until now) not been tested in vivo or in the peripheral nervous system. We were struck by the efficacy of GluCl v2.0 in silencing rodent sensory neurons both in vitro and ex vivo. In the absence of ligand, neurons expressing GluCl were electrically normal, but treatment with low levels of ivermectin evoked a substantive membrane conductance that rendered almost all neurons entirely resistant to evoked firing. Systems designed to inhibit neuronal activity are classically designed to hyperpolarize the membrane potential (Lechner et al., 2002;Slimko et al., 2002;Armbruster et al., 2007) and render the cell less likely to reach firing threshold. In CNS neurons, Cl À has a hyperpolarized Nernst potential due to low [Cl À ] i levels and therefore an increased conductance will hyperpolarize the resting membrane potential (Lerchner et al., 2007). DRG neurons are unique in that they actively maintain higher [Cl À ] i through activity of several Cl À co-transporters (Sung et al., 2000). Consistent with this we found that GluCl activation did not hyperpolarize the membrane potential; however, we still noted substantial silencing efficacy. This may be explained by the marked reduction in membrane resistance associated with GluCl activation, resembling the phenomena of shunting inhibition observed in endogenous neural systems (Segev, 1990;Borg-Graham et al., 1998). Despite this, the precise mechanism of silencing would prove challenging to demonstrate empirically.
Ca 2 + influx to the soma in response to natural stimuli [menthol and capsaicin, which activate subsets of small diameter DRG cells (Caterina et al., 1997;McKemy et al., 2002)] was reduced in GluCl-activated neurons. Given the long pseudo-unipolar axons of sensory neurons, there are advantages to a silencing system that can act on neurites as well as at the level of the soma. Using a microfluidic platform in which soma and neurites can be isolated and treated independently, we show that ivermectin application to sensory neurites alone is sufficient to prevent action potential propagation to the soma following peripheral depolarization.
With a view to translational applications and to test whether there were cross-species differences in silencing efficacy, we investigated whether the GluCl system could silence human iPSC-derived sensory neurons. We used a previously published differentiation protocol known to generate sensory neurons (Chambers et al., 2012) that has been shown to be a valuable model of human pain disorders (Young et al., 2014;Cao et al., 2016). Ivermectin treatment of human iPSC-derived sensory neurons expressing GluCl evokes a membrane conductance and silences these neurons to a similar extent as their rodent analogues.
Ivermectin does not readily cross the blood-brain barrier (Schinkel et al., 1994), leading previous studies of the GluCl system in the CNS to use relatively high dosing regimes (Lerchner et al., 2007). When administering ivermectin in vivo we opted to use the lower end of the dose range (5 mg/kg) as our target was peripheral neurons. This lower dose reduces the likelihood of off-target effects given that at high doses ivermectin can bind to GABA A receptors (Dawson et al., 2000). Doses of ivermectin 510 mg/kg have been shown not to have CNS depressant effects (Trailovic and Nedeljkovic, 2011). Reassuringly we did not observe any effects of ivermectin at a dose of 5 mg/kg on sensory or motor function in control animals (either in the naïve state or following nerve injury). In experimental animals expressing GluCl, a single dose of ivermectin resulted in a significant elevation of reflex withdrawal thresholds to noxious mechanical and thermal stimuli lasting 3 days. This is similar to the long-lasting behavioural effects previously observed following use of the GluCl v1.0 system in the striatum (Lerchner et al., 2007) and is thought to possibly reflect slow dissociation of ivermectin from GluCl (Slimko et al., 2002) and/or long-lasting retention of ivermectin in tissue (Lerchner et al., 2007). Importantly we noted that not only were the effects reversible but could also be reproduced following a subsequent treatment. We did not observe any motor deficits.
Finally, we tested the efficacy of GluCl in the spared nerve injury model of neuropathic pain (Decosterd and Woolf, 2000). Importantly, we treated with ivermectin at 1 week post-injury, a time when mechanical and cold hypersensitivity is well established. Ivermectin could significantly ameliorate mechanical pain-related hypersensitivity as well as reversing cold allodynia. This effect was reversible and reproducible up to 3 weeks post-injury, which was the last time point of testing. There was a significant correlation between the proportion of DRG cells transduced with GluCl and the degree of reversal of mechanical painrelated hypersensitivity. Consistent with these behavioural findings, we found that when assessed ex vivo ivermectin had successfully suppressed both ectopic and evoked activity in those sensory neurons expressing GluCl. Given the selectivity of AAV delivery for sensory neurons these findings emphasize the importance of aberrant primary afferent input in the maintenance of neuropathic pain (Vaso et al., 2014). The past decade has seen exciting developments in the identification of ion channels/signalling pathways contributing to the development and maintenance of neuropathic pain (Waxman and Zamponi, 2014;Habib et al., 2015). A caveat to therapeutically targeting individual molecules is the proposed degeneracy in the development of neuronal hyperexcitability following nerve injury (Ratte et al., 2014). GluCl activation represents a strategy to circumvent this problem by silencing neuronal activity regardless of inherent dysregulation in the system. This property may make the system also appropriate for other neurological disorders whereby dampening of neuronal excitability could prove efficacious, such as epilepsy (Avaliani et al., 2016). This is the first demonstration of efficacy of a chemogenetic system in a chronic pain model. There is some evidence that expression of the designer inhibitory G-proteincoupled receptor hM4Di (inhibitory designer receptor exclusively activated by designer drug DREADD) (Armbruster et al., 2007) in primary afferents can alter acute pain thresholds in vivo Saloman et al., 2016), however although widely used in the CNS its applicability to peripheral sensory neurons is questionable. Mouse sensory neurons do not exhibit functional GIRK currents (Nockemann et al., 2013), thought to be the endogenous mediator of hM4Di silencing (Armbruster et al., 2007). This contrasts with rat and human sensory neurons wherein at least a proportion of sensory neurons do express GIRK channels (Nockemann et al., 2013). In recordings from mouse DRG cells expressing hM4Di the reported efficacy was limited with 550% of such neurons showing a change in excitability (Saloman et al., 2016). These authors have also shown that hM4Di expression in primary afferents alters Na + and Ca 2 + currents in the absence of the ligand (Clozapine-N-oxide, CNO) and inhibits the analgesic effect of endogenous inhibitory G-protein-coupled receptors (a major confound if this were ever to be used therapeutically). Finally there are also concerns about the safety of the receptor ligand, CNO (Jann et al., 1994;Chang et al., 1998;Maclaren et al., 2016).
Other forms of neuromodulation include electrical stimulation devices (Colloca et al., 2017) and optogenetics (which has not yet been applied to patients). Optogenetics offers exciting opportunities but also faces many challenges, largely based on safely delivering sufficient illumination to target tissues (Montgomery et al., 2016). The chemogenetic approach has potential advantages, for instance in the ability to silence deep afferents inaccessible to light, which are a key source of ectopic activity (Michaelis et al., 2000;Kirillova et al., 2011), without the need of implants. One disadvantage is that ivermectin-mediated silencing in vivo exhibits temporal imprecision; however, the fact that behavioural efficacy is maintained for 3 days following a single dose could be advantageous when treating disorders of persistent hyperexcitability such as neuropathic pain. Neuromodulation mediated by electrical stimulation is used in the treatment of neuropathic pain (Kumar and Rizvi, 2014). However, electrode insertion is invasive, positive sensory symptoms may be troubling and efficacy is limited.
There would be significant challenges in developing a clinical therapeutic using our approach, mainly relating to delivery of the GluCl channel to sensory neurons in patients. Implicit in the findings from our nerve injury experiment is the fact that GluCl efficacy is directly correlated with the level of transduction achieved. AAV is, however, well tolerated in humans, can transfect post-mitotic neurons with long lasting expression (Hadaczek et al., 2010) and is already used in clinical trials (Kotterman and Schaffer, 2014). GluCl is modified from a Caenorhabiditis elegans protein and so the risk of an immune response in humans would need to be carefully considered. This is a major challenge for the field of AAV gene therapy, but is particularly pertinent when expressing a non-self transgene in the periphery (Montgomery et al., 2016). It is encouraging that long-term GluCl expression does not cause overt injury to DRG neurons and most significantly does not result in cutaneous denervation in our mouse models. However, strategies to modulate any potential immune response [e.g. tolerance induction or immune suppression (Mingozzi et al., 2003;Sack and Herzog, 2009)] to GluCl would have to be considered as part of any translational pathway. There is evidence that ectopic activity develops in both small and large diameter sensory afferents and that both populations contribute to neuropathic pain (Ossipov et al., 1999;Tarpley et al., 2004;Xu et al., 2015), hence our logic in targeting all sensory neurons. In certain situations, such as inflammatory pain (Abrahamsen et al., 2008;Brenneis et al., 2013), there would be an argument to restrict GluCl expression to small diameter nociceptive afferents that would necessitate development of effective promoters restricted to this population. The need to generate two independent AAVs expressing each subunit is a disadvantage when trying to silence a broad range of sensory neurons; however, certain paradigms can exploit this by using two independent promoters to express the subunits in an intersectional manner (Haubensak et al., 2010). Our data provide proof-of-concept that this chemogenetic approach can be used to define primary afferent contributions to persistent pain states and furthermore has translational potential to suppress hyperexcitability and treat neuropathic pain.