Controlled activation of cortical astrocytes can reverse neuropathic chronic pain


 Chronic pain is a major public health problem that currently lacks effective treatment options. Here, we report a novel combination therapy that can effectively reverse chronic pain induced by nerve injury in mice. By combing transient nerve block to inhibit noxious afferent input from injured peripheral nerves, with transient concurrent activation of astrocytes in the somatosensory cortex (S1) by either transcranial direct current stimulation (tDCS) or via the chemogenetic DREADD system, we could reverse allodynia previously established by partial sciatic nerve ligation (PSL). Activation of astrocytes initiated spine plasticity to reduce synapses formed shortly after PSL. The cure from allodynia persisted long after ceasing active treatment. Thus, our study represents the first report of a robust, readily translatable approach for treating chronic pain that capitalizes on the causative interplay between noxious afferents, sensitized central neuronal circuits and astrocytic-activation induced plasticity.


Introduction
Chronic pain remains as one of the leading causes of global disability, affecting around 20% of the adult population in the USA 1 . Novel approaches to reverse the mechanisms causing the transition from acute to chronic pain are likely to have the greatest therapeutic impact 2 . Injury-induced maladaptive neural circuit plasticity at both spinal and cortical levels, amplifies the noxious input and can render other sensory input as painful (allodynia) 3,4,5 . Innocuous circuits in the somatosensory cortex (S1) which normally encode the location, intensity, and duration of nociceptive stimuli 6 , are transformed during the early post-injury period to become activated by both nociceptive and innocuous stimuli 7,8 . Astrocytes are closely associated with neural circuit microstructure, and through both physical contacts and via secreted molecules can regulate a number of aspects of synaptic function, including synaptogenesis, spinogenesis, modulation of synaptic plasticity and the elimination of spines and synapses 9,10 . Some of these astrocytic signalling pathways responsible for spinogenesis in development are reactivated following peripheral nerve injury to promote dendritic spine turnover, and thereby contribute to this remodelling of cortical pain circuits 11 . We proposed that corrective remodelling of these S1 circuit synapses may effectively reverse chronic pain. We achieved this by blocking the noxious peripheral afferent 4 inputs while concurrently augmenting the activity of cortical astrocytes using transcranial direct current stimulation (tDCS) or the Designer Receptors Exclusively Activated by Designer Drugs (DREADD) system. This transient therapy induced rewiring of cortical circuit synapses and achieved a remission from allodynia that lasted well beyond the end of active treatment.

tDCS-TTX therapy reverses allodynia in neuropathic model mice
tDCS has been widely applied to humans for a variety of neurological disorders, typically applied at current intensities between 1-4 mA over multiple sessions of 15-30 mins duration 12 . Although tDCS is devoid of serious adverse effects, evidence for its efficacy in treating chronic pain is poor -at best providing only modest and temporary relief 12, 13,14 . Conventional tDCS is believed to act via modulating spontaneous neuronal activity, but recent rodent studies demonstrated tDCS induced synchronized and widespread astrocytic Ca 2+ transients in the absence of neuronal activation 15 . Using a similar protocol, we confirmed that a single tDCS session (0.1 mA for 10 minutes) applied over the cortex of awake mice caused a significant and sustained increase in the frequency and amplitude of Ca 2+ transients in S1 astrocytes ( Fig. 1, a to c). 5 We asked if modulating astrocyte activity could impact on chronic pain behaviour, and used partial sciatic nerve ligation (PSL) 16 to induce a sustained allodynia. tDCS was subsequently applied for a week at 8-hourly intervals, all 2 weeks following the PSL injury. We further reasoned that reducing noxious afferent activity from the injured nerve may also be required to reverse chronic allodynia, so simultaneously delivered tetrodotoxin (TTX) locally to the sciatic nerve via an implanted Elvax drug elution cuff (Fig. 1f). Prior to any treatment, all mice displayed mechanical allodynia following PSL as shown by decreases in paw withdrawal thresholds (Fig 1d). TTX application returned these thresholds back to control levels, indicating relief from pain due to afferent input blockade. In control mice, as TTX elution subsided 11 , thresholds again decreased to post PSL levels, indicating return of mechanical allodynia. However, in PSL mice with combined tDCS, paw withdrawal thresholds were sustained at pre-injury levels, indicating reversal of allodynia (Fig. 1d).
Thermal allodynia was also reversed by combined tDCS and TTX, without any effects on withdrawal thresholds in the uninjured contralateral paw ( Supplementary Fig. 4).
Conventional therapeutic tDCS, i.e., without simultaneous blockade of noxious inputs with TTX, was ineffective in providing lasting pain relief (Fig. 1e). Together our results show that a brief treatment regime combining tDCS configured to induce S1 astrocytic 6 activation, combined with TTX-blockade of noxious afferents, i.e., tDCS-TTX therapy, is able to mediate a lasting adaptive response to reverse chronic pain.
Specific chemogenic activation of S1 astrocytes also reverses allodynia tDCS is known to broadly activate astrocytes across the cortex both ipsilateral and contralateral to the stimulating anode 15 . To more specifically determine the locus of the therapeutic effect, we used a different and more targeted approach using the DREADD system 17 (Fig.2a). We expressed the hM3D receptor in S1 astrocytes by localised AAV injection (Fig. 2b), achieving expression in ~89% of S1 astrocytes (392 astrocytes from 5 mice). A single intraperitoneal injection of clozapine N-oxide (CNO; 1.0 mg/kg, i.p.), which activates this hM3D receptor, also increased astrocytic Ca 2+ transients. The increase in the Ca 2+ transient response was specific to S1 astrocytes; the activity of astrocytes in M1 was not affected by CNO ( Supplementary Fig. 5). Repeated CNO administration induced up-regulation of glial fibrillary acidic protein (GFAP) in S1 astrocytes ( Supplementary Fig. 6), indicating they adopted an activated phenotype.
Consistent with the rapid distribution of CNO 18 , astrocytic Ca 2+ activity peaked immediately after administration but then showed a sustained increase from 3 hours onwards (Fig. 2c), perhaps reflecting prolonged signalling or accumulation in brain 7 tissues 18 . Given that these sustained increases in astrocytic Ca 2+ were comparable to those seen with tDCS, we used this CNO dose in mice with established PSL-induced allodynia. We administered CNO at 8-hourly intervals for 1 week, whilst again simultaneously suppressing noxious afferent input from the injured sciatic nerve with local TTX elution from the Elvax cuff (Fig. 2a). As before, TTX administration immediately restored paw withdrawal thresholds to pre-injury levels ( Fig. 2d), but this was only transient in control mice treated with concurrent saline injections. In contrast, mice given TTX and concurrent CNO injections continued to maintain pre-injury withdrawal thresholds indicating a sustained reversal of allodynia, and this relief could last for at least 2 months in some mice (Figs. 2, d and e). Importantly, CNO treatment by itself (with astrocytic mCherry expression as a control for hM3D) did not reverse allodynia, and similarly there was no effect of CNO in mice expressing hM3D in astrocytes but not given concurrent TTX ( Supplementary Fig. 7a). Furthermore, lower doses of CNO that caused significantly smaller astrocytic activation, did not reverse the allodynia following PSL (Supplementary Fig. 8). Thus, a transient post-injury regime of CNO-TTX therapy or tDCS-TTX therapy are both able to prevent long-term allodynia, indicating that the key locus of the therapeutic effect is the selective activation of S1 astrocytes during the period when noxious afferent input is reduced. 8 The therapy preferentially eliminates dendrite spines associated with noxious circuits We next examined the mechanisms by which the combination therapy was able to induce this seemingly permanent reversal of the chronic pain response. Previous studies have demonstrated that transient and discrete astrocyte activation in the first week following PSL triggers a period of increased spine formation and elimination in S1 11,19 , which may represent maladaptive plasticity of cortical neural circuits that engenders nociceptive responses to previous innocuous sensory stimuli. We therefore hypothesized that subsequently augmenting astrocyte activity, coupled with an absence of noxious afferent input, could re-awaken spine plasticity and induce an adaptive rewiring of this maladapted circuitry. We therefore imaged S1 layer 5 pyramidal neuron dendritic spines, starting prior to PSL and again during and after the combination therapy, either CNO-TTX (Fig. 3a) or tDCS-TTX ( Supplementary Fig. 9a). Spine dynamics were significantly increased by CNO-TTX (Fig. 3, b to c) and tDCS-TTX therapy (Supplementary Fig. 9), with an increase in spine elimination and/or formation rates as compared with control mice (Figs. 3, d and e, Supplementary Fig.10). But spine density kept the same level before, during and after CNO-TTX therapy ( Supplementary Fig.  10e). We then stratified eliminated spines into those that were present prior to PSL (pre-PSL spines), and those that appeared just after PSL (early post-PSL spines), reasoning that pre-PSL spines were more likely to represent neuronal circuits for normal somatosensory processing, whereas early post-PSL spines would be more relevant to PSL-triggered maladaptive noxious circuits. Early post-PSL spine elimination was three times higher than pre-PSL spine elimination in CNO-treated mice, but equally labile in control mice (Fig. 3f). Closer examination of the temporal pattern of elimination of these early post-PSL spines (Fig. 3g) showed that while most were eliminated during the long pre-treatment period in all mice, the proportion of persistent spines eliminated during the subsequent treatment period was significantly higher in the CNO mice. These results are consistent with the idea that activated astrocytes provide a temporary window of plasticity where dendritic spines associated with maladaptive noxious circuits (rendered inactive by afferent input blockade) are preferentially eliminated.
Consistently, S1 neuronal activity in PSL mice, and the correlations between neuronal firing within these S1 neurons, was modestly but significantly decreased by CNO-TTX therapy (Fig. 3h, Supplementary Fig. 11), suggesting a reduced functional connectivity in these post-PSL noxious circuits concurrent with the spine elimination. We propose that this preferential loss of early post-PSL spines represents a targeted pruning of the 10 aberrant neural connections that critically contribute to the chronic tactile and thermal allodynia.

Preclinical evaluation of combined tDCS and lidocaine therapy
While TTX effectively blocked the noxious afferent inputs during the therapeutic window, it is too toxic to be integrated into clinical application. Therefore, we evaluated the use of a local anaesthetic, applying lidocaine by osmotic pump (2 mg/kg/hr) from day 0 to 3 ( Fig. 4a). While the effects of lidocaine on paw withdrawal thresholds were weaker than TTX, combined transient lidocaine and tDCS therapy still caused a sustained and significant reduction in the extent of allodynia that outlasted the treatment period (Fig. 4b). Combination tDCS-lidocaine therapy did not adversely affect motor coordination in the rotarod assay ( Fig. 4c), or locomotion and anxiety-like behaviours in the open-field test (Fig. 4d).

Discussion
This study demonstrates that astrocytes can be positively leveraged to treat chronic pain.
Specifically, transiently enhancing astrocytic Ca 2+ activity, either through tDCS or the DREADD system, whilst simultaneously blocking noxious afferent input from the site 11 of injury, enables increased circuit remodelling which we propose preferentially eliminates maladapted noxious circuits and thereby permanently eliminates mechanical allodynia and returns tactile sensation to normal. This is a conceptual shift in chronic pain treatment, from targeting neuronal excitability and synaptic transmission that reduces the consequences of hyperactive circuits, to targeting astrocytes and cortical circuit plasticity, that may reverse some of the mechanisms by which acute pain transitions to chronic pain.
Our study provides further support for the growing idea that aberrent neural circuit plasticity in S1 significantly underpins intractable nociceptive perceptions in chronic pain 11,20 . Hyperactivity of L2/3 20 , loss of effective local circuit inhibiton 21 , and activation of astrocytes to mediate spine plasticity 11,19 have all been observed during early chronic pain development in rodent models and prophylactically reducing cortical neuronal and astrocyte activity has been shown to prevent robust chronic pain development. Furthermore, enhanced activity of S1 affects other brain areas related to chronic pain, e.g. anterior cingulate cortex 20 . Our current report shows that targeting S1 circuits weeks after allodynia has developed can still ameliorate pain symptoms and reverse the injury-associated S1 plasticity. We propose that the tDCS-TTX treatment strategy reduces connectivity in S1 noxious circuits. Other spinal and brain regions are undoubtedly important for chronic pain, and it will be informative to see if and how their activities change in response to therapeutic S1 remodelling.
A key component of this therapeutic strategy was the concurrent application of TTX or lidocaine. We believe the doses and delivery systems used fortuitously and selectively reduced noxious afferent input without large impacts on tactile and motor fibres. A reduction of afferent input can very effectively weaken the synaptic representation of those afferents in cortical areas 22 . We propose reduction in nociceptive afferent activity by TTX weakens the maladapted cortical circuits responding to the painful tactile stimuli, exposing these weakened synapses to be eliminated, by astrocyte or microglial phagocytosis 23,24 . Because tactile sensation remains, circuits responding to "normal" sensation can be strengthened, returning S1 representations and activity back to control. Concurrent astrocyte activity may either facilitate synapse removal, or strengthen rewiring of "normal" circuits 25,26 . The current clinical use of nerve block approaches similarly targets excessive nociceptive efferent activity 27 whilst sparing tactile and motor function, and hence may be suitable to complement tDCS to selectively weaken aberrant maladapted cortical noxious circuits.
Our study suggests combined tDCS and afferent nerve blockade warrants further clinical investigation as a therapy for pain. tDCS is being widely trialled for 13 different neurocognitive applications, particularly depression, stroke recovery and chronic pain, and at typical current intensities is considered safe 12 . Based on our results in mice, we would envisage twice or thrice daily stimulation over one week with the anode placed over the S1 approximate to the injured region, and a modest stimulation intensity to preferentially activate astrocytes rather than neurons. Regional peripheral nerve block via a catheter implanted under ultrasound-guided surgery is used commonly for postoperative analgesia 28,29 and we would also envisage concurrent application of a long-lasting local anaesthetic bolus via this route. Alternatively, continuous infusion via a controlled pump for 4-6 days has also been trialled to treat different chronic pain syndromes 30,31 . However, it should be recognized that our current study only used young male mice, many pain phenotypes differ between sexes with females typically reporting greater pain responses 32,33 . A further limitation of our study regards translating efficacy from a single pain model in one mouse strain to humans; for example, pregabalin has larger and broader analgesic effects in rodent chronic pain models than observed clinically 34 .
In conclusion, we report a transient combination therapy that reverses mechanical allodynia in mice. We propose that activating cortical astrocytes while reducing peripheral noxious inputs engineer synaptic plasticity that breaks down 14 inappropriate neural connections formed during the transition from acute to chronic pain. Given that both nerve block and tDCS are both readily utilized in the clinic, it seems feasible to translate our discovery into clinical practice for the treatment of intractable chronic pain associated with a well defined peripheral injury.

Care and use of animals
All animal experiments were approved by the Institutional Animal Care and Use Committee of the National Institutes for Natural Sciences, Okazaki, Japan. Every effort was made to minimize the suffering and number of animals used.
All animal experiments were conducted using male, 8-10 week old C57BL/6 mice, M-line mice (C57BL/6 mice expressing enhanced green fluorescent protein (EGFP) in sparse subsets of cortical neurons under the thy-1 promoter) 35 or mGFAP-Cre mice (C57BL/6 mice expressing Cre recombinase in astrocytes under the mouse GFAP promoter (astrocyte specific promoter)) 36 . All mice were housed with ad libitum access to standard rodent chow and water under a 12-hour light-dark cycle. Mice were randomly separated to treatment or control groups.

Partial sciatic nerve ligation (PSL) and behavioural testing
The PSL model is a well-established animal model of chronic pain that is characterized by protracted touch-evoked allodynia and hyperalgesia 16 . PSL was performed by ligating the dorsal one-third to half of the right sciatic nerve with 8-0 silk suture under 1.5-1.8% isoflurane anaesthesia. 16 Mechanical-tactile allodynia was observed for 2 weeks post-PSL before further experiments. Mechanical allodynia and hyperalgesia were assessed manually using the von Frey filament test. All testing was performed during the 12hr daylight cycle. Prior to von Frey testing, mice were habituated individually for 30 minutes in the testing chamber which consisted of a small plastic cage with a mesh floor. The von Frey filaments were applied through the mesh floor so as to perpendicularly touch the hind paw with sufficient force to cause slight filament buckling. If this elicited withdrawal, licking or shaking of the paw by the mice, a positive response (pain) was recorded.
Filaments were tested with increasing then decreasing force to establish that which caused a median 50% paw withdrawal threshold 37 . All von Frey testing was performed in a blinded fashion with the experimenter unaware of the treatments the mice had received. There were no differences in the baseline thresholds for mechanical sensitivity between the three strains of mice used (C57BL/6, M-line and mGFAP-Cre).
Thermal sensitivity was assessed following injury (post-PSL) and after treatment (at days 8 and 14) using the Plantar test 38 with an infrared heat stimulus and an automated recording of latency to paw withdrawal (Hargreaves Apparatus, ugo baile, Italy). As previously described 39 , mice were placed in the thermal sensitivity enclosure for 60 mins for each of 2 days prior to testing to habituate to the plantar test environment. When mice positioned their hind paw on the infrared emitter/detector, heat stimulation started and continued until the paw was withdrawn. The latency to paw withdrawal was observed over 5 trials, with the averaged value excluded the two trials with the smallest and largest withdrawal times.
The accelerating rotarod test was performed to assess motor coordination as The open field test was performed to assess locomotor activity and anxietylike behaviour. As previously described 42 , mice were placed in a 40 × 40 cm chamber and exploratory behaviours recorded for 10 mins and subsequently analyzed using an automated tracking system (Lab Squirred Pty Ltd, Sydney, Australia). The open field was divided into an inner (20 × 20 cm) and outer zone, to assess anxiety-like behaviour.
We measured total locomotion speed, total distance travelled and the total duration spent in the inner zone.

Preparation and implantation of TTX mixed with Elvax, and of the lidocaine osmotic pump
As described previously 43 , TTX was applied to the right sciatic nerve using an ethylenevinyl acetate copolymer (Elvax) carrier. Briefly, 100 mg/mL Elvax solution was prepared by dissolving Elvax beads (Dupont) in dichloromethane, before adding 3 mM TTX to give a final TTX concentration of 300 μM. This TTX-Elvax solution was then stirred for 1 hour to ensure thorough mixing before pouring onto a pre-cooled glass dish, which had been incubated for 2-3 hours at -80 o C, and stored at -20 o C. This allowed for the complete evaporation of the dichloromethane and for the TTX-Elvax solution to solidify into a sheet that could then be applied around the right sciatic nerve just proximal to the PSL injury (under 1.5-1.8% isoflurane anaesthesia). The estimated duration for drug delivery using Elvax is 5-6 days 11 .
Lidocaine (2 mg/kg/hr in saline) was applied using an osmotic pump (0.5 μl/h, 1007D, Alzet, CA) implanted subcutaneously in the back with the tip of the osmotic pump implanted close to the sciatic nerve ligation. After 3 days, the osmotic pump was removed. Implantation and removal surgery occurred under isoflurane anaesthesia with mice.

Generation of Adeno-Associated Viruses
pAAV-GFAP-mCherry was prepared by generating a linearized vector from pAAV-GFAP-hM3D(Gq)-mCherry (Addgene, #50478) using restriction enzymes with subsequent vector end blunting and nick sealing mediated by DNA polymerase I Large

AAV injection
Mice used for the DREADD-behaviour only experiments were injected with AAV2/5-GFAP-hM3D(Gq)-mCherry as follows, 1-2 weeks prior to the PSL operation. Mice were anesthetized (i.p.) using a ketamine (70 mg/kg) and xylazine (10.5 mg/kg) mixture and mounted into a stereotaxic frame (NARISHIGE, Japan). Approximately 500 nL of AAV was injected into the left S1 at stereotaxic coordinates 0.5 mm posterior and 1.5 mm lateral to Bregma, and at a depth of 300 μm below the brain surface. The injection 21 pipettes were pulled from filament-containing glass capillaries (GDC-1, NARISHIGE, Japan) and AAV slowly expelled (over 10 mins) using pneumatic pressure (IM 300 Microinjector, Narishige Scientific Instrument Lab., Tokyo, Japan). ab5804, Merck, Germany). Immunocomplexes were visualized by chemiluminescent detection using HRP-conjugated goat anti-rabbit/anti-mouse secondary antibodies (1:300; Santa Cruz Biotechnology). Images were acquired using an A1R confocal microscope (Nikon, Tokyo, Japan) with NIS-elements software (Nikon, Tokyo, Japan) under a ×20 objective lens (PLanSApo, NA = 0.75) with a z-step size of 0.5 μm. The total number of mCherry-expressing cells co-localized with S100β or NeuN in the S1 22 was manually counted using the Cell Counter plugin (credit Durt De Vos) within the ImageJ software environment (NIH).

Open-skull chronic cranial window implantation
Mice used for in vivo imaging were injected with the relevant AAV's and implanted with a chronic cranial window according to the following procedure. The cranial window surgery was undertaken two weeks prior to the PSL operation.
Mice were anesthetized (i.p.) using a ketamine (70 mg/kg) and xylazine (10.5 mg/kg) mixture. The scalp skin was incised and the entire skull surface was waterproofed with tissue adhesive (3M Vetbond, 3M, MN, USA). A custom-made metal head plate was then directly attached to the skull using resin cement (Estecem II, Tokuyama Dental Corporation, Tokyo, Japan) and dental cement (Fuji Lute BC, GC Corporation, Tokyo, Japan). After the cements had cured, the entire skull surface was covered with dental adhesive (Super-Bond with Catalyst V, Monomer and Polymer, Sun Medical Corporation, Shiga, Japan) which acted as both a reinforcing and waterproofing agent.
Implantation of the cranial window was performed the next day. Mice were anaesthetized with 1.0-1.2 % isoflurane and secured in a stereotaxic frame via the 23 attached head plate. For S1 imaging, a circular 2-3 mm diameter craniotomy was then drilled over the hind-limb area of the left S1 (craniotomy centre at 0.5 mm posterior and 1.5 mm lateral to Bregma). Following this, approximately 500 nL of AAVs (GFAP-hM3D-mCherry, GFAP-GCaMP6f, GFPA-mCherry) were injected as described above into the centre of the craniotomy site at a depth of 300 µm from the brain surface.
Comparative M1 and S1 imaging used a single rectangular (3 × 2 mm) craniotomy through which approximately 500 nL of AAVs (CamKII-GCaMP6f (UPENN), GFAP-hM3D-mCherry, GFAP-GCaMP6f) were injected into M1 and S1 areas at a depth of 300 µm from the brain surface ( Supplementary Fig. 8a). The craniotomy sites were covered with a double glass coverslip (Matsunami Glass, Osaka, Japan) which consisted of a 2 mm diameter coverslip fused to a 4.5 mm diameter coverslip, or a 2 × 3 mm coverslip fused to a 4 × 5 mm coverslip. The coverslip was fixed to the skull using a mixture of dental cement (Fuji Lute BC, GC Corporation, Tokyo, Japan) and dental adhesive (Super-Bond with Catalyst V, Monomer and Polymer, Sun Medical Corporation, Shiga, Japan). Ca 2+ imaging of astrocytes and neurons in S1 or M1 for DREADD and tDCS assessment 24 Ca 2+ imaging was performed 2 weeks after cranial window implantation using a multiphoton microscope (Nikon A1R MP, Nikon, Tokyo, Japan) fitted with a 25x water immersion objective lens (Nikon Apo LWD 25x/1.10w, Nikon, Tokyo, Japan) and a Ti:Sapphire laser (MaiTai DeepSee, Spectra Physics, CA, USA), tuned to 950 nm for 2photon excitation of GCaMP6f.
During the imaging session, mice were secured in a custom-built frame via their surgically attached head plate. mGFAP-Cre and C57BL6/J mice anaesthetized with isoflurane (2% for induction, 1% for maintenance) were used to assess CNO-induced Ca 2+ activity. Awake C57BL6/J mice were used to assess tDCS-induced Ca 2+ activity, after habituation to the head fixation frame for 5 days prior to imaging session by daily, 15-minute fixation with free access to water. To determine neural circuit connectivity, Ca 2+ fluorescence activity was recorded from the soma of layer 5 pyramidal neurons in mice freely moving on a moveable running baseplate while head-fixed 45 . Mice were habituated to the experimental room, head fixation and baseplate with 10-minute sessions for 2 days prior to imaging.

Transcranial direct current stimulation (tDCS)
Transcranial direct current stimulation (tDCS) has traditionally been used to noninvasively modulate neuronal activity. In this study, however, we used a tDCS protocol to selectively activate astrocytes 15 . In mice used for tDCS-behaviour alone experiments, PSL was performed 1 week after attaching the metal head plate but 2 weeks before preparing the skull for tDCS stimulation. In mice used for tDCS-imaging experiments, cranial window implantation and AAV injection were performed 1 day after attaching a metal head plate and 2 weeks before preparation for tDCS. PSL was not performed in these mice.
To prepare for tDCS, mice were anesthetized (i.p.) using a ketamine (70 mg/kg) and xylazine (10.5 mg/kg) mixture. The scalp skin was incised and the entire skull surface was waterproofed with a tissue adhesive (3M Vetbond, 3M, MN, USA). A custom-made metal head plate was then directly attached to the skull using resin cement (Estecem II, Tokuyama Dental Corporation, Tokyo, Japan) and dental cement (Fuji Lute BC, GC Corporation, Tokyo, Japan). After the cements had cured, the entire skull surface was covered with dental adhesive (Super-Bond with Catalyst V, Monomer and Polymer, Sun Medical Corporation, Shiga, Japan) which acted as both a reinforcing and waterproofing agent. 27 On the first day of tDCS therapy (day 0), the cathode and anode sites for tDCS were prepared. For the cathode, 0.5 mm diameter silver wire (The Nilaco corporation, Tokyo, Japan) was subcutaneously implanted in the neck. For the anode, the dental adhesive was removed from a 5 mm 2 diameter circular area of the skull 2 mm posterior to the left S1 ( Supplementary Fig. 1). Between tDCS sessions, this anode skull site was covered with silicon to keep the skull clean and moist. During each tDCS session, a silver wire was placed within this anode skull site which was filled with conductive gel (Gelaid, Nihon Kohden, Japan). tDCS therapy consisted of ten-minute individual sessions performed at 8hourly intervals over 1 week (21 tDCS sessions in total). During each tDCS session, a constant 0.1 mA of direct current was applied from the anode to the cathode over the 10 minutes using a stimulus isolator (Nihon Kohden, Japan) ( Supplementary Fig. 1).
During this time, mice were awake but secured in a custom-made frame via their attached head plate. Control mice were subjected to the same procedure without any current stimulation. In experiments where the effect of tDCS on astrocytic or neuronal activity was assessed by simultaneous imaging, the constant current varied between each session from 0.01 mA to 1 mA.

Long-term in vivo 2-photon imaging of spine dynamics
In these experiments, M-line mice were injected with the relevant AAV's and implanted with a chronic cranial window as described above. The first imaging session was performed 14-20 days after cranial window implantation and thus preceded the PSL operation by 3 days. Subsequent imaging sessions were performed at 14, 10 and 7 days prior to the first CNO injection (day 0) and on days 0, 4, 7, 11 and 14 (Fig. 3a).
During each imaging session, mice were anesthetized using 1.0-1.5% isoflurane. The imaging area position was determined as previously described (11) by using intrinsic optical signal imaging and identifying the mCherry expressing area. In vivo 2-photon imaging was performed using a multiphoton microscope fitted with a 25x water immersion objective lens and a Ti:Sapphire laser tuned to 800 nm and 950 nm for 2-photon excitation of mCherry and EGFP, respectively. In order to perform reliable long-term imaging of the same specific layer 5 pyramidal neuron dendrite over multiple imaging sessions, the dendritic area of interest was first identified using lowmagnification imaging (512 × 512 pixels, 0.99 µm/pixel, 2-µm z step). Highmagnification (512 × 512 pixels, 0.12 µm/pixel, 0.5-µm z step) was then used to quantify the specific dendritic region morphology. 29 For analysis, all images were first imported into the ImageJ software environment (NIH) in order to create 3D image stacks for each dendrite, with motion artefacts corrected using TurboReg. The 3D image stacks were then imported into the AIVIA software environment (DRvision Technologies LLC, DC, USA) which was able to automatically identify and track individual spines on the same section of dendrite across all imaging sessions (with manual cross-checking). All types of dendritic protrusions were included for analysis, except for those which were located in close proximity to the distal tip of a dendritic branch, due to previous reports of their significantly greater instability 11 . Spine parameters were quantified by other experimenters who did not perform therapy and spine imaging and were blinded to the treatment/control group.
When comparing 2 successive imaging sessions, spines which were present in the latter session but not the former session were referred to as "formed" or "gained" spines whereas spines which were present in the former session but not in a latter session were referred to as "lost" spines. The spine formation rate between 2 successive imaging sessions was calculated as the number of formed spines divided by the total number of spines counted in the former session. Spine elimination rate between 2 successive imaging sessions was calculated as the number of lost spines divided by the 30 total number of spines counted in the former session. Spine turnover rate between 2 successive imaging sessions was an average of these values -calculated as the sum of the number of formed spines and lost spines, divided by twice the total number of spines counted in the former session. Spine density was calculated as the total number of spines counted, divided by the length of the dendrite. Spine densities were normalized to that counted on day 0.

Statistical analysis
All statistical analyses were performed using SPSSver22 (IBM, NY, USA) or MATLAB (The Mathworks, Inc). Comparisons between different sample groups within the same treatment cohort were performed using one-way repeated ANOVA or two-way repeated ANOVA followed by the Bonferroni post-hoc test. Non-parametric multiple comparisons between different treatment cohorts were performed using the Kruskal-Wallis H test followed by the Bonferroni post-hoc test. Comparisons between different treatment cohorts were performed using the Chi-squared test. Single variable comparisons were performed using paired, unpaired t-tests and Wilcoxon rank sum test.
Values of p < 0.05 were considered to be statistically significant. All values were reported as mean +/-2 standard error.

Data and materials availability
All data associated with this study are present in the paper or Supplementary Materials.   b Repeated imaging of the same dendritic segment from a layer 5 pyramidal neurons in the left S1 before (day -3 to day 0), during (day 4 to day 7) and after (day11 to day 14) CNO treatment. Arrowheads indicate spine formation (magenta) and spine elimination (blue). Scale bar = 5 µm.
c to e Spine dynamics during and after CNO (with TTX) treatment. The period during CNO-TTX or saline-TTX administration is indicated by grey shading. Asterisks indicate the level of statistical significance when comparing saline and CNO treatment as follows: *p < 0.05, **p < 0.01. c Normalized spine turnover rates; in saline-treated (white; n = 16 dendrites from 7 mice) and CNO-treated mice (green; n = 19 dendrites from 5 mice). Spine turnover rate was calculated as the average of spine formation and elimination rates. Comparisons between CNO and saline used a two-way repeated 53 ANOVA, the interaction effect: F(4, 132) = 1.461, p = 0.218, the main effect for group: F(1, 33) = 6.445, P = 0.016); followed by Bonferroni post hoc test, day 0 (p = 1.0), day 4 (p = 0.031), day 7 (p = 0.385), day 11 (p = 0.114), and day 14 (p = 0.032). d Normalized spine elimination rates in saline-treated (white; n = 16 dendrites from 7 mice) and CNO-treated mice (blue; n = 19 dendrites from 5 mice). Spine elimination rate was calculated as the number of spines that were lost between 2 successive imaging sessions, divided by the total number of spines counted in the first of these imaging sessions, and normalized to the pre-injury elimination rate. Comparisons between CNO and saline was tested using a two-way repeated ANOVA, the interaction effect: F(4, 132) = 1.852, p = 0.123, the main effect of group: F(1, 33) = 18.477, p = 0.000; followed by Bonferroni post-hoc test, day 0 (p = 1.000), day 4 (p = 0.002), day 7 (p = 0.018), day 11 (p = 0.285), and day 14 (p = 0.008). e Normalized spine formation rates in saline-treated (white; n = 16 dendrites from 7 mice) and CNO-treated mice (pink; n = 19 dendrites from 5 mice). Spine formation rate was similarly calculated as the number of spines that were formed between 2 successive imaging sessions, divided by the total number of spines counted in the prior imaging session, normalized to pre injury rate.
f CNO-TTX treatment was more likely to be associated with elimination of spines formed during the period just after PSL. Pre-PSL spines were defined as those which were observed both before the PSL operation and before the start of CNO-TTX therapy.
Early post-PSL spines were defined as those which only appeared in the week following PSL and which were also still present before the start of CNO-TTX therapy. Pre-PSL spine counts in CNO-and saline-treated mice were 33 spines, 19 dendrites from 5 CNO-treated mice and 17 spines, 16 dendrites from 7 saline-treated mice, respectively. "Before" spines were those which disappeared prior to the start of CNO-TTX therapy, i.e., lost at the day -3 or day 0 time points. "During" spines were those which disappeared during CNO-TTX therapy, i.e., lost at the day 4 or day 7 time points.
"After" spines were those which disappeared in the week after CNO-TTX therapy had finished, i.e., lost at the day 11 or day 14 time point. "Persistent" spines were those which were observed at all subsequent time points following their first appearance. c tDCS-treatment does not effect motor performance on the rotarod test. Graph shows the latency to fall off the escalating rotarod for tDCS/lidocaine, and lidocaine treated mice, before, during and after tDCS treatment. Statistical significance between tDCStreated mice and control mice was tested using a two-way repeated measures ANOVA,