Impaired Excitatory Drive to Spinal Gabaergic Neurons of Neuropathic Mice

Adequate pain sensitivity requires a delicate balance between excitation and inhibition in the dorsal horn of the spinal cord. This balance is severely impaired in neuropathy leading to enhanced pain sensations (hyperalgesia). The underlying mechanisms remain elusive. Here we explored the hypothesis that the excitatory drive to spinal GABAergic neurons might be impaired in neuropathic animals. Transgenic adult mice expressing EGFP under the promoter for GAD67 underwent either chronic constriction injury of the sciatic nerve or sham surgery. In transverse slices from lumbar spinal cord we performed whole-cell patch-clamp recordings from identified GABAergic neurons in lamina II. In neuropathic animals rates of mEPSC were reduced indicating diminished global excitatory input. This downregulation of excitatory drive required a rise in postsynaptic Ca2+. Neither the density and morphology of dendritic spines on GABAergic neurons nor the number of excitatory synapses contacting GABAergic neurons were affected by neuropathy. In contrast, paired-pulse ratio of Aδ- or C-fiber-evoked monosynaptic EPSCs following dorsal root stimulation was increased in neuropathic animals suggesting reduced neurotransmitter release from primary afferents. Our data indicate that peripheral neuropathy triggers Ca2+-dependent signaling pathways in spinal GABAergic neurons. This leads to a global downregulation of the excitatory drive to GABAergic neurons. The downregulation involves a presynaptic mechanism and also applies to the excitation of GABAergic neurons by presumably nociceptive Aδ- and C-fibers. This then leads to an inadequately low recruitment of inhibitory interneurons during nociception. We suggest that this previously unrecognized mechanism of impaired spinal inhibition contributes to hyperalgesia in neuropathy.


Introduction
Impaired GABAergic inhibition in the spinal cord may contribute to neuropathic pain which is characterized by increased sensitivity to thermal and mechanical stimuli (thermal and mechanical hyperalgesia) [1,2]. Blockade of GABAergic inhibition in the spinal cord mimics some symptoms of neuropathic pain [3], and activation of spinal GABA receptors can alleviate neuropathic pain in animals [4] and in humans [5]. The mechanisms underlying impaired GABAergic inhibition in the spinal cord in neuropathy are not fully understood but a number of different models have been proposed [6][7][8]. In their classical ''gate control theory of pain'' Melzack and Wall [9] suggested that inhibitory interneurons in the superficial spinal dorsal horn ''function as a gate control system''. Nociceptive afferent fibers would directly depress these inhibitory neurons which ''holds the gate in a relatively open position''. Thus, the ''output of the T(ransmission) cell rises'' [9] (see Fig. 1A). Experimental evidence for a direct depression of inhibitory interneurons by nociceptive nerve fibers is, however, lacking as nociceptive afferent fibers excite, rather than depress spinal GABAergic neurons [8,10]. Here we tested a different, but closely related hypothesis. We speculated that under conditions of a neuropathy the excitatory drive to inhibitory GABAergic neurons might be reduced. A reduced excitation would functionally be equivalent to an increase in depression of inhibitory neurons and could thus equally well open a gate for nociceptive messages to the brain (Fig. 1B) and thus lead to mechanical and thermal hyperalgesia.

Animals
Mice were treated in strict accordance with directive 2010/63/ EU of the European Parliament and of the council of the EU. The protocol was approved by the Austrian Federal Ministry of Science and Research (Permit Number: 66.009/097-C/GT/2007). All surgery was performed under isoflurane anesthesia, and all efforts were made to minimize suffering.
Male, 8 -10 weeks old, homozygous transgenic mice, expressing EGFP under the control of the promoter for GAD67 (GIN mice, [11]) (Jackson Laboratories, USA; strain name: FVB-TgN (GadGFP)45704Swn) were used in this study. Animals were kept and interbred in local facilities with food and water supply ad libitum.

Nerve ligation
Animals underwent either a sham surgery or a modified version [8] of chronic constriction injury (CCI) surgery, a model for neuropathic pain [12]. Mice were deeply anesthetized with isoflurane and the left sciatic nerve was exposed at the mid-thigh level. Proximal to the trifurcation the nerve was freed of adhering tissue and three ligatures (7-0 prolene) were tied around it with about 1 mm spacing in between. Ligatures were tied until they elicited a twitch in the hind paw. Muscle and skin incision were then closed. Sham-treated mice, that underwent the same procedure except ligation, were used as control animals.

Behavioral tests
Mechanical thresholds were assessed with calibrated von Frey monofilaments with incremental stiffness (Stoelting, USA) according to the up-and-down method [13]. The 50% threshold was calculated, which indicates the force of von Frey hair at which an animal reacts in 50% of the presentations [14].
Thermal nociceptive thresholds were determined by responses to a radiant heat source focused on the plantar surface of the rats hindpaw with a Plantar Test Instrument (Ugo Basile, Italy) [15]. The paw withdrawal latency was recorded by a digital timer with a cut-off of 30 s.

Spinal cord slice preparation
Ten to 13 days after surgery mice were anesthetized with isoflurane and the lumbar spinal cord was removed. The spinal cord was transferred to cold (,4uC), oxygenated incubation solution composed of (mM): NaCl 95, MgSO 4 7, CaCl 2 0.5, KH 2 PO 4 1.2, KCl 1.8, NaHCO 3 26, glucose 15, sucrose 50, oxygenated with 95% O 2 , 5% CO 2 ; pH 7.4, measured osmolarity 305 -320 mosmol?l 21 . After removal of all roots except dorsal roots L4 -L6 (ipsilateral to the surgery), the spinal cord was cut on a DSK microslicer (DTK-1000, Dosaka, Japan) into ,500 mm thick transverse spinal cord slices with a dorsal root attached. For two-photon laser-scanning microscopy parasagittal lumbar spinal cord slices were prepared on the microslicer at ,300 mm thickness. After preparation spinal cord slices were transferred to 30uC incubation solution and kept there for at least 1 h.

Electrophysiology
Single transverse slices were transferred to a recording chamber and superfused with oxygenated recording solution at 3 ml?min 21 at 33uC. The recording solution was identical to the incubation solution except for (in mM): NaCl 127; CaCl 2 2.4; MgSO 4 1.3 and sucrose 0. Lamina II dorsal horn neurons were visualized with infrared light using a cooled CCD camera (PCO, Germany). EGFP expressing neurons were identified by epifluorescence microscopy. Neurons located within a distance of 20 to 100 mm from the dorsal white matter were regarded as lamina II neurons [16].

mEPSC recordings
Miniature excitatory postsynaptic currents (mEPSCs) were measured in the presence of tetrodotoxin (TTX, 1 mM), D-2amino-5-phosphonovaleric acid (D-AP5, 50 mM; both Ascent Scientific), strychnine (4 mM) and bicuculline (10 mM; both Sigma-Aldrich) from minute 6 -10 after establishing the whole-cell configuration. At this time interval following whole-cell conformation establishment mEPSC frequency had stabilized. Clamp potential was set at -75 mV. Series resistance was controlled before and after mEPSC-recordings. Only cells with stable series resistances # 30 MV were used for analysis.
Data analysis was done offline by an investigator blinded to the treatment groups using MiniAnalysis Software (Synaptosoft, USA). Miniature postsynaptic currents were first detected automatically by the software using an amplitude threshold of 10 pA and an area threshold of 15 fC. All events were then visually reexamined. Any noise that spuriously met the trigger specifications was rejected (see also: [17]).

Paired-pulse recordings following dorsal root stimulation
Dorsal roots were stimulated via a suction electrode connected to a constant current stimulus isolator (A 360R, WPI, USA) at 0.1 ms pulse width. Excitatory postsynaptic currents (EPSCs) were recorded in the presence of strychnine (4 mM) and bicuculline (10 mM) and classified to be either Ad-or C-fiber-evoked according to their latency and threshold. Constant latencies and absence of failures during 10 Hz stimulation (for Ad-fibers) or 1 Hz stimulation (for C-fibers) were used as criteria for monosynaptic transmission [18]. Thresholds of stimulation intensity of Adand C-fibers did not differ significantly between sham-and CCItreated mice: Ad-fibers (sham: 203632 mA, CCI: 258644 mA), Cfibers (sham: 1.760.2 mA, CCI: 2.160.2 mA). The stimulation intensity was set to 200% of threshold values. Paired stimuli were applied at an interstimulus interval of 50 ms, 300 ms and 500 ms, respectively. Pairs of pulses were delivered at an interval of 15 s. For analysis at least 10 traces were averaged, the amplitudes of the first (P1) and the second (P2) EPSC were measured and the paired pulse ratio (PPR) was calculated as PPR = P2NP1 21 .

Immunohistochemical detection of c-Fos
Tissue was prepared 11 days after sham-or CCI treatment. Mice were briefly anesthetized with isoflurane and the hindpaw of the operated side was immersed into 52uC hot water for 20 seconds to induce c-Fos expression in spinal cord dorsal horn. Two hours later mice were deeply anesthetized with isoflurane and perfused with 0.9% saline followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB) (pH 7.4). Spinal cords were removed, postfixed in the same fixative overnight, cryoprotected by immersion in 20% sucrose in 0.1 M PB for 48h at 4uC and frozen in isopentane at -80uC.
Beginning from L6, the whole lumbar spinal cord was cut in transverse sections on a freezing microtome (CM 3050; Leica, Germany) at a thickness of 7 mm. Sections were serially mounted onto SuperFrost Plus slides (Menzel, Germany) and air dried.
Secondary antibodies used for immunofluorescence were all raised in donkey. Rinsed sections were incubated for 3 h at room temperature with anti-rabbit IgG conjugated to Cy5 and antimouse IgG conjugated to Cy3 (Jackson ImmunoResarch, USA; 1:200 in PBS). Each second series of slides was incubated with anti-rabbit IgG conjugated to Cy3 (only c-Fos).
After washing sections were covered with glass slides in a glycerol based medium containing Mowiol 4-88 (Calbiochem) and Propyl gallate (Sigma-Aldrich) to prevent fading.
Sections were viewed and photographed with a fluorescence microscope (Olympus BX51) equipped with a CCD camera (Olympus DP50) and measurements were performed with analySIS-Software (Soft Imaging Systems, Germany).
Digital images of the ipsilateral dorsal horn of the fourth lumbar segments (L4) were acquired with the 20x objective for neurons (NeuN), GABAergic cells (EGFP) and c-Fos. Quantification was accomplished in 10 to 16 randomly chosen sections of L4 at least 40 mm apart from one another.

Immunohistochemical detection of PSD-95, GAD67 and Synaptophysin
Eleven days after sham-or CCI treatment the mice were transcardially perfused as described above. The spinal cord was dissected out, postfixed for 4 hours, and after cryoprotection overnight snap frozen in isopentane at -80uC. Per animal four (5 mm thick) transverse sections of the L4 segment were cut on the cryomicrotome, mounted onto glass slides and air dried. For the staining of PSD-95 a modified antigen retrieval [19] by incubating the sections for 2 min at 37uC with Pepsin (Dako, Denmark) was necessary. After rinsing in PBS and blocking in 0.1 M PBS containing 0.1% Tween 80 for 30 min, sections were incubated over night at 4uC in the following primary antibodies: goat anti-PSD-95 (Abcam, 1:100) and mouse anti-GAD67 (Chemicon, 1:1000), diluted in blocking medium. Rinsed sections were incubated for 1 hour in rabbit anti-goat IgG conjugated to Cy5 (1:100) and rabbit anti-mouse IgG conjugated to Cy2 (1:200). Both secondary antibodies were purchased at Jackson Immuno Research. After rinsing, sections were incubated in monoclonal mouse anti-Synaptophysin labeled with OysterH 550 (1:1000, Synaptic Systems) for 1 hour. Coverslipping of the sections was performed as mentioned above.
Four sections from each animal were used for image acquisition. In each section two imaging windows (format: 102461024 pixel, zoom factor: 5, image size: 46.89 mm646.89 mm) within lamina II of the spinal cord dorsal horn were scanned ipsi-and contralateral to the surgery side, respectively. Confocal settings (laser power, gain and offset) were identical for all scans. Images were three times line averaged and scanned sequentially (in order to avoid fluorescent bleed-through) with a 63x oil-immersion lens at a zseparation of 0.3 mm on a Leica TCS SP5 confocal microscope. Pinhole was set at 1 Airy unit (100.26 mm), generating a calculated section thickness of 0.772 mm.
Analysis was performed by an experimenter blind to the treatment groups using ImageJ software (National Institutes of Health, USA). Images were processed by the rolling-ball algorithm at a radius of 10 pixels to subtract background. We used two different approaches to quantify colocalization of the three markers GAD67, PSD-95 and synaptophysin in superficial spinal dorsal horn. To perform a pixel intensity spatial correlation analysis we calculated Mander's colocalization coefficients pairwise for all three channels after automatic thresholding [20] using the JACoP plugin of ImageJ. The second approach was to generate a mask for each channel with automatic thresholds. The three masks were combined pairwise by Boolean operations. In addition the overlay of all three markers was quantified. We measured the overlaying areas in relative numbers as a function of the total signal area.

Two-photon laser-scanning microscopy
For two-photon laser-scanning microscopy GABAergic neurons were identified by epifluorescence and patch-clamped in the whole-cell configuration as described above. Since the EGFPlabeling did not fully access all dendritic branches and spines, the fluorescent dye calcein (1 mM, Invitrogen, USA) was added to the internal solution. The fluorescent dye was allowed to diffuse 10 minutes into the neurons before imaging. Imaging was performed on a two-photon laser-scanning microscope consisting of a Leica DM LFS A microscope (Leica, Germany) and a femtosecond Tisapphire laser (Chameleon-XR, Coherent, Germany). Excitation light (wavelength 800 nm) was focused by a 63x water immersion objective (0.9 NA). Scanning and image acquisition were controlled with Leica Confocal Software (LCS v.2.61). Emitted light was collected by non-descanned detectors. 3D volume samples of selected segments of the neurons were scanned, depending on the neuron, in areas from 10246256 pixels to 102461024 pixels, at 0.3 mm distance in z-direction.

Three dimensional analysis of dendritic spines
Spine analysis was performed using a 3D reconstruction software (described in detail in [21,22]) which allows measuring the geometric parameters of dendritic spines from confocal microscopic image stacks. The basic approach is the sequential comparison between the raw microscope image and an estimated image of a geometric model including dendrites and spines. The optimization process is repeated until an optimal match is achieved. Unbranched dendritic segments were selected from confocal image stacks and a 3-D median filter was used to reduce overall image noise. In contrast to the standard step by step approaches, where deconvolution is applied to compensate degradation caused by the point spread function of the microscope prior to reconstructing the spines, our system applied a vectorbased parametric geometric model in order to reduce the number of free parameters. The deconvolution was included as a later step during the model fitting process.
First, a raw model, including roughly estimated center axes for both, the dendritic segment and all protruding spines, was initiated. Each center axis was approximated by a set of model points interconnected by linear interpolation. The radius along this estimated axis was approximated by local radii, which were set at the level of each model point and also interconnected by linear interpolation. The tracking of dendrites and spines was based on a growing model approach. Starting with an initial element of at least two model points, which were positioned by the observer for each dendrite and each spine, the axes ''grow'' by adding new elements. Tracking was performed under visual control and interactive corrections were allowed at all stages of the tracing process.
In the final step the model was readjusted for an optimized match to the original raw microscope image. A convolved image of the raw model was generated by sampling the model with the same resolution as the microscope image and convolving it by the microscope point spread function, which resulted in a simulated microscope image. By iterative comparison of this image with the original microscope image the model parameters were adjusted and the best fitted model was then used for data analysis. The following parameters were calculated from the adapted geometric model to measure the shape of dendritic spines: volume, surface, length and averaged diameter. In addition, the density of spines was calculated for each dendritic segment.

Statistics
Analysis of the data was performed using SigmaPlot 11 (Systat). All values are means 6 s.e.m (if not stated otherwise). Unpaired Student's t-test (two-tailed), the non-parametric Mann-Whitney rank sum test, the one-way analysis of variance (ANOVA) and the two-way ANOVA, followed by Tukey's post-hoc test, were used for statistical comparison where appropriate. The significance level was set at 0.05.

Behavior
All tested CCI-operated but none of the tested sham treated mice expressed thermal ( Fig. 2A) and mechanical hyperalgesia (Fig. 2B) throughout the observation period of 9 days. mEPSC rate but not amplitude is reduced in CCI-treated animals We assessed the global excitatory input to identified GABAergic neurons in control (Fig. 3A) and in neuropathic animals (Fig. 3B). We recorded mEPSCs from EGFP-labeled GABAergic neurons in spinal cord lamina II. The mean rate (events ? s 21 ) of mEPSCs recorded from EGFP-labeled neurons in sham treated mice was 1.1460.19 (n = 26). In the CCI treated group rate of mEPSCs was significantly lower (0.5160.08, n = 34, P , 0.001, Fig. 3C and D). In contrast the mean EPSC amplitude was not significantly different in sham treated animals (29.562.5 pA) as compared to CCI-operated mice (26.061.5 pA, P.0.1, Fig. 3E and F).
Two mechanisms could well explain this decrease in mEPSC rate: a reduced number of synapses or a decrease of neurotransmitter release probability from presynaptic terminals. These two possibilities were explored next.
CCI-surgery does not change the density or morphology of dendritic spines of GABAergic lamina II interneurons nor the colocalization of GAD67 with markers of excitatory synapses We first tested if the number of excitatory synapses converging onto GABAergic neurons might be lower in CCI-animals. Dendritic spines are the predominant locations of excitatory synapses [23,24]. Their shape correlates with the functional state of a synapse [25]. As an estimate for the number of excitatory synapses of GABAergic neurons we used two-photon laserscanning microscopy to image and analyze 12 neurons in lamina II from CCI-treated mice and 13 neurons from a sham-treated control group (Fig. 4). Neither the density nor morphological parameters of dendritic spines in GABAergic neurons were significantly different between groups (Table 1).
In addition we performed immunohistochemical triple staining and confocal microscopy in five animals per group for GAD67 positive neurons in combination with PSD-95, an established marker for excitatory AMPAergic synapses in lamina II [26,27], and the presynaptic marker synaptophysin (Fig. 5). We used the isoform GAD67 of the GABA synthesizing enzyme glutamic acid decarboxylase (and the promoter for EGFP expression in the transgenic mice) as a marker for inhibitory GABAergic spinal neurons because the intensity of the EGFP labeling was too weak as compared to the synaptic markers. None of the quantification methods used (Mander's colocalization coefficients or analyzing the relative area of overlaid markers) did reveal any statistically significant differences between sham-and CCI-treated animals in any combination tested (Table 2).
Thus, the present data do not provide any evidence for either a loss of excitatory synapses or a morphological change in dendritic spines of spinal GABAergic neurons in animals with a CCI of the sciatic nerve.

Reduced transmitter release probability at terminals of Ad-and C-fibers contacting GABAergic neurons following CCI
We then tested if the release probability at excitatory synapses contacting GABAergic neurons might be reduced in CCI animals. As an estimate for release probability we evaluated the paired pulse ratio (PPR) of EPSCs evoked from presumably nociceptive Ad-and C-fiber afferents. Only EPSCs evoked monosynaptically from either Ad-fiber or from C-fiber stimulation were analyzed ( Fig. 6A and D). Paired pulse EPSC recordings were made from 21 EGFP-labeled neurons in sham-treated animals and from 20 EGFP-labeled neurons in CCI animals with monosynaptic input from Ad-fibers. In addition 16 neurons each in sham-treated and in CCI animals were studied with monosynaptic C-fiber input. The stimulation intensities to induce EPSCs were not significantly different in CCI-treated animals as compared to sham-treated controls for Ad-fibers (sham: 203632 mA, CCI: 258644 mA) or for C-fibers (sham: 1.760.2 mA, CCI: 2.160.2 mA). All recorded neurons expressed paired pulse depression but at significantly different levels ( Fig. 6B and E). The PPRs of both, Ad-and Cfiber-evoked responses were significantly higher in neurons recorded from CCI animals as compared to controls (Fig. 6C and F). The increase of the PPR in neuropathic animals indicates a decreased release probability at excitatory synapses between Ador C-fibers and spinal GABAergic neurons.
One could speculate that the mean amplitudes of evoked EPSCs could also be decreased in slices taken from animals that underwent a CCI-surgery. This was however not the case: We found no significant differences when analyzing the amplitudes of the first responses evoked by either Ad-(sham: 207635 pA; CCI: 189625 pA) or C-fiber stimulation (sham: 202633; CCI: 192662). This finding is not surprising, since the amplitudes of evoked EPSCs do not only depend on the synaptic release probability and stimulus strength but also on variable experimental conditions (e.g. different numbers of active synapses due to inevitable preparational differences, different arborization of dendritic trees within the slice preparation), resulting in a high variance. Ca 2+ -dependent signaling in GABAergic neurons presynaptically controls excitatory input CCI-treatment leads to an increase in Ca 2+ concentration in spinal dorsal horn neurons [28]. We thus asked if a Ca 2+dependent synthesis and/or release of an inhibitory retrograde messenger in GABAergic neurons could account for the reduced global excitatory drive [29]. We added the Ca 2+ chelator BAPTA to the recording pipette to prevent any rise in intracellular Ca 2+ ions of the GABAergic neurons under study. Under this condition, neither mEPSC rate ( Fig. 7A and B) nor amplitude ( Fig. 7C and D) nor the PPRs of both Ad- (Fig. 7E) and C-fiber-evoked (Fig. 7F) responses were significantly different between CCI-and shamtreated mice.
These findings indicate that a rise in postsynaptic Ca 2+ concentration is necessary to presynaptically depress excitatory synaptic drive to GABAergic neurons in CCI-treated animals suggesting the involvement of an inhibitory retrograde messenger.
Various classes of molecules have been found to act as retrograde messengers inhibiting the release of neurotransmitters. This includes lipid-derived messengers such as endocannabinoids acting on cannabinoid-1 (CB 1 ) receptors, gases such as nitric oxide (NO) and conventional neurotransmitters including GABA acting on GABA B receptors [29][30][31][32][33][34]. We next explored these possibilities.

Endocannabinoids acting on CB 1 receptors do not depress excitatory drive to GABAergic neurons
We first tested if endocannabinoids are tonically released by GABAergic neurons and depress excitatory drive to GABAergic neurons via the CB 1 receptor. If so, one would expect that blocking CB 1 receptors would lead to an increase in mEPSC rate. Neither in sham-treated animals nor in CCI-mice mEPSCs were affected by CB 1 receptor blockade with AM 251 (Table 3). Likewise, bath application of CB 1 receptor agonist ACEA failed to affect mEPSC rates in slices from naïve animals. Thus endocannabinoids acting on presynaptic CB 1 receptors are not the cause for reduced excitatory drive to GABAergic neurons in CCIoperated mice.

Neither GABA nor nitric oxide modulate global excitatory drive to GABAergic neurons
We then tested if bath application of putative retrograde messengers GABA, or the NO-donor SNAP is capable of reducing mEPSC rates in normal animals. Bath application of neither GABA (10 mM, n = 4) nor SNAP (200 mM, n = 5) had any effect on mEPSC rates (Table 3). Thus, neither GABA nor NO is likely the retrograde messenger that may reduce global excitatory drive of GABAergic neurons.

c-Fos activation following noxious thermal stimulus
We next asked if the reduced excitatory drive from primary afferent nociceptive nerve fibers onto spinal GABAergic neurons in slices obtained from CCI-animals has any functional meaning in vivo. According to our hypothesis noxious stimulation should activate fewer GABAergic neurons in CCI-treated animals as compared to sham-treated animals. We used the expression of c-Fos protein as a marker for neuronal activity [35]. In CCI-treated animals the number of EGFP-labeled neurons was not different as compared to sham-treated mice. A noxious stimulation induced, however, c-Fos in significantly fewer GABAergic neurons in the medial half of lamina II in the CCI group as compared to controls (Table 4 and Fig. 8). These findings indicate that in neuropathic  animals noxious stimuli indeed fail to activate as many spinal GABAergic neurons as they normally do in control animals. The overall c-Fos expression as well as the expression in non-EGFP neurons only was not different between groups. Since non-EGFP neurons constitute a very heterogeneous group including excitatory and inhibitory, even GABAergic neurons, we cannot draw any firm conclusion from this finding.

Discussion
The present study revealed a novel mechanism of impaired spinal inhibition likely contributing to neuropathic pain: the global downregulation of excitatory synaptic drive to GABAergic neurons. Furthermore, the excitatory drive from Ad-and C-fibers was reduced thus causing an impaired recruitment of inhibitory neurons during nociception. The resulting attenuation of feedforward inhibition may open a ''spinal gate'' of nociception as illustrated in Fig. 1B, thereby contributing to pain amplification in neuropathic animals and perhaps in pain patients.
Spinal GABAergic neurons are predisposed to function as gatekeepers for nociception. About 30% of the interneurons located in spinal lamina II are GABAergic [36,37]. Blocking spinal GABA receptors lowers pain thresholds [3], and spinal application of GABA receptor agonists increases pain thresholds in pathological pain states such as neuropathic pain [4,5].
A number of potential mechanisms that could lead to impaired GABAergic inhibition in nociceptive pathways have been explored. For example, cell death of spinal inhibitory neurons has been demonstrated in several animal models for neuropathic pain [38]. Other studies have, however, shown that the symptoms of neuropathic pain also arise in the absence of any detectable cell death [39,40]. Previous findings from our group suggest that the excitability of spinal GABAergic neurons is unchanged in animals with CCI [8]. GABAergic output is impaired in animals with a peripheral nerve injury, as demonstrated by a reduction of primary afferent evoked inhibitory postsynaptic currents [6]. In addition, CCI has been shown to lead to a shift of the anion gradient in some lamina I neurons. Thereby, the action of GABA is reduced and shifted towards excitation in neuropathy [7].
Here we studied the excitatory input to spinal GABAergic neurons which were identified by the expression of EGFP under the promoter of GAD 67. This labels a representative population  Table 2. Quantification of colocalization of GAD67 with the postsynaptic marker PSD-95 and the presynaptic marker synaptophysin in spinal lamina II.
of GABAergic neurons with respect to their electrophysiological, neurochemical and morphological characteristics [16,41]. CCI induced a significant decrease of mEPSC rate but not amplitude in spinal lamina II GABAergic interneurons in the present study. This demonstrates that CCI leads to a decrease in global excitatory drive to identified GABAergic neurons. Our data are in line with other findings showing a CCI-induced decrease of mEPSC rate in presumably inhibitory interneurons, as judged by their action potential firing pattern and cell morphology, whereas other spinal neuronal subgroups exhibited enhanced mEPSC rates [42,43].
The rate of mEPSCs reflects the release probability of neurotransmitter from presynaptic terminals given the number of synapses is constant [44]. In the central nervous system the number of active synapses can change e.g. after induction of synaptic long-term potentiation [45]. Thus, the reduced mEPSC rate in GABAergic neurons found in this study could either be due to a reduced release probability of neurotransmitter from presynaptic terminals or to a reduced number of excitatory synapses at GABAergic neurons. Assuming that the majority of excitatory synapses terminate on dendritic spines [23,24], a loss of spines should reflect decreased excitatory input [46]. Dendritic spines are highly plastic and have been shown to alter in size and shape within minutes in response to changes in the pattern or the intensity of synaptic input (for review see [25]). Accordingly, the quantification of dendritic spines can be used as a measure of the number of excitatory synapses. In contrast to some regions in the brain GABAergic neurons in the spinal cord do form dendritic spines [47]. In our sample of 3D-reconstruction all GABAergic neurons (n = 25) displayed dendritic spines. These neurons represent a random selection of EGFP-labeled neurons, which was not used for electrophysiology, since calcein filling reduced the quality of the recordings. We therefore could not quantify the proportion of spine bearing neurons in our electrophysiological experiments. In parasagittal slices, neither density nor length, diameter, surface, or volume of dendritic spines on spinal lamina II GABAergic neurons were altered in CCI-treated animals as compared to sham-operated mice. Dendritic spines on GABAergic neurons in lamina II of the spinal cord have not been studied extensively yet. Furthermore our results do not exclude that the number of excitatory synapses contacting dendritic shafts [48]   might have dropped in CCI-treated animals. To also account for these synapses we performed confocal microscopy of immunohistochemical stainings for GAD67 neuronal structures, PSD95, a marker for excitatory synapses [26,27], and the presynaptic maker synaptophysin. These stainings also revealed no differences between CCI-treated and control mice. Thus, we found no evidence that the number of excitatory synapses on spinal GABAergic neurons is altered in neuropathic animals.
We next tested if release probability of neurotransmitter from these primary afferent fibers terminating onto lamina II GABAergic neurons may be changed. An inverse relationship exists between the probability of neurotransmitter release and the PPR [49,50]. In CCI-treated animals the PPR calculated for synapses of primary afferent Ad-or C-fibers with spinal GABAergic neurons was always higher as compared to sham controls suggesting a reduced release probability. One can, however, not fully exclude a potential postsynaptic contribution to changes in PPR [51]. Collectively, both phenomena PPR and mEPSCs point to a decrease in release probability under different experimental conditions (i.e. action potential dependent and action potential independent release of neurotransmitter).
A potential explanation for a reduced release probability at synapses is the release of a retrograde, inhibitory messenger by the postsynaptic neuron. This often requires a rise in postsynaptic Ca 2+ ion concentration (for review see [29]). In animals with CCI the intracellular Ca 2+ concentrations in neurons of the superficial and deep spinal dorsal horn are indeed enhanced [28]. Our data now demonstrate that an enhanced Ca 2+ ion concentration in GABAergic neurons is indispensable for the depression of their global excitatory drive. The requirement of postsynaptic Ca 2+ rise in combination with a reduced release probability strongly suggests the involvement of a retrograde inhibitory messenger.
Possible candidates include endocannabinoids, which are released following intracellular Ca 2+ rise and act on presynaptic CB 1 receptors thereby reducing neurotransmitter release at excitatory synapses [31]. CB 1 receptors are abundantly expressed in the spinal cord dorsal horn [52,53]. Our data demonstrate, however, that neither the CB 1 receptor agonist ACEA nor the receptor antagonist AM 251 exerted any detectable effects on the mEPSC rates. GABA, which can act as a retrograde messenger as well inhibiting presynaptic neurotransmitter release via GABA B receptors [30], or NO, a retrograde messenger in the CNS shown to increase [33,34] or in some cases decrease [32] transmitter release, did not alter mEPSC rates either. This is in line with findings that nitric oxide synthase is only expressed in a small subset of EGFP-expressing GABAergic neurons in the spinal superficial laminae [16,41]. The present data thus make it unlikely that endocannabinoids acting on CB 1 receptors, GABA acting on GABA B receptors or NO account for the diminished global excitatory drive onto GABAergic neurons in neuropathic animals. Future studies will be needed to clarify the nature of the retrograde signaling.
We further showed that activation of GABAergic neurons in lamina II of the spinal dorsal horn is significantly impaired in CCIanimals as compared to controls. We used the expression of c-Fos protein as a marker for nociceptive activity in spinal dorsal horn at the single cell level [35]. After a CCI-induced initial rise c-Fos expression returns to normal in laminae I and II within 10 to 14 days [54,55]. In addition, noxious heat-evoked c-Fos expression in laminae I and II are not different in control versus CCI-treated animals when the global expression is assessed [56]. We performed our experiments on day eleven post-surgery and our results on the global c-Fos expression in lamina II are in full agreement with those previous studies. Only when we quantified c-Fos expression selectively in identified GABAergic neurons, we found a reduced expression in the CCI-treated animals indicating that activation of spinal GABAergic neurons is severely reduced in neuropathic animals.
In summary, we showed that peripheral neuropathy triggers Ca 2+ -dependent signaling pathways in spinal GABAergic neurons, resulting in a global downregulation of their excitatory drive. In addition, monosynaptic excitation of GABAergic neurons by Adand C-fibers during noxious stimulation is reduced, thus impairing normal feedforward inhibition and opening a spinal gate for nociceptive messages.  Table 4. c-Fos expression in GABAergic neurons following a noxious heat stimulus was reduced in CCI-treated mice.