Spinal Inhibition of GABAB Receptors by the Extracellular Matrix Protein Fibulin-2 in Neuropathic Rats

In the central nervous system, the inhibitory GABAB receptor is the archetype of heterodimeric G protein-coupled receptors (GPCRs). Receptor interaction with partner proteins has emerged as a novel mechanism to alter GPCR signaling in pathophysiological conditions. We propose here that GABAB activity is inhibited through the specific binding of fibulin-2, an extracellular matrix protein, to the B1a subunit in a rat model of neuropathic pain. We demonstrate that fibulin-2 hampers GABAB activation, presumably through decreasing agonist-induced conformational changes. Fibulin-2 regulates the GABAB-mediated presynaptic inhibition of neurotransmitter release and weakens the GABAB-mediated inhibitory effect in neuronal cell culture. In the dorsal spinal cord of neuropathic rats, fibulin-2 is overexpressed and colocalized with B1a. Fibulin-2 may thus interact with presynaptic GABAB receptors, including those on nociceptive afferents. By applying anti-fibulin-2 siRNA in vivo, we enhanced the antinociceptive effect of intrathecal baclofen in neuropathic rats, thus demonstrating that fibulin-2 limits the action of GABAB agonists in vivo. Taken together, our data provide an example of an endogenous regulation of GABAB receptor by extracellular matrix proteins and demonstrate its functional impact on pathophysiological processes of pain sensitization.


INTRODUCTION
G protein-coupled receptors (GPCRs) modulate a wide range of physiological processes and new drug candidates are continually being developed for selective GPCRs (Lagerstrom and Schioth, 2008;Hauser et al., 2017;Topiol, 2018). Functional GABAB receptors are expressed as obligate heterodimers of the two subunits GABAB1 (B1) and GABAB2 (B2) (Jones et al., 1998;Kaupmann et al., 1998), and GABAB-mediated inhibition is an essential control of neuronal activity throughout the central nervous system. GABAB receptor subunits bind several partner proteins (Bettler et al., 2004;Pin and Bettler, 2016;Fritzius et al., 2017) that regulate the receptor activation in pathophysiological conditions such as chronic pain (Laffray et al., 2007;Laffray et al., 2012;Zemoura et al., 2016;Malcangio, 2018). Among these binding partners, extracellular matrix (ECM) proteins have been shown to regulate postsynaptic GABAB-mediated inhibition (Saghatelyan et al., 2003). However, the importance of such interaction in a pathophysiological context has not been established yet.
In the central nervous system, aggregates of ECM molecules appear as so-called perineuronal nets that surround soma and dendrites (Fawcett et al., 2019). These nets are heterogeneous in structure and composition, incorporating proteins derived from both neuronal and glial cells. The expression of ECM proteins is developmentally and spatially regulated in the central nervous system. Some of these molecules are up-regulated following adult nerve injury (Quraishe et al., 2018;Roumazeilles et al., 2018) and ECM proteins have been implicated in synaptic differentiation and plasticity . The emerging mechanisms comprise clustering of receptors and changes in their composition and function (Dityatev and Schachner, 2006;Ferrer-Ferrer and Dityatev, 2018). To date, beside the specific class of adhesion GPCR (Langenhan et al., 2016) only very rare studies have considered interactions between ECM proteins and GPCRs (Saghatelyan et al., 2003;Yeh et al., 2008) and the mechanisms by which ECM proteins regulate GPCR activation in neurons remains largely unknown.
In this context, the present study focuses on the role of fibulin-2, an extracellular binding partner of the B1a variant of GABAB1 subunit (Blein et al., 2004) that is mainly found in presynaptic compartments (Vigot et al., 2006). We demonstrate that fibulin-2 up-regulation hampers GABAB activation and signaling by agonists. In contrast, we show that the loss of fibulin-2 facilitates agonist-induced activation of the GABAB receptor and decreases the efficiency of antagonists. Therefore, we suggest that fibulin-2 may limit conformational changes that are necessary for the receptor activation, and facilitate an inactive conformation of the receptor.
As a functional consequence of this mechanism, we demonstrate in a primary cell culture model that fibulin-2 regulates the GABAB-mediated presynaptic inhibition of neurotransmitter release. We further show the physiological impact of this mechanism on neuropathic pain sensitization in the dorsal horn of the rat spinal cord. Our data indicated fibulin-2 up-regulation in the dorsal spinal cord of neuropathic rats. In such chronic pain conditions, the efficiency of GABABmediated inhibition was improved by knocking-down the protein expression with in vivo application of anti-fibulin-2 small interference (si) RNA.
Finally, our demonstration that the fibulin-2 endogenous extracellular matrix protein inhibits GPCR activity in a pain context highlights a novel mechanism of disinhibition, and provides the groundwork for further studies of pathophysiological regulations.

Primary Cell Cultures
Neuronal cells were obtained from rats (Sprague Dawley) at 18 days of embryonic development. The embryos were delivered by cesarean section from deeply anesthetized rats and killed by decapitation. Pieces of lumbar region of the spinal cord or cortex were exposed to a 1X trypsin solution (Gibco) for 25 min at 37 • C. These pieces were then mechanically dissociated by forcing them through fine-tipped pipettes several times. Cells were plated on glass coverslips at a density of 150,000/100 µL. The glass coverslips were previously coated overnight at 37 • C with a 0.1 mg/mL polylysine solution for cortex and 1 mg/mL for spinal cord (Sigma, Saint Louis, MO, United States) and for 2 h with a 0.05 mg/mL laminine solution (Sigma). The cultures were kept in Serum-free Neurobasal TM medium (Gibco), supplemented with B27 (Gibco) and Glutamax (Gibco). Half of the medium was changed twice a week.
The efficacy of siRNA to downregulate the expression of fibulin-2 mRNA was assessed by SYBRGreen-based quantitative reverse transcription -polymerase chain reaction (qRT-PCR) both on cell cultures (data not shown) and after intrathecal injections in-vivo (see Results section, Expression of Fibulin-2 in vivo).

qRT-PCR
Expression analysis of fibulin-1 and fibulin-2 mRNA was performed with the DyNAmoTM SYBR Green qPCR kit (Finnzymes, Espoo, Finland). Triplicate runs were performed and Succinate dehydrogenase Complex, Subunit A (SDHA) was used as a normalizer. The relative level of expression was calculated using the comparative (2 CT) method (Livak and Schmittgen, 2001).

Two-Photon-Fluorescence Lifetime Imaging Measurement (2P-FLIM)
The interaction between GFP-tagged proteins and t-dimer DsRed-tagged proteins was studied as described previously (Laffray et al., 2007(Laffray et al., , 2012 using quantitative FRET (Fluorescence or Förster Resonance Energy Transfer) determination with FLIM (Fluorescence Lifetime Imaging Measurements) using Time Correlated-Single Photon Counting (Becker and Hickl, Berlin, Germany). To assess protein interactions, we calculated a relative FRET efficiency for each acquisition as follow: FRET Efficiency (%) = (τ Dmean -τ DA )/τ Dmean × 100 where τ is the time constant of the exponential fit with τDmean being the mean lifetime of the donor fluorophore (GFP) when expressed alone and τDA being the lifetime of the donor fluorophore in the presence of the acceptor (t-dimer DsRed). Results were expressed as a mean FRET efficiency (n = 21 cells or more in each condition) ± SEM.
Due to the mono-exponential decay of GFP and their spectra compatibility, the GFP-DsRed couple is very well suited for quantitative FRET determination with FLIM (Fluorescence Lifetime Imaging Measurements) (Tramier et al., 2002;Stubbs et al., 2005). Moreover, the t-dimer DsRed is a tandem of monomeric red fluorescent protein that matures rapidly, does not form aggregates, and has minimal emission when excited at 900 nm, a wavelength optimal for GFP and minimizing the excitation of t-dimer DsRed (Campbell et al., 2002).
The interaction between GFP-tagged protein and t-dimer DsRed-tagged protein was studied using two-photon-Fluorescence Lifetime Imaging Measurement (2P-FLIM) with the SPCM and SPC Image software (Becker & Hickl).
For GFP excitation, we used a two-photon pulsed excitation (890 nm), given by the Coherent Mira 900-F laser 5W (Coherent inc. Laser Group, Santa Clara, CA, United States). All acquisitions were undertaken with a 100x, 1.4NA, Leica oil objective lens on a Leica TCS SP2 inverted confocal microscope. The laser repetition frequency was 76 MHz which gave a 12 ns temporal window for lifetime measurements. Photon detection was performed in cells using the SPRC160, a homemade system based on a 16 channel multi-anode photomultiplier head (PML16, Becker&Hickl) and a concave holographic grating allowing spectral dispersion on the detector. The Becker & Hickl TCSPC 730 card (Becker & Hickl, Berlin, Germany) determined the time between fluorophore excitation and photon emission, as well as routing information associated to spectral range. This card was driven by the SPCM software which allowed fluorescence decay curve measurements using a single-spot mode; and fluorescence decay curve fits, allowing lifetime determination were obtained using the SPC Image software (both software from Becker & Hickl). The choice between mono-exponential decay and bi-exponential decay fitting was done using the reduced χ2 parameter. Lifetime values were extracted from three spectral channels, from 494 to 542 nm, and averaged. To ensure reliability of our measurements, we checked the lifetime stability upon this spectral range. As a second control, the first channels were used to ensure no autofluorescence in measured cells, which otherwise could result in artefactual FRET positive response.

Co-immunoprecipitation
The co-immunoprecipitation experiments were performed in COS-7 cells after transfection, according to the previously described procedure (Laffray et al., 2012). Briefly, total protein extract was homogenized in ice-cold PBS containing 1% CHAPS, 25 mM Hepes and 150 mM NaCl, and protease inhibitors, pH 7.4. Homogenates were incubated overnight with 50 µl of magnetic beads (protein-A sepharose, Sigma) at 4 • C. After centrifugation, the supernatants were transferred to the tube containing 5 µg of specific anti-myc (Roche) or anti-flag (Sigma) antibodies. Then, samples were washed twice, incubated in Laemmli buffer (Sigma) and subjected to Western blot detection. After electrophoresis using sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE), samples were transferred to polyvinylidene difluoride (PVDF) membranes. Blots were blocked in 5% skimmed milk for 30 min and then incubated with antibodies (mouse anti-myc, 1/1000, Roche; guinea pig anti-flag, Sigma; mouse anti-GABAB2, 1/1000, Eurobio, Les Ulis, France) at 4 • C overnight. Then detection was performed with the appropriate HRP-conjugated secondary antibodies at 1/1000 for 1 h30 (anti-mouse or anti-guineapig, Dako products, Agilent). Immunoreactivity was developed using the enhanced chemiluminescent method and visualized with a Syngene device (ChemiGenius 2XE model; Synoptics Ltd, Cambridge, United Kingdom).

Immunohistochemistry on Tissue Sections
Rats were perfused with 4% paraformaldehyde. For electron microscopy, glutaraldehyde (0.2%) was added to the fixative solution. The lumbar spinal cord was rapidly dissected out, and post-fixed for 2 h in 4% paraformaldehyde. Tissue samples were then processed for light or electron microscopy.
Immunostaining of the spinal cord was visualized under a SPE confocal microscope (Leica Microsystems). Images to be compared were collected during the same session using identical scanning settings. They were then imported into "ImageJ" free software (version 1.42q) (NIH, Bethesda, MA, United States) for quantitative analysis. Background was subtracted by thresholding. For fibulin-2 and GABAB1a intensity assessment, the mean gray level corresponding to fluorescence intensity was measured (Zaky et al., 2018). Results were expressed as a percentage of the intensity in sham animals. Counting was performed on 5 sections per rat, 8 (fibulin-2) or 4 (GABAB1a) rats per condition (sham or SNL). The extent of colocalization between fibulin-2 and GABAB1a was quantitatively assessed using a semi-automatized procedure with the Image J software (Landry et al., 2004;Zaky et al., 2018). After thresholding, colocalization area was assessed in the entire fields acquired in the dorsal horn of the spinal cord (4 sections per rat, 8 rats per condition). Results were expressed as a percentage of GABAB1a total labeling that is seen in the area.
Miniature EPSCs were recorded in voltage clamp mode (holding potential −60 mV) in the presence of 10 −6 M Tetrodotoxin (TTX). In such experimental conditions, no action potential could be evoked following injection of positive current into the recorded neuron, demonstrating a complete blockade of action potential propagation. Signals were amplified using an Axoclamp 2B amplifier and digitized at 10 kHz. Individual events were extracted from raw data using mini-analysis software 1 over a duration of 2 min and visually filtered according to their general shape.

Membrane Preparation
Primary cell cultures of rat cortex (8 million of cells/mL) were homogenized using a Teflon glass grinder (10 up-and-down strokes at 1500 rpm) in 5 ml of homogenization buffer (1 mM EGTA, 3 mM MgCl 2 and 50 mM Tris-HCl, pH 7.4) supplemented with 0.25 mM sucrose. The crude homogenate was centrifuged for 5 min at 500 rpm (4 • C) and the supernatant was recentrifuged for 10 min at 18000 rpm (4 • C). The resultant pellet was washed in 4 ml of homogenization buffer and recentrifuged in similar conditions. Aliquots of protein were stored at -80 • C until assay. Protein content was measured according to the method of Bowery (2006) using bovine serum albumin (BSA) as standard.

[ 35 S]GTPγS Binding Assays
Membrane aliquots were thawed and resuspended in a buffer containing 1mM EGTA, 3 mM MgCl 2 , 100 mM NaCl, 3 mU/ml adenosine deaminase and 50 mM Tris-HCl, at pH 7.4. The incubation was started by addition of the membrane suspension (400 microliters of membranes in each tube at a protein concentration of 0.1 mg/ml) to the previous mixture of 0.5 nM [ 35 S]GTPγS, 50 µM GDP and appropriate concentrations of drugs and was performed at 30 • C for 120 min with shaking. Incubations were terminated by adding 3 ml of icecold resuspension buffer followed by rapid filtration through Whatman GF/C filters presoaked in the same buffer. The filters were rinsed twice with ice-cold resuspension buffer, transferred to vials containing 5 ml of OptiPhase HiSafe II cocktail and the radioactivity trapped was determined by liquid scintillation spectrometry (Packard 2200CA). Non-specific binding was defined as the remaining [ 35 S]GTPγS binding in the presence of 10 µM unlabelled GTPγS. Non-specific binding was subtracted from total bound radioactivity to determine [ 35 S]GTPγS specific binding.

Competition Binding Assay
The mixture for the binding assay contained Hanks' Balanced Salt Solution (HBSS) (NaCl, 8 g/l; CaCl 2 , 0.14 g/l; KCl, 0.40 g/l; NaHCO 3 , 0.35 g/l; glucose, 1 g/l; MgCl 2 6H 2 O, 0.10 g/l; KH 2 PO 4 , 0.06 g/l; MgSO 4 7H 2 O, 0.10 g/l and Na 2 HPO 4 , 0.05 g/l, pH 7.4), the membrane suspension (2.07 µg of membrane proteins) and 2 nM [ 3 H]-CGP54626 with or without different concentrations of baclofen in a final volume of 0.5 ml. The mixture was incubated at 5 • C for 20 min, and then the reaction was terminated by adding ice-cold wash buffer followed by rapid filtration through Whatman GF/C filters (presoaked with 0.33% polyethyleneimine with 50 mM Tris-Cl, pH 7.4, for 30 min). The filters were rinsed twice with ice-cold wash buffer, transferred to vials containing 5 ml of OptiPhase HiSafe II cocktail and the radioactivity trapped was determined by liquid scintillation spectrometry (Packard 2200CA).

Animal Models and Behavioral Tests
Adult male Wistar rats (250-300 g; Charles River France, St Aubin les Elbeuf, France) were used. The experiments followed the ethical guidelines of the International Association for the Study of Pain and were approved by the local ethics committee in Bordeaux (AP 1/04/2005). Persistent neuropathic pain in rats was evoked by ligation of the right L5-L6 spinal nerve (Spinal Nerve Ligation, SNL) (Decosterd and Woolf, 2000). Mechanical response thresholds were monitored in SNL, and sham-operated (sciatic nerve was exposed but the L5 and L6 sciatic nerves were not ligated) animals at 1 day before surgery (reference value for each animal), and at day 12 (before and after intrathecal injections of baclofen) and at day 15 (24 h after the last siRNA injection). Rats were placed in the testing cage 1 h before the test for habituation. The withdrawal threshold of the leg on the operated side was determined in response to mechanical stimuli applied to the plantar surface of the foot. The limb withdrawal threshold was measured by an electronic device (Bioseb, France) that was derived from Von Frey filaments (Kalmar et al., 2003;Lever et al., 2003). Sham-operated rats were used as controls.

In vivo Intrathecal Injections and Mechanical Allodynia
For intrathecal injections of baclofen or RNAs (anti-fibulin-2 siRNA or mismatch RNA), sham and SNL rats were implanted with a catheter (PE-10; Phymep, France) inserted into the subarachnoid space (Fossat et al., 2007). The threshold to mechanical stimulation was monitored before and after surgery, and after the various injections. Results were expressed as percentages of changes in the threshold of neuropathic rats. Rats were tested for mechanical allodynia before surgery (day-1, reference value set to 100%). The threshold to mechanical stimulation was monitored again in sham-operated and SNL animals 12 days after surgery (day 12), after baclofen injection (1 µg in 10 µL; Sigma B5399) on the same day (day 12, baclofen), on day 15 after injection of anti-fibulin-2 siRNA or mismatch RNA (day 15, RNA), and finally after baclofen injection on the same day (day 15, RNA + baclofen). Anti-fibulin-2 siRNA (siRNA) and mismatch RNA (mmRNA) were injected daily from day 12 to day 14 according to a published protocol (Luo et al., 2005) and Neuromics instructions (Neuromics, Northfield, MN, United States). In some experiments, saclofen (30 µg, Sigma S166) or CGP55845 (10 µg, Tocris 5584) were co-applied in vivo with baclofen.

Statistical Analysis
Statistical analyses were conducted using SigmaPlot 12.0 software (SigmaStat, Systat Software Inc, San Jose, CA, United States). Results were presented as mean ± standard error of the mean (SEM). The Student's t-test was used for twosample comparisons. For multiple sample comparisons, one-way ANOVA or two-Way ANOVA followed by a Tukey post hoc test was performed for FRET studies. One-way ANOVA followed by a Tukey test was performed for behavioral studies. Regarding the electrophysiology study, for all comparisons of cumulative distributions, 2-sample Kolmogorov-Smirnov test was used. When not indicated otherwise, the Tukey test was used. p < 0.05 was considered significant.
Since various cell types normally secrete fibulin-2 in the extracellular matrix, we investigated the possible role of endogenous fibulin-2 in restraining GABAB conformational changes. For this purpose, we incubated spinal cultures with anti-fibulin-2 siRNA (without overexpressing fibulin-2) ( Figure 2B). When fibulin-2 endogenous expression was blocked with siRNA, FRET interactions were increased in both unstimulated and baclofen-activated GABAB receptor. In contrast, mismatch RNA had no effects on energy transfer between GABAB subunits.
Taken together, these data show that fibulin-2 hampers conformational changes of the GABAB receptor, both constitutively and upon agonist application. These changes in the GABAB conformation correlate with fibulin-2-induced loss in G-protein coupling. Finally, we showed with a siRNAbased strategy that endogenous fibulin-2 exerts a tonic inhibitory effect on GABAB activation as revealed by FRET analysis and GTP binding assay.

Mechanism of Fibulin-2 Action
With regards to the mechanisms accounting for fibulin-2induced impairment of GABAB activation, we first confirmed the interaction of the ECM protein with the GABAB receptor. Our attempts to perform co-immunoprecipitation of endogenous fibulin-2 and GABAB receptor subunits failed and the detection of endogenous fibulin-2 in western blot remained below the level of sensitivity, probably due to technical issues with immunoblotting of fibulin-2. We then switched to a cell culture approach and performed immunoprecipitation after transfection of tagged-proteins in COS-7 cells. Our study demonstrated that co-immunoprecipitation of myc-GABAB1a with an antimyc antibody retained fibulin-2-Flag ( Figure 3A, left panel) in cultures that were triple transfected with myc-GABAB1a, HA-GABAB2, and fibulin-2-Flag. The reverse experiments indicated that fibulin-2-Flag also precipitated myc-GABAB1a in the same culture model (Figure 3A, right panel). Interestingly, the coimmunoprecipitation of myc-GABAB1a and fibulin-2-Flag was partly lost in double transfected cell cultures, lacking the GABAB2-expressing plasmid ( Figure 3A). This suggested that most of myc-GABAB1a was unable to bind fibulin-2-Flag, most probably because it was not exposed to the plasma membrane in this condition, remaining trapped in the endoplasmic reticulum due to the relatively low amount of endogenous GABAB2 that normally ensures proper trafficking of the receptor (White et al., 1998;Kuner et al., 1999;Margeta-Mitrovic et al., 2000). A control experiment was conducted in cell cultures triple transfected with myc-GABAB1b, GABAB2 and fibulin-2-Flag ( Figure 3A). No co-immunoprecipitation between myc-GABAB1b and fibulin-2-Flag was detected in agreement with the lack of fibulin-2-induced changes in GABAB1b/GABAB2 FRET interactions (see Figure 2D). This confirmed that fibulin-2 interacts with GABAB1a but not GABAB1b which lacks the fibulin-2-interacting sushi-domain.
We then considered that fibulin-2 may trigger a possible loss of the heterodimeric state of the GABAB receptor as previously demonstrated for the B1b/B2 post-synaptic subtype (Laffray et al., 2007). Indeed, heterodimer dissociation would lead to similar changes in FRET interactions and GTP binding activity. We tested this hypothesis with co-immunoprecipitation of myc-GABAB1a and HA-GABAB2 subunits transfected in COS-7 cells in the presence or in the absence of fibulin-2 ( Figure 3B). No differences could be shown in B1a/B2 association in the different conditions tested, whereby ruling out the possibility that fibulin-2 alters the dimeric state of the GABAB receptor.

Functional Effects of Fibulin-2 on Inhibitory Transmission
Then, we asked whether fibulin-2 modifies B1a/B2-mediated inhibition. Since the B1a subunit is mainly found in presynaptic compartments (Vigot et al., 2006) we focused on presynaptic GABAB inhibition on spinal neuron cultures (Figure 4). To further investigate the role of fibulin-2 in the control of synaptic transmission, we analyzed the effects of baclofen (10 µM) and saclofen (100 µM) on the amplitude and frequency distributions of miniature excitatory EPSCs recorded in the presence of TTX (1 µM). In preparations transfected with mismatch RNA (Figure 4A1), comparison of the mEPSC amplitude distribution ( Figure 4A2) revealed no significant change between control and baclofen conditions (Kolmogorov-Smirnov test, KS). Moreover, the kinetics of averaged mEPSCs was not affected by baclofen application (see inset in Figure 4A2). Overall, in 9 recorded neurons, the average mEPSC amplitude remained unaltered by baclofen application (Figure 4A3, paired t-test). By contrast, the inter-event interval was significantly shifted toward the right (Figure 4A4), indicating a decrease in mEPSC frequency (KS test, p < 0.001, n = 9). Further saclofen addition (100 µM), in the continuous presence of baclofen, induced a recovery of mEPSCs frequency in 7 out of 9 recorded cells. Overall (Figure 4A5), baclofen induced a significant decrease in mEPSCs frequency to 58.2% ± 6.6 of control (p < 0.05, paired t-test). Whereas in the presence of saclofen, the average mEPSC frequency returned to control value (resp 91% ± 9). These results demonstrate a clear presynaptic inhibition of excitatory synaptic transmission in cultured spinal neurons. Similar results were obtained in control preparations as well as preparations exposed to transfecting agent alone (n = 9 and n = 7 respectively, data not shown).
In preparations transfected with fibulin-2 siRNA ( Figure 4B1), baclofen had no effect on the amplitude distribution ( Figure 4B2, KS test), and the kinetics of averaged mEPSCs remained unchanged (inset in B2). No statistical effect of baclofen on average mEPSCs amplitude could be observed ( Figure 4B3, paired t-test, n = 7). Baclofen induced a shift in the frequency distribution curve (Figure 4B4, p < 0.001), indicating a clear presynaptic inhibition of mEPSCs. However, in contrast with preparations transfected with mismatch siRNA, further saclofen application (100 µM) failed to antagonize baclofen effects (Figure 4B2, KS test), on mEPSCs frequency. Overall (Figure 4B5), baclofen application induced a decrease in mEPSCs frequency to 44.4 ± 11.4% of control values in 7 out of 7 cells (paired t-test, p < 0.05). In 5 out of these 7 neurons, further application of saclofen did not alter mEPSC frequency (54 ± 0.18% of control, paired-t-test, p < 0.05), showing that the antagonist loses its capacity to block presynaptic baclofen-induced activation of the GABAB receptor at the dose used when fibulin-2 is knocked-down.
Finally, in preparations transfected with fibulin-2 siRNA, the average mEPSC frequency was significantly lower than in preparations transfected with mismatch RNA (Figure 4C1). Importantly, this effect was preserved in the presence of baclofen ( Figure 4C1). This result indicates that the absence of fibulin-2 facilitates the action of baclofen and the presynaptic inhibitory effect of GABAB signaling. In contrast, no change was observed in the mean mEPSC amplitude (Figure 4C2).

Expression of Fibulin-2 in vivo
We next investigated whether changes in fibulin-2 expression can alter GABAB inhibition in vivo. Since GABAB is involved in the modulation of nociceptive transmission (Bowery, 2006) we studied fibulin-2 expression in rats with spinal nerve ligation (SNL) that induced neuropathic pain. Real-time RT-PCR showed a marked increase in fibulin-2 mRNA levels in the dorsal spinal cord of SNL rats ipsilateral to the injury, as compared to sham animals ( Figure 5A; One-way ANOVA, F (3,21) = 11.67, p < 0.01; 141.58% ± 7.2 in SNL rats; p < 0.01 vs. sham). Fibulin-2 overexpression was limited to the spinal cord since no increase was noticed in the dorsal root ganglia ( Figure 5B). However, the spinal cell types that secretes fibulin-2 remain to establish. Overexpression was specific of fibulin-2 since the expression of the related fibulin-1 showed no changes ( Figure 5C).
We checked fibulin-2 upregulation at the protein level with immunohistochemistry. Immunostaining was hardly detectable in control animal dorsal horn (Figure 5D1, left). Intensity was much higher in SNL rats (Figure 5D1, right) where discrete staining was observed. Changes in immunostaining intensity were quantified and showed a 60% increase in the ipsilateral dorsal horn of SNL rats as compared to sham animals ( Figure 5D2; 160.7% ± 23.7 of the sham intensity; p < 0.01, t-test). No changes were detected in the contralateral dorsal horn (Figure 5D2), or in the ventral horn (data not shown).
We further compared the distribution of fibulin-2 with B1a subunit (Figure 6), and CGRP (Figure 7), a marker of peptidergic nociceptive inputs with light and electron microscopy double labeling experiments.
both B1a and fibulin-2 labeling were also found independently, as single labeling, both intracellularly and at the membrane.
Regarding fibulin-2/CGRP double labeling, both signals were seen most often in close vicinity with light microscopy (Figures 7a-f). Electron microscopy confirmed that fibulin-2 is localized in the vicinity of the plasma membrane, around sensory nerve endings filled with CGRP-containing secretory granules (Figures 7g,h). These data suggested fibulin-2 interaction with GABAB receptors is located, at least partly, on nociceptive nerve endings in the spinal dorsal horn.

Functional Role of Fibulin-2 on Neuropathic Pain Sensitization
Finally, to assess the functional role of fibulin-2 on GABAB inhibition in chronic pain conditions, we tested the behavioral effect of anti-fibulin-2 siRNA intrathecal injection in SNL neuropathic rats (Figure 8). Using the von Frey test, we compared the effects of topical intrathecal application of baclofen and GABAB antagonist on paw withdrawal threshold following intrathecal treatment with anti-fibulin-2 siRNA or mismatch RNA.
One-way ANOVA indicated an effect on the withdrawal threshold [ Figure 8A; F (5,30) = 32.85, p < 0.01]. In the absence of baclofen, anti-fibulin-2 siRNA produces a slight, though significant, analgesic effect ( Figure 8A; 40.4% ± 1.4 in SNL vs. 48.9% ± 1.9 in SNL + siRNA; p < 0.05). The injection of mmRNA does not significantly affect the withdrawal threshold (Figure 8A; 40.4% ± 1.4 in SNL vs. 42.1% ± 2.3 in SNL + mmRNA; p < 0.05). Interestingly, the baclofen treatments enhance the effects of antifibulin-2 siRNA. More particularly, post hoc analysis shows that the baclofen plus siRNA treatment is more efficient than baclofen alone, or baclofen plus mmRNA to alleviate pain. Surgery resulted in a ∼60% decrease in paw withdrawal threshold ( Figure 8A). The injection of vehicle (iFect) after surgery did not further change the mechanical threshold of neuropathic rats (data not shown). The injection of baclofen at day 12 after surgery induced an increase in mechanical threshold indicating GABAB-mediated antinociceptive effects (Figure 8A; 40.4% ± 1.4 in SNL vs. 66.8% ± 1.4 in SNL + baclo; p < 0.01). This response to baclofen was significantly increased in SNL rats treated with antifibulin-2 siRNA (Figure 8A; 66.8% ± 1.4 in SNL + baclo vs. 87.5% ± 3.9 in SNL siRNA + baclo, p < 0.01). In contrast, the mean threshold for SNL animals was significantly increased upon siRNA treatment after baclofen application. Changes in response to baclofen were not observed in animals injected with mmRNA ( Figure 8A; 66.8% ± 1.4 in SNL + baclo vs. 72.0% ± 2.6 in SNL mmRNA + baclo, p > 0.05) showing paw withdrawal threshold similar to those observed in neuropathic rats before treatment.
The co-injection of baclofen and saclofen, in SNL rats, reversed the effects of baclofen alone (Figure 8B; 128.12% ± 2.62 in baclofen iFect vs. 108.79% ± 7.3 in baclofen plus saclofen iFect, p < 0.01). In the baclofen plus saclofen iFect group, paw withdrawal threshold was similar to that measured in SNL rats without injection, indicating the persistence of mechanical allodynia. In contrast, after anti-fibulin-2 siRNA application, saclofen was unable to significantly block baclofen effects. The increase of paw withdrawal threshold remained statistically unchanged (Figure 8B; 155.7% ± 9.4 in baclofen siRNA vs. 142.56% ± 6 in baclofen plus saclofen siRNA, p > 0.05). In both "baclofen" and "baclofen plus saclofen" groups, the paw withdrawal threshold was higher than after the control iFect injection. The injection of mmRNA did not change the effects of baclofen alone, or baclofen plus saclofen injection, and the withdrawal threshold remained different between the two groups ( Figure 8B; 128.5% ± 5.75 in baclofen mmRNA vs. 103.2% ± 6.7 in baclofen plus saclofen mmRNA; p< 0.05). These data demonstrate that fibulin-2 limits GABAB receptor inhibition in vivo. Knocking down fibulin-2 in vivo facilitates conformational changes and activation of the receptor, therefore enhancing GABAB-induced alleviation of pain.

GABAB1a Interaction With Fibulin-2
GABAB interacting proteins were demonstrated to control several aspects of the receptor functions (reviewed in Xu et al., 2014) including receptor dimerization [14.3.3ζ (Couve et al., 2001;Laffray et al., 2012)], intracellular targeting [MUPP1 (Balasubramanian et al., 2007), CHOP (Sauter et al., 2005)], signaling transduction and desensitization [ATF4 (Vernon et al.,FIGURE 5 | Fibulin-2 is upregulated in the spinal cord of neuropathic rats. (A) qRT-PCR analysis of fibulin-2 mRNA in the ipsilateral spinal dorsal horn of sham rats, neuropathic rats (SNL), and SNL rats injected with anti-fibulin-2 siRNA (SNL + siRNA) or with mismatch RNA (SNL + mmRNA). Data are expressed as percentage of sham ± SEM. Fibulin-2 expression increased in ipsilateral dorsal horn of neuropathic rats. This increase was efficiently prevented by siRNA injection. The injection of mmRNA had no effect on fibulin-2 expression. *p < 0.05, **p < 0.01 "sham" vs. "SNL"; n.s.: p > 0.05. (B) qRT-PCR analysis of mRNA of fibulin-2 in the ipsilateral dorsal root ganglia of sham and neuropathic (SNL) rats. Data are expressed as percentage of sham ± SEM. No significant changes are seen (n.s.: p > 0.05 "sham" vs. "SNL). (C) qRT-PCR analysis of mRNA of fibulin-1 in the ipsilateral spinal dorsal horn of sham and neuropathic (SNL) rats. Data are expressed as percentage of sham ± SEM. No significant changes are observed (n.s.: p > 0.05 "sham" vs. "SNL). (D1) Immunohistochemistry for endogenous fibulin-2 in the ipsilateral dorsal horn of sham (a), and neuropathic (SNL) (b) rats. The dotted line indicates the dorsal limit of the spinal cord section. Framed areas of the lamina II (II) in (a,b) are displayed at higher magnification in (c,d), respectively. Fibulin-2 expression was very low in sham animals (arrowheads in c) whereas an intense staining could be seen in the SNL group (arrowheads in d). (D2) The quantification of the signal intensity confirmed the significant difference in the ipsilateral dorsal horn between sham and SNL groups. In contrast, no changes is seen in the contralateral dorsal horn (n = 3 sections from 7 animals in sham and SNL groups). **p < 0.01; n.s.: p > 0.05. FIGURE 6 | Endogenous co-expression of B1a and fibulin-2 in sham and neuropathic rats. (A1) Immunohistochemistry for endogenous B1a (green) and fibulin-2 (red) in lamina II (II) of the dorsal horn of sham (a), and neuropathic (SNL) (b) rats. The dotted line indicates the dorsal limit of the spinal cord section. (A2) Micrographs are higher magnification of the framed areas in (A1) and correspond to sham (a-c), and SNL (d-f) rats. In sham animals, B1a subunit (green, arrows) and fibulin-2 (red, arrowheads) are distributed throughout the dorsal horn but show very little colocalization. In contrast, colocalization was more frequent in SNL rats (double arrowheads). However, some single labeling was also found for B1a (arrow) and fibulin-2. Bar = 15 µm (same for all). (B) Quantification of colocalization in the ipsilateral dorsal horn, expressed as the percentage of B1a labeling colocalized with fibulin-2 (n = 8 sections from 3 animals in the sham group; n = 7 sections from 3 animals in the SNL group). * * p < 0.01. (C) Double immunogold labeling for B1a (15 nm colloidal gold diameter) and fibulin-2 (5 nm colloidal gold diameter) in the ipsilateral dorsal horn of SNL rats. Both proteins colocalized at the plasma membrane of spinal neurons (double arrowheads in a,b). Gold particles association was found essentially at extra-synaptic sites (b, the synapse is indicated with large arrows). Single fibulin-2 labeling was also found in the vicinity of the plasma membrane (a, arrow). A, axonal nerve ending; bar = 100 nm.
FIGURE 7 | Endogenous co-expression of fibulin-2 and CGRP in neuropathic rats. (a-f) Immunohistochemistry for endogenous fibulin-2 and CGRP in lamina II of the dorsal horn of neuropathic (a-f) rats. Framed areas of the lamina II (II) in a to c are displayed at higher magnification in (d-f), respectively. CGRP (green) and fibulin-2 (red) staining was frequently seen overlapping or in close apposition (double arrowheads). Bar = 15 µm (same for all). (g,h) Double immunogold labeling for fibulin-2 (15 nm colloidal gold diameter) and CGRP (5 nm colloidal gold diameter) in the ipsilateral dorsal horn of neuropathic rats. CGRP was found in secretory granules (d,e, arrows). Secretory granules-containing processes were surrounded by fibulin-2 labeling, at the plasma membrane (d,e, arrowheads), mostly at extra-synaptic sites (large arrows in d). Bar = 100 nm in (d); 50 nm in (e). (Perroy et al., 2003)]. Only few interactions between partner proteins and the GABAB1a sushi domains have been described. GABAB1a sushi domains play important roles in receptor trafficking by stabilizing the GABAB1a/GABAB2 dimer at the cell surface (Biermann et al., 2010;Hannan et al., 2012). But the possibility that the sushi domains contribute to signaling transduction and physiological effects of the GABAB receptor has received only limited attention so far. However, it has been demonstrated that targeting GABAB1a sushi domains may impair GABAB receptor function (Tiao et al., 2008).

2001), GRK4
Actually, the most remarkable interaction was recently described between the sushi domains and the shed amyloid-β precursor protein ectodomain APP and its physiological effect was to regulate GABAB1a-mediated regulation of synaptic transmission (Rice et al., 2019).
The fibulin-2 is an extracellular matrix protein that has been shown to interact with the GABAB1a sushi domains with yeast two-hybrids and pull-down assays (Calver et al., 2002;Blein et al., 2004). Fibulin-2 modulates spinal axon growth trajectories during development (Schaeffer et al., 2018). It also mediates proneurogenic effects of transforming growth factor-beta1 in adult neural stem cells (Radice et al., 2015). Fibulin2 expression in the central nervous system remains largely unexplored but recent studies indicate it is produced by astrocytes (Radice et al., 2015;Schaeffer et al., 2018). In the present study we identified the colocalization between the GABAB receptor and the fibulin-2 in situ in the dorsal spinal cord, and we provide evidence for a direct interaction in vitro after the overexpression of tagged GABAB and fibulin-2 proteins.

Regulation of GABAB Receptor Activation by Fibulin-2
We assessed the effects of fibulin-2 on GABAB receptor activation using FLIM measurement of FRET interactions as previously described (Laffray et al., 2012). FRET provides an accurate measure of possible changes in the distance (<10 nm) and/or orientation between the fluorophores. Hence, it allows to study conformational changes between fluorescently tagged GABAB1a (GABAB1a-GFP) and GABAB2 (GABAB2-DsRed) receptor subunits (Fowler et al., 2007;Laffray et al., 2007). Baclofen-induced increase in FRET efficiency between GABAB1a and GABAB2 was demonstrated by others (Laviv et al., 2010). In contrast, our own studies indicated that the FRET efficiency between GABAB1b and GABAB2 is not, or slightly affected by baclofen (Laffray et al., 2012) (see also Figure 2D of the present study), probably because the absence of sushi domains in the GABAB1b subunit that changes the distance and/or the orientation of the GFP fused at the N-terminal of the B1a/b subunits.
proteins, such as 14-3-3 (Laffray et al., 2012), fibulin-2 is unlikely to regulate the dimeric state of the receptor according to the co-immunoprecipitation results. Therefore, we favored the hypothesis that fibulin-2 regulates the receptor conformation. In line with this hypothesis, the higher FRET efficacy upon anti-fibulin-2 siRNA treatment indicated that less constraints apply to the GABAB receptor subunits in the absence of fibulin-2, thus increasing the probability of conformational changes and receptor activation. Importantly, these data also demonstrated that endogenous fibulin-2 exerted tonic inhibition of GABAB activation.
Other examples of interactions with extracellular matrix proteins have been shown to regulate receptor activation. Tenascin R interacts with the B1b subunit and blocks postsynaptic-mediated GABAB inhibition on pyramidal neurons of the hippocampus via the HNK-1 carbohydrate (Saghatelyan et al., 2003). The NMDA glutamate receptor interacts with the matrix protein Reelin that regulates NMDA synaptic retention and surface distribution (Groc et al., 2007). The activity of AMPA receptors and their function in short-term synaptic plasticity also depends on the interaction between receptors and extracellular matrix proteins (Frischknecht et al., 2009).
It may be hypothesized that fibulin-2 acts through two different modes of action, i.e., alteration of ligand affinity or receptor conformation. However, fibulin-2 has different effects on the action of agonists (GABA and baclofen) and competitive antagonists (saclofen and CGP), although they interact with the same binding site. Hence, fibulin-2 is unlikely to simply modify ligand binding to the receptor. This is further confirmed by the competition binding assay ( Figure 3C) that indicates there is no changes in the ligand affinity. The extracellular matrix protein may rather alter the receptor conformation, thus changing the GG-protein-mediated effects of agonists.
One can propose that the interaction between fibulin-2 and the Sushi domains themselves influence the efficiency of ligands. Indeed, a regulatory role of the Sushi domains has been shown for glucagon and VIP receptors where the N-terminal extracellular domain of receptor contains highly conserved amino acid residues which are essential for its intrinsic binding activity (Carruthers et al., 1994;Couvineau et al., 1995).
Another possible mechanism relies on fibulin-2 dimerization (Sasaki et al., 1997). As a dimer, fibulin-2 could then bridge two GABAB heterodimers with each other. The formation of such higher order GABAB oligomers (Pin et al., 2009;Comps-Agrar et al., 2011) could account for limitations of structural changes and of GABAB activation.
Alternatively, the interaction between fibulin-2 and the Sushi domains may also interfere with the receptor trafficking and regulate GABAB inhibition as demonstrated for ATF4, another interactor (Corona et al., 2018).

Consequences of Fibulin-2/GABAB Interaction in Pain Sensitization
Nerve injury leading to pain sensitization has already been shown to induce up-regulation of extracellular matrix proteins in adult rats. More precisely, peripheral lesions increase the expression of fibronectin in the dorsal spinal cord, ipsilateral to the lesion (Nasu-Tada et al., 2006;Tsuda et al., 2008). Such injury-induced overexpression of matrix proteins is involved in pain sensitization by driving an increase of P2X4 receptor expression on microglia (Tsuda et al., 2003(Tsuda et al., , 2008Ulmann et al., 2008). Interestingly, a recent transcriptomics study indicates that the extracellular matrix pathways are the most largely regulated pathways in animal models of chronic pain, and these data are corroborated by data on human low back pain (Parisien et al., 2019), thus strengthening the hypothesis that the extracellular matrix plays a major role in pain sensitization processes.
The B1a/B2 heterodimer is mainly distributed in the presynaptic compartment (Vigot et al., 2006;Biermann et al., 2010). In superficial laminae of the spinal cord, the GABAB receptor presynaptic inhibition is more effective to suppress mechanical noxious transmission than innocuous transmission, which may account for a part of the mechanism of the analgesic effects of baclofen (Fukuhara et al., 2013). These results suggest that GABAB receptors tonically inhibit glutamate release from primary sensory fibers at a subset of synapses in deep dorsal horn, being more specific of the early phase of synaptic excitation (Salio et al., 2017). In agreement with this presynaptic localization, we showed that fibulin-2 controlled presynaptic inhibition mediated by the GABAB receptor. Therefore, we conclude that fibulin-2 exerts it action through interactions with the presynaptic receptor subtypes, although post-synaptic effect cannot be ruled out. In the spinal dorsal horn, we also showed that fibulin-2 is expressed in superficial laminae, in the vicinity of CGRP-containing nerve endings. The colocalization between these markers indicates that fibulin-2 can exert its regulatory effect on primary nociceptive afferents. Fibulin-2 up-regulation could thus limit the GABABmediated suppression of neurotransmitter release by sensory primary afferents in the superficial laminae of the dorsal horn. Fibulin-2 could also impede GABAB presynaptic inhibition on interneurons. In line with this role of the extracellular matrix protein, the knockdown of fibulin-2 enhances the efficacy of baclofen to alleviate mechanical allodynia in neuropathic rats. Our morphological data showed that only a subset of B1a subunit is associated to fibulin-2 at the cell surface. In line with previous reports (Frischknecht et al., 2009), our study suggests that the extracellular matrix protein could trap receptors, and control their activation in specific membrane domains. It may also restrain and filter the exchange of receptors between different subcellular compartments.
Taken together, our data indicate that fibulin-2 overexpression induces structural changes in the GABAB receptor that restrain its activation by the endogenous ligand, GABA, thus allowing a fine tuning of the receptor activation in sub-domains of the plasma membrane.
The regulation by the extracellular matrix may also limit the efficiency of therapeutic strategies aiming to stimulate directly the GABAB receptor. Our data are in line with the actual recommendations for clinical use of baclofen. In fact, baclofen administration is not proposed in EFNS guidelines of pharmacological treatment of neuropathic pain (Attal et al., 2006) and its general effectiveness as an analgesic is limited (Bowery et al., 2002). Administration of positive allosteric modulators may increase the effectiveness of baclofen treatment. However, recent studies point to a limited efficacy of GABAB allosteric modulators such as rac-BHFF in neuropathic mice (Zemoura et al., 2016;Malcangio, 2018). To overcome this limitation, co-application of inhibitors of extracellular matrix proteins, or positive allosteric modulators (Zemoura et al., 2016), together with baclofen, might provide a new mean to enhance baclofen efficiency in clinics, without using high doses that produce unwanted side effects such as sedation.
Together with the recent elucidation of other mechanisms involving reduced GABA release (Moore et al., 2002) or loss of GABAA inhibition (Coull et al., 2005;Knabl et al., 2008) our data support the view that disinhibition dramatically twists the excitability of spinal neurons and leads to pain sensitization (Woolf and Salter, 2000). Targeting GABAB associated proteins, and especially extracellular matrix proteins, may be of therapeutic interest by enhancing the action of classical pain killers.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The animal study was reviewed and approved by Comité local d'éthique, Université de Bordeaux.

AUTHOR CONTRIBUTIONS
M-AP, ML, RR-P, and FN conceived the study and planned the experiments. M-AP performed the FRET and immunohistochemistry experiments, and the pain behavior studies. GB-G performed the binding and the GTPγS assays. YL and FF performed the patch-clamp recordings. AF performed the qRT-PCR experiments. RB-B performed the co-immunoprecipitation experiments and the pain behavior studies. M-AP, ML, RR-P, YL, and FN wrote the manuscript. All authors contributed to the article and approved the submitted version.