In the rat cerebellum, Golgi cells receive serotonin-evoked inputs from Lugaro cells (L-IPSCs), in addition to spontaneous inhibitory inputs (S-IPSCs). In the present study, we analyze the pharmacology of these IPSCs and show that S-IPSCs are purely GABAergic events occurring at basket and stellate cell synapses, whereas L-IPSCs are mediated by GABA and glycine. Corelease of the two transmitters at Lugaro cell synapses is suggested by the fact that both GABAA and glycine receptors open during individual L-IPSCs. Double immunocytochemical stainings demonstrate that GABAergic and glycinergic markers are coexpressed in Lugaro cell axonal varicosities, together with the mixed vesicular inhibitory amino acid transporter. Lugaro cell varicosities are found apposed to glycine receptor (GlyR) clusters that are localized on Golgi cell dendrites and participate in postsynaptic complexes containing GABAA receptors (GABAARs) and the anchoring protein gephyrin. GABAAR and GlyR/gephyrin appear to form segregated clusters within individual postsynaptic loci. Basket and stellate cell varicosities do not face GlyR clusters. For the first time the characteristics of GABA and glycine cotransmission are compared with those of GABAergic transmission at identified inhibitory synapses converging onto the same postsynaptic neuron. The ratio of the decay times of L-IPSCs and of S-IPSCs is a constant value among Golgi cells. This indicates that, despite a high cell-to-cell variability of the overall IPSC decay kinetics, postsynaptic Golgi cells coregulate the kinetics of their two main inhibitory inputs. The glycinergic component of L-IPSCs is responsible for their slower decay, suggesting that glycinergic transmission plays a role in tuning the IPSC kinetics in neuronal networks.
GABAergic and glycinergic neurotransmission coexist in many structures of the CNS. Colocalization of the two neurotransmitters is prominent in the spinal cord (Todd and Sullivan, 1990; Ornung et al., 1994; Taal and Holstege, 1994; Yang et al., 1997) and other sensory and motor centers (Wenthold et al., 1987; Ottersen et al., 1988; Wentzel et al., 1993). The observation of GABAergic terminals facing glycine receptor (GlyR) clusters (Triller et al., 1987) and of mixed boutons presynaptic to domains containing GABAA receptor (GABAAR) and probably GlyR (Todd et al., 1996) first suggested that the two transmitters might be coreleased at the same synapse. Indeed, coactivation of the GABAAR and GlyR at a single contact, and by the content of a single vesicle, was demonstrated recently at interneuron–motoneuron connections (Jonas et al., 1998; O'Brien and Berger, 1999). Copackaging of the two neurotransmitters in vesicles is probably achieved by a single vesicular inhibitory amino acid transporter (VIAAT) (McIntire et al., 1997; Sagné et al., 1997).
In the spinal cord and brainstem, the only places where cotransmission of GABA and glycine is formally demonstrated (Jonas et al., 1998;O'Brien and Berger, 1999), pure glycinergic transmission and pure GABAergic transmission also occur (Ornung et al., 1994; Yang et al., 1997). Thus, networks of inhibitory interneurons using GABA and glycine as their transmitters have a complex and versatile functional organization. Analyzing this organization at the cellular and synaptic levels is a prerequisite to understand the specific roles played by GABA and glycine. The cerebellar cortex provides a good model for this study, because it is composed of a small number of well defined cell types (Palay and Chan-Palay, 1974) and contains both neurotransmitters. GABA is the predominant inhibitory neurotransmitter in this structure, but glycine is also found in inhibitory interneurons of the granule cell layer and in fibers of the molecular layer (Wilkin et al., 1981;Ottersen et al., 1987, 1988). The neuronal glycine transporter GlyT2 (Borowsky et al., 1993; Liu et al., 1993), a marker for glycinergic neurons (Poyatos et al., 1997; Spike et al., 1997), is expressed in the granular layer, in Golgi cell axonal varicosities, and in some varicose fibers of the molecular layer (Zafra et al., 1995). Several GlyR subunit mRNAs are detected in the cerebellar cortex (Malosio et al., 1991; Racca et al., 1998), and functional glycine receptors are present in immature granule cells (Kaneda et al., 1995) and in Golgi cells (Dieudonné, 1995).
In a previous study (Dieudonné and Dumoulin, 2000) we demonstrated the existence of a new connection between the cerebellar Lugaro cells and the Golgi cells. In the present work we show that Golgi cells receive both pure GABAergic inhibition from basket cells (S-IPSCs) and mixed GABAergic and glycinergic inhibition from Lugaro cells (L-IPSCs). Lugaro cells corelease the two transmitters on apposed postsynaptic arrays of GABAA and glycine receptors, the kinetics of which is coregulated by the postsynaptic Golgi cell. Therefore, the specific functions of GABA and glycine transmission may result from the different kinetics of their IPSCs.
MATERIALS AND METHODS
Slice preparation. Cerebellar thin slices were prepared from male Wistar rats, aged 11–21 d, following the method originally described by Llinás and Sugimori (1980) with slight modifications (Llano et al., 1991). Briefly, the cerebellum was dissected and immediately cooled to 0°C. A parasagittal cut was made in the paravermis, and parasagittal slices (180–300 μm thick) were cut from the vermis with a microslicer (Dosaka, Kyoto, Japan, or Leika, Nussloch, Germany). The slices were kept at 34°C for 1–9 hr before being transferred to the recording chamber. On some occasions slices were allowed to cool slowly from 34°C to room temperature 1 hr after slicing.
Slices were visualized using a 40× water-immersion objective (0.75 numerical aperture; Axioskop; Carl Zeiss) and infrared optics (illumination filter of 750 nm and a Sony CCD camera from which the infrared blocking filter had been removed). Golgi cells were selected for recording both on visual criteria, as explained previously (Dieudonné, 1995), and on the basis of their characteristic passive electrical properties (Dieudonné, 1998). Recordings were restricted to the Golgi cells of lobules I–VIII. The lobules IX and X, which belong functionally to the vestibulocerebellum, were discarded for this study to limit the variability associated with the specialization of the different regions of the cerebellar cortex.
Patch-clamp recording. All experiments were performed at room temperature (20–25°C). The recording chamber was continuously perfused at a rate of 1.5 ml/min with a saline solution, pH 7.4, containing (in mm): 125 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, 1.25 NaH2PO4, 26 NaHCO3, and 25 glucose, bubbled with a mix of 95% O2 and 5% CO2. The same solution was used during dissection and slicing. Strychnine (Sigma, St. Louis, MO) and gabazine (Research Biochemicals, Natick, MA) were bath applied. Serotonin (Sigma) was also bath applied, usually for 3 min, and washed for at least 15 min between applications to avoid desensitization.
Cells were recorded with an Axopatch 200B (Axon Instruments) in the voltage-clamp mode of the amplifier and held at −70 mV. The access resistance and pipette capacitance were carefully compensated. Pipettes had a resistance of 2–4 MΩ and were filled with the following internal solution (in mm): 142 CsCl, 10 HEPES, 1 EGTA, 5 MgCl2, 0.1 CaCl2, 4 Na-ATP, and 0.4 Na-GTP; pH was adjusted to 7.3 withN-methyl-d-glucamine. Recordings of spontaneous synaptic activity were stored on a digital audio tape (DTR-1204; BioLogic) and digitized off-line.
Electrical stimulations were performed using a patch pipette (3 MΩ) filled with the following solution (in mm): 140 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, and 30 HEPES; pH was adjusted to 7.3 with NaOH. Minimal stimulations of basket cell axons could be obtained in the vicinity of Purkinje cell bodies with voltage pulses of a few volts and 30–50 μsec duration.
Data analysis. PClamp softwares (Axon Instruments) were used for the acquisition of all recordings. Data were filtered at 2 kHz and digitized at 10 kHz. Spontaneous synaptic currents were detected automatically using the ACS software kindly provided by P. Vincent (Institut des Neurosciences, Paris). All detected events were subjected to visual inspection, and EPSCs were discarded on the basis of their fast decay time course [their time constant is faster than 3 msec (Dieudonné, 1998)]. Selected events could then be transferred in a file with the format of clampex files and further analyzed using the Clampfit program of the PClamp package. To fit the decay of the IPSCs, the offset was forced to 0 pA, and two exponential functions were tried first. The fit was accepted when the ratio of the time constants was >3 or when it was between 2.5 and 3, if the two components were of comparable amplitude (20–80% of the decay). In other cases a single exponential function was fitted to the data. More precise values for the time constants of decay were obtained by fitting the integral of the averaged IPSCs with one or two exponential functions. These values were used for the linear regression plots (see Fig. 8).
Fluorescence immunocytochemistry. Female Sprague Dawley rats (200 gm; Janvier) were deeply anesthetized with pentobarbital (60 mg/kg body weight, i.p.) and perfused intracardially with 4% paraformaldehyde (PFA) in PBS, pH 7.4. The cerebellum was removed, post-fixed for 12–15 hr in 4% PFA, and cut into 30-μm-thick sections with a vibratome in sagittal or transverse planes. Sections were immersed for 20 min in 50 mm ammonium chloride in PBS and rinsed. An additional fixation with a mixture of methanol:acetic acid (95:5) for 10 min at −20°C was performed when using the mAb 4a. After a preincubation step of 30 min in 0.1% gelatin and 0.1% Triton X-100 in PBS, overnight incubation was performed at 4°C in the latter solution with a monoclonal anti-GlyRa/b antibody alone (mAb 4a; 1:200) (Pfeiffer et al., 1984) or in combination with a polyclonal antibody against (1) VIAAT (1:500) (Dumoulin et al., 1999), (2) the GABAARγ2 subunit (1:100) (Somogyi et al., 1996), (3) calretinin (1:2000; Swant, Bellinzona, Switzerland) (Schwaller et al., 1993), or (4) GlyT2 (1:2500; Chemicon). The anti-GABAARγ2 antibody was also combined with (5) a monoclonal anti-gephyrin antibody (mAb 7a; 1:200; Boehringer Mannheim, Indianapolis, IN) (Pfeiffer et al., 1984). A monoclonal anti-GAD65 antibody (1:500; Boehringer Mannheim) (Gottlieb et al., 1986) was combined with (6) the anti-calretinin antibody or (7) the anti-GlyT2 antibody. Finally, two of the polyclonal antibodies, raised in different species, were combined: the anti-calretinin antibody (rabbit) and the anti-GlyT2 antibody (goat) (8). After rinses, sections were incubated for 2 hr with the secondary antibodies. The mAb 4a was revealed by an indocarbocyanine (CY3)-coupled goat anti-mouse IgG. For the revelation of different combinations (1–8), the CY3-coupled goat anti-mouse IgG was associated with an FITC-coupled goat anti-rabbit IgG for combination 1 or with a biotin-coupled goat anti-rabbit IgG for combinations 2, 3, and 6. For the latter combinations, an additional incubation of 90 min in FITC-coupled streptavidin was performed after the rinses. For combination 4, we used a CY3-coupled donkey anti-mouse IgG and an FITC-coupled donkey anti-sheep IgG; for combination 5, we used an FITC-coupled goat anti-mouse IgG and a CY3-coupled goat anti-rabbit IgG; and for combination 7, we used an FITC-coupled horse anti-mouse IgG and a CY3-coupled donkey anti-sheep IgG. Finally, combination 8 was revealed using an FITC-coupled donkey anti-rabbit IgG and a CY3-coupled donkey anti-sheep IgG. All secondary antibodies (Jackson ImmunoResearch, West Grove, PA) were used at 1:200. Sections were mounted with Vectashield (Vector Laboratories, Burlingame, CA) before examination with a standard epifluorescent microscope (Leica DMR). No staining was observed when primary antibodies were omitted. In addition, sections from spinal cord were used as a control for the specificity of antibody labeling.
Immunocytochemical data quantification. To investigate the proportion of Lugaro axonal varicosities that were GABAergic, glycinergic, or both, we quantified the results of the calretinin–GAD65, calretinin–GlyT2, and GlyT2–GAD65 double immunostainings. Only Lugaro cell varicosities that could be identified along a fiber were counted on digitalized images. The images corresponding to the second marker were then superimposed, and the colocalization was analyzed. In each case, between 150 and 200 varicosities (n) were counted, and the results are expressed as a percentage. We similarly quantified on GlyT2–GlyRα/β double immunostainings the proportion of Lugaro axonal varicosities associated with the GlyR and, conversely, the proportion of GlyR clusters associated with GlyT2-positive terminals.
Double stainings of GlyRα/β and GABAARγ2 subunits revealed an association without colocalization of the two subunits. The absence of colocalization is not caused by an image shift between the two wavelengths. First, strict colocalizations are displayed (see Fig. 4, all panels). Second, as noted in Results, GABAARγ2 clusters were often surrounded by, and not juxtaposed to, GlyRα/β clusters. Extensive quantification of the apposition could not be achieved for two technical reasons. First, because of the uneven detection of GABAARγ2 clusters in the sample, when combined with methanol fixation, areas with strong staining had to be selected for observation. Second, proper quantification should involve a measure of the distance between clusters along Golgi cell dendrites, which is impossible to perform without staining the dendrite itself.
Pharmacological profile of spontaneous and serotonin-evoked IPSCs recorded from Golgi cells
IPSCs were recorded from Golgi cells in the whole-cell configuration of the patch-clamp technique. As described previously (Dieudonné and Dumoulin, 2000), Golgi cells receive a spontaneous inhibitory input composed of small IPSCs (S-IPSCs) occurring at low frequency and a serotonin (5-HT)-dependent input composed of large IPSCs evoked by bath application of serotonin (L-IPSCs). Whereas L-IPSCs can be attributed to the activation by 5-HT of the Lugaro neurons, a neglected interneuron type of the cerebellar cortex (Lugaro, 1894), the synaptic origin of S-IPSCs is not known. A characteristic feature of L-IPSCs is their occurrence at regular intervals, which reflect the rhythmic firing of presynaptic Lugaro cells at several Hertz. The periodicity of L-IPSCs and, in many cells, their large amplitude allow them to be distinguished from S-IPSCs, which are always small and occur at random intervals. In any cell the frequency of L-IPSCs can be calculated simply by subtracting the frequency in the baseline, before serotonin application, from the frequency of IPSCs in the presence of serotonin (Dieudonné and Dumoulin, 2000).
To determine the nature of the receptors involved in S-IPSCs and in L-IPSCs, we performed bath applications of gabazine (20–30 μm) and strychnine (0.3–1 μm), which are selective inhibitors of GABAA and glycine receptors, respectively, in Golgi cells (Dieudonné, 1995). Gabazine blocked entirely the S-IPSCs in 11 of 12 Golgi cells (Fig.1 A). In the remaining cell the gabazine-resistant S-IPSCs were small (<30 pA). Their frequency matched the frequency of a distinct population of large IPSCs (recorded in control conditions), which were similar in amplitude to L-IPSCs evoked in this cell by serotonin (data not shown). This suggests that gabazine-resistant S-IPSCs in this cell represent the spontaneous activity of a presynaptic Lugaro cell.
To strengthen the conclusion that the vast majority of S-IPSCs are purely GABAergic, lower concentrations (2 μm) of gabazine were used in two cells. The frequency of S-IPSCs was decreased from 10 to 0.07 Hz, and the mean amplitude of the remaining IPSCs was 3.4 times smaller than that of control IPSCs. The threshold for the detection of the IPSCs was set at −6 pA, at which the simultaneous opening of two glycine receptors would be detected. It can be concluded that >99% of the S-IPSCs are mediated by GABAAreceptors only.
When serotonin was applied in the continuous presence of 30 μm gabazine, it evoked IPSCs in 11 of 14 Golgi cells (Fig. 1 A). These 5-HT-evoked IPSCs were blocked by 300 nm strychnine (n = 3), confirming that they are glycinergic IPSCs (Fig. 1 B). Their mean amplitude was 34 ± 20 pA (mean ± SD;n = 10). Reciprocally, serotonin evoked GABAergic IPSCs in 9 of 13 Golgi cells recorded in the presence of 1 μm strychnine (Fig. 1 C). These L-IPSCs had a mean amplitude of 90 ± 80 pA (mean ± SD;n = 9). Therefore Golgi cells receive purely GABAergic spontaneous IPSCs and glycinergic (L-IPSCsGly) and GABAergic (L-IPSCsGABA) IPSCs evoked by serotonin.
Basket cell→Golgi cell synapses are GABAergic
Synapses between basket cells and Golgi cells have been unequivocally identified at the electron microscopic level (Palay and Chan-Palay, 1974). In a previous paper (Dieudonné and Dumoulin, 2000), we proposed that S-IPSCs recorded in Golgi cells reflect the release of transmitter at these synapses, either miniature or evoked by the spontaneous firing of basket cells in the slice (Vincent et al., 1992; Kondo and Marty, 1998). Extracellular electrical stimulations have been used to confirm this point.
At the age of the animals used in this study, the pinceau-like structures around the cell body and the initial axonal segment of the Purkinje cells are not yet formed, but basket cell axons have already converged onto Purkinje cell bodies (Altman and Bayer, 1997). A relatively selective stimulation of basket cell axons can thus be obtained by placing the stimulating electrode at the apex of the Purkinje cell bodies. Low-intensity stimulations at this location could in all instances evoke IPSC in Golgi cells (Fig.2 A), even when the stimulation electrode was moved to Purkinje cell bodies located 200–300 μm away from the recorded Golgi cell. Increasing the stimulation intensity induced the recruitment of multiple fibers, indicating the presence of several presynaptic axons in the vicinity of the stimulating electrode (Fig. 2 B). Attempts to evoke IPSCs from random points of the low molecular layer and in regions of the Purkinje cell layer where Purkinje cell bodies had not been preserved were not always successful, even when using high-intensity stimulations. The electrically evoked IPSCs, as S-IPSCs, were completely blocked by gabazine (Fig. 2 C). We conclude that basket cells form functional inhibitory contacts with Golgi cells and that the corresponding IPSCs are entirely mediated by GABAA receptors.
Unitary L-IPSCs involve the activation of both GABAAand glycine receptors
The mixed sensitivity of the L-IPSCs to gabazine and strychnine contrasts with the pure GABAergic nature of the S-IPSCs and suggests that GABA and glycine can both be released by Lugaro cells. It has been shown that a small number of Lugaro cells presynaptic to a given Golgi cell are preserved in slices (Dieudonné and Dumoulin, 2000). In many cases a Golgi cell receives the input from zero or one Lugaro cell (Dieudonné and Dumoulin, 2000). Nevertheless, similar proportions of Golgi cells receive L-IPSCsGly in the presence of gabazine (79% of 14 cells tested), L-IPSCsGABA in the presence of strychnine (70%; n = 13), or L-IPSCs when serotonin is applied in the absence of inhibitors (73%; n = 26). This suggests that even when a single presynaptic Lugaro cell is preserved, this cell is able to release both GABA and glycine.
In three cells in which gabazine was added during the application of 5-HT, a glycinergic component of the IPSCs remained. These glycinergic IPSCs disappeared after washout of serotonin from the bath, confirming that they represent L-IPSCs. The converse experiment, that is, applying strychnine during the effect of serotonin, is more difficult to analyze because S-IPSCs are not blocked by strychnine and are mixed with L-IPSCs, the amplitude of which has been decreased by strychnine. In two of these experiments, in which S-IPSCs were sufficiently infrequent to be neglected, strychnine blocked by 28 and 32% the amplitude of the L-IPSCs without any significant effect on their frequency (98 and 101%). These experiments indicate that most L-IPSCs involve the activation of both GABAA and glycine receptors and that cotransmission is thus the rule at Lugaro cell→Golgi cell synapses.
To confirm this point, we analyzed in detail an experiment (Fig.3) in which a single presynaptic Lugaro cell was activated by serotonin. In this Golgi cell, the amplitude of L-IPSCs was so much larger than the amplitude of S-IPSCs that they could easily and reliably be discriminated (Fig. 3 A,C). Whereas >98% of the S-IPSCs were <60 pA (Fig. 3 B,top, arrows), the amplitude of L-IPSCs could be as large as 200 pA. As expected, S-IPSCs during the control period and small (<60 pA) IPSCs during serotonin application occurred both at random intervals and at similar frequencies (Fig. 3 D). In contrast, the distribution of the intervals between L-IPSCs (IPSCSs > 60 pA in the presence of serotonin) was well described by two Gaussian peaks with a preferred frequency of 5.6 Hz (Fig.3 E, left). This implies that a single presynaptic Lugaro cell is responsible for all of the serotonin-evoked L-IPSCs [see Dieudonné and Dumoulin (2000) for the validity of this criterion]. Although S-IPSCs are not mistaken for L-IPSCs with this threshold method, the contrary is not true because 21% of the large IPSCs were separated by two times the elementary interval (second Gaussian peak), indicating that 17% of the L-IPSCs were <60 pA and have been pooled with S-IPSCs.
In this cell, application of gabazine, after the effect of serotonin had reached its steady state, reduced the frequency of occurrence (Fig.3 A, bottom) of the IPSCs. As expected, spontaneous and serotonin-evoked activities were differentially affected. Gabazine completely blocked the small and random S-IPSCs, accounting for the reduced frequency of IPSCs, but it decreased the frequency of the rhythmic L-IPSCs only from 5.8 to 5 Hz (Fig.3 B). The intervals between the remaining (glycinergic) L-IPSCs showed a multi-Gaussian distribution with equidistant peaks separated by 135 msec, corresponding to a preferred frequency of 7.4 Hz (Fig. 3 E, right). This increase in the rhythmic frequency is caused by a disinhibition of the presynaptic Lugaro cell by gabazine (S. Dieudonné, unpublished results). Inter-IPSC intervals of two and three times the 135 msec period result from the fact that the glycinergic component of some of the IPSCs is so small that it falls under the detection threshold. The number of events in the second and third peaks of the histogram indicates that only 32% of the rhythmic events fell under the detection threshold (−10 pA; which is four glycine channel openings) and that for most of them the sum of two or three glycine channel openings seems to occur (Fig.3 B, asterisks). These results indicate that for the majority of the L-IPSCs both GABA and glycine are coreleased by the presynaptic Lugaro neuron.
Whether activation of both GABAAR and GlyR occurs at the same synaptic contacts cannot be inferred from these experiments. Because miniature glycinergic IPSCs are not recorded in this preparation, we could not use them to study whether a single vesicle can release GABA and glycine, as was done in the spinal cord (Jonas et al., 1998) and brainstem (O'Brien and Berger, 1999). We therefore performed immunohistochemical stainings to determine whether markers for GABA and glycine transmission are colocalized at all Lugaro cell→Golgi cell contacts or whether they are segregated at separate subsets of contacts.
Lugaro cell axon terminals express both GABAergic and glycinergic markers
Lugaro cells have typical ovoid somata lying in the upper granular layer, with long dendrites running in the parasagittal plane below and between the Purkinje cell bodies (Lainé and Axelrad, 1996;Dieudonné and Dumoulin, 2000). Two axonal plexi have been described, a parasagittal one that contacts basket cells, stellate cells, and possibly Golgi cells (Lainé and Axelrad, 1996, 1998) and a transverse one, oriented in the same direction as the parallel fibers, that establishes synaptic contacts with Golgi cells and possibly with basket and stellate cells (Dieudonné and Dumoulin, 2000). In a previous study, we demonstrated that this transverse plexus of Lugaro axons is composed of thick, calretinin-immunoreactive (-IR) varicose fibers running in the low molecular layer (Dieudonné and Dumoulin, 2000). Calretinin detection was therefore performed in a set of double-immunohistochemistry experiments performed on sections of the vermis cut transversally to the parasagittal plane, parallel to the axis of the folia. The majority of the varicosities (90.4%;n = 188) of the calretinin-positive axons were found to be immunoreactive for GAD65 (Fig.4 A), as suggested by previous observations indicating a GABAergic phenotype of Lugaro cells (Aoki et al., 1986; Gabbott et al., 1986; Dieudonné and Dumoulin, 2000), and 90.7% (n = 152) of them were also positive for the neuronal membrane glycine transporter GlyT2 (Fig.4 C). The varicosities identified by their GlyT2-IR were also immunoreactive for GAD65 (90.4%; n= 168) (Fig. 4 D).
The Lugaro cell parasagittal axonal plexus could not be unambiguously identified using anti-calretinin antibodies, because of the small diameter of its fibers and of a background staining of the parallel fibers. This background is very faint in the low molecular layer, where Lugaro transverse axons run, but becomes more important in the rest of the molecular layer, where the parasagittal Lugaro plexus extends (Fig.4 A). However, the parasagittal axons could be identified by their GlyT2 immunoreactivity as a sparse plexus of thin parasagittal fibers that can be noticed in both parasagittal and transverse sections (Fig. 4 E). The varicosities of the latter plexus were also GAD65-IR (87.3%; n = 182) (Fig. 4 E). Moreover, all GAD65-labeled terminals in transverse or sagittal planes were immunoreactive for VIAAT (data not shown), the mixed GABA and glycine vesicular transporter present in GABAergic, glycinergic, and mixed GABA- and glycine-containing terminals (Chaudhry et al., 1998; Dumoulin et al., 1999). Thus, the Lugaro cell is a mixed inhibitory neuron that possesses adequate machinery for GABA and glycine accumulation in, and release from, axonal varicosities.
Cerebellar Golgi cells express glycine receptors
Because Golgi cells receive a glycinergic input, we have investigated the distribution of GlyR in the cerebellar cortex in sections of the vermis cut in the sagittal plane, which allows the best visualization of the Golgi cell dendritic arborization. Immunoreactivity for GlyRα/β was found in all cerebellar lobules, in both granular and molecular layers (Fig.5 A). In the granular layer, GlyRα/β immunoreactivity was detected at some glomerular structures, most presumably corresponding to dendrites of unipolar brush cells (UBCs) (Mugnaini et al., 1994), and on neurons of the granular layer, the neuritic processes of which extended into the molecular layer. These latter neurons had the characteristic morphology and distribution of Golgi cells (Fig. 5 B). A diffuse cytoplasmic staining was observed within their cell body (Fig.5 C), and both basolateral and apical dendrites displayed a “punctated” GlyRα/β labeling (Fig. 5 B,D). In the molecular layer, Golgi cell dendrites were the only GlyRα/β-IR profiles. They could often be followed from the cell body to their end at the pial surface (Fig. 5 B). In this layer, simultaneous detection of GlyRα/β and VIAAT, used as a specific marker of inhibitory terminals, showed a close apposition between the two immunoreactivities (Fig. 5 E). This indicated that GlyR-IR clusters have a postsynaptic location on Golgi cell dendrites; Golgi cells are therefore the only ubiquitous cell type in the cerebellar cortex to receive an inhibitory glycinergic input (UBCs, associated with vestibular afferents, are concentrated in specific lobules).
Simultaneous detection of glycine and GABAA receptors on Golgi cell dendrites
Because Lugaro cells release both GABA and glycine, we studied whether GABAARs colocalized with GlyRs on the dendrites of Golgi cells. An antibody against the GABAARγ2 subunit was chosen for this experiment because this subunit is involved in the postsynaptic localization of major GABAAR subtypes (Essrich et al., 1998). In agreement with previous reports in the literature (Fritschy et al., 1992; Gutiérrez et al., 1994; Sassoe-Pognetto et al., 2000), GABAARγ2 immunoreactivity was found in the three layers of the cerebellar cortex (data not shown). In parasagittal sections of the vermis, in which Golgi cell dendrites can be followed throughout their course in the molecular layer, simultaneous detection of GlyRα/β and GABAARγ2 subunits showed that the two immunoreactivities were associated on Golgi cell dendrites (Fig. 6 A). However, the two receptors displayed a variegated pattern of immunoreactivities rather than an exact overlapping (Fig. 6 A,inset). GlyR-IR clusters without apparent association with GABAAR immunoreactivity were exceptionally found on Golgi cell dendrites (data not shown). This may result from reduced GABAARγ2 epitope detection by this antibody when an additional methanol fixation of the sections was used to detect GlyR immunoreactivity, as compared with paraformaldehyde fixation alone.
Gephyrin is a cytoplasmic protein associated with, but not exclusively located at, glycinergic synapses in the CNS (Sassoe-Pognetto et al., 1995, 1999, 2000; Craig et al., 1996; Giustetto et al., 1998;Lévi et al., 1999). In contrast with what is seen for GlyR, immunodetection of gephyrin showed a widespread labeling of the molecular layer that was not selective for Golgi cell dendrites. Dendrites outlined by gephyrin-IR clusters could be followed from the Golgi cell body to the molecular layer but were often lost there because of the staining of multiple profiles in this layer (data not shown). Double detection showed association, but not colocalization, of GABAARγ2 immunoreactivity with gephyrin-positive clusters in profiles that had the course and location of Golgi cell dendrites between or just above Purkinje cell bodies (Fig. 6 B).
From these results, we conclude that cotransmission involving GABAAR and GlyR occurs at single synaptic sites on Golgi cell dendrites. The juxtaposition of the two receptors on Golgi cell dendrites in both the low and the upper molecular layer indicates that the parasagittal plexus of the Lugaro cell axons contributes to the inhibition of Golgi cells, as shown by Lainéand Axelrad (1998), in addition to the transverse plexus, which is restricted to the lower one-third of the molecular layer (seeDieudonné and Dumoulin, 2000).
Lugaro cell axon varicosities are apposed to glycine receptors on Golgi cell dendrites
To confirm that Lugaro cells are the interneurons presynaptic to the mixed GlyR–GABAAR clusters present on Golgi cell dendrites, we performed simultaneous immunodetection of calretinin or GlyT2 and GlyRα/β. GlyRα/β clusters were found associated with calretinin-positive (Fig. 6 C) and GlyT2-positive terminals (Fig. 6 D,E). Quantification of the latter shows that 69.8% (n = 200) of the GlyT2-positive terminals in the transverse plexus associate with GlyRα/β clusters, but only 53.4% (n = 177) of these terminals do in the parasagittal plexus. These results confirm a major participation of the Lugaro transverse plexus in the connection with Golgi cells and a significant participation of the parasagittal axons to it, although less important. Besides, the number of associations in the transverse plane may have been underestimated because of the difficulty of focusing on the same plane the Lugaro axon and the Golgi dendrites, which have perpendicular orientations. For both plexi, the remaining terminals, which are not apposed to GlyRα/β clusters, are thought to represent contacts with basket and stellate cells, which were demonstrated at the electron microscopic level (Lainé and Axelrad, 1998). Thus, GlyT2 would also be present in axon terminals where transmission is purely GABAergic. Conversely, it was found that most GlyRα/β-immunoreactive clusters present in Golgi cell dendrites in the molecular layer were apposed to GlyT2 immunoreactivities in both transverse (89.5%; n = 192) and parasagittal (86.8%; n = 175) sections.
Finally, to confirm the absence of a GlyR component at basket cell→Golgi cell synapses, we performed a double immunodetection of parvalbumin [a marker of molecular layer interneurons and Purkinje cells (Kosaka et al., 1993)] and GlyR. In contrast to the results obtained with Lugaro cell markers, parvalbumin-positive profiles were not associated with GlyR clusters on Golgi cell dendrites (Fig.6 F).
Taken together, the morphological and electrophysiological results establish that the mixed serotonin-induced IPSCs recorded in Golgi cells are caused by corelease of GABA and glycine by the Lugaro cell axonal varicosities onto variegated postsynaptic arrays of GlyR and GABAAR located on Golgi cell dendrites. In contrast, pure GABAergic transmission at basket cell→Golgi cell synapses accounts for the spontaneous IPSCs recorded from Golgi cells.
Kinetic properties of S-IPSCs and L-IPSCs
The convergence of two identified inputs, one purely GABAergic and one mixed GABAergic and glycinergic, on the same postsynaptic cell offers for the first time the possibility to assess the specific function of glycinergic transmission in central structures dominated by GABAergic inhibitory transmission. For this purpose the decay kinetics of S-IPSCs and L-IPSCs was studied in 26 Golgi cells. IPSCs that were well separated from adjacent synaptic events were selected and averaged during the control period (S-IPSCs) and during serotonin application (L-IPSCs). L-IPSCs were distinguished from contaminating S-IPSCs by their larger amplitude (as in Fig. 3), and only L-IPSCs larger than the largest S-IPSC recorded before addition of serotonin were selected. The time course of the mean IPSCs was quantified by measuring either their half-width or the ratio of their total charge over their peak amplitude. Both values showed that the decay kinetics of both types of IPSCs was very variable from cell to cell (Fig.7 A,B). The half-width varied from 9.6 to 29.0 msec for S-IPSCs (18 ± 5 msec; mean ± SD;n = 26) and from 14 to 41 msec for the L-IPSCs (25 ± 8 msec; mean ± SD; n = 13). The charge-over-peak ratio varied from 19 to 50 msec for S-IPSCs (36 ± 10 msec) and from 22 to 65 msec for the L-IPSCs (42 ± 14 msec). In granule cells of the cerebellar cortex, maturation of GABAergic synapses during the third postnatal week has been associated with a decrease of the decay time constants of the GABAergic IPSCs (Tia et al., 1996; Carlson et al., 1998). However, maturation is not the cause of the large variability of IPSCs kinetics in Golgi cells. Over all of the experiments, there was no significant correlation between the age of the animal and the kinetics of the IPSCs (Fig.7 C). Furthermore, the S-IPSCs recorded from Golgi cells at postnatal day 13 have half decay times ranging from 12 to 27 msec, a dispersion as high as that of the overall population.
The half decay times of S-IPSCs and L-IPSCs, although highly variable from cell to cell, were in fixed quantitative relation to each other when recorded in the same cell (Fig. 7 D). In all of the cells, the L-IPSCs decayed more slowly than did the S-IPSCs, and the ratio of their time course was independent of the time constant of the IPSCs. This is illustrated in Figure 7 E, where the half decay time of the L-IPSCs is plotted against the half decay time of the S-IPSCs in the same cells. The half decay times are linearly correlated (p < 0.0001) (Fig. 7 E) with a slope of 1.28. For the charge-over-peak ratios, the regression slope was 1.34 (p < 0.0001). Therefore the decay time course of the IPSCs appears to be a highly variable parameter coregulated by the postsynaptic Golgi cells at the various inhibitory synapses converging onto them.
The kinetics of the GABAA and of the glycinergic components of inhibitory synapses is coregulated
It appeared likely that the consistent kinetic difference between the S-IPSCs and L-IPSCs was caused by the presence of a glycinergic component in the later synaptic events. We hence compared the kinetics of the GABAergic and glycinergic components of the IPSCs in individual Golgi cells. The successive pharmacological isolation of the two components in the same cell is difficult and unreliable because of the slow and partial reversibility of GABAA and glycine inhibitors in slices. Therefore S-IPSCs (purely GABAergic) were always recorded in control conditions, to serve as a kinetic standard, before recording L-IPSCs (induced by serotonin) in the presence of either gabazine (30 μm) or strychnine (1 μm). We found that L-IPSCsGly had a significantly slower decay than did S-IPSCs (Fig.8 C) with mean ratios of 1.46 ± 0.14 (mean ± SEM; n = 7) and 1.35 ± 0.13 (mean ± SEM; n = 7) for half-width and charge-over-peak measurements, respectively.
A more detailed analysis was performed. The decay of GABAergic S-IPSCs and of the GABAergic component of L-IPSCs (L-IPSCsGABA) was fitted by the sum of two exponential functions (Fig. 8 A). The fast time constants were 14 ± 3 msec for the S-IPSCs and 16 ± 5 msec for the L-IPSCsGABA, and the slow time constants were 53 ± 15 and 46 ± 16 msec, respectively (n = 6). Fast and slow time constants measured in the same cell for both types of IPSCs were linearly correlated with slopes of 1.16 and 0.86, respectively (Fig. 8 B). The ratio of the amplitudes of the fast and slow components were 1.2 ± 0.2 and 1 ± 0.2 for S-IPSCs and L-IPSCs, respectively (p > 0.3, t test). In contrast, the decay of the L-IPSCsGly was appropriately fitted by a single-exponential function in all of the cells studied (n = 7). Its time constant was intermediate between the fast and the slow time constants of decay of the S-IPSCs recorded in the same cells, with regression slopes of 2.3 and 0.68, respectively (Fig. 8 D). We conclude that the decay of the L-IPSCsGly differs from the decay of the L-IPSCsGABA and of S-IPSCs, maintaining the L-IPSCs and S-IPSCs decay time course in a fixed ratio. We propose that the slower decay kinetics of L-IPSCsGlyconstitutes a significant functional difference between the mixed Lugaro cell→Golgi cell synapses and the pure GABAergic synapses made by basket cells onto the same Golgi cells.
The mixed pharmacology of individual L-IPSCs recorded from Golgi cells indicates that Lugaro cells release both GABA and glycine. The experiment described in Figure 3 demonstrates that a single presynaptic Lugaro cell can release both GABA and glycine at each action potential. Corelease is most likely a general rule at Lugaro cell→Golgi cell synapses because the mean amplitude of L-IPSCs [130 pA (Dieudonné and Dumoulin, 2000)] is the sum of the amplitude of L-IPSCsGABA (90 pA) and of L-IPSCsGly (34 pA). These electrophysiological data are supported by the immunohistochemical demonstration of the presence of mixed arrays of GABAAR and GlyR at postsynaptic loci on Golgi cell dendrites, in front of Lugaro cell axonal varicosities. Colocalized immunoreactivities for both GAD65 and GlyT2 in the varicosities of the Lugaro cells suggest that GABA and glycine are indeed coreleased at single synaptic contacts. GlyT2 can accumulate much higher concentrations of glycine in the cytoplasm of neurons than can those achieved by the glial homolog GlyT1 (Roux and Supplisson, 2000). These high glycine concentrations are probably necessary to compete with GABA for transport by the nonselective vesicular transporter VIAAT (Sagné et al., 1997) and to achieve a substantial loading of the synaptic vesicles with glycine. The data presented in this paper offer the first evidence of the occurrence of functional cotransmission by GABA and glycine outside the spinal cord (Jonas et al., 1998) and brainstem motor nuclei (O'Brien and Berger, 1999), suggesting that inhibitory cotransmission may be more widely used in the CNS than suspected previously.
Receptor expression selectivity
Immunocytochemistry experiments have shown that GABAAR and GlyR subunits can segregate at different synapses on the same postsynaptic cell (Sassoe-Pognetto et al., 1995; Koulen et al., 1996). We show here that GlyR and GABAARγ2 are present at Lugaro cell→Golgi cell synapses but not at basket cell→Golgi cell contacts and therefore that receptors are segregated according to the identity of the presynaptic element. The presynaptic terminal must thus be instructive for the formation of the adequate postsynaptic site. This idea has been illustrated recently in the hippocampal pyramidal cells where α1 and α2 GABAAR subunits segregate to synapses where two different presynaptic interneurons are involved (Nyiri et al., 2001). The cellular processes involved in this mechanism are still under investigation; they probably involve signals mediated by cadherins and/or catenins (Fannon and Colman, 1996; Benson and Tanaka, 1998; Kohmura et al., 1998) or agrin (Böse et al., 2000). In addition, the present results show that >90% of the Lugaro varicosities displayed both GlyT2 and GAD65 immunoreactivities whereas only 73% of them were apposed to GlyR clusters. Thus, segregation of phenotypes does not seem to occur in this example of inhibitory terminals. This also further indicates that the postsynaptic element may select the type of inhibition it receives by avoiding expression of, or being unable to express, the receptors for one of the transmitters coreleased. Indeed, synapses made by the Lugaro cell axons onto basket and stellate cells lacked postsynaptic GlyR, because the latter is only expressed on Golgi cell dendrites in the molecular layer. The nature of both presynaptic and postsynaptic elements is therefore important in determining the molecular composition of a given inhibitory synapse.
Organization of mixed receptor arrays
Gephyrin has been first considered to be a marker of glycinergic synapses. It is now established that gephyrin is also associated with postsynaptic GABAAR in the absence of GlyR bothin vitro and in vivo (Sassoe-Pognetto et al., 1995, 1999, 2000; Craig et al., 1996; Giustetto et al., 1998). In the majority of synapses, the presence of GABAARγ2 subunits seems to be a critical factor for the association of GABAAR with gephyrin. Loss of gephyrin prevents synaptic clustering of γ2-containing GABAAR (Kneussel et al., 1999), and loss of GABAARγ2 subunits highly decreases the synaptic localization of GABAAR and gephyrin in several brain regions (Essrich et al., 1998), a perturbation that can be rescued by overexpression of the γ3 subunit (Baer et al., 1999). Our data indicate that GABAAR, gephyrin, and GlyR can accumulate at the same postsynaptic sites. Similar results have been found in spinal cord cultures, in which GABAAR β3 or γ2 subunits, gephyrin, and GlyR colocalize at the same synapses (Lévi et al., 1999; Dumoulin et al., 2000). At Lugaro cell→Golgi cell synapses, gephyrin and GlyR appear to be preferentially located on the edge of GABAARγ2 postsynaptic domains. However, gephyrin and GABAAR are found to be exactly colocalized at the majority of inhibitory synapses in the molecular layer (Sassoe-Pognetto et al., 2000). This indicates that the organization of the postsynaptic domains of receptors at the Lugaro cell→Golgi cell synapses is different from the organization at the majority of inhibitory synapses in the molecular layer, which are GABAergic synapses made by Purkinje cell recurrent collateral axons and basket and stellate cells.
Segregation of gephyrin or GlyR and GABAAR immunoreactivities inside the same postsynaptic domain has been described at other synapses. At Golgi cell→granule cell contacts in the granular layer, gephyrin clusters are embedded in GABAAR domains (Sassoe-Pognetto et al., 2000). Partial or complete segregation of GABAARβ3 subunit and gephyrin labelings has also been observed at the ultrastructural level in putative mixed synapses in the spinal cord (Todd et al., 1996), and in the rat retina, gephyrin and GABAARα2 subunit were found in the same synapse but did not colocalize (Sassoe-Pognetto et al., 1995). Segregation of GABAAR and gephyrin at some synapses suggests that γ-containing GABAARs are not recruited by a direct interaction with gephyrin. This is coherent with the lack of evidence of direct protein–protein interaction between gephyrin and GABAAR (Meyer et al., 1995).
Altough our data strongly suggest that GABAAR and GlyR clusters are segregated at mixed synapses onto Golgi cells, further ultrastructural investigation would be necessary to determine the spatial organization of mixed synapses, in particular to demonstrate formally that the two receptor clusters are facing the same presynaptic varicosity. The detailed organization of mixed synapses may have important functional consequences because the two types of receptor clusters may be located at different distances from the release sites. Consequently, they might be differentially saturated after synaptic release of the two transmitters, as occurs in GABA and glycine cotransmission in the dorsal horn layer I of the spinal cord (Chéry and de Koninck, 1999). In a more extreme model, GABAAR and GlyR clusters could face different active zones within the same presynaptic bouton, as suggested in the deep cerebellar nuclei (Chen and Hillman, 1993). In both cases the presynaptic neuron could modify rapidly the relative activation of the two types of postsynaptic receptors by changing its release probability. On a slower time scale, the presynaptic neuron could regulate the relative occupancy of both types of postsynaptic receptors by changing the concentration of GABA and glycine inside the synaptic vesicles.
Postsynaptic relevance of inhibitory cotransmission at Lugaro cell→Golgi cell synapses
What is the functional consequence of the variability of the decay time constants of the IPSCs at inhibitory synapses onto Golgi cells? Many data from the literature suggest that there is a correlation between the decay time constant of IPSCs and the period of oscillations in rhythmic neuronal networks (Traub et al., 1998). The difference of IPSC kinetics between Golgi cells may indicate that different parts of the cerebellum use different working frequencies. This would fit with the organization of the cerebellar vermis into lobules or groups of lobules receiving different afferences (Altman and Bayer, 1997) and with the recording of field oscillations at different frequencies in different cerebellar lobules (Pellerin and Lamarre, 1997; Timofeev and Steriade, 1997; Hartmann and Bower, 1998). Alternatively all parts of the cortex may have a mixed population of Golgi cells, each cell being optimally efficient when the network is oscillating at a frequency adapted to the kinetics of its IPSCs.
In motoneurons, GABAergic IPSCs (IPSCsGABA) decay more slowly than do glycinergic IPSCs (IPSCsGly) and can be further slowed by acting on allosteric modulatory sites (Jonas et al., 1998). This kinetic difference constitutes the key factor for the demonstration that both receptors are activated during miniature IPSCs. In the cerebellum, IPSCsGABAhave a biexponential decay, and IPSCsGly have a single-exponential decay of time constant intermediate between the two time constants of IPSCsGABA. IPSCsGly are 1.4 times slower than are IPSCsGABA (half decay time), approximately the inverse of the ratio in the spinal cord (where IPSCsGly are faster). This number is relatively modest compared with the dispersion of the decay time course between Golgi cells, but it may represent a significant functional difference. The modulation of the decay time course of GABAergic IPSCs by benzodiazepine, although not more profound, is known to have important effects on the function of neuronal networks. In this hypothesis, the main consequence of glycinergic transmission onto Golgi cells would be to give a slower decay time course to Lugaro cell IPSCs, compared with basket cell IPSCs. This would enhance the duration of the inhibition of Golgi cells by Lugaro cells, and therefore the disinhibition of granule cells, and would regulate the number of active granule cells. Because of the presence of putative glycinergic or mixed GABA and glycine neurons in many parts of the CNS, glycinergic transmission may constitute a target for new, more specific, drugs similar in their action to benzodiazepines.
A.D. was supported by BIOMEDII BMH4CT 972374 and the Institut National de la Santé et de la Recherche Médicale (U497). S.D. was supported by a fellowship from the Association Française contre les Myopathies, BIOMEDII BMH4CT 972374, the Ecole Normale Supérieure, and the Centre National de la Recherche Scientifique (Unité Mixte de Recherche 8544). We are grateful to Dr. P. Ascher for his support and critical reading of this manuscript. We also acknowledge Drs. H. Betz (Frankfurt) and W. Sieghart (Vienna) for providing the anti-GlyRα/β and anti-GABAARγ2 antibodies, respectively.
Correspondence should be addressed to Dr. Stéphane Dieudonné, Laboratoire de Neurobiologie, Unité Mixte de Recherche 8544, Ecole Normale Supérieure, 46, rue d'Ulm, 75005 Paris, France. E-mail:.