Presynaptic Calcium Signalling in Cerebellar Mossy Fibres

Whole-cell recordings were obtained from mossy fibre terminals in adult turtles in order to characterize the basic membrane properties. Calcium imaging of presynaptic calcium signals was carried out in order to analyse calcium dynamics and presynaptic GABA B inhibition. A tetrodotoxin (TTX)-sensitive fast Na+ spike faithfully followed repetitive depolarizing pulses with little change in spike duration or amplitude, while a strong outward rectification dominated responses to long-lasting depolarizations. High-threshold calcium spikes were uncovered following addition of potassium channel blockers. Calcium imaging using Calcium-Green dextran revealed a stimulus-evoked all-or-none TTX-sensitive calcium signal in simple and complex rosettes. All compartments of a complex rosette were activated during electrical activation of the mossy fibre, while individual simple and complex rosettes along an axon appeared to be isolated from one another in terms of calcium signalling. CGP55845 application showed that GABA B receptors mediated presynaptic inhibition of the calcium signal over the entire firing frequency range of mossy fibres. A paired-pulse depression of the calcium signal lasting more than 1 s affected burst firing in mossy fibres; this paired-pulse depression was reduced by GABA B antagonists. While our results indicated that a presynaptic rosette electrophysiologically functioned as a unit, topical GABA application showed that calcium signals in the branches of complex rosettes could be modulated locally, suggesting that cerebellar glomeruli may be dynamically sub-compartmentalized due to ongoing inhibition mediated by Golgi cells. This could provide a fine-grained control of mossy fibre-granule cell information transfer and synaptic plasticity within a mossy fibre rosette.


INTRODUCTION
Mossy fi bres constitute one of the important input pathways to the cerebellum (Eccles et al., 1966). The mossy fi bre rosette is the presynaptic element in the cerebellar glomerulus (Cajal, 1888), a unit of information transfer, which additionally is formed by granule cell dendrites and GABAergic Golgi cell axonal terminals. In order to understand input processing to the cerebellum, the physiological properties of the glomerulus are of importance, and the anatomical organization suggests that each glomerulus is a semi-autonomous unit, isolated in chemical terms from the surrounding cellular elements by a glial sheath (Eccles et al., 1967;Palay and Chan-Palay, 1974;Jakab and Hamori, 1988). The presynaptic rosettes are classifi ed as "simple" or "complex" depending on whether they are spherical or have a complex, branched structure (Cajal, 1888(Cajal, , 1926Craigie, 1926;Mugnaini et al., 1974;Palay and Chan-Palay, 1974). Modulation of mossy fi bre Ca 2+ infl ux can occur due to presynaptic GABA B receptors controlling calcium infl ux during low-frequency mossy fi bre activation; this effect was only noticeable following prolonged Golgi cell activation in slices (Mitchell and Silver, 2000). However, taking advantage of the more intact network available in the turtle cerebellum preparation, we wished to further analyse the timecourse of presynaptic GABA B modulation.
Presynaptic modulation could be considered glomeruli-specifi c processes, controlled by the local level of synaptic activity in each glomerulus. However, the large extent and branched structure of

PREPARATION
Adult (carapace length 20-25 cm) turtles (Trachemys scripta elegans) were deeply anesthetized with pentobarbitone (100 mg/kg, im) and decapitated. The surgical procedures comply with Danish legislation and are approved by the controlling body under the Danish Ministry of Justice. The brain was dissected out and transferred to oxygenated Ringers fl uid. For electrophysiological experiments, the cerebellum was then isolated. "Blind" patch recordings (cf. Blanton et al., 1989) were obtained either in the intact cerebellum or in 800 µm parasagittal slices cut on a Vibratome. For visually guided patch recordings from mossy fi bres, 250 µm thick parasagittal slices were used. Dissections and experiments were done at room temperature (22-25°C) which is within the physiological temperature range for turtles (∼15-35°C; Gatten, 1974). For calcium imaging experiments, the cerebellum and the brainstem were isolated and loading of calcium indicator was performed as described below.

ELECTROPHYSIOLOGY
Recording electrodes (15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25) were fi lled with a solution of (in mM): 127 KCH 3 SO 4 or K-gluconate, 1.53 Mg-gluconate, 3.7 MgCl 2 , 5 HEPES, 5 Na + -HEPES and 2 Na 2 -ATP, pH adjusted to 7.5 with KOH (all chemicals from Sigma). The data presented here were not adjusted for the liquid junction potential between recording electrode and bath. An Ag/AgCl electrode (Clark Electromedical, UK) was used as reference electrode in the bath. Tetrodotoxin (Alomone Labs), picrotoxin (Sigma) and CGP55845 (Tocris, UK) were added to the Ringer solution as indicated in the text. 4-Aminopyridine (4-AP; Sigma) and CoCl 2 (Sigma) were added to Ringer solution for at least 20 min each, before the records shown were taken. In other brain regions, GABA A receptors modulate presynaptic release (Engelman and MacDermott, 2004;Alle and Geiger, 2007) and Golgi cells use GABA as a transmitter (Ito, 1984;Dugué et al., 2005;Simat et al., 2007). In order to avoid interference from ionotropic GABA receptors, of particular concern during local GABA application (see below), our experiments were conducted in the presence of picrotoxin, a non-competitive antagonist of primarily GABA A-Rs (Yang et al., 2007), as indicated in the text (cf. Mitchell and Silver, 2000). The tissue was weighed down in the recording chamber by small silver wires. In some cases the thicker slices were glued to a Sylgard (Dow Corning, NJ, USA) support using cyanoacrylate glue. Patch electrodes were placed by micromanipulators (Luigs and Neumann, Germany) in the granule cell layer under visual guidance, using a Zeiss Axioscope FS equipped with 2× or 4× dry objectives for blind recordings and a 63×/0.9 water-immersion objective for visually-guided patching of individual axons using Dodt Infracontrast (Luigs and Neumann). Data were recorded in current clamp mode using an Axoclamp 2A (DC-10 kHz; Axon Instruments, CA, USA) and subsequently digitized using the Digidata A/D board and pClamp (both Axon Instruments) at digitization rates at or above 10 kHz.
Electrical stimulation of mossy fi bres was achieved either by using a pair of twisted, Tefl on-coated platinum-iridium wires (100 µm diameter) using pulses of 50 µs duration (up to 300 µA), and placed in the granule cell layer in the same parasagittal plane as the mossy fi bre recorded from, or by two Ag/AgCl electrodes (1 ms pulses, up to 10 mA) placed at opposite ends of the cerebellum along the sagittal axis. Both types of stimulus electrodes were connected to a constant current source (Isolator 11; Axon Instruments, CA, USA).
For anatomical identifi cation of axons recorded from electrophysiologically, biocytin (5 mM) or Lucifer Yellow CH (5 mM) was included in the pipette solution. For biocytin staining, the tissue was fi xed overnight in 4% paraformaldehyde with 0.2% glutaraldehyde, subsequently incubated with avidin-biotin complex (DAKO, Denmark) and Triton X (Sigma), and reacted with diaminobenzidine (Sigma) (Horikawa and Armstrong, 1988). The tissue was then dehydrated and mounted (Permount, Sigma) on standard microscope slides with coverslips. In tissue processed this way, signifi cant shrinkage occur, as reported previously (Grace and Llinás, 1985); in the turtle cerebellum, dimensions are reduced by about 50% (not shown). No attempt to correct for shrinkage was made in the fi gures presented here. For illustration, Neurolucida (Microbrightfi eld, MA, USA) was used to trace fi bres in order to provide an overview of the axon trajectory in the granule cell layer. For Lucifer Yellow CH staining, the axons were in some cases (visually guided patch recordings) imaged alive, whereas for blind patch recordings the tissue was fi xed in 4% paraformaldehyde overnight. In some cases the tissue was recut to thinner slices using a Vibratome before dehydration; in other cases the axon was photographed directly in the intact cerebellum following dehydration and clearing using standard histological procedures.
For acceptance into this study, the axons had to display a stable resting potential when no bias current was passed and had to be able to generate Na + -spikes repetitively when stimulated with trains of short depolarizing pulses at 10-100 Hz. A total of 17 axons were accepted; two of these recorded by visually guided patching in thin slices. The remaining experiments were done by blind patch recordings in the granule cell layer, since visualisation of mossy fi bre rosettes was diffi cult in the granule cell layer, due to presence of heavily myelinated axons in this layer, which in turtles also functions as the white matter found in the mammalian cerebellar cortex . All axons were silent unless intracellular or extracellular stimulation was applied, and then only one action potential was evoked per depolarizing stimulus (see Results). This was in contrast to Golgi and granule cells in the granule cell layer which displayed spontaneous and evoked synaptic potentials and robust repetitive Na spike fi ring, spontaneously and/or in response to depolarization (Midtgaard, 1992;Gabbiani et al., 1994). Seven mossy fi bres out of the 17 recordings included here were recovered for anatomical identifi cation; all recordings conformed to February 2010 | Volume 4 | Article 1 | 3 Thomsen et al. Cerebellar mossy fi bres the electrophysiological criteria described above. Membrane time constants were measured as the time taken to reach 63% of the maximum voltage defl ection during a small-amplitude hyperpolarizing current pulse, and spike amplitudes were measured from the resting membrane potential.

CALCIUM IMAGING
For anterograde labelling of mossy fi bres, the calcium indicator Calcium Green-1 dextran (molecular weight 5 kD; Molecular Probes, OR, USA) was dissolved in physiological saline at a concentration of 40 mM. Two microlitres of this solution was pressureinjected into the brainstem and the cerebellar peduncles using a glass pipette with a ∼20 µm diameter tip (Beierlein et al., 2004). Both sides of the brainstem were injected, primarily corresponding to the position of the vestibular nuclei along the fourth ventricle (Künzle, 1983a) up to the peduncles. However, it was commonly observed that the dye spread to other areas in the brainstem, making it likely that fi bres other than vestibulo-cerebellar axons were labeled in this study (Künzle, 1982(Künzle, , 1983b. The brainstem-cerebellum preparation was left to incubate overnight at 5°C in a sealed container with Ringer solution. This ensured adequate time for the dye to fi ll the mossy fi bres. The following day the cerebellum was carefully separated from the brainstem and placed in the recording chamber ventral side up. Fine silver wires were placed transversely over the peduncles and near the caudal tip of the cerebellum for mechanical stabilization. For fl uorescence imaging, a Polychrome light source (TILL Photonics, Germany) and a Sensicam (640 × 480 pixels; PCO, Germany) were used together with 63×/0.9 water immersion objective (Zeiss, Germany). This resulted in each pixel covering an area of 0.17 µm on each side at full camera resolution. However, 2 × 2 pixel binning was commonly used, and for high-speed imaging (up to 100 Hz) 4 × 4 or 8 × 8 pixel binning was applied. In some cases a 0.5× lens was added in front of the camera to enlarge the fi eld of view. Axons were readily visible from the ventral surface of the cerebellum through the ependymal cell layer, and could sometimes be followed for up to ∼1 mm. As indicated in Figure 1, rosettes were often widely spaced along the axon. Mossy fi bres were identifi ed based on their predominantly parasagittal course and the occasional occurrence of complex rosettes. In contrast, Purkinje cell axons are placed deeper in the cerebellum, just below the Purkinje cell layer, and display few boutons in the granule cell layer ; furthermore, the labelling procedure is commonly expected only to label axons anterogradely (Beierlein et al., 2004). The mossy fi bres usually had a diameter of ∼1-2 µm, consistent with anatomical data . In this way we could image activity-related fl uorescence signals from presynaptic terminals of myelinated CNS axons in the intact adult cerebellum. Forty mossy fi bres were included in this study. Axons were accepted for analysis if they displayed a robust increase in fl uorescence when the fi bres were activated electrically, and if they could follow repetitive activation at 10-100 Hz. For the intensity of fl uorescence excitation used in the present experiments, bleaching was measured as the decrease in fl uorescence during a 5-s long illumination without electrical stimulation, and was 0.58% per second of illumination (n = 4, SD ± 0.43%). The data were not corrected for this bleaching. A series of pictures was collected for each stimulation, consisting of a number of reference pictures, followed by pictures accompanied by electrical stimulation. The electrical stimulation was synchronised with picture acquisition. Background fl uorescence was measured in a region away from the axon of interest, and for each picture the background was subtracted from the raw fl uorescence of the axon under study. Four to six prestimulus pictures were averaged and used as reference. dF/F (%) was calculated as 100 × (F active − F reference )/ F reference . Images were analysed using custom written routines in IDL (ITT Visual Information Solutions); additional data analysis was performed using Origin (Microcal, MA, USA), including analysis of fl uorescence decay time constants. Decay time constants were considered to be bi-exponential if the fast and the slow time constants differed by more than a factor of four. The data shown here are for the most part an average of the responses to four identical stimulations. For illustration, pseudocolor dF/F images were spatially smoothed using a 3 × 3 median fi lter. Mossy fi bre complex rosettes often extend substantially in the z-axis. It was therefore rare to fi nd a complex rosette that could be imaged in toto in one focal plane. Therefore, Z-stacks were recorded by stepping the microscope 1 µm in the focus axis by a PC program controlling the Luigs and Neumann microscope z-axis drive. For each focal plane, one or several picture series were acquired during electrical stimulation. For illustration of rosette anatomy, a maximum intensity projection of the z-stack was used as a basis. However, for clarity, additional picture montage was undertaken using data from individual focal planes as indicated in the text.

LOCAL GABA APPLICATION
In order to analyse the local processing within a cerebellar glomerulus, GABA was applied locally prior to electrical activation of the mossy fi bre. GABA (1 mM; Sigma) was dissolved in Ringers fl uid and pressure ejected from a glass pipette (tip diameter ∼5 µm) using 15-25 ms air pulses (0.5-1 psi). To control for ejection of GABA, 10 µM Alexa-568 (Molecular Probes) was added to the pipette content. Using excitation at 488 nm and a bandpass (center: 535 nm, bandwidth: 45 nm; Omega Optical) emission fi lter for calcium imaging, the pressure and puff-duration used produced a minimal increase in background fl uorescence due to Alexa; this puff-related signal was corrected using the fl uorescence signal from pressure ejections not accompanied by electrical stimulations. Care was taken not to include glomeruli where movement artefacts caused by the GABA ejection interfered with image recording and analysis. Data from four glomeruli were accepted into this paper.

STATISTICS
Data are expressed as means ± standard deviation (SD), and signifi cance levels were tested using Student's paired t-test. Unless otherwise noted, the signifi cance level used was 0.05.

ELECTROPHYSIOLOGICAL PROPERTIES OF MOSSY FIBRE ROSETTES
The basic cerebellar circuitry, including the glomerular structure, is highly conserved in different species (Cajal, 1888(Cajal, , 1926Craigie, 1926;Mugnaini et al., 1974;Palay and Chan-Palay, 1974). In the turtle cerebellum, mossy fi bres mainly course along the anterior-posterior direction in the granule cell layer following their entry through the cerebellar peduncles , but fi bres can make signifi cant course changes in the coronal plane (Figure 1).  Mossy fi bres are myelinated, but myelination is interrupted at the presynaptic rosettes Palay and Chan-Palay, 1974). Extracellular recordings have previously been made from elements in the turtle cerebellar granule cell layer, including mossy fi bres (Walsh et al., 1972(Walsh et al., , 1974. We obtained whole cell recordings from mossy fi bres by blind patch in the intact cerebellum (Blanton et al., 1989) in 15 cases and by visually guided patch in thin slices in two cases. The fi bres were identifi ed by the inclusion of biocytin or a fl uorescent dye in the recording pipette and their distinct electrophysiological properties (see Materials and Methods; Figure 1A: whole mount fi bre; Figure 1C: visually patched fi bre; Movie in Supplementary Material). The recordings from visually identifi ed rosettes revealed similar fi ring patterns as blind recordings. For a subset of randomly selected fi bres, the resting potential was −53.4 mV (SD ± 5.3; n = 7), the input resistance for hyperpolarizing current pulses at rest was 1064 MOhm (SD ± 285; n = 7) and the February 2010 | Volume 4 | Article 1 | 5 Thomsen et al. Cerebellar mossy fi bres membrane time constant was 22.3 ms (SD ± 16.3; n = 7). A strong outward rectifi cation was present just above the resting potential and subthreshold for spike initiation, while hyperpolarizing current steps revealed a much larger steady state input resistance and a pronounced inward rectifi cation, followed by a repolarizing overshoot (Figures 1D,F). The inward and outward rectifying properties are similar to the responses seen in other types of axons (Bostock, 1995;Geiger and Jonas, 2000), including cerebellar mossy fi bres in mammals (Rancz et al., 2007). In addition, at identifi ed presynaptic terminals, a local response subthreshold for spike initiation was observed ( Figure 1E). Spike-amplitude was 66.9 mV (SD ± 14.6, n = 7) and duration 1.22 ms (half-width; SD ± 0.38; n = 7). The basic electrophysiological responses of mossy fi bres have been described by Rancz et al. (2007), and extracellular recordings have been used to characterize membrane conductances in mammalian mossy fi bre terminals (Maffei et al., 2002). In order to characterize the underlying conductances using whole-cell recordings, we analysed the pharmacology of some fundamental electrophysiological properties found in mossy fi bres. TTX (10 µM) abolished the local potential and the fast spike (n = 6; Figures 2A,B).
Following addition of TTX, the strong outward rectifi cation remained, with little evidence of regenerative inward currents ( Figures 2B,C). However, addition of 4-AP (1 mM), a potassium channel blocker, reduced the strong outward rectifi cation and revealed a large spike and subthreshold responses (n = 4; Figure 2D). This spike and the subthreshold responses were abolished by addition of the calcium channel blocker Co 2+ (2 mM, n = 3; Figure 2E), indicating that they were regenerative responses due to voltage-gated calcium channels. This is consistent with previous data showing the presence of high-threshold voltage gated (N, P/Q-type) calcium channels in mammalian mossy fi bre terminals (Mitchell and Silver, 2000;Maffei et al., 2002) and other presynaptic terminals such as the calyx of Held (see Schneggenburger and Forsythe, 2006 for review). The effect of 4-AP on mossy fi bre axon and bouton excitability is consistent with effects on potassium currents seen in other axons and presynaptic terminals (cf. Southan and Robertson, 1998;Tan and Llano, 1999;Dodson et al., 2003;Ishikawa et al., 2003;Schneggenburger and Forsythe, 2006).
Extracellular stimulation evoked an all-or-none Na + spike ( Figure 3A). Using long depolarizing current pulses, only one spike was evoked at the onset of stimulation (Figures 1C and 2A). However, mossy fi bres followed high-frequency repetitive stimulation when short stimulation pulses were used (Figures 3C-E). Turtle mossy fi bres started to show spike initiation failures when the stimulus frequency exceeded ∼125 Hz. Thus, the stimulation frequencies routinely used in this study (∼10-100 Hz) cover most of the mossy fi bre fi ring frequency range in turtles.
Mammalian mossy fi bres in vivo may fi re for prolonged periods (van Kan et al., 1993), while intense burst activity may sometimes occur at very high frequencies (Jörntell and Ekerot, 2006). It was therefore of interest to characterize the spiking pattern of mossy fi bres during continuous repetitive activation and during shorter bursts. Long-lasting repetitive spiking in hippocampal mossy fi bres increases spike width and transmitter release due a cumulative inactivation of I-A type potassium currents (Geiger and Jonas, 2000). Since such currents may be 4-AP sensitive and may therefore be present in cerebellar mossy fi bres (Figure 2), it was of interest to analyse the Na + action potentials during repetitive Na + spike fi ring. Little or no change in spike amplitude and duration was observed at stimulation frequencies between 10 and 100 Hz, and number of stimulations up to 40 (Figures 3D,E). Thus, compared to hippocampal mossy fi bres (Geiger and Jonas, 2000), little activity-dependent modulation of February 2010 | Volume 4 | Article 1 | 6 Thomsen et al. Cerebellar mossy fi bres electrophysiological spike parameters seemed to occur in cerebellar mossy fi bres for stimulation frequencies covering a substantial part of the fi ring frequency range. One electrophysiological effect of repetitive spiking was notable, however. Prolonged spiking was followed by an afterhyperpolarization (AHP; Figures 3D,F), lasting up to more than 5 s depending on the frequency and number of preceding Na + spikes. The AHP was abolished by TTX together with the spiking (not shown; n = 6). Such an AHP is commonly found in axons, due to Na + /K + pump activity (Gordon et al., 1990;Morita et al., 1993;Debanne, 2004;Kim et al., 2007). In principle, long-lasting AHPs could lead to spike propagation failure (for review, see Koester and Sakmann, 2000;Debanne, 2004). However, the mossy fi bres followed stimulation with little sign of spike initiation failure (Figures 3D,E). In conclusion, adult cerebellar mossy fi bres seemed electrophysiologically capable of reliable information transfer with little activity-dependent changes in spike parameters or spiking patterns over the range of fi ring frequencies used in our subsequent imaging experiments.

Fluorescence imaging of simple and complex mossy fi bre rosettes
Next, we wished to examine the local calcium dynamics at individual presynaptic rosettes, since this may provide a basis for local modulation of information transfer from mossy fi bres to granule and Golgi cells. Therefore, we fi lled mossy fi bres with dextrancoupled Calcium Green-1 (Beierlein et al., 2004; see Materials and Methods) in order to image fl uorescence signals from individual rosettes during electrical stimulation of mossy fi bres.

Spatial distribution of calcium signals.
Complex rosettes were defi ned as structures having side branches from the main axon (Figures 1 and 4A; Mugnaini et al., 1974). Extracellular electrical stimulation of mossy fi bres elicited an all-none-increase in fl uorescence signal when using single shocks and when using trains of stimuli ( Figure 4B). The stimulus-evoked increase in fl uorescence was abolished by TTX ( Figure 4B). Together, the results in Figure 4B indicate that the fl uorescence signal was due to all-or-none Na +dependent action potentials. Spontaneous calcium transients were not observed, in line with an absence of spontaneous spiking in patch-recordings from the axons. Complex rosettes showed a parallel increase in fl uorescence from the part of the rosette along the axon as well as from the side branches and satellite boutons (Figures 4C,D). No stimulus-evoked increase in fl uorescence was apparent over the axon proper (Figures 4C,D). This was observed for single stimuli as well as for trains of stimuli, and suggests that a complex rosette functions as a unit with regard to action potential activation of the different sub-compartments, i.e. branches and satellite boutons. Simple rosettes were enlargements in-line with the axon (Figures 1A and 4E; Mugnaini et al., 1974) or slightly offset in relation to the axon (Figures 1A and 4F). In both cases, stimulus-evoked calcium transients were confi ned to the anatomically defi ned rosette, with little spread along the axon. Complex rosettes could display intricate 3-dimensional branching and/or could extend for up to about 100 µm along the axon (cf. Mugnaini et al., 1974), and such rosettes were diffi cult to image in toto. However, in other cases smaller complex rosettes could be illustrated more  Figure 4G). Several small side branches were given off (see also Movie in Supplementary Material; Mugnaini et al., 1974), some them projecting out of the focal plane. Stimulus-evoked fl uorescence signals were readily detected in the small side-branches ( Figure 4G, right panel). Terminal rosettes ( Figure 4H) were found in six cases. They often branched elaborately and were diffi cult to visualize in their full extent Palay and Chan-Palay, 1974). Other terminal rosettes were more confi ned (cf. Figure 1A; cf. Mugnaini et al., 1974). Figure 4H shows a terminal rosette consisting of a series of bouton-like structures, with occasional fi ne branches terminating in small boutons. In a few terminal rosettes, such small "satellite" boutons were detected up to 10-15 µm from a parent axon, apparently connected to it by a very fi ne fi bre, sometimes with several boutons en passage (cf. Palay and Chan-Palay, 1974). However, the fi ne threadlike fi bre was often diffi cult to image in its full extent. As for the complex rosettes en passant, all side branches and small "satellite" boutons of terminal rosettes showed stimulus-induced fl uorescence signals ( Figure 4H, right panel).
In conclusion, mossy fi bres show action potential-evoked calcium signals in simple and complex rosettes. Side branches and satellite boutons showed robust calcium signals, and all compartments appeared to be activated in parallel (cf. Figures 4C,D). All mossy fi ber regions which displayed a stimulus-evoked calcium signal corresponded to the anatomical description of complex and simple rosettes , while regions corresponding to the axon proper, i.e. the myelinated part where only spike conduction takes place, showed little or no calcium signals. The data also showed that calcium signalling in simple and complex rosettes was highly localised, as relatively sharp transitions (Figures 4E-G) between zones with fl uorescence increases and the axon were commonly observed. This indicates that individual simple and complex  rosettes along a fi bre (Figure 1) are effectively isolated in terms of calcium signalling since the average distance between rosettes is 15-20 µm Palay and Chan-Palay, 1974). In this aspect these relatively thick myelinated CNS axons appear similar to fi ne unmyelinated axons (Koester and Sakmann, 2000;Collin et al., 2005). For single shocks, the relative change in fl uorescence (dF/F) was for simple rosettes 11.2% (SD ± 9.2, n = 13). Following stimulation, only two out of nine boutons (22%) analysed showed a mono-exponential fl uorescence decay, while in other boutons a bi-exponential decay was best fi tted, with the fast time constant of 156 ms (SD ± 68, n = 5) and a slow decay time-constant was 3711 ms (SD ± 2540, n = 5). The proportion of boutons with single-versus bi-exponential fl uorescence decay was comparable to fi ndings in other brain regions (35%: Koester and Sakmann, 2000). For a short train of stimuli (fi ve shocks, 50 Hz) in simple rosettes, the fast decay time constant was 160 ms (SD ± 60, n = 5) and the slow decay time-constant was 1349 ms (SD ± 606, n = 5). For complex rosettes, the fast decay time constant was 136 ms (SD ± 101, n = 4) and the slow decay time-constant was 1414 ms (SD ± 1024, n = 4). All the data were obtained in the presence of CGP55845 (60 µM), in order to block presynaptic GABA B receptors (see below), and picrotoxin (25 µM).

Responses to repetitive mossy fi bre activation
In order to examine the time-dependence of the fl uorescence signal, fi bres were stimulated repetitively at rates covering a wide range of fi ring frequencies while the signal was measured at high frame rates (Figures 5A,B).
The fl uorescence response increased with the number of stimuli at 100 Hz, close to the maximum frequency for turtle mossy fi bres. In order to resolve the contributions to each response in a stimulus train, the fi bres were activated at 50 Hz, while being imaged at high speed (100 Hz; Figure 5B). Each mossy fi bre activation gave a noticeable increase in the fl uorescence signal (Figures 5B and  6A). To summarize these fi ndings, stimulus-response curves were constructed for different numbers of stimuli at low stimulation frequency (12.5 Hz), medium frequency (50 Hz) and near-maximal frequency (100 Hz; Figure 5C). At 12.5 Hz, a near-plateau in fl uorescence change was eventually reached after 30 to 40 stimulations, but responses to individual stimuli could still be seen (Figure 6A), consistent with the individual spike response to each stimulus (Figures 3D,E). A similar tendency to plateauing of the calcium signal has been shown to be due to an increased rate of calcium removal from the cytosol at high calcium concentrations (Ohnuma et al., 1999). The fl uorescence signal did not reach a plateau during long-lasting activation at 50-100 Hz, but the rate of rise of the dF/F signal decreased somewhat during the stimulus train. The continued increase in dF/F during long-lasting intense stimulation suggests that calcium indicator saturation (Beierlein et al., 2004) was not reached. In order to simplify the analysis in most of the experiments below, the number of stimuli was chosen to be well below the levelling-off of the response curves seen at the higher frequencies. In conclusion, mossy fi bres produce spike-burst related fl uorescence signals where the contribution of each spike can be discerned, and the imaging system was able to follow the responses over a range of activation frequencies covering a large part of the frequency range of turtle mossy fi bre fi ring.

GABA B RECEPTOR MODULATION OF MOSSY FIBRE CALCIUM SIGNALLING
In slices, activation of GABA-B receptors reduces presynaptic calcium currents in hippocampal (Wu and Saggau, 1995) and cerebellar mossy fi bres, resulting in a diminished mossy fi bre excitatory postsynaptic current (EPSC) in granule cells (Mitchell and Silver, 2000). In the cerebellum, this effect was most pronounced at low frequencies of mossy fi bre activity, with little effect above ∼10 Hz (Mitchell and Silver, 2000). Spillover of GABA from Golgi cell axonal terminals were most likely responsible for this presynaptic inhibition, and long-lasting (more than 1 s) high-frequency fi ring of a single Golgi cell was required. We wished to examine the requirements for presynaptic GABA B modulation of mossy fi bres in the intact cerebellar network. Baclofen (100 µM) inhibited the stimulus-evoked peak fl uorescence signal from rosettes by 8% (fi ve stimuli at 50 Hz; SD ± 3; n = 3 boutons; p < 0.005; cf. Mitchell and Silver, 2000). Addition of CGP55845 showed that low-frequency activation of mossy fi bres (12.5 Hz; Figure 6A) was affected by GABA B inhibition. The GABA-source was most likely Golgi cells being activated during the mossy fi bre stimulation in a fast feedforward manner (Kanichay and Silver, 2008).The onset of the GABA B effect was analysed by examining the individual responses to mossy fi bre activation ( Figure 6B) using data from high speed calcium imaging (Figure 6A). Each peak value was the average of four separate sweeps used for statistical comparison. On average, for all fi bres, the effect of GABA B inhibition was evident after 180 ms (SD ± 57). Inhibition was evident already at the fi rst stimulus in one rosette, and after 260 ms in all rosettes (Figure 6). The onset time of the GABA B effect was further tested using higher stimulation frequencies, in order to cover the frequency range of turtle mossy fi bres. Analysis of the responses to stimulation at 50 and 100 Hz showed that GABA B receptors could control the very early part of the fl uorescence response (Figures 6D,F), in line with the observation that CGP55845 affected a single mossy fi bre response in some cases. On average, traces in CGP55845 differed significantly (p < 0.05) from control traces after 68.8 ms (n = 5; SD ± 41) at 50 Hz and 47.1 ms (n = 8; SD ± 38) at 100 Hz stimulation frequency. The experimental results indicate that GABA B receptor activation was effective over the entire range of mossy fi bre fi ring frequencies and maintained an effect during continuous mossy fi bre activity. Furthermore, our results show that presynaptic GABA B inhibition in cerebellar glomeruli can have a rapid onset within a few tens of milliseconds. Mossy fi bres are active in vivo both continuously and as shorter lasting high-frequency bursts (van Kan et al., 1993;Garwicz et al., 1998;Jörntell and Ekerot, 2006). We therefore examined the effect of GABA B-Rs on shorter bursts. In particular, shorter bursts allowed to test the duration of the GABA B effect by a pairedpulse stimulation paradigm. A pronounced paired-pulse depression of the second burst occurred at interstimulus intervals up to 1 s (Figures 6G,H) in 11 axons out of 15 tested at this long interval. For all intervals, paired pulse depression was observed in 17/18 simple rosettes and in 6/7 complex rosettes. Consequently, the paired-pulse data from the two types of rosettes were combined for subsequent analysis. Using the peak amplitude of the fi rst burst in control medium and with CGP55845 as reference, respectively, the amplitude of the second burst was for 50 Hz intra-burst stimulation 0.71 (SD ± 0.14; n = 6 axons) at 200 ms inter-burst interval, 0.79 (SD ± 0.13; n = 11 axons) at 500 ms and 0.76 (SD ± 0.19; n = 10 axons) at 1 second in control medium, and 0.93 (SD ± 0.13), 0.99 (SD ± 0.16) and 1.01 (SD ± 0.15) in the presence of CGP55845 (50 µM). The differences between control and sweeps in the presence of CGP55945 were signifi cant at p = 0.005. For intra-burst stimulation at 100 Hz (Figure 6G), the amplitude of the second burst was 0.90 (SD ± 0.28; n = 8) at 500 ms intra-burst interval in control conditions, and 0.94 (SD ± 0.22) in the presence of 50 µM CGP55845. The difference was signifi cant at p = 0.005. Note that this analysis may appear to underestimate the effect of GABA B inhibition on the second burst, as the fi rst burst often increased in amplitude when CGP55845 was applied (Figures 6C-H; on average to 1.7 times of control; n = 15 fi bres). The long-lasting effect of the fi rst burst suggests that short bursts of 50-100 Hz Golgi cell spiking observed in vivo (van Kan et al., 1993) may lead to prolonged aftereffects. In conclusion, GABA B receptor activation resulted in a paired-pulse depression of mossy fi bres inputs in the adult, intact cerebellum. This sets a time window for ongoing presynaptic integration of mossy fi bre information in the cerebellum.

LOCAL CALCIUM SIGNALLING IN COMPLEX ROSETTES
The large size and often structure of complex rosettes suggest that local modulation of calcium dependent processes such as transmitter release could take place, since calcium diffusion is restricted in axons (Collin et al., 2005). Localised modulation of calcium entry in a rosette could in principle occur due to localised presynaptic plasticity or local modulatory effects due to presynaptic transmitter effects. For instance, differential activity of Golgi cells innervating the same glomerulus (Fox et al., 1967;Mugnaini, 1972) could result in a non-uniform extracellular GABA signal in a glomerulus. This could lead to differential calcium infl ux in the rosette innervating that glomerulus. In order to test this possibility, we pressure-applied GABA to one part of large complex rosettes before electrical stimulation of the axon (Figure 7), using a stimulus interval of 300 ms, similar to the time frame for a robust paired-pulse depression (cf. Figures 6G,H). Picrotoxin was present throughout in order to block any potential presynaptic GABA A receptors (Engelman and MacDermott, 2004;Alle and Geiger, 2007).
The rosette responded to a burst of stimuli with a fl uorescence increase at all locations. However, local GABA application reduced the fl uorescence signal at locations close to the puff-pipette (Figures 7A-E). The inhibitory effect of GABA application was not due to lack of GABA-responsiveness in parts of the rosette, as paired-pulse depression (cf. Figure 6) was observed at all locations, and was attenuated by CGP55845 ( Figure 7F). The maximal local effect of GABA in four complex rosettes was 40.7% (SD ± 7.2, n = 4) reduction of dF/F compared to control. In conclusion, a complex rosette may be functionally subdivided with regard to local GABA control of calcium signalling.

DISCUSSION
We analysed the electrophysiological responses and the calcium signalling in cerebellar mossy fi bres in the adult, intact cerebellum at physiological temperature for turtles, i.e. 15-35°C. Whole-cell recordings from rosettes characterized the basic electrophysiological properties, including TTX-sensitive subthreshold responses and repetitive Na + spike fi ring. Addition of 4-AP revealed voltage-sensitive calcium responses, identifi ed by their sensitivity to Co 2+ . In the intact cerebellum, calcium imaging revealed Na + spike-associated calcium signals from simple and complex rosettes. This calcium signal was reduced by GABA B-R activation over entire fi ring frequency range of mossy fi bres in turtles (up to ∼125 Hz). In the presence of picrotoxin, local GABA application showed that calcium infl ux in a complex rosette could be regulated locally, suggesting that large rosettes may be functionally compartmentalized by GABAB receptor activation.

ELECTROPHYSIOLOGICAL PROPERTIES OF MOSSY FIBRES
The whole-cell recordings revealed that adult turtle cerebellar mossy fi bre electrophysiology was similar to young mammalian cerebellar mossy fi bres (Maffei et al., 2002;Rancz et al., 2007) and other axons in the PNS and the CNS (Bostock, 1995;Geiger and Jonas, 2000). For instance, our results (Figure 2) indicate that high-threshold voltagegated calcium channels are present in turtle mossy fi bre terminals, consistent with previous results in mammals (Mitchell and Silver, 2000;Maffei et al., 2002). Further points of species similarity are suggested by a strong, persistent outward rectifi cation activated just above resting membrane potential, and an inward rectifi cation which was seen during hyperpolarization (Figure 1; Bostock, 1995;Geiger and Jonas, 2000;Rancz et al., 2007). In mammalian hippocampal axons, A-like currents have been reported to infl uence spike propagation (Debanne, 2004) and duration (Geiger and Jonas, 2000). However, while 4-AP increased excitability (Figure 2), suggesting that A-like conductances could be present in cerebellar mossy fi bres, repetitive Na + spike fi ring did not reveal a signifi cant increase in spike-width for the stimulation parameters used here (Figure 3), in contrast to the cumulative I-A inactivation which presumably affects all release sites in hippocampal mossy fi bres (Geiger and Jonas, 2000). The effect of 4-AP may, however, also indicate that other types of potassium channels are present (Southan and Robertson, 1998;Tan and Llano, 1999;Dodson et al., 2003;Ishikawa et al., 2003); for instance, TEA-sensitive currents have previously been shown to control excitability in mammalian mossy fi bre terminals (Maffei et al., 2003). Further experiments are needed to clarify this issue. Propagation failure at branch-points may be due to Acurrents (Debanne, 2004), but we did not routinely observe axonal February 2010 | Volume 4 | Article 1 | 11 Thomsen et al. Cerebellar mossy fi bres

FIGURE 7 | Local inhibition of calcium signalling in complex rosettes by focal activation of GABA B receptors. (A)
Complex rosette branching off a mossy fi bre axon (cf. Figure 1). The position of the GABA pipette is indicated diagrammatically. For each region of interest indicated by the square boxes 1-4, the fl uorescence signals to fi ve electrical stimuli at 50 Hz (black arrowhead below dF/F traces) are shown as control and 300 ms after a GABA puff (open arrowhead). Each trace is the average of at least three trials. In location 4 (C), GABA had no effect, whereas in location 1-3 (B), (D, E) prior GABA application depressed the fl uorescence response to subsequent electrical activation. The results suggest that proximal regions affected by GABA (1, 2) did not prevent the action potentials from propagating to more distal regions (4, (C)) unaffected by GABA. All regions showed a CGP55845-sensitive paired-pulse depression at inter-burst intervals of 500 ms (F), 1-4; panel numbers correspond to regions in (A). The rosette was confi ned to the side branches: the axon showed no fl uorescence signal increase, while the whole extent of the side branches gave a fl uorescence response to electrical stimulation, similar to other complex rosettes shown. The size of this rosette was comparable to the lateral extent of rosettes in fi xed and dehydrated tissue (Figure 1). In other experiments (not shown) similar results were obtained in complex rosettes along the axon: local GABA application at one end of long rosette (cf. Mugnaini et al., 1974)  In our experiments, GABA B receptors modulated mossy fi bre calcium signals at high fi ring frequencies (Figure 6), where little or no GABA B receptor effect on mossy fi bre-granule cell transmission was observed previously (Mitchell and Silver, 2000). This apparent difference could be due to species or age differences. Future experiments combining presynaptic mossy fi bre calcium imaging with electrical recordings from corresponding postsynaptic granule cells may resolve this issue. Recent results in mammals show that the GABA concentration in a glomerulus is suffi cient to activate GABA B-Rs at Golgi cell axonal terminals (Mapelli et al., 2009), and extracellular GABA in a glomerulus may not only result from Golgi cell spiking, but also result from nonspike related release , resulting in phasic as well as tonic GABA A-R mediated inhibition of granule cells (Rossi and Hamann, 1998). These data may explain the results (Figure 6) indicating that the mossy fi bre calcium signal occasionally appeared to be under tonic inhibition. The GABA B-R activation was long-lasting (Figure 6), giving rise to a paired-pulse depression similar to GABA B effects in other brain regions (Wachowiak et al., 2005). Since mossy fi bres may fi re in short bursts (van Kan et al., 1993;Garwicz et al., 1998;Jörntell and Ekerot, 2006), this paired-pulse depression may be particularly relevant for modulation of information transfer at the mossy fi bre synapse and the occurrence of temporal contrast, in line with the adaptation of sensory inputs to other brain regions. Our results also indicate that presynaptic GABA B-Rs can have a signifi cant effect during the maintained mossy fi bre fi ring commonly observed in vivo (van Kan et al., 1993).
An interesting possibility is that the presynaptic GABA B-Rs could infl uence the induction and presynaptic expression of longlasting changes in the strength of mossy fi bre-granule cell synapses, such as long-term potentiation (LTP) or long-term depression (LTD) (Maffei et al., 2002;Sola et al., 2004;Gall et al., 2005;Pan et al., 2009).
Presynaptic modulation (cf. Schneggenburger and Forsythe, 2006) of cerebellar mossy fi bres appears to be more complex than GABA B-R mediated control of calcium infl ux: activation of A 1 adenosine receptors reduced transmitter release without a noticable effect on presynaptic calcium currents (Maffei et al., 2002). In contrast to these reductions of mossy fi bre output, nitric oxide provides -through LTP induction -a persistent increase in presynaptic currents at mossy fi bre terminals (Maffei et al., 2003).
Golgi cells may not only affect mossy fi bres by indirect means such as spillover (Mitchell and Silver, 2000), but very rare but direct synaptic connections to mossy fi bres rosettes have been reported (Palay and Chan-Palay, 1974). The signifi cance of these connections is unknown, but they could produce highly localised GABA A-R, glycine-R and GABA B-R mediated modulation of a rosette, similar to results involving GABA A-Rs in hippocampal axons (Alle and Geiger, 2007).
The often extensive branching and large extent of many complex rosettes (Figure 1; Mugnaini et al., 1974;Palay and Chan-Palay, 1974) suggested that localised signal processing may take place within a rosette, in analogy with local calcium signalling in dendritic branches and spines. A dynamic compartmentalization of mossy fi bre rosettes could occur due to, e.g. localised GABAB-R activation. In order to test this, GABA was applied locally to only a part branching, even though mossy fi bres in turtles  and mammals (Wu et al., 1999;Sultan, 2001) possess collaterals; consequently, this leaves open the possibility for information fi ltering at these sites. Another mechanism for activity-dependent changes could be propagation failure due to spike-induced AHPs, as seen in a variety of vertebrate and invertebrate axons (for review, see Debanne, 2004). While cerebellar mossy fi bres and other vertebrate axons possess such AHP (Figure 3; Gordon et al., 1990;Morita et al., 1993;Kim et al., 2007), spike failures during prolonged activity were not observed in our experiments (Figure 3). In conclusion, our data indicate that the electrophysiological properties of adult mossy fi bres ensure the distribution of the incoming information to the presynaptic terminals with a high degree of robustness, over the entire range of fi ring frequencies. This electrophysiological information shows, where comparable data exist (Maffei et al., 2002;Rancz et al., 2007), that the properties of mossy fi bres are evolutionarily highly conserved. In addition, in the present study, it provided a background for the subsequent calcium imaging experiments.
In all of the complex rosettes imaged with calcium indicator, spiking evoked an all-or none fl uorescence signal from all sidebranches in parallel (Figure 4). Thus, we did not detect regions that would not be activated during action potentials. This suggests that a complex rosette is electrically compact during Na + spiking, with uniform activation of all parts of the rosette, including the fi ne branches and satellite boutons.

GABA B MODULATION OF ROSETTE CALCIUM SIGNALLING
We examined GABA B-R modulation of mossy fi bre terminals in the intact cerebellar circuitry. Previous slice experiments in mammals established that mossy fi bre Ca 2+ -channels are inhibited by baclofen and Golgi cell-mediated GABA B-R activation (Mitchell and Silver, 2000). In these experiments, single Golgi cells had to be activated at high fi ring frequencies for at least 1 s before an effect on the MF-granule cell EPSP was detectable, and maximum EPSC depression was observed after ∼5 s of Golgi cell activity. Shorter (250 ms) high-frequency spiking of Golgi cells did not have any effect (Mitchell and Silver, 2000). In our experiments, the average onset time of mossy fi bre GABA B inhibition was ∼50-70 ms. In some cases (∼40%) no depression of mossy fi bre output transmission was observed (Mitchell and Silver, 2000), while we observed a GABA B-R effect in 16 out of 21 mossy fi bres tested (24% nonresponders to addition of GABA B receptor antagonist). The reasons for the differences in responsiveness and onset kinetics between our data and the slice experiments are not clear, since the distribution of GABA B receptors in turtles is similar to mammals (Albin and Gilman, 1992), but in our experiments it is likely that several Golgi cells projecting to a glomerulus (Fox et al., 1967;Mugnaini, 1972) are activated by the electrical stimulus, perhaps creating a larger spillover of GABA than the activation of individual Golgi cells by Mitchell and Silver (2000). GABA diffusion out of the slice preparation may result in a less effective spillover, and severed Golgi cell axons may account for the occasional failure to activate depression of the mossy fi bre-granule cell transmission (Mitchell and Silver, 2000). Finally, the slice data were obtained in very young rats (P12-13), while our data were obtained in adult animals. The glomerulus, including the glial wrapping, undergoes postnatal developmental maturation (Hamori and Somogyi, 1983;Mitchell and Silver, 2000). This may result in differences in extracel-physiological temperature (cf. Mitchell and Silver, 2000) opens the possibility that a large glomerulus may be functionally compartmentalized by local differences in extracellular GABA concentration, depending on the spatial activity pattern of Golgi cells. In this way, presynaptic inhibition by Golgi cell-mediated feedback-and lateral-inhibition to a particular glomerulus may fi nely regulate the mossy fi bre input to specifi c granule cell subpopulations contacting this glomerulus (Figure 8). Evidence has previously been obtained in the cat that Golgi cell inhibition of separate mossy fi bre inputs could be due to anatomically very specifi c distribution patterns of Golgi cell axons (Precht and Llinás, 1969), but it is presently not clear if this may be related to the scenario proposed in Figure 8. Compartmentalization of a glomerulus by Golgi cell activity ( Figure 8B) may hypothetically lead to differential LTP/ LTD induction (cf. Gall et al., 2005) at different mossy fi bre-granule cell synapses in this glomerulus (cf. D'Angelo et al., 1999), due to spatially specifi c and varying degrees of Golgi cell-mediated inhibition (cf. Eccles et al., 1966;Precht and Llinás, 1969;Gabbiani et al., 1994) of voltage-dependent calcium infl ux in different granule cells belonging to the glomerulus. This would increase the cerebellar information storage capacity at the glomerular level compared to a scenario where each glomerulus functions as a unit. This hypothesis could be directly tested in future LTP/LTD-induction experiments by recording simultaneously from two granule cells belonging to the same glomerulus.
While in many cases each presynaptic bouton in the CNS contacts only one postsynaptic cell, often with several release sites, as for instance the calyx of Held (Schneggenburger and Neher, 2005), some types of boutons may contact several cells, for instance the cerebellar mossy fi bre rosette Palay and Chan-Palay, 1974). It will be of interest to see whether neighboring release sites to different postsynaptic cells may be individually regulated, both in terms of presynaptic modulation of transmitter release and in terms of synaptic plasticity. To this end, development of new calcium probes (Qiu et al., 2008), perhaps targetable to active zones, would be of importance, and to further analyse the spatial activation of branched, complex presynaptic terminals, voltage sensitive indicators (Qui et al., 2008)

SUPPLEMENTARY VIDEO
Sequence of images of a living mossy fi bre fi lled with Lucifer Yellow in a slice of the cerebellar cortex. Each image was taken at a different depth from the cut surface of the slice using computer-controlled stepping of the microscope z-axis. Z-axis distance between images was 2 micrometers. Same mossy fi bre as shown in fi gures 1C-E.  Figure 7G. (B) In another scenario, no or marginal overlap of Golgi cell axons occur within a large glomerulus. Different compartments of a large complex rosette could in this way be subject to local presynaptic inhibition by different Golgi cell populations. This indicates that control of input to the cerebellar cortex could be more fi ne-grained than suggested by the scenario in (A), where an entire glomerulus functions as an input unit. Future anatomical analysis may allow to differentiate between these alternatives. In (B), the relative activity levels (indicated by the number and spacing of small vertical lines close to each cell) of Golgi cells may dynamically control presynaptic calcium infl ux in different branches of a rosette. Feedback as well as lateral-inhibition by Golgi cells could functionally compartmentalize the glomerulus. Note that the Golgi cells in the diagrams (A) and (B) represent two different cell populations and are not an indication of the actual number of Golgi cells innervating a glomerulus. Note also that the innervation scenarios depicted in (A) and (B) could be considered to illustrate opposite ends of a spectrum. of a rosette, resulting in a regional decrease in the stimulus-evoked calcium signals (Figure 7). In contrast, activation of the mossy fi bre spike invariably resulted in a widespread, uniform rosette calcium signal. More focal electrical stimulation in the granule cell layer than we could achieve in the present study may allow to test whether differential activation of the Golgi cells projecting to a glomerulus (Fox et al., 1967;Mugnaini, 1972) can induce compartmentalization of the calcium signalling in a rosette. Analysis of calcium signals from cerebellar mossy fi bre and squid synaptosomes (Serulle et al., 2007) has indicated how, ultimately, presynaptic calcium microdomains set the lower limit for local presynaptic calcium signalling and compartmentalization at a scale corresponding to the size and separation of active zones (Xu-Friedman and Regehr, 2003). Pressure-applied GABA, as in our experiments, spreads for comparatively much longer distances. Future experiments with more focal GABA application using optical GABA-uncaging (Trigo et al., 2009) might allow to analyse the spatial relationship between GABA B-receptor activation and local regulation of presynaptic calcium infl ux in fi ner detail, on a spatial scale of a few micrometers, taking into account the extracellular diffusion of GABA within a glomerulus (cf. Rossi and Hamann, 1998).
If the axonal territories of Golgi cells do not overlap completely within a large complex glomerulus (Figure 8; cf. Figures 1 and 7), the restricted extracellular GABA diffusion within a glomerulus at