In decerebrated, nonanesthetized cats, we made intracellular whole-cell recordings and extracellular cell-attached recordings from granule cells in the cerebellar C3 zone. Spontaneous EPSPs had large, relatively constant peak amplitudes, whereas IPSPs were small and did not appear to contribute substantially to synaptic integration at a short time scale. In many cases, the EPSPs of individual mossy fiber synapses appeared to be separable by their peak amplitudes. A substantial proportion of our granule cells had small receptive fields on the forelimb skin. Skin stimulation evoked explosive responses in which the constituent EPSPs were analyzed. In the rising phase of the response, our analyses indicated a participation of three to four different mossy fiber synapses, corresponding to the total number of mossy fiber afferents. The cutaneous receptive fields of the driven EPSPs overlapped, indicating an absence of convergence of mossy fibers activated from different receptive fields. Also in granule cells activated by joint movements did we find indications that different afferents were driven by the same type of input. Regardless of input type, the temporal patterns of granule cell spike activity, both spontaneous and evoked, appeared to primarily follow the activity in the presynaptic mossy fibers, although much of the nonsynchronized mossy fiber input was filtered out. In contrast to the prevailing theories of granule cell function, our results suggest a function of granule cells as signal-to-noise enhancing threshold elements, rather than as sparse coding pattern discriminators or temporal pattern generators.
Cerebellar granule cells are only 5–8 μm in diameter and have only four dendrites (range, 3–5), which receive one excitatory mossy fiber synapse each (Palay and Chan-Palay, 1974; Cathala et al., 2003). Therefore, they offer an ideal opportunity to study the contribution of individual synaptic afferents in the integration process. From in vitro studies, we know that the excitatory synaptic potentials are large (Silver et al., 1992; D'Angelo et al., 1995) and that inhibitory input from Golgi cells has a strong tonic component (Brickley et al., 1996; Rossi and Hamann, 1998). A study in the anesthetized rat has essentially confirmed these findings in in vivo whole-cell recordings (Chadderton et al., 2004).
However, the integrative properties of granule cells when the afferents are driven by synaptic input in a natural form remain primarily unknown. In the present study, we perform intracellular recordings with the whole-cell patch-clamp technique from granule cells in the C3 zone of the decerebrate, nonanesthetized cat. The C3 zone is one of the cerebellar sagittal zones (Apps and Garwicz, 2005) in which the cutaneously activated mossy fibers are easily driven and have particularly small and well defined peripheral receptive fields on the forelimb (Garwicz et al., 1998). In this system, mossy fibers with similar or overlapping receptive fields are terminating together in a somatotopical organization, and the activation kinetics on skin stimulation has been studied (Garwicz et al., 1998). Also, Golgi cells with specific cutaneous receptive fields are distributed according to the same organization (Ekerot and Jörntell, 2001), and because their axons are distributed locally, cutaneously activated inhibitory input to granule cells may be expected to arise from about the same skin area as the excitatory input.
Important issues that could be addressed with these recordings are as follows: (1) the behavior of small neurons when receiving natural spatiotemporal patterns of excitatory/inhibitory synaptic input in a nonanesthetized in vivo preparation; (2) the information content of granule cell–parallel fiber spikes: what spatiotemporal transformations do they represent (i.e., what are the convergence patterns of excitatory and inhibitory afferents on granule cells and how is the temporal pattern of mossy fiber input transformed to granule cell spikes that are conducted up to the molecular layer in the parallel fibers)?
These are fundamental issues for cerebellar physiology that also will provide an evaluation of current models of cerebellar granule cell function. Our data provide little support for a role of granule cells as pattern discriminators, sparse coders, and/or temporal pattern generators but instead suggest that granule cells have a simpler, more straightforward function as signal-to-noise-enhancing threshold detectors.
Materials and Methods
Adult cats were prepared as described previously (Ekerot and Jörntell, 2001; Jörntell and Ekerot, 2002, 2003). Briefly, after an initial anesthesia with propofol (Diprivan; Zeneca, Macclesfield Cheshire, UK), the animals were decerebrated at the intercollicular level and mounted in a stereotaxic frame. The animals were artificially ventilated, and the end-expiratory CO2, blood pressure, and rectal temperature were monitored continuously and maintained within physiological limits. Drainage of CSF, pneumothorax, and clamping the spinal processes of a few cervical and lumbar vertebral bodies served to increase the mechanical stability of the preparation. Our EEG recordings were characterized by a background of periodic 1–4 Hz oscillatory activity, periodically interrupted by large-amplitude 7–14 Hz spindle oscillations lasting for ≥0.5 s. These forms of EEG activities are normally associated with deep stages of sleep (cf. Niedermeyer and Lopes da Silva, 1993). The pattern of EEG activity and the blood pressure remained stable, also on noxious stimulation, throughout experiments.
Recordings and stimulation.
The initial delineation of the forelimb area of the C3 zone in the cerebellar anterior lobe and the continuous monitoring of the general condition in the sensitive mossy fiber→granule cell→parallel fiber pathway were performed as described previously (Ekerot and Jörntell, 2001; Jörntell and Ekerot, 2002). Also, the general recording procedures and the procedures for placing stimulation electrodes in the inferior olive and among the superficial parallel fibers have been described in detail previously (Jörntell and Ekerot, 2003). Electrical stimulation of mossy fiber afferents was used to study the properties of directly activated mossy fiber synapses. This stimulation was made with tungsten-in-glass microelectrodes inserted in another sagittally aligned folium of the C3 zone (transfolial activation), around the anterior interposed nucleus (supranuclear), or in the lateral reticular nucleus (LRN).
Patch-clamp electrodes were pulled to 13–30 MΩ from borosilicate glass capillaries on a vertical puller (NN-850; Narishige, Tokyo, Japan) and filled with a solution containing (in mm) 135 K-gluconate, 7 KCl, 0.1 CaCl2, 10 HEPES, and 2 Mg-ATP, or with an alternative solution containing (in mm) 135 K-gluconate, 7 KCl, 2 EGTA, 10 HEPES, and 2 Mg-ATP. The solution was titrated to pH 7.35–7.40 using KOH. In some cases, 1–3% neurobiotin (Vector Laboratories, Burlingame, CA) or biocytin (Sigma, St. Louis, MO) was included in the recording solution to allow postmortem morphological reconstructions of the recorded neurons. Histological procedures for biocytin were as described previously (Jörntell and Ekerot, 2003). Sections stained with neurobiotin were processed using streptavidin Alexa-488 conjugate (Invitrogen, Eugene, OR) in 1% bovine serum albumin overnight. Stained cells were analyzed in a confocal microscope (LSM 510; Zeiss, Oberkochen, Germany). The junction potential between the electrode solution and the extracellular (EC) environment of the cerebellum was measured to 5–7 mV in two experiments [this value is in agreement with that of Margrie et al. (2002), who used a similar K-gluconate-based solution in vivo] but was not compensated for in the displays. The steps to measure the junction potential were (1) place a recording electrode approximately 0.5 mm above the surface of the cerebellar cortex, (2) replace the paraffin oil covering the brain with the electrode solution (note that the earth electrodes were placed in the neck muscles and hence not in contact with the electrode solution), and (3) measure the difference in DC potential between the electrode solution and the EC environment by moving the electrode a few millimeters down in the cerebellum.
Granule cell recordings were obtained by first lowering the patch-clamp electrode under high positive pressure to a location just beneath the first Purkinje cell layer (PCL). The depth of the PCL in the recording area was repeatedly confirmed by EC recordings of Purkinje cells [identified by the coexistence of simple and complex spikes (Ekerot and Jörntell, 2001)] made en passage. Once in the granule cell layer, the high positive pressure was relieved to more moderate levels, and the electrode was slowly advanced until a substantial increase in tip resistance occurred. At that point, a hyperpolarizing current of 0.1–0.9 nA and a negative pressure was applied to establish a gigaohm seal (0.5–10 GΩ) between the electrode and the recorded cell. Once established, brief increases in negative pressure were used to break the seal and gain contact with the interior of the cell. All analysis performed in this study, except when stated otherwise, was made in granule cells that were prevented from firing spontaneously by using mild hyperpolarizing currents of 0.01–0.10 nA. The amount of bias current needed to prevent cells from firing spontaneously, the maximal EPSP amplitudes, and the amplitudes of occasional evoked spikes were monitored continuously to check the quality of the recording. Any substantial deviations from the maximal values for the neuron and the normal values for the population were taken as a sign of poor recording conditions, and such neurons/periods were excluded from the analysis. For example, if bias currents needed to control a cell exceeded 0.10 nA, spike amplitude and time course often looked abnormal, and such cells were immediately discarded. In general, the difficulties consisted of establishing a high seal and making a controlled break-in. If these stages were successful, they resulted very often in high-quality recordings, and <10% of these recordings had to be discarded. As the recordings approached an end, they could either terminate suddenly and completely or there would be a sudden reduction in quality associated with a depolarization and change in spike appearance. Such periods were also excluded from analysis.
EC recordings in the cell-attached mode were made with the same type of patch-clamp electrodes using a light suction to improve the isolation of the unit. A few EC recordings from mossy fiber terminals were made with fine tungsten-in-glass electrodes (exposed tip, 3–5 μm) as described previously (Garwicz et al., 1998; Ekerot and Jörntell, 2001).
To construct peristimulus histograms of activity evoked from the skin during manual stimulation, we used a strain-gauge device mounted on the index finger of the investigator (Jörntell and Ekerot, 2002, 2003). The signal of the strain-gauge device was used mainly as a trigger signal and as a general indication that approximately the same amount of force was applied during stimulations. For each recorded unit with cutaneous input, we attempted to identify the submodality of the input. Units responding to bending of hairs and air puffs from the mouth of the investigator were classified as hair units, whereas units not responding to such stimulation but still firing vigorously on light skin touch were classified as non-hair cutaneous units.
Analysis of EPSPs and IPSPs.
All analyses of synaptic inputs were made at a membrane potential of approximately −70 mV, except where stated otherwise. Identification of EPSP-like events and measurements of their peak amplitudes were made semiautomatically using home-made software: the software found and displayed an event, and the user accepted/rejected it and identified its peak amplitude. This semiautomatic procedure was checked manually to accurately find 99–100% of all EPSPs that exceeded 1.3 mV in peak amplitude. For the data used in the frequency distribution histograms for spontaneous EPSPs with specific amplitudes, we excluded EPSPs that were preceded by another EPSP of similar or greater amplitude by <10 ms because of the risk that repetitive activation could lead to a depression of the synaptic response. To minimize the risk that coincident EPSPs provided a source of error for the measurement of unitary peak amplitudes, the time course of each analyzed EPSP was checked and fitted to the time courses of other EPSPs that had previously been analyzed in the cell. If the time course of an EPSP could not be fitted to those of the other EPSPs with similar amplitude, that event was rejected. Note that the very fast rise times of the EPSPs (Table 1) made it possible to determine the peak amplitudes also on the decay phases of other EPSPs, because the baseline slope was much slower in comparison. In such cases, the EPSP amplitude was measured between the starting point and the peak amplitude of the EPSP.
The experimental procedures were approved in advance by the local Swedish Animal Research Ethics Committee.
In the forelimb area of the cerebellar C3 zone, we made intracellular recordings from 43 granule cells, all of which were scanned for somatosensory input from the ipsilateral forelimb. Twelve of these cells were morphologically identified after successful histological recovery after staining with biocytin or neurobiotin (Fig. 1). Another 20 granule cells were stained after EC recording (see below). Of the 32 stained granule cells, 24 had four dendrites per dendritic endings (“claws”), four had three dendritic endings, and four had five dendritic endings. Other recordings were identified as granule cells by their high input resistance (Table 1) and by their lack of excitatory responses to climbing fiber and parallel fiber activation. Note that the values of input resistances and membrane time constants (Table 1) are substantially lower than those of Chadderton et al. (2004), which could be explained by (1) a higher background activity in mossy fibers caused by the absence of anesthesia in our preparation and (2) the fact that we used more mature animals in which the tonic inhibitory activity would be expected to be stronger (Brickley et al., 1996; Wall and Usowicz, 1997). In addition, granule cells were characterized by very large spontaneous EPSPs and a lack of sizeable fast IPSPs.
To analyze individual mossy fiber EPSPs during peripheral stimulation, it was important to evaluate how constant the peak amplitudes of single mossy fiber inputs were. To this end, we used threshold stimulation with a metal microelectrode placed in the granule layer/white matter of an adjacent folium (transfolial). The stimulating microelectrode was sagittally aligned with the recording microelectrode, and in the beginning of the experiment, it was verified that stimulation evoked local field potentials in the recording area. In line with previous in vitro studies (Silver et al., 1992, 1996; D'Angelo et al., 1995; Brickley et al., 2001), we found that the peak amplitudes of unitary mossy fiber EPSPs evoked by electrical stimulation had a low variability (Fig. 1 A,B; Table 1). Note that mossy fiber EPSCs in vitro have been shown to be multiquantal and consecutive responses vary in amplitude in discrete steps when recorded at room temperature (Wall and Usowicz, 1998). Thus, the variability in the number of quanta released seems to be lower in our in vivo preparation. Both spontaneous and evoked EPSPs had fast rise times and large peak amplitudes (Fig. 1, Table 1). Paired stimulation at short interstimulus intervals resulted in a substantial depression of the second evoked EPSP (Fig. 1 C,D). The average maximal depression (63%) was remarkably similar to values obtained in vitro (Xu-Friedman and Regehr, 2003), but the average time constant of the recovery, calculated from our pooled data, was much faster in vivo (8 ms compared with 35 ms in vitro).
In granule cells in which spontaneous activity was recorded for a sufficient amount of time (≥200 s; n = 16), we tested whether the frequency distribution histograms of EPSP amplitudes could reveal individual mossy fiber inputs. We found that such histograms were always characterized by three to five peaks, but the amount of separation between peaks varied. Examples of cells in which the population of EPSP amplitudes was distinctly separated into groups are shown for granule cells with four dendrites (Fig. 1 E–G) and for one granule cell with five dendrites (Fig. 1 H). In addition, an example of a granule with less clear separation of EPSP amplitudes is shown in Figure 1 I. In Figure 1 E, hyperpolarization down to −90 mV was used to increase EPSP amplitudes and hence amplify the separation. In other cases, the cells were held at approximately −70 mV. In all cases, there was a group of small spontaneous EPSP-like events of unclear nature with amplitudes just above detection level. Within this group, we sometimes found EPSPs with a distinctly slower time course (see Fig. 7 A), which has also been found for granule cell EPSCs in vitro, where they have been attributed to spillover activation of AMPA receptors (DiGregorio et al., 2002). At low intensities, the spike responses to rectangular current pulses varied with the spontaneous EPSPs activity, but stronger current pulses elicited intense, regular spike trains with little sign of adaptation (Fig. 1 F–I).
Another conspicuous feature of granule cell recordings was a lack of substantial fast IPSPs at resting potential (−60 to −50 mV) and when the cells were depolarized (up to −20 mV) with a positive bias current. The most prominent inhibitory responses were observed on stimulation of skin areas in the vicinity of the excitatory receptive field of surrounding granule cells, where excitatory responses were not evoked (Fig. 2 A,B). However, such stimulation failed to evoke substantial fast IPSPs (defined as hyperpolarizing potentials of >1.0 mV and time-to-peak <5 ms), even though the averaged response proved to contain an IPSP with a slow time course and the Golgi cells within this area typically responded to such stimulation (Fig. 2 A). The peak amplitude of averaged inhibitory responses evoked by electrical skin stimulation varied with the membrane potential, seemingly in a linear manner with a reversal potential approximately −75 mV (Fig. 2 C). Note that the low amplitudes of fast IPSPs were probably not attributable to our recording conditions, because we under the same conditions have recorded sizeable fast IPSPs in stellate and basket cells at rest (Jörntell and Ekerot, 2003). The discrepancy with the data of Chadderton et al. (2004) on this point could be as a result of the absence of anesthesia in our preparation or to the differences in age (our animals were adult, whereas Chadderton et al. used 18- to 27-d-old rats). The relative lack of fast IPSPs in our preparation do, however, clearly not contradict findings from a series of in vitro studies, in which it has been suggested that the role of Golgi cell inhibition in granule cells is to set the excitability level on a longer time scale using a predominantly tonic form of inhibition (Brickley et al., 1996; Wall and Usowicz, 1997; Rossi and Hamann, 1998; Hamann et al., 2002).
Granule cells had different types of peripheral input
Once a granule cell recording was established, the peripheral activation of the spike was investigated qualitatively by manual stimulation to determine to what extent it was activated by cutaneous and joint movement-related input on the forelimb. In line with the known pathways of mossy fiber input to this area of the cerebellar cortex, we found granule cells to be of four types: cutanoeusly activated, strongly activated by joint movement, moderately/weakly activated by joint movements, or not activated from the periphery. For reasons discussed later, the inputs to the activated cells are likely to be from the exteroceptive cuneocerebellar tract (E-CCT), the proprioceptive component of the cuneocerebellar tract (P-CCT), and the LRN. Cells that could not be driven probably received mossy fiber input from pontine nuclei (Brodal and Bjaalie, 1992; King et al., 1998) normally driven by the cerebral cortex, a connection that was transected by the decerebration.
Granule cells with cutaneous input were the most common type. Figure 3, A and B, illustrates the responsiveness to cutaneous stimulation of a mossy fiber and a granule cell, respectively, in EC recordings. The raw data and the histograms of spike activation on repeated stimulation illustrate a low spontaneous activity and extremely strong responses on peripheral activation, which were characteristic for mossy fibers (Garwicz et al., 1998) and granule cells with this type of input. Figure 3 also illustrates the general differences in background and evoked synaptic activity between granule cells with different types of input. Generally, strong peripheral activation was associated with low spontaneous activity and vice versa. Strong peripheral activation was also associated with large shifts in the membrane potential produced when EPSPs occurring at high frequencies summated, whereas weaker responses produced a less marked shift in membrane potential. In cells with no input, but with spontaneous spike firing (Fig. 3 F), underlying spontaneous depolarizations caused by spontaneous coincidences of background EPSP activity appeared to be responsible for the spike firing.
Cutaneous receptive fields
Because of the virtual absence of spontaneous activity and the extreme intensity of the evoked responses, the cutaneous receptive field for spike activation was readily delineated. These receptive fields were similar to those reported previously for mossy fiber terminals (Garwicz et al., 1998) with respect to size and location on the skin (Fig. 4 A). Just like in mossy fibers, a sensitivity gradient within the receptive field [defined by the presence of skin areas within the receptive field from which <70% of the maximal response was evoked (compare Fig. 4 B)] was sometimes observed also in granule cells. To quantify the receptive field for EPSPs, cells were hyperpolarized down to −70 mV to prevent spiking. Input was tested from several small skin areas, in turn, by using repeated brief, light strokes. A strain gauge mounted on the fingertip of the investigator was used to obtain data on stimulus onset and duration and to control that the force applied to the different skin areas did not differ substantially (Fig. 4 B,D). The peristimulus histograms (Fig. 4 C) and the integrated synaptic responses (Fig. 4 D) indicated that EPSPs could be evoked from one small skin area only. To quantify this observation, we compared the average net integrated synaptic response evoked from within the receptive field and from adjacent skin areas located outside the receptive field (Fig. 4 D). Adjacent skin sites were located on either side of the receptive field, either along the proximo-distal or the radio-ulnar axis.
Analysis of peripheral activation of individual EPSPs
In cutaneously activated granule cells, the initial part of the response evoked by electrical skin stimulation in the center of the receptive field consisted of a near instantaneous depolarizing step of 15–25 mV (from −70 mV) (Figs. 5 ⇓–7) (i.e., severalfold larger than the maximal amplitudes of the unitary EPSPs) (Table 1). These responses always had a constant response onset latency time of 6–8 ms, identifying the afferent pathway as the E-CCT (Cooke et al., 1971). Weaker responses being composed of only one unitary EPSP could sometimes be evoked from the outskirts of the receptive field (Fig. 5 B), with longer, yet constant, response latencies.
Figure 6 illustrates additional examination of the composition of responses evoked by electrical skin stimulation. Morphological identification and the frequency distribution of spontaneous EPSP amplitudes indicated that this cell had four dendrites (Fig. 6 A,B). Electrical skin stimulation elicited a strong, stereotyped EPSP response (Fig. 6 C). We set out to analyze the minimal number of mossy fiber synapses that could underlie the evoked response by first defining a time window within which a simulated response must reach the peak amplitude of the evoked response (Fig. 6 D). Averages of EPSPs within the four different groups of amplitudes were used to simulate, by addition, evoked responses composed of activity in different mossy fiber afferents (Fig. 6 E–G). First, we started with the single EPSP with the largest peak amplitude, but neither 500 nor 1000 Hz EPSP activation of this afferent came even close of being capable to reproduce the evoked response. We also noted that because of the paired-pulse depression ratios at 1–3 ms intervals (60–63%), 1 ms intervals were required for temporal summation to gain additional depolarization during the EPSP train. Therefore, as a next step, we added the activation of other EPSPs (ii–iv) also at 1000 Hz (Fig. 6 G). We found that only when we simulated a near simultaneous activation of all four EPSPs at 1000 Hz was it possible to reproduce the depolarization observed in the evoked response. It can be noted that a double activation at 1000 Hz of all four simulated EPSPs was enough to reach the same level as in the evoked response (Fig. 6 G, bottom), whereas an additional 1000 Hz activation made the simulated response “overshoot” its target. In nonsimulated evoked response (Fig. 6 D), it is likely that the plateau depolarization obtained is attributable to asynchronous, lower-frequency activation of the different mossy fiber afferents.
We performed this analysis in six cells in which the evoked responses reached a peak within 3–4 ms after onset. The results indicated that simultaneous activation of 3.51 ± 0.51 different mossy fiber afferents was required to reproduce the observed responses. Note that this measure is likely to be conservative, because mossy fibers typically do not respond as intensely as in our simulation on electrical skin stimulation (Garwicz et al., 1998) and because no additional depression beyond the first pair of EPSPs was integrated in the simulation.
In cutaneously activated cells, the first few milliseconds of the evoked responses seemed to be built up of several individual EPSP-like events. The time courses and amplitudes of these events, delineated by inflection points on the time–voltage curve, could be fitted with averages of individual synaptic inputs recorded in the cell (Fig. 7). The derived voltage signal showed that the inflection points used to define the onset times of EPSPs clearly differed from the range of the baseline noise (Fig. 7 C). This was always confirmed to be the case when the rising phases were analyzed with respect to EPSP composition. Of special interest were responses in which two or more EPSPs were initiated within 1 ms of each other, or in which consecutive EPSPs had increasing response amplitudes. In such response phases, the EPSPs had to be caused by activity in different mossy fiber afferents, because mossy fibers do not fire faster than 1000 Hz in vivo (van Kan et al., 1993; Garwicz et al., 1998) and unitary EPSPs depress substantially when activated repetitively at short intervals (Fig. 1 D). In a systematic analysis of EPSP-like events in the rising phases of the evoked responses, we combined these two criteria to obtain an estimate of the theoretically minimal number of individual synapses activated in the initial 5 ms of responses evoked by electrical or manual stimulation. This analysis was performed for 10–20 consecutive responses in all 10 cells in which we obtained a sufficient amount of responses recorded at approximately −70 mV. We found that the initial part of the response contained 3.2 ± 0.64 EPSPs (range, 2–4). A total of 2.6 ± 0.71 of these (range, 1–4) had to be attributable to the activity in different mossy fiber synapses, because they occurred at higher rates than 1000 Hz or because consecutive EPSPs had increasing amplitude.
Granule cells with noncutaneous input
Within the group of granule cells with deep input, cells activated from individual digits or the wrist joint typically had strong responses and therefore probably received their input from the P-CCT (Cooke et al., 1971). Among the cells that were activated from the elbow or shoulder joints, some responded much less distinctly and faithfully than others and were, in some cases, only activated by squeezing of deep structures. Their peripheral activation was similar to that found for cells in the LRN (Oscarsson and Rosén, 1966; Clendenin et al., 1974), and when tested (n = 10), these cells were also found to be activated by electrical stimulation in the LRN. Granule cells with noncutaneous input were spontaneously active (Fig. 2, Table 2). However, the onset of the response was less intense than in cutaneously activated granule cells. For example, the net integrated synaptic responses were only a little more than double the background activity (Table 3). The analysis performed for cutaneously activated granule cells was hence not possible. Nevertheless, during stimulation by appropriate joint movements, we found that EPSPs of all amplitudes were driven by the same type of input. This is shown as examples for one cell with strong activation and for another cell with weaker, less direct activation on joint movement (Fig. 8). In the cell with strong activation, we used a weaker, long-lasting stimulation to obtain a low level of tonic activation during which the amplitudes of individual EPSPs were relatively unperturbed (Fig. 8 C). Using stronger stimulation, the initial part of the response was a strong phasic response with EPSP summation creating a depolarizing shift in the membrane potential (Fig. 8 A). Such shifts in membrane potential were observed also in other cells with strong activation by joint movements. The amplitude of the DC shift exceeded the peak amplitude of the maximal unitary EPSP (6–7 mV) by 58 ± 11% (mean ± SD; n = 5) and could therefore not be attributable to the activation of a single EPSP only. (For comparison, in cells with cutaneous input, the DC shift exceeded the maximal unitary EPSP amplitude by 204 ± 24% (n = 6). In a cell with weaker, indirect activation on elbow extension, peristimulus histograms showed that all EPSPs, as defined in the amplitude frequency histogram, were driven by this input (Fig. 8 D–F).
Properties of cell-attached recordings
In addition to the intracellular recordings, we made 65 EC granule cell recordings using the cell-attached recording mode. Twenty of the EC recordings were morphologically identified as granule cells using the juxtacellular labeling technique (Pinault, 1996). Apart from the absence of spike responses to inferior olivary stimulation, EC granule cell recordings were distinguished from EC Golgi cell recordings by their shorter spike durations [mean ± SD: 0.87 ± 0.16 ms for 25 EC granule cell recordings, 1.5 ± 0.12 ms for 5 EC Golgi cells; the latter value is in agreement with van Kan et al. (1993)], the much shorter recording distance over which the spike could be followed, and their more irregular spontaneous firing rate (cf. Holtzman et al., 2006). They were also distinguished from mossy fiber recordings by the absence of a “glomerulus potential” (Walsh et al., 1974), their longer spike durations [the duration of EC mossy fiber spikes is <0.5 ms (Fig. 2 A, inset) (van Kan et al., 1993; Garwicz et al., 1998)], and, in 70% of the EC recordings, by the observation of small EPSP-like events during some stage of the recording.
Spontaneous and evoked spike activity in granule cells with different inputs
In the total material of IC and EC recorded granule cells, our tests indicated that all 108 cells except two were activated by one type of somatosensory input only. Furthermore, there was a specific depth distribution of granule cells depending on the type of input they received (Table 4). Cutaneously activated granule cells were, by far, the most common type. The firing characteristics of these cells were similar to those of mossy fibers of the E-CCT (Garwicz et al., 1998; Ekerot and Jörntell, 2003) in that the spontaneous activity was low, in granule cells zero, but appropriate peripheral activation evoked very strong spike firing throughout the duration of the stimulation (Figs. 2, 9; Table 3). However, the extreme firing frequencies observed is likely to primarily reflect that the manual skin stimulation used is an optimal activation of these inputs. During behavior, these inputs may be used for cutaneous proprioception (see Discussion), which probably will activate the cells in a lower-frequency range. Also, proprioceptive mossy fibers of the LRN (Clendenin et al., 1974) and the P-CCT (Cooke et al., 1971) can fire at 1 kHz under appropriate peripheral activation, but activation under behavior seems to consist in much more moderate firing rates (50–300 Hz) (van Kan et al., 1993).
In the majority of cutaneously activated granule cells, we even found a submodality-specific input, because some cells were activated by skin hairs only, whereas others were not activated from hairs at all. Note also that the spontaneous EPSP frequency in granule cells was much higher than the spontaneous spike firing frequency, which at least for cutaneously activated granule cells that lacked spontaneous spike firing directly shows that activation of more than one EPSP was required to fire the cell (Table 2).
We found a general relationship between the activity of spontaneous EPSPs and spontaneous spikes (Table 2), suggesting that the spike activity of granule cells primarily was a reflection of the ongoing synaptic input. To further test this, we first compared the spike activation in granule cells that were phasically activated on brief stimulation of skin hairs. The peristimulus histograms of the evoked responses were remarkably congruent between different granule cells, and the response onset latency times were identical (Fig. 9 B). However, mossy fibers carrying this type of information are typically rapidly adapting (Garwicz et al., 1998), and longer-duration stimulations resulted in spike trains with adapting firing frequencies (Fig. 9 C). Granule cells with tonic responses to skin stimulation were few, but as can be seen in the example in Figure 9 D, these cells responded with sustained spike trains for the duration of the stimulation. Among granule cells activated by joint movements, a tonic response component was more common. Also, in this case, was there a sustained spike firing for the duration of the stimulation, with no obvious consistent temporal pattern after the initial part of the response (Fig. 9 E).
In the present study, we investigated the peripheral input in intracellular whole-cell and EC cell-attached granule cell recordings in a nonanesthetized in vivo preparation. In granule cells activated by skin input, we found that all cutaneous synaptic input was driven from small overlapping receptive fields. In these cells, two different types of analyses each indicated a contribution to the evoked response from at least three or four mossy fiber synapses (i.e., probably all mossy fiber synapses available). We also found that different granule cells were driven by different types of input and that they, overall, had a specific depth distribution depending on the input they received.
All granule cells had a background of EPSP activity, which is in line with our previous observations in this preparation that all mossy fibers are spontaneously active (Garwicz et al., 1998; Ekerot and Jörntell, 2001) (Fig. 3 A). However, some granule cells did not fire any spontaneous spikes, indicating that a single EPSP is not enough to fire these cells, which was also reported by Chadderton et al. (2004). Overall, there was a correlation between the spontaneous activities of spikes and EPSPs (Table 2), but the spike activity typically was much lower than the EPSP activity, indicating that much of the mossy fiber activity is filtered out. Note that the absence of spontaneous spike activity in many granule cells activated from the skin was probably attributable to the special situation of the preparation, with the animal being completely still. We frequently noted that the cutaneous input was so sensitive that passive movement of the forelimb, without touching the receptive field, was enough to make the cell start firing (Fig. 3 B). It is likely that the high sensitivity of the receptors in the hair follicles and/or skin can make them fire even as the skin stretches during joint bending and, in that sense, mediate the function of cutaneous proprioception (cf. Edin and Johansson, 1995; Edin, 2004).
In both intracellular and EC recordings, we found that granule cell spikes were driven by one type of somatosensory input and, in the cells activated by cutaneous input, from one receptive field only and even from one submodality only (hair/non-hair). We found little evidence for convergence of mossy fibers activated by different types of input or mossy fibers activated from different receptive fields (different sites on the skin or different specific joints). This finding did not come as a surprise because there is an orderly zonal-microzonal termination of many mossy fiber tracts. Mossy fibers of the E-CCT have specific receptive fields and are distributed approximately along the longitudinal microzonal organization previously described for climbing fibers with specific cutaneous receptive fields (Ekerot and Larson, 1980; Ekerot et al., 1991; Garwicz et al., 1998). Therefore, granule cells within a microzone are mainly reached by mossy fibers with the same receptive field, and, accordingly, we found that granule cell receptive fields were similar to those previously reported for mossy fiber terminals (Garwicz et al., 1998). It should be emphasized that an absence of integration of different cutaneous inputs in granule cells is strongly supported by the fact that also the cutaneous parallel fiber receptive fields in Purkinje cells and interneurons are similar to those of single mossy fibers (Ekerot and Jörntell, 2001; Jörntell and Ekerot, 2002, 2003).
In addition to the microzonal organization described previously for mossy fiber terminals, we now found that granule cells also had a specific depth distribution depending on the type of input they received (Table 4). These observations suggest a detailed topographical organization within the three-dimensional space of the granule layer that further argues against a convergence of different types of mossy fiber inputs in granule cells.
Peripherally driven synaptic inputs
In cells with strong responses to skin stimulation or joint movements, evoked responses consisted in a near instantaneous DC shift of 10–25 mV (from −70 mV in membrane potential) that was sustained by a high-frequency activation of EPSPs. Because of the depression of the EPSP amplitudes at high-frequency activation, only in the initial phase of the response was it possible to use amplitude identification of EPSPs to analyze the contribution of individual mossy fiber afferents. For granule cells with cutaneous input, we made two different types of analyses to obtain a measure of the theoretical minimal number of EPSPs contributing to the evoked response. As shown in Figure 6, by using the average time course of different EPSPs identified in the distribution histograms, a simulation of the initial phase of the response indicated that more than three different mossy fiber afferents firing in near synchrony at nearly 1000 Hz had to participate in the response. Considering that some granule cells had only three dendrites, this analysis indicated that all available mossy fiber inputs had to participate in the response. The other type of analysis was more conservative in that it included only EPSPs discernible by inflection points in the rising phase of the response, and only if they occurred within 1 ms of each other or if consecutive EPSPs had increasing amplitudes. Still, this analysis indicated that at least three different afferents contributed to the strong responses evoked in granule cells activated via the E-CCT.
Granule cells not activated via the E-CCT were divided into three types: (1) cells strongly activated by joint movements; (2) cells weakly/moderately and/or indirectly activated by joint movements/skin stimulation; and (3) cells with no peripheral input. Because of their activation properties, the driving mossy fiber systems were suggested to be (1) the P-CCT, (2) the LRN, and (3) the parts of the pontine nuclei that normally are driven by pathways from the cerebral cortex but that were deafferented by the brainstem transection of our preparation. For cells strongly activated by joint movements, we could make another type of analysis of the rising phase of the response. The depolarization of this phase exceeded the peak amplitudes of the largest EPSPs by >50%. This was an indication that at least more than one EPSP had to contribute to the response, because the EPSP of a single afferent could not summate to reach this level of depolarization. We also observed in all cells driven by joint movements that EPSPs of all different amplitudes were driven by the same stimulation (Fig. 8). Importantly, in none of these cells did we observe any cutaneous input of the E-CCT type, even though this was routinely tested for many skin sites both manually and electrically.
Theories of cerebellar granule cell function
Our results point in the direction that granule cells function as signal-to-noise-enhancing threshold elements with a single, specific type of input. This would be a new model of granule cell function that contrasts with the prevailing models of granule cell function as pattern discriminators/sparse coders or temporal pattern generators, models that have never before been evaluated in actual granule cell recordings.
The concepts of pattern discrimination and sparse coding in granule cells have heavily influenced current models about the function of the cerebellar granule layer (Marr, 1969; Albus, 1971; Tyrrell and Willshaw, 1992; Houk et al., 1996; Schweighofer et al., 1998, 2001; Wolpert et al., 1998; Chadderton et al., 2004; Semyanov et al., 2004). The underlying assumption is that there is a divergent distribution of mossy fiber input to the granule layer so that the individual granule cell samples different types of mossy fiber input and therefore works as an associative element. Sparse coding, meaning that granule cells are silent under resting conditions and fire just a few spikes when the adequate combination of inputs activates the cell, was thought to be a consequence of this organization. Our findings disagree with the sparse coding-pattern discrimination concept in that (1) we did not find a convergence of different types of mossy fiber inputs in granule cells, (2) granule cells with some types of mossy fiber input were not silent at rest, and (3) all granule cells with strong peripheral input fired tremendously strong spike trains when activated appropriately.
A different theoretical viewpoint on the function of the granule cell stems from attempts to explain how the cerebellum might contribute to the timing of responses in classical conditioning and other behavioral paradigms. Various theories of granule cells as temporal pattern generators have been presented, most notably the variations on delay-line coding (for review, see Medina et al., 2000; Medina and Mauk, 2000; Ohyama et al., 2003). These theories were not supported in our recordings, in which granule cells were activated by repeated, uniform stimulations (Fig. 9). In addition, electrical skin stimulation always produced exactly timed EPSP responses (Fig. 5) that looked very similar in different cells. Overall, from our experience of recording from the afferent mossy fibers activated by cutaneous input and joint movement input (Garwicz et al., 1998), we believe that the activation patterns we now observed in granule cells primarily reflect the synchronized activity in presynaptic mossy fibers driven by similar input.
This work was supported by the Swedish Research Council (projects K2005-04X-14780-03A and K2006-04X-08291-19-3), the Segerfalk Foundation, the Swedish Medical Society, Crafoordska stiftelsen, and Magnus Bergwalls Stiftelse.
- Correspondence should be addressed to Dr. Henrik Jörntell, Department of Experimental Medical Sciences, Section for Neuroscience, BMC F10, Tornavägen 10, SE-221 84 Lund, Sweden.