Current source density correlates of cerebellar Golgi and Purkinje cell responses to tactile input

Abstract

The overall circuitry of the cerebellar cortex has been known for over a century, but the function of many synaptic connections remains poorly characterized in vivo. We used a one-dimensional multielectrode probe to estimate the current source density (CSD) of Crus IIa in response to perioral tactile stimuli in anesthetized rats and to correlate current sinks and sources to changes in the spike rate of corecorded Golgi and Purkinje cells. The punctate stimuli evoked two distinct early waves of excitation (at <10 and ∼20 ms) associated with current sinks in the granular layer. The second wave was putatively of corticopontine origin, and its associated sink was located higher in the granular layer than the first trigeminal sink. The distinctive patterns of granular-layer sinks correlated with the spike responses of corecorded Golgi cells. In general, Golgi cell spike responses could be linearly reconstructed from the CSD profile. A dip in simple-spike activity of coregistered Purkinje cells correlated with a current source deep in the molecular layer, probably generated by basket cell synapses, interspersed between sparse early sinks presumably generated by synapses from granule cells. The late (>30 ms) enhancement of simple-spike activity in Purkinje cells was characterized by the absence of simultaneous sinks in the granular layer and by the suppression of corecorded Golgi cell activity, pointing at inhibition of Golgi cells by Purkinje axon collaterals as a likely mechanism of late Purkinje cell excitation.

for a long time, the cerebellum has been considered to be primarily, if not exclusively, a motor organ as lesions cause symptoms such as ataxia, dysmetria, tremor, and hypotonia (Holmes 1939). Research of the past decades revealed more of its true nature: a neural substrate processing massive sensory information and serving not only motor, but also nonmotor regions (Bower 1997; Bower and Woolston 1983; Gao et al. 1996; Hartmann and Bower 2001; Jueptner and Weiller 1998; O'Reilly et al. 2008). This does not come as a surprise if one considers that the cerebellum contains more neurons than the rest of the brain and is strongly connected with the cerebral cortex (Ros et al. 2009; Rowland et al. 2010; Schmahmann and Pandya 1997). From comparisons of fiber counts in the different cerebellar pedunculi, 90% of the afferent mossy fibers have been estimated to be of neocorticopontine origin in humans (Tomasch 1969).

The afferent pulses conveyed by mossy fibers reach the output neurons of the cerebellar cortex, the inhibitory Purkinje cells (PCs), disynaptically via the granule cells, either at synapses on the ascending granule axon segment or on the terminal parallel fibers (Gundappa-Sulur et al. 1999; Walter et al. 2009). PCs respond to peripheral stimulation with a variety of sequences of enhancement and suppression of their baseline level of simple-spike (SS) activity (Bower and Woolston 1983; Holtzman et al. 2006; Jaeger and Bower 1994).

Spanning both the granular and molecular layers, the inhibitory Golgi cells (GoCs) are ideally placed to control the processing of incoming information (D'Angelo 2008). Hence, unlike PCs, GoCs receive also excitation from mossy fibers (Cesana et al. 2009; Kanichay and Silver 2008). This monosynaptic connection has been proposed to evoke the short latency (less than 5–10 ms), high-fidelity spike response to tactile peripheral stimulation (Vos et al. 1999b). However, several questions remain unresolved. The short-latency response peak of GoCs often presents as a doublet or triplet of precisely timed spikes. GoCs exhibit also a later response peak (at ∼20 ms), followed by a suppression lasting several tens of milliseconds. Morissette and Bower (1996) suggested the later response peak to be of corticopontine origin, as it disappeared after S1 inactivation and as its variable latency strongly correlated with the S1 response latency at the single trial level. However, a modeling study could not exclude the slowly propagating parallel fibers as a secondary source of excitation (Volny-Luraghi et al. 2002), and alternative pathways exist (see discussion). In addition, both the initial burst-like response and the long-lasting suppression could reflect the activation of afferent synapses or, alternatively, be generated by intrinsic membrane mechanisms (Solinas et al. 2007). Moreover, despite their stereotypical response patterns to peripheral stimulation, GoCs have larger receptive fields and a lower response selectivity than mossy fibers (Prsa et al. 2009). This loss of specificity was attributed to their presumed wide input via parallel fibers (De Schutter et al. 2000; Vos et al. 2000).

To better explain the sequence of responses in GoCs and PCs, we combined the extracellular recording of neuronal spiking activity with the localization of synaptic currents through current source density analysis (CSD) of local field potentials (LFPs). CSD analysis is a proven method for locating synaptic currents in vivo (Mitzdorf 1985; Nicholson and Freeman 1975; Pettersen et al. 2006), but after early successful attempts in the cerebellar cortex (Kwan and Murphy 1974; Nicholson and Llinás 1975), it was largely abandoned there due to methodological concerns about its ability to distinguish synaptic from passive dipole currents. Indeed, current conservation requires synaptic current entering the cell to be balanced by an outward current of equal magnitude. Although this passive leak or capacitive current is usually more diffuse than the active synaptic current, CSD analysis cannot tell them apart. A simultaneous peak or dip on the peristimulus spike time histogram (PSTH) can help locate the active sink or source of the dipole. In the present study, we isolated synaptic current sinks and sources through their correlation with corecorded PC and GoC spike responses.

MATERIALS AND METHODS

Surgical Procedures

Twenty-eight adult male Wistar rats (weight: 250–350 g) were used in this study, following procedures reported previously (Vos et al. 1999a,b). In short, in 23 animals, anesthesia was induced intraperitoneally with a mixture of ketamine hydrochloride (75 mg/kg) and xylazine hydrochloride (3.9 mg/kg). Toe-pinch reflexes were used to control the level of anesthesia and supplemental doses (of induction dose) given intramuscularly as needed. The rat's head was fixed in a stereotaxic frame (Kopf Instruments) with nonpuncturing earbars and a standard mouthpiece. Body temperature was maintained at 37–38°C using a homeothermic blanket system (Harvard Apparatus). A midline incision was made, the periost removed, and the skull cleaned. Removing the cranial muscles exposed the caudal part of the skull. A rongeur was used to remove the squamous part of the occipital bone to expose Crus I and II of the left hemicerebellum. This area was leveled by tilting the rat's head downward. After puncturing the cisterna magna to reduce cerebellar pulsation, the dura was locally incised and reflected. The exposed cerebellar surface was protected with agar (2%, 37°C).

Five of the rats were initially anesthetized with isoflurane (1.3–1.6% in O₂, 1 l/min flow rate), and in four of these rats the same stimulus protocol was repeated before and 1 h after the switch to ketamine/xylazine anesthesia.

Multielectrode Recordings

Extracellular single-unit recordings were made with a 16-channel silicon probe (15-μm thickness, 3- or 5-mm length; NeuroNexus Technologies). The recording sites (13.3 × 13.3 μm) were evenly spaced along the probe at distances of 50 μm, and their impedances were stable at 1.5 MΩ. The probe had a tapering width from 123 μm near the top of the shank to 33 μm at the tip electrode contact and was positioned parallel to the parallel fibers and perpendicularly onto the surface of Crus IIa with a Kopf Instruments Micropositioner.

All 16 channels were simultaneously processed (gain: 4,000–10,000; filtering: 400-Hz-to-20-kHz band-pass), digitized, and discriminated by a PC-controlled Multichannel Acquisition Processor (MNAP; Plexon; 25-kHz sampling rate). Spike waveforms were discriminated using a real-time time/voltage window. Signals were also made audible to improve spike classification. Recordings of waveforms and spike trains were stored for further off-line analysis.

LFPs were obtained through band-pass filtering (0.7–170 Hz) and sampled at 1 kHz. However, we took advantage of a systematic submillisecond drift of the clock signal for stimulus delivery with respect to the phase of the LFP sampling signal, which allowed us to calculate compound LFPs after many stimulus repetitions with an effective sampling rate of 25 kHz (see below).

Identification of GoCs and PCs

The probes were advanced or repenetrated until single units could be discriminated on 1 or more of the channels. Along the probe, we sampled 17 pairs of PCs and GoCs, 3 pairs of GoCs, a single GoC (n = 4), or a single PC (n = 5) under ketamine/xylazine anesthesia. A pair of GoCs and 3 single GoCs were sampled under isoflurane anesthesia.

Units were identified and classified as GoCs (n = 32) or PCs (n = 22) based on generally accepted parameters for in vivo recordings in ketamine/xylazine-anesthetized rats (Simpson et al. 2005; Vos et al. 1999a,b). PCs were identified by the simultaneous presence of SSs and complex spikes (CSs). The average SS firing rate was 38 spikes/s (SD: 19) under ketamine/xylazine anesthesia with a median interspike interval (ISI) of 18 ms (SD: 7; Fig. 1A). Single-unit quality of the PC recordings was assured by the absence of ISIs shorter than the PC refractory period (2 ms). In some cases, the CS of the PC was lost after retuning of the probe depth to isolate a GoC at another channel. The waveform of CSs is generated more dendritically than that of the somatic SSs, and the simultaneous detection of CSs and SSs is only possible over a small tuning distance. However, we never classified a unit as a PC if no CSs were present before retuning the probe depth. This, together with the stability of the distribution of the SS ISIs and the characteristic sound of SSs on the audio monitor, assured us to be recording from the same PC as the one initially isolated.

Fig. 1.

Electrophysiological identification of Golgi cells (GoCs) and Purkinje cells (PCs) on a 1-dimensional (1D) silicon probe. A: median interspike interval (ISI) distribution of the recorded GoCs (black) and PCs (gray) in 10-ms bins. The inset shows the median ISI of different units as they were recorded at different positions along the silicon probe shank (y-axis; 1 = electrode tip; 16 = base). B: ISI distribution (1-ms bins) of PC simple spikes (SSs; top; recording k-080408) and a GoC (bottom; recording k-190308). The insets show superimposed spikes of the respective units (horizontal scale bar = 0.1 ms). CS, complex spike.

All GoCs were recorded in the granular layer as judged from the typical background noise. GoCs are known to fire spontaneously in a typical syncopated cadence and to have a rather broad ISI distribution lacking high-frequency components (Simpson et al. 2005; Vos et al. 1999b). The identification of these units as GoCs has recently been confirmed by juxtacellular labeling in a study by Holtzman et al. (2006). Note that Lugaro cells, located in the upper granular layer, respond to peripheral stimuli with prolonged (>100 ms) increases in spike rate (Van Welie and Haüsser 2009), a response pattern that was never observed in our GoCs. The spontaneous frequency recorded under ketamine/xylazine anesthesia in our sample of GoCs ranged from 2 to 26 spikes/s with an average of 7.5 (SD: 5.5) and a median ISI of 99 ms (SD: 43; Fig. 1A). The insets to Fig. 1B show extracellularly recorded spike waveforms of a GoC and a PC.

As an additional in vivo control of the recording quality of the 16 electrodes along the single-shank probes, we sought for systematic differences between the electrodes in their capacity of capturing spikes (see inset to Fig. 1A). With electrode number as the grouping variable, we did not find significant differences in the average spontaneous firing frequency or the median ISI (ANOVA on ranks).

Peripheral Stimulation

We used a custom-built speaker probe (Fig. 2) to deliver tactile stimuli to the perioral area (upper lip, lower lip, and furry buccal pad), vibrissae, or forepaw. (Note that forepaw responses were indistinguishable from facial responses, apart from an increase in latency of 2.3 ms on average, and that we denote early response waves invariably as trigeminal, although forepaw responses obviously are spinal in origin.) The probe, a 25-gauge rod glued inside a hypodermic needle (25-gauge), was connected to a 2.25-in. speaker cone that was positioned with a magnetic stand. A diode prevented speaker oscillations when driven with a rectangular 2-ms pulse. Onset delay (to 50% of probe displacement) was 0.6 ms, rise time 0.4 ms (10–90% of probe displacement), and probe pulse width 1 ms (at 50% probe displacement). Maximal probe displacement was 0.5 mm. Jitter was negligible (<2%), and the time course of probe deflection had a high fidelity and reproducibility (Fig. 2B), as measured with a piezo film sensor [LDT1-028k; Measurement Specialties (MEAS)]. The stimulus protocol was under LabVIEW control (National Instruments).

Fig. 2.

Stimulation procedures. A: experimental setup: 2-ms tactile stimuli were delivered to the rat's snout or forepaw with a speaker-driven stimulator. Single (“a”) or doublet (“b”; 40- to 50-ms interstimulus interval) stimuli were presented at a repetition rate of 2 Hz. Extracellular signals were registered and processed with a Multichannel Acquisition Processor (MNAP). B: detail of the speaker-driven stimulus, measured with a piezo film sensor. The almost complete overlap of the curves for the mean and the mean ± SE demonstrates the reproducibility of the probe deflection (n = 225).

Stimuli were delivered in 400 successive trials at a repetition rate of ∼2 Hz. In 3 rats, a series of double-pulse experiments was performed with an interstimulus interval of 40 or 50 ms, repeated at 2 Hz for 200 trials (400 stimuli in total; Fig. 2A). As noted above, although single-trial LFPs were sampled at 1 kHz, the submillisecond drift of the stimulation clock allowed us to construct compound LFP and CSD diagrams with an effective resolution of 40 μs after averaging over all 200 or 400 trials. This technique is similar to the use of staggering stimulus times in fMRI experiments to enhance the effective sampling rate (Henson et al. 1999).

Histology

Rats were euthanized with a lethal dose of sodium pentobarbital (120 mg/kg ip). Because the currents needed to make electrolytic lesions in cerebellar cortex (0.4 mA, 10 s) were damaging to the electrodes, compromising their reuse, the recording positions were histologically verified in fewer than half of the cases (8 of 22 rats), using cryostat coronal sections (40 μm) stained with cresyl violet (Fig. 5A). Note, however, that the coregistered PCs and the typical reversal in polarity of the LFPs (see results) allowed us to precisely localize the border between the molecular and granular layer in all experiments. The thickness of the molecular layer, measured during recordings as the distance from surface penetration to the detection of PC SS activity, was ∼300 μm.

Fig. 5.

CSD and PSTHs of a corecorded GoC and PC after doublet stimulation of the upper lip (recording k-050808; ketamine/xylazine anesthesia). A: histological reconstruction of the probe position in the cerebellar cortex (scale bar = 50 μm). B: CSD response profile with depths of corecorded PC and GoC indicated by dashed (PC) and dash-dotted (GoC) lines. As in Fig. 4, upward arrows indicate sinks in the GL associated with the trigeminal and putatively corticopontine waves of excitation. Note that the putatively corticopontine sink (at 20 ms) is located closer to the PL and absent after the 2nd stimulus of the pair. The evoked sinks high in the ML were less prominent than in Fig. 4. Many of the GL sinks had complementary sources in the ML, but a strong source deep in the ML (open arrowhead) coincided with the short-latency inhibition on the PC PSTH (C; open arrowhead). C: corresponding PSTHs (binwidth: 1 ms) of a paired GoC-PC recording. The PC SS PSTH shows a short-latency excitatory peak (SLE) followed by short-latency inhibition (SLI; open arrowhead), whereas the CS PSTH does not show an evoked response. The GoC PSTH shows a sharp trigeminal and weaker putatively corticopontine peak, the latter almost completely absent following the 2nd stimulus of the doublet.

Ethical Considerations

All efforts were taken to reduce animal discomfort and to minimize the number of animals used in this study. All procedures were in accordance with federal laws and approved by the ethical committee for animal experimentation of the University of Antwerp.

Data Analysis

PSTHs for GoCs and PCs were constructed in 1-ms bins, and further statistical analysis was performed with a 5-ms binwidth for the GoC silent period and the PC long-lasting excitation. Excitatory responses or, conversely, a reduction in firing rate were considered significant if the PSTH counts rose above, or fell below, the 95% confidence interval (CI) of the mean firing rate, which was calculated over all 100-ms prestimulus intervals. The excitatory response latency was measured at the response peak unless otherwise stated. In the case of a reduction in firing rate, the first bin of this interval was taken as response latency.

The one-dimensional (1D) depth profile of the CSD was estimated as the negative of the second-order difference of the LFP with respect to depth, multiplied by the tissue conductivity (Mitzdorf 1985). (The boundary traces at electrodes 1 and 16 were either neglected when their exact position was uncertain or duplicated to obtain 16 CSD traces.) This sign convention makes current sinks positive (corresponding to excitatory synaptic currents; yellow to red in diagrams) and current sources negative (inhibitory currents; blue to violet). By averaging the CSD diagram over 400 trials, the effective resolution was increased to 40 μs, as noted above. We further assumed cerebellar cortex to have a homogeneous tissue conductivity of 0.23 S/m without a capacitive component (Okada et al. 1994). We also applied other formulas involving spatial smoothing (Freeman and Nicholson 1975) or expansion of the CSD as a truncated series of principal components (Di et al. 1990) but found too much detail was lost.

Current sinks and sources were considered significant if they crossed the threshold of background activity (calculated in the same manner as for the PSTH). Latencies of CSD responses of interest were calculated with respect to the peak amplitudes. CSD profiles were visualized in MATLAB (MathWorks) using bilinear interpolation.

Except when stated, the statistics in this paper are expressed as means and SD.

Linear Reconstruction of PSTHs from the CSD Depth Profile

To examine how well the synaptic currents, as reflected by the CSD profile, determined the instantaneous spike rate of simultaneously recorded GoCs and PCs, we applied optimization techniques to reproduce each PSTH as a weighted sum of the 16 CSD traces. The CSD profile was averaged in 1-ms bins, each trace normalized to represent a vector of unit length, and an occasional low-quality signal from a defective electrode replaced by Gaussian noise. A 1-ms shift of both signals was allowed. Both unconstrained and constrained optimization techniques were applied.

Principal component analysis.

The 16 CSD traces were regarded as basis vectors in an n-dimensional space, n being the number of time points. The reconstruction of the PSTH amounts to calculating the coordinates of the PSTH in that space. This can be simplified when the basis vectors are orthonormal, in which case the coordinates are the inner products of the PSTH with each basis vector. We used principal component analysis to form such an orthonormal basis and retransformed the coordinates to weights of the original CSD traces.

Linear programming.

In a 2nd approach, we used the MATLAB lsqlin function to constrain all weights to be positive. Here, the weights were optimized by minimizing the error on a system of 41 equations in 16 variables.

The quality of reconstruction was expressed as the correlation coefficient between the actual and the reconstructed PSTH. This statistic offers the advantage of being independent of the mean level of activity, as the reconstructed PSTHs could have bins with negative counts. For a better visual comparison with the actual PSTHs in Figs. 10 and 11, the reconstructed PSTHs were incremented and negative counts thresholded. The quality of reconstruction was mostly indistinguishable for both procedures, and the weight vectors had similar profiles. Both the LFP and CSD signals could be used as basis for the reconstruction, the former usually being numerically more stable.

Fig. 10.

Evoked GoC responses can be reconstructed from the CSD profile. Reconstruction of the PSTHs of a GoC (recorded at 5th electrode position) through a linear combination of the CSD traces (recording k-160108; ketamine/xylazine anesthesia). The recorded (black bars) and reconstructed (white bars) PSTHs are compared for 5 stimulus conditions (LL, lower lip stimulation; UL1–4, 4 positions of upper lip stimulation) and 2 reconstruction methods (left vs. right column). In the left column, an optimized weight vector was calculated with principal component analysis (PCA) for each stimulus condition separately. The resulting 5 weight vectors are overlaid in the left bottom panel. In the right column, a single weight vector (bottom panel) was simultaneously calculated for all stimulus conditions (by applying PCA to a concatenated trace of PSTHs and CSDs). Each panel shows fit of reconstruction expressed as a correlation coefficient. Vertical scale bars: 50 counts per millisecond (collected over 400 trials in 1-ms bins).

Fig. 11.

Recorded PSTHs (black columns) and PSTHs reconstructed from the CSD profiles (white columns) for the 4 PCs illustrated in the previous figures. Vertical scale bars: 25 spikes, collected over 400 trials in bins of 1-ms width. Each panel shows fit of reconstruction expressed as a correlation coefficient.

To assess its significance, the correlation coefficient was compared with those obtained for 50 reconstructions of the respective PSTH from noise and expressed as a z-score (number of SD above the mean).

RESULTS

We used single-shank silicon probes to simultaneously register in Crus IIa, LFPs, and GoC and/or PC spikes evoked by tactile stimulation of perioral and paw receptive fields in anesthetized rats. Estimation of the CSD allowed us to correlate the synaptic activation patterns in the granular and molecular layers with the PSTH response profiles of GoCs and PCs.

We first assess the error resulting from a 1D CSD analysis and thereafter describe the hallmarks of the LFP and CSD depth profiles through a few case descriptions. These cases should not be considered as typical of certain classes, but together they illustrate all response features used in further analysis. Thereafter, we present a statistical analysis of the CSD profiles and the spike responses of GoCs and PCs and of the correlations between both types of signals. Finally, we show that in particular the GoC spike response could be almost completely reconstructed from the CSD traces.

Validity of the 1D Derivation of Current Sources and Sinks

A full 3D calculation of the current density (I) requires the 2nd derivative of the potential (φ) be calculated along 3 orthogonal axes,

$$-I={\sigma }_x\frac{{\partial }^2\varphi }{\partial x^2}+{\sigma }_y\frac{{\partial }^2\varphi }{\partial y^2}+{\sigma }_z\frac{{\partial }^2\varphi }{\partial z^2},$$ (1)

where x, y, and z can conveniently be assumed to represent the dimensions perpendicular to the cortical surface (x), along the parallel fiber axis (y), and along the sagittal axis (z), respectively. As stated above, we took σ_x = σ_y = σ_z = 0.23 S/m. The 1D reconstruction envisaged in the present study measures only the 1st term. We compared this approximation with the 2D derivation (sum of the x- and y-terms) from recordings on a multishaft probe aligned along the parallel-fiber axis (the direction along which current is most likely to spread). The probe had 4 shafts 125 μm apart, each carrying 4 channels at 100-μm spacing. We applied CSD analysis to the central 4 of the 16 channels at 2 depths (encompassing the molecular and granular layers) and 2 positions of the probe, yielding 16 data points in total. Figure 3 shows that the 1D reconstructions were within 32% deviation of the 2D reconstructions for all but 3 data points. Given the large distance between the shafts, these deviations may be an overestimate. Note that the sign never reversed, hence sources and sinks kept their polarity. When the magnitudes (absolute values) of the x- and y-terms were compared, the 1st term (depth) produced 83% of their sum. Extending this to all 3 dimensions, and assuming equal magnitudes for the y- and z-terms, would still result in a 73% contribution of the depth axis to the potential variation, justifying a 1D derivation as a 1st approximation.

Fig. 3.

Control current calculations on a 2D probe. Comparison of 1D and 2D calculations of the current source density (CSD) from local field potentials (LFPs) recorded on a 2D probe comprising 4 shafts of 4 electrodes each. The probe was aligned along the parallel-fiber axis (experiment k-110810), and the recordings were repeated at 2 depths (molecular vs. granular layer) and at 2 positions of the penetration. Stimulation of the upper lip evoked LFPs of similar waveform and amplitude on all 4 shafts, and the CSD was calculated for the central 4 channels. Each data point represents the average CSD during 20 ms after stimulation, calculated using either a 1D or 2D derivation. Positive and negative values represent sinks and sources, respectively. For the 3 pairs of data points in open symbols, the 1D deviation from the 2D reconstruction was >32%. The other 13 data pairs produced an average 1D error of 12%.

Punctate Peripheral Stimulation Evokes Two Waves of Excitation in Cerebellar Cortex

Confirming and extending previous results (Morissette and Bower 1996; Vos et al. 1999b), the LFP and CSD diagrams of all recordings under ketamine/xylazine anesthesia (n = 65 stimulus conditions for 29 electrode positions in 23 rats) showed 2 waves of excitation in the granular layer spreading across the cortex: a 1st wave arriving within 10 ms after stimulation and a 2nd wave after ∼20 ms.

These two waves are clearly visible in the LFP diagram of Fig. 4A as strong negativities in the granular layer (upward arrows at 8.3 and 17.8 ms), immediately followed in each case by a weaker negativity in the molecular layer (downward arrows at 9.8 and 18.6 ms). In each wave, the granular-layer negativity (upward arrow) corresponds to the N1/N2 peak previously reported in vivo (Eccles et al. 1967) and in vitro (Maffei et al. 2002) and attributed to pre- and postsynaptic currents at synapses from mossy fibers to granule and Golgi cells. The molecular-layer negativity (downward arrow) corresponds to the N3 peak attributed to synaptic currents evoked by granule cell synapses on ascending axons or parallel fibers (Armstrong and Drew 1980; Bengtsson and Jörntell 2007). In accordance with this, the CSD diagram (Fig. 4, B and D) showed strong current sinks at these time instants in the granular and molecular layer, respectively. Note also that the LFP reversed sign across the cortex, at about the border between the granular and molecular layers (Fig. 4, A and B), proving that the LFPs were generated intracortically and not spread from an external source by passive volume conduction. Typically, in all recordings, the signals relaxed after ∼30 ms and then reversed sign, the potential in the granular layer now turning positive and the molecular layer negative.

Fig. 4.

Two waves of excitation can be distinguished by their different refractoriness to doublet (paired-pulse) stimulation. LFPs, CSD, and PC peristimulus spike time histograms (PSTHs) calculated from 200 trials of paired-pulse (A–D) or 400 trials of single-pulse (E) stimulation of the upper lip (recording k-230708; ketamine/xylazine anesthesia). Here and in the following figures, red dotted lines highlight stimulation time. A: averaged LFP traces show 2 waves of excitation apparent as negativities in the granular layer (GL; upward arrows) and spreading to the molecular layer (ML; downward arrows). After the 2nd stimulus of the pair, only the 1st wave occurs. The LFP reverses sign across the cortex, near the PC layer (PL). Time scale as in C. B: CSD response profile with overlying LFP traces recorded along the probe. Schematic of probe on the left shows electrode positions in micrometers relative to the probe base. Putative layers of the cerebellar cortex are indicated on the right (WM: white matter). White dashed line indicates the recording position of a simultaneously recorded PC. The asterisk indicates a strong CS-associated current sink in the ML. C: PSTHs of the PC response for SSs and CSs (binwidth: 1 ms). The SS response shows a short-latency excitation at 7 ms (SLEa) and 20 ms (SLEb) and a decrease in activity at the time CSs occur (CS peak latency: 32 ms). D: blow-up of the CSD response profile shown in B. Upward and downward arrows as in A. Vertical scale bar = 100 μm, time scale as in E. E: graphs of the CSD at electrode positions 250 μm (ML) and 550 μm (GL), after sorting all trials of a single-pulse experiment according to the occurrence of CSs in the PC response (solid line: CS trials; dashed line: trials without a CS). As explained in results, CS trials were characterized by an enhanced strength of the putatively corticopontine wave at 20 ms. The peak at 35 ms coincides with the CS-associated current sink in B and D. [Vertical scale bar = (0.23 S/m * 0.1 mV)/(50 μm)² or 9,200 A/m³].

The short latency of the first wave of excitation indicates a mono- or oligosynaptic trigeminocerebellar input. In contrast, the second wave had arguably a multisynaptic, neocorticopontine component. Indeed, paired stimulation with a 40-ms interval failed to evoke the second wave of excitation after the second stimulus (Fig. 4, A, B, and D), in accordance with the lack of activation observed in somatosensory cortex under similar stimulus conditions (Jellema et al. 2004; Morissette and Bower 1996). The neocortical origin is corroborated in this double-pulse experiment by the coregistered CSs, for which timing is modulated via an indirect neocorticoolivary pathway in Crus IIa of cerebellar cortex (Brown and Bower 2002; see discussion). The CSs evoked by the first pulse of the pair (see PSTH in Fig. 4C) were associated with a strong sink in the molecular layer (asterisk in Fig. 4B), and, as expected (Atkins and Apps 1997), the second pulse of the pair failed to evoke both CSs and the corresponding sink. In the single-pulse experiment (Fig. 4E), however, sorting of the trials according to the presence of a CS revealed a close relationship between the strength of the second wave of excitation, as expressed by the strength of its sink in the granular layer, and the strength of the CS-associated sink in the molecular layer. More precisely, the second sink (at 20 ms) was much stronger during trials evoking a CS, suggesting that both the second wave and the occurrence of a CS depended in the same manner on neocortical activation.

Figures 5, 6, and 8 illustrate three other cases showing additional response features, described in the next sections, such as an ascending pattern of current sinks in the granular layer, current sources associated with the GoC silent period, and sources and sinks in the molecular layer associated with PC spike responses.

Fig. 6.

CSD (A) and PSTHs of a corecorded GoC and PC (B) after single-pulse stimulation of the lower lip (recording k-110608; ketamine/xylazine anesthesia). Same format as in Fig. 5, except that the PC CS signal was lost after tuning probe depth to better isolate the GoC. As before, upward arrows indicate excitatory response components associated with the trigeminal and putatively corticopontine waves; the open arrowhead is associated with the SLI of the PC. Note in the GL the ascending current sink coincident with a multipeaked GoC PSTH during the 1st (trigeminal) wave and the long-lasting source during the GoC silent period.

Fig. 8.

Late-response components on the CSDs (A) and PSTHs of a corecorded GoC and PC (B). Single-pulse stimulation of the vibrissal pad (recording k-190308; ketamine/xylazine anesthesia). A: CSD response profile (spatial smoothing applied, binwidth: 5 ms) with depths of the corecorded PC and GoC indicated by dashed (PC) and dash-dotted (GoC) lines. B: PSTHs with 5-ms binwidth. Note the GoC silent period (open arrowhead) coincident with the PC LLE and the current source near the GoC recording position.

Simultaneous Recordings of LFPs and GoC Spikes Reveal a Stratified Input Pattern to GoCs

The GoC spike responses were comparable with those of previous reports with several clearly distinguishable short-latency response components, including the trigeminal/spinal (4–13 ms) and putatively corticopontine (14–29 ms) excitatory peaks, followed by a long-lasting inhibition (Figs. 5C, 6B, and 8; Holtzman et al. 2006; Vos et al. 1999b).

The current sinks in the granular layer were strongly correlated with the GoC excitatory response components (Fig. 5, B and C, and Fig. 6, A and B). We found a high correlation for response latency (peaks on GoC PSTH vs. current sink components in the granular layer; r = 0.97, P < 0.001; Fig. 7A) as well as for response amplitude (r = 0.42, P < 0.001). The trigeminal and putatively neocorticopontine current sinks in the granular layer had a stratified organization, the trigeminal sinks being located deeper in the granular layer compared with the putatively corticopontine sinks (P < 0.001, ANOVA on ranks; Fig. 7B). More precisely, with the GoC recording electrodes aligned to zero position, putatively corticopontine current sinks were located +43 μm (SD: 66) further toward the molecular layer, whereas the trigeminal sinks were located −45 μm (SD: 93) closer toward the white matter. Often, during the trigeminal wave itself, an ascending pattern of CSD current sinks could be seen to coincide with distinct peaks on the GoC PSTH, as shown in Fig. 6, A and B.

Fig. 7.

Population data showing the relationships between the current sinks and sources on the CSD profile and the spike responses of corecorded GoCs and PCs under ketamine/xylazine anesthesia. See results for statistics. A: relationship between the latency of current sinks in the GL and the latency of peaks on the GoC PSTH. Both trigeminal (early) and putatively corticopontine (late) components are correlated. B: relationship between the latency of sinks in the GL and their location relative to the position of a simultaneously recorded GoC. Zero position corresponds to the depth of the electrode recording the GoC spikes. Positive and negative values indicate locations closer to the ML and WM, respectively. C: average CSD profile from all recordings in which the PC showed short-latency inhibition on its PSTH (n = 30). CSD profiles were aligned at the PC recording depth (0 on vertical axis) and at the onset of the PC SLI (0 on horizontal axis). Note the strong current source (blue) coincident with the SLI. D: relationship between the duration of the GoC silent period as measured from the PSTH (GoC sil. per., vertical axis) and the duration of the concomitant current source in the GL on the CSD diagram (CSD sil. per., horizontal axis). E: paired recordings of GoCs and PCs show a correlation between the length of the GoC silent period and the length of the long-latency excitatory response (LLE). F: the same dataset shows a positive correlation between the latency of the GoC early (trigeminal) excitatory response peak and the latency of the SLEa and SLI components on the PC PSTH. (Linear fits shown in D–F.)

The GoC silent period is a known response component on tactile and air-puff stimulation (Holtzman et al. 2006; Tahon et al. 2005; Vos et al. 1999b). In our data, its onset latency measured 34 ms (SD: 12), and the spike rate was significantly decreased for a duration up to 87 ms (SD: 30). On the CSD diagrams, its counterpart in the granular layer was a current source (Figs. 5B and 6A) at a depth not significantly different from the GoC recording position (+18 μm; SD: 61). A significant correlation was found between the durations of the silent period and the concomitant source (r = 0.78, P < 0.001; Fig. 7D).

Double-pulse experiments (n = 3 in 2 rats with coregistered GoCs) with interstimulus intervals of 40–50 ms were performed to further investigate the origin of the second excitatory peak on the GoC PSTH, centered at ∼20 ms in Figs. 5C and 6B. In accordance with the absence of a corticopontine sink on the CSD, as mentioned above (Fig. 4B), the latter peak on the GoC PSTH was reduced by 91% (SD: 14) after the second pulse. In sharp contrast, the first (trigeminal) response peak was only reduced by 49% (SD: 21), this reduction reflecting the weaker excitability of GoCs during their silent period (Kanichay and Silver 2008; Tahon et al. 2005).

Simultaneous Recordings of LFPs and PCs Reveal Active Sources and Sinks in the Molecular Layer

PC SS responses (n = 51; ketamine/xylazine anesthesia) showed four different response components (Bower and Woolston 1983). Short-latency short-duration excitation (SLE; Fig. 4C) was seen in 55% of all recordings with a mean latency of 7.0 ms (SD: 1.7). A short-latency short-duration inhibition (SLI; Figs. 4C, 5B, and 6B) was present in 69% of cases with a latency of 11.4 ms (SD: 2.4) and lasting 2.8 ms (SD: 2.1). With a latency of 20.5 ms (SD: 2.4), we observed a second excitatory response (longer-latency, short-duration excitation or SLEb; Fig. 4C) in 51% of cases. Finally, a long-latency, long-duration excitatory response (LLE) was present in 78% of recordings with a latency of 42.2 ms (SD: 11.4) and lasting up to 143.6 ms after stimulation (SD: 74.7).

Correlating the PC SS response components with sinks and sources on the CSD profile of the molecular layer is complicated by the presence of passive current sources putatively generated at GoC dendrites and complementing their active granular-layer sinks. Nonetheless, the SLEa and SLEb were associated with full-blown current sinks in 7 out of 24 recordings in which we coregistered PCs with the entire depth of the molecular layer (see for instance Fig. 4, B and C). There was no significant CSD correlate of the LLE. The most conspicuous CSD signature of early PC responses, however, was a source located deep in the molecular layer and coincident with the SLI, present in 30 of 43 recordings. This source was for example particularly conspicuous in recording k-110608 (Fig. 6A), where it was so strong as to be accompanied by a sink higher in the molecular layer.

Figure 7C highlights this current source on the averaged CSD diagram from 30 recordings, aligned by SLI onset time and PC recording position. The timing of the source (at most a few milliseconds after the peak of the 1st wave), its coincidence with the dip in the SS PSTH, and its location deep in the molecular layer suggest that it is caused by inhibitory currents at synapses from basket cells to PCs.

Clear evoked PC CS responses were present in 5 out of 12 recordings in which we coregistered PC CSs with the entire depth of the molecular layer (see for instance Fig. 4C). Their PSTH peak had a latency of 34.4 ms (SD: 2.5). As illustrated in Fig. 4, evoked CSs were associated with a sink in the molecular layer at similar peak latencies in 4 out of those 5 recordings. Such a current sink was not observed when no CS response was found. Sinks associated with spontaneous CSs could also be distinguished from those of SSs by their location higher in the molecular layer (data not shown).

Paired Recordings of PCs and GoCs Show Strongly Correlated Response Components

Under ketamine/xylazine anesthesia, 17 pairs of PCs and GoCs were simultaneously recorded under 43 stimulus conditions with an average distance of 154 μm (SD: 89) along the probe. In this sample of paired recordings, the statistics of the PSTH response components were not different from those described in previous sections. An early excitatory (trigeminal) GoC response to tactile stimulation of perioral receptive fields was present in 93% of all cases, with a mean latency of 5.0 ms (SD: 1.4). PCs showed an early excitatory response (SLE; onset: 7.4 ms; SD: 1.8) in 47% of cases. A brief but significant decrease in SS activity (SLI; onset: 11.3 ms; SD: 2.4) was seen in 70% of cases. Pairwise correlations (Pearson) of onset latencies were highly significant (P < 0.001) between the trigeminal excitatory response on the GoC PSTH and both the SLEa (n = 19, r = 0.74) and the SLI (n = 28, r = 0.58) on the PC PSTH (Fig. 7F). A significant correlation (P = 0.025) could also be found within PCs between their SLEa and SLI. In all cases, the GoC trigeminal excitatory response preceded the PC SLEa with a latency of 2.6 ms (SD: 1.2) and the PC SLI with a latency of 6.2 ms (SD: 2.0). There was no significant correlation between the GoC response amplitude and the duration of the PC SLI.

In our dataset of paired recordings, the GoC silent period was present in 82% of cases with an onset time of 35 ms (SD: 11) and a duration of 90 ms (SD: 30). A long-lasting increase in PC SS output (LLE) on peripheral stimulation was seen in 91% of cases with an onset time of 35 ms (SD: 12) and a duration of 229 ms (SD: 77). To address the influence of the GoC silent period on the PC LLE, we compared all paired cases and found a significant positive correlation between the duration of both response components (r = 0.37, P = 0.01; Fig. 7E; see also Holtzman et al. 2006) that was disrupted by shuffling both datasets.

Figure 8 illustrates the spike response and the CSD profile (Fig. 8A) for the PC recording with the strongest LLE. Typically, the SS rate gradually increased over the interval coinciding with the silent period of the corecorded GoC. During this rising phase, the PC could fire in a very rhythmic pattern, as shown by the regular sequence of peaks on the PC PSTH (Fig. 8B). There were no signs of current sinks in the granular layer during the LLE (Fig. 8A). In contrast, the granular layer showed inhibitory sources at the GoC recording sites, consistent with the activation of inhibitory synapses on the soma or basal dendrites of GoCs.

Recordings Under Isoflurane Anesthesia Confirm the Different Origins of the Two Waves of Excitation

Holtzman et al. (2006) observed that GoC responses were largely invariant under different anesthetics, like pentobarbitone, α-chloralose, urethane, or ketamine/xylazine. Isoflurane, however, is a volatile anesthetic known to suppress thalamocortical neurons of the ventral posteromedial nucleus (Detsch et al. 1999) by potentiating their extrasynaptic GABA_A receptors (Jia et al. 2008). Under isoflurane anesthesia, none of the recorded GoCs (n = 5) showed a putatively corticopontine excitatory peak, but in all cases a trigeminal response component (3–12 ms) was present. In four experiments, including recordings from two GoCs, the CSD depth profiles could be compared before and after a switch from isoflurane to ketamine/xylazine anesthesia (wash out of 1 h). In all four cases, putative pontine current sinks were absent or weakly present under isoflurane anesthesia but became fully apparent under ketamine/xylazine anesthesia (Fig. 9). Apart from this, CSD profiles in the molecular layer were similar for the two types of anesthesia, indicating little differential effects on synaptic transmission.

Fig. 9.

Switching from isoflurane (left) to ketamine/xylazine anesthesia (right) distinguishes the 2 waves of excitation. CSDs and GoC PSTHs averaged over 400 trials of single-pulse stimulation of the lower lip (recording k-050810). White line indicates electrode position of the simultaneously recorded GoC. Isoflurane blocks the 2nd wave of excitation that is present at 15–22 ms during ketamine/xylazine anesthesia (shaded areas). Vertical scale bar = 100 μm.

GoC PSTHs Can Be Reconstructed from the LFP and CSD Profiles

The strong correlation described above between the spike responses of GoCs and their CSD correlates in the granular layer raises the question how well the PSTHs could be reconstructed from a linear combination of the CSD traces. For GoCs, it was always possible to find linear reconstructions that explained >90% of the variance of the PSTH in the 40-ms poststimulus interval (mean correlation coefficient: 0.94; SD: 0.02; range: 0.90–0.98; n = 29; see for instance Fig. 10, left column). Reconstructing each PSTH from noise yielded an average correlation of 0.61 (average SD: 0.08). From this analysis, the correlation coefficient of reconstructed GoC PSTHs was highly significant with an average z-score of 4.1 (SD: 0.5; range: 3.1–5.3).

GoCs have large receptive fields, and their PSTHs evoked at different stimulus locations typically vary in the strength and exact timing of the trigeminal and putatively corticopontine response components (Vos et al. 2000). These changes were associated with changes in the CSD profile in the present study. As a consequence, a single set of weights could be found for the present case that reconstructed all PSTHs with a correlation coefficient >0.85 (Fig. 10, right column; average z-score: 3.9, range: 3.5–4.1).

Reconstructing the PSTHs was always more difficult for PCs than for corecorded GoCs, typically explaining only 80% of their variance (mean correlation coefficient: 0.78; SD: 0.1; range: 0.59–0.94; n = 21; see Fig. 11). Reconstructing each PSTH from noise yielded an average correlation of 0.58 (average SD: 0.1) and an average z-score of 2.1 (SD: 1.2; range: 0.2–4.8). This greater difficulty in reconstructing the evoked responses of PCs can be attributed both to a weaker modulation of their spike rate after stimulation and to fluctuations in their background firing caused by intrinsic activity, invisible on the CSD diagram.

DISCUSSION

The Sequence of Synaptic Activations Probed with a 1D Electrode Array

We used CSD analysis along a 1D probe to localize the effective synaptic inputs to cerebellar GoCs and PCs after peripheral stimulation. Previous studies were unable to localize the synapses activated during the second wave of excitation at ∼20 ms after stimulation.

The results presented here are unequivocal: in all cases was the second response component associated with a current sink in the granular layer, presumably activated via the neocorticopontine pathway (Leergaard et al. 2000). The observed close correlation between these sinks and the spike responses of corecorded GoCs suggests that mossy fiber to GoC connections contributed to the sinks. This is consistent with the large size of GoCs and the vertical orientation of their dendrites as well as with the recent in vitro characterization of currents from mossy-fiber synapses (Kanichay and Silver 2008) and the retrograde tracing of some of the afferents to the pontine nuclei (Holtzman et al. 2009). Although granule cells outnumber GoCs by several orders of magnitude, they may contribute less than expected to the CSD signal because the ratio of synapses within a mossy-fiber glomerulus is inevitably much smaller, and the fields along their nonoriented dendrites are expected to undergo less positive interference.

Note that although the presence of a granular layer sink cannot prove the mossy-fiber origin of the second response peak, its absence would have forced us to consider the slowly propagating parallel fibers as the putative origin of the response delay.

Additionally, we found that trigeminal mossy fiber input was systematically located deeper in the granular layer than putatively corticopontine input evoked by the same stimulus. This finding adds a vertical stratification to the original finding of overlapping trigeminal and pontine receptive fields in the granular layer (Bower and Woolston 1983). A vertical organization of mossy-fiber input to the granular layer has recently been reported in the decerebrate cat (Jörntell and Ekerot 2006). Whereas in the latter study this resulted in modality-specific responses of granule cells, the present study and previous ones (Holtzman et al. 2006; Prsa et al. 2009; Vos et al. 2000) indicate that GoCs integrate synaptic input from multiple receptive fields. A layered projection pattern could result from the layered distribution of attractants or repellants (Solowska et al. 2002). Alternatively, corticopontine mossy fibers may excite GoCs disynaptically via the granule cells, for which ascending axons have been shown to make synapses on GoCs within the granular layer (Cesana et al. 2009). We cannot rule out, however, that a putatively pontine sink deeper in the granular layer was occluded by the early onset of inhibition of GoCs by PC collaterals. In the double-pulse experiment of Fig. 5, for instance, the corresponding source already set in before the expected timing of the (absent) pontine peak.

The origin of the CSD pattern in the molecular layer was more difficult to define, apart from a deep source coincident with the suppression of SSs. Clear signs of activation of granule cell synapses were sought based on their location, timing relative to granular-layer sinks, and association with PC responses. Although deflections on the LFP and CSD traces corresponding to the N3 negativity (Armstrong and Drew 1980; Eccles et al. 1967) were always observed (see Fig. 4A), they were mostly buried in the compensatory sources dominating the early molecular layer response, and full-blown sinks were found in only one-third of our recordings. Reasons for this low association may be partly methodological, such as the 1D CSD derivation, the averaging over many trials of cutaneous stimulation, a nonperpendicular penetration through deeply located, folded folia, and anesthesia (see below), or the more diffuse localization of these synapses for which excitatory currents are rapidly followed by inhibitory currents from interneurons (Mittmann et al. 2005). Nevertheless, in a few cases (Fig. 4), current sinks evoked after putatively corticopontine activation were colocalized with climbing-fiber currents, offering a potential substrate for learning theories.

Our paired recordings of GoCs and PCs showed significant correlations in the response characteristics of these neurons to tactile stimulation. All short-latency response components (within 20 ms after stimulation) most likely originated from mossy-fiber inputs (Vos et al. 1999b), exciting GoCs monosynaptically and PCs disynaptically, and inhibiting PCs a few milliseconds later via molecular-layer interneurons (Barmack and Yakhnitsa 2008; Bower and Woolston 1983).

We also showed that the late-response components, such as the GoC silent period and the PC long-lasting excitation (Holtzman et al. 2006; Vos et al. 1999b), were pairwise correlated, indicating that suppression of GoCs can disinhibit granule cells and facilitate overlying PCs in a time frame beyond 200 ms. Although the silent period has long been observed in other GoC preparations (Eccles et al. 1966), its function and origin have remained unclear. Proposed mechanisms are intrinsic membrane currents of the GoC (Solinas et al. 2007), activation of metabotropic glutamate receptors of type 2 (Watanabe et al. 1998), or the dynamics of the network (Dugué et al. 2009; Maex and De Schutter 1998). The clear current source we found in the granular layer near the GoC recording position and coincident with the silent period (Figs. 5, 6, and 8) would be consistent with an inhibitory synaptic current through the poorly characterized synapses from Purkinje axon collaterals on GoCs (Crook et al. 2007; Fox et al. 1967; Palay and Chan-Palay 1974).

Finally, although the constrained recording conditions did not allow us to map receptive fields exhaustively, our finding that short-latency inhibition was the most consistent PC response supports previous reports of broader inhibitory vs. excitatory receptive fields in PCs (Bower and Woolston 1983; Ekerot and Jörntell 2001; Santamaria et al. 2007).

Origins of Response Latencies

We first discuss the multisynaptic origin of the stimulus-evoked CSs and thereafter the different mossy-fiber pathways. In the present study, facial stimulation evoked CSs in Crus IIa, which is part of cortical zone C2, with a mean latency of 35 ms (see also Wise et al. 2010). In older studies, electrical stimulation of the face or infraorbital nerve was shown to generate discharges in the inferior olive with a latency of 19–30 ms (Cook and Wiesendanger 1976), and the earliest responses in the ipsilateral posterior cerebellum have latencies of about 17–20 ms (Armstrong and Drew 1980; Atkins and Apps 1997). Axonal transport studies showed that the trigeminorecipient zone of the inferior olive (dorsal accessory and principal olives, DAO-PO) contains neurons that project to Crura I and/or II (Huerta et al. 1983; Yatim et al. 1996). Nevertheless, no climbing fiber responses were found in the posterior lobe of the cerebellum of decerebrate rats (Armstrong and Drew 1980; but see also Bengtsson and Jörntell 2007), suggesting that convergence of neocortical inputs is needed to maintain or drive the excitability of the climbing fiber pathway. Dual recordings showed that neocortex actually entrains complex spiking in PCs (Brown and Bower 2002; Ros et al. 2009), and electrical stimulation of motor cortex evoked CSs in Crus IIa at latencies of 15–25 ms (Marshall and Lang 2004). These data substantiate a neocortical origin of the CSs recorded at a latency >30 ms in Fig. 4. Note that these CSs, which marked a sign reversal on the LFP profile (Fig. 4A) and the transition from a sink-dominated granular layer to a sink-dominated molecular layer (Fig. 4B), arrived later than the putative cerebropontine mossy-fiber projection. Hence, the mossy and climbing fiber projections, which have been suggested to coincide (Morissette and Bower 1996; Odeh et al. 2005), were not precisely synchronized in the present study.

As concerns the mossy-fiber-generated sinks in the granular layer and their accompanying peaks on GoC and PC PSTHs, the earliest peak, sometimes arriving after <5 ms, is almost certainly of trigeminal/spinal origin. Tracer studies and electrical-stimulation experiments have demonstrated trigeminocerebellar projections from the nucleus principalis and from pars oralis/interpolaris of subnucleus spinalis to Crus II (Falls and Alban 1986; Watson and Switzer 1978; Woolston et al. 1981; Yatim et al. 1996).

The 2nd peak (at ∼20 ms) has been attributed to a corticopontocerebellar projection (Morissette and Bower 1996; Vos et al. 2000) relayed via the somatotopy-preserving basal pontine nuclei (Leergaard et al. 2000; Serapide et al. 2001). This origin is supported by the disappearance of the 2nd peak by lidocaine inactivation of area S1 (Morissette and Bower 1996) and by isoflurane anesthesia and paired-pulse stimulation in the present study. Nevertheless, alternative pathways cannot be completely ruled out. Another cerebrocerebellar relay station, the nucleus reticularis tegmenti pontis, projects to Crus II in a stripe-like fashion (Serapide et al. 2002). Ascending trigeminocerebellar projections are also relayed over pontine nuclei (Swenson et al. 1984) and especially over the lateral reticular nucleus (LRN). After electrical stimulation of the LRN, Xu and Edgley (2010) observed latencies of excitatory responses ranging from 14.1 to 23.6 ms in 4 out of 27 GoCs. However, in Crus II, projections from pontine nuclei outnumber those from the LRN (Pijpers et al. 2006; Wu et al. 1999).

Origins of Response Variation

The question arises whether differences in response pattern among GoCs, and among PCs, can be explained by the microanatomic organization of the mossy-fiber projections. Mossy fibers originating in the trigeminal nuclei, basal pontine nuclei, and nucleus reticularis tegmenti pontis bifurcate to make multiple projections to Crus IIa, forming discontinuous patches or stripes (Pijpers et al. 2006; Serapide et al. 2002; Woolston et al. 1981), colocalized with climbing-fiber projections of similar body representations (Sugihara and Shinoda 2004).

Vos et al. (2000) described four response patterns in GoCs occurring with almost equal frequency. GoCs could exhibit only an early (trigeminal) response peak, only a late (putatively corticopontine) peak, or both peaks. The fourth and strongest response, in which the earlier peak appeared as a burst, could only be evoked from the most sensitive part of the receptive field, and the burst was attributed to doublet or triplet firing in a monosynaptically connected mossy fiber (Rancz et al. 2007).

The varying patterns of sinks in the granular layer observed in the present study, and in each case matching the peaks on the GoC PSTH, indicate that mossy rather than parallel fibers may shape the GoC receptive fields. However, because it was not possible to completely map the receptive fields within the recording time available, there was a bias toward tactile stimuli evoking stronger GoC responses, and it was previously assumed that primarily parallel-fiber-evoked GoC responses are weak (Vos et al. 1999b). Previous work demonstrating synchronization of GoCs along the parallel fiber axis (Volny-Luraghi et al. 2002; Vos et al. 1999a) also supports GoC activation by parallel fiber synapses. For the responses studied here, the broad activation (Fig. 10) implies a considerable convergence of mossy fibers onto GoCs and a loss of GoC selectivity (Prsa et al. 2009). Whereas the lack of a trigeminal or putatively corticopontine peak in some GoCs, or equivalently in the same GoC at particular stimulus positions, may be explained as a selectivity in the mossy-fiber projection pattern, this would imply as well that the trigeminal and pontine projections are not perfectly aligned.

An even greater response variability has been demonstrated for PCs in many studies (Bower and Woolston 1983; Holtzman et al. 2006; Wise et al. 2010). Despite their massive parallel-fiber input, PCs have been described to have small excitatory receptive fields. In the present study, the trigeminal and putatively corticopontine peaks of excitation were less outspoken in PCs than in GoCs (the strongest case being illustrated in Fig. 4). The inhibitory dip separating them, however, could be evoked from a wide receptive field, as could the long-latency excitation (Fig. 8). The long-latency long-lasting excitation has been argued to be, in terms of numbers of spikes, the most important component of the PC response (Holtzman et al. 2006).

Limitations in the Estimation and Interpretation of the LFP and CSD Profiles

Our 1D calculation of current sources and sinks assumes that all the observed voltages are built up by local transmembrane currents that continue their path in extracellular space exclusively along the depth of the cortex and in a medium with homogeneous resistivity. Hence, extracellular leak currents within a plane perpendicular to the electrode will conceal synaptic currents or be misinterpreted as being caused by synaptic activity. Nicholson and Freeman (1975) warned against this type of error, for which the main effect is to accentuate edges along the CSD profile and to attenuate elongated zones of constant current (see their Fig. 4). The discrete spacing of electrodes may generate a similar artifact on the CSD profile. However, many sinks in the granular layer were so strongly associated with peaks on the PSTHs of coregistered GoCs, as were sources in the molecular layer with dips on the PC PSTH, that we consider them as genuine signs of synaptic activity. Less localized sinks, such as those caused by granule cell synapses in the molecular layer, however, may have been underestimated in the present study. The above requirement of homogeneous resistivity is less stringent for the interpretation of the cerebellar CSD (see Okada et al. 1994).

The present study was not able to distinguish synaptic activity at the ascending segment of granule cell axons (Gundappa-Sulur et al. 1999) from that at parallel fibers. Multishaft recordings are needed to demonstrate a transverse displacement of sinks in the granular layer toward the molecular layer. However, in the present study, delayed response components on GoC PSTHs could always be explained by concomitant sinks in the granular layer, reflecting mossy fiber activity, without having to invoke a delayed propagation along parallel fibers. Although it has recently been shown that ketamine/xylazine anesthesia attenuates the fields caused by parallel-fiber and climbing-fiber responses (Bengtsson and Jörntell 2007), stimulus-evoked sinks were clearly present in the molecular layer of many recordings, associated with responses of SSs and CSs.

Technical Advances for Future Research

The present study demonstrates the feasibility of multielectrode, combined field/spike recordings in cerebellar cortex, opening the venue for several applications.

The probe design, only 15 μm thin, caused no dimpling of the cerebellar surface and had no influence on the quality of the GoC and PC single-unit recordings (Fig. 1). Because of the dense structure of the cerebellar cortex, this was a major concern and a requisite for future use of the technique in a chronic setting in the behaving rat.

Brain computer interface technology could benefit from LFP/CSD feature extraction as an add-on or alternative to the preferred invasive methodology of today where single- and multiunit extracellular recordings of mainly output neurons are used (for review, see Nicolelis and Lebedev 2009). Using LFP/CSD reduces electromyographic and high-frequency electronic noise (Csicsvari et al. 2003), is easier to capture, and lasts longer than single-unit recordings (Scherberger et al. 2005).

GRANTS

This work was supported by Fonds voor Wetenschappelijk Onderzoek (FWO) Grant G.0244.08 to E. De Schutter and R. Maex and by the University of Antwerp.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).