Effects of short-term synaptic plasticity in feedforward inhibitory circuits on cerebellar responses to repetitive sensory input

Abstract

Short-term plasticity occurs at synapses throughout the cerebellum, raising the question of how such plasticity affects cerebellar processing in vivo. To address this issue, we recorded responses of molecular layer interneurons (MLIs), Purkinje cells, neurons of the cerebellar nuclei (CbN), and mossy fibers in the CbN, in awake head-fixed female mice. During short trains of air puffs applied to the whisker pad with intervals from 25 to 200 ms, the first puff generated brief suppressions of spike probability in Purkinje cells and brief elevations in all other cell types, resulting in coincident excitation and disinhibition of CbN cells and large whisker protractions. Later puffs evoked smaller whisks and smaller responses in all cells, with the strongest decrease in the CbN. The reduction resulted from facilitation of EPSCs from parallel fiber axons of granule cells, decreasing net inhibition of individual Purkinje cells, and from activation of fewer MLIs, reducing inhibition across the Purkinje population. Downstream, the decrease in Purkinje-mediated disinhibition, in conjunction with depression of excitatory mossy fiber-to-CbN pathways, reduced net excitation of CbN cells. Sensory-evoked responses were transient and effectively transmitted synaptically, but movement-related responses were prolonged and progressively cancelled at successive stages of the circuit. Moreover, many MLI, Purkinje, and CbN cells had bilateral receptive fields. In these cells, changing the stimulus location restored responsiveness and increased whisk magnitudes. Thus, several types of cerebellar neurons can report stimulus changes without specifying stimulus parameters, thereby serving as event detectors that can facilitate movement in response to altered sensory inputs.

Introduction

A major role of the cerebellum is rapid sensorimotor integration, which involves coordinating responses to sensory input on the time scale of milliseconds, in movements ranging from simple reflexes to complex behavioral sequences (Thach 1968, 1970a, 1970b; Armstrong and Edgley, 1984a, 1984b; Medina and Lisberger, 2007; Bryant et al., 2010; Herzfeld et al., 2015; Chen et al., 2016; Brown and Raman, 2018; Brown et al., 2024; Gaffield et al., 2022; Becker and Person, 2019). Cerebellar circuits underlying such functions were originally noted for their systematic organization and apparent simplicity (Eccles et al., 1967; Marr 1969): To a first approximation, they consist of two feedforward inhibitory circuits, the mossy fiber → Golgi cell → granule cell (parallel fiber axon) pathway and the granule cell → molecular layer interneuron (MLI) → Purkinje cell pathway, both of which are nested within the mossy fiber → granule → Purkinje → cerebellar nuclei (CbN) circuit.

The computations performed by these pathways, however, lead to transformations not predicted by anatomical connectivity alone, as behaviorally evoked changes in spike rate do not consistently generate the expected neuronal responses. For instance, Purkinje cells strongly inhibit CbN cells (Ito et al., 1970; Jahnsen 1986; Aizenman et al., 1998), yet during continuous movements like reaching, locomotion, passive limb flexions, and whisking, increases in Purkinje cell firing rates do not reliably elicit drops in CbN cell spiking (Thach 1968, 1970a, 1970b; Armstrong and Edgley, 1984a, 1984b; McDevitt et al., 1987; Sarnaik and Raman, 2018, Brown and Raman, 2018; Chabrol et al., 2019). By contrast, in stimulus-evoked actions, such as whisker protractions in mice generated by brief tactile input to the whisker pad, Purkinje cell spiking is transiently suppressed, correlating with disinhibition of CbN neurons and amplified whisker movements, revealing the inverse spiking relationship predicted by the circuit (Brown and Raman, 2018; Brown et al., 2024). The sign of the Purkinje cell response, however, is not predicted from the anatomy, since parallel fiber excitation from granule cells stimulated by trigeminal mossy fibers should precede feedforward inhibition from MLIs by one synaptic delay. Thus, factors such as the interactions among convergent and divergent inputs, as well as the relative amplitudes and kinetics of synaptic responses, evidently contribute to the complexity of outputs.

Synaptic and intrinsic parameters, however, are themselves subject to short-term modulation, and in vitro studies in cerebellar slices reveal a rich palette of short-term plasticity. Mossy fiber synapses onto granule cells can depress or facilitate (Xu-Friedman and Regehr, 2003; Saviane and Silver, 2006; Chabrol et al., 2015); parallel fiber synapses onto Purkinje cells facilitate (Atluri and Regehr, 1996; Dittman et al., 2000; Valera et al., 2012) and the resulting EPSCs sum with MLI-mediated IPSCs to generate a variety of Purkinje cell responses (Dizon and Khodakhah, 2011; Grangeray-Vilmint et al., 2018; Brown et al., 2024); and mossy fiber synapses onto CbN cells depress strongly, while transmission from Purkinje cells remains robust even at high frequencies (Wu and Raman, 2017; Telgkamp et al., 2004). Additionally, temporal relationships among convergent inputs can change the efficacy of transmission (Person and Raman, 2012; Wu and Raman, 2017; Binda et al., 2023), raising the question of how short-term plasticity influences cerebellar output in vivo.

To investigate this question, here we recorded responses of MLIs, Purkinje cells, and CbN cells during short trains of air puffs applied to the whisker pads of awake head-fixed mice. All cell types generated reliable, transient responses to the initial stimulus, while responses to later stimuli were reduced, leading to decreased cerebellar output and smaller whisker protractions with successive puffs. The decreases resulted from differential short-term plasticity and changing relative strengths of excitation and inhibition at several synapses. In neurons with bilateral receptive fields, changing the side of stimulation restored responsiveness and increased whisk magnitude. Moreover, unlike brief sensory-related signals, slower movement-related modulations of firing rate were not reliably transmitted, leading to a high-pass filtering that allows cerebellar whisking circuits to report and respond to sensory change.

Materials and Methods

Experimental animals.

All experiments were conducted in accordance with institutional guidelines with methods approved and overseen by the Northwestern University IACUC. All mice were housed with standard care in an accredited veterinary facility with free access to food and water. For cerebellar slice recordings, experiments were performed on tissue from 3- to 5-week-old wild-type C57BL-6/J mice (n = 8, P21-32, Jackson Labs, RRID: IMSR JAX:000664). Both sexes were used, and sex was recorded. No differences in parameters studied were found between sexes and the data were pooled. For in vivo experiments in awake mice, females were used because they adapted more readily to the recording apparatus than males. For in vivo electrophysiological recordings, single-unit recordings were performed on 8- to 16-week-old F1 female offspring of Pcp2-cre (RRID: IMSR JAX:004146) and Ai32 (RRID: IMSR JAX:024109) mice, termed Pcp2-ChR2 mice (n=16). For in vivo optogenetic experiments, granule cell stimulation experiments were performed on 8- to 16-week-old F1 female offspring of Gabra6-cre (a kind gift of Dr. Yue Yang, Northwestern University) and Ai32 mice (n= 4, Jackson Labs) termed Gabra6-ChR2 mice. For in vivo voltage imaging, experiments were performed on 8- to 16-week-old, heterozygous female Nos1-cre mice (RRID: IMSR JAX:017526, n=5, Jackson Labs) virally expressing the voltage indicator pAce (Kannan et al., 2022) in MLIs.

Experimental Design and Statistical Analysis

Cerebellar slice recordings.

Mice were anaesthetized with isoflurane and transcardially perfused with warm artificial cerebrospinal fluid (ACSF, in mM: 123 NaCl, 3.5 KCl, 26 NaHCO₃, 1.25 NaH₂PO₄, 1 MgCl₂, 1.5 CaCl₂, and 10 D-glucose) bubbled with 95% O₂/5% CO₂ (36°C). Parasagittal slices (300 μm) were cut on a VT1200 vibratome (Leica), incubated in carbogenated ACSF for 30 minutes at 34°C, then recovered at room temperature (22°-25°C) before being transferred to the recording chamber (32°-35°C). Borosilicate pipettes (2-5 MΩ) were filled with (in mM): 105 KCH₃SO₃, 10.4 D-gluconic acid, 5.3 NaCl, 2 MgCl₂, 0.1 CaCl₂, 1 EGTA, 10 HEPES, 4 MgATP, 0.3 Tris GTP, 14 phosphocreatine di(Tris) salt, 4.8 QX314 (Tocris), 1 4-AP, 20 TEA. All chemicals were from Sigma-Aldrich. Whole cell recordings were made from crus I/II Purkinje cells. Data were acquired at 20 kHz with a Multiclamp 700B amplifier / Digidata 1550B and pClamp (Molecular Devices). Synaptic release from granule cells and MLIs was evoked by electrically stimulating parallel fibers with a glass theta stimulating electrode. Synaptic currents were recorded in Purkinje cells in voltage clamp at −70 mV for EPSCs and 0 mV for IPSCs. At least 4 trials were repeated for each cell and averaged.

Surgeries.

In preparation for in vivo experiments, mice were anesthetized with isoflurane (1-2%), locally injected with lidocaine (2%) under the scalp, and buprenorphine ER-Lab (1 mg/kg, subcutaneous, ZooPharm) and meloxicam (20 mg/kg, subcutaneous, Dechra) were pre-operatively administered for analgesia. A head plate was surgically implanted for head-fixation using dental cement (C&B-Metabond). Surgical microbind screws (1/16 SL; Fisher) were implanted in the parietal bones to improve the adherence of the headplate to the skull and thereby to reduce motion artifacts during recordings or imaging. A craniotomy was performed centered over crus I. For single-unit recordings, the craniotomy was centered at 6.25 mm caudal and 3.125 mm lateral to bregma for Purkinje cells, or at 6.1 mm caudal and 2.3 mm lateral to bregma for CbN cell recordings. For imaging experiments with the voltage indicator pAce, two 500 nL injections (AAV/DJ-CAG-DIO-pAce-Kv2.1, diluted 1/10, stock titre: 1.9x10e13 VG/mL) were made at separate locations in crus I of Nos1-cre mice at a depth of 500 μm. After injection, a 3 mm diameter, 0.15 mm thick circular coverslip (Harvard Apparatus) was secured over the craniotomy with dental cement. After ~10 days of recovery for non-imaging experiments, or 2-4 weeks of recovery and indicator expression for voltage imaging experiments, mice were habituated to head-fixation on the setup, by sitting on a platform for 1-2 hours per day for 2-3 days. After mice had been acclimated to the setup, experiments were performed on each mouse over a period of 5-10 days.

In vivo electrophysiology.

Loose cell-attached recordings of Purkinje cells or CbN cells were made in awake head-fixed mice. Borosilicate pipettes (3-6 MΩ) were filled with Tyrode’s solution (mM, 150 NaCl, 4 KCl, 2 CaCl₂, 2 MgCl₂, 10 HEPES, 10 glucose, pH 7.35 with NaOH). Recordings were made in voltage-clamp mode with a MultiClamp 700B amplifier, digitized with a Digidata 1550B, and acquired with pClamp software (Molecular Devices). Purkinje cell recordings were made from the dorsal surface of crus I and in the fissure between crus I/II, within the coordinates of 2.5-4 mm lateral and 5.8-6.6 mm caudal to bregma, at depths ranging 0.1-2.5 mm below the surface. Purkinje cell identity was confirmed by the presence of complex spikes in the recordings. Recordings from neurons in the lateral CbN were made 2.2-2.7 mm lateral and 5.7-6.7 mm caudal to bregma, and 1.75-3 mm below the surface of the cerebellum. Because Purkinje cells in the experimental Pcp2-ChR2 mice can be optogenetically activated, CbN cell identity was confirmed by the silencing of spiking when crus I/II Purkinje cells were stimulated for 20 ms with a 200 μm fiber optic (NA = 0.22, ThorLabs) coupled to a 465 nm LED (Doric Lenses) placed onto the surface of crus I. After each recording session, craniotomies were covered with Kwik Sil (World Precision Instruments).

Air puff and optogenetic stimulation.

For optogenetic activation of granule cells, an optical fiber (200 μM, NA=0.22, ThorLabs) was placed on the surface of crus I in Gabra6-ChR2 mice. Granule cells were stimulated with a 5-ms pulse of 465 nm light (4.5 mW, measured at the fiber tip with a Thorlabs PM100D).

For sensory stimulation of the whisker pad, air puffs (10 ms, 20 psi) were applied through a nozzle (1 mm inner diameter) gated by a solenoid valve. The nozzle was positioned ~1 cm from the whisker pad. Depending on the type of experiment, either single air puffs or trains of air puffs were applied to the contralateral and/or ipsilateral whisker pad with an inter-trial interval of 5–6 sec. The timing of the puffs was controlled by either a Master-8, Master-9, or Arduino. The puff reached the whisker pad ~8 ms after the trigger. Electrophysiological recordings for each cell were averaged for 46 ± 25 trials/condition.

Targeted illumination, 1-photon voltage imaging microscope.

Voltage-imaging of MLIs was done as in Brown et al., (2025). The emission light path of the one-photon microscope consisted of a 20X, 1NA water immersion objective with 0.17 mm coverslip correction (Zeiss) followed by a f=100 mm infinity corrected tube lens (TTL100-A, Thorlabs). The resultant magnification at the camera sensor was 12X. The excitation light path consisted of a 470 nm LED (M470L5, Thorlabs), a 1024x768 digital micromirror device (DMD, V-7000, Vialux), followed by a f=165 mm infinity corrected tube lens (TTL165-A, Thorlabs), which focused the DMD image at the sample plane and set the illumination field of view to be 500 μm x 670 μm. A GFP filter set was used to spectrally isolated the excitation and emission light. Voltage signals were imaged with a Veo 610 sCMOS camera (Vision Research).

Voltage imaging.

To identify neurons for recording, a thin bar of light was scanned across crus I while images were acquired. The acquired stack of images was then max Z-projected, producing a high-contrast image where indicator-labelled MLIs were clearly visible. MLI somata were selected for light-targeting by uploading a black and white region of interest (ROI) image mask to the DMD containing white spots at the locations of neuronal targets (spot diameter: 22.5 - 28.5 μm; >twice the diameter of an MLI). The ROI image underwent an affine transform to ensure that the DMD image projected onto the sample with the appropriate size and orientation. To obtain a high signal-to-noise ratio, the light power of the LED was set to provide ~100 mW/mm² at the level of the imaging plane. Although targeted illumination is susceptible to brain motion artefacts, the stability of the surgical preparation (described above) and the habituation of the mice to the setup minimized such error (Brown et al., 2025). One or two nonoverlapping fields of view were imaged in each mouse across one to two imaging sessions. Images were acquired at 3000-4000 frames per second (fps) with 12-bit resolution. To minimize photobleaching of pAce and to ensure that neurons were exposed to minimal light, imaging was performed episodically, with a trial duration of 1 second for experiments during which the interstimulus interval was varied or ~700 ms for experiments in which stimulus location was changed. Voltage imaging recordings for each cell were averaged for 50 ± 9 trials/condition (range = 30-61).

All image preprocessing, ROI delineation, and signal extraction were performed as in Brown et al., (2025). Raw imaging data underwent subpixel rigid motion correction by adjusting for the phase correlation shift between each frame and a filtered reference frame. Images were then down-sampled from a resolution of 1.5 μm to 2.25 μm to speed up subsequent processing. Since the signal of interest was the rapidly changing fluorescence associated with action potentials, slowly varying fluorescence was removed by subtracting a 10 ms moving median of each pixel. The spatial locations (x,y) of all putative action potentials in the imaging volume, detected as local 3D maxima, were aggregated into a 2D spatial histogram where individual neurons appeared as prominent peaks owing to the aggregation of their spikes in single bins (Chen et al., 2025). 41x41 pixel regions centered on each neuron were extracted at the time of each spike. The spatial footprint, or spike-triggered average (STA), of each neuron was computed by taking the median of its spike images. Traces of changes in signal over time were then obtained for the STA and neuropil filter for each neuron by matrix multiplication with the full image volume and neuropil signal was then removed from the total signal by linear regression. STA signals then underwent demixing and common reference removal (Brown et al., 2025).

Whisker imaging and tracking.

All whiskers were trimmed except C1 and C2 (ipsilateral to the recording site). Whiskers were illuminated from below with an infrared LED array and were imaged at 250 fps with a Genie Nano M640 NIR camera (Teledyne DALSA Inc.). Whiskers were tracked in Fiji (ImageJ) using the Ridge Detection plugin and whisker angle was extracted from the most rostral whisker in MATLAB (Brown et al, 2025). In the experiments involving optogenetic activation of granule cells, whiskers were tracked with DeepLabCut (https://www.mackenziemathislab.org/deeplabcut) and the angle was computed between 2 tracked points on the base of the rostral whisker relative to the midline. Analysis of whisker position was performed on 12 mice (> 300 trials/condition).

Spike sorting for awake recordings.

Simple and complex spikes were clustered based on the mean and standard deviation of the action potential waveform in a 4-ms window following the initial spike. Complex spikes clustered separately from simple spikes, owing to higher variance in the signal from the spikelets and post-spike depolarization. Each complex spike was then visually confirmed.

Data analysis.

All data were analyzed in MATLAB and plotted in IGOR (WaveMetrics). For each cell, mean peristimulus time histograms (PSTHs) were constructed. Spikes for all trials were aligned to stimulus onset (trigger time of the first puff or optogenetic pulse), summed in 2-ms bins, and divided by the number of trials. The brief bin width provided a measure of spike timing, reported as mean spike probability per bin.

In Gabra6-ChR2 mice subject to optogenetic stimulation, Purkinje cells were classified as light responsive if their simple spike probability changed relative to a 200 ms baseline in a single 2-ms bin between 0–14 ms post-stimulation. Cells with a single bin z-score of ≥1.96 (p<0.05, two-tailed) were classified as opto-elevated cells and those with a z-score of ≤ −1.96 were classified as opto-suppressed cells. Cells that responded with both a significant elevation and suppression were categorized according to the largest response. The change in spike probability (Δspk_prob) was calculated as the difference between the baseline (−200 ms to 0 ms) and the maximum or the minimum value in the response bins from the PSTH.

For experiments involving sensory stimulation, only Purkinje cells with ≥ 60 trials (≥20 per condition) were included, which was required for reliable classification of simple spike suppression. Because the lag between triggering the air puff and stimulation of the whisker pad was ~6-8 ms and the sensory signal is relayed in 6–8 ms (Brown et al., 2024), Purkinje cells were classified based on the deviation of the lowest value in the three PSTH bins between 12 and 18 ms after the first stimulus. If the lowest bin z-score was more negative than −1.65 (p<0.05, one-tailed) relative to a 200 ms-baseline, the cell was classified as “simple-spike suppressed.”

For calculation of Δspk_prob under the various experimental conditions, a 2-bin response window was selected according to the time at which the average suppression occurred in the population: 14–18 ms for the multiple-interval experiment and 12–16 ms for the contralateral/ipsilateral puff experiment. The mean value of these two bins was compared to the relevant baseline taken from a “rate trace” computed from a 10-bin (20-ms) rolling average of the PSTH. For complex spikes, PSTHs with 2-ms bins were smoothed as above to generate a rate trace for complex spikes, from which the peak was measured. Purkinje cells were classified as responsive at the level of complex spikes if they had a z-score ≥ 1.65 (p<0.05). These were not necessarily the same cells that showed simple spike suppressions.

To be included in the dataset, CbN cells required ≥ 15 trials (≥5 per condition). CbN cells were classified as responsive if the z-score of their spike probability in a single bin 10–18 ms after the first stimulus was ≥ 1.65 (p<0.05), relative to a 200-ms baseline. MLIs in the dataset underwent ≥ 30 trials of each tested condition. MLIs were defined as responders if the z-score in a single bin 12–24 ms after the first stimulus was ≥ 1.65 (p<0.05), relative to a 200-ms baseline. For responses during trains, the baseline and absolute peak probability, respectively, were calculated as the mean of bins 0–10 ms post-stimulus (before the signal reached MLIs) and as the maximum bin 12–24 ms post-stimulus. The Δspk_prob was calculated as the difference between the baseline and absolute peak probability.

To measure mossy fiber responses, cell-attached records were first zeroed by subtracting a rolling mean of 1 second. After spike detection, traces were blanked 0.6 ms pre- and post-spike to remove the associated transient. After removal of spikes, the filtered signal was aligned to the stimulus, and all trials were averaged. The mean mossy fiber transient associated with sensory input was calculated as the difference between the maximum value 0–12 ms post stimulus and the preceding minimum value in the same window.

Statistics.

Data are reported as mean ± SEM unless noted otherwise. Statistical significance was assessed with paired Wilcoxon signed rank tests, unless indicated as unpaired.

Results

Responses of Purkinje and CbN cells to repetitive tactile stimulation.

Whisker-specific tactile information is relayed to the cerebellum via mossy fibers from the trigeminal nuclei, which excite glutamatergic granule cells. Granule cells monosynaptically excite Purkinje cells directly through their parallel fiber axons and provide disynaptic feedforward inhibition through MLIs (Figure 1A). Despite the predicted sequence of excitation followed by inhibition, however, loose-cell attached recordings from crus I/II Purkinje cells in awake head-fixed mice (Figure 1B) previously showed that, in ~50% of cells, the first response to an air puff applied to the whisker pad, which evokes a reflexive whisker protraction, is a well-timed, brief (2-4 ms) suppression of simple spikes. This transient response is followed by a gradually increasing and decreasing prolonged increase in simple spike firing associated with whisking (Brown and Raman, 2018; Brown et al., 2024). Repeating this experiment and generating mean peristimulus time histograms (PSTHs) from spike rasters (Figure 1C), with 2-ms bins to illustrate spike timing (Brown and Raman, 2018, Methods), verified that the spike probability fell from 0.153 ± 0.010 to 0.010 ± 0.014 in the 14-18 ms window after the puff was triggered (Materials & Methods) in 22/50 cells (44%, Figure 1D, light blue). These cells were therefore classified as “simple-spike suppressed”; the remaining 28 “non-suppressed” cells responded with a short-latency spike probability increase or lacked detectable responses (Figure 1D, dark blue). Suppressions were followed by a slower increase in complex spike probability 10-30 ms later than simple spike suppression (max, 0.046±0.005); a similarly timed but smaller increase was present in non-suppressed cells (max, 0.020±0.004, p<0.001, unpaired, Figure 1D, black, grey). The reproducible timing of the simple spike inhibition confirms that tactile input generates a brief, synchronous gap in simple spiking across the population of suppressed Purkinje cells. Moreover, the presence of this gap without a preceding elevation of spike probability indicates that the inhibition of Purkinje cells by MLIs must outweigh the excitation from parallel fibers, despite the extra synaptic delay (Brown et al., 2024).

Figure 1. Purkinje cell responses to trains of repetitive sensory input.

A) Simplified diagram of the cerebellar circuit. Mossy fibers carry sensory and motor-related input to the cerebellum, forming excitatory synapses on granule cells (GrC) and neurons of the cerebellar nuclei (CbN). Parallel fiber axons of granule cells excite molecular layer interneurons (MLI) and Purkinje cells. Purkinje cells inhibit CbN cells. The + and − signs denote excitatory and inhibitory synapses, respectively.

B) Top, Single unit recording from a Purkinje cell showing simple spikes (SS) and complex spikes (CxS). Bottom, Schematic of the experimental setup for in vivo recordings during head-fixed, puff-evoked whisking. Recordings were made from crus I/II, puffs were applied to the contralateral whisker pad, and the ipsilateral whiskers were tracked with high-speed videography.

C) Sample spike raster of a Purkinje cell across 90 trials. The puff was applied at time 0. Simple spikes are shown as blue ticks and complex spikes as grey circles. Note that the clustering of complex spikes 30-40 ms after the puff is readily apparent but the earlier well-timed gap in simple spikes is more difficult to discern by eye.

D) Top, puff-evoked whisker movement averaged across all trials in all cells. Upward deflection indicates protraction. Bottom, 2-ms population PSTH of simple spikes of simple-spike suppressed (n=22) and non-suppressed (n=28) Purkinje cells. Simple spikes are shown as light blue bars for suppressed and a dark blue line for non-suppressed cells. Complex spike PSTHs are superimposed as grey bars and black line, respectively.

E) Population PSTH of simple-spike suppressed Purkinje cells aligned to time of the first puff of the train for each interstimulus interval (ISI, 200, 100, 50, and 25 ms). Bin width, 2 ms. Simple spikes (blue). Unlike panel D, complex spikes are plotted as bars (grey) that sum with the simple spikes to provide a measure of total spiking.

F) Change in spike probability for simple spikes (SS) for the first and second puff for each ISI from data in E. Individual cells, open symbols; and mean, closed symbols.

G) Mean ± SEM values for change in spike probability for each puff in each train.

H) Complex spike (CxS) response probability plotted as a smoothed PSTH for each ISI.

I) Mean ± SEM change in complex spike probability (ΔCxS) for each puff in each train.

Since synaptic responses are subject to short-term modulation, we next recorded responses to repetitive stimuli by cells classified as suppressed. Trains of 5 air puffs with interstimulus intervals of 200, 100, 50, and 25 ms were applied to the whisker pad (Figure 1E). In all cases, suppressions were evident at the onset of the train. The mean PSTHs across all cells indicated that subsequent responses became smaller with progressively shorter intervals. Since suppressions superimpose relatively linearly on movement-associated rate changes (Brown and Raman, 2018; Brown et. al. 2024), responses were quantified on a cell-by-cell basis as the change in spike probability (Δspk_prob) relative to the smoothed rate trace, (Materials & Methods); hence, more negative numbers indicate larger suppressions (Figure 1F). Compared to the first puff, the Δspk_prob evoked by the second puff was virtually abolished after an interval of 25 ms, falling to 7% of the first response (puff₁ vs. puff₂, from −0.045 ± 0.010 to 0.003 ± 0.009, p=0.004, N=22). For the 50-ms interval, it dropped to 47% (−0.049 ± 0.011 to −0.023 ± 0.007, p=0.03) and for 100 ms, to 53% (−0.059 ± 0.007 to −0.031 ± 0.008, p=0.002). With a 200 ms interval, however, the Δspk_prob largely recovered, reaching 82% of the original value (−0.050 ± 0.009 to −0.041 ± 0.007, p=0.16). The mean values from the cell-by-cell analysis for each interval indicated that the responses fell by ~60% over about 4 stimuli (Figure 1G). Thus, suppression is strongest in response to the initial puff and decreases with repetition.

Complex spike probability also fell with repetitive stimulation, remaining near 50% on the second stimulus for intervals of 200 and 100 ms, but dropping to <20% or undetectable for intervals of 50 and 25 ms, respectively (Δspk_prob for puff₁ vs puff₂: 200 ms, 0.036±0.004 vs 0.021±0.002; 100 ms, 0.035±0.003 vs 0.017±0.002; 50 ms, 0.036±0.003 vs 0.006±0.001; Figure 1H, 1I).

The simple spike suppressions are of interest because they reflect a synchronous population response, which can intensify transmission to downstream neurons receiving convergent Purkinje input (Person and Raman, 2012). We therefore repeated experiments while recording from CbN cells. These cells responded to the first puff with strong, transient increases in spike probability, which reliably produced a double peak in the PSTH (Figure 2A). The increase in spike probability began ~2-4 ms before the Purkinje spike suppression, likely owing to direct excitation by trigeminal mossy fibers, and increased strongly during the period of transient disinhibition (Brown and Raman, 2018). With successive stimulation, CbN response magnitudes diminished greatly (Figure 2A). As in Purkinje cells, an interval of 25 ms strongly reduced the Δspk_prob to the second puff, which fell to 34% (from 0.390 ± 0.043 to 0.132 ± 0.018, p<0.001, N=25). The drop in response amplitude occurred over fewer stimuli than in the Purkinje cells (Figure 2B), however, giving a relative Δspk_prob to the second puff of 43% for the 50-ms interval, (from 0.392 ± 0.045 to 0.167 ± 0.022, p<0.001), 45% for the 100-ms interval (from 0.384 ± 0.043 to 0.174 ± 0.020, p<0.001), and 51% for the 200-ms interval (from 0.380 ± 0.039 to 0.193 ± 0.022, p<0.001). Thus, the differential response to the first and later puffs is intensified in the CbN, in a manner that is relatively weakly interval-dependent on the time scales tested (Figure 2C). Nevertheless, the relative changes in spike probability in CbN and Purkinje cells are well correlated, largely owing to the drop after the first stimulus (r² = 0.77, Figure 2D)

Figure 2. CbN cell responses to trains of repetitive sensory input.

A) Population PSTHs of CbN cell responders (n=25) for each ISI (200, 100, 50, and 25 ms, as labeled). Bin width, 2 ms.

B) Change in spike probability for the first and second puff for each ISI from data in A. Individual cells, open symbols; and mean, closed symbols.

C) Mean ± SEM values for change in spike probability for each puff in each train.

D) Change in spike probability for CbN cells vs. Purkinje (Pkj) cells. Dotted line, linear fit with slope = −5.525 and intercept = 0.055. Note that the relatively steep slope of the fit is driven primarily by the first stimulus, indicating that the CbN cell responses are less strongly dependent on ISI than the Purkinje cell responses.

E) Mean evoked whisker movements during trains of puffs with each ISI; shorter intervals evoke smaller protractions. Resting position is equivalent in all cases, and traces are offset for comparison.

Experimentally reducing the output of CbN cells decreases the magnitude of reflexive whisker protractions (Brown and Raman, 2018; Brown et al., 2025). Here, whisker deflections became smaller after the first puff in the train, providing further evidence that the size of the sensory-evoked movement is influenced by the magnitude of CbN output (Figure 2E).

Filtering of signals by short-term synaptic plasticity.

The observation that a series of identical stimuli generates progressively smaller, or adapting, responses motivated us to investigate the synaptic transformations occurring on these time scales at successive stages of the cerebellar circuit. We first considered the basis for the decreasing simple spike suppressions in Purkinje cells. Since these suppressions are likely driven by inhibition from molecular layer interneurons (MLIs), we tested the extent to which MLIs respond reliably to repetitive tactile input by using in vivo voltage imaging to record action potentials of MLIs (Brown et al., 2025) during trains of tactile stimuli.

MLIs were selectively labelled with the genetically encoded voltage indicator pAce (Kannan et al., 2022) expressed in Nos1-cre mice. During voltage imaging from cohorts of 62-142 cells (Figure 3A, 3B), trains of 4 puffs were applied to the whisker pad with the same interstimulus intervals as in the previous experiments (Figure 3C). Averaging trials from all cells revealed that a single air puff to the whisker pad resulted in a response profile with two components: a shortlatency, transient increase of spike probability, which was triggered by the sensory input, followed by a later, broadly timed increase, which correlated with the evoked whisker movement. The first, transient phase reflects spiking synchronized on the time scale of milliseconds across the population and the second, slow phase reflects an increase in spike rate, i.e., more spikes occurring non-synchronously over a longer period of time (Brown et al., 2025). Unlike Purkinje cells, nearly all MLIs responded to the puff (n=337/356, 95%; baseline vs peak spike probability, 0.033 ± 0.001 vs 0.166 ± 0.004, p<0.001). These cells were included in further analyses.

Figure 3. MLI responses to trains of repetitive sensory input.

A) MLIs in a single field of view in crus I of a head-fixed mouse. MLIs are expressing the voltage indicator pAce, illuminated with targeted 1-photon light for voltage imaging.

B) Simultaneous recording of spikes from a population of MLIs (30/78 shown), imaged at 4000 fps, on a single trial. Red line and arrow denote time of air puff application.

C) Population PSTH of MLI responders (n = 337) for each ISI (200, 100, 50, and 25 ms, as labeled). Bin width, 2 ms.

D) Mean ± SEM values for peak spike probability for each puff in each train.

E) As in D for baseline spike probability.

F) As in D for change in spike probability, calculated as the difference between baseline and peak.

G) Change in spike probability for Purkinje (Pkj) cells vs. either the peak (open symbols) or change in (solid symbols) spike probability for MLIs. Red lines, linear fits, with slopes of −0.014 and −0.550 and intercepts of −0.025 and 0.019, respectively, indicating that Purkinje cell responses correlate more strongly with transient changes in MLI spiking.

With successive puffs, MLIs showed transient spike probability increases after each stimulus, superimposed on the slow changes in spike rate (Figure 3C). The data were first quantified as absolute peak spike probability following each puff, which stayed high and even increased slightly with the briefest intervals (Figure 3D). These peak probabilities were largely driven by the underlying movement-associated increase in spike rate, however, evident as the elevated baseline before the transient responses (Figure 3E). Subtracting the baseline gave an estimate of the Δspk_prob associated with each transient, which dropped with repeated stimulation (puff₁ vs puff₂; 200 ms, 0.133 ± 0.004 vs. 0.088 ± 0.003; 100 ms, 0.123 ± 0.004 vs. 0.087 ± 0.003; 50 ms, 0.126 ± 0.004 vs. 0.083 ± 0.003; 25 ms, 0.122 ± 0.004 vs. 0.076 ± 0.003; all p<0.001; Figure 3F). This transient change in inhibition correlated better with the change in simple spike suppression in the Purkinje cell targets than did the absolute amount of inhibition (Purkinje Δspk_prob vs. MLI Δspk_prob, r² = 0.69; vs. absolute MLI spk_prob, r² = 0.0003; Figure 3G). Thus, with repetitive puffs, a decreasing proportion of MLIs responding to the sensory stimulation may weaken the suppression-inducing transient inhibition seen across the population of Purkinje cells.

This idea, however, is predicated on the inference that within individual granule-MLI-Purkinje microcircuits, inhibition can dominate over excitation, despite the additional synaptic delay (Brown et al., 2024). If so, then direct activation of granule cells should be sufficient to elicit a suppression. We therefore recorded Purkinje cell responses to brief (5-ms), optogenetic stimulation of granule cells in Gabra6-ChR2 mice, on the assumption that afferent parallel fibers would be stimulated by chance in a subset of recordings (Figure S1A). Plotting z-scores of transient spike probability relative to mean probability 0-14 ms post-stimulation revealed that 41% of Purkinje cells (n=31/76 cells) did not respond to the applied optogenetic stimulus (Figure S1B), suggesting that in these cases, the activated granule cells did not impinge on the recorded Purkinje neuron. In the remaining 45 cells, optogenetic stimulation indeed affected Purkinje spiking. In 42% of these Purkinje cells, simple spiking increased (“opto-elevated” n=19/45, Figure 4A, left, Figure S1C, left, S1D). Owing to jitter in response time of the opto-elevated cells, the Δspk_prob in the population PSTH was only 0.034 (Figure 4B, left), although cell-by-cell analysis indicated that the maximum Δspk_prob was 0.102±0.011 (Figure 4B, right).

Figure 4. Feedforward inhibition and short-term plasticity of cerebellar synapses.

A) Heat map of z-scores representing deviations from the 200 ms baseline, calculated from PSTHs of Purkinje cell simple spikes after optogenetic stimulation of granule cells at time 0. Left, opto-elevated (n=19) and right, opto-suppressed (n=26) Purkinje cells. Note that opto-suppressed cells tend to display more extreme deviations than opto-elevated cells.

B) Left, population PSTHs of simple spikes from the heat maps in (A) for opto-suppressed cells (light blue bars) and opto-elevated cells (dark blue line). Right, change in simple spike probability for individual cells open symbols, and mean ± SEM (closed symbols).

C) Top, mean ± SEM whisker position during optogenetic stimulation, indicating that the 5-ms direct stimulation of granule cells had no consistent effect on whisker position. Bottom, population PSTHs of complex spike (CxS) probability aligned to the optogenetic stimulus for opto-suppressed cells (grey), and opto-elevated cells (black), indicating that the 5-ms direct stimulation of granule cells had no consistent effect on complex spiking.

D) In vitro whole-cell voltage-clamp recording from a Purkinje cell, showing postsynaptic currents evoked by parallel fiber stimulation with trains applied with the same ISIs as in vivo (left to right, 200, 100, 50, and 25 ms). Each panel shows a pair of traces, the upper one recorded at a holding potential of 0 mV for IPSCs, and the lower one recorded at a holding potential of −70 mV for EPSCs.

E) Mean ± SEM values for short-term plasticity for EPSCs (left) and IPSCs (right) in Purkinje cells (n=11) from experiments as in (D). Peak synaptic current amplitudes (PSC_n) were normalized to the peak of the first current (PSC₁). Points above and below the dotted lines indicate short-term facilitation and short-term depression, respectively.

F) Recovery of Purkinje cell synaptic currents from short-term plasticity evoked by single stimuli. Top, Pairs of electrical stimuli were applied to parallel fibers with a range of interstimulus intervals (in 20-ms increments, 100-ms interval depicted). Bottom, Overlaid traces of pairs of IPSCs (upper traces) or EPSCs (lower traces) evoked by the stimulation. Response to the 100-ms interval, grey, other responses, black. For IPSCs, the relatively constant amplitudes at all intervals indicate the lack of plasticity induced by the first stimulus; for EPSCs, the large second response indicates facilitation, which decreases when the second stimulus is applied at larger intervals.

G) Mean ± SEM values for recovery from short-term plasticity, for EPSCs (filled symbols) and IPSCs (open symbols) in Purkinje cells (n=7) from experiments as in (F). Peak synaptic current amplitudes evoked by the second pulse (PSC₂) were normalized to the peak of the first current (PSC₁). Points above and below the dotted line indicate short-term facilitation and short-term depression, respectively. Note x-axis break with final point at 1000 ms. The time course of recovery from facilitation for EPSCs was fit with a single exponential (red line) with τ = 69 ms.

H) The current associated with extracellular mossy fiber response (black) recorded in vivo from the CbN responses to all first puffs, after excision of CbN spikes. The asterisk (*) indicates the early transient triggered by the sensory stimulus. The corresponding grand mean PSTH of all CbN responses to the first puff in the experiments of Figure 2A is superimposed for comparison (red).

I) Averaged mossy fiber responses for each train, as in (H), with ISIs as labeled. Asterisks (*) denote the early transient.

J) Relative amplitudes of transient mossy fiber responses from data in (I). Peak responses were normalized to the first response in each train. Note that the response to the second stimulus in the 25-ms train is lost in the late component of the mossy fiber signal and is therefore absent.

Of particular interest, however, was that optogenetic activation of granule cells decreased spiking in over half the population (58%) of the affected Purkinje cells (“opto-suppressed” n=26, Figure 4A, right; Figure 4B, left, Figure S1C, right,). Their maximum cell-by-cell Δspk_prob was −0.086 ± 0.005 (−0.054 in the population PSTH, Figure 4B, right). Optogenetic stimulation did not evoke a measurable whisker response, nor did it affect complex spiking (Figure 4C). Together, these results demonstrate that disynaptic inhibition from MLIs can indeed curtail monosynaptic excitation from granule cells sufficiently to suppress simple spikes with minimal preceding spike probability elevation. Thus, the I/E ratio in Purkinje cells of resting mice can effectively invert the sign of granule cell output.

Parallel fiber synapses onto Purkinje cells, however, facilitate strongly (Atluri and Regehr, 1996, Dittman et al, 2000, Brown et al., 2024), suggesting that I/E ratio is likely to change with repetitive stimuli. To quantify the extent and time scale of changes in the I/E ratio, we measured short-term facilitation and recovery in vitro in cerebellar slices. Whole-cell recordings were made from Purkinje cells voltage-clamped at −70 mV or 0 mV to measure parallel-fiber-mediated EPSCs or MLI-mediated IPSCs, respectively, evoked by trains of electrical stimuli applied to parallel fibers at the same intervals tested in vivo (Figure 4D). Quantifying the degree of facilitation or depression relative to the first response (Figure 4E) demonstrated that, with 200-ms intervals, EPSCs depressed only slightly by the final pulse (n=11 cells). With 100-, 50-, and 25-ms intervals, facilitation of EPSCs was evident, and became much stronger with more rapid stimulation, reaching an increase of 87.2% on the last three pulses with the 25-ms interval. In the same cells with the same afferent stimulation, however, IPSCs remained within 22.2% of the initial response regardless of stimulus number or frequency. Together these data suggest that in the in vivo experiments, in the subset of granule-MLI-Purkinje microcircuits that do not drop out over the course of the train, the I/E ratio likely decreases in target Purkinje cells.

To examine the time course of recovery from short-term plasticity, pairs of electrical stimuli were delivered to parallel fibers with intervals ranging from 20 to 200 ms (Figure 4F). The first pulse induces plasticity, and the second pulse reveals how much facilitation or depression remains. With a 20-ms interval, the second pf-EPSC was more than double the amplitude of the first (EPSC₂/EPSC₁ = 2.16 ± 0.13, n=7, Figure 4G). With increasing interstimulus intervals, the second EPSC gradually decreased, recovering almost to baseline in 200 ms (EPSC₂/EPSC₁ = 1.2 ± 0.02; τ_recovery = 69 ms, Figure 4G). In contrast, IPSCs in the same cells showed little plasticity over the same intervals, staying within 15% of the initial pulse over the range of tested interstimulus intervals (IPSC₂/IPSC₁: 20-ms ISI, 0.86 ± 0.07; 200-ms ISI, 0.99 ± 0.06, Figure 4G). Thus, the differential short-term plasticity of parallel fibers and MLIs likely reinforces and/or intensifies the reduction in Purkinje simple spike suppression by repeated sensory stimuli.

Downstream CbN cells may also experience a change in I/E ratio resulting from synaptic inhibition from Purkinje cells and excitation from mossy fibers. To estimate the strength of excitation to CbN cells by mossy fiber input on successive puffs in vivo, we examined the loose cell-attached records after digitally removing action potentials (Materials & Methods), which reveals the underlying mossy-fiber dependent synaptic response (Brown and Raman, 2018). Each tactile stimulus activated a response with two components: first, a brief transient, and second, a slowly rising and slowly decaying phase (Figure 4H). The transient, which is appropriately timed to drive the Δspk_prob in CbN cells, depressed with successive stimuli, and the extent of depression increased as the interstimulus interval was shortened (Figure 4I, 4J). The decrement in excitation and the reduction in disinhibition thus likely combine to generate the decrease in CbN output with repeated puffs.

Detection of changes in sensory stimuli.

If adaptation of cerebellar output during repeated sensory input indeed results from short-term plasticity, rather than intrinsic refractory periods or spike frequency accommodation, then cerebellar responses might be restored by changes in sensory input that activate distinct synapses. In this context, a relevant observation is the high proportion of puff-responsive crus I/II Purkinje cells when stimuli are applied only ipsilaterally in some studies and only contralaterally in others (Brown and Raman, 2018; Zempolich et al., 2021; Brown et al., 2024), suggesting that some cells have bilateral receptive fields. Information from such widely separated regions of the sensory periphery seems likely to be carried through distinct mossy fiber pathways, possibly activating different populations of granule cells. If the parallel fibers relaying such information converge onto common MLI or Purkinje cell targets, then output patterns might be altered by changes of stimulation site.

To investigate this idea, we voltage imaged the responses of MLIs to air puffs applied to the contralateral and/or ipsilateral whisker pad (Figure 5A). The majority of MLI neurons (n=139/175, 79%) responded to contralateral as well as to ipsilateral puffs, indicative of bilateral receptive fields (Figure 5B). In these cells, simultaneous application of puffs to both sides elicited responses that were larger than puffs applied to either side alone (peak Δspk_prob relative to baseline: contra, 0.175 ± 0.009, ipsi, 0.167 ± 0.008, combined = 0.246 ± 0.012, ipsi vs contra, p=0.39, ipsi or contra vs. combined p<0.001; Figure 5C, Figure 5E). The greater response to the combined puffs could arise either from unilaterally sensitive granule cells converging at the level of MLIs, or from converging ipsilateral and contralateral mossy fibers activating bilateral granule cells with higher probability.

Figure 5. MLI responses to change of stimulus location.

A) Schematic of experimental setup. During voltage-imaging of MLIs in crus I, puffs were applied to the contralateral or ipsilateral whisker pad, separately or simultaneously, and the ipsilateral whiskers were tracked.

B) Identification of unilaterally and bilaterally sensitive MLIs. Maximum z-score responses of each MLI (n=175) to the ipsilateral vs. the contralateral puff. Dotted lines, cutoff for p<0.05. Of 156 contralaterally responsive and 150 ipsilaterally responsive cells, 139 cells responded to both sides.

C) Left, Population PSTH of bilaterally responsive MLIs (n=139) to puffs applied contralaterally (purple bars) or ipsilaterally (black lines). Right, MLI response to simultaneous (“combined”) contralateral and ipsilateral stimulation.

D) Top panels, stimulus protocols indicating puff locations for 4 puffs with 25-ms ISIs. In all cases, the first 3 puffs were contralateral, and fourth puff location varied. Bottom panels, Population PSTHs of bilaterally responsive MLIs when the location of the fourth puff was contralateral (left), ipsilateral (middle), or combined (right).

E) Mean ± SEM values for change in spike probability for each of the four puffs when the location of the fourth puff was contralateral (open circles), ipsilateral (closed circles) or combined (open triangles). Responses to single stimuli that were ipsilateral (Black open circle) or combined (orange circle) are plotted as single points.

As above, when trains of air puffs were applied with 25-ms intervals to the contralateral whisker pad, the transient increase in spike probability decreased to 39% over successive stimuli (Figure 5D, left). The train was then repeated, but with the fourth puff applied ipsilaterally instead of contralaterally (Figure 5D, middle). The switch was sufficient to restore the transient to 71% of its initial value, i.e., far higher than observed with the fourth contralateral puff (contra vs ipsi, Δspk_prob, 0.069 ± 0.004 vs. 0.122 ± 0.006, p<0.001, Figure 5D, middle, Figure 5E). When the fourth stimulus consisted of a combined puff, the transient response recovered to 77% (Δspk_prob, 0.137 ± 0.006). This value was again greater than the response to a fourth contralateral puff (p<0.001) and had a slightly higher amplitude than the ipsilateral puff applied alone (p=0.02; Figure 5D, right, Figure 5E). The restoration of the transient response with ipsilateral input demonstrates that MLI spiking did not fatigue with repetitive stimulation, but that synaptic depression likely occurred upstream in the circuit generating the contralateral response.

We then repeated this experiment recording the responses of crus I/II Purkinje cells. With loose-cell-attached recordings, 69/78 cells (88%) responded with either simple spike suppressions or elevations to puffs to at least one location; 38/69 cells (55%) were bilaterally sensitive regardless of response polarity, with 13/69 cells (19%) reliably suppressing to both sides (Figure 6A). During recordings from these 13 Purkinje cells (Figure 6B, Figure 6C), trains of contralateral air puffs applied with 25-ms intervals again showed reduced suppressions (Figure 6D), but switching to an ipsilateral fourth puff restored the suppression to 119% of the first response, (Figure 6E, left). Likewise, when the fourth stimulus was a combined puff, suppression was restored to 110% (Figure 6E, right). Quantifying the suppressions on a cell-by-cell basis showed the differences in suppression on the fourth puff (Figure 6F, Δspk_prob, contra vs ipsi, −0.025 ± 0.079 vs. −0.075 ± 0.018, p=0.01, vs. combined, −0.069 ± 0.013, p=0.001). Thus, while the subset of Purkinje cells with lateralized receptive fields may encode information about the location of the input, in cells with bilateral receptive fields, the suppression signal would not convey what input occurred but simply that an input arrived. Such bilaterally sensitive Purkinje cells thus appear well-suited to be event detectors, responding to changes in sensory patterns without specifying stimulus characteristics.

Figure 6. Purkinje cell responses to change of stimulus location.

A) Identification of unilaterally and bilaterally sensitive Purkinje cells. Greatest negative z-score responses of each Purkinje cell (n=78) to the ipsilateral vs. the contralateral puff. Black dotted lines, zeroes; red dotted lines, cutoff for p<0.05 for suppression.

B) Heat maps of z-scores calculated from PSTHs of simple spikes for each of the 13 bilaterally sensitive Purkinje cells that were suppressed by puffs applied contralaterally (top) and ipsilaterally (bottom).

C) Population PSTHs of simple spikes from the heat maps in (B) for bilaterally sensitive Purkinje cells in response to the contralateral puff (blue bars) or ipsilateral puff (black line).

D) Top, stimulus protocol indicating times of 4 contralateral puffs with 25-ms ISIs. Bottom, Population PSTHs of bilaterally responsive Purkinje cells in response to contralateral stimulation.

E) As in (D) when the location of the fourth puff was switched to be ipsilateral (left) or combined (right).

F) Mean ± SEM values for change in simple spike probability for each puff in the train when the location of the fourth puff was contralateral (open circles), ipsilateral (closed circles), or combined (open triangles).

G) Population PSTH of complex spikes in Purkinje cells, as in the three experiments of (D) and (E), with the fourth puff was contralateral (yellow), ipsilateral (black line), or combined (orange line). The response to the ipsilateral puff (grey line) alone (from the same cells) is included for comparison with the fourth puff.

Bilateral sensitivity of Purkinje cells at the level of complex spikes was more widespread than that measured for simple spikes (n=57/78 cells, 73%). Ipsilateral puffs, however, yielded peak complex spike probabilities that were nearly twice that of contralateral responses (Δspk_prob, 0.053 ± 0.004 vs. 0.030 ± 0.002, p<0.001, Figure 6G). When the stimulation site was switched from contralateral to ipsilateral on the fourth puff in the train, or both were given together, the change in complex spike probability to the final puff was 0.019 ± 0.003 (ipsi) and 0.019 ± 0.002 (combined), as compared to 0.003 ± 0.0005 (p<0.001) with continued contralateral stimulation. The magnitude of the peak response probability to an ipsilateral puff after previous contralateral stimulation, however, was only 35.2% of the response amplitude to an ipsilateral puff applied alone (Figure 6G), which may reflect constraints on the maximal firing rate of the single afferent climbing fiber (Hashimoto et al., 2009, Busch and Hansel 2023). Consequently, complex spikes may be limited as detectors of stimulus changes on time scales of less than 100 ms.

Effects of stimulus changes in CbN cells.

Lastly, we recorded from CbN cells during stimulus trains, varying the location of the fourth puff. Nearly all CbN cells (N=39/40) strongly increased their firing regardless of whether the puff was applied contralaterally or ipsilaterally (Figure 7A, 7B, 7C). The prevalence of bilateral responses may arise from convergence of Purkinje cells with distinct receptive fields (e.g., Herzfeld et al., 2015) as well as from direct innervation of individual CbN cells by both ipsilateral and contralateral mossy fibers.

Figure 7. CbN cell responses to change of stimulus location.

A) Identification of unilaterally and bilaterally sensitive CbN cells. Maximum z-score response of each CbN cell (n=40) to the ipsilateral vs. contralateral puff.

B) Heat maps of z-scores calculated from PSTHs of spikes for each of the 39 bilaterally sensitive CbN cells that were elevated by puffs applied contralaterally (top) and ipsilaterally (bottom).

C) Population PSTH of bilaterally sensitive CbN cells (n=39) in response to the contralateral (red bars) or ipsilateral puff (black line).

D) Top, stimulus protocol indicating times of 4 contralateral puffs with 25-ms ISIs. Bottom, Population PSTHs of bilaterally responsive CbN cells in response to contralateral stimulation.

E) As in (D) when the location of the fourth puff was switched to be ipsilateral (left) or combined (right).

F) Mean ± SEM values for change in spike probability for each puff when the location of the fourth puff was contralateral (open circles), ipsilateral (closed circles) or combined (open triangles).

G) Average whisker movements, aligned to the time of the first puff, when the location of the fourth puff was contralateral (grey), ipsilateral (black), or combined (orange).

As above, when trains of air puffs were applied to the contralateral whisker pad, the transient increase in spike probability decreased over successive stimuli (Figure 7D). The side-switching experiment gave results comparable to those in MLIs and Purkinje cells: When an ipsilateral fourth puff followed three contralateral puffs, a large transient was restored (76% of the initial response, Figure 7E left), which was significantly greater than that evoked by a fourth contralateral puff (Δspk_prob, contra vs ipsi, 0.161 ± 0.018 vs. 0.316 ± 0.031, p<0.001). A similar result was obtained with the combined fourth puff, which reached 82% of the initial response (Δspk_prob, 0.358 ± 0.032; fourth puff, contra vs. combined, p<0.001; Figure 7E, right, Figure 7F). Thus, like bilaterally sensitive Purkinje cells, CbN cells effectively respond to changes in tactile input.

The amplitudes of associated whisker protractions and CbN cell output were also correlated; with switches in stimulus site, forward whisks were enlarged, even when the whisker was already protracted (Figure 7G). Since CbN output is required for reflexive whisks to reach their full amplitude (Brown and Raman, 2018), this result suggests that the restoration of CbN output contributes to the strengthened motor response upon introduction of a distinct stimulus into a chain of repetitive sensory events.

Discussion

These results provide insight into how sensory signals are transformed by short-term plasticity, convergence, and divergence at several cerebellar synapses. Air puffs applied to the whisker pad reliably elicit brief increases in spike probability in MLIs, which transiently inhibit Purkinje cells. CbN cells are simultaneously disinhibited by the synchronously silenced, converging Purkinje cells and briefly excited by mossy fibers, favoring whisker protraction. During short regular trains of tactile input with 25- to 200-ms intervals, the transient sensory responses of Purkinje cells and target CbN cells decrease, contributing to reduce the size of whisker protractions evoked by repetitive stimulation. Unlike sensory puff-evoked spiking, which is well-timed, motor command-related information associated with whisking is relayed as broadly timed changes in spike rate. These relatively low-frequency signals are progressively reduced at successive levels of the circuit, such that CbN cells exhibit only transient sensory responses. These firing patterns are consistent with different forms of feedforward inhibition at each stage of processing, which offer a mechanistic explanation for the adaptation of motor output to repeated sensory inputs. A subset of Purkinje cells and nearly all MLIs and CbN cells had bilateral receptive fields. After adaptation induced by repetitive puffs, switching stimulus location to another region of the receptive field restored sensory responses and whisk magnitude, suggesting that neuronal signals are tuned to detect and respond to stimulus changes, even without specifying what or where the stimulus may be.

Short-term plasticity in cerebellar circuits.

Short-term plasticity throughout the cerebellum has been described in vitro (Xu-Friedman and Regehr, 2003; Saviane and Silver, 2006; Chabrol et al., 2015; Wu and Raman, 2017; Atluri and Regehr, 1996; Dittman et al., 2000; Beierlein et al., 2007; Grangeray-Vilmint et al., 2018; Dorgans et al., 2019; Brown et al., 2024; Telgkamp and Raman, 2002; Telgkamp et al., 2004; Najac and Raman, 2015). Although synaptic properties between certain cell types are consistent (e.g., parallel fiber synapses onto Purkinje cells reliably facilitate), plasticity profiles can vary widely even between the same classes of cells. Such variance is sometimes lower within relatively homogeneous subsets of heterogeneous afferents (Chabrol et al., 2015), suggesting that plasticity patterns relate to the informational content of the synapses.

How these patterns combine in vivo depends on interactions within microcircuits, many of which include feedforward inhibition. Feedforward pathways, in which a presynaptic neuron excites a postsynaptic target and also branches to stimulate an intervening interneuron that inhibits the target, often curtail, time, or otherwise “sculpt” net excitatory signals (Buzsáki 1984; Lawrence & McBain, 2003). Such patterns are seen in Purkinje cells in vitro (Mittmann et al., 2005) and in vivo as spike probability increases in ~50% of cells sensitive to air puffs (Brown et al., 2024).

By contrast, in ~50% of responsive Purkinje cells in vivo, tactile stimulation suppresses simple spikes without preceding spike probability elevation. Thus, the I/E ratio in individual Purkinje cells can strongly favor inhibition, effectively inverting the sign of the microcircuit. Given the strong short-term facilitation of parallel fibers and minimal plasticity of MLIs, inhibition likely dominates in microcircuits with the least active granule cells (Brown et al., 2024), e.g., in resting mice, which are not engaging in overt cerebellar (crus I) behavior. With repeated stimulation, however, facilitation tips the balance away from inhibition. Short-term plasticity within these microcircuits can thereby produce the strongest spike suppression at train onset.

Such differentiation, or high-pass filtering, can also be achieved through synaptic depression (Abbott et al., 1997; Abbott and Regehr, 2004; Silver 2010), spike frequency accommodation (e.g., Zhang and Trussell, 1994; Brew and Forsythe, 1995; Rathouz and Trussell. 1998), and sensory adaptation (Müller et al., 1999; Cook et al., 2003; Li and Glickfeld, 2023). At synapses onto Purkinje cells, however, the onset responses are inhibitory rather than excitatory, setting the stage for downstream disinhibition, and repetitive stimulation can even act as a toggle switch, converting the sign of the net feedforward input from inhibition to excitation.

Although the microcircuit culminating on each Purkinje cell independently functions as a differentiator, upstream factors intensify this effect across the population. The corresponding circuit targeting granule cells (with presynaptic mossy fibers and Golgi cells) generates net excitation, manifested here as the puff-evoked transient increase in MLI spike probability. This transient diminishes with successive sensory stimulation. Since parallel fiber-to-MLI synapses are reliable and facilitating (Carter and Regehr, 2002; Nahir and Jahr, 2013; Dorgans et al., 2019), the decreasing response likely does not reflect insufficient excitatory input to MLIs. Nor are MLIs likely to be actively suppressed, since MLI-MLI inhibition occurs with a delay (Häusser and Clark, 1997). Instead, synaptic depression of mossy fibers may lead fewer granule cells to fire with repetitive stimulation (Chabrol et al., 2015).

Additionally, Golgi cells, which produce well-timed responses to mossy fiber stimulation in cerebellar slices, may time granule cell spikes through classical feedforward inhibition (Kanichay et al., 2008). In anaesthetized rats, however, Golgi-mediated inhibition resulting from somatosensory stimulation can actually jitter excitation-driven spikes in granule cells (Duguid et al., 2015). Moreover, the long (50-ms) afterhyperpolarizations of Golgi cells limit both the rate and precision of repetitive firing of granule cells (Forti et al., 2006). These effects, along with feedback from granule cells to Golgi cells, may further disperse Golgi spike timing and thereby prolong inhibition of target granule cells. As granule cells are inhibited, fewer MLIs are activated, and the proportion of Purkinje cells with simple spike suppression will also fall, reducing synchronous suppression of Purkinje cells across the population.

The circuits involving CbN cells, at which mossy fibers and Purkinje cells converge, are typical in that a puff stimulus activating mossy fiber afferents is net excitatory, evoking spikes. The nominally inhibitory Purkinje-cell leg of these circuits, however, is briefly disinhibitory at sensory onset owing to the upstream sign-inversion, such that the two branches of the feedforward circuit reinforce one another. (Notably, despite short-latency excitation in half the afferent Purkinje cells, puff-evoked transient decreases of firing probability in CbN cells were not observed). As Purkinje disinhibition decreases over the course of repeated stimuli, mossy fiber-mediated excitation falls as well, further sharpening the distinction between the initial and repeated puffs. Since CbN output augments whisker movements (Brown and Raman, 2018; Brown et al., 2025), this changing output is consistent with a cerebellar role in adaptation of whisking reflexes.

Differential filtering of sensory and motor components

Parallel fibers also carry motor command-related signals (Stone and Lisberger, 1990; Powell et al., 2015; Chen et al., 2016, 2017; Knogler et al., 2017; Giovannucci et al., 2017; Harmon et al., 2017; Brown and Raman, 2018; Wagner et al., 2019). In MLIs, a slow rise and fall of spike probability, reflective of non-synchronous firing across the population, is evident 50–100 ms post-puff, in association with whisker movement. Consistent with convergence of sensory and motor streams, individual MLIs show both slow firing rate changes as well as brief, well-timed puff responses (Brown et al., 2025). Interestingly, although the well-timed spikes evoke the expected postsynaptic inhibition, the same Purkinje cell targets increase their spike rates between 50 and 100 ms. Thus, Purkinje cell spiking is reliably suppressed during the transient but not the sustained MLI response. These results suggest that motor-related, parallel fiber inputs onto neighboring Purkinje cells are sufficiently numerous, strong, and/or facilitated to override the substantial inhibition from MLIs. The summed parallel fiber excitation and MLI inhibition, dispersed over the time course of the whisker protraction, thus may contribute to the cancellation associated with cerebellum-like circuits (Marr 1969; Bell 1986; Requarth and Sawtell, 2011; Brooks and Cullen, 2013).

Distinct sensory and motor-related responses are also present in CbN cells, which show a small rate increase 50–100 ms post-puff during the first whisker protraction. Their PSTHs thereby resemble those of MLIs, but with a greater disparity between the peak magnitudes of the transient and broadly timed responses. With repeated stimulation, dwindling transient responses in MLIs and Purkinje cells ride atop prolonged spike rate increases. Strikingly, this motor-related slow component is nearly absent in CbN cells, such that successive stimuli elicit only transient spiking. Plasticity, convergence, and population temporal codes thus yield effective high-pass filtering even when low-frequency signals are large, through amplification of sensory-evoked transients and cancellation of slow motor-command-related changes in firing rate.

Bilateral receptive fields.

A subset of Purkinje cells and most MLIs and CbN cells had simple spike receptive fields comprising both whisker pads (Bower and Woolston, 1983; Holtzman et al., 2006; Brown and Raman, 2018), raising the question of what computational capacity is gained upon loss of stimulus specificity. The present data show that these bilaterally sensitive neurons effectively transmit switches in the location of sensory input within their large receptive fields. Thus, they can operate as event detectors, identifying change without encoding what that change may be. Whether specificity of movements can be generated by such mechanisms remains uncertain, but for the relatively simple stimuli and responses studied here, such circuits appear well-suited for the rapid detection of and reaction to new inputs.

Significance Statement.

Neural coordination of rapid motor responses to new sensory information entails adaptation to repeated stimuli while retaining the ability to respond to stimulus changes. Here, we measured action potentials of several types of neurons that form the cerebellar circuit while applying puffs of air to the mouse whisker pad that evoke whisker movements. The results reveal patterns of activity that intensify responses to temporal and spatial changes in sensory input and attenuate responses to repetitive stimulation, in a way that correlates with movement magnitude. The combined effect of convergence, divergence, and short-term synaptic changes across several synapses ultimately permits the cerebellum to detect sensory changes and rapidly adjust movements within tens of milliseconds.