Prolonging the postcomplex spike pause speeds eyeblink conditioning

Abstract

Climbing fiber input to the cerebellum is believed to serve as a teaching signal during associative, cerebellum-dependent forms of motor learning. However, it is not understood how this neural pathway coordinates changes in cerebellar circuitry during learning. Here, we use pharmacological manipulations to prolong the postcomplex spike pause, a component of the climbing fiber signal in Purkinje neurons, and show that these manipulations enhance the rate of learning in classical eyelid conditioning. Our findings elucidate an unappreciated aspect of the climbing fiber teaching signal, and are consistent with a model in which convergent postcomplex spike pauses drive learning-related plasticity in the deep cerebellar nucleus. They also suggest a physiological mechanism that could modulate motor learning rates.

Identifying the cellular mechanisms responsible for memory formation is a central challenge in neuroscience. In the cerebellum, the formation of associative motor memories is thought to involve changes in circuit function in at least two locations: the cerebellar cortex and the deep cerebellar nuclei (DCN) (1–4). It has also been suggested that memory formation occurs in stages, with plasticity occurring first in the cortex followed by plasticity in the DCN (5–7). These sites within the circuit have been hypothesized to encode distinct aspects of motor memories, with the DCN necessary for the expression and long-term storage of the memory (8–12) and for savings (13), whereas circuits in the cortex contribute to more refined features such as the timing of learned movements (14–18).

Activity in climbing fibers (CFs), the axonal projection of neurons in the inferior olive (IO), has been proposed to serve as a teaching or error signal for the formation of associative, cerebellum-dependent memories (19–22). In a variety of systems, CF activity has been shown to encode unexpected events and errors in movement, information appropriate for instructing circuit changes during motor learning (23–26). Also consistent with a teaching role are the observations that direct electrical stimulation of IO/CFs can substitute for (27), that inhibition and lesion of the IO/CFs can prevent associative learning (28), and that lesions of the IO in trained animals result in extinction with continued paired training (29). This evidence strongly supports the view that the IO/CF is the essential teaching or reinforcing system for learning of all discrete responses with an aversive stimulus (30). However, despite extensive support implicating CFs in cerebellar learning, we lack a mechanistic understanding of how CFs coordinate circuit plasticity at a cellular level. For example, it is not known whether CF activity serves as an instructive signal that drives the sorts of associative plasticity hypothesized to occur in the DCN (2, 3).

Here, we describe experiments that focus on an underappreciated aspect of the CF response in Purkinje neurons (PNs). CF input to PNs generates an extremely large, excitatory postsynaptic depolarization throughout the PN dendrite, which in turn triggers a burst response termed the complex spike. Under baseline conditions, PNs intrinsically fire at extremely regular rates averaging around 50 spikes per second (31, 32). Thus, complex spikes occur within this ongoing spontaneous activity and when they occur they lead to a transient pause termed the postcomplex spike pause (post-CxSp pause). Considering that converging PNs tonically inhibit DCN neurons, we hypothesize that CF-induced pauses in the firing of groups of PNs may represent a CF-dependent teaching signal in the DCN. To test this idea, we identified two different pharmacological agents that can prolong the post-CxSp pause. When applied during training in a classical eyelid conditioning paradigm, both of these agents lead to an increased rate of associative learning. These results support an important role for the post-CxSp pause in transmitting learning-relevant information to the DCN, and provide a potential therapeutic target site for pathologies presenting deficits in cerebellar learning and function.

Results

Two Independent Pharmacological Strategies for Prolongation of the Postcomplex Spike Pause.

Small-conductance, calcium-activated potassium channels (SK channels) are activated by the large calcium influx that occurs during a complex spike (33), and it has been demonstrated that the extent of dendritic calcium spiking influences the duration of the post-CxSp pause, an effect most likely mediated by activation of SK channels (34). We thus hypothesized that enhancement of SK channel function by 1-EBIO, a positive modulator of calcium activated potassium channel function, would prolong the post-CxSp pause in PNs (35, 36). To noninvasively monitor complex spikes and firing, we made extracellular recordings from single PNs in acute slices in the presence of the GABA_A receptor antagonist picrotoxin (100 μM). Stimulation of the single climbing fiber input to the PN (0.125 Hz, stimulation currents just above threshold) elicited complex spikes and transient post-CxSp pauses in simple spike firing and bath application of 20 μM 1-EBIO prolonged these pauses (Fig. 1 A and B). The mean pause duration before 1-EBIO application was 32 ± 9 ms, and 1-EBIO increased the pause duration by 53 ± 8% (48 ± 34 ms, n = 9, P < 0.005). Concluding each experiment, application of the AMPA-receptor antagonist DNQX (20 μM) confirmed that the complex spike-associated pauses were due to actions of the excitatory CF (Fig. 1 A and B).

Fig. 1.

1-EBIO and ZD7288 prolong postcomplex spike pauses. (A) Example traces showing single trial responses to CF stimulation (arrow) under the indicated conditions. (B) Spike rasters aligned to the CF stimulus (arrow) demonstrating control responses in control (red) and in 1-EBIO (blue). The hashmarks indicate a 13.3-min gap between control and 1-EBIO conditions. DNQX (black) was applied at the end of the experiments to confirm that all effects were due to CF input. (C) Average cumulative net spike counts (see Methods) for six PNs indicate that significantly more spikes are “lost” in response to the CF input in 1-EBIO (blue) versus control (red). Responses are blocked entirely by DNQX (black). Light shading represents SEM. (D) Changes in net spike count per CF stimulus measured at t = 400–600 ms after the CF stimulus. (E) As in C, average cumulative net spike counts for six PNs exposed to ZD7288 (green) compared with control (red) and DNQX (black) conditions. (F) Steady-state changes for ZD7288 experiments. Asterisks denote P values of P < 0.0001.

To capture the complicated dynamics with which PNs can resume regular spiking after a complex spike, we quantified the prolongation of post-CxSp pause by calculating the net number of “spikes lost” in response to each complex spike (see Methods). This analysis demonstrated that 1-EBIO significantly lengthened the post-CxSp pause (3.85 ± 1.35 spikes versus 1.70 ± 0.58 spikes in control, P < 0.0001, n = 6). Effects for 1-EBIO are summarized in a peri-stimulus time plot indicating the average number of spikes lost per CxSp across all PNs (Fig. 1C). This result is also reflected in a histogram quantifying the mean number of spikes lost per CxSp at 800–900 ms after the CxSp stimulus (Fig. 1E). These analyses indicate that no net change in spike count was observed in the presence of DNQX confirming that the loss of spikes was due to the CxSp (Fig. 1 C and E).

We conducted additional measurements to show that 1-EBIO has specific effects on the post-CxSp pause. 1-EBIO has no observable effect on the complex spike waveform, as neither an analysis of the number of spikelets per complex spike (control, 3.5 ± 0.29; 1-EBIO, 3.75 ± 0.48; P = 0.64) nor of the complex spike duration (control, 4.5 ± 0.6 ms; 1-EBIO, 5.2 ± 0.9 ms; P = 0.21) showed significant differences. We also determined that 1-EBIO has only moderate effects on baseline PN spiking rate, reducing it from 47 ± 0.7 Hz in control to 40 ± 5.5 Hz. We attribute this moderate slowing to the fact that regular CF input is likely to elevate basal calcium levels slightly. Prior work has found that, under some conditions, 1-EBIO has no effect on basal spike rate (36), but in others mildly reduces firing rate in the absence of CF input (37).

In an effort to identify an additional compound that we could use to prolong the post-CxSp pause we turned to a reagent that we expected would achieve the same cellular effects, but by a mechanism distinct from 1-EBIO. Hyperpolarization-activated cation (HCN) channels are highly expressed in cerebellar PNs and both HCN1 gene deletion and blockade of HCN1 channels leads to bistable firing behavior in PNs, characterized by alternating periods of normal pacemaking and periods of silence (38, 39). Such observations suggest that upon strong hyperpolarization, PNs rely on HCN1 channels to help resume the more depolarized membrane potential at which they pacemake. With this in mind, we evaluated low concentrations of ZD7288, an antagonist for HCN channels. We chose a concentration of ZD7288 (1 μM) that we determined in separate experiments to cause ∼50% inhibition of the HCN-mediated current (I_h). As summarized in Fig. 1 E and F, ZD7288 prolonged the length of the post-CxSp pause in every recorded neuron and resulted in significantly longer post-CxSp pauses (mean pause duration in control was 141 ± 35 ms and in ZD 7288 was 305 ± 82 ms reflecting a 216 ± 59% increase in duration, P < 0.0075, n = 6). This was reflected in more spikes “lost” in the cumulative peri-complex spike histogram (11.4 ± 3.2 spikes, versus 2.06 ± 0.79 spikes in control, P < 0.001, n = 6; Fig. 1 D and F). The more prolonged phase over which spike counts decrease in Fig. 1D is reflective of the fact that ZD7288 led to variable pause duration in different cells and this panel is an average of six PNs some with shorter and some with longer pauses. This concentration of ZD7288 did not affect baseline firing frequency (control: 27.1 ± 4.2 Hz; ZD7288 23.1 ± 4.6 Hz), nor did it affect the complex spike waveform (number of spikelets in the complex spike: control, 2.75 ± 0.5; 1-EBIO, 3.25 ± 0.5; P = 0.5; complex spike duration: control, 4.7 ± 0.67 ms; 1-EBIO, 4.5.2 ± 0.72 ms; P = 0.9). Taken together, these in vitro experiments validate two mechanistically distinct pharmacological strategies for prolonging the post-CxSp pause that can be used to test the how manipulations of such signals affect behavior.

Infusion of Each Drug During Eyeblink Training Results in Faster Acquisition of Conditioned Eyeblink.

To assess the effects of prolonging post-CxSp pause on associative cerebellar learning, we tested the influence of 1-EBIO and ZD7288 in an eyeblink conditioning paradigm. Rats were trained in a classic 300 ms-delay conditioning paradigm (Fig. 2A), using a coterminating tone as the conditioning stimulus and a peri-orbital shock as the unconditioned stimulus. Ten minutes before training each day we injected 1 μL of either 1-EBIO (n = 12), ZD7288 (n = 8), or a saline control (n = 8) into lobule HVI of the cerebellum via a surgically implanted cannula. Electrodes implanted into the obicularis oculi muscle relayed electromyogram signals that corresponded to eyelid muscle movements allowing us to monitor eye closure (Fig. 2B).

Fig. 2.

Experimental parameters for conditioning experiments. (A) (Upper) The learning schedule consisting of 2 d of habituation during which EMGs were recorded but no stimulus delivered, 6 d of acquisition (90 paired US–CS trials and 10 CS-alone trials delivered in blocks of 9 and 1), and 3 d of extinction during which only tone was delivered. (Lower) The trial structure consisting of a 300-ms, 2-KHz tone whose onset precedes and offset coterminates with a 100-μs peri-orbital shock. Consecutive trials were separated by randomly generated 20- to 40-s intervals. (B) A raw and processed (see Methods) EMG signal from an acquisition trial showing the emergence of the conditioned response.

Training sessions consisted of 6 d of associative conditioning with 90 tone-plus-shock pairs and 10 tone-alone trials delivered each day. By day 6, all rats, regardless of the drug they received, reached the same performance plateau in terms of the percentage of conditioned response (59% 1-EBIO, 65% ZD7288, and 65% saline control, Fig. 3 A and B). During the first 4 d of training, however, both the rats receiving 1-EBIO and those receiving ZD7288 had significantly higher % conditioned response, as determined by repeated-measures ANOVA analyses (1-EBIO: days 1–3 P < 0.01, day 4 P < 0.05, and ZD7288: days 1–3, P < 0.05), compared with saline control animals. Early in the training paradigm, animals receiving either 1-EBIO or ZD7288 also showed a higher rate of acquisition of conditioned responses compared with the saline-injected control counterparts, an effect that was significant for 1-EBIO in the first 2 d of conditioning (Fig. 3C).

Fig. 3.

Infusion of 1-EBIO or ZD7288 results in faster acquisition of conditioned responses in associative eyelid conditioning. (A and B) Percent conditioned responses (CRs) for each day. Day 0 is a habituation day, days 1–6 are training (acquisition) days, and Ex1-3 are extinction days. Animals received either 1-EBIO (n = 12), ZD7288 (n = 8), or saline (n = 8) via an implanted cannula 10 min before training on acquisition days. Asterisks denote significant differences from control at P < 0.05, and double asterisks at P < 0.01 by repeated-measures ANOVA. (C) Mean rate of CR acquisition on each training day (% change per day) as a function of experimental group. The double asterisk indicates a significant difference from saline at P < 0.01 and the single asterisk at P < 0.05. (D) Example EMG recordings from the habituation day, acquisition days 2 and 6, and the last day of extinction.

Discussion

Our results identify an important contribution by the post-CxSp pause to the rate of cerebellum-dependent learning. We found that enhancement of this specific component of the CF response enhances learning. Classical cerebellar learning theory proposes that the CF input to the cerebellar cortex carries unconditioned stimulus (US) or “teaching” information, which triggers forms of circuit plasticity that underlie learning (3, 19, 21). However, at a cellular level, most attention has focused on only one aspect of CF-mediated signals, the complex spike burst in PNs. Complex spikes in PNs are believed to be essential for the induction of several forms of synaptic plasticity within PNs, including parallel fiber–PN long-term depression (3). Our findings suggest that a slower, inhibitory phase of the CF signal, the post-CxSp pause, plays a significant role in role in circuit plasticity.

Our data support the idea that CF-induced pauses in PN spontaneous firing may be important inductive signals for associative forms of plasticity in the deep cerebellar nuclei. It is known from in vivo imaging and multielectrode array measurements that temporally correlated complex spikes occur in parasagittally oriented modules of PNs within the cerebellar cortex (40–43), and recent evidence is consistent with the idea that the CF-associated teaching signal is encoded by temporally correlated complex spike activity (44). The axons of PNs are inhibitory and converge on single DCN neurons, thus prolonged post-CxSp pauses occurring synchronously in a module of PNs could provide an extended window of disinhibition to the DCN cells receiving such input (45, 46). This pattern of activity would have appropriate properties for driving long-term potentiation in mossy fiber-to-DCN synapses (47). Moreover, convergent post-CxSp pauses to DCN neurons would render CF teaching signals distinctive from spontaneous CF input or strong MF inputs, an important property for CFs to confer an unambiguous teaching signal.

By their nature, pharmacological manipulations raise concerns due to the possibility of nonspecific effects; in addition, HCN and SK channels are present on neurons other than PNs. The fact that both drugs had a similar effect on learning makes the possibility of nonspecific actions unlikely. In considering whether the drugs might act on other neurons within the circuit, it is important to realize that both compounds reduce rather than increase excitability and all forms of synaptic plasticity in the cerebellar cortex that have been described to date are triggered by increases in excitability (48). For these reasons, it is difficult to conceive of alternative mechanisms that would lead to increased learning.

Although we have targeted the drugs to the cerebellar cortex, it is possible that the DCN could be affected. Within the DCN, rebound excitability following hyperpolarization can trigger a robust increase in the strength of mossy fiber inputs (47) and HCN channels could, in principle, promote such rebound excitability. However, if HCN channels are involved, their inhibition would be predicted to impair plasticity. Thus, as for the cerebellar cortex, it is difficult to identify alternative mechanisms of action common to both 1-EBIO and ZD7288 that would explain our findings.

In summary, we have validated two pharmacological tools and have used them in an eyeblink conditioning paradigm to show that the duration of post-CxSp pauses in PN firing impact the rate of learning. These results provide additional insight into the cellular signals that drive learning in a canonical associative learning circuit and suggest a network basis for delivery of a teaching signal to the DCN, a long hypothesized site of circuit plasticity in cerebellar learning (2, 3).

Methods

All experimental procedures were approved by the University of California at Los Angeles (UCLA), Institutional Animal Committee on Use and Care (IACUC).

Electrophysiology.

Extracellular recordings in either voltage clamp (V_pipette = 0 mV) or current clamp (I_pipette = 0) mode were made with 2–5 MΩ resistance pipettes in mouse parasagittal cerebellar slices (200–300 μm). A bipolar theta stimulation electrode was used to directly stimulate single climbing fiber axons. All experiments were conducted at near physiological temperatures (32–34 °C) in an artificial cerebrospinal fluid (aCSF) composed of: 119 mM NaCl, 26 mM NaHCO₃, 11 mM glucose, 2.5 mM KCl, 2.5 mM CaCl₂, 1.3 mM MgCl₂, and 1 mM NaH₂PO₄ and saturated with 95% O₂ and 5% CO₂. In addition, the GABA_A receptor antagonist picrotoxin (100 μM) was added to the aCSF in all experiments. 1-EBIO (20 μM, a positive modulator of SK channels), ZD7288 [1 μM, an HCN (I_h) channel antagonist], and DNQX (20 μM, an AMPA-receptor antagonist) were obtained from Tocris Biosciences and bath applied as indicated.

Cannula and Eyelid Electrode Implant Surgery.

All sterotactic surgeries were carried out using aseptic techniques approved by the UCLA IACUC. Two- to 3-mo-old Sprague–Dawley rats were anesthetized using a ketamine (75–85 mg/kg) and xylazine (5–10 mg/kg) mixture delivered via an i.p. injection. Once anesthetized, the head was shaved and disinfected, and the animal was secured in a stereotaxic apparatus (David Kopf). Two stainless steel anchoring screws were secured to the skull, and a stainless steel cannula guide (Plastics One) was inserted through a small craniotomy into the left cerebellum lobule HVI (stereotaxic coordinates from Bregma: anterior–posterior = −10.85, medial–lateral = 3.0, dorsal–ventral = −4.0), and secured with dental cement. A homemade headstage composed of a plastic pedestal (Plastics One) with five stainless steel electrodes emerging was secured with a drop of dental cement, and four of the electrodes passed through the upper left eyelid muscle, with a small hook at the end to secure the electrode wire in the muscle. The fifth electrode was wrapped around one of the anchoring screws to serve as a ground. The skull was then covered with a layer of dental cement, and sterile suture was used to close the skin around the implants. Finally, triple antibiotic gel was applied the wound. A dummy cannula (Plastics One) was inserted and secured to the cannula guide to prevent clogging. Rats were given the analgesic Buprenex (0.05 mg/kg, i.m.) and the anti-inflammatory carpofen (5 mg/kg, i.m.), and a week to recover.

Delay Eyelid Conditioning.

On training acquisition days, 1 μL of either 20 mM 1-EBIO, 5 mM ZD7288, or 0.9% saline was microinjected using an 11Plus syringe (Harvard Apparatus) 10 min before beginning the session. The following stimulus parameters were used: conditioning stimulus (CS) was a 300-ms, 2-kHz tone, and the US was a 10-ms unipolar electrical stimulation that coterminated with the tone. The intertrial interval was randomized between 20 and 40 s. Training occurred over 11 d: 2 d of habituation (no stimuli), 6 d of acquisition training (CS–US paired trials), and 3 d of extinction (CS only). During training, nine CS–US paired trials preceded a CS-only trial, and this set of 10 trials was repeated 10 times for a total of 100 trials per day. Eye movements were measured by electromyogram (EMG) signals recorded by the electrodes implanted into the upper eyelid muscle during surgery. EMG signals were amplified at ×10K gain, digitized at 10 kHz, and filtered between 0.1 and 1 kHz.

Eyelid conditioning experiments were done in a custom built system, with some preliminary experiments completed in the laboratory of R.F.T. at the University of Southern California. These preliminary experiments involved infusion of ZD7288 into rabbit cerebellar lobule HVI with the experimenter blind to treatment and showed results similar to those described in Fig. 3 (49).

Data and Statistical Analyses.

Electrophysiology data were analyzed using Igor Pro (Wavemetrics) and Microsoft Excel. To determine the number of spikes “lost” during the post-CxSp pause, poststimulus time histograms (PSTH) were computed and integrated. A linear fit to the baseline range (a window from −100 to −5 ms relative to the CF stimulus) of the integral was extrapolated over the entire duration and subtracted from the integrated line to yield the corrected cumulative spike probability. From this, we extracted information about the net number of spikes “lost” during the post-CxSp pause (50). Averaged data were compared by using Student’s t test in Microsoft Excel.

Custom routines, written in LabView (National Instruments) were used for EMG acquisition and analysis. A small subset of trials (2% per animal on average) were excluded on the bases of (i) activity in a 400-ms pretone window or (ii) startle responses determined by above threshold activity within the first 100 ms following tone onset. Conditioned responses were counted when integrated EMG traces exceeded a threshold criterion in the last 200 ms of the conditioned stimulus. Repeated-measures ANOVA was done in Stata.