Activation of the climbing fiber input powerfully excites cerebellar Purkinje cells via hundreds of widespread dendritic synapses, triggering dendritic spikes as well as a characteristic high-frequency burst of somatic spikes known as the complex spike. To investigate the relationship between dendritic spikes and the spikelets within the somatic complex spike, and to evaluate the importance of the dendritic distribution of climbing fiber synapses, we made simultaneous somatic and dendritic patch-clamp recordings from Purkinje cells in cerebellar slices. Injection of large climbing fiber-like synaptic conductances at the soma using dynamic clamp was sufficient to reproduce the complex spike, independently of dendritic spikes, indicating that neither a dendritic synaptic distribution nor dendritic spikes are required. Furthermore, we found that dendritic spikes are not directly linked to spikelets in the complex spike, and that each dendritic spike is associated with only 0.24 ± 0.09 extra somatic spikelets. Rather, we demonstrate that dendritic spikes regulate the pause in firing that follows the complex spike. Finally, using dual somatic and axonal recording, we show that all spikelets in the complex spike are axonally generated. Thus, complex spike generation proceeds relatively independently of dendritic spikes, reflecting the dual functional role of climbing fiber input: triggering plasticity at dendritic synapses and generating a distinct output signal in the axon. The encoding of dendritic spiking by the post-complex spike pause provides a novel computational function for dendritic spikes, which could serve to link these two roles at the level of the target neurons in the deep cerebellar nuclei.
Adult Purkinje cells each receive input from a single climbing fiber (CF), which makes hundreds of synapses across its highly branched main dendrites (Palay and Chan-Palay, 1974). CF activation produces a distinctive high-frequency burst of spikes: the complex spike (Eccles et al., 1966; Fujita, 1968; Llinás and Sugimori, 1980b), which is thought to represent a critical signal for the operation of the cerebellar cortex, conveying both timing information (Welsh and Llinás, 1997) and triggering synaptic plasticity (Gilbert and Thach, 1977; Hansel et al., 2001; Ito, 2001). Understanding complex spike generation is therefore essential for understanding information processing and plasticity in Purkinje cells.
The mechanism of complex spike generation has been debated for decades (Schmolesky et al., 2002). It is known that, similarly to simple spikes (Stuart and Häusser, 1994; Clark et al., 2005; Khaliq and Raman, 2006), the first spike of the complex spike is generated in the axon (Stuart and Häusser, 1994); both backpropagate only weakly into the dendrites (Llinás and Sugimori, 1980b; Stuart and Häusser, 1994). Purkinje cell dendrites can, however, generate prominent dendritic calcium spikes, which can propagate toward the soma (Eccles et al., 1966; Llinás et al., 1968; Llinas and Nicholson, 1971). CF input activates dendritic spikes, often in a burst-like pattern (Fujita, 1968; Llinás and Sugimori, 1980b), suggesting an alternative view to the classical hypothesis that the somatic spikelets in the complex spike are generated in the axon (Granit and Phillips, 1956; Eccles et al., 1966). The difficulty in suppressing spikelets in the complex spike by antidromic spike activation (Martinez et al., 1971), together with the finding that secondary spikelets can be modulated independently of the first spike of the complex spike (Campbell et al., 1983), suggested that spikelets in the complex spike could reflect dendritic spikes. However, the fact that somatic bursts of spikes can be triggered without significant dendritic calcium entry, and that inhibition of dendritic spikes only has a weak effect on the somatic complex spike waveform (Callaway et al., 1995; Callaway and Ross, 1997), strongly argues against this interpretation. Nevertheless, the contribution of individual dendritic spikes to the complex spike and the origin of the spikelets within the complex spike remain unresolved. This is important because in other neuronal types, strong links can exist between dendritic spikes and axonal action potentials. In mitral cells and in hippocampal and cortical pyramidal cells, dendritic spikes can trigger one or more axonal action potentials (Chen et al., 1997; Stuart et al., 1997; Golding and Spruston, 1998; Larkum et al., 1999, 2001; Ariav et al., 2003). Moreover, backpropagating axonal action potentials can themselves promote dendritic spikes, a reciprocal interaction that can lead to a burst of axonal action potentials (Pinsky and Rinzel, 1994; Mainen and Sejnowski, 1996; Larkum et al., 1999, 2001; Doiron et al., 2002; Turner et al., 2002; Williams and Stuart, 2002). We have therefore used simultaneous somatic and dendritic or somatic and axonal recording to directly investigate the link between dendritic, somatic, and axonal events during complex spike generation in Purkinje cells.
Materials and Methods
Parasagittal slices (200–300 μm) of postnatal day 18 (P18)–P24 rat cerebellum were made according to standard techniques (Geiger et al., 2002; Davie et al., 2006). All procedures were approved by the United Kingdom Home Office. Artificial CSF (ACSF) for both slicing and recording contained the following (in mm): 125 NaCl, 2.5 KCl, 26 NaHCO3, 1.25 NaH2PO4, 25 glucose, 1 MgCl2, and 2 CaCl2, and was bubbled with 5% carbon dioxide, 95% oxygen. Slices were continuously superfused with ACSF during the experiment, and all experiments were performed at 34–35°C.
Simultaneous whole-cell patch-clamp recordings from both soma and dendrites were made under direct visual control (Davie et al., 2006) using differential interference contrast optics (Zeiss Axioskop). Patch electrodes were pulled to resistances of 4–5 MΩ (for the soma) or 7–10 MΩ (for the dendrites) from thick-walled, filamented borosilicate glass. Internal solution (280–285 mOsm) contained the following (in mm): 133 KMeSO4, 7.4 KCl, 0.3 MgCl2, 10 HEPES, 0.1 EGTA, 3 Na2ATP, and 0.3 Na2GTP, pH adjusted to 7.2 with KOH, to which 0.5% biocytin was usually added. Current-clamp recordings were made using both AxoClamp 2B and 2A amplifiers (Molecular Devices), and were low-pass filtered at 3 kHz before being sampled at 20–50 kHz using Axograph 4.9 software (Axograph Scientific) and an ITC-18 interface (Instrutech).
CF input was stimulated by placing a patch electrode filled with ACSF in the granule cell layer (0.2 ms, 20–80 V square pulses; 1 Hz stimulation frequency). The location of the stimulation electrode and the stimulation intensity were adjusted until an isolated CF input was excited without evoking an antidromic action potential in the Purkinje cell axon. Complex spikes preceded by a simple spike within ∼1.5 ms were excluded from analysis because these were found to have a disrupted complex spike pattern. In some experiments, the cell was taken to threshold for generating multiple dendritic spikes in response to CF stimulation by varying the tonic current injection at the soma (−600 to 900 pA).
For simultaneous somatic and axonal recordings, whole-cell recordings were first made from Purkinje cells using the above internal solution containing 45–80 μm AlexaFluor 488, and the CF input was located using an extracellular stimulating electrode as above. The somatic pipette was then retracted carefully to form an outside-out patch, thereby resealing the cell membrane. Two patch electrodes containing 150 mm NaCl were then used to make simultaneous loose cell-attached recordings in voltage-clamp mode from the soma and the proximal axon and to record the capacitive currents generated by action potentials (sampling rate, 100 kHz). The peak of the capacitive current corresponds to the time of the fastest rise of the local action potential and therefore can be used to indicate the relative timing of spikes in the soma and the axon (Clark et al., 2005). Biocytin staining was used for post hoc recovery of cell morphologies.
This method (Robinson and Kawai, 1993; Sharp et al., 1993) was implemented using custom analog hardware (SM-1; Cambridge Conductance). Synaptic-like conductances were applied to the soma using two patch-clamp electrodes (one recording membrane potential, the other injecting current) to avoid bridge balance errors (Williams, 2004). Waveforms were biexponential, with τrise = 0.3 ms, τdecay = 3 ms, peak amplitude of 100–500 nS, and a reversal potential of 0 mV. In some exploratory experiments, the effect of conductance time course was investigated, varying τrise from 0.2 to 0.5 ms and τdecay from 2 to 5 ms. Complex spike-like events were produced over this entire 2.5-fold range of kinetics, and changing kinetics did not produce substantial improvements in the match to the physiological complex spike. In some experiments, dendritic spikes were evoked during somatic dynamic clamp; these spikes were triggered by dendritic current injection (biexponential current waveform, τrise = 0.5 ms and τdecay = 5 ms, peak amplitude of 2–5 nA together with tonic current of 0–330 pA).
Data were analyzed using custom programs, together with NeuroMatic data analysis software (http://www.neuromatic.thinkrandom.com/), in Igor Pro (Wavemetrics). The amplitudes of spikelets in the complex spike were measured with respect to baseline membrane potential (averaged over 50 ms before the complex spike), and the resulting amplitudes were sometimes normalized to the amplitude of the first spike in the complex spike. Somatic spike threshold was taken as the point at which the second derivative of the spike reached 10% of its peak value during the rising phase of the spike; threshold amplitude was measured as the voltage at this point minus the minimum voltage reached in the preceding trough. Dendritic spike width was measured at half the height of the spike peak above the preceding trough in membrane potential. Attenuation was measured in records in which extra dendritic spikes did not trigger extra somatic spikelets, allowing the depolarization caused by the propagated dendritic spike to be seen in isolation. The amplitude of the propagated event was found by subtracting from these records the average somatic waveform of corresponding responses of a similar somatic spiking pattern but without the extra dendritic spike (see Fig. 5A). Analogously, the amplitude of the dendritic spike was found by subtracting the average dendritic waveform of the same dataset from the individual records with extra dendritic spikes. Attenuation was calculated as the somatic event amplitude as a percentage of the dendritic event amplitude. Spontaneous interspike intervals (ISIs) were calculated as the average of three consecutive intervals between simple spikes occurring in the same trace as the post-complex spike pause. Post-complex spike afterhyperpolarizations (AHPs) were quantified by finding the minimum voltage during the post-complex spike pause in spontaneous firing and subtracting mean baseline membrane potential (as measured above). Differences between groups were tested for statistical significance using Student's t test (paired or unpaired as appropriate). Data were reported as the mean [in some cases weighted by the number of observations in each dataset (Bland and Kerry, 1998)] ± SEM unless otherwise indicated.
Somatic dynamic clamp can reproduce the complex spike
Activation of CF input to cerebellar Purkinje cells triggers the highly reproducible pattern of somatic spikes known as the complex spike (Fig. 1A,B) (Eccles et al., 1966; Llinás and Sugimori, 1980a). To determine whether the dendritic distribution of CF synaptic contacts is necessary to generate the distinctive spiking pattern of the complex spike, we examined whether concentrating the synaptic conductance at the soma is sufficient to mimic the complex spike waveform. Dynamic clamp was used to inject a biexponential conductance, based on the physiological CF synaptic conductance (Perkel et al., 1990; Llano et al., 1991; Silver et al., 1998; Wadiche and Jahr, 2001), via simultaneous dual somatic patch-clamp recordings (Fig. 1C). By adjusting the amplitude of the somatic dynamic-clamp conductance (Fig. 1D), a remarkably good match could be achieved with the CF-evoked complex spike (Fig. 1B). This matching was achievable in most cells, not only in neurons exhibiting complex spikes consisting of relatively few spikes (Fig. 1A), but also for neurons with elaborate complex spikes consisting of ≥5 spikes (see supplemental Fig. 1, available at www.jneurosci.org as supplemental material). The mean value of the “optimal” synaptic conductance was 170 ± 58 nS (range, 80–325 nS; n = 40 cells), comparable with the conductance values estimated using somatic voltage-clamp recording of CF EPSCs [200 ± 23 nS; n = 16 (Silver et al., 1998); p = 0.63]. Decreasing the synaptic conductance from the optimal value reduced the number of evoked spikes in the complex spike, whereas increasing the conductance ultimately led to inactivation of spiking, associated with voltage ripples on top of an envelope of depolarization. This depolarization rarely exceeded −30 mV, even at conductance values of 500 nS, thus maintaining a driving force for the synaptic conductance. The “optimal” synaptic conductance produced a highly quantitative match to the physiological complex spike in the same cells, with the number of spikes being indistinguishable (3.1 ± 0.12 vs 3.1 ± 0.12 spikes; n = 40 cells; p = 0.62; paired t test) between physiological and synthetic complex spikes. Furthermore, details of the spike waveform could also be closely matched, with amplitudes of individual spikelets being matched with an average error of only 18.0 ± 2.0%, and spike times matched with only 14.0 ± 1.5% error (n = 40).
Together, these observations demonstrate that the distribution of the CF synaptic conductance across the dendrites is not necessary to permit burst firing or to maintain synaptic driving force, because the complex spike can be reproduced simply by concentrating a similar synaptic conductance at the soma.
Dendritic spikes are not necessary for generating a complex spike
In some cell types, bursts of spikes at the soma are triggered by activation of dendritic calcium spikes (Larkum et al., 1999; Magee and Carruth, 1999; Williams and Stuart, 1999). Physiological CF input typically triggers one to three dendritic calcium spikes (Fujita, 1968; Llinás and Sugimori, 1980b; Crepel et al., 1981). It is not clear whether these dendritic spikes are necessary to generate complex spikes. To address this question, we made simultaneous dendritic recordings (104 ± 6 μm from the soma; range, 60–135 μm) during the “optimal” somatic dynamic clamp for mimicking the physiological complex spike (Fig. 2A). Activation of the CF produced rapidly rising, large-amplitude dendritic calcium spikes during the somatic complex spike (Fig. 2B) (Llinás and Sugimori, 1980b; Crepel et al., 1981). However, the somatic dynamic-clamp input that mimicked the complex spike in the same cell was not associated with generation of dendritic calcium spikes (Fig. 2C) (13 of 13 cells): only a small, slow dendritic depolarization was observed (Fig. 2D) (peak amplitude, 4.6 ± 0.5 mV vs 40.2 ± 1.2 mV for the physiological CF input; p < 10−12; peak dV/dt = 6.3 ± 0.4 mV/ms vs 60.2 ± 4.4 mV/ms for the physiological CF input; p < 10−7), reflecting the passive backpropagation of the somatic activity.
Contribution of dendritic calcium spikes to somatic output during the complex spike
These dendritic recordings have demonstrated that the somatic complex spike can be generated independently of a dendritic calcium spike, i.e., that dendritic calcium spikes are not necessary for complex spike generation. How then do dendritic calcium spikes contribute to the complex spike? We addressed this question by taking advantage of the stochastic nature of spike generation at threshold, where, by setting the tonic current injection at the appropriate value, the dendritic CF response could alternate by chance between one and two (or sometimes between two and three) dendritic spikes. In this way, the effect of adding a dendritic calcium spike on the somatic complex spike could be assessed. Using this approach, we found that an extra dendritic calcium spike had a remarkably small effect on the somatic complex spike waveform. Two examples are shown in Figure 3. In the Purkinje cell shown in Figure 3A, the addition of a second dendritic calcium spike did not produce any additional spikelets in the somatic complex spike, only a small additional depolarization (∼5 mV) after the last spikelet (Fig. 3B). In contrast, in the Purkinje cell shown in Figure 3C, an extra dendritic calcium spike could, in some trials, trigger an additional spikelet in the somatic complex spike (Fig. 3D). On average, 0.81 ± 0.10 additional somatic spikes were induced by the extra dendritic calcium spike in this neuron (Fig. 3E) (p < 10−6). Thus, even in a cell in which a calcium spike had relatively strong effects, it added less than one spike on average to the somatic complex spike. Across the population of cells, an extra dendritic calcium spike triggered only 0.24 ± 0.09 additional spikes at the soma (Fig. 3E) (weighted mean ± SEM; n = 21; p < 0.01), and of these we predict (Monsivais et al., 2005) that only ∼25% might propagate down the axon. The strength of the effect of an extra dendritic spike was not dependent on age (P18–P24, r = 0.05; p = 0.82) or on dendritic recording distance (range = 60–155 μm; r = 0.2; p = 0.37).
We performed an additional set of experiments to investigate the influence of the distributed synaptic conductance on the propagation of the dendritic spike to the soma. During somatic dynamic-clamp experiments, which we demonstrate above can mimic a complex spike, we “added” varying numbers of dendritic spikes at threshold by injecting current via a dendritic recording electrode (Fig. 4A,B). The dendritic spikes triggered in this way had a similar, but even weaker effect than physiologically triggered dendritic spikes; adding a single dendritic spike triggered only 0.1 ± 0.1 additional spikelets in the somatic complex spike (Fig. 4C) (not significantly different; n = 19; p = 0.37). This indicates that the influence of dendritic spikes on somatic spiking is weak even when the CF synaptic conductance is not widely distributed in the dendritic tree, along the path of propagation.
Together, these experiments directly demonstrate that dendritic calcium spikes have only a weak effect on the somatic complex spike, and are not responsible for producing the somatic burst of spikes.
Determinants of dendritic spike influence on somatic spiking
Why is the dendritic calcium spike so ineffective in triggering somatic spikelets? There are several reasons for this. First, dendritic spikes in Purkinje cells are relatively small and brief. We estimated the amplitude of dendritic spikes triggered by CF input relative to just-subthreshold events on alternating trials (Fig. 5A), giving a mean peak amplitude of only 10.6 ± 1.4 mV, and a half-width of 1.6 ± 0.12 ms (dendritic recording at 115 ± 11 μm from the soma; n = 9). Second, dendritic spikes experience substantial attenuation between the dendrite and the soma. Dendritic spikes were attenuated to 34.5 ± 4.6% of their original amplitude by the time they reached the soma (Fig. 5A,B) (n = 9; dendritic spikes measured at 115 ± 11 μm from the soma; note that this measure provides an upper limit of the degree of attenuation at this distance during a complex spike, because we rejected the occasional events that triggered somatic spikes). Third, dendritic spikes often arrived at the soma within the apparent refractory period for somatic spikes. When the dendritic spike occurred <0.8 ms after a preceding somatic spikelet, no additional somatic spikelet was triggered. Furthermore, although later dendritic spikes could successfully trigger somatic spikelets, there was a sigmoidal relationship between the amplitude of these somatic spikelets and the latency of the dendritic spike (as measured in Fig. 5C). This was observed both within individual datasets (Fig. 5D; see also corresponding traces in Fig. 3D) and across all datasets (Fig. 5E), where the half-maximum of the relationship was at 1.4 ms. However, even when the dendritic spike occurred well outside the apparent somatic refractory period, it was still often unable to trigger an additional somatic spikelet. This was because the somatic depolarization caused by the dendritic spike was below threshold for generating a somatic spikelet (weighted mean subthreshold depolarization, 4.1 ± 0.8 mV higher than previous trough vs spike threshold, 6.2 ± 0.5 mV higher; n = 7; p < 0.03, paired t test). Dendritic spikes triggered during somatic dynamic clamp were similarly often found to be within a somatic refractory period (supplemental Fig. 2, available at www.jneurosci.org as supplemental material) or below spike threshold (mean subthreshold depolarization, 1.3 ± 0.2 mV vs spike threshold, 1.9 ± 0.7 mV; n = 6; p < 0.04, paired t test).
This combination of characteristics explains why dendritic spikes have a relatively weak influence on somatic spikelet generation. As a consequence, there was no consistent temporal relationship between a dendritic spike and somatic spikelets. Specifically, the peaks of individual dendritic spikes could either occur both before or after a somatic complex spikelet it appeared to promote (supplemental Fig. 3A, available at www.jneurosci.org as supplemental material). This indicates that somatic spikelets are not simply reflections of forward-propagated dendritic spikes. In fact, the timing of extra somatic spike appears to be strongly influenced by intrinsic axosomatic properties. The later a dendritic spike occurs after the penultimate somatic spike, the earlier the extra spike occurs relative to the dendritic spike (supplemental Fig. 3B, available at www.jneurosci.org as supplemental material). If the somatic ISI was independent of the dendritic spike time, the slope of this relationship would be −1 (conversely, if it was entirely dependent on the dendritic spike time, the slope would be 0): we find a slope of −0.54, indicating that the somatic spike timing was influenced, but not strictly determined, by the dendritic spike time.
Dendritic spikes triggered by the CF modulate pauses in Purkinje cell output
We have demonstrated that dendritic spikes evoked by CF input have a remarkably weak electrical effect on the somatic complex spike waveform, and thus on the number of spikes transmitted down the axon directly associated with the complex spike. CF activation is also associated with a pause in spontaneous spiking after the complex spike in vivo (Latham and Paul, 1970; Bloedel and Roberts, 1971; Armstrong and Rawson, 1979; Sato et al., 1992). The mechanisms underlying the generation and duration of this pause are not well understood. Armstrong and Rawson (1979) found that pause length in vivo depends on the spontaneous simple spike firing rate, and we found this to hold true in vitro. Only at fairly low spontaneous firing rates (>20–25 ms ISI) did the length of the pause exceed one spontaneous ISI (data not shown). Furthermore, by examining our somatic and dendritic recordings, we found that the duration of the pause also strongly depends on the number of dendritic calcium spikes triggered by CF input (Fig. 6A). An extra dendritic spike lengthened the post-complex spike pause by 15.2 ± 2.6 ms (Fig. 6B) (weighted mean ± SEM; p < 0.001), representing a 52 ± 11% (Fig. 6C) increase in pause duration, or a further 0.56 ± 14 ISIs quiescence before spontaneous spiking resumed. The effect of the extra dendritic spike on the pause was independent of the number of extra spikelets added to the somatic complex spike (r = 0.36; p = 0.30), and was also independent of the spontaneous firing rate (r = −0.35; p = 0.29). The effect of dendritic spikes on the post-complex spike pause was also assessed by comparing CF-evoked events to those generated by somatic dynamic clamp, and so lacking dendritic spikes (supplemental Fig. 4A, available at www.jneurosci.org as supplemental material). The pause after climbing fiber input that evoked only one dendritic spike was, on average, 3.8 ± 1.6 ms longer than that after the matching dynamic-clamp-evoked complex spikes, which did not evoke a dendritic spike (p < 0.05; n = 11) (supplemental Fig. 4B, available at www.jneurosci.org as supplemental material), corresponding to a 31 ± 12% increase in pause duration, or a further 0.22 ± 0.09 ISIs (supplemental Fig. 4C, available at www.jneurosci.org as supplemental material).
One way in which dendritic calcium spikes might influence the post-complex spike pause is through activation of hyperpolarizing conductances (e.g., calcium-activated potassium channels). We therefore measured the amplitude of post-complex spike AHPs (Fig. 7A). A second dendritic spike increased dendritic AHPs by 3.4 ± 0.7 mV (Fig. 7B,C) (weighted mean ± SEM; p < 0.001); a similar change in somatic AHPs was also seen. Pause duration was correlated with the amplitude of the dendritic AHP across all datasets (Fig. 7D) (n = 11; r = −0.89; p < 10−16), and an extra dendritic spike caused both larger AHPs and longer pauses.
These data suggest that, from the point of view of the postsynaptic deep cerebellar nuclei (DCN) neurons, the most salient effect of dendritic spikes may be to increase the pause after the complex spike rather than the number and timing of spikelets in the complex spike itself.
Axonal origin of somatic complex spikes
The results described above show that dendritic spikes are not directly linked to the spikelets in the complex spike. By elimination, the spikelets must therefore arise either in the soma or in the axon. This is important to discriminate, given that the powerful CF input is almost exclusively dendritic. To resolve this question, we made simultaneous cell-attached recordings from the soma and proximal axon of Purkinje cells (≤75 μm from the soma) while evoking CF input. This enabled us to directly compare the relative timing at the soma and axon of each spikelet component of the complex spike. In all cases, the spikelets occurred first in the axon (Fig. 8). When somatic and axonal spikes were averaged by triggering off the somatic spike, the axonal spike preceded the somatic spike for all spikelets, by 0.03 ± 0.01 ms (n = 8 cells), 0.15 ± 0.04 ms (n = 7), 0.14 ± 0.03 ms (n = 5), and 0.14 ± 0.03 ms (n = 5) for the first, second, third, and fourth spikelets in the complex spike, respectively. The fact that the axonal spike precedes the somatic spike in each case indicates that all the spikelets within the complex spike are initiated in the axon.
Our results provide direct evidence that all spikelets within the complex spike of Purkinje cells originate in the axon. The prominent dendritic spikes triggered by CF activation play only a very minor and indirect role in the generation of this stereotyped axonal burst. Dendritic spikes propagate poorly to the soma, where they were often below spike threshold or within the refractory period, and thus usually failed to trigger an additional axosomatic spike. Consistent with these findings, we found that a distributed pattern of dendritic synaptic input, although required for triggering dendritic calcium spikes, is not necessary for generation of the complex spike burst. These results support a model of the Purkinje cell in which the dendrites are functionally separate from the axon and soma during the complex spike. We identify a new role for dendritic spikes as regulators of the post-complex spike pause, complementing their roles as local triggers for synaptic plasticity and modulators of intrinsic firing.
Axonal initiation of the complex spike
Our paired recordings from the soma and axon resolve the question of the origin of the complex spike (Schmolesky et al., 2002) by directly demonstrating that all the spikelets are initiated in the axon, similarly to spontaneous simple spikes (Stuart and Häusser, 1994; Clark et al., 2005). The axonal origin of the spikelets is surprising, given the enormous CF-triggered current concentrated in the soma and proximal dendrites. In other neurons, although the lowest-threshold site for initiation of sodium action potentials is typically in the axon (Coombs et al., 1957; Colbert and Johnston, 1996; Palmer and Stuart, 2006), strong synaptic input can shift the initiation site from the axon to the dendrites (Mainen et al., 1995; Chen et al., 1997; Stuart et al., 1997; Golding and Spruston, 1998). In Purkinje cells, however, dendrites are unable to support initiation or active backpropagation of sodium spikes, because of their low Na+ channel density (Stuart and Häusser, 1994), and their elaborate geometry (Vetter et al., 2001; Bekkers and Häusser, 2007). In contrast, the axon is more favorable for spike initiation because it is protected from the substantial capacitive load of the soma and dendrites by its high axial resistance (Dodge and Cooley, 1973; Mainen et al., 1995; Colbert and Pan, 2002), potentially aided by increased densities or altered properties of axonal Na+ channels (Mainen et al., 1995; Colbert and Pan, 2002; Boiko et al., 2003; Kole et al., 2008). The next step will be to identify the precise location of the initiation site of the complex spike within the axon.
Somatic dynamic clamp reproduces the complex spike
By mimicking the complex spike waveform with a somatic synaptic-like conductance, we have shown that neither dendritic spikes nor a dendritic distribution of synaptic conductance is required for complex spike generation. Surprisingly, the shunt produced by concentrating the entire synaptic conductance at the soma does not prevent the complex spike burst, generating spiking similar to the rapid bursts produced by somatic current injection (Callaway and Ross, 1997; Swensen and Bean, 2003; McKay et al., 2005). This is due in part to the axonal initiation site of spikelets, which provides electrical isolation from the somatic shunt. Furthermore, the powerful and fast somatic voltage-gated Na+ conductances in Purkinje cells (Raman and Bean, 1999), together with the K+ conductances (Raman and Bean, 1999; Edgerton and Reinhart, 2003; Martina et al., 2003; McKay and Turner, 2004; Zagha et al., 2008), which balance large depolarizing currents (Swensen and Bean, 2005; Zagha et al., 2008) and help maintain high-frequency firing, appear to predominate over the synaptic conductance during complex spike generation.
Propagation of dendritic spikes
We demonstrate that the dendritic spikes triggered by CF input do not directly trigger somatic spikelets in the complex spike and influence its generation only weakly. This parallels the relatively weak effect of dendritic inhibition on the somatic complex spike (Callaway et al., 1995). We show that the low efficacy of dendritic spikes in Purkinje neurons is attributable to a combination of factors. First, they are relatively brief and small, in striking contrast to the ∼80 mV, ∼60 ms dendritic calcium spikes in layer 5 pyramidal cells (Zhu, 2000). Furthermore, their dendrosomatic propagation is substantially attenuated, attributable in part to the unfavorable impedance mismatches of the highly branching geometry of the Purkinje cell dendritic tree (Vetter et al., 2001), and they may be further diminished by somatic active shunting conductances (Häusser et al., 2001). In addition, the strong synaptic and voltage-gated dendritic conductances active during the complex spike, although initially triggering dendritic spikes, may serve to restrict their spread, reducing the current transmitted to the soma. However, dendritic spikes generated in the absence of distributed dendritic conductance are attenuated to a similar or even greater extent (Fig. 4) (Rancz and Häusser, 2006), suggesting that widespread dendritic spike initiation during the complex spike may actually aid propagation to the soma by reducing the axial current between branches and thus partially mitigating the impedance mismatch. Finally, we demonstrate that the highly attenuated dendritic spike is usually subthreshold when it reaches the soma, or within the refractory period of the previous spikelet. Together, these factors explain why dendritic spikes in Purkinje cells, consistent with previous work (Callaway et al., 1995), have a far weaker influence on axonal spiking than in mitral neurons (Chen et al., 1997), or in cortical and hippocampal pyramidal neurons (Schiller et al., 1997; Golding and Spruston, 1998; Larkum et al., 1999), in which a dendritic calcium spike can contribute up to 2.5 output spikes (Williams and Stuart, 2002).
Functional role of dendritic spikes triggered by CF input
We have demonstrated a new functional role for dendritic spikes in Purkinje cells: regulation of the pause in axonal output after the complex spike. This pause is a well known feature of the CF response in vivo (Latham and Paul, 1970; Bloedel and Roberts, 1971; Armstrong and Rawson, 1979; Sato et al., 1992; Barmack and Yakhnitsa, 2008), but its underlying mechanisms are unknown. Our findings suggest that the calcium entry associated with CF-triggered dendritic spikes (Ross and Werman, 1987; Miyakawa et al., 1992) activates calcium-dependent potassium conductances (Hounsgaard and Midtgaard, 1989; Edgerton and Reinhart, 2003; Rancz and Häusser, 2006), causing hyperpolarization and a prolonged pause before resumption of spontaneous firing, analogous to the recently described pause in spiking triggered by strong parallel fiber input (Steuber et al., 2007). This suggestion is further supported by the finding that disruption of Purkinje cell calcium buffering (Servais et al., 2005) alters post-complex spike pause duration. Thus, pause duration reports the number of dendritic spikes triggered by CF input. The pause is likely to be a potent signal to downstream DCN neurons [perhaps more so than the complex spike itself, which is often poorly propagated down the axon (Khaliq and Raman, 2005; Monsivais et al., 2005)], given that the relief of DCN cells from persistent inhibition triggers rebound firing (Aizenman and Linden, 1999; McKay et al., 2005), which is both a salient electrical signal and a trigger for plasticity at the DCN (Nelson et al., 2003; Pugh and Raman, 2006).
This new role for dendritic spikes in regulating the post-complex spike pause complements the well established functions linked to the large dendritic calcium influx they generate (Ross and Werman, 1987; Miyakawa et al., 1992). This biochemical signal can trigger long-term depression of CF input and plays an important role in short-term and long-term regulation of parallel fiber and inhibitory synaptic strength (Hansel et al., 2001; Brenowitz and Regehr, 2005). CF activity also regulates intrinsic firing behaviors in ways likely to involve calcium influx, such as modulating spontaneous firing rate and pattern (Cerminara and Rawson, 2004; McKay et al., 2007).
A new perspective on Purkinje cell function
The CF input to the Purkinje cell provides a timing signal for online motor control (Welsh and Llinás, 1997; Kitazawa et al., 1998) and an error signal, which enables synaptic plasticity at parallel fiber synapses and mossy fiber–DCN synapses (Gilbert and Thach, 1977; Kitazawa et al., 1998; Pugh and Raman, 2006). We demonstrate that CF input briefly divides the Purkinje cell into two weakly coupled functional compartments, a dendritic and an axosomatic compartment, which reflect these distinct roles. By distributing CF input in the dendrites, synaptic and active dendritic currents are efficiently isolated from the shunt of the Purkinje cell's specialized high-frequency firing mechanisms (Stuart and Häusser, 1994; Raman and Bean, 1999; Häusser et al., 2001). This allows dendritic spike generation and the associated, plasticity-triggering calcium influx to occur in parallel with burst generation at the soma. The mutual isolation of the two compartments allows the number of dendritic calcium spikes, and the resulting downstream dendritic effects of CF input, to be modulated by parallel fiber or inhibitory input without strongly influencing the burst pattern at the soma (Callaway et al., 1995). In contrast, the regulation of the post-complex spike pause by dendritic spikes triggered by the CF input reveals a new link between plasticity in the Purkinje cell dendritic tree, Purkinje cell output pattern, and the resulting plasticity at downstream targets.
This work was supported by grants from the Wellcome Trust, European Commission, and the Gatsby Foundation. J.D. was supported by the Wellcome Trust 4-year PhD Programme in Neuroscience at University College London. We are grateful to Mickey London, Ede Rancz, Arnd Roth, Spencer Smith, and Ikuko Smith for helpful discussions and for their comments on this manuscript.
- Correspondence should be addressed to Michael Häusser, Wolfson Institute for Biomedical Research, University College London, Gower Street, London WC1E 6BT, UK.
- Aizenman and Linden, 1999.↵
- Ariav et al., 2003.↵
- Armstrong and Rawson, 1979.↵
- Barmack and Yakhnitsa, 2008.↵
- Bekkers and Häusser, 2007.↵
- Bland and Kerry, 1998.↵
- Bloedel and Roberts, 1971.↵
- Boiko et al., 2003.↵
- Brenowitz and Regehr, 2005.↵
- Callaway and Ross, 1997.↵
- Callaway et al., 1995.↵
- Campbell et al., 1983.↵
- Cerminara and Rawson, 2004.↵
- Chen et al., 1997.↵
- Clark et al., 2005.↵
- Colbert and Johnston, 1996.↵
- Colbert and Pan, 2002.↵
- Coombs et al., 1957.↵
- Crepel et al., 1981.↵
- Davie et al., 2006.↵
- Dodge and Cooley, 1973.↵
- Doiron et al., 2002.↵
- Eccles et al., 1966.↵
- Edgerton and Reinhart, 2003.↵
- Fujita, 1968.↵
- Geiger et al., 2002.↵
- Gilbert and Thach, 1977.↵
- Golding and Spruston, 1998.↵
- Granit and Phillips, 1956.↵
- Hansel et al., 2001.↵
- Häusser et al., 2001.↵
- Hounsgaard and Midtgaard, 1989.↵
- Ito, 2001.↵
- Khaliq and Raman, 2005.↵
- Khaliq and Raman, 2006.↵
- Kitazawa et al., 1998.↵
- Kole et al., 2008.↵
- Larkum et al., 1999.↵
- Larkum et al., 2001.↵
- Latham and Paul, 1970.↵
- Llano et al., 1991.↵
- Llinas and Nicholson, 1971.↵
- Llinás and Sugimori, 1980a.↵
- Llinás and Sugimori, 1980b.↵
- Llinás et al., 1968.↵
- Magee and Carruth, 1999.↵
- Mainen and Sejnowski, 1996.↵
- Mainen et al., 1995.↵
- Martina et al., 2003.↵
- Martinez et al., 1971.↵
- McKay and Turner, 2004.↵
- McKay et al., 2005.↵
- McKay et al., 2007.↵
- Miyakawa et al., 1992.↵
- Monsivais et al., 2005.↵
- Nelson et al., 2003.↵
- Palay and Chan-Palay, 1974.↵
- Palmer and Stuart, 2006.↵
- Perkel et al., 1990.↵
- Pinsky and Rinzel, 1994.↵
- Pugh and Raman, 2006.↵
- Raman and Bean, 1999.↵
- Rancz and Häusser, 2006.↵
- Robinson and Kawai, 1993.↵
- Ross and Werman, 1987.↵
- Sato et al., 1992.↵
- Schiller et al., 1997.↵
- Schmolesky et al., 2002.↵
- Servais et al., 2005.↵
- Sharp et al., 1993.↵
- Silver et al., 1998.↵
- Steuber et al., 2007.↵
- Stuart and Häusser, 1994.↵
- Stuart et al., 1997.↵
- Swensen and Bean, 2003.↵
- Swensen and Bean, 2005.↵
- Turner et al., 2002.↵
- Vetter et al., 2001.↵
- Wadiche and Jahr, 2001.↵
- Welsh and Llinás, 1997.↵
- Williams, 2004.↵
- Williams and Stuart, 1999.↵
- Williams and Stuart, 2002.↵
- Zagha et al., 2008.↵
- Zhu, 2000.↵