Complex spikes are high-frequency bursts of Na+ spikes, often riding on a slower Ca2+-dependent waveform. Although complex spikes may propagate into axons, given their unusual shape it is not clear how reliably these bursts reach nerve terminals, whether their spikes are efficiently transmitted as a cluster of postsynaptic responses, or what function is served by such a concentrated postsynaptic signal. We examined these questions by recording from synaptically coupled pairs of cartwheel cells, neurons which fire complex spikes and form an inhibitory network in the dorsal cochlear nucleus. Complex spikes in the presynaptic soma were reliably propagated to nerve terminals and elicited powerful, temporally precise postsynaptic responses. Single presynaptic neurons could prevent their postsynaptic partner from firing complex but not simple spikes, dramatically reducing dendritic Ca2+ signals in the postsynaptic neuron. We suggest that rapid transmission of complex spikes may control the susceptibility of neighboring neurons to Ca2+-dependent plasticity.
The reliability of information transfer between neurons depends in large part on the capacity of action potentials to propagate down the axon, invade boutons, and induce neurotransmitter release. This has generally been considered a high-fidelity process, and studies in hippocampus, cortex, and cerebellum support that idea (Mackenzie and Murphy, 1998; Williams and Stuart, 1999; Cox et al., 2000; Forti et al., 2000; Koester and Sakmann, 2000; Raastad and Shepherd, 2003; Brenowitz and Regehr, 2007). Other studies, however, indicate that somatic action potentials can fail to initiate synaptic transmission under conditions that alter voltage-gated ion channel availability, such as high-frequency firing or changes in somatic membrane potential (Debanne et al., 1997; Debanne, 2004; Meeks et al., 2005).
Complex spikes are brief, high-frequency bursts of Na+ action potentials, often driven by a slow, Ca2+ or synaptically mediated depolarization (Schmolesky et al., 2002; Kim and Trussell, 2007), and are produced throughout the brain (Deschênes et al., 1982; Chagnac-Amitai et al., 1990; Niespodziany and Poulain, 1995; Harris et al., 2001; Jung et al., 2001; Athanassiadis et al., 2005). The high firing frequencies and underlying waveform of complex spikes suggest that the action potentials comprising a complex spike, termed spikelets, might fail to initiate synaptic transmission. Consistent with this, recent studies in cerebellar Purkinje cells have shown that 30–60% of complex spike spikelets fail to propagate down the axon (Khaliq and Raman, 2005; Monsivais et al., 2005). It is not known, however, whether additional failures occur at the synapse or what effect failures have on postsynaptic excitability.
The dorsal cochlear nucleus (DCN) is a cerebellum-like brainstem auditory nucleus that functions in sound localization (Oertel and Young, 2004). Cartwheel cells are DCN interneurons that receive parallel fiber input and share genetic, developmental, and physiological properties with cerebellar Purkinje cells (Mugnaini et al., 1987; Berrebi et al., 1990; Ivanova and Yuasa, 1998). Like Purkinje cells, cartwheel cells fire complex spikes (Zhang and Oertel, 1993; Manis et al., 1994). Moreover, parallel fiber inputs to cartwheel cells undergo long-term plasticity that is highly dependent on the timing of presynaptic and postsynaptic activity (Tzounopoulos et al., 2004, 2007). Unlike Purkinje cells, cartwheel cells project locally, inhibiting neighboring neurons including other cartwheel cells (Berrebi and Mugnaini, 1991; Golding and Oertel, 1996, 1997; Rubio and Juiz, 2004). By recording from synaptically coupled pairs of nearby cartwheel cells, we showed previously that complex spike spikelets sometimes fail to elicit IPSCs (Tzounopoulos et al., 2004). Here, we use this approach to investigate the fidelity of unitary connections formed by complex spiking neurons. We identified properties of complex spikes that promote transmission failures and found that these failures can occur presynaptically. Surprisingly, the postsynaptic charge flux evoked by complex spikes was rather insensitive to transmission failures, suggesting that complex spikes provide a distinct signal to postsynaptic neurons. Furthermore, we showed that complex spikes inhibit postsynaptic firing over a broad time window and that this significantly alters postsynaptic calcium signaling.
Materials and Methods
A colony of transgenic mice expressing green fluorescent protein (GFP) in interneurons (GIN mice) (Oliva et al., 2000) backcrossed against a C57BL/6J (Jackson Laboratories) background was established and maintained. Breeding pairs consisted of one hemizygous and one wild type. Hemizygous offspring had high expression of GFP in their spinal cords, and neonates [postnatal day 0 (P0)–P2] were reliably phenotyped by checking for fluorescence under blue light. All procedures were conducted in accordance with guidelines set by the Oregon Health and Science University Institutional Animal Care and Use Committee.
Recordings were made from 210-μm-thick coronal brainstem slices prepared from P16–P22 GIN mice and their nontransgenic littermates. The transgene had no apparent effect on any measured parameter and, therefore, the data from these two groups were pooled. Slices were cut and then incubated for 1 h in 34°C ACSF composed of (in mm) 130 NaCl, 3 KCl, 2.4 CaCl2, 1.3 MgSO4, 1.2 KH2PO4, 20 NaHCO3, 3 Na-HEPES, and 10 glucose, bubbled with 5%CO2/95%O2, osmolality ∼305 mOsm. All recordings were made at 33–34°C with ACSF that was supplemented with 10 μm 6,7-dinitroquinoxaline-2,3-dione (DNQX) or 10 μm 2,3-dihydroxy-6-nitro-7-sulfonyl-benzo[f]quinoxaline (NBQX) to block excitatory activity. To examine the pharmacology of synaptic transmission between cartwheel cells, 500 nm strychnine and 20 μm 2-(3-carboxypropyl)-3-amino-6-(4-methoxyphenyl)-pyridazinium bromide (SR95531) were added directly to the bath ACSF. Electrodes (2–4 MΩ) were filled with an intracellular solution composed of (in mm) 113 K-gluconate, 9 HEPES, 4.5 MgCl2, 0.1 EGTA, 14 Tris2-phosphocreatine, 4 Na2-ATP, and 0.3 Tris-GTP, with osmolality brought to ∼290 mOsm with sucrose and pH adjusted to 7.25 with KOH. To improve IPSC detection, ECl was increased in some experiments by adding 6.6 mm KCl to the intracellular solution in place of an equal amount of K-gluconate. Data were corrected for a 13 mV junction potential.
Neurons were visualized with infrared videomicroscopy and gradient contrast optics (Dodt et al., 2002). A mercury lamp and an enhanced GFP (EGFP) filter set were used to view EGFP-expressing neurons. Data were acquired at 20–50 kHz and low-pass filtered at 3–10 kHz using a MultiClamp 700B amplifier, Digidata 1322A digitizer, and Clampex 9 software. In voltage-clamp recordings, cells were held at −73 to −93 mV and series resistance (<30 MΩ) was usually 80% compensated. In current-clamp recordings, current was injected to maintain the membrane potential near −78 mV and bridge balance and pipette capacitance neutralization were set and adjusted throughout the experiment. Except where noted, action potentials were elicited with square current pulses in current clamp.
Strychnine, SR95531, DNQX, NBQX, and dl-APV were from Ascent Scientific. All others were from Sigma-Aldrich.
Electrophysiological data analysis.
Data were analyzed using Clampfit 9 and custom algorithms written in Igor Pro (Wavemetrics). Each spikelet in a complex spike that produced an IPSC was counted as a transmission success; spikelets that did not produce an IPSC were counted as failures. To correlate spikelet properties with transmission probabilities, spikelets were first sorted into groups on the basis of whether they produced transmission successes or failures. The data for these two groups were binned, and for each bin transmission probability = successes/(successes + failures). Transmission probabilities were then fit with a sigmoidal function of the form y = 1/(1 + exp(−(x − Xhalf)/k)). Synaptic jitter was equal to the SD of the presynaptic spike to IPSC latency (from peak of presynaptic spike to time of 20% rise of IPSC). To compare IPSC areas across pairs, we examined pairs in which we were able to evoke different numbers (e.g., spikelet doublets, triplets, etc.) of postsynaptic IPSCs (>10 trials per condition, n = 9 pairs). For each pair, the mean area for each IPSC group was calculated and normalized to the mean area for the three-IPSCs group. Data were then averaged across pairs. To simulate EPSPs in current clamp, we examined the time course of spontaneous EPSCs measured in voltage clamp. EPSCs were well fit by the double exponential function EPSC (t) = Amp(1 − e−t/τrise)(e−t/τdecay), where Amp was variable and corresponds to the amplitude of the EPSC, τrise = 0.30 ± 0.05 ms, and τdecay = 1.59 ± 0.13 ms. This function was used to generate stimulus files that directed the current clamp to inject current with an EPSC-like waveform. Amp was gradually increased until the EPSC-like stimulus was large enough to reliably evoke simple spikes or complex spikes (two to four spikelets per stimulus). Significant differences (p < 0.05) were detected with t tests or one-way ANOVAs with Tukey's post hoc test. Data are mean ± SEM except where noted.
Two-photon imaging and analysis.
For imaging experiments, slices were prepared from P17–P22 ICR mice (Harlan) using the procedures detailed above. Experiments were performed at 32–34°C. EGTA was omitted from the internal solution and replaced with 20 μm Alexa Fluor 594 hydrazide and 200 μm Fluo-5F (bouton imaging experiments) or 200 μm Fluo-4FF (somatodendritic imaging experiments; Invitrogen). A Ti:sapphire pulsed laser (Chameleon Ultra II; Coherent) tuned to 810 nm powered a Prairie Technologies Ultima two-photon imaging system. Laser intensity was controlled with an electro-optical modulator (350–80 LA; Conoptics). Cells were imaged with a 60×, 0.9 numerical aperture (NA) objective (Olympus). Both epifluorescence and transfluorescence signals were captured (the latter using a 1.4 NA oil-immersion condenser and custom substage detectors). Fluorescence was split into red (R; Alexa 594) and green (G; Fluo-5F, Fluo-4FF) channels using dichroic mirrors and bandpass filters (epifluorescence: 575 DCXR, HQ525/70, HQ607/45; transfluorescence: T560LPXR, ET510/80, ET620/60; Chroma) and focused on multialkali photomultiplier tubes (R9110; Hamamatsu). Processes were imaged in line-scan mode (2.2 ms/line, including mirror flyback), and complex spike, simple spike, and spike train stimuli were interleaved. Data are presented as averages of 20–40 (for dendritic recordings) or 60–80 (for axonal recordings) events per site and expressed as 100 × (G/R)/(G/R)max, where (G/R)max was the maximal fluorescence in saturating Ca2+ (2 mm) (Yasuda et al., 2004). Electrophysiological data were recorded at 5 kHz and filtered at 2 kHz using an Axopatch 200B amplifier and acquired with an ITC-18 interface (Instrutech) using custom software written in Igor Pro. Spikes were evoked by somatic current injection (complex spikes: 3–500 pA, 10 ms; simple spikes: 1 nA, 2 ms) or by stimulus electrode excitement of parallel fiber inputs in the presence of 500 nm strychnine and 20 μm SR95531 (NBQX was omitted for this experiment). Hyperpolarizing steps were often injected to suppress complex spiking where appropriate (−2 to −500 pA, 10 ms duration, 2 ms delay from depolarizing onset). IPSPs were evoked via stimulus electrode excitement of inhibitory inputs in the presence of 10 μm NBQX and 100 μm dl-APV.
We used dual whole-cell recordings of synaptically coupled pairs of cartwheel cells to examine the fidelity with which somatic complex spikes produce postsynaptic responses. Slices were prepared from transgenic GIN mice or from their nontransgenic littermates. In GIN mice, the promoter for the GABA synthetic enzyme GAD67 drives EGFP expression in a subset of interneurons (Oliva et al., 2000). Most EGFP-labeled cells in the outer layers of the DCN were cartwheel cells, as confirmed by their location, their characteristic round somata and spiny dendrites (Fig. 1A) (Wouterlood and Mugnaini, 1984), and their capacity to generate complex spikes (Zhang and Oertel, 1993; Manis et al., 1994). This EGFP labeling facilitated the localization of cartwheel cells and was particularly useful for identifying pairs of nearby cartwheel cells.
If a complex spike initiated by a current pulse in one cartwheel cell elicited IPSCs in another cartwheel cell, we classified that pair as synaptically coupled. Of the 152 cartwheel cell pairs from which we recorded, 62 pairs were synaptically coupled, indicating a minimal probability of connectivity of 41%. Synaptic transmission between cartwheel cells appeared to be predominately glycinergic, as the glycine receptor antagonist strychnine blocked cartwheel cell to cartwheel cell IPSCs by ∼85%. The GABAA receptor antagonist SR95531 (gabazine) blocked the remaining current (Fig. 1B,C). The finding that cartwheel cells corelease GABA is consistent with our observation that the GAD67 promoter drives EGFP expression in these cells in GIN mice, and with previous studies (Mugnaini, 1985; Kolston et al., 1992; Gates et al., 1996).
Spikelet transmission failures
Most complex spikes consisted of a large initial spikelet followed by one to three smaller spikelets. The early spikelets in a complex spike almost always led to a postsynaptic response, but later spikelets frequently failed to do so (Fig. 2). Across a population of 1435 trials from 27 cartwheel cell pairs, the probability that the first or second spikelet would produce an IPSC was >98%. For the third spikelet, the probability dropped to 87%, and for the fourth, to 65% (Fig. 2B). When data were sorted according to pairs in which the presynaptic cell fired complex spikes with two, three, or four spikelets, it became clear that most transmission failures occurred at the final spikelet (Fig. 2C).
This differs from cerebellar Purkinje cells, in which the second spikelet in complex spikes was the most failure prone (Khaliq and Raman, 2005; Monsivais et al., 2005). To investigate the basis of this difference, we examined the complex spike waveform, testing several properties of individual spikelets for correlations between spikelet parameters and transmission successes or failures. Spikelet peak amplitude and maximal rate of rise were positively correlated with the probability that a spikelet would yield an IPSC (Fig. 3A). For spikelets after the initial spikelet in a complex spike, the maximal extent of membrane repolarization preceding the spikelet (repolarization potential) was negatively correlated with transmission probability. Comparisons between spikelet properties revealed that peak amplitude, maximal rate of rise, and the preceding repolarization potential covaried and are reasonable predictors of transmission success (Fig. 3B). Interestingly, the interspikelet interval (ISI) did not correlate with transmission probability, nor did it covary with any of the other spikelet properties we measured. There did appear to be a trend toward failures occurring at low ISI values (Fig. 3B, bottom, clusters of failures between 2 and 4 ms ISI), but a number of failures were observed when ISI values were quite large. It is likely that the slow depolarization underlying the complex spike obscured the expected relationship between ISI and spikelet transmission probability, such that the repolarization potential preceding a spikelet exerted greater influence on the transmission probability of the subsequent spikelet than did the ISI. For example, a spikelet preceded by a short ISI and a fairly negative repolarization potential was more likely to elicit an IPSC than a spikelet preceded by a long ISI with a depolarized repolarization potential. Thus, spikelet shape, more than spikelet timing, correlates with transmission probability.
If the amplitude of the underlying slow depolarization contributes significantly to the spikelet shape, then a spike burst separated from the underlying slow depolarization characterizing complex spikes should possess higher transmission fidelity than a regular complex spike. To test this hypothesis, we delivered trains of simple spikes with frequencies matched to ISIs determined for complex spikes in a given presynaptic cell. Because high-frequency stimuli elicited in current-clamped cells almost always evoked complex spikes, simple spike trains were generated in voltage clamp with brief depolarizing voltage steps that were just large enough to reliably evoke escaping spikes (0.2 ms steps from −78 mV to +27 to +57 mV). These simple spike trains produced IPSCs similar to those elicited by complex spikes (Fig. 4A,B). The transmission fidelity of complex-spike-matched simple spike trains was nearly perfect and much higher than that of complex spikes measured within the same cartwheel cell pairs or across a population of pairs (Fig. 4C). Cartwheel cell pairs were also able to transmit to postsynaptic cells simple spike trains elicited at frequencies of up to 500 Hz (Fig. 4D). Indeed, three of four pairs followed three or more spikes at 500 Hz before exhibiting transmission failures. These results indicate that the failure of a spikelet to produce an IPSC does not reflect an intrinsic limit on axonal propagation or synaptic release. Rather, they suggest that slow, small spikelets do not always invade the axon.
Failures occur before presynaptic terminals
Action potentials that invade boutons evoke relatively uniform elevations in local Ca2+ (Geiger and Jonas, 2000; Koester and Sakmann, 2000; Bischofberger et al., 2002; Koester and Johnston, 2005; Brenowitz and Regehr, 2007). Ca2+ transients from trains of action potentials can be predicted by the linear sum of individual spike Ca2+ waveforms, offset by relative spike timing within the train (Koester and Sakmann, 2000; Brenowitz and Regehr, 2007). We used this relationship to test whether spikelet transmission failures occur upstream of presynaptic boutons. Cartwheel cells were current clamped and filled via the patch pipette with the red dye Alexa Fluor 594 (20 μm) and the Ca2+ indicator Fluo-5F (200 μm). The axon was imaged with two-photon microscopy as it projected from the soma, and the first one or two en passant boutons encountered were used for subsequent Ca2+ imaging (n = 5 cells, 9 boutons, 104 ± 19 μm from soma) (Fig. 5A). A simple spike elicited in the soma produced a bouton Ca2+ transient whose decay was fit with a single exponential function (Fig. 5B). Ca2+ transients from trains of simple spikes (six spikes at 50 Hz) were well fit by the linear sum of single-spike exponentials offset by 20 ms (Fig. 5C,F); however, complex spikes often evoked sublinear Ca2+ transients (Fig. 5D). In six of nine boutons, the observed Ca2+ transient was smaller than predicted from the sum of the exponential functions, suggesting that some spikelets failed to invade the bouton (Fig. 5F). Furthermore, the amplitudes of the Ca2+ transients for all nine boutons were positively correlated with the maximal rate of rise of the final spikelet in the complex spike, with the slowest spikelets yielding the lowest observed/predicted ΔG/R ratios (r = 0.96) (data not shown). This is consistent with our earlier observation that slowly rising spikelets are less likely to be transmitted (Fig. 3). The amplitude of the complex spike Ca2+ transient was always smaller than that evoked by the simple spike train and was always within the linear range of the Ca2+ indicator (Yasuda et al., 2004), thus excluding errors resulting from dye saturation. This result suggests that spikelet transmission failures occur before invading presynaptic terminals, probably at the level of the axon initial segment.
The slow, Ca2+-mediated depolarization underlying the complex spike can be elicited even when Na+ channels are blocked by tetrodotoxin (TTX) (Golding and Oertel, 1997; Kim and Trussell, 2007). It is possible that this waveform might propagate into the axon and alter Ca2+ levels in axonal boutons. We examined this possibility and found no change in bouton Ca2+ levels after stimulating cartwheel cells to fire complex spikes in the presence of TTX (Fig. 5E,G), indicating that, as in Purkinje cells, Ca2+ spikes do not propagate (Monsivais et al., 2005).
Reproducibility of postsynaptic responses to complex spikes
In the analyses thus far, postsynaptic responses to complex spikes have been treated as all-or-nothing events. The information content of a complex spike, however, will depend not only on its ability to elicit IPSCs but also on the temporal fidelity and reproducibility of those IPSCs. Accordingly, we investigated the trial-to-trial variability in the postsynaptic responses to complex spikes. An overlay of 15 trials from one cartwheel-to-cartwheel cell pair indicates that the time between a presynaptic spikelet and postsynaptic IPSC is brief (Fig. 6A), whereas a view of the same IPSCs on an expanded time scale demonstrates that there is little variability in the relative timing between presynaptic and postsynaptic events (Fig. 6C). Across 27 pairs, the latency from the peak of the first spike in a presynaptic complex spike to the 20% rise in the resulting IPSC was 0.58 ± 0.10 ms (SD) (Fig. 6B). Synaptic jitter, a measure of the spread of latency values, was 0.05 ± 0.02 ms (SD) (Fig. 6B). The initial postsynaptic response to a presynaptic complex spike therefore occurs rapidly and with a remarkably high degree of temporal precision.
Within cartwheel cell pairs, we routinely observed that IPSCs elicited by presynaptic complex spikes rose to similar peak amplitudes in repeated trials (Fig. 7A). We quantified this by calculating the coefficient of variation (CV = SD/mean) of IPSC peak amplitude for each IPSC in a complex IPSC (Fig. 7B). Spikelet transmission failures were either excluded from the data set or included as zero amplitude events. By comparing results from these two conditions, it was evident that transmission failures at the fourth spikelet in a complex spike had a large effect on peak amplitude variability, whereas failures at other spikelet positions did not, probably because failures at earlier positions are less common.
The charge carried by a complex IPSC, as measured by the area under the IPSC, may be a more informative measure than IPSC peak amplitude of the overall effects of a complex spike on the excitability of the postsynaptic cell. We measured the area under IPSCs resulting from simple spikes and from complex spikes with two to four spikelets. Although a particular presynaptic cartwheel cell tended to produce complex spikes with a consistent number of spikelets, our data set included nine pairs in which the presynaptic cell produced complex spikes with 2–3 or 3–4 spikelets. From this data set, we found that three IPSCs were needed to double the charge resulting from a single IPSC, indicating that IPSCs after the initial IPSC yield sublinear increases in total charge (Fig. 7C). This is because successive IPSCs are smaller. This sublinearity, combined with random amplitude fluctuation of each event from trial to trial, had striking effects on minimizing variability in charge transfer and on the effects of spikelet failures. Indeed, the variability of IPSC charge was far lower than the variability of IPSC peak amplitudes, with CV values of <20%, and was constant as the number of spikelets contributing to the complex IPSC increased from one to four (Fig. 7D). When complex IPSCs containing transmission failures were included in the data set, the failures did not decrease the average charge carried by the complex IPSCs or increase the CV. These results suggest that, despite spikelet transmission failures, complex spikes from a given cartwheel cell elicit reproducible charge flux in postsynaptic cells.
Complex spikes inhibit postsynaptic excitability
Cartwheel cells receive excitatory inputs from parallel fibers that extend through the molecular layer of the DCN. Given a likely scenario in which a synaptically coupled pair of cartwheel cells receives excitatory inputs from separate groups of parallel fibers and each group drives its respective cell to fire, we hypothesized that a presynaptic complex spike would more effectively inhibit firing in the postsynaptic cell than a presynaptic simple spike. To test this, we attempted to selectively stimulate one cartwheel cell in a pair by placing a stimulus electrode at several nearby points in the molecular layer. With this approach, we were unable to stimulate parallel fibers synapsing onto one cartwheel cell without also stimulating parallel fibers synapsing onto neighboring cartwheel cells (data not shown), presumably because of the high density of parallel fibers in the molecular layer and the relatively large area that a stimulating electrode excites. We therefore mimicked cell-selective parallel fiber input by injecting depolarizing current into current-clamped cartwheel cells using an EPSC-like waveform (see Materials and Methods). The amplitude of these currents was set so as to reliably elicit a simple spike or complex spike in the presynaptic cell and a complex spike in the postsynaptic cell. Stimuli were delivered to the postsynaptic cell at a fixed time and presynaptic stimuli were delivered within a window ±20 ms from the onset of the postsynaptic stimulus. Presynaptic simple spikes and complex spikes reliably inhibited postsynaptic complex spikes when they were evoked at the same time as the postsynaptic stimulus (Fig. 8A,B). When presynaptic stimuli were offset in time, we found that complex spikes inhibited postsynaptic firing throughout a time window at least 4 ms broader than that generated by presynaptic simple spikes (Fig. 8C). Presynaptic simple spikes and complex spikes were both able to reduce the number of spikelets in the postsynaptic complex spike, often converting complex spikes into simple spikes. We did not observe any instances in which presynaptic firing blocked the firing of the initial spikelet in the postsynaptic cell. This suggests that the response of a cartwheel cell to parallel fiber input that is just above the threshold for complex spike initiation is highly dependent on whether or not a neighboring cartwheel cell provides coincident inhibitory input. Thus, individual cartwheel cells may exert a strong influence over the generation of complex spikes in neighboring cartwheel cells.
Inhibition that can convert a complex spike to a simple spike might influence postsynaptic excitability in multiple ways. For example, in cerebellar Purkinje cells, complex spikes induce Ca2+ influx throughout the dendritic arbor (for review, see Schmolesky et al., 2002). Moreover, spike-timing dependent synaptic plasticity in cartwheel cell dendrites is strongly dependent on postsynaptic Ca2+ levels (Tzounopoulos et al., 2004, 2007). We therefore compared the Ca2+ transients produced by simple spikes and complex spikes in cartwheel cell dendrites. Complex spikes were elicited by somatic current injection and by stimulation of parallel fiber inputs with a stimulus electrode placed in the molecular layer (Fig. 8D). The resulting complex spikes generated Ca2+ transients in the soma and throughout the dendrites (Fig. 8E). Across several cartwheel cells, complex spikes repeatedly elicited large Ca2+ transients in the somatodendritic compartment, whereas simple spikes, elicited by somatic current injection, produced little Ca2+ influx (Fig. 8F). The magnitude of complex-spike-generated Ca2+ transients was relatively insensitive to the method used for evoking the complex spike (r2 = 0.90, linear regression of somatically evoked vs synaptically evoked complex spike Ca2+ transients), suggesting that complex spikes evoked by somatic current injection mimic the somatodendritic Ca2+ flux produced by synaptically evoked complex spikes.
We next examined whether inhibition that converts a complex spike to a simple spike reduces somatodendritic Ca2+ transients. Complex spikes were generated through somatic current injection and IPSPs were evoked via a stimulus electrode placed in the molecular layer. When current was injected to elicit a complex spike, large Ca2+ transients were observed in the soma and dendrites (Fig. 8G,H). When an IPSP was evoked at the same time as the somatic current injection or 2 ms preceding the somatic current injection, the complex spike stimulus yielded a simple spike and a markedly smaller Ca2+ transient. Across five cells, simple spikes produced by the combination of a complex spike stimulus and an IPSP resulted in Ca2+ transients that were significantly smaller than the Ca2+ transients from full complex spikes (p < 0.01). Furthermore, these Ca2+ transients were not significantly different from those generated by simple spikes that were evoked by brief somatic current injection followed by a hyperpolarizing current to prevent complex spike firing (p = 0.14). These results indicate that blockade of complex spike firing by neighboring cartwheel cells must yield a sharp reduction in Ca2+ signaling throughout the dendritic arbor.
The cartwheel cell, with its relatively short axon and high connectivity rate, offered a unique opportunity to probe the impact of complex spikes on inhibitory synaptic transmission. By recording from synaptically coupled pairs of cartwheel cells, we were able to correlate several aspects of spikelet size and shape with the amplitude and timing of postsynaptic events. We found that a high-frequency burst of spikes can enter the axon, efficiently propagate to nerve terminals, and be transmitted as IPSCs with a high degree of temporal precision. Failures to propagate all components of the complex spike appear to originate at invasion of the axon, as axons and terminals were able to sustain transmission of very high-frequency simple spikes. Remarkably, when failures occurred, they had no impact on the trial-to-trial variation in inhibitory charge. We believe this is the result of the characteristics of IPSC summation: because individual IPSCs appear to saturate in amplitude, later events contribute relatively less charge to the IPSC complex, so that their periodic failure has little impact on inhibition.
Mechanisms underlying transmission failures
The failure of later spikelets in complex spikes to elicit a postsynaptic response differs substantially from the situation in cerebellar Purkinje cell complex spikes, where the second spikelet was the most prone to propagation failures measured in the axon (Khaliq and Raman, 2005; Monsivais et al., 2005). In both cartwheel cells and Purkinje cells, spikelets that were small and slow to rise generated most transmission failures (Khaliq and Raman, 2005; Monsivais et al., 2005). In cartwheel cells, the extent to which the membrane potential repolarized between spikelets corresponded with the probability that the next spikelet would yield an IPSC (repolarization potential was not examined in Purkinje cells). These results suggest that transmission failures occurred when Na+ channels inactivated, thereby limiting their ability to contribute to subsequent spikelets.
High firing frequencies can also promote the inactivation of voltage-gated ion channels (Debanne, 2004). In cartwheel cells, however, ISIs did not correlate with transmission probability, and cartwheel cell pairs were able to follow short bursts of simple spikes at frequencies up to 500 Hz and at frequencies matched to those observed within complex spikes. Thus, failures during complex spikes cannot be ascribed to limits on the frequency of axonal spike conduction or on the ability of the synapses to reliably transmit. Two-photon imaging experiments confirmed that transmission failures occur between the soma and proximal axonal boutons. It seems likely therefore that such failures arise at the axon initial segment, probably because of the progressive Ca2+-dependent depolarization that triggers each spikelet (Kim and Trussell, 2007). Purkinje neuron complex spikes do not feature this more gradual rise because they are driven by a large and fast synaptic conductance. Thus, the second spikelet is the smallest, has the shortest ISI, and is the most likely to fail in the axon (Khaliq and Raman, 2005; Monsivais et al., 2005).
Axonal branch points might contribute to action potential propagation failures (Lüscher and Shiner, 1990; Debanne, 2004). Cartwheel cell axons branch profusely (Berrebi and Mugnaini, 1991), and, in our Ca2+ imaging experiments, the most proximal boutons were always found after the first axonal branch point. It is therefore possible that some of the failures we observed occurred at this branch point. The presence of branch points, however, does not necessarily correlate with transmission failures. Two imaging studies in cerebral cortex showed that action potential propagation was quite secure, even when multiple branch points were traversed (Cox et al., 2000; Koester and Sakmann, 2000). Moreover, we found that high-frequency simple spikes propagated well. Thus, our results are most consistent with the hypothesis that transmission failures reflect failures of somatic spikelets to propagate into the axon.
Temporal precision and reproducibility of postsynaptic response to complex spikes
We found that synapses between cartwheel cells exhibited a high degree of temporal precision, with a latency of ∼0.6 ms and jitter of ∼0.05 ms. Compared with other synapses, where similar measurements have been made using conditions comparable with ours (i.e., paired recordings in slice preparations at or near physiological temperature), these values appear to be quite small. In cerebellum, synapses formed by Golgi cells have latencies of 1.05–1.1 ms and jitters of 0.1–0.3 ms (Dugué et al., 2005). At cortical synapses, reported latencies range from 0.6–4 ms and jitters from 0.35 to 0.9 ms (Markram et al., 1997; Feldmeyer et al., 1999, 2002, 2006; Gabernet et al., 2005; Boudkkazi et al., 2007). In the hippocampus, basket cell to pyramidal cell synapses exhibit a latency of 1.46 ms and jitter of 0.31 ms (Neu et al., 2007), whereas connections between mossy fiber boutons and CA3 pyramidal neurons had a latency of 0.7 ms (Geiger and Jonas, 2000). Thus, to our knowledge, cartwheel cell synapses have a latency as brief as any previously measured in these brain regions. This is not surprising, given that cartwheel cell action potentials probably do not travel far before reaching boutons. The jitter at cartwheel cell synapses, however, appears to be exceptionally low. This suggests that Ca2+ influx and vesicular release are tightly coupled in cartwheel cell boutons. Interestingly, inhibitory synapses between cerebellar Purkinje cells appear quite different, with longer synaptic latencies and a higher frequency of transmission failure (Orduz and Llano, 2007).
IPSCs elicited by complex spikes varied little from trial to trial. In the presynaptic bouton, this low variability might reflect high release probability, multivesicular release, and/or multiple release sites (Wadiche and Jahr, 2001; Foster et al., 2002; Oertner et al., 2002; Telgkamp et al., 2004; Christie and Jahr, 2006). Across the synapse, low variability could indicate saturation of receptor binding sites (Wadiche and Jahr, 2001; Foster et al., 2002; Harrison and Jahr, 2003). Surprisingly, transmission failures had little impact on the postsynaptic charge flux elicited by complex spikes. Several factors might contribute to this. First, late IPSCs contributed less charge than the initial IPSC. Second, most transmission failures occurred at the final spikelet. Third, the variability in charge flux from trial to trial even when all spikelets yielded IPSCs might be larger than, and therefore occlude, the variability introduced by occasional transmission failures. Thus, the properties of the cartwheel-to-cartwheel cell synapse seem to compensate for spikelet transmission failures, thereby helping to maintain the identity of complex spikes as they cross the synapse.
Impact of complex spike signaling on postsynaptic excitability
Using paired recordings of near-neighbor cells, 41% of cartwheel cells were synaptically coupled. Thus, cartwheel cells mediate feedforward inhibition to principal cells and form a connected network of interneurons. Understanding how one cartwheel cell modulates the excitability of another will be critical to appreciating how this network operates. Previous studies indicated that at least some cartwheel-to-cartwheel cell connections are depolarizing (Golding and Oertel, 1996; Tzounopoulos et al., 2004), but did not indicate how much a presynaptic cell can influence the membrane potential of a postsynaptic cell or how this would differ with simple spikes and complex spikes. We found that presynaptic simple spikes and complex spikes generated IPSPs powerful enough to convert postsynaptic complex spikes to simple spikes. Presynaptic complex spikes, by reliably eliciting multiple IPSPs, inhibited postsynaptic excitability over a broader time window than did simple spikes. Considered in the context of the cartwheel cell network, this capacity to inhibit neighboring cells would stabilize activity within the network, thereby moderating the inhibition forwarded to the output cells of the DCN. It is currently not known whether connections between cartwheel cells control activity selectively within or between isofrequency bands. Thus, interpretation of our work with respect to the DCN circuit will require additional analysis of patterns of axonal projections.
The impact of the conversion of complex spike to simple spike firing has a profound effect on Ca2+ signaling in the soma and dendrites of cartwheel cells. Imaging experiments indicated that complex spikes generate Ca2+ transients that extend throughout the dendritic arbor, increasing in size with distance from the soma, whereas individual simple spikes, produced by brief somatic current injection or stronger somatic current injection coincident with an evoked IPSP, generated little increase in dendritic Ca2+ (see also, Molitor and Manis, 2003). Parallel fiber synapses onto cartwheel cells undergo spike-timing-dependent long-term plasticity that is strongly dependent on postsynaptic Ca2+ (Tzounopoulos et al., 2004, 2007). Moreover, dendritic depolarization leads to depolarization-induced suppression of excitability that is dependent on Ca2+-induced release of endocannabinoids. Together, these data suggest a novel function of lateral inhibition in which activity in inhibitory interneurons might regulate induction of synaptic plasticity at a neighbor's excitatory synapses by preventing complex spike generation. The high temporal precision of cartwheel-to-cartwheel cell synapses is therefore well suited to the regulation of spike-timing dependent plasticity. In vivo support for this idea has recently been obtained in the cerebellum-like electrosensory lobe of mormyrid electric fish (Sawtell et al., 2007). There, medium ganglion cells feature two spike types, a fast axon spike and a broad dendritic spike. The latter play an essential role in behaviorally relevant spike-timing-dependent plasticity, are thought to elicit Ca2+ signals throughout the dendrites, and are selectively gated by inhibitory input. Thus, rapid interneuronal control of Ca2+ signals may be a general mechanism for regulation of plasticity in cerebellum-like structures.
This work was supported by National Institutes of Health (NIH) Grants DC004450 and NS028901, NIH Training Grants NS007381 (M.T.R., K.J.B.), DC005945 (M.T.R.), and DK007680 (K.J.B.), and a Tartar Trust fellowship (K.J.B.). We thank Sascha du Lac for providing GIN mice.
- Correspondence should be addressed to Laurence O. Trussell, Oregon Health & Science University, 3181 Southwest Sam Jackson Park Road, L335A, Portland, OR 97239.