|Home | About | Journals | Submit | Contact Us | Français|
Purkinje neurons are the output cells of the cerebellar cortex and generate spikes in two distinct modes, known as simple and complex spikes. Revealing the point of origin of these action potentials, and how they conduct into local axon collaterals, is important for understanding local and distal neuronal processing and communication. By utilizing a recent improvement in voltage sensitive dye imaging technique that provided exceptional spatial and temporal resolution, we were able to resolve the region of spike initiation as well as follow spike propagation into axon collaterals for each action potential initiated on single trials. All fast action potentials, for both simple and complex spikes, whether occurring spontaneously or in response to a somatic current pulse or synaptic input, initiated in the axon initial segment. At discharge frequencies of less than approximately 250 Hz, spikes propagated faithfully through the axon and axon collaterals, in a saltatory manner. Propagation failures were only observed for very high frequencies or for the spikelets associated with complex spikes. These results demonstrate that the axon initial segment is a critical decision point in Purkinje cell processing and that the properties of axon branch points are adjusted to maintain faithful transmission.
Communication between most neurons in the central nervous system entails the conversion of synaptic potentials into changes in the rate or timing of action potentials. How this conversion is achieved, and how faithfully the information is transmitted once it is encoded, is a fundamental question in neuroscience. Electrophysiological investigations in vertebrate neurons have demonstrated that typically action potentials initiate at the axon initial segment (Coombs et al., 1957; Hausser et al., 1995; Stuart et al., 1997; Martina et al., 2000; Palmer and Stuart, 2006; Shu et al., 2007). However, investigations of subicular cortical pyramidal cells or cerebellar Purkinje neurons have suggested that spikes may initiate in the 1st node of Ranvier (Colbert and Johnston, 1996; Clark et al., 2005). More recent pharmacological investigations suggest that Purkinje cells, like most other cell types, do initiate action potentials in the AIS (Khaliq and Raman, 2006). One difficulty in determining the initiation site of action potentials is the inability, using traditional recording techniques, to record along the somatic-axonal axis with high spatial resolution.
Cerebellar Purkinje cells generate trains of simple spikes spontaneously, owing to their intrinsic membrane properties (Raman and Bean, 1999), modulated by excitatory (e.g. parallel fiber) and inhibitory synaptic inputs. In contrast, activation of climbing fibers from the inferior olive results in the generation of a large excitatory postsynaptic potential that initiates a complex spike, consisting of one or two full action potentials, between which appear one to three additional partially inactivated spikes, known as spikelets (Ito and Simpson, 1971; Campbell and Hesslow, 1986; Khaliq and Raman, 2005; Davie et al., 2008). Both the spikes and the spikelets occurring in response to the large somato-dendritic EPSP are initiated in the axon, although precisely where in the axon is unknown (Stuart and Hausser, 1994; Khaliq and Raman, 2005; Monsivais et al., 2005; Davie et al., 2008).
The pattern of spikes experienced by presynaptic boutons depends critically upon the propagation of these action potentials along the axon and its collaterals. Propagation failures have been demonstrated in invertebrate preparations and in some mammalian systems (Yau, 1976; Grossman et al., 1979b, a; Smith, 1980; Antic et al., 2000; Debanne, 2004). Regarding the frequency of axonal failures, studies utilizing calcium imaging techniques have demonstrated that failures of axon propagation are rare in pyramidal cells of the neocortex and hippocampus (Mackenzie and Murphy, 1998; Cox et al., 2000; Koester and Sakmann, 2000). These studies are limited to frequencies of approximately 100 Hz or less, owing to the slow kinetics of Ca2+ signaling. Extracellular recordings from cortical axons have suggested that propagation failures may be common during high frequency activity (Meeks et al., 2005), although whole cell axonal recordings have failed to confirm conduction failures in the main axon (Shu et al., 2007), potentially due to the high S/R of whole cell relative to extracellular recordings. In cerebellar Purkinje neurons, propagation failures along the main axon have been observed at frequencies in excess of 250 Hz, and during spikelet generation (Khaliq and Raman, 2005; Monsivais et al., 2005). Whether or not failures occur in axon collaterals has not been directly addressed, owing to the difficulty of recording from these small structures.
Here, through the use of a recent significant advance in voltage sensitive dye imaging (Canepari et al., 2010; Holthoff et al., 2010), we provide direct evidence that in cerebellar Purkinje neurons all action potentials initiate in the AIS and during simple spike activity propagate faithfully down the main axon and into local axon collaterals at frequencies of < 250 Hz. At frequencies above normal physiological rates, or during the generation of spikelets of the complex spike, spike initiation and/or propagation failures can be observed in the main axon proximal to the first collateral. These modifications of waveform output from Purkinje neurons and their targets will have a significant influence on cerebellar function and dysfunction.
We prepared parasagittal slices of cerebellar vermis from mice (L7-tau-GFP [gift of Sascha du Lac, Salk Institute] or pcp2-EGFP [Jackson Laboratories]) in accordance with national and institutional guidelines. Male or female mice (2 to 5 weeks of age) were deeply anesthetized with sodium pentobartital (50 mg/kg) and euthanized through decapitation. The brain was rapidly removed and placed in ice cold (<5° C) cutting solution containing (in mM): 85 NaCl, 2.5 KCl, 3.5 MgSO4, 0.5 CaCl2, 25 NaHCO3, 1.25 NaH2PO4, 10 glucose, 3 myo-inositol, 3 Na-pyruvate, 0.5 L-Ascorbic acid, 75 sucrose, and aerated with 95% O2, 5% CO2 to a final pH of 7.4. We dissected the cerebellar vermis, glued it to an ice cold stage on a Leica microslicer, and cut 200–400 µm slices in ice-cold cutting solution. The sections were placed in an incubator at 35° C for 0.5 h in the cutting solution, and were then transferred to a chamber containing ACSF (in mM: 126 NaCl, 3.0 KCl, 1.0 MgSO4, 2.0 CaCl2, 26 NaHCO3, 1.25 NaH2PO4, 10 glucose, 3 myo-inositol, 2 Na-pyruvate, 0.4 L-Ascorbic acid) at room temperature for at least 0.5 h prior to dye loading and recording in a submersion chamber.
The voltage sensitive dye (VSD) imaging system is constructed around an upright microscope (Olympus BX51WI) equipped to switch easily between three different imaging systems: 1) confocal epi-fluorescence for cell selection and morphological reconstruction (Yokogawa CSU-XAH system, Solamere Tech., Salt Lake City, UT); 2) IR-DIC video microscopy for establishing whole cell recordings; and 3) wide-field epi-fluorescence for voltage sensitive dye imaging (Figure 1). Initially, slices were examined with the confocal imaging system for morphological completeness of their dendritic and especially axonal components, using a 3 line Argon Laser at 50 mW, 488 nm (National Laser Company, Salt Lake City, UT) (Figure 1; supplemental movies 1, 2). Both pcp2-EGFP and L7-tau-GFP mice gave satisfactory to excellent results when examining dendritic and axonal morphologies, although the L7-tau-GFP mice were superior for examining axonal projections, but inferior to pcp2-EGFP mice for examining dendritic morphologies (supplemental movies 1, 2). This process of cell selection illuminated individual neurons typically for less than 1 minute and was determined to be non-toxic. Illuminating Purkinje neurons in this manner, while performing whole cell somatic recordings, for a period of 10 minutes failed to reveal any detectible change in their membrane potential, input resistance, membrane time constant, or action potential characteristics (e.g. threshold, height, width, dV/dt) (not shown).
The signal-to-noise ratio in voltage sensitive dye recording depends, among other factors, on the amount of dye filling the cell, the morphology of the cell, and the proximity of the cell to the surface of the slice. Because we image in wide field mode, all structures to be simultaneously monitored must lie close to the plane of focus. The closer the imaged structures lie to the surface of the slice, the less the image is degraded by light scattering in the tissue. The selection of neurons with processes that are intact and anatomically appropriate for imaging has been one of the major limiting steps in experiments using voltage sensitive dye recording because thin neuronal processes are not visible under DIC video-microscopy. By utilizing transgenic mouse lines expressing GFP or EGFP, we were able to select neurons with optimal morphology prior to dye loading, greatly increasing the efficiency of these experiments. The GFP and EGFP fluorescence does not interfere with voltage sensitive dye signals due to the non-overlapping emission spectra of these two fluorophores.
We loaded selected neurons with voltage dye with a glass patch electrode by filling the tip with dye-free internal solution (in mM: 130 K-gluconate, 7 KCl, 4 ATP-Mg, 0.3 GTP-Na, 10 Phosphocreatin-Na, 10 HEPES, 0.4% biocytin; adjusted to pH 7.4 and 284 mol/kg), and then back-filled with the dye-containing internal solution (JPW3028, 400–800 µM) and performing whole cell somatic recordings from the selected neuron under DIC optics (Figure 1B). The dye JPW3028 is a doubly positively charged analog of the aminostryryl-pyridinium series of lipophilic voltage sensitive dyes available from Molecular Probes (Eugene, OR) as D-6923. This electrochromic dye does not significantly increase the membrane capacitance of the labeled neuron as evident from a number of control measurements showing that the waveform of the electrically recorded action potentials remains unaltered after intracellular application of the dye (e. g. (Canepari et al., 2007)). The physical basis for the lack of the capacitive load effect, characteristic for some FRET as well as some protein voltage probes, is that the interaction of the electric field with electrochromic dyes leads to a charge movement across a very small fraction of the membrane field (the size of the chromophore; (Blunck et al., 2005)). Estimation of cell membrane capacitance in voltage clamp throughout dye loading showed no significant increase (p>0.05, n=4). In addition, parameters monitoring spike kinetics (width, maximum slope and minimum slope) were not different for cells loaded with dye and cells without dye (p>0.05; n=5, each group). Dye-free solution in the tip is necessary because this highly lipophilic dye would otherwise spill onto the slice prior to sealing and patching the cell of interest. Any dye bound to membrane outside the cell of interest increases the background fluorescence (F) and thus degrades recording sensitivity. The selected cell was patched and stained at room temperature for 20–50 minutes by allowing passive diffusion of the dye into the cell. After staining, we carefully detached the patch electrode from the neuron forming an outside-out patch after which the slice is incubated at room temperature for 1–1.5 h, allowing the dye to diffuse from the soma into distal processes. The dye diffuses slowly compared with other dyes of the same size/weight due to its high affinity for lipid membranes. Prior to recording, the perfusion solution temperature was typically increased to physiological level (35–36° C; a small number of recordings were obtained at room temperature) and we re-patch the stained cell with a voltage sensitive dye-free electrode (although the patch pipette now contained 0.4% biocytin for morphological reconstruction) for simultaneous optical and electrophysiology recording.
Cells filled with the voltage sensitive dye were illuminated with a 450 mW frequency-doubled diode-pumped Nd:YVO4 low noise laser emitting at 532 nm (MLL532, Changchun New Industries Optoelectronics Tech. Co., Ltd., Changchun, China). The laser beam was directed to a light guide coupled to the microscope via a single-port epifluorescence condenser (TILL Photonics GmbH, Gräfelfing, Germany) designed to overfill the back aperture of the objective. In this way, near uniform illumination of the object plane was attained. Illumination of the soma was limited by the microscope field-stop in order to minimize unnecessary damage to the recorded neuron. Fluorescence from the stained cell was collected through the objective (60x water immersion, 1.0 NA, Nikon), long pass filtered (RG610 HP, Omega Optical, Brattleboro, VT), and projected onto a 80×80 pixel high speed CCD camera with a 10 kHz maximum frame rate (NeuroCCD-SMQ, RedShirtImaging LLC, Decatur, GA) mounted at the front camera port of the trinocular tube. At the magnification that we used, each pixel of the camera corresponded to a square region 4 µm per side. The calculation of action potential initiation point used the middle of the pixel (e.g. 2 µm) as the value for statistical purposes. Changes in voltage-sensitive dye fluorescence were imaged during simple spikes evoked with single 0.5 ms current pulses injected into the soma, or trains of 20 pulses delivered at 100 to 350 Hz or during spontaneously generated trains of action potentials. Complex spikes were evoked by stimulating climbing fibers with 100 µS, 50–200 µA local stimulation in the dendritic layer approximately 50 µm lateral to the recorded neuron. Each trial lasted approximately 20–60 ms, and the voltage-sensitive dye was illuminated only during this period. We could obtain between 15 and 25 trials before changes in the electrophysiological properties of the recorded neuron (action potential amplitude and duration, resting membrane potential, resting input resistance) were evident, in which case the experiment was terminated.
The S/N obtained during voltage sensitive dye imagining varied from cell to cell depending on the concentration of dye and proximity of the axon to the surface of the slice (e.g see supplemental Figure S2). Thus, to achieve comparable S/N (> 6:1) ratio across experiments, we averaged fluorescence changes from groups of neighboring pixels. Although we typically could obtain this S/N ratio in single trial recordings, in some cases we carried out temporal averaging of a small numbers (4–32) of trials (supplemental Figure S2). To avoid smearing in the trial-to-trial average due to temporal jitter, for each trial we aligned the signal in each pixel to the peak electrical (somatic recordings) or optical (averaged over the entire axon) signal. A low pass binomial filter (1 pass) was applied to reduce high frequency noise. Within trial bleaching and low frequency noise were removed by subtracting an exponential or polynomial function fit to portions of the fluorescence signal containing no spikes (Figure S2C). The □F/F signal from different parts of the neuron varies owing to factors other than voltage (e.g. density of dye, partitioning of dye in inner and outer membranes, depth of process in tissue, etc.). To compare the timing of action potentials in different regions of the neuron, we scaled all spikes to the same baseline and peak height (Figure S2D). The waveform of fluorescence action potential signals shown in figure 2, figure 3, and figure 4 were interpolated to 100 kHz using a cubic spline; raw data points (collected at 10 kHz) are superimposed on the interpolated curve for traces shown on an expanded time scale (see Figure S2F). We determined the location of spike initiation as the portion of the axon where the change in fluorescence corresponding to propagating spikes first reaches half-maximum amplitude. Because the 10 kHz frame rate captured only 3–4 data points on the rising phase of each spike (Figure 2), we fit a line between the two data points surrounding the rising phase half height and solved for the time at which the spike reached half-maximum amplitude (“onset latency”). This method was superior to using either the foot or the peak of the action potentials, since both of these are characterized by a slowly changed membrane potential, allowing for substantial temporal jitter to be introduced by small variations in the accuracy of amplitude measurement.
When examining the probability of spike propagation, we were able to reduce the sampling rate to 5 kHz and examine a larger set of pixels, in order to image a larger portion of the axon and collaterals. This sample rate is sufficient to accurately detect action potentials in the imaged structures, even on single trials. Spike failures were considered to have occurred when the action potential fell below the noise level for all points at and distal to the point in question, and when previous action potentials were observed to give rise to the normal S/N ratio of at least 6 at these same locations. All analyses were performed in NeuroPlex (RedShirtImaging, LLC, Decatur, GA) and Matlab. At the end of the experiment, the axonal arbor of the neuron was recorded either by obtaining a z-stack of the arbor with the confocal microscope, or through processing of the tissue to reveal biocytin.
Movies of action potential initiation and propagation were generated by performing a pixel by pixel (4 µm square) average, aligned in time according to the peak of each action potential (averaged over the entire axon). A mask was applied to exhibit only those pixels lying over the neuronal process of interest. A cubic spline interpolation at 100 kHz was then applied to the average data points (e.g. see Figure S2F). Each pixel was scaled to the same average baseline and peak amplitudes (from 0 to 1). A color code was applied such that the most negative value was blue, the peak (peak of the spike) value was yellow and the half-way value was red (Figure S4A). Although data points were collected every 100 µs, we used the cubic spline interpolation to extrapolate the likely values of relative membrane potential at 50 µs intervals (Figure S4A). The average data was filtered with a low pass (2850 cutoff frequency) Gaussian filter (Figure S4B). The resulting spatial distribution of relative voltages over time was then subjected to spatial smoothing by averaging the values of each pixel with that of the immediately adjacent pixels (that carried signal) in a 3×3 array (3×3 mean filter). This averaging was done for each pixel according to the raw values at the neighboring pixels. From 1 to 5 iterations of this spatial smoothing were performed (Figure S4C, D).
The spikelets of complex spikes could be of relatively small amplitude, which added significantly to the challenge of detection of the propagation of these events and determining the region of propagation failure. This is particularly true since there is temporal overlap of the spike/spikelet waveforms within a complex spike. We considered a spikelet to have failed to propagate down the axon if it became undetectable at the furthest locations of the main axon (e.g. Figure 6,Figure 7 and supplemental Figure 5). We found that the 1st spikelet in the complex spike propagates in the main axon and axon collaterals with a low probability (<6%). To investigate spatial characteristics of spikelet failure, we plotted amplitude of the first and second spikelets, normalized to the height of the first spike, as a function of distance from the soma (n=5 cells).
For measurement of spikelet initiation, we averaged across trials (3–18) and across groups of pixels (6–8) to achieve a S/N appropriate for tracking spikelets, which were often <60% of the amplitude of a full spike. Separate averages were obtained for each spike and spikelet by aligning trials to the peak of each spike and spikelet recorded in current clamp. We then defined times windows corresponding the first spike, first spikelet, and second spikelet, and divided the maximum values found in the spikelet windows by the amplitude of the first spike, which always propagated. Since the first spike and subsequent spikelet occurred close in time (up to 1000 Hz), the spikelet was often superimposed on the falling phase of the first spike. To correct for offset imposed by the falling phase of the first spike on the amplitude of the first spikelet, we subtracted a double exponential function fit to the falling phase of the first spike before measuring amplitude of the first spikelet. An estimate of the noise was obtained by dividing the peak value measured in a time window preceding the complex spike by the amplitude of the first spike. Spikes were considered to fail when the ratio of the spikelet peak to the amplitude of the 1st spike fell to the level of the noise. For two cells, the peak ratio for the second spikelet never returned to baseline due to the inclusion of rare trials in which the second spikelet propagated. By averaging only trials in which the second spike failed to propagate, we calculated that the amplitude of the spikelet dropped to the level of the noise at 94+/−24 microns from the soma.
To examine action potential initiation and propagation in cerebellar Purkinje neurons using voltage sensitive dye imaging, it was important to first obtain a cell in which a significant portion of the main axon and collaterals were intact and within approximately 50 µm of the upper surface of the slice. This process of cell selection was greatly enhanced by examining, using a spinning disk confocal system, Purkinje cells in cerebellar slices from transgenic mice lines in which these cells contain a high density of a fluorescent protein label (GFP or EGFP) (Figure 1; Supplemental movies 1,2; see Methods). Examining morphological features of Purkinje neurons using GFP fluorescence did not reveal any significant adverse effects even with 10 minutes of continuous imaging (see Methods). Once a cell was selected in which the main axon and collaterals could be followed for at least 300 µm, the neuron was patched under DIC optics and loaded with the voltage sensitive dye JPW3028 as previously described (Antic et al., 1999; Djurisic et al., 2004; Canepari et al., 2008).
Following the labeling period, a whole cell recording was again obtained from the soma of the stained neuron. Only cells that exhibited no significant change in their electrophysiological properties (e.g. input resistance, time constant, action potential amplitude, width and peak dV/dt) between the 1st and 2nd whole cell recordings were accepted for further experimentation. In these neurons, the electrical signals were recorded with the somatic patch electrode while the initiation and propagation of action potentials were simultaneously monitored optically at multiple locations along the main axon and collaterals using a high speed (up to 10,000 Hz frame rate) CCD camera (see Methods). We took advantage of the recent critical improvement in voltage-sensitive dye imaging technique based on monochromatic, laser light excitation (532 nm; 450 mW) in the wide-field epi-fluorescence microscopy mode. The combination of an increase in light intensity (beyond the level that can be achieved with an arc-lamp) and the use of near optimal monochromatic excitation wavelengths resulted in a dramatic improvement in the sensitivity of voltage imaging (expressed as the S/N) by a factor that varied from 10 – 40 in different measurements (Holthoff et al., 2010; Canepari et al., 2010). This significant increase in S/N allows us to examine single trial data, typically with minimal spatial averaging and filtering (see Methods). To avoid effects of photodynamic damage, we monitored the width of the action potential from trial to trial, both in the soma and in the fluorescence signal averaged over the whole axon (supplemental Figure 6). Trials were excluded from analysis if they exhibited significant changes in spike width, reflecting potential damage to sodium and potassium channels.
To compare spike initiation timing, we chose the time of half maximal rise during the up-stroke of the action potential at each point along the axon (Figure 2). The half maximal point was chosen because this was a more reliable parameter than either the peak or the foot of the action potential for two main reasons. First, the accuracy of measurements based on the comparison of time-to-half-amplitude is better than the 0.1 ms sampling interval because the measurements are made, using linear interpolation, along the most rapidly changing phase of the signal. Second, noise in the signal can cause an error in the time of the transition to a given signal amplitude; the greater the rate of change of the signal amplitude as it traverses a given amplitude level, the less will be the error in transition time (Malmstadt et al., 1974; Ross and Krauthamer, 1984). These factors favor the time-to-half-amplitude, which is approximately the point of maximal influx of Na+, as the most reliable measure of spike timing (see Methods).
Examination of the timing of simple spike initiation, generated in response to the intrasomatic injection of a short duration (0.5 msec) depolarizing current pulse, with voltage sensitive dye imaging revealed the half maximal amplitude point of the rising phase of the action potential to occur earliest in the axon initial segment (AIS) at an average distance of 16.4+/−1.5 µm from the soma (n=11 cells). This spike then propagated down the main axon toward and beyond the first node of Ranvier at an average rate of 1.03 +/−0.28 meters per second (Figure 2,A-E; n=11). In many of our axonal recordings, the first node of Ranvier was identifiable as the location of the first branch point, giving rise to an axon collateral (Figure 2A,Figure 3A; see (Chan-Palay, 1971; Clark et al., 2005; Orduz and Llano, 2007)). Similar results were also obtained when examining simple spikes that occurred spontaneously at frequencies of 23–105 Hz (see supplemental Figure 1). One advantage of the imaging technique used here is that in many recordings of repetitive discharges the S/N ratio was sufficient to allow us to determine the site of spike initiation for every action potential in a train in single-trial recordings. At all frequencies of simple spike activity examined (0.1–250 Hz), with spikes generated either spontaneously or through the repetitive intrasomatic injection of short current pulses, we found that the action potentials initiated at the AIS (Figure 3B,C; n=345 spikes in 15 cells).
Comparison of the action potential timing in the AIS and the first node of Ranvier demonstrated a consistent latency difference (AIS spiking prior to 1st node) for all action potentials in a spike train (Figure 3). However, this does not mean that action potentials propagated smoothly along the axon. Indeed, in approximately half of the recorded cells we found that the 1st node of Ranvier reached half-maximum AP amplitude prior to immediately adjacent axonal regions. A typical result is shown in Figure 4A, where a mask has been applied to restrict the analysis to only those pixels lying over the center of the axonal processes. The simultaneous presentation of the temporal and spatial dynamics of action potential initiation and propagation is facilitated by a color-coded display. In Figure 4, a relative scale was applied separately to each optical channel (CCD pixel) with the minimum value of the signal amplitude (resting membrane potential) represented by blue/green, the peak of the action potential by yellow, and the half-maximal point by red. Each frame in the sequence shown in Figure 4A is a color-coded representation of the spatial distribution of relative membrane potential at one point in time with individual frames separated by 50 µs. Such a high temporal resolution was derived from action potential waveforms recorded from multiple locations at a frame rate of 10 kHz followed by 20–100 kHz interpolation of the cubic spline fit (see also Figure 2B and supplemental Figure 2). The spatiotemporal distribution of the half-amplitude signal (red) revealed the initiation of action potentials in the AIS followed by propagation down the axon as well as back-propagation into the soma. In approximately half of the neurons in which the 1st node of Ranvier was clearly resolved (n=8/15), we observed that the AIS spike was rapidly followed by spike generation in the first node, followed by the generation of action potentials at a slight lag at the axonal regions between the AIS and 1st node (Figure 4A,B). In the experiment shown in Figure 4, a second “hot spot” of spike initiation was observed distally to the 1st node of Ranvier and corresponded to the origin of a second axon collateral (Figure 4A). Again, the spike conducted rapidly along the main axon between the site of origin of the 1st and 2nd axon collaterals, suggesting that both branch points correspond to nodes of Ranvier (Figure 4A,B). A similar conclusion about the initiation and spread of the AP can be reached based on the spatiotemporal distribution of the peak of the AP signal (yellow). Thus, the action potentials in 8/15 neurons were observed to propagate down the main axon in a “saltatory” manner (see supplemental movies 3,4). Of the seven axons that did not clearly exhibit this characteristic, 4/7 had onset latency measurements that were too noisy in some portion of the axon to clearly discern a local minimum at the first node. The other three axons exhibited a smooth propagation of action potential conduction along the main axon, suggesting that saltatory conduction had not occurred in this small subgroup of cells (3/15).
The amplitude versus time derivative of action potentials reveal that they rise very rapidly, with the rate of change becoming much slower near, and falling to zero at, the peak of the spike (McCormick et al., 1985). Because of this, the membrane potential spends a significant fraction of the duration of the action potential near its peak voltage. One consequence of this can be seen by examining the spatial distribution of relative membrane potential over time during the propagation of the action potential in Figure 4B. Although there are clear and large differences in the membrane potential at the AIS versus the 1st node of Ranvier at each point of time during the rising phase of the action potential in the AIS, once the spike has peaked at this location, the membrane potential at the 1st node of Ranvier is able to “catch up” (Figure 4B). This is because at the 1st node of Ranvier the membrane potential is rapidly rising during the initial phase of local spike initiation, while at the AIS, the membrane potential is already near the peak of the spike and is slowly changing (Figure 4C). Thus, the proportion of the axon that is within a similar membrane potential range changes during different phases of the spike. During the rapidly rising phase (0 to 200 µs), the spatial profile of the membrane potential is quite non-uniform, while near the peak of the spike (250–400 µs), large segments of the axon between the soma and the 1st and even 2nd nodes are simultaneously depolarized to within a narrow membrane potential range (see Figure 4B).
The high spatial and temporal resolution of the voltage imaging technique utilized here provides a unique way to examine possible failures in AP initiation and propagation, especially propagation across branch points, even during high frequency discharges (Figure 5). Typically, the main axon of Purkinje neurons issues a primary collateral within approximately 60–150 µm, and sometimes a second primary collateral could be identified at a distance of 200–300 µm from the soma. These collaterals turn towards the cell body layer, ultimately giving rise to synaptic contacts onto nearby Purkinje neurons and other cell types (Hamori and Szentagothai, 1968; Larramendi and Lemkey-Johnston, 1970; Chan-Palay, 1971; De Camilli et al., 1984; Orduz and Llano, 2007; Watt et al., 2009). These primary collaterals can give rise to secondary, tertiary, and higher order collaterals (Figure 5).
In a series of measurements, we were able to monitor the initiation and propagation of simple action potential discharges in the main axon as well as the primary and secondary collaterals (n=562 stimuli in 6 cells; Figure 5) at frequencies of 100–350 Hz. At frequencies below approximately 300 Hz, the intrasomatic injection of short duration current pulses resulted in the reliable initiation of action potentials in the AIS (Figure 5B; n=13 failures out of 283 spikes; 4.6%). At frequencies between 300 and 350 Hz, however, we observed a slightly higher initiation failure rate (n=47 failures/279 spikes; 16.8%) although these failures to initiate an action potential did not occur evenly throughout all neurons (cf. Figure 5A, B). Although not tested, we assume that these initiation failures were the result of the relative refractory period for spike initiation in the AIS (Hasenstaub et al., 2005).
All action potentials that were successfully initiated at the AIS by the intrasomatic current pulse successfully propagated down the main axon and nearly all successfully invaded the axon collaterals. We observed failures to invade the axon collateral on only 8/502 action potentials (1.6% branch point failure rate), and all of these failures occurred in the axon of one neuron and only at frequencies of >/=300 Hz (see supplemental Figure 3). These rare failures were associated with the generation of a spike of unusually low amplitude in the AIS, suggesting that partial inactivation of the Na+ currents in this portion of the axon had occurred (Khaliq and Raman, 2005; Monsivais et al., 2005). Thus, the propagation of action potentials down the main axon and into primary and secondary branches appeared to be highly reliable, especially at frequencies that are typical for simple spike discharge in vivo (< 250 Hz; (Thach, 1968; Buttner and Waespe, 1984; Medina and Lisberger, 2009)) (Figure 5B).
In another series of measurements, we monitored and analyzed the initiation and propagation pattern of complex spikes in the axonal arbor. Complex spikes are generated in Purkinje neurons in response to activation of their climbing fiber inputs, and typically consist of a full amplitude action potential followed by 1–3 sharp depolarizing potentials of smaller amplitude called spikelets. Sometimes, the final action potential of the complex spike is again of larger amplitude (Figure 6A, Figure 7B) (Khaliq and Raman, 2005; Davie et al., 2008). We examined the point of initiation and axonal propagation of the first full action potentials as well as the subsequent spikelets (Figure 6,Figure 7). As with simple spikes, the action potentials of the complex spike, both full and partial amplitude, occurred first in the AIS (Figure 6A–C). Interestingly, however, the spikes initiated closer to the soma than those of simple spike activity. The first full spike of the complex spike initiated at 3.6 +/− 1.2 µm from the soma, while the first spikelet initiated at 5.2 +/− 1.5 µm and the second at 9.2 +/− 3.3 µm along the AIS. The initiation of spikes near the soma during complex spikes is likely the result of the large depolarization induced by the massive EPSP generated in the soma and dendrite in response to climbing fiber stimulation.
The first spike of the complex, as a rule, propagated faithfully down the axon and collaterals (Figure 6A,Figure 7B). Spikelets, on the other hand, exhibited a variable probability of propagation down the main axon and consequently into the collaterals. In the cells illustrated in Figure 6 and Figure 7, the first spikelet fails to propagate down the main axon and collaterals, while the second spikelet, which is larger in amplitude than the first, propagates with a relatively high probability (Figure 6, Figure 7; supplemental Figure 5). Spikelet propagation probability was measured in single trials. The point at which the first spikelet failed to propagate (defined as the point at which the spikelet peak could not be distinguished from the noise; see Methods) was relatively consistent and occurred at 40–130 µm from the edge of the soma (X=94+/−24 µm; n=5 cells). Spikelets that did not fail to propagate typically increased in amplitude after the 1st node of Ranvier, and thus presumably became full amplitude action potentials (Figure 6,Figure 7). We did not observe failures of propagation along primary or secondary collaterals if the spike or spikelet propagated successfully in the main axon (Figure 7). We also did not observe any failures of propagation of spikes or spikelets (either during the complex spike or simple spike activity) to propagate into the soma, which has been observed in some cases in cortical pyramidal cells (Shu et al., 2007). Finally, although spikelets sporadically occurred at very high frequency (up to 1000 Hz) in some neurons, the average rate of spikes propagating down the axon during a complex spike was 275+/−16 Hz (n=58 complex spikes in 6 neurons).
We have demonstrated here that all fast action potentials, whether generated as simple spikes or as part of a complex spike, initiate in the AIS. Simple spikes then propagate, in a saltatory manner and almost without fail at frequencies within the physiological range (< approximately 250 Hz; (Thach, 1968; Buttner and Waespe, 1984; Kahlon and Lisberger, 2000) ), down the main axon and into primary and secondary axon collaterals. The smaller spikelets of a complex spike often fail to propagate down the main axon, as observed previously (Ito and Simpson, 1971; Campbell and Hesslow, 1986; Khaliq and Raman, 2005; Monsivais et al., 2005). Spike initiation failures mostly occurred at high frequencies (>/= 350 Hz), and were primarily associated with an inability of the axon initial segment to recover sufficiently from the refractory period following the preceding spike. Axon collateral branch point failures were rarely observed under our conditions.
In mammalian neurons action potentials typically (but not always) start in the axon initial segment. Cerebellar Purkinje neurons initiate action potentials, of both simple and complex spike variety, in their axons (Stuart and Hausser, 1994; Clark et al., 2005; Monsivais et al., 2005; Khaliq and Raman, 2006). The precise location of this initiation, however, has been to some degree unclear. Immunocytochemical localization of Nav1.6 and Nav1.1 channels reveal a high density in the Purkinje cell AIS, which occupies approximately the first 10–20 µm of the axon (Jenkins and Bennett, 2001). Immediately following the AIS, the main axon is myelinated, with the 1st node of Ranvier occurring approximately 60–150 µm from the cell body (Gianola and Rossi, 2004; Clark et al., 2005; Khaliq and Raman, 2006; Orduz and Llano, 2007). Nodes of Ranvier are associated with a high density of Nav1.6 Na+ channels (Caldwell et al., 2000; Duflocq et al., 2008) and thus represent highly excitable regions. Extrapolation of the action potential initiation point, using group data consisting of extracellular recordings from single axonal locations simultaneous with whole cell recordings from the soma of Purkinje neurons, has suggested that spikes may initiate in the 1st node of Ranvier, as opposed to the AIS (Clark et al., 2005).
Pharmacological investigations, on the other hand, suggest that action potentials of Purkinje neurons do indeed initiate in the AIS (Khaliq and Raman, 2006). Thus, reduction of Na+ currents in the AIS, but not the 1st node of Ranvier, results in strong effects on action potential threshold and repetitive spiking characteristics. Our data provide direct and clear evidence that the AIS is the point of initiation of Na+ dependent action potentials in cerebellar Purkinje neurons. On average, the region of the AIS that initiated an action potential earliest during simple spike activity was 16 µm from the soma, near the end of the AIS and the beginning of myelination. In other cell types, this region of the AIS has the lowest local threshold owing in part to a high density of Nav1.6 channels characterized with an activation threshold that is 10–15 mV more negative than other types of Na+ channels (Colbert and Pan, 2002; Boiko et al., 2003; Hu et al., 2009). Another contributing factor is the small size of this compartment with a small capacitive and conductance load, such that the entry of Na+ ions is able to have a rapid and strong effect on local membrane potential. We directly observed saltatory conduction of the action potential in approximately half of cells recorded. In these cases the 1st node of Ranvier discharged immediately after the AIS, but before the intervening axonal membrane. This “jumping” of the action potential from the AIS to the 1st node may help to explain prior difficulties in observing the precise location of action potential initiation in Purkinje neurons, since this will significantly shorten the interval between AIS and 1st node action potentials. During complex spikes, action potentials initiated closer to the soma, with the first spike initiating very near the cell body, and later spikelets initiating a few µm further down the AIS. Presumably, this near somatic initiation of the action potentials of the complex spike results from the fast and large depolarization of the somatic-dendritic membrane potential owing to the large EPSP generated by the climbing fiber stimulation.
Electrophysiological investigations of invertebrate axons, as well as computational studies (Zhou and Chiu, 2001), have suggested that action potential propagation may fail at branch points (Grossman et al., 1979b, a; Smith, 1980; Debanne, 2004). Direct tests of axonal conduction failures in axon collaterals, or at branch points, in the mammalian brain have been difficult to obtain owing to their small size. Simultaneous somatic and extracellular axonal recordings in cerebellar Purkinje cells reveal that conduction failures along the main axon only occur at frequencies greater than approximately 250 Hz, and often only after prolonged discharge (Khaliq and Raman, 2005; Monsivais et al., 2005). Although we observed spike initiation failures at these high frequencies, we found that once a spike was initiated in the AIS, it nearly always propagated down the main axon, and only rarely did it fail to propagate into axon collaterals. Our lack of observation of axon conduction failures, even at high frequencies, may be because we did not examine the effects of prolonged high frequency discharge, as was done by other investigators (Khaliq and Raman, 2005; Monsivais et al., 2005; Khaliq and Raman, 2006). Extracellular recordings from Purkinje neurons in awake, behaving animals indicate that simple spikes typically occur at frequencies of 30–120 Hz, with occasional peaks in firing rate in excess of 250 Hz (Thach, 1968; Buttner and Waespe, 1984; Kahlon and Lisberger, 2000). Thus, failures of propagation of simple spike activity along the main, myelinated, axon and axon collaterals are likely to be rare under normal conditions.
The main axon of Purkinje neurons gives rise to the first axon collateral approximately 60–150 µm from the soma. This axonal process typically bends back towards the cell body layer, and gives rise to higher order collaterals that innervate other Purkinje, Golgi, and Basket cells (Hamori and Szentagothai, 1968; Larramendi and Lemkey-Johnston, 1970; De Camilli et al., 1984; Watt et al., 2009). Failure of the action potential to propagate into the primary axon collateral was rare and associated with a smaller amplitude action potential in the main axon proximal to the branch point, even through this spike propagated along the main axon past the branch point (see supplemental Figure 3). The smaller amplitude of the action potential in the main axon suggests that these spikes were compromised by falling within the spike refractory period, and thus may have not generated enough local current flow in the branch point to initiate a spike in the primary collateral. In other cells, the branch point was observed to boost the action potential, discharging before immediately adjacent regions of the axon, a pattern consistent with branch points containing a high density of Na+ channels (Caldwell et al., 2000; Duflocq et al., 2008). Presumably, once action potentials generated in the AIS are successful at initiating a spike at the 1st axonal branch point, the boosting of this spike by the local high density of Na+ channels insures the propagation of these spike into the axon collaterals as well as down the main axon. Thus, not only is the AIS a critical decision point for the transformation of synaptic activity (and intrinsic conductances) into action potential activity, but the region between the AIS and 1st node of Ranvier functions, particularly during complex spikes, as a critical region for deciding which spikes are propagated and which are not.
Prior electrophysiological recordings from Purkinje cell axons during complex spike generation indicate that many spikelets, especially those of smaller amplitude, may not propagate down the main axon (Ito and Simpson, 1971; Campbell and Hesslow, 1986; Khaliq and Raman, 2005; Monsivais et al., 2005). Similarly, we found that the small spikelets often do not successfully initiate an action potential in the 1st node of Ranvier following their generation in the AIS, and thus fail to propagate down the main axon or into axon collaterals. Using high resolution Vm-imaging, we provide direct evidence for the precise location of both spikelet initiation (see above) and failure. Spikelets, which are small in amplitude, are presumably generated by a significantly reduced inward flux of Na+ ions at the AIS. During failures of propagation, this reduced Na+ current is insufficient to provide enough axial current flow down the main axon to bring the 1st node of Ranvier to spike threshold, and thus the spikelet fails to propagate between the AIS and the 1st node (see results). In vivo, climbing fiber activation typically results in at least two successfully propagating axonal action potentials at 200–300 Hz, followed by a brief pause in action potential activity (Bloedel and Roberts, 1971; Sato et al., 1992). Owing to the presence of a large low threshold Ca2+ current in deep nuclear neurons (Llinas and Muhlethaler, 1988; Aizenman and Linden, 1999), the synchronized discharge and pause of several Purkinje neurons may result in a pause followed by enhanced rebound activity in the output of the cerebellum (McDevitt et al., 1987).
Our results confirm the high fidelity of action potential conduction in mammalian axons (Cox et al., 2000; Koester and Sakmann, 2000; Raastad and Shepherd, 2003; Shu et al., 2007 ), with relatively rare failures in the main axon occurring as a reduction in AIS spike amplitude (and presumably inward current flow) which subsequently results in an inability of the AIS spike to initiate a full action potential in the 1st node of Ranvier. During simple spike activity, this appears to have relatively little functional consequence, since propagation failures down the main axon and into axon collaterals were almost completely absent at physiological frequencies. During complex spikes, however, the failure of AIS-1st node propagation provides a variable filter on the output of the cerebellar cortex onto their target neurons. Our results also demonstrate that in cerebellar Purkinje neurons, as in other mammalian cell types, the axon initial segment is the key location for the initiation and propagation of fast, Na+ dependent action potentials and represents a key location for the conversion of synaptic potentials into alterations in neuronal firing rate and timing.
We thank Sascha du Lac (Salk Institute) for the L7-tau-GFP mice. This work was supported by NIH (DAM & DZ) and NSF (DZ), by an NSF graduate research fellowship (AF) as well as the Kavli Institute for Neuroscience at Yale.