Action potential fidelity during normal and epileptiform activity in paired soma–axon recordings from rat hippocampus

Abstract

Although action potential initiation and propagation are fundamental to nervous system function, there are few direct electrophysiological observations of propagating action potentials in small unmyelinated fibres, such as the axons within mammalian hippocampus. To circumvent limitations of previous studies that relied on extracellular stimulation, we performed dual recordings: whole-cell recordings from hippocampal CA3 pyramidal cell somas and extracellular recordings from their axons, up to 800 μm away. During brief spike trains under normal conditions, axonal spikes were more resistant to amplitude reduction than somatic spikes. Axonal amplitude depression was greatest at the axon initial segment < 150 μm from the soma, and initiation occurred ∼75 μm from the soma. Although prior studies, which failed to verify spike initiation, suggested substantial axonal depression during seizure-associated extracellular K⁺ ([K⁺]_o) rises, we found that 8 mm[K⁺]_o caused relatively small decreases in axonal spike amplitude during brief spike trains. However, during sustained, epileptiform spiking induced in 8 mm[K⁺]_o, axonal waveforms decreased significantly in peak amplitude. During epileptiform spiking, bursts of two or more action potentials > 20 Hz failed to propagate in most cases. In normal [K⁺]_o at 25 and 32°C, spiking superimposed on sustained somatic depolarization, but not spiking alone, produced similar axonal changes as the epileptiform activity. These results highlight the likely importance of steady-state inactivation of axonal channels in maintaining action potential fidelity. Such changes in axonal propagation properties could encode information and/or serve as an endogenous brake on seizure propagation.

Proper neural function requires reliable transmission to downstream targets. Accordingly, axonal action potential propagation is an area of much historical and recent interest (Swadlow et al. 1980; Debanne, 2004). Propagation may be highly faithful (Allen & Stevens, 1994; Mackenzie et al. 1996; Cox et al. 2000; Forti et al. 2000; Koester & Sakmann, 2000; Raastad & Shepherd, 2003), but also unfaithful in certain physiological conditions (Dyball et al. 1988; Ducreux et al. 1993; Luscher et al. 1994; Debanne et al. 1997) and pathophysiological conditions (Malenka et al. 1981; Poolos et al. 1987; Soleng et al. 2003a; Meeks & Mennerick, 2004). Clearly it is important to understand the factors that promote and compromise axonal signalling.

Previous studies of axonal signalling in the hippocampus are clouded by technical and interpretive limitations. Several previous studies found that mild extracellular potassium concentration ([K⁺]_o) elevations depress propagating action potentials (Malenka et al. 1981; Poolos et al. 1987; Meeks & Mennerick, 2004). However, these studies are complicated by the paradoxical effects of [K⁺]_o rises on hippocampal excitability (Hablitz & Lundervold, 1981). [K⁺]_o rises can initiate seizures (Zuckermann & Glaser, 1968; Korn et al. 1987; Chamberlin & Dingledine, 1988; Somjen & Muller, 2000), so it is not clear whether depression of axonal spikes (Malenka et al. 1981; Poolos et al. 1987; Meeks & Mennerick, 2004) is directly associated with [K⁺]_o rises themselves or with secondary effects. Many studies, because they employ extracellular stimulation, cannot distinguish initiation changes from conduction changes. The ectopic nature and location of these stimulation methods may also unintentionally alter spike conduction properties. Furthermore, compound afferent volley depression (Malenka et al. 1981; Poolos et al. 1987; Soleng et al. 2003a; Meeks & Mennerick, 2004) during repetitive stimulation could result from binary failure in some axons and/or amplitude reduction across all fibres. Finally, the cell types giving rise to compound (Balestrino et al. 1986; Poolos et al. 1987) or single-fibre (Raastad & Shepherd, 2003; Meeks & Mennerick, 2004; Soleng et al. 2004) action potentials are also unknown. To circumvent these limitations, we simultaneously recorded intracellular somatic and extracellular axonal action potentials from hippocampal CA3 neurones up to 800 μm apart. This technique allows observation of action potentials as they initiate near the soma and propagate down the axon.

Here we investigated axons in baseline conditions and elevated [K⁺]_o conditions previously shown to compromise axonal signalling. Under baseline conditions, we found that axons propagated spikes > 99% of the time, and did so without exhibiting waveform changes evident at the soma. We also found that axonal recordings greater than 150 μm from the soma were the most resistant to train-dependent amplitude depression, while initial segment spikes showed depression similar to somatic spikes. In 8 mm[K⁺]_o, we observed minimal axonal spike amplitude reduction during brief trains initiated at the soma, and found no evidence for increased failures. However, many CA3 neurones experienced long-lasting plateau potentials (PPs) resembling epileptiform activity (McCormick & Contreras, 2001). PPs often occurred in response to brief depolarizing current injections in elevated [K⁺]_o. During PPs, axonal spike amplitude was strongly depressed despite lower overall spiking frequency and even when somatic spike waveforms were preserved. In addition to the overall decrease in the amplitude of axonal spike waveforms, burst spikes (> 20 Hz) often failed to propagate altogether. Axonal waveform changes were replicated in baseline [K⁺]_o by PPs and sustained somatic current injections, but not by prolonged spiking in the absence of steady-state somatic depolarization. Thus, axonal signals decrease significantly when CA3 somata are depolarized above action potential threshold for extended periods of time. This mechanism may limit seizure propagation by interrupting network over-excitation.

Methods

Slice preparation

Sprague-Dawley rats, 12–16 days old, were anaesthetized with halothane by inhalation and decapitated. Animal use protocols were approved by the Washington University Animal Studies Committee. Brains were dissected into oxygenated ice-cold slicing buffer containing (mm): 87 NaCl, 25 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 0.5 CaCl₂, 3.0 MgCl₂, 25 glucose, 75 sucrose, and 0.5 kynurenic acid. Transverse slices of 300 μm were made through the hippocampus using a vibratome (World Precision Instruments, Sarasota, FL, USA). Slices were immediately transferred to artificial cerebrospinal fluid (aCSF) containing 2 mm kynurenic acid at 34°C for 30 min, and then incubated at room temperature (22–25°C) for at least 30 min prior to recording. The aCSF contained (mm): 125 NaCl, 25 NaHCO₃, 3 myo-inositol, 3 sodium pyruvate, 2.5 KCl, 1.25 NaH₂PO₄, 2 mm CaCl₂, 1 mm MgCl₂, 0.4 ascorbic acid, 10 glucose. Salts and compounds were obtained from Sigma-Aldrich (Saint Louis, MO, USA) unless otherwise noted.

Electrophysiological recording

All recordings were made in baseline aCSF containing ionotropic glutamate receptor antagonists (1 μm 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulphonamide (NBQX) and 25 μmd-(−)-2-amino-5-phosphonopentanoic acid (d-AP5; Tocris-Cookson, Ellisville, MO, USA)) and a GABA_A receptor antagonist (100 μm picrotoxin) to eliminate synaptic activity and related secondary spiking. CA3 somata were identified under infrared differential interference contrast (IR-DIC) optics on an upright fluorescence-capable microscope (E600FN, Nikon Instruments, Tokyo, Japan) viewed with an infrared camera (DAGE-MTI, Michigan City, IN, USA). Patch and extracellular recording pipettes were pulled from borosilicate glass (World Precision Instruments) with a Flaming/Brown micropipette puller (model P-87, Sutter Instrument Company, Novato, CA, USA). Internal solution contained (mm): 130 potassium gluconate, 4 NaCl, 0.5 CaCl₂, 5 EGTA, 10 Hepes, 4 Mg-ATP, 0.3 Na₂-GTP, and 10–200 μm AlexaFluor 488 or AlexaFluor 568 hydrazide (Molecular Probes, Eugene, OR, USA). Open-tip pipette resistance in the bath was typically between 4 and 7 MΩ. Seal formation and whole-cell patch configuration were achieved in voltage-clamp mode. Somatic recordings were obtained using an Axoclamp 2B amplifier or a MultiClamp 700B amplifier (Axon Instruments, Union City, CA, USA). Bridge balance and pipette capacitance compensation were controlled either manually with the Axoclamp 2B amplifier or using Multi-Clamp 700B Commander software attached to the MultiClamp 700B amplifier (Axon Instruments). Signals were acquired at 20–100 kHz, filtered using a 10 kHz (soma) or 3 kHz (axon) 8-pole Bessel filter and digitized using a DigiData 1322A 16-bit A/D converter (Axon Instruments) running on a Pentium 4-based personal computer running pCLAMP 9 acquisition software (Axon Instruments).

CA3 neurones were current-clamped to their initial resting potentials (∼−70 mV) in baseline solution while the AlexaFluor 488 hydrazide filled the cell for 10–20 min. During this time cells were stimulated by injecting depolarizing currents (10–400 pA, 0.75–1 s in duration) to determine their baseline firing properties. Meanwhile, a second patch pipette for extracellular recording (tip resistance 1.5–5 MΩ) containing aCSF and 0.2 mg ml⁻¹ (0.02%) Lucifer Yellow CH (LY) was positioned under dim fluorescence near the soma (Figs 3, 4 and 5), a proximal segment of axon (Fig. 5), or an axon branch > 150 μm from the soma travelling in the direction of CA2 (thus likely giving rise to the Schaffer collateral axons). Upon contact with the axon, suction was applied, resulting in a 2- to 3-fold increase in pipette resistance to 5–15 MΩ, which remained stable within 10–20% over the recording period. After achieving the dual somatic–axonal recordings, somas were stimulated using depolarizing current steps. In cases where cells depolarized by more than 5 mV or spike frequencies in response to current injection changed by more than 10% following establishment of axon recordings, we assumed cell damage from phototoxicity or mechanical trauma, and recordings were aborted (less than ∼15% of all recordings). Recordings where axonal signals in baseline conditions were less than twice the peak–peak noise amplitude were also discarded. Loose-seal somatic and axonal recordings were performed in voltage-clamp mode of the patch amplifier at a command potential of 0 mV. Clamp current signals were inverted and scaled by the resistance of the recording electrode for display as extracellular voltage. For 8 mm[K⁺]_o studies, we analysed only those recordings in which axonal signals recovered to baseline levels after washout to baseline [K⁺]_o.

Figure 3. Comparison of intracellular somatic and extracellular action potential measurements.

A, intracellular record of a train of action potentials elicited by a 1 s, 100 pA current injection at the soma from a resting potential of −70 mV. Trains at 10–30 Hz showed little accommodation at this stimulus strength and duration. B, plot of the first derivative of the somatic trace in A. C, extracellular ‘loose-seal’ recording at the soma of the cell shown in panels A and B. D–F, first action potentials from the train shown in A–C, respectively. Dotted vertical line indicates the peak of the somatic derivative (dV/dt). The extracellular waveforms peak at the same time as intracellular dV/dt, serving as a further indication that local capacitive currents (I_c=C_mdV/dt) dominate early portions of the extracellular waveform (see Fig. 4 for further discussion). G–I, intracellular somatic membrane potential, derivative (dV/dt), and extracellular axonal recordings in response to a 0.75 s, 200 pA current injection at the soma from a resting potential of −74 mV. Note the pronounced depression of maximum dV/dt later in the train. J–L, first action potentials from the trains shown in G–I. Despite the difference in signal/noise ratio, the polarity (positive) and kinetics are comparable to extracellular recordings at the soma. Vertical bar indicates the time of maximum somatic dV/dt. The peak of the axonal action potential occurs with a delay of 0.94 ms. Axonal latency varied with recording location (see Fig. 5C). Recording distance from the neurone/axon pair in G–L was measured in 3-dimensions to be 422 μm.

Figure 4. Single axon recordings reflect local membrane currents.

A, extracellular recordings taken at short distances from the soma display negative local potentials (inset) and decay exponentially with a length constant of 12.5 μm. The exponential fit to this curve is shown by the dashed line. The horizontal dotted line represents detection threshold. Inset: when the extracellular suction electrode was brought into contact with the somatic membrane, the ‘loose seal’ configuration was achieved as seen in Fig. 3A–F (pipette resistance: 6.8 MΩ). A positive signal is measured, similar in polarity and kinetics to axonal signals. The polarity and kinetics indicate this initial signal results from capacitive current flow across the membrane. Ionic currents are likely to contribute to later portions of the waveform (Jack et al. 1975). When displaced small distances from the somatic membrane and suction applied to maintain relative pipette resistance (3.5–4.2 MΩ), local potentials reverse direction, indicating a flow of positive charges (source current) out of the extracellular electrode during somatic spiking. The pipette thus contributes to the local source currents within 25–30 μm of the soma. Traces represent spike-triggered averages of 8 isolated somatic spikes. Calibration bar represents 2 ms and 100 μV. B, extracellular potential when the extracellular electrode was placed ∼5 μm from a visualized axon, 567 μm from the soma. Suction was applied (pipette resistance changed from ∼3 to ∼6 MΩ) as during axonal recording. No extracellular waveform was observed. Traces represent spike-triggered averages of 8 individual spikes. C, when the axon was brought into the lumen of the extracellular electrode following suction (pipette resistance 6.5 MΩ), a local positive potential was recorded at specific latencies from the somatic spike (see Fig. 5C).

Figure 5. Dependence of spike waveforms on spike history and recording distance.

A, spiking-dependent depression of somatic and axonal signals during brief action potential trains. Values are normalized to the first action potential in the train. Somatic slope (filled bars) was depressed severely by action potential 10 (P < 0.01, n = 42 records from 36 neurones with mean frequency 23 ± 1 Hz). Axons from recordings immediately adjacent to the soma up to 800 μm away from the soma were pooled to yield estimates of axonal amplitude depression, which was much lower than somatic slope depression (open bars, P < 0.01, n = 36). B, the main plot (open symbols) represents axonal spike 10 amplitude depression during 15–30 Hz trains, plotted as a function of distance of the recording from the soma. The dotted line represents the degree of amplitude depression from 24 recordings performed > 150 μm from the soma. The inset (filled symbols) represents the same records after somatic slope depression was subtracted. Therefore, a value of zero indicates that somatic and axonal depression were exactly the same. The data were fitted by a single exponential rise to a maximum (continuous line, length constant: 102 μm). As indicated in Fig. 3, recordings < 30 μm from the soma may have been partially affected by local somatic signals. C, plot of axon–soma latency versus axonal recording distance. Negative values indicate the axon signal peaked before somatic dV/dt. Recordings up to 500 μm from the soma were fit with a bi-linear function of the form y=y_o+a(x−b) +c|x−b|. In this equation, b represents the distance corresponding to the nadir point (the average site of spike initiation). Antidromic conduction velocity can be calculated as 1/(c−a) and orthodromic conduction velocity 1/(c+a). Fit values for this data set reported an initiation site at 74.2 μm from the soma, an antidromic conduction velocity of 0.23 m s⁻¹, and orthodromic conduction velocity of 0.37 m s⁻¹.

Imaging and immunocytochemistry

Slices were fixed in 4% paraformaldehyde for 20–30 min immediately following recording. Filled cells were imaged using the Nikon C1 confocal system running on a Nikon Eclipse TE300 inverted microscope. Filled cells were reconstructed in three dimensions using Nikon C1 software, and three-dimensional measurements of axonal recording distance were made using MetaMorph 6 software (Universal Imaging, Downingtown, PA, USA).

For studies of myelination (Fig. 2), cells were filled with 4 mg ml⁻¹ (0.4%) biocytin in addition to AlexaFluor 568 hydrazide. Following fixation, slices were transferred to block solution consisting of PBS containing 0.1% Triton X-100 and 2% goat serum for 30–90 min. Slices were then treated with an antibody against myelin basic protein at 4°C overnight (1: 1000, SMI-99, Sternberger Monoclonals, Lutherville, MD, USA). We visualized biocytin and the antimyelin antibody using streptavidin-conjugated Cy3 (1: 1000) and antimouse IgGγ2b conjugated to AlexaFluor 488 (1: 1000), respectively (Molecular Probes).

Figure 2. Lack of myelination of CA3 pyramidal cell axons.

The primary image shows a CA3 neurone filled with biocytin. The neurone was visualized with streptavidin-conjugated Cy3 (red), and myelin was visualized with an antibody against myelin basic protein and secondary antibody conjugated to AlexaFluor 488 (green). Although occasional oligodendrocytes are present in the region (green labelling), we observed no colocalization of myelin staining with axons from the CA3 neurones studied (pseudocoloured pink to highlight, n = 22). The example is from a postnatal day 15 animal. The inset shows myelin basic protein staining in the corpus callosum from the same slice at the same depth of focus, exposure, and gain settings as the primary panel. Both images represent a confocal projection 50 μm thick from the surface of the slice.

Data analysis and statistics

Somatic and axonal action potentials were analysed using MiniAnalysis software (Synaptosoft, Decatur, GA, USA). Axonal spikes occurring outside an 8 ms window surrounding the peak of the somatic action potential were assumed to arise from axons other than the axon of interest in the extracellular recording pipette, and these spikes were eliminated from display and analysis. Extracellular axonal action potential amplitude was calculated by subtracting the trough from the peak of the action potential signal as previously reported (Raastad & Shepherd, 2003; Meeks & Mennerick, 2004). Somatic action potential amplitude was measured as total voltage difference between prestimulation resting potential and peak action potential height. Somatic spike threshold was calculated based on the second derivative of the action potential waveform implemented by MiniAnalysis software. We found that our extracellular axonal signal amplitude was dominated by fast capacitive components (see Results and Figs 3 and 4). Since capacitive currents are proportional to dV/dt, we compared extracellular axonal amplitudes to the first derivative of the intracellularly recorded membrane potential with respect to time (Johnston & Wu, 1995). We calculated the derivative of somatic records in two ways. First, we calculated action potential slope by linear regression fit to the rising phase of the action potential (MiniAnalysis, Synaptosoft). Second, we calculated the derivative (dV/dt) of the membrane potential trace using pCLAMP 9.2 software (Axon Instruments). Action potential slope measurements and peak–trough measurements from the derivative traces produced indistinguishable results (correlation between peak dV/dt and linear regression of rising phase: r²= 0.96, P < 1 × 10⁻¹⁰, slope value: 1.04). In all analyses, we used somatic rising-phase slope values (from linear regression) for summary statistics. Derivative traces are used for graphical comparison with axonal signals.

Changes in somatic and axonal action potentials were calculated by normalizing the amplitude or slope of individual action potentials to the corresponding initial action potential values from 6–10 trials within a given condition. Depression of axonal signals always refers to depression of the peak–peak amplitude. When axonal peak–peak amplitude within the 8 ms window around somatic spikes was less than the mean plus 2 standard deviations of the peak–peak noise, we considered the event to be a detection failure, cf. Khaliq & Raman, (2005), but not necessarily a true propagation failure (Fig. 10). Spike-triggered average waveforms were created by aligning eight or more axonal signals based on the time of the somatic peak voltage (Fig. 4) or peak somatic dV/dt (Fig. 9). Statistical significance was determined using paired, two-tailed Student's t test and ANOVA, where applicable. Significance level was set at P < 0.05. All error bars represent standard errors of the mean. P-values of correlation are based on the r² value of the fitted function.

Figure 10. Somatic burst spikes during PPs fail to propagate.

A, raw extracellular spikes from a paired somatic–axonal recording during a PP. Axonal traces were aligned based on the peak of somatic dV/dt of spikes during a PP in 8 mm[K⁺]_o. Condition labels appear to the left of each trace. ‘Initial’ spike represents the first spike in response to the 1 s current injection. Individual ‘non-burst’ and ‘burst’ detection failures are indistinguishable from noise. Scale bar: horizontal 1 ms, vertical 100 μV. B, somatic dV/dt-triggered averages of 18 axonal responses in each of the categories to the left in A except the initial spike, which is reproduced for reference. Spike-triggered averaging for ‘non-burst’ spikes revealed detectable, but strongly diminished, axonal waveforms. In contrast, spike-triggered averaging from ‘burst’ spikes revealed no detectable waveform (compare with averaged noise, bottom trace), indicating propagation failure. C, scatter plot of individual amplitudes for burst (filled circles) and non-burst (open circles) events during a single PP. We observed no failures before spike 60 in this PP (5.1 s from the initial spike). The horizontal dotted line represents detection threshold. As seen in the raw traces, burst and non-burst detection failures are not easily distinguishable from one another on an individual basis. D, average amplitudes from successful detections and failures during the PP. Non-burst detection failures, while having an average amplitude below detection threshold (horizontal dotted line), had a significantly larger amplitude than burst failures. Asterisk indicates P < 0.01. Results are representative of 2 of 4 plateau cells with sufficient numbers of failure events from bursts and non-bursts for spike-triggered averaging. The remaining two cells had no discernible waveform in spike-aligned averages from either non-burst failures or burst failures, indicating that outright failure is also possible during non-burst spikes in some cells.

Figure 9. Summary of the effect of PPs induced in 8 mm[K⁺]_o on somatic and axonal action potentials.

A, summary of somatic and axonal responses from 6 neurones that underwent PPs. Somatic slope depression in this data set was significantly reduced in 8 mm[K⁺]_o, both at spike 10 and during PPs (filled bars, P < 0.02 for both, n = 6). However, there was no additional slope depression on average during PPs than was seen by spike 10 of direct stimulation (P= 0.70, n = 6). In contrast, axonal amplitude was significantly reduced during PPs compared with spike 10 preceding the PP (open bars, P= 0.01, n = 6). B, effect of burst phenomena on axonal amplitude depression during PPs. Action potential bursts during PPs contained high-frequency spiking (> 20 Hz), and showed the most severe somatic slope and axonal amplitude reductions (see Fig. 8B, no. 4). Somatic action potentials occurring during PPs, but not involved in burst events (‘non-burst’ spikes) showed no significant change in slope depression (filled bars) with respect to the 10th spike before PP generation (P= 0.67). Despite this, axonal spike amplitude (open bars) during ‘non-burst’ events was significantly more depressed than the 10th spike preceding the PP (P < 0.01, n = 6). Somatic action potentials occurring during burst events (defined by instantaneous frequency > 20 Hz) showed significantly more slope depression and axonal amplitude depression than non-burst events during the PP (both P < 0.01). It should be noted that the strong axonal amplitude depression seen during bursts reduced these signals to the level of detection, and our measurements may represent an under-estimate of the actual axonal signal depression (Fig. 10). C, summary of the relationship between absolute somatic slope and axon amplitude before and during PPs. Somatic spikes with slopes below 90 mV ms⁻¹ during PPs (dark grey bars) were associated with more severe axonal amplitude depression than spikes with the same slopes in 2.5 mm[K⁺]_o (black bars) or 8 mm[K⁺]_o before PPs (light grey bars, P < 0.05, n = 6). At slope values above 90 mV ms⁻¹, axonal spike amplitudes in the 3 conditions were similar (P > 0.25, n = 6).

Results

Axonal action potentials are highly resistant to waveform reductions ≥ 150 μm from the soma

We studied action potential initiation and propagation in axons from CA3 pyramidal neurones in transverse P12–P16 rat hippocampal slices. We successfully recorded 36 CA3 soma–axon pairs. An example recording is shown in Fig. 1. Neurones were filled with 10–200 μm AlexaFluor 488 hydrazide for 10–20 min to allow for dye diffusion before achieving dual somatic–axonal recording configuration (Fig. 1A). The average distance from the cell body to the site of axonal action potential recording was 288 ± 35 μm (n = 34; we had no recording distance data for two cells). In the presence of ionotropic glutamate and GABA receptor antagonists, we induced action potential trains at the soma by injecting 50–400 pA depolarizing currents for 0.75–1 s from a resting potential of ∼−71 mV (Fig. 1B). Action potential threshold measured in the soma was −40 ± 1 mV and changed by 8 mV over the first 10 action potentials of the train (to −32 ± 2 mV at action potential 10, n = 12). Axon recordings showed distinct, time-locked action potential signals in response to somatic spiking (Fig. 1C, also Fig. 3).

Figure 1. Paired CA3 soma–axon recordings.

A, confocal reconstruction of a CA3 pyramidal neurone filled with AlexaFluor 488 hydrazide. Image represents a maximum projection created from a 61 μm thick confocal image series. The asterisk marks the site of extracellular axon recording, measured in three dimensions at 641 μm from the soma. B, simultaneous somatic and axonal records from the imaged cell in A recorded in baseline conditions (2.5 mm[K⁺]_o). A 15 Hz action potential train was elicited by injecting a 90 pA depolarizing current at the soma (bottom trace). Extracellular action potential waveforms recorded at the axon were time locked to the somatic action potentials and were > 99% faithful in baseline conditions at all frequencies studied (up to ∼35 Hz). C, expanded display of the first action potential in the recording in B.

Two groups using similar techniques to those used in the present studies recently examined spike propagation in myelinated cerebellar Purkinje cell axons (Khaliq & Raman, 2005; Monsivais et al. 2005). To place our experiments in the context of this work, we examined myelination status of the axons that we studied. Fills of CA3 pyramidal cells with biocytin were followed by fixation and staining with an antibody against myelin basic protein. Although strong staining in nearby fibre tracts was observed (Fig. 2 inset), we observed no myelin staining associated with any of the 22 CA3 axons examined, although staining of occasional presumed oligodendrocytes in stratum oriens of CA3 was evident (Fig. 2). These results are broadly consistent with previous ultrastructural studies of Schaffer collateral axons in rat CA1 stratum radiatum, which have shown that the fibres are unmyelinated (Westrum & Blackstad, 1962). Previous studies appear not to have explicitly examined the myelination of CA3 axons in the region up to 800 μm from the soma targeted in the present study. Consistent with the staining, slopes of plots relating the timing between spikes arriving at the somatic and axonal recording pipettes yielded an estimated orthodromic conduction velocity in the fibres we studied of ∼0.37 m s⁻¹ at 25°C (Fig. 5C), similar to measurements in unmyelinated Schaffer collaterals (Soleng et al. 2003b).

Because axonal action potentials were recorded extracellularly, these signals should be proportional to total local membrane current. The initial positive polarity of the signals suggested the major component of the early waveform is capacitive and therefore nearly proportional to the first derivative of intracellular membrane potential (Jack et al. 1975; Johnston & Wu, 1995). In support of this assertion, differentiation of current-clamp records from the soma produced similar waveforms to extracellular axonal records (Fig. 3D–F and J–L). As a further direct empirical test of the appropriateness of using somatic slope or derivative measures, we performed simultaneous loose-seal and whole-cell recordings from individual somata. These recordings clearly showed that the extracellular spike amplitude depressed comparably to the soma derivative trace rather than to the somatic raw voltage trace (Fig. 3A–C). Overall, the 10th peak of the calculated derivative depressed by 50 ± 9% while loose-seal somatic peaks depressed by 50 ± 8% (P= 0.73, n = 5). Furthermore, the peak of the loose-seal waveform corresponded temporally to the peak of the soma derivative trace (Fig. 3D–F), corroborating the correspondence between these two measures.

It is possible that strong somatic currents could contaminate axonal recordings. We explicitly examined this possibility and found that upon moving a loose-seal pipette away from the soma while maintaining electrode resistance, the signal reversed in polarity and decreased with increasing distance from the soma. This negative signal became unmeasurable by 30 μm from the soma (Fig. 4A). When the extracellular electrode was positioned within 5 μm from distal axon segments and suction applied, there was often no contact made with the axon (verified visually). When spiking was initiated at the soma in this near-axon configuration, no detectable waveforms were present, even when noise was reduced by spike-triggered averaging (Fig. 4B). When the electrode was subsequently brought into direct contact with the axon and suction applied, a positive-going waveform was immediately observed (Fig. 4C). Therefore, somatic and other distal current sinks are unlikely to contaminate axonal records except in the most proximal recordings. It is possible that axonal recordings may have other complications, including the geometry of the isolated axonal segment and the number of ion channels in the isolated segment, but the waveforms were quite stereotyped in polarity and overall shape. We therefore conclude that the most appropriate comparisons between extracellular loose-seal and intracellular somatic records are between the extracellular amplitude and the peak dV/dt or rising phase slope of the intracellular voltage record.

Axonal recordings during 15–30 Hz trains were (> 99%) faithful in baseline conditions, showing essentially no failures of propagation. We found that depression of axonal spike amplitudes during these trains reached 15 ± 2% in axons by the 10th action potential (Fig. 5A, open bars, mean frequency 23 ± 1 Hz). This value represents slightly more amplitude reduction compared to our previous study of Schaffer collateral fibres (Meeks & Mennerick, 2004). Although the amount of spiking-dependent depression was similar for the raw amplitude of somatic action potentials (10 ± 1% by action potential 10), comparison with somatic slope showed that axonal action potentials were much less sensitive to train-dependent amplitude depression than soma slope, which depressed by 50 ± 2% by the 10th action potential (Fig. 5A, filled bars, P < 0.01, n = 42 records from 36 soma–axon pairs). These results suggest that different mechanisms are likely to contribute to waveform reductions in the soma versus the axon in CA3 pyramidal neurones.

We wondered whether axonal amplitude depression during these spikes might be attributed to distance, because the most distal of our recordings had several axonal branch points intervening between the soma and the site of recording. Branch points may be sites of weakening of the axonal spike (Swadlow et al. 1980). Rather than weakening with distance, the axonal spikes were actually least affected 150 μm from the soma and beyond (Fig. 5B). Average axonal spike depression at spike 10 in the most distal recordings was only 10 ± 1%, consistent with our previous ‘blind’ measurements of Schaffer collateral spikes (Meeks & Mennerick, 2004). By contrast, recordings made near the soma exhibited the most amplitude depression, similar to that reached by somatic slopes (Fig. 5B). Because some of the variability in relative axonal spike amplitude depression in Fig. 5B could be caused by frequency differences among the trains (15–30 Hz spike trains are pooled), we calculated the difference between the reduction in axonal amplitude and somatic slope from each recording to better isolate the effect of distance on the axonal signal (Fig. 5B inset). This plot clearly reinforces the conclusion that axonal amplitude depression is highest near the soma and becomes less severe with distance. This increase in spike strength with distance occurs despite an average of five axonal branch points in the most distal recordings (600–800 μm). These data may be consistent with a relatively distal initiation site (Colbert & Johnston, 1996; Stuart et al. 1997; Monsivais et al. 2005), and a weakening of the signal as it encounters the large capacitive load of the soma (Swadlow et al. 1980), although a small amount of the weakening of the most proximal recordings could be caused by source field currents from the soma (Fig. 3A).

We also examined latency differences between soma and axon signals (Fig. 5C). We compared the timing of the peak amplitude of axonal signals with the time of maximum rise of somatic signals. We found that for proximal axonal recordings, the axonal spike preceded the somatic spike, as expected if the axonal recording site is more proximal to the initiation site than the soma (Clark et al. 2005). We plotted latency versus recording distance for 31 recording distances up to 500 μm from the soma and found that the maximum negative latency, corresponding to initiation site, occurred 74 μm from the soma. These results are similar to previous results from myelinated axons (Clark et al. 2005) suggesting that initiation occurs distal to the axon initial segment. For all subsequent experiments, axonal recordings > 150 μm from the soma were used to evaluate axonal signals, beyond the segment of axon showing depression of extracellular signals and beyond the apparent initiation site.

As another test of the idea that axonal signals are less susceptible to amplitude depression than somatic slopes, we investigated the influence of instantaneous frequency on action potentials at the soma and in axon recordings > 150 μm from the soma (Fig. 6). The first and second action potentials of trains were separated by 10–100 ms, depending on depolarizing current amplitude and individual cell membrane properties. These initial bursts often occurred at higher instantaneous frequencies than subsequent action potentials and correspondingly exhibited more waveform variation than later spikes in the train. Somatic spike amplitude was reduced by 15 ± 1% at the smallest interspike intervals (ISIs) and gradually increased with increasing ISI (Fig. 6A and D, grey circles). Similarly, somatic action potential slope was strongly depressed at high frequencies but recovered with an exponential time course dependent on ISI (Fig. 6B and D, black circles, time constant of single exponential fit: 19.5 ms). By contrast, axonal action potential amplitude was completely independent of ISI (Fig. 6C and D, open squares, r² of linear fit = 0.004, P= 0.92). These results bolster the conclusion that propagating action potential waveforms are much less depressed than somatically recorded action potentials, even when both raw somatic spike amplitude and somatic slope are significantly depressed.

Figure 6. Axon spikes are more resistant to waveform decreases than somatic spikes at high instantaneous frequencies.

A, soma current clamp plots from a CA3 neurone with spontaneous variations in interspike interval (ISI) between action potentials 1 and 2 induced by DC injection. At the smallest ISIs, somatic action potential amplitude was reduced and recovered with increasing ISI. B, first derivative plot of the trace shown in A. Whereas somatic amplitude was reduced by < 20% at the smallest ISIs, these amplitude reductions were accompanied by slope reduction of > 50%. C, extracellular axonal recording from the cell in A. Axonal action potential amplitude was not depressed, even at the smallest spontaneous ISIs. D, averaged ISI-dependent changes from 279 spontaneous variations in ISI from 18 neurones. Responses were binned with 5 ms intervals with the exception of the last 2 ISI bins (60–70 ms and 70–105 ms, respectively). Soma amplitude (grey circles) and slope (•) recovered with increasing ISI, whereas axonal amplitude (□) showed no substantial depression at any of these intervals.

Axonal action potentials resist waveform decreases during brief action potential trains in 8 mm [K⁺]_o

Our previous results from Schaffer collateral axons indicated that 20–50 Hz trains of action potentials became depressed and prone to failures when exposed to 8 mm[K⁺]_o (Meeks & Mennerick, 2004). The behaviour of axons in 8 mm[K⁺]_o is of interest because these [K⁺]_o levels are typically reached during seizures and effects on spike propagation could have important consequences for seizure maintenance and/or propagation (Somjen, 2002). However, the extracellular stimulation techniques employed previously had a number of limitations (see Introduction and Discussion). For instance, previous studies could neither conclusively verify action potential initiation, nor identify the cells of origin of the recorded axons.

In the present experiments, we examined the effect of elevated [K⁺]_o on both the soma and the axon in the paired somatic–axonal configuration. [K⁺]_o at 8 mm caused an average depolarization of 11 ± 1 mV from −74 mV to −63 mV (n = 12). Initial somatic action potential threshold was not detectably affected in this set of neurones by this [K⁺]_o rise (−39 ± 1 mV in baseline compared to −40 ± 1 mV in 8 mm[K⁺]_o). The amplitude of isolated action potentials decreased by 13 ± 2 mV in elevated [K⁺]_o compared with 2.5 mm[K⁺]_o. Of the decrease, 11 ± 1 mV was caused by steady-state membrane depolarization; the absolute peak decreased by only 2 ± 1 mV (P < 0.01, n = 12). The slope of the initial action potential decreased by 4 ± 1% compared to baseline values (from 154 ± 6 mV ms⁻¹ in 2.5 mm[K⁺]_o to 147 ± 6 mV ms⁻¹ in 8 mm[K⁺]_o, P= 0.01, n = 12). These data show that 8 mm[K⁺]_o did not strongly alter isolated (initial) somatic action potentials with respect to baseline.

We adjusted somatic current injection amplitude to elicit 15–30 Hz action potential trains to match trains recorded in baseline [K⁺]_o. The peak amplitude of somatic spike 10 of trains in elevated [K⁺]_o decreased 13 ± 1% relative to the initial spike in the train, similar to the spike 10 decrease observed in 2.5 mm[K⁺]_o (12 ± 2%, n = 10). The slopes of these somatic spikes were similarly affected by 8 mm[K⁺]_o (59 ± 4% reduction at AP 10 in 8 mmversus 54 ± 3% in 2.5 mm[K⁺]_o, Fig. 7B and D, filled and open circles, respectively, P= 0.15, n = 10). Again, we saw little change in the characteristics of somatic action potentials in spike trains during 8 mm[K⁺]_o exposure.

Figure 7. Effects of elevated [K⁺]_o on action potential trains.

A, somatic recording of an action potential train induced by a 50 pA step depolarization in 8 mm[K⁺]_o. Somatic resting membrane potential was −64 mV. B, first derivative plot of the action potential train in A. C, axonal action potential recording resulting from the action potential trains indicated in A and B. D, averaged responses of 10 cells at action potentials 1–10 from somata (•) and axons (▪) during action potential trains induced in 8 mm[K⁺]_o. As seen in baseline [K⁺]_o, somatic slope depression remained significantly greater than axonal amplitude depression (P < 0.01, n = 10). ○ and □, somatic slopes and axonal amplitudes (respectively) of action potentials 1–10 from these 10 neurones in 2.5 mm[K⁺]_o. E, instantaneous frequency response of somata and axons from action potentials 1 and 2 in 8 mm[K⁺]_o. Somatic slope (•) and axonal amplitude depression (▪) were slightly greater across all ISIs than controls (○ and □, respectively, P < 0.05, n = 10). Axonal spikes began to show frequency-dependent amplitude depression at ISIs < 25 ms in 8 mm[K⁺]_o, but did not depress more than 20%.

Axons, perhaps unexpectedly, did not show strong amplitude reductions during brief trains compared with 2.5 mm[K⁺]_o (16 ± 5% depression at spike 10 in elevated [K⁺]_o compared to 12 ± 2%, Fig. 7C and D, filled and open squares, respectively, P= 0.40, n = 10). At this [K⁺]_o, axonal action potentials were relatively independent of instantaneous frequency, while somatic slope was again significantly depressed at small ISIs (Fig. 7E). Some reduction (< 20%) of axonal spike amplitude was seen at ISIs < 25 ms (Fig. 7E, filled squares, P < 0.02), but we observed no evidence for increased axonal failures. The lack of increased failures differs from previous results in Schaffer collateral axons (Meeks & Mennerick, 2004). It therefore seems likely that the failure increases observed in past studies (Poolos et al. 1987; Meeks & Mennerick, 2004) were a result of initiation failure, since initiation could not be verified with the extracellular fibre stimulation previously employed.

These data indicate that action potential propagation 150–800 μm from the soma remains intact in the presence of [K⁺]_o elevations alone, with slight decreases in axonal amplitude at the smallest ISIs. Although 8 mm[K⁺]_o alone did not strongly alter action potential conduction during brief trains, elevated [K⁺]_o exposure in many cases induced epileptiform action potential discharges in CA3 neurones superimposed on depolarizations resembling plateau potentials (Segal, 1994; Fraser & MacVicar, 1996). The occurrence of these events, which had features of ictal-like discharges (Segal, 1994; McCormick & Contreras, 2001), afforded us the opportunity to examine how action potential fidelity is affected by epileptiform pathophysiological firing in the hippocampus.

CA3 axonal spike amplitudes depress severely during K⁺-induced PPs

Of CA3 neurones studied in 8 mm[K⁺]_o, 36% responded to somatic action potential initiation with long-lasting PPs. PPs lasting 23 ± 3 s occurred after the 1 s depolarizing pulse (Fig. 8), followed by a sustained membrane hyperpolarization of 5–10 mV lasting 40–80 s during which PPs could not be induced (data not shown). Like similar events recorded in CA1 pyramidal cells (Fraser & MacVicar, 1996) and subicular pyramidal neurones (Stewart & Wong, 1993), the PP usually ended suddenly, followed by a period of slow hyperpolarization (e.g. Fig. 8A). During PPs, spike generation was persistent, but at a lower average frequency than during the initial, stimulus-induced train (13 ± 2 Hz versus 21 ± 5 Hz, P < 0.01, n = 6). The PPs and associated discharges were blocked by 200 μm Cd²⁺ but were insensitive to atropine (data not shown; cf. Fraser & MacVicar, 1996). An example somatic/axonal recording during a PP event is shown in Fig. 8, with summary data shown in Fig. 9. All axonal recordings were made at sites > 150 μms from the soma. We observed a slight increase in somatic slope reduction at spike 10 just before PP development compared to baseline spike 10 (56 ± 5% reduction versus 43 ± 3% in 2.5 mm[K⁺]_o, P < 0.01, n = 6). During PPs, we observed a wide variation in somatic slope (discussed below), but average slope reduction was not significantly different from reductions observed at spike 10 before PP development (59 ± 6% reduction during PPs versus 56 ± 5% at spike 10 in 8 mm[K⁺]_o, P= 0.39, n = 6). As observed in the absence of PPs, the 10th axonal spike amplitudes during the initial 1 s stimulus were not significantly depressed in this group of cells (20 ± 5% depression versus 12 ± 4% depression at spike 10 in 2.5 mm[K⁺]_o, P= 0.08, n = 6, Fig. 8B, no. 1; Fig. 9A, open bars). However, during the PP event, axonal spike amplitudes were strongly reduced compared to the 10th action potential just before PPs (mean depression 41 ± 8%versus 20 ± 5%, P < 0.01, n = 6, Fig. 9A). In summary, despite the similar somatic slope depression before and during PPs in 8 mm[K⁺]_o, we only observed strong axonal amplitude depression during PPs.

Figure 8. Plateau potential (PP) generation in CA3 neurones in response to 1 s depolarizing current injection in 8 mm[K⁺]_o.

A, CA3 somata often responded to stimulation in elevated [K⁺]_o with 10–50 s PPs. During these PPs, the soma fired action potentials at 5–50 Hz, with average spike frequency only 13 ± 2 Hz (n = 6). B–D, somatic and axonal action potentials on a smaller time scale. Recordings 1–5 represent the segments indicated by the numbered bars below the trace in A. B, axonal action potentials decreased in amplitude during PPs (nos 2–5), but not before (no. 1). Asterisks mark peak times of somatically detected spikes. C, soma derivative plot shows severe dV/dt depression during PPs, especially during burst events (nos 2 and 4). D, somatic action potential amplitude from current clamp traces shows much less severe depression in absolute amplitude during PPs, even during burst events.

We frequently observed burst-like events during PPs where two to eight spikes occurred at a high frequency (> 20 Hz) and with severe somatic slope reduction (> 80%, Fig. 8B, no. 4). These events appeared similar to burst events observed previously in CA3 (Wong & Prince, 1981), subicular pyramidal neurones (Stewart & Wong, 1993; Jung et al. 1997), and cortical pyramidal neurones (Schwindt & Crill, 1999). These bursts can be induced by dendritic depolarization and require Ca²⁺ entry (Schwindt & Crill, 1999), although the precise mechanisms for their generation remain in question (Lazarewicz et al. 2002).

We sought to determine whether bursts accounted for the axonal amplitude depression observed in Fig. 9A. We separated burst action potential events from non-burst events during PPs based on instantaneous action potential frequency (Fig. 9B). We used a criterion of two or more spikes occurring at > 20 Hz to define bursts. The mean frequency for burst events was 31 ± 3 Hz and for non-burst events, 4 ± 4 Hz. Somatic slope was significantly depressed during these bursts compared with non-burst action potentials within the PP (86 ± 5% slope reduction in bursts versus 54 ± 6% otherwise, P < 0.01, Fig. 9B, filled bars). Axonal spike amplitudes from burst events were significantly more depressed than non-burst spikes during the same PP (50 ± 8% depression for bursts versus 36 ± 7% during non-bursts, Fig. 9B, open bars, P < 0.01, n = 6). Especially during bursts, axonal action potential amplitudes dropped below the threshold for detection (e.g. Fig. 8B, no. 4), but peak-to-trough amplitude values were still assigned to these events in our analysis. Because noise amplitude, rather than a zero, was assigned to detection failures, the 50% average reduction of axonal action potential amplitude likely represents an underestimate of the actual axonal spike depression, and some events may represent true propagation failures, a possibility examined further in Fig. 10. Regardless of the proportion of complete propagation failures, graded decreases in the amplitude of the axonal waveform were clearly important in the axonal changes (Fig. 8), in contrast to binary failures of propagation. Non-burst events during PPs, which had similar relative somatic slope changes to non-PP spikes, showed significant axonal extracellular spike amplitude reductions (36 ± 7% reduction during PPs versus 20 ± 5% at action potential 10 before PP induction, P < 0.01, Fig. 9B). We also found no correlation between distance of recording from the soma (range: 150–760 μm) with the amount of signal reduction (P > 0.36, n = 6). Overall, these results suggest that burst events alone do not account for the axonal spike reductions observed during PPs.

To rule out that our relative measurements of somatic slope reduction may have obscured more pronounced deficits in absolute slope, we compared axonal amplitude depression in various conditions based on absolute somatic slope (Fig. 9C). We found that axonal action potentials depress during PPs at absolute slope values (< 90 mV ms⁻¹) which cause little amplitude depression in the axon in the absence of PPs (Fig. 9C, P < 0.05, n = 6). The small number of action potentials during PPs that had large somatic slopes (> 90 mV ms⁻¹) were not significantly depressed in amplitude compared to baseline spikes or spikes preceding PPs (Fig. 9C, P > 0.25, n = 6). Therefore, neither absolute depression nor relative depression of somatic slope can fully account for the depression of axonal spike amplitudes during PPs.

To explicitly examine the role of propagation failures in the changes in axonal conduction observed, we performed spike-triggered averaging on axonal signals that met our criteria for detection failures (< 2 standard deviations above the mean noise). Five neurones in our data set had sufficient numbers of detection failures (> 8) to allow averaging. Figure 10A shows raw traces from one of these cells. The traces represent the initial spike obtained at the onset of the 1 s current injection used to initiate the PP and subsequent examples of axonal successes and detection failures during the PP. Detection failures from burst and non-burst somatic spikes were difficult to confidently distinguish from noise on an individual basis. Figure 10B shows averages of 18 events in each category obtained by aligning the times of maximum somatic dV/dt. Figure 10C shows a running plot of detection failures of both burst and non-burst action potentials. It is clear from the spike-triggered average waveforms in Fig. 10B and the plot in Fig. 10C that the mean of non-burst spike detection failures was larger than the mean of the burst detection failures. We found that in the five cells examined, 69 ± 13% of burst spikes failed to reach detection threshold, while 11 ± 4% of non-burst spikes failed detection. The proportion of non-burst spikes failing to exceed detection threshold increased with increasing PP duration (Figs 8 and 12). Of all detection failures, 72 ± 8% were accounted for by bursts. Therefore, we conclude that burst spikes from CA3 pyramidal neurones are highly prone to propagation failure, similar to recent reports from myelinated Purkinje cell axons (Khaliq & Raman, 2005; Monsivais et al. 2005).

Figure 12. PP-like depolarization events induced in 2.5 mm[K⁺]_o are sufficient to decrease axonal spike amplitudes.

A, binned record from an example neurone during 3 PPs induced in 2.5 mm[K⁺]_o utilizing the protocol shown in Fig. 11A. The somatic membrane potential was depolarized using a steady-state current injection to −58 mV followed by a 1 s DC pulse to induce a brief 20 Hz action potential train. This spiking was followed by PPs lasting 22 ± 2 s. Soma slope (•) decreased strongly during the 1 s pulse (preceding the arrow), followed by a partial recovery associated with the decrease in spike frequency from 20 to 8 Hz. Axonal spike amplitude (□) became progressively depressed throughout this period, and did not recover along with somatic slope during the transition to PP spiking. The horizontal dotted line indicates axonal detection threshold. B, binned record from the same neurone in A in response to a 25 s sustained current injection from a resting potential of −72 mV (see Fig. 11B). The nature of the depolarizing pulse did not induce high-frequency action potential spiking initially, but the average frequency was similar to the other method (9 Hz compared to 8 Hz during the PP in A). The decline in somatic slope (•) and axonal amplitude (□) was similar to the other method. Grey diamonds indicate the axonal spike amplitude depression caused by 10 Hz pulse trains as shown in Fig. 11C. Note this depression was substantially smaller than either of the other conditions. C, summary of responses of 5 neurones exposed to the spiking paradigms represented in Figs 8 and 11. Axonal spike amplitude depression seen using the methods in Fig. 11A and B were statistically indistinguishable from 8 mm[K⁺]_o-induced PPs (filled bars: somatic slope reduction; open bars: axonal action potential depression). Ten-hertz, 200 pulse trains in these 5 neurones (rightward-most open bar) reported similar spike amplitude reduction at spikes 175–200 to reductions seen at the 10th spike before PP induction. This value was statistically less affected than all sustained depolarization conditions (P < 0.05 for all, n = 5).

These data illustrate the decreases evident in axonal signals during PPs. The effect of these events on action potential conduction in the first 800 μm of the axon appears to be twofold. First, PPs reduce axonal spike amplitude, an effect not seen during brief action potential trains, even when absolute or relative somatic slope reduction is matched. Secondly, PPs seem to facilitate an increase in ‘bursting’ behaviour at the soma. During these burst events, axonal spikes are most severely depressed, and in many cases fail to propagate entirely. Ignoring the contribution of bursting events, the influence of PPs on axonal action potential conduction is especially surprising in light of the lower overall frequency of spiking during PPs. We considered several potential mechanisms that might underlie the changes seen in axonal spikes during PPs. First, the presence of depolarizing [K⁺]_o across the entire plasma membrane of the cell, in combination with long-lasting somatic depolarization or repetitive firing, may affect the ability of axons to maintain consistent waveform during sustained activity. Second, long-lasting depolarization of the soma alone may affect conduction along the axon. Third, axonal action potentials may be more sensitive to sustained firing than somata. We investigate these possibilities below.

Axonal action potentials depress during sustained somatic depolarization in 2.5 mm[K⁺]_o

We were able to induce PPs and epileptiform action potential discharges in CA3 pyramidal cells without exposing them to elevated [K⁺]_o by injecting sustained DC to bring the somatic membrane potential between −65 and −45 mV (below spike threshold, mean V_m−57 ± 3 mV, n = 5). One-second current steps superimposed on these steady injections induced PPs similar to those seen in elevated [K⁺]_o (mean PP duration 13 ± 4 s, mean spike frequency 11 ± 3 Hz; Figs 11A and 12A). Axonal spike amplitudes (recording distances > 150 μm) during these PPs depressed similarly to PPs in 8 mm[K⁺]_o (33 ± 6%versus 41 ± 8%, P= 0.37, n = 5, Fig. 12A and C), indicating that elevated [K⁺]_o is not necessary for the axonal action potential changes seen above. These results exclude an important role for the spatially uniform depolarization generated by elevated [K⁺]_o in causing either PPs or the axonal spike amplitude depression.

Figure 11. Stimulation of PPs and PP-like events in 2.5 mm[K⁺]_o.

A, steady-state depolarization of the soma to a membrane potential between −45 and −65 mV (below spike threshold), followed by 1 s depolarizing steps (horizontal bar) resulted in PPs between 5 and 30 s in duration. B, from a membrane potential of ∼−70 mV, 25 s constant current injections caused plateau-like spiking. C, from a membrane potential of ∼−70 mV, brief (1–3 ms), high-intensity (1–3 nA) current pulses were delivered to the soma at 10 Hz. Although V_m did not recover to initial V_m between spikes, the steady-state V_m between spikes using pulse trains was well below that seen during the stimuli shown in A and B. All example traces were recorded from the same cell in 2.5 mm[K⁺]_o. Summary of somatic and axonal responses is shown in Fig. 12.

As another method to generate sustained spiking, we applied 25 s suprathreshold current injections to the soma from resting potentials of ∼−70 mV (mean frequency 13 ± 1 Hz; Figs 11B and 12B). This stimulation also induced similar axonal spike amplitude reductions to those seen during PPs (33 ± 6% amplitude reduction during PPs elicited as in Fig. 11A and 34 ± 3% reduction when elicited as in Fig. 11B, P= 0.37 and P= 0.42, respectively, compared to 8 mm[K⁺]_o PPs, n = 5, Fig. 12C). Notably, axonal spike amplitude decayed smoothly over the period of sustained depolarization with both stimulus protocols (Fig. 12A and B). Changes in axonal signals were largely divorced from changes in somatic waveform (Fig. 12A and B), perhaps indicating that the steady depolarization, rather than spiking per se, was playing an important role in the axonal changes. As with PPs, we found no significant correlation between distance of axonal recording and the degree of amplitude depression (P= 0.71, n = 5). We also measured changes in somatic and axonal spikes at higher temperature (32°). In three cells challenged with 25 s sustained depolarizations, somatic rising slope decreased by 34 ± 10% at 25°C and 34 ± 13% at 32°C (P= 0.86). Average axonal spike amplitude during the sustained depolarization was depressed by 56 ± 16% at 25°C and 57 ± 15% at 32°C (P= 0.94).

To directly test whether spiking alone, without a steady depolarization, could elicit axonal changes, we delivered 200 suprathreshold current pulses (1–3 ms, 1–3 nA) at 10 Hz to elicit fixed-frequency action potentials at the soma (Figs 11C and 12B, grey diamonds). During these trains, the envelope somatic depolarization was much weaker than during the other methods used to cause or simulate PPs in 2.5 mm[K⁺]_o (−59 ± 2 mV for the paradigm in Fig. 11C versus−35 ± 3 mV and −36 ± 2 mV for Fig. 11A and B, respectively, P < 0.01, n = 5). Trains of 200 action potentials at 10 Hz from a resting potential of ∼−70 mV failed to depress axonal spike amplitudes to the level achieved by PPs or by sustained depolarization to the soma with direct current injection (20 ± 3% depression at spikes 175–200 of 10 Hz pulse, P < 0.05 compared to others, n = 5, Fig. 12C). The axonal amplitude depression seen at the end of these long pulse trains was similar to the depression seen during brief 20 Hz trains. This result indicates that sustained spike generation alone is not sufficient to recreate the axonal amplitude reductions seen during PPs. Only conditions which elicited sustained membrane potential depolarization at the soma were able to depress axonal spike amplitudes by more than 20%.

Discussion

Our results highlight some conditions under which action potential conduction is highly reliable and conditions under which axonal signals are weakened. With brief action potential trains from any resting potential, the axon shows stronger signal preservation than the soma and initial segment. By contrast, with longer trains involving sustained somatic membrane depolarization, extracellular axonal spikes decrease in amplitude relative to the soma. Our results strongly suggest that epileptiform activity decreases the amplitude of extracellular axonal spikes and leads to propagation failures in bursting spikes. We also conclude from Figs 11 and 12 that during spike trains in normal conditions, action potential waveform varies substantially, with the steady-state somatic membrane potential serving as an arbiter of propagation strength. In the case of seizures, it remains possible that the observed decreases in axonal signals are simply an epiphenomenon of a pathological activity state. On the other hand, a more interesting possibility is that decreased axonal signal strength could represent an endogenous protective mechanism to limit seizure propagation. In the case of normal function, changes in spike waveform could impart more information content to axonal signals than frequency coding alone. The general results are in agreement with two recent reports employing similar techniques to myelinated fibres of cerebellar Purkinje neurones (Khaliq & Raman, 2005; Monsivais et al. 2005).

Clearly, under basal conditions, axonal spikes in CA3 pyramidal cells were resistant to factors, including high-frequency firing, which depressed somatic spike amplitude and somatic rising-phase slope. Sustained spiking, as long as it occurred as a result of transient depolarization rather than sustained depolarization, also failed to significantly alter axonal spikes. Therefore, action potentials triggered by transient physiological stimuli, like fast synaptic potentials, are unlikely to be associated with significant signal alteration in the axon. Also, when brief spike trains were superimposed on small, steady depolarizations of < 15 mV, like those induced by 8 mm[K⁺]_o, axonal signal fidelity was maintained. On the other hand, when spiking occurred on an envelope of sustained depolarization (e.g. pathophysiological PPs or tonic current injection), significant degradation of the propagating spike occurred, largely divorced from changes in somatic spike waveforms.

Do the axonal changes that occur during sustained depolarization represent a meaningful change in signalling? We suggest that they do. During PPs, burst spiking resulted in an average axonal spike amplitude below our detection limits (typically ∼30–50% of control amplitude). Even when spike-triggered averaging, which increased signal-to-noise ratio, was employed, we observed no underlying axonal waveform during these bursts. Previous studies have suggested that extracellular amplitude decreases exceeding 30% may result in propagation failure (Raastad & Shepherd, 2003). Although it is possible that minimal amplitude spikes propagate, we conclude that burst spikes during epileptiform activity do not propagate reliably to downstream terminals. Analyses of changes in non-burst spikes are more complicated because we suggest that changes in our primary measure, axonal signal amplitude, reflect changes in the intracellular spike rise time. This could mean that axonal signals broaden with little attenuation in the peak amplitude, an effect that in isolation may increase downstream neurotransmission (Sabatini & Regehr, 1997; Borst & Sakmann, 1999; Qian & Saggau, 1999). However, similar waveform changes in extracellular axonal signals have been previously associated with downstream synaptic depression (Meeks & Mennerick, 2004). It therefore remains likely that the impact of these axonal changes is depression of downstream transmitter release. Other studies also suggest that similar waveform changes during repetitive spiking or mild Na⁺ channel inhibition are likely to depress Ca²⁺ influx (Bischofberger & Jonas, 1997) and transmitter release (Sabatini & Regehr, 1997; Borst & Sakmann, 1999; Qian & Saggau, 1999; Prakriya & Mennerick, 2000).

Our previous study of axonal conduction fidelity found more evidence for overt failure of axonal action potentials during rises in extracellular [K⁺]_o (Meeks & Mennerick, 2004), and these previous results are consistent with even older work suggesting axonal initiation or conduction deficits in the presence of elevated [K⁺]_o (Malenka et al. 1981; Poolos et al. 1987). This discrepancy highlights the power of the present techniques, particularly our ability to verify action potential initiation, which has not been possible in past studies. It is likely that failures in past studies resulted from failure of the extracellular stimulating shock to initiate action potentials rather than from propagation failure. On the other hand, we cannot completely exclude other potential explanations including undetected PPs in past work, anatomical differences in the axons (e.g. variations in the synaptic varicosities or caliber of axon), and the contributions of severed, and therefore damaged, fibres to past results.

Our results suggest that spikes associated with similar somatic waveform can either be propagated with high extracellular amplitude (e.g. Figs 4, 5, 6, 7) or depressed amplitude (e.g. Figs 8, 9, 10 and 12) in the axon, depending on depolarization status of the soma. The change in axonal signals is likely to be mediated by accumulated inactivation of axonal channels (Monsivais et al. 2005), while the strong reductions in somatic slope and initial segment waveforms with even brief trains of activity (e.g. Figs 5 and 6) may result from conductances (e.g. potassium conductances), not present in the axon. Distinct functional properties of initial segment channels relative to somatic channels have been noted in some cases (Colbert & Pan, 2002). Our work suggests that the initial segment of CA3 axons behaves more like the soma (Figs 4 and 5B and C), despite these potential functional differences in the channels of the initial segment. Our conclusions are also consistent with work suggesting that initiation may be downstream of the initial segment (Colbert & Johnston, 1996; Stuart et al. 1997; Monsivais et al. 2005) with the spike backpropagated with some alteration toward the soma through the initial segment.

Given that somatic depolarization state influences axonal fidelity, what are the implications of this relationship for cell function? We have shown that sustained depolarization during epileptiform activity promotes failures of burst spikes and diminishes regular spike waveforms. These changes seem likely to represent a brake on seizure activity in the hippocampus, one of the most seizure prone areas of the nervous system. The mechanisms elucidated here may also help to shed light on the effects of current clinical treatments involving neuronal stimulation. Deep brain stimulation, electroconvulsive therapy, and transcranial magnetic stimulation are increasingly important clinically but the mechanisms remain poorly understood at the cellular level (Breit et al. 2004; Di Lazzaro et al. 2004; McDonald et al. 2004; McIntyre et al. 2004). Our results suggest that sustained depolarization of somata may lead to decreased signalling from stimulated regions, similar to some previous modelling results (McIntyre et al. 2004). Finally, in terms of normal cell function, we predict that the switch to a low fidelity mode of signalling is probably elicited by slow neuromodulators such as acetylcholine (Fraser & MacVicar, 1996), which, by simple depolarization of the soma, may shift neurones from a high-fidelity mode of propagation to an altered mode of propagation.