Abstract
Action potentials (spikes) are regenerated at each node of Ranvier during saltatory transmission along a myelinated axon. The high density of voltage-gated sodium channels required by nodes to reliably transmit spikes increases the risk of ectopic spike generation in the axon. Here we show that ectopic spiking is avoided because KV1 channels prevent nodes from responding to slow depolarization; instead, axons respond selectively to rapid depolarization because KV1 channels implement a high-pass filter. To characterize this filter, we compared spike initiation properties in the soma and axon of CA1 pyramidal neurons from mice of both sexes, using spatially restricted photoactivation of channelrhodopsin-2 (ChR2) to evoke spikes in either region while simultaneously recording at the soma. Somatic photostimulation evoked repetitive spiking whereas axonal photostimulation evoked transient spiking. Blocking KV1 channels converted the axon photostimulation response to repetitive spiking and encouraged spontaneous ectopic spike initiation in the axon. According to computational modeling, the high-pass filter implemented by KV1 channels matches the axial current waveform associated with saltatory conduction, enabling axons to faithfully transmit digital signals by maximizing their signal-to-noise ratio for this task. Specifically, a node generates a single spike only when rapidly depolarized, which is precisely what occurs during saltatory conduction when a pulse of axial current (triggered by a spike occurring at the upstream node) reaches the next node. The soma and axon use distinct spike initiation mechanisms (filters) appropriate for the task required of each region, namely, analog-to-digital transduction in the soma versus digital signal transmission in the axon.
Significance Statement
Neurons use action potentials, or spikes, to transmit information reliably over long distances. Spikes can be initiated through different dynamical mechanisms depending on the types of ion channels involved. The input required to evoke spikes differs depending on spike initiation dynamics. Using targeted optogenetic stimulation to evoke spikes in different subcellular compartments, we show that the soma and axon of pyramidal neurons use distinct spike initiation mechanisms suited to the distinct role of each compartment. Specifically, the soma uses a low-pass filter supporting analog-to-digital transduction whereas the axon uses KV1 channels to implement a high-pass filter seemingly optimized for transmitting spikes. Importantly, the high-pass filter prevents the axon from generating ectopic spikes if slowly depolarized.
Introduction
Whereas analog signals such as graded depolarization are well suited for efficient computing, digital signals such as spikes are better for transmitting information because of their resilience to noise (Sarpeshkar, 1998). When conducted along a myelinated axon, spikes are actively regenerated at each node whereas subthreshold depolarization decays. In theory, subthreshold depolarization could be boosted but actively offsetting attenuation to deliver an analog signal with its original amplitude is less feasible than reliably transmitting spikes whose firing rate is proportional to the original depolarization.
Axons do more than simply transmit spikes (Clark and Hausser, 2006; Bucher and Goaillard, 2011; Debanne et al., 2011; Zbili and Debanne, 2019) but even spike transmission remains incompletely understood. For instance, graded voltage changes travel short distances down an axon and modulate spike-evoked synaptic output (Alle and Geiger, 2006; Shu et al., 2006); this represents analog modulation of a digital signal rather than transmission of an analog signal per se, akin to spike width affecting synaptic output (Shapiro et al., 1980). But how do graded voltage changes travel down an axon without disrupting spikes being transmitted down the same axon? Why does axonal depolarization not evoke ectopic spikes? Ectopic spikes would severely corrupt transmission of spikes originating at the soma or axon initial segment (AIS) by adding “noise” spikes that collide with and thus subtract “signal” spikes. The high density of sodium channels required for reliable transmission (Hallermann et al., 2012; Hu and Jonas, 2014) predicts a high risk of ectopic spiking, yet ectopic spikes occur rarely.
Such questions have gone largely untested because the small size of axons makes them difficult to stimulate and record intracellularly. Extracellular recordings and voltage imaging are useful alternatives to intracellular recordings but do not offer equivalent stimulation capabilities (see below). The bleb formed at the cut end of an axon can be patched, but whether excitability is affected by the initial cut or subsequent bleb formation remains unclear. Sustained depolarization of axon blebs evokes transient spiking (Shu et al., 2007b; Hu and Jonas, 2014; Kole and Popovic, 2016; Thome et al., 2018), reminiscent of class 3 spiking (Hodgkin, 1948). But unlike the transient spiking evoked at blebs far (>110 mm) from the soma, sustained depolarization of blebs formed within the axon initial segment (AIS, <100 mm from soma) evokes repetitive spiking (Shu et al., 2007a), proving that not all blebs spike transiently and suggesting, instead, that intrinsic excitability differs between the AIS and distal axon. Transient spiking has also been reported when injecting current into boutons (Geiger and Jonas, 2000; Bischofberger et al., 2006; Ohura and Kamiya, 2018; Byczkowicz et al., 2019; Martinello et al., 2019) or into the calyx of Held (Dodson et al., 2003; Ishikawa et al., 2003; Sierksma and Borst, 2017), although excitability of these structures may be specialized to support synaptic transmission. That said, some studies (Dodson et al., 2003; Ishikawa et al., 2003; Shu et al., 2007b) and older work on peripheral axons (Kocsis et al., 1983; Poulter et al., 1989) converted axons from transient to repetitive spiking by blocking potassium channels with 4-aminopyridine (4-AP).
While recording from the soma of CA1 pyramidal neurons, we evoked spikes optogenetically from different subcellular locations using spatially restricted photostimulation (Zhu et al., 2015). Unlike electric field stimulation, which uses brief pulses to modulate membrane potential indirectly via effects on extracellular voltage (Rattay, 1999), optogenetic stimulation can produce sustained depolarization via sustained transmembrane current, thus avoiding the need for intracellular access to inject current. By optogenetically evoking spikes in a particular region, one can infer the spike initiation dynamics of that region without recording locally, since the resulting spikes propagate and can thus be recorded from a remote location amenable to patching, like the soma. Thus, by combining somatic recordings with targeted optogenetic stimulation and pharmacological manipulations, plus computational modeling, we show that axons use KV1 channels to optimize their intrinsic excitability to reliably transmit spikes.
Materials and Methods
Experiments
Animals
All procedures were approved by the Animal Care Committee at The Hospital for Sick Children (protocol #65769). CamKII-Cre mice (JAX# 005359) were crossed with floxed ChR2-eYFP (Ai32) mice (JAX# 012569) to express ChR2 tagged with YFP in CA1 pyramidal neurons. The Ai32 mouse uses the CAG promoter to drive ChR2 expression. No axonal abnormalities were observed (assessed for up to 80 d) in pyramidal neurons virally transfected with ChR2 driven by this promoter, whereas mild abnormalities were observed with the stronger CamKII promoter, and significant abnormalities were observed when introducing ChR2 via in utero electroporation (Miyashita et al., 2013). We used transgenic mice instead of virus, using the CamKII promoter to drive expression of Cre recombinase which, in turn, allows expression of ChR2. In our hands, Ai32 mice expressing ChR2 in primary sensory neurons exhibited no differences in excitability when compared back-to-back with wild-type mice (Xie et al., 2024), nor have we observed any behavioral deficits. Overall, we cannot exclude that expressing ChR2 affects axon physiology but we have no evidence suggestive of such effects and past data should assuage concerns.
Slice preparation
Adult mice (6–8 weeks old) of either sex were anesthetized with isoflurane and decapitated. No sex differences were observed and data were therefore pooled. The brain was rapidly removed and placed in ice-cold oxygenated (95% O2 and 5% CO2) sucrose-substituted artificial cerebrospinal fluid (aCSF) containing the following (in mM): 252 sucrose, 2.5 KCl, 2 CaCl2, 2 MgCl2, 10 d-glucose, 26 NaHCO3, 1.25 NaH2PO4, and 5 kynurenic acid. Parasagittal slices (350 µm) were prepared in sucrose-substituted aCSF with a microtome (VT-1000S, Leica) and were then kept in normal oxygenated aCSF (126 mM NaCl instead of sucrose and without kynurenic acid) until recording. Slices were transferred to a recording chamber constantly perfused at ∼2 ml/min rate with oxygenated (95% O2 and 5% CO2) aCSF heated to 31 ± 1°C and viewed with an Axio Examiner microscope (Zeiss) using a 40×, 0.75 NA water immersion objective (N-Achroplan, Zeiss) and Infrared CCD camera (IR-1000, Dage-MTI). Cells were visualized with a long-pass filter (OG590) in the transmitted light path to avoid activating ChR2 while patching.
Recording
Pyramidal neurons in the CA1 region of the hippocampus were recorded in whole-cell mode using an Axopatch 200B amplifier (Molecular Devices) with >75% series resistance compensation. Micropipettes (5–7 MΩ) were pulled from borosilicate glass capillaries (O.D., 1.5 mm; I.D., 0.86 mm; Sutter Instrument) using a PC-10 puller (Narishige). The patch pipette solution contained the following (in mM): 125 KMeSO4, 5 KCl, 10 HEPES, 2 MgCl2, 4 adenosine triphosphate (ATP; Sigma-Aldrich), and 0.4 guanosine triphosphate (GTP; Sigma-Aldrich); pH was adjusted to 7.2–7.3 with KOH. Reported values of membrane potential are corrected for a liquid junction potential of 9 mV. Resting membrane potential and input resistance were monitored during experiments and neurons experiencing >20% change in either were excluded. Recordings were low-pass filtered at 2 kHz and digitized at 20 kHz using a Power1401 DAQ and Signal software (v7, Cambridge Electronic Design, CED).
Stimulation
For current-based stimulation, square steps were injected into the soma through the patch pipette. For optogenetic stimulation, a 455 nm LED (Thorlabs, M455L4) with a Zeiss collimation adapter was used to deliver blue light through the epifluorescent port of the microscope; the field diaphragm was partially closed, limiting the photostimulated area to a spot ∼20 µm in diameter. Light intensity at the specimen plane was measured with a S170C photodiode power sensor and PM100D power meter (Thorlabs). Photostimulus intensity was controlled via a T-cube LED driver (Thorlabs, LEDD1B) modulated by computer using the Power1401 DAQ (CED). Photostimulus steps were preceded by a strong 200-ms-long prepulse to inactivate ChR2 prior to the step (Fig. 2A). Extracellular electrical stimulation was applied using a DS3 Isolated Current Stimulator (Digitimer) to deliver constant current via bipolar electrodes positioned across the axon bundle (alveus) just distal to the photostimulation site. Pulses 100 μs in duration were increased in amplitude (0.5–3.0 mA) until spikes were recorded. Timing of all stimuli was controlled by computer using Signal 6.02 (CED). For each stimulus modality, threshold (Th) is defined as the minimum intensity required to evoke a spike.
Because the axon was not visualized in experiments, the length of axon between the soma and photostimulation site could not be measured; instead, distance to the photostimulation site was measured as the shortest length (i.e., straight line) between the soma and photostimulation site. For consistency, the same convention was applied to stimulation data. Based on the model geometry, the length along the axon
Drug application
Fast synaptic transmission was blocked in all experiments by bath applying 20 μM CNQX (Hello Bio), 40 μM d-APV (Hello Bio), and 10 μM bicuculline (Hello Bio). In some experiments, voltage-gated sodium channels were blocked by puffing tetrodotoxin (TTX; Alomone) onto the axon; for this, 20 μM TTX was dissolved in HEPES-buffered aCSF and applied for 20–1,000 ms via a puff pipette positioned just above the region of interest, with pressure adjusted to 2–4 psi using a picospritzer (Toohey Company). To block sodium channels intracellularly, the pipette tip was filled with drug-free solution but 1 mM QX-314 (Tocris) was added to the intracellular solution in the shank of the pipette. To block voltage-gated potassium channels predicted to modulate spike initiation (viz., KV1 channels), 200 μM 4-AP (Sigma-Aldrich) was added to the bath. At concentrations ≤300 μM, 4-AP selectively blocks KV1.1, 1.2, 1.5, and 2.1 (Coetzee et al., 1999); KV2.1 activates only at suprathreshold voltages, i.e., during the spike (VanDongen et al., 1990; Malin and Nerbonne, 2002), and is not expressed in the axon beyond the AIS (Kirizs et al., 2014), and so its blockade could not influence spike initiation (as we confirmed with simulations). We bath applied 4-AP because, consistent with past observations (Hu and Bean, 2018), preliminary testing suggested that puffing the drug was ineffective because blockade of KV channels occurred slowly.
Experimental design and statistical analyses
Normality was tested with the Shapiro–Wilk test. Data are summarized as median and interquartile range (IQR) because of their non-normal distribution. Appropriate nonparametric tests were applied and are reported in the text. All analysis was conducted using SigmaPlot 11 (SYSTAT).
Modeling
A multicompartment model of a mouse CA1 pyramidal neuron with a detailed myelinated axon was constructed using NEURON v8.0 (Hines and Carnevale, 1997). This model is based on the CA1 pyramidal neuron model by Shah et al. (2008), but changes were made to its morphology and ion channel densities. All code will be made freely available at https://modeldb.science/2018262.
Morphology
The cell dimensions were uniformly shrunk to account for the difference in the average size of rat (original model) and mouse (experiments) CA1 pyramidal neurons (Routh et al., 2009). The reconstructed model reproduces morphological features (e.g., total surface area, total dendritic length, somatic surface area) of mouse CA1 pyramidal neurons (Benavides-Piccione et al., 2020; Table 1). The axon used by Shah et al. (2008) was replaced by a more detailed axon containing an axon hillock, AIS, myelin sheaths, and nodes of Ranvier. Inside the myelin sheath, each internode comprised paranode, juxtaparanode, and internode regions (Guo et al., 2018; Table 2). The axon had 17 nodes and a total length of ∼1,800 µm.
Morphological properties of the model neuron compared with published values
Morphological parameters of the model's axon
Passive properties
The specific membrane capacitance was 1 µF/cm2 except in myelinated regions, where it was 0.1 µF/cm2. For all simulations, the resting membrane potential and temperature was set to the same value as in experiments, 31°C. The axial resistivity (Ri) was 150 Ω·cm for soma, axon hillock, and dendrites and 50 Ω·cm for the AIS and axon. Membrane leak conductance was 0.028 mS/cm2 for soma and dendrites and 1 mS/cm2 for nodes (Table 3).
Passive electrical properties of model neuron
Active properties
Conductances in our model included NaV1.6, KV7, KV1, delayed rectifier K+, A-type K+, AHP, H current, and T-type Ca2+ current. Channel kinetics were taken from published models (see Table 4 for ModelDB accession numbers). The Na+ channel kinetics differ between the soma and axon to include slow inactivation. The A-type K+ channel in the distal and proximal apical dendrites had different activation properties: vhalfn = −1 or 11 mV for distal and proximal dendrites, respectively. Guided by previously described spatial variations in conductance densities, the density of each conductance for each region was adjusted to reproduce our experimental results (Table 4). In brief, the density of NaV1.6 in the AIS and nodes was 13× and 3× higher, respectively, than that in soma (Gold et al., 2007). Ion channel densities were generally low in internodes. Density of KV7 channels was higher in nodes compared with internodes, and the density of KV1 channels was higher in the juxtaparanode (Krishnan and Kiernan, 2013). The model also included ChR2 (see below).
Type, ModelDB accession number, and density (in mS/cm2) of voltage-gated channels used in the model neuron
Simulation
The ICLAMP point process in NEURON was used to inject current at different points in the neuron model. Based on Foutz et al. (2012), the photostimulus was modeled in Python and interfaced with the ChR2 model in NEURON. In brief, the ChR2 model has four states, two open (O1, O2) and two closed (C1, C2), whose rate of change is determined by the following:
Spike-triggered average and covariance analysis
In simulations, noisy current was injected into a node of Ranvier at the middle of the intact axon or at the middle of the AIS. Noise was modeled as an Ornstein–Uhlenbeck process (Uhlenbeck and Ornstein, 1930):
The original unfiltered noise was used for all analyses. The spike-triggered average (STA) was calculated as the average of all stimulus segments preceding a spike. The mean (μ) was subtracted from the STA to show stimulus fluctuations around zero. For the spike-triggered covariance (STC), the covariance matrix of the spike-triggered stimuli was formed based on the following equation (Samengo and Gollisch, 2013):
Surrogate data were obtained by shuffling the spike train (Ratté et al., 2015). Diagonalization of
Axial current calculation
To calculate the axial current waveform a node receives during spike propagation, current was injected into the soma, and the intracellular and extracellular voltages were recorded from a paranode adjacent to (just upstream of) a node in the middle of the axon (Fig. 9A, diagram). Axial current (I1) was measured in mA based on Ohm's law:
Results
How does an axon transmit spikes reliably without generating additional spikes ectopically?
Spikes originating in or near the soma are normally transmitted down the axon without additional spikes being generated ectopically in the axon, but achieving this is not trivial. To explore the required properties, we simulated various ion channel combinations in a conductance-based multicompartment model of a CA1 pyramidal neuron with a myelinated axon ∼1,800 µm in length. Figure 1A illustrates the model neuron's morphology and highlights the stimulation and recording sites. In a model axon with a high density of sodium channels but reduced potassium channels (Fig. 1B), sustained somatic depolarization evoked spikes that originated at/near the soma and propagated down the axon; however, additional spikes originated in the axon (i.e., ectopically) and propagated down the axon but not back into the soma (Fig. 1B, left). Injecting current in the axon at node 6 evoked spikes originating in the axon; all spikes propagated orthodromically down the axon but only a subset propagated antidromically into the soma (Fig. 1B, right). When sodium channel density in the axon was reduced to prevent spikes being initiated by axon depolarization (Fig. 1C, right), somatic depolarization could still evoke repetitive spiking in the soma but those spikes failed to propagate down the axon (Fig. 1C, left). Systematically varying sodium channel density in the axon revealed a narrow range (green arrow) in which spikes originating at/near the soma were reliably propagated down the axon without ectopic spikes being generated in the axon (Fig. 1D). Unlike reducing sodium channel density, increasing potassium channel density enabled reliable propagation of spikes originating in/near the soma (Fig. 1E, left) and limited the response to abrupt axon depolarization to a single ectopic spike and altogether prevented ectopic spiking during gradual axon depolarization (Fig. 1E, right). Densities of both KV1 and KV7 were varied in these simulations, but subsequent work (Figs. 7, 8) identified KV1 as being critical. We sought to answer whether central axons use KV1 channels to mitigate ectopic spike generation without compromising reliable spike transmission.
Limited ion channel combinations enable reliable spike propagation without ectopic spike generation. A, Morphology of model neuron. Stimulation and recording sites are illustrated. Current was injected into either the soma or node 6. Voltage was recorded from the soma (black), node 6 (orange), and node 14 (green). B, Axon model with a high sodium channel density and low potassium channel density. Channel densities are reported as % of baseline value (see Materials and Methods). Somatic current injection evoked spikes originating at/near the soma that propagated down the axon (left). An additional spike originated in the axon (i.e., ectopically) yet failed to invade the soma (gray inset). Current injection in the axon (node 6) evoked spikes that propagated down the axon, with some also traveling back into the soma (right). C, Axon model with low sodium and potassium channel densities. Spikes evoked by somatic current injection failed to propagate down the axon (left). Current injection in the axon did not evoke spikes (right). D, Effect of varying sodium channel density in axon model with low potassium channel density. Response pattern over range of sodium channel densities. There is a narrow range (green arrow) in which spikes originating at/near the soma propagated reliably along the axon without generating ectopic spikes (green). Propagation failed (red) or ectopic spikes were generated (blue) for lower and higher densities, respectively. Gray insets highlight successful spike propagation but with reduced amplitude and speed (a,b) for comparison with ideal performance (c). E, Axon model with high sodium and potassium channel densities. Somatic current injection evoked spikes that were reliably propagated down the axon (left). Abrupt axonal depolarization evoked only a single spike (top right) and gradual axonal depolarization evoked no spikes (bottom right).
Spike initiation properties differ qualitatively between the soma/AIS and distal axon
To study axon excitability experimentally without recording from the axon, we photostimulated the axon of CA1 pyramidal neurons expressing ChR2 while recording from the soma (Fig. 2A). Photostimuli were spatially restricted to a spot ∼20 µm in diameter and were targeted to either the axon as it runs in the alveus or, for comparison, to the soma. The axon of recorded cells could not be visualized and so the photostimulus was moved systematically away from the soma while monitoring responses recorded at the soma. The sustained photostimulus step was always preceded by a prepulse to remove the inactivating component of the ChR2 current, thus rendering the step-evoked photocurrent relatively “square” (Fig. 2A); the prepulse is not shown in later figures illustrating responses measured in current clamp. Synaptic transmission was blocked by bath-applied CNQX, d-APV, and bicuculline to prevent the direct neuronal response to photostimulation being obscured by photo-evoked synaptic input.
Sustained depolarization of the soma or axon evokes different spiking patterns. A, Strategy. Photostimulation was targeted to the soma or axon of CA1 pyramidal neurons while simultaneously recording from the soma. In simulations, voltage was recorded at additional loci including the axon initial segment (AIS) and distal axon. Distance to the photostimulation site is calculated as the shortest distance from the soma, which underestimates the total length of axon (see Materials and Methods). Top schematic shows axon of CA1 neuron projecting to the subiculum via the alveus. Trace shows sample voltage-clamp recording of response to somatic photostimulation. Photostimuli always included a prepulse to remove the inactivating component of the ChR2 response; all subsequent traces (in current clamp) illustrate only the response to the principal stimulus, whose intensity was varied (see Materials and Methods). B, Sample experimental data. Somatic current injection (left) evoked spiking patterns very similar to those evoked by photostimuli targeted to the soma (middle). In both cases, increasing the stimulus amplitude above threshold (Th.) evoked repetitive spiking. Photostimulating the distal axon (right) evoked a single spike even as stimulus intensity was increased. Gray insets show dV / dt plotted against V to show difference in voltage trajectory at spike initiation. Curvature of the voltage trajectory was not quantified, but the qualitative difference between smooth and abrupt spike onsets was generally clear and corresponded to different values of voltage threshold. Voltage threshold (red arrow) is determined as voltage where dV / dt (normalized to peak velocity during spike upstroke) exceeds a 1% change (dotted red line) from baseline (dotted white line). Spikes evoked by somatic stimulation had a smooth onset and a high (depolarized) voltage threshold (∼−47 mV) whereas spikes evoked by axonal stimulation had a more abrupt onset and low voltage threshold (∼−70 mV). C, Sample simulation data. Like in experiments, current injection (left) or photostimulation (middle) of the soma at 1.5× and 2× threshold evoked repetitive firing whereas photostimulating the axon evoked a single spike (right). Spike onset (based on somatic voltage) was smooth for spikes evoked by somatic stimulation or abrupt for axonal stimulation (gray insets). Voltage recorded at different locations (beige insets) confirm that “smooth” onset spikes originate (i.e., are seen first) in the AIS whereas “abrupt” onset spikes originate in the axon.
In current clamp, both current injection and photostimulation evoked repetitive spiking when applied at the soma (Fig. 2B, left, center), whereas photostimulating the axon far (>600 μm) from the soma evoked a single spike (Fig. 2B, right). The neuron model with KV1 channels in the axon (Fig. 1E) reproduced the difference in spiking patterns depending on stimulation site (Fig. 2C). Beyond differences in spiking pattern, comparing the spike waveform itself revealed differences relevant for inferring the site of spike initiation. For somatic recordings, spikes originating near the soma have a smooth onset whereas spikes that propagate to the soma (from a remote site of origin) have an abrupt, or kinky, onset (McCormick et al., 2007; Thome et al., 2018). The voltage trajectory at spike onset is not an intrinsic feature of the spike but, rather, depends on the spatial proximity between spike initiation and the recording site: A spike has a smooth onset when recorded near its initiation site (i.e., when recording and initiation sites are electrically coupled) whereas the same spike has an abrupt onset when recorded at a site to which it has propagated (Telenczuk et al., 2017). Therefore, the smooth onset of spikes evoked by somatic stimulation confirms that those spikes originate at/near the soma whereas the abrupt onset of spikes evoked by axon photostimulation confirms that those spikes arrive from the axon, presumably originating at/near the photostimulation site (see gray insets in Fig. 2B showing dV/dt at spike threshold). Simulations confirmed that smooth-onset spikes originate in the AIS, adjacent to the soma, whereas abrupt-onset spikes originate in the distal axon (Fig. 2C). Since we always recorded from the soma during experiments, we identified spikes of somatic origin by their smooth onset and hereafter refer to them as SO spikes, whereas we identified spikes of axonal origin by their abrupt onset and hereafter refer to them as AO spikes.
Axon photostimulation closer (<600 μm) to the soma often evoked multiple spikes but, notably, only the first spike had the abrupt onset characteristic of an AO spike, suggesting that later spikes are SO spikes (Fig. 3A). Somatic depolarization was also evident under these stimulus conditions. Somatic depolarization (and SO spikes) could be driven by current originating from on-target activation of ChR2 in the axon that spreads electrotonically to the soma or by off-target activation of ChR2 on dendrites. Photostimulating the alveus at sites equidistant from the soma toward or away from the subiculum (directions designated as + and −, respectively), evoked comparable ChR2 current measured in voltage clamp (Fig. 3B). Given the absence of the axon in the − direction (Yang et al., 2014), the observed symmetry supports off-target activation of dendritic ChR2. Furthermore, the axon length constant is relatively short according to simulations (Fig. 3C), arguing against much contribution from on-target activation of axonal ChR2 (see also Discussion). We therefore incorporated light scatter into simulated photostimuli, using data in Figure 3B to tune the degree of scattering (Fig. 3D). Figure 3E shows the spatial extent of ChR2 activation before (top right) and after (bottom left) incorporating light scatter. Importantly, the dendrites, soma, and axon, respectively, constitute 81, 2, and 17% of the model neuron's total surface area and, moreover, dendrites are densely branched within a small volume, unlike the axon. Therefore, even a small amount of scattered light reaching the dendrites and causing weak ChR2 activation can, cumulatively, evoke a large ChR2 current in the dendrites according to our simulations.
Repetitive spiking during axon photostimulation arises from scattered light triggering spikes in the AIS. A, In experiments, photostimulating the axon closer to the soma evoked somatic depolarization and multiple spikes, but only the first spike had an abrupt onset characteristic of an AO spike (gray insets compare dV / dt at spike onset). B, Voltage-clamp recordings of ChR2 current evoked by photostimulation of the alveus at different distances from the soma in the direction toward (+) or away from (−) the subiculum. The axon is present only in the + direction yet photostimuli in either direction evoked similar current, suggesting that scattered light reaches and activates ChR2 in the dendrites of the recorded cell, as opposed to current spreading electronically to the soma from ChR2 activated in the axon. C, Axon length constant (λ). Voltage change (ΔV) caused by injecting depolarizing or hyperpolarizing current at the soma was measured at the soma and at nodes located at different distances along the axon. Data were fit by the equation
Given the plausibility of scattered light causing off-target activation of ChR2, we retested our model neuron with photostimuli that included light scatter. Axon photostimuli with scatter evoked a mixture of AO and SO spikes (Fig. 4A, right), unlike the single AO spike observed with truly focal photostimuli (Fig. 4A, left). The exact pattern of SO and AO spikes depended on the position of the photostimulus along the axon. For instance, a scattered photostimulus aimed at 713 µm from the soma evoked an SO spike at lower intensities than required for an AO spike (Fig. 4A, compare spikes c and a), but when the scattered photostimulus was shifted farther down the axon, an AO spike was evoked at a lower photostimulus intensity than the SO spike (Fig. 4B) because AO photothreshold was insensitive to distance whereas SO photothreshold increased with increasing distance from the soma. Shifting the scattered photostimulus even farther down the axon continued this trend such that only AO spikes were observed for experimentally realistic photostimulus intensities (Fig. 4C). Repetitive SO spikes occurred when scattered photostimulation was applied to the soma (Fig. 4D), like during focal photostimulation (compare Fig. 2C), but at a slightly lower threshold. These simulation results are entirely consistent with current clamp recordings of photo-evoked responses.
Light scatter reproduces AO and SO spikes during axon photostimulation in model. A, Potostimulating the model axon 713 μm from soma without light scatter evoked a single spike even at 2xTh (left). Photostimulating the same location with scattered light evoked somatic depolarization and repetitive spiking (right) but only the first spike (a) was AO whereas subsequent spikes (b) were SO based on comparison of voltage recorded from different sites (beige insets). Photostimulating at an intensity <AO photothreshold evoked a single SO spike (c). B, Photostimulating farther from the soma (813 μm) with light scatter did not evoke SO spikes below AO photothreshold (unlike panel A) but suprathreshold photostimulation evoked an AO spike followed by several SO spikes. Notably, the AO threshold did not vary with distance from the soma, unlike the SO threshold. C, Axon photostimulation with light scatter at 1,011 μm from the soma evoked on a single AO spike, like more proximal axon photostimuli without light scatter. D, Photostimulating the soma with light scatter decreased the SO threshold and increased the number of spikes (compare with Fig. 2C, middle).
Overall, our data show that axons respond to sustained depolarization with a single AO spike, consistent with responses in model axons containing KV1. However, despite our efforts to restrict photostimulation to the axon, even a small amount of scattered light appeared to activate ChR2 on the dendrites, depolarizing the soma and often evoking SO spikes. Because SO spikes may obscure the pattern of AO spikes, additional experiments were conducted to better separate AO and SO spikes, and thereby verify if the axon spiking response is indeed transient.
Somatic hyperpolarization and local sodium channel blockade differentiate SO and AO spikes
Consistent with past work showing that initiation and antidromic propagation of AO spikes into the soma are not affected by somatic hyperpolarization (Bahner et al., 2011; Thome et al., 2018), we found that AO spikes evoked by axon photostimulation or electric field stimulation (just distal to the photostimulation site) were unaffected by somatic hyperpolarization, whereas subsequent SO spikes were abolished (Fig. 5A). Spikes evoked by axon photostimulation had the same abrupt onset as spikes evoked by electric field stimulation (see insets showing dV/dt at spike threshold). Systematic comparison of responses with prestimulus somatic membrane potential (Vsoma) adjusted to −70 or −90 mV confirmed that SO spikes were abolished by somatic hyperpolarization whereas the AO spike remained (Fig. 5B). Also, hyperpolarizing the soma significantly increased the somatic photostimulus intensity required to initiate an SO spike from a median of 7.8 mW/mm2 (IQR = 6.2–16.4) to 23.4 mW/mm2 (IQR = 15.4–62.4; Wilcoxon signed rank test, Z = 3.18, p < 0.001) but did not affect the minimum axon photostimulus intensity required to initiate AO spikes (Z = −1.34, p = 0.5; Fig. 5C).
Separation of SO and AO spikes by somatic hyperpolarization. A, Hyperpolarizing the somatic membrane potential (Vsoma) abolished SO spikes evoked by scattered light during axon photostimulation without affecting the AO spike. Vsoma is indicated beside each trace. For comparison, electric field stimulation was applied to the alveus to evoke an AO spike which, like the AO spike evoked by axon photostimulation, was not abolished by somatic hyperpolarization. Gray insets compare dV / dt at onset of AO spikes evoked by each type of axonal stimulus. B, Effect of adjusting Vsoma to −70 or −90 mV. All SO spikes evoked by axon photostimulation at the AO threshold were abolished by hyperpolarization whereas the AO spike remained (n = 5 cells). C, Effect of Vsoma on photothreshold. The minimum photostimulus intensity required to evoke SO or AO spikes was found by titrating the axon photostimulus intensity with Vsoma adjusted to −70 or −90 mV. SO photothreshold (left) was significantly increased (Wilcoxon signed rank test, Z = 3.18, p < 0.001, n = 12 cells) from 7.8 mW/mm2 (median) at −70 mV to 23.4 mW/mm2 at −90 mV whereas the AO photothreshold was not significantly altered (Z = −1.34, p = 0.5, n = 5 cells) at 156.1 mW/mm2 and actually decreased in two neurons. Box plots (gray) report the median and interquartile range. D, Phase portraits illustrate the relationship between voltage and its first derivative (dV / dt, top) and second derivative (d2V / dt2, bottom) during AO spike. The AO spike evoked from −90 mV (green) exhibited a noticeable inflection in its rising phase (red *) compared with AO spiked evoked from −70 mV (black). This inflection reflects the additional time required to charge the somatic membrane (via axial current arriving from the axon) before somatic sodium channels start to activate. Note that during an SO spike, axial current from the axon is absent (or minimal) and spike initiation relies on injected current or transmembrane ChR2 current to drive the depolarization required to activate local sodium channels.
Effects of somatic hyperpolarization on the voltage trajectory of AO spikes recorded in the soma further attest to the axonal origin of AO spikes. As an AO spike propagates into the soma, axial current flowing from the axon rapidly charges the somatic membrane, activating local voltage-dependent sodium channels whose transmembrane inward current perpetuates the upswing of the spike. If the soma is hyperpolarized, greater charging of the somatic membrane by axial current is required before local voltage-gated sodium channels are eventually activated, leading to a delay that is evident as a notch in the early voltage trajectory, which is most evident by considering “acceleration” in V (i.e., d2V/dt2) plotted against V (Fig. 5D).
To further distinguish SO and AO spikes, we preferentially blocked voltage-gated sodium channels in the soma or axon, respectively. To block somatic channels preferentially, QX-314 was applied intracellularly via the patch pipette; the axon photostimulus was configured prior to establishing the whole-cell recording configuration. Most SO spikes were promptly abolished shortly after breakthrough, as the somatic cytosol was dialyzed with QX-314, and all SO spikes were lost several minutes before the AO spike was eventually blocked, as QX-314 slowly spread into the axon (Fig. 6A). This pattern was observed in 5 of 5 neurons tested with QX-314 and was reproduced in the model neuron by progressively reducing the density of voltage-gated sodium channels in the soma, AIS, and axon (Fig. 6B; see also Table 5). To block AO spikes preferentially, TTX was puffed onto the axon near the photostimulation site. The AO spike evoked by axon photostimulation was abolished by TTX whereas SO spikes remained and actually occurred earlier after stimulus onset (Fig. 6C). This pattern was observed in 3 of 3 neurons tested with axonally puffed TTX and was reproduced in the model neuron by reducing the density of voltage-gated sodium channels in the axon (Fig. 6D).
Separation of SO and AO spikes by localized sodium channel blockade. A, Applying QX-314 via the patch pipette gradually eliminated all SO spikes long before the AO spike was lost. Time 0 corresponds to when the cell membrane was ruptured, allowing QX-314 to enter the somatic cytosol from the patch pipette. B, In simulations, reducing sodium channel density in the soma yielded a similar progressive loss of SO spikes before the eventual loss of the AO spike. Changes in sodium channel density are summarized in Table 5. Beige insets depict the voltage at the soma (black), AIS (pink), and the axon (purple) showing that AO spike failed to be regenerated at the soma as sodium channel density was reduced. C, Puffing TTX locally on the axon eliminated the AO spike (a) without reducing SO spikes. In fact the first SO spike (b) occurred with a shorter latency in the absence of the AO spike. Gray insets confirm that the first spike before TTX is AO (a) whereas the first spike after TTX is SO (b). D, In simulations, reducing sodium channel density in the axon prevented axon photostimulation from evoking an AO spike without preventing scattered light from evoking SO spikes. Like for experiments, the first SO spikes occurred earlier in the absence of the AO spike. Beige insets confirms that first spike before TTX is AO whereas the first spike after TTX is SO.
Na+ channel reduction (% from baseline) at time points i–v in Figure 6B
KV1 channels regulate spike initiation in the axon
Having established that the axon, unlike the soma/AIS, responds to sustained depolarization with a single spike, we proceeded to explore the basis for this spiking pattern, following up on preliminary simulations in Figure 1. Reducing KV1 density converted the model axon photoresponse from transient to repetitive spiking (Fig. 7A), where all spikes recorded in the soma now had an abrupt onset. Recording voltage at multiple points along the model axon confirmed that all spikes originated in the axon (Fig. 7A, inset). In experiments, applying a low-dose of 4-AP to block KV1 channels caused axon photostimulation to evoke repetitive spiking, where several of the late spikes clearly had an abrupt onset characteristic of AO spikes (Fig. 7B). In the presence of 4-AP, axon photostimulation evoked repetitive AO spikes in 8 of 10 cells tested, whereas repetitive AO spikes were observed in just 1 of 15 cells tested in control conditions (Fig. 7C; χ2 = 14.0, p = 0.00018). Moreover, the minimum axon photostimulus intensity required to evoke AO spikes dropped significantly, from a median of 93.6 mW/mm2 (IQR = 78.0–171.1) in control conditions to 39.0 mW/mm2 (IQR = 21.5–115.1) in 4-AP (Fig. 7D; Mann–Whitney test, U = 20, p = 0.023). It is also notable that spikes with abrupt onsets were sometimes observed to occur spontaneously in 4-AP and could not be abolished by somatic hyperpolarization (Fig. 7E). Blockade of synaptic transmission (see Materials and Methods) prevented the epileptiform network activity that would otherwise occur with bath-applied 4-AP.
Blockade of KV1 channels encourages repetitive AO spikes in the axon. A, KV1 channel density was reduced in the axon model and the number of spikes was determined for different axon photostimulus intensities. Blocking KV1 channels converted the response to axon photostimulation from a single spike to repetitive spiking. All spikes were AO (see insets). B, Applying 4-AP to block KV1 channel caused repetitive AO spiking in response to axon stimulation, consistent with simulations. C, In control conditions, only 1 of 15 neurons tested ever exhibited >1 AO spike during axon photostimulation, whereas 8 of 10 neurons exhibited repetitive AO spikes in the presence of 4-AP, which is a highly significant change in proportion (χ2 = 14.0, p = 0.00018). D, AO photothreshold was significantly reduced by 4-AP (Mann–Whitney test, U = 20, p = 0.023, n = 15 and 7 cells for control and 4-AP, respectively), consistent with simulations (panel A). Box plots (gray) summarize median and IQR of group data. E, Example of spontaneous AO spikes. These spikes, with their characteristic abrupt onset, could not be abolished by somatic hyperpolarization.
Impact of axon spike initiation properties on information transmission
According to simulations, the axon initiates a spike only if its voltage reaches threshold faster than KV1 channels activate (Fig. 8A), consistent with previous work on the effects of low-threshold K currents at somatic responsiveness (Reyes et al., 1994; Svirskis et al., 2004; Prescott et al., 2008; Higgs and Spain, 2011; Ratté et al., 2015; Azevedo and Wilson, 2017; Baldassano and MacLeod, 2022). By extension, slowly ramped photostimuli failed to evoke AO spikes (Fig. 8B), consistent with current injection data in Figure 1E. To compare the input waveform best suited to evoke spikes in the AIS and axon, we injected noisy current (τ = 5 ms) into our model neuron in each region and calculated the STA and spike-triggered covariance (Fig. 8C). The AIS (left) exhibited a broad monophasic STA and one significant eigenvalue, consistent with a low-pass filter, whereas the axon (right) exhibited a narrow biphasic STA and two significant eigenvalues, consistent with a high-pass filter. When KV1 channels were removed, the axon exhibited a monophasic STA and a single significant eigenvalue (Fig. 8D), like the AIS, thus demonstrating that KV1 channels mediate the high-pass filter. Projecting spike-triggered stimuli onto eigenvectors (Fig. 8C, bottom panels) shows that spike initiation in the AIS depends exclusively on the amount of depolarization (feature 1), whereas the axon with KV1 channels is also sensitive to the rate of depolarization (feature 2). Figure 8E illustrates the practical consequences of these local differences in excitability. From these results, we conclude that CA1 pyramidal cells behave as integrators because of the low-pass filtering/class 1 excitability of their AIS but their axons exhibit high-pass filtering/class 3 excitability.
KV1 channels implement a high-pass filter that explains transient spiking in the axon. A, Plotting voltage (black) against KV1 conductance (red) shows that axon voltage must reach spike threshold before KV1 channels strongly activate. Zooming in on the perithreshold voltage range (yellow) shows that a spike initiates only if the voltage trajectory escapes to the right of the separatrix (purple); a just-subthreshold voltage trajectory (gray) is shown for comparison. The separatrix reflects a race between fast-activating inward and slower-activating outward currents. B, Ramped axon photostimulation did not evoke an AO spike despite exceeding the intensity at which an AO spike is evoked by abrupt-onset photostimulation at the same location. C, Noisy current (τ = 5 ms) was injected into the AIS or axon node to calculate the spike-triggered average (STA, black) and spike-triggered covariance. Insets show eigenvalues from spike-triggered stimuli (circles) and surrogate data (black line). For the AIS (left), the STA was broad and monophasic, and there was only one significant eigenvalue (orange). For the axon (right), the STA was narrower and biphasic, and there were two significant eigenvalues (orange and green). The latter pattern is characteristic of a high-pass filter. Bottom panels show spike-triggered stimuli (black) and surrogate data (gray) projected onto features (eigenvectors) corresponding to the two smallest eigenvalues. Dimensions of the black cloud relative to the gray cloud reflect which stimulus features influence spike initiation and argue that spike initiation in the AIS depends solely on the amount of depolarization whereas spike initiation in the axon depends jointly on the amount and rate of depolarization. D, KV1 channels implement the high-pass filter. When KV1 channels in the axon were reduced by 90%, the STA was monophasic and only one eigenvalue was significant, like the AIS and unlike the intact axon in C. E, Sample responses of axon and AIS reveal sensitivity to different stimulus features. Sample traces (top) show voltage responses at the axon (gray) and AIS (black) to noisy current (red, τ = 0.5 ms, σ = 0.22 nA and μ = 0.1 nA) injected at the axon or soma, respectively. Noise was injected at the soma rather than directly into the AIS to model the effect of synaptic current passing through the soma to reach the AIS and being low-pass filtered along the way. The voltage response in the AIS is much smoother than in the axon. Whereas abrupt increases in input (gray brackets) drive the axon to spike, a sustained increase (black bracket) is required to activate the AIS. Brackets are shown to scale underneath STAs (bottom) calculated from ∼10,000 spikes for the corresponding condition. This noise is still slower than inputs typically experienced by the axon (Fig. 9A) but is useful for making comparisons since both the AIS and axon can respond to it (Fig. 9C).
To explore the consequences of the KV1-mediated high-pass filtering observed in our axon model, we measured the axial current experienced by a node during saltatory spike propagation, which is the natural “input” to which each node must respond by initiating its own spike. Unlike the comparatively slow, sustained synaptic depolarization received by the AIS (Fig. 8E), nodes experience a short but intense pulse of axial current (Fig. 9A) each time a spike propagates down the axon. That pulse is faster than the axon STA in Figure 8C but, notably, the STA also depends on noise kinetics. Testing the axon with faster noise (τ = 0.5 ms) revealed a faster STA (Fig. 9B) that matches the axial current kinetics. When stimulated with fast noise (Fig. 9C, top), low-pass filtering prevented any spikes from initiating at the AIS (black), unlike the axon's vigorous response (gray). Slow noise (τ = 50 ms) produced the opposite pattern (Fig. 9C, bottom), confirming that the axon and AIS are tuned very differently, consistent with the distinct signals they normally process (Fig. 9D). In other words, the slow inputs that normally drive spiking in the AIS cannot activate the axon and, conversely, the fast inputs that drive spiking in the axon cannot activate the AIS, at least not if arriving via the soma.
The high-pass filter implemented by KV1 channels matches the axial current waveform. A, Axial current during spike propagation. Pink trace shows the axial current experienced by a node when a spike occurs at the immediate upstream node. The time axis is reversed to correspond to left-to-right flow of axial current in schematic (top), which illustrates how axial current is calculated (see Materials and Methods for details). B, STA and STC analysis for axon using faster noise (τ = 0.5 ms) that approximates axial current kinetics. The STA was narrower, consistent with axial current waveform in A. There was only one significant eigenvalue, suggesting that rate of depolarization is unimportant, but this is because all stimulus fluctuations are fast; indeed, these fluctuations are too fast to evoke spikes in the AIS (see C). The axial current waveform (from A) projected onto the feature space (pink, mean ± standard deviation for 10 trials) differs significantly from 0 (t9 = 4.56, p = 0.0014, 9 one-sample t test). C, Sample responses to fast (τ = 0.5 ms) and slow (τ = 50 ms) noise. For fast noise (top), the axon spiked vigorously and exhibited large voltage fluctuations between spikes, whereas the AIS was quiescent and its voltage fluctuations were blunted. For slow noise (bottom), the axon was quiescent whereas the AIS spiked regularly despite input fluctuations. Since input to the AIS normally arrives via the soma, noise was injected into the soma to assess the AIS response (like in Fig. 8C). Noise was injected into the axon to assess the axon response. σ = 0.22 nA and μ = 0.1 nA for all cases except slow noise applied to the soma, where μ was reduced to 0.05 nA to reduce the spike rate. D, Schematic of region-specific processing of neural signals. The neural signal can exist in analog or digitals forms, as graded potentials or spikes, respectively. The dendrites and soma compute in analog (red) but the AIS converts (transduces) the signal to digital (purple). The axon then capitalizes on resilience of digital signals to reliably transmit information over long distances (cyan) before the presynaptic terminals convert the signals back to analog (purple). By matching the axial current waveform, the high-pass filtering implemented by KV1 channels supports reliable spike propagation without encouraging ectopic spike generation.
Discussion
By optogenetically stimulating the intact axon of CA1 pyramidal neurons while recording evoked spiking at the soma, we showed that the axon and soma/AIS have distinct spike initiation properties. Whereas the soma of a CA1 pyramidal neuron responds to sustained depolarization with repetitive spiking, its axon responds with transient spiking (Fig. 2). The transient spiking pattern is due to KV1 channels (Fig. 7), which implement a high-pass filter. High-pass filtering ensures that the axon initiates a spike only if it is rapidly depolarized (Fig. 8), thus preventing slow depolarization from triggering ectopic spikes. According to our computational modeling, the filter implemented by KV1 matches the axial current waveform associated with saltatory spike propagation (Fig. 9). This in turn suggests that axon excitability is optimized for the purpose of digital signal transmission, namely, to transmit spikes originating in the AIS. This is very different from the AIS, which is responsible for transducing graded synaptic depolarization into de novo spikes. Overall, these results argue that different regions of a neuron have intrinsic excitability that is specialized for the function of that region (Fig. 9D).
Despite our efforts to restrict photostimuli to the axon to selectively evoke axon origin (AO) spikes, light scattering was found to activate ChR2 on dendrites, triggering unintended somatic origin (SO) spikes (Fig. 3). This complicated the interpretation of experimental data, but the identities of AO and SO spikes were verified by their differential sensitivity to somatic hyperpolarization (Fig. 5) and local sodium channel blockade (Fig. 6). Furthermore, simulations showed that the axon has a short length constant (λ), meaning ChR2 current activated in the axon cannot spread appreciably to the soma (see below).
In myelinated axons, spikes are transmitted by being regenerated at consecutive nodes of Ranvier. Upon spike regeneration at a node, the resulting axial current propagates along the internodal regions toward the downstream nodes and back toward the AIS and soma. This axial current depends on various intrinsic and extrinsic factors (Abdollahi and Prescott, 2024). Estimates of the axonal λ have typically ranged from 400 to 550 μm (Alle and Geiger, 2006; Shu et al., 2006; Kole et al., 2007) but have reached as high as 800 μm (Alle and Geiger, 2008). Our passive axon model had a λ value in this range (∼520 μm; Fig. 3C), but λ dropped to 300–350 μm in the axon model with KV1 channels, which is similar to other measurements (150–350 μm; Sasaki et al., 2012; Zbili et al., 2020, 2021). Differences in λ depending on whether depolarizing or hyperpolarizing pulses are applied is due to KV1, which preferentially attenuates the spread of depolarizing current; even hyperpolarizing current spreads less than in a passive axon because some KV1 is active at resting membrane potential (Hu and Bean, 2018). Past models (Shu et al., 2006) did not include KV1 channels. Experimental conditions are also a factor; for example, Alle and Geiger (2006) measured λ while hyperpolarizing the cell to −80 mV, which inactivates KV1 and lengthens λ according to our simulations (Fig. 3C). Since axon recordings are normally conducted from blebs, we also simulated bleb recording conditions by terminating the model axon at different lengths from the soma and measuring
As discussed above, voltage-gated channels can attenuate or amplify graded depolarization, which is of course well established in dendrites (Spruston et al., 2016). But sodium channel activation during saltatory spike propagation is best conceptualized as a signal restoration process—axial current (a decrementing analog signal) is converted back to a spike (an all-or-none digital signal) at each node, as opposed to axial current being intermittently boosted. Repeated interconversion between analog and digital signals prevents noise accumulation (Sarpeshkar, 1998). The high density of Na channels localized to each node is important for restoring the digital signal upon receipt of an analog signal, but the appropriate analog signal must be differentiated from noise that might otherwise trigger new, inappropriate (i.e., ectopic) spikes (Fig. 1).
We found that axons, unlike the AIS, are sensitive to two features of their input, namely, the amount and rate of depolarization. Indeed, the axon behaves as a high-pass filter by exclusively responding to high amplitude (intense) and high frequency (abrupt) stimuli (Fig. 8). Calculating the axial current received by each node during spike propagation showed that the axial current is indeed an intense, abrupt pulse (Fig. 9). Matching each node's spike initiation filter to the axial current waveform associated with a propagated spike optimizes the fidelity of spike transmission insofar as filter matching increases the signal-to-noise ratio (Turin, 1960). In comparison, graded synaptic depolarizations are not simultaneously intense and abrupt but instead tend to vary gradually, in part because of somatodendritic filtering. The axon will not respond with ectopic spikes to this sort of input; in contrast, the AIS responds to this graded input with spiking whose rate is proportional to the depolarization. Conversely, the AIS is limited in its capacity to respond to high-frequency signals because of its proximity to the soma which, because of its large surface area, charges slowly. The high- and low-pass filtering properties of the axon and AIS are well suited for their respective functions (Fig. 9D). Pathological conditions such as demyelination can disrupt the intensity and kinetics of the signal transmitted along the axon, causing nodes to fail to reliably propagate spikes or, depending on compensatory changes (like downregulation of KV channels), to produce ectopic spikes (Coggan et al., 2010).
KV1 channels reduce the energy efficiency of spikes (Hallermann et al., 2012) but, given the ubiquitous expression of KV1 channels in axons, those channels must serve a function whose value offsets the increased metabolic cost. Our findings reveal that KV1 channels critically shape axon filter properties (Fig. 8). KV2.1 channels, although likely blocked by low-dose 4-AP, cannot account for our observations (see Materials and Methods). The juxtaparanodes of myelinated axons are known to have a high density of KV1 channels (Wang et al., 1993). Input must be intense and abrupt to activate Na channels in the nodes before KV1 channels in the juxtaparanodes get activated. When the input is gradual, KV1 channels activate and prevent the axon from generating a spike, even if the input is intense. This is consistent with studies showing that lack of KV1 channels in the axon results in hyperexcitability (Smart et al., 1998; Brew et al., 2003). Restricted localization of KV1 channels in juxtaparanode regions is crucial for axonal excitability. For example, in demyelinated optic nerves, KV1 channels spread away from the nodes, disturbing spike propagation (Bagchi et al., 2014). On the other hand, dispersion of KV1 channels away from the node can suppress hyperexcitability arising from other compensatory changes (Calvo et al., 2016).
Digital signals, because of their resilience to noise, are well suited for transmitting information reliably whereas analog signals are more appropriate for efficient computing (Mead, 1990; Sarpeshkar, 1998). Neurons exploit both types of signals, conducting computations in analog in their dendrites and soma and then transmitting the resultant signals in a digital form. Indeed, spikes cost more energy and convey less information than graded potentials (Sengupta et al., 2014), but the benefits presumably outweigh the costs. Analog-to-digital transduction at the AIS dictates neuronal spiking patterns and coding properties, and it is, therefore, appropriate that plasticity of the AIS has attracted substantial attention (Kole and Stuart, 2012). Converting spike trains into synaptic transmitter release at the other end of the axon offers another important locus for plasticity, in this case representing digital-to-analog transduction. By comparison, transmitting spikes along the intervening axon may seem mundane, but transmitting signals with high fidelity over long distances despite noisy conditions, noisy components, and other biological constraints (Laughlin and Sejnowski, 2003) is not trivial. Our results suggest that axons use KV1 channels to implement a filter matched to the signal intended for transmission, thus supporting saltatory conduction of spike originating at the AIS. By doing this, axons can maintain a high density of sodium channels (as required for rapid and reliable transmission) without the risk of generating ectopic spikes. Proper neurological function hinges on this subtle yet elegant aspect of axonal design.
Footnotes
This work was funded by the Canadian Institutes of Health Research (CIHR) Foundation Grant 167276 to S.A.P.
↵*N.A. and Y-F.X. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Steven A. Prescott at steve.prescott{at}sickkids.ca.