| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Department of Neurobiology, Kavli Institute for Neuroscience, Yale University School of Medicine, New Haven, Connecticut
Submitted 30 August 2006; accepted in final form 2 November 2006
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
; Colbert and Pan 2002; Mainen et al. 1995; Milojkovic et al. 2005; Stuart et al. 1997a,b). Which compartment is the first to initiate action potentials and how far these spikes propagate and interact with the other compartments may have a significant effect on short- and long-term cortical function (Golding et al. 2002; Larkum et al. 1999a; Svoboda et al. 1997; Waters and Helmchen 2004). In addition, the different compartments of cortical neurons may interact to generate complex patterns of spike generation, such as burst discharges, prolonged depolarizations, and oscillations (Helmchen et al. 1999; Kim and Connors 1993; Larkum and Zhu 2002; Larkum et al. 2001; Stuart et al. 1997b; Wang 1999).
Investigations of even low-frequency single-spike initiation (the simplest form of action-potential generation) in cortical pyramidal cells have yielded some conflicting results, although the general consensus is that in response to somatic depolarizing current pulses or local, electrical activation of threshold synaptic inputs, fast spikes are initiated in the axon initial segment followed by back propagation into the soma and apical dendrite (Palmer and Stuart 2006; Stuart et al. 1997a,b). This finding is in agreement with classical studies in spinal motoneurons (Coombs et al. 1957; Fuortes et al. 1957), although some studies have suggested that the first node of Ranvier, and not the axon initial segment, is the typical site of action-potential initiation in some cell types (Clark et al. 2005; Colbert and Johnston 1996; Colbert and Pan 2002).
Although these studies indicate that the axon initial segment is the preferred site of spike initiation under the relatively quiescent conditions of the in vitro slice, the site of fast spike initiation under conditions when the cortical network is active is still relatively unexplored. In vivo, cortical pyramidal neurons receive tens of thousands of synaptic inputs, which often culminate in prolonged depolarizations and significant increases in membrane conductances of both dendritic and somatic compartments (Destexhe et al. 2001; Steriade et al. 2001). One example of complex barrages of synaptic activity is the up state of the slow oscillation that occurs during slow wave sleep or anesthesia (Haider et al. 2006; Steriade et al. 1993, 2001). The synaptic barrages arriving during the up state display properties that are similar, but not identical, to those of the waking state (Steriade et al. 2001). The properties of fast spike initiation in cortical pyramidal cells during ongoing barrages of synaptic activity, and how these give rise to spike threshold variability (Azouz and Gray 1999), are not well understood. Simultaneous dendritic and somatic extracellular recordings in vivo indicate that action potentials are typically back-propagating from the soma into the apical dendrites, although the precise location of initiation of these spikes could not be discerned (Buzsaki and Kandel 1998). Simultaneous in vivo dendritic Ca2+ imaging and somatic recordings suggest that layer 2/3 pyramidal cells may generate apical dendritic spikes only after somatic spikes (Svoboda et al. 1997, 1999), whereas the apical dendrites of layer 5 pyramidal neurons may occasionally generate Ca2+ transients (presumed action potentials) in isolation of somatic spikes (Helmchen et al. 1999). Indeed, in vitro studies have suggested that action potentials may initiate first from dendritic compartments during strong and/or synchronous synaptic stimulation to localized or distributed portions of the dendritic tree (Gasparini et al. 2004; Golding and Spruston 1998; Larkum et al. 1999a,b; Losonczy and Magee 2006; Milojkovic et al. 2005; Regehr et al. 1993; Schiller et al. 1997), and these dendritic spikes may or may not propagate completely to the soma (Jarsky et al. 2005; Larkum and Zhu 2002; Larkum et al. 2001; Losonczy and Magee 2006; Milojkovic et al. 2005; Schiller et al. 1997). Dendritic and somatic interactions are even more complex during the generation of bursts of action potentials, in which rapid action potentials may be initiated in the axonal/somatic compartments, whereas the generation of slower spikes or other active depolarizing events in the dendrites provide the prolonged depolarization that drives the burst (Helmchen et al. 1999; Kim and Connors 1993; Larkum and Zhu 2002; Larkum et al. 2001; Stuart et al. 1997b; Wang 1999).
These results suggest that the site of spike initiation in regular spiking cortical pyramidal cells may vary according to the nature of the initiating stimulus, with threshold stimuli preferring initiation in the axon (initial segment or 1st node of Ranvier) and strong, synchronized, or distal stimuli occasionally initiating spikes from the dendrites. Given the possibility of spikes being initiated in variable portions of the neuron, it is not clear where these action potentials will be initiated under the various conditions occurring in vivo, including during ongoing synaptic barrages (either spontaneous or evoked by sensory stimuli), or under pathological conditions, such as during epileptic seizures. Here, as a first step toward investigating this question, we examine the properties of spike initiation in cortical pyramidal cells during the spontaneous generation of up states in vitro and in vivo, in response to sensory stimuli in vivo, during epileptiform discharges in vitro, and during the intradendritic and intrasomatic injection of complex patterns of current in vitro. We provide evidence that under all of these conditions, fast action potentials are preferentially initiated from the axonal initial segment and propagate reliably down the axon.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
1 h of incubation, slices were transferred to a submerged chamber in which both a bottom and top grid were used to lift and hold the slices in position and that contained the ACSF solution as in the preceding text. Cortical neurons were visualized with an upright infrared-differential interference contrast (IR-DIC) microscope (Zeiss Axioskop 2 FS plus). A light-sensitive camera (OLY-150, Olympus) was used for tracing the fluorescent axonal profiles. For only those recordings where the slow oscillation or epileptic activity was examined in the submerged chamber, the ACSF solution was modified to contain (in mM) 1 MgSO4, 1 CaCl2, and 3.5 KCl (Sanchez-Vives and McCormick 2000; Shu et al. 2003a,b). The membrane potentials in our whole cell recordings were not corrected for Donnan liquid junction potential, which is between 5 and 15 mV (Fricker et al. 1999).
Whole cell recordings were achieved simultaneously from either the soma and the apical dendrite or the soma and the cut end of the main axon using a Multiclamp 700B or Axoclamp 2B amplifier (Molecular Devices, Sunnyvale, CA). Patch pipettes were formed on a Sutter Instruments (Novato, CA) P-97 microelectrode puller from borosilicate glass (1B200-4, WPI, Sarasota, FL). Pipettes for somatic recording had an impedance of 5–6 M
and were filled with an intracellular solution that contained (in mM) 140 KGluconate, 3 KCl, 2 MgCl2, 2 Na2ATP, and 10 HEPES, pH 7.2 with KOH (288 mosM). Calcium buffer included in the whole cell pipette was 0.2 mM EGTA. Only regular spiking neurons were analyzed in this study.
Alexa Fluor 488 (100 µM) and biocytin (0.2%) were added to the pipette solution for tracing and labeling the recorded pyramidal cells. For simultaneous somatic and axonal (prefrontal cortex) or somatic and dendritic (typically in somatosensory cortex) whole cell recordings, 
10 min after somatic whole cell recording was established, the course of the main axon or apical dendrite was examined under the fluorescent microscope equipped with a x40 water-immersion objective and a magnifier of up to x2. Only those pyramidal neurons in which the apical dendrite or axon was preserved and came to the upper surface of the slice were used in this portion of our study. Patch pipettes for whole cell dendritic or axonal recording were filled with a similar intracellular solution but without fluorescent dye added; these had an impedance of 9–15 M. The pipette was advanced to the apical dendrite or the cut end of the axon with a positive pressure of 65 mbar and guided by switching back and forth between the fluorescent and DIC images with the total time the cell was exposed to fluorescence being kept to <20 s to minimize damage (our whole cell recordings from the soma did not reveal any evidence of changes in the electrophysiological properties of the recorded neurons during this brief exposure to fluorescence). The apical dendrite or the bleb formed at the cut end of the axon was then pressed by the pipette tip and negative pressure was applied to form a seal. As soon as a tight seal (>10 G) was achieved, pulses of brief suction were applied to break the patch, and the whole cell configuration could be easily obtained. Thereafter steps of positive and negative current were injected to the soma to examine the intrinsic membrane properties of the recorded neuron. During the whole period of simultaneous somatic and dendritic, or somatic and axonal recordings, access resistance was monitored frequently; recordings with access resistance >25 M for somatic recording, 80 M for axonal or dendritic recording, were discarded. Bridge balance and capacitance neutralization were carefully adjusted before and after every experimental protocol. After a recording was completed, the slice was transferred to 4% paraformaldehyde in 0.1 M phosphate buffer for subsequent immunostaining and visualization.
All computations were performed in Spike2 (Cambridge Electronic Design, Cambridge, UK) and MATLAB (MathWorks, Bethesda, MD). All error bars were calculated as the mean ± SE, and data in the text are reported as means ± SD.
A number of experiments were performed with a dynamic-clamp technique using a DAP-5216a board (Microstar Laboratory). Noisy conductances were constructed according to an Ornstein-Uhlenbeck (colored noise) model (Destexhe et al. 2001). The SD (and, concurrently, the amplitude) of the conductance was adjusted to give a 10-mV peak-to-peak membrane potential deviation, and the holding current was adjusted to keep the cell just near spike threshold.
The in vivo intracellular recordings analyzed here were from previously published studies performed in the primary visual cortex of halothane anesthetized cats (Nowak et al. 2005) or ketamine-xylazine-anesthetized ferrets (Haider et al. 2006). Analysis was restricted to those neurons classified as regular spiking, according to standard criteria (McCormick et al. 1985; Nowak et al. 2003).
Anatomical methods
Male ferrets, 7-wk or 15-mo old, were used for examination of the axon region of myelination, mainly in the ferret prefrontal cortex.
MYELIN STAINING.
BULK SILVER STAINING.
Because silver staining can make visible even the thinnest fibers of myelin in the cortex and it has been successfully used in various animal species, to visualize the overall pattern of myelination in the ferret brain, we used a modified version of the silver staining of myelin technique described by Gallyas (1979) sing 40-µm-thick, formalin fixed, vibratome cut brain sections mounted on gelatin subbed glass slides (Fig. 5A).
|
GOLGI STAINING.
For Golgi staining (Fig. 5B, supplemental Fig. 2), we used a commercially available kit that has combined and improved the Golgi-Cox and the classical rapid Golgi methods. This kit, known as the FD Rapid GolgiStain (FD NeuroTechnologies, Ellicott City, MD) is convenient to use because it is highly reliable and results in excellent staining of unmyelinated axons while preserving staining of somata and dendritic spines. Also, this method works well in older specimens, an important feature since a commonly cited drawback of Golgi impregnation is that it has been sometimes observed to work well only in young animals (Millhouse 1981).
|
Model methods
Our computational model (Fig. 6) was implemented usingNEURON 5.8 (Hines and Carnevale 1997) and was based on the multi-compartmental model of the full dendritic and somatic structure of a layer 5 cortical pyramidal cell (Fig. 1D in Mainen and Sejnowski 1996) coupled with a reconstruction of cortical pyramidal cell axons according to Binzegger et al. (2005). We used this model recently to examine the passive propagation properties of intracortical axons (Shu et al. 2006). The temperature of the modeled cell was 37°C.
|
|
AXON.
Our model of the axon began with that of Mainen et al. (1995) and was subsequently modified to include either a reduced axonal arbor that was of similar length and branching pattern as that of our biocytin-filled neurons in vitro or a more complete axonal arbor modeled after that of Binzegger et al. (2005). In addition, we modified membrane capacitance and ionic conductances to match the spike initiation properties of our layer 5 pyramidal cells (McCormick, Shu, and Yu, unpublished results).
MAIN AXON. The soma connects to the axon hillock, which tapers from 2.4 to 0.92 µm and has a length of 10 µm. The hillock is followed by a 40-µm initial segment of 0.92 µm diam. After the initial segment, there is a 300-µm length of unmyelinated axon (0.55 µm in diameter) based on our examinations on the main axons of prefrontal cortical neurons in 7-wk-old ferrets. For modeling the full axonal arbor, the next portion of the main axon is myelinated and 3,000 µm in length with internode distances of 200 µm. The diameter of the internodal axon is 0.92 µm. The diameter of axonal nodes is 0.46 µm, and their length is 1 µm.
AXON COLLATERALS OF FULL LAYER 5 PYRAMIDAL CELL.
All collaterals were assumed to be unmyelinated (Shu et al. 2006). There are eight first-order collaterals (diameter: 0.46 µm), 23 second-order collaterals (diameter: 0.37 µm), and 131 third-, fourth-, and fifth-order collaterals (diameter: 0.37 µm). The total length of collaterals is 64 mm. Including the main axon, the total axon length is 66.7 mm.
AXON COLLATERALS OF "SLICE" LAYER 5 PYRAMIDAL CELL. For the slice neuron simulation, the dendritic tree and soma are the same as in the simulated full neuron, but there are six first-order collaterals, 22 second-order collaterals, totaling 18.7 mm in length consisting of 35 sections composed of 3,293 compartments. With the main axon, the total axonal length for the slice neuron is 19.2 mm.
ELECTRICAL PROPERTIES.
The membrane capacitance Cm for the soma, dendrites, unmyelinated axon, and axon collaterals is modeled as 0.7 µF /cm2, whereas for myelinated axon it is 0.04 µF /cm2. The axial resistance for dendrite and axon is 100 m (Mainen and Sejnowski 1996). The membrane time constant of the soma, dendrite, hillock and initial segment is 21 ms, whereas for the main axon and collaterals the time constant ranges from 2 to 10.5 ms.
CHANNEL DISTRIBUTIONS. The transient Na+ current is present in all parts of the modeled cell and its density is high in the initial segment, hillock, and node (7,500 pS/µm2) but lower in the soma (750 pS/µm2) and dendrites (20 pS/µm2). In unmyelinated model axon and collaterals, Na+ conductance is between 750 and 2,500 pS/µm2 (the thinner the collateral, the lower the value of the Na+ conductance). The density of Na+ conductance was low (20 pS/µm2) in the myelinated portions of the axon. The reversal potential of Na+ in our model is 60 mV. The fast, voltage activated K+ current, IKv is present in the model hillock, initial segment, and node (1,000 pS/µm2). The density is lower for soma (400 pS/µm2), unmyelinated model axon (500 pS/µm2), and axon collaterals (200 pS/µm2 for 1st-order collaterals, 100 pS/µm2 for higher-order collaterals). The reversal potential of K+ is –90 mV.
The slow noninactivating potassium current (M-current; Ikm), high-voltage activated Ca2+ current, ICa and one Ca2+-dependent K+ current, IkCa, are distributed throughout the soma and dendrites (Km conductance 0.1, Ca conductance 0.3, and Kca conductance is 3 pS/µm2) as per Mainen et al. (Mainen and Sejnowski 1996). The reversal potential of Ca is 140 mV. Background leak current is distributed throughout the cell. In the soma, dendrites, hillock, initial segment, and myelinated axon, gleak = 0.0000333 S/cm2. In the unmyelinated axon and collaterals, gleak = 0.00006667 S/cm2. In the nodes, gleak = 0.02 S/cm2. The value of the leak conductance is modified to give time constants that are similar to what we observed in experiments. The reversal potential of the leak current is –70 mV. For the specific rate functions for the different ionic channels, see Mainen and Sejnowski (1996).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
) spherical 3–6 µm bleb owing to being cut during the slice procedure (Fig. 1D). Here we perform whole cell recordings from the axon, apical dendrite, and soma to examine the properties of action-potential generation in layer 5 cortical regular spiking pyramidal cells in the prefrontal cortex in response to current injections and during spontaneous synaptic activity of the up state (Shu et al. 2003b, 2006) as well as during epileptiform discharges. We then compare our results with action potentials recorded in vivo during up states and in response to visual stimulation.
Whole cell recordings were obtained simultaneously from either the cell body and apical dendrite (n = 7) or cell body and axon (n = 16) of cortical pyramidal cells during spontaneous up and down states in ferret prefrontal or somatosensory cortical slices maintained in vitro (Fig. 1). These up states are characterized by 5- to 15-mV amplitude, 0.25- to 1-s duration barrages of synaptic activity containing approximately an equal mixture of excitatory and inhibitory potentials (Fig. 1, B and E) (Haider et al. 2006; Shu et al. 2003b). This synaptic activity arises from the discharge of neighboring neurons and results in the activation of action potentials in the soma at the rate of 5–31 Hz (10-90th percentiles) with an average discharge rate of 11.2 ± 3.2 (SD) Hz (n = 7; supplementary Fig. 1A1 ).
Simultaneous recordings from pyramidal somata and apical dendrites (50–580 µm from the soma) during the generation of spontaneous up states revealed that action potentials were generated in both the cell body and apical dendrite. Examination of the timing of these action potentials revealed that all action potentials were generated in the soma prior to the apical dendrite (n = 637 spikes in 7 neurons; Fig. 1C) even at all frequencies of action-potential generation from 5 to 75 Hz. The dendritic action potentials decreased in amplitude and broadened in duration with increases in distance from the cell body (not shown). Consistent with previous findings, these recordings revealed that action potentials propagate from the soma up the apical dendrite at a speed of 0.66 ± 0.18 m/s (n = 6) (Larkum et al. 2001; Stuart et al. 1997a).
Simultaneous recordings from the axon and soma of pyramidal cells during the spontaneous generation of up states revealed a spike timing relationship that varied with the distance of the axonal recording site from the soma (n = 16). Recordings in which action potentials arrived at the axonal recording site prior to the soma were obtained at axonal distances of between 40 and 157 µm (Fig. 1, D–G), whereas axonal recordings that exhibited spikes occurring after those in the soma were recorded at distances of 111–311 µm (see following text; Fig. 1G). Plotting the time difference between the axonal and somatic spike times versus location of the axonal recording site revealed a relationship that fell to zero difference at a distance of 80–110 µm (Fig. 1G). If the action potentials were to propagate orthodromically and antidromically from their site of origin with equal speed, then spikes that arrive at the same time at the axonal and somatic recording sites are likely to have initiated at a point halfway between the two. Therefore these results suggest that during spontaneous up states, action potentials are initiated in the axon somewhere between 40 and 55 µm from the soma. Our computational model suggests, however, that action potentials propagate orthodromically away from their site of origin at an average speed of 0.83 m/s, whereas they propagate antidromically toward the cell body at an average speed of 0.7 m/s (see Fig. 6). These simulations suggest that our present method of estimation of the site of spike initiation should be compensated by 10 µm toward the soma.
Action potentials initiated by artificial up states
To control the precise location of the arrival of the currents initiating action potentials, we used the dynamic-clamp technique to inject artificial barrages of synaptic activity in the apical dendrite or soma while performing dual dendritic-somatic or somatic-axonal recordings (n = 10; Fig. 2). As with naturally occurring barrages of synaptic activity, these artificial barrages resulted in action potentials that occurred first in the soma followed by propagation into the apical dendrite (Fig. 2A), at all frequencies of action-potential generation from 2 to 34 Hz (supplementary Fig. 1B), for both somatic and apical dendritic injections. The injection of artificial up states into the soma resulted in action potentials, the timing difference of which between the soma and axon (Fig. 2, C and D) exhibited a zero difference at 110 µm (Fig. 1G), indicating that these spikes are initiated at 55 µm from the cell body, in the axon initial segment. Similar results were obtained with action potentials generated in response to current pulses and during epileptiform activity (see Fig. 1G and following text).
Responses to current pulse injection
By initiating spikes with the intracellular injection of short-duration current pulses at either the somatic or axonal recording sites and examining the timing relationship of these action potentials versus distance between the recording locations, we were also able to calculate the apparent action-potential initiation site using simple algebra (Fig. 3). Our calculation revealed that in response to somatic current pulses action potentials initiate 46 ± 5.6 µm (n = 16 cells) down the axon from the edge of the soma. This location, however, should be compensated for possible differences in antidromic versus orthodromic conduction velocities in our experiments (see text above).
|
5 nA), action-potential generation occasionally failed in either the axonal or somatic recording site (Fig. 4D). In addition, action potentials were observed to fractionate into components, which were especially pronounced when dV/dt of the action potential was examined (Fig. 4F). There appeared to be at least three components to these action potentials. The smallest component, as recorded from the soma or distal axon, preceded spike generation in both the distal axonal and somatic recording sites (Fig. 4F). Because of this timing, and the presence of this smaller spike even when action potentials in the soma and distal axonal recording sites failed (Fig. 4F, *), we presume that this component represents the generation of an action potential in the axon initial segment. This IS spike was followed, in the soma, by the initiation of a somatodendritic (SD) spike. Similarly, in the distal axon (DA), local action-potential generation was also preceded by the IS spike (Fig. 4F).
|
; Waxman 1995). As mentioned in the preceding text, prolonged depolarization of the proximal axon (22–97 µm from the soma; n = 6) resulted in repetitive spike discharge, similar to that seen with somatic depolarization (not shown). Thus the soma, proximal axon, and distal axon all exhibited distinct electrophysiological features. Properties of myelination of axons in ferret prefrontal cortex
Previous reports have compared the site of spike initiation with the pattern of myelination to demonstrate that spikes are initiated either in the axon initial segment or the first node of Ranvier (Clark et al. 2005; Palmer and Stuart 2006). Here we also compared the pattern of myelination in layer 5 pyramidal neurons in the ferret prefrontal cortex with our calculation of the spike initiation point. The initial portions of the principal axon of cortical pyramidal cells contains four components that we will consider here: the axon hillock, which is characterized by the initial tapered component of the axon at its origin (typically the cell body but on occasion a dendrite); the axon initial segment; the myelinated segment of the axon; and the first node after myelination. To determine these components in layer 5 cortical pyramidal cells of the ferret prefrontal cortex, we examined the myelination of these axons through both Golgi-stained material (Fig. 5B) as well as immunofluorescence staining against myelin on neurons that were also filled with biocytin and Alexa-488 (Fig. 5, C and D; supplementary Fig. 2).
First, we examined the overall pattern of myelination of the 7-wk-old ferret prefrontal cortex through silver-stain visualization of myelin (Fig. 5A; METHODS). This method revealed dark staining only within the white matter, with relatively little staining in the medial or lateral banks of the prefrontal cortex (Fig. 5A; n = 6 ferrets). In contrast, silver-stain based visualization of myelination in the primary visual cortex from the same animals revealed typical bands of myelin in both the white matter as well as within layers 4 and 5 (not shown).
To examine the issue of myelination of layer 5 pyramidal cell axons in the ferret prefrontal cortex further, we examined Golgi-stained material (see METHODS). Golgi staining reveals axons only if they are unmyelinated (Fairen et al. 1977). Therefore the point at which axons emanating from the cell body lose Golgi staining should indicate the approximate beginning of myelination. Examination of Golgi stained sections revealed that in the 7-wk-old ferret prefrontal cortex the principal axon extends for a long distance (average of 212 ± 118 µm; n = 38) from the soma before staining is lost. Examination of Golgi-stained material from the prefrontal cortex of older ferrets (15 mo) revealed that these also appear to be unmyelinated for a long distance (138 ± 103 µm; n = 46) although this distance is less than in 7-wk-old animals. The long length of axon that is visualizable with Golgi staining in the ferret prefrontal cortex is in contrast to that previously reported for many neocortical pyramidal cells (Farinas and DeFelipe 1991). Indeed, when we examined Golgi-stained layer 5 pyramidal cells in the ferret primary visual cortex, the axons became unlabeled at a much shorter distance (36 ± 19 µm in 7 wk old; n = 30; 54 ± 46 µm in 15 mo old; n = 30; supplementary Fig. 2).
To examine the location of myelination more directly, we performed immunofluorescent staining of myelin basic protein (MBP) on prefrontal cortical layer 5 neurons that had been previously filled with biocytin and a fluorescent dye (Alexa 488; 7-wk-old ferrets). Of 15 neurons examined, the axons of 12 were cut at the surface of the slice at an average distance of 289 ± 144 µm before myelination became evident. The axons of the three remaining cells became myelinated at an average distance of 453 ± 87 µm, corresponding roughly with the location of their entry into the white matter (Fig. 5, C and D). These results indicate that in layer 5 pyramidal neurons of the ferret prefrontal cortex, action potentials are initiated in the initial, unmyelinated portion of the axon and that this location does not correspond to the beginning of myelination as it does in some preparations (Palmer and Stuart 2006).
Location of first axon collateral
The first axon collateral is often associated with the first node of Ranvier (Fraher and Kaar 1984; Sloper and Powell 1979), which has been suggested to be the location of action-potential generation in some cell types (Clark et al. 2005; Colbert and Johnston 1996; Coombs et al. 1957). Here we examined the distance from the cell body to the first axon collateral in Golgi-stained sections of ferret prefrontal cortex. Of 38 cells examined in 7-wk-old ferrets, 26 exhibited at least one axon collateral, and the average distance to the first collateral from the soma was 104.3 ± 32.1 µm. This distance is similar to that obtained by examining the distance to the first axon collateral in biocytin/Alexa-488-filled pyramidal cells (112.5 ± 25.7; n = 14). In adult (15 mo old) ferrets, the distance to the first axon collateral in Golgi-stained sections was similar (105 ± 26 µm; n = 10). Thus the calculated location of action-potential initiation is significantly proximal to the location of the first axon collaterals in our layer 5 pyramidal neurons.
Computational model of action-potential initiation and propagation
One possible modifying factor in calculating the location of spike initiation is the assumption that orthodromic and antidromic spikes propagate at the same velocity from their site of origin. We examined this using computational models of either a restricted, "slice-like" axonal arbor, or a complete arbor (Mainen and Sejnowski 1996; Mainen et al. 1995; Shu et al. 2006). In both the partial and complete axonal arbor models, spikes initiated at 38 µm from the root of the axon in response to somatic current injection (Fig. 6). Increasing the axonal density of gNa+, starting with levels equal to that of the soma, moved the action-potential initiation point along the axon away from the soma. This progression down the axon exhibited a plateau of 40 µm (not shown). In our model of axonal action-potential generation, the spikes were found to propagate orthodromically down the axon and antidromically toward the cell body at different rates (Fig. 6C). In simulations in which the main axon terminated at lengths similar to those in our actual recordings, we calculated that this difference in orthodromic and antidromic propagation velocity will cause the calculated point of initiation in real recordings to be on average 10 µm farther down the axon than the actual initiation point. Including this correction factor results in an estimated initiation point of between 30 and 45 µm from the soma in our recordings from real neurons.
Using current pulses to initiate action potentials at the cut end of the axon can also result in a miscalculation of the point of initiation of action potentials. This is because the spikes recorded at, for example, 300 µm from the soma propagate at different velocities according to whether they are initiated locally and propagate antidromically or initiated at the axon initial segment and propagate orthodromically (Fig. 6C). Antidromic spikes propagate more quickly at their site of initiation (300 µm from the soma in this model), owing to the depolarization of the axon by the current pulse during antidromic activation. In contrast, initiation of spikes with somatic current pulse did not result in such a large effect at the site of spike initiation (axon initial segment). The differences in average propagation velocity suggest that our algebraic calculations from the data obtained in real neurons (Fig. 3) may estimate the location of spike initiation to be distal to the actual site of spike initiation by 8 µm, resulting in a corrected location of 38 µm from the soma.
Spike threshold varies with distance from the soma
To examine the possibility that the axon initial segment has a lower threshold for action-potential initiation, we plotted the membrane potential at which spikes were initiated with intra-axonal current pulses versus distance along the axon from the soma (Fig. 7). This plot revealed a relationship in which the threshold increased with distance down the axon, after a distance of 100 µm (Fig. 7; n = 20 axonal recordings). Plotting the best linear fit for all points distal to 50 µm and the best linear fit for points 
75 µm revealed in intersection at 60 µm. Although there is considerable variation in this plot of spike threshold (each point represents a different cell), the data suggest that the axonal region up to 100 µm from the soma has a lower spike threshold than the more distal axon.
|
Recently we have demonstrated that somatic recordings of action potentials that originate in the axon reveal an unusually rapid rate of rise at the foot of the spike (McCormick et al. 2006), giving rise to the so-called action-potential "kink" (Naundorf et al. 2006). Here we examined whether or not this was also true for action potentials that were initiated in the apical dendrite of layer 5 pyramidal cells (Fig. 8). Through the intracellular injection of large (0.8–2 nA; 100–500 ms) current pulses into the proximal to middle (150–420 µm) portions of the apical dendrite, we were able to cause the dendritic initiation of fast action potentials that then propagated to the soma. Examination of these somatic spikes revealed a relatively slow, or smooth, transition at the initiation point of the spike (Fig. 8A; n = 105 spikes in 4 cells). In contrast, examination of somatic action potentials generated in response to somatic current injection and that were known, from dual somatic/axonal whole cell recordings, to have originated in the axon exhibited a prominent kink at spike initiation (Fig. 8B; n = 2671 spikes in 13 cells). Plotting the slope of the phase plot at spike initiation for action potentials known to be initiated in the axon or the apical dendrite revealed very distinct distributions (Fig. 8E). Next we compared the distributions of the slopes at spike initiation of action potentials occurring in vivo during either up states in the prefrontal cortex of the ferret (Haider et al. 2006) or during visual stimulation in the cat primary visual cortex (Fig. 8E) (Nowak et al. 2005). Interestingly, both spikes initiated during up states in the ferret prefrontal cortex (n = 2364 spikes in 11 cells) and spikes initiated by visual stimuli in the cat primary visual cortex exhibited a prominent kink at spike onset (Fig. 8C; n = 4,298 spikes in 10 cells). This result was quantified by measuring the slope of the phase plot of spike dV/dt versus membrane potential (Naundorf et al. 2006; McCormick et al. 2006) (Fig. 8D). The distributions of the slope of the phase plot at spike initiation for in vivo spikes during up states as well as visually evoked spikes were very similar to that for in vitro somatic recordings of axonally initiated spikes in layer 5 pyramidal cells (Fig. 8E). In contrast, the distributions of in vivo spike phase plot slopes were completely different from those for spikes known to have originated in the apical dendrite in vitro (Fig. 8E). These results indicate that spikes occurring in cortical neurons in vivo, either during spontaneous up states in anesthetized ferrets or during visual stimulation in anesthetized cats, are primarily initiated in the axon.
|
Previous studies have suggested that dendritic spike initiation may occur when the dendrites are strongly and/or synchronously depolarized (Golding and Spruston 1998; Kloosterman et al. 2001; Stuart et al. 1997b; Turner et al. 1991). In addition, the pattern and properties of spike generation in axons may be very important in the generation of epileptiform activity (Meeks et al. 2005; Noebels and Prince 1978). Here we examined the point of initiation of fast action potentials during the generation of epileptiform activity in prefrontal cortical slices. After the bath application of 50 µM picrotoxin, and in the presence of 1 mM Mg2+ and 1 mM Ca2+ (Sanchez-Vives and McCormick 2000; Shu et al. 2003a,3b), cortical slices generated prolonged (0.4–8 s) epileptiform bursts consisting of an initial large depolarization and inactivation of somatic spikes, followed by repetitive bursts of action potentials (Fig. 9). This epileptiform activity is known to be generated through recurrent excitatory networks within the cortical slice and is associated with very strong barrages of excitatory and inhibitory (although GABAA receptors are blocked here) synaptic activity (reviewed in McCormick and Contreras 2001).
|
45 µm from the soma of these epileptiform spikes (Fig. 1G). Simultaneous whole cell recording from the soma and proximal to distal parts of the axon (60–300 µm) during epileptiform activity demonstrated that these portions of the axon followed action-potential generation in the soma on a nearly one to one fashion with failures of spike initiation and conduction being rare (Figs. 9 and 10; n = 6). The large depolarization (paroxysmal depolarization shift, PDS) associated with epileptiform bursts was also present in the axon, although the amplitude of this decreased with distance from the soma (Figs. 9 and 10). The large depolarizations, particularly the first PDS, associated with epileptiform activity could strongly reduce the amplitude of spikes generated in the soma and, to some degree, in the axon initial segment (Fig. 10). However, the number of action potentials generated in the proximal and distal axon during the initial PDS was often greater than in the soma (Figs. 9, C and D, and 10B).
|
); this is in direct contrast to our results here. To test the possibility that this discrepancy may result from differences in the recording techniques, we recorded extracellularly (as per Meeks et al. 2005) from the same axons that we had previously performed whole cell recordings and compared the results. We found that axonal action potentials recorded extracellularly could suffer from a significant signal to noise problem. Many of the spikes that were previously observed with axonal whole cell recording disappeared into the background noise of the recording when recorded extracellularly (cf. Fig. 9, C and D). |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
) indicate that this ongoing synaptic activity can propagate long distances (>0.5 mm) along these processes. Which compartment then will initiate the fast action potentials that propagate down the axon in response to this synaptic bombardment?
Prior in vitro studies suggest that fast action-potential initiation will most often occur in the axon in pyramidal cells, followed by back propagation into the somatic and dendritic compartments (Stuart et al. 1997a,b). However, exactly which compartment of the axon will initiate the action potential has been an open question, with studies favoring both the axon initial segment (Palmer and Stuart 2006; Stuart et al. 1997a) as well as the first node of Ranvier (Clark et al. 2005; Colbert and Johnston 1996; Colbert and Pan 2002). A recent imaging study using voltage-sensitive dye indicates that action potentials initiated in response to current pulses and single excitatory postsynaptic potentials (EPSPs) evoked with local electrical stimulation occur first 35 µm down the main axon from the soma (Palmer and Stuart 2006). To put this in perspective, it is worth reviewing the structure of cortical axons. Axons from pyramidal neurons typically arise from the cell body, although they can also emanate from a basal dendrite (Palay et al. 1968). The transition between the axon and the soma/dendrite from which it originates is associated with a conical, 5 µm long, transition segment, the axon hillock. The axon initial segment follows the axon hillock. It is typically a 17- to 40-µm long segment of axon that on the electron microscopic level contains a layer of dense granular material underneath the membrane, microtubules in the cytoplasm, and scattered clusters of ribosomes (Farinas and DeFelipe 1991; Inda et al. 2006; Palay et al. 1968). Immunocytochemical localization of Na+ channels, K+ channels, and cytoskeleton and adhesion molecules spectrin 
IV and Caspr2 reveal a high density in the axon initial segment (as well as at nodes of Ranvier) (Inda et al. 2006). Often, but not always, the end of the axon initial segment (as defined by the loss of the submembranous structures typical of the initial segment on the EM level) is associated with the beginning of myelination (Farinas and DeFelipe 1991; Palay et al. 1968). In the layer 5 neurons studied here, the main axons were unmyelinated for approximately 200 µm (see Fig. 5, supplementary Fig. 2). Our results indicate that action potentials typically initiate in a region that is 35–50 µm from the soma, long before the axons are myelinated in these cells. Although Palmer and Stuart (2006) have suggested that the action-potential initiation site corresponds to the beginning of myelination, these authors also observed a proximal initiation site at early postnatal ages when axons are unmyelinated. Together, these and our results suggest that the region of myelination by itself does not determine the site of spike initiation. Immunocytochemical studies suggest that the axon initial segment may contain a high density of Na+ and K+ channels, causing it to have a low threshold for action-potential initiation (Inda et al. 2006; Komada and Soriano 2002). Our high-resolution recordings of action potentials in the axon initial segment reveal a very high rate of rise for these spikes, indicating a relatively high local density of Na+ channels (McCormick et al. 2006). In contrast, direct recordings of Na+ and K+ channels from the initial segment suggest an alternative interpretation. The Na+ channels at the axon initial segment may have a more negative activation curve than those in the soma (Colbert and Pan 2002). In addition to the presence of a high density of Na+ channels, and/or the presence of Na+ with low threshold, the electrical characteristics of the axon initial segment are conducive to the initiation of action potentials. The large capacitive and conductance load of the soma and dendrites decreases with increased distance down the axon, thereby decreasing the amount of current required to initiate an action potential. As synaptic activity arriving in the dendrites and soma of the neuron propagates down the axon (Alle and Geiger 2006; Shu et al. 2006), the axon initial segment forms the location that is most favorable to action-potential initiation. This region of the cell is strongly affected by the summated membrane potential arriving from the various dendritic branches and soma (Shu et al. 2006) and its electrical properties (e.g., channel densities and properties, electrical isolation, etc.) facilitate the production of action potentials.
Where do action potentials initiate in vivo? This is a difficult question to answer because simultaneous recordings with high temporal resolution from different portions of pyramidal cells, which also include the axon, in vivo are not yet possible. Prior extracellular recording studies indicate that action potentials typically are back propagating from the soma into the apical dendrites (Buzsaki and Kandel 1998), indicating that the spikes initiated somewhere within the soma, basal dendrites, or axon compartments. One property of action potentials offers an interesting possibility to decipher the region of origin of action potentials in vivo. Recently it was noted that spikes occurring in cortical neurons exhibit a prominent and characteristic rapid rate of rise at their initiation, giving rise to a so-called "kink" (Naundorf et al. 2006) (Fig. 8). Detailed recordings and simulations of cortical pyramidal cells reveal that this kink in the action potential arises from spike initiation in the axon (McCormick et al. 2006), owing in part to the high density of Na+ channels apparently present in the axon initial segment (Inda et al. 2006; Komada and Soriano 2002). Our recordings reveal that spikes initiated in the apical dendrite do not share this property of axonally initiated action potentials (see Fig. 8). When we examined the properties of spikes occurring in vivo, during either spontaneous up states in anesthetized ferrets or visual stimulation in anesthetized cats, we found a distribution of rate of rise at the foot of the spike that was very similar to spikes initiated in the axon in layer 5 pyramidal neurons in vitro. The strong overlap in these distributions suggests that spikes that we recorded in vivo were preferentially initiated in the axon. The small difference in spike slope distributions (see Fig. 8) between in vivo and in vitro suggests that a small percentage of spikes in vivo may have originated in the dendrites. However, it is also possible that these differences arise from comparing the spikes of different cell types (e.g., layer 5 pyramids in vitro versus multiple cell types in vivo) or differences in important physiological parameters between in vivo and in vitro (e.g., modulators, ion concentrations, anesthetics
) or with a current pulse delivered at 300 µm down the axon (
). Depolarization of the end of the axon results in rapid propagation of spikes in this region, and we calculate that differences in the orthodromic and antidromic propagation speeds result in an 8-µm mismatch between the estimated and actual location of action-potential initiation.
