INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

During visually evoked gamma (30-70 Hz) oscillations in cats, recorded extracellularly, single neurons were observed to burst at gamma frequency, with high-intraburst firing frequency (Gray 1994). Such a firing pattern, variously called "fast rhythmic bursting" (FRB) or "chattering," was demonstrated in intracellularly recorded cells in superficial cortical layers of in vivo cat cortex, some of which were shown to be spiny (hence presumably excitatory) neurons (Gray and McCormick1996). These latter chattering cells responded to both visual stimulation and sufficiently large depolarizing pulses, with fast rhythmic bursts. In cat cortex in vivo, deep thalamic-projecting neurons, and also aspiny (presumed inhibitory) neurons, could also exhibit fast rhythmic bursting during depolarizing current pulses of appropriate amplitude (Steriade et al. 1998); intralaminar thalamocortical neurons in vivo can also discharge in this pattern (Steriade et al. 1993). FRB occurs intermittently in so-called stuttering cells, which are GABAergic (Gupta et al. 2000). Steriade et al. (1998) further noted high-frequency runs of presumed excitatory postsynaptic potentials (EPSPs) in cells capable of FRB, as if some FRB neurons were presynaptic to other FRB neurons. An obvious question is then: to what extent do FRB neurons contribute to the generation of (as opposed to just participation in) gamma oscillations? An understanding of the intrinsic mechanisms of FRB could be helpful (but clearly is not sufficient) in answering this question.

Rhythmic bursting, often in the gamma range, has also been observed in neocortical slices from adult ferrets (Brumberg et al. 2000) and (at somewhat lower frequencies) from cats (Nishimura et al. 2001). Interesting observations of Brumberg et al. (2000) included the following. 1) FRB could be evoked by metabotropic glutamate receptor activation. [This is interesting because metabotropic glutamate receptors are critically involved in gamma oscillations evoked by electrical stimulation; additionally, activation of metabotropic glutamate receptors, in the CA1 hippocampal region and superficial neocortex in vitro, can evoke gamma oscillations (Gillies et al. 2002; Whittington et al. 1995, 1997). These receptors might also play a role in sensory-activated gamma in vivo]. 2) FRB can also be evoked by current pulses, as in vivo. 3) FRB seems to depend on generation of a fast spike afterdepolarization (ADP). 4) FRB persists in 0 [Ca²⁺], 2 mM [Mn²⁺] media. 5) FRB is suppressed by reduction of Na⁺ conductances. 6) in contrast, the sea anemone toxin ATX II, which appears to enhance persistent Na⁺ conductance [at least in soma/dendritic membrane (Brand et al. 2000; Mantegazza et al. 1998)], favors FRB. Brumberg et al. (2000) suggested that the spike ADP could be enhanced by persistent Na⁺ conductance, thereby promoting FRB; this idea had previously been incorporated into a two-compartment model of Wang (1999), in which persistent g_Na was located in the dendritic compartment and a fast spike-generating mechanism in the somatic compartment. Interestingly, Brumberg et al. (2000) noted that when rhythmic spiking was converted to fast rhythmic bursting, by prolonged intracellular current injection, the width of action potentials increased, and the maximal rate of fall of the action potentials decreasedas if a K⁺ current might have been reduced; these authors did not, however, explore which K⁺ current could be involved.

The in vitro study of Nishimura et al. (2001) likewise showed that the spike ADP in layer III sensorimotor cortical neurons is not blocked by g_Ca reduction, that the ADP is enhanced by intracellular calcium chelation, and that the ADP is Na⁺-dependent. Furthermore, replacing extracellular Ca²⁺ with Mn²⁺ could convert a rhythmic firing pattern to an FRB-like pattern. Friedman and Gutnick (1989) had previously shown that intracellular 1,2-bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid (BAPTA), in neocortical neurons, enhanced spike ADPs and could induce bursting that was sometimes rhythmic.

Here, we build on these previous results. We construct a detailed model of a layer 2/3 pyramidal neuron that, as a form of calibration, replicates recordings of dendritic fast and slow Ca²⁺-dependent spikes, under conditions of steady dendritic depolarization. It was discovered (serendipitously) that the model would generate fast rhythmic bursting under either of two conditions, which could occur exclusively or in combination: 1) when persistent Na⁺ conductance is increased (as shown previously by others, as noted above); or 2) when a fast voltage- and [Ca²⁺]-dependent conductance, g_K(C), is reduced. The model also suggests that the spike ADP can result in part from decremental antidromic conduction of an axonal spike, suggesting another means by which a Na⁺ conductance could influence FRB, without necessarily involving a persistent Na⁺ conductance. Experiments in rat neocortical slices show that a number of manipulations can convert rhythmic spiking to FRB (although at frequencies below 30 Hz, in our experimental conditions); such manipulations include the previously known (from other preparations) buffering of intracellular [Ca²⁺], and enhancing persistent Na⁺ conductance. In addition, FRB could be evoked by blocking BK channels (that are presumed to mediate g_K(C)), even independently of any enhancement of persistent Na⁺ conductance. The data suggest that at least some of the spikes in each burst could originate in the axon distal to the initial segment. We further suggest that both Na⁺-dependent axonal excitability and also reduction of g_K(C) are important in FRB.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

Simulation methods

The formal approach to simulating the layer 2/3 cortical pyramidal neuron was similar to that used to simulate a CA3 hippocampal pyramidal cell (Traub et al. 1994), but most of the detailsin cell geometry and passive and active parameterswere different. In this study, voltages "V " are transmembrane potentials, in contradistinction to earlier studies (e.g., Traub and Miles 1995; Traub et al. 1994) in which "V " signified "voltage relative to resting potential."

OVERALL MODEL STRUCTURE. Model structure was qualitatively based on a drawing in Major (1992), his Fig. 3.14, but with a limited number of compartments and much greater symmetry than in the real neuron. The model (Fig. 1) has eight equivalent basal dendrites, four equivalent apical obliques, an apical shaft, and equivalent superficial apical branches. There are 68 soma-dendritic compartments and 6 axonal compartments. The most proximal basal and oblique compartments, along with the two perisomatic apical shaft compartments, are called "proximal dendrites"; the outer 24 superficial apical compartments are called "distal dendrites."

View larger version (16K): [in this window] [in a new window]   Fig. 1. Compartmental structure of model layer 2/3 pyramidal neuron. There are 68 soma-dendritic compartments, with 8 basal dendrites and 4 apical oblique dendrites. The axon contains 6 compartments.

GEOMETRICAL PARAMETERS. The soma was a cylinder with a length of 15 µm and radius of 8 µm. All dendritic compartments had a length of 50 µm. The radius of basal and oblique dendrites was 0.5 µm. The radius of the apical shaft was 4 µm, tapering to 2 µm, while distal apical branches had a radius of 0.8 µm. The surface area of each dendritic compartment was taken to be 4rl (r = radius, l = length), rather than 2rl, to allow for the contribution of spines to area. Total surface area of the soma/dendrites was 35,940 µm². The most proximal axonal compartment had a length of 25 µm and a radius of 0.9 µm. The other axonal compartments had a length of 50 µm, and a radius starting at 0.7 µm, tapering to 0.5 µm.

PASSIVE PARAMETERS. Membrane capacitance C_m was 0.9 µF/cm²; membrane resistivity R_m was 50,000 -cm² for soma-dendrites and 1,000 -cm² for the axon, and internal resistivity R_i was 250 -cm for soma-dendrites and 100 -cm for the axon. R_input measured at the soma, with all active currents blocked, was 69.4 M.

DYNAMICS OF VOLTAGE AND [CA²⁺] BEHAVIOR. The equations describing dynamics are standard and are here summarized briefly. As units, we shall use mV, ms, nF, µS, nA. First, the discrete version of the cable equationan approximation to the original partial differential cable equationdescribes the evolution of voltage in each compartment k

(1)

In Eq. 1, C_k is the capacitance of compartment k and V_k the transmembrane voltage. The sum is taken over all compartments m that are connected to compartment k. _m,k is the conductance (internal) between the respective compartments (with the assumption that the extracellular space is isopotential). I_ionic,k is the transmembrane ionic current for compartment k: one must be careful about the sign of this term, as inward currents (which depolarize the membrane) are, by convention, negative. For the membrane potential to increase during an inward current, we need the minus sign before I_ionic,k.

(2)

View larger version (24K): [in this window] [in a new window]   Fig. 2. Simulation of dendritic Na+ and Ca2+ spikes in response to dendritic current pulse. Response of model neuron to depolarizing current pulses, injected into apical dendrite (compartment just proximal to the terminal branches); there is a 50-M (20 nS) shunt at the injection site. A 1.5-nA pulse (left) induces a large calcium spike in the dendrite, generating a burst of fast spikes in the soma; this is followed by a series of somatic spikes that backpropagate (as seen on expanded traces, not shown) to the dendritic site. A 2.5-nA pulse (right) induces short trains of small fast dendritic spikes (backpropagated from full somatic spikes), interspersed with broader calcium spikes, possessing superimposed small fast spikes (inset). Persistent gNa blocked (DNa(P) = 0); DK(C) = 1.6.

(3)

(4)

1

(5)

MAIN PARAMETERS VARIED, AND THEIR NOTATION. Preliminary simulations were run, with stimulation of the model neuron via current pulses or antidromic stimulation (using a 0.4-nA, 0.8-ms depolarizing pulse delivered to a distal axonal compartment). Current pulses were also delivered to soma and to various compartments in the dendrites, sometimes in the presence of an additional leak conductance at the stimulated site. Minor adjustments to the rate functions were tested, but mainly we wished to find values of the conductance densities that would produce somatic action potentials capable of "back-propagating" to the dendrites and also that would produce dendritic calcium spikes. Needless to say, the resulting distribution of conductance densities is not unique. The conductance densities used in this paper are given in the APPENDIX. Note that the density of g_Na(F) on the soma and proximal dendrites in this model is much higher than its density in the remaining parts of the dendrites, a situation in contrast to some models (e.g., Mainen et al. 1995); with a maximum density of 400 mS/cm² for g_Na(F) in the axon (enough to make our model axon extremely excitable), increased perisomatic g_Na(F) densities were required for a somatic spike to develop.

INTEGRATION METHOD. A second-order Taylor series (explicit) method was used, as in previous publications (e.g., Traub et al. 1994). The integration time step, dt, was 0.004 ms.

COMPUTING ENGINE. Programs were written in FORTRAN, and simulations were either run on a single "wide node" (equivalent to a UNIX workstation) of an IBM SP2 parallel computer or else 12 different simulations (each with a distinct value of some parameter) were run simultaneously on the 12 nodes of the parallel machine.

Experimental methods

Slices (450-µm-thick) of temporal cortex were obtained from artificial cerebrospinal fluid (ACSF) perfused brains of adult male Wistar rats, anesthetized with isoflurane followed by ketamine/xylazine injection. The animal was perfused once all pain reflexes had disappeared. Slices were maintained in an interface chamber and perfused with artificial cerebrospinal fluid containing the following (in mM): 133 NaCl, 18 NaHCO₃, 3 KCl, 1.25 NaH₂PO₄, 10 D-glucose, 1 MgCl₂, 1.7 CaCl₂, equilibrated with 95% O₂-5% CO₂, pH 7.2 at 35°C. Sharp electrode recordings were taken from layer 2/3 pyramidal neurons at the level of soma or apical dendrite.

SHARP ELECTRODE RECORDINGS. Membrane potential recordings were obtained from 36 layer 2/3 pyramidal cell somata, 4 fast-spiking (presumed interneuronal) neurons, and 12 presumed dendrites (identified by their characteristic attenuated fast spikes, lack of fast spike AHP, and easy induction of broad bursts of spikes on depolarization). Microelectrodes (resistance 30-90 M) were filled with 1.5 M potassium acetate or potassium methysulfate alone or in conjunction with 0.3 mM of the calcium-chelating agent BAPTA (Sigma, UK). Somatic recordings were examined for evidence of rhythmic bursting activity after long (>10 s) periods of tonic depolarizing current injection (0.5-1.0 nA) (see Brumberg et al. 2000); fast rhythmic bursting was not observed without this depolarization, during experiments with BAPTA. With BAPTA-filled sharp electrode recordings, somatic resting membrane potential (RMP) was 68 ± 4 mV, and somatic input resistance was 45 ± 5 M. For presumed dendrites, RMP was 60 ± 8 mV and input resistance was 62 ± 4 M.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

NA⁺ AND CA²⁺ ELECTROGENESIS IN RESPONSE TO DENDRITIC DEPOLARIZATION. Injections of depolarizing current pulses into an apical dendrite of the model evoke either trains of small, fast Na⁺ action potentials (Fig. 2, left) or else small, fast action potentials that are intermixed with (and superimposed on) slower high-threshold Ca²⁺-mediated action potentials (Fig. 2, right). Both of these patterns resemble patterns reported previously in experimental dendritic recordings of layer 2/3 neocortical pyramidal neurons (Amitai et al. 1993) (and also other types of pyramidal cell, e.g., Kamondi et al. 1989). In our model, all of the fast spikesthose that are between and also those that are superimposed on the slow spikesare initiated perisomatically and backpropagate into the dendrites; this is the case both for tonic intrasomatic and for tonic intradendritic current injections (data not shown).

View larger version (29K): [in this window] [in a new window]   Fig. 3. Experimental recording of dendritic Na+ and Ca2+ spikes. Dendritic response to depolarizing current injection. A: example traces taken from a layer 2/3 dendrite in control conditions (i.e., absence of any drug in perfusate or pipette solution). Spontaneous, low-amplitude sharp spikes are generated at resting membrane potential. Depolarization to -50 mV induces rapid trains of small, sharp spikes interspersed with large slow spikes. Further depolarization causes additional reduction in sharp spike amplitude, but results in long (>5 s) trains of slow spikes. Scale bars 20 mV, 120 ms. B: strong depolarization (+0.5 to +0.7 nA) generates rhythmic trains of slow spikes at theta frequency. Example trace shows 1-s epoch of theta activity during injection of +0.5 nA. Mean membrane potential was 42 mV. Below is the autocorrelation for a 3-s epoch of activity including the period shown in the example. Scale bars 30 mV, 100 ms. C: high-pass filtering (cut-off 20 Hz) reveals rhythmicity in bursts of fast spikes within the gamma range. Example trace is the high-pass filtered component of B. Autocorrelation shows clear rhythmicity at 38 Hz. Scale bars 10 mV, 120 ms. Note the resemblance of these experimental dendritic recordings with the simulated dendritic traces in Fig. 2.

IN RESPONSE TO INCREASING SOMATIC DEPOLARIZATION, RHYTHMIC DOUBLETS AND BURSTS DEVELOP, THEN RAPID TONIC FIRING. The response of the model neuron to somatic depolarizing currents was somewhat different from dendritic depolarizing currents, even when persistent g_Na was blocked (Fig. 4). As the current increased, rhythmic firingwhich started as single isolated spikesbecame associated with spike doublets, and brief bursts, with interburst frequencies at ~20 to ~40 Hz, and within-burst spike intervals ~5 ms (1.1 nA) to ~4-4.5 ms (1.5 nA). Between bursts, there is a hyperpolarization which is mostly generated by g_K(M), a voltage-dependent K⁺ conductance, and the K⁺ conductance of highest density in the model dendrites. [Recall that in previous experiments (Brumberg et al. 2000), the between-burst hyperpolarizations were "actively generated".] Still larger depolarizing currents produced high-frequency tonic firing. The overall pattern of behavior is similar to that reported by Steriade et al. (1998). [See also Fig. 3.7 in Steriade (2001).]

View larger version (26K): [in this window] [in a new window]   Fig. 4. Appearance of rhythmic bursting and then fast tonic firing with increasing somatic depolarization:simulation. Increasing somatic depolarizing currents induce bursts of increasing frequency and number of action potentials/burst, and eventually high-frequency tonic firing. DNa(P) = 0, DK(C) = 1.3.

View larger version (24K): [in this window] [in a new window]   Fig. 5. Current-frequency (fI) curve for a model neuron that is rhythmic spiking (RS) over most of the stimulus range. A: fI curve was approximately linear up to a somatic depolarizing current of 1.05 nA. A 1.2-nA current evoked singlets alternating with doublets, and 1.35- and 1.5-nA currents evoked continuous doublets. B: 2 examples of firing patterns, at points shown in the graph in A. DNa(P) = 0, DK(C) = 1.6.

View larger version (23K): [in this window] [in a new window]   Fig. 6. fI curve for a model neuron that generates doublets and bursts. A: burst frequency increases linearly from 0.15- to 0.75-nA stimulating currents. At stimulating currents of 0.9 nA, the cell switched from doublets to multiplets, and then increasing currents produced more spikes/burst, while overall burst frequency remained approximately constant. DNa(P) = 0.7, DK(C) = 1.6.

PERSISTENT G_NA FAVORS FAST RHYTHMIC BURSTING IN THE MODEL. Other authors have suggested a role of persistent g_Na in fast rhythmic bursting based on both theoretical (Wang 1999) and experimental (Brumberg et al. 2000) considerations. In our model also, persistent g_Na favors the occurrence of rhythmic bursting, with brief ADPs following the bursts (Fig. 7). (Nevertheless, as will be shown, persistent g_Na is not required for fast rhythmic bursting to occur in this model.) The simulations in Fig. 7 show an interesting feature: some of the spike doublets (middle traces) show a spikelet between the two large action potentials (e.g., arrow in Fig. 7B). Similar spikelets were seen in other simulations (not shown), either with tonic current injection or after antidromic stimulation; in these cases, it could be shown (see also Fig. 9) that the somatic spikelet originated as a full axonal spike that was decrementally conducted to the soma and that decremented further with propagation into the dendrites; it could be further shown that during tonic somatic current injection, when the axon was disconnected from the soma, somatic firing could still be induced, but without spikelets (data not shown).

View larger version (25K): [in this window] [in a new window]   Fig. 7. Increasing persistent gNa favors fast rhythmic bursting:simulation. Simulations were performed with high gK(C) density (DK(C) = 1.6). A: as persistent gNa density is increased, the firing pattern evolves from rhythmic single spikes to rhythmic bursting, with increasing numbers of spikes per burst. B: details of action potentials. Note the spikelet in the middle (arrow). In this model, such spikelets originate from extra axonal spikes. [Similar spikelets leading into action potentials were observed by Draguhn et al. (1998) and Schmitz et al. (2001) in low [Ca2+] media and were attributed to decremental antidromic invasion.] In addition, in this model (e.g., Fig. 2), tonic stimulation of either soma, or of dendrites themselves, leads to spike initiation in the axon, not in the dendrites.

EXPERIMENTALLY, SNAP INDUCES RHYTHMIC DOUBLETS, SOMETIMES WITH INTERSPERSED SPIKELETS. Bath application of 100 µM SNAP [a nitric oxide (NO) donor that enhances persistent g_Na (Hammarstrom and Gage 1999), 5 experiments] resulted in occasional spontaneous burst discharges at resting membrane potential in all layer II neurons tested after 2 h (RMP 62 ± 3 mV, n = 12). Injection of depolarizing current to maintain membrane potential at 55 mV generated repetitive single spiking in control conditions (spike frequency 14 ± 4 Hz, n = 12). After 2-h exposure to SNAP, rhythmic bursting was seen at this membrane potential in 8/12 cells tested (Fig. 8A). Bursting consisted of double spikes occurring at a frequency of 20 ± 5 Hz (n = 8) with an interspike frequency of 224 ± 12 Hz (n = 12). Spike doublets were accompanied by an ADP during rhythmic bursting. No differences were seen in spike widths at half-height for control, single spikes, and the first spike in a burst [control width 0.85 ± 0.05 ms (50 events per n = 12 cells), during rhythmic bursting 0.85 ± 0.7 ms (50 events per n = 8 cells)]. There was also no difference between the widths of the first and second spikes during a burst [second spike width at half-height 0.88 ± 0.06 ms (50 events per n = 8 cells)]. The maximum hyperpolarization following the first spike was significantly reduced when comparing control, single repetitive spikes (4.5 ± 0.8 mV from the base of the action potential, 50 events per n = 12 cells) with bursts (3.0 ± 0.2 mV, 50 events per n = 8 cells, P < 0.05).

View larger version (23K): [in this window] [in a new window]   Fig. 8. The NO donor S-nitroso-N-acetylpenicillamine (SNAP, 100 µM) induces fast rhythmic bursting that is suppressed by phenytoin and that is associated with spikelets that appear to derive from axons. A: example 1-s traces from layer 2 temporal neocortical neuron. Top: pattern of repetitive firing when this neuron was depolarized to an average membrane potential of 55 mV. Bottom: repetitive burst firing in the same neuron 2 h after addition of SNAP (100 µM) to the bathing medium. Bursting occurred spontaneously at resting membrane potential (62 mV in this case), but example shows the pattern of activity at 55 mV for comparison with control. B: multiple spikes in the presence of SNAP are generated by an afterdepolarization (ADP) compared with control. Both bursting and the ADP were abolished by bath application of 120 µM phenytoin. C: in the presence of SNAP (phenytoin absent), spontaneous bursts at resting membrane potentials were often associated with partial spikes. Injection of strong depolarizing current into the cell soma through the recording electrode abolished full spikes but brief trains of partial spikes remained (top trace). Similarly, hyperpolarization to 90 mV abolished full spikes, while leaving brief trains of partial spikes. D: during spontaneous bursting induced by SNAP, dendritic recordings show brief depolarizations, without either full or partial spikes. Injection of +0.1-nA current induced single dendritic spikes on the peak of the depolarizations, but not bursts. The dendritic spikes, when they occurred, were broader than even pairs of spikelets recorded at the soma, suggesting that somatic spikelets were not initiated 1:1 by dendritic spikes. (Phenytoin was absent in this experiment.)

BURSTS OF SOMATIC SPIKELETS IN SNAP COULD PLAUSIBLY ARISE FROM AXONAL BURSTS. The simulations in Fig. 9 support the idea that SNAP-induced spontaneous runs of somatic spikelets arise in the axon, rather than in dendrites. [The reader will recall that dendritic current injections (Figs. 2 and 3) lead to dendritic spikelets, but notin the model at leastto somatic spikelets.] In Fig. 9, we injected small current pulses into the model axon, producing brief fast trains of axonal spikes. As expected from similar simulations of CA3 pyramidal cells (Draguhn et al. 1998), axonal spike trains produced somatic spikelets. If the soma was not too hyperpolarized (Fig. 9A), one or more full somatic spikes could occur as well (cf. Fig. 8C); full somatic spikes did not occur when the soma was sufficiently hyperpolarized (Fig. 9B, cf. Fig. 8C). The backpropagating somatic spike in Fig. 9A led to a broadened, attenuated dendritic spike, resembling the experimental dendritic spike of Fig. 8D.

View larger version (20K): [in this window] [in a new window]   Fig. 9. Spontaneous runs of somatic spikelets (as in Fig. 8C) could, in principle, result from axonal spike bursts:simulation. A: an 18-ms current pulse was passed into the distal axon when the soma was at 70 mV. The resulting burst of axonal action potentials leads to a somatic spike, a large spikelet, and a run of smaller spikelets. A broad attenuated action potential is induced in an apical branch compartment (just past the 2nd branch point), resembling the dendritic potential in Fig. 8D. B: application of the same current stimulus when the soma was at 85 mV led to fewer axonal spikes, and only spikelets at the soma. The dendrite displays only a slow potential with superimposed rippling. DNa(P) = 1.0, DK(C) = 1.0.

AXONAL SPIKES CAN, IN PRINCIPLE, LEAD TO SPIKE ADPS. The simulation of Fig. 10 shows that an axonal spike can, in principle, lead to a spike ADP, even without persistent g_Na. This phenomenon could contribute to the relatively sharp spike ADPs that are sometimes shown in the literature, in neurons capable of FRB (e.g., Fig. 3.7B1 in Steriade 2001).

View larger version (13K): [in this window] [in a new window]   Fig. 10. A fast spike ADP can be generated primarily in the axon, without requiring persistent gNa:simulation. The model neuron was stimulated antidromically, with a current pulse to a distal axonal compartment. The broad somatic spike (i.e., broad relative to an axonal spike) re-excites the axon, generating a "reflected" spike (as shown to occur experimentally in CA3 hippocampal pyramidal neurons; Traub et al. 1994). The reflected axonal spike induces a brief spike ADP in the soma (arrow) [cf. Figs. 7 and 8, and also Brumberg et al. (2000) (their Fig. 3A)]. The somatic spike ADP is not generated in the apical dendrite: the spike ADP in the distal apical shaft is smaller, slower, and peaks later than the somatic ADP, i.e., the dendritic spike ADP is backpropagated. Persistent Na+ conductance is absent in this simulation. DNa(P) = 0.7, DK(C) = 1.6.

BLOCKING G_CA FAVORS SPIKE DOUBLETS AND FAST RHYTHMIC BURSTING. Both Brumberg et al. (2000) and Nishimura et al. (2001) used ionic manipulations to show that blocking g_Ca not only fails to suppress rhythmic bursting, but may even enhance it. In our model as well (data not shown), progressive blockade of g_Ca converted rhythmic single action potentials to rhythmic doublets, and then to rhythmic bursts. These simulations were done without persistent g_Na, and with a relatively high-density of g_K(C) (D_Na(P) = 0, D_K(C) = 1.6).

REDUCING INTRACELLULAR [CA²⁺] FAVORS SPIKE DOUBLETS AND FAST RHYTHMIC BURSTING. We also performed simulations in which a fixed ceiling was imposed on [Ca²⁺]_i, whose units, in the model, are arbitrary (see METHODS). As this ceiling was reduced during repeated injections of the same depolarizing current pulse (data not shown), first rhythmic doublets appeared, and then rhythmic bursts. Similar behavior was observed experimentally (data not shown) as BAPTA (0.3 mM) entered a layer 2/3 neuron from the recording electrodealthough BAPTA acts as a buffer of [Ca²⁺]_i rather than imposing an exact ceiling. BAPTA induced rhythmic bursting in all regular spiking cells tested (n = 10). The mean interspike interval during bursts induced in this way was 5 ± 1 ms, and burst frequency was 12-24 Hz. Friedman and Gutnick (1989) had previously shown a pair of bursts in a neocortical neuron injected with EGTA. None of four fast-spiking neurons developed fast rhythmic bursting on BAPTA injection.

REDUCING G_K(C) FAVORS SPIKE DOUBLETS AND FAST RHYTHMIC BURSTING. The fact that bursting, in the model and in real cells becomes more intense with g_Ca reduction, or with intracellular [Ca²⁺] reduction, suggests that it is suppression of one or more Ca²⁺-gated K⁺ conductances that underlies fast rhythmic bursting, at least under some conditions. But which one(s)? We were not able to induce fast rhythmic bursting in regular-spiking model neurons solely by manipulation of the slow AHP conductance, g_K(AHP) (data not shown). On the other hand, simulated reductions of the fast voltage- and Ca²⁺-gated conductance g_K(C) (Kang et al. 1996)which is fast enough to contribute to action potential repolarization in hippocampal pyramidal neurons (Shao et al. 1999)did lead to a transition from rhythmic spike to fast rhythmic bursting (Fig. 11A), in a pattern similar to that seen with reduction of g_Ca, or with reduction of intracellular [Ca²⁺]. Spikelets were sometimes observed during the course of simulated bursts induced by reduction of g_K(C).

View larger version (43K): [in this window] [in a new window]   Fig. 11. Reducing gK(C) density favors fast rhythmic bursting. A: simulation showing the emergence of fast rhythmic bursting as gK(C) is reduced. The reduction in conductance density was applied uniformly across compartments. Note that persistent Na+ conductance is absent in these simulations: DNa(P) = 0 (calibration: 100 ms, 50 mV). B: experimental data from a layer 2 temporal cortical neuron. The control trace shows repetitive firing during depolarization to 54 mV. Middle: 21-Hz rhythmic bursting in the same neuron, 30 min after bath application of the BK channel blocker iberiotoxin (IbTx, 50 nM). Bottom: further addition of SNAP (100 µM) leads to slowing of the burst rate, along with more spikes per burst (calibration: 100 ms, 50 mV). C: IbTx-induced rhythmic bursting persisted >1 h. Bath application of phenytoin (120 µM) reduced burst frequency, but failed to abolish bursting (calibration: 100 ms, 50 mV). D: multiple action potential generation in the presence of IbTx was accompanied by an ADP, an increase in spike half-width, and a decrease in initial spike afterhyperpolarization. Phenytoin failed to antagonize either the ADP or the spike broadening (calibration: 5 ms, 50 mV).

DURING THE COURSE OF A BURST, SOMATIC SPIKES DEVELOP INCREASING INFLECTIONS ON THEIR UPSTROKES (AS ALSO SHOWN BY BRUMBERG ET AL. (2000), THEIR FIG. 11A). The data so far suggest that fast membrane processes, on a few-millisecond time scale, are critical in allowing fast rhythmic bursting to occur. Such fast processes suggest involvement of the axon in this unique firing behavior. We noted in simulations that the very fast g_Na and g_K(DR) used in this model led to a high degree of axonal excitability. The distal axon, for example, would fire two spikes on occasion to the soma's single spike (e.g., Figs. 9 and 10); or, during somatic spike triplets, the axon might fire five times (data not shown). Some simulations (e.g., Fig. 7B) showed spikelets and notched action potentials (cf. Draguhn et al. 1998; Schmitz et al. 2001), and spikelets could also be observed in experimental recordings (e.g., Fig. 8C). It was therefore interesting to noticeboth in simulations and in confirmatory experiments (cf. Brumberg et al. 2000)that the upstrokes of somatic action potentials became progressively more inflected during the course of a burst, consistent with axonal initiation of the spikes (data not shown). This axonal initiation was directly confirmed in simulations. The progressively increasing somatic spike inflections probably result from increasing soma/dendritic K⁺ conductances and Na⁺ channel inactivation that develop during the burst.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

Intrinsic oscillatory properties at gamma frequency, both subthreshold and chattering, in neocortical neurons have been suggested to be important in the generation of network gamma oscillations in the cortex (Gray and McCormick 1996; Llinás et al. 1991; Nuñez et al. 1992; Steriade et al. 1998). For this reason alone, an understanding of the cellular mechanisms is essential. Understanding of cellular mechanismsspecifically, which currents are involved in producing particular firing behaviorsis additionally important, because such understanding allows one to make predictions as to how the currents (and thus the firing behaviors) might be regulated by the brain in vivo. It is noteworthy, for example, that persistent Na⁺ conductance can be influenced by NO (Hammarstrom and Gage 1999), while BK channels are regulated by protein kinase A (Dworetzky et al. 1996). Here, we have concentrated on fast rhythmic burstingfiring patterns of full action potentials that decrement only slightly in amplituderather than on subthreshold behavior.

We have constructed a multi-compartment, multi-conductance model of a layer 2/3 neocortical neuron that uses voltage-clamp kinetic data for simulating many of the channel types (e.g., Martina et al. 1997). Dendritic electrogenesis in this model looks reasonably realistic (Figs. 2 and 3). Contrary to our expectations when building this model (which was not originally intended to simulate fast rhythmic bursting), the model did indeed produce fast rhythmic bursting of a form similar to that seen in vivo (Steriade et al. 1998). In addition, the model agrees well with experimental data on rhythmic doublets and bursts, obtained from adult rat neocortical slices with sharp electrodes. It therefore makes sense to examine the predictions made by, and interpretations suggested by, this model as to the cellular mechanisms of fast rhythmic bursting.

What the model suggests is that one important set of membrane properties contributing to fast rhythmic bursting may be the kinetics of fast Na⁺ and fast K⁺ channels, especially in the axon. These properties alone (data not shown) allow spike ADPs to occur, when the axonal membrane is depolarized, and promote a ready tendency to produce spike multiplets. Persistent g_Na contributes to this tendency, not only because of its limited (in the model, nonexistent) inactivation, but also because of its relatively low activation threshold (French et al. 1990; Kay et al. 1998). Experimental data reported here [as well as previously (Brumberg et al. 2000)] are also consistent with a role for persistent g_Na in fast rhythmic bursting. In addition, the tendency for axonal spike multiplets to invade the soma could be regulated, we suggest, in part by g_K(C), a fast voltage- and calcium-gated conductance. g_K(C) is likely present in soma and dendrites (Kang et al. 1996); the fact that g_Ca is present in Purkinje cell axons (Callewaert et al. 1996) makes it conceivable that g_K(C) is present in axons as well, although we did not include axonal g_K(C) in the model. In the model, reduction of soma/dendritic g_K(C) could induce fast rhythmic bursting (in response to depolarizing stimuli) even in the absence of persistent g_Na, as could reducing calcium entry or intracellular calcium concentration. Dendritic depolarizations do not appear to contribute in a significant way, at least in the model. Experimentally, bath application of the BK channel blocker (i.e., g_K(C) blocker) IbTx also induced fast rhythmic bursting that is resistant to phenytoin, both as the model predicts. Presumably, the toxin is acting uniformly on somatic, dendritic, and axonal (if such exist) locations of BK channels.

Both in the model and in the experiment, fast rhythmic bursting could be accompanied by somatic spikelets that did not originate in apical dendrites and that (in the model, at least) could be shown to originate in the axon. These observations are also consistent with the hypothesis that axonal excitability contributes to fast rhythmic bursting.

In the model, it is additionally important that there be a voltage-dependent (and calcium-independent) K⁺ conductance, of relatively large amplitude and appropriate kinetics, to regulate the interburst interval. In our simulations, this role is played by "g_K(M)," but we have no experimental data concerning the presence or absence of cholinergic regulation of this conductance; if no cholinergic regulation turns out to be present, it does not influence the basic ideas of the model, provided the conductance has appropriate amplitude and kinetics.

Intraburst intervals, both in model and in experiments reported here, are often not as short as those seen in other in vitro preparations (e.g., Brumberg et al. 2000) or in vivo (Gray and McCormick 1996; Steriade et al. 1998). The reasons for this are not clear. One possibility in vivo is that chattering occurs in the presence of metabotropic activation that would tend to reduce leak (and other) K⁺ conductances. In addition, both our (isolated) model neurons and experimentally recorded neurons rarely burst at frequencies over 30 Hz. It should be emphasized, however, that previously reported chattering cells mayin response to current injectionsdischarge bursts in the 10-30 Hz range (e.g., Figure 3F of Gray and McCormick 1996). Because of the limited intraburst firing frequencies in our data, however, it is appropriate to remain cautious as to how, or if, our data apply to chattering in vivo.

If fast rhythmic bursting is of importance for in vivo gamma rhythms, one expects that the intrinsic oscillation could be gated by inhibitory postsynaptic potentials (IPSPs), as well as by intracellular depolarization: such gating is characteristic of virtually all in vitro network gamma rhythms (Fisahn et al. 1998; Whittington et al. 1995, 1997); it is not apparent how tight network synchrony could occur without regulation by IPSPs (reviewed in Traub et al. 1999). If it is additionally true that axonal properties are critical for chattering to occur, as we suggest, then axo-axonic inhibitory neurons would be expected to play an important role in network gamma oscillations. Preliminary simulations have been performed with a network of RS, FRB, fast-spiking (FS) interneurons, and low-threshold spiking (LTS) interneurons (Gibson et al. 1999). In the simulations, some FS interneurons inhibit principal cell axon initial segments, while others inhibit somata and proximal dendrites; the LTS interneurons inhibited dendrites. These simulations confirm that model FRB cells can participate in synchronized gamma oscillations at frequencies 50 Hz.