Action potentials and limit eveles wT
successive peaks should therefore have heights of 18.0, 5.4, and 1.62. Figure9.8 shows that
thisisabout right, although there are too few spikes inthis simulation about a thousand)
to accurately delineate the higher order peaks
95 Human and mammal
n cortical neurons
Equation (9.7) provides an accurate and tractable description of the Hodgkin-Husley
‘equations for action potentials in the squid axon. However, the squid axon is unusual in
havingonly one Na” and one K” current, Asa consequence. the squid axon cannot fireat
rates below about 175 spikes/s and produces only a modest increase in spike rate with,
increasing input current (see Fig. 9.3), The vast majority of neurons in other animals also
possess a rapid, transient K” current that permits the cel to fire at very low spike rates,
‘witha long latency to the first spike when the input current is low (Rogawski, 1985). This
current, known as /,, Was first characterized and added to the Hodgkin-Huxley model by
Connor, Walter, and McKown (1977). fs currents are Found ina wide range of neurons,
including human and mammahan neocortical neurons (Avoli and Williamson, 1996;
Gutnick and Cll, 1995). Because of its ubiquity and also because neurons with an Ix
current exhibit a mathematically distinctive bifurcation from the resting to the spike
generation stave, let us extend our action potential model to incorporate
‘As itlethe Na aud K* cussents in the syuid axon, Conus e¢e/-(1977)imwrpurated a
mathematically complex, transcendental description of the /4 current. More recently,
Rose and Hindmarsh (1989) demonstrated that many effects of the /, current could be
approximated by making the equation for the recovery variable R quadratic. Let us
therefore examine the dynamical consequences of a quadratic dR/dr equation. The
equations are
80 (uri ¢ 4750 433812) 70.48) ~ 268074 098)+1
ARN ay 297 40.79 | 0.330 1 0.38)) (9.10)
*
Soms
The capacitance C = 1,0 Fjem? and therefore has not been written explicitly. These
‘equations have parameter values chosen to provide a good approximation to the action
potentials produced by human neocortical neurons (Wilson, 1999), (Human neurons
have been studied intracellularly in tissue, usually from the temporal lobe, removed
doring epilepsy surgery.) Note that the d K/dr equation s wntten as the sum of a hnear and
quadratic voltage term, which may be interpreted qualitatively as the normal /« and the
transtent /« current contributions.
Figure 9.9 shows the state spave for (A) / =O and (B) = 0.5nA from MatLab script
HamanNeurou.an, For 1 = 0 there are three steady states at «, and c. The resting state at
4, where V— 0.75 (~7SmV), is an asymptotically stable node, while b and ¢ are an
unstable saddle point and an unstable spiral point respectively. Threshold is reached for
(0.10) when the stable node (a) and unstable saddle point (b) coalesce and vanish at a148 Spikes, decisions, and actions
a 5
Lim yo
f 7
Ll
ae 0s as
v
Py
¢
o|
v
50
10
10 ° 70
Fig.9.9- Phase planes for (910) for /=0 (A) and /=0.5(B) In A the olins intersect in thee steady sates
fam asympouialysiabie ne (a, Sale phat (aml a sable anal pit () Ks B the se and
shows the bifurcation diagram for (9.10) as a function of £
eA gata a Sea eye mcd peal pad i
bifurcation, which is therefore very different froma Hopf bifurcation, Thisleaves only the
tanstable spital print eaten wich ah
cation diagram in Fig. 9.9C, where limit eyeles surround the single unstable spiral point
for 1 >A, while three steady states, including the resting state at the stable node, exist
between A and B. Because a saddle point and a node coalesce and vanish at A thisisealled
4 saddle-node bifurcation, and the limit cycle that emerges in such eases will have a finite
amplitude at its inception This is very different from a Hopf bifurcation where the limit
cycle must originate with infinitesimal amplitude when a single steady state changes from
stable to unstable, Not all saddle-node bifurcations produce limit eyeles. and simulations
are generally required to determine whether or nota limit eyele does occur.
Figure 9.10 illustrates the spike train produced by eqn (9.10) for f= 0.22nA. Note the
{wo charaeteristicchanges produced by the simulated /a current: along latency tothe fist
spike (about 200 ms), and a much lower firing rate (about 5.0 spikes/s) than that produced
byeither (9.3) or(9.7) and shown in Fig. 9.3. In fact, just above threshold (J = 0.2149 nA),
‘pele energes, This is evident ia the bifureAction potentials and limit eyelos 149
8
Potertil (n¥)
‘0100 200300400 500. 600700 600 000 1000
“Time ms)
Fe. 9.10 Spike tain fom (9.10 for
22mA. Note lone latency and lw spike rate compared to Fig 94
(0.10) produces latencies of more than | sand spike rates below | spike/s! Both the lower
spike rates and the long latencies result from the fact that the dR/dt = Oisocline lies very
close to the limit cycle inthe subthreshold region marked ‘sub’ in Fig. 9.9B. This means
that dRjdr will be very small in this subthreshold region, so the interval between spikes is
seatly increased relative to the case where the dR/dr isocline is far away (compare with
Fig. 9.2). The greatly reduced spike rates produced by x have the functional significance
of extending spike rate coding over a much greater dynamic range (Connor et al., 1977
Rozawski, 1985).
‘Neurons that can be stimulated to fire at arbitrarily low frequency due to saddle-node
bifurvations are termed Class I neurons, while those that can only bein firing ata rela-
tively high frequency resulting from a subcritical Hopf bifurcation are called Class I
neurons (Ermentrout, 1998). Almost all mammalian ncurons arc Class I, while the squid
axon is of Clase IL
‘The cortex of humans and other mammals is currently thought to contain four major
types of neurons, each of which is characterized by its distinctive nonlinear dynamical
properties. Spike shapes for two of these cel types. regular-spiking (RS) and fast-spiking
(FS). are plotted in Fig. 9.1 1C (McCormick et a.. 1985). Action potentials of RS neurons,
havea rapid rise but a much slower decay phase. FS cells, on the other hand. havea similar
tise ut a much more rapid decay phase than regular-spiking cells, and in consequence, the
width of the pikes is significantly narrower. In humans, spike widths at half amplitude
average 0.95ms for RS cells and 0.60ms for FS cells (Wilson, 1999). The model in
eqn (9.10) has already been optimized to approximate the size and shape of RS neuron
action potentials obtained from human neocortical neurons, and a comparison between
model results and human data (Foehring et a., 1991) is shown in Fig. 9.11. The model
Provides an accurate quantitative fit to the spike shape and, with the addition ofa slow
potential to be discussed in the next chapter, to the spike ratesay well, Equation (9.10) can
also provide an excellent quantitative approximation to the action potentials of PS cellsif
just one parancter is changed: the recovery time constant must be reduced to 7_ = 2.1 ms,150 Spikes, decisions, and actions
40
a 8
A eran
7 Ean (9.10)
o
Fr
8d
Time (see) Time (ms)
Potential (mv)
Regular—spiking
[\
vm 4. \.---------
1.0ms
Fig. 9.11 Repularspiking (RS) and fust-piking (FS) neurons, (A) Spike shape ofa human RS neuron (ata
from Foehsing eal 1991) compared to spike [om (9.10) (B) RS and FS spike shapes produced by changing
4 (910) from 5.6m (RS) to 2 ms (FS) (C) FS and RS spike shapes recorded from neocortical neuro
(eeproduced wih permission, McCormick etal 1985).
{his speed-up of the recovery phase by a factor of 2.) produces the FS action potential
plotted in Fig. 9.11B, You can verity this by changing TauR in the MatLab script,
‘HumanNNeuronan, Nute tac the FS spike is wut ouily narsywer but also slightly reduced in
hight relative to the RS cell, and this is also evident in the data in Fig. 9.11C. Thisisa
‘dynamical consequence of the faster R variable in the equations,
9.6 Subharmoi
resonance and phase shifts
‘When the stimulus to linear system is periodic with frequeney w the response is always
periodic at the same frequency. In nonlinear systems, however. limit cyeles can respondat
subhiarmonics of the stimulus frequeney, a phenomenon known as subharmonic reson
ance. Cochlear neurons in the auditory system exhibit subharmonic resonance in
response to pure tones of appropriate frequencies (Kiang et al., 1965). The phenomenon
of subharmonic resonance may be illustrated by the response of (9.7) to the periodic
stimulus = Asin(2nw). If you run the MatLab script HHnoisem with A = 0.25,
SDnoise =0 (noise is not relevant here) and w = 200 Hz, the result at the top of Fig. 9.12