ctrophysiological Examples τ (V) < 0.

Membrane Voltage (mV)

0.8 0

Activation Variable
0.7 -10
τ (V) < 0.1

Huxley description Simplified

of dynamicsneuron modelspotential
of membrane (Izhikevich, parts of chapters 3 and 4)
and voltage-gated 0.6 -20 τ (V) < 0.01
0.5 -30
Instantaneous
an be reduced to a one-dimensional system when all transmembrane 0.4 -40

ave fast kinetics. For the sakeand

One-dimensional of illustration, let us models
two-dimensional consider a space- 0.3 -50

rane having leak current and a fast voltage-gated current Ifast having 0.2 -60

Assume
variable p, a neuron membrane with one leak and one fast current: 0.1

0
-70

-80
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9

Time (ms) Time (ms)

Leak IL Ifast
z }| { z }| {
C V̇ = ° gL (V ° EL )Figure
° g p (V
3.1:° E) (3.2) (3.2, 3.3) converges to that of
Solution of the full system
ṗ = (p1 (V ) ° p)/ø (V )
one-dimensional system (3.4) as ø (V(3.3)
)!0

If t(V) <<1,
nless parameters C =p changes
1, gL =much faster
1, and g than
= 1.V (time scalethat
Suppose separation).
the gat-We can consider that in
thefaster
time than
scale that reduce
V varies, p is the two-dimensional
already in equilibrium system (3.2, 3.3) to a one-dimensional equat
3) is much the voltage kinetics (3.2), which means that the
e time constant ø (V ) is very small, i.e. ø (V ) ø 1, in the entire biophys- instantaneous Ifast
dp
nge. Then, the gating process may be treated as being instantaneous, z }| {
≅0 ⇒ p = p∞ (V ) One-Dimensional Systems
C V̇ = °gL (V ° EL ) ° g p1 (V ) (V ° E) .
totic value p = pdt 1 (V ) may be used in the voltage equation (3.2) to
20
This reduction introduces a small error of the order ø (V ) ø 1, as on
m(t) 55 10 V(t)
Fig. 3.1. τ (V) < 0.5
Membrane Voltage (mV)

-10 Since theτ (V)hypothetical

< 0.1 current Ifast can be either inward (E > EL
-20
(E < EL ), τand (V) < the
0.01 gating process can be either activation (p is m, as in
Solution for different values
-30
Huxley model) or inactivation (p of
Instantaneous is the
h),time
there are four fundamentally diff
constant
-40

-50
for Ifast (V ), which we summarize in Fig. 3.2 and elaborate on below.
-60

-70

1 2 3 4 5 6 7 8 9 10
-80
0 1 2 3 4 5 6 7
Current8 9 10
Figure 3.2: Four fundamental examples
Time (ms) Time (ms) inward outward gated currents with one gating variab
 at every point V where F (V ) is negative,
- the derivative
+
V̇ is F(V)=-I(V)/C
negative, and hence
above; the

curr
presented see also Fig.

derivative of membr
-1 negative feedback, positive feedback,
resonant current 0amplifying current
state variable V decreases. In contrast,INa,pat every point where F (V ) is positive, V̇ is
IExample:
and the Instantaneous V persistent Nagreater
current
(V)

Leak + instantaneous positive,

Na,p
state variable
-2
increases; the the value of F (V ), the faster
V increases. Thus, the direction throughout
-60 -40 -20
the rest of this
of movement
0 20 40 60
ofchapter.
the-50state (Some biologists
variable
-60 -40 -20 0 20
V 40,refer
and
60
tohence
transienttheNa+ curr
ur introduction into dynamical systems,
evolution we dynamical
of the will use thewith
Isystem,
-model
Na,pvery slow inactivation
ispotential
membrane determined
(mV)
asby
beingthepersistent,
sign sinceVfunction
ofpotential,
membrane the the currentFdoes
(mV) (V ).not change m
on the time scale a of 1 sec.) We obtain the experimental b parameter values
The right-hand side of
instantaneous INa,p the I leak -model (3.1) or the I Na,p -model (3.5) in Fig. 3.8 is the
z current-voltage
steady-state }| 3.5: a. I-V
Figure { (I-V) C relation,
= 10ofµF
relations IL (Vcurrent,
the, leak I) =
or0IpA (V
LIL )+I
,,fast Na + (VgL = ) respectively,
current, 19ImS , EL taken
= °67 mV ,
Na,p Na , and combined
C V̇ = I ° gL (V ° EL ) ° gNa m1 (V ) (V ° ENa ) g = 74 mS (3.5)
, V = 1.5 mV , k = 16 mV , E = 60 mV
with the minuscurrentsign, I(V
see) Fig.
= IL (V 3.5.+ IPositive
) Na Na (V ) in the values
INa,p of the
1/2-model right-hand
(3.5). Dots denote side I0 (VF) (V
data)Nafrom
mean
layer 5 pyramidaltocella in
negative I-V, corresponding net ratinward
visual cortex. current b. Thethat right-hand
depolarizes side ofthe the membrane.
INa,p -model
using whole-cell patch clamp recordings of a layer 5 pyramidal neuron in the vi
(3.5).
Conversely, negative
m1 (V ) = 1/(1 + exp {(V1/2 ° V )/k}) values mean
cortex of positive
a rat at I-V,
room corresponding
temperature. We to
prove a net
in Ex. outward
3.3.8 and current
illustrate in Fig.
that hyperpolarizes the membrane. that the model approximates the action potential upstroke dynamics of this neuro
dynamical systemsThe are model’s
usually written in the I(V
I-V relation, form), is depicted in Fig. 3.5a. Due to the nega
conductance region in the I-V curve, this one-dimensional model can exhibit a num
This is a one-dimensional dynamical system
3.2.2 Equilibria of interestingV̇ non-linear = F (V ) , V (0) = V 2 R ,
phenomena, 0such as bistability, i.e. co-existence of res
(3.6)

for example, and excited states. From a mathematical point of view, bistability occurs beca
The next step in the qualitative analysis
the right-hand of
V̇ =side
°80any° Vdynamical
function , in Vthe =system
°20 , isequation
(0)differential to find(3.5),
its equilibria
depicted in Fig. 3
The steady-state
or restorpoints,
equilibrium
i.e., thefor the variable
is
values
where V is a scalar
not V (membrane
monotonic.
oftime-dependent
the state In Fig.
variable
variable
3.6 potential
wherewe
denoting
depict
the
at rest)state
typical
current
is given
voltage
of
time
the
by
courses of the m
system,
(3.5) with two values of injected dc-current I and 16 different initial conditions.
V̇ = Vt = dV qualitative
/dt is its derivative
behavior in withFig.respect
3.6a isto time bistable:
clearly t, F is a depending
scalar functionon the(its
initial condit
F the
(V )trajectory
output is one-dimensional) = 0 that (V is an
determines equilibrium).
the evolution of the system,
of the membrane potential goes either up to the excited e.g., the right- state or d
hand side of (3.5)to the divided
restingby state.
C; see InFig. 3.5b. V0the
contrast, 2 Rbehavior
is an initial condition,
in Fig. 3.6b is and R is
monostable, since
A phase plane plotOne-Dimensional
At each(plot of F(V) versus
such point
the V̇
real =
line, V) willstate
0,
i.e.,
Systemsrestingthe
a line givedoes
state
of both
real notthe
variable
numbers Vequilibrium
(R
exist.
n
does
would not
be
The goal of the points
change.
the and its
In
n-dimensional
dynamical the63stability
context
real
system space). of
theory reviewed in
In
membrane potential dynamics, the context of dynamical
chapterequilibria systems,
correspond
is to understand the real
why andto line
how R
the is called
thepoints phase
behaviorwhere line
depends or
the state line
on steady-
the initial condit
stable (phase space or andstate space
unstable
the for Rn )oftothe
parameters stress
system.the fact that each point in R corresponds
state I-Vequilibrium
curvetopasses zero. At
a certain, possibly
each such point there is a balance of the inward and
inadmissible state of the system, and each state of the system
equilibrium
outward currents so that tothe net transmembrane current is zero, and the membrane
F(V)
corresponds a certain F(V)in R. For example, the state of the Ohmic membrane
point
voltage does not change.
(3.1) is just its3.2
(Incidentally,
membrane Dynamical the part
potential l ībra
V 2 Systems
R. Thein theofLatin
state word aequil model
the Hodgkin-Huxley ībrium
means balance).(see VSect. 2.3) is the four-dimensional vector (V, m, n, h) 2 R . The state of the INa,p -
4

model (3.5) is In general,

again its membrane V potential
dynamical systemsVcan 2 R,
bebecause
continuous or discrete,
the value m = mdepending on whe
1 (V ) is
The IK - and Ih -modelsthey mentioned
areby
in
described
Sect. 3.1 can have only one equilibrium
by differential or difference equations. Continuous one-dimensio
be-
unequivocally defined V.
cause their I-V relationsWhen allI(V ) are monotonic
parameters are constant, increasing
the dynamical
Figure
functions.
3.9: The system
The
sign ofisthecalledcorresponding
autonomous.
slope,
functions F (Vslope
negative )When
are monotonic ofdecreasing
at least one positivethe parameters
slope and can= Fhave (V ), only
is time-dependent,
∏ 0 theone
determines zero.
system
the isstability
non-autonomous,
F (V)<0denoted
In contrast, the INa,p as V̇ = F (V,
- and Ft).
IKir -models can of
(V)>0 have many equilibria because their I-V
the equilibrium.
To solve (3.6) means to find a function V (t) whose initial value is V (0) = V0
curves are not monotonic, and hence there is a possibility for multiple intersections
and whose derivative is F (V (t)) at each moment t ∏ 0. For example, the function
with the Stability
V -axis.Vof(t)
For example,
an= equilibrium
V isthere are three
determined by theequilibria
signs of theinfunction Fig. 3.8b
0 + at is an explicit analytical solution to the dynamical system V̇ = a. The
corresponding
F around it. to
 grap

grap
stable 20
61 -50
equilibrium
0
-100
-100 -50 0 -50 0 50
ak-model INa,p-model membrane potential, V (mV) membrane potential, V (mV)
100
Which has two stable equilibrium values for V (at
80
IL+INa
around -50 mV and 35 mV)

graph of F(V)=V
0 40
60 stable equilibrium
=0 V<0 -10 V(t) 30
40
-20 20

membrane potential, V (mV)

membrane potential, V (mV)

20
um -30 10
0 V(t)

-50 0 -50 0 50 Bistability -40 0

-50 -10
potential, V (mV) membrane potential, V (mV)
-60 -20
stable equilibrium
-70 -30
unstable equilibrium
40 -80 -40
stable equilibrium -90 -50
30 stable equilibrium
-100 -60
20 0 5 0 5
membrane potential, V (mV)

time (ms) time (ms)

10 a b
V(t)
Stability of equilibria
0 can also be calculated by linearization of the dynamics around the
Figure 3.8: Graphs of the right-hand side functions of equations (3.1) and (3.5) and
equilibrium value-10 corresponding numerical solutions starting from various initial conditions.
-20
stable equilibrium

V = Veq + ΔV -30 analytical solution of (3.6), see the dots in Fig. 3.7. The approximation has a no-
unstable equilibrium F '(Veq ) >use
ticeable error of order h, so scientific software packages, such as MATLAB, 0 more Unstable
! ! ! -40
V = Veq +Δ V =-50F(Veq + ΔV )
sophisticated high-precision numerical methods.
In many cases, however, we do not need exact solutions, but rather qualitative
stable equilibrium
understanding of the behavior of (3.6) and how it depends on parameters and the
5
-60
! ≈ F(V ) +ΔV 0 )
dF(V 52 initial state V0 . For example, we might be interested in the number ofeqequilibrium Stable F '(V ) < 0
me (ms)Δ V time +o(ΔV
(ms) ) = F '(Veq )ΔV
eq V=Veq (rest) points the system could have, whether the equilibria are stable, their attraction
a dV b domains, etc.
F '(V )t F '(Veq ) = 0
right-hand
ΔV side functions
(t) = ΔV of equations eq
(0)e (3.1) and (3.5) and
Need to include higher order terms in Taylor expansion
olutions starting from various initial conditions.3.2.1 Geometrical analysis
 One-Dimensional Systems

Unstable equilibria separate attraction domains and act as thresholds

One-Dimensional Systems

Unstable V
Equilibrium
V state A threshold state B

F(V)
Attraction domain Attraction domain F(V)
of state A of state B

Attraction Domain
subthreshold
perturbation

Attraction Domain
threshold
state A
3.12: Two attraction domains in a one-dimensional system are separated by the
e equilibrium.
Attraction domain: set of initial conditions
eshold) that converge
perturbations inthe
leave time tovariable
state a specific stable
in the attraction domain of the state B
te, whileequilibrium
large (superthreshold) perturbations initiate the regenerative process
upstroke of an action potential, and the voltage variable becomes attracted to the superthreshold
state. Generation of the action potential must be completed via repolarization, perturbation
moves V back to the rest state. Typically, repolarization occurs because of a
ly slow inactivation of Na+ current and/or slow activation of an outward K+
, which are not taken into account in the one-dimensional system (3.5). To
threshold
for such processes, we consider two-dimensional systems in the next chapter. state A
all that the parameters of the INa,p -model (3.5) were obtained from a cortical
dal neuron. In Fig. 3.15, left, we stimulate (in vitro) the cortical neuron by short
) strong pulses of current to reset its membrane potential to various initial values
erpret the results using the INa,p -model. Since activation of Na+ current is not state B
aneous in real neurons, we allow variable m to converge to m1 (V ), and ignore
ms transient activity that follows each pulse. We also ignore the initial segment
 Exercise: Depolarize the initial membrane potential of the leak+instantaneous Nap model at
different values and verify 76
that the unstable equilibrium acts as a threshold.
One-Dimensional Systems

Bifurcations: Qualitative changes in the structure of equilibrium points as we change

ensionalaSystems
parameter (change of stability, appearance
73 or disappearance of equilibria)
bistability I=0 V(t)
100 40 excited state

membrane potential, V (mV)

F(V)
Bistability 80 20

60 0
40
-20
20
-40 threshold
rest
0
rest threshold -60

bifurcation I=16
100 40 V(t) excited state
F(V)

membrane potential, V (mV)

Bifurcation 80 20

60 0
40
-20
20
-40
0
tangent point -60

monostability I=60
100 40 V(t)
Monostability excited state
F(V)
80 membrane potential, V (mV) 20

60 0
40
-20
20 Slow transition to the
-40
0 high V equilibrium
-60
-50 0
membrane potential, V (mV)
50 0 1 2 3 4 5 (‘ghost’ of the previous
time (ms)
.22: Mechanistic illustration of a bifurcation as a change of the landscape. low stable equilibrium)
 100 ms
nal Systems 81
ms
on
Exercise: Starting in the low equilibrium value81(V=-52.51 mV), add FaIshort
(Vsn , I) 6= 0 pulse
current (at(10
I= Isnto
ms) ), the
leakage-fast Na,p, with I=20 pA and I=30 pA. Where is the threshold to switch to the high equilibrium value?
Now change the pulse where
duration to a longer
slow transition one. F I denotes
Explain the derivative
the reason of F with
of the observed respect to I. Geometricall
behavior.
-60 mV
means that as I changes past Isn , the graph of F approaches, touches, an
43.1 pA intersects the V axis.
0 pA Bifurcation diagrams
I (V)
29:(V) Plotslatency
A 400-ms of equilibria
in a layeras a function
5 pyramidal of
neuron aofparameter
Saddle-node
rat visualbifurcation
cortex. results
Normal in appearance
forms or disappearance of a pair of equ
Figure 3.31: Equilibria are intersec-
as in Fig. 3.26. None of the six examples on the right-hand side of Fig. 3.27 can un
-40 Figure tions of
3.31: the steady-state
Equilibria I-V curve
are intersec-
unsta
ble eq a saddle-nodeI = bifurcation because at least one of the conditions above is violate
-60 -40 -20 40 ofI1the
0 uilibria20tions (V ) steady-state
and a horizontal
I-V line
curve const. Minimal mathematical representations
membrane potential, V (mV)
I1 (V ) and a horizontal line The number of conditions involving strict equality (“=”) is called the co-dim
membrane potential, V (mV)

I = const.
-20 0
ane potential, V (mV)
20 40
of a bifurcation. The saddle-node(in terms of polynomials)
bifurcation of the because there
has co-dimension-1
-45
saddle-node
esce and annihilate each other. As the parameter one
(fold) bifurcation
condition
varies involving
from right to “=”,different
and the types of bifurcations.
other two Saddle-
conditions involve inequalities
bria — one stable and one unstable — appear from
annihilate each other. As the parameter variesCo-dimension-1 a single point.
from right to Thus, node bifurcations correspond to a
bifurcations can be reliably observed in systems with one param
he stable
direction ofone
movement of—the bifurcation parameter, the saddle-node
ne
plains
and
-50
unstable appear from a single point. Thus, quadratic
It is an easy exercise to check that normal form
the one-dimensional system
tion ofdisappearance
movement uor
ilib appearance
ofeqthe bifurcation of
ria a new stable
parameter, state. In any case,
the saddle-node
stable
behavior of the
sappearance systems changes
or appearance exactly
of a new at the
stable bifurcation
state. point.
In any case,
r of the systems changes exactly at the bifurcation point. V̇ = I + V 2
-55
urcations 0 and I-V 5 recordings
10 15 20 25 Check that the bifurcation diagram of this
injected dc-current I, (pA) is at saddle-node bifurcation when V = 0 and I = 0 (please, check all three condi
ons
erminingandsaddle-node
I-V recordingsbifurcation diagrams of neurons may be a daunting normal form is typical of the saddle-node.
task. However, it is a trivial exercise when
This system
the bifurcation
is called theSystems
One-Dimensional
parameter
topological normal form for saddle-node bifurcation
Figure 3.30:bifurcation
Bifurcation diagram ofneurons
the system in Fig. 3.26.
saddle-node
dc-current The saddle-node diagrams bifurcation
of diagram
may be a of aas the ofonethis system are topologically equivalent to those depicted in Fig
daunting
phasesuchportraits
owever, it isI.a trivial
In thisexercise
case, the whenbifurcation diagram,
the bifurcation parameter
Inneuron can bebifurcation
obtained from the I-V
justI.the steady-state I-V relation I1 (V except
) plotted that theI-V relation
bifurcation occurs at I = 0, and not at I = 16.
ent this case, the diagram, suchonasthe
the(I,
any neocortical neurons, such as the one in Fig. 3.29, to generate repetitive
V200
one )-plane.
uation charateristic
e steady-state I-V relation I1and
(V ) plotted on the (I, V )-plane. I=16
steady-state current (pA)

entials with small frequency, how it predicts that all such0 neurons,
C V̇ = I ° I1 (V ) = 0
as dynamical systems, reside near saddle-node bifurcations. I=-100
C V̇ = Iif°and
s an equilibrium I1 (V
only) =if0the net membrane current, I ° I-200 1 (V ), is
ple, equilibria of the INa,p -model are solutions of the equation-400
Bifurcation diagramnet membrane current, I ° I1 (V ), is
ilibrium if and only if the I (V)
ilibria of the INa,p -model areIsolutions of the equation
1 (V ) -600
z }|
tep in the geometrical bifurcation analysis of one-dimensional { systems is Figure 3.31: Equilibria are inte
= I ° (gL (V
s of 0zbifurcation °I1E(V
diagrams,L ) )+which
gNa m1we (V )(V
do °E
in Na ))3.30
Fig. , for the-800
saddle-node tions of the steady-state I-V c
}| {
shown
I ° (g in
(V Fig.
° E 3.26.
) + g Tom1 draw
(V the°bifurcation
)(V E )) , how diagram, we determine-100 -80 the-60 -40 -20 0 20 40 I1 (V ) and a horizontal line I = c
L L Na
directly from (3.5). In Fig. 3.31 we illustrate Na to find the equilibria membrane potential, V (mV)
f the stable and unstable equilibria for each value of the parameter I and
We plot the steady-state I-V curve I1 (V ) and draw a horizontal line
asromwhite or black
(3.5). In Fig.circles
3.31 inwethe (I, V ) plane
illustrate how to in find
Fig. the
3.30.equilibria
The equilibria form
 Possible bifurcations in dynamical systems with only 1 variable
Described by normal forms (simplest polynomials for f(x) that are structurally equivalent to each
type of bifurcation).

Saddle-node:
dx dx dx
Normal form: dt dt dt

dx
= r + x2
dt x x x

r<0 r =0 r >0
Unstable eq. points
xeq
Saddle-node bifurcation
Bifurcation diagram:
r
Stable eq. points
Transcritical bifurcation
xeq
dx
= rx − x 2
dt
r

Pitchfork bifurcation
dx dx
= rx − x 3 = rx + x 3
dt dt
xeq xeq

r r

Supercritical
Subcritical
 Bistability can generate hysteresis Increase and then decrease a parameter does not take
68 you to the original point. One-Dimensional Systems
One-Dimensional Systems

40 40
states repolarization
excited
20 20
(another variable)
membrane potential, V (mV)

membrane potential, V (mV)

saddle-node
(fold) bifurcation
0 thre
sho
ld s
superthreshold
tate perturbation
0
Exercise: ‘Experimentally’ prove
s
-20
hy
pe
rp
-20 IK
hysteresis in the leak-fast Nap model
-40 -40
starting from strongly negative injected
ol
saddle-nodeariza
tio
(fold) bifurcation n

re pstr
tio
n lariza

u
depo
current and applying a ‘ramp’ current,
ge ok
-60 threshold -60
rest state

ne e
ra
measuring the value of the voltage at
tiv
-80 tes
I (V) -80
sta I Na
e
t 16
res
-100 -100
each ramp value.
-120 -890 -120 excited
-1000 -500 0 -1000 -500 0
injected dc-current,
Figure 3.14: Mechanistic I (pA) of the mechanism of generation
illustration injected dc-current,
of anI (pA)
action po-
tential.
Figure 3.32: Bifurcation diagram of the INa,p -model (3.5). 68 One-Dimensional
Note that although the leakage-fastNa model reproduces the upstroke We need another variable for
in membrane potential of the neuron, it never recovers the low voltage recovery
e phase portrait of the system, as we illustrate in Fig. 3.32, right. Each point where
e branches
strong fold (max or min of I1 (V 40 )) corresponds to a saddle-node bifurcation. repolarization
20 mV
nce theredepolarization
are two such folds, 1 ms at I = 1630pA and at I = °890 stable equilibrium
pA, there are two saddle- (another variable)
de bifurcations in the system. The first one, studied in Fig. 3.25, corresponds to the
20
membrane potential (mV)

superthreshold
sappearance of the rest state. The other 10 one, illustrated in Fig. 3.33, corresponds perturbation
IK
membrane potential (mV)

+30 mV 0
the disappearance of the excited state. It occurs because I becomes so negative hy
pe
rp
action -10
at the Na+ inward current is
ol
no longer
potentials strong enough to balance the leak outward ariz
at
io tion

re pstr
-20 n lariza
rrent and the negative injected dc-current to keep the membrane in the depolarized

u
depo

ge ok
threshold
rest state

ne e
-30
xcited) state.-40 mV

ra
unstable equilibrium

tiv
-40
I Na

e
Below the reader can find more examples -50
of bifurcation analysis of the INa,p - and
ir -models, which have non-monotonic
pulses of current -60 I-V relations and stable
canequilibrium
exhibit multi-stability
0 0.2 0.4 0.6 0.8 1 1.2
states. The IK - and Ih -models have monotonic I-V relations and hence only one
time (ms) excited
uilibrium state. These models cannot have saddle-node bifurcations, as the reader