From Compartmental models to

Point Neurons
Axon hillock
 Point Neurons
General Form :

dV
  g i xi i yiqi (Vi  V )  I Synaptic
p
C
dt i

dx x (Vm )  x

dt  x (Vm )
I synaptic   g si (t )(Vi  V )
i
 Two Dimensional Neurons
Enables phase plane analysis

Most important variants

 Fitzhugh-Nagumo Model
 Morris-Lecar Model

Software: XPPAUT
 Fitzhugh-Nagumo Model (1)
A simplification of HH model
dV
C  g Na m 3h (VNa  V )  g K n 4 (VK  V )  g L (VL  V )  I ext
dt
dn
n  n  n
dt
dm
m   m  m
dt
dh
h   h  h
dt

m is much faster than the others:

m  m
 Fitzhugh-Nagumo Model (2)
Eliminating the fast dynamics
dV 3
C  g Na m h (VNa  V )  g K n 4 (VK  V )  g L (VL  V )  I ext
dt
dn
n  n  n
dt
dh
h   h  h
dt

Blue: Full system

Red: Reduced System
 Fitzhugh-Nagumo Model (3)
Eliminating h
dV 3
C  g Na m h (VNa  V )  g K n 4 (VK  V )  g L (VL  V )  I ext
dt
dn
n  n  n
dt
dh
h   h  h
dt

h ( t )  0 . 8  n (t )
 Fitzhugh-Nagumo Model (4)
Fitzhugh-Nagumo 2D model:
dV 3
C  g Na m (0.8  n) (VNa  V )  g K n 4 (VK  V )  g L (VL  V )  I ext
dt
dn
n  n  n
dt

dn dV

dt dt
V : Fast System
n : Slow System
 Fitzhugh-Nagumo Model (5)
Nullclines:

n  0 V  0
n nullcline
V  0 n  0 V  0
n  0
V  0

Orbit or
Spiral? V  0 V nullcline

n  0
 Fitzhugh-Nagumo Model (6)
Fitzhugh- Nagumo equations:

dV V3
V  W  I 2

dt 3 1.5

dW
  (V  a  bW ) 1

dt 0.5

a  0 . 7 , b  0.8 W 0

  0.08 -0.5

-1

-1.5

-2
-3 -2 -1 0 1 2 3

Qualitatively capyures the properties of the exact model

 Analysis of Fitzhugh-Nagumo
System (1)
dV V3 a  0 .7 , b  0 .8
V  W  I
dt 3   0.08
dW
  (V  a  bW )
dt

(1  V 2 ) -1 
Jacobian: A 
 -b 

Fixed point for I=0 : V=-1.20, W=0.625

Eigen valu es of A (2  (V 2  1  b )  (V 2  1)b    0) :
1, 2  0.5  0.42i
The fixed point is: Stable Spiral
 Analysis of Fitzhugh-Nagumo
System (2)
Response of the resting system (I=0) to a current pulse:

1.5

0.5

0
W
-0.5

-1

-1.5

-2
-3 -2 -1 0 1 2 3

V
 Analysis of Fitzhugh-Nagumo
System (3)
Response of the resting system (I=0) to a current pulse:
2

1.5

0.5

W
-0.5

-1

-1.5

-2
-3 -2 -1 0 1 2 3

Threshold is due to fast sodium gating (V nullcline)

Hyperpolarization and its termination is due to sodium/potassium channel

 Analysis of Fitzhugh-Nagumo
System (4)
Response to a steady current :
dV V3 a  0 .7 , b  0 .8
V  W  I
dt 3
  0.08
dW
  (V  a  bW )
dt
(1  V 2 ) -1 
Jacobian: A 
 -b 

Fixed point for I=1 : V=0.41, W=1.39

Eigen valu es of A (2  (V 2  1  b )  (V 2  1)b    0) :
1, 2  0.41  0.32i
The fixed point is: Unstable spiral
 Analysis of Fitzhugh-Nagumo
System (4)
Response of the resting system (I=1) to a steady current:

1.5

0.5

W 0

-0.5

-1 Stable
Limit Cycle
-1.5

-2
-3 -2 -1 0 1 2 3

V
 Analysis of Fitzhugh-Nagumo
System (4)
Response of the resting system (I=1) to a steady current:
2

1.5

0.5

W 0

-0.5

-1

-1.5

-2
-3 -2 -1 0 1 2 3

V
 Bifurcation
By increasing the parameter
 I the stable fix point renders unstable
 A stable limit cycle appears

If by changing a parameter qualitative

behavior of a system changes, this
phenomenon is called bifurcation and the
parameter is called bifurcation parameter
Onset of oscillation with non-zero
frequency
In the resting (I=0) the fixed point is a stable spiral

The imaginary part of the eigenvalue is not zero and the

real part is negative

In the bifurcation, the fixed point loses its stability  the

real part of eigen value becomes positive and the
imaginary part remains non-zero
The frequency of oscillation is proportional to the magnitude
of the imaginary part of the eigenvalue
By increasing I, the oscillation onset starts with non-zero
frequency
Hopf Bifurcation
IF response of Fitzhugh-Nagumo
model

30

0
0 0.25 0.5 0.75
1 I
 Neuron Type I / Type II
Gain functions of type I and Type II neurons

I II

Neural Coding

Type I : Axon Hillock of most neurons

Type II: Axons of Most neurons, whole body of
non-adaptive cortical interneurons, the spinal
neurons
 Question
What happens if we feed the Fitzhugh-
Nagumo neuron with a strong inhibitory
pulse?
Post inhibitory rebound spike

V
Morris - Lecar Model : an example
of type I neuron

Originally proposed for barnacle muscles.

Fast system is the dynamics of Ca ions

Slow system is the dynamics of K ions

 Morris-Lecar Model (1)
6

dV
C   I ionic (V , W )  I
4

dt 3

dW W (V )  W
2

 1

dt  W (V ) 0
-50 -40 -30 -20 -10 0 10 20 30 40 50

I ionic (V , W )  GCa m (V )(V  ECa )  GKW (V  E K )  Gl (V  El )

 V 1
m (V )  0.51  tan 
1

 1.5 
0.9

0.8

 V 
0.7

W (V )  0.51  tan 
0.6

 30 
0.5

0.4

5 0.3

 W (V )  0.2

cosh V
0.1

60 0
-50 -40 -30 -20 -10 0 10 20 30 40 50

V
Morris-Lecar Model (2)

Stable Spiral
 Morris-Lecar Model (3)
Current Pulse

Unstable node
Saddle node

Stable node Separatrix

 Morris-Lecar Model (4)

Unstable node
Saddle node

Stable node

Saddle Node Bifurcation

IF response of Morris-Lecar model

30

0
0 0.25 0.5 0.75
1 I
 Reminder
HH Model is Type II

Stevens-Conner Model is type I

 A hyper polarization activated current
changes the dynamics
 Bursting Neurons
Adding another slow process (Eugene Izhikevich
2000)
 Three dimensional phase plane
Integrate-and-Fire Neuron (1)
Maybe the most popular neural model

One of the oldest models (Lapicque 1907)

Although very simple, captures almost all of the

important properties of the cortical neuron

Divides the dynamics of the neuron into two

regimes
 Sub Threshold
 Supra Threshold
Integrate-and-Fire Neuron (2)
Sub-Threshold:
 The HH equations show
that in sub-threshold
regime, sodium and
potassium active channels
are almost close
 The corresponding terms
can be neglected in the
voltage equation
dV 3
C  g Na m h (VNa  V )  g K n 4 (VK  V )  g L (VL  V )  I ext
dt
dV
C  g L (VL  V )  I ext
dt
Integrate-and-Fire Neuron (3)
Sub Threshold: C
dV
 g L (VL  V )  I ext
dt
 Linear ODE
 With on input (Iext=0)Stable fixed point at (V=Vl)
 2D modes also demonstrate this behavior:

Unstable node
Saddle node

Stable node Separatrix

Integrate-and-Fire Neuron (4)
Supra threshold:
 The shape of the action potentials are more or
less the same
 At the synapse, the action potential events

translate into transmitter release

  As far as neuronal communication is

concerned, the exact shape of the action

potentials is not important,
rather its time of occurrence is important
 Integrate-and-Fire Neuron (5)
Supra Threshold:
 If the voltage hits the threshold at time t0:
1) a spike at time t0 will be registered
2) The membrane potential will be reset to a reset
value (Vreset)
The system will remain there for a refractory period
(t ref)
t0
Vth

V
Vreset

t
Integrate-and-Fire Neuron (6)
dV
if V(t)  th : C  g L (VL  V )  I ext
dt

t   registered spikes     (t  t k )

if V(t)  Vth  k

V ([t , t  t ref ])  Vreset

 Integrate-and-Fire Neuron (7)
Response to a Steady
Current
dV
C  g L (VL  V )  I ext , (V (0)  Vr )
dt
t g L t g L
I ext
V (t )  (1  e C
)  (Vr  VL ) e C  VL
gL
t th g L tth g L
I
Vth  ext (1  e C
)  (Vr  VL ) e C
 VL
gL
C  I ext  g L (Vr  VL ) 
tth  ln  
g L  I ext  g L (Vth  VL ) 
isi  tth  t ref
1 1 1
f   
isi tth  tref C  I ext  g L (Vr  VL ) 
ln    t ref
g L  I ext  g L (Vth  VL ) 
 Variants of Integrate-and-fire (IF)
neuron
dV dV
Non-Leaky IF: C  g L (VL  V )  I ext (t )  C  I ext (t )
dt dt
g L 0

Adaptive IF:
 dV
C dt  ( g L  g adapt )(VL  V )  I ext
if V(t)  th : 
 dg adapt
  g adapt
 adapt dt

t   registered spikes     (t  t k )
 k

if V(t)  Vth V ([t , t  t ref ])  Vreset

 g adapt (t )  g adapt (t )  Ginc

Spike-Response model
 Firing rate neurons
Individual spikes are not modeled, rather the
firing rate of the neurons are modeled.

Assumes that the information is coded by the

firing rate of the neurons and individual spikes
are not important
 Firing rate neurons (1)
The free membrane voltage equation (no threshold
implemented): dV
C  g L (VL  V ) I ext (t )
dt
The firing rate is a function of the free membrane
voltage: f  g (V )

g is usually a monotonically increasing function. These

models mostly differ in the choice of g.
 Firing rate neurons (2)
Wilson-Cowan model
 Usually two neurons (excitatory/inhibitory)

E   E  g (aE  bI  iput E )

I   I  g (cE  dI  iput E ) Input E

b
1
g ( x) 
1  ex d I E a
g(x)

Input I c

x
 Firing Rate Neurons (3)
Linear-Threshold model:
dV
C  g L (VL  V )  I ext , f  G (V )
dt f
0 if V  Vth
G (V )  
aV , a  0 if V  Vth
V

Based on the observation of the gain function

in cortical neurons:
f

100 Hz
Physiological
Range
I
 Steady State model
The input current and the firing rate of the
neuron are constant in time:
f i  g ( I ext
i i
), I ext   wi , j f j
j

1
g(I ) 
1  e  I

Used in Artificial Neural Networks

(Perceptrons…)
 Binary neuron
Depending on the input, the neuron is either
on (1) or off (0): f  g ( I i ), I i   w f
i ext ext i, j j
j

0 if I  
g (i )  
1 if I  

Ising neuron (+1/-1):

f i  g ( I ext
i i
), I ext   wi , j f j
 Used in Hopfield network j

 1 if I  
g (i )  
 1 if I  
 Overview of the neural models:
Biological Reality Numerical Simulation
Detailed conductance based models (HH)

Reduced conductance based models (Morris-Lecar)

Two Dimensional Neurons (Fitzhugh-Nagumo)

Integrate-and-Fire Models

Firing rate Neurons (Wilson-Cowan)

Steady-State models

Binary Neurons (Ising) Analytical Solution

Artificial