HAL Id: cel-01626745
https://hal.inria.fr/cel-01626745
Submitted on 31 Oct 2017
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Bruno Cessac, Dora Karvouniari
To cite this version:
Bruno Cessac, Dora Karvouniari. A mathematical approach to retinal waves. Doctoral. Chile. 2017.
�cel-01626745�
A mathematical approach to
retinal waves
B. Cessac, D. Karvouniari
Biovision team, INRIA, Sophia
Antipolis
Acknowledgement
L. Gil, INLN, Sophia Antipolis
O. Marre, Institut de la Vision, Paris
S. Picaud, Institut de la Vision, Paris
Matthias Hennig, University of Edinburgh
Evelyne Sernagor, Newcastle University
2
Visual
cortex
LGN
Visual system and retinal waves
Retinal structure
3
Visual
cortex
Retinal structure
LGN
Visual system and retinal waves
Spontaneous spatio-temporal ‘’waves’’
during development
§
Disappear short after birth
Spatio-Temporal evolution of a
retinal wave in mice during
development, Maccione et al. 2014
4
5
Stages of Retinal Waves During
Development
Stage I
•
Formation of
retina circuitry
•
Chemical
synapses not
formed yet
•
Gap juction-mediated
6
Stages of Retinal Waves During
Development
Stage I
Stage II
•
Formation of
retina circuitry
•
Chemical
synapses not
formed yet
•
Gap juction-mediated
•
Retinotopic mapping
•
Nicotinic Acetylcholine Receptors
(nAChR)
7
Stages of Retinal Waves During
Development
Stage I
Stage II
Stage III
•
Formation of
retina circuitry
•
Chemical
synapses not
formed yet
•
Gap juction-mediated
•
Retinotopic mapping
•
Nicotinic Acetylcholine Receptors
(nAChR)
•
Disappear when
vision is functional
•
Glutamate –
AMPA receptors
groups.oist.jp/dnu/
8
Focus on stage II retinal waves
Stage II
(cholinergic
retinal waves)
Zheng et al. 2006
Stage II retinal waves are well studied
experimentally and there is already existing
work on their modelling
Starburst Amacrine
Cell (SAC)
Goal of this lecture
Idea:
Identify generic mechanisms generating stage II
retinal waves and analyze them mathematically.
Goal of this lecture
10
Motivation
:
• Retinal waves instruct the shaping of the visual
system
Idea:
Identify generic mechanisms generating stage II
retinal waves and analyze them mathematically.
Goal of this lecture
11
Motivation
:
• Retinal waves instruct the shaping of the visual
system
• Understanding the mechanisms that generate
them may help to control them
• New mathematical problems and new techniques.
Idea:
Identify generic mechanisms generating stage II
retinal waves and analyze them mathematically.
12
12
Experiments for the emergence of retinal waves
Experiment for isolated neurons,
Zheng et al., 2006, Nature
Retinal waves require three components:
i)
Spontaneous bursting activity
13
13
Experiment for isolated neurons,
Zheng et al., 2006, Nature
Retinal waves require three components:
i)
Spontaneous bursting activity
ii)
Refractory mechanism (
slow After HyperPolarisation- sAHP
)
14
14
Experiment for coupled and isolated
neurons, Zheng et al., 2006, Nature
Retinal waves require three components:
i)
Spontaneous bursting activity
ii)
Refractory mechanism (
slow After HyperPolarisation- sAHP
)
iii)
Coupling (through Acetylcholine neurotransmitter)
Coupled neurons
synchronize
Coupled neurons
synchronize
Experiments for the emergence of retinal waves
15
15
Experiment for coupled and isolated
neurons, Zheng et al., 2006, Nature
Retinal waves require three components:
i)
Spontaneous bursting activity
ii)
Refractory mechanism (
slow After HyperPolarisation- sAHP
)
iii)
Coupling (through Acetylcholine neurotransmitter)
Coupled neurons
synchronize
Isolated Neurons burst
independently
Coupled neurons
synchronize
Experiments for the emergence of retinal waves
From the non linear physics perspective a retinal wave is
a spatiotemporal activity which:
• is generated by the conjunction of
local (cell level)
nonlinear characteristics and network effects
16
Mathematical characterization
of Retinal Waves
From the non linear physics perspective a retinal wave is
a spatiotemporal activity which:
• is generated by the conjunction of
local (cell level)
nonlinear characteristics and network effects
• can have different
forms
(fronts, spiral, standing
waves, spatio-temporal chaos)
17
Mathematical characterization
of Retinal Waves
From the non linear physics perspective a retinal wave is
a spatiotemporal activity which:
• is generated by the conjunction of
local (cell level)
nonlinear characteristics and network effects
• can have different
forms
(fronts, spiral, standing
waves, spatio-temporal chaos)
• is
induced by generic mechanisms
that can be
studied in the context of dynamical systems theory
18
Mathematical characterization
of Retinal Waves
From the non linear physics perspective a retinal wave is
a spatiotemporal activity which:
• is generated by the conjunction of
local (cell level)
nonlinear characteristics and network effects
• can have different
forms
(fronts, spiral, standing
waves, spatio-temporal chaos)
• is
induced by generic mechanisms
that can be
studied in the context of dynamical systems theory
• can be associated with
bifurcations
19
Mathematical characterization
of Retinal Waves
• Model
biophysically
the dynamics of
isolated neurons
(SACs)
• Carefully model all currents involved
• Fix parameters from experiments
20
Our strategy
• Model
biophysically
the dynamics of
isolated neurons
(SACs)
• Carefully model all currents involved
• Fix parameters from experiments
•
Bifurcation analysis
of the non-linear dynamical
system for isolated neurons
21
Our strategy
• Model
biophysically
the dynamics of
isolated neurons
(SACs)
• Carefully model all currents involved
• Fix parameters from experiments
•
Bifurcation analysis
of the non-linear dynamical
system for isolated neurons
• Extract a
generic biophysical mechanism
of
spontaneous bursting activity for SACs
22
Our strategy
• Model
biophysically
the dynamics of
isolated neurons
(SACs)
• Carefully model all currents involved
• Fix parameters from experiments
•
Bifurcation analysis
of the non-linear dynamical
system for isolated neurons
• Extract a
generic biophysical mechanism
of
spontaneous bursting activity for SACs
• Model the network interactions:
• Study the effects of synaptic coupling to
synchrony and wave initiation
• Search for biophysical parameters which control
spatiotemporal patterns
23
Our strategy
Modelling the bursting of individual Starburst
Amacrine Cells
Bruno Cessac and Dora Karvouniari
Biovision Team, INRIA Sophia Antipolis,France.
1
From biophysics to modelling
2
Model analysis
Biophysics of bursting SAC
Retinal waves require three components:
Spontaneous bursting activity
Refractory mechanism (slow After HyperPolarisation- sAHP)
Coupling (through Acetylcholine neurotransmitter)
Biophysics of bursting SAC
Figure:
Zheng et al., Nature, 2006. Grey curve: Fast oscillations due to
voltage-gated Ca
+2
channels and Ca
+2
dependent K
+
channels. Black
Biophysics of bursting SAC
C
m
dV
dt
= −g
L
(V − V
L
) − g
C
M
∞
(V )(V − V
C
) − g
K
N(V − V
K
)
+I
ex
(•)
τ
N
dN
dt
= Λ(V )(N
∞
(V ) − N)
(1)
Morris-Lecar model with fast Potassium.
Dynamics
C
m
dV
dt
= −g
L
(V − V
L
) − g
C
M
∞
(V )(V − V
C
) − g
K
N(V − V
K
)
−g
sAHP
R
4
(V − V
K
)
τ
N
dN
dt
= Λ(V )(N
∞
(V ) − N)
τ
C
dC
dt
= −
H
α
XCC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
S
dS
dt
= α
S
(1 − S)C
4
− S
τ
R
dR
dt
= α
R
S
(1 − R) − R
(2)
Dynamics
C
m
dV
dt
= −g
L
(V − V
L
) − g
C
M
∞
(V )(V − V
C
) − g
K
N(V − V
K
)
−g
sAHP
R
4
G
sAHP
(R)(V − V
K
)
τ
N
dN
dt
= Λ(V )(N
∞
(V ) − N)
τ
C
dC
dt
= −
H
α
XCC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
S
dS
dt
= α
S
(1 − S)C
4
− S
τ
R
dR
dt
= α
R
S
(1 − R) − R
(3)
Dynamics
Leak time scale: τ
L
=
g
C
L.
Conductances rescaling: ˜
g
X
=
g
g
XL.
τ
L
dV
dt
= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
τ
N
dN
dt
= Λ(V )(N
∞
(V ) − N)
τ
C
dC
dt
= −
H
α
CXC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
S
dS
dt
= α
S
(1 − S)C
4
− S
τ
R
dR
dt
= α
R
S
(1 − R) − R
Multi-scale dynamics
Fast V , N. τ
L
= 11 ms, τ
N
= 5 ms.
Medium C . τ
C
= 2 s.
Fast-scale
t
f
=
τ
t
L⇒ dt
= τ
L
dt
f
dV
dt
f= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
τ
Nτ
LdN
dt
f= Λ(V )(N
∞
(V ) − N)
τ
Cτ
LdC
dt
f= −
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
Sτ
LdS
dt
f= α
S
(1 − S)C
4
− S
τ
Rτ
LdR
dt
f= α
R
S
(1 − R) − R
Fast-scale
Set ǫ
X
=
τ
Lτ
X, X = C , R, S.
ǫ
C
=
11
2
×
10
−
3
∼
5.5 × 10
−
3
,
ǫ
S
= ǫ
R
=
11
44
×
10
−
3
∼
2.5 × 10
−
4
dV
dt
f= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
dN
dt
f=
τ
Lτ
NΛ(V )(N
∞
(V ) − N)
dC
dt
f= ǫ
C
h
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
i
dS
dt
f= ǫ
S
α
S
(1 − S)C
4
− S
dR
dt
f= ǫ
R
[ α
R
S(1 − R) − R ]
Fast-scale: approximation
dV
dt
f= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
dN
dt
f=
τ
Lτ
NΛ(V )(N
∞
(V ) − N)
dC
dt
f= 0
dS
dt
f= 0
dR
dt
f= 0
Morris-Lecar model analysis
dV
dt
f= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
dN
dt
f=
τ
Lτ
NΛ(V )(N
∞
(V ) − N)
(4)
Morris-Lecar model analysis
dV
dt
f= −˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
) + I
ext
dN
dt
f=
τ
Lτ
NΛ(V )(N
∞
(V ) − N)
(5)
Bifurcation analysis when varying the parameter I
ext
.
Morris-Lecar model analysis
The saddle-node bifurcation
Figure:
From https://elmer.unibas.ch/pendulum/bif.htm and
The Hopf bifurcation
Figure:
From Navarro et al, ”Control of the Hopf bifurcation in the
Takens-Bogdanov bifurcation”, CDC2008
The homoclinic bifurcation
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
The complete movie
Karvouniari et al, ICMNS, 2016
The complete movie
The complete movie
Back to Zhou experiment
Figure:
Zheng et al., Nature, 2006. Grey curve: Fast oscillations due to
voltage-gated Ca
+2
channels and Ca
+2
dependent K
+
channels. Black
Back to Zhou experiment
Back to Zhou experiment
Karvouniari et al, submitted to Plos Comp. Bio.
0 50 100 150 Time (ms) −50 0 V (mV) Oscillations Block Ca++ Block K+
Figure:
Left. Green.Application of a current step pulse of 150 pA for 60ms. Orange: Oscillations disappearwhen Ca+2related conductances are set to zero. Red. Blocking the oscillations upon setting the voltage-gated K + conductance to zero. Right: Zheng et al. 2006 experimental figure. Grey curve: Fast oscillations due to voltage-gated Ca+2channels and Ca+2dependent K+channels. Black curve: Application of Cd+blocking Ca+2 related channels.
Dynamics of bursting
τ
L
dV
dt
=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
τ
N
dN
dt
=
Λ(V )(N
∞
(V ) − N)
τ
C
dC
dt
=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
S
dS
dt
=
α
S
(1 − S)C
4
− S
τ
R
dR
dt
=
α
R
S(1 − R) − R
Fast V , N. τ
L
= 11 ms, τ
N
= 5 ms.
Medium C . τ
C
= 2 s.
Slow S , R. τ
R
= τ
S
= 44 s.
Slow time-scale
t
s
=
τ
t
S⇒ dt
= τ
S
dt
s
τ
Lτ
SdV
dt
s=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
τ
Nτ
SdN
dt
s=
Λ(V )(N
∞
(V ) − N)
τ
Cτ
SdC
dt
s=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
s=
α
S
(1 − S)C
4
− S
dR
dt
s=
α
R
S(1 − R) − R
Slow time-scale
ǫ
V
=
τ
Lτ
S∼
2.5 × 10
−
4
,
ǫ
N
=
τ
Nτ
S∼
1 × 10
−
3
ǫ
C
=
τ
Cτ
S∼
4.5 × 10
−
2
ǫ
V
dV
dt
s=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
ǫ
N
dN
dt
s=
Λ(V )(N
∞
(V ) − N)
ǫ
C
dC
dt
s=
−
H
α
XCC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
s=
α
S
(1 − S)C
4
− S
dR
dt
s=
α
R
S(1 − R) − R
Slow time-scale approximation
0
=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
0
=
Λ(V )(N
∞
(V ) − N)
0
=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
s=
α
S
(1 − S)C
4
− S
dR
dt
s=
α
R
S
(1 − R) − R
Steady state
V
=
V
ML
(V ) + ˜
G
sAHP
(R)V
K
˜
g
ML
(V ) + ˜
G
sAHP
(R)
where:
V
ML
(V ) = V
L
+ ˜
g
C
M
∞
(V )V
C
+ ˜
g
K
N
∞
(V )V
K
;
˜
g
ML
(V )
= 1 + ˜
g
C
M
∞
(V ) + ˜
g
K
N
∞
(V ),
Steady state
V
follows ”adiabatically” R.
V ≡ f
(V , R) =
V
ML
(V ) + ˜
G
sAHP
(R)V
K
˜
g
ML
(V ) + ˜
G
sAHP
(R)
(6)
Steady state
V
follows ”adiabatically” R.
V ≡ f
(V , R) =
V
ML
(V ) + ˜
G
sAHP
(R)V
K
˜
g
ML
(V ) + ˜
G
sAHP
(R)
(6)
Implicit function theorem.
Wherever
∂f
∂V
6= 1 V follows a unique and differentiable branch
V ≡ V
(R). At points R where
∂f
∂V
= 1 this branch may disappear
Hint
V ≡ f
(V , R) =
V
ML
(V ) + ˜
G
sAHP
(R)V
K
˜
g
ML
(V ) + ˜
G
sAHP
(R)
(7)
Taking the differential gives:
dV
1 −
∂f
∂V
=
∂f
∂R
dR
⇒
dV
dR
=
∂f
∂R
1 −
∂V
∂f
Morris-Lecar model analysis
Medium-scale
t
m
=
τ
t
C⇒ dt
= τ
C
dt
m
τ
Lτ
CdV
dt
m=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
τ
Nτ
CdN
dt
m=
Λ(V )(N
∞
(V ) − N)
dC
dt
m=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
τ
Sτ
CdS
dt
m=
α
S
(1 − S)C
4
− S
τ
Rτ
CdR
dt
m=
α
R
S
(1 − R) − R
Medium-scale
ǫ
V
=
τ
Lτ
C∼
5 × 10
−
3
,
ǫ
N
=
τ
Nτ
C∼
2.5 × 10
−
3
ǫ
S
= ǫ
R
=
τ
Cτ
S∼
4.5 × 10
−
2
ǫ
V
dt
dV
m=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
ǫ
N
dt
dN
m=
Λ(V )(N
∞
(V ) − N)
dC
dt
=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
m=
ǫ
S
α
S
(1 − S)C
4
− S
dR
dt
m=
ǫ
R
[ α
R
S
(1 − R) − R ]
Medium-scale approximation
0
=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
0
=
Λ(V )(N
∞
(V ) − N)
dC
dt
m=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
=
0
dR
dt
=
0
Medium-scale approximation
At this time scale:
R, S
are constant;
V
follows the variations of R;
dC
dt
m= −
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
Medium-scale approximation
Rapid (medium time scale between upper/lower branch) ”Outer
dynamics”.
Slow time-scale approximation
0
=
−˜
g
L
(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − ˜
g
K
N(V − V
K
)
− ˜
G
sAHP
(R)(V − V
K
)
0
=
Λ(V )(N
∞
(V ) − N)
0
=
−
H
Xα
CC
+ C
0
− δ
C
g
C
M
∞
(V )(V − V
C
)
dS
dt
s=
α
S
(1 − S)C
4
− S
dR
dt
s=
α
R
S
(1 − R) − R
”Inner dynamics”
V
follows R;
C
follows V ;
Note P(R) ≡ C
4
, the average value of C
4
for a given value of
R.
⇒
Dynamics on the slow branches.
dR
dt
s= α
R
S
(1 − R) − R
dS
dt
s= α
S
(1 − S)P(R) − S
”Inner dynamics”
Nullclines.
R
Nullcline:
dR
dt
s= 0 ⇒
The vector field is vertical on the line S =
α
R(1−R)
R
S
Nullcline:
dt
dS
s
= 0 ⇒
The vector field is horizontal on the line S =
α
SP(R)
Conclusions
Conclusions
We have now a model reproducing individual SACs bursting.
Here SACs burst periodically with a frequency determined by
the parameters of the model (mainly τ
R
).
Conclusions
We have now a model reproducing individual SACs bursting.
Here SACs burst periodically with a frequency determined by
the parameters of the model (mainly τ
R
).
A small change of parameters make them into a rest state
where they burst only upon an external excitation (noise).
Conclusions
We have now a model reproducing individual SACs bursting.
Here SACs burst periodically with a frequency determined by
the parameters of the model (mainly τ
R
).
A small change of parameters make them into a rest state
where they burst only upon an external excitation (noise).
What happens when they cells are coupled with
Acetylcholine ?
Modelling stage II retinal waves
Bruno Cessac and Dora Karvouniari
Biovision Team, INRIA Sophia Antipolis,France.
15-01-2017
1
Coupling SACs with Acetylcholine
2
Dynamical systems analysis
3
A simplified setting to mathematically study retinal waves
4
Generalisations
Coupling SACs with Acetylcholine
Coupling SACs with Acetylcholine
Coupling SACs with Acetylcholine
Figure:
Nicotinic receptors and conductance.
Coupling SACs with Acetylcholine
Coupling SACs with Acetylcholine
Figure:
Equation of membrane potential.
Main currents
G
sAHP
(R
i
) = g
sAHP
R
i
4
,
(1)
sAHP conductance,
G
A
({A
k
}
k∈B
i) = g
A
X
k∈B
iU(A
k
),
(2)
Ach synaptic conductance with:
U
(A) =
A
2
γ
A
+ A
2
,
(3)
B
i
is the set of index of neurons connected to i.
Dynamics
C
m
dV
dt
i= −g
L
(V
i
− V
L
) − g
C
M
∞
(V
i
)(V
i
− V
C
) − g
K
N
i
(V
i
− V
K
)
−G
sAHP
(R
i
)(V
i
− V
K
) − G
A
({A
k
}
k∈B
i)(V
i
− V
A
)
τ
N
dN
dt
i= Λ(V
i
)(N
∞
(V
i
) − N
i
)
τ
C
dC
dt
i= −
H
α
CXC
i
+ C
0
− δ
C
g
C
M
∞
(V
i
)(V
i
− V
C
)
τ
S
dS
dt
i= α
S
(1 − S
i
)C
i
4
− S
i
τ
R
dR
dt
i= α
R
S
i
(1 − R
i
) − R
i
d A
idt
= −µA
i
+ β
A
T
A
(V
i
),
(4)
Dynamics
Leak time scale: τ
L
=
g
C
L. Ach time scale: τ
A
=
µ
1
.
Conductances rescaling: ˜
g
X
=
g
g
XL.
τ
L
dV
dt
i=
−(V
i
− V
L
) − ˜
g
C
M
∞
(V
i
)(V
i
− V
C
) − ˜
g
K
N
i
(V
i
− V
K
)
− ˜
G
sAHP
(R
i
)(V
i
− V
K
) − ˜
G
A
({A
k
}
k∈B
i)(V
i
− V
A
);
τ
N
dN
dt
i=
Λ(V
i
)(N
∞
(V
i
) − N
i
);
τ
C
dC
dt
i=
−
H
α
CXC
i
+ C
0
− δ
C
g
C
M
∞
(V
i
)(V
i
− V
C
);
τ
S
dS
dt
i=
α
S
(1 − S
i
)C
i
4
− S
i
;
τ
R
dR
dt
i=
α
R
S
i
(1 − R
i
) − R
i
;
τ
A
d A
dt
i=
−A
i
+
β
µ
AT
A
(V
i
);
Multi-scale dynamics
Fast V , N. τ
L
= 11 ms, τ
N
= 5 ms.
Medium C , A. τ
C
= 2 s, τ
A
= 1.86 s.
A simplified setting to mathematically study retinal waves
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
A simplified setting to mathematically study retinal waves
Main hypotheses:
1) We neglect the fast Potassium current. V
i
has only two stable states:
A simplified setting to mathematically study retinal waves
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
Main hypotheses:
2) To switch from low to high state, neuron i needs an external excitatory
current. This can be:
The Ach current coming from neighbours;
A ”shot noise” current of the form:
g
N
B
i
(V
i
− V
C
)
where B ∈ {0, 1} is a Bernoulli variable with probability
Prob
[ B
i
(s) = 1, s ∈ [t, t + dt ] = pdt, with p ∼
τ
1
R.
A simplified setting to mathematically study retinal waves
Main hypotheses:
3) sAHP has a simplified form (does not depend on Calcium). We only
keep the variables S, R where S depends directly on V :
τ
S
d S
i
dt
= −S
i
+ γG (V
i
)
with:
G
(V
i
) =
1
1 + e
−
κ(V −V
0)
.
A simplified setting to mathematically study retinal waves
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
Main hypotheses:
4) Acetylcholine synapses are
assumed
not to be yet functional
(Ford-Feller, 2011). The main source of Ach coupling is Ach diffusion.
τ
A
d A
i
dt
= −A
i
+
β
A
µ
T
A
(V
i
) + D∆A
i
;
where ∆ is the Laplacian on a square lattice.
Boundary conditions are ignored.
A simplified setting to mathematically study retinal waves
1
Two state voltage.
2Shot noise.
3
Simplified form for sAHP.
4Ach diffusion.
Dynamics
From now we consider that variables depend continuously on space
⇒
dt
d
→
∂t
∂
τ
L
∂V
∂t
=
−(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) + ˜
g
N
B(V − V
C
)
− ˜
G
sAHP
(R)(V − V
K
) − ˜
g
A
U
(A)(V − V
A
);
τ
S
∂S
∂t
=
−S + γG (V );
τ
R
∂R
∂t
=
α
R
S
(1 − R) − R;
τ
A
∂A
∂t
=
−A +
β
µ
AT
A
(V ) + D∆A(V );
Medium-scale
t
m
=
τ
t
A⇒ dt = τ
A
dt
m
ǫ
V
=
τ
τ
LA∼ 5 × 10
−
3
;
ǫ
S
= ǫ
R
=
τ
τ
AS∼ 4.5 × 10
−
2
ǫ
V
∂t
∂V
m=
−(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) − g
N
B(V − V
C
)
− ˜
G
sAHP
(R)(V − V
K
) − ˜
g
A
U
(A)(V − V
A
);
∂S
∂t
m=
ǫ
S
[ −S + γG (V ) ] ;
∂R
∂t
m=
ǫ
R
[ α
R
S(1 − R) − R ] ;
∂A
∂t
m=
−A +
β
Aµ
T
A
(V ) + D∆A;
Medium-scale approximation
0
=
−(V − V
L
) − ˜
g
C
M
∞
(V )(V − V
C
) + g
N
B(V − V
C
)
− ˜
G
sAHP
(R)(V − V
K
) − ˜
g
A
U
(A)(V − V
A
);
∂S
∂t
m=
0;
∂R
∂t
m=
0;
∂A
∂t
m=
−A +
β
Aµ
T
A
(V ) + D∆A;
Medium-scale approximation
At this time scale:
R, S
are constant;
V
follows the variations of R, A;
V
=
V
L
+ ˜
g
C
M
∞
(V )V
C
+ ˜
g
N
BV
C
+ ˜
G
sAHP
(R)V
K
+ ˜
g
A
U(A)V
A
1 + ˜
g
C
M
∞
(V ) + ˜
g
N
B
+ ˜
G
sAHP
(R) + ˜
g
A
U
(A)
Acetylcholine front propagation
We consider here a propagation in 1 dimension.
∂A
∂t
m
= −A +
β
A
µ
T
A
(V ) + D∆A ⇒
∂A
∂t
m
= −A +
β
A
µ
T
A
(V ) + D
∂
2
A
∂x
2
Spatially homogeneous state
Low voltage state. If all neurons are in the low voltage state, V
−
,
∆A = 0 ⇒ A reaches a stationary homogeneous state on the
medium time scale.
A
−
=
β
A
µ
T
A
(f (R, A
−
,
0)),
where 0 means that B = 0 everywhere (no shot noise current).
In this state, the S production G (V
−
) ∼ 0 so there is no sAHP on
Spatially homogeneous state
High voltage state. If all neurons are in the high voltage state,
V
+
, ∆A = 0 ⇒ A reaches a stationary homogeneous state on the
medium time scale.
A
+
=
β
A
µ
T
A
(f (R, A
+
, B)).
Here, the state can be reached either due to shot noise or to Ach
current.
In this state, the S production G (V
+
) >> 0 so there is an sAHP
on the slow time scale.
Acetylcholine front propagation
Local perturbation of the low state. At some point x
0
we
induce a shot noise current at time 0 (B(x
0
,
0) = 1).
This raises the voltage to its upper value V
+
.
This increases the Ach production at x
0
.
Ach diffuses to neighbours and induces an excitatory current.
If g
A
conductance is large enough this current raises the
neighbours voltage to V
+
.
This can lead to a propagating wave.
Under which conditions?
Acetylcholine front propagation
Moving frame:
ξ
= x − ct
m
where c is a constant ⇒ dξ = dx − cdt
m
.
For a function A(x, t
m
) ≡ A(ξ) we have:
dA
=
∂A
∂ξ
d ξ
=
∂A
∂ξ
( dx − cdt
m
) ⇒
∂A
∂x
=
∂A
∂ξ
;
;
∂
2
A
∂x
2
=
∂
2
A
∂ξ
2
;
∂A
∂t
m
= −c
∂A
∂ξ
−c
∂A
∂ξ
= −A +
β
A
µ
T
A
(f (R, A, B)) + D
∂
2
A
∂ξ
2
Acetylcholine front propagation
Potential well:
∂V
∂A
= A −
β
Aµ
T
A
(f (R, A, B))
V(A) =
A
2
2
−
β
A
µ
Z
T
A
(f (R, a, B))da + K
Rest states:
∂V
∂A
= 0.
Newton equation
D
∂
2
A
∂ξ
2
= −c
∂A
∂ξ
−
∂V
∂A
.
We note
∂A
∂ξ
≡ ˙
A;
∂A
∂ξ
22≡ ¨
A
D ¨
A
= −c ˙
A
−
∂V
Acetylcholine front propagation
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
D ¨
A
= −c ˙
A
−
∂V
∂A
.
Formally, this is the equation of motion of a particle, with mass D, in a
potential well V, with a friction coefficient c.
V has 3 extrema, given by
∂V
∂A
= A −
β
Aµ
T
A
(f (R, A, B)) = 0 ⇒:
A
=
β
A
µ
T
A
(f (R, A, B))
These are the fixed points of the medium scale local dynamics (∆A = 0).
Maxima of V correspond to stable fixed points.
Effect of sAHP
Figure:
Bifurcation diagram in the space I − V .
Effect of sAHP
Effect of sAHP
Figure:
Pulse propagation.
Effect of sAHP
Time and space scales
Ach raising and propagation, controlled by g
Ach
and D
(Diffusion);
sAHP controlled by g
sAHP
(Local);
Time of excitation controlled by characteristic times and
conductances;
Time of propagation controlled by c depending on the other
parameters.
Generalisations
Relaxing hypotheses
1
Two state voltage.
2Shot noise.
3
Simplified form for sAHP.
4Ach diffusion.
Relaxing hypotheses
1
Two state voltage. ⇒ Fast bursting in the high voltage state.
2Shot noise.
3
Simplified form for sAHP.
4Ach diffusion.
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
Relaxing hypotheses
1
Two state voltage. ⇒ Fast bursting in the high voltage state.
2Shot noise. ⇒ Spontaneous bursting.
3
Simplified form for sAHP.
4Ach diffusion.
Relaxing hypotheses
1
Two state voltage. ⇒ Fast bursting in the high voltage state.
2Shot noise. ⇒ Spontaneous bursting.
3
Simplified form for sAHP. Role of Ca ?
4Ach diffusion.
Bruno Cessac and Dora Karvouniari Modelling stage II retinal waves
Relaxing hypotheses
1
Two state voltage. ⇒ Fast bursting in the high voltage state.
2Shot noise. ⇒ Spontaneous bursting.
3