positive feedback and synchronized bursts in neuronal...
TRANSCRIPT
Positive Feedback and Synchronized Bursts in Neuronal Cultures
陳志強
C. K. Chan
Institute of Physics, Academia Sinica
National Central University, Taiwan
中央研究院 物理研究所
KIAS 2015
Collaborators P. Y. Lai, Central Univ.
Y. T. Huang, Central Univ.
C. T. Huang,
Academia Sinica
C. C. Chen,
Academia Sinica
Moore’s Law
https://humanswlord.files.wordpress.com/2014/01/moores-law-graph-gif.png
Neuromorphic Computation IBM: True North Chip (2014)
1 million programmable neurons and 256 million programmable synapses.
5.4 billion transistors at 28nm, < 70 mw, size of a stamp
http://www.research.ibm.com/articles/brain-chip.shtml
Living Networks
• Connections
- small world, scale free, random …
- 2D, 3D
- feed forward, recurrent …
• Synaptic Dynamics
- facilitation
- depression
Dynamics of Neural Cultures: Synchronized Bursting
Multi-Electrode Array
http://www.multichannelsystems.com/
Synchronized Burst (SB) of a neuronal network • Neuronal network : excitable network
• Brain, slice, dissociated neuronal network
• SB : synchronization + neuron burst
5 s
40 mV
Wagenaar, PRE 73, 1539 (2006)
(discrete repeated firings)
firing rate
raster plot
Cultures play rich synchronized burst (SB) dynamics
Daniel A. Wagenaar , BMC Neuroscience2006, 7:11
firing rate change with days
Dissociated neuronal cultured network is a good material to study SB
High density culture recorded by MEA
Basic Questions
• What are the basic dynamics of these living neuronal network?
• Can memories be written and read back from them?
• Can these networks perform useful functions?
SB has hidden information?
Gordon the Robot (2008) https://www.youtube.com/watch?v=1-0eZytv6Qk
Why we are not seeing more of Gordon?
Outline
• Properties of Synchronized Bursts
• Mean field model of firing rate with synaptic plasticity (TM model)
• Extended TM model (TMX)
• Implications of TMX model
• Summary
Neuronal cultures and MEA Wistar rat E17 cortical cells
Density : 2500 cells/mm2 ; Area : 5 mm2
Coating : 0.1 % PEI
Medium : DMEM + 5%FBS + 5%HS
Incubator : 37 ℃ and 5% CO2
3 – 5 neurons / electrode
200 μm 40 μm
MEA60 system (MEA1060-Inv-BC + Qwane ITO-200 chip)
Sampling rate : 20K Hz Chip : 60 electrodes (8×8)
Diameter : 40 μm Spacing : 200 μm
13 DIV 9 DIV
Synchronized burst firing histogram (SBFH)
Co
un
ts /
5 m
s
spike detection
0 s 90 s
Raster plot of a SB
τIBI
elec
tro
des
Firing rate histogram
τB
Characteristic Patterns
Reverberation Is Enhanced by Decreasing Magnesium Ions
Increasing NMDAR Conductance
Reverberation Is Suppressed by Increasing Glutamate
Increasing System Noise
Reverberation Is Enhanced by Increasing Bicuculline
Decreasing Inhibition Connections
Summary
DIV ↑ noise ↑ (Glu ↑)
Conductance
↑ (Mg↓)
Inhibition ↓ (BIC ↑)
Firing rate ↑ ↑ ↑ ↑
Bursting rate ↑ ↑ ↑ ―
Burst duration
↓ ↓ ↑ ↑
reverberation ↓ ↓ ↑ ↑
Demo of positive feedback in the Panther CGS765 tube VCA
https://www.youtube.com/watch?v=S6_AF28BrQM
Mean field description
E(t)
Network
E’(t)
synchronized burst
similar activity
rate model
mean-field model
Positive Feedback
If E’(t) > E(t)
Recurrent connection
Short term synaptic facilitation
ISRN Neurology, Sumiko Mochida,Volume 2011 (2011), Article ID 919043, 7 pages
[Ca2
+ ]
Resting Ca2+
~ 100 nM
Residual Ca2+
~ 1 μM
1st action potential 2nd action potential
(a)
(b)
Ca2+
1st action potential Residual Ca2+ 2nd action potential Resting Ca2+
Short term synaptic depression
Blitz, Dawn M. et al. Nat. Rev. Neurosci. (5), 2004
Tsodyks-Markram (TM) model I
𝑑𝑥
𝑑𝑡=
1−𝑥
𝜏𝐷− 𝑢𝑥𝐸
𝑑𝑢
𝑑𝑡=
𝑈−𝑢
𝜏𝑓+ 𝑈(1 − 𝑢)𝐸
Firing rate
Depression
Facilitation
𝑑𝐸
𝑑𝑡=
1
𝜏[−𝐸 + 𝑏 𝑙𝑛 1 + 𝑒
𝐽𝑢𝑥𝐸−𝐼0𝑏 ]
Cortes, Jesus M., et al. PNAS 110.41 (2013): 16610-16615.
Tsodyks-Markram (TM) model II
High firing Steady state
Oscillatory steady state
Low firing steady state
(will not stop!)
I0 = -1
I0 = -1.3
I0 = -2
TMX model
𝑑𝑥
𝑑𝑡=
𝜒0−𝑥
𝜏𝐷− 𝑢𝑥𝐸
𝑑𝑢
𝑑𝑡=
𝑈−𝑢
𝜏𝑓+ 𝑈(1 − 𝑢)𝐸
𝑑𝜒0
𝑑𝑡=
𝑋0−𝜒0
𝜏𝑥− 𝛽𝐸 𝜏𝑥 : recovery time constant of 𝜒0 (20 s)
β : slow depletion rate (0.01)
𝑥 : fraction of available resources
𝑢 : release probability
𝜒0 : baseline of 𝑥
𝑈 : baseline of 𝑢 (0.3)
𝜏𝐷 : recovery time constant of 𝑥 (0.15 s)
𝜏𝐹 : recovery time constant of 𝑢 (1.5 s)
𝐽 : strength of recurrent connections
𝑏 : neuronal gain function (1.5)
𝐸 : firing rate
𝐼0 : inhibition current (-1.3)
𝑋0 : maximum available resources slow
depletion
firing rate
available resource
release probability
𝜏 : recovery time constant of 𝑥 (20 s)
𝑑𝐸
𝑑𝑡=
1
𝜏[−𝐸 + 𝑏 𝑙𝑛 1 + 𝑒
𝐽𝑢𝑥𝐸−𝐼0𝑏 ]
Simulated SBFH at different DIV
J = 5.8, 𝜏𝐷 = 0.15 s and U = 0.3.
J = 4.8, 𝜏𝐷 = 0.2 s and U = 0.28.
J = 6.8, 𝜏𝐷 = 0.1 s and U = 0.32
χ0 is the main factor to terminate the SB
X0 : Maximum available resources
SB initiates by a big enough positive feedback
𝑑𝐸
𝑑𝑡=
1
𝜏[−𝐸 + 𝑏 𝑙𝑛 1 + 𝑒
𝐽𝑢𝑥𝐸−𝐼0𝑏 ]
How does SB generate?
Faster recovery controls the firing patterns
E(H
z)
E(H
z)
𝑑𝑥
𝑑𝑡=
𝜒0−𝑥
𝜏𝐷− 𝑢𝑥𝐸
𝑑𝑢
𝑑𝑡=
𝑈−𝑢
𝜏𝑓+ 𝑈(1 − 𝑢)𝐸
How do reverberations generate?
χ0 is the main factor to terminate the SB
𝑑𝜒0
𝑑𝑡=
𝑋0−𝜒0
𝜏𝑥− 𝛽𝐸
How does SB terminate?
Physical Origin of slow recovery
Role of Astrocytes?
Astrocyte modulates neuronal activity by various gliotransmitters
KERRI SMITH, NATURE(468), Nov. 2010
Glial mechanism for Slow recycling
Blitz, Dawn M. et al. Nat. Rev. Neurosci. (5), 2004
Ratio of Glia to Neurons More glia higher intelligence ?
Exp Neurol. 1985 Apr;88(1):198-204.
Human astrocytes are larger and more complex than those of infra primate mammals, suggesting that their role in neural processing has expanded with evolution. To assess the cell-autonomous and species selective properties of human glia, we engrafted human glial progenitor cells (GPCs) into neonatal immuno deficient mice. Upon maturation, the recipient brains exhibited large numbers and high proportions of both human glial progenitors and astrocytes. The engrafted human glia were gap-junction coupled to host astroglia, yet retained the size and pleomorphism of hominid astroglia, and propagated Ca2+ signals 3-fold faster than their hosts. Long-term potentiation (LTP) was sharply enhanced in the human glial chimeric mice, as was their learning, as assessed by Barnes maze navigation, object-location memory, and both contextual and tone fear conditioning. Mice allografted with murine GPCs showed no enhancement of either LTP or learning. These findings indicate that human glia differentially enhance both activity-dependent plasticity and learning in mice
Cell Stem Cell 12, 342–353, March 7, 2013 ª2013 Elsevier Inc.
Implications
• SB might be a sign of too much feedback (connections) similar to some types of epilepsy
• Slow recycling mechanism might be carried out by astrocytes
• Training protocol should be done in a low active neural network. • Decrease number of glia
• 3D cultures
• Control network topology
• Need glia in True North?