1. Influence of Connectivity on Activity
Levels in Patterned Neuronal Networks
Sankaraleengam Alagapan
Wheeler Lab
J Crayton Pruitt Family Department of
Biomedical Engineering
2. Brain on a Chip
• Confluence of Technologies
– In vitro neural culture,
Microelectrode Arrays
(MEAs), Substrate MEA Dissociated Neuronal Culture
Modification, Microfluidic
Devices
• Simplified small scale
model of brain PDL –PLL Pattern Microtunnel Device
• Useful in drug screening, in
vitro models of pathologies,
basic neuroscience
Extracellular recordings from MEA
Introduction 2
3. Brain on a Chip
• Patterning Patterned Network – 4 Connect
Control the structure of the
network – amount of Patterned Network – 8 Connect
convergence and divergence
• Microtunnel devices
Control the direction of
information flow- create sub-
networks where one drives the
other
• Understand the influence of
structure on network function Microtunnel device
Pictures by Eric and Kucku
Introduction 3
4. Connectivity in Neuronal Networks
• Structural Connectivity
– Anatomical Connections - Studied using staining, tracing
etc
• Functional Connectivity
– Statistical measure of temporal correlations in activity –
“Things that are wired together fire together” – E.g. –
Correlation, Coherence, Mutual Information
• Effective Connectivity
– Gives an idea of which region of the network drives and
which region is being driven – Combination of both
structure and function – E.g. Transfer Entropy, Granger
Causality
Background 4
5. Connectivity and Function
• Structure plays an important role in enabling a
particular function in vivo. E.g., Cerebellum
• Small world architecture develops naturally in
dissociated cultures and this architecture plays a role
in the self-sustained activity patterns observed in
such cultures (Srinivas et al. 2007)
Background 5
6. Connectivity and Activity Level
• Which network is more active? i.e., which network
will have higher average firing rate?
Homeostatic
Plasticity
=
More interconnections – Degree of Fewer interconnections – Degree of
Connectivity is high Connectivity is low
Intuition
Hypothesis 6
8. Hypothesis
Activity level in neuronal networks is governed
by the overall strength of connectivity in the
network
Hypothesis 8
9. Specific Aims
• Obtain an optimal measure of strength connectivity
by comparing different measures on data from living
networks
• Study the relation between connectivity strength and
the degree of convergence in the network.
• Study the effect of stimulation on connectivity
strengths and influence of degree of convergence on
this effect
Aims 9
10. Specific Aims
• Obtain an optimal measure of strength connectivity
by comparing different measures on data from living
networks
• Study the relation between connectivity strength and
the degree of convergence in the network.
• Study the effect of stimulation on connectivity
strengths and influence of degree of convergence on
this effect
Aims 10
11. Measures of Connectivity
Requirements for an Optimal Measure:
• Reveal the underlying structure as clearly as
possible
– Should measure the strength of connection between two
neurons/nodes in both directions
– Should eliminate the effects of other neurons/nodes as
much as possible
Cross Correlogram
Joint Entropy
Granger Causality
Background 11
12. Measures of Connectivity
Cross Correlogram:
• Measures/shows how the
spikes of one neuron is
distributed in time with
respect to another.
• Peaks measure of the
strength of connection
• Delay corresponding to
peaks Idea of direction of
information flow
Background 12
13. Measures of Connectivity
Joint Entropy:
Entropy measure of the cross inter-spike intervals (cISI)
between two spike trains X,Y
n
JE ( X , Y ) = −∑ p (cISI k ).log 2 ( p (cISI k ))
k =1
p(cISIk) is the estimated probability of cISIk
Background 13
14. Measures of Connectivity
Granger Causality:
• Suppose X and Y are 2 time series modeled as
autoregressive processes, Y G-causes X if the including past
of Y in modeling X decreases the variability of residuals in
the model and vice versa.
• The amount by which the variability is reduced gives a
measure of strength and direction is revealed in the relative
strengths
• Conditional Granger Causality: Same idea as Granger, with
both X and Y conditioned on another variable Z which
might influence the two time series
Background 14
15. Measures of Connectivity
• Garofalo et al (2009) compared the performance of
Crosscorrelogram(CC), Mutual Information (MI),
Joint Entropy(JE) and Transfer Entropy(TE) in
simulated networks
– Showed TE performed better than the other measures
– MI had the worst performance
• Barnett et al (2009) proved mathematically that TE
and Granger Causality are the same measure for
Gaussian Processes
Background 15
16. Aim 1 Experiment 1
Line Pattern
A B C
Why line patterns?
• A unique structure which can constrain neurons in such a
way that strength of AB > strength of AC more often
Research Design 16
17. Aim 1 Experiment 1
• Construct line patterned networks
• Record spontaneous activity at ~DIV 21
• Measure connectivity strengths
• Check consistency among measures
Validation
• Stimulate spontaneously active nodes and observe
evoked responses from other active nodes
• If response is evoked consistently from other nodes,
the stimulated node is connected with these nodes
Research Design 17
18. Patterned Line Networks
1 2 3 4 5 DIV 3 6
• Patterned Networks with Line
2
Patterns
3
• Activity Recorded DIV 24
4
5
6
27 37 57
7
8
9
310
10
Preliminary Results 18
21. Aim 1 Experiment 2
• Alternate Approach: Use of two-welled microtunnel devices
• Plating cells in wells with few days interval leads to axon
growth through tunnels predominantly in one direction
• Strength of A B > Strength of B A i.e., Network A
affects network B more than network B affecting network A
• Electrodes under microtunnels capture axonal propagation
of action potential and these act as a model of two
interacting nodes and measures can
Microwell B
(Output Well)
be tested in this model
DIV 10
Microtunnels
Microwell A
(Input Well)
DIV 20
Arrow indicates direction
100 µm of growth of axons
Research Design 21
22. Microtunnel Data
• Conditional Granger Causality and Cross Correlograms
• Microtunnel Devices – Tunnel Data
Preliminary Results 22
23. Microwell Data
• When bin size = 1ms, interactions in microwell not
evident (lesser causal values)
Bin size 10 ms Bin size 1 ms
Tunnels
Preliminary Results 23
24. Microwell Data
• When bin size = 10ms, interactions between wells
have causal values higher than those within tunnels
Tunnels
Causal values from A B
greater than BA
Preliminary Results 24
25. Specific Aims
• Obtain an optimal measure of strength connectivity
by comparing different measures on data from living
networks
• Study the relation between connectivity strength and
the degree of convergence in the network.
• Study the effect of stimulation on connectivity
strengths and influence of degree of connectivity on
this effect
Aims 25
26. Convergence and Connectivity
strength
• Higher convergence More pathways between
nodes More possibility for correlated activity
Higher connection strength
Mean Connection Strength ∝ Convergence
• Convergence controlled in patterned networks and
connection strengths can be compared
Research Design 26
27. Aim 2 Experiment
• Construct patterned networks with different
convergence 2,4,8 and random
• Spontaneous and evoked activity from DIV 21
• Compute mean connectivity strengths for each
network using the measures
• Test for statistically significant difference between
connectivity strengths of different patterns
• Compute mean firing rate for each network
• Test for Connectivity strength = k x Convergence
Research Design 27
28. Specific Aims
• Obtain an optimal measure of strength connectivity
by comparing different measures on data from living
networks
• Study the relation between connectivity strength and
the degree of convergence in the network.
• Study the effect of stimulation on connectivity
strengths and influence of degree of connectivity on
this effect
Aims 28
29. Stimulation of cultured networks
• Activity dependent plasticity in neurons: “Things
that fire together, wire together” (Hebbian Theory)
• Electrical stimulation analogous to external stimuli
and has been used to induce LTP (High Frequency)
and LTD (Low Frequency) in slices
• Induce a change in plasticity (connection strength)
in in vitro dissociated networks through stimulation
Background 29
30. Stimulation induced change in firing
rate
• Jimbo et al. (1999)
stimulated in vitro
networks with a tetanus
Potentiated pulse and found that there
was a pathway specific
long term change in the
firing rate of neurons
Depressed
• Change in the post
synaptic currents
confirming a change in the
plasticity of the synapses
Background 30
31. Stimulation induced change in
connectivity measures - GC
• Cadotte et al (2008)
repeated Jimbo’s experiment
and used Granger Causality/
Conditional Granger
Causality to measure the
changes in the network
• Confirmed Jimbo’s results
as well showed change in
Granger/Conditional
Granger values before and
after tetanus
Background 31
32. Aim 3 Experiment
• Measure connectivity strengths in networks of Aim
2
• Induce change in connection strength using tetanic
stimulation
• Measure connectivity strengths again and compare
against pre-tetanic connectivity strength
• Compare change in connectivity strengths in the
different networks
Research Design 32
33. References
Cadotte AJ, Demarse TB, He P, Ding M. Causal Measures of Structure and Plasticity in Simulated and Living
Neural Networks. PLOS One. 2008;3(10).
Jimbo Y, Tateno T, Robinson HP. Simultaneous induction of pathway-specific potentiation and depression in
networks of cortical neurons. Biophysical journal. 1999;76(2):670-8.
Srinivas KV, Jain R, Saurav S, Sikdar SK. Small-world network topology of hippocampal neuronal network
is lost, in an in vitro glutamate injury model of epilepsy. The European journal of neuroscience.
2007;25(11):3276-86.
Garofalo M, Nieus T, Massobrio P, Martinoia S. Evaluation of the performance of information theory-based
methods and cross-correlation to estimate the functional connectivity in cortical networks. PloS one.
2009;4(8):e6482.
Perkel DH, Gerstein GL, Moore GP. Neuronal spike trains and stochastic point processes. II. Simultaneous
spike trains. Biophysical journal. 1967;7(4):419-40.
Dworak B, Varghese K, Pan L, Brewer G, Wheeler BC. Creating Unidirectional Neural Networks on a Chip.
In: Proceedings of MEA2010. Reutlingen, Germany; 2010:320-21.
References 33