Understanding the role of astrocytic GABA in simulated neural
networks*
Kerstin Lenk, Eero Risnen, and Jari AK Hyttinen, Member, IEEE
Abstract Astrocytes actively influence the behavior of the
surrounding neuronal network including changes of the
synaptic plasticity and neuronal excitability. These dynamics
are altered in diseases like Alzheimers, where the release of the
gliotransmitter GABA is increased by affected, so called
reactive astrocytes. In this paper, we aim to simulate a neural
network with altered astrocytic GABA release. Therefore, we
use our developed neuron-astrocyte model, called INEXA,
which includes astrocyte controlled tripartite synapses and the
astrocyte-astrocyte interaction. Our results show that GABA
released by astrocytes may be responsible for synchronous
inhibition of postsynaptic neurons. With increased GABA
inhibition, the spike and burst rate decreased while the burst
duration and spikes per burst remain similar. To our
knowledge, it is the first time that the effect of this
gliotransmitter to the neural network was simulated.
I. INTRODUCTION
Astrocytes are known to affect the synaptic neuronal
transmission and blood flow. However, role and functions of
astrocytes in the neuronal communication on network level in
health and disease calls for new insights. Astrocytes support
the supply of nutrients to the neurons, and are responsible for
the liquor regulation in the brain [1], [2]. They are involved
in the processing, transfer and storage of information by the
nervous system [3], [4]. This glia cell type releases
neurotransmitters to inuence neuronal function. Most
known neurotransmitters, which have an influence on
astrocytes, are glutamate, ATP (adenosine triphosphate) and
GABA (-Aminobutyric acid). These gliotransmitters can
change the synaptic plasticity and neuronal excitability [5],
[6]. These dynamics are altered in diseases for example like
epilepsy, Huntingtons and Alzheimers [6][8]. In
Alzheimers disease, the gliotransmitter GABA is more
released by affected, so called reactive, astrocytes [6], [7],
and thus, inhibits excitatory synaptic transmission. This leads
to impaired learning and memory. In this paper, we aim to
simulate a neural network with altered astrocytic GABA
*Research supported by the 3DNeuroN project in the European Unions
Seventh Framework Programme, Future and Emerging Technologies, grant
agreement no296590, and by TEKES- the Finnish funding agency for
innovation (Human Spare Part 2 Project). E.R.s project has received
funding from the European Unions Horizon 2020 research and innovation
programme under the Marie Sklodowska-Curie grant agreement
No 642563.
K. Lenk and J.A.K. Hyttinen are with Tampere University of
Technology, BioMediTech, 33500 Tampere, Finland (corresponding
authors e-mail: kerstin.lenk@tut.fi).
E. Risnen is with Institute for Complex Systems (ISC), National
Research Council (CNR), Sesto Fiorentino, Italy, and Universit degli Studi
di Firenze, via G. Sansone 1, 50019 Sesto Fiorentino, Italy
(raisanen.eero@unifi.it).
978-1-4577-0220-4/16/$31.00 2016 IEEE
release meaning that we present simulation results of healthy
and pathological states. Our model simulates the network and
spiking activity as seen in cultures on an in vitro
multielectrode arrays (MEA) [9]. Therefore, we have
developed the INEXA model [10], [11] which includes a
neural network and the presynapse-astrocyte communication
to model the network effects of altered GABA dynamics. To
our knowledge, it is the first time that the effect of this
gliotransmitter to the neural network, where one astrocyte is
connected to several hundred synapses, is simulated.
II. METHODS
The basis of the model was the spiking neuronal network
model INEX [10] which consists of inhibitory and excitatory
neurons. The probability of each neuron to spike follows an
inhomogeneous Poisson process. In order to model tripartite
synapses, we used a modified version of presynapse astrocyte
interface by De Pitt et al. for excitatory synapses [12] which
is based on Tsodyks-Markram model of synaptic activity
[13]. We made further modification to the presynaptic model
that enables astrocytes to increase or decrease synaptic
strength based on gliotransmission introduced by De Pitt et
al. This modification takes into account different time scales
of different transmitters, and thus, the effect gliotransmission
depends on time scales. Astrocytes IP3 and calcium were
modeled using simple exponential equations where the
calcium release follows the IP3 release with a small delay. In
order to combine the synaptic inputs from all the synapses
enwrapped by a single astrocyte, the local astrocytic
responses to each synapse were summed into a global
astrocyte calcium response. The propagation of calcium
waves in the astrocyte network was then modeled according
to the simplified UAR calcium signaling model (including
the three states of an astrocyte: (U)nactivated, (A)ctivated or
(R)efractory) introduced by Lallouette et al. [14]. Astrocytes,
when activated, signal back locally to their connected
synapses by releasing glutamate. When a whole astrocyte
becomes activated, it releases GABA to all its enwrapped
synapses. The interplay of these integrated models is shown
in Figure 1.
The additional inhibitory pathway representing the release
of astrocytic GABA was added from astrocytes to
postsynaptic neurons through each excitatory synapse
controlled by the astrocyte as follows. Equation (1) describes
the mean firing rate for excitatory neurons i:
i(tk)=max(ci+jyijsj(tk-1)+jyGABAAij(tk-1),0), (1)
where c is the basic activity/ noise of each neuron, y the
synaptic strength and s=1 or 0, if there was a spike in the
last time slice or not, respectively. Furthermore, Aij denotes
if an astrocyte, which is connected to a neuron via synapse
6121
ij, is in an active state. This indicates that the astrocyte is number of active astrocytes remains more constant over time
near enough that the GABA released by the astrocyte has an when the GABA effect is higher (Figure 3-5 D)).
inhibitory effect to the postsynaptic neuron. The more
connections there are, the closer and more encased in the
astrocyte the neuron is, and thus, the stronger the effect. The
strength of the inhibition is determined by the parameter
yGABA that was used as a variable to model the GABA effect
in altered dynamics scenarios.

Figure 2. Distribution of the astrocytes and neurons on a virtual

multielectrode array. The x- and the y-axis represent the distance in
millimeter. Blue dots represent astrocytes, blue lines astrocyte connections,
green dots excitatory neurons and red dots inhibitory neurons. Neuronal
connections are omitted for clarity of the image.

Figure 1. Scheme of the interplay between presynapse, postsynapse, and

the astrocyte. In the presence of an astrocyte the signal is not transferred
unidirectional between the pre- and the postsynapse. Rather the presynaptic
neurotransmitters influence the astrocyte, which in turn has an effect on
both synaptic terminals. 1) Synaptic basal release probability; 2) Amount of
calcium bound to the sensors; 3) Available neurotransmitters; 4) Amount of
neurotransmitter released.

We simulated a 2D network modelling a neural cell

culture similarly as those cultures used on an in vitro MEA
[15] with 200 excitatory and 50 inhibitory neurons with about
26 per cent connectivity as well as 107 astrocytes. On
average, an astrocyte is connected to about 120 nearby
excitatory synapses (Figure 2. ). The network topology was
defined with rule based stochastic process. The simulated
spike trains had a length of 5 minutes. We set yGABA = [-
0.01; -0.1; -0.3] meaning that the neuronal network activity is Figure 3. Example of neuronal activity when astrocytic GABA inhibition
influenced by low, high and very high GABA inhibition. All is very low (yGABA = -0.01). A) spike trains of all neurons, B) summed
other parameter were the same for all simulations (for spikes per 5 ms over all neurons, C) spike trains of the first 30 neurons for
comparison: the excitatory synaptic strength y+ij = 0.8 and the more details, and D) number of active astrocytes.
inhibitory synaptic strength y-ij = -0.6). For each value of
yGABA, we did ten repetitions of the simulation.
For each of the simulations, we calculated the medians
and lower and upper quartiles of spike rate, burst rate, burst
duration, and average number of spikes per bursts. Briey, to
examine the intrinsic bursting, we used a modified version of
the burst analysis algorithm [16] that relies on the cumulative
moving average (CMA) and the skewness () of the
interspike interval (ISI) histogram.
III. RESULTS
An example of an astrocyte network and its connections
is given in Figure 2. Figure 3 to 5 show in A) the spike trains
of all neurons, in B) the summed spikes per 5 ms over all
neurons, in C) the spike trains of the first 30 neurons to see a
more detailed structure in the patterns, and in D) the number Figure 4. Example of neuronal activity when astrocytic GABA inhibition
of simultaneously active astrocytes for yGABA = [-0.01; -0.1; - is high (yGABA = -0.1). A) spike trains of all neurons, B) summed spikes per
0.3] respectively. We see less spikes and bursts while the 5 ms over all neurons, C) spike trains of the first 30 neurons for more
details, and D) number of active astrocytes.
astrocytic GABA effect is increased (Figure 3-5 A)-C)). The

6122

Figure 5. Example of neuronal activity when astrocytic GABA inhibition
is very high (yGABA = -0.3). A) spike trains of all neurons, B) summed spikes
per 5 ms over all neurons, C) spike trains of the first 30 neurons for more
details, and D) number of active astrocytes.
Figure 6 Figure 6. shows the medians and lower and upper
quartiles of spike rate in spikes per minute, burst rate in
bursts per minute, burst duration in milliseconds, and
average number of spikes per bursts for yGABA = [-0.01; -
0.1; -0.3].
Figure 6. Boxplots for the spike rate, burst rate, burst duration, and
average number of spikes per bursts for yGABA = -0.01, yGABA = -0.1, and
yGABA = -0.3 respectively.
IV. DISCUSSION
We simulated neural networks with three different
strength of astrocytic GABA. As expected, the overall
neuronal network activity is reduced when the astrocytic
GABA effect is increased (Figure 3 to 6). This decrease can
be especially seen in the spike and burst rate. However, the
burst duration and spikes per burst remains somewhat stable
while the GABA inhibition rises. This means that we can see
longer spikes and more spikes per burst when GABA gets
stronger.
We run the simulations in time course of five minutes. In our
opinion, this is enough to showcase the dynamical feature of
the model, as the simulator does not contain mechanisms
6123
that function in time scales longer than tens of seconds. In
biological networks, astrocytes could have plasticity or some
other kind of memory that could have long-term effects, but
these kind of functions are outside the scope of our model
and this paper.
For computational reasons, the size of the model was
limited to 250 neurons and 107 astrocytes that physically
over the electrode area of a virtual MEA. Naturally, the
number of connections of each astrocyte to neighboring
synapses was reduced compared to in vitro networks [17],
[18].
Our assumption was that GABA released by astrocytes
may be responsible for synchronous inhibition of
postsynaptic neurons [19][22]; this can be also seen in
Figure 4. This means that astrocytes are a tonic inhibitory
feedback system. The tonic inhibition reduces the systems
overall activity until it is again between the boundaries set by
the astrocyte dynamics of our model. While in low release
networks the restriction of activity is done by reducing the
activity of everything slightly, the high release shuts down
some neurons which results in reduction of network activity
and the temporary shutdown of single neurons (Figure 4 and
5). In Alzheimers disease, this would lead to memory
impairment [6], [7]. Comparing Figure 5 against Figure 3 and
4, we can also see an alteration of the astrocyte dynamics. In
Figure 5, where the GABA activity is enhanced, the number
of active astrocytes remains constant over time.
In the future, we would like to compare our findings with
in vitro MEA data from human inducted pluripotent stem
cells from Alzheimers patients. Furthermore, we would like
to investigate the role of astrocytic GABA in epilepsy and see
if the model can produce epileptic signaling with altering
GABA.
V. CONCLUSION
We showed that our neural model INEXA is able to
reproduce effects by different astrocytic GABA amounts to
the neuronal network. When the amount of GABA is high
enough, the neuronal network responds with synchronous
bursts.
ACKNOWLEDGMENT
All authors thank Emre Kapucu and Inkeri Vornanen for
the development and modification of the CMA algorithm.
REFERENCES
[1] M. Amiri and N. Hosseinmardi, Astrocyte-neuron interaction as
a mechanism responsible for generation of neural synchrony: a
study based on modeling and experiments, J. Comput. Neurosci.,
vol. 34, pp. 489504, Jun. 2013.
[2] H. Kimelberg and M. Nedergaard, Functions of astrocytes and
their potential as therapeutic targets, Neurotherapeutics, vol. 7,
no. 4, pp. 338353, 2010.
[3] A. Araque and M. Navarrete, Glial cells in neuronal network
 function., Philos. Trans. R. Soc. Lond. B. Biol. Sci., vol. 365, pp.
23752381, Aug. 2010.
[4] S. R. McIver, M. Faideau, and P. G. Haydon, Neural-Immune
Interactions in Brain Function and Alcohol Related Disorders.
Boston, MA: Springer US, 2013.
[5] S. Paixo and R. Klein, Neuron-astrocyte communication and
synaptic plasticity, Curr. Opin. Neurobiol., vol. 20, no. 4, pp.
466473, 2010.
[6] L. Ben Haim, M. C. Sauvage, K. Ceyzriat, and J. F. Curtin,
Elusive roles for reactive astrocytes in neurodegenerative
diseases Edited by: Citation:, vol. 9, no. August, pp. 127,
2015.
Astrocyte-Derived GABA in Cultured Embryonic Rat
Hippocampal Neurons, J. Neurophysiol., vol. 84, pp. 13921403,
2000.
[20] A. S. Kozlov, M. Angulo, E. Audinat, and S. Charpak, Target
cell-specific modulation of neuronal activity by astrocytes, Proc.
Natl. Acad. Sci. U.S.A., vol. 103, no. 43, pp. 1005810063, 2006.
[21] M. C. Angulo, K. Le Meur, A. S. Kozlov, S. Charpak, and E.
Audinat, GABA , a forgotten gliotransmitter, Prog. Neurobiol.,
vol. 86, pp. 297303, 2008.
[22] B.-E. Yoon and C. J. Lee, GABA as a rising gliotransmitter.,
Front. Neural Circuits, vol. 8, no. December, p. 141, 2014.

[7] O. Yarishkin, J. Lee, S. Jo, E. M. Hwang, and C. J. Lee,

Disinhibitory Action of Astrocytic GABA at the Perforant Path
to Dentate Gyrus Granule Neuron Synapse Reverses to Inhibitory
in Alzheimers Disease Model, Exp. Neurobiol., vol. 24, no. 3,
pp. 211218, 2015.

[8] A. M. Wjtowicz, A. Dvorzhak, M. Semtner, and R. Grantyn,

Reduced tonic inhibition in striatal output neurons from
Huntington mice due to loss of astrocytic GABA release through
GAT-3., Front. Neural Circuits, vol. 7, no. November, pp. 112,
2013.

[9] B. C. Wheeler, Y. Nam, and B. C. Wheeler, In vitro

microelectrode array technology and neural recordings., Crit.
Rev. Biomed. Eng., vol. 39, no. 1, pp. 4561, 2011.

[10] K. Lenk, A simple phenomenological neuronal model with

inhibitory and excitatory synapses, in Proceedings of the 5th
international conference on Advances in nonlinear speech
processing, 2011, pp. 232238.

[11] E. A. Risnen, K. Lenk, and J. A. K. Hyttinen, Combining

spiking neuronal network model with presynaptic and astrocyte
interface models, in Frontiers in Neuroinformatics, 2014, no. 10.

[12] M. De Pitt, V. Volman, H. Berry, and E. Ben-Jacob, A tale of

two stories: astrocyte regulation of synaptic depression and
facilitation., Plos Comput. Biol., vol. 7, no. 12, p. e1002293,
Dec. 2011.

[13] M. V. Tsodyks and H. Markram, The neural code between

neocortical pyramidal neurons depends on neurotransmitter
release probability, Proc. Natl. Acad. Sci., vol. 94, no. 2, pp.
719723, Jan. 1997.

[14] J. Lallouette, M. De Pitt, E. Ben-Jacob, and H. Berry, Sparse

short-distance connections enhance calcium wave propagation in
a 3D model of astrocyte networks., Front. Comput. Neurosci.,
vol. 8, pp. 118, Jan. 2014.

[15] G. Wallach, J. Lallouette, N. Herzog, M. De Pitt, E. Ben Jacob,

H. Berry, and Y. Hanein, Glutamate mediated astrocytic filtering
of neuronal activity, PLoS Comput. Biol., vol. 10, no. 12, p.
e1003964, 2014.

[16] F. Kapucu, J. Tanskanen, J. E. Mikkonen, L. Yl-Outinen, S.

Narkilahti, and J. A. K. Hyttinen, Burst analysis tool for
developing neuronal networks exhibiting highly varying action
potential dynamics, Front. Comput. Neurosci., vol. 6, pp. 114,
Jan. 2012.

[17] M. M. Halassa, T. Fellin, and P. G. Haydon, The tripartite

synapse: roles for gliotransmission in health and disease, Trends
Mol. Med., vol. 13, no. 2, pp. 5463, 2007.

[18] E. a Bushong, M. E. Martone, Y. Z. Jones, and M. H. Ellisman,

Protoplasmic astrocytes in CA1 stratum radiatum occupy
separate anatomical domains., J. Neurosci., vol. 22, no. 1, pp.
183192, 2002.

[19] Q. Liu, A. E. Schaffner, Y. H. Chang, D. Maric, and J. L. Barker,

Persistent Activation of GABA_A Receptor/ Cl^- Channels by

6124