Time-Delay-Induced Phase-Transition To Synchrony in Coupled Bursting Neurons
Time-Delay-Induced Phase-Transition To Synchrony in Coupled Bursting Neurons
Time-Delay-Induced Phase-Transition To Synchrony in Coupled Bursting Neurons
Signal transmission time delays in a network of nonlinear oscillators are known to be responsible for
a variety of interesting dynamic behaviors including phase-flip transitions leading to synchrony or
out of synchrony. Here, we uncover that phase-flip transitions are general phenomena and can occur
in a network of coupled bursting neurons with a variety of coupling types. The transitions are
marked by nonlinear changes in both temporal and phase-space characteristics of the coupled
system. We demonstrate these phase-transitions with Hindmarsh-Rose and Leech-Heart interneuron
models and discuss the implications of these results in understanding collective dynamics of
C 2011 American Institute of Physics.
bursting neurons in the brain. V
[doi:10.1063/1.3584822]
A large body of experimental work on brain activity has comparable to time-scales of neuronal oscillations are known
demonstrated that phase synchronization of neuronal to have significant effects in the ensemble activity of neu-
oscillations is the basis for various percepts and actions, rons. Thus, in modeling studies of neurons and networks, the
such as perceptual decision-making, attention and mem- influence of time delays on the ensemble activity has
ory processes, awareness, sensory-motor, or multisensory received a great deal of attention recently.4–16 In networks of
integration. Synchronized neuronal oscillations can occur coupled neurons, time delays have been shown to affect not
in neurons from a small brain region to a large-scale net- only the amplitude dynamics generating instabilities,8 oscil-
work of distributed brain regions. Synchrony in networks lation death,11,12 enhancement or suppression of synchron-
of spatially distributed neurons involves signal transmis- ized oscillations,4–6,17 or phase-coherent oscillations,9 but
sion time delays because of finite propagation speeds and also the phase dynamics leading the system to or out of syn-
axonal lengths. In recent years, many theoretical and
chrony.13–16,18–21 This time-delay-induced phase transition,
computational studies of nonlinear oscillators reported
marked by a relative phase change from zero to p and a dis-
various interesting effects of time-delays on phase syn-
continuous change of the average oscillating frequency, was
chrony. One of the important time-delay induced effects
is a phase-flip transition leading to synchrony or out of named as phase-flip bifurcation.22 Such time-delay induced
synchrony. Here, we confirm and extend this time-delay effects in phases and frequencies have not been systemati-
induced effect in a class of multi-time scale dynamical cally studied in different coupling types using multi-time
systems such as bursting neurons. We find that phase-flip scaled dynamical systems like coupled bursting neurons.
transitions are general phenomena and can occur in a At the phase-flip bifurcation point, the ensemble activity
network of coupled bursting neurons with a variety of of coupled periodic or chaotic oscillators changes from in-
coupling types. The transitions are marked by abrupt phase to out-of-phase (phase difference of p) oscillations (or
changes in both temporal and phase-space characteristics vice versa).22 This is accompanied by a discontinuous change
of the coupled system. We show that these phase-transi- in the average frequency of oscillations and also by a discon-
tions occur in networks of different types of bursting neu- tinuity in the largest negative Lyapunov exponent across a
ron models and discuss the implications of these findings. critical delay or coupling strength in time-delayed interac-
tion.19 The average frequency shows a nonlinear dependence
with time-delay and coupling strength. A time-delayed inter-
action was identified to be the necessary condition for this
I. INTRODUCTION
transition. This phenomenon was observed in various
Signal transmission time delays are the result of axonal dynamic regimes such as in oscillator death, periodic, quasi-
conduction and chemical synaptic processes and are inherent periodic and chaotic oscillations.19,22 Bursting neurons have
in networks of neurons in the brain. While the chemical syn- distinctly different multiple time scales of oscillations: spike
aptic time delays are small (2 ms), the axonal conduction activity with fast time scale and burst activity (consisting of
delays, which depend on the distance between neurons in the two or more spikes) with slow time-scales. How do bursting
brain, can reach upto tens of milliseconds.1–3 Time delays neurons respond to time-delayed interactions? Are there
time-delay induced phase-flip transitions to or out of syn-
a)
Electronic mail: [email protected]. chrony? What are the general characteristics of these
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-2 Adhikari, Prasad, and Dhamala Chaos 21, 023116 (2011)
FIG. 1. (Color online) Schematic of time-delayed neuronal interaction. Neuron 2 sends an action potential or a burst of action potentials to neuron 1 at time t,
and neuron 1 feels it at later time t þ s, where s is the time delay. This delay in a chemical coupling is due to axonal conduction and synaptic processes,
whereas time delay in electrical coupling is only due to axonal conduction. In modeling a time-delayed interaction from neuron 2 to neuron 1, the coupling
term to neuron 1 can take the delayed input from neuron 2. The coupling function f12 depends on this delayed input, both for electrical and chemical synaptic
couplings, where g is coupling strength, x1, x2 are potentials of neuron 1 and neuron 2, and CðÞ is 1 for positive (*) and 0 otherwise. A part of the schematic
was adapted from Izhikevich, Eugene M., Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting, figure 1.1, page 2, adapted, V C 2006
transitions in excitatory and inhibitory neuronal networks? In specify the manner in which the neurons i and j are coupled,
this paper, we investigate into these questions using two with HðXi ; Xj ; sÞ being a function of Xi(t) and Xj ðt sÞ. For
bursting neuron models: Hindmarsh-Rose (HR) neuron23 and a three-variable neuron model, Xi ¼ ½xi ; yi ; zi T and
Leech-Heart interneuron (IN).24,25 H ¼ ½gðxi ðtÞ; xj ðt sÞÞ; 0; 0T , where g is a coupling func-
Here, we show that the phase-flip bifurcation can also tion, and the superscript T denotes the transpose. If the neu-
occur in time-delayed coupled bursting neurons. But, in the rons i and j are connected via a gap junction (electrical
network of these neurons, the phase-transition from syn- coupling), then the coupling function takes the form
chrony to out-of-synchrony occurs with a different maximum gðxi ðtÞ; xj ðt sÞÞ ¼ ðxj ðt sÞ xi ðtÞÞ. For a coupling via a
relative phase, not p. The average frequency and the largest chemical synapse, it is hðxi ðtÞ; xj ðt sÞÞ ¼ ðxi ðtÞ Vis Þ
negative Lyapunov exponent show discontinuous changes at Cðxj ðt sÞÞ, where the reversal potential Vis can set the
a critical delay and coupling strength. These results hold true threshold for excitation or inhibition, and CðÞ is 1 for posi-
in a variety of couplings: electrical, and excitatory and inhib- tive (*) and 0 otherwise. If Vis > xi ðtÞ for all xi and t, the cou-
itory synaptic couplings, and also in a network of three or pling term is always positive and the synapse is excitatory,
more neurons. These results may help understand the mecha- i.e., the input to ith neuron via coupling can enhance the ac-
nisms of a long-distance synchrony of oscillations underly- tivity of this neuron. On the other hand, if Vis < xi ðtÞ for all
ing brain functions or dysfunctions. xi and t, the coupling term is negative and synapse is inhibi-
tory, i.e., the input to ith neuron via coupling can suppress its
II. METHODS AND RESULTS activity. Usually, CðÞ is approximated by a sigmoidal func-
tion: 1=½1 þ expfbðxj ðt sÞ Hs Þg, where b determines
A. Time-delayed coupled network model the slope of the function and Hs is the firing threshold. Fig. 1
We begin by formulating a general model for a network shows a time-delayed coupling setup with two neurons.
of neurons with time-delayed interactions. An example of a
time-delayed interaction between two neurons is shown in B. Neuron models
Fig. 1. Consider a coupled system of N bursting neurons. Both Hindmarsh-Rose neuron23,26 and Leech-Heart
Each neuron when isolated follows X_ ¼ FðXÞ, where X is an Interneuron24,25 models are known to produce bursting activ-
m-dimensional vector of dynamical variables such as mem- ity in extended ranges of system parameters. Here, to study
brane voltage and gating variables in Hodgkin-Huxley for- the time-delay induced phase-transitions in a bursting activ-
malism, and F(X) is the velocity field. The coupled system is ity, we have considered using them both. Three-variable HR
described by the following equations: neuron26 is a phenomenological model although the original
two-variable model23 was constructed to describe snail’s
X Ki
neuron dynamics. The IN neuron describes the electrical ac-
X_ i ¼ Fi ðXi Þ þ Aij HðXi ; Xj ; sÞ; i ¼ 1; …; N (1)
Ki j¼1 tivity of neurons that control Leech’s heart.
The HR neuron dynamics is described by the following
where subscript i in Xi and Fi represents i-th neuron. Here Ki set of equations:26,27
is the number of connections to the i-th neuron, namely its
x_ ¼ y ax3 þ bx2 þ z þ Iext
degree and 1 Ki < N, is the coupling strength, s > 0 is
the net time delay—the time for the action potential to propa- y_ ¼ c dx2 y (2)
gate along the axon connecting the pre-synaptic neuron j to z_ ¼ r½sðx x0 Þ z
the post-synaptic neuron i. The connection topology is given
as Aij ¼ 1 if the neurons i and j are connected to each other where x is the membrane potential, y is the fast current (re-
and Aij ¼ 0 otherwise. The coupling function H : Rm ! Rm covery variable), and z is the slow current also known as
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-3 Time-delay-induced phase-transition Chaos 21, 023116 (2011)
FIG. 2. (Color online) Bifurcation diagrams (upper panel) and first two Lyapunov exponents (lower panel) for Hindmarsh-Rose (HR) neuron (left panel) and
Leech-Heart Interneuron (IN) neuron (right panel) models. These results help us understand the qualitative behaviors of single neurons. Here, the bifurcation dia-
grams are obtained by computing the times Dt between successive spikes, including the time intervals between bursts of spikes. As the parameter, external cur-
rent to a neuron (Iext) is decreased from 4.0 to 1.5, HR neuron goes through a series of transitions: from period 1 to chaos via period-doubling cascade and back
to periodic behaviors. The maximum Lyapunov exponent (lower left, blue) shows that the HR system is chaotic approximately for the range 2:92 < Iext < 3:40.
shift
In the case of IN neuron, as the parameter, VK2 , is decreased from –0.01 to –0.025 volt, it goes through a series of period-adding bifurcation.31,32 There is a cha-
otic (or many period) regime around –0.024 volt, which belongs to one attractor state. There is another period-1 attractor at and around this parameter value.
This region shows bistability of chaotic attractor and period-1 attractor (see Fig. 3).
adaptation variable. Here, a ¼ 1.0, b ¼ 3.0, c ¼ 1.0, d ¼ 5.0, chaotic (or high-period periodic) states as shown in the bifur-
s ¼ 4.0, r ¼ 0.006, x0 ¼ –1.60, and Iext is the external current cation diagram and Lyapunov exponents (Fig. 2 right panel).
input. r is the ratio of fast/slow time scales. This system Typical phase-space portraits and corresponding potential
exhibits a multi-time-scale spike-burst chaotic behavior for time series of burst activity for both of these neuron models
2:92 < Iext < 3:4, as shown in the bifurcation diagram and are shown in Fig. 3. IN neuron is bistable—period 1 (shown
Lyapunov exponents (Fig. 2, left panel). in Fig. 3 with red) and chaotic state (or high period periodic
shift
The Leech-Heart Interneuron model has the following state, shown in blue) coexist around VK2 0:024 V. Both
equations of motion:25 of these oscillators show the spike-burst activity (multi-time
scale dynamical behaviors) in a wide range of bifurcation pa-
v_ ¼ 2½30 m2 ðv þ 0:07Þ þ 8ðv þ 0:046Þ rameters. There are two distinct time-scale of oscillations
þ 200f 3 ð150; 0:0305; vÞhðv 0:045Þ and the trajectories in the phase-space move slowly in one
(3) region whereas they move fast in the other region. The y–z
h_ ¼ 24:69½ f ð500; 0:0333; vÞ h plane along the firing threshold x ¼ –1 approximately sepa-
shift rates these two regions in the HR neuron, and v 0:04 V
m_ ¼ 4½f ð83; 0:018 þ VK2 ; vÞ m;
approximately separates in the IN neuron case. Different
where v is the membrane potential, h and m are membrane time scales of oscillations can also be seen in voltage traces:
channel gating variables. h is associated with fast ionic cur- slow motion when the membrane potential is below the firing
rent such as flow of sodium ions and m is the slow gating threshold and fast when it is above [voltage traces in Fig. 3
variable. f ða; b; vÞ ¼ 1=½1 þ expðaðb þ vÞÞ is a Boltzmann (lower panel)]. In non-delayed coupled systems consisting of
function which describes the kinetics of activation/inactiva- these bursting neuron types, synchrony in individual bursts
shift
tion of ionic currents. VK2 is an experimentally accessible can be easily achieved through coupling of individual neu-
bifurcation parameter that is a deviation from the average rons as a precursor to a complete synchrony,27 and by vary-
potential v ¼ –0.018 V corresponding to the semi-activated ing coupling strengths, or modifying connections of neuronal
potassium channel at f ¼ 1/2. IN neuron has periodic and subnetworks.28
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-4 Adhikari, Prasad, and Dhamala Chaos 21, 023116 (2011)
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-5 Time-delay-induced phase-transition Chaos 21, 023116 (2011)
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-6 Adhikari, Prasad, and Dhamala Chaos 21, 023116 (2011)
neurons generalize a network consisting larger number of that the phase-flip transitions are primarily due to the intrin-
neurons. Fig. 7 in the top panel shows an example of four sic time-scales of the system and time delays in couplings.
neurons connected in a ring with excitatory synaptic cou- Thus, these results of phase-flip transitions in weakly
pling. As time-delay s changes from 0.25 to 0.30, the system coupled systems can be expected to remain robust to varia-
goes from out-of-phase state to in-phase state. Here, the out- tions of number of nodes or the system size.
of-phase state is the most stable when the phase-difference
between the nearest neighbor is 2p=N, where N is the num-
III. CONCLUSIONS
ber of oscillators, and in this example it is 4. For a network
of eight oscillators, this phase difference becomes 2p=8. We In summary, we uncover a phenomenon of phase-flip
have used the similarity function as described in Ref. 33 to bifurcation in networks of bursting neurons with a variety of
find the phase-differences between neurons. Here, the neu- time-delayed coupling types and topology. Unlike the previ-
rons in the periphery close to each other in a ring topology ous findings in systems of periodic or chaotic oscillators,
have p=2 phase difference whereas off-diagonal neurons far- coupled bursting neurons undergo time-delay-induced phase
thest from each other are out-of-phase by p. These network transitions to asynchrony out of synchrony by a phase differ-
results extend the previous findings19 from two oscillators. ence less than p. The transitions are marked by abrupt
The phase-flip transitions can be demonstrated even in changes in the second negative Lyapunov exponent, average
weakly coupled phase-oscillators.34 Consistent with such frequency, and relative phase difference. In the case of near-
finding, our results of weakly coupled oscillators indicate est neighbor excitatory coupling, oscillators next to each
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp
023116-7 Time-delay-induced phase-transition Chaos 21, 023116 (2011)
9
other are at an equal phase difference around a circle: for X. Liang, M. Tang, M. Dhamala, and Z. Liu, Phys. Rev. E. 80, 066202 (2009).
10
three oscillators, the phase difference between neighbors is X. Liang, M. Dhamala, L. Zaho, and Z. Liu, Phys. Rev. E. 82, 010902 (2010).
11
A. Prasad, M. Dhamala, B. M. Adhikari, and R. Ramaswamy, Phys. Rev.
2p=3; for four, it is 2p=4, which generalizes to N oscillators E. 81, 0027201 (2010).
with phase difference of 2p=N. Time-delays are unavoidable 12
A. Prasad, M. Dhamala, B. M. Adhikari, and R. Ramaswamy, Phys. Rev.
in spatially distributed dynamical systems like the neurons in E. 82, 0027201 (2010).
13
Q. Wang, Z. Duan, M. Perc, and G. Chen, Europhys. Lett. 83, 50008 (2008).
the brain and can play significant roles in large-scale 14
Q. Wang, M. Perc, Z. Duan, and G. Chen, Phys. Rev. E 80, 026206 (2009).
synchronized oscillations in the brain. Time-delayed cou- 15
Q. Wang, M. Perc, Z. Duan, and G. Chen, Physica A 389, 3299 (2010).
16
pling can, for example, induce synchronization between two Q. Wang, G. Chen, and M. Perc, PLoS ONE 6, e15851 (2011).
17
cells with intrinsic antiphase bursting activities. Even in mu- N. Buric, K. Todorovic, and N. Vasovic, Phys. Rev. E. 78, 036211 (2008).
18
H. G. Schuster and P. Wagner, Prog. Theor. Phys. 81, 939 (1989).
tual inhibition among multiple neurons, synchrony of burst- 19
A. Prasad, S. K. Dana, R. Karnatak, J. Kurths, B. Blasius, and R. Ramasw-
spike activities can occur with time-delayed coupling. These amy, Chaos 18, 023111 (2008).
20
time-delay-induced transitions may help explain the co-exis- C. Masoller, M. C. Torrent, and J. Jarcia-Ojalvo, Philos. Trans. R. Soc.
tence of various frequency rhythms in different parts of a London, Ser. A 367, 3255 (2009).
21
B. Ermentrout and T.-W. Ko, Philos. Trans. R. Soc. London, Ser. A 367,
large neuronal network. 1097 (2009).
22
A. Prasad, J. Kurths, S. K. Dana, and R. Ramaswamy, Phys. Rev. E. 74,
ACKNOWLEDGMENTS 035204 (2006).
23
J. L. Hindmarsh and R. M. Rose, Nature 296, 162 (1982).
24
The author M.D. would like to thank Akira Sakurai and G. S. Cymbalyuk and R. L. Calabrese, Neurocomputing 38–40, 159 (2001).
25
Gennady Cymbalyuk for their valuable discussions and NSF G. S. Cymbalyuk and A. Shilnikov, J. Comput. Neurosci. 18, 255 (2005).
26
J. L. Hindmarsh and R. M. Rose, Proc. R. Soc. London, Ser. B 221, 87 (1984).
for the funding support (NSF Career Award BCS 0955037). 27
M. Dhamala, V. K. Jirsa, and M. Ding, Phys. Rev. Lett. 92, 028101 (2004).
28
X. Sun, J. Lei, M. Perc, J. Kurths, and G. Chen, Chaos 21, 016110 (2011).
1 29
H. A. Swadlow, J. Neurophysiol. 54, 1346 (1985). S. Jalil, I. Belykh, and A. Shilnikov, Phys. Rev. E 81, 045201(R) (2010).
2 30
E. R. Kandel, J. H. Schwartz, and T. M. Jessell, Principles of Neural Sci- E. M. Izhikevich, Dynamical Systems in Neuroscience: The Geometry of
ence, 3rd ed. (Elsevier, New York, 1991). Excitability and Bursting (MIT Press, Cambridge, MA, 2007).
3 31
H. A. Swadlow, J. Neurophysiol. 71, 437 (1985). A. Shilnikov, R. L. Calabrese, and G. Cymbalyuk, Phys. Rev. E. 71,
4
M. Dhamala, V. K. Jirsa, and M. Ding, Phys. Rev. Lett. 92, 074104 (2004). 056214 (2005).
5 32
M. G. Rosemblum and A. S. Pikovsky, Phys. Rev. Lett. 92, 114102 (2004). P. Channell, G. S. Cymbalyuk, and A. Shilnikov, Phys. Rev. Lett. 98,
6
E. Rossoni, Y. Chen, M. Ding, and J. Feng, Phys. Rev. E 71, 061904 (2005). 134101 (2007).
7 33
P. Gong and C. van Leeuwen, Phys. Rev. Lett. 98, 048104 (2007). M. G. Rosenblum, A. S. Pikovsky, and J. Kurths, Phys. Rev. Lett. 78,
8
A. Ghosh, Y. Rho, A. R. McIntosh, R. Kotter, and V. K. Jirsa, PLOS Com- 4193 (1997).
34
put. Biol. 4(10), e1000196 (2008). V. V. Klinshov and V. I. Nekorkin, Chaos, Solitons, Fractals 44, 98 (2011).
Author complimentary copy. Redistribution subject to AIP license or copyright, see http://cha.aip.org/cha/copyright.jsp