Journal of Theoretical Biology 370 (2015) 157170

Contents lists available at ScienceDirect

Journal of Theoretical Biology

journal homepage: www.elsevier.com/locate/yjtbi

Improving desynchronization of parkinsonian neuronal network via

triplet-structure coordinated reset stimulation
Denggui Fan, Qingyun Wang n
Department of Dynamics and Control, Beihang University, Beijing 100191, China

H I G H L I G H T S

We propose a new triplet-structure coordinated reset stimulation (CRS) paradigm.

CRS can signicantly desynchronize the neuronal network of Parkinson's disease.
Synaptic plasticity can greatly enhance the CRS-induced desynchronization.
The closed-loop CRS can improve desynchronization and reliability.

art ic l e i nf o

a b s t r a c t

Article history:
Received 2 August 2014
Received in revised form
7 December 2014
Accepted 28 January 2015
Available online 7 February 2015

We investigate how the triplet-structure coordinated reset stimulations (CRS), which acts on the GPe,
STN and GPi within the basal ganglia-thalamocortical motor circuit, can destabilize the strong
synchronous state and improve the reliability of thalamic relay in the parkinsonian network. It is
shown that compared with the permanent (1:0 ONOFF) CRS or the classic deep brain stimulation
paradigm, the periodic m:n ONOFF CRS (i.e., m ON-cycles stimulation followed by n OFF-cycles
stimulation) can signicantly desynchronize the neuronal network of Parkinson's disease, and evidently
improve the delity of thalamic relay. In addition, the CRS-induced desynchronization can be greatly
enhanced when the STN subpopulation within the pathologic network is subjected to the synaptic
plasticity. Furthermore, the desynchronization and reliability can also be further improved as the closedloop CRS strategy is introduced. The obtained results can be helpful for us to understand the
pathophysiology mechanism of Parkinson's disease, even though the feasibility of CRS still needs to be
explored in clinic.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Desynchronization control
Stimulation paradigms
Synaptic plasticity

1. Introduction
Even though the mechanism of neurological disorder such as
Parkinson's disease or essential tremor remains unclear, experimental
results have indicated that these motor disorders can be characterized
by the excessively strong pathological synchronization (Nini et al.,
1995; Brown et al., 2001; Levy et al., 2002; Timmermann et al., 2002;
Hammond et al., 2007; Smirnov et al., 2008). Therefore, desynchronizing neuronal activity can be a desirable way to suppress the pathological synchronization and restore the normal brain function. For a
clinical purpose, several desynchronizing techniques have been developed by means of nonlinear dynamics (Dovzhenok et al., 2013; Tass,
2003a), which are based upon either the feedback loop (Dovzhenok
et al., 2013) or the phase resetting principle (Tass, 2003b), e.g.,
coordinated reset stimulation.

Corresponding author.
E-mail address: nmqingyun@163.com (Q. Wang).

http://dx.doi.org/10.1016/j.jtbi.2015.01.040
0022-5193/& 2015 Elsevier Ltd. All rights reserved.

It is well known that several deep brain stimulation (DBS) techniques including classic open-loop DBS (Rubin and Terman, 2004;
Khn et al., 2008; Meissner et al., 2005) or closed-loop DBS (Feng
et al., 2007a; Little et al., 2013; Rosin et al., 2011) have been designed
to deliver the stimulation signal to the subcortical basal ganglia (BG)
via implanted electrodes. This can clinically improve the hypokinetic
motor symptoms of patients with idiopathic Parkinson's disease (PD)
or essential tremor and animal models of PD. However, because of its
continuous strong stimuli or xed stimulation target, standard DBS
cannot sufciently restore the motor function of brain, and can lead to
some prominent side effect, e.g., damaging the brain tissue or
worsening the motor ability as the delivery of stimulation signals
ends up. Resultantly, new DBS strategy must be further developed to
obtain maximal control effects.
Coordinated reset stimulation (CRS) is a very promisingly desynchronizing stimulation method, where brief high-frequency (HF)
pulse trains (i.e., bursts) are delivered via different sites rather than
continuously on a xed site. Recently, CRS has received much
attention and appears to be the best candidate of clinical therapy

158

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

(Hauptmann and Tass, 2009; Lysyansky et al., 2011; Popovych and

Tass, 2012; Tass et al., 2012). It can be performed in both open and
closed loop mode (Tass, 2003a). However, this strategy has not been
implemented in PD patients. To our knowledge, it is only proposed
in ex-vivo experiments (Tass et al., 2009) or theoretical studies (Tass
and Majtanik, 2006) to counteract the pathological synchronization.
In addition, most of previous studies are mainly based on the subnetwork of STN or STN-GPe. Recently, some interesting works set out
to focus on the study of behaviors within the global basal gangliathalamocortical network (Meijer et al., 2011; Guo and Rubin, 2011).
Meijer et al. (2011) presented a computational model of a thalamocortical relay neuron to explore the basal ganglia-thalamocortical loop
behavior in the context of Parkinson's disease and deep brain stimulation (DBS). Guo and Rubin (2011) established the basal gangliathalamocortical network to investigate the TC reliability as CRS is
applied to STN neuron. However, it is still not fully clear that how the
desynchronizing effect of CRS can globally inuence the behavior of
downstream thalamic neurons. To extend the previous studies, we
will propose a new CRS paradigm to explore the effect of CRS on PD
network by using the network composed of the basal gangliathalamocortical motor circuit.
In the basal ganglia-thalamocortical motor circuit, different stimulation targets, whether the stimulation is current standard DBS or
newly proposed phase resetting CRS, are used to treat the neurological disorder including PD. The DBS targets lie in the area of BG,
typically in the STN, GPi or GPe (Saxena, 2011) for PD patients. As a
typical case, the STN is usually a standard anatomical target for DBS.
During the stimulation, the stimulation signal is usually conducted
on a xed target area or the same structure in the basal ganglia. And,
the stimulation strength is always excessive and disruptive because
of the characteristic of high-frequency(HF) or long-lasting procedure
of DBS. For clinical application, the stimulation parameter is very
crucial with the stimulation being effective enough, but not too
strong for the purpose of energy-saving and avoiding side effect (e.g.,
caused by current spread). For this purpose, we propose a new CRS
protocol, which implies that we can not use one or more electrodes
on a xed area permanently. This method will simultaneously apply
three micro-electrodes into three different basal ganglia nuclei (see
Fig. 1): GPe, STN and GPi, respectively. For each nucleus, a brief pulse
train (i.e., bursts) is delivered via one of three electrodes.
Based on the proposed stimulation protocol, we can globally
investigate the desynchronizing effect of CRS on the network of the
basal ganglia-thalamocortical motor circuit. Here, we mainly consider
the effect of intermittent m:n ONOFF CRS, in which, m cycles ON
stimulation are followed by n cycles OFF stimulation. In addition, we
also show that how the CRS affects the relay behavior of thalamic
neurons, where reliability of information transmission will be viewed
as a measuring standard. Meanwhile, it is shown that both CRS with
synaptic plasticity and the closed-loop can inuence the desynchronizing effect and reliability performance of the network model for
the basal ganglia-thalamocortical motor circuit.
The paper is organized as follows. In Section 2, we introduce
the computational model of network dynamics. The simulation
results of CRS on the computational network model are presented
in Section 3. Finally, conclusion is made in Section 4.

connected framework that is rstly proposed in Terman et al.

(2002). Here, we will reconstruct the synaptic structure and connectivity of the network architecture as illustrated in Fig. 1, which has
been developed by Feng et al. (2007a). In particular, this network
consists of four parts (see Fig. 1): the spheres denote the eight GPe
neurons, the cylinders are the eight STN neurons, and the cubes
describe the eight GPi neurons, together with two thalamocortical
relay (TC) model neurons. Each GPe neuron inhibits two STN neurons, which skips its three nearest STN neurons, as well as sends
inhibitions to the two immediate neighboring GPe neurons and one
GPi neuron. Each STN neuron sends excitations to the nearest single
GPi neuron, and GPe neuron in the circle of neurons, respectively.
Each GPi neuron inhibits only one TC neuron, but each TC neuron
receives synaptic inhibitions from four GPi neurons (e.g., 1st TC
neuron receives inhibitory inputs from 1st, 2nd, 5th and 6th GPi
neurons). In addition, GPe neurons receive striatal inhibitory inputs,
and TC neurons receive excitatory sensorimotor (SM) signals (see
Rubin and Terman, 2004). And, all three basal ganglia nuclei, GPe,
GPi and STN, receive excitatory HF stimulation (HFS) inputs from
three different micro-electrodes (see Fig. 1), respectively. All model
parameters can be found in Refs. (Terman et al., 2002; Rubin and
Terman, 2004; Feng, 2007a,b; Dovzhenok et al., 2013), where
g GPe-GPe and I app GPe, i.e., self-synaptic conductance within the
GPe and constant input from striatum to GPe, are two essential
parameters in changing the healthy state of neural systems. Physiologically and anatomically, the PD-like state of neural system can
be induced by the degeneration of dopaminergic neurons within
the substantia nigra pars compacta (SNc). This is because the
degeneration of dopaminergic neurons can rst directly decrease
the release of dopamine, and further indirectly affect the downstream synaptic connections of dopaminergic modulation within
the basal ganglia-thalamocortical motor circuit. On the other hand,
both the g GPe-GPe and I app GPe are modulated by the dopamine
(Dovzhenok et al., 2013). When the other parameters are xed,
tuning the values of g GPe-GPe and I app GPe can produce the
transition between parkinsonian state and normal state (Terman
et al., 2002; Feng, 2007a,b; Dovzhenok et al., 2013). Especially, the
larger values of both g GPe-GPe and I app GPe correspond to the
normal state of system, but the smaller values correspond to the
parkinsonian state. Accordingly, in the numerical experiments these
correspond to the random or synchronous dynamics.

2.2. Model for each neuron type

The well known HodgkinHuxley (HH) type equations are used
to model dynamical behaviors of the STN, GPe and GPi neurons.
We take the canonical form of HH equations as follows.
X ion
CmV 0 
I k  I syn I app I CRS
1
k

where V is the membrane potential, C m 1 F cm  2 is the

membrane capacitance.
X ion
I k I L I K I Na I T I Ca I AHP
2
k

2. Description of model
2.1. The network model
We use the computational model based on basal gangliathalamocortical motor circuit, which was initially originated in Refs.
Terman et al. (2002) and Rubin and Terman (2004), and then
revised by Feng et al. (2007a) for the purpose of searching for new
DBS waveforms. Connection of the network model is the sparsely

is the total ionic current through the membrane, where I L g L

V  V L is the leak current, I K g K n4 V V K is the potassium
current, I Na g Na m31 V hV  V Na is the sodium current, I T g T a31
V rV  V Ca is the low-threshold T-type Ca2 current, I Ca g Ca s21
V V  V Ca is the high-threshold Ca2 current and I AHP g AHP
V  V K Ca=Ca kAHP is the Ca2 -activated, after-hyperpolarization K current. Among these ionic currents, gx represents the
channel conductance and Vx represents the reversal potential for
the ion x.

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

159

Fig. 1. The sparsely connected, structured network framework adopted from Terman et al. (2002) and Rubin and Terman (2004). Lines ending with arrows (green) and
closed ellipses (red, rufous or black) indicate the excitatory and inhibitory synaptic connections, respectively. GPe, STN and GPi neurons are shown in spheres, cylinders and
cubes, respectively. TC neurons are shown in two yellow squares with pink edge. Each GPe neuron inhibits two STN neurons which skip three STN neurons nearest to the
GPe, as well as sends inhibitions to the immediate neighboring two GPe neurons and one GPi neuron. Each STN neuron sends excitations to both the nearest single GPi
neuron and GPe neuron in the circle of neurons. Each GPi neuron inhibits only one TC neuron, but each TC neuron receives synaptic inhibitions from four GPi neurons (e.g.,
1st TC neuron receives inhibitory input from 1st, 2nd, 5th and 6th GPi neurons, etc.). As shown in Terman et al. (2002) and Rubin and Terman (2004), the network framework
shows a periodic structure so that neurons 1st and 8th in each circle are neighbors, etc. The triplet of micro-electrodes applied on the three different structures as colored
(1-black: GPe, 2-red: STN, 3-blue: GPi), and the inset is signals through the three electrodes. (For interpretation of the references to color in this gure caption, the reader is
referred to the web version of this article.)

Note that in STN neurons, I T g T a31 V b1 rV  V Ca is mod2

ied by replacing the r with b1 r, where
b1 r

1
r  b

1 exp b

1
1 exp  bb

is the steady-state of gating variable, which contributes to the

rebound bursts (Terman et al., 2002). However, the general form
can be employed to describe the other steady-states of the
voltage-dependent gating variables in the ionic current equations:
x1 r

1
1 exp  Vtx x

where x m; h; n; r; a; s, both x and x are constants. The slow

gating activation and inactivation variables are described as the
following general equation:
x0 x x1 xt=x

where x h; n; r represents the sodium, potassium and T-current

inactivation, respectively. x is a constant parameter, and x is the
time constant, which obeys the rst-order kinetics:

x 0x

x

Vt 
1 exp  x

where 0x and 1x are constants. Note that Eq. (6) only holds for the
case of x h; n. And r does not obey Eq. (6), and it is considered as
a constant (Terman et al., 2002). Because of the governing action of
intracellular Ca2 concentration on the after-hyperpolarization
current, the Ca follows the dynamics equation:
Ca  I Ca  I T  kCa Cat;
0

where is a constant related to the buffer effect of calcium, and kCa

is the constant representing the calcium pump rate.
I app is an applied constant current, which can adjust the neuron
ring property. Especially, for GPe neuron, I app represents the
striatum input. I syn is the synaptic current, which can be written by
the form of I - . This represents the direction of the synaptic
current from neuron to neuron . I CRS is the intermittent multistructure HFS current injected into the structures of basal ganglia
by micro-electrodes.
Thalamic neurons or TC cells are also modeled by the Hodgkin
Huxley type equations, but more simple than those for the basal
ganglia (Rubin and Terman, 2004). There are only three state
variables, V Th ; hTh and r Th for the reduced TC neurons (Rinzel, 1985):
C m V 0Th I L  I Na  I K  I T  I GPi-Th I SM
0

hTh h1 V Th  hTh =h V Th
r 0Th r 1 V Th  hTh =r V Th

8
9
10

where C m 1 F cm , I L g L V Th  V L ,
V Th  V Na , I K g K 0:751 hTh 4 V Th  V K
V Th  V T . I GPi-Th is the synaptic current sent by GPi to TC. Note
that the gating variables hTh and r Th have the state equations similar
Th
to (5), but their time constants Th
h and r differ slightly. Here
2

Th
h

1
Th
bh
aTh
h

I Na g Na m31 V Th hTh
and I T g T a21 V Th r Th

11

with



V 46
aTh
h 0:128 exp 
18

12

160

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

and
4
Th


bh
1 exp  V 523


Th
r 0:4 28 exp 

13

V 25
:
10:5

14

I SM is the sensorimotor input, which is usually modeled as a period

pulse train with the form:
I SM iSM H sin 2 t=SM  1  H sin 2 t SM =SM 

15

where H is the Heaviside bi-value step function, that is H(x)1, if x 4 0

and H(x)0, if x r 0. Usually, SM 50 ms and iSM 5 pA m  2 are
taken as the period and amplitude of I SM , respectively. SM 5 ms is
the positive input duration which determines the duty circle of the
pulse train.
2.3. Synaptic currents
I syn in Eq. (1) is the current due to synaptic coupling, whose
general form is
X
I - g - V  V - sk
16
k

where / represents the pre-/post-synaptic neuron type, which

can be the GPe, GPi, STN or TC neurons. Here, TC neuron only
receives inhibitory synaptic input from GPi, and does not set
synaptic input to any other neurons. g - is the maximal synaptic
conductance, V - is the synaptic reversal potential, synaptic
variable sk is the synaptic conductance of the kth pre-synaptic
P
neuron and the term k sk sums over the pre-synaptic neurons. For
example, in network model considered here, each GPe neuron
receives excitatory input from a single STN neuron and inhibitory
inputs from two other neighboring GPe neurons in the same
structure. Hence, the synaptic current in the membrane voltage
equation for the ith GPe has the following expression:
I i-GPe GPe g STN-GPe V GPe t  V STN-GPe  siSTN t
1
1
t siGPe
t
g GPe-GPe V GPe t  V GPe-GPe  siGPe

17
where, is the STN or GPe.
According to Rubin and Terman (2004), the s obeys the oneorder dynamical differential equation of the following form:
s0 A 1  s H 1 V   B s :

18

where
1


H 1 V
H
1 exp  V 
H
is the smooth approximation of the bi-value step function H,
H
where and H are constant parameters. A , B and are also
constants contributing to the synaptic currents.

the stimulation electrodes inserted in the three different structures

(i.e., GPe, STN and GPi) are sequentially activated via delivering a
brief pulse train of length T crs =3, respectively. The delay , as shown
in Fig. 2(b) (i.e., permanent or periodical 1 : 0 ONOFF CRS), between
the onsets of two neighboring HFS pulse trains is T crs =3, which is
precisely the duration of HFS current received by each structure
during a complete CRS period T crs . The m : n ONOFF CRS, that is m
ON-circles stimulation is followed by n OFF-circles stimulation, has
been demonstrated as an effective and weak stimulation protocol to
affect the level of synchrony for the neuronal populations. It is
evident that the number n of OFF-cycles, i.e., the stimulation-free
cycles, can affect the strength of ONOFF CRS. For example, the
larger the number n, the more the stimulation-free cycles. This will
reduce the stimulation duration of CRS in the entire stimulation
process, hence lessening the average stimulation frequency. As an
example, the case of m 3; n 2 (i.e., 3:2 ONOFF CRS), is mainly
considered in this paper. As shown in Fig. 2(c), during the periodical
3:2 ONOFF CRS, 3 ON-circles stimulation is always followed by 2
OFF-circles stimulation. It is the 2 OFF-circles stimulation that can
inuence the phase-resetting action of CRS and the average stimulation strength.
When these three electrodes are activated sequentially, the
impact of this new CRS signals on jth neuron among eight neurons
of each structure for BG can be modeled in the form:
I jCRS k I Stim

3
X

Dk ; k ; jk tPt

k1

k 1; 2; 3; j 1; 8

19

where I Stim 50 pA m , k 1, 2, 3 represent three different

stimulation micro-electrodes applied to GPe, STN and GPi, respectively (see Fig. 1). The spatial prole function (see Fig. 2(d)) is
described as follows:
!
j  k 2
Dk ; k ; j exp 
20
2 2k
2

It determines the stimulation strength delivered on the jth neuron

via kth micro-electrode. We take k 4.5 and k 3 for k 1, 2, 3.
k t in Eq. (19) is the indicator function:

1 kth electrode is active at t;
k t
21
0 otherwise
P(t) is the HF pulse train having the form of rectangular pulse with
unity amplitude:
Pt H sin 2 t=T p  1  H sin 2 t p =Tp 

22

where Tp 6 ms is the period of pulse, p T p =10 0:6 ms is the

duration of positive current. Thus, Eq. (19), together with Eqs.
(20)(22), provides the stimulation distribution of the whole
neuronal populations under various cases of m : n ONOFF CRS
including permanent, i.e., 1:0 ONOFF CRS.

3. Results
2.4. CR stimulation setup
In what follows, we introduce CRS protocol based on the above
network model for the basal ganglia-thalamocortical motor circuit.
Different from the classic DBS (see Fig. 2(a)) or the previous CRS
protocol (Hauptmann and Tass, 2009; Lysyansky et al., 2011; Popovych
and Tass, 2012; Tass et al., 2012), the new CRS that we introduce is the
intermittent HFS pulse current which is sequentially administered to
the different structures of basal ganglia at different times through
three different micro-electrodes (see Figs. 1, 2(b) and 2(c)).
For a complete stimulation cycle of length T crs , where T crs closes
to the mean period T of the collective oscillation for each structure,

Pathological neuronal network with PD (i.e., basal gangliathalamocortical motor circuit with abnormal synaptic coupling)
can be characterized by excessively synchronous activity of STN
neurons, which can contribute to the hypokinetic motor symptoms.
Hence, desynchronizing this abnormal synchronization can possibly
restore the action capability of patients with PD. As an effective
surgical therapy for the PD, the technology of deep brain HFS has
been made for practical application. However, classic DBS has
attached more importance on the ring pattern or ring frequency
of the system, instead of its synchronous activity. Extending previous studies, we will investigate the effect of the proposed CR-HFS

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

161

Classic DBS
Tcrs
GPe
STN

GPi

Permanent CRS
3Tcrs ON

2T

crs

OFF

GPe
STN
GPi
3 : 2 ON OFF CRS
CRS Strength

1
0.8
0.6
0.4

Neuron ( i )
Fig. 2. Stimulation paradigms: (a) Classic DBS, i.e., continuous stimulation with high-frequency. (b) Three stimulation signals (high-frequency (HF) bursts) of the permanent
CRS protocol for the k 3 stimulation structures. T crs =3 corresponding to the delay between the onsets of two neighboring HFS pulse trains, i.e., the duration of HFS
current received by each structure during a complete CRS period T crs , is shown in line with double arrows. Color indicates which structure is stimulated by the corresponding
HF stimulation. (c) 3:2 ONOFF CRS with 3 ON-cycles stimulation followed by 2 OFF-cycles stimulation. (d) Spatial prole function Dk ; k ; j; k 1; 2; 3; j 1; 8, of the
stimulation current decay for the 8 neurons within each stimulation structure. (For interpretation of the references to color in this gure caption, the reader is referred to the
web version of this article.)

protocol above on desynchronization of the STN neuronal subpopulation within the network model.

3.1. Desynchronization effect of CRS-ON

To observe quantitatively the desynchronous level of CRS on
the network model, we introduce a desynchronization metric by
using the standard deviation (see e.g., Gao et al., 2001), which can
be calculated by the form:
8
T p
1 X
>
>
>

m;
>
>
T m1
<

"
#2
N
N
>
>
1X
1X
>
2
i
i
>

x
m

x
m
>
:
Ni1
Ni1

23

where m and T represent the time step index and the total number
of time steps, respectively. i is the index of STN neuron and N 8
is the total number of STN neurons. x V n or rn represents the
membrane potential or slow variable of STN neuron model, where
the subscript n indicates the STN neuron (similarly hereinafter).
Particularly, can also be used as a superior quantity to effectively
measure the spatiotemporal desynchronization of neuronal subpopulation, thus revealing the different desynchronizing levels under
the application of CRS. It is obvious from Eq. (23) that the larger the
, the more desynchronous the neuronal network.
Because the essential bursting synchronization of STN neurons in
beta-band can occur, we mainly consider the slow variable r located
in each STN neuron model to estimate the effect of CRS on the
parkinsonian network. Fig. 3(a) presents the results about values
of with rn for various m : n ONOFF CRS protocols, together with

two planes representing the normal or pathologic state of network

model, respectively.
As shown in Fig. 3(a), without CRS, the pathologic state plane
locates lower position in the space, representing the less value of
(namely reecting the higher synchronous or lower desynchronous level of network). However, the normal state plane is just the
opposite, representing the higher desynchronization or healthy
state of patients with PD. In the presence of CRS, there are various
desynchronizing degrees, relative to the pathologic state and for
different pairs of m; n, because the m : n ONOFF CRS has elevated
the pathologic state plane towards the normal one. Similar results
about values of with Vn are shown in Fig. 3(b) except for the case
of 2 : 2 ONOFF CRS, which fails to increase the value of , but
dropping down the plane of pathologic state.
In order to globally and qualitatively show the effect of m : n
ONOFF CRS on the parkinsonian neuronal network, as an example,
we select the case of m : n 3 : 2, which has a better desynchronous
level, to compare with the normal cases and pathologic states. It can
be seen in Fig. 4(a) that in the absence of stimulation, the pathologic
neuronal network of parkinsonian state evolves into two clusters of
bursting synchrony in the STN neuronal subpopulation as well as the
GPi and GPe (internal and external segment of the globus pallidus)
neuronal subpopulations, which contributes to the rhythmic inhibition of GPi to TC neurons by means of yielding rhythmic excitation
from STN to GPi. However, in the presence of 3:2 ONOFF CRS (see
Fig. 4(b)), the two clusters of bursting synchrony gradually diminish
and eventually become into the random distribution, which represents the clear desynchronizing state as it occurs experimentally in
healthy state (see also Fig. 4(c)).
The CRS is essentially weak stimulation. However, it is important
that the CRS protocol can reset the phase of dynamics, hence
breaking the phase-locked synchrony and leading to a more

162

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

0.04

STN

Normal State

0.035

(rn )

0.03
0.025
0.02

m : n ONOFF CRS

0.015

GPi

0.01
Pathologic State

0.005
5

4
4

12

GPe

Normal State

500

11.5

600

700

800

900

1000

800

900

1000

800

900

1000

t (ms)

(Vn)

11
m : n ONOFF CRS

10.5
10

STN

Pathologic State

9.5
9
8.5
5

3
m

GPi

Fig. 3. Two different desynchronization scales (a) r n and (b) V n of the STN
subpopulation under the m : n ONOFF CRS with different combination (m,n) are
shown between the pathologic state plane (light green) and the normal state plane
(light grey). (For interpretation of the references to color in this gure caption, the
reader is referred to the web version of this article.)

uncorrelated dynamics. This means that CRS can change the

collective ring pattern of the neural populations and make the
random distribution of neuronal activity come into being. A concrete example of the desynchronizing action of CRS on the tightly
correlated dynamics is shown in Figs. 5 and 6. It can be seen that
before stimulation (see Fig. 5(a)), the whole slow variables, r in ;
i 1; 8, of the STN subpopulation in pathologic neuronal network,
intertwine together and gradually separate into two heavily linked
clusters (bursting synchrony), which can be also clearly illustrated
by time traces of two STN neurons (see Fig. 6(a)). It is expected,
nevertheless, that during the stimulation of 3:2 ONOFF CRS, the
connections within the slow variables of STN neurons become loose
and weak (see Fig. 5(b)), and the corresponding time traces (see
Fig. 6(b)) become mutually independent.

3.2. Phase planes of STN cell constant input

For the excessive bursting synchrony of STN subpopulation in
the parkinsonian network model and the desynchronization action
of CRS on the pathologic state, in this section, dynamical method is
adopted to explain the underlying mechanism of bursting activity
and the effect of CRS on which, by making use of the STN neuron
model. The uncoupled model subthalamic nucleus (STN) neuron is
governed by the following ve rst-order ordinary differential

GPe

500

600

700
t (ms)

STN

GPi

GPe

500

600

700
t (ms)

Fig. 4. It can be observed that 3:2 ONOFF CRS applied on the triplet of GPe, STN
and GPi, within the basal ganglia-thalamocortical motor circuit, can dissociate the
strong synchronous subpopulation. (a) Pathologic state, i.e., strong bursting
synchrony before stimulation, (b) during the 3:2 ONOFF CRS and (c) normal state.

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

Pathologic State (rn)

0.25

0.2

0.15

0.1

0.05

100

200

300

400

500

600

700

800

900 1000

600

700

800

900 1000

t (ms)

3 : 2 ONOFF CRS (rn)

0.25

0.2

0.15

0.1

0.05
0

100

200

300

400

500
t (ms)

Fig. 5. Time traces of slow variables rn for the 8 STN neurons within the
parkinsonian network model (a) before stimulation (pathologic state) and
(b) during the 3:2 ONOFF CRS. The insets in the (a) and (b) show the enlargement
of the parts bounded by two vertical dashed lines.

equations:
!
8
P ion
>
1
>
> V0

I
I

I
;
app
>
GPe-STN
CRS
k
<
Cm
k
>
x0 x x1  xt=x ; x h; n; r
>
>
>
:
Ca0  I Ca  I T  kCa Cat

24

where we have neglected the subscript STN in the ve dependent

variables, and the other notations are as in Eqs. (1)(3). V is the
membrane voltage, Ca is the Ca2 concentration, h; n; r represent
the sodium, potassium and T-current inactivation, respectively. I CRS
is the control strategy used here. I GPe-STN is the inhibitory synaptic
current received by one STN neuron from two GPe neurons, which,
according to Eq. (16), can be expressed by the form: I GPe-STN
g GPe-STN V STN  V GPe-STN siGPe t sjGPe t, j i jj 4; i; j 1; 8.
The phase plane can provide us a helpful vision to understand
the dynamical properties of model solutions and analyze how these
solutions change with different synaptic inputs. We can plot
nullclines of both the membrane potential V and slow variable r,
V-nullcline and r-nullcline, on the same 2D-plane (see Figs. 7(a)
and (b)) to further explore the obtained dynamics.
The method introduced here requires that both I CRS and skGPe t;
k 1; 8, are constant parameters, i.e., I CRS  I nCRS and skGPe t  snGPe .
In particular, snGPe can be viewed as the average of the skGPe t,
k 1; 8, with respect to the time and all the GPe neurons. Initially,
we assume that I nCRS 0 in this case. During our simulations (see
Figs. 7(a) and (b)), different nullclines for different values of snGPe are
plotted to observe the responses of the STN neuron. It can be seen that

163

the r-nullcline does not change, whereas the V-nullcline is getting

lower as the snGPe increases. This results from the value of snGPe that
corresponds to the inhibitory input.
As mentioned above, the snGPe can change the nullclines of fast
variable V. And then, the intersections of nullclines can also be
changed. Because these intersections correspond to the steadystates of dynamics system, snGPe can eventually affect the dynamics
of STN model neurons. This leads us to pay much more attention on
the effect of snGPe on the dynamical behaviors. As shown in Fig. 8, the
bifurcation diagram of V with snGPe is plotted as snGPe changes in the
interval [0, 1]. In particular, it is shown that at snGPe  0:2, there is a
supercritical Hopf bifurcation point H1. In addition, the system has
also two neutral saddle bifurcation points, H2, H3, and a point
denoting the limit point (fold) bifurcation, LP. From the theory of
dynamics system, near the supercritical Hopf bifurcation point, H1,
the steady-state solution of the system will switch from a stable
equilibrium point to an unstable one, and a stable limit cycle will
also appear. Especially, as snGPe is close enough to H1 from the right
neighborhood of H1, the system will show oscillation with low
amplitude and high frequency. However, the synaptic conductance
skGPe t is time-dependent and, particularly in parkinsonian conditions, will uctuate on and off. Hence, snGPe t, which is averaged
over the interval [0, t] and all the 8 GPe neurons, is also timedependent. And in pathologic case, snGPe t can also uctuate largely
(around 0.29 during our simulation). For convenience, we still
follow the notation of snGPe in the following studies.
For the pathological state, as uctuation of snGPe is large, the trajectory of system will periodically switch from the high-frequency
and low-amplitude state (corresponding to small snGPe ) to the one
with low-frequency and high-amplitude (corresponding to large
snGPe ). This nally leads to the bursting behavior of the system (see
Fig. 6). It is known that the bursting behavior has been experimentally proven to be dominant pathologic activity (Beurrier et al.,
2002) in neuronal populations relevant for motor disorders including PD. However, the application of HFS can signicantly reduce the
uctuation of snGPe by narrowing the region where skGPe t uctuates.
In addition, HFS can also elevate the value of snGPe . Thus, HFS
can eventually keep the system in a state with almost continuous
spiking activity. It can be further observed that for the STN-DBS
case, there is a larger snGPe ( 0.36 in our simulation) than both the
parkinsonian (ditto, i.e., 0.29) and 3:2 ONOFF CRS cases ( 0.33
in our simulation). So, the system shows a spiking behavior with
both high frequency and large amplitude. This implies an enhanced
activity in stimulation area of STN (Hashimoto et al., 2003). Because
CRS is essentially a weak stimulation, the system state corresponding to 3:2 ONOFF CRS is neither a bursting pattern as the parkinsonian case nor a spiking pattern as the STN-DBS case, but an
alternating state between bursting and spiking patterns. In addition,
CRS can also reset the phase of system variable. So, under the
condition of 3:2 ONOFF CRS, the phases of STN model variables
can be randomly dissociated during a periodic circle, and gradually
reaches to a desynchronizing state.
3.3. Performance evaluation of CRS-ON
Normal-state network models of the basal ganglia-thalamocortical
motor circuit generally show good level of reliability, i.e., the faithfulness of thalamic relay in response to SM inputs. However, in
parkinsonian case, a bad or missed spike (referring to such an event
that the TC neuron responds to SM input so excessively or insufciently that it can re more than one action potential or fails to re) is
more likely to happen. Hence, the effective control algorithms should
be developed to guarantee the spikes of the thalamic cells within
pathologic network to re faithfully, as the thalamic cells receive
inputs from SM. The network model can thereby restore the normal
level of reliability for thalamic relay. Here, two types of paradigms, CRS

164

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

V1(STN)

50
0
50
100

100

200

300

400

500

600

700

800

900

1000

100

200

300

400

500

600

700

800

900

1000

100

200

300

400

500

600

700

800

900

1000

r1,2(STN)

0.25
0.2
0.15
0.1

V2(STN)

50
0
50
100

t (ms)

V1(STN)

50
0
50
100

100

200

300

400

500

600

700

800

900

1000

100

200

300

400

500

600

700

800

900

1000

100

200

300

400

500

600

700

800

900

1000

r1,2(STN)

0.25
0.2
0.15
0.1

V2(STN)

50
0
50
100

t (ms)

Fig. 6. The voltages for the 1st (black curves, up box) and 2nd (red curves, down box) STN neurons, together with their slow variables (black and red curves, middle box) in
the parkinsonian network model (a) before stimulation and (b) during the 3:2 ONOFF CRS. (For interpretation of the references to color in this gure caption, the reader is
referred to the web version of this article.)

and classic DBS, are used to explore, and compare the stimulation
performance on the parkinsonian network model. In order to quantitatively understand the control effect of these stimulation paradigms,
we rstly introduce a reliability-based metric as follows:
Rel 1 

N bad N missed
N SM

25

where N bad or Nmissed is the number of bad or missed spikes of TC

neuron in response to SM input, respectively. Rel represents the
fraction of successful responses under the stimuli of N SM SM inputs.
Because Rel can show the level of reliability for thalamic relay,
we can consider it as a measure of the performance. Results on the
reliability, Rel, of the network model both at rest (i.e., pathologic
and normal state) and parkinsonian case with stimulation control
(i.e., classic DBS, permanent CRS and 3:2 ONOFF CRS) are shown

in Fig. 9. It can be seen that the network model of normal state

exhibits the best level of reliability Z 0:9, while the pathologic
state with no stimulation has the worst level of reliability ( 0.4).
However, as the stimulation control is applied, the reliability of
pathologic network model can be increased signicantly. Furthermore, it is shown in Fig. 9 that for both the 3:2 ONOFF CRS and
the DBS-STN, the performance of reliability is slightly higher than
that of the permanent CRS. Specically, for the normal case and
the pathologic case with no stimulation or with the 3:2 ONOFF
CRS applied, the corresponding responses of thalamic voltage to
the SM inputs can be seen in Fig. 10, which shows that 3:2 ONOFF
CRS can partly restore the relayed delity of the parkinsonian case.
In addition, it is known that HFS, which can be applied on the
patients with PD via implanted micro-electrodes, are actually
cathodicanodic biphasic square current pulses. Because of the

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

165

20
25

0.15

rnullcline

30

0.1
35

Vnullclines

0.05

H1

40

H2

LP

45

0.05

50

0.1

55

H3

60

0.15
70

60

50

40

30

20

65

V (mV)
70

0.1

0.2

0.3

0.4

0.5

0.6

0.7

s *GPe
0.06

Pathologic State (s * = 0.29)

GPe

0.05

3 : 2 ONOFF CRS (s *GPe = 0.33)

0.04

Fig. 8. The one-parameter bifurcation diagram of membrane potential for the STN
neuron, V, with respect to the synaptic conductance, snGPe . H1, corresponding to the
supercritical Hopf bifurcation point around snGPe 0:2, H2, H3, neutral saddle points,
and LP, fold bifurcation point, shown in red dots. (For interpretation of the
references to color in this gure caption, the reader is referred to the web version
of this article.)

STNDBS (s * = 0.36)
GPe

0.03
3:2
ON OFF
CRS

0.02
0.01

Permanent
(1 : 0
ON OFF)
CRS

0
60

55

50

45

40

V (mV)

STN DBS

Fig. 7. Nullclines for the STN model. (a) The r-nullcline (black curve) is shown
along with three different V-nullclines. Each V-nullcline corresponding to a
different value of snGPe as colored (red: snGPe 0:29, green: snGPe 0:33, blue:
snGPe 0:36). (b) The enlargement of the part bounded in the box with dashed
edge in (a). (For interpretation of the references to color in this gure caption, the
reader is referred to the web version of this article.)

complexity, risk and expensiveness of surgical treatment for

changing the battery, the energy-saving is also the crucial and
essential work for designing control algorithm. For example, for
any two stimulation protocols having the matchable values of Rel,
the one which expends less current is more encouragingly to be
adopted. So, the amount of current stimuli can be viewed as
another measure for evaluating the performance of various stimulation paradigms, which can be calculated by using the root
mean square (RMS) value of I crs m:
1
I crs
RMS p  J I crs m J 2
N

26

where I crs is the CRS current, similar to the one in Eq. (1), J  J
represents the 2-norm, m and N represent the time step index and
the total number of time steps, respectively. It can be observed in
Fig. 9 that as the 3:2 ONOFF CRS is applied on the pathologic
state, not only the good score of Rel can be achieved by the system,
but also the least stimulation current can the system consume,
compared with the cases of permanent CRS and classic DBS.
Hence, it can be concluded that the 3:2 ONOFF CRS paradigm
can be the potential candidate for HFS therapy.
In what follows, sGPi , inhibitory synaptic conductance from GPi
to TC neurons, will also be considered, which can indirectly and
substantially contribute to the failure or success of thalamic relay

Pathologic
State

Normal
State

10

20

30

40

50

60

70

80

90

100

Fig. 9. The effect of m : n ONOFF CRS on the parkinsonian system. Comparison of

current expenditure (ICRS , deep blue bars), reliability score (Rel, multiplied by 100
for convenience of comparison, light green bars) and averaged synaptic conductance (snGPi , multiplied by 100 for convenience of comparison, rufous bars) for the
different cases of the parkinsonian network (with DBS or CRS being applied) and
normal network. (For interpretation of the references to color in this gure caption,
the reader is referred to the web version of this article.)

(Rubin and Terman, 2004) by means of I GPi-Th (see Eq. (16)). The
simulation results (see Fig. 9, multiplied by 100 to facilitate comparison) show that snGPi , the average of siGPi t, i 1; 8; t A 0; 1000
with respect to the time and all the GPi neurons, never falls below the
constant snGPi 0:3 either in the normal case or parkinsonian state
with different m : n ONOFF CRS or STN-DBS paradigms applied,
where CRS has the better performance towards the synaptic conductance of normal state. Nevertheless, the snGPi of pathologic case
with no stimulation is obviously lower than 0.3.
From the above analysis, it can be known that comprehensive
factors without being limited to the three aspects mentioned above
should be taken into account to identify the physiologically healthy
level of the network model and assess the control performance of
various stimulation paradigms on pathologic neural network. It is

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

TC (mV)

166

TC (mV)

SM
2
(pA m )

TC (mV)

SM
2
(pA m )

SM
2
(pA m )
0

200

400

600

800

1000

1200

1400

1600

1800

2000

Time (ms)
Fig. 10. The time traces of thalamic voltages (blue curves). Thalamic voltage response for system in the (a) normal case, (b) parkinsonian case and (c) pathologic state with
3:2 ONOFF CRS being applied, respectively. SM inputs (red square waves) are shown below each response. (For interpretation of the references to color in this gure caption,
the reader is referred to the web version of this article.)

clear in Fig. 9 that for the pathologic cases, both snGPi and reliability
are inferior to any other case. In addition, we can also observe that
during the application of various stimulation paradigms on the
pathologic network model, the 3:2 ONOFF CRS case has not only
the comparable superiority, but also the least current expenditure.
In a word, the optimal control program for PD should be a trade-off
among all the factors of the network model, including reliability,
Rel, current expenditure, I crs
RMS and the averaged synaptic conductance, snGPi .
3.4. Phase planes of TC cell constant input
From the dynamic point of view, Rubin and Terman (2004)
investigated the action of classic DBS on TC neuron model and
concluded that DBS can change the nullclines of fast variable V,
hence inuencing the dynamical activity of TC neurons. In this
section, likewise, we attempt to explore the effect of CRS on the TC
neuron model, compared to the case of classic DBS. Similar to the
analysis for the phase plane of uncoupled STN, the uncoupled
model thalamocortical relay (TC) neuron is governed by the
following three rst-order ordinary differential equations:
8
1
0
>
>
>
< V C m  I L  I Na  I K  I T  I GPi-Th I SM ;
0
27
h h1 V  h=h V;
>
>
>
: r 0 r V  h= V
1

where, all the notations are as in Eqs. (8)(10), except for eliding the
subscript Th. V is the membrane voltage, h; r represent the sodium
and T-current inactivation, respectively. Note that the reliability of
thalamic relay is heavily dependent on the availability of IT (Rubin and

Terman, 2004; Schiff, 2010). The nullclines for the fast variable V and
slow variable r are shown in Figs. 11(a) and (b). Generally, in
excitatory neuron models, there exist N-shaped nullclines (e.g., see
Figs. 11(a) and (b): red, pink, green and blue lines) for the fast variable
V (Schiff, 2010) and S-shaped nullclines (e.g., see Figs. 11(a) and (b):
black lines) for the slow variable r (Rubin and Terman, 2004). I SM is
the input of SM to the TC neuron. I GPi-TC is the inhibitory synaptic
current received by one TC neuron from four GPi neurons, which,
according to Eq. (16), can be expressed by the form:
1
4
5
I GPi-TC g GPi-TC V TC  V GPi-TC siGPi siGPi
siGPi
siGPi
, i 1; 3.
The phase-plane method introduced here requires that both I SM
and skGPi t, k 1; 8 are constant parameters, i.e., I SM  I nSM and
skGPi t  snGPi . Particularly, snGPi can be viewed as the average of
skGPi t, k 1; 8 with respect to the time and all the GPi neurons.
Initially, we assume I nSM 0. As shown in Figs. 11(a) and (b), the
r-nullcline does not change, but the V-nullclines will change for
the different levels of snGPi in the normal state, parkinsonian case
with 3:2 ONOFF CRS or classic DBS applied and parkinsonian case
with no stimulation, respectively. This can be used to observe the
responses of the TC neuron. It is still worth noting that each
nullcline is calculated based on the sum of four snGPi , corresponding
to the total synaptic level of each case.
The synaptic conductance skGPi t is periodical square wave
under the parkinsonian condition of network model (Rubin and
Terman, 2004; Schiff, 2010). The corresponding average, snGPi , is a
relatively small positive constant which is far away from the one of
normal case. Hence, the intersection of r-nullcline and V-nullcline
corresponding to the pathologic case with no stimulation is lower
than the other cases (see Figs. 11(a) and (b)). However, because the
T-current underlies the reliability of TC neurons, the system will
physiologically require enough T-current, by obtaining higher r, to

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

dysfunctional synaptic plasticity. Interestingly, Lourens et al. have used

a biologically plausible STN-GPe network model with the synaptic
plasticity to investigate the effect of continuous stimulation and
CR-stimulation on the Parkinson's disease (Lourens et al., 2011;
Lourens, 2013). Their results show that synaptic plasticity can facilitate
the training of the network to re in a less synchronized manner as a
short-duration desynchronizing stimulation is applied sufciently
long, and has sufciently high amplitude. Whereas, when a continuous stimulation is employed, the synaptic plasticity has a negative
effect, where the network becomes more pathological as long as the
stimulation is switched off. Thus, it is reasonable in our network
model to consider the synaptic plasticity of the excitatory connections
sent by N STN neurons to the GPe or GPi subpopulation. The network
model introduced with the synaptic plasticity is mainly characterized
by the revised form of synaptic coupling currents:

1
0.9

rnullcline

0.8
0.7

0.6
0.5
Vnullclines

0.4
0.3
0.2
0.1
0
100

90

80

70

60

50

40

30

V (mV)

3 : 2 ONOFF CRS
STNDBS

0.45

Pathologic State

Sij P e  j IBIij j =P  D e  j IBIij j =D

0.4

1X k
I i-j g i-j V j  V i-j
S s
N k ij i

0.35
0.3
0.25
0.2
84

82

80

28

where Sij is the coefcient indicating the synaptic impact from the ith
neuron to the jth one. The changes of the coupling coefcient Sij, as
shown in Fig. 12(a), can be controlled by the following symmetric and
smooth function (Hauptmann and Tass, 2009; Pster and Gerstner,
2006):

Normal State

0.5

167

78

76

74

72

V (mV)
Fig. 11. Nullclines for the TC neuron model. (a) The r-nullcline (S-shaped, black
curve) is shown along with four different V-nullclines (N-shaped, colored curves).
Each corresponding to a different case as colored (red: normal state, pink: 3:2 ON
OFF CRS, green: STN-DBS, blue: pathologic state). (b) The enlargement of the part
bounded in the box with dashed edge in (a). (For interpretation of the references to
color in this gure caption, the reader is referred to the web version of this article.)

overturn the peaks in the V-nullclines (i.e., the knees of these

curves), when the intersections of nullclines for the fast variable V
and slow variable r are low on the phase plane (Schiff, 2010). With
the skGPi t periodically oscillating, the availability of T-current will
also uctuate on and off. But, HFS can suppress the uctuation and
keep r in a moderate level, which can guarantee TC neurons to re
reliably in response to the SM input. By applying HFS, as shown in
Fig. 11(a) and (b), the snGPi will also be elevated, hence raising the
V-nullclines corresponding to the cases of classic DBS and 3:2
ONOFF CRS, respectively. And, the V-nullcline in the case of 3:2
ONOFF CRS is closer to the normal one (corresponding to the
largest value of snGPi ) or farther away from the parkinsonian one
(corresponding to the smallest value of snGPi ) than that of the classic
DBS case. This implies that 3:2 ONOFF CRS has the better control
effect on the parkinsonian network model.
3.5. Open-loop CRS with synaptic plasticity
Recent experimental results (Shen et al., 2003) have demonstrated
that there exists strong synaptic plasticity within the excitatory
connections of subthalamic nucleus (STN) subpopulation. In addition,
model-based simulation analysis (Schroll et al., 2014) has also conrmed that both the motor impairments and pathway imbalances
emerging in Parkinson's disease can indeed be explained by the

29

where P 0:038, D 0:02, P 10 ms, D 25 ms are constant

parameters. IBIij is the inter-burst interval of different neurons within
the coupled neuronal population. That is, the onsets of bursts will be
used to control the changes of the synaptic coupling strength. This is
because the network model of parkinsonian case is characterized by
pathologic bursting activity (Beurrier et al., 2002). Particularly, STN
neurons are busters in this case. From Eq. (29) (see also Fig. 12(a)), it
can be known that the synaptic coupling strength between any two
neurons would be potentiated as the bursts in these neurons are close
enough to each other (e.g., IBIij o 10 ms), whereas weakly related
bursts (e.g., IBIij 4 20 ms) can depotentiate the synaptic connection.
This is evident, physiologically. The synchronized neuronal population,
which serves as a model for the disease state, will get adapted to each
other and learn the abnormally strong synaptic interplay. But the
desynchronized neuronal population, which serves as the model for
the healthy state, will alienate from each other, leading to a faithful
unlearning of synaptic weights.
The net plasticity effect reects the overall level of potentiation
or depotentiation for the synaptic coupling strength over a large
series of bursts. Within the interval  ; , the net plasticity
effect can be calculated according to the function S (Hauptmann
and Tass, 2009):
Z
1
S
Sij d
30
2 
As shown in Fig. 12(b), for small (e.g., o 30 ms), i.e., synchronizing stimulation, the net synaptic effect shows a synaptic potentiation, whereas the large , which corresponds to desynchronizing
stimulation, leads to a synaptic depotentiation.
The simulation results on the pathologic network model with
synaptic plasticity is shown in Fig. 13. It can be seen (Fig. 13(a)) that
under the control of 3:2 ONOFF CRS, the network model can display
fully random neural ring, compared with the previous results (see
Fig. 4). This shows the adequate uncorrelation within STN neurons,
i.e., the fairly desynchronizing state, which is consistent with the
previous works (Lourens et al., 2011; Lourens, 2013). In addition, the
inuence of synaptic plasticity on the reliability of thalamic relay is
also shown in Fig. 13(b). When the parkinsonian network model is
applied by the control of 3:2 ONOFF CRS and subjected to the
synaptic plasticity, the SM inputs can be accurately relayed by the TC
neurons, reaching to a perfect reliability of thalamic relay. Hence, as

168

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

20

x 10

15

S ij

10

5
100

80

60

40

20

20

40

60

80

100

IBI ij (ms)

20

x 10

is that whenever an action potential is recorded from either the

M1 or GPi, xed latency of 80 ms later, a single pulse or seven
continuous pulses (a short train with 130 HZ) will be delivered to
the GPi through a pair of electrodes. In addition, Schiff (2010,
2012) investigated the effect of proportional feedback control laws
on the reliability of TC neuron and Guo and Rubin (2011) surveyed
the action of proportional feedback control laws under the multisite stimulation of the STN. Motivated by this, we will project a
simple control law based on the proportional feedback method,
which can be used as a specic closed-loop control protocol.
As mentioned in the previous sections, the synaptic conductances, skGPi t k 1; 8, from GPi to TC neurons, are timedependent. Especially, in parkinsonian case, all the skGPi t will
uctuate sharply, which can thereby lead to a uctuant averaged
synaptic conductance, snGPi t, over all the 8 GPi neurons. Obviously,
snGPi t is still time-dependent. Thus, it is instructive to restore the
normal brain function by eliminating the pathologic uctuation
and attempting to draw this uctuant wave near the normal
temporal average. For this purpose, we design a proportional
control law:
(

15

S()

10

10

20

30

40

50

60

70

80

90

100

(ms)

Fig. 12. (a) Symmetrical plasticity curve, adapted from Hauptmann and Tass
(2009): for all pairs ij of neurons, which receive excitations from STN within the
network model, the onsets of their bursts are detected and the inter-burst intervals,
IBIijs, are calculated. The plotted curve with respect to IBIij reects the change of
synaptic strength. (b) The net plasticity effect calculated according to Eq. (30) (see
also Hauptmann and Tass (2009)). (For interpretation of the references to color in
this gure caption, the reader is referred to the web version of this article.)

shown in Figs. 13(a) and (b), synaptic plasticity can strengthen the
effect of CRS not only on the desynchronization of neuronal population, but also on the reliability of thalamic relay. The reason for this
may be elusive from the dynamic point of view, but it is physiologically signicant that synaptic plasticity can further improve the
ability of pathological system to learn or repair towards the
healthy state.
3.6. Closed-loop CRS
Experimental evidence has qualitatively shown that closedloop DBS is superior to the open-loop strategy in ameliorating
parkinsonism (Rosin et al., 2011). Crucially, the CRS can be
performed in both an open-loop and a closed-loop model (Tass,
2003a). This implies that CRS is possibly the most promising
candidate for clinical treatment (Lysyansky et al., 2011). Therefore,
it is still well worth exploring the effect of CRS on the desynchronization of network model and the faithfulness of thalamic relay
under the condition of closed-loop strategy (see Fig. 14). This can
provide a more profound insight into the pathophysiology of
Parkinson's disease. However, searching for an effective adaptive
strategy is still the key to design the closed-loop feedback control
algorithm. The method used in the experiment (Rosin et al., 2011)

I CRS t kgain   H ;
snGPi t  sn  T

31

where H is the heaviside function (see also Eq. (15)), kgain is a

constant representing the proportional feedback gain and T is the
threshold to determine when the CRS turns on. In addition, sn is
the average of the synaptic conductances from GPi to TC neurons
in the network model with normal condition. During our simulation, we take kgain 250 or 300 pA m  2 and T 0.05, respectively. And sn is calculated over t 0; 1000 and all the GPi
neurons. It can be known from Eq. (31) that as the difference
between snGPi t and sn is lower than the threshold, i.e., o 0,
ICRS(t) vanishes. The rationale for this, neurophysiologically, is in
the self-regulation of biological systems. In addition, if 4 0,
there exists positive correlation between and ICRS(t). This represents the feedback control of closed-loop strategy.
As shown in Fig. 14, the 901-clockwise rotated convex-shaped
box with thin dot-dashed edge represents the model of basal
ganglia-thalamocortical motor circuit (adapted from Rubin and
Terman, 2004), which can be called as the physiological innerloop. Outside this box, it is the closed-loop control circuit (denoted
by short and thick grey arrowheads), which can be viewed as the
physical outer-loop. The closed-loop controller rst detects the
averaged signal from the inner-loop about the synaptic conductance received by TC neurons from GPi, then copes with the
obtained signal using the proportional control law provided above
(Eq. (31)), and nally determines whether the CRS current should
be applied on the stimulation structures of the inner-loop and
simultaneously completes the control activity.
For comparison, Fig. 15 presents the results on desynchronization
effect, reliability and current expenditure of parkinsonian case under
the conditions of open-loop 3:2 ONOFF CRS and closed-loop 3:2 ON
OFF CRS with kgain 250 or 300 pA m  2 . As shown in Fig. 15, the
system obtains the best faithfulness of thalamic relay and the largest
desynchronizing effect in the case of closed-loop 3:2 ONOFF CRS with
kgain 300 pA m  2 , but this process will need expending more
stimulation currents than the other cases. Moreover, the system, under
the
condition
of
closed-loop
3:2
ONOFF
CRS
with
kgain 250 pA m  2 , shows slightly less reliability than that of the
open-loop 3:2 ONOFF CRS case, but it also presents better desynchronizing effect and less stimulation currents expenditure. Also, in
the case of open-loop 3:2 ONOFF CRS, the system achieves moderate
reliability and current consumption, but the worst desynchronizing
effect. Overall, the closed-loop stimulation paradigms have the signicantly better desynchronizing effect on the pathological neural

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

169

8
7

STN

6
5
4
3
2
1
500

600

700

800

900

1000

t (ms)
0

TC(mV)

20
40
60
80

100

200

300

400

500
t (ms)

600

700

800

900

1000

Fig. 13. The CRS effect of the parkinsonian system subject to plasticity. (a) The raster picture reects the stronger desynchronizing effect and (b) the thalamic voltage (blue
curve) responds more faithfully to the inputs of SM (red square wave) in the parkinsonian case with 3:2 ONOFF CRS applied and subject to the synaptic plasticity.
(For interpretation of the references to color in this gure caption, the reader is referred to the web version of this article.)

Fig. 14. The schematic diagram of closed-loop CRS of the parkinsonian system. The
basal ganglia-thalamocortical motor circuit (also called physiological inner-loop) is
shown in the 901-clockwise rotated convex-shaped box with thin dot-dashed edge
(see also Rubin and Terman, 2004). Inhibitory connections are shown in blue
dashed arrows, excitatory connections are shown in red dashed arrows. The closedloop control circuit (also called physical outer-loop) is denoted by short and thick
grey arrowheads. Excitatory CRS is shown in green arrows. The averaged signal
from the inner-loop about the synaptic conductance received by thalamus from GPi
is rstly detected by the controller and then coped with using the proportional
control law of the controller (Eq. (31)). Finally, the results returned by the controller
are used to determine whether the system should be subject to the excitatory CRS
feedback drive (green arrows). (For interpretation of the references to color in this
gure caption, the reader is referred to the web version of this article.)

Fig. 15. The effect of closed-loop CRS on the parkinsonian system. Comparison of
current expenditure (I CRS , rufous bars), reliability score (Rel, multiplied by 100 for
convenience of comparison, light green bars) and desynchronization scale (r n ,
multiplied by 1000 for convenience of comparison, deep blue bars) in the
parkinsonian case applied by the open-loop 3:2 ONOFF CRS and closed-loop 3:2
ONOFF CRS with kgain 250 or kgain 300 pA m  2 , respectively. (For interpretation of the references to color in this gure caption, the reader is referred to the
web version of this article.)

network, which is consistent with the experimental results (Rosin

et al., 2011). However, the advantage of closed-loop paradigms
strongly depends on the choice of optimal parameters and actual
clinical demands. For example, in some cases, the closed-loop stimulation paradigm can get the reasonable desynchronizing effect and good

170

D. Fan, Q. Wang / Journal of Theoretical Biology 370 (2015) 157170

reliability, but costs more currents. In other cases, even though it needs
expending less currents and gains better desynchronizing effect, the
reliability can not be what we expect. Therefore, the results presented
in our simulation can provide a profound insight into the HFS therapy
for PD. It seems that, without considering the choice of optimal
parameters, it is incorrect to believe that one approach is always
superior to the others.

4. Conclusion
In this paper, we have used a sparsely connected, structured
network model of Parkinson's disease (Rubin and Terman, 2004) to
explore the impact of the newly proposed CRS strategy on the
pathologic synchronization and the faithfulness of thalamic relay.
Different from the previous stimulation paradigms, the new type of
CRS protocol proposed here is performed on three different structures GPe, STN and GPi within the basal ganglia-thalamocortical
motor circuit, respectively, with a triplet of electrodes. The model
was tuned, referring to Rubin and Terman (2004), to generate
parkinsonian behavior, which can feature excessively synchronized
bursting patterns red by the neurons of STN subpopulation, worse
faithfulness of thalamic relay, and lower average synaptic conductance due to the large uctuations. Results have shown that, as CRS,
especially the m : n ONOFF CRS, is applied, the pathological
features indicating parkinsonian state can be greatly improved.
The results are expected to help the understanding of the pathophysiology of PD, even though the feasibility of CRS still needs to be
clinically or experimentally tested. In addition, due to the synaptic
plasticity within the excitatory connections of the STN subpopulation, we have also attempted to explore the impact of CRS on
desynchronization of parkinsonian networks with the synaptic
plasticity. It is shown that compared to the case without the synaptic
plasticity, the synaptic plasticity can endow the proposed CRS with
the ability of stronger desynchronization and better improvement of
the reliability. Also, we have considered the effect of closed-loop
strategy, which has conrmed the superiority of closed-loop protocol to the open-loop one. Our results have revealed that the
advantage of closed-loop strategy is strongly dependent on the
choice of optical parameters. In the future, the obtained results can
have crucial guiding for searching the potential candidate of clinical
application.

Acknowledgments
This work was supported by the National Natural Science Foundation of China (Nos. 11325208 and 11172017), the Research Fund for
the Doctoral Program of Higher Education (No. 20121102110014).
References
Beurrier, C., Garcia, L., Bioulac, B., Hammond, C., 2002. Subthalamic nucleus: a clock
inside basal ganglia?. Thal. Relat. Syst. 2 (01), 18.
Brown, P., Oliviero, A., Mazzone, P., Insola, A., Tonali, P., DiLazzaro, V., 2001.
Dopamine dependency of oscillations between subthalamic nucleus and
pallidum in Parkinson's disease. J. Neurosci. 21 (3), 10331038.
Dovzhenok, A., Park, C., Worth, R.M., Rubchinsky, L.L., 2013. Failure of delayed
feedback deep brain stimulation for intermittent pathological synchronization
in Parkinson's disease. PloS One 8 (3), e58264.
Feng, X.J., Greenwald, B., Rabitz, H., Shea-Brown, E., Kosut, R., 2007a. Toward closedloop optimization of deep brain stimulation for Parkinson's disease: concepts
and lessons from a computational model. J. Neural Eng. 4 (2), L14.
Feng, X.J., Shea-Brown, E., Greenwald, B., Kosut, R., Rabitz, H., 2007b. Optimal deep
brain stimulation of the subthalamic nucleusa computational study. J. Comput.
Neurosci. 23 (3), 265282.
Gao, Z., Hu, B., Hu, G., 2001. Stochastic resonance of small-world networks. Phys.
Rev. E 65 (1), 016209.

Guo, Y., Rubin, J.E., 2011. Multi-site stimulation of subthalamic nucleus diminishes
thalamocortical relay errors in a biophysical network model. Neural Netw. 24
(6), 602616.
Hammond, C., Bergman, H., Brown, P., 2007. Pathological synchronization in
Parkinson's disease: networks, models and treatments. Trends Neurosci. 30
(7), 357364.
Hashimoto, T., Elder, C.M., Okun, M.S., Patrick, S.K., Vitek, J.L., 2003. Stimulation of
the subthalamic nucleus changes the ring pattern of pallidal neurons.
J. Neurosci. 23 (5), 19161923.
Hauptmann, C., Tass, P.A., 2009. Cumulative and after-effects of short and weak
coordinated reset stimulation: a modeling study. J. Neural Eng. 6 (1), 016004.
Khn, A.A., Kempf, F., Brcke, C., Doyle, L.G., Martinez-Torres, I., Pogosyan, A.,
Brown, P., 2008. High-frequency stimulation of the subthalamic nucleus
suppresses oscillatory activity in patients with Parkinson's disease in parallel
with improvement in motor performance. J. Neurosci. 28 (24), 61656173.
Levy, R., Ashby, P., Hutchison, W.D., Lang, A.E., Lozano, A.M., Dostrovsky, J.O., 2002.
Dependence of subthalamic nucleus oscillations on movement and dopamine
in Parkinsons disease. Brain 125 (6), 11961209.
Little, S., Pogosyan, A., Neal, S., Zavala, B., Zrinzo, L., Hariz, M., Brown, P., 2013.
Adaptive deep brain stimulation in advanced Parkinson's disease. Ann. Neurol.
74 (3), 449457.
Lourens, M.A.J., 2013. Neural Network Dynamics in Parkinson's Disease. Doctoral
Dissertation, University of Twente.
Lourens, M.A., Nirody, J.A., Meijer, H.G., Heida, T., van Gils, S.A., 2011. The effect of
spike time dependent plasticity on activity patterns in the basal ganglia. BMC
Neurosci. 12 (Suppl 1), 351.
Lysyansky, B., Popovych, O.V., Tass, P.A., 2011. Desynchronizing anti-resonance effect
of m: n ONOFF coordinated reset stimulation. J. Neural Eng. 8 (3), 036019.
Meijer, H.G.E., Krupa, M., Cagnan, H., Lourens, M.A.J., Heida, T., Martens, H.C.F., Bour,
L.J., van Gils, S.A., 2011. From Parkinsonian thalamic activity to restoring
thalamic relay using deep brain stimulation: new insights from computational
modeling. J. Neural Eng. 8 (6), 066005.
Meissner, W., Leblois, A., Hansel, D., Bioulac, B., Gross, C.E., Benazzouz, A., Boraud, T.,
2005. Subthalamic high frequency stimulation resets subthalamic ring and
reduces abnormal oscillations. Brain 128 (10), 23722382.
Nini, A., Feingold, A., Slovin, H., Bergman, H., 1995. Neurons in the globus pallidus do
not show correlated activity in the normal monkey, but phase-locked oscillations
appear in the MPTP model of parkinsonism. J. Neurophysiol. 74 (4), 18001805.
Pster, J.P., Gerstner, W., 2006. Triplets of spikes in a model of spike timingdependent plasticity. J. Neurosci. 26 (38), 96739682.
Popovych, O.V., Tass, P.A., 2012. Desynchronizing electrical and sensory coordinated
reset neuromodulation. Front. Human Neurosci., 6.
Rinzel, J., 1985. Excitation dynamics: insights from simplied membrane models.
Fed. Proc. 15 (44), 29442946.
Rosin, B., Slovik, M., Mitelman, R., Rivlin-Etzion, M., Haber, S.N., Israel, Z., Bergman,
H., 2011. Closed-loop deep brain stimulation is superior in ameliorating
parkinsonism. Neuron 72 (2), 370384.
Rubin, J.E., Terman, D., 2004. High frequency stimulation of the subthalamic
nucleus eliminates pathological thalamic rhythmicity in a computational
model. J. Comput. Neurosci. 16 (3), 211235.
Saxena, S., 2011. Neuronal Activity of the Globus Pallidus Internus: Correlates
Related to Cued Movements, Parkinson's Disease, and Deep Brain Stimulation.
Doctoral Dissertation, Johns Hopkins University.
Schiff, S.J., 2010. Towards model-based control of Parkinson's disease. Philos. Trans.
R. Soc. A: Math. Phys. Eng. Sci. 368 (1918), 22692308.
Schiff, S.J., 2012. Neural Control Engineering: The Emerging Intersection Between
Control Theory and Neuroscience. MIT Press.
Schroll, H., Vitay, J., Hamker, F.H., 2014. Dysfunctional and compensatory synaptic
plasticity in Parkinson's disease. Eur. J. Neurosci. 39 (4), 688702.
Shen, K.Z., Zhu, Z.T., Munhall, A., Johnson, S.W., 2003. Synaptic plasticity in rat subthalamic
nucleus induced by high-frequency stimulation. Synapse 50 (4), 314319.
Smirnov, D.A., Barnikol, U.B., Barnikol, T.T., Bezruchko, B.P., Hauptmann, C., Bhrle,
C., Tass, P.A., 2008. The generation of Parkinsonian tremor as revealed by
directional coupling analysis. EPL (Europhys. Lett.) 83 (2), 20003.
Tass, P.A., 2003a. A model of desynchronizing deep brain stimulation with a
demand-controlled coordinated reset of neural subpopulations. Biol. Cybern.
89 (2), 8188.
Tass, P.A., 2003b. Desynchronization by means of a coordinated reset of neural subpopulations a novel technique for demand-controlled deep brain stimulation.
Progr. Theor. Phys. Suppl. 150, 281296.
Tass, P.A., Majtanik, M., 2006. Long-term anti-kindling effects of desynchronizing
brain stimulation: a theoretical study. Biol. Cybern. 94 (1), 5866.
Tass, P.A., Silchenko, A.N., Hauptmann, C., Barnikol, U.B., Speckmann, E.J., 2009.
Long-lasting desynchronization in rat hippocampal slice induced by coordinated reset stimulation. Phys. Rev. E 80 (1), 011902.
Tass, P.A., Qin, L., Hauptmann, C., Dovero, S., Bezard, E., Boraud, T., Meissner, W.G.,
2012. Coordinated reset has sustained aftereffects in Parkinsonian monkeys.
Ann. Neurol. 72 (5), 816820.
Terman, D., Rubin, J.E., Yew, A.C., Wilson, C.J., 2002. Activity patterns in a model for
the subthalamopallidal network of the basal ganglia. J. Neurosci. 22 (7),
29632976.
Timmermann, L., Gross, J., Dirks, M., Volkmann, J., Freund, H.J., Schnitzler, A., 2002.
The cerebral oscillatory network of parkinsonian resting tremor. Brain 126 (1),
199212.