Computational analysis of subthalamic nucleus and lenticular fasciculus activation during
therapeutic deep brain stimulation
Computational analysis of STN DBS
Svjetlana Miocinovic1,2, Martin Parent3, Christopher R. Butson2, Philip J. Hahn2, Gary S. Russo4,
Jerrold L. Vitek4, Cameron C. McIntyre1,2
1Department of Biomedical Engineering, Case Western Reserve University, Cleveland, Ohio
2Department of Biomedical Engineering, Cleveland Clinic Foundation, Cleveland, Ohio
3Centre de Recherche Université Laval Robert-Giffard, Beauport, Québec, Canada
4Department of Neuroscience, Cleveland Clinic Foundation, Cleveland, Ohio
Cameron C. McIntyre, Ph.D.
Department of Biomedical Engineering
Cleveland Clinic Foundation
9500 Euclid Avenue ND20
Cleveland, OH, 44195
Phone: (216) 445-3264
Fax: (216) 444-9198
Page 1 of 42
Articles in PresS. J Neurophysiol (May 31, 2006). doi:10.1152/jn.00305.2006
Copyright © 2006 by the American Physiological Society.
The subthalamic nucleus (STN) is the most common target for the treatment of
Parkinson’s disease (PD) with deep brain stimulation (DBS). DBS of the globus pallidus internus
(GPi) is also effective in the treatment of PD. The output fibers of the GPi that form the
lenticular fasciculus pass in close proximity to STN DBS electrodes. In turn, both STN
projection neurons and GPi fibers of passage represent possible therapeutic targets of DBS in the
STN region. We built a comprehensive computational model of STN DBS in parkinsonian
macaques to study the effects of stimulation in a controlled environment. The model consisted of
three fundamental components: 1) a 3D anatomical model of the macaque basal ganglia, 2) a
finite element model of the DBS electrode and electric field transmitted to the tissue medium,
and 3) multi-compartment biophysical models of STN projection neurons, GPi fibers of passage
and internal capsule fibers of passage. Populations of neurons were positioned within the 3D
anatomical model. Neurons were stimulated with electrode positions and stimulation parameters
defined as clinically effective in two parkinsonian monkeys. The model predicted axonal
activation of STN neurons and GPi fibers during STN DBS. Model predictions regarding the
degree of GPi fiber activation matched well with experimental recordings in both monkeys. Only
axonal activation of the STN neurons showed a statistically significant increase in both monkeys
when comparing clinically effective and ineffective stimulation. Nonetheless, both neural targets
may play important roles in the therapeutic mechanisms of STN DBS.
Keywords: Parkinson’s disease; globus pallidus; electric field model; neuron model; clinical
efficacy; parameter selection
Page 2 of 42
Deep brain stimulation (DBS) has become an established clinical therapy for advanced
Parkinson’s disease (PD) (Obeso et al., 2001). Chronic high frequency electrical stimulation of
subcortical structures can provide more than 50% improvement in clinical ratings of motor
symptoms (Walter and Vitek 2004). However, the impressive clinical efficacy achieved by DBS
has occurred without a clear understanding of the therapeutic mechanisms of action. In addition,
a number of adverse effects can be generated by DBS including sensorimotor impairments,
involuntary movements (stimulation-induced dyskinesias), as well as speech, mood and
cognitive disturbances (Okun et al. 2005; Volkmann et al. 2002). Often these side-effects can be
avoided or alleviated with proper adjustment of the stimulation settings (Krack et al. 2003).
However, further improvements in the engineering design and clinical implementation of DBS
technology will rely on addressing a number of questions on the effects of DBS on the nervous
system. The fundamental goal of this project was to enhance our understanding of the target
neural elements of the stimulation.
The subthalamic nucleus (STN) represents the most common anatomical target for DBS
treatment of PD (Limousin et al. 1998). Electrodes placed in the STN are surrounded by several
neural types (local projection neurons and their axons, fibers of passage, afferent inputs, etc.),
but knowledge of the response properties of these different neural types to DBS is limited.
Therapeutic DBS electrode contacts are typically located in the region of the dorsal STN,
lenticular fasciculus (LF or Forel’s field H2) and zona incerta (Voges et al. 2002; Saint-Cyr et al.
2002; Starr et al. 2002; Hamel et al. 2003; Yelnik et al. 2003; Zonenshayn et al. 2004; Nowinski
et al. 2005). STN projection neurons send their highly collateralized axons to the globus pallidus,
striatum and substantia nigra (Sato et al. 2000). The LF courses just dorsal to the STN and is
Page 3 of 42
composed of fibers from the internal segment of the globus pallidus (GPi), which carry output
from the basal ganglia to the thalamus (Parent et al. 2001; Parent and Parent 2004). Given that
GPi DBS provides similar therapeutic benefits as STN DBS (Burchiel et al. 1999; Obeso et al.
2001; Rodriguez-Oroz et al. 2005), both STN projection neurons and pallidothalamic (GPi)
fibers represent viable candidates as the therapeutic target of the stimulation (Parent and Parent
2004). However, the axons of local projection neurons and fibers of passage respond at similar
extracellular stimulation thresholds (Ranck 1975; Nowak and Bullier 1998; McIntyre and Grill
1999) making it difficult to determine which neuron types are activated during STN DBS.
The extent of neural activation generated by extracellular stimulation depends on the
stimulation parameters, electrode and tissue electrical properties, and the position and orientation
of neural elements with respect to the electrode (Ranck 1975; Tehovnik 1996; McIntyre et al.
2004a). To address these issues in the context of STN DBS, we developed a comprehensive
computer model with detailed representation of the 3D neuroanatomy, the time-dependent
electric field generated by DBS electrodes, and the underlying biophysics that regulate the neural
response to stimulation. We tested our hypothesis that both STN projection neurons and GPi
fibers of passage are activated during clinically effective STN DBS. This would imply that both
neural types could play a role in the therapeutic mechanisms of STN DBS, and prompt further
investigation into electrode localization and stimulation parameter selection techniques for
optimizing the stimulation to individual subjects.
We customized our modeling framework to analyze neural activation in two parkinsonian
macaques implanted with chronic scaled clinical DBS systems (Hashimoto et al. 2003). Our goal
was to theoretically reproduce the experimental effects of STN DBS that improved parkinsonian
symptoms (bradykinesia and rigidity) in the two monkeys. Our first aim was to quantify the
Page 4 of 42
proportion of STN projection neurons and GPi fibers of passage that were activated during
clinically effective and ineffective stimulation. We found a significant increase in axonal
activation of STN projection neurons during clinically effective compared to ineffective
stimulation. Considerable GPi fiber activation was observed in only one of the two monkeys.
Single unit extracellular recordings of short latency, presumably antidromic, GPi activation
during STN DBS in both monkeys support our model predictions. The second aim was to
analyze how changes in electrode location affected neural activation. We found that sub-
millimeter shifts can alter neural activation, highlighting the importance of precise electrode
positioning. The final aim was to investigate the effects of stimulation-induced trans-synaptic
inhibition on somatic and axonal firing in STN projection neurons during STN DBS. We found
that somatic firing was reduced during DBS compared to the spontaneous pre-stimulation
activity, as seen experimentally (Welter et al. 2004; Filali et al. 2004; Meissner et al. 2005).
However, axonal activation was largely unaffected by the somatic inhibition and the axon easily
fired at the stimulation frequency (McIntyre et al. 2004a). Preliminary portions of this work have
been presented in abstract form (Miocinovic et al. 2004).
The implementation of chronic scaled clinical DBS systems in parkinsonian macaques
provides the foundation for detailed study of the therapeutic mechanisms of the stimulation
(Hashimoto et al., 2003; Elder et al. 2005). We coupled our experimental results with detailed
computer modeling to provide new insight into the cellular effects of STN DBS. We developed
computational models customized to two parkinsonian macaques implanted with STN DBS
systems. Each model consisted of three fundamental components: 1) a 3D anatomical model of
Page 5 of 42
the monkey basal ganglia, 2) a finite element model of the DBS electrode and electric field
transmitted to the tissue medium, and 3) multi-compartment biophysical models of reconstructed
STN projection neurons, GPi fibers of passage and internal capsule fibers of passage.
Populations of the neuron models were placed within context of the 3D anatomical model and
the histologically defined DBS electrode positions. The DBS electric field model was then
applied to the neuron models, thereby allowing for theoretical prediction of the neural response
to the stimulation.
We built a three-dimensional (3D) reconstruction of the basal ganglia for monkey R7160
(Macaca mulatta) (Hashimoto et al., 2003). Digital atlas templates of the macaque brain (Martin
and Bowden 2000) were warped in Edgewarp v3.28 (Bookstein 1990) to histological brain slices
to identify borders of nuclei not visible directly in the Nissl stained sections (Fig. 1). Nuclei of
interest were outlined in the 2D warped atlas slices, spaced in 1 mm increments. 3D volumes
were created by interpolating between these contour lines using the graphical modeling program
Rhinoceros v3.0 (McNeal & Associates, Seattle, WA). The resulting 3D brain atlas provided an
anatomically realistic virtual space to position the DBS electrode and the neuron models (Fig. 1).
The DBS electrode trajectory was reconstructed from the histological slices, and it was verified
that it matched the electrode position drawing in the top row of Figure 1 from Hashimoto et al.
(2003). The brain of the second monkey (R370) was not available; therefore, we used the same
3D atlas, but manually adjusted the DBS electrode position until the sagittal cross-section from
the 3D atlas matched the rendering in the bottom row of Figure 1 from Hashimoto et al. (2003).
Page 6 of 42
The 3D geometry of a longtailed macaque’s STN projection neuron (Macaca
fascicularis) was reconstructed using biotin dextran amine labeling and axonal tracing as
described by Sato et al. (2000) (Fig. 2). The neuron geometry was defined using Neurolucida
(MicroBrightField, Inc., Williston, VT) and converted for display in Rhinoceros and our 3D
brain atlas. Axonal tracing experiments (Sato et al. 2000) have revealed that STN projection
neurons course either dorsally along the ventral border of the thalamus or ventrally along the
lateral border of the STN on their way to the globus pallidus. To account for this anatomical
variability, the original neuron reconstruction was used to create two additional STN neuron
geometries with alternative axonal paths (Fig. 3A). The GPi axon geometry was based on the
description of lenticular fasciculus trajectory from Parent and Parent (2004). The LF fibers
emerged dorsomedially from the GPi, crossed the IC at a level dorsal to the STN, turned caudally
to run along the dorsal border of the STN in Forel’s field H2, joined the ansa lenticularis in
Forel’s field H, continued through Forel’s field H1, and terminated in the ventral thalamus (Fig.
The internal capsule defines the lateral border of the STN. Consequently, motor evoked
responses from activation of the corticospinal tract (CST) can be elicited with relatively low
thresholds during STN stimulation. To provide a gross level of model validation and connection
to behavioral measurements we incorporated CST axon trajectories into our anatomical
framework. From the level of dorsal thalamus, the CST fibers coursed ventrally at an
approximately 20 degree anterior-to-posterior angle (Fig. 3C).
The three STN neuron geometries, along with the GPi and CST axon trajectories were
placed within the 3D atlas in their respective anatomically realistic positions and orientations
Page 7 of 42
(Fig. 3). Neural populations of the STN neurons, GPi fibers and CST fibers were created by
copying the five basic geometries and distributing them randomly within the atlas, while still
keeping each within their respective anatomical boundaries. In total, three such populations were
created by randomly shifting neurons by ± 250 µm in any direction, and manually repositioning
those that ended up outside their anatomical boundaries. After the populations were established,
a DBS electrode was placed within the STN. The cells were defined as damaged if any part of
the axon, soma or substantial portion of the dendritic tree intersected with the electrode. We
started with 100 STN neurons, 80 GPi fibers and 70 CST fibers, removed all the damaged cells
and then randomly removed additional cells so that the final count for each of the three
populations was 50 cells.
We built multi-compartment cable models of STN projection neurons, GPi fibers of
passage and CST fibers of passage using NEURON v5.8 (Hines and Carnevale 1997). The STN
soma, initial segment and dendrites contained channel dynamics of the rat subthalamic projection
neuron originally developed by Gillies and Willshaw (2006), where dendritic channel
conductance densities scale linearly with distance from the soma (Fig. 2C). To better fit the
membrane dynamics to our neuron geometry and the firing pattern of STN neurons in a
parkinsonian monkey, we modified the original conductances in the following way: the calcium-
activated potassium channel was increased by 80%, the fast potassium rectifier was increased by
20% (soma and initial segment only), and the fast acting sodium channel was decreased by 25%
in the soma and initial segment, and by 35% in the dendrites. The model temperature was set to
36°C to simulate in-vivo conditions. The STN neuron model had a resting potential of
Page 8 of 42
approximately -54 mV and spontaneous tonic firing of 32 Hz consistent with the rate of
36.5±10.8 Hz recorded by Wichmann et al (2002) in the parkinsonian macaque. The neuron
firing rate increased with increasing amplitude of intracellular depolarizing current. The slope of
the model frequency-intensity (f-I) curve was 0.26 Hz/pA, lower than slopes recorded in rat brain
slices (~0.54 Hz/pA in Bevan and Wilson 1999). Despite this discrepancy the model neuron was
able to fire at more than 200 Hz, well above the range of clinically used DBS frequencies (100-
185 Hz). The input resistance measured as the slope of the I-V curve while neuron was in a
hyperpolarized state (-59.6 mV resting potential) and injected with small currents (-0.2 nA to
0.05 nA) was 35 MΩ, twice the value of resistances recorded in rat STN neurons in-vivo (mean
18 MΩ, range 9-28 MΩ in Kita et al. 1983). This discrepancy could be due to a lack of synaptic
inputs into the dendritic tree, differences between monkey and rat STN neuron morphology, or
imperfect measurement of dendritic diameters during histological 3D reconstruction resulting in
underestimate of the surface area. Average rat in-vitro measurements range from 146-200 MΩ
(Nakanishi et al. 1987; Beurrier et al. 1999) placing the model neuron within the range of in-vivo
and in-vitro recordings. The membrane time constant estimated from the 1/e point of the
membrane potential change induced by a low intensity hyperpolarizing current pulse was 6 ms
consistent with in-vivo rat measurements (6±2 ms in Kita et al. 1983).
The axon of the STN neuron as well as the GPi and CST axons were based on the
myelinated axon model originally described by McIntyre et al. (2002). The fiber diameter was
set to 2 µm and the individual segment dimensions and ion channel conductances were adjusted
as previously described (McIntyre et al. 2004a). The axonal resting potential was set to -65 mV.
The GPi axon was induced to fire tonically at 80 Hz by injecting short current pulses at the GPi
terminal end of the axon to mimic parkinsonian macaque GPi firing rate of 80.1±20.2 Hz
Page 9 of 42
(Wichmann et al. 2002). There were no synaptic connections or interactions between any of the
different neurons in the model system.
In some simulations trans-synaptic GABAa input to the STN neurons was added to
evaluate the influence of stimulation-induced trans-synaptic conductances on somatic activity
during high frequency stimulation. This was simplistically modeled as an inhibitory postsynaptic
current applied only to the central compartment of the cell body. The time course and amplitude
of the GABAa synaptic conductance was modeled with experimentally defined first-order
kinetics of the transmitter binding to postsynaptic receptors (Destexhe et al. 1994a,b). Since
afferent inputs (i.e. axonal terminals) are typically more excitable than the passing axons
(McIntyre et al. 2004a; Anderson et al. 2006), in all applicable simulations we assumed that
synaptic inhibition was always activated by each extracellular stimulus pulse, regardless of the
cell position with respect to the electrode.
Electric Field Model
An axisymmetric finite element model (FEM) of the DBS electrode was created in
FEMLAB 3.1 (COMSOL, Inc.,Burlington, MA) to calculate potentials generated in the tissue
medium by the electrode. The stimulating lead was a scaled-down version of the chronic DBS
electrode used in humans (Model 3387, Medtronic Inc., Minneapolis, MN) and consisted of four
metal contacts each with a diameter of 0.75 mm, height of 0.50 mm, and separation between
contacts of 0.50 mm. The volume conductor was 5 cm x 5 cm in size and grounded at the
boundaries. The bulk conductivity of the tissue medium was 0.2 S/m (Ranck 1963), and a 0.25
mm encapsulation sheath (0.1 S/m) surrounded the electrode shaft (Grill and Mortimer 1994;
Butson et al. 2006). The electrode shaft was modeled as an insulator (1e-6 S/m) and each
Page 10 of 42
electrode contact as a conductor (1e6 S/m). Voltage sources were specified at two electrode
contacts for bipolar stimulation. The model had variable resolution mesh with a total of 32,640
mesh elements that increased in size with increasing distance from the electrode. Voltage values
within the volume were determined from the Poisson equation, which was solved using direct
matrix inversion (UMFPACK solver) (Fig. 4).
The stimulus waveform produced by the Itrell II (Medtronic Inc., Minneapolis, MN)
implantable pulse generator (IPG), as used in human DBS and the monkey experiments, is a
voltage-controlled biphasic asymmetric square pulse. However, the actual stimulus delivered to
the brain tissue is modified by the electrode capacitance. To account for the effects of electrode
capacitance, a Fourier finite element model (FEM) was utilized as described in Butson and
McIntyre (2005). The four general steps of this method are briefly described here. First, the IPG
stimulus waveform was constructed in the time domain. It was then converted to frequency
domain using discrete Fourier transform (DFT) in Matlab (Mathworks, Natick, MA). Third, the
FEM model was solved at each component frequency of the DFT (1024 frequencies between 0
and 50kHz). The result at each frequency was scaled and phase shifted using the DFT
magnitudes and phases. Finally, an inverse Fourier transform was performed to obtain the
stimulus waveform in the time domain. The electrode was modeled as purely capacitive (0.65
µF), and adjusted to account for the smaller surface area of monkey electrode contacts compared
to the human DBS electrode (Butson and McIntyre 2005).
We coupled the finite element electric field model with the multi-compartment neuron
models to enable quantitative stimulation predictions in the context of the 3D neuroanatomy.
Page 11 of 42
This coupling was accomplished by applying extracellular voltages from the electric field model
to each compartment of each neuron model and simulating the biophysical response (action
potential signaling) of each neuron over time (McIntyre et al. 2004a) (Figs. 4, 5). The magnitude
of the applied extracellular voltage was dependent on the stimulus amplitude, stimulus
waveform, and the compartment’s distance from the electrode. At each time step of the
simulation the extracellular voltage at each neural compartment was updated to a value
determined by the time-dependent stimulus train delivered to the tissue medium (Figs. 4, 5).
Clinical efficacy for various DBS parameter settings was established in two parkinsonian
macaques using behavioral tests (for details see Hashimoto et al. 2003). In both monkeys, R7160
and R370, the electrode was positioned in the posterior STN at a 20 degree angle in the sagittal
plane. In monkey R7160, bipolar stimulation (contact 0 cathode, contact 2 anode) at 136 Hz and
210 µs pulse width produced consistent improvement in rigidity and bradykinesia at 1.8 V
amplitude (clinically effective stimulation), but not at 1.4 V (clinically ineffective stimulation).
In monkey R370 bipolar stimulation (contact 2 cathode, contact 0 anode) at 136 Hz and 90 µs
pulse width was clinically effective at 3 V amplitude and clinically ineffective at 2 V. Thus,
these various stimulation parameters were applied to our model system. The fundamental
differences between the model simulations for the two monkeys were the electrode position and
stimulation parameters. In addition, the two electrode positions resulted in somewhat different
neural populations because different neurons were ‘destroyed’ by the electrodes. The model
neurons were stimulated with a train of 25 pulses. Longer train durations (1 second; 136 pulses)
did not impact neural response to stimulation (Fig. 5B).
We did not observe any model neurons that exhibited a blocking of axonal firing from the
direct application of the DBS electric field; therefore, we concentrated our analysis on the
Page 12 of 42
excitatory response of the stimulation (Fig. 5A). Those that produced orthodromically
propagating action potentials in response to more than 80% of the stimulus pulses were
considered to be activated. This percentage was chosen because most neurons responded either
to none or to 20 or more of the 25 stimulus pulses. Some activated neurons did not respond to all
25 pulses because in certain cases the stimulus pulse was delivered immediately following a
spontaneous action potential (originating in the soma), while the axon was still in the refractory
state. Percentages of activated neurons were averaged over three randomized populations.
Student’s t-test (one-tailed; p<0.05) were performed to compare the averages for clinically
ineffective and effective stimulation conditions.
Experimental Recordings of GPi Activity During STN DBS
The experimental recording procedure and data collection from the two monkeys used in
this study has been described in detail elsewhere (Hashimoto et al., 2002; 2003; Elder et al.,
2005). Briefly, single unit neural activity was recorded extracellularly from GPi identified
neurons using glass-coated platinum–iridium microelectrodes (impedances of 0.4–0.8 MΩ at 2
kHz). Recording penetrations were made in parasagittal planes moving rostral to caudal at an
angle of 70° to the orbitomeatal line. Neurons (n = 12 for monkey R7160 and n=27 for monkey
R370) were recoded for 25-35 sec during clinically effective STN DBS at 136 Hz. Stimulus
artifact template subtraction methods (Hashimoto et al., 2002) and in-house software developed
in MATLAB v7.0 (Mathworks Inc., Natick, MA) were used for the neural signal analysis.
Peristimulus time histograms were constructed with a 0.2 ms bin size. The histograms used
336,373 spikes resulting from 644,929 stimuli in 39 GPi cells. Stimuli for which no spike was
recorded in the interstimulus interval were not considered. Probability distributions for R7160
Page 13 of 42
and R370 were compared using a χ2 goodness of fit test (p<0.05). Cells were further classified
as having an early response if more than 15% of the stimulus responses occurred at a latency of
less than 1.5 ms. Significance of differences in the proportion of early response cells was found
by χ2 test (p<0.05).
We calculated levels of axonal activation for populations of STN projection neuron, GPi
fibers of passage (lenticular fasciculus) and CST fibers of passage during clinically effective and
ineffective STN DBS in two parkinsonian macaques. We evaluated four general aspects of our
model system. First, activation of CST fibers at muscle contraction thresholds were analyzed to
provide a gross degree of model validation. Second, we evaluated the neural response to
therapeutic stimulation using three separate randomized populations of STN neurons and GPi
fibers, and we correlated the model predictions to single-unit microelectrode recordings from the
two monkeys during STN DBS. Third, the sensitivity of neural activation to electrode position
was assessed by moving the electrode by 0.25 mm in four directions in the horizontal plane. And
finally, the effects of DBS on STN somatic firing were evaluated in a model with stimulation-
induced inhibitory trans-synaptic conductances.
Activation of the Internal Capsule with DBS
To address the experimental predictability of our model system we evaluated the
activation of CST fibers. Visually determined muscle contraction thresholds in the two monkeys,
were 3V for R7160 and 3.5V for R370. These stimulation parameters resulted in activation of
11±1% and 9±3% of CST fibers in our R7160 and R370 models, respectively, averaged over
Page 14 of 42
three random populations (mean±SD). In addition, clinically effective stimulation parameters
resulted in minimal activation of CST fibers, 0% and 5±2% in the R7160 and R370 models,
respectively. These results indicate that the model exhibited appreciable increases in CST fiber
activation when comparing stimuli that experimentally resulted in no visible muscle contraction
(clinically effective) and stimuli where muscle contractions were observed (CST threshold).
This provides some evidence that the voltage spread in the tissue around the electrode was
accurately predicted by the finite element model, and that the neuron models fire at realistic
Activation of the STN and LF with STN DBS
In the context of STN DBS, both STN projection neurons and GPi fibers of passage in
the LF represent viable candidates as the therapeutic target of the stimulation. The GPi is an
output nucleus of the basal ganglia and the STN modulates basal ganglia output through
excitatory projections into the GPi. We defined DBS induced activation of STN projection
neurons as the generation of an action potential anywhere in the neuron, resulting from the
applied electric field, which propagated to the axonal terminal. When stimulating STN projection
neurons with DBS, action potential initiation always took place in the myelinated axon, and if an
action potential was induced by the stimulation it always propagated to the axon terminal (Fig.
Our simulations predict activation of both GPi fibers and STN neurons during STN DBS
(Fig. 6). These results were consistent over three randomized populations of neurons.
Stimulation parameters that failed to improve parkinsonian symptoms in monkey R7160 (1.4 V,
210 µs, 136 Hz), activated 29±2% of STN projection neurons, 9±4% of GPi fibers of passage
Page 15 of 42