disease, we acutely recorded the single-unit activity of 274 ventral intermediate/ventral oralis posterior motor thalamus (Vim/Vop)
were recorded while the patients performed a target-tracking motor task using a cursor controlled by a haptic glove. We observed that
modulations in firing rate of a substantial number of neurons in both Vim/Vop and STN represented target onset, movement onset/
direction, and hand tremor. Neurons in both areas exhibited rhythmic oscillations and pairwise synchrony. Notably, all tremor-
schemes. Even in the presence of this pathological activity, linear models were able to extract motor parameters from ensemble dis-
charges. Based on these findings, we propose that chronic multielectrode recordings from Vim/Vop and STN could prove useful for
Neurosurgical implantation of deep brain stimulation (DBS)
(PD) and essential tremor (ET) (Ondo et al., 1998; Koller et al.,
(STN; for PD patients) and the ventral intermediate nucleus of
thalamus (Vim; for ET patients). Both structures are involved in
motor control (Parent and Hazrati, 1995; Guillery and Sherman,
2002). The dorsolateral STN receives afferents from motor cor-
tex, premotor cortex, and supplementary motor areas (Parent
and Hazrati, 1995; Hamani et al., 2004). Vim projects to these
et al., 1990, 1994, 2002; Raeva et al., 1999; Magarin ˜os-Ascone et
al., 2000; Magnin et al., 2000; Rodriguez-Oroz et al., 2001;
2003; Williams et al., 2005). Many reports have examined single-
unit activity in these regions with respect to tremor (Lenz et al.,
Magnin et al., 2000; Rodriguez-Oroz et al., 2001; Brodkey et al.,
2004; Hua and Lenz, 2005; Amtage et al., 2008) and pathological
synchronous oscillations (Levy et al., 2000, 2002; Amirnovin et
in these studies was low (?2), providing limited information
about the pathological activity of larger neuronal ensembles.
Ensembles of simultaneously recorded neurons have been
used to enable brain-machine interfaces (BMIs) for neuropros-
al., 2000; Nicolelis, 2001; Serruya et al., 2002; Taylor et al., 2002;
lis and Lebedev, 2009; Lebedev et al., 2011; O’Doherty et al.,
2011). However, recordings in humans have rarely been ex-
Author contributions: T.L.H., A.M.F., M.A.L., D.A.T., and M.A.L.N. designed research; T.L.H., A.M.F., and D.A.T.
DGE-1106401-004 to A.M.F.; the VA Merit Review Award, and NIH Grants R21NS066115 and RO1AG037599 to
Correspondence should be addressed to Miguel A. L. Nicolelis, 311 Research Drive, Bryan Research Building,
8620 • TheJournalofNeuroscience,June20,2012 • 32(25):8620–8632
tracted from neuronal ensembles (Kennedy and Bakay, 1998;
(Patil et al., 2004), in which our laboratory demonstrated the
from neuronal ensembles recorded in motor thalamus [Vim/
ventral oralis posterior motor thalamus (Vop)] or STN during
DBS surgery to decode modulations of hand force during a one-
dimensional target-tracking task.
We now examine neuronal population firing patterns during a
voluntary motor task in a new sample of 25 human patients. While
targets, up to 23 subcortical neurons were recorded simultaneously
and acutely in Vim/Vop and STN to elucidate the relationship be-
tween neuronal modulations, rhythmic oscillations, and neuronal
synchrony. A linear decoder model was applied to reconstruct cur-
sor position from spiking activity. Neurons were classified by oscil-
latory firing patterns, tremor association, synchrony, and tuning to
Intraoperative recordings were conducted in 25 human patients under-
going placement of therapeutic DBS implants in either Vim or STN. All
studies were approved by the Duke University Institutional Review
Board and human ethics committees, and all participating patients un-
derstood and signed all required consent forms.
Patient characteristics and operative plan. All patients selected for this
10 male, 1 female) DBS electrode implantation surgery. Patients whose
akinetic/rigid variant with on/off fluctuations and dyskinesia) were can-
didates for implantation in STN. Both groups of patients were off their
medications before and during surgery.
Patients first underwent Leksell frame placement, followed by a MRI
scan to localize the implantation target. For Vim patients, the target
was typically estimated according to anterior–posterior commissure
(AC-PC) criteria, located ?5–6 mm in front of the PC, on the AC-PC
line with a lateral measure depending on the width of the third ventricle
first penetration for approaching Vim was the localization pass from a
frontal burr hole, with the upper 5 mm of the recording track near the
caudalis at the most posterior extent. Typically, the upper 5 mm was the
best for multineuron recordings, reflecting more of the anterior motor
thalamus (Vop rather than Vim), as shown in Figure 1A. For STN pa-
state imaging. The initial target was located at 11–12 mm from the mid-
line, 2–3 mm posterior to the midpoint of the AC-PC line, and 4 mm
first performed to define the borders of the STN, according to standard
electrophysiological criteria, with the goal of attaining at least 5.5–6 mm
of STN. Typically, 2–3 passes were performed. Localization was per-
formed using single-channel tungsten microelectrodes.
For both targets, once single-unit recordings had been performed for
localization, a 32-channel Pt/Ir microwire (35 ?m diameter) array (Ad-
Tech Medical Instrument) was passed to the appropriate depth via an
outer cannula (Patil et al., 2004), where significant activity was noted
with the single-unit electrode. After allowing a few minutes for initial
the cannula. Once the number of clearly distinguishable single units was
maximized, the microwire array was left in place. At each electrode
depth, the patient was instructed to proceed with the voluntary motor
crowire array was removed and the DBS treatment electrodes were im-
planted. As a clinical routine, a brain computed tomography scan was
performed within 12 h of the surgery procedure, and in no instance was
a hemorrhage or other complication noted. Hence, the clinical risks of
temporary placement of the 32 channel microwire array were demon-
more, in our research, we have used this electrode in many (N ? 72)
patients over several years with no post-op evidence of hemorrhage.
Electrophysiological recording. The 32-channel microwire recordings
were performed with a Plexon MAP system. Since this study was per-
formed intraoperatively during electrophysiological mapping of the im-
(Vc, the sensory nucleus). The exact angle of the trajectory varied slightly from patient to
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles J.Neurosci.,June20,2012 • 32(25):8620–8632 • 8621
epochs (up to 8, mean ? 3.8), between which the electrode depths were
altered. For each recording epoch, single units were sorted offline using
custom software developed in-house. Extracted spikes consisted of 32
samples each, sampled at 40 kHz, aligned on the crossing of a linear
as the microdrive corrupted the recording traces) or threshold was oth-
erwise changed during a recording epoch, the record was broken into
separate epochs. Recorded neuronal discharges on a given channel that
could not be sorted and isolated as single units were classified as belong-
ing to a multiunit, indicating a collection of individual discharges whose
identity could not be firmly established. Figure 1B shows the sorted unit
waveforms from a single recording session. We estimated the signal-to-
by an estimate of the noise variance (Bankman et al., 1993), yielding a
mean SNR of 4.69 for all sorted units.
Voluntary motor task. Patients were placed in a supine, semisitting
position in front of a computer monitor. A 5DT Systems Data Glove 5
Ultra haptic glove was placed over the hand contralateral to the micro-
electrode array. This glove was used to measure flexion/extension of the
fingers, sampled at 1 kHz. The average flexion/extension signal, which
in front of the patient for high visibility. Patients were trained to modu-
late the opening and closing of their hand to acquire targets by moving the
cursor into a box placed randomly along a horizontal line (Fig. 2A).
The required targetholdtimewas200ms.Oncethetargetwasacquired,
the box disappeared for 300 ms before reappearing in a new random
extent of the screen. Therefore, movements did not strictly alternate
between left and right; successive jumps would frequently occur in the
During the preliminary training/calibration phase, the cursor gain,
offset, and target box size were systematically calibrated by the experi-
menters to compensate for variations in physical ability. Specifically, in
to cursor motion was increased. In patients with pronounced hand
tremor, the size of the target was increased. Finally, the offset was set so
of the screen. During the recordings that followed, the length of individual
motor task sessions varied depending on electrophysiological recording
quality and the level of patient fatigue. Figure 2B shows a representative
Neuronal tuning to target and movement. All subsequent data analyses
were performed using Matlab (MathWorks). We use the term “tuning”
to refer to modulation of neuronal firing rates that is correlated to an
external parameter and “tuning strength” to refer to the extent of those
For all sorted single units and multiunits, perievent time histograms
(PETHs) of neuronal activity (Awiszus, 1997) were generated using one
time. PETHs triggered on target appearance were constructed using a
window beginning 0.5 s before each event trigger and ending 1.5 s after,
ment time. Movement time was defined as the moment at which the
cursor crossed the midpoint between initial cursor position (at target
appearance) and the endpoint target position. We chose this standard as
occasional incorrect movements. Regardless of reaction time, this event
trigger was locked to movement, being in close proximity to the point of
movement times ?200 ms (premature movement) or ?1000 ms (inat-
sition trials were chosen for further analysis.
Neuronal tuning to either target or movement was determined by
PETHs generated by uncorrelated triggers. We calculated significance
using the one-sample Kuiper’s test (Kuiper, 1962; Batschelet, 1981; Zar,
1999), a nonparametric test related to the Kolmogorov–Smirnov (K-S)
K-S test, Kuiper’s test is equally sensitive throughout the distribution, a
useful property in scenarios in which the locations of the peak modula-
tions are not known a priori. Variations of the K-S test have been used
previously in the significance evaluation of neuronal PETHs (Ghazanfar
et al., 2001; Wiest et al., 2005; Gutierrez et al., 2006). In this study, we
used Kuiper’s test to distinguish an observed distribution of event-
triggered spike times from the null hypothesis (uniform probability dis-
tribution). Kuiper’s test requires the calculation of the maximum
positive and negative deviations of the observed PETH cumulative
distribution function (CDF) from a uniform distribution CDF (ramp
function); the sum of these two deviations is the statistic V: V ?
The Kuiper statistic, K, is a normalized version of V, taking into ac-
count the size of the observed sample size N, in this case, the number of
binnedspikes:(K ? VN1/ 2? 0.155 ? 0.24N?1/ 2).Todistinguish
the test statistic Kobsfrom the null hypothesis, we generated a boot-
strapped distribution of 1000 simulated Kuiper statistics (Ksim). Prelim-
inary analysis determined shuffling of spike timestamps to be a
suboptimal control; the process eliminates spike autocorrelations from
the bootstrap distribution, thereby potentially biasing the evaluation of
the observed distribution in favor of significance. Instead, each value of
Ksimwas calculated using a PETH constructed from the original spike
the events (target appearance, movement) are randomly and indepen-
dently assigned to decorrelate them from the spiking data. For each
sorted unit, the resulting bootstrapped distribution of Ksimwas used to
Units were deemed to be tuned to task events (target or movement
onset) using the threshold p ? 0.05. These units exhibited temporal
modulations in firing rate relative to newly appearing targets and/or
target-directed movements. Tuning strength was defined as the z-score
of the observed PETH relative to the bootstrap distribution.
ing of the hand to actuate a one-dimensional cursor toward randomly appearing targets. B,
Example off-line prediction of hand/cursor position. Prediction was obtained using a linear
Kalman filter with a 500 s training period, from Patient M (Vim/Vop). Forty-eight units (16
8622 • J.Neurosci.,June20,2012 • 32(25):8620–8632 Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles
Directional tuning. The directional tuning of each sorted unit or mul-
tiunit was defined as the difference in neuronal response for leftward
versus rightward movements. Significance of directional tuning was deter-
the maximum positive and negative deviations between the spike CDFs for
max[CDFsample 2?CDFsample 1].
The Kuiper statistic K was calculated from V using the same equation
as the one-sample Kuiper’s test, but in the case of the two-sample
Two PETHs triggered on movement time were generated, one for
leftward movements and one for rightward movements. As with the
one-sample Kuiper’s test, each unit’s Kobsstatistic was compared with a
bootstrapped distribution of Ksimgenerated from randomized trigger
times. Units were deemed to be directionally tuned using the threshold
p ? 0.05. Tuning strength was defined as the z-score of the observed
PETHs relative to the bootstrap distribution.
not permitted under our approved experimental protocol, haptic/posi-
tion tracking have been used repeatedly to analyze tremor (Bardorfer et
al., 2001; Su et al., 2003; Vinjamuri et al., 2009), as has accelerometry
(Ghika et al., 1993; Grimaldi et al., 2007; Birdno et al., 2008).
For all sorted single units, perievent phase histograms (PEPHs) of
event trigger, similar to the approach of Lebedev et al. (1994). Tremor
of 100–2000 ms (0.5–10 Hz) were analyzed. To exclude the impact of
voluntary movements, hand velocity peaks occurring within 250 ms of a
movement trigger were excluded. Each tremor period was defined in
units of phase, with neuronal spike activity captured into 100 bins of
equivalent phase aperture (3.6° each). The zero phase for each cycle was
defined by a local maximum in hand velocity. Only sorted units with at
least 500 valid tremor periods were chosen for further analysis.
Each resulting PEPH was a measurement of neuronal firing rate with
respect to tremor phase. The one-sample Kuiper’s test, in addition to
possessing uniform sensitivity, is also rotationally invariant, meaning
that the arbitrary choice of zero phase has no effect on the assessment of
statistical significance. For analysis of tremor tuning, we generated a
bootstrapped distribution of 1000 simulated Kuiper statistics (Ksim);
each was calculated using a PEPH constructed from trials whose binned
spike counts were circularly rotated by uniformly random phase offsets.
For each analyzed unit, the bootstrapped distribution was used to pro-
observed PEPH relative to the bootstrap distribution.
Oscillatory neurons. For all sorted units with at least 1000 extracted
spikes, we used Welch’s method (Oppenheim and Schafer, 1975) with
eight nonoverlapping segments to determine the spike train autopower
spectral density. The power spectra were smoothed using a 0.5 Hz rect-
angular sliding window. For each unit, the peak autopower frequency
was determined in the 1–25 Hz range, with frequency content ?1 Hz
discarded for the remainder of the analysis. For the peak frequency, we
determined the SNR by dividing peak power by the mean power (as-
of comparison, the same spectral analysis was performed on hand accel-
eration traces for all recorded sessions.
Preliminary analysis indicated a functional separation of peak fre-
to be dominated by low-frequency power and were therefore judged not
As in previous studies (Lenz et al., 1988; Amtage et al., 2008), only units
in either the spike train or hand acceleration autopower spectra was
determined by calculating the maximum power concentrated in a 1 Hz
band within the physiologically relevant 2.5–7.5 Hz window. “Peaked-
power in the 1–25 Hz band.
In addition to established linear methods, we developed a heterodyne
method for detecting nonstationary or wide-bandwidth synchronized
activity between each neuron’s firing rate and associated hand velocity.
This method, inspired by the frequency shifting scheme used in radio
frequency transceivers, was applied to all sorted units that fulfilled the
selection criteria for both target tuning analysis and oscillatory analysis.
The spike train and hand velocity recordings were first bandpass filtered
(fourth-order Butterworth, zero phase method) between 2 and 12 Hz,
then multiplied, yielding a third time series from which to extract spectral
modulated components in both neuronal firing rate and hand velocity are
transferred to DC. The ratio of spectral energy (Eobs) from 0 to 0.125 Hz
(signal) over that from 0.25–2 Hz (baseline) was considered as a metric of
heterodyne correlation between neuronal activity and hand movement.
generated by shuffling spike timestamps 1000 times and repeating the
above analysis. Since all low-frequency information is filtered out before
multiplication, shuffling timestamps was determined to be a suitable
control. For all single units, the resulting bootstrapped distribution was
tremor if the heterodyne correlation between neuronal activity and
movement was significant using the threshold p ? 0.05. Heterodyne
relative to the bootstrap distribution.
Efficacy of neuronal recordings for kinematic predictions. Prediction al-
gorithms from the BMI literature were applied to subcortical neuronal
populations to extract behavioral parameters. Several algorithms were
tested, including the linear Kalman filter, unscented Kalman filter, and
the Wiener filter. Figure 2B shows an example off-line prediction for a
30 s window of task performance.
Since all three algorithms achieved the same approximate fidelity in
Wiener filters were fit by binning neuronal data into 100 ms time slices
training was performed by the random selection of 50% of these time
slices; predictions were then made on a distinct random 25%. This pro-
cess was repeated with 100 draws of fit and predict time slices. Correla-
tion coefficient (R) between predicted hand position and actual hand
position was measured for each of the draws; the mean correlation coef-
are reported for all sessions with at least 50 presented targets.
each subset ensemble size N, we performed 1000 draws of random en-
semble subset and Wiener filter fit and prediction; the R values for these
draws were averaged to form a smooth neuron dropping curve. Follow-
ing Wessberg et al. (2000), the resulting curve was then fit to the follow-
ing hyperbolic function to extrapolate the performance results to larger
ensemble sizes: R2? cN/(1 ? cN).
Neuronal synchrony. The use of simultaneous ensemble recordings
mine the statistical significance of the synchrony between two neurons,
we analyzed the cross-correlation peak between pairs of spike trains.
Pairs were analyzed if they each contained at least 100 spikes and corre-
sponded to a session with at least 50 targets. The cross-correlation coef-
ficient was first calculated for the observed spike trains of the two
neurons, then smoothed using a 5 ms rectangular sliding window. The
observed test statistic Cobswas defined as the peak coefficient in the ?10
ms time lag range. Bootstrap simulations (n ? 1000) of the two spike
trains were generated by convolving the spike trains with a Gaussian
kernel (? ? 250 ms) and then generating new spike trains via an inho-
correlated high-frequency content in the bootstrap distribution while
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsemblesJ.Neurosci.,June20,2012 • 32(25):8620–8632 • 8623
Csim. The bootstrapped distribution was used to produce a p value, and
neuron pairs were deemed to be significantly synchronous using the
threshold p ? 0.05.
joint peristimulus time histograms (JPSTHs) for neuron pairs, as origi-
nally proposed by Aertsen et al. (1989). Our JPSTHs were adjusted by
subtracting the shift predictor histogram and normalizing (bin-by-bin)
as the “true normalization” of the JPSTH.
A total of 25 DBS implantation patients were examined. In these
patients, we simultaneously recorded from ensembles of up to 23
well isolated neurons from either Vim/Vop or STN, depending on
in terms of duration and target acquisition rate, as limited by indi-
vidual patient pathology and motivation. Neurons from these sub-
to target, movement, direction, and tremor. Moreover, neuronal
STN cells (N ? 168) exhibited a higher (p ? 0.01, Mann–
Whitney U test) mean firing rate than Vim/Vop cells (N ? 83):
both cases). In both subcortical areas, we found substantial pop-
ulations of oscillatory neurons, as well as neurons strongly tuned
to target, movement, direction, and tremor. Furthermore, neu-
rons in both subcortical areas tended to show tuning to multiple
parameters (Table 1) rather than belonging to disjoint sets. For
example, the number of Vim/Vop cells tuned to both target and
exhibit synchrony. Curiously, all tremor-associated neurons ex-
hibited synchrony within the recorded neuronal ensemble.
movement onset (Table 2). Of all single units tested, 29.2% of 168
target appearance. Both of these percentages represent signif-
icant populations (binomial test, p ? ? 0.001 in both cases).
Figure 3A shows example PETHs for three highly responsive
1000 ms range were discarded for movement tuning, and only
sessions with at least 50 valid trials were subjected to further
statistical analysis. Because of the additional reaction-time crite-
rion, fewer single units were analyzed for movement tuning than
of 26 STN cells were found to be tuned to movement. Both of
these percentages represent statistically significant populations
ple PETHs for three highly responsive neurons.
trials, target tuning strength significantly predicted movement tun-
ing strength (? ? 0.70, p ? ? 0.001 for Vim/Vop; ? ? 0.71, p ? ?
that a larger-than-expected number of neurons in both subcortical
target appearance and movement time. However, visual inspec-
tion of some tuned units indicated a clear decoupling of the neu-
ral encoding of target appearance and movement. Sorted raster
plots from example neurons are shown in Figure 4, A and C.
showing two clear bands of increased spike density. In both pan-
clearly related to target appearance (?450 ms postappearance).
The diagonal bands have a near-unity slope, indicating a clear
time-locked relationship between neuronal activity and move-
TargetMovement DirectionTremor OscillatoryHeterodyne
ParameterUnittype Area No.ofunits #Tuned
8624 • J.Neurosci.,June20,2012 • 32(25):8620–8632Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles
?300 ms after the defined movement time. From these data, it
can be concluded that these neurons were tuned to both target
appearance and movement; they modulated their firing rates in
relation to both events.
Modest differences were seen in the aggregate response patterns
movement (Fig. 5). Both cell types exhibited a mean response that
rently with movement (Fig. 5B). Differences in the relative lags for
Vim/Vop and STN neuronal activation likely reflect the position of
intention, with signals arriving before motor cortex activation,
whereas the collaterals from motor cortex to STN deliver signals at
to target appearance; the same was also true for movement tuning
Example PETHs. A, Strongly tuned units to target appearance. i, Vim/Vop cell,
time. Time along the x-axis is relative to target appearance. A and C show spike raster plots
depicts movement time. For generation of the color plots, the data were smoothed using a
Separation between single-unit response to target appearance and movement
cortical areas using two event triggers: target appearance (A), and movement time (B). Re-
ported time is relative to the event trigger. Before aggregation, individual PETHs were
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsemblesJ.Neurosci.,June20,2012 • 32(25):8620–8632 • 8625
Another metric of interest for the behavioral responsiveness of
subcortical neurons was directional tuning; Table 2 gives the di-
rectional tuning results for both single units and multiunits. Of
STN cells were found to exhibit directional tuning. Both of these
percentages represent statistically significant populations (bino-
mial test, p ? 0.001 for Vim/Vop, p ? 0.01 for STN). Figure 3C
shows example PETHs for three strongly tuned neurons. Similar
proportions of Vim/Vop and STN cells were tuned to direction
(two-tailed Fisher’s exact test, p ? 0.05).
Despite the clear separation in the neuronal response to left-
ward and rightward movements, note the transient regions of
convergence that occurred in Figure 3C. In Figure 3Ciii, for ex-
ample, the neuronal responses to each direction converged just
before movement. For many tuned neurons in both Vim/Vop
and STN, the degree of directional modulation varied through-
out the temporal window.
For both Vim/Vop and STN cells, we found strong positive
of both target tuning and movement tuning. When controlling
for the number of session trials, target tuning strength signifi-
cantly predicted directional tuning strength (? ? 0.16, p ? 0.05
for Vim/Vop; ? ? 0.38, p ? 0.01 for STN). Similarly, movement
tuning strength significantly predicted directional tuning
classification result in Table 1.
Whereas single units are identifiable as distinct neurons, a mul-
of analyzed multiunits were tuned to target, movement, and di-
rection (binomial test, p ? 0.01 in all cases). A substantial num-
ber of these tuned multiunits were found on the same recorded
the number of session trials, the target tuning strength of single
units significantly predicted the target tuning strength of same-
channel multiunits (? ? 0.18, p ? 0.01). This confirms the pres-
ence of correlated tuning in nearby neurons. Thus, a substantial
amount of encoded information was present in subcortical mul-
tiunits, arguing for the potential inclusion of these signals in fu-
ture analyses of ensemble activity.
To identify potentially pathological neurons within the recorded
subcortical populations, we analyzed the tremor sensitivity of
single units using the discussed PEPH approach; the results are
cells and 15.9% of 82 STN cells were found to be correlated to
sensitive neurons. These results demonstrate that for highly
tuned units, the dependence of spike rate on tremor phase re-
mained stable throughout the recording session (Fig. 6), even if
proportions of Vim/Vop and STN cells were tuned to tremor
(two-tailed Fisher’s exact test, p ? 0.05).
For Vim/Vop cells (but not STN cells), we found a positive
correlation between the strength of tremor tuning and that of
directional tuning. When controlling for the number of session
trials, directional tuning strength significantly predicted tremor
tuning (? ? 0.34, p ? 0.05). However, we found no relationship
between tremor tuning and either undirected target or move-
ment tuning (p ? 0.05 for all cases).
However, these results do not distinguish whether these
ative of tremor.
To explore neuronal oscillations in Vim/Vop and STN and their
relationship to patient pathology, we inspected the autopower
spectra of single-unit spike trains for strong frequency peaks,
yielding peak frequency and SNR (Fig. 7A). Of all tested single
units with SNR ? 2, the distribution of peak frequencies showed
a clear bimodal distribution with a border between low- and
high-frequency oscillations at ?2.5 Hz (Fig. 7B).
Spike train spectra from Vim/Vop and STN neurons also
tended to possess large amounts of energy at low frequencies,
suggestive of 1/f (pink) noise. This power-law distribution has
patient attention and arousal. The observed distribution may ex-
1/f noise from those exhibiting strong oscillations in the tremor-
power within this frequency range were eligible to be classified as
of 274 Vim/Vop cells and 17.9% of 123 STN cells were classified as
oscillatory. No difference was seen in the proportions of oscillatory
Vim/Vop and STN cells (two-tailed Fisher’s exact test, p ? 0.05).
Figure 8 shows example interspike interval (ISI) plots for three
highly oscillatory cells. Note that all three ISI histograms exhibit
Figure 9A shows the smoothed autopower spectra of spike
8626 • J.Neurosci.,June20,2012 • 32(25):8620–8632 Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles
ized horizontal trace corresponding to a distinct unit. From this
figure, one can visually identify some of the highly oscillatory
units as well as observe the congruity between multiple units
ized spectra for Vim/Vop and STN cells (Fig. 9B) is statistically
significant (?2test, p ? ? 0.001). From Figure 9B, it is clear that
STN cells tended to concentrate power at a lower frequency (3
rather than 4 Hz).
The pairwise classification results in Table 1 reject the notion
that oscillatory neurons and behaviorally tuned neurons form
disjoint sets. Furthermore, we found no relationship between
spike autopower peakedness and the strength of any of the three
(target, movement, direction) behavioral tuning metrics (p ?
anticorrelation suggests that the sets of behavioral neurons and
oscillatory neurons are far from disjointed. Instead, they appear
to exist as overlapping populations.
Our next analysis intended to uncover a relationship between
However, we did not find any clear relationship. Linear regres-
sion analysis revealed no relationship between peak frequency
(2.5–7.5 Hz range) of spike train autopower spectra and corre-
sponding hand acceleration autopower spectra (p ? 0.05 for
sharpness of the two spectra for Vim/Vop neurons (p ? 0.05),
but we did observe a marginally significant positive relationship
spectra for the spike train autopower and hand acceleration au-
topower of three highly oscillatory units. For all three cells (rep-
and tremor tuning strength for Vim/Vop cells (p ? 0.1 for STN
cells). These findings call into question the presumed causal lin-
ear relationship between the two, suggesting the possibility of an
elusive nonlinear relationship.
analyzed single units. Units with peak frequency ?10 Hz (17.4% of all units) are not
SNR mostly below the classification threshold of 2. B, Histogram showing the number of
oscillatory (SNR ? 2) units at each peak frequency. A clear separation between two
normalized. The horizontal traces are grouped by patient and recording area. The color bar
represents arbitrary units of normalized energy density. B, Mean autopower spectra from all
dow and normalized before aggregation. Each of the dashed traces represents a bootstrap
A, Smoothed autopower of spike trains for all sorted units with sufficient spike
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsemblesJ.Neurosci.,June20,2012 • 32(25):8620–8632 • 8627
Our heterodyne decoding analysis further explored this rela-
tionship by applying a nonlinear frequency shifting approach to
the autopower spectra. Of 239 analyzed single units, 13.3% of
Vim/Vop cells and 17.3% of STN cells were found to be tremor
associated via heterodyne decoding (Table 5). Both of these per-
centages represent statistically significant populations (binomial
test, p ? ? 0.001), but the difference between them is not signifi-
cant (two-tailed Fisher’s exact test, p ? 0.05).
tremor correlations than spectral peak analysis if the tremor sig-
nal is wide-bandwidth or prone to phase changes. Indeed, this
schema may better serve to explain the relationship between the
oscillatory activity of neurons and observed tremor. In fact, this
can explain the similarity in the proportions of tuned neurons in
Tables 3 and 5. Furthermore, Vim/Vop firing indicated a strong
positive correlation between tremor tuning strength, identified
using PEPHs, and heterodyne tremor tuning strength (? ? 0.31,
p ? 0.001). This relationship was marginally significant in STN
yses for Vim/Vop cells.
We also performed off-line predictions of cursor motion using
the recorded ensembles. The correlation coefficient (mean ?
1.98 SE) for each of the sessions is shown in Figure 11. Although
the predictions varied greatly across sessions and patients, the
results compared favorably with our previous study (Patil et al.,
was chosen for further analysis, and neuron dropping curves
were generated for these two sessions and fitted to a hyperbolic
function (Wessberg et al., 2000). Extrapolation of the hyperbolic
fit produced estimates of the approximate ensemble sizes re-
quired to achieve R2? 0.9: 106 Vim/Vop neurons or 397 STN
We analyzed neuronal synchrony in pairs of sorted units and
investigated how its prevalence varied across subcortical areas;
the results are given in Table 6. Using the cross-correlation ap-
proach, 43.0% of Vim/Vop pairs and 25.8% of STN pairs were
found to be significantly synchronous. Both of these percentages
represent significant populations (binomial test, p ? ? 0.001 for
both cases). Figure 12A shows example cross-correlation plots
for three highly synchronous pairs, while Figure 12B shows the
i and ii, clearly shows temporal synchronization along the diago-
in Figure 12Biii.
We found highly significant differences between the Vim/
Vop and STN in terms of the proportions of synchronous
synchronous than STN pairs (two-tailed Fisher’s exact test,
p ? ? 0.001).
It has been reported that the level of tremor in parkinsonian
patients is positively correlated to the degree of pairwise syn-
chrony among STN cells (Levy et al., 2000). To test the relation-
ship between tremor tuning and local synchrony, we compared
the subpopulation of both Vim/Vop and STN neurons that were
synchronous with at least one other neuron in their respective
(PEPH method). Only neurons fulfilling the criteria of both in-
7. The observed proportions are significantly different (two-
tailed Fisher’s exact test, p ? 0.01), indicating a clear interaction
nized to at least one other unit were tuned to tremor, whereas no
unsynchronized units were tuned to tremor.
highly oscillatory units. A, Vim/Vop cell, Patient W. B, STN cell, Patient V. C, Vim/Vop cell,
The best session for each subcortical area (Vim/Vop, STN) was chosen for further analysis.
Area No.ofpairs No.ofsynchronous
8628 • J.Neurosci.,June20,2012 • 32(25):8620–8632 Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles
neurons (either Vim/Vop or STN) in 25 patients, recorded in
patients who performed visually guided hand movements. To
the representation of motor parameters in these neurons, as well
as activity thought to be related to pathological states—tremor
sensitivity, oscillations, and pairwise synchrony. The present
al., 2004). We further propose that chronic subcortical microelec-
trode technology could serve as the basis of a new generation of
neuroprosthetic devices aimed at both monitoring and actively re-
Neurons in both subcortical areas (Vim/
Vop and STN) were found to encode fea-
tures of patient motor behavior, both
voluntary (target tracking) and involun-
a substantial number of neurons were
found to be tuned to target appearance,
movement onset, and movement direc-
tion (Table 2). This finding likely indi-
cates that neurons in both structures are
broadly tuned across multiple modalities
and muscle groups. This observation is
consistent with previous studies in which
STN neurons have been reported to have
large receptive fields that respond to mul-
of STN cells tuned to movement in a sim-
ple two-dimensional joystick task (Wil-
liams et al., 2005). It has been reported
that 42% of STN neurons respond to pas-
these, 25% responded to multiple joint
2003), whereas 51% of Vim/Vop neurons
lapping 10% of these cells tuned to voli-
tional movements (Lenz et al., 1990).
ited bimodal encoding of target and move-
ment (Fig. 4), demonstrating what appears
to be superposition of two independent
neural representations. Furthermore, we
ing to target appearance and to movement
onset in both structures. Additionally, di-
related to movement tuning strength,
indicating that individual neurons encoded
rons from both subcortical areas exhibited rate modulations based
Our analysis of involuntary motor activity (tremor) yielded
phase histograms demonstrating clear phase-locking between neu-
for STN). The literature reports a large range in the prevalence of
tremor-related cells. For STN, researchers have reported 11%
and 52% (Amtage et al., 2008). For Vim/Vop, researchers have re-
ported 34% (Lenz et al., 1988), 35.6% (Zirh et al., 1998), and 51%
(Hua and Lenz, 2005). These disparities are probably due to differ-
ences in recording parameters and classification methodology. For
example, whereas many researchers have classified tremor tuning
and those tuned to parameters of voluntary motor behavior (Ta-
tremor tuning strength and directional tuning strength (Vim/
Vop cells only). These results are consistent with earlier studies
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsemblesJ.Neurosci.,June20,2012 • 32(25):8620–8632 • 8629
reported a large proportion of tremor-related STN cells to be
simultaneously related to voluntary movements either through
motor or sensory loops.
Within a tremor-relevant frequency range (2.5–7.5 Hz), we ob-
iting strong oscillations (Table 4). Furthermore, we found that
the mean autopower spectra for recorded Vim/Vop cells had a
higher peak frequency than that of recorded STN cells (Fig. 9B).
This finding is consistent with reports of higher mean tremor
(Deuschl et al., 1998).
Whereas Rodriguez-Oroz et al. (2001) reported that oscilla-
tory cells in STN did not represent movements, we found mod-
erate overlap (Table 1) between Vim/Vop and STN neurons
exhibiting oscillations and those tuned to voluntary behavioral
parameters (target, movement, direction). This important find-
ing indicates that there is not a strict dichotomy between patho-
logical neurons and those encoding motor signals.
Similarly, we found overlap but no significant correlation be-
tween the presence of oscillatory patterns and tremor tuning in
both subcortical areas. This finding corroborates the claims of
earlier studies of ventral thalamus and STN (Magnin et al., 2000;
Rodriguez-Oroz et al., 2001), in which the sets of tremor-related
neurons and oscillatory neurons show modest intersection. In
other words, not all tremor-related neurons exhibited oscilla-
tions, and some oscillatory neurons exhibited no clear associa-
tion with tremor.
Furthermore, spectral peak detection methods revealed no
relationship between spike train and hand acceleration (Fig. 10),
ship between neuronal oscillations and hand tremor. We hypothe-
sized that for some cells, spike train and tremor comodulated with
that a substantial number of cells in both subcortical areas may be
nal synchrony within the basal ganglia (Heimer et al., 2006). In
erally consistent with reports of pairwise synchrony in the
majority of analyzed STN pairs (Levy et al., 2000, 2002). How-
pairs much further apart than the submillimeter separation in
Levy et al. (2000, 2002). Furthermore, Vim/Vop neuron pairs
in the ensemble were found to be tremor-tuned; no unsynchro-
nized cells were tremor-tuned (Table 7). This finding corrobo-
rates reports that the prevalence of STN pairs synchronized at
high frequencies is correlated to the degree of parkinsonian
tremor (Levy et al., 2000) and that patients without observable
tremor do not exhibit high-frequency STN synchrony (Levy et
al., 2002). Others have reported correlations between STN syn-
chrony and bradykinesia/rigidity (Weinberger et al., 2009). Our
findings have extended Levy et al.’s (2000, 2002) results to com-
binatorial pairs in neuronal ensembles from both STN and Vim/
Vop. It remains unclear to what degree network synchrony is
indicative of tremor pathology, although the effective disruption
of unstable network activity by DBS stimulation suggests a
itations, whereas long-term recordings from subcortical struc-
tations are performed every year with minimal risk (Bronstein et
yield more stable extracellular recordings (Porada et al., 2000;
Kru ¨ger et al., 2010) by mitigating the problem of electrode mi-
cromotion seen in cortical implants.
bring about the viability of subcortical BMI systems, first dis-
cussed in our previous study (Patil et al., 2004). Our reported
offline prediction results from our best Vim/Vop and STN ses-
sions are comparable to those reported from selected rhesus ma-
2000; Carmena et al., 2003). However, hyperbolic extrapolation
to achieve prediction fidelity above R2? 0.9. This would of
course necessitate the design of microelectrode arrays for
cannula-based implantation with more recording sensors and
demonstrable long-term safety and recording efficacy.
We propose that clinical studies using chronic ensemble re-
cordings in humans will permit both the continued study of the
neurophysiological mechanisms involved in motor control as
well as long-term monitoring of pathological activity. Chronic
recordings will facilitate continuous examination of changes in
tuning, oscillations, and synchrony as a function of the patient’s
symptomatic state both on and off treatment. Specifically, this
wealth of electrophysiological data may well be used to instruct
the improvement of closed-loop DBS systems that are currently
in initial stages of development (Rosin et al., 2011; Rouse et al.,
We have shown that within the STN and Vim/Vop thalamus,
there is an idiopathic mixture of pathology and behavior tuning;
dominant approach of delivering high-frequency DBS through
macroelectrodes. Subcortical ensembles remain an untapped re-
Movement-related neurons of the subthalamic nucleus in patients with
Parkinson disease. J Neurosurg 97:1167–1172.
AertsenAM,GersteinGL,HabibMK,PalmG (1989) Dynamicsofneuronal
Amirnovin R, Williams ZM, Cosgrove GR, Eskandar EN (2004) Visually
guided movements suppress subthalamic oscillations in Parkinson’s dis-
ease patients. J Neurosci 24:11302–11306.
Amtage F, Henschel K, Schelter B, Vesper J, Timmer J, Lu ¨cking CH, Hellwig
B (2008) Tremor-correlated neuronal activity in the subthalamic nu-
cleus of Parkinsonian patients. Neurosci Lett 442:195–199.
Awiszus F (1997) Spike train analysis. J Neurosci Methods 74:155–166.
Bankman IN, Johnson KO, Schneider W (1993) Optimal detection, classi-
8630 • J.Neurosci.,June20,2012 • 32(25):8620–8632Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles
fication, and superposition resolution in neural waveform recordings.
IEEE Trans Biomed Eng 40:836–841.
Bardorfer A, Munih M, Zupan A, Primozic A (2001) Upper limb motion
analysis using haptic interface. IEEE/ASME Trans Mechatronics
Batschelet E (1981) Circular statistics in biology. New York: Academic.
Benazzouz A, Breit S, Koudsie A, Pollak P, Krack P, Benabid AL (2002)
Intraoperative microrecordings of the subthalamic nucleus in Parkin-
son’s disease. Mov Disord 17 [Suppl 3]:S145–S149.
Birdno MJ, Kuncel AM, Dorval AD, Turner DA, Grill WM (2008) Tremor
varies as a function of the temporal regularity of deep brain stimulation.
Brodkey JA, Tasker RR, Hamani C, McAndrews MP, Dostrovsky JO, Lozano
AM (2004) Tremor cells in the human thalamus: differences among
neurological disorders. J Neurosurg 101:43–47.
Bronstein JM, Tagliati M, Alterman RL, Lozano AM, Volkmann J, Stefani A,
RE, DeLong MR (2011) Deep brain stimulation for Parkinson disease:
an expert consensus and review of key issues. Arch Neurol 68:165.
Carmena JM, Lebedev MA, Crist RE, O’Doherty JE, Santucci DM, Dimitrov
DF, Patil PG, Henriquez CS, Nicolelis MA (2003) Learning to control a
brain-machine interface for reaching and grasping by primates. PLoS
ChapinJK,MoxonKA,MarkowitzRS,NicolelisMA (1999) Real-timecon-
trol of a robot arm using simultaneously recorded neurons in the motor
cortex. Nat Neurosci 2:664–670.
Davidsen J, Schuster HG (2002) Simple model for 1/f(alpha) noise. Phys
Rev E Stat Nonlin Soft Matter Phys 65:026120.
Deuschl G, Bain P, Brin M (1998) Consensus statement of the movement
Deuschl G, Schade-Brittinger C, Krack P, Volkmann J, Scha ¨fer H, Bo ¨tzel K,
Herzog J, Hilker R, Klebe S, Kloss M, Koy J, Krause M, Kupsch A, Lorenz
D,etal. (2006) Arandomizedtrialofdeep-brainstimulationforParkin-
son’s disease. N Engl J Med 355:896–908.
Fetz EE (2007) Volitional control of neural activity: implications for brain-
computer interfaces. J Physiol 579:571–579.
Ghazanfar AA, Krupa DJ, Nicolelis MA (2001) Role of cortical feedback in
sensory thalamic neurons. Exp Brain Res 141:88–100.
Ghika J, Wiegner AW, Fang JJ, Davies L, Young RR, Growdon JH (1993)
Portable system for quantifying motor abnormalities in Parkinsons-
disease. IEEE Trans Biomed Eng 40:276–283.
Optical deconstruction of parkinsonian neural circuitry. Science
Grimaldi G, Piet L, Manto M (2007) Effects of wrist oscillations on con-
tralateral neurological postural tremor using a new myohaptic device
(‘wristalyzer’). In: Proceedings of the 4th IEEE-EMBS, pp. 44–48. Cam-
Guillery RW, Sherman SM (2002) The thalamus as a monitor of motor
outputs. Philos Trans R Soc Lond B Biol Sci 357:1809–1821.
Gutierrez R, Carmena JM, Nicolelis MA, Simon SA (2006) Orbitofrontal
ensemble activity monitors licking and distinguishes among natural re-
wards. J Neurophysiol 95:119–133.
Hamani C, Saint-Cyr JA, Fraser J, Kaplitt M, Lozano AM (2004) The sub-
thalamic nucleus in the context of movement disorders. Brain 127:4–20.
Heimer G, Rivlin M, Israel Z, Bergman H (2006) Synchronizing activity of
basal ganglia and pathophysiology of Parkinson’s disease. J Neural
Transm Suppl 70:17–20.
Hochberg LR, Serruya MD, Friehs GM, Mukand JA, Saleh M, Caplan AH,
BrannerA,ChenD,PennRD,DonoghueJP (2006) Neuronalensemble
control of prosthetic devices by a human with tetraplegia. Nature
Hua SE, Lenz FA (2005) Posture-related oscillations in human cerebellar
thalamus in essential tremor are enabled by voluntary motor circuits.
J Neurophysiol 93:117–127.
Kennedy PR, Bakay RA (1998) Restoration of neural output from a para-
lyzed patient by a direct brain connection. Neuroreport 9:1707–1711.
Kennedy PR, Bakay RA, Moore MM, Adams K, Goldwaithe J (2000) Direct
control of a computer from the human central nervous system. IEEE
Trans Rehabil Eng 8:198–202.
Kennedy PR, Kirby MT, Moore MM, King B, Mallory A (2004) Computer
control using human intracortical local field potentials. IEEE Trans Neu-
ral Syst Rehabil Eng 12:339–344.
Koller WC, Lyons KE, Wilkinson SB, Troster AI, Pahwa R (2001) Long-
term safety and efficacy of unilateral deep brain stimulation of the thala-
mus in essential tremor. Mov Disord 16:464–468.
Kru ¨gerJ,CaruanaF,VoltaRD,RizzolattiG (2010) Sevenyearsofrecording
from monkey cortex with a chronically implanted multiple microelec-
trode. Front Neuroeng 3:6.
Kuiper NH (1962) Tests concerning random points on a circle. Proc
Koninkl Neder Akad Wetensch 63:38–47.
Kumar R, Lozano AM, Sime E, Lang AE (2003) Long-term follow-up of
thalamic deep brain stimulation for essential and parkinsonian tremor.
Lebedev MA, Denton JM, Nelson RJ (1994) Vibration-entrained and pre-
movement activity in monkey primary somatosensory cortex. J Neuro-
Lebedev MA, Tate AJ, Hanson TL, Li Z, O’Doherty JE, Winans JA, Ifft PJ,
Zhuang KZ, Fitzsimmons NA, Schwarz DA, Fuller AM, An JH, Nicolelis
MA (2011) Future developments in brain-machine interface research.
Clinics (Sao Paulo) 66 [Suppl 1]:25–32.
Lenz FA, Tasker RR, Kwan HC, Schnider S, Kwong R, Murayama Y,
Dostrovsky JO, Murphy JT (1988) Single unit analysis of the human
ventral thalamic nuclear group: correlation of thalamic “tremor cells”
with the 3–6 Hz component of parkinsonian tremor. J Neurosci
Lenz FA, Kwan HC, Dostrovsky JO, Tasker RR, Murphy JT, Lenz YE (1990)
correlated with movement. Brain 113:1795–1821.
Lenz FA, Kwan HC, Martin RL, Tasker RR, Dostrovsky JO, Lenz YE (1994)
related activity in functionally identified cells. Brain 117:531–543.
Lenz FA, Jaeger CJ, Seike MS, Lin YC, Reich SG (2002) Single-neuron anal-
ysis of human thalamus in patients with intention tremor and other clin-
ical signs of cerebellar disease. J Neurophysiol 87:2084–2094.
LevyR,HutchisonWD,LozanoAM,DostrovskyJO (2000) High-frequency
synchronization of neuronal activity in the subthalamic nucleus of par-
kinsonian patients with limb tremor. J Neurosci 20:7766–7775.
Levy R, Ashby P, Hutchison WD, Lang AE, Lozano AM, Dostrovsky JO
(2002) Dependence of subthalamic nucleus oscillations on movement
and dopamine in Parkinson’s disease. Brain 125:1196–1209.
Magarin ˜os-Ascone CM, Figueiras-Mendez R, Riva-Meana C, Co ´rdoba-
Ferna ´ndez A (2000) Subthalamic neuron activity related to tremor and
movement in Parkinson’s disease. Eur J Neurosci 12:2597–2607.
Magnin M, Morel A, Jeanmonod D (2000) Single-unit analysis of the palli-
dum, thalamus and subthalamic nucleus in parkinsonian patients. Neu-
Marsden JF, Limousin-Dowsey P, Ashby P, Pollak P, Brown P (2001) Sub-
thalamic nucleus, sensorimotor cortex and muscle interrelationships in
Parkinson’s disease. Brain 124:378–388.
Nicolelis MA (2001) Actions from thoughts. Nature 409:403–407.
NicolelisMA,LebedevMA (2009) Principlesofneuralensemblephysiology
underlying the operation of brain-machine interfaces. Nat Rev Neurosci
MA (2011) Active tactile exploration using a brain-machine-brain in-
terface. Nature 479:228–231.
Ondo W, Jankovic J, Schwartz K, Almaguer M, Simpson RK (1998) Unilat-
eral thalamic deep brain stimulation for refractory essential tremor and
Parkinson’s disease tremor. Neurology 51:1063–1069.
Oppenheim AV, Schafer RW (1975) Digital signal processing. Englewood
Cliffs, NJ: Prentice-Hall.
Parent A, Hazrati LN (1995) Functional anatomy of the basal ganglia. II.
The place of subthalamic nucleus and external pallidum in basal ganglia
circuitry. Brain Res Brain Res Rev 20:128–154.
Patil PG, Turner DA (2008) The development of brain-machine interface
neuroprosthetic devices. Neurotherapeutics 5:137–146.
Patil PG, Carmena JM, Nicolelis MA, Turner DA (2004) Ensemble record-
a brain-machine interface. Neurosurgery 55:27–35; discussion 35–38.
Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsemblesJ.Neurosci.,June20,2012 • 32(25):8620–8632 • 8631
Porada I, Bondar I, Spatz WB, Kru ¨ger J (2000) Rabbit and monkey visual Download full-text
cortex: more than a year of recording with up to 64 microelectrodes.
J Neurosci Methods 95:13–28.
Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I (2005) Invariant
visual representation by single neurons in the human brain. Nature
RaevaS,VainbergN,TikhonovY,TsetlinI (1999) Analysisofevokedactiv-
ity patterns of human thalamic ventrolateral neurons during verbally
ordered voluntary movements. Neuroscience 88:377–392.
J, DeLong MR, Obeso JA (2001) The subthalamic nucleus in Parkin-
Rodriguez-Oroz MC, Obeso JA, Lang AE, Houeto JL, Pollak P, Rehncrona S,
Kulisevsky J, Albanese A, Volkmann J, Hariz MI, Quinn NP, Speelman
JD, Guridi J, Zamarbide I, Gironell A, Molet J, Pascual-Sedano B, Pidoux
B, Bonnet AM, Agid Y, et al. (2005) Bilateral deep brain stimulation in
Parkinson’s disease: a multicentre study with 4 years follow-up. Brain
Rosin B, Slovik M, Mitelman R, Rivlin-Etzion M, Haber SN, Israel Z, Vaadia
E, Bergman H (2011) Closed-loop deep brain stimulation is superior in
ameliorating parkinsonism. Neuron 72:370–384.
Rouse AG, Stanslaski SR, Cong P, Jensen RM, Afshar P, Ullestad D, Gupta R,
Molnar GF, Moran DW, Denison TJ (2011) A chronic generalized bi-
directional brain-machine interface. J Neural Eng 8:036018.
Instant neural control of a movement signal. Nature 416:141–142.
Su Y, Allen CR, Geng D, Burn D, Brechany U, Bell GD, Rowland R (2003)
3-D motion system (“data-gloves”): application for Parkinson’s disease.
IEEE Trans Instr Meas 52:662–674.
Taylor DM, Tillery SI, Schwartz AB (2002) Direct cortical control of 3D
neuroprosthetic devices. Science 296:1829–1832.
Theodosopoulos PV, Marks WJ Jr, Christine C, Starr PA (2003) Locations
of movement-related cells in the human subthalamic nucleus in Parkin-
son’s disease. Mov Disord 18:791–798.
Vinjamuri R, Crammond DJ, Kondziolka D, Lee HN, Mao ZH (2009) Ex-
traction of sources of tremor in hand movements of patients with move-
ment disorders. IEEE Trans Inf Technol Biomed 13:49–56.
WeinbergerM,HutchisonWD,DostrovskyJO (2009) Pathologicalsubtha-
akinesia? Exp Neurol 219:58–61.
J, Biggs SJ, Srinivasan MA, Nicolelis MA (2000) Real-time prediction of
hand trajectory by ensembles of cortical neurons in primates. Nature
Wiest MC, Bentley N, Nicolelis MA (2005) Heterogeneous integration of
bilateral whisker signals by neurons in primary somatosensory cortex of
awake rats. J Neurophysiol 93:2966–2973.
Williams ZM, Neimat JS, Cosgrove GR, Eskandar EN (2005) Timing and
direction selectivity of subthalamic and pallidal neurons in patients with
Parkinson disease. Exp Brain Res 162:407–416.
Zar JH (1999) Biostatistical analysis, 4th edition. Upper Saddle River, NJ:
Zirh TA, Lenz FA, Reich SG, Dougherty PM (1998) Patterns of bursting
occurring in thalamic cells during parkinsonian tremor. Neuroscience
8632 • J.Neurosci.,June20,2012 • 32(25):8620–8632 Hanson,Fulleretal.•PopulationAnalysisofHumanSubcorticalEnsembles