Articles in PresS. J Neurophysiol (April 15, 2015). doi:10.1152/jn.00988.2014
Cortical drive of low-frequency oscillations
in the human nucleus accumbens
during action selection
1
2
3
4
5
6
Max-Philipp Stenner1,2,*, Vladimir Litvak1, Robb B. Rutledge1,5, Tino Zaehle2, Friedhelm
C. Schmitt2, Jürgen Voges3,4, Hans-Jochen Heinze2,4, Raymond J. Dolan1,5
7
8
9
10
11
12
13
1Wellcome
Trust Centre for Neuroimaging, University College London, 12 Queen Square,
London WC1N 3BG, United Kingdom
2Department
of Neurology, Otto-von-Guericke-University Magdeburg, Leipziger Str. 44,
39120 Magdeburg, Germany
3Department
14
15
4Department
16
17
5Max
of Stereotactic Neurosurgery, Otto-von-Guericke-University Magdeburg,
Leipziger Str. 44, 39120 Magdeburg, Germany
of Behavioral Neurology, Leibniz Institute for Neurobiology, Brenneckestr.
6, 39118 Magdeburg, Germany
Planck University College London Centre for Computational Psychiatry and Ageing
Research, London WC1B 5EH, United Kingdom
18
*correspondence: Wellcome Trust Centre for Neuroimaging, 12 Queen Square, WC1N
19
3BG, UK, telephone: 0044-20 3448 4377, email: [email protected]
20
21
Running title: Cortico-accumbens coupling during action selection
22
23
24
25
1
Copyright © 2015 by the American Physiological Society.
26
Abstract
27
The nucleus accumbens is thought to contribute to action selection by integrating
28
behaviourally relevant information from multiple regions including prefrontal cortex.
29
Studies in rodents suggest that information flow to the nucleus accumbens may be
30
regulated via task-dependent oscillatory coupling between regions. During instrumental
31
behaviour, local field potentials in the rat nucleus accumbens and prefrontal cortex are
32
coupled at delta frequencies (Gruber et al., 2009a), possibly mediating suppression of
33
afferent input from other areas and thereby supporting cortical control (Calhoon and
34
O’Donnell, 2013). Here, we demonstrate low-frequency cortico-accumbens coupling in
35
humans, both at rest and during a decision-making task. We recorded local field
36
potentials (LFP) from the nucleus accumbens in six epilepsy patients who underwent
37
implantation of deep brain stimulation electrodes. All patients showed significant
38
coherence and phase-synchronization between LFP and surface EEG at delta and low
39
theta frequencies. While the direction of this coupling indexed by Granger causality
40
varied between subjects in the resting-state data, all patients showed a cortical drive of
41
the nucleus accumbens during action selection in a decision-making task. In three
42
patients this was accompanied by a significant coherence increase over baseline. Our
43
results suggest that low-frequency cortico-accumbens coupling represents a highly
44
conserved regulatory mechanism for action selection.
45
46
Keywords
47
nucleus accumbens, action selection, local field potentials, synchronization, Deep Brain
48
Stimulation
2
49
Introduction
50
The nucleus accumbens contributes to the selection of goal-directed actions by
51
integrating behaviourally relevant information from the hippocampus, the amygdala
52
and prefrontal cortex, among other areas (Grace, 2000; Goto and Grace, 2008). These
53
inputs are likely integrated as a function of current task requirements (Gruber et al.,
54
2009a; Calhoon and O’Donnell, 2013; Stuber, 2013). Previous work suggests that this
55
task-dependent integration may be regulated by inter-area coupling of oscillatory
56
synaptic potentials. In rodents, theta oscillations (around 7 Hz) in the ventromedial
57
striatum (which includes the nucleus accumbens) are coherent with hippocampal theta
58
oscillations during active exploration (Gruber et al., 2009a) and in maze tasks (Berke et
59
al., 2004; van der Meer and Redish, 2011). Entrainment of ventral striatal single units to
60
the hippocampal theta rhythm has been associated with combined contextual (spatial)
61
and motivational information (Tabuchi et al., 2000; van der Meer and Redish, 2011).
62
Together, these results have led to the proposal that theta coupling between the
63
hippocampus and the ventral striatum/nucleus accumbens may regulate how
64
contextual information influences action selection (Grace, 2000; Gruber et al., 2009a;
65
Pennartz et al., 2009).
66
67
Instrumental behaviour, on the other hand, has been associated with oscillatory
68
coupling of the nucleus accumbens with prefrontal cortex at delta frequencies (1-4 Hz).
69
Gruber et al. (2009) found strong delta coherence between the nucleus accumbens core
70
and prefrontal cortex, accompanied by enhanced single-unit cross-correlation, when
71
rats engaged in instrumental behaviour as compared to periods of active exploration.
72
Because theta coherence between the nucleus accumbens core and the hippocampus
3
73
was simultaneously reduced, the authors suggested that cortico-accumbens coupling
74
may override a hippocampal influence on neuronal excitability in the nucleus
75
accumbens (Calhoon and O’Donnell, 2013), thereby promoting behavioural flexibility.
76
Translation of these findings to humans has remained speculative, largely because
77
electrophysiological data from the human nucleus accumbens are rare. Such data have
78
recently become more tractable via recordings from deep brain stimulation (DBS)
79
electrodes, implanted in the nucleus accumbens for treatment of neurological or
80
neuropsychiatric disorders.
81
82
Here, we examined functional and directed connectivity between cortex and the nucleus
83
accumbens in six patients treated with DBS for drug-resistant epilepsy. We studied
84
coherence, phase-synchronization and Granger causality both at rest and during a
85
value-based decision-making task. Given reports of cortico-accumbens connectivity at
86
delta frequencies in rodents (Gruber et al., 2009a), we expected to find low-frequency
87
coupling between cortical EEG and local field potentials (LFP) in the nucleus
88
accumbens. Furthermore, we predicted that the strength of this coupling would
89
increase during the action selection stage of our decision-making task, accompanied by
90
evidence for a cortical drive of low-frequency oscillations in the nucleus accumbens, as
91
demonstrated in rodents (Gruber et al., 2009a; Calhoon and O’Donnell, 2013). Scalp
92
recordings may reflect hippocampal signals in addition to neocortical signals (e.g.
93
Kaplan et al., 2012). Here, we could rule out a substantial hippocampal contribution to
94
the observed LFP/EEG coupling in two of the patients who had undergone unilateral
95
resection of the hippocampus prior to DBS surgery.
4
96
Methods
97
Patients
98
Six patients with pharmacoresistant partial epilepsy participated in the study (mean ±
99
SD age: 39.5 ± 8.6 years; three females; all right handed; see table 1 for details). All
100
patients participated in in-house protocols aiming to objectify safety and potential anti-
101
ictal efficacy of deep brain stimulation of the nucleus accumbens. Clinical outcomes
102
have been reported elsewhere (Schmitt et al., 2014). Both this clinical trial and the
103
experiments reported here were approved by the institutional review board of the
104
University of Magdeburg (registration number 03/08), and all patients gave written
105
informed consent.
106
107
Electrode implantation
108
All patients underwent stereotactically guided surgery with implantation of quadripolar
109
electrodes in the nucleus accumbens and anterior thalamus of each hemisphere
110
(Medtronic, model 3387). Planning of electrode placement and surgical procedures
111
were performed as described in detail in previous work (Voges et al., 2002; Zaehle et al.,
112
2013). For the nucleus accumbens, standardized coordinates were 2 mm rostral to the
113
anterior border of the anterior commissure at the level of the mid-sagittal plane, 3–4
114
mm ventral and 6–8 mm lateral of the midline. For each patient, these standardized
115
coordinates were adjusted according to pre-surgical MRI, taking into account the
116
anatomy of the vertical limb of Broca’s diagonal band as an important landmark.
117
Specifically, electrodes were placed 2–2.5 mm lateral of the vertical limb of Broca’s
118
band, in the caudo-medial part of the nucleus accumbens. This area is thought to
5
119
correspond to the remnant of the shell area in primates (Sturm et al., 2003), as defined
120
by histochemical criteria. The position of each electrode was confirmed by
121
intraoperative stereotactic X-ray images and postoperative CT. At least two contacts of
122
each quadripolar electrode were placed inside the nucleus accumbens, covering those
123
parts equivalent to the nucleus accumbens shell and core regions in rodents.
124
125
Following surgery, electrode leads were externalized for six days, allowing test
126
stimulation with different parameters as well as recordings from depth electrodes in
127
various psychological tasks and at rest. Subsequently, electrode cables were connected
128
to an impulse generator located beneath the left pectoral muscle.
129
130
Procedure & task
131
Data were collected on two separate days between the third and fifth day post surgery,
132
with one dataset collected at rest and one collected during a decision-making task. Four
133
patients contributed both resting-state data and task-related data. For one patient (P5),
134
only resting-state data were collected, and for another patient (P6), only task-related
135
data were collected. Consequently, each dataset comprised five patients. For the resting-
136
state recordings, patients sat comfortably in an upright position and were asked to
137
remain as still as possible, with their eyes open. Three to ten minutes of resting-state
138
data were recorded.
139
6
140
The task dataset was recorded during a standard economic decision-making task,
141
presented on a laptop computer. The nucleus accumbens is active in a variety of tasks
142
that involve action selection (Nicola, 2007). We used a task for which strong responses
143
are reliably recorded from the human nucleus accumbens during action selection in
144
previous functional MRI studies (Tom et al., 2007; Rutledge et al., 2014). Activity of
145
single neurons in the nucleus accumbens can be used to predict subsequent actions in a
146
financial decision-making task (Patel et al., 2012), even after controlling for the
147
probability of reward. Because the objective value of the gamble in our task changes
148
from trial to trial, making appropriate actions requires subjects to determine the
149
subjective value of each available gamble. This makes such an economic decision task
150
ideal for studying goal-directed actions. Furthermore, this task allowed us to
151
additionally test for value and reward prediction error signals that might relate to
152
dopamine, which have been identified in this area using other techniques (Patel et al.,
153
2012; Hart et al., 2014). These investigations are the subject of a separate study.
154
155
The task required patients to decide whether to accept or reject a monetary gamble
156
offer. If accepted, a gamble could result in a monetary gain or loss with equal
157
probabilities. Each patient made 200 choices between this risky gamble option and a
158
safe option, which was worth 0 euros. There were five different win amounts for the
159
risky option (25, 40, 55, 75, 100 euro cents). Loss amounts for the risky option (5-200
160
euro cents) were determined by multiplying gain amounts by 20 different multipliers,
161
ranging from 0.5 to 5 to accommodate a wide range of gain-loss sensitivity. The risky
162
gamble option and the safe option were represented by three numbers on the screen,
163
each presented in black font in the centre of a white rectangle (figure 2A; the screen
7
164
background colour was black). Two of the three numbers were presented on one side of
165
the screen (left or right) and corresponded to the possible gain and loss of the current
166
gamble offer. The potential loss amount, indicated by a negative sign, and the possible
167
win amount were presented slightly below and above the horizontal meridian of the
168
screen, respectively, at equal eccentricities from the centre of the screen. The third
169
number, representing the safe option, was always zero and was presented on the
170
opposite side of the screen, on the horizontal meridian. The side of the screen
171
(left/right) on which each of the two options (safe option and gamble) was presented
172
was counterbalanced across trials.
173
174
Each trial started with the presentation of the two options on the screen. Patients chose
175
the option presented on the left or on the right by pressing one of two keys on a
176
keyboard with their left or right hand, respectively (left “Ctrl” and “Enter” on the
177
number block, respectively). There was no time limit for this decision. If patients chose
178
to gamble, the outcome was randomly determined by the computer and displayed after
179
a 2 s delay period for 1.5 s. The inter-trial interval (ITI) was jittered between 1.5 and 2 s.
180
The outcome of each trial counted for real money. Subjects were endowed with 15
181
euros at the start of the experiment and, in addition to this endowment, earned an
182
average of 15.66 euros (range, 3.73 to 27.38 euros), which was paid out at the end of the
183
experiment. Before the main experiment, patients practised the task and became
184
familiar with the range of possible wins and losses for 50 trials, which did not count for
185
real money.
186
8
187
Recording and analyses of local field potentials
188
LFP- and surface EEG-recording
189
Local field potentials were recorded continuously from four contacts on each of the four
190
DBS-electrodes (one in the nucleus accumbens and one in the anterior thalamic nucleus
191
of each hemisphere). Electrode contacts (platinum-iridium) were 1.5 mm wide and
192
spaced 1.5 mm apart (edge-to-edge distance). Simultaneous to LFP recordings, data
193
from two to five surface EEG electrodes (gold plated silver electrodes) were also
194
recorded (no online filters). Surface electrode number and placement varied between
195
patients due to post-surgical head bandages. The only surface EEG channel that was
196
available across all patients and both datasets was Cz (positioned according to the 10-
197
20 system).
198
199
Data were digitized at a sampling rate of 512 Hz using a Walter Graphtek system.
200
Following previous work (Litvak et al., 2012), we analysed LFP data from DBS-
201
electrodes in a bipolar montage, in which each contact is referenced to one
202
neighbouring contact. This results in three channels per DBS-electrode (i.e., each of the
203
three most ventral contacts was referenced against its dorsal neighbour). This bipolar
204
montage maximizes spatial specificity for the nucleus accumbens by minimizing volume
205
conduction effects from distant sources. Surface EEG channels were referenced against
206
the left ear lobe.
207
208
Preprocessing
9
209
LFP and surface EEG data were analysed using FieldTrip (Oostenveld et al., 2011) and
210
Matlab (version R2012a; Mathworks). Line noise was removed using a narrow-band
211
bandstop-filter (4th order, two-pass Butterworth filter). Visual inspection showed a slow
212
drift at the very beginning of the resting-state data, likely due to a long time constant of
213
the amplifier. This drift was removed using a high-pass filter (6th order, two-pass
214
Butterworth filter; cutoff frequency: 0.5 Hz). For each patient, resting-state data were
215
then epoched into consecutive segments of 2 seconds each (yielding 80 to 284 epochs
216
per patient). For the task dataset, data were epoched from 1.5 seconds before the onset
217
of the two options on the screen to 5 seconds after patients indicated their choices by
218
pressing a key (this corresponds to 3 seconds after the onset of the outcome in trials in
219
which patients chose to gamble). Because movement artefacts were more likely during
220
the task than during the resting-state recordings (for which patients only had to sit
221
still), a simple artefact rejection routine was added for the task dataset. Specifically, a
222
trial was rejected if the maximum amplitude variance across all available channels in
223
that trial exceeded a set threshold, as implemented by standard options in FieldTrip.
224
This resulted, on average, in a rejection of 12.5% of all task-related trials.
225
226
Time windows for distinct task stages
227
Our primary interest in the task dataset was to test for a modulation of cortico-
228
accumbens connectivity during different stages of the task. Specifically, we asked how
229
connectivity changed as patients decided between the two options and anticipated and
230
subsequently saw the outcome of a chosen gamble. To compare connectivity during the
231
decision-, anticipation- and outcome stage to a baseline, we divided each trial into four
232
consecutive time windows, each equal to 1.5 seconds in duration (figure 2A). The 1.5
10
233
seconds prior to the onset of the two options on the screen served as a baseline (“ITI” in
234
figure 2A). The decision stage was defined as the time interval 1.5 seconds before
235
patients pressed a key to choose one of the two options. Trials in which patients
236
responded to the presentation of the options faster than 1.5 seconds were excluded
237
from analysis (of all four task stages). Given response times in the task (median: 2.09 s
238
(P1), 3.04 s (P2), 1.52 s (P3), 2.11 s (P4) and 1.46 s (P6)), this time interval of 1.5 s is
239
likely to include (aspects of) the decision process. A pre-response interval of 1.5 s also
240
allowed us to use analysis time windows of equal duration across the four task stages
241
(note that, across trials, the shortest inter-trial interval available as a baseline was 1.5
242
s). Accordingly, the anticipation stage of the task was defined as the 1.5 seconds that
243
preceded the onset of the gamble outcome, while the 1.5 seconds after outcome
244
presentation corresponded to the outcome stage.
245
246
By design, trials in which patients chose the safe option differed fundamentally from
247
gamble trials in that, in the former, patients knew immediately the outcome of their
248
decision. Consequently, anticipation- and outcome stages do not exist in trials in which
249
the safe option was chosen, at least not in the same sense as in gamble trials. Because
250
the connectivity metrics used here are sensitive to the number of trials used for
251
analysis, trials in which patients chose the safe option were excluded from the
252
comparison of the four task stages (note that we report a control analysis of our main
253
contrast of interest – comparing the decision stage to the baseline – which does include
254
trials in which the safe option was chosen). The numbers of trials available for all four
255
task stages, i.e., trials in which subjects chose to gamble and responded with a minimum
256
reaction time of 1.5 seconds, were 66 (P1), 102 (P2), 48 (P3), 103 (P4) and 63 (P6).
11
257
258
Functional and directed connectivity
259
Because the brain and surrounding tissues have no relevant capacitance across the
260
frequency range of interest in EEG (<<1 kHz), signals measured at surface electrodes
261
instantaneously reflect underlying cortical generators (e.g., Penny et al., 2011; Schomer
262
and Lopes da Silva, 2012). Because of this, connectivity between LFP recorded via DBS
263
electrodes and surface EEG can be interpreted as cortico-subcortical coupling.
264
265
We first computed the spectral coherence between the signal at Cz and the signal at
266
each of the six accumbens channels (three for each hemisphere). Coherence indexes the
267
consistency of a linear phase- and amplitude-relation between two signals across trials
268
(Walter, 1963). Coherence is defined as the absolute value of the trial-averaged cross-
269
spectrum, normalized by the square root of the product of the two trial-averaged
270
autospectra. Coherence ranges from 0 to 1, with higher values indicating higher
271
consistency of a phase- and amplitude-relation between two signals. Coherence was
272
computed across trials from the cross- and auto-spectra after obtaining a fast Fourier
273
transform of the Hanning-tapered data for each trial/resting-state epoch.
274
275
Because coherence intermingles phase and amplitude, we tested for significant phase-
276
synchronization between cortex (Cz) and the nucleus accumbens with a separate
277
metric. To test for phase-synchronization, the weighted phase-lag index (wPLI) was
278
computed, a recent extension (Vinck et al., 2011) of the phase-lag index (PLI; Stam,
279
Nolte, & Daffertshofer, 2007). The (signed) PLI is defined as the trial-averaged sign of
12
280
the imaginary component of the cross-spectrum and ranges from -1 to 1. In cases where
281
two signals show little or no (non-zero) phase-synchronization, the PLI tends towards
282
zero. Values closer to -1 or 1, on the other hand, indicate that phase lags or leads
283
between two signals are not equiprobable across trials, thus reflecting a consistent
284
phase-relation. A further improvement in the sensitivity of the PLI has recently been
285
demonstrated by weighting the sign of the imaginary part of the cross-spectrum by its
286
magnitude (i.e., by the π/2 component of the phase difference between signals; Vinck et
287
al., 2011). Here, the wPLI was computed on the basis of the single-trial Fourier spectra
288
using standard settings in FieldTrip.
289
290
To determine directionality of information flow between cortex and the nucleus
291
accumbens, we computed a nonparametric variant of pairwise Granger causality
292
(Dhamala et al., 2008). Pairwise Granger causality indexes the contribution of one time
293
series to forecasting a second time series, over and above the variance already
294
explained by the past information of the latter. In the frequency domain, the
295
computation of Granger causality is based on the spectral transfer function and the
296
noise covariance matrix, which can be obtained either from an autoregressive model or,
297
alternatively, nonparametrically, from a factorization of the spectral matrix. The latter
298
approach avoids any assumptions regarding a specific autoregressive model order and
299
was used here. Nonparametric Granger causality was computed from single-trial
300
Fourier spectra in FieldTrip, using standard settings (i.e., a factorization based on the
301
Wilson algorithm (Wilson, 1972)).
302
13
303
Delays between surface EEG and nucleus accumbens LFP were obtained from a
304
regression of phase differences on frequency, following previous work (Rosenberg et al.,
305
1989; Riddle and Baker, 2005). The rationale behind a phase-frequency regression is
306
that a constant conduction delay should lead to a linear increase in phase differences
307
across a frequency range that shows significant coupling. Phase differences for the
308
combination of each nucleus accumbens channel with Cz were obtained from the cross-
309
spectra of each patient. Because we expected similar delays for all nucleus accumbens
310
channels, phase differences were pooled across channels (this also increases the
311
number of data points available for the regression analysis; see Riddle & Baker, 2005,
312
for a similar pooling of channels). For each patient, phase differences were then linearly
313
regressed on frequency across the low-frequency range that showed significant
314
coherence in that particular patient (figure 1A). The slope of the regression line in ms
315
was used as a proxy for conduction delays. Because frequency windows with significant
316
coherence were quite narrow in some patients, particularly in P3 and P4, phase
317
differences were computed across resting-state epochs of 6.7 seconds each to increase
318
the frequency resolution and, thereby, the number of frequency bins available for the
319
regression analysis. This value was chosen to achieve a high frequency resolution while
320
keeping the number of epochs available for the estimation of the cross-spectrum as high
321
as possible.
322
323
Only two patients showed a significant fit of the regression line and acceptable R2-
324
values (both p<.0005; R2-values of .074 and .16), presumably because their resting-state
325
recordings were longest (~10 mins each). In the other three (3-5 mins of recordings),
326
the fit was far from significant (all p>.2, R2<.01), possibly due to an insufficient number
14
327
of trials available to estimate the cross-spectrum with sufficiently low noise (≤40 trials
328
each) and a relatively low number of frequency bins for the regression (as low as 11).
329
Since data segments for the different task stages were only 1.5 seconds long, delays
330
were only computed from the resting-state data.
331
332
For all connectivity analyses, the lowest frequency and the frequency resolution were
333
defined as the inverse of the trial/epoch length in seconds. For the resting-state dataset,
334
coherence spectra, wPLI spectra and Granger causality spectra were explored over a
335
broad frequency range (up to 100 Hz). Having established a frequency window of
336
interest in our analysis of the resting-state data, we tested for a task-dependent
337
modulation of cortico-accumbens connectivity in this frequency window (≤6 Hz). Note
338
that this frequency window of interest was obtained from the independent analysis of
339
the resting-state dataset. Since we had no prior hypothesis regarding any difference
340
between the left and right nucleus accumbens, or between ventral and dorsal channels,
341
mean spectra across all six nucleus accumbens channels were used for statistical
342
analysis. We show data for the left and right nucleus accumbens separately in figure 3.
343
Separate analyses of the three DBS channels of each electrode showed no systematic
344
differences across individuals.
345
346
Statistics
347
Due to the moderate size of our study sample (five patients for each dataset) statistical
348
group-level inference is inappropriate. Instead, we report statistical results for
349
individual patients and note commonalities and differences between single-patient
15
350
statistical results. All statistical analyses were based on nonparametric resampling tests.
351
To test for significant peaks in the coherence- and wPLI spectra, the trial order of single-
352
trial Fourier spectra was randomly shuffled for the nucleus accumbens, but not for Cz
353
(the same shuffled trial order was used for all six nucleus accumbens channels).
354
Coherence and the wPLI were then re-computed from the shuffled data. This was
355
repeated for at least 4000 iterations, each with a unique, shuffled trial order for the
356
nucleus accumbens. For each frequency bin, a p-value was computed as the proportion
357
of iterations whose (coherence- or wPLI-) values exceeded the value obtained from the
358
original data. These p-values were corrected for multiple comparisons across frequency
359
bins using Bonferroni correction (between 0 and 100 Hz for the resting-state data and
360
≤6 Hz for the task data). Bonferroni-corrected p-values below .025 were considered
361
significant (corresponding to a two-sided test). To test for significant peaks in the
362
coherence spectrum, a one-sided threshold (.05) was used because we were only
363
interested in coherence which was larger for the original data than the shuffled data.
364
365
To test for significant peaks in the Granger causality spectra, we compared spectra
366
computed from the original data to Granger causality obtained after reversing the order
367
of time bins within each channel (effectively reversing the time axis in all channels), as
368
recently suggested by Haufe et al. (2013). Because Granger causality is sensitive to
369
temporal order, reversing the time axis should also reverse the direction of information
370
flow as indexed by Granger causality. In contrast, any spurious causality, for example,
371
due to common coloured noise or differences in the signal-to-noise-ratio between
372
channels (e.g., Nolte et al., 2008), should not be affected by a reversal of the time axis
373
(Haufe et al., 2012). To test whether Granger causality was significantly greater when
16
374
computed from the original vs. the time-reversed data, a nonparametric permutation
375
test was used. For each trial, Fourier spectra from the original and the time-reversed
376
data were first randomly assigned to two separate pools of trials. Granger causality was
377
then re-computed for each pool. This was repeated for at least 1000 iterations, each
378
with a unique permutation. For each frequency bin, a p-value was obtained as the
379
proportion of permutations which resulted in a Granger causality difference that was
380
larger than the observed difference between the original data and the time-reversed
381
data. These p-values were corrected for multiple comparisons across frequency bins
382
using Bonferroni correction (between 0 and 100 Hz for the resting-state data and ≤6 Hz
383
for the task data). Since we were only interested in Granger causality peaks which were
384
larger for the original than the time-reversed data, a one-sided significance threshold
385
was used (Bonferroni-corrected p-values below .05).
386
387
Permutation tests were also used to test for a significant modulation of coherence
388
during the decision-making task. Here, we tested for changes in cortico-accumbens
389
coherence during each of the decision-, anticipation- and outcome stages of the task,
390
each in relation to baseline. As a baseline, we used the inter-trial interval instead of our
391
resting-state data because the two datasets were recorded on two separate days, so that
392
potentially confounding factors, e.g., differences in tiredness, were not controlled for. Having
393
established a frequency window of interest in the analysis of the resting-state data (≤6
394
Hz), p-values were obtained from the permutation distribution after averaging
395
coherence values across this frequency window of interest. p-values were Bonferroni-
396
corrected for the number of contrasts (3) and considered significant if they fell below
397
.025 (corresponding to a two-sided test). The same procedure was used to test for
17
398
differences in cortico-accumbens connectivity between the left and right nucleus
399
accumbens and for differences between cortico-accumbens and cortico-thalamic
400
coherence.
18
401
Results
402
Cortico-accumbens connectivity at rest
403
We first tested for significant coherence between surface-channel EEG (at Cz) and LFP
404
from the nucleus accumbens at rest. A broad frequency range was explored (up to 100
405
Hz). As expected, we found significant peaks in the coherence spectrum at low
406
frequencies (between 2.5 to 5 Hz). Remarkably, a significant low-frequency peak was
407
present in each of the five patients (figure 1A; Bonferroni-corrected for multiple
408
comparisons across all frequency bins up to 100 Hz). Peak frequencies and peak
409
morphology varied across patients, with some patients (P1, P2 and P5), showing more
410
than one distinct peak at low (delta/theta) frequencies. Up to 100 Hz, no other
411
significant peaks were identified which were consistent across patients (P2 showed a
412
separate peak at beta frequencies). When comparing cortico-accumbens vs. cortico-
413
thalamic coherence, averaged across frequencies ≤ 6 Hz, cortico-accumbens coherence was
414
significantly stronger in three patients (P2, P3 and P5; all p<.001). The remaining two patients
415
showed trends in the same direction (P4: p=.07; P1: p=.15).
416
417
Coherence intermingles the amplitude- and phase-relation between two signals. To test
418
specifically for cortico-accumbens phase-synchronization, we computed the weighted
419
phase-lag index (wPLI; Vinck et al., 2011), a measure of a consistent (non-zero) phase-
420
relation between two signals. Results closely resembled our coherence findings. Each of
421
the five patients showed a significant low-frequency (3 to 5 Hz) peak in the wPLI
422
spectrum (figure 1B; Bonferroni-corrected for multiple comparisons across all
423
frequency bins up to 100 Hz). In one patient (P2), the sign of the wPLI separated three
19
424
distinct peaks in the spectrum, one at delta/theta frequencies, as in all other patients,
425
and one each at higher theta frequencies and in the beta band.
426
427
Both spectral coherence and the wPLI measure functional connectivity. To study the
428
direction of information flow between cortex and the nucleus accumbens at rest, a
429
nonparametric variant of spectral Granger causality was computed. Significant peaks at
430
low frequencies (< ~10 Hz) were identified in the Granger causal spectrum of each
431
patient (figure 1C; Bonferroni-corrected for multiple comparisons across all frequency
432
bins up to 100 Hz). The direction of information flow, from cortex to the nucleus
433
accumbens or vice versa, varied between patients. Two patients (P2 and P4) showed a
434
predominantly subcortical drive (nucleus accumbens leading). In one patient (P1), we
435
observed a predominantly cortical drive. Patient P5 showed a cortical drive at ~2.5 Hz
436
and a subcortical drive at ~9 Hz, and in patient P3, a clear distinction between
437
directions was difficult to establish.
438
439
Delays between surface EEG and nucleus accumbens LFP were estimated from the slope
440
of a linear regression of phase differences on frequency across the range that showed
441
significant coherence peaks (figure 1A). Only two patients (P1 and P2) showed a
442
significant fit of the linear regression (both p<.0005; R2-values of .074 and .16), while
443
the fit was not significant in the remaining three patients (all p>.2, R2<.01; see methods
444
for a discussion of possible reasons). In P1, the estimated delay from cortex to the
445
nucleus accumbens was +38 ms, while in P2, the delay was -100 ms. These delays are in
446
line with the Granger causality results from these two patients (figure 1C). Patient P1,
20
447
who showed a predominantly cortical drive, had a cortico-accumbens delay of +38 ms.
448
The magnitude of this delay is longer than the (likely monosynaptic) conduction delays
449
of ≤20 ms observed in rats (see discussion; McGinty & Grace, 2009). In P2, on the other
450
hand, Granger causality revealed a predominantly subcortical drive, consistent with a
451
delay of -100 ms and a polysynaptic influence of the nucleus accumbens on cortex.
452
453
Taken together, our resting-state analyses confirmed that cortical EEG signals (at Cz)
454
and LFP from the nucleus accumbens are coupled at low (delta and low theta)
455
frequencies. At rest, the direction of this coupling varies between individuals. One
456
possible explanation for this heterogeneity is that, by definition, resting-state
457
recordings do not control which task a specific brain network is currently engaged in.
458
To study how cortico-accumbens connectivity is modulated during decision-making, we
459
analysed functional and directed connectivity during a standard decision-making task.
460
461
Cortico-accumbens connectivity during decision-making
462
During the task, patients chose between a certain amount of money (0 euros) and a
463
risky gamble option which could lead to a monetary win or loss (figure 2A).
464
Behavioural analysis confirmed that patients paid attention to the task and chose
465
between options on the basis of the current gamble offer. Specifically, we tested
466
whether the proportion of trials in which patients chose to gamble was higher when the
467
mathematical expected value of the risky gamble option was relatively high vs. low (as
468
defined by a median split). A nonparametric permutation test (1000 unique
469
permutations) confirmed that this was the case in four of the five patients (% gamble
21
470
chosen, high vs. low expected value: 59% vs. 33% (P1), 68% vs. 53% (P2), 78% vs. 47%
471
(P3), 93% vs. 70% (P6); all permutation p-values < .025, corresponding to a two-sided
472
test). Only one patient (P4) chose the gamble offer equally often when its expected
473
value was relatively high vs. low (76% vs. 76%). Subjects earned, on average, an
474
additional 15.66 euros (range, 3.73 to 27.38 euros). In contrast, a random chooser
475
would earn an additional 8.83 euros on average, and four of the five patients earned
476
more than that. Patients made their choices with a mean median response time of 2.09
477
seconds (range of median response times, 1.46-3.04 seconds). On average, they chose to
478
gamble in 65.3% of trials. There was considerable variability in preferences across
479
subjects (range, 46-81.5%), as expected in a task like this (Tom et al., 2007). Four of the
480
five patients were consistent in their preference for choosing gambles, as evidenced by
481
permutation tests that compared the tendency to choose a gamble in the first vs. second
482
half of the experiment. Patients P1, P2, P4 and P6 chose a similar proportion of gambles
483
in the first and second half (p> .09, threshold for significance at .025), while patient P3
484
chose significantly more gambles in the second than in the first half, p=.005).
485
486
We were interested how direction and strength of connectivity between cortex and the
487
nucleus accumbens changed during the course of a trial, specifically in the low-
488
frequency window of interest that we established in our resting-state analyses (≤ 6 Hz).
489
To this end, low-frequency coherence values during the decision-, anticipation- and
490
outcome stages of the task were compared to a baseline (the inter-trial interval). Figure
491
2B shows coherence spectra during all four stages of the task (the colour-coding
492
corresponds to the colours in figure 2A). Four of the five patients (P2 to P6) showed at
493
least one peak in the frequency window of interest (≤6 Hz) in at least one of the four
22
494
task stages. Peak frequencies were slightly lower than at rest (around 2 Hz).
495
Numerically, the highest coherence peak occurred during the decision stage of the task
496
in all five patients, followed by the anticipation stage in four of the five patients.
497
Coherence below 6 Hz was significantly higher during the decision stage as compared to
498
baseline in three of the five patients (P2, P3 and P6; all p-values < .025 after Bonferroni-
499
correction). The patient whose coherence- and wPLI spectra at rest revealed two
500
distinct peaks at delta and theta frequencies (P2, figure 1A and B) also showed two
501
corresponding coherence peaks during the task. When comparing the anticipation- and
502
outcome stage to baseline, we found a significant enhancement of coherence across the
503
frequency window of interest only in one patient and one task stage (the outcome stage
504
in P1). Trials in which patients chose the safe option (0 euros) were excluded from the
505
analysis of task coherence to this point to avoid differences in trial numbers (see
506
methods). However, when computing coherence across all trials and comparing the
507
decision stage to baseline, we found a very similar pattern of results (with P2, P3 and P6
508
showing a significant enhancement of low-frequency coherence during the decision
509
stage; P4 showed a trend for significance of .031). In contrast to this increase in cortico-
510
accumbens coherence during the action selection stage of the task, no task-modulation of
511
cortico-thalamic coherence ≤6 Hz was observed in any of the five patients (all p>.3).
512
513
We tested whether this enhancement of functional connectivity during the decision
514
stage coincided with a cortical drive of low-frequency oscillations in the nucleus
515
accumbens, as predicted from the animal literature (Gruber et al., 2009a; Calhoon and
516
O’Donnell, 2013). Granger causality spectra confirmed that all patients showed a
517
significant cortical drive at frequencies similar to their coherence peaks (except P3, who
23
518
showed a trend of p=.08; figure 2C, Bonferroni-corrected for multiple comparisons
519
across all frequency bins ≤ 6 Hz). For P2, there was a significant cortical drive at ~2.5 Hz
520
and a significant subcortical drive at ~5.5 Hz, which, together, mapped well onto the
521
two distinct low-frequency peaks observed in the coherence spectra of this patient at
522
rest and during the task.
523
524
EEG/LFP connectivity does not reflect hippocampus/nucleus accumbens coupling
525
Previous studies suggest that scalp recordings may reflect hippocampal signals in
526
addition to neocortical signals (e.g. Kaplan et al., 2012). In rodents, oscillations in the
527
hippocampus and the nucleus accumbens are coupled at theta frequencies (Berke et al.,
528
2004; Gruber et al., 2009a). Accordingly, the EEG/LFP connectivity observed here could,
529
in principle, reflect connectivity between the nucleus accumbens and the hippocampus,
530
rather than the neocortex. To address a potential hippocampal influence on our
531
connectivity findings, we made use of the fact that two of the patients had undergone
532
surgical resection of the medial temporal lobe to treat their epilepsy. Patients P3 and P4
533
had undergone right and left temporal lobe resection, respectively, three and nine years
534
before DBS surgery. In non-human primates (unlike in rodents) hippocampal
535
projections to the nucleus accumbens do not cross to the contralateral hemisphere
536
(Friedman et al., 2002). Under the assumption that this holds for humans (see
537
discussion), we tested for differences in delta/theta coherence as a function of
538
hemisphere (operated vs. non operated hemisphere). If the hippocampus made a
539
substantial contribution to the observed low-frequency coherence, we would expect
540
reduced coherence peaks in the operated hemisphere, both at rest and during the task.
24
541
If there was no substantial hippocampal contribution, removal of the hippocampus
542
should have little impact.
543
544
Patient P4 indeed showed a significant reduction in delta/theta coherence in the left
545
(operated) hemisphere at rest relative to the right (not operated) hemisphere
546
(resampling p-value <.025). Importantly, however, this patient showed no difference
547
between hemispheres during the decision-stage of the task, where a similar coherence
548
peak was observed in both hemispheres (figures 3A and B). For patient P3, who had
549
undergone right temporal lobe resection, we found significantly stronger coherence in
550
the operated hemisphere during the task (resampling p-value <.025) and no difference
551
at rest. In addition, patient P2, who had not undergone any surgery of the medial
552
temporal lobe, also showed significantly stronger coherence (at rest) in the right vs. left
553
hemisphere, similarly to P4 and P3.
554
555
556
557
25
558
Discussion
559
We demonstrate oscillatory coupling between local field potentials in the human
560
nucleus accumbens and surface EEG at delta and low theta frequencies, both at rest and
561
during a decision-making task. During the action selection stage of the task, this
562
coupling consists of a cortical drive of ~2 Hz oscillations in the nucleus accumbens,
563
accompanied by an increase in cortico-accumbens coherence.
564
565
Low-frequency oscillations and nucleus accumbens network function
566
Low-frequency coupling of local field potentials and cross-correlation of single units in
567
the hippocampus and the nucleus accumbens are well established in rodents (Tabuchi
568
et al., 2000; Berke et al., 2004; Gruber et al., 2009a; van der Meer and Redish, 2011).
569
Entrainment of nucleus accumbens neurons to the hippocampal theta rhythm is thought
570
to convey contextual (spatial) information (Tabuchi et al., 2000; van der Meer and
571
Redish, 2011), possibly constraining afferent drive from prefrontal cortex during action
572
selection (Grace, 2000; Goto and Grace, 2008). Conversely, recent studies show that
573
prefrontal input to the nucleus accumbens can suppress excitatory drive from the
574
hippocampus. Specifically, Calhoon & O’Donnell (2013) reported that 50 Hz train
575
stimulation of medial prefrontal cortex attenuated excitatory post-synaptic potentials in
576
the nucleus accumbens that were elicited via hippocampal projections. Because train
577
stimulation is thought to mimic burst firing of prefrontal neurons during instrumental
578
behaviour (Peters et al., 2005), the authors argued that naturally occurring afferent
579
input from prefrontal cortex, specifically during instrumental action, may also suppress
580
hippocampal drive, possibly via inhibitory interneurons. During instrumental
26
581
behaviour, local field potentials in the nucleus accumbens core and prefrontal cortex are
582
coupled at delta frequencies (1-4 Hz), accompanied by a decrease in theta coupling of
583
the nucleus accumbens with the hippocampus (Gruber et al., 2009a). Taken together,
584
this suggests that cortico-accumbens coupling at delta frequencies may be involved in
585
aspects of instrumental action, at least in rodents. Translation of these findings to
586
human physiology has so far been speculative because electrophysiological data from
587
the human nucleus accumbens is rare.
588
589
Here, we demonstrate cortico-accumbens coupling at low frequencies in humans. In
590
particular, we find a cortical drive of 2-3 Hz oscillations in the nucleus accumbens
591
during the action selection stage of a standard decision-making task, accompanied by an
592
increase in 2 Hz coherence between cortex and the nucleus accumbens. The delay
593
between cortex and the nucleus accumbens was estimated to be ~38 ms from a phase-
594
frequency regression of the resting-state data of one patient (which allowed for
595
sufficiently long epochs, see methods). This delay is longer than the (monosynaptic)
596
conduction delays between frontal cortex and the nucleus accumbens reported in rats
597
(≤20 ms; McGinty & Grace, 2009). One possible reason for this is that cortico-accumbens
598
coherence may involve an additional step of synaptic transmission. A plausible
599
candidate for mediating cortico-accumbens coherence is the pool of inhibitory
600
interneurons in the nucleus accumbens. Calhoon & O’Donnell (2013) reported evidence
601
for an involvement of GABAergic neurotransmission in the heterosynaptic suppression
602
of hippocampal input by prefrontal stimulation in rats, consistent with reports that
603
prefrontal stimulation activates nucleus accumbens interneurons (Gruber et al., 2009b).
604
Interneurons account only for a minority of striatal neurons (Jiang and North, 1991),
27
605
and their direct contribution to local field potentials (and, thus, to the observed cortico-
606
accumbens coherence) is therefore likely to be small. Taken together, intermediate
607
transmission via interneurons would agree with findings in rodents and could account
608
for the delay between surface EEG and LFP observed here. However, we note that our
609
estimation of the conduction delay should be treated with caution since it is based on
610
data of a single patient. Furthermore, previous studies have shown that delays
611
computed on the basis of phase differences may not correspond well to conduction
612
delays (Williams and Baker, 2009).
613
614
A role of low-frequency oscillations in the nucleus accumbens, in the range of theta
615
frequencies, in human decision-making has been proposed before. Cohen et al. (2009)
616
reported a transient increase in phase-synchronization at ~4-8 Hz approximately 250
617
to 550 ms after the outcome in a reversal learning task, both between the left and right
618
nucleus accumbens and between the nucleus accumbens and a frontal EEG electrode. In
619
contrast to this transient phenomenon (~1 cycle), we report cortico-accumbens
620
connectivity over much longer segments (1.5 to 2 seconds) and at lower frequencies (2-
621
5 Hz) which, importantly, changes in strength and direction during action selection. The
622
cortical drive consistently observed during action selection in our study is in agreement
623
with a previous report of a Granger-causal influence of cortex on the nucleus accumbens
624
in humans (Cohen et al., 2012). However, that study focused on connectivity during
625
outcome anticipation in a task that had no action selection component, and found
626
connectivity across a broad frequency range (of up to 20 Hz, possibly due to a limited
627
frequency resolution, as the authors note). In contrast, we demonstrate an increase in
28
628
low-frequency coupling and a consistent cortical drive during the action selection stage
629
of our decision-making task.
630
631
Cortical sources involved in low-frequency coupling with the nucleus accumbens
632
We studied connectivity between surface EEG recorded at electrode Cz and LFP from
633
the nucleus accumbens. Given that surface EEG signals are subject to a high degree of
634
volume conduction and distortion of the electric field by the skull, EEG signals at central
635
sensors such as Cz likely contain components from widely distributed cortical sources.
636
We addressed the possibility of a substantial contribution of hippocampal signals,
637
recorded at surface electrodes, to the observed EEG/LFP connectivity in two patients
638
who had undergone unilateral resection of the medial temporal lobe in the past. Using
639
anterograde and retrograde tracer techniques, Friedman et al. (2002) demonstrated
640
that anatomical connections between the hippocampus and the nucleus accumbens in
641
macaque monkeys do not cross to the contralateral hemisphere. We are not aware of
642
any anatomical studies in humans that have addressed this issue. Assuming that these
643
results also hold in humans, we tested whether ipsilateral removal of the hippocampus
644
systematically reduced the observed EEG/LFP coherence. Since this was not the case, a
645
substantial hippocampal contribution seems unlikely. It is possible, however, that the
646
nucleus accumbens of the operated hemisphere is indirectly coupled with the
647
contralateral hippocampus, e.g. via coupling between the nucleus accumbens of each
648
hemisphere (Cohen et al., 2009). Note that a neocortical contribution to the observed
649
EEG/LFP connectivity does not imply that a single cortical source is consistently
650
involved. Indeed, we find differences in coupling between our resting-state data and the
651
task data, both in peak frequencies (which are slightly lower during the task) and in
29
652
direction (with a consistent cortical drive during the task, but not at rest). This may
653
indicate that slightly different cortical sources or neuronal populations are involved in
654
cortico-accumbens coupling as a function of the task. Future studies that combine high-
655
density EEG or MEG with simultaneous DBS-recordings may reveal the precise cortical
656
topography of this coupling.
657
Limitations
658
Invasive recordings from the human nucleus accumbens are only available when there
659
is a clinical indication for DBS therapy. Here, we studied data recorded from patients
660
with epilepsy. All patients underwent video-EEG-monitoring, which documented a focal,
661
i.e., cortical, epileptogenic zone and excluded other epilepsy syndromes. In animal
662
models (Deransart et al., 2000) as well as in epilepsy patients (Schmitt et al., 2014), the
663
nucleus accumbens can be involved in seizure propagation. Accordingly, the clinical
664
rationale underlying DBS of the nucleus accumbens in epilepsy is a suppression of
665
seizure propagation, not of an epileptogenic focus. However, we cannot exclude the
666
possibility that the nucleus accumbens functions differently in these patients as
667
compared to healthy individuals. During recordings, patients were constantly under
668
supervision. No seizure was observed or reported in any case. In addition, no inter-ictal
669
epileptic activity was observed in the recorded data. Epileptic activity in the nucleus
670
accumbens as a potential confounding factor can therefore be excluded.
671
672
It is important to note that all six patients were on treatment with anticonvulsant drugs
673
during data acquisition (table 1). The predominant mechanisms of action of these drugs
674
consist either in a blockade of ion channels (LCM, LTG, ZNS, CBZ, OXC) and/or in an
30
675
interference with GABAergic neurotransmission (STP, Clozabam). While we cannot rule
676
out the possibility that these drugs changed cortico-accumbens connectivity, we note
677
that the heterogeneity in medication across patients (table 1) contrasts with the high
678
consistency of our findings. In particular, patients whose medication involved drugs
679
known to enhance or mimic GABAergic neurotransmission (especially Clobazam and
680
STP in P2, possibly also CBZ in P2 and P3) showed similar results as all other patients,
681
whose medication acts primarily via ion channels (LCM and LTG in patients P4, P5 and
682
P6; but see Cunningham & Jones, 2000, for a potential GABAergic enhancement by LTG,
683
although this has been questioned, see Braga, Anderjaska, Post, & Li, 2002).
684
685
Our study depends on a precise placement of DBS electrodes into the nucleus
686
accumbens. Despite high surgical expertise, we cannot fully exclude slight imprecision
687
in electrode placement. Since safety considerations forbid post-surgical MRI, the
688
location of the electrodes can only be assessed by combining pre-surgical MRI with
689
post-surgical computer tomography. Following this approach, coordinates of the most
690
ventral contacts in our study, averaged across all patients, are [7 6 -9] (right
691
hemisphere) and [-7 6 -9] (left hemisphere; normalized to MNI space). However, this
692
analysis is subject to some inaccuracy itself, since it relies on two separate imaging
693
techniques and data from two different days.
694
695
We cannot exclude the possibility that discrete structural changes to the nucleus
696
accumbens during electrode implantation, such as trauma or oedema, may have
697
influenced coherence and Granger causality spectra. Structural changes due to electrode
31
698
implantation are thought to be responsible for the so-called ‘stun effect’, i.e., a transient
699
amelioration of symptoms in some patients with Parkinson’s Disease in the first days
700
after DBS surgery, before the stimulator is switched on (e.g., Eusebio and Brown, 2009).
701
However, it is difficult to explain a frequency-specific task modulation, as demonstrated
702
in our study (figure 2), purely on the basis of such structural alterations.
703
704
In summary, we demonstrate highly consistent, oscillatory coupling between cortex and
705
the human nucleus accumbens at delta and low theta frequencies. During action
706
selection, this coupling turns into a cortical drive of delta oscillations in the nucleus
707
accumbens. In view of a similar coupling in rodents during instrumental behaviour, and
708
in light of the high consistency of our findings across subjects, cortico-accumbens delta
709
coupling may represent a highly conserved mechanism of regulating neuronal
710
excitability and, ultimately, action selection in the nucleus accumbens.
711
32
712
References
713
714
Berke JD, Okatan M, Skurski J, Eichenbaum HB. Oscillatory entrainment of striatal
neurons in freely moving rats. Neuron 43: 883–96, 2004.
715
716
717
718
Braga MFM, Anderjaska VA, Post RM, Li H. Lamotrigine reduces spontaneous and
evoked GABA A receptor-mediated synaptic transmission in the basolateral amygdala:
implications for its effects in seizure and affective disorders. Neuropharmacology 42:
522–529, 2002.
719
720
Calhoon GG, O’Donnell P. Closing the gate in the limbic striatum: prefrontal
suppression of hippocampal and thalamic inputs. Neuron 78: 181–90, 2013.
721
722
723
Cohen MX, Axmacher N, Lenartz D, Elger CE, Sturm V, Schlaepfer TE. Nuclei
accumbens phase synchrony predicts decision-making reversals following negative
feedback. J Neurosci Off J Soc Neurosci 29: 7591–8, 2009.
724
725
726
727
Cohen MX, Bour L, Mantione M, Figee M, Vink M, Tijssen M a J, van Rootselaar A-F,
van den Munckhof P, Schuurman PR, Denys D. Top-down-directed synchrony from
medial frontal cortex to nucleus accumbens during reward anticipation. Hum Brain
Mapp 33: 246–52, 2012.
728
729
730
731
Cunningham MO, Jones RS. The anticonvulsant, lamotrigine decreases spontaneous
glutamate release but increases spontaneous GABA release in the rat entorhinal cortex
in
vitro.
[Online].
Neuropharmacology
39:
2139–46,
2000.
http://www.ncbi.nlm.nih.gov/pubmed/10963757.
732
733
734
Deransart C, Riban V, Le B-T, Marescaus C, Depaulis A. Dopamine in the striatum
modulates seizures in a genetic model of absence epilepsy in the rat. Neursocience 100:
335–344, 2000.
735
736
Dhamala M, Rangarajan G, Ding M. Analyzing information flow in brain networks with
nonparametric Granger causality. Neuroimage 41: 354–62, 2008.
737
738
Eusebio A, Brown P. Synchronisation in the beta frequency-band - The bad boy of
parkinsonism or an innocent bystander? Exp Neurol 217: 1–3, 2009.
739
740
741
Friedman DP, Aggleton JP, Saunders RC. Comparison of hippocampal, amygdala, and
perirhinal projections to the nucleus accumbens: combined anterograde and retrograde
tracing study in the Macaque brain. J Comp Neurol 450: 345–65, 2002.
742
743
Goto Y, Grace A a. Limbic and cortical information processing in the nucleus
accumbens. Trends Neurosci 31: 552–8, 2008.
744
745
746
Grace AA. Gating of information flow within the limbic system and the pathophysiology
of
schizophrenia.
[Online].
Brain
Res
Rev
31:
330–41,
2000.
http://www.ncbi.nlm.nih.gov/pubmed/10719160.
33
747
748
Gruber AJ, Hussain RJ, O’Donnell P. The nucleus accumbens: a switchboard for goaldirected behaviors. PLoS One 4: e5062, 2009a.
749
750
Gruber AJ, Powell EM, O’Donnell P. Cortically activated interneurons shape spatial
aspects of cortico-accumbens processing. J Neurophysiol 101: 1876–82, 2009b.
751
752
753
Hart AS, Rutledge RB, Glimcher PW, Phillips PEM. Phasic dopamine release in the rat
nucleus accumbens symmetrically encodes a reward prediction error term. J Neurosci
34: 698–704, 2014.
754
755
Haufe S, Nikulin V V, Müller K-R, Nolte G. A critical assessment of connectivity
measures for EEG data: a simulation study. Neuroimage 64: 120–33, 2013.
756
757
758
759
Haufe S, Nikulin V V, Nolte G. Alleviating the Influence of Weak Data Asymmetries on
Granger-Causal Analyses. In: Latent Variable Analysis and Signal Separation: Lecture
Notes in Computer Science, edited by Theis F, Cichocki A, Yeredor A, Zibulevsky M. 2012,
p. 25–33.
760
761
Jiang BYZ, North RA. Membrane Properties and Synaptic Responses of Rat Striatal
Neurones in vitro. J Physiol 443: 533–553, 1991.
762
763
764
Kaplan R, Doeller CF, Barnes GR, Litvak V, Düzel E, Bandettini P a, Burgess N.
Movement-related theta rhythm in humans: coordinating self-directed hippocampal
learning. PLoS Biol 10: e1001267, 2012.
765
766
767
768
769
Litvak V, Eusebio A, Jha A, Oostenveld R, Barnes G, Foltynie T, Limousin P, Zrinzo
L, Hariz MI, Friston K, Brown P. Movement-related changes in local and long-range
synchronization
in
Parkinson’s
disease
revealed
by
simultaneous
magnetoencephalography and intracranial recordings. J Neurosci Off J Soc Neurosci 32:
10541–53, 2012.
770
771
772
McGinty VB, Grace A a. Timing-dependent regulation of evoked spiking in nucleus
accumbens neurons by integration of limbic and prefrontal cortical inputs. J
Neurophysiol 101: 1823–35, 2009.
773
774
Van der Meer M a a, Redish a D. Theta phase precession in rat ventral striatum links
place and reward information. J Neurosci 31: 2843–54, 2011.
775
776
Nicola SM. The nucleus accumbens as part of a basal ganglia action selection circuit.
Psychopharmacology (Berl) 191: 521–50, 2007.
777
778
779
Nolte G, Ziehe A, Nikulin V V, Schlögel A, Krämer N, Brismar T, Müller K-R. Robustly
estimating the flow direction of information in complex physical systems. Phys Rev Lett
100, 2008.
780
781
782
Oostenveld R, Fries P, Maris E, Schoffelen J-M. FieldTrip: Open source software for
advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell
Neurosci 2011: 156869, 2011.
34
783
784
785
Patel SR, Sheth S a, Mian MK, Gale JT, Greenberg BD, Dougherty DD, Eskandar EN.
Single-neuron responses in the human nucleus accumbens during a financial decisionmaking task. J Neurosci 32: 7311–5, 2012.
786
787
788
Pennartz CM a, Berke JD, Graybiel AM, Ito R, Lansink CS, van der Meer M, Redish a
D, Smith KS, Voorn P. Corticostriatal Interactions during Learning, Memory Processing,
and Decision Making. J Neurosci 29: 12831–8, 2009.
789
790
Penny WD, Friston KJ, Ashburner JT, Kiebel SJ, Nichols TE, editors. Statistical
Parametric Mapping: The Analysis of Functional Brain Images. Academic Press, 2011.
791
792
793
Peters YM, O’Donnell P, Carelli RM. Prefrontal cortical cell firing during maintenance,
extinction, and reinstatement of goal-directed behavior for natural reward. Synapse 56:
74–83, 2005.
794
795
Riddle CN, Baker SN. Manipulation of peripheral neural feedback loops alters human
corticomuscular coherence. J Physiol 566: 625–39, 2005.
796
797
798
Rosenberg JR, Amjad AM, Breeze P, Brillinger DR, Halliday DM. The Fourier
Approach to the Identification of Functional Coupling between Neuronal Spike Trains.
Prog Biophys molec Biol 53: 1–31, 1989.
799
800
801
Rutledge RB, Skandali N, Dayan P, Dolan RJ. A computational and neural model of
momentary subjective well-being. Proc. Natl. Acad. Sci. ( August 4, 2014). doi:
10.1073/pnas.1407535111.
802
803
804
Schmitt FC, Voges J, Heinze H-J, Zaehle T, Holtkamp M, Kowski AB. Safety and
feasibility of nucleus accumbens stimulation in five patients with epilepsy. J Neurol 261:
1477–84, 2014.
805
806
Schomer DL, Lopes da Silva F. Niedermeyer’s Electroencephalography: Basic Principles,
Clinical Applications, and Related Fields. 6th ed. Lippincott Williams & Wilkins, 2012.
807
808
809
Stam CJ, Nolte G, Daffertshofer A. Phase lag index: assessment of functional
connectivity from multi channel EEG and MEG with diminished bias from common
sources. Hum Brain Mapp 28: 1178–93, 2007.
810
Stuber GD. Cortical operation of the ventral striatal switchboard. Neuron 78: 6–7, 2013.
811
812
813
Sturm V, Lenartz D, Koulousakis A, Treuer H, Herholz K, Klein JC, Klosterkötter J.
The nucleus accumbens: a target for deep brain stimulation in obsessive–compulsiveand anxiety-disorders. J Chem Neuroanat 26: 293–299, 2003.
814
815
816
Tabuchi ET, Mulder a B, Wiener SI. Position and behavioral modulation of
synchronization of hippocampal and accumbens neuronal discharges in freely moving
rats. Hippocampus 10: 717–28, 2000.
817
818
Tom SM, Fox CR, Trepel C, Poldrack R a. The neural basis of loss aversion in decisionmaking under risk. Science 315: 515–8, 2007.
35
819
820
821
Vinck M, Oostenveld R, van Wingerden M, Battaglia F, Pennartz CM a. An improved
index of phase-synchronization for electrophysiological data in the presence of volumeconduction, noise and sample-size bias. Neuroimage 55: 1548–65, 2011.
822
823
824
825
Voges J, Volkmann J, Allert N, Lehrke R, Koulousakis A, Freund H-J, Sturm V.
Bilateral high-frequency stimulation in the subthalamic nucleus for the treatment of
Parkinson disease: correlation of therapeutic effect with anatomical electrode position. J
Neurosurg 96: 269–79, 2002.
826
827
828
Walter DO. Spectral Analysis for Electroencephalograms: Mathematical Determination
of Neurophysiological Relationships from Records of Limited Duration. Exp Neurol 8:
155–181, 1963.
829
830
831
Williams ER, Baker SN. Circuits generating corticomuscular coherence investigated
using a biophysically based computational model. I. Descending systems. J Neurophysiol
101: 31–41, 2009.
832
833
Wilson GT. The Factorization of Matricial Spectral Densities. SIAM J Appl Math 23: 420–
426, 1972.
834
835
836
Zaehle T, Bauch EM, Hinrichs H, Schmitt FC, Voges J, Heinze H-J, Bunzeck N.
Nucleus accumbens activity dissociates different forms of salience: evidence from
human intracranial recordings. J Neurosci 33: 8764–71, 2013.
837
36
838
Acknowledgements
839
We thank Christian Kluge and Dominik Bach for valuable suggestions regarding data
840
analysis and interpretation. This work was supported by the Wellcome Trust (Ray
841
Dolan Senior Investigator Award 098362/Z/12/Z). MPS was supported by a scholarship
842
from the German Research Foundation (Deutsche Forschungsgemeinschaft DFG, STE
843
2091/1-2). RBR was supported by the Max Planck Society. TZ, JV and HJH received
844
funding from the German Research Foundation (Deutsche Forschungsgemeinschaft,
845
Sonderforschungsbereich
846
Neuroimaging is supported by core funding from the Wellcome Trust 091593/Z/10/Z.
847
The authors declare no competing financial interests.
SFB-779
TPA2).
37
The
Wellcome
Trust
Centre
for
848
Tables
849
850
851
852
853
Table 1. Patient data.
gender/age/
disease duration epilepsy
ID
(years)
syndrome
P1
854
855
856
857
M /41/12
etiology
seizure
lateralisation
seizure
onset
AED
surface EEG
(restingstate/ task)
multifocal
cryptogenic
bilateral
frontotemporal
LCM 400 mg
ZNS 400 mg
FCz, Cz/
Oz, Fpz, Cz
left, possibly
bilateral
bifrontal
and left
mediotemporal
OXC 900 mg
T1, T2, Fpz,
Clobazam 5 mg Oz, Cz/Oz, Fpz,
STP 4500 mg
Cz, POz
P2
M/32/31
multifocal
genetic
(SCNA1)
P3
F/44/30
multifocal
hippocampal
sclerosis*
bilateral
temporal
CBZ 1200 mg
Oz, Fpz, Cz/
Oz, Fpz, Cz
P4
M/40/31
focal
left
hippocampal
sclerosis†
left
temporal
LTG 400 mg
LCM 400 mg
Oz, Fpz, Cz/
Oz, Fpz, Cz
P5
F/28/12
multifocal
cryptogenic
bilateral
temporal
LTG 200 mg
LCM 200 mg
Fpz, Cz/
no task data
P6
F/52/17
focal
cryptogenic
left
temporal
LTG 250 mg
LCM 400 mg
no restingstate data/Oz,
Fpz, Cz
LCM – Lacosamide; ZNS – Zonisamide; OXC – Oxcarbazepine; STP – Stiripentol; CBZ –
Carbamazepine; LTG – Lamotrigine;
*underwent right temporal lobe resection 3 years before DBS surgery
†underwent left temporal lobe resection 9 years before DBS surgery
38
858
Figure legends
859
860
Figure 1 Coherence, phase-synchronization and Granger causality between cortex
861
(EEG surface electrode at Cz) and the nucleus accumbens in resting-state data.
862
Each row shows data for one of the five patients (P1 to P5). All x-axes represent
863
frequency in Hz. The light grey vertical shading indicates frequency bins at which the
864
original spectrum (black line) is significantly different from the trial-shuffled spectra (A
865
and B) or the time-reversed spectrum (C, dotted line). The dark grey shading in A and B
866
corresponds to two standard deviations around the mean of the trial-shuffled spectra.
867
Numbers next to peaks correspond to peak frequencies. A, Coherence and B, the
868
weighted phase-lag index (wPLI), a measure of (non-zero) phase-synchronization, for
869
all six patients. In C, the left column shows Granger causality spectra representing a
870
cortical drive of the nucleus accumbens, while the right column shows coupling in the
871
opposite direction (nucleus accumbens leading cortex). All patients show significant
872
peaks in the coherence- and wPLI spectra at low (delta and theta) frequencies (A and B).
873
The direction of cortico-accumbens coupling at rest varies across patients (C).
874
875
Figure 2 Coherence and Granger causality between cortex (Cz) and the nucleus
876
accumbens during the decision-making task. A, Schematic of one trial of the task
877
with distinct task stages (all intervals are equal to 1.5 seconds in duration). ITI, inter-
878
trial interval. B, Coherence as a function of frequency during all four stages of the task,
879
colour-coded as in A (black: baseline/ITI, red: decision stage, green: anticipation stage,
880
blue: outcome stage). Coherence is significantly enhanced during the decision stage as
39
881
compared to baseline in three patients (P2, P3 and P6). Numbers next to peaks
882
correspond to peak frequencies. Each row shows data for one of the five patients (P1 to
883
P6). C, Granger causality spectra for the decision stage of the task. As in figure 1, C, the
884
left column shows a Granger causal influence of cortex on the nucleus accumbens, while
885
the right column shows coupling in the opposite direction (nucleus accumbens leading
886
cortex). The dotted line shows Granger causality after reversing the time axis at all
887
channels. The grey vertical shading indicates frequency bins at which the original
888
spectrum (black line) is significantly greater than the spectrum computed after
889
reversing the time axis. Four of the five patients show a significant cortical drive of delta
890
oscillations in the nucleus accumbens (trend in P3).
891
892
Figure 3 Coherence between cortex (Cz) and the nucleus accumbens, separately
893
for each hemisphere, both at rest and during the decision stage of the task. In all
894
plots, the dotted line represents coherence spectra between cortex and the right
895
nucleus accumbens, and the solid line corresponds to coherence with the left nucleus
896
accumbens. A, Coherence at rest. B, Coherence during the decision stage of the task. The
897
boxes mark the two patients (P3 and P4) who had undergone resection of the right (P3)
898
and left (P4) medial temporal lobe in the past. Cortico-accumbens coherence is not
899
altered systematically by unilateral resection of the hippocampus (P3 and P4).
900
901
40
A
B
C
P1
AC
AC
P1
P1
P2
P2
4 Hz
P3
P3
5 Hz
P4
P4
3 Hz
P5
P5
4 Hz
2.5 Hz
5 Hz
Cz
3.5 Hz
P2
P4
4 Hz
P5
3 Hz
frequency [Hz]
Granger causality
P3
coherence
5 Hz
phase-synchronization (wPLI)
7.5 Hz
Cz
A
ITI
decision
B
C
anticipation
Cz
outcome
AC AC
Cz
P1
P1
0.3
1.5 Hz
0.1
5
10
frequency
15
P2
2 Hz
P2
0.3
2.5 Hz
5.5 Hz
coherence
5
10
frequency
2 Hz
0.3
15
P3
0.1
5
10
frequency
2 Hz
0.3
15
P4
Granger causality
0.1
P3
(2 Hz)
P4
2.5 Hz
0.1
5
10
frequency
2 Hz
0.3
15
P6
P6
2 Hz
0.1
5
10
frequency
15
frequency [Hz]
0.3
P2
15
P3
5
10
frequency
15
15
P4
0.3
0.1
15
frequency [Hz]
5
10
frequency
0.5
15
P6
0.3
0.1
5
10
frequency
P5
0.1
5
10
frequency
0.5
coherence
coherence
coherence
15
0.5
0.3
0.1
0.3
0.1
5
10
frequency
0.5
P4
0.3
5
10
frequency
15
0.3
0.1
0.5
0.1
5
10
frequency
15
5
10
frequency
15
decision
stage (task)
P1
5
10
frequency
0.5
P3
0.1
coherence
15
0.5
0.3
0.1
5
10
frequency
coherence
coherence
P2
0.3
0.1
0.5
coherence
0.3
0.5
coherence
coherence
coherence
P1
left temporal
lobe resection
resting-state
0.5
B
right temporal
lobe resection
right nucleus accumbens
coherence
left nucleus accumbens
coherence
A