Functional neural dynamics underlying auditory event-related N1 and N1 suppression response.

Abstract

Presenting tone triplets of identical stimuli preceded by silent intervals of 30 s produces a series of three N1 averaged event-related potentials (ERPs), the first being of greater amplitude (non-suppressed N1) than the second and third ones (suppressed N1). Maximal statistically independent components (ICs) of single-trial multi-electrode scalp EEG responses to triplets were obtained by ICA algorithm, and then each IC was searched for underlying brain structures by LORETA inverse solution, and for oscillatory contributions by time-frequency analysis. Non-suppressed N1 cortical mechanisms were broken down into five ICs, grouped in two time-windows (early-onset and late-onset) involving the participation of temporal, frontal and parietal structures, and sub-serving EEG oscillatory contributions of power enhancement and putative phase concentration of mainly theta, alpha and low beta bands. Suppressed N1 was due to the modulation of two above-mentioned early-onset ICs, involving temporal structures only, and mainly sub-serving oscillatory contributions of phase concentration of theta and alpha. Present results, showing quantifiable changes of IC descriptors - i.e. time window of activation, implied structures and oscillatory contributions - extracted from two distinct brain functional situations (non-suppressed versus suppressed N1), give support to the view that ICA is not merely a statistical "latent variables" model when applied to ERPs, but could help to capture underlying specific function subunits of brain dynamics.

Full text

Page 1

This article was published in an Elsevier journal. The attached copy
is furnished to the author for non-commercial research and
education use, including for instruction at the author’s institution,
sharing with colleagues and providing to institution administration.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Page 2

Author's personal copy
suggested that large-scale interaction of these neural contributions
Functional neural dynamics underlying auditory event-related
N1 and N1 suppression response
C. Grau,a,⁎Ll. Fuentemilla,aand J. Marco-Pallarésb
aNeurodynamics Laboratory, Department of Psychiatry and Clinical Psychobiology, Pg Vall d’Hebrón, 171, 08035, University of Barcelona, Catalonia, Spain
bOtto-Von-Guericke-Universität Magdeburg, Institut Für Psychologie II, Germany
Received 4 December 2006; revised 7 February 2007; accepted 13 March 2007
Available online 30 March 2007
Presenting tone triplets of identical stimuli preceded by silent intervals
of 30 s produces a series of three N1 averaged event-related potentials
(ERPs), the first being of greater amplitude (non-suppressed N1) than
the second and third ones (suppressed N1). Maximal statistically
independent components (ICs) of single-trial multi-electrode scalp
EEG responses to triplets were obtained by ICA algorithm, and then
each IC was searched for underlying brain structures by LORETA
inverse solution, and for oscillatory contributions by time–frequency
analysis. Non-suppressed N1 cortical mechanisms were broken down
into five ICs, grouped in two time-windows (early-onset and late-
onset) involving the participation of temporal, frontal and parietal
structures, and sub-serving EEG oscillatory contributions of power
enhancement and putative phase concentration of mainly theta, alpha
and low beta bands. Suppressed N1 was due to the modulation of two
above-mentioned early-onset ICs, involving temporal structures only,
and mainly sub-serving oscillatory contributions of phase concentra-
tion of theta and alpha. Present results, showing quantifiable changes
of IC descriptors – i.e. time window of activation, implied structures
and oscillatory contributions – extracted from two distinct brain
functional situations (non-suppressed versus suppressed N1), give
support to the view that ICA is not merely a statistical “latent
variables” model when applied to ERPs, but could help to capture
underlying specific function subunits of brain dynamics.
© 2007 Elsevier Inc. All rights reserved.
Introduction
Brain electrical responses time-locked to events (stimulus,
response or behavioural task) have been studied for decades by
averaging the event-locked single trials to obtain event-related
potentials (ERPs). Although averaged ERPs resemble basic waves,
they undergo complex neural dynamic activity that is likely to arise
from multiple temporarily and spatially distributed specialized
brain sources (Bressler and Kelso, 2001). Hence, it has been
to ERPs gives rise to their functional integration, which might lead
to support of specific function activities (Friston, 2005). In
addition, the relevance of oscillatory activity as a framework for
the functional integration of a distributed neural activity has been
pointed out (Fries, 2005). Indeed, converging recent evidence
showed that a distributed functionally conformed network of
neurons might be transiently linked by reciprocal dynamic
connections (Varela et al., 2001), thus resulting in multiple
feedback loops that give rise to oscillatory neural-based behaviour
(Bressler, 2002). Therefore, based on these arguments, underlying
neural functional properties given by an ERP are unlikely to be
captured solely by studying the representation of the temporal
course and scalp features of the averaged electrical activity, but
might require identification of their related temporal evolution,
cortical topography and oscillatory behaviour (Fogelson et al.,
2006), which can only be captured at the single-trial level.
The concept of ERPs as being composed of orchestrated
clusters of distributed neural contributions, which transiently
integrate to support specific functional activities, has been recently
enlarged on the basis of ICA (Independent Component Analysis)
results on multi-electrode EEG data (Makeig et al., 2002; Penny et
al., 2002). Accordingly, ERPs could be broken down into
maximal statistically independent contributions (independent
components, ICs), which might be characterised by specific
temporo-structural (Makeig et al., 2002; Marco-Pallarés et al.,
2005) and frequency domain features (Makeig et al., 2004a). This
leads to the proposition that the properties of maximal statistical
independence and the multifaceted physiological characterization
(in the time-, space- and frequency-domains) of each IC strongly
suggest that they might also pick up specific functional brain
subunits. However, at present, there is no direct evidence that
neural activity identified by an IC intrinsically modulate to
functional experimental manipulations, though this would,
unequivocally, transfer meaningful functional entity to the given
neural activity.
The first objective of this study was to provide evidence that
separate specific functional activity of brain computing captured by
ICs and their descriptors (i.e. time–structures–frequency) show
quantifiable changes in distinct functional situations. To this end,
www.elsevier.com/locate/ynimg
NeuroImage 36 (2007) 522–531
⁎Corresponding author. Fax: +34 93 403 44 24.
E-mail address: carlesgrau@ub.edu (C. Grau).
Available online on ScienceDirect (www.sciencedirect.com).
1053-8119/$ - see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.neuroimage.2007.03.027

Page 3

Author's personal copy
we selected auditory N1, which in front of identical external
repeated stimulation shows reduced amplitude (attenuation or
repetitive suppression) (Näätänen, 1992; Friston, 2005). This
reduced amplitude is associated with the prevention of the inflow
of redundant information from the environment into daily life
(Näätänen, 1992). The advantage of this experimental paradigm
resides in that, as the stimuli of the sequence are physically
identical, functional changes sub-serving N1 amplitude reduction
other than the ones experienced by its intrinsic neural substrate can
be discarded.
The second objective of this work is to advance on the
characterization of functional neural dynamics underlying auditory
event-related N1 and N1 suppression response. Auditory N1 ERP
involves a complex neural network of several dynamic properties
described by temporal, spatial and spectral contributions. Hence, in
the time domain, scalp electromagnetic (MEG) (Loveless et al.,
1996; McEvoy et al., 1997) studies showed distinct sequential N1
contribution activations, referred to as “early” (∼80 ms) and “late”
N1 (∼130 ms) (Näätänen, 1992). Further, in the spatial domain,
N1 are sub-served by a multi-generator process (Naatanen and
Picton, 1987), involving temporal (Hari et al., 1980), frontal
(Knight et al., 1980), cingulate (Tzourio et al., 1997) and parietal
regions (Knight et al., 1980). And, in the frequency domain, the
study of the underlying auditory N1 neural oscillatory activity
showed the involvement of a multi-frequency phenomenon
enclosing enhanced spectral power and phase-resetting of theta
and alpha frequency bands at scalp EEG (Jansen et al., 2003;
Fuentemilla et al., 2006). In addition, it has been argued that
suppressed N1 might be due to the functional modulation of basic
N1 processes, such as distinct EEG recovery cycles of N1 “early”
and “late” contributions (Loveless et al., 1996), to decreased
auditory cortical neuron activity (Rosburg et al., 2006) or to a
reduction in scalp theta and alpha phase synchrony together
with a disappearance of a concomitant spectral power modula-
tion (Fuentemilla et al., 2006). However, to date, no integrated
approach has made a comparative study of auditory N1 and N1
suppressed addressing the behaviour of time–structure–frequency
ICdescriptorsrelatingtoN1andN1repetitivesuppression,thoughit
could provide new insights in the study of N1 functional neural
dynamics.
Here, we tackle these objectives by applying a 3-step approach
analysis. The first two steps were based on the combined
Independent Component Analysis (ICA) (Bell and Sejnowski,
1995; Jung et al., 2001)+Low-Resolution Tomography (LORETA)
(Pascual-Marqui et al., 1994; Pascual-Marqui, 1999) proposed by
Marco-Pallarés et al. (2005). ICA has been used to “blindly”
identify neural modes describing activity in concurrent electro-
magnetic activity that is spatially fixed and temporally independent
(Makeig et al., 1996), while LORETA analysis of the spatial maps
associated with each IC provided solutions to the inverse problem
of the neural source location of each component (Marco-Pallarés
et al., 2005). In addition, the spectral content of each IC will be
studied by computing a single-trial time–frequency analysis
(Makeig et al., 2004b).
Material and methods
Subjects
16 right-handed healthy subjects (7 female), 25.4±1.6 years
(range, 20–28 years), after complete description of the work, gave
their written consent to participate in the study. All subjects had no
history of head injury, neurological disease, audiological problems,
severe medical illness or drug abuse. The experiment complied
with the Code of Ethics of the World Medical Association
(Declaration of Helsinki) and was approved by the Ethics
Committee of the University of Barcelona.
Stimuli and procedure
Stimuli paradigm consisted of trains of three pure sine-wave
tones (1000 Hz, 90 dB, 10 ms rise/fall) where, first two tones (S1
and S2) were standard sounds (75 ms duration) and third tone was
randomly (P=0.5) a standard (S3) or deviant sound (Dev) (25 ms
duration). Intra-train interval was 584 ms and inter-train interval
was 30 s. Only those stimuli trains containing three standard tones
were analyzed in this study.
Participants sat in a comfortable chair in a dimly lit and
electrically and acoustically shielded booth. During the EEG
recording, each subject was instructed to perform a visual
irrelevant task (reading) to ignore the auditory stimuli and to
avoid extra eye movements and blinking.
EEG recording
EEG activity (0.1–70 Hz bandpass; 50 Hz notch filtered)
was acquired by a 32-channel amplifier (Synamps, Neuroscan
Inc.) at a sampling rate of 500 Hz. EEG electrodes followed the
10–20 position system (FP1, Oz, FP2, F7, F3, Fz, F4, F8, T3,
C3, Cz, C4, T4, T5, P3, Pz, P4, T6) with ten additional
electrodes (FC1, FC2, FT3, FT4, M1, M2, IM1, IM2, TP3,
TP4, CP1 and CP2) referenced to the tip of the nose.
Impedances were maintained below 5 kΩ during experiment.
Ocular movements (EOG) were registered from two electrodes
at the outer canthus of each eye. A single ground electrode was
attached at AFz.
Event-related potential data analysis
3000-ms EEG epochs were averaged off-line, which included
1000-ms pre-stimulus as a baseline period and S1, S2 and S3
standard tone responses. Exceeding ±100 μV in EEG and/or EOG
epochs were excluded automatically. No digital off-line band-pass
filter was used for ERP extraction. A total of 823 trials were
obtained for the analysis, in which each subject contributed with
69±14 trials (mean±SEM).
Auditory N1 was identified as the largest negative contribution
between the 80 and 180 ms time window of the ERPs to the three
standard sounds. In each case, the baseline-to-peak value was taken
as the magnitude (μV) of the response.
N1 statistical analysis (one-way analysis of variance (ANOVA))
was carried out on mean N1 amplitude measurements at mid-
line frontal (Fz), central (Cz) and parietal (Pz) electrode
locations plus left (C3) and right (C4) central electrodes with
respect to the pre-stimuli baseline (1000 ms). N1 suppression
was assessed by the repeated-measures ANOVA calculation to
N1 amplitudes after S1, S2 and S3, in which the factors
Electrode (Fz, C3, Cz, C4 and Pz)×Condition (N1 to S1, S2
or S3) to the N1 peak values were compared. Greenhouse–
Geisser correction to P values was used when appropriate. Post-
hoc comparisons by Student’s t-test for paired samples were
made to N1 peak values (S1, S2 and S3) within each elec-
523 C. Grau et al. / NeuroImage 36 (2007) 522–531

Page 4

Author's personal copy
trode when the within-subject Condition factor was statistically
significant.
Independent component analysis tomography imaging
A ICA-LORETA analysis was applied to the data as proposed
in Marco-Pallarés et al. (2005). To that, firstly, data was demeaned
and whitened and then the Infomax algorithm was used in
independent component analysis of concatenated single-trial data
from all subjects (Makeig et al., 1996) and, separately, to each
single-subject trials. IC extraction was performed using the
EEGLAB (v4.5) software under Matlab 7.0, as described by
Delorme and Makeig (2004), to EEG epochs of −100 to 200 ms of
each S1, S2 and S3 stimuli. An IC was considered responsible for
N1 generation following these constraints: (a) it had a sequence of
at least 12 consecutive sampling points (24 ms) (Guthrie and
Buchwald, 1991; Grau et al., 1998; Marco-Pallarés et al., 2005)
that exceed the threshold cut-off values after computing bootstrap
distributions (P<0.05), extracted randomly from baseline data and
applied 200 times were used, and (b) the activated window
(P<0.05) coincided with N1 ERP latency range (80–180 ms) of
any of the three standard sounds (S1, S2 or S3) responses.
Second, a LORETA analysis was performed with the scalp
maps associated with selected ICs that passed restrictions (a) and
(b). Only the values greater than 2.5 times the standard deviation of
the standardized data (in the LORETA spatial solution) were
accepted as being activated (Marco-Pallarés et al., 2005; Gomez
et al., 2006).
Event-Related Spectral Perturbations (ERSPs) and Inter-Trial
Coherence (ITC)
In a third step of the analysis, for each IC extracted previously,
a trial-by-trial time–frequency analysis was computed (1000 ms
before S1 to 2000 ms after) by using a sinusoidal moving Hanning-
windowed wavelet with linear increased cycles, from 2 cycles for
the lowest frequency (3.9 Hz) to 13 cycles to the highest frequency
(48.8 Hz) analysed.
Changes in event-related spectral power response were
computed by the Event-Related Spectral Perturbation (ERSP)
index (Makeig, 1993) (1),
ERSPðf ;tÞ ¼1
n
X
n
k¼1
ðFkð f ;tÞÞ2
ð1Þ
where, for n trials, Fk( f, t) is the spectral estimate of trial k at
frequency f and time t. ERSP showed mean time–frequency values
that exceeds significant cut-off threshold extracted from pre-
stimulus baseline period.
Event-locked EEG partial phase coherence was computed by
the Inter-Trial Coherence index (ITC) (2), analogous to the “phase-
locking index” (Tallon-Baudry et al., 1996),
?????
where || represents the complex norm. ITC values ranges from 0 to
1, wherein values near to 1 implies almost perfect phase
coincidence across trials. To determine the threshold signification
of both ERSP and ITC, bootstrap distributions (P<0.01), extracted
randomly from baseline data and applied 200 times were used
(Makeig et al., 2002). ERSP and ITC analysis was computed by
using EEGLAB v4.5 (Delorme and Makeig, 2004) under Matlab 7
(The Mathworks, Inc.).
ITCðf ;tÞ ¼1
n
X
n
k¼1
Fkð f ;tÞ
jFkð f ;tÞj
?????
ð2Þ
Results
Event-related potentials
All tones (S1, S2 and S3) elicited significant N1 ERP to all
electrode locations studied (P<0.001). 4 subjects were excluded for
the analysis due to technical problems on 4 electrodes (CP1, TP3,
CP3andTP4).Fig.1depictsthegrandaverageofERPwaveformsto
each stimulus across 12 subjects at Fz, C3, Cz, C4 and Pz electrodes
and the scalp topography of the N1 peak amplitude to each stimuli.
N1 attenuation was measured by N1 decreased amplitude across
Fig. 1. (a) All subjects' EEG epoch (−1000 to 2000 ms) grand average epoch in response to S1 (red line), S2 (green line) and S3 (blue line) stimuli at Cz. Grey
boxes at N1 responses correspond to the EEG time windows computed by ICA. (b) Grand average of Fz, C3, C4 and Pz after S1, S2 and S3 stimuli. (c) Scalp
voltage plot of the grand average response after S1, S2 and S3 at N1 peak (identified at N1 vertex at Cz electrode). (For interpretation of the references to colour
in this figure legend, the reader is referred to the web version of this article.)
524 C. Grau et al. / NeuroImage 36 (2007) 522–531

Page 5

Author's personal copy
each electrode (F(2,22)=15.16, P<0.001) in which a main
Electrode×Condition interaction effect was observed (F(8,88)=
3.46, P<0.05). This suggested that decreased N1 amplitude was not
equivalent at all electrode locations. Post-hoc comparisons were
made at each repetition of N1 peak value (S1, S2 and S3) and
electrode position. While N1 to S1 was higher than N1 to S2 and S3
(P<0.01)exceptS1vs.S2atFz(P<0.05),nosignificantdifferences
were encountered between N1 amplitudes to S2 and to S3 (P>0.1).
These results are extended to all electrode locations (S1 vs. S2 and
S3, P<0.05; S2 vs. S3, P>0.1).
Independent components of N1 and N1 suppressed: early and late
onset N1 subcomponents
Three independent components (cFTl, cFa and cTl) activated in
response to the S1 stimulus (Figs. 2 and 3), but not in response to
repeated S2and S3stimuli,while two other componentswere present
after all S1, S2 and S3 stimuli presentation (cTrl and cTr) (Fig. 3). IC
contributions to ERP variance raised up to 76.3% (Fig. 4a).
Component cFTl (fronto-left temporal) (Fig. 2) showed a late-
onset pattern and was significant from 144 to 174 ms stimulus
onset (P<0.05), which encloses the second half of N1 to S1
stimulus, but was not significant after S2 and S3 stimuli (P>0.05).
Brain sources underlying cFTl encompassed left temporal and right
inferior frontal cortex (P<0.05), and its time–frequency analysis
showed that theta, alpha and low beta frequency ranges (3.9–
19.5 Hz) underlie this component, by both significant (P<0.01)
enhanced spectral power and ITC increase after S1.
Component cFa (frontal anterior) (Fig. 2) manifested a late
onset start and was significant after S1 stimulus onset from 129 to
162 ms (P<0.05), also contributing to the second half of the N1
latency range. No significant participation of this component was
encountered in response to S2 and S3 stimuli (P>0.05). Brain
sources of component cFa were placed in anterior cingulate and
right inferior frontal cortex (P<0.05). cFa time–frequency analysis
showed similar patterns as cFTl after S1, i.e., significant (P<0.01)
enhancement to both spectral power and ITC within theta, alpha
and low beta (3.9–19.5 Hz) frequency bands. Moreover, a small,
Fig. 2. Late-onset N1 ICs (cFTl and cFa). Each IC had: (a) its scalp spatial distribution, (b) the associated cortical source location (P<0.05), (c) its temporal
evolutionwithinN1temporalwindowappearance–colouredbarsbelowaxis denotesignificant(P<0.05)activation,as comparedto baselineperiod– and(d) its
significant (P<0.01) spectral power (ERSP) and phase alignment (ITC) associated when it was significantly activated, as shown in (c). The apparent ERSP and
ITC temporal non-coincidence could be due to a smearing effect of wavelet transformation (Lakatos et al., 2005). (For interpretation of the references to colour in
this figure legend, the reader is referred to the web version of this article.)
525 C. Grau et al. / NeuroImage 36 (2007) 522–531