Content uploaded by Corinna Haenschel
Author content
All content in this area was uploaded by Corinna Haenschel
Content may be subject to copyright.
Gamma and beta frequency oscillations in response
to novel auditory stimuli: A comparison of human
electroencephalogram (EEG) data with
in vitro
models
Corinna Haenschel*
†
, Torsten Baldeweg
‡
, Rodney J. Croft*, Miles Whittington
§
, and John Gruzelier*
*Division of Neuroscience and Psychological Medicine, Imperial College School of Medicine, London W6 8RP, United Kingdom; ‡Institute of Child Health and
Great Ormond Street Hospital for Sick Children, University College London, London WC1N 2AP, United Kingdom; and §School of Biomedical Sciences,
University of Leeds, Leeds LS2 9NL, United Kingdom
Communicated by Nancy J. Kopell, Boston University, Boston, MA, April 10, 2000 (received for review December 15, 1999)
Investigations using hippocampal slices maintained in vitro have
demonstrated that bursts of oscillatory field potentials in the
gamma frequency range (30–80 Hz) are followed by a slower
oscillation in the beta 1 range (12–20 Hz). In this study, we
demonstrate that a comparable gamma-to-beta transition is seen
in the human electroencephalogram (EEG) in response to novel
auditory stimuli. Correlations between gamma and beta 1 activity
revealed a high degree of interdependence of synchronized oscil-
lations in these bands in the human EEG. Evoked (stimulus-locked)
gamma oscillations preceded beta 1 oscillations in response to
novel stimuli, suggesting that this may be analogous to the
gamma-to-beta shift observed in vitro. Beta 1 oscillations were the
earliest discriminatory responses to show enhancement to novel
stimuli, preceding changes in the broad-band event-related poten-
tial (mismatch negativity). Later peaks of induced beta activity over
the parietal cortex were always accompanied by an underlying
gamma frequency oscillation as seen in vitro. A further analogy
between in vitro and human recordings was that both gamma and
beta oscillations habituated markedly after the initial novel stim-
ulus presentation.
Gamma and beta frequency oscillations occur in the neocor-
tex in response to sensory stimuli over a range of modalities
(1, 2). Evidence is accumulating that gamma oscillations are
involved in feature binding and associational memory (3, 4). The
mechanism behind these cortical oscillations remains to be
elucidated, but research has demonstrated that, at a cellular and
network level, cortical gamma activity can be generated by
specific neuronal subtypes (5) and networks of interconnected
inhibitory interneurons (6). At a larger network level, focal
cortical gamma oscillations can be elicited upstream from pri-
mary sensory pathways by tetanic stimulation of the thalamic
reticular nucleus (7). Previous studies using cortical and hip-
pocampal slice preparations have demonstrated that experimen-
tal gamma oscillations can occur spontaneously for long periods
of time during activation of metabotropic cholinergic receptors
(8) and can be induced transiently by activation of metabotropic
glutamate receptors or by bursts of afferent stimulation (6, 9, 10).
Experimentally, beta frequency oscillations are generated
following periods of synchronous gamma frequency activity (9,
10). Their generation depends on the period-by-period poten-
tiation of recurrent excitatory synaptic potentials, which occurs
concurrently with a recovery of postspike afterhyperpolarization
following the initial stimulus. The beta oscillations appeared to
manifest as subharmonics of the preceding gamma oscillation,
with pyramidal cells firing on only ever y second or third period
of a continuing subthreshold, inhibition-based gamma oscilla-
tion. As a consequence, all cells involved in the beta network will
skip the same beats. These oscillations have been suggested to
represent a dynamic method of generating and recalling stimu-
lus-specific assemblies of neurons. In addition, recent efforts to
model beta oscillations have shown that their generation en-
hances the range of temporal delays between elements in a
neuronal assembly within which synchronization is possible (11).
This observation is complementary to the demonstration that
long-range (multimodal) sensory coding is performed at beta
frequencies, whereas local synchronization occurs at gamma
frequencies (12).
Transient bursts of gamma oscillations in the human electro-
encephalogram (EEG) can be detected in response to sensory
stimulation (13). They can either be tightly time and phase
locked to the stimulus (termed stimulus-evoked gamma oscilla-
tions) (14), or they may occur with variable latency (termed
stimulus-induced gamma oscillations) (15). Stimulus-evoked
gamma oscillations are contained in the averaged evoked po-
tential and can be extracted by band-pass filtering (14). Such
short-latency gamma-band oscillations have been found in the
auditory potential within 100 ms from stimulus onset (16–19). In
contrast, stimulus-induced gamma oscillations disappear in the
average evoked potential because of the jitter in latency from
one trial to the next. They have to be extracted by using methods
that distinguish between phase-locked and non-phase-locked
activity (20). In this study, we used the method of event-related
synchronization and desynchronization (21).
Induced gamma activity has been found in response to sensory
stimuli in the human EEG (15, 20). The functional significance
of evoked and induced gamma oscillations still remains unclear.
However, it has been suggested that the evoked gamma band
response may reflect synchronously active neural assemblies
(feature binding) or may signal the precise temporal relationship
of concurrently incoming stimuli (20). The later induced gamma
oscillation is thought to reflect object representation (3, 20) or
the activation of associative memories (4, 22).
Recent in vitro studies have demonstrated that stimulus-
induced gamma and beta oscillations habituate markedly on
repeated stimulation (23). The aim of this study was to investi-
gate the relationship between gamma and beta oscillations in the
human EEG in response to novel and repeated auditory stimuli
designed to mimic the in vitro paradigm. We tested the hypoth-
esis that novel stimuli differing in frequency from a repeated
sequence of tones would elicit a similar burst of correlated
gamma and beta 1 activity. Furthermore, we expected that these
responses, particularly with regard to induced oscillations, would
habituate rapidly with stimulus repetition.
Methods
Subjects. Ten participants (three females, seven males), 20–35
years old (mean, 21.5), with normal hearing were tested. All
Abbreviations: EEG, electroencephalogram; ERP, event-related potential; GFP, global field
power; MMN, mismatch negativity.
†To whom reprint requests should be addressed at: Cognitive Neuroscience and Behavior,
Imperial College School of Medicine, London W6 8RP, UK. E-mail: C.Haenschel@ic.ac.uk.
The publication costs of this article were defrayed in part by page charge payment. This
article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C.
§1734 solely to indicate this fact.
Article published online before print: Proc. Natl. Acad. Sci. USA, 10.1073兾pnas.120162397.
Article and publication date are at www.pnas.org兾cgi兾doi兾10.1073兾pnas.120162397
PNAS
兩
June 20, 2000
兩
vol. 97
兩
no. 13
兩
7645–7650
PSYCHOLOGY
participants were free of neurological and psychiatric disorders
and had no history of hearing impairment. Ethical approval was
obtained and all participants provided written consent before
testing. Subjects were seated in an armchair with head support
in a sound-attenuated and electrically shielded testing chamber.
Instructions were given to relax completely with eyes closed.
Stimuli. Pure sinusoidal tones were generated with a Neurosoft
Sound program and delivered binaurally through headphones by
the Stim interface system (Neuroscan Labs, Sterling, VA).
Stimuli were presented in a series of 40 trials. A single trial
consisted of a sequence of 8 tones (randomized up to a maximum
of 16) of one constant frequency, of which the first eight were
analyzed. The tone frequency altered randomly between trials
from the lower frequency limit of 100 Hz to the upper limit of
5000 Hz (from 100 to 1000 Hz in steps of 100 Hz and between
1000 and 5000 Hz in steps of 200 Hz). The tones were 50 ms long
(rise and fall time 5 ms), had an intensity of 95 dB (sound
pressure level), and were presented with a constant stimulus
onset asynchrony of 2 s. The intertrial interval was also 2 s.
EEG Recording. An ECI Electro-Cap containing 28 electrodes
arranged in accordance with the international 10–20 system was
fitted to the subject’s head and a ground electrode was placed 1.5
cm in front of the central frontal electrode. A reference elec-
trode was attached to the nose. Vertical and horizontal electro-
oculogram electrodes were placed above and below the right eye
and laterally from both eyes. Recording, digitization, and pro-
cessing of the EEG data were carried out with a SynAmps
amplifier and Neuroscan system version 4.0. The EEG was
recorded at a sampling rate of 500 Hz with a system bandpass
between 0 and 100 Hz.
Data Analysis. EEG data were averaged in intervals from 500 ms
before the stimulus and up to 1,000 ms after stimulus onset and
were baseline corrected from ⫺250 to 0 ms before stimulus for
each stimulus. First, EEG epochs were excluded automatically if
amplitudes exceeded ⫾100
V and were then visually inspected
for more subtle artifacts, such as muscle contamination.
Broad-band event-related potential (ERP) components were
filtered between 0.5 and 30 Hz before further processing. Peak
amplitudes and latencies of N1 and P3 components were defined
in the following time intervals: 80–150 ms and 250– 450 ms,
respectively. Analysis of oscillatory responses distinguished be-
tween phase-locked and non-phase-locked activity. In both sets
of analyses, the cut-off frequencies for the following frequency
bands were as follows: alpha, 8–12 Hz; beta 1, 12–20 Hz; beta 2,
20–30 Hz; and gamma, 30–50 Hz. No specific predictions were
made for changes in the alpha band. However, it was included in
the analysis to exclude the possibility that changes in the gamma
band may be artifacts of alpha harmonics (16).
Stimulus-evoked activity was obtained from the averaged
evoked potential of each subject by bandpass filtering with a
two-pole Butterworth, zero-phase-shift digital filter (⫺6dB
edges at cut-off frequencies, slope 24 dB兾octave). For analysis of
evoked oscillatory activity, we used a global field power (GFP)
analysis as previously suggested to improve the signal-to-noise
ratio of multichannel data (17). GFP is defined as the standard
deviation across multiple channels as a function of time. It is a
root mean square measure that quantifies the spatial potential
field sampled over the scalp. A peak of GFP reflects a maximum
of the total underlying brain activity that contributes to the
surface potential field (24). GFP was used to measure the mean
and peak amplitude as well as latency for stimulus-evoked
gamma, beta 1, beta 2, and alpha bands in the 20- to 150-ms time
window.
The power of induced activity was calculated by using the
Instantaneous Frequency Analysis module of Scan 4.0 software
(Neuroscan Labs) based on the method of complex demodula-
tion (25, 26). First, the center frequency (CF) and cut-off
frequency were selected for a zero-phase bandpass filter. Com-
plex demodulation then produced a complex time series [with
real part cos(2
䡠CF䡠t) and imaginary part sin(2
䡠CF䡠t)] from the
original real-time series. This had the effect of shifting the entire
spectrum of the original time series to 0 Hz. Both the real and
imaginary parts were low-pass filtered. Then the envelope of the
center frequency activity was computed as the modulus of the
complex time series after filtering (in units of power in
V
2
). To
compute the induced band power statistic across all epochs, the
modulus squared was summed across all epochs, and the real and
imaginary parts were summed across all epochs (at each time
point within the epoch). The mean across epochs at each time
point (average evoked potential) was explicitly removed for the
power computations. The result is the event-related induced
band power. Mean and peak amplitude as well as latency of the
induced oscillations were identified in the time window between
200 and 400 ms and 600 and 800 ms in response to each of the
eight tones.
Statistical Analysis. The statistical analyses were based on the a
priori hypotheses that response habituation in gamma and beta
oscillations occurs immediately following the first stimulus as
observed in the in vitro experiments (23). Therefore, repeated-
measures ANOVA compared the response to stimulus 1 with the
mean of subsequent stimuli 2–8 (within-subject factor stimulus).
To account for possible topographical effects on broad-band
ERP components N1 and P3 as well as on induced activity in
each frequency band, the following within-subject factors were
also included: anterior–posterior (n⫽2) and side (n⫽3: left,
center, right). These six regions included the following elec-
trodes: left anterior region F3, F7, T3, C3; central anterior region
CZ, FZ; right anterior region F4, F8, T4, C4; left posterior region
P3, T5, PO1; central posterior region PZ, OZ; and right posterior
region P4, T6, PO2.
Topographical information was not available for the GFP
values (evoked activity). Therefore, ttests were used to compare
mean GFP differences within each frequency band. Peak laten-
cies were compared to test the hypotheses that evoked and
induced gamma oscillations would precede beta activity. Bon-
ferroni correction was applied to correct for multiple compar-
isons across frequency bands.
Results
Broad-Band ERPs. The first stimulus in each trial evoked enhanced
N1 and P3 components (F
(1,9)
⫽11.4, P⬍0.001), F
(1,9)
⫽32.7,
P⬍0.001, respectively) (Fig. 1). The corresponding difference
waves revealed an early frontal negative wave peaking at 107 ms,
which exhibited the typical characteristics of a mismatch nega-
tivity (MMN or N2a) (27). In contrast to the later N2b compo-
nent, which was maximal at the vertex, MMN showed inverted
peaks at mastoid electrodes (data not shown). With large
frequency separation between stimuli, the MMN temporally
overlaps the N1 (28, 29). The P3 component in the difference
wave showed a vertex (CZ) amplitude maximum at 322 ms and
a later parietal peak at 350 ms.
Evoked Oscillatory Activity. The gamma response to stimulus 1
consisted of a negative-positive-negative complex peaking at 45
ms (SD 16) with a fronto-central distribution (Figs. 2 and 3). This
complex showed inverted peaks at temporal leads, in line with
previous evidence suggesting bilateral sources within the supe-
rior temporal plane (13). The evoked beta 1 and beta 2 responses
showed a similar distribution, but with longer peak latencies (74
ms, SD 27 and 59 ms, SD 26, respectively). GFP measurements
indicated a steady rise in beta 1 power during the initial burst of
gamma activity. As this initial stimulus-locked gamma burst
7646
兩
www.pnas.org Haenschel et al.
subsided, it was replaced by a strong oscillatory response within
the beta 1 band. There was a delay of approximately two periods
of gamma oscillation before the onset of beta 1 activity. The
mean evoked response to stimulus 1 was significantly larger
compared with subsequent stimuli (Figs. 2 and 4) for beta 1
(t
(9)
⫽2.36, P⫽0.042), but not for beta 2 (t
(9)
⫽1.73, P⫽0.118)
and gamma (t
(9)
⫽1.29, P⫽0.230).
As predicted, activity in the gamma band peaked significantly
earlier than in beta 1 (t
(9)
⫽3.60, P⫽0.006), but not in beta 2
(t
(9)
⫽1.88, P⫽0.093). The same was true for peak latencies to
the subsequent stimuli 2–8 (gamma: 49 ms, SD 12; beta 1: 65 ms,
SD 11; beta 2: 50 ms, SD 13; t
(9)
⫽2.61, P⫽0.028; t
(9)
⫽0.17,
P⫽0.867, respectively).
It is important to note that there were no significant stimulus
effects for evoked alpha activity (t
(9)
⫽0.60, P⫽0.563). Also,
alpha peaked later (102 ms, SD 40) in comparison to the other
frequency bands, and this difference was significant (stimulus 1:
t
(9)
⬎2.84, P⬍0.019; stimuli 2–8: t
(9)
⬎3.99, P⬍0.003).
Induced Oscillatory Activity. Two distinct peaks of induced activity
with a central posterior distribution were seen in response to
stimulus 1. As predicted, the first peak of oscillatory-induced
activity (200 – 400 ms; Fig. 2) was larger in response to stimulus
1 compared with subsequent stimuli for beta 1 (stimulus effect,
F
(1,9)
⫽6.2, P⫽0.035); beta 2, (F
(1,9)
⫽7.1, P⫽0.029) and in
trend for gamma: F
(1,9)
⫽4.9, P⫽0.054). Gamma and beta 1
activity were larger over posterior compared with anterior
regions for the first but not for subsequent stimuli (interaction
effect stimulus by anterior-posterior) for gamma: F
(1,9)
⫽6.1,
P⫽0.036; and beta 1: F
(1,9)
⫽5.4, P⫽0.045). The habituation
of the oscillatory response within these frequency bands oc-
curred after the initial novel stimulus and was not graded
throughout consecutive stimuli (Figs. 2 and 4). Similar changes
occurred in the second peak of induced gamma and beta 1
activity between 600 and 800 ms poststimulus (stimulus effect:
for gamma: F
(1,9)
⫽7.7, P⫽0.022, beta 1: F
(1,9)
⫽5.5, P⫽0.044),
beta 2 (F
(1,9)
⫽4.5, P⫽0.064). However, in contrast to the first
peak, this later peak was still evident in response to subsequent
stimuli. For beta 1 this peak was also more prominent at
posterior sites (F
(1,9)
⫽7.4, P⫽0.023).
In contrast to the pattern of peak latencies of the evoked
oscillatory activity where gamma led beta 1, peaks of induced
beta activity appeared to superimpose on a prolonged period of
induced gamma frequency activity (Fig. 5). The peak latencies of
these two induced peaks at the point of maximum (Fig. 2) in the
central parietal electrode were as follows: for the first peak:
gamma 330 ms, SD 49; beta 1: 288 ms, SD 50; and beta 2: 301
ms, SD 65; and for the second peak: gamma: 684 ms, SD 70;
beta 1: 684 ms, SD 50; beta 2: 663 ms, SD 45. There were no
significant differences between gamma and beta latencies (t
(9)
⬍1.84, P⬎0.099, t
(9)
⬍0.83, P⬎0.427, for early and late
peaks, respectively).
In addition to the predicted changes in gamma and beta
oscillations, there was also a significant stimulus effect in the
alpha band (F
(1,9)
⫽9.0, P⫽0.015) for peak 1 and in trend
(F
(1,9)
⫽4.7, P⫽0.059) for peak 2 (Fig. 2). While the latency of
the early alpha peak was not different from that of gamma and
beta (t
(9)
⬍1.24, P⬎0.246), the late alpha activity occurred with
a delay of more than 100 ms (t
(9)
⬍5.46, P⬍0.001).
Correlation Between Gamma and Beta Activity. Finally, we investi-
gated whether the magnitude of gamma activity correlated with
the magnitude of beta 1 activity as predicted from in vitro studies
(9, 10, 23). We expected that such a correlation would apply
particularly to the first, novel, stimulus in each stimulus trial.
This prediction was confirmed for both stimulus-evoked (GFP)
Fig. 1. Broad-band ERPs shown in response to the first stimulus and stimuli
2–8 (Left) and corresponding difference waves (Right) for fronto-central (FZ,
Top), vertex (CZ, Middle), and centro-parietal (PZ, Bottom) electrodes. MMN,
mismatch negativity. Note the prominent P3 potential at 320 ms with a vertex
maximum compared with the more posterior peak at 350 ms. The former
resembles the ‘‘novelty P3’’ (P3a) described by Knight and Nakada (30).
Fig. 2. (A) Stimulus-locked oscillations. Grand average auditory evoked
responses to stimulus 1 and stimuli 2– 8 are shown for gamma (30–50 Hz), beta
1 (12–20 Hz), beta 2 (20 –30 Hz), and alpha bands (8–12 Hz). Global Field Power
(GFP) quantifies the spatial potential field over the scalp. Note that the GFP
values exhibit twice the frequency of the bandpass frequency because of
rectification of the data when computing root-mean-square values. (B) In-
duced activity. Time windows used for statistical analysis of the two induced
peaks are indicated in shading in the bottom traces.
Haenschel et al. PNAS
兩
June 20, 2000
兩
vol. 97
兩
no. 13
兩
7647
PSYCHOLOGY
activity and for the early peak of the induced oscillator y activity.
For evoked GFP this correlation was significant for stimulus 1
(gamma–beta 1: Spearman’s r⫽0.73, P⫽0.016), but not for
subsequent stimuli (gamma–beta 1: r⫽0.04, P⫽0.907). There
was no significant correlation between gamma and beta 2 activity
(for stimulus 1: r⫽0.51, P⫽0.130; for stimuli 2–8: r⫽0.09, P⫽
0.803). The early induced gamma correlated significantly with
beta 1 for the first and subsequent stimuli (r⫽0.70, P⫽0.025;
r⫽0.71, P⫽0.022, respectively). For the late peak this
correlation was not significant for stimulus 1 (r⫽0.41, P⫽
0.244); however, it was highly significant for subsequent stimuli
2–8 (r⫽0.95, P⬍0.001). This different pattern of correlation
suggests that the late induced activity may subserve processes in
addition to the initial encoding of novel stimuli.
Discussion
In vitro studies have characterized a robust pattern of oscillatory
responses to afferent stimulation in the hippocampus (9, 10). The
response takes the form of an initial period of gamma oscillation
followed by a transition, of variable duration, to a beta frequency
oscillation. In addition, recent studies have shown that this
pattern of oscillatory response habituated markedly when stimuli
were presented less than 1 min apart (23). Results from the
present study demonstrated that a very similar pattern of tran-
sition from gamma to beta oscillations occurs in response to
auditory stimulation in the human EEG.
Three observations are critically involved in supporting this
statement. First, the initial evoked oscillatory response to novel
stimuli took the form of a gamma oscillation, which was either
replaced by or superimposed on a longer-lasting beta frequency
oscillation (gamma-to-beta shift). For the emergence of oscil-
latory EEG activity at the scalp a certain level of underlying
neuronal synchronization as well as the recruitment of pyramidal
cells is necessary. This indicates a superthreshold origin of these
oscillations. The transition from gamma to beta oscillation was
related to the magnitude of this initial superthreshold gamma
oscillation. Second, gamma oscillations were always present
(evoked or induced) concurrently with bursts of beta activity.
Fig. 3. Topographical maps of the evoked and induced activity for the first stimulus (A) and the mean of stimuli 2–8 (B) are shown for all four frequency bands.
For evoked activity, maps were computed for the two positive peaks around the GFP maximum. For induced activity, maps are shown for the maxima of the early
and late induced peaks. Note that the maps of evoked beta 1 exhibit an early fronto-central and a later centro-parietal distribution. Polarity information is not
included in the distribution of induced peaks.
Fig. 4. Mean and standard error are shown for the evoked and induced
activity in response to the first eight stimuli in each trial. Response amplitude
in the evoked and induced beta 1 band differed significantly between the first,
novel, stimulus and the mean of all subsequent stimuli.
7648
兩
www.pnas.org Haenschel et al.
The beta activity that correlated with gamma (30 –50 Hz) was in
the beta 1 range (12–20 Hz) not the beta 2 range (20 –30 Hz).
This observation suggests that the beta 1 oscillations represent
a subharmonic of the on-going gamma and not simply a slowing-
down of the gamma frequency response, and accordingly cor-
responds to what is seen in vitro (9, 10). Finally, in the present
protocol novel auditory stimuli were followed by repeated
identical stimuli. Both gamma and beta oscillations habituated
markedly after the initial novel presentation. This effect was
particularly apparent for oscillations in the beta 1 band, also
consistent with in vitro obser vations (23).
The above comparisons have yielded some intriguing similar-
ities. However, conclusions based on comparison between in
vitro preparations and the human EEG must remain tentative at
present in view of considerable differences between the meth-
ods. For example, the activity recorded in vitro was predomi-
nantly induced in nature (i.e., not strictly phase-locked to the
stimulus), compared with a mixture of both stimulus-evoked and
-induced activity in the human recordings. Also, the exact nature
of the stimulation protocol used in both studies is analogous at
best. We used loud auditory stimuli to resemble the high
stimulation intensity needed to elicit beta oscillation in the
hippocampal slice. For the human experiment, sudden changes
in tone frequency were assumed to provide stimulus novelty. In
contrast, during the in vitro recordings novelty was thought to be
served by alternating from proximal to distal stimulation (re-
cording) sites in the rat hippocampal CA1 region (23). Also, the
time scale of stimulation was different for the human experiment
and the recording in the slice, reflecting optimal conditions for
eliciting oscillatory activity in these two different experimental
situations. It is also likely that there is at least some difference
in the generation of fast oscillations in the hippocampal cortex
(archicortex) and neocortex, and that there are species differ-
ences between rodents and humans. Furthermore, although the
correlation between single-cell and population activity is firmly
established (10), the sources of such activity in the human EEG
are less well established (13, 20). Nevertheless, the prediction
derived from the experimental and the theoretical model pro-
vides a firm basis for the understanding of analogous phenomena
in the human EEG.
The fronto-central scalp distribution of the early evoked
gamma and beta oscillations was consistent with previous studies
implicating bilateral dipole sources located in the auditory cortex
within the superior temporal gyrus (13). The later induced
gamma and beta oscillations were elicited with peaks at 300 ms
and 680 ms over central posterior electrodes, suggesting sources
within the parietal association areas. Ver y similar bimodal peaks
of induced gamma activity at 300 ms and 500 ms were obser ved
by Tallon-Baudry and Bertrand (20) over parietal regions in an
auditory target detection task. Notwithstanding that the induced
activity was extracted after rectification of the EEG (thereby
losing polarity information), it is still evident that separate
regions were involved in the generation of early evoked com-
pared with late induced activity. Our data do not directly address
the question of the functional interaction between early activa-
tion of the auditory cortex and the later activity over the
posterior association cortex implicated in novelty detection and
attention (30, 31). It is therefore of interest that the evoked beta
1 activity, occurring initially with a frontal distribution, appeared
to shift to a centro-parietal maximum (Fig. 3) during the later
phase of this evoked oscillation. Importantly, this shift can be
observed only for evoked beta 1 activity, not for gamma and beta
2. Although no firm conclusion about the cerebral generators
can be drawn from surface EEG data without combining them
with source localization methods, it will be import ant to establish
whether this topographical shift is related to the long-range
synchronizing properties of beta oscillations as predicted by
Koppell et al. (11).
Experimental protocols similar to the protocol used in the
present study commonly elicit MMN potentials between 80 and
200 ms over the frontal scalp as the earliest discriminatory ERP
(29, 32). MMN reflects an automatic (preattentive) comparison
in sensory memory between the incoming stimulus and a mem-
ory trace of the preceding stimuli. With large frequency sepa-
ration between the stimuli as in this experiment, the MMN is
superimposed on the N1 component (28, 29), reflecting overlap
of two separate components: refractoriness of the N1 generator
and a true MMN. MMN and N1 peaked at approximately 105 ms
poststimulus (shown as difference waves in Fig. 1), which was 30
ms later than the peaks of beta 1 activity, and 60 ms later than
gamma. No significant stimulus effects were observed in the
gamma response. The significant stimulus effect observed in
beta 1, therefore represents an even earlier discriminatory
response than the MMN. Post-hoc comparison showed that the
beta 1 GFP latency was significantly shorter than the N1 peak
latency (t
(9)
⫽3.32, P⫽0.009). Thus, the present results are in
agreement with previous reports that failed to find significant
stimulus effects in the gamma band response before 100 ms
(33–35). However, the current data showed that stimulus-
specific encoding of novel or deviant stimuli resulted in en-
hanced beta 1 oscillations, which has not been investigated
previously. Nevertheless, there is evidence that attention and
expectancy can enhance this early gamma response, suggesting
a role for top-down modulation (36–38).
The relationship between gamma and beta oscillations points
to a number of cognitive correlates. Sokolow (39) and Darrow
et al. (40) discovered that beta oscillations appeared in the
human EEG in response to novel stimuli that also elicited an
‘‘orienting response.’’ It remains to be investigated whether the
electrodermal orienting response (30, 41) shows a pattern of
habituation similar to that of the gamma–beta frequency shifts
recorded in this study.
It has been argued that the early evoked gamma band response
could reflect the initial coactivation of neuronal assemblies
representing specific stimulus features (binding) (20). A change
Fig. 5. (Upper) The temporal sequence of the early (evoked) and late
(induced) oscillatory activity in the gamma and beta 1 band is illustrated. Note
that the shift from gamma to beta 1 occurred at the initial phase-locked
response, whereas the induced peaks were not significantly different in
latency. The time scale and voltage scaling for evoked and induced oscillations
differ. (Lower) Schematic summary of the temporal sequence using a common
time-scale, showing that the gamma-to-beta transition as seen in vitro was
observed only for the early evoked (phase-locked) peak. There were no
significant differences between the late induced gamma and beta 1 bands in
peak latency.
Haenschel et al. PNAS
兩
June 20, 2000
兩
vol. 97
兩
no. 13
兩
7649
PSYCHOLOGY
in such feature configuration by a novel stimulus would be
detected at the level of the auditory cortex. The present data
suggest that this detection mechanism is reflected in a burst of
beta 1 oscillations. A similar role for gamma-induced beta
oscillations in novelty detection was postulated on the basis of in
vitro data (9). In the latter case synchronous beta oscillations
were generated as a consequence of potentiation of excitatory
synaptic connections between pyramidal cells afforded by the
initial synchronous gamma oscillation. In the hippocampal slice
model this phenomenon takes approximately 10 periods of
gamma oscillation to produce the required potentiation. In the
present study this appears to occur over 2–5 periods of gamma.
This temporal difference may reflect the greater occurrence of
recurrent excitatory synaptic connectivity in the neocortex in
comparison with the hippocampus.
The following scenario of processing steps can be postulated.
Stimulus novelty detected in the auditory cortex is ‘‘flagged’’ to
posterior association areas by means of a synchronous oscillation
in the beta 1 range, which then entrains local inhibitory networks
at gamma frequency (11). This may explain why induced beta 1
activity over the parietal cortex peaked slightly earlier than
gamma. It still needs to be established whether this mechanism
is specific to the beta frequency range. In human experiments
thalamo–cortical interactions in the alpha range are likely to
occur. This may account for the observed stimulus effects on
induced alpha activity. However, as in other studies (reviewed in
ref. 20), differences in latency and topography suggest that the
gamma and beta oscillations are not artifacts of alpha harmonics
as previously argued (16).
In conclusion, despite differences in method there was a
considerable degree of consistency between the two levels of
experimental analysis. This study demonstrates that the pattern
of interplay between gamma and beta (specifically beta 1)
oscillations is similar in neocortical responses to auditory stim-
ulation to that seen in vitro in the hippocampal slice preparation.
Both show (i) a degree of dependence on gamma oscillations for
the generation of beta responses and (ii) a very similar pattern
of habituation to repeated stimulation, which suggest that they
may be involved in processes associated with encoding into
sensory memory both at the cellular level (synaptic potentiation)
and at the cognitive level (ERP: MMN). These data demonstrate
a robust phenomenon allowing the study of the interrelationship
between gamma and beta oscillations in humans. The similarities
between these findings and observations from in vitro studies
suggest that a combination of these approaches will facilitate
detailed pharmacological and psychophysiological investigations
into the mechanisms of the generation and function of fast
oscillations in the central nervous system.
We thank Dr. Mark Pflieger for helpful discussion and support w ith the
instantaneous frequency analysis. This study was supported by grants
from the Institut fu¨r Grenzgebiete der Psychologie und Psychohygiene,
Freiburg, Germany, and The Wellcome Trust.
1. Barth, D. S. & MacDonald, K. D. (1996) Nature (London) 383, 78 – 81.
2. Roelfsema, P. R., Engel, A. K., Konig, P. & Singer, W. (1997) Nature (London)
385, 157–161.
3. Rodriguez, E., George, N., Lachaux, J. P., Martinerie, J., Renault, B. & Varela,
F. J. (1999) Nature (London) 397, 430– 433.
4. Miltner, W. H., Braun, C., Arnold, M., Witte, H. & Taub, E. (1999) Nature
(London) 397, 434– 436.
5. Gray, C. M. & McCormick, D. A. (1996) Science 274, 109–113.
6. Whittington, M. A., Traub, R. D. & Jefferys, J. G. (1995) Nature (London) 373,
612–615.
7. MacDonald, K. D., Fifkova, E., Jones, M. S. & Barth, D. S. (1998) J. Neuro-
physiol. 79, 474–477.
8. Buhl, E. H., Tamas, G. & Fisahn, A. (1998) J. Physiol. (London) 513, 117–126.
9. Whittington, M. A., Traub, R. D., Faulkner, H. J., Stanford, I. M. & Jeffreys,
J. G. R. (1997) Proc. Natl. Acad. Sci. USA 94, 12198–12203.
10. Traub, R. D., Whittington, M. A., Buhl, E., Jeffreys, J. G. R. & Faulkner H. J.
(1999) J. Neurosci. 19, 1088–1105.
11. Kopell, N., Ermentrout, B. E., Whittington, M. A. & Traub, R. D. (2000) Proc.
Natl. Acad. Sci. USA 97, 1867–1872.
12. Von Stein, A., Rappelsberger, P., Sarnthein, J. & Petsche, H. (1999) Cerebral
Cortex 9, 137–150.
13. Pantev, C., Makeig, S., Hoke, M., Galambos, R., Hampson, S. & Gallen, C.
(1991) Proc. Natl. Acad. Sci. USA 88, 8996–9000.
14. Galambos, R., Makeig, S. & Talmachoff, P. J. (1981) Proc. Natl. Acad. Sci. USA
78, 2643–2647.
15. Tallon-Baudry, C., Bertrand, O., Delpuech, C. & Pernier, J. (1996) J. Neurosci.
16, 4240– 4249.
16. Juergens, E., Roesler, F., Henninghausen, E. & Heil, M. (1995) Neuroreport 6,
813–816.
17. Clementz, B. A., Blumenfeld, L. D. & Cobb, S. (1997) Neuroreport 8, 3889 –
3893.
18. Basar, E., Rosen, B., Basar-Eroglu, C. & Greitschus, F. (1987) Int. J. Neurosci.
33, 103–117.
19. Karakas, S. & Basar, E. (1998) Int. J. Psychophysiol. 31, 13–31.
20. Tallon-Baudr y, C. & Bertrand, O. (1999) Trends Cogn. Sci. 3, 151–162.
21. Pfurtscheller, G. (1988) Electroencephalogr. Clin. Neurophysiol. 70, 190–193.
22. Pulvermueller, F., Keil, A. & Elbert, T. (1999) Trends Cogn. Sci. 3, 250–252.
23. Doheny, H. C., Faulkner, H. J., Gruzelier, J. H. & Whittington, M. A. (2000)
J. Physiol., in press.
24. Lehmann, D. & Skrandies, W. (1986) Electroencephalogr. Clin. Neurophysiol.
48, 609– 621.
25. Thatcher, R. W., Toro, C., Pf lieger, M. E. & Hallet, M. (1994) in Functional
Neuroimaging: Technical Foundations, ed. Thatcher, R. W. (Academic, San
Diego), pp. 269–278.
26. Thatcher, R. W. (1995) J. Neuroimaging 5, 35– 45.
27. Naatanen, R. (1992) Attention and Brain Function (Erlbaum, Hillsdale, NJ).
28. Scherg, M., Vasjar, J. & Picton, T. W. (1989) J. Cogn. Neurosci. 1, 336–355.
29. Baldeweg, T., Richardson, A., Watkins, S., Foale, C. & Gruzelier, J. (1999)
Ann. Neurol. 45, 495–503.
30. Knight, R. T. & Nakada, T. (1998) Rev. Neurosci. 9, 57–70.
31. Yamaguchi, S. & Knight, R. T. (1991) J. Neurosci. 11, 2039 –2054.
32. Naatanen, R. & Winkler, I. (1999) Psychol. Bull. 125, 826 –859.
33. Tiitinen, H., Sinkkonen, J., May, P. & Naatanen, R. (1994) Neuroreport 6,
190–192.
34. Joliot, M., Ribary, U. & Llinas, R. (1994) Proc. Natl. Acad. Sci . USA 91,
11748–11751.
35. Marshall, L., Moelle, M. & Bartsch, P. (1996) Neuroreport 7, 1517–1520.
36. Tiitinen, H., Sinkkonen, J., Reinikainen, K., Alho, K., Lavikainen, J. &
Naatanen, R. (1993) Nature (London)364, 59–60.
37. Makeig, S. & Jung, T. P. (1996) Brain Res. Cogn. Brain Res. 4, 15–25.
38. Herrmann, C. S., Mecklinger, A., Pfeifer, E. (1999) Clin. Neurophysiol. 110,
636– 642.
39. Sokolow, N. E. (1963) Annu. Rev. Physiol. 25, 545–580.
40. Darrow, C. W., Vieth, R. N. & Wilson, J. (1957) Science 126, 74 –75.
41. Gruzelier, J., Eves, F., Connolly, J. & Hirsch, S. (1981) Biol. Psychol. 12,
187–209.
7650
兩
www.pnas.org Haenschel et al.
