tent.
Cross-frequency phase synchronization: A brain mechanism of memory matching and attention
Cross-frequency phase synchronization: A brain mechanism of
memory matching and attention
Paul Sauseng,
a,b
Wolfgang Klimesch,
a, Walter R. Gruber,
a
and Niels Birbaumer
b
a
Department of Psychology, University of Salzburg, Austria
b
Institute of Medical Psychology, Eberhard-Karls-University Tuebingen, Germany
Received 2 August 2007; revised 24 October 2007; accepted 13 November 2007
Available online 3 December 2007
Spatial attention amplifies the neural response, i.e. spike rates, brain
metabolism, and oscillatory activity at gamma frequency (beyond
30 Hz). In this study we show that when a visual target is attended
enhanced synchrony between gamma phase (30 to 50 Hz) and theta
phase (4 to 7 Hz), representing bottom-up and top-down activity,
respectively, can be observed. This is interpreted as memory matching
between incoming visual information and stored (top-down) informa-
tion. The results highlight the function of oscillatory brain activity in
the integration of memory and attention processes. This seems to be
true in particular for theta oscillations showing increased interregional
phase-coupling. We conclude that memory information is stored within
a distributed theta network and it is matched with an incoming sensory
trace at posterior brain areas.
� 2007 Elsevier Inc. All rights reserved.
Introduction
Although the capacity of the visual system is impressingly large,
it is nonetheless limited. Since we are continuously exposed to a
tremendous amount of visual information, it is of crucial importance
to select the relevant information. It has been shown that visual
attention amplifies the evoked neural response (event-related poten-
tials and/or BOLD signal) to selected items leading to increased
efficiency of neuronal processing of attended information by the
way of faster response times or higher detection rates (
Hillyard and
Anllo-Vento, 1998; Hillyard et al., 1998; Luck et al., 1997; Kastner
and Ungerleider, 2000; Ungerleider et al., 1998; Desimone, 1998
).
This amplification of neuronal responses by attention was also
reported for oscillatory brain activity. Increased power of fast
rhythmic responses at gamma frequency (beyond 30 Hz) can be
observed during processing of attended vs. unattended stimuli
independent of sensory modality (
Fries et al., 2001; M�ller et al.,
2000; Gruber et al., 1999; Bauer et al., 2006; Kaiser and
Lutzenberger, 2005; Fell et al., 2003a; Steinmetz et al., 2000
).
If we draw attention to an external stimulus it is important to
hold a representation of the anticipated stimulus in mind. Thus, in
most attention tasks working memory plays an important role. It
was emphasized that attention and working memory share common
cortical networks (
Kastner and Ungerleider, 2000; Ungerleider et
al., 1998; Desimone, 1998
). Frontal and parietal brain areas seem to
be involved in both processes. Slow brain rhythms, particularly
theta oscillations (around 6 Hz), appear to be associated with
memory processes (
Jensen and Tesche, 2002; Kahana, 2006;
Klimesch et al., 1996; Sarnthein et al., 1998; O'Keefe and Burgess,
1999; Raghavachari et al., 2001; Sederberg et al., 2003; Seager
et al., 2002
). It has been suggested that the function of large,
distributed networks is associated with slow oscillations, such as
theta and alpha (
Von Stein and Sarnthein, 2000; Sauseng et al.,
2002, 2005a; Schack et al., 2005
). In contrast, high frequency
oscillations, such as gamma, are related to neural processes in more
local networks (
Von Stein and Sarnthein, 2000
). This leads to the
consideration that the integration between top-down processes
guided by a complex working memory system and the bottom-up
processing of perceptual information may be reflected by a dynamic
interaction between theta and high frequency oscillations. There is
supportive evidence for this view: (i) in working memory tasks the
involvement of theta oscillations, particularly in a fronto-parietal
network, is reported (
Sarnthein et al., 1998; Sauseng et al., 2004,
2005a; Kopp et al., 2006
). In addition, it was observed that gamma
activity also is increased in short-term memory tasks (
Kahana,
2006; Tallon-Baudry et al., 1998; Lutzenberger et al., 2002; Kaiser
et al., 2003; Sederberg et al., 2003; Howard et al., 2003; Osipova
et al., 2006
). (ii) It is well established that the phase of theta
oscillations is functionally related to gamma activity during
memory tasks. This is theoretically elaborated and tested with
neural networks (
Lisman, 2005; Jensen and Lisman, 2005; Lisman
and Idart, 1995; Jensen, 2004; Jensen, 2006
) as well as in human
EEG studies (
Fell et al., 2003b; Schack et al., 2002; Burgess and
Ali, 2002; Palva et al., 2005; Mormann et al., 2005; Demiralp et al.,
2007; Canolty et al., 2006
).
www.elsevier.com/locate/ynimg
NeuroImage 40 (2008) 308
317
Corresponding author. Department of Psychology, University of
Salzburg, Hellbrunnerstr. 34, A-5020 Salzburg, Austria, Europe. Fax: +43
662 8044 5126.
E-mail address:
wolfgang.klimesch@sbg.ac.at
(W. Klimesch).
Available online on ScienceDirect (www.sciencedirect.com).
1053-8119/$ - see front matter � 2007 Elsevier Inc. All rights reserved.
doi:
10.1016/j.neuroimage.2007.11.032 Based on this evidence, here we show that a shift of spatial
attention modulates phase-synchronization between theta and
gamma activity in the parieto-occipital cortex. When an internal
representation of an expected stimulus meets a matching sensory
input, theta activity (reflecting top-down processes) and gamma
oscillations (representing bottom-up processes;
Singer, 1993
;
Busch et al., 2004
) become synchronized in phase. This might
explain the functional interplay between working memory and
attention.
Methods
Participants
29 healthy volunteers participated in the experiment after giving
written informed consent. EEG data of 7 subjects were excluded
from analysis due to artifacts caused by eye-blinks and horizontal
eye movements. The remaining sample of 22 participants consisted
of 3 men and 19 women with a mean age of 23.6 years.
Experimental procedures
Participants performed a cued visual attention task (
Posner,
1980; Hillyard et al., 1994
). They had to fixate the centre of a
computer monitor (indicated by a fixation cross) throughout the
whole experiment. At the beginning of each trial an arrow
(1.2 � 0.6�) either pointing to the right or to the left was foveally
presented for 34 ms. Subjects were instructed to focus their
attention to the cued hemi-field without moving their eyes to the
target location. After an interval with a duration between 600 and
800 ms, a target was presented for 50 ms. Targets were white bars
on black background and were shown 6.5� either right or left from
the centre of the computer monitor. Subjects had to indicate by
button press whether the bar was small (1 � 1.9�) or large (1 � 2.2�).
Frequencies for small and large targets were 50% and were equally
distributed to the different experimental conditions. A total of 1024
trials was run. In half of them attention was cued to the right and in
the other half attention was cued to the left hemi-field. In 75% of the
trials cue and target location were congruent (valid cue condition)
and the remaining 25% were incongruent (invalid cue condition).
To analyze approximately the same number of epochs for both
conditions, only a third of the trials was randomly chosen from the
valid cue condition.
Prior to the EEG experiment a training session consisting of 50
trials was run. The EEG experiment was started immediately after
the training session.
EEG recordings and analysis
Using a Synamps 32-channel amplifier (Neuroscan Inc.), EEG
was recorded from 30 Ag
AgCl electrodes at a sampling rate of
250 Hz. Impedance was kept below 15 kOhm. EOG correction was
applied and data were visually inspected for artifacts. Data were
segmented in intervals of 2000 ms (1000 ms preceding target onset
to 1000 ms post-stimulus) for each condition separately. After arti-
fact rejection at least 80 trials remained for further analysis in each
subject and each experimental condition.
By averaging over trials event-related potentials (ERPs) were
obtained. This was done independently of cue validity, i.e. all trials
in which a target was presented in the left visual hemifield and all
trials with right hemifield target presentation were averaged sepa-
rately (the reason for that was that the following source localization
should apply equally well for the valid and invalid conditions).
Using BESA 5.1 (MEGIS Software Inc.) dipole source localization
was run for the early ERP components. The time window between
100 and 250 ms after target onset was used for source localization.
In each subject a bilateral symmetric dipole pair was set and fitted in
position and orientation to minimize residual variance. Then,
applying in each subject the resulting individual dipole model,
source wave forms for the two dipoles were calculated for each
single trial. This was done to reduce EEG data from 30 scalp
electrodes to only two channels, i.e. one left and one right
hemispheric dipolar source. All further analyses were then run using
the estimated source wave forms.
Using Matlab 7.0.1 (Mathworks) Gabor expansion was applied
to the single trials to obtain phase and amplitude information for 50
frequency bins between 1 and 50 Hz with a distance of 1 Hz
between center frequencies.
Gabor expansion is applied to transf