Opinion

Working memory and neural

oscillations: alpha–gamma versus
theta–gamma codes for distinct
WM information?
Frédéric Roux1 and Peter J. Uhlhaas2,3,4
1
BCBL, Basque Center for Cognition, Brain and Language, Paseo Mikeletegi 69, Donostia/San Sebastian, 20009, Spain
2
Institute of Neuroscience and Psychology, University of Glasgow, 58 Hillhead Street, Glasgow, G12 8QB, UK
3
Department of Neurophysiology, Max-Planck Institute for Brain Research, Deutschordenstrasse 46, Frankfurt am Main, 60528,
Germany
4
Ernst-Strüngmann Institute (ESI) for Neuroscience, in Cooperation with Max-Planck Society, Deutschordenstrasse 46, Frankfurt
am Main, 60528, Germany

Neural oscillations at different frequencies have recently [4]. This proposal has recently received support through
been related to a wide range of basic and higher cognitive observations that rhythmic activity at low (theta, alpha)
processes. One possible role of oscillatory activity is to and high (beta, gamma) frequencies facilitates the forma-
assure the maintenance of information in working mem- tion of coherently organized groups of neurons via the
ory (WM). Here we review the possibility that rhythmic establishment of transient temporal correlations [5–9].
activity at theta, alpha, and gamma frequencies serve Although rhythmic patterns of activity were first dem-
distinct functional roles during WM maintenance. Specif- onstrated by investigators at the beginning of the 20th
ically, we propose that gamma-band oscillations are ge- century, a relationship with behavior was only established
nerically involved in the maintenance of WM information. 50 years later in seminal work by Singer and colleagues
By contrast, alpha-band activity reflects the active inhibi- [10]. Specifically, these findings revealed that action poten-
tion of task-irrelevant information, whereas theta-band tials generated by cortical cells in the primary visual cortex
oscillations underlie the organization of sequentially or- (V1) are phase-locked to the gamma rhythm in response to
dered WM items. Finally, we address the role of cross- a visual stimulus, thereby providing evidence of a relation-
frequency coupling (CFC) in enabling alpha–gamma and ship between the phase of neuronal oscillations (see Glos-
theta–gamma codes for distinct WM information. sary) and the temporal organization of neuronal activity
(for a different perspective on phase-coding through gam-
Neuronal oscillations and working memory ma-band oscillations see [11]). This finding is supported by
Working memory (WM) involves the ability to maintain several studies that have demonstrated that neuronal
and manipulate information over short periods of time and oscillations enable efficient transmission and coding of
can be subdivided into the initial encoding of information information in distributed neuronal populations [7,12].
and maintenance and retrieval of WM items (Figure 1A) In relation to WM, several groups have demonstrated a
[1]. Because WM is centrally involved in many aspects of
higher cognitive functions, a substantial amount of work
has been dedicated to identifying the neuronal substrates Glossary
of different WM processes [2,3]. Electrocorticography (ECOG): measurement of electrical brain signals using
A central question in WM research is how groups of electrodes that are implanted subdurally on the surface of the brain.
neurons represent and sustain items in the absence of Electro/magnetoencephalography (EEG/MEG): non-invasive methods for
studying brain function that reflect the electrical activity of neuronal popula-
sensory inputs. One possibility is that reverberating neu- tions with millisecond temporal resolution.
ronal activity in distributed cell assemblies underlies WM Local field potential (LFP): electric potential in the extracellular space around
neurons. LFP is a widely available signal in many recording configurations,
maintenance, as originally proposed by Hebb (Figure 1B)
ranging from single-electrode recordings to multi-electrode arrays.
Long-range synchrony: synchronizations between widely separated brain
regions (> 2 cm) as reflected, for example, in phase synchrony.
Corresponding author: Uhlhaas, P.J. (peter.uhlhaas@glasgow.ac.uk).
Neuronal oscillations: prominent feature of spontaneous and task-related brain
Keywords: oscillations; cell assemblies; working memory; MEG/EEG; cross-frequency
activity that occur at the level of single units, local field potentials (LFPs), and
coupling; cognition.
EEG/MEG recordings. The traditional view is that neuronal oscillations reflect
1364-6613/$ – see front matter inhibition-based fluctuations of neuronal activity that emerge from the
ß 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tics.2013.10.010 synchronous activation of large neuronal ensembles.
Phase: way of quantifying the difference between two oscillations according to
some feature (peak or trough) of one of the oscillation with respect to the other.
Spectral power: reflects the amplitude of neural oscillations computed through
a time–frequency transformation (TFT).

16 Trends in Cognitive Sciences, January 2014, Vol. 18, No. 1

 Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

(A) Encoding (C)

Sample
items

Maintenance x
Baseline Cue Delay Saccade
Delay (D)
100 0.22

Frequency [Hz]

Coherence
50
Retrieval
Test
item 0 0
-0.5 0 0.5 1 1.5 2
Time [s]
Response

(B) (E)

12
13

1,4 5,9 Neuron #1

2,14 6,10 Neuron #2

3,11
Neuron #3
8
Neuron #4
7,15
Neuron #5

θ Rhythm

TRENDS in Cognitive Sciences

Figure 1. Working memory (WM), neuronal oscillations, and cell assemblies. (A) The three stages of WM comprise encoding, maintenance, and retrieval phases. (B)
Schematic representation of a Hebbian cell assembly. Arrows represent a sequence of neural cell assemblies in which each cell assembly fires according to the numbers on
top of each arrow. Once each cell assembly in the loop has fired, the sequence begins again with the first cell assembly. Thus, sequential reactivation of cell assemblies
creates a loop (reverberating circuit) that could support sustained neuronal spiking activity during WM maintenance. Modified from [4]. (C,D) Example of a delayed
occulomotor response task. The cue indicates the spatial position on the screen that has to be memorized. After the delay period, the monkey has to perform a saccade
towards the memorized location. The enhancement of spike-field coherence during the delay period reflects the temporal alignment of anatomically distributed neurons
with the phase of local field potential (LFP) gamma-band activity. This relationship could underlie the formation of reverberating cell assemblies in parietal cortex during
WM retention. Modified from [13]. (E) Theta-based activation of cell assemblies in the hippocampus: electrode positions (upper panel) and raster plots (lower left and right
panels) of spiking activity for five neurons in the hippocampus. Vertical lines indicate troughs of theta activity (bottom trace). After reordering of the data via a stochastic
search for precise temporal relationships, individual cell assemblies become visible around the troughs of theta activity (right lower panel). Modified from [81].

correlation between the timing of neuronal spiking, the specifically involved in the temporal organization of WM
phase of oscillatory activity in local field potentials (LFPs), items. By contrast, oscillatory activity at alpha frequencies
and WM delay activity (Figure 1C–E) [13,14]. is not relevant for WM information per se, but has a critical
Evidence of a relationship between neuronal oscilla- role in protecting WM items from non-relevant informa-
tions and the maintenance of WM items in humans has tion. As a result, distinct forms of WM information rely on
been provided by electro-/magnetoencephalographic (EEG/ different oscillatory networks that are grouped through
MEG) and electrocorticographic (ECOG) recordings. These cross-frequency coupling (CFC) into different WM codes.
studies have reported enhanced amplitude and synchrony
of oscillatory activity and WM load-specific modulations at WM delay activity and gamma-band oscillations
different frequencies, in particular in theta (4–7 Hz), alpha Initial research focused on the relationship between gam-
(8–13 Hz), and gamma (30–200 Hz) ranges (Table 1). How- ma-band activity and perceptual processes [15]. However,
ever, the functional role of these different frequencies, as it soon became clear that high-frequency oscillations are
well as their relationship to distinct WM processes, has not restricted to visual responses but also occur during a
remained unclear. wide range of cognitive and executive processes, including
Here we propose a framework for integrating the differ- WM. Indeed, gamma-band oscillations occur in all cortical
ent findings on neuronal oscillations during WM mainte- areas and most subcortical structures and are particularly
nance. Specifically, we suggest that gamma-band prominent in superficial layers [16].
oscillations reflect a generic mechanism for active mainte- Experimental and theoretical studies suggest that gam-
nance of WM information, whereas theta-band activity is ma-band oscillations in the range 30–100 Hz emerge from
17
Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

Table 1. Overview of EEG/MEG and ECOG studies of WM maintenancea

Study Method WM paradigm Frequency Localization of WM effects
[44] EEG Sternberg task u (4–7.5 Hz) ACC
a (7.5–14 Hz) Occipito–parietal cortex
[20] EEG Visual DMS a (9–12 Hz) Occipital channels
b (15–20 Hz) Occipital and frontal channels
g (20–80 Hz) Occipital and temporal channels
[55] ECOG Sternberg task u (5–9 Hz) Frontal and temporal cortex
[56] MEG Sternberg task u (7–8.5 Hz) Frontal channels
[40] EEG Sternberg task a (9–12 Hz) Posterior and central channels
[26] MEG Auditory spatial g (50–70 Hz) Frontal, parietal, and temporal sensors
[25] MEG Auditory spatial g (55–80 Hz) Temporal and prefrontal sensors
[28] ECOG Sternberg task u (6.73 Hz) OTC, parietal cortex, and frontal cortex
g (38 Hz)
[41] EEG Visual DMS a (8–12 Hz) Posterior electrodes
[43] MEG Auditory Sternberg task u/a (6–12 Hz) Posterior, central, and left anterior electrodes
[45] MEG Visual DMS task a (8–13 Hz) Occipital lobe
g (30–100 Hz) Occipital lobe
[42] MEG Auditory DMS a (9–12 Hz) Superior temporal cortex
[22] MEG Memory saccade task a (8–12 Hz) Posterior parietal and occipital sensors
b (13–25 Hz) Posterior channels
g (60–90 Hz) Posterior channels
[23] MEG Auditory DMS g (50–90 Hz) Parieto–occipital sensors
[24] MEG Auditory DMS g (55–85 Hz) Frontal, occipital, and central sensors
[46] EEG Visual DMS u (4–8 Hz) Parieto–occipital sensors
a (8–12 Hz)
g (50–70 Hz)
[82] EEG Sternberg task u (3.7–7.5 Hz) Frontal electrodes
a (7.5–13.7 Hz) Posterior electrodes
[69] ECOG Sternberg task u (4–8 Hz) Hippocampus
g (25–100 Hz) Hippocampus
[27] MEG Somatosensory DMS a (8–14 Hz) Parietal cortex and right SII
g (65–150 Hz) SII and frontal cortex
[29] ECOG Sternberg task g (48–90 Hz) Hippocampus
[72] EEG Temporal order/DMS u (5–7 Hz) Prefrontal electrodes
a (9–12 Hz) Parietal/occipital electrodes
[33] MEG Visual DMS g (80–100 Hz) Parietal cortex
[31] MEG Visual DMS u (5–9 Hz) Distributed network of regions including
a (10–14 Hz) occipital, parietal, temporal, and frontal
b (15–30 Hz) cortex
g (30–150 Hz)
[21] MEG Visual DMS task a (10–14 Hz) Premotor cortex, left PFC, IPL
g (60–80 Hz)
[65] MEG Sternberg task a (8–12 Hz) Parieto–occipital sensors
[71] EEG Sternberg task u (5–8 Hz) Left frontal electrodes
a (9–12 Hz) Posterior
b (14–28 Hz) Left posterior
Visual DMS g (30–50 Hz) Right frontal electrodes
a
Abbreviations: ACC, anterior cingulate cortex; DMS, delayed match to sample task; IPL, intraparietal lobe; OTC, occipito–temporal cortex; PFC, prefrontal cortex; PPC,
posterior parietal cortex; SII, somatosensory cortex.

the interplay between GABAergic interneurons and glu- band activity may not reflect an oscillatory process but
tamatergic inputs mediated by NMDA and AMPA recep- instead could be more closely related to multi-unit activity
tors [17]. GABAergic interneurons, especially those [19].
expressing the calcium-binding protein parvalbumin First evidence of a link between the maintenance of WM
(PV), play a particularly important role in the generation items and gamma-band oscillations was reported by Tal-
of high-frequency oscillations because of their fast-spiking lon-Baudry and colleagues [20], who observed a sustained
characteristics and the short time constants of synaptic increase in oscillatory activity in the 24–60-Hz frequency
interactions mediated by these cells [18]. Gamma-band range in EEG recordings during a visual delayed-match-to-
oscillations above 100 Hz are likely to be mediated by sample (DMS) task. Subsequently, a large body of evidence
different generating mechanisms, such as gap-junction- has supported these findings for visuospatial [21,22], au-
mediated transmission [17]. Moreover, broad gamma- ditory [23–26], and somatosensory WM maintenance [27].
18
Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

Importantly, several studies demonstrated a parametric sis is supported by studies involving source localization of
relationship between the number of items to be memorized EEG/MEG data that have demonstrated an increase in
(WM load) and the amplitude of gamma-band oscillations alpha-band oscillations outside core regions of the WM
[21,28–31]. network [21,27]. Haegens et al. observed an increase in
In addition, recent advances in source localization of alpha band activity over task-irrelevant regions during a
EEG/MEG data [32] have allowed insights into the spatial somatosensory WM task that was strongest during suc-
layout of the underlying generators during WM. WM-load- cessful WM performance [27]. Moreover, Roux et al. ob-
related gamma-band activity has been localized to key served a positive correlation between 10–14-Hz delay
nodes of the WM network, including parietal and prefron- activity and reaction times, suggesting that increased
tal cortices [21,30,31,33]. This evidence is complemented alpha-band activity does not support efficient WM perfor-
by ECOG data highlighting the WM-load-dependent mod- mance [21].
ulation of gamma-band oscillations in frontal [28] and
hippocampal regions [29]. Distinct roles for gamma- versus alpha-band oscillations
Finally, a functional relationship between gamma-band during WM?
activity and maintenance of WM items is supported by the Manipulation of WM load alone may not be sufficient to
fact that (i) different gamma-band frequencies underlie the disentangle the relationship between alpha/gamma-band
maintenance of distinct WM items [23], (ii) gamma-band oscillations and WM maintenance because an increase in
activity correlates with behavioral indexes of WM perfor- the number of WM items is also associated with increases
mance [21,24], and (iii) single-trial fluctuations of gamma- in attentional demands [48], which can influence oscillato-
band activity predict the number of WM items [21,34]. ry activity [49]. Therefore, experimental designs may be
required that allow differentiation between the manipula-
WM delay activity and alpha-band oscillations tion of attentional resources and WM load, for example
Alpha oscillations (8–13 Hz) were the first rhythm to be through paradigms in which distracting information is
discovered by Berger in 1924 [35]. One brain region that is introduced while the number of task-relevant WM items
crucially involved in the generation of alpha-band oscilla- is kept constant.
tions is the thalamus [36]. In addition, alpha oscillations In a study by Sauseng and colleagues, task-relevant and
have been recorded in subcortical areas, such as the hip- task-irrelevant items were indicated by a cue that pointed
pocampus and the reticular formation, and are particularly towards either the right or the left side of a visual array,
prominent in the deeper layers (layer V) [16]. thus making information in either the left or the right
It is well established that alpha rhythms result from hemifield task-irrelevant (Figure 2A) [46]. The results
reciprocal interactions between excitatory and inhibitory showed that alpha-band activity was modulated by the
neurons whereby the synchronization is also stabilized by number of irrelevant items (distractors) and lateralized to
gap junctions among inhibitory interneurons. The suscep- the task-irrelevant hemifield (Figure 2B). Crucially, repet-
tibility of these networks to engagement in alpha rhythms itive transcranial magnetic stimulation (rTMS) at alpha
is in turn modulated by cholinergic and serotonergic mech- frequencies (10 Hz) over the task-irrelevant hemisphere
anisms and by glutamatergic afferents acting via metabo- was associated with an improvement in WM capacity
tropic receptors [37]. (Figure 2C), thereby supporting the possibility of a causal
Although initially thought to reflect cortical idling, a relationship between alpha oscillations and the inhibition
growing body of evidence critically implicates rhythmic of task-irrelevant information.
activity in the alpha band in cortical communication and Similarly, Roux et al. tested whether the amplitude of
cognition [38,39]. According to the inhibition timing hy- oscillatory activity in a particular frequency band reflects
pothesis [39], alpha oscillations may play an active role the maintenance of task-relevant WM items by comparing
during cognitive processes through the inhibition of task- the amplitude of oscillatory activity in response to three
irrelevant brain regions, for example. relevant items, three relevant and three irrelevant items
In agreement with this perspective, alpha-band activity (distractor condition), and six relevant items in a DMS
has also been prominently observed during WM tasks. paradigm (Figure 2C) [21]. The results showed a specific
Jensen et al. examined EEG data during the retention relationship between the amplitude of 60–80-Hz activity
interval for a modified Sternberg task and found pro- and the number of relevant WM items (Figure 2D–G),
nounced activity at 9–12 Hz [40]. Similar to gamma-band which was not observed for alpha-band oscillations. The
activity, alpha-band oscillations have been observed in distinct relationship between alpha/gamma-band oscilla-
several sensory modalities during WM maintenance tions and maintenance of relevant WM items was further
[27,41,42]. Moreover a load-dependent increase has been supported by source localization results, which indicated
demonstrated [43], although evidence of an opposite rela- that 60–80-Hz activity occurred in parietal and prefrontal
tionship has been reported as well [44]. cortices and covaried with WM load, whereas 10–14-Hz
Despite the consistent involvement of alpha-band oscil- activity was localized to the premotor cortex, suggesting a
lations in WM maintenance, the functional interpretation role for alpha-band oscillations in the inhibition of motor
is conflicting. Although some studies have related delay plans during the delay period.
activity in the alpha band to the functional inhibition of
task-irrelevant brain areas [22,40,45,46], others have Theta-band oscillations during WM maintenance
linked alpha-band activity directly to processes underly- An additional important observation from our study [21]
ing WM maintenance [31,41,43,47]. The former hypothe- and previous work [27,31,45] is the absence of theta-band
19
 Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

(A) (B) Load 2 Load 3 Load 4 Load 6

Cue

Memory -1.5 [μV2] 1.5

array
(C) 3.5 *
200 * Key:
Delay * real rTMS
100 3.0

Memory capacity [K]

sham rTMS
Test
array
900
2.5

Tim
e[ 2000
ms 2.0
]

1.5
Ipsilateral Contralateral Vertex

Baseline Encoding Delay Retrieval Response

(D)

400 120 400

ms 0m ms
s

Le BA9
70 +/- 10 Hz
Key:
Load 3
(E) (Gamma) Distractor
Load 6
(G) 5 (%)
120
Frequency [Hz]

8.5 0.4 s
p < 0.001
Change [%]

80 Delay
Rel.

40 Right IPL
-8.5 70 +/- 10 Hz
(Gamma)
-0.5 0 0.4 1.6 2
(H) 5 (%)

(F) 0.4 s p < 0.001

Delay
18
Frequency [Hz]

35
Right BA6
Change [%]

14
12 +/- 2 Hz
Rel.

(Alpha)
10
(I) 40 (%)
-35
-0.5 0 0.4 1.6 2 0.4 s p < 0.001
Time [s]
Delay
TRENDS in Cognitive Sciences

Figure 2. Alpha/gamma-band activity during working memory (WM) maintenance. (A) WM task used by Sauseng et al. [46]. A cue is presented at the beginning of each trial
that indicates in which part of the screen participants have to memorize the presented items. After a delay period, participants are instructed to report whether the items
presented in the WM array have changed by pressing a button. (B) Topographies of alpha power showing modulation of lateralized (ipsilateral minus contralateral) alpha-
band activity with increasing WM load. Alpha-band power is consistently enhanced over electrodes in the ipsilateral hemisphere compared to those contralateral to the
attended side. (C) Effects of repetitive transcranial magnetic stimulation (rTMS) at alpha frequencies (10 Hz) on WM capacity. rTMS over the ipsilateral hemisphere improves
WM capacity compared to sham TMS and the control condition. By contrast, rTMS over the contralateral hemisphere leads to a reduction in WM capacity compared to
sham TMS and the control condition. Panels B and C modified from [46]. (D) Visual–spatial WM task used by Roux et al. [21]. Participants were instructed to memorize the
spatial positions of red items only and to ignore the blue items. (E,F) Spectrograms of sensor activity for (E) lower and (F) higher frequencies showing enhanced oscillatory
activity at alpha (10–14 Hz) and gamma (60–80 Hz) frequencies during the delay period. Pseudo-colors indicate the relative increase in task-related activity compared to
baseline activity. (G,H) Time courses of 60–80-Hz activity in the prefrontal cortex (BA9) and intraparietal lobe (IPL). The source reconstruction of delay activity suggests the
possibility that gamma power in the IPL and PFC reflects different aspects of WM maintenance. This is supported by the fact that gamma-band activity in parietal cortex was
modulated by the number of stimuli presented in the memory array but not by the number of relevant items, which suggests that modulation of 60–80-Hz activity reflects
scanning and maintenance of the spatial positions of possible WM locations. (I) Time courses of 10–14-Hz activity in the premotor cortex (BA6) showing that alpha-band
activity was not modulated by the number of task-relevant items. Panels E–I are modified from [21].

20
 Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

activity, raising the question whether theta-band activity

(A) Test item
is relevant for WM processes.
Response
Theta oscillations (4–8 Hz) represent one of best-studied
Next trial rhythms in the mammalian brain [50]. Theta oscillations
+ B M F S M + are particularly prominent in the hippocampus–enthorinal
Key system, but also occur in extrahippocampal regions, such
Item list Delay Feedback as the prefrontal cortex, amygdala, and visual cortex. In
(B) the hippocampus, theta oscillations are generated by an
20 6 x10 6
interplay of glutamatergic and GABAergic neurons. In
addition, GABAergic inputs are modulated by cholinergic

Spectral power
Frequency [Hz]

15 inputs from the septum, which possibly acts as a pacemak-

er for theta activity [51].
10 Studies in rodents revealed a close relationship between
the occurrence of theta oscillations in the hippocampus,
5
locomotion, and the place-specific firing of hippocampal
0 pyramidal cells (place cells), which have been implicated
0 5 10
(C) Time [s] in the coding of spatial positions through phase precession
[52]. In addition, a large body of work indicates that the
2 items
4.5 hippocampus plays a crucial role in the formation and recall
Log10(θ-Power)

3 items
4 items
of episodic and declarative memory [53]. More specifically, it
has been proposed that theta-band oscillations are related to
WM capacity limits of approximately seven items [54].
3.5 Consistent with this evidence, several studies have
revealed pronounced theta-band activity during WM main-
-2 0 2 4 6 8 tenance. Raghavachari et al. obtained ECOG data from
Time [s]
epileptic patients while they performed a Sternberg task
[55]. Time–frequency analysis indicated the occurrence of
(D) theta-band activity in several cortical sites throughout the
task, including during the delay period, which increased
Gamma [38.05Hz] Power

1.5 systematically with WM load. These results were supported

by a MEG study in which frontal theta activity increased
parametrically with the number of items retained in WM
1
(Figure 3A,B) [56], and Moran et al. observed a complex
3 items 4 items relationship between theta-band activity in hippocampal,
2 items frontal, and parietal regions and WM capacity [57].
0.5
1 2 3 4 5 6
Time bin Long-range synchronization during WM maintenance
In addition to amplitude modulation of neural oscillations
within local circuits, long-range synchronization between
(E)
neuronal assembles, as assessed through phase or coher-
ence measures, has been demonstrated in several studies
Frequency for amplitude [Hz]

1 Item 2 Items 4 Items

60 0.5 [30,47,58,59]. Long-range synchronization could support
Modulaon index

50 computations during WM maintenance within circum-

40 scribed brain regions by establishing transient neural
communication to facilitate information transfer within
30
distributed WM networks [60]. Consistent with this hy-
20 pothesis, there is cumulative evidence of synchronization
0
5 10 5 10 5 10 at theta, alpha, and gamma frequencies during WM main-
Frequency for phase [Hz] tenance that has been related to top-down control [59],
TRENDS in Cognitive Sciences manipulation of WM items [61], integration of WM sub-
processes [62], and synaptic plasticity [60].
Figure 3. Theta–gamma band activity during working memory (WM) maintenance.
(A) Schematic representation of a Sternberg paradigm used by Raghavachari et al.
[55]. Participants have to memorize a list of letters that are presented sequentially. Neural oscillations and distinct WM memory contents
After a brief delay period, a probe item is presented and participants have to
The current evidence suggests that oscillations at theta-,
indicate whether the probe belongs to the list of memorized items. (B)
Spectrogram of electrocorticographic (ECOG) signals recorded in human alpha- and gamma-band frequencies are the dominant
participants showing enhanced oscillatory activity at theta frequencies (6–9 Hz)
during the delay period (black bar). (C) Time courses of theta power for differing
list length. Theta power remains elevated above baseline levels throughout the sizes. Panel D modified from [28]. (E) Empirical evidence of the occurrence of
sequential encoding of individual letters (gray bar) and during the retention phase theta–gamma coupling during the maintenance of multiple sequentially ordered
(black bar). Importantly, a list size effect can be detected from the modulation of WM items. Co-modulograms of ECOG signals indicate list-length-dependent
theta power during the encoding and delay phases. Panels B and C modified from modulation of theta–gamma phase–amplitude coupling (PAC) during the delay
[55]. (D) Time course of gamma-band activity as a function of list size. A similar period of a Sternberg task. Pseudo-colors denote the strength of PAC. Panel E
effect can be observed in the modulation of gamma-band activity for different list modified from [69].

21
Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

rhythms engaged during WM maintenance, albeit with

Box 1. Cross-frequency coupling
potentially important differences with regard to their
functional role and task-dependent modulation. An over- WM may constitute a prominent example of a cognitive process in
which cross-frequency coupling (CFC) occurs between two fre-
view of recent studies that investigated neural oscilla-
quency ranges, an issue that is attracting increased attention in
tions during WM delay activity (Table 1) suggests that cognitive neuroscience [66] because it may play a crucial role in
theta activity occurs preferentially in tasks that involve neuronal communication. This is supported by numerous observa-
sequential coding of multiple WM items, such as during tions that slower brain rhythms occur simultaneously with faster
the Sternberg paradigm (Figure 3A), whereas alpha oscil- oscillations during WM [18,20,21,23], and different forms of CFC
may exist between these rhythms [3,71]. One form of CFC is phase–
lations tend to occur during tasks that require mainte- amplitude coupling (PAC), whereby the amplitude of a higher
nance of simultaneously presented visual or spatial frequency is modulated by the phase of a slower rhythm. Other
information. forms include amplitude–amplitude coupling and n:m phase locking
This distinction may be considered compatible with [71], which refer to power and phase co-fluctuations between
evidence on the two specialized storage systems for WM different frequencies. For the generation of theta–gamma and
alpha–gamma codes during WM delay activity, we consider both
information: the phonological loop and the visuospatial phase–phase and PAC as possible ways to establish cross-frequency
sketch pad [1]. The former deals with the rehearsal of interactions. Importantly, the modulation of local high-frequency
verbal/acoustic information and consists of two parts: a activity through the phase of a slower rhythm has been proposed as
short-term phonological store for auditory memory traces a putative mechanism for the synchronization of fast oscillations
across distant sites [3].
that are subject to rapid decay, and a subvocal rehearsal
component or articulatory loop that can consolidate mem-
ory traces through sequential reactivation. Accordingly, structures [5] and thus could represent an effective repre-
sequential encoding, such as during Sternberg tasks, could sentational format for WM information.
activate the phonological loop to support the maintenance
of sequenced WM items through subvocal rehearsal, which Distinct theta/gamma and alpha/gamma codes for WM
results in a corresponding increase in theta activity during information?
the delay period. By contrast, it has been proposed that the Importantly, theta/alpha rhythms do not occur in isolation
visuospatial sketchpad is responsible for the storage and but are frequently accompanied by activity in higher fre-
rehearsal of visual and spatial WM items. In analogy to its quency ranges, in particular at gamma frequencies
phonological counterpart, the sketchpad contains a storage [21,27,31,45]. This observation is consistent with increas-
component, the visual cache, and a rehearsal component, ing evidence that neural oscillations form a hierarchical
the inner scribe [1]. system in which different rhythms exhibit CFC [66], sug-
This multicomponent model of WM is consistent with gesting that populations of neurons oscillating at different
the distinct functional roles and anatomical regions in- frequencies interact with each other, forming nested as-
volved in the generation of theta/alpha-band oscillations. semblies (Box 1) [63].
Theta activity is prominently involved not only in the Cross-frequency interactions between theta- and gam-
hippocampal–entorhinal system but also in cortical ma-band oscillations are a prominent feature not only of
regions such as the prefrontal cortex. One of the core hippocampal circuits [67] but also of other structures [68]
functions attributed to theta-band oscillations in the hip- and thus could provide a code for representing multiple
pocampal system is the temporal integration of cell assem- and sequentially ordered WM items in which cycles of
blies [52]. Although first demonstrated for the tracking of gamma-band oscillations are coordinated through an un-
spatial positions, the same mechanism may also support derlying theta rhythm. Evidence of such a theta–gamma
the representation and consolidation of sequentially orga- code [63] during WM has been reported by Axmacher et al.
nized memory traces [63]. [69]. The authors examined intracranial EEG recordings in
Alpha-band activity, by contrast, is most frequently human epilepsy patients to demonstrate that maintenance
observed in sensory regions and the thalamus, and is of multiple items in WM is accompanied by load-dependent
prominently involved in the discrete sampling and gating theta–gamma coupling in the hippocampus (Figure 3E).
of visual scenes [64]. Accordingly, modulation of alpha Similarly, Holz et al. observed enhanced theta–gamma
activity may enable the gating of task-relevant WM items, coupling for correctly identified visual WM items in EEG
for example during suppression of distracting information data in a DMS paradigm [70].
during WM encoding and maintenance. In this context, In contrast to theta–gamma interactions, evidence of an
recruitment of alpha-band oscillations for inhibition of alpha–gamma code during DMS tasks that involve main-
distracting information can also occur during tasks requir- tenance of sensory information has not yet been investi-
ing the maintenance of sequential information [65], gated. However, evidence of a differential relationship
highlighting that theta/alpha frequencies are closely between WM content and oscillatory activity at theta
linked to the task demands during WM and not limited and alpha frequencies has been provided by recent EEG
to the format of WM representations. studies [71,72]. Specifically, these data show that if the
In this framework, gamma-band oscillations represent a content of WM is changed from multiple sequentially
generic mechanism for the representation of individual ordered items to discrete visual [47] or spatial information
WM items, irrespective of WM content and format. This [49], theta activity is replaced by alpha activity.
is because the synchronization of neuronal discharges at Another important finding from these studies is that
gamma frequencies supports the integration of neurons theta and alpha activities were localized over distinct EEG
into cell assemblies in different cortical and subcortical recordings sites, thereby supporting the view that WM
22
 Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

due to the greater number of gamma cycles that can be

Box 2. Outstanding questions accommodated within a theta versus alpha cycle. Indeed,
there is preliminary evidence to suggest that WM capacity
 What is the role of long-range synchronization in WM and how limitations are directly related to the nesting of theta/
does it relate to CFC? gamma-band oscillations [54].
 Is beta-band activity relevant for WM and what is its function?
 Which oscillation frequencies are important for manipulation of
WM items? WM and neural oscillations: towards an integrated
 What is the relationship between theta and gamma/alpha codes paradigm
during WM maintenance and encoding and retrieval stages of Although a large body of psychophysical studies and inves-
WM? tigations with fMRI and event-related potentials (ERPs) of
 What is role of subcortical structures, such as the thalamus, in the
WM processes have been conducted [3,48,75–78], we be-
generation of neural oscillations during WM maintenance?
 Has theta activity in the rodent hippocampus a similar role in lieve that further inquiries into the involvement of rhyth-
human memory processes? mic activity at different temporal and spatial scales could
add an important, complementary perspective on the
mechanisms, subprocesses, and brain regions involved in
content may also influence the recruitment of distinct WM WM (Box 2).
networks. This hypothesis is supported by evidence from This is because neural oscillations represent a physio-
ECOG recordings showing that theta–gamma phase–am- logical signal that reflects the ensemble activity of neurons
plitude coupling (PAC) occurs preferentially in fronto– that have been hypothesized to underlie WM maintenance.
temporal regions during non-visual tasks, whereas al- In addition, time–frequency analyses allow a decomposi-
pha–gamma PAC is most likely observed in parietal–oc- tion of distinct spectral bands that can be linked to differ-
cipital networks during the processing of visual ent cognitive processes involved in WM. This is supported
information [73]. by the current review that suggests distinct roles of differ-
The possibility that theta–gamma versus alpha–gamma ent oscillation frequencies in WM subprocesses.
codes underlie distinct WM contents is also supported by Finally, an important advantage of neural oscillations is
psychological evidence of different capacity limits for se- that comparable indexes of rhythmic activity can be
quentially versus simultaneously presented WM material obtained across different species and spatial scales, which
[74]. It is conceivable that this difference is a manifestation allows for direct comparisons between human data as
of the varying length of the gamma duty cycle, that is, the assessed with EEG/MEG and invasive recordings in
number of gamma cycles that can occur within a fixed rodents, for example. This possibility can also facilitate
temporal window [38]. In other words, the increased ca- in searching for the mechanisms underlying neural oscilla-
pacity for sequentially presented WM material could be tions and associated cognitive processes, and thus in

Input Frequency Network Process

Allocaon of aenon
PFC
Readout of
relevant items
Visual-spaal Alpha-Gamma PMC
Inhibion of task-
irrelevant items
PC

Consolidaon
PFC
Segmentaon
of informaon
Sequenal Theta-Gamma MTL
Rehearsal
HI

TRENDS in Cognitive Sciences

Figure 4. Oscillatory networks for working memory (WM) maintenance. Overview of oscillatory networks for sequential and sensory WM information. For sensory
information, alpha–gamma networks occur predominantly in the prefrontal cortex (PFC) and parietal cortex (PC) and subcortical regions such as the thalamus (PMC), which
has been prominently implicated in the gating of sensory information and the generation of alpha oscillations [21,36,79]. In this framework, gamma-band activity underlies
the active maintenance and read-out of relevant WM items, whereas alpha-band activity is predominantly involved in the inhibition of task-irrelevant WM items. For the
coding of sequential information, theta–gamma band oscillations occur predominantly in medial temporal lobe and frontal cortices. It is important to note that these
networks can overlap, as in cases in which distracting information is introduced into a sequential WM task.

23
Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

testing of mechanistic relationships between physiological 11 Lisman, J. and Buzsaki, G. (2008) A neural coding scheme formed by
the combined function of gamma and theta oscillations. Schizophr.
and behavioral phenomena.
Bull. 34, 974–980
12 Bosman, C.A. et al. (2012) Attentional stimulus selection through
Concluding remarks selective synchronization between monkey visual areas. Neuron 75,
We have provided an overview of the potential role of 875–888
distinct oscillation frequencies in WM information. Over- 13 Pesaran, B. et al. (2002) Temporal structure in neuronal activity during
working memory in macaque parietal cortex. Nat. Neurosci. 5, 805–811
all, there is consistent evidence to suggest that theta-,
14 Siegel, M. et al. (2009) Phase-dependent neuronal coding of objects in
alpha-, and gamma-band frequencies are functionally rel- short-term memory. Proc. Natl. Acad. Sci. U.S.A. 106, 21341–21346
evant for maintenance of WM items, as evidenced by close 15 Uhlhaas, P.J. et al. (2009) Neural synchrony in cortical networks:
relations with WM capacity and detection rates. More history, concept and current status. Front. Integr. Neurosci. 3, 17
specifically, the data suggest that different types of WM 16 Buffalo, E.A. et al. (2011) Laminar differences in gamma and alpha
coherence in the ventral stream. Proc. Natl. Acad. Sci. U.S.A. 108,
information require distinct oscillatory codes. Accordingly, 11262–11267
maintenance of sensory–spatial WM items involves an 17 Buzsaki, G. and Wang, X.J. (2012) Mechanisms of gamma oscillations.
alpha–gamma code, whereas sequential WM information Annu. Rev. Neurosci. 35, 203–225
relies on theta/gamma-band oscillations, which also 18 Sohal, V.S. et al. (2009) Parvalbumin neurons and gamma rhythms
emerge from different functional networks (Figure 4). This enhance cortical circuit performance. Nature 459, 698–702
19 Uhlhaas, P.J. et al. (2011) A new look at gamma? High- (>60 Hz)
hypothesis is supported by recent evidence suggesting that gamma-band activity in cortical networks: function, mechanisms and
alpha–gamma coupling may be related to thalamo–cortical impairment. Prog. Biophys. Mol. Biol. 105, 14–28
circuits [79], whereas theta–gamma coupling has been 20 Tallon-Baudry, C. et al. (1998) Induced gamma-band activity during
associated with fronto–hippocampal networks [63]. the delay of a visual short-term memory task in humans. J. Neurosci.
18, 4244–4254
Testing of this hypothesis could involve experiments
21 Roux, F. et al. (2012) Gamma-band activity in human prefrontal cortex
with different types of WM information to demonstrate codes for the number of items maintained during working memory. J.
dynamic switching between theta–gamma and alpha– Neurosci. 32, 12411–12420
gamma networks, accompanied by advanced source analy- 22 Medendorp, W.P. et al. (2007) Oscillatory activity in human parietal
sis to identify the underlying generators. Moreover, recent and occipital cortex shows hemispheric lateralization and memory effects
in a delayed double-step saccade task. Cereb. Cortex 17, 2364–2374
advances in the entrainment of neural oscillations through
23 Kaiser, J. et al. (2008) Distinct gamma-band components reflect the
either transcranial magnetic or transcranial alternating short-term memory maintenance of different sound lateralization
current stimulation (TMS/tACS) stimulation indicate that angles. Cereb. Cortex 18, 2286–2295
it is also possible to bias rhythmic activity and associated 24 Kaiser, J. et al. (2009) Task- and performance-related modulation of
cognitive processes [80]. Accordingly, differential entrain- domain-specific auditory short-term memory representations in the
gamma-band. Neuroimage 46, 1127–1136
ment at theta/alpha/gamma frequencies can be used to bias 25 Kaiser, J. et al. (2003) Dynamics of gamma-band activity in human
specific WM processes and key nodes of the WM network to magnetoencephalogram during auditory pattern working memory.
establish causal relations between neural oscillations and Neuroimage 20, 816–827
cognition. 26 Lutzenberger, W. et al. (2002) Dynamics of gamma-band activity
during an audiospatial working memory task in humans. J.
Neurosci. 22, 5630–5638
Acknowledgments 27 Haegens, S. et al. (2010) Somatosensory working memory performance
This work was supported by the LOEWE Grant ‘Neuronale Koordination in humans depends on both engagement and disengagement of regions
Forschungsschwerpunkt Frankfurt’. in a distributed network. Hum. Brain Mapp. 31, 26–35
28 Howard, M.W. et al. (2003) Gamma oscillations correlate with working
References memory load in humans. Cereb. Cortex 13, 1369–1374
1 Baddeley, A. (2012) Working memory: theories, models, and 29 van Vugt, M.K. et al. (2010) Hippocampal gamma oscillations increase
controversies. Annu. Rev. Psychol. 63, 1–29 with memory load. J. Neurosci. 30, 2694–2699
2 Wager, T.D. and Smith, E.E. (2003) Neuroimaging studies of 30 Palva, J.M. et al. (2010) Neuronal synchrony reveals working memory
working memory: a meta-analysis. Cogn. Affect. Behav. Neurosci. 3, networks, predicts individual memory capacity. Proc. Natl. Acad. Sci.
255–274 U.S.A. 107, 7580–7585
3 Luck, S.J. and Vogel, E.K. (2013) Visual working memory capacity: 31 Palva, S. et al. (2011) Localization of cortical phase and amplitude
from psychophysics and neurobiology to individual differences. Trends dynamics during visual working memory encoding and retention. J.
Cogn. Sci. 17, 391–400 Neurosci. 31, 5013–5025
4 Hebb, D.O. (1949) The Organization of Behaviour: A 32 Palva, S. and Palva, J.M. (2012) Discovering oscillatory interaction
Neuropsychological Theory, Wiley networks with M/EEG: challenges and breakthroughs. Trends Cogn.
5 Singer, W. (2009) Distributed processing and temporal codes in Sci. 16, 219–230
neuronal networks. Cogn. Neurodyn. 3, 189–196 33 Linden, D.E. et al. (2012) Mapping brain activation and information
6 Salazar, R.F. et al. (2012) Content-specific fronto-parietal during category-specific visual working memory. J. Neurophysiol. 107,
synchronization during visual working memory. Science 338, 1097– 628–639
1100 34 Polania, R. et al. (2012) Noninvasively decoding the contents of visual
7 Havenith, M.N. et al. (2011) Synchrony makes neurons fire in sequence, working memory in the human prefrontal cortex within high-gamma
and stimulus properties determine who is ahead. J. Neurosci. 31, 8570– oscillatory patterns. J. Cogn. Neurosci. 24, 304–314
8584 35 Berger, H. (1929) Ueber das Elektroencephalogramm des Menschen.
8 Pastalkova, E. et al. (2008) Internally generated cell assembly Arch. Psychiatr. Nervenkrankheiten 87, 527–570
sequences in the rat hippocampus. Science 321, 1322–1327 36 Roux, F.W. et al. (2012) The phase of thalamic alpha activity modulates
9 Buschman, T.J. et al. (2012) Synchronous oscillatory neural ensembles cortical gamma-band activity: evidence from resting state MEG
for rules in the prefrontal cortex. Neuron 76, 838–846 recordings. J. Neurosci. 32, 12411–12420
10 Gray, C.M. et al. (1989) Oscillatory responses in cat visual cortex 37 Uhlhaas, P.J. et al. (2008) The role of oscillations and synchrony in
exhibit inter-columnar synchronization which reflects global cortical networks and their putative relevance for the pathophysiology
stimulus properties. Nature 338, 334–337 of schizophrenia. Schizophr. Bull. 34, 927–943

24
Opinion Trends in Cognitive Sciences January 2014, Vol. 18, No. 1

38 Jensen, O. and Mazaheri, A. (2010) Shaping functional architecture by 61 Schack, B. et al. (2005) Phase synchronization between theta and upper
oscillatory alpha activity: gating by inhibition. Front. Hum. Neurosci. 4, alpha oscillations in a working memory task. Int. J. Psychophysiol. 57,
186 105–114
39 Klimesch, W. et al. (2007) EEG alpha oscillations: the inhibition-timing 62 Sauseng, P. et al. (2010) Control mechanisms in working memory: a
hypothesis. Brain Res. Rev. 53, 63–88 possible function of EEG theta oscillations. Neurosci. Biobehav. Rev.
40 Jensen, O. et al. (2002) Oscillations in the alpha band (9–12 Hz) 34, 1015–1022
increase with memory load during retention in a short-term 63 Lisman, J.E. and Jensen, O. (2013) The theta-gamma neural code.
memory task. Cereb. Cortex 12, 877–882 Neuron 77, 1002–1016
41 Herrmann, C.S. et al. (2004) Phase-locking and amplitude modulations 64 VanRullen, R. and Koch, C. (2003) Is perception discrete or continuous?
of EEG alpha: two measures reflect different cognitive processes in a Trends Cogn. Sci. 7, 207–213
working memory task. Exp. Psychol. 51, 311–318 65 Bonnefond, M. and Jensen, O. (2012) Alpha oscillations serve to protect
42 Kaiser, J. et al. (2007) Alpha synchronization during auditory spatial working memory maintenance against anticipated distracters. Curr.
short-term memory. Neuroreport 18, 1129–1132 Biol. 22, 1969–1974
43 Leiberg, S. et al. (2006) Effects of memory load on cortical oscillatory 66 Canolty, R.T. and Knight, R.T. (2010) The functional role of cross-
activity during auditory pattern working memory. Brain Res. 1120, frequency coupling. Trends Cogn. Sci. 14, 506–515
131–140 67 Belluscio, M.A. et al. (2012) Cross-frequency phase-phase coupling
44 Gevins, A. et al. (1997) High-resolution EEG mapping of cortical between theta and gamma oscillations in the hippocampus. J.
activation related to working memory: effects of task difficulty, type Neurosci. 32, 423–435
of processing, and practice. Cereb. Cortex 7, 374–385 68 van der Meij, R. et al. (2012) Phase–amplitude coupling in human
45 Jokisch, D. and Jensen, O. (2007) Modulation of gamma and alpha electrocorticography is spatially distributed and phase diverse. J.
activity during a working memory task engaging the dorsal or ventral Neurosci. 32, 111–123
stream. J. Neurosci. 27, 3244–3251 69 Axmacher, N. et al. (2010) Cross-frequency coupling supports multi-
46 Sauseng, P. et al. (2009) Brain oscillatory substrates of visual short- item working memory in the human hippocampus. Proc. Natl. Acad.
term memory capacity. Curr. Biol. 19, 1846–1852 Sci. U.S.A. 107, 3228–3233
47 Sauseng, P. et al. (2005) EEG alpha synchronization and functional 70 Holz, E.M. et al. (2010) Theta-gamma phase synchronization during
coupling during top-down processing in a working memory task. Hum. memory matching in visual working memory. Neuroimage 52, 326–335
Brain Mapp. 26, 148–155 71 Roberts, B.M. et al. (2013) Oscillatory activity during maintenance of
48 Mayer, J.S. et al. (2007) Common neural substrates for visual working spatial and temporal information in working memory.
memory and attention. Neuroimage 36, 441–453 Neuropsychologia 51, 349–357
49 Fries, P. et al. (2001) Modulation of oscillatory neuronal synchro- 72 Hsieh, L.T. et al. (2011) Neural oscillations associated with item and
nization by selective visual attention. Science 291, 1560–1563 temporal order maintenance in working memory. J. Neurosci. 31,
50 Colgin, L.L. (2013) Mechanisms and functions of theta rhythms. Annu. 10803–10810
Rev. Neurosci. 36, 295–312 73 Voytek, B. et al. (2010) Shifts in gamma phase-amplitude coupling
51 Pignatelli, M. et al. (2012) Neural circuits underlying the generation of frequency from theta to alpha over posterior cortex during visual tasks.
theta oscillations. J. Physiol. Paris 106, 81–92 Front. Hum. Neurosci. 4, 191
52 Buzsaki, G. and Moser, E.I. (2013) Memory, navigation and theta 74 Cowan, N. (2001) The magical number 4 in short-term memory: a
rhythm in the hippocampal-entorhinal system. Nat. Neurosci. 16, reconsideration of mental storage capacity. Behav. Brain Sci. 24, 87–114
130–138 75 Anderson, D.E. et al. (2013) A common discrete resource for visual
53 Lega, B.C. et al. (2012) Human hippocampal theta oscillations and the working memory and visual search. Psychol. Sci. 24, 929–938
formation of episodic memories. Hippocampus 22, 748–761 76 Emrich, S.M. et al. (2013) Distributed patterns of activity in sensory
54 Jensen, O. and Lisman, J.E. (1998) An oscillatory short-term memory cortex reflect the precision of multiple items maintained in visual
buffer model can account for data on the Sternberg task. J. Neurosci. short-term memory. J. Neurosci. 33, 6516–6523
18, 10688–10699 77 Riggall, A.C. and Postle, B.R. (2012) The relationship between working
55 Raghavachari, S. et al. (2001) Gating of human theta oscillations by a memory storage and elevated activity as measured with functional
working memory task. J. Neurosci. 21, 3175–3183 magnetic resonance imaging. J. Neurosci. 32, 12990–12998
56 Jensen, O. and Tesche, C.D. (2002) Frontal theta activity in humans 78 Bledowski, C. et al. (2010) Basic operations in working memory:
increases with memory load in a working memory task. Eur. J. contributions from functional imaging studies. Behav. Brain Res.
Neurosci. 15, 1395–1399 214, 172–179
57 Moran, R.J. et al. (2010) Peak frequency in the theta and alpha bands 79 Saalmann, Y.B. et al. (2012) The pulvinar regulates information
correlates with human working memory capacity. Front. Hum. transmission between cortical areas based on attention demands.
Neurosci. 4, 200 Science 337, 753–756
58 Sauseng, P. et al. (2004) Theta coupling in the human 80 Thut, G. et al. (2011) Entrainment of perceptually relevant brain
electroencephalogram during a working memory task. Neurosci. oscillations by non-invasive rhythmic stimulation of the human
Lett. 354, 123–126 brain. Front. Psychol. 2, 170
59 Crespo-Garcia, M. et al. (2013) Working memory processes are 81 Harris, K.D. et al. (2003) Organization of cell assemblies in the
mediated by local and long-range synchronization of alpha hippocampus. Nature 424, 552–556
oscillations. J. Cogn. Neurosci. 25, 1343–1357 82 Scheeringa, R. et al. (2009) Trial-by-trial coupling between EEG and BOLD
60 Fell, J. and Axmacher, N. (2011) The role of phase synchronization in identifies networks related to alpha and theta EEG power increases
memory processes. Nat. Rev. Neurosci. 12, 105–118 during working memory maintenance. Neuroimage. 44, 1224–1238

25