Bernstein Center for Computational Neuroscience and Berlin Center for Advanced Neuroimaging and Clinic for Neurology, Charité Universitätsmedizin,
corporate member of Freie Universität Berlin, Humboldt Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany.
Berlin School of Mind and
Brain, Humboldt Universität, Berlin, Germany.
Cluster of Excellence NeuroCure, Charité Universitätsmedizin, corporate member of Freie Universität
Berlin, Humboldt Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany.
Department of Psychology, Humboldt Universität zu Berlin, Berlin,
SFB 940 Volition and Cognitive Control, Technische Universität Dresden, Dresden, Germany.
These authors contributed equally:
Thomas B. Christophel and Polina Iamshchinina. *e-mail: firstname.lastname@example.org
Items held in working memory can be either attended or not,
depending on their current behavioral relevance. It has been
suggested that unattended contents might be solely retained
in an activity-silent form. Instead, we demonstrate here that
encoding unattended contents involves a division of labor.
While visual cortex only maintains attended items, intrapari-
etal areas and the frontal eye fields represent both attended
and unattended items.
The short-term retention of sensory stimuli in working memory
is fundamental to human cognition
. A wide range of primate elec-
trophysiology and human imaging studies have reported content-
selective brain signals that encode working memory contents across
. Such persistent, stimulus-selective activity has been
observed in multiple regions across the cortical sheet, including
sensory, parietal and frontal regions
. Recently, however, it has been
postulated that working memory can be retained in an ‘activity-
. In this model, working memory contents are believed
to be retained by changes in synaptic weights rather than neuro
. In line with this, several studies have recently reported
absence of persistent stimulus-selective activity when items are
held in memory, but are currently not behaviorally relevant
currently nonprioritized items are frequently referred to as ‘unat
tended memory items’ (UMIs), as opposed to ‘attended memory
. These results suggest that attended memory items
are retained actively while unattended memory items are retained
in an activity-silent form.
However, the absence of content-selective signals for unattended
items observed in prior work
might reflect a lack of sensitivity in
the experimental procedures. For example, these studies used small
numbers of subjects, trained their classification models on attended
items in separate one-item tasks and only analyzed limited sets of
voxels or electrodes, leaving it possible that unattended items might
be represented in other brain areas or using an orthogonal neural
. Here we test directly whether brain regions in not only sen-
sory but also parietal and frontal cortex contain memory represen-
tations during the delay phase for unattended stimuli. We acquired
functional MRI data from a large pool of subjects (n = 87) while
they were memorizing orientation stimuli (Supplementary Fig. 1).
We used a working memory design that allowed us to separately
identify representations of attended and unattended stimuli. In each
trial, participants first memorized the orientation of two gratings
(Fig. 1). After presentation of these stimuli, a retrocue indicated
which of the two gratings would be tested in an upcoming change
discrimination task following an extended delay, which is the main
retention interval in our design. Then, after this memory test, a sec
ond retrocue was shown that selected either the same or the other
orientation for a second memory test. Such a two-stage retention
task forces participants to maintain the orientations of both gratings
until the second retrocue, but prioritizes and thus directs attention
to the first retrocued item (AMI) while minimizing attention on the
other item (UMI)
We used a variant of multivariate pattern analysis (cvMANOVA;
see Methods for details)
to identify which brain regions encoded
the memorized orientations for attended and unattended items.
The experiment was designed to optimize the ability to detect mem
ory information in the main retention interval following the first
retrocue (Fig. 1 and see Methods for details). We analyzed stimuli
in each hemifield separately to account for differences in retinotopic
location. Our analysis focused on the set of regions where prior
work indicated the presence of persistent stimulus-selective activity
for orientations when attention was not manipulated
: visual cor-
tex (V1–V4), intraparietal sulcus (IPS0–IPS5) and the frontal eye
fields (FEF; Fig. 2a).
In early visual cortex, we found reliable information about
attended memory items (Fig. 2b; one-tailed one-sample t test;
= 3.37, P = 0.000558) whereas we found no significant informa-
tion for unattended items (one-tailed one-sample t test; t
P = 0.423091, lower 95th percentile confidence interval, corrected
) = 0.01). Information was also significantly higher for
attended than for unattended items (two-tailed paired-sample
t test; t
= 2.65, P = 0.009467, lower CI
= –0.012, difference
in mean pattern distinctness (Δ D) CI
= [0.007, 0.048]). This
finding closely resembles previous reports that unattended memory
items are not accompanied by delay-period information in percep
tually driven brain regions
. Our data, as shown in Supplementary
Fig. 2, suggests that this attention effect is primarily driven by V1.
It is worth noting that we cannot exclude the possibility that more
sensitive methods might reveal information for unattended items
also in visual cortex. Furthermore, whether attended or unattended
items can be decoded in the current study might depend on using a
larger sample size than in prior work (Supplementary Fig. 3).
Regardless, if our analyses had focused exclusively on these visual
brain regions, we might have concluded that working memory
representations for unattended stimuli are silent during the delay.
Cortical specialization for attended versus
unattended working memory
Thomas B. Christophel
*, Polina Iamshchinina
, Chang Yan
, Carsten Allefeld
NATURE NEUROSCIENCE | VOL 21 | APRIL 2018 | 494–496 | www.nature.com/natureneuroscience
© 2018 Nature America Inc., part of Springer Nature. All rights reserved.