Psychology
2013. Vol.4, No.8, 655-662
Published Online August 2013 in SciRes (http://www.scirp.org/journal/psych) http://dx.doi.org/10.4236/psych.2013.48093
Visual Working Memory in Human Cortex
Brian Barton, Alyssa A. Brewer
Department of Cognitive Sciences, University of California, Irvine, USA
Received May 18th, 2013; revised June 28th, 2013; accepted July 9th, 2013
Copyright © 2013 Brian Barton, Alyssa A. Brewer. This is an open access article distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium,
provided the original work is properly cited.
Visual working memory (VWM) is the ability to maintain visual information in a readily available and
easily updated state. Converging evidence has revealed that VWM capacity is limited by the number of
maintained objects, which is about 3 - 4 for the average human. Recent work suggests that VWM capacity
is also limited by the resolution required to maintain objects, which is tied to the objects’ inherent com-
plexity. Electroencephalogram (EEG) studies using the Contralateral Delay Activity (CDA) paradigm
have revealed that cortical representations of VWM are at a minimum loosely organized like the primary
visual system, such that the left side of space is represented in the right hemisphere, and vice versa. Re-
cent functional magnetic resonance imaging (fMRI) work shows that the number of objects is maintained
by representations in the inferior intraparietal sulcus (IPS) along dorsal parietal cortex, whereas the reso-
lution of these maintained objects is subserved by the superior IPS and the lateral occipital complex
(LOC). These areas overlap with recently-discovered, retinotopically-organized visual field maps (VFMs)
spanning the IPS (IPS-0/1/2/3/4/5), and potentially maps in lateral occipital cortex, such as LO-1/2, and/or
TO-1/2 (hMT+). Other fMRI studies have implicated early VFMs in posterior occipital cortex, suggesting
that visual areas V1-hV4 are recruited to represent information in VWM. Insight into whether and how
these VFMs subserve VWM may illuminate the nature of VWM. In addition, understanding the nature of
these maps may allow a greater investigation into individual differences among subjects and even be-
tween hemispheres within subjects.
Keywords: Visual Working Memory; Visual Field Maps; EEG; fMRI
Behavioral Measurements
Visual working memory (VWM) is the ability to maintain
visual information in a readily available and easily updated
state. Despite our rich visual experience, VWM has a limited
capacity and represents only a small fraction of the available
visual scene. Under the right conditions, one can miss changes
happening across wide swaths of the visual scene (Simons &
Ambinder, 2005). Even with a complete sensory trace of visual
information, such as iconic memory, only a subset of the in-
formation can be accurately reported (Sperling, 1960). Many
researchers have found evidence across a variety of tasks that
VWM capacity is limited to representations of about 3 - 4 ob-
jects (Sperling, 1960; Pashler, 1988; Irwin, 1992; Luck &
Vogel, 1997; Cowan, 2001; Vogel, Woodman et al., 2001; Awh,
Barton et al., 2007; Scolari, Vogel et al., 2008; Zhang & Luck,
2008; Barton, Ester et al., 2009).
The most robust and now most popular of these tasks is the
change detection task, which has three stages. The first is the
encoding stage, where an array of objects is briefly presented
(usually for a period between 100 and 500 ms), and the subject
must encode as many of the objects as they can into VWM.
Next, there is a short delay, generally about 1000 ms, when the
subject must maintain the objects in VWM. Finally, the test
array in which one object may have changed is presented, and
the subject must indicate with a button press whether the test
array is the same as or different from the sample array. A
common variation probes a single object rather than displaying
the entire array of objects at test, to reduce the likelihood of
counting and grouping strategies (Figure 1).
The change detection task allows for the estimation of the
number of objects that can be simultaneously held in VWM,
using the k formula developed by Pashler (1988) and refined by
Cowan (2001). The formula can be written as:
k = (Hit Rate + Correct Rejection Rate − 1) * Set Size (1)
The k formula assumes that there are no limitations in the
encoding or test portions of the task, such that errors are due
only to the lack of objects being maintained in the maintenance
stage. It is important to emphasize that there are other factors
that could affect performance which would be interpreted as
differences in VWM capacity k, but which in fact have nothing
to do with VWM maintenance. As a result, one must be careful
in interpreting values of k not to draw aberrant conclusions
about VWM capacity.
Luck and Vogel (1997) employed the change detection task
and k formula to great effect, showing that the number of ob-
jects the average human could maintain in VWM was 3 - 4.
Furthermore, by presenting a variety of different numbers of
features for each object, they also showed that the number of
objects represented does not depend on the number of features
attributed to each object, but only the number of objects main-
tained (Luck & Vogel, 1997). Neurological correlates to this
number limit of VWM capacity show sustained cortical activity

B. BARTON, A. A. BREWER
Figure 1.
Change detection with simple and complex objects. Typical change
detection tasks are two-alternative forced-choice tasks in which a sub-
ject is presented with an array of objects to encode into VWM (usually
for ~100 - 500 ms), maintains those representations once they are no
longer visible (usually for ~1000 ms), and is then asked whether one
probed object is the same or different from the sample object in the
same position (usually until the subject responds, or for ~1000 - 2000
ms). Here, a set size of four and a probe array of one are represented.
However, set size is often varied, and entire arrays are often presented
at the probe, although still only one test object can change. (a) Simple
colored square stimuli, typically drawn from about 6 distinct hues.
Changes are always low in similarity (e.g., a green square changes to
red), and thus require low resolution to make an accurate comparison.
Complex stimuli adapted from Barton & Brewer (2010) are displayed
in (b) and (c), which have both one of 6 or so overall hues like the
colored squares, but the sides of each cube have three shades, creating 6
possible shading patterns as well. Thus, a change in overall hue (b)
results in equivalent performance as with colored squares (a), because
both are low similarity changes. In contrast, a change of shading pattern,
but not overall hue (c), results in worse performance, because higher
resolution is required to make an accurate comparison between the
sample object and test probe.
corresponding to the number of objects maintained in VWM
and have been demonstrated in EEG (Vogel & Machizawa,
2004; Vogel, McCollough et al., 2005; Drew, McCollough et
al., 2006; McCollough, Machizawa et al., 2007) and fMRI (Todd
& Marois, 2004; Todd & Marois, 2005; Xu & Chun, 2006). In
addition, individual differences in VWM number capacity have
been demonstrated behaviorally (Awh, Barton et al., 2007),
with EEG (Vogel & Machizawa, 2004; Vogel, McCollough et
al., 2005), and with fMRI (Todd & Marois, 2005).
Alvarez and Cavanagh (2004) then showed that the number
of objects that can be maintained in VWM also depends on the
complexity of the maintained objects. Using the change detec-
tion task and k formula with objects of varying levels of com-
plexity, they showed that an independent method of assessing
complexity (search rate among similar distracters) is strongly
correlated with the estimation of the maximum number of ob-
jects of a certain type that could be maintained. Consequently,
VWM number capacity was estimated to be lower for complex
objects than simple ones. Neurological correlates that show sus-
tained cortical activity corresponding to changes in the com-
plexity of objects maintained in VWM have similarly been
demonstrated in EEG (Gao, Li et al., 2009) and fMRI (Xu &
Chun, 2006).
These two disparate findings have been resolved by Awh,
Barton, & Vogel (2007), who revealed that performance be-
tween simple and complex objects is identical, so long as the
level of similarity between the test and sample object is large.
Comparisons of simple objects are not limited by resolution
because the sample and test are so dissimilar that the limiting
factor is only k, the number of objects that can be held simulta-
neously in VWM. Normally, complex objects are limited by
resolution, because they are so similar to one another within
each category that the resolution of each object is too low to
make accurate comparisons. If one makes the changes large
enough such that resolution is no longer a limiting factor by
making categorical changes between complex objects, per-
formance is limited by k, as with simple objects (Figure 1).
Thus, it appears that there are two limitations of VWM per-
formance: the number of objects simultaneously held and the
resolution at which those objects are represented.
Discrete or Flexible Resource Allocation?
The idea that a resolution resource exists for VWM begs the
question: how is it allocated across maintained objects? Two
scenarios arise as likely possibilities. The first is a flexible re-
source allocation model. In this model, resolution could be
allocated differentially to objects of different complexity, such
that a simple object might not be allocated much resolution, but
an object of higher complexity, more demanding to represent,
might be allocated more resolution. The second model is a rigid
resource allocation model, in which objects up to a certain
maximum, k, are represented with a certain resolution, regard-
less of complexity.
Recently, evidence favoring the discrete resource allocation
model came when Zhang & Luck (2008) used a procedure that
provides independent estimates of the number and resolution of
representations in VWM to demonstrate that a subset of avail-
able objects are maintained and no information is retained
about other objects. Further, they showed that giving a pre-cue
to indicate which of the sample objects was most likely to
change increased the likelihood that the cued object was en-
coded, but not the likelihood that it was allocated more resolu-
tion. The idea is that if resources could be flexibly allocated,
subjects would have a large incentive to allocate more resolu-
tion to the pre-cued item. Barton, Ester, & Awh (2009) demon-
strated complimentary results that argue in favor of a discrete
resource allocation model. They presented arrays of objects to
encode in a change detection task that varied in the amount of
overall complexity in the display and found that change detec-
tion performance with any given object in the display did not
depend on the complexity of the other objects in the display.
Also in line with the predictions of a rigid resource allocation
model, maintained object resolution plateaus after the item limit,
k, is reached (Anderson, Vogel et al., 2011).
However, evidence in favor of a flexible resource allocation
model came with the demonstration that drawing attention to an
object with a flash changes the resolution at which it is encoded
into VWM (Bays & Husain, 2008). Not only that, but it was
demonstrated that some of the error in the Zhang & Luck (2008)
task was due to a mnemonic mismatch between color and loca-
tion information, not simply random guessing (Bays, Catalao et
656

B. BARTON, A. A. BREWER
al., 2009). Furthermore, other studies have found no evidence
for a limit in the number of representations that can be held in
VWM in general (Keshvari, van den Berg et al., 2013), or at
least for certain conditions (Sligte, Scholte et al., 2008).
When directly compared, it appears that the discrete resource
allocation model slightly more accurately fits the data of this
study relative to the flexible resource allocation model (Rouder,
Morey et al., 2008). Although much is made of the differences
between these models, their similarities are often overlooked. It
is agreed in general that resolution decreases as the number of
objects maintained increases (Zhang & Luck, 2008; Barton,
Ester et al., 2009; Bays, Catalao et al., 2009; Bays, Gorgoraptis
et al., 2011). Both models propose that resolution decreases as
set size increases, because the same amount of resources are
split among many objects, but one proposes that those re-
sources are evenly distributed across a limited number of dis-
crete chunks, while the other proposes that the chunks are
unlimited and their resolution can vary. Although one can ima-
gine somewhat different predictions for neural implementa-
tion made by each model, it may be most useful for current
questions of cortical organization to focus on their similarities
and how the common features interact with the retinotopic or-
ganization of the visual system.
Electroencephalography
Electroencephalography is the measure of the electrical ac-
tivity originating in the brains of human subjects, recorded at
numerous electrode sites placed on their scalps. EEG offers
excellent insight into the temporal resolution (on the order of
milliseconds) of electrical brain activity. However, many sources
contribute to the activity recorded at each electrode, so its spa-
tial resolution is coarse and generally relies on converging evi-
dence (Clark, Fan et al., 1994; Luck, 1999; Luck, Woodman et
al., 2000; McCollough, Machizawa et al., 2007).
A technique which has been applied to studies of VWM is a
specialized form of the event-related potential (ERP) measure-
ment. ERPs are very small electrical signals generated in corti-
cal regions in response to specific events or stimuli (Blackwood
& Muir, 1990; Sur & Sinha, 2009). The ERP related to a par-
ticular type of stimulus is measured by having subjects re-
peatedly respond to the stimulus hundreds or thousands of
times, and then averaging the signal across trials and subjects in
order to reveal a common waveform associated with the stimu-
lus (Luck, 1999). A specialized ERP component used to study
VWM is known as Contralateral Delay Activity (CDA) and
measures sustained electrical activity during the delay period of
a change detection task. The CDA requires objects to be pre-
sented in both hemifields, but held in VWM in one hemifield
and not in the other. Then, the ERP waveform recorded from
electrode sites in the posterior parietal, occipital, and occipi-
tal-temporal regions of the hemisphere ipsilateral to the remem-
bered objects is subtracted from that of the contralateral hemi-
sphere, resulting in a difference waveform, which constitutes
the CDA (Vogel & Machizawa, 2004; Vogel, McCollough et al.,
2005; Drew, McCollough et al., 2006; McCollough, Machi-
zawa et al., 2007; Gao, Li et al., 2009).
Originally, the CDA revealed sustained activity commensu-
rate to the number of objects represented in VWM. Using sim-
ple colored squares in a change detection task, the CDA activity
increased as set sizes increased, until the subjects’ working
memory capacity was exceeded, at which point the CDA rea-
ched a plateau (Vogel & Machizawa, 2004; McCollough, Ma-
chizawa et al., 2007). More recently, the CDA has been shown
to reflect the amount of resolution allocated to objects main-
tained in VWM. In a study comparing the CDA when simple or
complex objects are maintained in a change detection task,
electrode sites in posterior parietal, occipital, and occipito-tem-
poral cortex showed greater activity for complex than simple
objects (Gao, Li et al., 2009). While activity at the same elec-
trode sites increased as the number of maintained simple ob-
jects increased, the activity remained indistinguishable between
set sizes for complex objects. It is important to note that the
VWM capacity estimate for these complex objects was meas-
ured to be two objects, so no difference is expected between set
sizes two and four (Alvarez & Cavanagh, 2004). The authors
interpreted these data as consistent with a flexible resource al-
location model; however, the results can equivalently be ex-
plained by a rigid resource allocation model.
Functional Magnetic Resonance Imaging
When neurons are active, the ratio of oxygenated hemoglo-
bin to deoxygenated hemoglobin in nearby blood increases after
a few seconds (known as the hemodynamic response). Func-
tional magnetic resonance imaging (fMRI) is a technique that
takes advantage of such blood oxygen-level dependent (BOLD)
activity in the brain by applying a strong magnetic field and
encoding positions in space with slight differences in magnetic
field strength and phase (Ogawa, Tank et al., 1992). This allows
for a very specific spatial readout of brain activity (on the order
of 1 mm3), but the hemodynamic response is on the order of
seconds, which limits the ability of fMRI to distinguish fine
temporal differences (Logothetis, Pauls et al., 2001; Logothetis
& Wandell, 2004; Serences, 2004; Shmuel, Augath et al., 2006;
Poldrack, Fletcher et al., 2008; Schridde, Khubchandani et al.,
2008; Schummers, Yu et al., 2008; Chen & Parrish, 2009).
Initial fMRI studies of VWM focused on the role of the
frontal lobe, in regions such as dorsolateral and ventrolateral
prefrontal cortex (PFC). Much of the focus was trying to parcel
out “spatial” vs. “non-spatial” VWM storage abilities, which re-
turned mixed results that have since been reevaluated (Jonides,
Smith et al., 1993; D’Esposito & Postle, 1999; Postle, Zarahn et
al., 2000; Postle, Awh et al., 2004; Roth, Serences et al., 2006).
These areas are now largely considered to be involved in the
encoding-, manipulation-, and response-related aspects of VWM
(Postle, Berger et al., 2000; Rypma, Berger et al., 2002) or
attentional feedback control (Curtis & D’Esposito, 2003), op-
erations normally ascribed to the “central executive” (Baddeley
& Hitch, 1974; Baddeley, 1996; Baddeley, 2000; Baddeley,
2003). There is little evidence that these frontal areas, though
vital to the VWM network, are the regions that give rise to the
number and resolution limits on VWM capacity (Postle, 2006).
In the search for regions which subserve VWM capacity lim-
its, researchers have observed three regions of cortex whose
activity during a VWM task is commensurate with either the
number of or the resolution required to represent objects in
VWM. The first region identified spanned the intraparietal
sulcus (IPS) in the dorsal parietal cortex bilaterally and was
shown to have BOLD signal modulation positively correlated
with the number of objects maintained in VWM during the
delay period of a change detection task (Todd & Marois, 2004;
Todd & Marois, 2005).
Further research led to the discovery that inferior IPS in both

B. BARTON, A. A. BREWER
hemispheres showed activity dependent upon the number of
maintained objects, but superior IPS and part of the lateral-
occipital complex (LOC) in both hemispheres responded in-
stead to the complexity of maintained objects (Xu & Chun,
2006). These regions were identified using either simple or
complex objects in a change detection task. Behavioral VWM
capacity was estimated for simple and complex objects, and
then BOLD modulation was compared to behavioral estimates
across LOC, superior and inferior IPS. Superior IPS and LOC
were shown to have BOLD signal modulation corresponding to
the behavioral measurements of VWM capacity. Inferior IPS
showed activity corresponding to the number of objects pre-
sented, regardless of complexity.
The authors argue that inferior IPS is an area of object indi-
viduation based on location, leading to the VWM number ca-
pacity limit found in simple change detection tasks. In contrast,
they argue, superior IPS and LOC are areas responsible for
object identification and binding, such that a smaller number of
complex objects can be represented than simple objects, in
favor of a two-stage, flexible resource allocation model (Xu &
Chun, 2006; Xu & Chun, 2007; Xu & Chun, 2009). However, it
is also possible that the results could be equivalently ex-
plained by a slightly altered version of the discrete resource
allocation model.
Recent studies have also implicated V1-hV4 in VWM proc-
essing, perhaps representing information that is related to the
sensory information processed in each map (Harrison & Tong,
2009; Sligte, Scholte et al., 2009; Ester, Anderson et al., 2013).
Studies such as these call into question the idea that a particular
location of cortex is responsible for VWM, but rather it is likely
that VWM is a distributed aspect of the entire visual system.
Perhaps, like the visual system, representations of different
types of information are represented in different portions of
cortex.
Visual Field Maps
Visual information first enters primate cortex in primary vis-
ual cortex (area V1), located in the posterior occipital lobe, for
low-level processing of visual details (Tootell, Hadjikhani et al.
1998). From there, visual information is sent to other visual
areas, which can be differentiated by their unique cytoarchitec-
tonic structures, connectivity, visual field maps (VFMs), and
functional processing (e.g., (Van Essen, Newsome et al., 1984;
Felleman & Van Essen, 1991; Wandell, Dumoulin et al., 2007;
Brewer & Barton, 2012)). Many visual areas were first identi-
fied in monkey using primarily single and multiple unit re-
cording techniques, and since, many homologues have been
discovered in human (e.g., (Brewer, Press et al., 2002; Sereno
& Tootell, 2005; Wandell, Dumoulin et al., 2007; Kolster,
Peeters et al., 2010)). The most compelling evidence for visual
areas in humans are VFMs, often called retinotopic maps, be-
cause they follow the organization of the retina. That is, nearby
neurons within a VFM analyze properties of nearby points of an
image on the retina (e.g., (Wandell, 1999; Wandell, Dumoulin
et al., 2007; Brewer & Barton, 2012)).
VFMs are routinely measured using standard traveling wave
methods (Engel, Rumelhart et al., 1994; Sereno, Dale et al.,
1995; DeYoe, Carman et al., 1996; Engel, Glover et al., 1997;
Wandell, 1999; Wandell, Dumoulin et al., 2007; Brewer & Bar-
ton, 2012). In this method, subjects view a periodic stimulus
that moves smoothly through the visual field, which creates a
traveling wave of activity within retinotopic VFMs. From the
fMRI response time series, the position of the stimulus within
the visual field that is optimal for evoking a response from each
cortical location can be identified. Retinotopic VFMs are meas-
ured with respect to two visual field dimensions: eccentricity
and polar angle. Rotating wedge and expanding ring stimuli
consisting of high contrast, flickering checkerboard patterns are
typically used to measure gradients of visual field polar angle
and eccentricity, respectively.
Each of the first few VFMs (V1, V2, and V3) can be identi-
fied as containing a representation of a full hemifield of visual
space, with each hemisphere representing the contralateral he-
mifield (Wandell, 1999; Dougherty, Koch et al., 2003; Wandell,
Dumoulin et al., 2007; Brewer & Barton, 2012). VFM V1 and
the adjacent VFMs V2 and V3 contribute to a confluent foveal
representation at the occipital pole (Schira, Tyler et al., 2009).
All of these areas have a lower visual field representation lo-
cated on the dorsal part of the posterior occipital lobe and an
upper visual field representation on the ventral surface. Beyond
these early VFMs in the posterior occipital lobe, visual cortex is
loosely organized into anatomically distinct dorsal and ventral
“streams” (Morel & Bullier, 1990; Baizer, Ungerleider et al.,
1991). Several dorsal and ventral human VFMs have been
mapped using fMRI, and they contain abutting upper and lower
visual field representations and distinct foveal representations
(Tootell, Mendola et al., 1997; Press, Brewer et al., 2001; Wade,
Brewer et al., 2002; Brewer, Liu et al., 2005).
Beyond the dorsal lower vertical meridian representation of
V3 lies a string of hemifield maps running from the transverse
occipital sulcus (TOS) up along the medial wall of the intrapa-
rietal sulcus (IPS). The first maps bordering V3d are V3A and
V3B, which share a discrete foveal representation within the
TOS (Tootell, Mendola et al., 1997; Smith, Greenlee et al.,
1998; Press, Brewer et al., 2001). A series of recently described
maps extend from IPS-0 (also called V7 (Tootell, Hadjikhani et
al., 1998)) along the medial wall of the IPS, reversing smoothly
into several hemifields from IPS-1 to IPS-5 (Sereno, Pitzalis et
al., 2001; Silver, Ress et al., 2005; Swisher, Halko et al., 2007;
Wandell, Dumoulin et al., 2007; Konen & Kastner, 2008).
Weak foveal representations for each map fall along the fundus
of the IPS.
In addition, two sets of VFMs (LO-1, LO-2, TO-1, and TO-2)
positioned on the dorsal part of lateral occipital cortex anterior
to V3d were recently described (Larsson & Heeger, 2006;
Amano, Wandell et al., 2009). LO-1 and 2 share the confluent
foveal representation with V1, 2 and 3, while TO-1 and 2 share
a distinct foveal representation. LO-1 and 2 overlap with part of
object-selective LOC, while TO-1 and 2 overlap with motion-
selective areas MT and MST, respectively. Additional VFMs
TO-3 and TO-4 have recently been described just inferior to
TO-1 and TO-2 (Kolster, Peeters et al., 2010). Several recent
studies and reviews can provide more extensive details on the
current state of visual field mapping (Brewer & Barton, 2012),
the travelling wave methodology (Engel, 2012), population re-
captive field modeling for VFM measurements (Dumoulin &
Wandell, 2008), and the history of VFM measurements (Wan-
dell & Winawer, 2011).
The recently discovered IPS VFMs are strongly modulated
by attention, and they appear to overlap with the superior and
inferior IPS regions reported by Xu & Chun (2006) to subserve
VWM. Also, it is likely that the region of lateral occipital cor-
tex implicated in VFM falls within one or several of the lateral
658

B. BARTON, A. A. BREWER
VFMs (LO-1/2, TO-1/2) (Barton & Brewer, 2010). If these
regions do overlap, the retinotopic organization of the VFMs
has important implications for the functional properties of these
VWM areas. Indeed, studies of early VFMs suggest that VWM
is represented in a retinotopic fashion in the early visual system,
so it is likely that this relationship is maintained throughout the
visual hierarchy.
Individual Differences
In the search for behavioral effects, ERP signatures, and
brain regions corresponding to VWM capacity limits, individ-
ual differences are often overlooked as noise in the data that is
muddying an otherwise clean study. If published, reports on
individual differences of VWM capacity are usually relegated
to less prestigious journals after the primary findings have
snagged the lion’s share of the attention. However, individual
differences in VWM capacity continue to arise and are becom-
ing more and more important in the ongoing debates of VWM
capacity limitations (Vogel & Machizawa, 2004; Todd & Mar-
ois, 2005; Vogel, McCollough et al., 2005; Awh, Barton et al.,
2007).
Much of the debate over VWM capacity limitations ignores
the dissociation between the ability to maintain a number of
objects in VWM and the ability to allocate resolution to those
objects. Awh et al. (2007) revealed that performance across
independent number-limited change detection tasks is strongly
correlated within subjects. Correspondingly, performance on
independent resolution-limited tasks is highly correlated. How-
ever, performance on number-limited tasks does not correlate
with performance on resolution-limited tasks. Together, these
results suggest that there are two independent abilities being
tested in these tasks, one for number and one for resolution,
which vary within and between subjects.
Todd & Marois reported group data (Todd & Marois, 2004)
and individual differences (Todd & Marois, 2005) for their
change detection measurements that revealed BOLD modula-
tion in IPS corresponding to the number of objects maintained
in VWM. Todd & Marois (2005) reported that the difference in
BOLD modulation in relevant regions between the set size at
which a subject’s VWM number capacity is reached and set
size 1 correlates with each subject’s VWM capacity estimate. In
other words, subjects with a larger VWM capacity as measured
behaviorally also show larger BOLD modulations in cortex in
response to the VWM task. This finding demonstrated that the
BOLD signal can be used as a marker for VWM number capac-
ity in individual subjects.
VWM number capacity limits have further been shown to
correlate with limits of the number of objects one can track in a
multiple object tracking (MOT) task. Also, it has been reported
that similar regions of cortex respond in a load-based manner
during MOT and VWM tasks, and thus MOT may be subject to
the same number capacity limit (Culham, Brandt et al., 1998;
Culham, Cavanagh et al., 2001; Culham & Kanwisher, 2001;
Cavanagh & Alvarez, 2005). The MOT task is similar to a
change detection task, except that subjects must actively track
objects rather than maintain them between the sample and test
(Pylyshyn & Storm, 1988; Drew & Vogel, 2008). Interesting
evidence has been presented using MOT tasks that show dif-
ferences in the number of objects that can be tracked depending
on whether the objects span both hemifields or are contained in
one hemifield (Cavanagh & Alvarez, 2005). If one takes the
assumption of a shared number capacity limit between VWM
and MOT tasks, this is strong evidence that the total capacity
limit is divided between the two hemispheres. In that case, one
would expect that there may be individual differences between
hemispheres within a subject.
Finally, what if there were little to no variability in the VWM
capacity number limit among individuals or within hemispheres?
It is possible that some unaccounted-for effect may underlie
what appear to be individual differences in the estimation of
VWM capacity number limit. Vogel et al. (Vogel, McCollough
et al., 2005) had a very intriguing study using the CDA, where
they compared subjects with high and low VWM capacity
number limits. Using a change detection task with simple col-
ored squares, they presented two or four target objects to re-
member, or two target objects and two distracter objects. They
found that subjects with high capacity VWM showed equiva-
lent CDA amplitude for two targets and two targets with two
distracters, but subjects with low capacity VWM showed equi-
valent amplitude for four targets and two targets with two dis-
tracters. In an even more striking manipulation, they asked
subjects to remember two targets for half of the delay period,
and then either add two targets (append red) or exclude two
distracters (exclude green) that appear during the delay period.
High capacity subjects successfully appended targets and ex-
cluded distracters, whereas low capacity subjects successfully
appended targets, but also appended distracters. Together, these
results suggest that perhaps lower estimates of VWM number
capacity may actually be due to an inability to deal with irrele-
vant objects, and not a lower maximum number of maintained
objects. Based on this work, it is possible that all subjects have
very similar VWM number capacities, but have trouble with
other aspects of the task, such as successfully locking onto a
manageable number of objects to maintain.
Conclusion and Future Directions
VWM capacity has been defined by two factors: the number
of objects maintained and the resolution at which those objects
are represented. The number of objects the average human can
maintain seems to be about 3 - 4, and the resolution allocated to
each of those objects declines as the number of objects main-
tained increases. Resolution appears to be allocated in a rigid
manner, such that resolution is evenly distributed across main-
tained objects, regardless of those objects’ complexity. Al-
though a flexible resource allocation model has also gained
traction, we suggest that the two models have more similarities
than differences, and it is those similarities which should drive
future research into the neural underpinnings of VWM.
These capacity limits have been demonstrated in a variety of
behavioral measures, and the CDA reveals patterns of electrical
cortical activity that show dissociable indices of number and
resolution. Functional MRI results suggest that the number ca-
pacity limit arises in inferior IPS, while the resolution capacity
limit arises in superior IPS and LOC, though other aspects of
these objects (e.g., color) may be represented in other parts of
visual cortex. The cortical regions involved with capacity limi-
tation overlap with several recently-discovered VFMs, indicat-
ing that they are retinotopically organized. Thus, the limitations
measured in VWM tasks may arise from the properties of the
underlying organization of these VFMs. This idea also presents
intriguing evidence that VWM may recruit the visual system to
represent objects.

B. BARTON, A. A. BREWER
Future lines of research should address specifically which of
the VWM areas fall into which of the retinotopically-organized
VFMs. It is possible that some of the VWM areas fall near
retinotopic maps and not within them, but it is difficult to tell at
this point because these regions have not been compared for
overlap within individual subjects across visual cortex. In addi-
tion, most of the VWM capacity limitation literature ignores in-
dividual differences within and between subjects, and there is a
wealth of opportunity to demonstrate just how subjects differ in
their capacity limitations.
Acknowledgements
B.B. wrote the manuscript, which was then revised by B.B.
and A.A.B.
REFERENCES
Alvarez, G. A., & Cavanagh, P. (2004). The capacity of visual short-
term memory is set both by visual information load and by number
of objects. Psychological Sc i e n c e , 15, 106-111.
doi:10.1111/j.0963-7214.2004.01502006.x
Amano, K., Wandell, B. A. et al. (2009). Visual field maps, population
receptive field sizes, and visual field coverage in the human MT+
complex. Journal of Neurophysiology , 102, 2704-2718.
doi:10.1152/jn.00102.2009
Anderson, D. E., Vogel, E. K. et al. (2011). Precision in visual working
memory reaches a stable plateau when individual item limits are ex-
ceeded. Journal of Neuros ci en c e, 31, 1128-1138.
doi:10.1523/JNEUROSCI.4125-10.2011
Awh, E., Barton, B. et al. (2007). Visual working memory represents a
fixed number of items regardless of complexity. Psychological Sci-
ence, 18, 622-628. doi:10.1111/j.1467-9280.2007.01949.x
Baddeley, A. (1996). Exploring the central executive. The Quarterly
Journal of Experimental Psychology Section A: Human Experimental
Psychology, 49, 5-28.
Baddeley, A. (2000). The episodic buffer: A new component of work-
ing memory? Trends in Cognitive Sciences, 4, 417-423.
doi:10.1016/S1364-6613(00)01538-2
Baddeley, A. (2003). Working memory: Looking back and looking
forward. Nature Reviews Neuroscience, 4, 829-839.
doi:10.1038/nrn1201
Baddeley, A. D., & Hitch, G. (1974). Working memory. In H. B.
Gordon (Ed.), Psychology of learning and motivation (Vol. 8, pp. 47-
89). Waltham, MA: Academic Press.
Baizer, J. S., Ungerleider, L. G. et al. (1991). Organization of visual
inputs to the inferior temporal and posterior parietal cortex in ma-
caques. Journal of Neuroscienc e , 11, 168-190.
Barton, B., & Brewer, A. A. (2010). Visual working memory capacity
in retinotopic cortex: Number, resolution, and population receptive
fields. Journal of V i s ion, 10, 729. doi:10.1167/10.7.729
Barton, B., Ester, E. F. et al. (2009). Discrete resource allocation in
visual working memory. Journal of Experimental Psychology: Hu-
man Perception and Performance, 35 , 1359-1367.
doi:10.1037/a0015792
Bays, P. M., & Husain, M. (2008). Dynamic shifts of limited working
memory resources in human vision. Science, 321, 851-854.
doi:10.1126/science.1158023
Bays, P. M., Gorgoraptis, N. et al. (2011). Temporal dynamics of en-
coding, storage, and reallocation of visual working memory. Journal
of Vision, 11, Article 6. doi:10.1167/11.10.6
Bays, P. M., Catalao, R. F. et al. (2009). The precision of visual work-
ing memory is set by allocation of a shared resource. Journal of Vi-
sion, 9, Article 7.
Blackwood, D. H., & Muir, W. J. (1990). Cognitive brain potentials and
their application. British Journal of Psychiatry Supplements, 9, 96-
101.
Brewer, A. A., & Barton, B. (2012). Visual field map organization in
human visual cortex. In S. Molotchnikoff (Ed.), Visual Cortex, In-
Tech, 29-60.
Brewer, A. A., Liu, J. et al. (2005). Visual field maps and stimulus se-
lectivity in human ventral occipital cortex. Nature Neuroscience, 8,
1102-1109. doi:10.1038/nn1507
Brewer, A. A., Press, W. A. et al. (2002). Visual areas in macaque
cortex measured using functional magnetic resonance imaging. Jour-
nal of Neuroscience, 22, 10416-10426.
Cavanagh, P., & Alvarez, G. A. (2005). Tracking multiple targets with
multifocal attention. Trends in Cognitiv e S c ie n c e s , 9, 349-354.
doi:10.1016/j.tics.2005.05.009
Chen, Y., & Parrish, T. B. (2009). Caffeine’s effects on cerebrovascular
reactivity and coupling between cerebral blood flow and oxygen me-
tabolism. Neuroimage, 44, 647-652.
doi:10.1016/j.neuroimage.2008.09.057
Clark, V. P., Fan, S. et al. (1994). Identification of early visual evoked
potential generators by retinotopic and topographic analyses. Human
Brain Mapping, 2, 170-187. doi:10.1002/hbm.460020306
Cowan, N. (2001). The magical number 4 in short-term memory: A
reconsideration of mental storage capacity. Behavioral and Brain
Sciences, 24, 87-114. doi:10.1017/S0140525X01003922
Culham, J. C., & Kanwisher, N. G. (2001). Neuroimaging of cognitive
functions in human parietal cortex. Current Opinion in Neurobiology,
11, 157-163. doi:10.1016/S0959-4388(00)00191-4
Culham, J. C., Cavanagh, P. et al. (2001). Attention response functions:
Characterizing brain areas using fMRI activation during parametric
variations of attentional load. Neuron, 32, 737-745.
doi:10.1016/S0896-6273(01)00499-8
Culham, J. C., Brandt, S. A. et al. (1998). Cortical fMRI activation pro-
duced by attentive tracking of moving targets. Journal of Neurophy-
siology, 80, 2657-2670.
Curtis, C. E., & D’Esposito, M. (2003). Persistent activity in the pre-
frontal cortex during working memory. Trends in Cognitive Sciences,
7, 415-423. doi:10.1016/S1364-6613(03)00197-9
D’Esposito, M., & Postle, B. R. (1999). The dependence of span and
delayed-response performance on prefrontal cortex. Neuropsycholo-
gia, 37, 1303-1315. doi:10.1016/S0028-3932(99)00021-4
DeYoe, E. A., Carman, G. J. et al. (1996). Mapping striate and extra-
striate visual areas in human cerebral cortex. Proceedings of the Na-
tional Academy of Sciences of the United States of America, 93,
2382-2386. doi:10.1073/pnas.93.6.2382
Dougherty, R. F., Koch, V. M. et al. (2003). Visual field representa-
tions and locations of visual areas V1/2/3 in human visual cortex.
Journal of Vision, 3, 586-598. doi:10.1167/3.10.1
Drew, T., & Vogel, E. K. (2008). Neural measures of individual differ-
ences in selecting and tracking multiple moving objects. Journal of
Neuroscience, 28, 4183-4191.
doi:10.1523/JNEUROSCI.0556-08.2008
Drew, T. W., McCollough, A. W. et al. (2006). Event-related potential
measures of visual working memory. Clinical EEG & Neuroscience,
37, 286-291. doi:10.1177/155005940603700405
Dumoulin, S. O., & Wandell, B. A. (2008). Population receptive field
estimates in human visual cortex. Neuroimage, 39, 647-660.
doi:10.1016/j.neuroimage.2007.09.034
Engel, S. A. (2012). The development and use of phase-encoded func-
tional MRI designs. Neuroi mage , 62, 1195-1200.
doi:10.1016/j.neuroimage.2011.09.059
Engel, S. A., Rumelhart, D. E. et al. (1994). fMRI of human visual cor-
tex. Nature, 369, 525. doi:10.1038/369525a0
Engel, S. A., Glover, G. H. et al. (1997). Retinotopic organization in
human visual cortex and the spatial precision of functional MRI. Ce-
reb Cortex, 7, 181-192. doi:10.1093/cercor/7.2.181
Ester, E. F., Anderson, D. E. et al. (2013). A neural measure of preci-
sion in visual working memory. Journal of Cognitive Neuroscience,
25, 754-761. doi:10.1162/jocn_a_00357
Felleman, D. J., & Van Essen, D. C. (1991). Distributed hierarchical
processing in the primate cerebral cortex. Cerebral Cortex, 1, 1-47.
doi:10.1093/cercor/1.1.1
Gao, Z., Li, J. et al. (2009). Storing fine detailed information in visual
working memory—Evidence from event-related potentials. Journal
of Vision, 9, 1-12. doi:10.1167/9.7.17
660

B. BARTON, A. A. BREWER
Harrison, S. A., & Tong, F. (2009). Decoding reveals the contents of
visual working memory in early visual areas. Nature, 458, 632-635.
doi:10.1038/nature07832
Irwin, D. E. (1992). Memory for position and identity across eye
movements. Journal o f Experimental Psychology, 18, 307-317.
Jonides, J., Smith, E. E. et al. (1993). Spatial working memory in hu-
mans as revealed by PET. Nature, 363, 623-625.
doi:10.1038/363623a0
Keshvari, S., van den Berg, R. et al. (2013). No evidence for an item
limit in change detection. PLoS Computational B i ol ogy, 9, 1.
doi:10.1371/journal.pcbi.1002927
Kolster, H., Peeters, R. et al. (2010). The retinotopic organization of the
human middle temporal area MT/V5 and its cortical neighbors. The
Journal of Neuroscie n c e, 30, 9801-9820.
doi:10.1523/JNEUROSCI.2069-10.2010
Konen, C. S., & Kastner, S. (2008). Representation of eye movements
and stimulus motion in topographically organized areas of human
posterior parietal cortex. The Journal of Neuroscience, 28, 8361-
8375. doi:10.1523/JNEUROSCI.1930-08.2008
Larsson, J., & Heeger, D. J. (2006). Two retinotopic visual areas in
human lateral occipital cortex. The Journal of Neuroscience, 26,
13128-13142. doi:10.1523/JNEUROSCI.1657-06.2006
Logothetis, N. K., & Wandell, B. A. (2004). Interpreting the BOLD
signal. Annual Revie w o f Physiology, 66, 735-769.
doi:10.1146/annurev.physiol.66.082602.092845
Logothetis, N. K., Pauls, J. et al. (2001). Neurophysiological investiga-
tion of the basis of the fMRI signal. Nature, 412, 150-157.
doi:10.1038/35084005
Luck, S. J. (1999). Direct and indirect integration of event-related po-
tentials, functional magnetic resonance images, and single-unit re-
cordings. Human Brain Mapping, 8, 115-201.
doi:10.1002/(SICI)1097-0193(1999)8:2/3<115::AID-HBM8>3.0.CO
;2-3
Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working
memory for features and conjunctions. Nature, 390, 279-281.
doi:10.1038/36846
Luck, S. J., Woodman, G. F. et al. (2000). Event-related potential stud-
ies of attention. Trends i n Cognitive Sciences, 4 , 432-440.
doi:10.1016/S1364-6613(00)01545-X
McCollough, A. W., Machizawa, M. G. et al. (2007). Electrophysio-
logical measures of maintaining representations in visual working
memory. Cortex, 43, 77-94. doi:10.1016/S0010-9452(08)70447-7
Morel, A., & Bullier J. (1990). Anatomical segregation of two cortical
visual pathways in the macaque monkey. Visual Neuroscience, 4,
555-578. doi:10.1017/S0952523800005769
Ogawa, S., Tank, D. et al. (1992). Intrinsic signal changes accompany-
ing sensory stimulation: Functional brain mapping with magnetic re-
sonance imaging. Proceedings of the National Academy of Sciences
of the United States of America, 8 9 , 591-5955.
doi:10.1073/pnas.89.13.5951
Pashler, H. (1988). Familiarity and visual change detection. Perception
& Psychophysics, 44, 369-378. doi:10.3758/BF03210419
Poldrack, R. A., Fletcher, P. C. et al. (2008). Guidelines for reporting
an fMRI study. Neuroimage, 40, 409-414.
doi:10.1016/j.neuroimage.2007.11.048
Postle, B. R. (2006). Working memory as an emergent property of the
mind and brain. Neuroscience, 139, 23-38.
doi:10.1016/j.neuroscience.2005.06.005
Postle, B. R., Awh, E. et al. (2004). The where and how of attention-
based rehearsal in spatial working memory. Brain Research. Cogni-
tive Brain Research, 20, 194-205.
doi:10.1016/j.cogbrainres.2004.02.008
Postle, B. R., Zarahn, E. et al. (2000). Using event-related fMRI to
assess delay-period activity during performance of spatial and non-
spatial working memory tasks. Brain Research Pr oto cols , 5, 57-66.
doi:10.1016/S1385-299X(99)00053-7
Postle, B. R., Berger, J. S. et al. (2000). Activity in human frontal cor-
tex associated with spatial working memory and saccadic behavior.
Journal of Cognitive Neuroscience, 12, 2-14.
doi:10.1162/089892900564028
Press, W. A., Brewer, A. A. et al. (2001). Visual areas and spatial
summation in human visual cortex. Vision Research, 41, 1321-1332.
doi:10.1016/S0042-6989(01)00074-8
Pylyshyn, Z. W., & Storm, R. W. (1988). Tracking multiple independ-
ent targets: Evidence for a parallel tracking mechanism. Spatial Vi-
sion, 3, 179-197. doi:10.1163/156856888X00122
Roth, J. K., Serences, J. T. et al. (2006). Neural system for controlling
the contents of object working memory in humans. Cerebral Cortex,
16, 1595-1603. doi:10.1093/cercor/bhj096
Rouder, J. N., Morey, R. D. et al. (2008). An assessment of fixed-ca-
pacity models of visual working memory. Proceedings of the Na-
tional Academy of Sciences of the United States of America, 105,
5975-5979. doi:10.1073/pnas.0711295105
Rypma, B., Berger, J. S. et al. (2002). The influence of working-mem-
ory demand and subject performance on prefrontal cortical activity.
Journal of Cognitive Neuroscience, 14, 721-731.
doi:10.1162/08989290260138627
Schira, M. M., Tyler, C. W. et al. (2009). The foveal confluence in
human visual cortex. The Journal of Neuroscience, 29, 9050-9058.
doi:10.1523/JNEUROSCI.1760-09.2009
Schridde, U., Khubchandani, M. et al. (2008). Negative BOLD with
large increases in neuronal activity. Cerebral Cortex, 18, 1814-1827.
doi:10.1093/cercor/bhm208
Schummers, J., Yu, H. et al. (2008). Tuned responses of astrocytes and
their influence on hemodynamic signals in the visual cortex. Science,
320, 1638-1643. doi:10.1126/science.1156120
Scolari, M., Vogel, E. K. et al. (2008). Perceptual expertise enhances
the resolution but not the number of representations in working
memory. Psychonomic Bulletin and Review, 15, 215-222.
doi:10.3758/PBR.15.1.215
Serences, J. T. (2004). A comparison of methods for characterizing the
event-related BOLD timeseries in rapid fMRI. Neuroimage, 21,
1690-1700. doi:10.1016/j.neuroimage.2003.12.021
Sereno, M. I., & Tootell, R. B. (2005). From monkeys to humans: What
do we now know about brain homologies? Current Opinion in Neu-
robiology, 15, 135-144. doi:10.1016/j.conb.2005.03.014
Sereno, M. I., Dale, A. M. et al. (1995). Borders of multiple human
visual areas in humans revealed by functional MRI. Science, 268,
889-893. doi:10.1126/science.7754376
Sereno, M. I., Pitzalis, S. et al. (2001). Mapping of contralateral space
in retinotopic coordinates by a parietal cortical area in humans. Sci-
ence, 294, 1350-1354. doi:10.1126/science.1063695
Shmuel, A., Augath, M. et al. (2006). Negative functional MRI re-
sponse correlates with decreases in neuronal activity in monkey vis-
ual area V1. Nature Neuroscience, 9, 569-577.
doi:10.1038/nn1675
Silver, M. A., Ress, D. et al. (2005). Topographic maps of visual spatial
attention in human parietal cortex. Journal of Neurophysiology, 94,
1358-1371. doi:10.1152/jn.01316.2004
Simons, D. J., & Ambinder, M. S. (2005). Change blindness. Current
Directions in Psychological Sci e n c e , 14, 44-48.
doi:10.1111/j.0963-7214.2005.00332.x
Sligte, I. G., Scholte, H. S. et al. (2008). Are there multiple visual
short-term memory stores? PLoS One, 3, e1699.
doi:10.1371/journal.pone.0001699
Sligte, I. G., Scholte, H. S. et al. (2009). V4 Activity predicts the
strength of visual short-term memory representations. Journal of
Neuroscience, 29, 7432-7438.
doi:10.1523/JNEUROSCI.0784-09.2009
Smith, A. T., Greenlee, M. W. et al. (1998). The processing of first- and
second-order motion in human visual cortex assessed by functional
magnetic resonance imaging (fMRI). The Journal of Neuroscience,
18, 3816-3830.
Sperling, G. (1960). The information available in brief visual presenta-
tions. Psychological Monographs: General and Applied, 74 , 1-29.
doi:10.1037/h0093759
Sur, S., & Sinha, V. K. (2009). Event-related potential: An overview.
Indian Journal of Psychiatry, 18 , 70-73.
doi:10.4103/0972-6748.57865
Swisher, J. D., Halko, M. A. et al. (2007). Visual topography of human
intraparietal sulcus. The Journal of Neuroscience, 27, 5326-5337.
doi:10.1523/JNEUROSCI.0991-07.2007

B. BARTON, A. A. BREWER
662
Todd, J. J., & Marois, R. (2004). Capacity limit of visual short-term
memory in human posterior parietal cortex. Nature, 428, 751-754.
doi:10.1038/nature02466
Todd, J. J., & Marois, R. (2005). Posterior parietal cortex activity pre-
dicts individual differences in visual short-term memory capacity.
Cognitive, Affective, & Behavioral Neuroscience, 5, 144-155.
doi:10.3758/CABN.5.2.144
Tootell, R. B., Mendola, J. D. et al. (1997). Functional analysis of V3A
and related areas in human visual cortex. The Journal of Neurosci-
ence, 17, 7060-7078. doi:10.1016/S0896-6273(00)80659-5
Tootell, R. B., Hadjikhani, N. et al. (1998). The retinotopy of visual
spatial attention. Neuron, 21, 1409-1422.
Tootell, R. B., Hadjikhani, N. K. et al. (1998). Functional analysis of
primary visual cortex (V1) in humans. Proceedings of the National
Academy of Sciences of the Unite d S t at e s o f America, 95, 811-817.
doi:10.1073/pnas.95.3.811
Van Essen, D. C., Newsome, W. T. et al. (1984). The visual field rep-
resentation in striate cortex of the macaque monkey: Asymmetries,
anisotropies, and individual variability. Vision Research, 24, 429-
448. doi:10.1016/0042-6989(84)90041-5
Vogel, E. K., & Machizawa, M. G. (2004). Neural activity predicts
individual differences in visual working memory capacity. Nature,
428, 748-751. doi:10.1038/nature02447
Vogel, E. K., McCollough, A. W. et al. (2005). Neural measures reveal
individual differences in controlling access to working memory. Na-
ture, 438, 500-503. doi:10.1038/nature04171
Vogel, E. K., Woodman, G. F. et al. (2001). Storage of features, con-
junctions and objects in visual working memory. Journal of Experi-
mental Psychology: Human Perception and Performance, 27, 92-
114. doi:10.1037/0096-1523.27.1.92
Wade, A. R., Brewer, A. A. et al. (2002). Functional measurements of
human ventral occipital cortex: Retinotopy and colour. Philosophical
Transactions of the Royal Society of London. Series B: Biological
Sciences, 357, 963-973. doi:10.1098/rstb.2002.1108
Wandell, B. A. (1999). Computational neuroimaging of human visual
cortex. Annual Review of Neuroscience, 22, 145-173.
doi:10.1146/annurev.neuro.22.1.145
Wandell, B. A., & Winawer, J. (2011). Imaging retinotopic maps in the
human brain. Vision Research, 51, 718-737.
doi:10.1016/j.visres.2010.08.004
Wandell, B. A., Dumoulin, S. O. et al. (2007). Visual field maps in
human cortex. Neuron, 56, 366-383.
doi:10.1016/j.neuron.2007.10.012
Xu, Y., & Chun, M. M. (2006). Dissociable neural mechanisms sup-
porting visual short-term memory for objects. Nature, 440, 91-95.
doi:10.1038/nature04262
Xu, Y., & Chun, M. M. (2007). Visual grouping in human parietal
cortex. Proceedings of the National Academy of Sciences of the
United States of America, 104, 18766-18771.
doi:10.1073/pnas.0705618104
Xu, Y., & Chun, M. M. (2009). Selecting and perceiving multiple vis-
ual objects. Trends in C ognitive Sciences, 13, 167-174.
doi:10.1016/j.tics.2009.01.008
Zhang, W., & Luck, S. J. (2008). Discrete fixed-resolution representa-
tions in visual working memory. Nature, 453, 233-235.
doi:10.1038/nature06860