The Journal of Neuroscience, January 9, 2013 • 33(2):523–531 • 523

Behavioral/Cognitive

Parietal Cortex Mediates Conscious Perception of Illusory

Gestalt
Natalia Zaretskaya,1 Stuart Anstis,2 and Andreas Bartels1
1Vision and Cognition Laboratory, Centre for Integrative Neuroscience, University of Tübingen, D-72076 Tübingen, Germany, and 2Department of
Psychology, University of California, San Diego, La Jolla, California 92093

Grouping local elements into a holistic percept, also known as spatial binding, is crucial for meaningful perception. Previous studies have
shown that neurons in early visual areas V1 and V2 can signal complex grouping-related information, such as illusory contours or
object-border ownerships. However, relatively little is known about higher-level processes contributing to these signals and mediating
global Gestalt perception. We used a novel bistable motion illusion that induced alternating and mutually exclusive vivid conscious
experiences of either dynamic illusory contours forming a global Gestalt or moving ungrouped local elements while the visual stimulation
remained the same. fMRI in healthy human volunteers revealed that activity fluctuations in two sites of the parietal cortex, the superior
parietal lobe and the anterior intraparietal sulcus (aIPS), correlated specifically with the perception of the grouped illusory Gestalt as
opposed to perception of ungrouped local elements. We then disturbed activity at these two sites in the same participants using trans-
cranial magnetic stimulation (TMS). TMS over aIPS led to a selective shortening of the duration of the global Gestalt percept, with no effect
on that of local elements. The results suggest that aIPS activity is directly involved in the process of spatial binding during effortless
viewing in the healthy brain. Conscious perception of global Gestalt is therefore associated with aIPS function, similar to attention and
perceptual selection.

Introduction al., 2000; Fang et al., 2008). Similarly, when objects occlude each
Visual binding of local elements into a global Gestalt is crucial for other, some neurons in early visual cortex correctly assign border
normal vision. It allows us to perceive scenes and objects as a ownerships, segregating background from foreground (Zhou et
whole rather than as a meaningless collection of individual fea- al., 2000; Qiu and von der Heydt, 2005), with consistent results in
tures. A failure of this ability can lead to the rare symptom of the human (Fang et al., 2009).
simultanagnosia. Patients with this symptom can recognize dis- It has not been established which regions provide the top-
tinct objects of the visual scene or name distinct building blocks down feedback thought to underlie these responses (Zhang and
of hierarchical shapes but cannot perceive the global picture of a von der Heydt, 2010). Physiological evidence suggests that
scene, hampering their daily lives (Balint, 1909; Wolpert, 1924). grouping responses are linked to mechanisms of attentional se-
Because visual grouping happens automatically and effort- lection (Qiu et al., 2007; Fang et al., 2009; Zhang and von der
lessly, it has been suggested to occur before attentional selection Heydt, 2010), a function associated with the anterior intraparietal
(Davis and Driver, 1994; Mattingley et al., 1997). In accord with sulcus (aIPS) (Corbetta and Shulman, 2002). This is consistent
this, most primate electrophysiology and human imaging studies with bilateral lesions in posterior parietal cortex leading to simul-
have addressed Gestalt perception in early and mid-level visual tanagnosia and with substrates involved in guiding attention to
regions. For example, neurons in early visual areas can be mod- global or local aspects of hierarchical stimuli (Fink et al., 1996;
ulated by grouping-related cues, such as interrupted line seg- Mevorach et al., 2006; Romei et al., 2011). However, higher-level
ments forming illusory contours (von der Heydt et al., 1984; neural substrates involved in automatic and spontaneous percep-
Grosof et al., 1993), with corresponding signals found using hu- tual grouping have not been studied in the healthy brain before.
man imaging (Hirsch et al., 1995; Mendola et al., 1999; Seghier et Here we used a novel paradigm that allowed us to investigate
neural mechanisms involved in automatic and spontaneous per-
Received June 19, 2012; revised Oct. 24, 2012; accepted Oct. 27, 2012. ceptual grouping across space in healthy individuals. We used a
Author contributions: N.Z., S.A., and A.B. designed research; N.Z. performed research; N.Z. and A.B. contributed bistable stimulus that led to alternations between the perception
unpublished reagents/analytic tools; N.Z. analyzed data; N.Z. and A.B. wrote the paper. of local, isolated objects and of a grouped illusory Gestalt impres-
This work was funded by the Centre for Integrative Neuroscience Tübingen. We thank Frank Mühlbauer and
Simon Klein for their help in eye tracking and TMS experiments and Stephan de la Rosa for advice on statistical
sion (Anstis and Kim, 2011) (see Fig. 1A). Importantly, it was the
analysis. conscious percept that alternated spontaneously between two
The authors declare no competing financial interests. states rather than instructed attention that focused on one of two
Correspondence should be addressed to Andreas Bartels, Vision and Cognition Laboratory, Centre for Integrative always perceived aspects of a hierarchically grouped stimulus
Neuroscience, University of Tübingen, Otfried-Müller-Strasse 25, 72076 Tübingen, Germany. E-mail:
[email protected].
(Mevorach et al., 2006). Therefore, this stimulus was ideally
DOI:10.1523/JNEUROSCI.2905-12.2013 suited to identify neural mechanisms involved in genuine, auto-
Copyright © 2013 the authors 0270-6474/13/330523-09$15.00/0 matic and perceptually explicit Gestalt formation. We first pre-
524 • J. Neurosci., January 9, 2013 • 33(2):523–531 Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception

sented the stimulus to human volunteers

undergoing fMRI and correlated subjec-
tive perceptual reports of global Gestalt
and of local elements with neural activity.
We then applied transcranial magnetic
stimulation (TMS) over the activated sites
to examine how disruption of their activ-
ity affected global and local perceptual
periods.

Materials and Methods

Participants
A total of 18 healthy subjects (21–33 years old,
nine females, two left-handed, one author)
participated in this study. All had normal or
corrected-to-normal vision and no history of
epilepsy and other neurological disorders. All
subjects gave written informed consent before
participation. The study was conducted ac-
cording to the Declaration of Helsinki and was
approved by the ethics committee of the Uni-
versity Clinic Tübingen.

fMRI Experiment
fMRI stimuli and procedure. The stimuli are il-
lustrated in Figure 1A. They were generated us-
ing Cogent 1.27 (John Romaya, Wellcome
Department of Imaging Neuroscience, Univer-
sity College London, London, UK) on a Win-
dows personal computer running MATLAB
2006b (MathWorks) and presented via a lin-
earized projector with a resolution of 1280 ⫻
1024 and a refresh rate of 60 Hz. Four pairs of
dots were presented on a gray background (90
cd/m 2) at 100% contrast (i.e., maximal lumi-
nance for white dots and minimal luminance
for black dots). All dots had the same contrast
polarity (either black or white). Individual dots
had a size of 0.5°, the distance between dot cen-
ters of each dot pair was 2°, and the distance
between the screen center and each dot pair
center was 5°.
Subjects were asked to fixate a red dot at the
center of the screen and to report their current
spontaneously occurring percept by pressing
and holding down one of the two buttons (one
for global, one for local percepts) with fingers
of their right hand. They were instructed not to
press any button if unsure of their percept.
Brain activity was measured on a 3 tesla Sie-
mens Magnetom TIM system using blood ox-
ygenation level-dependent (BOLD) contrast
with a gradient-echo planar imaging sequence
(repetition time, 2300 ms; echo time, 40 ms; 33
interleaved acquired horizontal slices with in-
plane resolution of 3 ⫻ 3 mm 2; slice thickness,
2.5 mm; gap, 0.5 mm; flip angle, 79°). Each Figure 1. Illustration of stimuli and fMRI paradigm. A, Illustration of the bistable motion stimulus and the two percep-
fMRI run consisted of four periods of 1.5 min tual states. The stimulus consisted of four pairs of dots that moved in-phase on circular paths. Perception alternated
stimulus viewing intermitted by 15 s of fixation spontaneously between two different interpretations: local motion of dot pairs or global motion of two illusory squares
without any stimulus (Fig. 1B). Between dis- (Anstis and Kim, 2011). Subjects fixated a central fixation dot throughout all experiments (for fixation accuracies, see Fig.
tinct 1.5 min periods, the dot-rotation direc- 5). B, fMRI paradigm: 20 stimulus blocks of 1.5 min were separated by blanks (fixation only) of 15 s. During each stimulus
tion and dot-contrast polarity was randomly block, dots were shown in either black or white. Subjects indicated periods of global or local percepts by button presses. C,
varied. The first four volumes were discarded Histograms showing the distributions of normalized dominance durations (normalized by the mean percept duration of
to allow for T1 equilibration effects. This re- each session) for global and local percepts as reported during the fMRI recordings across all subjects, superimposed with
sulted in a total of 187 volumes per run and a fitted gamma functions.
total run duration of 7 min (Fig. 1B). Addition-
ally, a high-resolution structural T1-weighted
image was acquired. Each subject underwent
Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception J. Neurosci., January 9, 2013 • 33(2):523–531 • 525

five to six experimental runs. The speed of dot

rotation was adjusted after each run in an at-
tempt to balance the duration of local and
global motion percepts (with faster rotation
tending to increase local percepts), resulting in
an average ⫾ SD speed of 2.49 ⫾ 0.20 rotations
per second across subjects. Note that this ma-
nipulation did not affect the results, because
run-specific effects were modeled away in the
general linear model (GLM) analysis.
fMRI data analyses. Data were analyzed us-
ing SPM5 (Wellcome Department of Imaging
Neuroscience). Echo planar imaging volumes
were slice-time corrected, motion corrected,
normalized to Montreal Neurological Institute
(MNI) space, and spatially smoothed (Gaussian
filter with 6 mm full-width at half-maximum
for individual and 12 mm for group analysis).
Each subject was analyzed separately using
a standard GLM approach, followed by a
random-effects group analysis. Subject’s re-
ports of global and local percepts were used to
build two regressors modeling onsets of local
and global percepts. Additionally, onsets of the
baseline periods and of the stimulus presenta-
tion (at the beginning of each run and at the
end of each baseline period) were included as
third and fourth regressors. The latter were
used to account for effects related to the ap-
pearance of the stimulus on the screen. General
effects of each run (speed or other unspecific
effects) were modeled with a regressor consist-
ing of a column of ones.
The analysis of fMRI data involved two
steps. First, we identified voxels that were gen-
erally activated by the stimulus compared with
baseline, using the contrast [global:1, local:1,
baseline:⫺2], i.e., voxels in which any linear
combination of activity during global and local
percepts was greater than baseline. This con-
trast would later serve as a mask for statistical
correction of the more specific contrasts. This
contrast was applied in every subject and then
on the group level using a random-effects anal-
ysis with a liberal threshold of p ⬍ 0.05 (uncor-
rected). After this, to identify voxels involved
in perceptual grouping, we performed a voxel-
wise comparison between global and local per-
cept onsets in every individual subject and on
the group level. The random-effects group
analysis for the global versus local comparison
was thresholded at p ⬍ 0.05, familywise error
(FWE) corrected for the volume of stimulus-
activated voxels identified in the first step.
In addition to the analysis described above,
we also tested whether the modeled onsets fully
accounted for the differential activity in global
versus local conditions or whether additional
regressors modeling the steady-state percep-
Figure 2. fMRI activity related to visual stimulation and to global and local perceptual states. A, Brain activity related to both tual periods explained additional variance in
perceptual states compared with fixation only at p ⬍ 0.05 uncorrected, shown on an inflated template brain. The volume of the the data. We repeated the fMRI analysis us-
activated voxels was used for statistical correction of the comparisons between global and local perceptual states. Peak coordinates ing a new GLM that included regressors for
and their significance levels are given in Table 1. B, Activity related to the global perceptual state across the whole brain is shown both onsets and steady-state durations of
in red-to-yellow, and that related to the local percept is in blue. Voxels in the aIPS survived whole-brain FWE correction ( p ⬍ 0.05); each condition, thus accounting for onset-
both parietal clusters (aIPS and SPL) survived FWE correction for the mask shown in A. For details, see Table 2. For illustration, maps specific variance in addition to steady-state-
are shown at p ⬍ 0.0001 (uncorrected). C, Parietal activation for the contrast “global versus local,” shown for three representative related variance.
participants. LH, Left hemisphere; RH, right hemisphere. Finally, we examined the behavioral data to
ascertain that differences in brain activity be-
tween global and local perceptual states cannot
526 • J. Neurosci., January 9, 2013 • 33(2):523–531 Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception

Figure 3. TMS stimulation sites and experimental paradigm. A, The three sites targeted by TMS shown on a three-dimensional reconstruction of the brain and on the head surface of one
representative participant. B, Experimental design of the TMS experiment. Each of the three TMS sites depicted in A was targeted on a separate day, counterbalanced across subjects. On each day,
two runs of behavioral experiments were recorded before and four runs after TMS of the target site.

be explained by potential differences in durations of transitions from defined vertex (the highest point on a subject’s head; see Fig. 3A) as a
global to local perceptual state and vice versa, as it has been reported control stimulation site to account for the unspecific effects of
recently for bistable perception (Knapen et al., 2011). Transitions were stimulation.
defined as periods during which no button was pressed. Periods during TMS experiment procedure. Subjects were seated in a chair resting their
which both buttons were pressed simultaneously were also counted as head on a chin rest. The stimulus was presented on a Dell 2007FP mon-
transitions. Median durations of each transition type (i.e., global to local itor with a resolution of 1600 ⫻ 1200 pixels and a refresh rate of 60 Hz at
or vice versa) were compared in a paired t test across subjects. a distance of 40 cm away from the observer. The stimulus had the same
configuration as during the fMRI experiment with a larger size (dot size,
TMS experiment
1°; screen center to dot pair distance, 10°; distance between paired dots,
In the subsequent TMS experiment, we tested whether disturbing activity
of areas that were significantly more active during global compared with 2°). The dots were black throughout all TMS experiments, presented on
local percepts would affect perceptual grouping processes. a gray screen (91 cd/m 2). Each experimental run consisted of four 50 s
Planning of the stimulation sites. Based on the consistent fMRI results trials separated by 10 s of rest (Fig. 3B). During rest, subjects were allowed
across the individual subjects, which were reflected in the group statistics to close their eyes and look freely around on the screen.
(see Results and Fig. 2), the two right parietal sites that were more acti- Each of the three TMS sites was tested on a separate day, with the order
vated with global compared with local percepts were chosen for stimula- of the sites randomized across subjects (Fig. 3B). On the first day, sub-
tion: the aIPS and the superior parietal lobe (SPL). These sites were jects’ active motor threshold was determined (defined as the TMS inten-
localized in each subject’s right hemisphere using individual fMRI results sity at which 5 of 10 pulses over the left motor cortex elicit a visible muscle
(Figs. 2C, 3A). The average MNI coordinates of the two stimulated sites twitch while subjects hold their right-hand fingertips contracted), and
were as follows: mean ⫾ SD of the SPL: x ⫽ 20.70 ⫾ 5.28, y ⫽ ⫺66.67 ⫾ several sessions were undertaken to adjust the stimulus velocity to opti-
7.06, z ⫽ 50.03 ⫾ 4.30; mean ⫾ SD of the aIPS: x ⫽ 32.46 ⫾ 4.39, y ⫽ mize the balance between local and global percepts. The individual sub-
⫺46.66 ⫾ 7.01, z ⫽ 48.79 ⫾ 7.18 (note that small discrepancies of these ject’s speed (mean ⫾ SD, 2.58 ⫾ 0.11 rotations per second) was then kept
coordinates with those reported for the random-effects analysis are at- constant for all of the following days. For TMS coil positioning, the
tributable to differences in the averaging approach and smoothing in- subject’s head was then coregistered with their individual structural MR
volved in each method). In addition to the functional sites, we also image using surface markers and a camera-based stereotactic neuronavi-
Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception J. Neurosci., January 9, 2013 • 33(2):523–531 • 527

gation system (LOCALITE). After registration, subjects completed four Table 1. Brain regions, coordinates, cluster sizes, peak z values, and p values of
runs with the same instructions as described above for the fMRI experi- voxel clusters preferring either global or local perceptual states compared with
ment. The first two runs of each experimental day served to accustom gray baseline in the whole-brain random-effects fMRI group analysis
subjects to the setting and to achieve a stable baseline and were later Cluster size z p p
discarded, because perceptual durations in binocular rivalry, as well as in Brain region x, y, z (mm) (voxels) (peak) (uncorrected) (FWE-corrected)
our bistable global–local stimulus are known to change significantly
within the first few minutes of viewing before they reach a more steady L. extrastriate cortex ⫺50, ⫺74, 00 1799 5.45 ⬍0.001 0.001
level (Suzuki and Grabowecky, 2007; Anstis and Kim, 2011). All remain- R. extrastriate cortex 46, ⫺68, ⫺02 1835 4.65 0.001 0.033
ing runs entered the analysis described below. After the fourth run, 48 s of R. SPL/aIPS 28, ⫺56, 54 445 2.89 0.002 0.995
continuous theta burst stimulation (cTBS) (three pulses at 50 Hz re- L. aIPS ⫺26, ⫺56, 52 60 2.22 0.013 1.000
peated every 0.2 s, resulting in 600 pulses in total) was applied over one of Corpus collosum 04, 00, 22 2 1.75 0.04 1.000
the three locations: right SPL, right aIPS, or vertex. cTBS has been shown The activated volume, thresholded at p ⬍ 0.05 (uncorrected), was used for statistical correction in the subsequent
to have an inhibitory effect on neural function for several minutes after analysis. Anatomical locations were identified with the help of the SPM Anatomy toolbox (Simon Eickhoff, Research
Center Jülich, Jülich, Germany). L., Left; R., right.
its application (Huang et al., 2005). Pulses were delivered by a figure-of-
eight coil (MC-B70) connected to a MagPro X100 stimulator (MagVen-
ture). Intensity of the stimulation was set to 80% of individual active
motor threshold, resulting in a mean ⫾ SD intensity of 25.89 ⫾ 4.64% of
global percept lasted on average 7.02 ⫾ 3.69 s (mean ⫾ SD) and
stimulator output. Immediately after stimulation, subjects completed
another four runs.
those of the local percept 5.68 ⫾ 2.25 s. The distributions of
TMS data analyses. First, the onset times and durations of all global durations of both percepts were well fitted by a gamma function
and local percepts were determined. After this, the durations of the first [maximum likelihood fit coefficient of determination (r 2)
global and the first local percepts within every 50 s trial were identified global ⫽ 0.987, local ⫽ 0.983; Fig. 1C]. Observers reported the
and averaged separately for trials before and after TMS. We used the first transitions in this bistable display to be perceptually immediate.
global and local percepts of every trial to ensure that the number of data The average median ⫾ SD of transition periods across subjects
points was the same for each subject in every condition despite the vari- was 50.7 ⫾ 34.0 ms for global percepts and 52.0 ⫾ 37.3 ms for
able percept durations (the number of perceptual phases that could be local percepts, respectively, with no significant difference be-
accommodated into a 50 s trial varied from 2 to 22). This also helped us tween them (t(17) ⫽ ⫺0.202, p ⫽ 0.843). Therefore, differential
to minimize time-dependent effects within each trial (Anstis and Kim,
durations of transition periods from local to global and vice versa
2011). Trials before TMS were regarded as “baseline.” Therefore, for
every percept type (global, local), the average percept duration after the
cannot account for neural activity differences. This stands in con-
TMS was normalized by its baseline. This resulted in two values per trast to binocular rivalry, in which transition periods are much
subject: normalized duration of the local percept and normalized dura- longer and highly variable (Knapen et al., 2011).
tion of the global percept. Each value represents the perceptual duration Analysis of eye-tracking data recorded throughout all fMRI
after TMS relative to that preceding TMS. These values were used to sessions revealed no differences in fixation accuracy during per-
conduct a repeated-measures one-way ANOVA over different TMS sites ception of global Gestalt compared with ungrouped items (aver-
separately for each percept type. age distance from fixation, t(17) ⫽ ⫺1.26, p ⫽ 0.226; variance of
the eye position, t(17) ⫽ ⫺1.570, p ⫽ 0.135; see also Fig. 5).
Eye tracking
To examine whether patterns of eye movements during fixation were
related to perceiving local and global interpretations of the stimulus, we fMRI: activity fluctuations correlating with global
monitored eye position throughout all fMRI and TMS experiments in Gestalt perception
every subject. During fMRI experiments, pupil position was acquired fMRI activity related to both perceptual states compared with
using the infrared camera-based ASL EyeTrac 6 Eye Tracking System gray baseline periods involved a bilateral site corresponding to
(Applied Science Laboratories) at a sampling rate of 60 Hz. During TMS putative extrastriate area V5⫹/MT⫹ and two superior parietal
experiments, eye position was monitored using a pupil-glint vector with regions, shown in Figure 2A (Table 1 lists peak coordinates and
an Arrington Viewpoint infrared eye-tracking system at a sampling rate
statistical information of each cluster). The comparison of brain
of 220 Hz (Arrington Research).
The preprocessing of eye-position data acquired during fMRI and
activity related to global percepts compared with local ones re-
TMS experiments included blink interpolation using the nearest- vealed greater activity of areas of the right posterior parietal cor-
neighbor method and smoothing of x and y gaze positions with a 200 ms tex, as well as right and left putamen (not visible in the figure),
running average window. After this, the absolute distance of gaze from shown in Figure 2B and Table 2. The maximum was located in the
the fixation dot was calculated for each sampled time point. For the fMRI aIPS, with a second peak in the SPL. Both peaks were significant
experiment, these distances were sorted according to whether they at p ⬍ 0.05, FWE corrected for the volume of voxels active during
stemmed from local or global perceptual states and used to compute two both percepts compared with baseline as shown in Figure 2A.
values: (1) the average distance between the gaze position and the fixation Both parietal sites were also reliably activated in each individual
dot, and (2) the variance, which quantifies the variability in the eye subject, reaching p ⬍ 0.05 (FWE whole-brain corrected) in 12 of
position relative to the fixation dot. These two values were computed for
18 subjects, as illustrated in Figure 2C. In the inverse comparison
each subject and each percept type. Two-tailed paired-sample t tests
between global and local conditions were performed for the average and
of local perceptual states with global ones, the right fusiform
the variance. For the TMS experiment, distances were sorted according gyrus was most active, with its activation cluster extending into
to percept type and stimulation site. After this, the average and the vari- early visual cortex (Fig. 2B; Table 2 lists coordinates of all regions
ance values of each subject, normalized by their pre-TMS baseline values, differentially involved in either direction).
were submitted to two one-way ANOVAs, one for global and one for The comparison of the steady-state global versus local condi-
local percept, with “stimulation site” as factor. tions in a GLM that included separate regressors for onsets and
steady state of each percept revealed the same right aIPS cluster
Results that was found in the original GLM that modeled onsets only, yet
fMRI: behavior and eye movements with somewhat reduced significance (cluster size of 167 voxels at
Analysis of behavioral responses during fMRI scanning showed p ⬍ 0.001 uncorrected, with peak coordinates of x ⫽ 32, y ⫽ ⫺48,
that, across all sessions of all 18 subjects, the durations of the z ⫽ 70). This finding suggests that not only onsets of global
528 • J. Neurosci., January 9, 2013 • 33(2):523–531 Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception

Table 2. Brain regions, coordinates, cluster sizes, peak z values, and p values of voxel clusters preferring either global-to-local perceptual states or local-to-global
perceptual states in the whole-brain random-effects fMRI group analysis
Cluster size
Brain region x, y, z (mm) (voxels) z (peak) p (FWE-corrected mask) p (FWE-corrected whole brain)
Global ⬎ local
R. aIPS/area 2 34, ⫺40, 52 169 5.46 0.0001 0.002
R. SPL 22, ⫺60, 58 55 3.69 0.0349 0.303
L. putamen ⫺28, ⫺04, ⫺04 275 5.26 n/a 0.004
R. putamen 26, ⫺04, 10 309 4.80 n/a 0.028
Local ⬎ global
R./L. fusiform gyrus ⫾30, ⫺71, ⫺09 3004 4.71 0.003 0.0400
L. rolandic operculum ⫺46, ⫺04, 04 288 5.21 n/a 0.0053
L. area 6 ⫺14, 02, 66 2100 4.85 n/a 0.0235
L. insula lobe ⫺32, 26, 00 265 4.85 n/a 0.0238
L. postcentral gyrus ⫺46, ⫺20, 62 238 4.76 n/a 0.0328
R. inferior frontal gyrus 40, 26, 04 799 4.65 n/a 0.0490
(pars triangularis)
L. inferior frontal gyrus ⫺44, 06, 28 292 4.51 0.008 0.0825
(pars opercularis)
R. precentral gyrus 48, 04, 44 90 4.09 0.05 0.2987
To match Figure 2, clusters reaching a threshold of p ⬍ 0.0001 (uncorrected) and surviving p ⬍ 0.05 at the cluster level are included. p values at FWE correction for the whole brain as well as for the volume of visually activated voxels are
provided. Anatomical locations were identified with the help of the SPM Anatomy toolbox (Simon Eickhoff, Research Center Jülich). *n/a indicates voxels outside of the mask volume. L., Left; R., right.

percepts but also their continuous presence were correlated with

elevated activity of the right posterior parietal cluster.

TMS: interfering with perception of global Gestalt

Because perception of the local elements and of the global Gestalt
excluded each other and because the physical stimulus remained
the same during both perceptual states, the activity we observed
in fMRI must be related to the perception of grouped Gestalt.
However, the fMRI data leave it open whether the parietal regions
were activated as a consequence of the global Gestalt perception
(e.g., attributable to involuntary attention drawn to it) or
whether they were involved in its generation. We therefore tested
whether interfering with neural processing in the activated pari-
etal areas would selectively impair Gestalt perception. We mea- Figure 4. Perceptual effects of TMS on global and local states. Effects of TMS at each of the
three stimulation sites on global (A) and local (B) percept durations. Bars show normalized
sured global and local percept durations before and after
dominance durations after TMS (mean ⫾ SE across subjects). Normalization consisted of divid-
application of an inhibitory TMS protocol (cTBS) over three sites ing the post-TMS dominance duration by the pre-TMS dominance duration. Note that, consis-
(aIPS, SPL, and, as control site, vertex), with experiments per- tent with previous behavioral observations (Anstis and Kim, 2011), the duration of the global
formed on separate and counterbalanced days for each site (see percept increased with time, and this effect was overridden and reversed after right aIPS stim-
Materials and Methods). Figure 4 shows normalized percept du- ulation. There were no effects for the local percept. Error bars indicate SEM. *p ⬍ 0.006 (Bon-
rations (i.e., duration post-TMS/pre-TMS) obtained for each ferroni corrected).
stimulation site, separately for global and local percepts. There
was a significant effect of stimulation site for the durations of the immediate (average median ⫾ SD across subjects was 52.33 ⫾
global Gestalt percept (F(2,34) ⫽ 5.02, p ⫽ 0.012) but not for the 33.98 ms).
local one (F(2,34) ⫽ 0.13, p ⫽ 0.877). Post hoc tests showed that the Finally, we performed an analysis to see whether the behav-
effect on global durations was driven by TMS over the aIPS (Fig. ioral effect size of TMS application correlated with the strength of
4), which significantly and substantially (24%) reduced the global the BOLD signal of the stimulated site, as we observed in a pre-
percept duration after stimulus onset compared with TMS over vious study (Zaretskaya et al., 2010). As a measure of fMRI effects,
the vertex control condition (t(17) ⫽ ⫺3.655, p ⫽ 0.006 Bonfer- we used the difference between mean ␤ estimates in global and
roni corrected for three comparisons), with no effect for SPL local conditions of the fMRI experiment in a 6 ⫻ 6 ⫻ 6 mm
(t(17) ⫽ ⫺1.374, p ⫽ 0.287 uncorrected). The effects of aIPS– region of interest around the individual aIPS coordinates of each
TMS on global durations were significantly larger than those on subject. TMS effects were defined as difference between effects of
local durations (one-sided paired t test for aIPS minus vertex stimulation on global and local perceptual states (percentage
between global and local percepts, t(17) ⫽ 2.185, p ⫽ 0.0215). change in global minus percentage change in local). However, no
As in the fMRI experiment, there were no differences in fixa- significant correlation was found (r ⫽ ⫺0.040, p ⫽ 0.874).
tion accuracy across different conditions of the TMS experiment
(average distance from fixation during global percept, F(2,34) ⫽ Discussion
0.176, p ⫽ 0.839; average distance from fixation during local Our bistable stimulus allowed us to study neural correlates of two
percept, F(2,34) ⫽ 0.861, p ⫽ 0.432; variance of the eye position entirely distinct visual percepts that alternated spontaneously:
during global percept, F(2,34) ⫽ 0.719, p ⫽ 0.500; variance during one consisted of spatially separate local elements, and the other
local percept, F(2,34) ⫽ 2.950, p ⫽ 0.066; for details, see Fig. 5 and grouped the elements into a global Gestalt. Importantly, the per-
Materials and Methods), and perceptual transitions remained cepts were vivid, mutually exclusive, and occurred without ex-
Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception J. Neurosci., January 9, 2013 • 33(2):523–531 • 529

typical simultanagnosic patients who usu-

ally fail, she could perceive the global as-
pect of hierarchical letter stimuli on half
of the trials but failed to do so on the other
half, with successful trials correlating with
parietal activity close to our aIPS region
(Himmelbach et al., 2009). Together with
our findings, this evidence suggests that
the same machinery is involved in vol-
untary attentional selection of global
stimulus aspects and in the spontaneous
formation of a global conscious percept
during effortless viewing.
However, a difference exists in that
voluntary attention to both global and lo-
cal aspects of hierarchical stimuli could be
biased by parietal TMS in previous studies
(Mevorach et al., 2006, 2010), whereas
during spontaneous perception, we could
only bias global but not local perception.
This divergence in results is likely to re-
flect a difference in mechanisms of per-
ceiving local elements that constitute a
whole as opposed to directing attention to
local features, which also requires the ac-
tive suppression of global aspects, the lat-
Figure 5. Eye movement data. Two-dimensional histograms of the eye position of one representative participant during global ter being an established parietal function
(left column) and local (right column) stimulus perception. Fixation data is shown separately for the fMRI (top row) and TMS (Mevorach et al., 2010). Consistent with
(bottom row) experiments. Group analysis of eye-position data using average distance of gaze from fixation as well as the variance our findings, perception of local elements
of this distance during both fMRI and TMS experiments revealed no difference in fixation accuracy across the different conditions. rather than attention to them has been as-
For details, see Materials and Methods. sociated with activity in the posterior part
of the fusiform cortex in the vicinity of
hV4 and of encroaching medial parts of
plicit voluntary attention or physical manipulations of the the lateral occipital complex, which are involved in processing
stimuli. fMRI experiments showed that activity in aIPS was cor- simple local shapes (Lerner et al., 2001; Hayworth and Bieder-
related with conscious perception of the whole compared with man, 2006; Vinberg and Grill-Spector, 2008).
perception of its parts. Disturbing this activity using an inhibitory Another group of previous studies manipulated physical con-
TMS protocol substantially shortened the perceptual periods of figurations, such as inducers of a Kanizsa shape, to induce global
the global Gestalt without affecting that of the local elements. or local percepts (Hirsch et al., 1995; Mendola et al., 1999; Seghier
These results indicate an important role of aIPS in perceptual et al., 2000; Kourtzi et al., 2003; Fang et al., 2008). Most of these
grouping. studies explicitly limited their data acquisition or analyses of
A number of previous studies asked subjects to actively guide grouping-related activity to visual cortex and were therefore un-
voluntary attention to either global or local aspects of already able to examine parietal involvement. However, similar to two of
grouped hierarchical structures (Fink et al., 1996; Mevorach et the reports with good control over low-level stimulus features
al., 2006) and to actively ignore the task-irrelevant part of the (Fang et al., 2008; McMains and Kastner, 2011), we observed a
stimulus (Mevorach et al., 2010). However, this approach does deactivation of early visual areas during Gestalt perception
not compare processes related to Gestalt formation and percep- relative to baseline and also relative to the “local” perceptual
tion with their absence, because Gestalt is always present and state (Fig. 2 A, B). Because our conditions were physically
perceived. These studies indicate that the right posterior parietal identical, this deactivation is likely to be attributable to top-
cortex, including aIPS, plays a direct and important role in direct- down modulation.
ing attention between the local and the global stimulus aspect Several hypothetical explanations may account for this. For
(Mevorach et al., 2006; Romei et al., 2011), a region previously one, it could be viewed in context of predictive coding (Rao and
associated with spatial attention (Corbetta and Shulman, 2002; Ballard, 1999; Hohwy et al., 2008). According to this model, the
Yantis and Serences, 2003) and perceptual selection. In contrast sensory information provided by early visual cortex is compared
to this approach, which investigated primarily directed attention, with higher-level predictions provided by extrastriate, parietal, or
our study relied on spontaneous and vivid perceptual alterna- prefrontal regions about a global or higher-level interpretation
tions, without changes in voluntary attention. The anatomically that may lead to the sensory input. Lower-level activity is can-
coincident location in the right IPS in those studies with that celled if it matches predictions of higher-level regions, such that
identified here therefore points to a convergent mechanism in- mainly prediction errors are propagated toward higher levels that
volved in directed attentional selection and in perception of Ge- can then adjust their predictions accordingly. In the present
stalt. A similar location has also been reported in a case of a rare study, aIPS would be the candidate site encoding the high-level
simultanagnosia patient. Unlike healthy humans who usually can prediction. Consistency between prediction and sensory input
perceive the global aspect of hierarchical letter stimuli and unlike would then reduce early visual activity, as observed here during
530 • J. Neurosci., January 9, 2013 • 33(2):523–531 Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception

global perception (Fig. 6). This is plausi-

ble mechanistically, because previous
studies have shown that aIPS can modu-
late early visual areas (Ruff et al., 2009).
Consistent with the present study and
with previous grouping studies, deactiva-
tion of striate cortex has also been re-
ported during form perception with
various shape stimuli (Murray et al.,
2002), as well as during viewing of coher-
ent or predictable motion as opposed to
random motion during viewing of ab-
stract as well as natural scenes (Bartels
et al., 2008; Alink et al., 2010).
An alternative explanation for deactiva-
tion of early visual areas that is potentially
related to predictive coding concerns the
so-called border-ownership responses.
Border-ownership responses have been
observed in a subset of edge-responsive
neurons in V1 and V2 (Zhou et al., 2000;
Dong et al., 2008). These neurons are
modulated by the side on which the object
in the foreground is located that “owns”
the edge. Because information regarding
object identity lies far beyond the field of
view (and beyond the functional proper-
ties) of these neurons, it must be conveyed
Figure 6. Interpretation of aIPS and visual cortex activity in context of predictive coding. The following reasoning could provide
to these neurons from higher levels. It is one possible interpretation of our data. During global perception, aIPS facilitates spatial binding of the sensory input and the
currently unknown which high-level area encoding of the global interpretation in terms of two large moving squares. This function is associated with higher activity in aIPS.
endows this special subset of edge- According to predictive coding theory, aIPS feeds its high-level interpretation/prediction of the sensory input back to early visual
responsive neurons in V1 and V2 with the cortex. Because the prediction matches the sensory input, the two cancel each other out, leaving minimal prediction error and low
required “knowledge” about object seg- activity in early visual cortex. Local perception arises when aIPS has low activity (or is disturbed by TMS) and does not spatially bind
mentation (Zhang and von der Heydt, or encode a global interpretation, leading to higher activity in early cortex because no prediction cancels the sensory input.
2010), but our results point to aIPS as a
potential source of this high-level grouping signal because it was 2011). Our evidence supports the notion of a shared anatom-
most strongly activated during the “global” compared with “lo- ical substrate in aIPS.
cal” perceptual state and because it was causally involved in form- Finally, aIPS has been implicated previously not only in atten-
ing the global percept. tion but also in perceptual selection during bistable perception
Intriguingly, this hypothesis is indirectly supported by elec- (Kleinschmidt et al., 1998; Lumer et al., 1998; Sterzer and Klein-
trophysiology and fMRI experiments that revealed an interaction schmidt, 2007; Carmel et al., 2010; Zaretskaya et al., 2010; Kanai
between the border-ownership response in early visual areas with et al., 2011). Importantly, however, TMS over this site changed
top-down attention (Qiu et al., 2007; Fang et al., 2009). These the frequency of perceptual alternations for both stimuli equally,
experiments showed that border-ownership responses were thus indicating that generic mechanisms of perceptual stability or
strongly modulated by attention (Qiu et al., 2007). This has led to selection had been disturbed (Carmel et al., 2010; Zaretskaya et
al., 2010; Kanai et al., 2011). One study interpreted this finding in
the suggestion that mechanisms related to figure– ground seg-
context of above-mentioned predictive coding theory, in that
mentation are tied to those of attentional selection (Qiu et al.,
aIPS may encode the higher-level prediction about the sensory
2007; Fang et al., 2009), which in turn is mediated by aIPS (Cor-
input, whereas a posterior site, in which TMS had opposite ef-
betta and Shulman, 2002). Our results therefore provide direct
fects, may encode the error signal (Kanai et al., 2010, 2011). At
support for the above suggestion. Notably, also monkey IPS has least our above interpretation of aIPS encoding the higher-level
been found to convey grouping responses that are attentionally prediction (two large moving squares) on what may have caused
modulated (Yokoi and Komatsu, 2009). the sensory input (rotating dots) is consistent with this: disturb-
The idea of a functional relationship between grouping and ing aIPS using TMS in the current study biased the alternation
attention also fits well with a recent study examining percep- process, selectively reducing the percept corresponding to the
tual competition between multiple items and the available high-level prediction/interpretation of the input (Fig. 6). How-
attentional resources that can modulate responses in early vi- ever, as explained above, our data point to visual cortex as the site
sual cortex. The study examined responses in early visual areas of prediction-error calculation, because it reduced activity during
V1–V4 to parametrically varied grouping strength of Kanizsa global perception, i.e., when the high-level prediction matched
inducers and found that grouping strength parametrically the sensory input and therefore cancelled the prediction error.
limited the amount of attentional modulation of early visual The above interpretations provide a theoretical framework for
responses, suggesting that the resources for grouping and at- the notion that the mechanisms of attention, binding, and per-
tentional selection may be shared (McMains and Kastner, ceptual selection might be closely related (Bartels, 2009), and the
Zaretskaya et al. • Parietal Cortex Mediates Gestalt Perception J. Neurosci., January 9, 2013 • 33(2):523–531 • 531

present findings suggest convergence of these functions in the gration of local features into global shapes: monkey and human FMRI
aIPS. studies. Neuron 37:333–346. CrossRef Medline
Lerner Y, Hendler T, Ben-Bashat D, Harel M, Malach R (2001) A hierarchi-
cal axis of object processing stages in the human visual cortex. Cereb
References Cortex 11:287–297. CrossRef Medline
Alink A, Schwiedrzik CM, Kohler A, Singer W, Muckli L (2010) Stimulus Lumer ED, Friston KJ, Rees G (1998) Neural correlates of perceptual rivalry
predictability reduces responses in primary visual cortex. J Neurosci 30: in the human brain. Science 280:1930 –1934. CrossRef Medline
2960 –2966. CrossRef Medline Mattingley JB, Davis G, Driver J (1997) Preattentive filling-in of visual sur-
Anstis S, Kim J (2011) Local versus global perception of ambiguous motion faces in parietal extinction. Science 275:671– 674. CrossRef Medline
displays. J Vis 11(3):13 1–12. CrossRef Medline McMains S, Kastner S (2011) Interactions of top-down and bottom-up
Balint R (1909) Seelenlähmung des Schauens, optische Ataxie, raümliche mechanisms in human visual cortex. J Neurosci 31:587–597. CrossRef
Störung der Aufmerksamkeit. Monatsschr Psychiat Neurol 25:51– 81. Medline
CrossRef Mendola JD, Dale AM, Fischl B, Liu AK, Tootell RB (1999) The representa-
Bartels A (2009) Visual perception: converging mechanisms of attention, tion of illusory and real contours in human cortical visual areas revealed
binding, and segmentation? Curr Biol 19:R300 –R302. CrossRef Medline by functional magnetic resonance imaging. J Neurosci 19:8560 – 8572.
Bartels A, Zeki S, Logothetis NK (2008) Natural vision reveals regional spe- Medline
Mevorach C, Humphreys GW, Shalev L (2006) Opposite biases in salience-
cialization to local motion and to contrast-invariant, global flow in the
based selection for the left and right posterior parietal cortex. Nat Neuro-
human brain. Cereb Cortex 18:705–717. CrossRef Medline
sci 9:740 –742. CrossRef Medline
Carmel D, Walsh V, Lavie N, Rees G (2010) Right parietal TMS shortens
Mevorach C, Hodsoll J, Allen H, Shalev L, Humphreys G (2010) Ignoring
dominance durations in binocular rivalry. Curr Biol 20:R799 –R800.
the elephant in the room: a neural circuit to downregulate salience. J Neu-
CrossRef Medline
rosci 30:6072– 6079. CrossRef Medline
Corbetta M, Shulman GL (2002) Control of goal-directed and stimulus-
Murray SO, Kersten D, Olshausen BA, Schrater P, Woods DL (2002) Shape
driven attention in the brain. Nat Rev Neurosci 3:201–215. CrossRef
perception reduces activity in human primary visual cortex. Proc Natl
Medline
Acad Sci U S A 99:15164 –15169. CrossRef Medline
Davis G, Driver J (1994) Parallel detection of Kanizsa subjective figures in
Qiu FT, von der Heydt R (2005) Figure and ground in the visual cortex: v2
the human visual system. Nature 371:791–793. CrossRef Medline
combines stereoscopic cues with gestalt rules. Neuron 47:155–166.
Dong Y, Mihalas S, Qiu F, von der Heydt R, Niebur E (2008) Synchrony and
CrossRef Medline
the binding problem in macaque visual cortex. J Vis 8(7):30 1–16.
Qiu FT, Sugihara T, von der Heydt R (2007) Figure-ground mechanisms
CrossRef Medline
provide structure for selective attention. Nat Neurosci 10:1492–1499.
Fang F, Kersten D, Murray SO (2008) Perceptual grouping and inverse CrossRef Medline
fMRI activity patterns in human visual cortex. J Vis 8(7):2 1–9. CrossRef Rao RP, Ballard DH (1999) Predictive coding in the visual cortex: a func-
Medline tional interpretation of some extra-classical receptive-field effects. Nat
Fang F, Boyaci H, Kersten D (2009) Border ownership selectivity in human Neurosci 2:79 – 87. CrossRef Medline
early visual cortex and its modulation by attention. J Neurosci 29:460 – Romei V, Driver J, Schyns PG, Thut G (2011) Rhythmic TMS over parietal
465. CrossRef Medline cortex links distinct brain frequencies to global versus local visual process-
Fink GR, Halligan PW, Marshall JC, Frith CD, Frackowiak RS, Dolan RJ ing. Curr Biol 21:334 –337. CrossRef Medline
(1996) Where in the brain does visual attention select the forest and the Ruff CC, Blankenburg F, Bjoertomt O, Bestmann S, Weiskopf N, Driver J
trees? Nature 382:626 – 628. CrossRef Medline (2009) Hemispheric differences in frontal and parietal influences on hu-
Grosof DH, Shapley RM, Hawken MJ (1993) Macaque V1 neurons can sig- man occipital cortex: direct confirmation with concurrent TMS-fMRI.
nal “illusory” contours. Nature 365:550 –552. CrossRef Medline J Cogn Neurosci 21:1146 –1161. CrossRef Medline
Hayworth KJ, Biederman I (2006) Neural evidence for intermediate repre- Seghier M, Dojat M, Delon-Martin C, Rubin C, Warnking J, Segebarth C,
sentations in object recognition. Vision Res 46:4024 – 4031. CrossRef Bullier J (2000) Moving illusory contours activate primary visual cortex:
Medline an fMRI study. Cereb Cortex 10:663– 670. CrossRef Medline
Himmelbach M, Erb M, Klockgether T, Moskau S, Karnath HO (2009) Sterzer P, Kleinschmidt A (2007) A neural basis for inference in perceptual
fMRI of global visual perception in simultanagnosia. Neuropsychologia ambiguity. Proc Natl Acad Sci U S A 104:323–328. CrossRef Medline
47:1173–1177. CrossRef Medline Suzuki S, Grabowecky M (2007) Long-term speeding in perceptual switches
Hirsch J, DeLaPaz RL, Relkin NR, Victor J, Kim K, Li T, Borden P, Rubin N, mediated by attention-dependent plasticity in cortical visual processing.
Shapley R (1995) Illusory contours activate specific regions in human Neuron 56:741–753. CrossRef Medline
visual cortex: evidence from functional magnetic resonance imaging. Vinberg J, Grill-Spector K (2008) Representation of shapes, edges, and sur-
Proc Natl Acad Sci U S A 92:6469 – 6473. CrossRef Medline faces across multiple cues in the human visual cortex. J Neurophysiol
Hohwy J, Roepstorff A, Friston K (2008) Predictive coding explains binoc- 99:1380 –1393. CrossRef Medline
ular rivalry: an epistemological review. Cognition 108:687–701. CrossRef von der Heydt R, Peterhans E, Baumgartner G (1984) Illusory contours and
Medline cortical neuron responses. Science 224:1260 –1262. CrossRef Medline
Huang YZ, Edwards MJ, Rounis E, Bhatia KP, Rothwell JC (2005) Theta Wolpert I (1924) Die Simultagnosie: Störung der Gesamtauffassung.
burst stimulation of the human motor cortex. Neuron 45:201–206. Zeitschrift für Gesamte Neurologie und Psychiatrie 93:397– 415. CrossRef
CrossRef Medline Yantis S, Serences JT (2003) Cortical mechanisms of space-based and
Kanai R, Bahrami B, Rees G (2010) Human parietal cortex structure pre- object-based attentional control. Curr Opin Neurobiol 13:187–193.
dicts individual differences in perceptual rivalry. Curr Biol 20:1626 –1630. CrossRef Medline
CrossRef Medline Yokoi I, Komatsu H (2009) Relationship between neural responses and vi-
Kanai R, Carmel D, Bahrami B, Rees G (2011) Structural and functional sual grouping in the monkey parietal cortex. J Neurosci 29:13210 –13221.
fractionation of right superior parietal cortex in bistable perception. Curr CrossRef Medline
Biol 21:R106 –R107. CrossRef Medline Zaretskaya N, Thielscher A, Logothetis NK, Bartels A (2010) Disrupting
Kleinschmidt A, Büchel C, Zeki S, Frackowiak RS (1998) Human brain ac- parietal function prolongs dominance durations in binocular rivalry.
tivity during spontaneously reversing perception of ambiguous figures. Curr Biol 20:2106 –2111. CrossRef Medline
Proc Biol Sci 265:2427–2433. CrossRef Medline Zhang NR, von der Heydt R (2010) Analysis of the context integration
Knapen T, Brascamp J, Pearson J, van Ee R, Blake R (2011) The role of mechanisms underlying figure-ground organization in the visual cortex.
frontal and parietal brain areas in bistable perception. J Neurosci 31: J Neurosci 30:6482– 6496. CrossRef Medline
10293–10301. CrossRef Medline Zhou H, Friedman HS, von der Heydt R (2000) Coding of border owner-
Kourtzi Z, Tolias AS, Altmann CF, Augath M, Logothetis NK (2003) Inte- ship in monkey visual cortex. J Neurosci 20:6594 – 6611. Medline