articleseidemannlab.cps.utexas.edu/~eyal/publications/chen... · 2012-05-11 · neuron article...
TRANSCRIPT
Neuron
Article
Attentional Modulations Relatedto Spatial Gating but Not to Allocationof Limited Resources in Primate V1Yuzhi Chen1 and Eyal Seidemann1,*1Center for Perceptual Systems, Department of Psychology and Section of Neurobiology, University of Texas at Austin, Austin,
TX 78712, USA
*Correspondence: [email protected] 10.1016/j.neuron.2012.03.033
SUMMARY
Attention can modulate neural responses in sensorycortical areas and improve behavioral performanceinperceptual tasks.However, the nature andpurposeof these modulations remain under debate. Herewe used voltage-sensitive dye imaging (VSDI) tomeasure V1 population responses while monkeysperformed a difficult detection task under focal ordistributed attention. We found that V1 responses atattended locations are significantly elevated relativeto actively ignored or irrelevant locations, consistentwith the hypothesis that an important goal of atten-tion in V1 is to highlight task-relevant information.Surprisingly, these modulations were indistinguish-able under focal and distributed attention, suggest-ing a minor or no role for attention as a mechanismfor allocating limited representational resources inV1. The response elevation at attended locations isadditive, is widespread, and starts shortly beforestimulus onset. This elevation could contribute tospatial gating by biasing competition in subsequentprocessing stages in favor of attended stimuli.
INTRODUCTION
Behavioral performance in perceptual tasks such as detection
and discrimination can be significantly improved by prior knowl-
edge regarding the stimulus’s location and features (e.g., Dun-
can, 1980; Posner, 1980). This improvement is thought to be
mediated by top-down attentional mechanisms that modulate
sensory representations based on task demands. Consistent
with this possibility, experiments using single-unit recordings in
behaving monkeys (e.g., Moran and Desimone, 1985; Haenny
et al., 1988; Motter, 1993; Treue and Maunsell, 1996; McAdams
andMaunsell, 1999; Seidemann and Newsome, 1999; Treue and
Martınez Trujillo, 1999; Reynolds et al., 2000; Bichot et al., 2005)
and fMRI in human subjects (e.g., Kastner et al., 1999; Ress
et al., 2000; Buracas and Boynton, 2007) revealed that task
demands can significantly modulate neural responses in visual
cortical areas. However, the purpose of these modulations is still
under debate.
It is commonly assumed that sensory systems have limited
representational resources and that the goal of attention is to
allocate these resources based on task demands (e.g., Broad-
bent, 1958). Selection, however, is necessary even under condi-
tions in which the sensory system’s ability to represent multiple
stimuli is not limited, because the task may require the subject
to use only a subset of the available stimuli and ignore others.
Therefore, another possible goal of attention is to gate task-
irrelevant stimuli in order to limit their access to circuits that
control behavior (e.g., Allport, 1993). These two possible goals
of attentions are distinct and, as discussed below, have different
predictions regarding the expected physiological effects of
attention. However, these two forms of attention are not mutually
exclusive and could both operate in the same cortical area.
To illustrate the differences between these two attentional
mechanisms, consider the following toy example (Figure 1).
You are presented with four coins. On half of the trials all four
coins are tails, and on the other half three are tails and one is
a head. Your task is to report whether a head is present, and if
so, where it is located. What makes the task difficult is that
instead of getting direct access to the coins, you observe a
‘‘noisy sensory representation’’ of each coin; consequently,
there is a probability that the observed coin face is different
from its true value. The fidelity of the sensory representation is
represented by the ‘‘probability of spontaneous flip’’ (pf, indi-
cated by the red bar near each coin).
Consider the following two versions of the task. In the focal-
attention version, you are cued in advance as to the only possible
coin location where the headmay have occurred (Figures 1B and
1D; cue indicated by blue square). In the distributed-attention
version, all four coin locations are cued, and therefore, the head
couldhaveoccurredat anyof these locations (Figures1Aand1C).
Now compare two scenarios, one in which your sensory repre-
sentation is limited (Figures 1C and 1D), and one in which it is
unlimited (Figures 1A and 1B). When the sensory representation
has limited resources, attention allocates these resources ac-
cording to the task, and the fidelity is high under focal attention
(pf = 0.1) and lower under distributed attention (pf = 0.15).
When the sensory representation is not limited, the fidelity under
both focal and distributed attention is the same (pf = 0.1).
Consider first the no-resource-limit case. Intuitively, even in
this case, the task is more difficult under distributed attention
than under focal attention. To see this, consider an example in
which the bottom right coin is a head that has not flipped.
Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc. 557
Figure 1. Toy Example of a Behavioral Task Illustrating the Signatures of Two Forms of Attentional Modulations: Allocation of Limited
Representational Resources and Gating of Task-Irrelevant Information
The goal of the task is to report whether Washington’s head is present and, if so, where. In half of the trials there is no head and in the other half there is one head.
Each coin has a spontaneous flip probability (red bar) that represents the probability that the observed coin face is different from the true coin face.
(A and C) Distributed attention.
(B and D) Focal attention.
The cue (blue square) indicates where the head could occur.
(A and B) No resource limit.
(C and D) Resource limit.
The number in each panel indicates the expected accuracy of an observer that uses the optimal strategy to perform this task. The optimal strategy for flip
probabilities% 0.2 is as follows: report ‘‘no head’’ if all observed coins are tails; otherwise, report ‘‘head’’ and select any of the observed head locations. See text
for additional details.
Neuron
Attentional Modulations in Primate V1
However, one of the other three coins has flipped and it is also
a head. In the distributed-attention case, you have to guess
which one of the two observed heads (if any) was originally
a head. On the other hand, in the focal-attention case, you
know that the only location where the head could have occurred
is the bottom right and, therefore, have a higher chance of re-
porting correctly that this location contains the head. Hence,
despite the equal fidelity of the representation in focal and
distributed attention, behavioral accuracy under distributed
attention will be lower. The numbers in each panel show the ex-
pected accuracy of an observer that uses an optimal strategy to
perform this task. The accuracy of this observer is reduced by
19% in the distributed attention task versus the focal attention
task. This example illustrates that a difference in accuracy
between focal and distributed attention is not, by itself, evidence
in favor of limited representational resources.
Now consider the case in which the sensory representation
has limited resources and the probability of spontaneous flip is
higher under distributed attention than under focal attention. It
is easy to see that this drop in fidelity must lead to an additional
drop in accuracy under distributed attention. In our toy example,
the accuracy of the optimal observer drops from 90% in the
focal-attention case to 60% in the distributed-attention case,
which is lower than the accuracy in the distributed-attention
case under the unlimited resource scenario.
558 Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc.
Our toy example shows that the signature of attention as a
mechanism for allocation of limited resources is enhanced neural
sensitivity to attended stimuli under focal attention versus
distributed attention (by increased signal and/or by reduced
noise).
As discussed above, another possible goal of attention is to
limit the behavioral impact of task-irrelevant stimuli by selectively
blocking irrelevant signals. In our toy example, we represent
this gating mechanism as the color of the coin. Coins at cued
locations are gold colored, while coins at uncued (and therefore
irrelevant) locations are silver colored. The color of the coin is
analogous to a neural bias that highlights task-relevant locations
and allows subsequent processing stages to selectively gate
task-irrelevant signals.
Our toy example shows that the signature of attention as
a mechanism for gating task-irrelevant information is a response
bias in favor of relevant (attended) versus irrelevant (ignored)
locations. This bias signal may be associated with enhanced
neural sensitivity at attended locations, but as long as the
enhanced sensitivity is the same under focal and distributed
attention, it would be inconsistent with a limited resource
mechanism.
To study these two forms of attention experimentally, we used
VSDI to measure V1 responses while monkeys performed a
detection task analogous to our toy example above. This task
Time
500ms 500ms 500~1200ms 300ms 300ms 300ms
Saccade to Target
Imaging with VSD
Fixation
Single-cue
Multiple-cues
Fixation Dimming
Mask Only
Mask+Target
A
EDCB
Figure 2. Task and Behavioral Performance under
Different Attentional Conditions
(A) Sequence of events within a single trial. After initial
fixation, either a single cue appeared at one of the four
possible locations (single cue) or four cues appeared at all
four locations (multiple cues). After an additional variable-
duration fixation period, dimming of the fixation point
provided a temporal cue that 300 ms later the visual
stimulus will appear. In half of the trials the visual stimulus
was composed of four 10% contrast masks of one orien-
tation (‘‘mask only’’). In the remaining trials, a low-contrast
(3.5%–4.5%) orthogonal target was added to one of the
masks (‘‘mask-plus-target’’). In target-present trials the
monkey was required to shift gaze to the target location as
soon as it detected the target. In target-absent trials the
monkey was required to maintain fixation.
(B) Average percent correct (PC) of the twomonkeys under
focal attention (single cue, red) and distributed attention
(multiple cues, blue). The percent correct in single-cue
trials was computed only for trials without a distracter.
(C) Average across experiments of median reaction times
in hit trials in the two attentional conditions.
(D) Same as (C) but in false-alarm trials.
(E) Average probability that the monkey would shift gaze to
the distracter location (PD) (the location opposite to the
cue) in trials with (magenta) and without (cyan) a distracter.
Here and in all other figures: N, the number of experiments
for each monkey; p, the p value of a paired t test across
experiments; error bars represent SEM.
Neuron
Attentional Modulations in Primate V1
(described below) allowed us to measure simultaneously the
behavioral and neurophysiological effects of both forms of atten-
tion. In single isolated neurons, VSDI signals are linearly related
to membrane potential across the entire physiological dynamic
range (e.g., Salzberg et al., 1973). In the primate cortex, recent
results suggest that the VSDI signal at any given location is
proportional to the summed membrane potential of a population
of neurons, integrated over a Gaussian-shaped area with stan-
dard deviation (SD) of �230 mm (Chen et al., 2012). Therefore,
attentional modulations measured with VSDI are likely to reflect
the inputs that V1 neurons receive from top-down circuits rather
than the attentional modulations of the spiking output of V1
neurons. Recent VSDI studies in behaving primates demonstrate
that VSDI is highly sensitive and can provide reliable information
about visual stimuli even below the subject’s behavioral detec-
tion threshold (Chen et al., 2006, 2008a). Therefore, VSDI al-
lowed us to studywhether, and how, these two forms of attention
modulate neural population responses over a large V1 region
with high sensitivity and at high spatial and temporal resolution.
RESULTS
We trained twomonkeys to detect a low-contrast oriented target
(analogous to ‘‘head’’ in our toy example) that appeared on half of
the trials on top of one of four evenly spaced higher-contrast
masks of orthogonal orientation (Figure 2A). We first used VSDI
to determine the layout of the retinotopic map in the imaged
area (Yang et al., 2007). We then positioned the stimuli so that
one of the four stimulus locations fell inside the receptive fields
of V1 neurons at the center of the imaged area. To report detec-
tion, themonkey had to shift gaze to target location within a short
time window following target onset. As in our toy example, at the
beginning of each trial, a cue indicated to the monkey whether to
attend to one of the four possible locations (single-cue, ‘‘focal-
attention’’) or to all four locations (multiple cues, ‘‘distributed-
attention’’). To ensure that themonkeywas ignoring the irrelevant
locations in focal attention trials, in half of those trials, a distracter
identical to the target could appear at the location opposite to the
cue. Themonkey had to ignore this distracter. Finally, blank trials
with no cue and no visual stimulus, and control trials with cue(s)
but no visual stimulus, were randomly intermixed with all other
trial types (see Experimental Procedures for additional details).
This task allowed us to measure V1 responses to the same
physical visual stimuli under three attentional states: when only
the location corresponding to the imaged area was cued (focal
attention, ‘‘attend-in’’), when it was one of four cued locations
(‘‘attend-distributed’’), and when another location was cued
and the imaged location had to be ignored (focal attention,
‘‘attend-out’’) (Figure 3A, top row). As illustrated in our toy
example (Figure 1), this task allowed us to examine the two
possible forms of attentional effects in V1. By comparing
responses in ‘‘attend-in’’ and ‘‘attend-distributed’’ states, we
tested the hypothesis that attention allocates limited representa-
tional resources in V1. By comparing responses in ‘‘attend-in’’
and ‘‘attend-out’’ states, we tested the hypothesis that attention
in V1 helps to spatially gate task-irrelevant signals.
Attentional Modulations of Behavioral PerformanceAs expected, the monkeys’ behavioral performance was signifi-
cantly better in terms of accuracy (Figure 2B) and reaction times
Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc. 559
σmajor
σminor
AB
C D E F
Figure 3. Spatial Patterns of V1 Population Responses under Three Attentional States in Monkey 1
(A) Top row shows a schematic diagram of the three attentional states. The black circle represents the receptive field (RF) of the neurons at the center of the
imaged area, while thewhite circles represent the cue. Att-in: the receptive field location is cued. Att-out: the cued location is outside of receptive field. Att-dstr: all
four locations are cued. The bottom two rows show the average response maps in mask (middle row) and mask-plus-target (bottom row) trials under the
attentional state indicated by the diagram in the top row. At each location the responses are averaged over the first 200ms after stimulus onset. The average in the
100 ms before stimulus onset is then subtracted from the amplitude at each location. The ellipsoidal region at the center of the imaged area shows the area
directly activated by themask or mask-plus-target. The response is anisotropic due to the anisotropy in themap of visual space in V1. The response is dominated
by the higher-contrast mask and is therefore very similar in mask and mask-plus-target trials. Note that in attend-out state the overall activity level in the area
surrounding the active region is lower than in attend-in and attend-distributed states.
(B) Two-dimensional (2D) fit to the spatial patterns of V1 population response. The responses were fitted with a 2D Gaussian plus a uniform baseline. The space
constants along the major and minor axis of the Gaussian are indicated. Top: 2D Gaussian fit to response to mask in attend-in state. Bottom: one-dimensional
horizontal slice through the two-dimensional fit, indicated by the black line in the top panel. The baseline, amplitude, and space constant (s) are indicated. For
comparison, the spatial response profile in a 1-mm-wide strip along this slice is indicated by a dashed line. The shaded area indicates ± 1 SEM across
experiments. Right: same as bottom but along a vertical slice.
(C–F) Mean and SEM of the fitted parameters across 11 experiments in monkey 1. The color represents the attentional state as indicated in the top row of (A).
Asterisks indicate significant differences (p < 0.05 based on paired t test across experiments). The only significant differences are between the magnitude of the
baseline in attend-out and the other two attentional states.
Neuron
Attentional Modulations in Primate V1
(Figures 2C and 2D) under focal attention than under distributed
attention (see also Figure S1 available online). If these differ-
ences in behavioral performance are mediated, at least partially,
by top-downmodulations in V1, and if target representation in V1
is a limited resource, wewould expect the VSDI-measured target
sensitivity to be higher under focal attention than under distrib-
uted attention.
In addition, inmost trials themonkeys successfully ignored the
distracter. However, the probability that themonkeys would shift
gaze to the location opposite to the cue in focal attention trials
was significantly higher in the presence of the distracter (Fig-
ure 2E), indicating that the monkeys were incapable of fully
ignoring the distracter. If selective gating of signals from ignored
locations is mediated, at least partially, by top-down modula-
tions in V1, we would expect the VSDI-measured V1 responses
to be biased in favor of attended versus ignored locations. Our
560 Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc.
next stepwas therefore to examine V1 responses under the three
attentional states.
Spatial Distribution of Attentional Modulations in V1We used VSDI to measure V1 population responses while the
monkeys performed the detection task. Figure 3A shows the
average spatial patterns of V1 population responses for each
of the two visual stimuli under the three attentional states in
monkey 1 (after subtracting the average responses in blank
trials). Consistent with our previous results (Chen et al., 2006,
2008a; Palmer et al., 2012), the visual stimuli activated a localized
ellipsoidal region that subtended multiple mm2 in V1. Because
target contrast (3.5%–4.5%) was lower than mask contrast
(10%), the response was dominated by the mask, consistent
with single-unit masking results (e.g., Busse et al., 2009)
and with the detrimental effect of the mask on the monkeys’
C
B
D
A
Figure 4. Summary of Attentional Effects in the Two MonkeysAverage and SEM of the normalized amplitudes of the Gaussian and baseline
components under the three attentional states. Because attentional modula-
tions were stimulus independent (see text), results from target-absent and
target-present trials were combined before fitting the spatial response maps.
The amplitudes of the Gaussian and baseline components were normalized by
the average amplitude of the Gaussian.
(A) Normalized amplitude of 2D Gaussian in monkey 1 is not affected by
attentional state.
(B) Normalized amplitude of baseline is significantly higher in attend-in and
attend-distributed than in attend-out in monkey 1.
(C and D) Same as (A) and (B) but for monkey 2. The overall quality of the VSDI
signals in monkey 2 was lower than inmonkey 1, which could have contributed
to the lower attentional effect in this monkey.
Neuron
Attentional Modulations in Primate V1
detection threshold. However, peak responses in target-present
trials were significantly higher than in target-absent trials (one-
tailed paired t test, p < 0.01 for both monkeys; combined across
all three attentional states). The spatial profile of the response
was similar in the three attentional states. However, the activity
over the entire imaged area was elevated in attend-in and
attend-distributed trials (Figure 3A, note the lighter colors in
attend-in and attend-distributed conditions). To quantitatively
analyze the attentional effects, we fitted the responses with
a two-dimensional (2D) Gaussian plus a spatially uniform base-
line (Figure 3B). These two spatial components provided
a good fit to the observed responses (r2 > 0.9 for all stimulus/
cue combinations in both monkeys).
The attentional state significantly modulated the spatially
uniform baseline component (Figure 3F) but had no significant
effect on the amplitude or the shape of the Gaussian component
(Figures 3C–3E). The baseline was elevated in attend-in and
attend-distributed conditions relative to attend-out condition,
which was indistinguishable from the baseline in blank condition
(trials with no cue and no visual stimulus). We obtained similar
results in monkey 2 (Figure S2).
To test whether the attentional state affected the target-
evoked response (difference between target-present and
target-absent response), we performed paired t tests on the
amplitudes of the target evoked response in the three attentional
states. None of the test showed a significant effect (p > 0.13).
We therefore combined the responses across the two visual
stimuli.
A summary of the results from the two monkeys (combined
across the two visual stimuli; Figure 4) reveals robust attentional
modulations of the baseline response which are several-fold
larger than the response evoked by the target. In both monkeys,
baseline activity was indistinguishable between attend-in and
attend-distributed (paired t test, monkey 1, p = 0.95; monkey
2, p = 0.57), but significantly lower in attend-out relative to
the other two attentional states (paired t test, monkey 1, p <
0.0052 for both tests; monkey 2, p < 0.015 for both tests); base-
line activity in attend-out and blank conditions was indistinguish-
able (paired t test, monkey 1, p = 0.29; monkey 2, p = 0.39), and
this was true at the location where the distracter could appear
(opposite to the cue) as well as the two other unattended
locations (Figure S3). The Gaussian amplitude was independent
of attentional state (paired t test, p > 0.094 for all tests). To
quantify this effect, we normalized all responses by the average
amplitude of the Gaussian component. The average normalized
amplitude of the attentional baseline elevation was 23% in
monkey 1 and 12% in monkey 2, while the average normalized
target-evoked response (additional response evoked by the
target in the presence of the mask) was only 4.7% in monkey 1
and 7.1% in monkey 2.
Attentional Modulations of Response VariabilityWhile responses under focal and distributed attention are the
same on average, it is still possible that attention enhances
neural sensitivity under focal attention by modulating neural
noise (Cohen and Maunsell, 2009; Mitchell et al., 2009). To
examine this possibility, we computed the SD of the response
amplitude across trials and the spatial correlations of the
response variability (Chen et al., 2006). Neither the SDs nor the
spatial correlations varied significantly with attentional state
(Figure 5), suggesting that in our task, attention does not lead
to significant changes in these noise properties at the population
level in V1.
Time Course of Attentional Modulations Relatedto Spatial GatingTo determine when the attentional modulations are initiated and
how they evolve over time, we compared the dynamics of the
baseline component in the three attentional states (see Experi-
mental Procedures). Our results show that the attentional
modulations start to build up about 100 ms before the stim-
ulus-evoked response (compare Figures 6A and 6D with Figures
6C and 6F) and about 200 ms after fixation point dimming.
Similar results were obtained in control trials in which no visual
stimulus was presented after the cue(s) (Figures 6B and 6E).
These modulations, therefore, are stimulus independent, are
preparatory in nature, and are timed to occur shortly before
stimulus onset.
Possible Confounding EffectsOur results suggest that top-down mechanisms can modulate
neural population responses in V1 based on stimulus relevance,
Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc. 561
B
D
A
C
Figure 5. Attentional State Does Not Change Properties of theVariability at the Level of Neural Populations in V1
(A) Standard deviation across trials of VSDI responses is not affected by
attentional state in monkey 1. VSDI response was averaged for each trial
during the 200 ms after stimulus onset in a 1 3 1 mm2 region centered at the
peak of the fitted 2D Gaussian. The standard deviation was normalized by the
fitted average peak response.
(B) Spatial correlations across trials between two locations as a function of
their spatial separation are not affected by attentional state in monkey 1. One
of the two locations was in the central 1 3 1 mm2 region; the other was
anywhere in the region-of-interest.
(C and D) Same as (A) and (B) but for monkey 2.
In (A) and (C), error bars represent SEM. In (B) and (D), shaded areas represent
SEM.
Neuron
Attentional Modulations in Primate V1
but before we can conclude that the elevated baseline reflects
a genuine top-down attentional signal, we have to rule out
several confounding effects.
First, it is possible that the observed baseline modulations are
due to direct visual response to the cue. This seems unlikely
because the interval between cue offset and stimulus onset
(800–1500 ms) was long and variable and because cue diameter
(3�) was much larger than the receptive fields of the neurons in
the imaged area (<1�). Nevertheless, to rule out this possibility,
we examined the time course of the baseline modulations
time-locked to cue offset. These time courses were indistin-
guishable between the three attentional states (Figure S4),
demonstrating that the attentional modulations (which are
time-locked to stimulus onset) are not due to direct visual
responses to the cue.
Second, systematic differences in fixational eye movements
between the attentional states could have contributed to the
observed variations in V1 responses. This possibility seems
unlikely given that the attentional modulations start before the
stimulus-evoked responses (Figure 6). Nevertheless, to rule out
this possibility, we compared several eye position statistics in
the three attentional states. Our results reveal no significant
differences in these statistics depending on the attentional
condition (Figure S5), providing further support for the top-
down nature of the observed modulations.
562 Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc.
DISCUSSION
What is the purpose of the observed attentional modulations?
One possible goal of attention is to allocate limited representa-
tional resources based on task demands (e.g., Broadbent,
1958). Another possible goal of attention is to limit the access
of task-irrelevant stimuli to circuits that control behavior (e.g., All-
port, 1993). If the representation of multiple visual targets in V1
was a limited resource that could be controlled by attention,
we would have expected V1 target sensitivity at attended loca-
tions to be higher under focal attention than under distributed
attention (Figures 1C and 1D). Our finding that V1 population
responses at attended locations are indistinguishable under
focal and distributed attention suggests that in our task, and at
the level of neural populations, target sensitivity in V1 may not
be a limited resource that can be enhanced by focal attention.
We find that behavioral performance is improved under focal
attention relative to distributed attention (Figures 2B–2D). As
illustrated by our toy example (Figure 1), behavioral improvement
under focal attention is expected even if V1 target sensitivity is
not limited and is identical in focal and distributed attention. A
simple analysis based on signal detection theory shows that
the observed behavioral improvement in accuracy under focal
attention is consistent with no changes in neural sensitivity under
focal and distributed attention (Suppl. Figure 6; see also Eckstein
et al., 2000; Palmer et al., 2000; Pestilli et al., 2011). This analysis,
therefore, provides further support to the hypothesis that in our
task, target sensitivity is not a limited resource that can be
enhanced by focal attention.
While our physiological and behavioral results appear to be
inconsistent with attention as a mechanism for allocating limited
resources in V1, we cannot rule out the possibility that such
a mechanism operates in V1 in other tasks. Similarly, we cannot
rule out the possibility that such a mechanism operates in V1 but
affects a small subset of the neurons, or enhances the responses
of some neurons while suppressing the responses of others,
leading to a population response that is unaltered. However,
what makes our negative result compelling is the contrast with
the robust gating-related attentional modulations obtained in
the same experiments. Therefore, we conclude that in our
task, and at the level of neural populations in V1, attentional
effects related to allocation of limited resources either are absent
or are much weaker than effects related to gating of irrelevant
stimuli.
Previous studies in behaving monkeys have demonstrated
that increased task difficulty can lead to enhanced neural
responses in area V4 (Spitzer et al., 1988; Boudreau et al.,
2006). Because our distributed attention trials weremore difficult
than the focal attention trials, increased vigilance in distributed
attention trials could have led to enhanced responses in V1
that masked a reduction in response in distributed versus focal
attention trials. However, it seems unlikely that such opposing
effects precisely canceled each other to produce indistinguish-
able responses in focal and distributed attention. In addition, it
is not clear whether similar effects of vigilance are present in
V1. Finally, in our task target contrast was near detection
threshold even in focal attention trials and the two trial types
were randomly intermixed. Therefore, it seems less likely that
B
E
A
D
C
F
Figure 6. Time Course of Attentional Modulations
and Stimulus-Evoked Responses
Time courses of the amplitudes of the Gaussian and
baseline components were normalized by the peak
amplitude of the Gaussian in the mask-plus-target
condition.
(A) Time courses of the normalized baseline component in
attend-in and attend-distributed trials in monkey 1. The
time course of the normalized baseline component in
attend-out trials was subtracted from the time courses in
the other two attentional states.
(B) Time course of normalized baseline component in
control trials with cue but no visual stimulus. In the
experiments in monkey 1 we only included control trials
with multiple cues. Therefore, we show the time course in
multiple-cue control trials after subtracting the time course
in blank trials.
(C) Time courses of the normalized Gaussian component
in mask (cyan) and mask-plus-target (magenta) in
monkey 1.
(D–F) Same as (A) and (B) but for monkey 2. In the exper-
iments in monkey 2 we included control trials for all three
attentional states, so in (E) we show the time course in
attend-in and attend-distributed control trials after sub-
tracting the time course in attend-out control trials. The
shaded areas represent ± 1 SEM.
Neuron
Attentional Modulations in Primate V1
differences in attentional load between focal and distributed
attention affected our results.
The observed differences between V1 responses at attended
and ignored locations are consistent with the hypothesis that an
important goal of attention in V1 is to limit the behavioral effect of
task-irrelevant visual stimuli. The elevated baseline at attended
locations could contribute to this selective spatial gating by
biasing competition in subsequent processing stages in favor
of task-relevant stimuli (Desimone and Duncan, 1995). If this
top-down signal itself was a limited resource, we would have ex-
pected to see differences between attentional modulations in
focal and distributed attention. However, the baseline elevation
is indistinguishable between focal and distributed attention,
demonstrating that the top-down mechanism mediating this
effect is not a limited resource (at least when the number of
possible locations is four).
Additive versus Nonadditive Attentional EffectsThe observed attentional modulations are additive and stimulus
independent. Because VSDI signals measure changes in
membrane potentials, this result implies that in our task, the
top-down input that V1 neurons receive is stimulus independent.
This is consistent with our findings that the attentional effect
starts before stimulus onset and can occur even when the visual
stimulus is absent. However, it is important to note that, due to
the nonlinear relationship between membrane potential and
spikes (reviewed in Priebe and Ferster, 2008), the additive
effects observed with VSDI could lead to nonadditive modula-
tions at the level of V1 spiking activity. Specifically, due to the
Vm-to-spikes nonlinearity, the baseline elevation of membrane
potential could have a reduced effect at the level of spiking
activity when the response is weak (e.g., before stimulus onset
or at locations far from stimulus center), and an enhanced effect
when the response is strong (e.g., near the peak of the stimulus
evoked response). The possible effects of such nonlinearity on
the predicted attentional modulations at the level of spiking
activity are illustrated in Figure 7. This nonlinearity could lead
to a small increase in the firing rates of V1 neurons in the absence
of visual stimulation. Such an effect could be difficult to detect
using single unit electrophysiology but may be more prominent
in population responses. Consistent with this possibility, a recent
study of attentional modulations at the level of single neurons in
V1 found no baseline modulations in the absence of the stimulus
or at very low stimulus contrasts (Thiele et al., 2009), while fMRI
results in V1 show clear baseline modulations even when the
stimulus is absent (e.g., Buracas and Boynton, 2007; Murray,
2008; Pestilli et al., 2011).
Temporal Dynamics of Attentional ModulationsThe attentional signals are initiated after fixation point dimming
and shortly before the visually evoked responses (Figure 6).
This result implies that top-down modulations can be anticipa-
tory in nature, consistent with previous studies (e.g., Ghose
and Maunsell, 2002), and do not require extrastriate cortex to
first process the visual stimulus. In fact, the attentional signal
can be present even in the absence of the visual stimulus
(Figures 6B and 6E), consistent with some electrophysiological
(Luck et al., 1997; Reynolds et al., 2000; Williford and Maunsell,
2006) and fMRI (Kastner et al., 1999; Ress et al., 2000) results.
Spatial Scale of Attentional ModulationsThe attentional modulations in V1 operate at a larger spatial scale
than the stimulus-evoked response. Because the baseline eleva-
tion extends beyond our imaged area, we cannot determine the
Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc. 563
BA
C
Figure 7. A Schematic Illustration of Possible
Effects of Attention at the Level of Spiking Activity
Given our VSDI Results
The figure shows the spatial profile (A) and temporal
dynamics (B) of attentional effects measured with VSDI
(top panels) and the predicted patterns of modulations at
the spiking level (lower panels) assuming a nonlinear
relation between VSDI and spiking activity similar to the
one observed between membrane potential and spiking
activity in single neurons (C). The attentional effect
(difference between attend-in and attend-out) is indicated
by the dashed black line. An additive effect in space and in
time at the level of membrane potential can lead to a
nonadditive (stimulus dependent) effect at the level of
spiking activity.
Neuron
Attentional Modulations in Primate V1
exact spatial extent of this top-down signal. However, since we
observed no baseline elevation in attend-out trials, the top-down
signal must have a limited spatial extent, at least in focal atten-
tion trials. In distributed attention trials, top-down signals could
activate simultaneously four broad but separate V1 regions
peaked at the representation of the possible stimulus locations,
consistent with a recent finding in humans (Muller et al., 2003).
Alternatively, a larger contiguous region could be activated,
such as the V1 region corresponding to a ring at target eccen-
tricity. The spatial extent and shape of the top-down attentional
signal could be addressed in future VSDI experiments by
systematically shifting the position of the stimuli relative to the
position of the receptive fields in the imaged area.
Relation to Prior Studies of Attentional Modulationsin V1 using Electrophysiology and fMRIOur results are relevant to the ongoing controversy regarding
the magnitude and nature of attentional effects in V1. Single-
unit studies can reveal consistent attentional modulations in
macaque V1 (e.g., Motter, 1993; Luck et al., 1997; Roelfsema
et al., 1998; Ito and Gilbert, 1999; McAdams and Maunsell,
1999; Marcus and Van Essen, 2002; Roberts et al., 2007; Thiele
et al., 2009). However, these effects tend to be weak (but see
Chen et al., 2008b) and delayed and are typically observed
only in the presence of visual stimulation. In contrast, brain
imaging studies using fMRI in human subjects reveal pro-
nounced attentional modulations in V1 (e.g., Kastner et al.,
1999; Ress et al., 2000; Buracas and Boynton, 2007; Pestilli
et al., 2011) that occur even in the absence of visual stimulation.
One possible explanation for this discrepancy is that fMRI BOLD
signals amplify attentional effects by pooling weak modulations
over large populations of neurons. A second possibility is that
attention operates differently in humans and in macaque
monkeys. Finally, it is possible that some attention related
BOLD signals reflect direct modulations of hemodynamic
564 Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc.
responses that are independent of local
neural activity (e.g., Sirotin and Das, 2009). The
robust attentional modulations of V1 population
responses reported here are consistent with
the first possibility and provide support to the
general hypothesis that responses that might
be weak and heterogeneous at the level of
single neurons could have a substantial impact at the level of
neural populations (for review, see Seidemann et al., 2009).
Summary and conclusionsIn summary, our results show that despite significant differences
in behavioral performance between focal and distributed atten-
tion, V1 responses at attended locations are indistinguishable
under these two attentional states. These results suggest that
in our task, the representation of visual targets in V1 is not
a limited resource that can be enhanced under focal attention.
However, our results reveal robust elevation of V1 activity based
on stimulus relevance. Responses are elevated over a large
region centered on the attended locations and are maintained
at a default low state at ignored locations. This additive elevation,
which is initiated shortly before stimulus onset, is likely to
contribute to the ability of subsequent processing stages to
selectively gate task-irrelevant sensory signals.
EXPERIMENTAL PROCEDURES
Behavioral Task and Visual Stimulus
Two monkeys were trained to detect a small oriented target that appeared at
one of four fixed locations on top of a background of four orthogonal masks
(Figure 2A). Each trial began when a small bright fixation spot (0.1� 3 0.1�)appeared at the center of the screen. The monkey was required to fixate the
spot for 500ms and thenwas cued for another 500ms to pay attention to either
one of the four locations (single cue) or to all four locations (multiple cues). The
cue was a 0.02� thick bright circular ring with diameter of 3� centered on the
possible target location. After an additional delay of 500–1200 ms after cue
offset, the fixation spot dimmed, indicating that the visual stimuli (mask or
mask-plus-target) would appear 300 ms later. The target was a Gabor patch
with s = 1/6�, spatial frequency = 2.76 cyc/�, phase = 90�, eccentricity =
2.5�–3�, and orientation = 0� for onemonkey and 90� for the other. The contrastof the target was near the monkey’s detection threshold (3.5%–4.5%). The
masks differed from the target only in their orientation, which was orthogonal
to that of the target, and their contrast which was 10%. The target was pre-
sented for 300 ms in half of trials, while the four background masks were pre-
sented at the same time in all trials. In order to receive a reward, the monkey
Neuron
Attentional Modulations in Primate V1
was required to make a saccadic eye movement to the target location in
target-present trials or to hold fixation during a period of 600 ms after stimulus
onset in target-absent trials. In addition, in half of the single-cue trials a dis-
tracter (identical to target) was introduced to the location opposite the cue.
The monkeys were required to ignore the distracter. We also included control
trials in which the cue was presented but no visual stimulus appeared and
blank trials in which no cue and no visual stimulus were presented. The
monkey was required to maintain fixation in these trials.
Visual stimuli were presented on a gamma-corrected high-end 21 inch color
display (Sony Trinitron GDM-F520) at a fixedmean luminance of 30 cd/m2. The
display subtended 20.5� 3 15.4� at a viewing distance of 108 cm and had
a pixel resolution of 1024 3 768, 30-bit color depth, and a refresh rate of
100 Hz. Behavioral measurements and data acquisition were controlled by
a PC running a software package for neurophysiological recordings from alert
animals (Reflective Computing). Eye movements were measured using an
infrared eye-tracking device (Dr. Bouis Inc.).
Optical Imaging with VSD
Our general experimental procedures for VSDI in behaving monkeys have
been described in detail elsewhere (Chen et al., 2006, 2008a). All procedures
were approved by the University of Texas Institutional Animal Care and Use
Committee and conformed to National Institutes of Health standards. Briefly,
we used oxonol voltage-sensitive dyes (RH 1838 or RH 1691) (Shoham
et al., 1999) to stain the cortical surface and an Imager 3001 system (Optical
Imaging) to image brain activity. Imaging data were collected using resolution
of 5043 504 pixels at 110 Hz and were further binned to 633 63 pixels tomini-
mize the contribution of shot noise, with each final pixel corresponding to
0.25 3 0.25 mm2 cortical area.
Analysis of Imaging Data
We completed 22 VSDI experiments from two hemispheres of monkey 1 and
25VSDI experiments fromonehemisphere ofmonkey2.We selected for further
analysis experiments in which the average amplitude of the peak spatial
response exceeded twice the response standard deviation across trials
(11 and 12 experiments from monkeys 1 and 2, respectively). The remaining
experimentshad lowsignal-to-noise ratios, usually attributable topoor staining.
VSDI analysis followed six steps for each experiment: (1) we removed trials
with aberrant VSDI responses (usually < 1% of total trials). In each trial, we
divided each frame into four quadrants, and average the fluorescence in
each quadrant. A trial was removed if the average fluorescence at any of the
quadrants and frames was out of ± 5 standard deviations across all trials. (2)
We normalized the response at each pixel by the average fluorescence across
all trials and frames. (3)We subtracted the average response in blank trials from
all individual trials. (4) We cropped all frames to an area of 103 8 mm2 with the
responsepeak near the center of the croppedarea. (5)Weestimated the spatial
response maps. In each trial and at each location, we averaged the response
within a 200 ms interval after stimulus onset, and then subtracted the average
response within a 100 ms interval before stimulus onset to obtain a spatial
response map. For each attentional state, we averaged the spatial response
maps across all corresponding trials irrespective of behavioral outcome and
then fitted the average map with a 2D Gaussian function R(x,y) = a*G(x,y) + b,
where G(x,y) was a Gaussian function and a and b were the amplitudes of the
Gaussian and baseline. (6) We estimated the time courses of the Gaussian
and the baseline. Because no significant difference was found in the Gaussian
component across the three attentional states, we defined the spatiotemporal
responses as R(x,y,t) = a(t)*G(x,y) + b(t), where a(t) and b(t) were the time
courses of Gaussian and baseline. We first averaged the spatial response
maps in step 5 across the three attentional states and fitted the average with
a2DGaussian function toobtainG(x,y). Then, for eachattentional state,wepro-
jected G(x,y) and the baseline to each frame to calculate a(t) and b(t). All data
analysis was performed in MATLAB (Mathworks).
SUPPLEMENTAL INFORMATION
Supplemental Information includes six figures and can be found with this
article online at doi:10.1016/j.neuron.2012.03.033.
ACKNOWLEDGMENTS
We than W. Geisler for the suggesting the toy example in Figure 1 and for
helpful discussions. We thank D. Ress, D. Heeger, J. Maunsell, and members
of the Seidemann lab for helpful comments and suggestions and T. Cakic for
technical support. This work was supported by National Eye Institute Grants
EY-016454 and EY-016752. Author contributions: Y.C. and E.S. designed
the experiments, Y.C. carried out the experiments and analyzed the data,
and Y.C. and E.S. wrote the paper.
Accepted: March 15, 2012
Published: May 9, 2012
REFERENCES
Allport, D.A. (1993). Attention and control: have we been asking the wrong
questions? A critical review of twenty-five years. In Attention and
Performance XIV, D.E. Meyer and S. Kornblum, eds. (Cambridge, MA: MIT
Press), pp. 183–218.
Bichot, N.P., Rossi, A.F., and Desimone, R. (2005). Parallel and serial neural
mechanisms for visual search in macaque area V4. Science 308, 529–534.
Boudreau, C.E., Williford, T.H., and Maunsell, J.H. (2006). Effects of task
difficulty and target likelihood in area V4 of macaque monkeys.
J. Neurophysiol. 96, 2377–2387.
Broadbent, D.E. (1958). Perception and Communication (London: Pergamon
Press).
Buracas, G.T., and Boynton, G.M. (2007). The effect of spatial attention on
contrast response functions in human visual cortex. J. Neurosci. 27, 93–97.
Busse, L., Wade, A.R., and Carandini, M. (2009). Representation of concurrent
stimuli by population activity in visual cortex. Neuron 64, 931–942.
Chen, Y., Geisler, W.S., and Seidemann, E. (2006). Optimal decoding of corre-
lated neural population responses in the primate visual cortex. Nat. Neurosci.
9, 1412–1420.
Chen, Y., Geisler, W.S., and Seidemann, E. (2008a). Optimal temporal decod-
ing of neural population responses in a reaction-time visual detection task.
J. Neurophysiol. 99, 1366–1379.
Chen, Y., Martinez-Conde, S., Macknik, S.L., Bereshpolova, Y., Swadlow,
H.A., and Alonso, J.M. (2008b). Task difficultymodulates the activity of specific
neuronal populations in primary visual cortex. Nat. Neurosci. 11, 974–982.
Chen, Y., Palmer, C.R., and Seidemann, E. (2012). The relationship between
voltage-sensitive dye imaging signals and spiking activity of neural popula-
tions in primate V1. J. Neurophysiol., in press. Published online March 14,
2012. 10.1152/jn.00977.2011.
Cohen, M.R., and Maunsell, J.H.R. (2009). Attention improves performance
primarily by reducing interneuronal correlations. Nat. Neurosci. 12, 1594–
1600.
Desimone, R., and Duncan, J. (1995). Neural mechanisms of selective visual
attention. Annu. Rev. Neurosci. 18, 193–222.
Duncan, J. (1980). The locus of interference in the perception of simultaneous
stimuli. Psychol. Rev. 87, 272–300.
Eckstein, M.P., Thomas, J.P., Palmer, J., and Shimozaki, S.S. (2000). A signal
detection model predicts the effects of set size on visual search accuracy for
feature, conjunction, triple conjunction, and disjunction displays. Percept.
Psychophys. 62, 425–451.
Ghose, G.M., and Maunsell, J.H.R. (2002). Attentional modulation in visual
cortex depends on task timing. Nature 419, 616–620.
Haenny, P.E., Maunsell, J.H.R., and Schiller, P.H. (1988). State dependent
activity in monkey visual cortex. II. Retinal and extraretinal factors in V4.
Exp. Brain Res. 69, 245–259.
Ito, M., and Gilbert, C.D. (1999). Attention modulates contextual influences in
the primary visual cortex of alert monkeys. Neuron 22, 593–604.
Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc. 565
Neuron
Attentional Modulations in Primate V1
Kastner, S., Pinsk, M.A., De Weerd, P., Desimone, R., and Ungerleider, L.G.
(1999). Increased activity in human visual cortex during directed attention in
the absence of visual stimulation. Neuron 22, 751–761.
Luck, S.J., Chelazzi, L., Hillyard, S.A., and Desimone, R. (1997). Neural mech-
anisms of spatial selective attention in areas V1, V2, and V4 of macaque visual
cortex. J. Neurophysiol. 77, 24–42.
Marcus, D.S., and Van Essen, D.C. (2002). Scene segmentation and attention
in primate cortical areas V1 and V2. J. Neurophysiol. 88, 2648–2658.
McAdams, C.J., and Maunsell, J.H.R. (1999). Effects of attention on orienta-
tion-tuning functions of single neurons in macaque cortical area V4.
J. Neurosci. 19, 431–441.
Mitchell, J.F., Sundberg, K.A., and Reynolds, J.H. (2009). Spatial attention
decorrelates intrinsic activity fluctuations in macaque area V4. Neuron 63,
879–888.
Moran, J., and Desimone, R. (1985). Selective attention gates visual process-
ing in the extrastriate cortex. Science 229, 782–784.
Motter, B.C. (1993). Focal attention produces spatially selective processing in
visual cortical areas V1, V2, and V4 in the presence of competing stimuli.
J. Neurophysiol. 70, 909–919.
Muller, M.M., Malinowski, P., Gruber, T., and Hillyard, S.A. (2003). Sustained
division of the attentional spotlight. Nature 424, 309–312.
Murray, S.O. (2008). The effects of spatial attention in early human visual
cortex are stimulus independent. J. Vis. 8, 2.1–2.11.
Palmer, J., Verghese, P., and Pavel, M. (2000). The psychophysics of visual
search. Vision Res. 40, 1227–1268.
Palmer, C.R., Chen, Y., and Seidemann, E. (2012). Uniform spatial spread of
population activity in primate parafoveal V1. J. Neurophysiol. 107, 1857–1867.
Pestilli, F., Carrasco, M., Heeger, D.J., and Gardner, J.L. (2011). Attentional
enhancement via selection and pooling of early sensory responses in human
visual cortex. Neuron 72, 832–846.
Posner, M.I. (1980). Orienting of attention. Q. J. Exp. Psychol. 32, 3–25.
Priebe, N.J., and Ferster, D. (2008). Inhibition, spike threshold, and stimulus
selectivity in primary visual cortex. Neuron 57, 482–497.
Ress, D., Backus, B.T., and Heeger, D.J. (2000). Activity in primary visual
cortex predicts performance in a visual detection task. Nat. Neurosci. 3,
940–945.
566 Neuron 74, 557–566, May 10, 2012 ª2012 Elsevier Inc.
Reynolds, J.H., Pasternak, T., and Desimone, R. (2000). Attention increases
sensitivity of V4 neurons. Neuron 26, 703–714.
Roberts, M., Delicato, L.S., Herrero, J., Gieselmann, M.A., and Thiele, A.
(2007). Attention alters spatial integration in macaque V1 in an eccentricity-
dependent manner. Nat. Neurosci. 10, 1483–1491.
Roelfsema, P.R., Lamme, V.A.F., and Spekreijse, H. (1998). Object-based
attention in the primary visual cortex of the macaque monkey. Nature 395,
376–381.
Salzberg, B.M., Davila, H.V., and Cohen, L.B. (1973). Optical recording of
impulses in individual neurones of an invertebrate central nervous system.
Nature 246, 508–509.
Seidemann, E., and Newsome, W.T. (1999). Effect of spatial attention on the
responses of area MT neurons. J. Neurophysiol. 81, 1783–1794.
Seidemann, E., Chen, Y., and Geisler, W.S. (2009). Encoding and decoding
with neural populations in the primate cortex. In The Cognitive
Neurosciences IV, M.S. Gazzaniga, ed. (Cambridge, MA: MIT Press),
pp. 419–434.
Shoham, D., Glaser, D.E., Arieli, A., Kenet, T., Wijnbergen, C., Toledo, Y.,
Hildesheim, R., and Grinvald, A. (1999). Imaging cortical dynamics at high
spatial and temporal resolution with novel blue voltage-sensitive dyes.
Neuron 24, 791–802.
Sirotin, Y.B., and Das, A. (2009). Anticipatory haemodynamic signals in
sensory cortex not predicted by local neuronal activity. Nature 457, 475–479.
Spitzer, H., Desimone, R., and Moran, J. (1988). Increased attention enhances
both behavioral and neuronal performance. Science 240, 338–340.
Thiele, A., Pooresmaeili, A., Delicato, L.S., Herrero, J.L., and Roelfsema, P.R.
(2009). Additive effects of attention and stimulus contrast in primary visual
cortex. Cereb. Cortex 19, 2970–2981.
Treue, S., andMartınez Trujillo, J.C. (1999). Feature-based attention influences
motion processing gain in macaque visual cortex. Nature 399, 575–579.
Treue, S., and Maunsell, J.H.R. (1996). Attentional modulation of visual motion
processing in cortical areas MT and MST. Nature 382, 539–541.
Williford, T., and Maunsell, J.H.R. (2006). Effects of spatial attention on
contrast response functions in macaque area V4. J. Neurophysiol. 96, 40–54.
Yang, Z., Heeger, D.J., and Seidemann, E. (2007). Rapid and precise retino-
topic mapping of the visual cortex obtained by voltage-sensitive dye imaging
in the behaving monkey. J. Neurophysiol. 98, 1002–1014.
Neuron, Volume 74
Supplemental Information
Attentional Modulations Related
to Spatial Gating but Not to Allocation
of Limited Resources in Primate V1
Yuzhi Chen and Eyal Seidemann Supplemental Inventory
- Figure S1 is related to Figure 2; it provides additional summary information regarding behavioral
performance
- Figure S2 is related to Figure 3; it is identical to figure 3 but for a second monkey
- Figure S3 is related to Figure 4; it provides additional breakdown of responses in ‘attend out’
trials based on the position of the cue relative to the receptive field of the imaged neurons
- Figure S4 is related to Figure 6; it provides evidence that the observed modulations in V1 are not
due to direct visual response to the cue
- Figure S5 is related to Figure 6; it provides evidence that the observed modulations in V1 are not
due to variations in eye positions
- Figure S6 is related to Figure 1; it shows that behavioral improvement in focal attention can be
explained without invoking a change in neural sensitivity/fidelity due to limited resources (as in
Fig. 1A,B)
Suppl. Fig. 1
A B C
Suppl. Fig. 2
σmajor
σminor
B
C D E F
A
Suppl. Fig. 3
A B
C D
Suppl. Fig. 4
B
A
Suppl. Fig. 5
B A C
D E F
Suppl. Fig. 6
SUPPLEMENTAL FIGURE LEGENDS Figure S1. Differences in behavioral performance between focal and distributed attention.(A) The
percentage of hits in focal attention trials (single-cue) is significantly higher than in distributed attention
trials (multiple-cues). (B) The percentage of false alarms in focal attention trials (single cue) is not
significantly different than in distributed attention trials (multiple-cues) even though there are four
possible target locations in distributed attention trials and only one possible target location in focal
attention trials. This result suggests that the monkeys are adjusting their criterion between focal and
distributed attention trials. (C) Rate of “wrong target” in distributed attention trials. “Wrong target” is
defined as a target-present distributed attention trial in which the monkey selected a location different
from the target location. N, the number of experiments for each monkey; p, the p-value of a paired t-
test across experiments; error bars, SEM.
Figure S2. Average spatial pattern of V1 population responses under three attentional states in monkey
2. (A) Top row shows a schematic diagram of the three attentional states. The black circle represents the
receptive field (RF) of the neurons at the center of the imaged area, while the white circles represent
the cue. Att-in: the receptive field location is cued. Att-out: the cued location is outside of receptive
field. Att-dstr: all four locations are cued. Bottom two rows show the average response maps in mask
(middle row) and mask-plus-target (bottom row) trials under the attentional state indicated by the
diagram in the top row. At each location the responses are averaged over the first 200 ms after stimulus
onset. The ellipsoidal region at the center of the imaged area shows the area directly activated by the
mask or mask-plus-target. The response is anisotropic due to the anisotropy in the map of visual space
in V1. The response is dominated by the higher contrast mask and is therefore very similar in mask and
mask-plus-target trials. Note that in attend-out state the overall activity level in the area surrounding
the active region is lower than in attend-in and attend-distributed states. (B) Two dimensional (2-D) fit
to the spatial patterns of V1 population response. The responses were fitted with a 2-D Gaussian plus a
uniform baseline. The space constants along the major and minor axis are indicated. Top – 2-D Gaussian
fit to response to mask in attend-in state. Bottom – one dimensional horizontal slice through the two-
dimensional fit, indicated by the black line in the top panel. The baseline, amplitude and space constant
(σ) are indicated. For comparison, the spatial response profile in a 1 mm wide strip along this slice is
indicated by a dashed line. The shaded area indicates ± one SEM across experiments. Right – same as
bottom but along a vertical slice. (C)-(F) Mean and SEM of the fitted parameters across 12 experiments
in monkey 2. The color represents the attentional state as indicated in top row of (A). Asterisks indicate
significant differences (p<0.05 based on paired t-test across experiments). The only significant
differences are between the magnitude of the baseline in attend-out and the other two attentional
states.
Figure S3. Summary of attentional effects in attend-out trials in the two monkeys. Attend out-trials were
divided into trials in which the cue was opposite to the imaged area (light green; the location where the
distracter could appear) and trials in which the cue was in one of the other two locations (dark green). In
both types of attend-out trials, the baseline is not significantly elevated relative to blank trials and the
stimulus-evoked Gaussian response is indistinguishable from all other attentional conditions. The format
is the same as in Fig. 4.
Figure S4. Time courses of the baseline component time locked to cue offset show no evidence for
direct visual responses to the cue during the imaging period and therefore cannot account for the
observed modulations in V1 responses. In each trial, we used the time course of the baseline
component in the first 300 ms of data acquisition (terminating 100 ms before stimulus onset). We
divided the time courses into 100 ms long bins based on the time from cue-offset to fixation point
dimming. We then averaged the time courses of trials in the same attentional states in each bin. The
figure shows data from three bins that contained most of the trials. Dashed vertical line indicates the
average dimming time relative to cue offset. While in some of the bins the results are noisy, there is no
systematic modulation that is time locked to cue offset. (A) Results from monkey 1. (B) Results from
monkey 2. The shaded areas represent ± one SEM.
Figure S5. Fixational eye movements do not differ significantly between attentional states and therefore
cannot account for the observed modulations of V1 responses. (A) Scatter plot of average eye position
in different condition in monkey 1. Each point indicates the average eye position across trials of the
same condition during the 200 ms before stimulus onset. The four color lines represent the directions
towards the corresponding cued positions. To simplify the display, the three experiments collected from
the left hemisphere in monkey 1 were excluded from this analysis, but the results are the same with
those experiments included. (B) Mean and SEM of the average standard deviation of the eyes during the
same 200 ms interval. (C). Mean and SEM of the peak eye velocity during the same 200 ms interval. (D)-
(F) Same as A-C but for monkey 2. All p values are based on a one-way anova. ‘S-cue’ – single cue; ‘M-
cue’ – multiple cues; ‘B’ – blank; ‘Pos’ – position.
Figure S6. Behavioral improvement in focal attention can be explained without invoking a change in
neural sensitivity/fidelity (as in Fig. 1A,B). To assess the expected behavioral improvement in focal
attention under the assumption of no resource limit, we used a simple model based on signal detection
theory (Green and Swets 1966). The underlying assumption in this model is that subjects base their
decision on a noisy sensory representation, with neural responses that are normally distributed (rather
than binary as in our toy example). The mean neural response in target-present trials is higher than in
target-absent trials, and this difference, in units of response standard deviation, controls the neural
sensitivity and therefore the subject’s accuracy. In the distributed attention case, the responses at all
four locations are assumed to be independent and to have the same signal and noise levels. The
simulation was performed as follows. In simulated focal attention trials, the response from the cued
location was compared to an optimal criterion to determine whether the target was present or absent.
Trials were considered correct if the response exceeded the criterion in target-present trial or was lower
than the criterion in target-absent trial. In simulated distributed attention trials, the maximum of the
responses at the four locations was compared to an optimal criterion. Target-present trials were
considered correct if the response at target location was maximal and exceeded the criterion. Target-
absent trials were considered correct if the maximal response was lower than the criterion. The optimal
criterion was selected to maximize accuracy. For each signal-to-noise ratio, we simulated the same
number of trials as in the real experiment and repeated the simulation 1000 times. The figure shows
percent correct for single-cue trials vs. for multiple-cues trials. The black curve represents the simulation
results, while the symbols represent the experimental results for the two monkeys. The cross represents
mean ± one SD for the experimental results. The shaded areas represent ± one SD for the simulation.
The change in accuracy between focal and distributed attention is not significantly different from the
prediction of the model, indicating that it is possible to account for the behavioral results without
invoking a change in sensitivity between focal and distributed attention. A key difference between this
model and our toy example (Fig. 1) is that the criterion is allowed to vary between focal and distributed
attention. A fixed criterion followed by a random selection between all locations that exceed the
criterions (rather than using the location with the maximum response) would make the two models
equivalent. Our behavioral results provide clear evidence that the monkeys changed their criteria
between focal and distributed attention trials (Supp. Fig. 1). To see this, consider the false alarm rate in
the two attentional states. If the criterion was fixed, we would expect to see a large increase in the false
alarm rate in the distributed attention case. Instead, the false alarm rate is similar in the two attentional
states.