| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Center for Visual Science and Department of Brain and Cognitive Sciences, University of Rochester, Rochester, 14623 2 Center for Neural Science, New York University, New York City, New York 10003
Submitted 6 January 2003; accepted in final form 17 April 2003
 |
ABSTRACT |
|---|
|
|
|
INTRODUCTION |
|---|
|
;
Maffei and Fiorentini 1976). A
pattern that extends beyond the receptive field often elicits a weaker
response than a pattern that just fills it. This surrounding zone of
inhibition is not generally considered part of the receptive field because
stimuli falling on it cannot by themselves elicit a response. The most
suppressive surrounding patterns usually have the same orientation and spatial
frequency as those to which the receptive field is best tuned
(Blakemore and Tobin 1972;
Cavanaugh et al. 2002b;
DeAngelis et al. 1994;
Gilbert and Wiesel 1990;
Levitt and Lund 1997;
Nelson and Frost 1978;
Sillito et al. 1995).
It has been argued that the surround helps figure-ground segregation by
suppressing signals within patches of uniform texture
(Bradley and Andersen 1998;
Lamme 1995;
Sillito et al. 1995;
Tanaka et al. 1986) and that
it adjusts the responsivity of a neuron to the ambient contrast in the
neighborhood of the receptive field
(Cavanaugh et al. 2002a). We
wondered if the surround might confer an additional benefit. In natural
images, the correlation between the statistics of two regions declines with
the separation of the regions (Simoncelli
and Schwartz 1999). If a pattern falling on the surround
consistently suppressed a neuron's response to a pattern of similar
orientation falling on the receptive field, it might repel the neuron's
orientation tuning curve, thereby reducing the correlation (redundancy) among
visual signals that arise from adjacent image regions of similar structure. In
this respect, the surround would act in the spatial domain in much the same
way that rapid contrast adaptation
(Müller et al. 1999) acts
in the time domain.
In this paper, we explore the mechanism of surround suppression to characterize its latency and persistence and to ask whether it acts by subtraction or division. We describe how reductions in response reduce redundancy of image representation among neurons with neighboring or overlapping receptive fields: specifically, we show that surround suppression changes orientation selectivity and that these orientation-specific changes in sensitivity reduce redundancy.
|
|
METHODS |
|---|
|
). Visual stimuli
Sinusoidal gratings were generated as described in Müller et al.
(2001). Displays often
required two gratings whose spatial and temporal characteristics could be
controlled independently. These were produced on a single monitor by dividing
the 256 available lookup table entries into two independent sets of 128
entries, one allocated to each pattern. The display was viewed from a distance
of 114–342 cm, depending on the resolving power of the neuron under
study, and its height subtended 11.1–3.8°. It was refreshed at 75
Hz, with the scan from top to bottom of the screen taking 
10 ms.
Characterizing receptive fields
The basic characteristics of receptive fields, including their positions
and dimensions, were established as described in Müller et al.
(2001). Briefly, receptive
fields were first mapped using a small patch of moving grating whose spatial
and temporal characteristics were continuously adjustable by the experimenter.
The preferred position, size, orientation, and spatial frequency derived from
this examination were then used as the starting points for systematic
measurement of each of these characteristics. The position of the receptive
field was established using a patch of moving grating, of length and width
approximately matching the receptive field, presented at a matrix of positions
centered on the estimated position and spaced 0.25 lengths and widths apart.
Having found the receptive field position, we then established the rectangle
(lying in the preferred orientation of the neuron) that best matched the
receptive field in size. To establish the optimal length, we used a series of
gratings of different lengths with width fixed at a preliminary estimate; to
establish the preferred width, we used a series of gratings of different
widths with length fixed at the length preferred. We then established the
optimal spatial phase using flashed gratings at a series of phases. If at any
stage in this sequence of measurements it appeared that some estimate was
incorrect, we repeated the sequence. To characterize the influence of the
region surrounding the receptive field, we presented a grating of optimal
size, of the highest contrast that did not saturate the response, together
with a surrounding grating that enclosed but did not overlap the rectangle
that bounded the receptive field and was also of high contrast. The outer
boundary of this surround usually extended to 2° from the center of the
receptive field—a region that preliminary observations showed to be
large enough to capture essentially all of the surround's influence (see also
Cavanaugh et al. 2002a;
Levitt and Lund 2002). In
control experiments, we established that a surround pattern presented alone
elicited no excitatory response from a neuron. If the control failed, the
inner boundary of the surround (chosen initially to abut the estimated
receptive field) was enlarged in stages, leaving a gap between the surround
and any pattern placed on the receptive field. Occasionally the surrounding
pattern presented alone slightly depressed the neuron's spontaneous discharge.
In such cases, we did not enlarge the surrounding pattern to leave a gap. We
studied neurons' responses to moving gratings and to stationary gratings in
optimal spatial phase. Respiration-induced eye movements
(Forte et al. 2002) were
occasionally large enough to prevent satisfactory measurement of the responses
of simple cells to stationary gratings of high spatial frequency.
Characterizing orientation selectivity
For parts of our analysis, we needed to characterize a change in the shape
of an orientation tuning curve, generally a deformation that shifted the peak
of the tuning curve and introduced an asymmetry. To capture this, we
calculated the center-of-mass of the tuning curve
iRi/Ri
where the Ri denote the responses to gratings in a range
of orientations i, either i 
{0, ±7,
±14, ±21, and ±28°} or i {0,
±18, ±36, ±54, ±72, and ±90°} where
0° is the nominal preferred orientation. For a neuron with a symmetrical,
unimodal tuning curve that is well sampled by the test orientations the
center-of-mass is simply the preferred orientation.
|
|
RESULTS |
|---|
|
Mechanism
Because the surround's influence is expressed only by modulating responses to patterns falling on the receptive field, we infer it from the change the surrounding pattern brings about in the response to an optimal grating on the receptive field. In principle, signals from outside the receptive field might originate from one neuron or from many—in the lateral geniculate nucleus (LGN), within V1, or in extrastriate cortex—and might act subtractively or divisively. A detailed understanding of the time course of surround effects would help us to approach these questions.
Rapid, transient influence
Because neurons of the type commonly encountered in cortex respond
transiently to a stationary stimulus
(Müller et al. 2001), we
would expect the influence of a stationary surround to be transient.
Figure 1 shows, for three
neurons, the average discharge rate during the first 100 ms of response to the
presentation of an optimal stationary grating in the receptive field as a
function of the time after surround onset at which this optimal grating was
presented. The graphs therefore trace out the time course of the surround's
transient influence. The response is most suppressed when center and surround
gratings are presented synchronously and recovers rapidly—in two cases
completely—as the interval between the onsets of the surround grating
and the central grating approaches 100 ms (A and B). For
asynchronies greater than that bringing about the weakest response
Rmin, the discharge rate R is well described by
an exponential recovery (Fig.
1, —)
 | (1) |
is the time constant of recovery. If
Rrf is defined as the response to the optimal receptive
field stimulus alone, the recovering responsivity can be conveniently
characterized by its time constant and by the ratio
(Rasympt –
Rrf)/(Rmin –
Rrf), which represents the extent to which the surround's
influence is sustained.
|
The measurements and analysis illustrated in
Fig. 1 were made on 16 neurons.
Figure 2 summarizes them.
Surround effects decay quickly, often to small values (negative values mean
that the asymptotic discharge rate in the presence of a surrounding grating
was higher than when the receptive field alone was stimulated). The
distribution of time constants is like that for the decay of responses to
stationary gratings confined to the receptive field
(Müller et al. 2001)
(Fig. 2), but the sustained
surround effects have a broader distribution of amplitudes.
|
For most neurons (12/16) on which we examined the effects of gratings presented asynchronously to the surround and receptive field, the surround exerted its greatest influence when it appeared simultaneously with the grating on the receptive field (e.g., Fig. 1). This implies that surround signals act on a neuron as rapidly as do those arriving through the receptive field. We explored this further by looking at the fine structure of responses to gratings falling on receptive field and surround.
We analyzed cumulative spike counts of the kind shown in
Fig. 3. Figure 3A shows, for a
complex cell that had no spontaneous activity, the cumulative count following
the presentation of an optimal grating on the receptive field, alone (—)
and concurrently with a surrounding grating of the same orientation (- -). The
surrounding grating completely extinguished the response. The progressive scan
of the monitor ensures that parts of the surrounding grating are displayed
before the stimulus to the receptive field, and parts are delivered later.
Assuming that the dominant surround signals originate 0.5° beyond the
receptive field (Born and Tootell
1991; Cavanaugh et al.
2002a; Levitt and Lund
2002; Maffei and Fiorentini
1976) and that all of these signals arise in the part stimulated
earliest, we estimate that the surround might be stimulated as much as 1.25 ms
earlier than the receptive field. Taking this as the precision of our
measurement, surround signals act as quickly as those arising through the
receptive field. Figure
3B shows a corresponding set of traces obtained with an
orthogonal surrounding grating that increased the neuron's
responsivity. The discharge departs from baseline sooner when the surrounding
grating is present. This might mean that surround signals are available before
signals that arrive through the receptive field first drive a neuron's
membrane potential to its spike threshold. To explore that issue, we looked at
the time course of the suppression brought about by a surround presented
after the onset of the grating falling on the receptive field. This
is shown, for another cell, in Fig.
3C. Measured in this way the suppressive influence of the
surround had a latency of 32 ms—faster than the response to a grating
confined to the receptive field of this neuron (42 ms) or any other (see
following text, Fig.
4B).
|
|
Measurements of the kind shown in Fig. 3 were made on 37 neurons and are summarized in Fig. 4. Figure 4A shows the average cumulative spike counts obtained from 37 neurons following the onset of a grating that just filled the receptive field (—) or covered both the receptive field and the surround (- -). Before averaging, the traces for individual neurons were aligned to the time of onset of the response to the optimal grating. The points at which the slopes of the stimulus-driven traces begin to differ from the slope of the trace of the maintained discharge (· · ·) and from each other show the latencies of the signals arising from the optimal grating and the pattern surrounding it respectively. Both signals begin abruptly and influence discharge at the same time.
Figure 4B shows more detail about the behavior of individual neurons. For each neuron, the absolute latency of response to an optimal grating that filled the receptive field is plotted against the absolute latency of the influence of a surrounding grating of matching orientation, measured by the method shown in Fig. 3. Most neurons receive surround signals and receptive field signals at the same time. For a few (such as that of Fig. 3C) the surround is faster; for a few others it is slower.
The more reliable the effect of the surround, the more likely it is to
exert its influence at the moment the response begins.
Figure 5, shows how the
latencies of surround signals (expressed relative to the latency of signals
originating within the receptive field) vary with the strength of the
surround's influence, expressed in d' units (Eq. 2,
measured using the 1st 100 ms of discharge).
Figure 5A shows the
latencies when the surrounding grating lay at the orientation optimal for the
receptive field; Fig.
5B shows latencies when the surrounding grating lay at
the orientation orthogonal to the optimal grating on the receptive field. For
most cells, a surrounding grating, whether suppressing or facilitating the
response, acts just as rapidly as a grating confined to the receptive field.
Only for those cells showing the least reliable surround influence was that
influence delayed by 
10 ms; similarly, some of cells showing the most
advanced surround influence were among the least reliable.
|
Divisive gain control
Our use of stationary gratings to stimulate the receptive field provides an opportunity to establish whether surround signals act through division or subtraction. A stationary grating presented to the receptive field brings about a brisk response that declines over the course of a few hundred milliseconds. To characterize the nature of the surround's effect (subtractive/divisive), we presented the surrounding grating, on separate trials, at a range of times shortly after presenting a stationary grating to the receptive field. If the surround acts subtractively, the presentation of a surrounding pattern at any time after the onset of the pattern on the receptive field will lead to a constant reduction in response. On the other hand, if the surround acts divisively, the effect of its presentation will shrink with the decaying response to the stationary pattern on the receptive field. The filled symbols in Fig. 6A show, for one neuron, the response after the onset of a stationary grating confined to the receptive field, sampled in 100-ms epochs, each beginning at the poststimulus time indicated by the symbols (some of these epochs overlap). Open symbols show the response to the same grating, sampled in the same way, but now from a different set of trials in each of which a surrounding grating was presented just before one of the sampling epochs (by a fixed interval equal to the neuron's onset latency), so that the added grating begins to influence discharge immediately before the beginning of the epoch indicated by the symbol. In other words, the open symbols show what would be the response to the presentation of a stationary grating in the receptive field, in the presence of a surround of constant potency. The dotted line shows the uppermost curve displaced so as to best fit the response obtained in the presence of the surrounding grating. This is what one might expect were the surround acting subtractively, but it does not characterize the data well.
|
The response in the presence of the surrounding grating is much better described by assuming that a newly presented surrounding grating has the same divisive effect at any time during the response to the grating in the receptive field. This is shown by the dashed line, which is a reduced-scale replica of the uppermost curve, fitted after discounting the maintained discharge.
We made measurements of this kind on n = 8 neurons and
fit subtractive and divisive models as in the preceding text. Fit quality was
measured by the percentage of variance in the data that each model left
unaccounted for (Fig.
6B): the model's mean-squared fit error divided by the
mean-squared error of the best-fit horizontal line. This quantity is zero for
perfect fits, 100 for fits that are just as good as the best horizontal line,
and potentially larger when data are noisy. We used a permutation test
(Edgington 1995) to determine
whether a subtractive or a divisive surround better explained the results. The
null hypothesis was that the distributions of fit quality for the two models
were the same (each model has exactly 1 free parameter). A simulated data set
was drawn randomly from that distribution by choosing each of the 2n
actual fit qualities without replacement. This was repeated 5000 times. To
assess statistical significance we computed a two-tailed P value, the
probability that a mean absolute difference between distributions of fit
qualities for the two models as simulated was greater than or equal to that
for the actual fits. Division explained significantly more of the variance
than subtraction (P < 0.01, permutation test). The divisive model
left 8% of the variance unaccounted for in the median case; the subtractive
model left 21%. For seven of these neurons the divisive model fit well,
leaving <15% of the variance unaccounted for
(Fig. 6B).
Influence on orientation selectivity
To establish in detail how the presence of a surrounding grating influences a neuron's orientation tuning, we measured responses to a set of optimally sized gratings of the preferred spatial frequency at a range of orientations, alone, and in the presence of each of two surrounding gratings that lay in different orientations slightly oblique to the preferred orientation.
We made measurements with stationary gratings in optimal phase as well as
with moving gratings. When using stationary gratings we measured the first 100
ms of response, which captures the onset transient
(Müller et al. 2001).
When using moving gratings, we also analyzed the onset transient (initial 100
ms of response), having first confirmed that this yielded the same results as
analyzing the whole response.
Figure 7, A and B, shows for two complex cells how the orientation of a surrounding grating influenced orientation selectivity measured with a grating confined to the receptive field. By depressing responsivity to gratings in the neighborhood of its own orientation and (for the neuron in Fig. 7B) increasing responsivity to gratings in other orientations, the surrounding grating repels the peak of the tuning curve. The surrounding grating also reduces the variability of the response to neighboring orientations (e.g., error bars in Fig. 7B). Figure 7C shows, for a simple cell, that orientation selectivity is not systematically affected by the orientation of a surrounding grating. Responses are suppressed, and orientation selectivity is even changed but without regard for the orientation of the surrounding grating.
|
To quantify the deformation in orientation tuning illustrated in
Fig. 7, we computed the
center-of-mass
iRi/Ri
(see METHODS) for all the cells on which we made measurements.
Figure 8 shows, for 32 complex
cells and 9 simple cells, the center of mass found in the presence of a
surrounding grating inclined to one side of the preferred orientation (usually
by 14°) against the center of mass found in the presence of a grating
inclined by the same amount to the other side of the preferred orientation.
For complex cells (
and 
) the surrounding grating often repels
(and never attracts) the center of mass. Exchanging one surrounding grating
for the other moves the center of mass by 4° on average. For simple cells
(
) a surrounding grating often reduced responsivity, but in 8/9 neurons
we studied, orientation tuning was not influenced by the orientation of the
surrounding grating. Thus in Fig.
8, 8/9 simple cells fall along the unit diagonal.
|
A surrounding grating selectively depresses the responsivity of a complex
cell to a central grating of similar orientation, making the tuning curve
locally steeper and less variable from trial to trial. Might it thus improve
the neuron's capacity to distinguish gratings of orientations near the
orientation of the surround? We examined this by estimating the
discriminability of two gratings, identical except for a 14° (or 36°)
difference in orientation. We measured a series of responses to each, and for
the resulting distributions of spike counts calculated d', the
difference between the mean counts, divided by the SD. When the two
distributions had different SDs, we used the root-mean-square SD, thus
 | (2) |
). We
measured responses twice, once with and once without, a surrounding grating
whose orientation matched one of the test gratings. Among the 32 neurons for
which we made this comparison the presence of the surrounding grating
increased d' significantly (t-value increased by
2) in 9 but decreased it significantly in 5. A surrounding pattern that
depresses responsivity thus appears to have no systematic effect on the
discriminability of patterns falling in the receptive field. |
|
DISCUSSION |
|---|
|
By depressing responsivity and changing a complex cell's tuning locally in
stimulus space, a surrounding grating will necessarily reduce correlation
among the discharge rates (as distinct from the often-discussed correlation in
spike times) of the population of neurons that respond to a particular
stimulus (e.g., a grating of a particular orientation). This will increase the
information transmitted by each spike
(Barlow and Földiák
1989). Consider how the responsiveness of a population of neurons
that have neighboring or overlapping receptive fields and are tuned to similar
orientations will be changed by a single grating that surrounds all their
receptive fields. We simulate such a population by taking the neurons from
Fig. 8 (32 complex cells, 9
simple cells, studied with 2 surround orientations) and normalizing their
actual tuning curves so that the surround has an orientation of 0°. This
results in a simulated population of neurons most with preferred orientation
–14, 0, or 14°. Figure
9A shows orientation tuning of two complex cells, each
measured with a grating confined to its receptive field, presented alone
(dotted traces) and with a surrounding grating at nominal orientation 0°
(solid traces with symbols). The surrounding grating sharply reduced each
complex cell's response to gratings with similar orientations, and pushed
apart the neurons' preferred orientations. Whenever a single grating extends
across the receptive fields of a population of complex cells, these changes in
tuning will make their responses less redundant.
|
To quantify the redundancy of the simulated population's response rates and
the surround's role in reducing it, we compute the pairwise cross-correlation
between the response rates of the population of neurons weighted according to
the probability of a stimulus of each orientation. Thus we define redundancy
(
) in the population's response to each grating of orientation 
falling in the receptive field as
 | (3) |
The measure can be used to show how surrounding gratings would reduce
redundancy among the response rates of the simulated population. We calculate
with and without a surround for gratings presented to the receptive
field at a range of orientations. Figure
9B (
) shows the average reduction in redundancy at
each orientation (relative to the orientation of the surrounding grating) for
pairs of neurons drawn from our population of 32 complex cells. Naturally,
redundancy is most reduced when firing rate is most reduced: when the
orientation of the grating falling on the receptive field is near that of the
surround. The orientation-dependent reduction in redundancy arises because the
surrounding grating changes the orientation selectivities of neurons. This
happens without reducing the discriminability of patterns falling on the
receptive field (see RESULTS). The redundancy among the responses
of our population of nine simple cells (open squares) was also reduced by the
surround but to a uniform value at every orientation. This is because, in
simple cells, surround stimulation scales down the entire orientation tuning
curve without systematically changing its shape
(Fig. 7C).
The reduced redundancy among the responses of the population of neurons
tuned to a particular orientation will increase the information transmitted by
each spike (Barlow 1990) and
saving energy (Lennie 2003).
Benefits will be greatest when parts of the image falling just outside the
receptive field (on the surround) have nearly the same structure as the parts
falling on the receptive field: the more a pattern falling on the receptive
field resembles those that surround it, the greater the reduction in
redundancy among the responses of the population. This behavior seems well
matched to the properties of natural scenes: Simoncelli and Schwartz
(1999) have shown, for a small
sample of natural images, that statistics of iso-oriented filters responding
to adjacent image regions are highly correlated, whereas those from
cross-oriented filters or more separated regions are less well correlated.
This encourages us to think of inhibitory lateral interaction as a phenomenon
that complements the rapid adaptation described by Müller et al.
(1999): both remove local
correlations from neuronal signals, one in time, the other in space.
Relation to other findings
Gilbert and Wiesel (1990)
and Sengpiel et al. (1997)
reported (sometimes large) changes in preferred orientations of neurons in cat
cortex brought about by surrounding gratings but did not find the consistent
repulsion of preferred orientation that we observed here in complex cells.
Knierim and Van Essen's
1992 observations on latencies of surround influences in awake
monkey (Table 1) are consistent with our finding that surrounds begin to
suppress discharge as soon as neurons begin responding, though Knierim and Van
Essen found significant suppression only after a delay. Zipser et al.
(1996) studied how V1 neurons
in awake monkeys responded to patterns that covered and extended beyond the
receptive field and were embedded in a larger surrounding pattern that had the
same or contrasting structure. The influence of such surrounding patterns is
expressed 80–100 ms after the beginning of the response to the central
pattern. Moreover, it influences responses only in the awake monkey
(Lamme et al. 1998). Because
of this, and the fact that even the central patterns used by Zipser et al.
would have stimulated the kind of surround we have characterized, we think
this long-latency phenomenon is unrelated to the one studied here. In
anesthetized monkeys, Bair et al.
(1999) stimulated receptive
field and surround independently with a continuous series of very briefly
presented gratings in four possible pairings of preferred and orthogonal
orientations and found that the surround's suppressive effect was delayed
20–30 ms. The temporal structure of stimulation—an uninterrupted
train of high-contrast stimuli—was quite unlike that provided by our
discrete trials with intervening blanks and probably results in neurons having
lower contrast gain. How that influences surround latency remains to be
determined.
Origin of surround signals
Two broad classes of accounts have been offered of the origin of surround
signals: they are conveyed via lateral connections within V1, such as those
described by Rockland and Lund
(1983) and Gilbert and Wiesel
(1983) or they are conveyed
through feedback connections from extrastriate cortex, an idea suggested by
the observation that some surround effects are expressed with long latencies
(Zipser et al. 1996).
In normally functioning visual cortex, synaptic delays are probably 5
ms (Maunsell and Gibson 1992).
Two or more of these, amounting to 10 ms, would be required for a feedback
signal from extrastriate cortex to influence a V1 neuron. If such a feedback
loop originated in V1 neurons of the kind we have studied here, signal
transmission time would be too long to explain the often-instantaneous action
of the surround. Could a feedback loop originate in V1 in signals that we have
not characterized? One possible source might be a class of short-latency relay
neurons that itself does not express any immediate surround influence and that
we have not studied. We think it unlikely that such neurons exist. Our method
for measuring latency (Fig. 3)
can identify the earliest times at which signals are present in V1 neurons
(and therefore available to any feedback loop). The shortest surround
latencies match the shortest latencies for stimuli in the receptive field and
thus too short to be plausibly the result of feedback
(Fig. 4B).
The short latency of surround signals, their circumscribed range (see
following text), and their asymmetrical/irregular weight around the receptive
field (Cavanaugh et al. 2002b;
Walker et al. 1999) make it
unlikely that they arise in feedback from extrastriate cortex. Might they
arise in connections from neurons within V1
(Das and Gilbert 1999)? In the
region of V1 from which we made our recordings strong surround signals, which
most often originate within 0.5° of the receptive field
(Born and Tootell 1991;
Maffei and Fiorentini 1976),
will most often need to be propagated over 2 mm or less because the
magnification factor is 4 mm/°
(Van Essen et al. 1984)
(Fig. 6). Some inhibitory
interneurons have axonal arbors as much as 800 µm across
(Lund and Wu 1997) and so
could carry signals from much or most of a neuron's surround. Signals from the
LGN could drive interneurons to give rise to surround signals that appear to
act as fast as those arising in the receptive field. Membrane depolarization
that would normally lead to a spike begins 10 ms before the spike arises
(Azouz and Gray 1999), so a
surround signal (either hyperpolarizing or depolarizing) that arrived 
10
ms later than the normal driving signal could alter the time at which the
spiking threshold was reached. Although delayed, that surround signal would
appear to act sooner than the normal drive to the receptive field. Presumed
inhibitory interneurons in visual cortex studied in vivo have shorter visual
latencies than presumed relay cells
(Mancilla et al. 1998). All
this can be explained by surround signals that arise not from feedback from
extrastriate cortex, but in connections from neurons within V1.
Relation to other contrast gain controls
Surround signals act divisively to regulate sensitivity
(Fig. 9) (also
Cavanaugh et al. 2002a) and
are just one class of several in cortex that regulate contrast gain but do not
themselves drive a cell. Cross-orientation inhibition
(Bonds 1989;
DeAngelis et al. 1992) and
null-phase inhibition (Geisler and
Albrecht 1992) within the receptive fields of simple cells act
this way. The latter phenomena can be explained by models that assume a
neuron's contrast sensitivity is regulated ("normalized") by a
pooled signal from a large number of neurons with overlying receptive fields
covering all orientations (Carandini et al.
1997; Heeger
1992). Because this pooled signal is isotropic, it probably does
not give rise to the surround inhibition studied here.
Rapid contrast adaptation (Müller
et al. 1999) is another phenomenon that regulates contrast gain
via a mechanism that also seems to be unlike that used by the surround.
Nevertheless, much like the surround it brings about pattern-selective changes
in the orientation tuning of complex cells only and does so without impairing
a complex cell's capacity to discriminate orientation. These shared
characteristics can be explained by supposing that the fundamental
gain-controlling operation, whether lateral influence from the surround or
rapid adaptation, originates in simple cells and that several simple cells
with a range of orientation preferences are the subunits that provide input to
each complex cell.
|
|
DISCLOSURES |
|---|
|
|
|
ACKNOWLEDGMENTS |
|---|
|
|
|
FOOTNOTES |
|---|
Address for reprint requests: J. R Müller, Howard Hughes Medical Institute and Dept. of Neurobiology, Fairchild D209, Stanford University School of Medicine, Stanford, CA 94305-5125 (E-mail: jim{at}monkeybiz.stanford.edu).
|
|
REFERENCES |
|---|
|
Bair W, Cavanaugh JR, and Movshon JA. The time course of contextual modulation in macaque V1 neurons. Invest Ophthalmol Vis Sci 40: S201, 1999.
Barlow HB. A theory about the functional role and synaptic mechanisms of visual aftereffects. In: Vision: Coding and Efficiency, edited by Blakemore C. Cambridge, UK: Cambridge Univ. Press, 1990, p. 363–375.
Barlow HB and Földiák P. Adaptation and decorrelation in the cortex. In: The Computing Neuron, edited by Durbin RM, Miall C, and Michison GJ. New York: Addison-Wesley, 1989, p. 54–72.
Blakemore C and Tobin EA. Lateral inhibition between orientation detectors in the cat's visual cortex. Exp Brain Res 15: 439–440, 1972.[Web of Science][Medline]
Bonds AB. Role of inhibition in the specification of orientation-selectivity of cells in the cat striate cortex. Vis Neurosci 2: 41–55, 1989.[Web of Science][Medline]
Born RT and
Tootell RBH. Single-unit and 2-deoxyglucose studies of side inhibition in
macaque striate cortex. Proc Natl Acad Sci USA
88: 7071–7075,
1991.
Bradley DC and
Andersen RA. Center-surround antagonism based on disparity in primate area
MT. J Neurosci 18:
7552–7565, 1998.
Carandini M,
Heeger DJ, and Movshon JA. Linearity and normalization in simple cells of
the macaque primary visual cortex. J Neurosci
17: 8621–8644,
1997.
Cavanaugh JR,
Bair W, and Movshon JA. Nature and interaction of signals from the
receptive field center and surround in macaque V1 neurons. J
Neurophysiol 88:
2530–2546, 2002a.
Cavanaugh JR,
Bair W, and Movshon JA. Selectivity and spatial distribution of signals
from the receptive field surround in macaque v1 neurons. J
Neurophysiol 88:
2547–2556, 2002b.
Das A and Gilbert CD. Topography of contextual modulations mediated by short-range interactions in primary visual cortex. Nature 399: 655–661, 1999.[Medline]
DeAngelis GC,
Freeman RD, and Ohzawa I. Length and width tuning of neurons in the cat's
primary visual cortex. J Neurophysiol
71: 347–374,
1994.
DeAngelis GC,
Robson JG, Ohzawa I, and Freeman RD. Organization of suppression in
receptive fields of neurons in cat visual cortex. J
Neurophysiol 68:
144–163, 1992.
Edgington ES. Randomization Tests. New York: Dekker, 1995.
Forte J, Peirce JW, Kraft JM, Krauskopf J, and Lennie P. Residual eye-movements in macaque and their effects on visual responses of neurons. Vis Neurosci 19: 31–38, 2002.[Web of Science][Medline]
Geisler WS and Albrecht DG. Cortical neurons: isolation of contrast gain control. Vision Res 32: 1409–1410, 1992.[Web of Science][Medline]
Gilbert CD and Wiesel TN. Clustered intrinsic connections in cat visual cortex. J Neurosci 3: 1116–1133, 1983.[Abstract]
Gilbert CD and Wiesel TN. The influence of contextual stimuli on the orientational selectivity of cells in primary visual cortex of the cat. Vision Res 30: 1689–1701, 1990.[Web of Science][Medline]
Green DM and Swets JA. Signal Detection Theory and Psychophysics. New York: Wiley, 1966.
Heeger DJ. Normalization of cell responses in cat striate cortex. Vis Neurosci 9: 181–197, 1992.[Web of Science][Medline]
Hubel DH and Wiesel TN. Binocular interaction in striate cortex of kittens reared with artificial squint. J Neurophysiol 28: 1040–1059, 1965.
Knierim JJ and
Van Essen DC. Neuronal responses to static texture patterns in area V1 of
the alert macaque monkey. J Neurophysiol
67: 961–980,
1992.
Lamme VAF. The neurophysiology of figure-ground segregation in primary visual cortex. J Neurosci 15: 1605–1615, 1995.[Abstract]
Lamme VAF,
Zipser K, and Spekreijse H. Figure-ground activity in primary visual
cortex is suppressed by anesthesia. Proc Natl Acad Sci
USA 95:
3263–3268, 1998.
Lennie P. The cost of cortical computation. Curr Biol 13: 493–497, 2003.[Web of Science][Medline]
Levitt JB and Lund JS. Contrast dependence of contextual effects in primate visual cortex. Nature 387: 73–76, 1997.[Medline]
Levitt JB and Lund JS. The spatial extent over which neurons in macaque striate cortex pool visual signals. Vis Neurosci 19: 439–452, 2002.[Web of Science][Medline]
Lund JS and Wu CQ. Local circuit neurons of macaque monkey striate cortex. IV. Neurons of laminae 1–3A. J Comp Neurol 384: 109–126, 1997.[Web of Science][Medline]
Maffei L and Fiorentini A. The unresponsive regions of visual cortical receptive fields. Vision Res 16: 1131–1139, 1976.[Web of Science][Medline]
Mancilla JG, Fowler M, and Ulinski PS. Responses of regular spiking and fast spiking cells in turtle visual cortex to light flashes. Vis Neurosci 15: 979–993, 1998.[Web of Science][Medline]
Maunsell JRH and Gibson JR. Visual response latencies in striate cortex of the macaque
monkey. J Neurophysiol 68:
1332–1344, 1992.
Müller JR,
Metha AB, Krauskopf J, and Lennie P. Information conveyed by onset
transients in responses of striate cortical neurons. J
Neurosci 21:
6978–6990, 2001.
Müller JR,
Metha AB, Krauskopf J, and Lennie P. Rapid adaptation in visual cortex to
the structure of images. Science
285: 1405–1408,
1999.
Nelson JI and Frost BJ. Orientation-selective inhibition from beyond the classical receptive field. Brain Res 139: 359–365, 1978.[Web of Science][Medline]
Rockland KS and Lund JS. Intrinsic laminar lattice connections in primate visual cortex. J Comp Neurol 216: 303–318, 1983.[Web of Science][Medline]
Sengpiel F, Sen A, and Blakemore C. Characteristics of surround inhibition in cat area 17. Exp Brain Res 116: 216–228, 1997.[Web of Science][Medline]
Sillito AM, Grieve KL, Jones HE, Cudeiro J, and Davis J. Visual cortical mechanisms detecting focal orientation discontinuities. Nature 378: 492–496, 1995.[Medline]
Simoncelli EP and Schwartz O. Image statistics and cortical normalization models. In: Advances in Neural Information Processing Systems, edited by Kearns MS, Solla SA, and Cohn DA. Cambridge, MA: MIT Press, 1999.
Tanaka K, Hikosaka K, Saito H, Yukie M, Fukada Y, and Iwai E. Analysis of local and wide-field movements in the superior temporal visual areas of the macaque monkey. J Neurosci 6: 134–144, 1986.[Abstract]
Van Essen DC, Newsome WT, and Maunsell JHR. The visual field representation in striate cortex of the macaque monkey: asymmetries, anisotropies, and individual variability. Vision Res 24: 429–448, 1984.[Web of Science][Medline]
Walker GA,
Ohzawa I, and Freeman RD. Asymmetric suppression outside the classical
receptive field of the visual cortex. J Neurosci
19: 10536–10553,
1999.
Zipser K, Lamme
VAF, and Schiller PH. Contextual modulation in primary visual cortex.
J Neurosci 16:
7376–7389, 1996.
This article has been cited by other articles:
|
|
 |
|
  S. P. MacEvoy, T. D. Hanks, and M. A. Paradiso Macaque V1 Activity During Natural Vision: Effects of Natural Scenes and Saccades J Neurophysiol, February 1, 2008; 99(2): 460 - 472. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. Tailby, S. G. Solomon, N. T. Dhruv, N. J. Majaj, S. H. Sokol, and P. Lennie A New Code for Contrast in the Primate Visual Pathway J. Neurosci., April 4, 2007; 27(14): 3904 - 3909. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Schwabe, K. Obermayer, A. Angelucci, and P. C. Bressloff The Role of Feedback in Shaping the Extra-Classical Receptive Field of Cortical Neurons: A Recurrent Network Model J. Neurosci., September 6, 2006; 26(36): 9117 - 9129. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. A. Smith, W. Bair, and J. A. Movshon Dynamics of suppression in macaque primary visual cortex. J. Neurosci., May 3, 2006; 26(18): 4826 - 4834. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. S. Webb, C. J. Tinsley, C. J. Vincent, and A. M. Derrington Spatial Distribution of Suppressive Signals Outside the Classical Receptive Field in Lateral Geniculate Nucleus J Neurophysiol, September 1, 2005; 94(3): 1789 - 1797. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Chen, Y. Dan, and C.-Y. Li Stimulation of non-classical receptive field enhances orientation selectivity in the cat J. Physiol., April 1, 2005; 564(1): 233 - 243. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. G. Solomon, J. W. Peirce, and P. Lennie The Impact of Suppressive Surrounds on Chromatic Properties of Cortical Neurons J. Neurosci., January 7, 2004; 24(1): 148 - 160. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. K. Murphy and K. D. Miller Multiplicative Gain Changes Are Induced by Excitation or Inhibition Alone J. Neurosci., November 5, 2003; 23(31): 10040 - 10051. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
TOP
ABSTRACT
INTRODUCTION
), and in the presence of the
surrounding grating alone (
). B:
average reduction brought about by a surround in the redundancy