The Spatial Distribution of the Antagonistic
Surround of MT/V5 Neurons
D.-K. Xiao, S. Raiguel, V. Marcar1 and G.A. Orban
The majority (217/325, 66%) of the neurons in the middle temporal
(MT) area/V5 show strong antagonistic surrounds, defined here by a
decrease of at least 50% in the summation curve. We mapped the
antagonistic surround in 145 such cells, using eight circularly
distributed surround stimulus patches (Surround Asymmetry Test,
SAT) and also mapped the surround in 51 of these 145 cells using a
grid consisting of 25 square patches (Surround Mapping Test, SMT).
Both tests showed that the angular surround distribution was
non-uniform in the majority of these neurons. In half the neurons, the
antagonistic surround was asymmetric, and arose from a single
region on one side of the excitatory receptive field (ERF). In another
quarter of the sample the surround was bilaterally symmetric, and
arose from a pair of regions on opposite sides of the ERF. Only the
remaining 20% showed a circularly symmetric surround distribution.
These three groups differed in their laminar distribution. The SMT
showed that, radially, the surround antagonism reached a maximum,
on average, at 1.5 times the ERF radius. Detailed comparisons of the
spatial relationships of excitatory and inhibitory regions of the RF
components shows that non-homogeneity of the surround influence
appears to be an intrinsic property of the surround. Such a property
may underly the extraction of the surface orientation and curvature
from speed patterns.
selectivities, at least with regard to color, where they seem to
underlie color constancy (Zeki, 1983).
In the motion domain, there are additional incentives for
creating antagonistic surrounds. Locomotion induces motion
over the retina but uniform motion — translation at the same
speed and direction over a wide region — contains little information. However, spatial changes in the velocity field contain a
great deal of information, not only about object trajectories in
three dimensions, but also about object shape — in both two and
three dimensions — and egomotion in three dimensions. Thus
antagonistic surrounds, insofar as they capture changes in the
velocity field, would prove very useful in the motion domain.
Indeed the existence of antagonistic surrounds is well
documented in area MT of the owl monkey as well as area MT/V5
of the macaque (Allman et al., 1985a,b; Tanaka et al., 1986;
Lagae et al., 1989; Born and Tootell 1992; Raiguel et al., 1995).
It is well established that area MT/V5 plays a major role in the
analysis of retinal motion (for review, see Orban, 1997). In fact,
the majority of MT/V5 neurons have some degree of antagonistic
surround. Tanaka et al. (1986) reported that 40% of the cells they
recorded in area MT showed a surround suppression of 50% or
more; Raiguel et al. (1995) found that >90% of the cells tested
showed a surround inhibition of at least 15%, with a median
inhibition of 63%. Furthermore, there is definite laminar
distribution of surround inf luences: many neurons in layer IV,
and to some extent layer V I, exhibit weak or no surround
antagonism (Raiguel et al., 1995). Although the occurrence of
antagonistic surrounds in area MT/V5 is well documented, their
properties have received relatively little attention. Tanaka et al.
(1986) compared the surround inf luence exerted by two
wedge-shaped surround stimuli either aligned with the axis of
preferred motion, or orthogonal to it. The two pairs of wedges
yielded similar levels of surround antagonism in any given
neuron. This has been accepted as evidence that the antagonistic
surrounds of MT/V5 neurons are spatially uniform.
However, the antagonistic surround as classically conceived,
in which the surround encircles the CRF, is useful mainly for
detecting discontinuities in the retinal velocity distribution. This
can be advantageous for figure–ground segregation and for
detecting the presence of a discrete object. Theoretical studies
(Droulez and Cornilleau-Pérès, 1990; Koenderink and Van
Doorn, 1992; Buracas and Albright, 1996) have shown that for
the extraction of information concerning three-dimensional
structure, asymmetric surrounds are much more effective. Thus,
the aim of the present experiments was to test the spatial
distribution of the surround inf luences. Since the question
centered upon the angular distribution of these surround
inf luences, we developed a test in which eight different
positions encircling the CRF were tested separately. In order to
obtain similar information at higher spatial resolution in both
the angular and radial direction, we introduced a second test in
Introduction
In the visual world, the retinal images of individual objects are
relatively restricted and hence whatever aspect of the image we
consider, whether color, texture, luminance, motion or depth,
the image is rarely uniform with respect to these characteristics.
We would therefore expect the visual system to be tuned for
processing spatial variations in motion, luminance or color,
rather than for uniform fields of these parameters. This is well
illustrated by the example of luminance, since even at the level
of the retina, the ganglion cells have a center-surround
organization as a means of attenuating signals of uniform
luminance. The selectivity for luminance differences is further
refined at the early cortical level by the powerful inhibitory
f lanks present in the receptive fields (RFs) of simple cells in area
V1 (Bishop et al., 1971). Non-uniformity associated with other
aspects of vision is detected at later stages of visual processing,
when the parameters relevant to that aspect are extracted, often
through analogous center-surround organizations. Inf luences
from beyond the classical receptive field (CRF) have been
documented even at the primary cortical levels (Orban et al.,
1987; Knierim and Van Essen, 1992; Sillito et al., 1995), but it is
in the extrastriate areas that surround inf luences become fully
expressed. Antagonistic surrounds have been documented in the
extrastriate area V4 where antagonism for spatial frequency,
orientation and even wavelength have been reported (Zeki,
1983; Desimone et al., 1985; Schein and Desimone, 1990; Ghose
et al., 1994). These neurons respond much more strongly to a
small patch of a monochrome grating or colored stimulus than
they do to a larger one. These surrounds can also give rise to new
Laboratorium voor Neuro- en Psychofysiologie, Katholieke
Universiteit Leuven, School of Medicine, Campus Gasthuisberg,
B-3000 Leuven, Belgium
1
Current address: Sektion Visuelle Sensorik, Universitäts —
Augenlinik, Waldhörnlestrasse 22, D-7202 Tübingen, Germany
Cerebral Cortex Oct/Nov 1997;7:662–677; 1047–3211/97/$4.00
which 25 different positions were tested for their interaction
with a small central stimulus. Both tests pointed to the same
conclusion: in most MT/V5 neurons the surround is not
uniformly distributed around the CRF. The results of this
investigation into the spatial distribution of surround inf luences
suggests that we must modify our concept of surround function
to include a broader range of functionality more in line with
theoretical predictions. Preliminary reports have been
published (Xiao et al., 1994, 1995).
Materials and Methods
The basic methods employed in this study were similar or identical to
those described in previous reports (Lagae et al. 1993, 1994; Raiguel et
al., 1995), thus only a brief description will be given. Modifications and
methods unique to the present study will be discussed fully.
Animal Preparation
Extracellular single-unit recordings were made in area MT/V5 of 15
anesthetized (Sufenta Forte 5 µg/kg/h) and paralyzed (Pavulon 0.4
mg/kg/h) male macaque monkeys (Macaca fascicularis) weighing
between 3.2 and 5.4 kg, prepared for acute recording. These animals
were also used in other studies that have been reported elsewhere. The
cortical area and laminar location of each cell recorded were later
confirmed histologically.
Visual Stimulation and General Procedure
Patches of random dot patterns, which consisted of white (48 cd/m2) dots
on a dark (0.2 cd/m2) background, moving coherently in the
frontoparallel plane, were the basis of our visual stimuli. Each test
consisted of a number of stimulus conditions differing either in patch
size, location, speed and/or direction of motion. All stimuli were
generated prior to testing as stored sequences of 512 × 512 pixel images
using a Microvax II Workstation. During testing, the stimuli were
displayed on a video monitor at a 100 Hz refresh rate by a Gould IP 9545
image computer, and presented in pseudorandom order. The monitor was
mounted on a device that enabled us to position the monitor freely in
three-dimensional space so that the line from the eye to the center of the
monitor screen was kept perpendicular to the screen surface. The
distance between the eye and the screen was either 0.57 m or 0.28 m,
depending on the size of the receptive field (RF) being tested. Dot density
in the stimuli was 2.5 dots per square degree with dots 0.35° in diameter
at the usual testing distance of 0.57 m, and all allusions to stimulus size or
speed refer to dimensions at this viewing distance. Random dots filled the
entire 25.6 × 25.6° area of the monitor at all times, but only the dots
within the stimulus moved during stimulus presentation.
Penetrations entered the brain midway between the superior
temporal and lunate sulci at ∼15 mm from the midline and running in the
parasagittal plane at an angle of 25–30° posterior to the vertical.
Glass-coated tungsten electrodes were used with a free tip of 7∼9 µm and
an in situ resistance of ∼10 MΩ. With later animals, magnetic resonance
images of the brain were used to facilitate planning the penetration. Small
electrolytic lesions were made along each penetration to aid in the
histological reconstruction of the electrode path and the identification of
the cortical area (Van Essen et al., 1981; Desimone and Ungerleider, 1986)
and layer of each neuron recorded (Raiguel et al., 1995).
During the experiment, area MT/V5 was identified physiologically by
the high proportion of directionally selective cells and by the visuotopic
organization of the RFs. After a single MT/V5 neuron was isolated, the
position and size of its RF, as well as its preferred direction and preferred
speed of stimulus motion, were approximated by hand plotting. We will
refer to this hand-plotted RF as the classical RF (CRF), according to the
standard practice of previous investigators (A llman et al., 1985a,b;
Tanaka et al. 1986). A CRF diameter was calculated as the square root of
the length times the width of the hand-plotted RF. The data presented in
this report, however, were obtained from the subsequent quantitative
tests using computer-controlled stimuli. The video monitor, on which the
stimuli were displayed, was initially centered upon the hand-plotted RF,
usually at a distance of 0.57 m from the eye. However, for cells with large
RFs or those preferring faster speeds of stimulus motion, a distance of
0.28 m was used. In the quantitative tests, cells were stimulated
monocularly, using the eye giving the stronger response. Spikes were
recorded over a period starting at 250 ms before the onset of stimulus
movement and ending 500 ms after the stimulus had stopped. On-line
data analysis provided the necessar y feedback for selecting the
appropriate stimuli during the experiment.
Quantitative Testing
The five quantitative tests employed in this investigation will be described
in the order of actual testing. The cell’s preferred direction and speed of
stimulus motion were first determined using the direction test which
consisted of 48 stimulus conditions comprising 16 different directions
covering 0–360° at 22.5° intervals, and three different speeds (2.5, 10
and 40°/s). In some of the later experiments, the three speeds tested were
5, 20 and 40°/s. The size of the stimulus used in this test was estimated
from hand plotting and controlled by masking the screen with a circular
cutout. Data were collected over at least five and usually 10 or more
stimulus presentations for this and all subsequent tests. Direction-tuning
curves were computed for each speed and the direction and speed of the
stimulus eliciting the best response were taken as the optima for the cell.
The monitor was then rotated so that the horizontal axis of the screen was
kept parallel to the cell’s preferred direction of stimulus motion, because
all the stimuli used in subsequent tests moved only along the horizontal
axis of the screen. The cell’s preferred speed of stimulus motion was
utilized in all stimuli in the subsequent tests.
The extent of the RF was mapped using the two-dimensional
position (P2D) test (Fig. 1A). Square patches (5 × 5°) of random dots,
moving in the cell’s preferred direction and speed of stimulus motion,
were presented at the 25 positions of a 5 × 5 grid. For cells with an RF
diameter <5°, stimulus squares measuring 2.5° on a side were used, and
for those with an RF diameter >15°, the monitor was placed at 0.28 m to
produce 10 × 10° patches. To visualize the RF, the response at each grid
position was normalized to the highest response level and plotted as a
function of stimulus position. Values between neighboring grid positions
were linearly interpolated and connected with isoresponsive contour
lines drawn at intervals of 12.5%. The resulting response contour was
displayed on-line as a color-coded response map. Since this quantitatively
determined, excitatory RF was much larger than the hand-plotted CRF,
we shall refer to the quantitatively determined RF as the Excitatory RF
(ERF), although previous publications (Raiguel et al., 1995; Xiao et al.,
1995, 1997) have used the term CRF for both the hand-plotted and the
quantitative entity. The center of mass, determined mathematically from
the response strength at all grid positions, was considered as the cell’s
ERF center. The stimulus display was then repositioned so that the center
of the screen coincided with the center of the cell’s ERF. If necessary, this
test was repeated several times until the display was centered to within a
half a degree and the entire RF was contained within the screen. For a
number of cells (17/145) with small RFs, these repetitions entailed testing
with smaller, 2.5° squares (Fig. 2) in the interest of more precise
centering, before returning to the standard size to cover as much of the
surround as possible in the subsequent tests.
The existence of an antagonistic surround was determined by the
summation test (Fig. 1B), in which eight circular stimuli of different sizes
and centered on the ERF center were presented. The diameters of the
stimuli covered a range from 3.2 to 25.6°, at 3.2° intervals. A decrease in
response level as the stimulus size increases beyond a certain point
indicates the presence of an antagonistic surround, and the amount of this
decrease ref lects the strength of the antagonistic surround. Only cells
with a strong surround, in which the response at maximum diameter was
suppressed to 50% or less of the maximum response, were further
examined using the surround tests. According to our estimate based on a
sample of 325 MT/V5 neurons, 66% of these neurons have a strong
surround as defined here. The summation test not only detects the
presence of a surround, but also provides crucial information as to the
distance from the ERF center at which the surround inf luences become
detectable. Although, the optimum diameter, which corresponds to the
peak of the summation curve, indicates the point where the antagonistic
surround inf luence first becomes detectable, the surround per se could
actually extend further towards the RF center and overlap to a greater
extent with the ERF.
To investigate the spatial distribution of the antagonistic surround
strength of MT/V5 cells, we designed two specific tests which presented
Cerebral Cortex Oct/Nov 1997, V 7 N 7 663
Response Calculation
The response evoked from a cell by a given stimulus was measured as the
median net average firing rate, defined as the median of the firing rates for
all presentations of that stimulus minus the spontaneous activity of that
cell in the same test. The firing rate for each presentation was taken as the
average discharge rate over a time period equal to the stimulus duration
but beginning 50 ms after the stimulus onset, to allow for the response
latency of the cell. The spontaneous activity was calculated from the
average discharge rates during the 250 ms period immediately preceding
each stimulus presentation. For any given test, the spontaneous activity
was then taken to be the median activity of all such periods in that test.
Responses calculated in this manner could be negative, if the stimulus
suppressed the spontaneous activity.
Figure 1. Stimulus diagrams. (A) Two-dimensional position test, (B) summation test,
(C) SAT, (D) SMT. Arrows indicate moving dots; dots in this diagram represent the
stationary dots which remained on the screen at all times; lines delineating borders of
stimuli were not present in the actual stimuli. SAT and SMT involve the presentation of
two stimuli, one fixed in the center, and one located peripherally.
two random dot patches simultaneously, one for ERF stimulation, the
other for activation of the antagonistic surround. In these two tests, the
dot motion in both the center and surround patches matched the
preferred direction and speed of the cell.
The first of these two tests was the Surround Asymmetry Test (SAT)
in which the cell was driven by a circular stimulus patch centered on the
ERF center, while a second circular patch was presented in one of eight
positions surrounding the center patch at 45° intervals (Fig. 1C). The
stimulus in the center was identical in every respect to the stimulus of
optimal diameter, that which had elicited the maximum response in the
summation test. The second, peripheral stimulus extended from the edge
of the display almost to the edge of the center stimulus, leaving a 1° gap
between. Thus, the two stimuli were clearly segregated, and as much of
the radial extent of the surround as possible was covered by the
peripheral stimuli. Two control conditions were also included, one of
which consisted of the center stimulus alone, while the other was the
full-screen, 25.6° stimulus centered on the ERF. All but five of the neurons
were tested at the standard distance of 0.57 m, the others at 0.28 m. With
the SAT, we were thus able to study the angular distribution of the
antagonistic surround. Since this was one of the principal tests in these
experiments, a somewhat higher number, at least seven presentations, of
all stimulus conditions were shown [median number of trials = 11, first
quartile (Q1) = 10, third quartile (Q3) = 11].
The second test was the Surround Mapping test (SMT), which
utilized the same 25-position grid as the P2D test. In this test, however,
the cell was always stimulated by a 5 ×5° stimulus square presented in the
central position (Fig. 1D), while a second stimulus square of the same size
was simultaneously presented in one of the remaining 24 positions. One
neuron was, however, tested at a distance of 0.28 m, using 10 × 10°
squares. The central stimulus was also presented alone, to serve as a
control. The median number of presentations in this second principal test
was nine trials (Q1 = 9, Q3 = 11). This test complemented the SAT, not
only because it mapped the surround in a more detailed fashion, but also
because it made it possible to study both the angular and radial
distribution of the surround. The advantage of the SAT over the SMT is
that the center stimulus is generally more nearly optimal in size and that
the larger size of the peripheral stimuli has the capacity to summon the
surround inf luences more effectively. In this way, the two tests, though
similar in concept, are complementary in terms of the information they
provide.
664 Spatial Distribution of Antagonistic Surround • Xiao et al.
Determination of Antagonistic Surround Strength
The strength of the antagonistic surround was quantified by the
difference in the response to center stimulus alone and the response to
center and surround stimuli combined. This difference was expressed as
a percentage of the response to the center stimulus alone according to the
formula S = 100 × (1 – Rcs/Rc), where S is the strength of the antagonistic
surround elicited by a particular surround stimulus, Rcs is the response to
the combination of the center and surround stimuli, and Rc is the
response to the center stimulus alone. Positive values indicate inhibitory
effects, with values >100 indicative of inhibition of spontaneous activity.
Negative values indicate facilitatory effects. To visualize the spatial
distribution of the antagonistic surround, the surround inf luence revealed
by the SAT was plotted in polar coordinates (Fig. 3C), and that revealed by
the SMT was depicted as an isoinhibitory contour map in a manner
analogous to the ERF map (Fig. 3D) except that here the inhibition was
not normalized with respect to the maximum inhibition, but rather is
expressed as a percentage of the center response.
Quantification and Classification of the Angular Distribution of
the Antagonistic Surround
To quantify the angular distribution of the antagonistic surround, two
selectivity indices (SIs) were computed (Vogels and Orban, 1994) for each
cell from the levels of surround antagonism generated by each of the eight
surround positions in the SAT, using a formula, based on circular
statistics, for calculating the length of the mean vector:
SI =
LM n Si ⋅ sinbαigOP2 + LM n Si ⋅ cosbαigOP2
MNi∑=1
PQ MNi∑=1
PQ
n
∑ Si
i =1
where n is the number of surround positions tested, S is the surround
inhibition elicited by the stimulus located at position i, and αi is the angle
between the direction of the stimulus motion and the line connecting the
ERF center to the center of the surround stimulus. In cases where the
surround antagonism was negative at one or more positions, the
minimum value was subtracted from the eight surround values before the
SI was calculated. We termed the first of the two SIs the Unimodal
Selectivity Index (USI), which was calculated from the actual αi values.
The USI is a measure of the degree of unimodality of the surround
antagonism, i.e. the tendency for the surround antagonism to be concentrated on one side of the RF. The second SI, the Bimodal Selectivity Index
(BSI), was calculated with αi values doubled, and ref lects the degree of
bimodal distribution, or the tendency for the surround antagonism to be
concentrated along an axis, on opposite sides of the RF. All SI calculations
were corrected for grouping (Batschelet, 1981). An SI value of 1 indicates
that only a single position (USI), or two positions along an axis (BSI), were
effective in modulating the neuron’s activity, and an SI value of 0 indicates
a uniform angular distribution of the surround antagonism.
The Rayleigh test (Batschelet, 1981), which tests the null hypothesis
that the data are distributed uniformly, and the two SIs were employed for
classifying the angular distribution of an antagonistic surround into one of
the three categories: asymmetric (Rayleigh test P < 0.05 and USI > BSI),
bilaterally symmetric (Rayleigh test P < 0.05 and BSI > USI) and circularly
symmetric (Rayleigh test P ≥ 0.05).
Figure 2. Two-dimensional position tests for neuron 7721, one obtained with 5.0 × 5.0° squares (A,B), and one obtained with 2.5 × 2.5° squares (C,D). Here and in subsequent
figures, columns of stimulus positions are indicated with numbers, rows by lowercase letters. (A) and (C) show peristimulus time histograms (PSTHs) for the respective tests, (B)
and (D) the ERF contour maps for these tests. In (A) and (C), horizontal bars indicate stimulus duration of 320 ms, vertical bar indicates 50 spikes/s in (A) and 100 spikes/s in (C).
Response levels in (B) and (D), relative to the respective maxima, are shown by gray levels indicated in the scale under (D); circle indicates center of mass of the ERF; cross indicates
the center of the display. Size scales are indicated on the right of the ERF maps. This neuron was recorded in layer IIIc, and its RF was located at 12.5° from the fixation point. The
estimated size of the ERF was similar using either stimulus size, but the centering could be assessed much more precisely using the smaller stimulus size. The monitor was moved
0.75° up following the test shown in (D).
To indicate the angular location of the most effective inhibitory region
of the surround, an optimal angle (OPA) for each cell was determined
from the inhibitory inf luence of the eight surround positions using the
formula for calculating the mean vector angle (Batschelet, 1981).
F n Si ⋅ sinbαig I
JJ
G∑
OPA = arctanG i =1
GG n Si ⋅ cos αi JJ
GH i∑=1 b g JK
where n is the number of surround positions tested, S is the surround
strength elicited by the surround stimulus located at position i, and αi is
the relative angle between the direction of the stimulus motion and
position i. For cells that have an asymmetric surround, the OPA is the
angle between the direction of motion and the position where inhibition
was maximal. For cells that have a bilaterally symmetric surround, the
OPA represents the angle of the axis through the two optimal surround
positions, relative to the direction of stimulus motion. For cells that have
a circularly symmetric surround, the OPA is not defined.
Analysis of SMT Data
Responses at each grid position were converted into percentage
inhibition or facilitation using the center response as the reference.
Values between neighboring grid positions were linearly interpolated and
connected with isoresponse contours drawn at intervals of 12.5%. The
resulting contour map or SMT map provides a graphic picture of the
surround distribution.
These SMT data were also fitted with mathematical functions
representing the three spatial distributions of the surround inf luences.
The appropriate model was chosen based on the appearance of the SMT
map, which in most cases was identical to the type of symmetry
determined by the SAT. In four cases, the initial choice of model proved
inappropriate, and a second, successful fitting was carried out using one
of the remaining two models. An asymmetric surround was modeled with
a generalized two-dimensional Gaussian, using the function described in
Raiguel et al. (1995). The free parameters were radial and angular
position of maximum inhibition, and the elongation and orientation of
the Gaussian as well as its height. The bilaterally symmetric surround was
modeled with a pair of circularly symmetric, two-dimensional generalized
Gaussians of equal diameter and height. The two curves were constrained
such that the placement of one was radially symmetric with respect to the
other about the ERF center, but they could otherwise be positioned at any
angle and at any distance from the center. Free parameters were thus the
angular position, radial distance and height of the pair of Gaussians. The
circularly symmetric surround was modeled as the difference of two
Cerebral Cortex Oct/Nov 1997, V 7 N 7 665
Figure 3. The result of the four tests: two-dimensional position tests (A), summation test (B), SAT test (C) and SMT test (D) for neuron 7916, recorded in layer IV and with an RF
eccentricity of 6.4°. The ERF map is shown in (A). Response levels as in Figure 2(B,D), optimal diameter (black circle) and CRF (black square) are drawn to scale and in their actual
positions. Angular positions are indicated, in 90° steps, relative to the preferred direction of motion (arrow). Half-height diameter was 8.0°, oblateness index 15, ERF orientation 125°
(see Appendix). The summation curve is plotted in (B); histograms to left are PSTHs for three representative diameters. Vertical bar at lower right shows scale of histograms in (B)
and (D), and corresponds to 100 spikes/s. Horizontal bars below PSTH here and in (C) and (D) indicate stimulus duration of 200 ms. From the summation curve were derived the
optimal diameter (3.2°), the percentage suppression (100%), and the radial extent (5°). (C) Polar plot of inhibition evoked by peripheral SAT stimuli at different angular positions. PSTHs
along the axes of the polar plot show the responses to the combined stimulus of the center and one of the eight peripheral positions. Response to the full-screen stimulus and center
stimulus alone are shown in the lower left and lower right respectively. Calibration bar indicates 100 spikes/s. The USI and BSI of this neuron were 0.36 and 0.12 respectively. OPA
(arrow) was 112°. Full-screen inhibition and maximum SAT inhibition both equal to 100%. Finally, (D) shows the SMT map. Inhibition, as a percentage of the center response, is plotted
as gray levels (scale at bottom). Representative PSTHs to right show the responses at the positions indicated. White star indicates the position of the maximum inhibition according
to the model fitted to the SMT data. From the SMT, we obtained an angle of 109°, a surround distance of 5.8°, corresponding to 1.45 ERF radii, and a maximum inhibition of 88%.
Both SAT and SMT indicated an asymmetric surround located above the RF.
circularly symmetric, two-dimensional generalized Gaussians with
common origin at the RF center and fixed diameter ratio, producing a
semi-toroidal shape resembling a split bagel. Free parameters consisted of
the height and diameter of the component Gaussians. The algorithm
converged to a satisfactory fit in all but two of the 51 neurons tested.
From these models, we extracted the radial distance of maximum
inhibition and, for non-circularly symmetric surrounds, the angular
position of the maximum inhibition. This angular position relative to the
direction of motion is referred to as the SMT angle. To describe the radial
position of the antagonistic surround, we define the surround distance as
the radial distance of the maximum surround inhibition. When expressed
in units equal to the radius of the ERF, we refer to this quantity as the
relative surround distance. The ERF radius was derived using the
half-height radius of a generalized Gaussian function fitted to the
responses of the P2D test (Raiguel et al., 1995). To quantify the
elongation of the ERF for each cell, we computed an oblateness index
(OBI) equal to 100 × (L – S)/L. The quantities L and S are the lengths of the
longer and shorter axes of the ERF and were derived using the half-height
radius of the two-dimensional Gaussian function (Raiguel et al., 1995).
Results
Database
In total, we recorded from 161 MT/V5 neurons, 145 of which
666 Spatial Distribution of Antagonistic Surround • Xiao et al.
were tested with the SAT, and 51 of these were also tested with
the SMT. The remaining 16 cells were tested only with a
preliminary version of the SAT employing 16 pie-shaped
segments containing a random dot field. The present paper
reports only on the data obtained from the 145 cells tested with
the SAT, although results obtained from the 16 preliminary tests
were very similar to those reported here. In general, the neurons
responded vigorously to the center stimulus of the SAT (median
= 30 spikes/s, Q1 = 16, Q3 = 55), and showed little spontaneous
activity (median = 0 spikes/s, Q1 = 0, Q3 = 0). Since we
specifically selected neurons having strong antagonistic surrounds, the vast majority of the cells in our sample were located
in the superficial layers. The laminar distribution was as follows:
8 neurons in layer II, 10 in IIIa, 30 in IIIb, 51 in IIIc, 15 in IV, 26
in V and 5 in VI. The eccentricities of ERF centers ranged from
0.7 to 31° (median = 10°, Q1 = 6.5°, Q3 = 14.6°), thus half were
located within the central 10° of the visual field. Half of the cells
(74/145) preferred a stimulus motion speed of 10 or 20°/s, a
smaller proportion (31/145) preferred a slower speed (2.5 or
5°/s), and the rest (40/145) preferred a faster speed (40°/s).
The sizes of the ERFs, measured as the half-height diameter of
a generalized Gaussian function fitted to the responses of the
Figure 4. Different surround positions evoke very different levels of suppression. The
minimum inhibition elicited by the peripheral stimuli in the SAT of each cell was plotted
against the maximum inhibition generated by the same set of peripheral stimuli.
Inhibitions are expressed as a percent of the center response. Arrows indicate the
median values. The dashed oblique line shows a minimum/maximum ratio of 1:2, the
solid diagonal line indicates equal inhibition. Solid horizontal line indicates a minimum
inhibition equal to zero, the stippled horizontal line a minimum inhibition equal to 15%.
P2D test (Raiguel et al., 1995), ranged from 3 to 18° (median =
9.8°, Q1 = 7.5°, Q3 = 13.5°). This quantitative measurement
exceeded the diameter of the hand-plotted CRF by an average
factor of 2.35. Using a lower response level of 25% yields an even
larger ERF/CRF ratio of 3.33. The neurons tested exhibited
strong antagonistic surrounds: their response was reduced by an
average of 88% of the maximum (Q1= 69%, Q3 = 100%) when
stimulated with the full-screen stimulus. The optimal diameter
derived from the summation curve was also on average much
smaller than the ERF 50% diameter, in agreement with our earlier
observations (Raiguel et al., 1995). As a consequence, the
optimal diameter averaged only 1.38 times the CRF diameter.
Since the inner edges of the peripheral SAT stimuli were 1°
removed from the center stimulus of optimal diameter, these
inner edges showed a slight overlap with the CRF in only 12% of
the sample. The position of the outer edge of the peripheral SAT
stimuli was also a consideration, since it was necessary to cover
a maximal amount of the surround. In most neurons, the display
size was sufficient to cover the entire antagonistic surround.
Indeed, the summation curve reached a non-zero plateau (see
Figs 6B and 8B,D,E) in 73% of the neurons, indicating that the
surround had been completely recruited by the largest stimuli
used. The summation curve was still decreasing at the largest
diameter tested, indicating incomplete coverage of the surround
at the maximum stimulus size, in a mere 4% of the sample. In the
remaining 23%, the response fell to zero towards the upper end
of the summation curve (see Figs 3B, 5B, 8A,C). In these
neurons, it cannot be determined from the summation curves
whether the surround was completely included in the region of
visual space tested, since responses measured extracellularly
cannot decrease below zero. A substantial part of the surround
was most likely included, however, since the excitation from the
ERF was completely suppressed.
The Antagonistic Surround of Most MT/V5 Neurons
Shows Spatial Heterogeneity
The antagonistic surround of most MT/V5 neurons was not
equally effective in all directions, which is to say that the
surround inhibition depended on the angular position of
the peripheral SAT stimulus. This is illustrated by cell 7916 in
Figure 3. This neuron had a relatively small RF: its CRF measured
3.2° in diameter compared to 8.0° for the 50% ERF diameter
(Fig. 3A). The summation test revealed a very strong surround
with a suppression of 100% at the largest diameter and a
response reduced to 10% of maximum at a diameter as small as
12.8° (Fig. 3B). The optimal diameter, which also measured
3.2°, was identical to the CRF size. Since the inner edge of the
SAT peripheral stimuli were positioned 1° from the center,
optimal-diameter stimulus, none of the peripheral SAT stimuli
overlapped with the CRF. Yet different surround regions, though
always producing inhibition when stimulated, differed substantially in the degree of inhibition produced, which ranged
from 20 to 100% (Fig. 3C). Since the center stimulus remained
unchanged in the eight SAT conditions, variations in response
levels are dependent solely upon the surround region stimulated
by the peripheral stimuli.
In the majority of the cells investigated, the degree of position
dependence exhibited by the surround antagonism was quite
striking, since the maximum inhibition averaged 72% (n = 145),
while the average minimum was 0%. This is illustrated in Figure
4, which plots, for each neuron, the minimum inhibition elicited
by the eight SAT conditions as a function of the maximum
inhibition. The high degree of surround heterogeneity was
unexpected, given the view held by previous investigators
(Tanaka et al., 1986) that surround inf luence is equal in all
angular positions. In >60% (90/145) of the cells studied, at least
one position had no inhibitory effect (minimum inhibition
<15%). A quarter of the cells (37/145) even showed facilitation
(minimum inhibition <–15%), which in 41% (15/37) of these
cases exceeded 50% of the center response. Even in the 40% of
the cells (55/145) where all eight surround positions showed at
least 15% inhibition, the ratio of the maximum to minimum
inhibition was >2 in almost all cases (median ratio = 3, Q1 = 2.33,
Q3 = 4.5).
Different Angular Distributions of the Antagonistic
Surrounds in MT/V5 Cells
Figure 3C shows that the surround inf luence is not only
heterogeneous, but that the surround antagonism systematically
depends on the angular position of the surround stimuli.
Positions 45, 90 and 135° elicited the strongest inhibition, while
positions 270 and 315° evoked the least inhibition. This neuron’s
surround was thus characterized by a single maximum located
near 90° (OPA = 112°). The asymmetric nature of this surround
is ref lected by the USI of 0.36 (Rayleigh test P < 0.001) and the
BSI of 0.12, which are typical values for neurons with an
asymmetric surround. Such neurons have a statistically
significant Rayleigh test and a BSI that is less than the USI. More
than half the neurons tested (74/145) behaved in this manner.
Figure 5 illustrates a different sort of behavior. Neuron 7805
had a CRF measuring 4° in diameter and an ERF 50% diameter of
13° (Fig. 5A). Notice, however, that the ERF was elongated along
the 125° axis, a feature the hand plotting had identified fairly
well. The summation test again revealed a strong surround, since
the response fell to zero at maximum diameter (Fig. 5B).
However, the decrease in response rate was much more gradual
than that in Figure 3B. The optimal diameter of 6.4° was larger
than the CRF and hence the SAT peripheral stimuli again did not
encroach upon the CRF. As in the preceding example, all
surround positions produced inhibitory effects ranging from 20
Cerebral Cortex Oct/Nov 1997, V 7 N 7 667
Figure 5. The result of the four tests: two-dimensional position tests (A), summation test (B), SAT test (C) and SMT test (D) for neuron 7805, recorded in layer V and with an RF
eccentricity of 11.9°. Conventions as in Figure 4. Calibration bars in (C) and between (B) and (D) indicate 100 spikes/s. Horizontal bars below PSTH in (B–D) indicate stimulus duration
of 200 ms. Half-height ERF diameter in (A) was 13°, oblateness index 43, ERF orientation 90° (see Appendix). The optimal diameter in (B) was equal to 6.4°, the percentage
suppression 100%, and the radial extent 6.6°. (C) The USI and BSI of this neuron were 0.19 and 0.44 respectively. OPA (heavy line) was 25°. Full-screen inhibition and maximum SAT
inhibition both equal to 100%. The two stars in the SMT map (D) indicate the position of the inhibition maxima according to the model. From the SMT, we obtained an angle of 48°,
a surround distance of 9.5° corresponding to 1.5 ERF radii, and a maximum inhibition of 82%. Both SAT and SMT indicated a bilaterally symmmetric surround located along the upper
right–lower left diagonal.
to 90% inhibition (Fig. 5C). In this example, however, there were
two strongly inhibitory regions positioned on opposite sides of
the RF, located between 180 and 225° and between 0 and 45°.
Thus, the position dependence of the surround antagonism is in
this case characterized by a double maximum. This corresponds
to a significant Rayleigh test, a large BSI (0.44 for this neuron)
and a smaller USI (0.19 for this neuron) and typifies the
one-quarter (39/145) of the MT/V5 neurons that had bilaterally
symmetric surrounds.
A final group of neurons, amounting to a mere fifth of the
sample (32/145), behaved as one would expect from the existing
literature. In cell 6604, for example, the surround inf luence was
uniform (Fig. 6). Its CRF was again clearly smaller than the ERF,
whose 50% diameter was 8.2°. The ERF was fairly elongated
along the 173° axis, a feature also recognized by the hand plot.
The summation test (Fig. 6B) revealed a surround nearly as
strong as in the two previous examples. The optimal diameter
was 6.4° and here, too, the SAT peripheral stimuli did not
encroach upon the CRF. In this example, however, surround
inhibition remained between 37 and 60% at all positions (Fig.
6C), which is statistically indistinguishable from a uniform
distribution (Rayleigh test, P > 0.1). Accordingly, both the USI
and BSI were very small (0.02 and 0.07 respectively). Small USI
and BSI values and a negative Rayleigh test are characteristic of
neurons with circularly symmetric surrounds. The relative rarity
of MT/V5 neurons that display this sort of uniform surround
inf luence is consistent with Figure 4, which shows relatively few
neurons with minimum/maximum SAT inhibition ratios <2.
668 Spatial Distribution of Antagonistic Surround • Xiao et al.
A Continuum of Angular Surround Distributions
The neurons illustrated in Figures 3, 5 and 6 were chosen to
illustrate the three main types of surround distributions we
obser ved, but there is in fact a continuum of angular distributions with all possible intermediate conditions between the
three major distribution types. This is demonstrated by Figure 7
which plots the BSI as a function of the USI for all 145 neurons
tested with the SAT. Although the neurons with asymmetric
surrounds (triangles) are indeed located in the lower right
portion of the diagram, those with bilaterally symmetric
surrounds (rectangles) in the upper left and the those with
circularly symmetric surround (circles) in the lower left, it is
clear that there is a continuum in both the USI and BSI distributions. The boundary between heterogeneous and uniform
surrounds corresponds roughly to an SI of 0.15. The many
neurons lying close to this boundary are intermediate cases
between circularly symmetric and asymmetric or bilaterally
symmetric surrounds. Neurons with larger SIs and lying near the
diagonal represent intermediate cases between asymmetric and
bilaterally symmetric surrounds.
Figure 8 illustrates some of the neurons with intermediate
surround distributions. The neuron in Figure 8A represents the
case with the lowest USI (0.13) that was still classified as having
an asymmetric surround. The surround position directly above
the ERF evoked an 80% inhibition in this neuron, that directly
below, only 38%. Figure 8B illustrates a neuron with a bilaterally
symmetric surround but with a low BSI (0.16). Here again, the
maximum SAT inhibition exceeded the minimum by a factor 2.
Figure 6. The result of the two-dimensional position tests (A), summation test (B), SAT test (C) and SMT test (D) for neuron 6604, recorded in layer II and with an RF eccentricity
of 31°. Conventions as in Figures 4 and 6. Calibration bars indicate 100 spikes/s. Horizontal bars below PSTH indicate stimulus duration of 320 ms. Half-height ERF diameter in (A)
was 8.2°, oblateness index 45, ERF orientation 173° (see Appendix). The optimal diameter in (B) was equal to 6.4°, the percentage suppression 97% and the radial extent 7.8°. (C)
The USI and BSI of this neuron were 0.02 and 0.07 respectively. Full-screen inhibition was 87% and maximum SAT inhibition 67%. The dashed circle in the SMT map (D) indicates
the position of the inhibition maximum according to the model. From the SMT we obtained a surround distance of 5.8°, corresponding to 1.4 ERF radii, and a maximum inhibition of
55%. Both SAT and SMT indicated a circularly symmetric surround.
The neuron in Figure 8C had an asymmetric surround but shows
some tendency towards bilateral symmetry (compare with Fig.
3C). Finally, the last two neurons illustrate weak surrounds (68%
suppression in Fig. 8D and 55% suppression in Fig. 8E), but their
asymmetry (Fig. 8D) or bilateral symmetry (Fig. 8E) still remains
obvious.
Non-uniform Surround Distributions: Overlap with ERF
or Genuine?
It is clear from Figures 3, 5, 6 and 8 that the SAT peripheral
stimuli certainly did not overlap with the CRF. In most cases,
however, they did overlap to some extent with the ERF, as
pointed out in an earlier publication (Raiguel et al., 1995).
Where the ERF is circular, as in the neurons of Figure 8C–E, the
overlap will be equal in all positions, and thus cannot give rise to
surround heterogeneities. Where the ERF is elongated, as in
Figures 5 and 6, the overlap will vary according to the position of
the SAT peripheral stimulus, driving the neuron to different
degrees. This is illustrated by the ERF shown in Figure 9A, where
the outlines of the SAT peripheral stimuli depicted in the figure
show an overlap visibly greater in the positions at 0 and 180°
than those at 90 and 270°. Hence, the different degrees of
interaction between central and peripheral SAT stimuli might be
due to differences in the amount of overlap between the
peripheral SAT stimuli and the ERF rather than to a genuine
heterogeneity in the distribution of the inhibitory input
generating the surround. In some neurons like 6604 (Fig. 6) the
Figure 7. Plot of BSI as a function of USI for all 145 MT/V5 cells tested with the SAT.
Circles indicate circularly symmetric surround cells, triangles depict asymmetric
surround cells and squares represent bilaterally symmetric cells. Open symbols refer to
the cells shown in Figures 3, 5, 6 and 8. Arrows indicate medians.
Cerebral Cortex Oct/Nov 1997, V 7 N 7 669
Figure 8. Results of two-dimensional position tests (P2D), SMT tests, SAT tests and summation tests for neurons 6615 (A), 7720 (B), 6316 (C), 8212 (D) and 7921 (E). Same
conventions as in Figures 3, 5 and 6, except that PSTHs not shown, the gray-level scale for SMT maps is expanded to include facilitation (negative values), and in (E) the circle
indicating the optimum diameter is shown in white for visibility. These cells were recorded in layers IIIb, IIIb, IIIc, IV and VI, at eccentricities of 6.8, 19.8, 0.8, 10, and 12° respectively.
Values derived from these tests are given in the Appendix.
670 Spatial Distribution of Antagonistic Surround • Xiao et al.
Figure 9. Relationship between overlap of peripheral SAT stimuli with the ERF and their inhibitory effect. (A). Position of eight peripheral SAT stimuli with respect to the ERF, CRF and
optimal diameter of neuron 6604 (same cell as in Fig. 7). Same conventions as in Figure 4A. (B) Distribution of Spearman rank correlation coefficient between the inhibitory effect of
the peripheral SAT stimuli and their overlap with the ERF for the 145 MT/V5 neurons tested with the SAT. Asterisks indicate bins containing significant correlation coefficients. Median
correlation coefficient equaled –0.05. In (A), the SAT peripheral stimuli did not overlap at all with the CRF but stimuli at positions 0 and 180° overlapped more with the ERF than those
at 90 and 270°. Correlation coefficient of the overlap with the inhibitory effect was 0.27 (P > 0.5).
overlap does not appear to determine the distribution of
surround inf luences simply because the surround was circularly
symmetric; it may, however, in others, such as in neuron 7805
(Fig. 5), where the surround axis is more or less orthogonal to
the axis of the ERF.
To investigate this question as it applies to all cells, we
calculated the overlap of every SAT peripheral stimulus with the
ERF for each neuron by integrating the ERF profile over the areas
covered by the SAT stimuli, and then correlated, on a
neuron-by-neuron basis, the overlap with the inhibition evoked
by the peripheral SAT stimuli. If overlap with the ERF really
contributes to the variations in surround inf luence, one would
expect to see a significant negative correlation between overlap
and inhibition, since greater overlap should result in stronger
activation and hence less apparent inhibitory interaction. For
the neuron shown in Figure 9A, whose SAT is illustrated in
Figure 6C, the correlation (Spearman rank) was only 0.27 (P >
0.52). The absence of any significant correlation is not surprising
since the surround was uniform. The distribution of the
correlation coefficients is shown for all 145 neurons in Figure 9B.
Clearly, there is no systematic bias towards negative correlations.
The median (–0.05) was close to zero and there were only nine
neurons in which the negative correlation reached significance,
about the same number as those showing a positive correlation.
Thus heterogeneity in the surround inf luence did not ref lect the
differences in overlap with the ERF for the majority of MT/V5
neurons, and hence must ref lect differences in the spatial
distribution of inhibitory inf luence.
Surround Strength
Up to now we have dealt only with the angular distribution of
surround inf luence, but the SAT also provides information
concerning the amount of inhibition arising from each part of
the surround. We wondered whether the maximum inhibition
evoked by a single SAT stimulus was related to the global
inf luence of the surround as measured with the full-screen
stimulus. Therefore we plotted the maximum SAT inhibition as a
function of the full-screen inhibition (Fig. 10). For most cells, the
full-screen inhibition was stronger than the maximum SAT
Figure 10. The optimal surround position elicits almost as much inhibition as the entire
surround. The maximum inhibition elicited by the peripheral SAT stimuli plotted against
the inhibition induced by the full-screen stimuli (Spearman rank correlation ρ = 0.52, P
< 10–6). Circles, triangles and squares represent circularly symmetric, asymmetric and
bilaterally symmetric surround cells respectively. The circled star indicates 26 data
points at (100, 100), including 20 asymmetric, five bilaterally symmetric and one
circularly symmetric surround. The arrows along the axis indicate the median values for
each group: dark arrow, asymmetric surround; open arrows, circularly symmetric;
shaded arrow, bilaterally symmetric.
inhibition, since most points fall below the diagonal in Figure 10.
Yet the inhibition elicited at the optimum SAT position was fairly
strong in most cells, and the median maximum SAT inhibition of
71% was rather close to the median full-screen inhibition of 88%.
In fact, the maximum SAT inhibition equaled or exceeded that
elicited by the full-screen stimulus in nearly 40% (57/145) of the
cells tested, even though the surround area covered by an SAT
stimulus comprises only about one-fifth to one-tenth of the area
covered by the full-screen stimulus.
Cerebral Cortex Oct/Nov 1997, V 7 N 7 671
Table 1
Cortical laminar distribution of different type of surround cells
Layer
Asymmetric surround (%)
Bilaterally symmetric surround (%)
Circularly symmetric surround (%)
Total number of cells
Total
II
IIIa
IIIb
IIIc
IV
V
VI
37.5
50.0
12.5
8
20.0
20.0
60.0
10
46.7
30.0
23.3
30
47.0
25.5
27.5
51
53.4
33.3
13.3
15
77.0
11.5
11.5
26
40.0
60.0
0.0
5
51
27
22
145
Table 2
Eccentricity distribution of different type of surround cells
Eccentricity (degrees)
Figure 11. The average position tuning curves for the three types of surround spatial
distribution. To prepare the average position tuning curves, inhibitions elicited by the
eight peripheral stimuli of the SAT were normalized to the maximum for each cell and
assigned a relative position angle, which is the angle between the position of the
peripheral stimulus inducing the maximum inhibition and the positions of those eight
peripheral stimuli. The normalized inhibitions for all cells of a given surround type are
averaged at each position and plotted as a function of the relative position angle.
Vertical bars indicate the standard errors. Data were analyzed with split-plot ANOVA
using type and position as factors. The two main effects and their interaction were
significant: F(2) = 21.11, P < 10–6; F(6) = 2.56, P < 0.02; and F(2,6) = 6.9,
P < 10–6.
Figure 12. Relationship between angular position of the surround as determined by
the SAT (OPA) and by the model fitted to the SMT data (SMT angle) for 33 MT/V5
neurons with non-circularly symmetric surrounds and for which a fit was obtained. The
linear correlation cofficient r = 0.92 (P < 10–6 ); the equation of the regression line
was y = 0.85x +5.1.
Figure 8 suggests a correlation between the maximum SAT
inhibition and the full-screen inhibition. The two neurons with
weak surrounds (Fig. 8D,E) in the summation test, where the
largest diameter stimulus is in effect a full-screen stimulus, also
had weaker maximum inhibition in the SAT (see Appendix).
Figure 10 shows that there was a correlation (Spearman rank
coefficient = 0.52, P < 10–6) between the full-screen inhibition
and maximum SAT inhibition for the entire sample, although a
high degree of scatter is apparent.
672 Spatial Distribution of Antagonistic Surround • Xiao et al.
Asymmetric surround (%)
Bilaterally symmetric surround (%)
Circularly symmetric surround (%)
Total number of cells
0–5
5–10
10–15
>15
50.0
20.0
30.0
20
52.0
30.7
17.3
52
47.4
29.0
23.6
38
54.4
22.8
22.8
35
Properties of Surround Types
Since there was a continuum in surround distributions, we
wondered how distinct the three groups of surrounds, which we
had defined somewhat arbitrarily, actually were. A side-by-side
comparison of the angular distributions of their respective of
surround inf luences would be helpful in this regard. This was
accomplished by comparing the average surround inhibition
position-tuning curves for the three categories (Fig. 11). A
position-tuning curve plots the relative inhibitions elicited by the
eight SAT surround stimuli as a function of the angle of the
surround position, relative to the OPA, for each cell. An average
position-tuning curve is then the average of all such curves for a
given class. For the asymmetric surround class, the average
position-tuning curve has a single peak at the optimum position,
and the inhibition at all positions displaced 105° or more from
the OPA remains <50% of this peak. The average position-tuning
curve for the bilaterally symmetric group shows two distinct
peaks, one at the optimum position and the other at the position
opposite the optimum. Here, the inhibition induced by stimuli
displaced 90° from the OPA fell to 40% of the optimum. In
contrast to the curves described above, the average positiontuning curve for the circularly symmetric group is visibly f latter,
with the inhibition remaining >65% of the peak value at all
angular positions. Statistical analysis (split-plot ANOVA; 100%
column not included) indicated that these three curves differed
significantly from one another, since the interaction between
factors ‘group’ and ‘position’ was significant (see legend to Fig.
11).
The three surround types also differed in their laminar
distributions (Table 1). Although the weaker surrounds in layer
IV and VI (Raiguel et al., 1995) meant that relatively few cells
were sampled there, layer V differed significantly (χ2 = 7.28, P <
0.01) from other layers in its high proportion of neurons with
asymmetric surrounds. In the most superficial layers, laminae II
and IIIa combined, the opposite was true: the proportion of
asymmetric surround neurons was smaller than in all other
layers, a difference which came close to significance (χ2 = 0.45,
P < 0.07). This was complimented by an increase in the
proportion of circularly symmetric surround neurons, which
was especially pronounced in sublamina IIIa. The majority of
neurons (6/10) in this layer had uniform surrounds, making it
statistically distinct from all other layers (χ2 = 6.77, P < 0.01). The
Figure 13. Distribution of the Spearman rank correlation coeffiecient between the
excitatory response in the P2D test and the inhibitory effect in the SMT for 51 MT/V5
neurons tested with the SMT. Median value = 0.01. Asterisk indicates a bin with
significant values.
small number of neurons render any conclusions tentative, but it
is interesting that this layer also has the narrowest surrounds, as
measured in the radial direction (Raiguel et al., 1995). While
there were clear-cut laminar differences between the surround
groups, there was no relationship between the eccentricity of
the RF and the type of surround (Table 2).
Finally, there was a systematic difference in surround strength
among the three surround types. Both the maximum SAT
inhibition and the full-screen inhibition were stronger for the
asymmetric surround group than for the other two groups. Both
differences proved statistically significant (Mann–Whitney U z =
2.48, P < 0.02; z = 2.50, P < 0.01, for maximum and full-screen
inhibition respectively). There was no difference in the degree
of ERF elongation among the three surround types. However,
while there was no relationship between the OPA and ERF
orientation for asymmetric surrounds, the average angle
between the OPA and the ERF orientation was significantly
greater than 45° (χ2 = 5.45, P < 0.02) for the bilaterally symmetric
surrounds. Thus in neurons with bilaterally symmetric surrounds
the axis of maximal inhibition tends to be orthogonal to the ERF
axis, as was the case in Figures 5 and 8B.
Angular Distributions of Surround Inf luence: SMT Test
For 51 neurons we also studied the spatial distribution of the
antagonistic surround using the SMT to map its location and
extent. Overall, SAT and SMT tests produced similar results, and
the spatial layout of the surround inf luence as measured with
SMT generally matched the angular distribution type derived
from SAT. This well illustrated by the example cells in Figures 3,
5 and 6. The SAT revealed an asymmetric surround in neuron
7916, with a maximum at 112° (Fig. 3C). Accordingly, the SMT
revealed a single strong inhibitory region above the ERF (Fig.
3D). Neuron 7805 had a bilaterally symmetric surround in the
SAT (Fig. 5C), and the SMT revealed two inhibitory regions,
symmetrical with respect to the ERF, in the positions expected
on the basis of the SAT (Fig. 5D). Finally, neuron 6604 gave a
circularly symmetric surround in the SAT (Fig. 6C) and the SMT
indeed revealed a near-perfect annular surround (Fig. 6D).
It was the general rule that neurons (34/49) had similar
surround distributions in both the SAT and SMT (Table 3). This
association was highly significant (χ2 = 34.6, df = 4, P < 10–4). There
were, however, some exceptions as indicated in Table 3. This is
not totally surprising since there really was a continuum of
Figure 14. Surround distances (from ERF center) as determined by the SMT. (A) The
distribution of the surround distance in terms of ERF radii for 49 MT/V5 cells for which
a fit could be obtained with the appropriate surround model. (B) The surround extent
extracted from the summation test plotted as a function of the surround distance. Filled
symbols indicate the 38 neurons for which there was a significant linear correlation
(r = 0.73, P < 10–6). Dashed line represents the regression line fitted to the 38 data
points. Open symbols represent the four neurons for which the SMT model failed to
capture the surround fully. For all 42 neurons, the linear correlation was still significant
(r = 0.46, P < 0.002).
Table 3
Comparison of surround types in SAT and SMT
SAT
Asym.
Bilat.
Circ.
Total
SMT
Asym.
Bilat.
Circ.
Total
15
3
1
19
5
10
1
16
3
2
9
14
23
15
11
49
Asym. = asymmetric surround; Bilat. = bilaterally symmetric surround; Circ. = circularly
symmetric surround.
surround distributions in the SAT, and borderline cases may be
expected to fall on opposite sides of the criterion in different tests.
This is the case in some of the examples shown in Figure 8. The
neuron in Figure 8A was classified by the SAT as having an
asymmetric surround, but the USI was relatively small. Not
unexpectedly, the neuron had a surround map that was best fitted
with a circularly symmetric surround model. In a similar vein, the
neuron in Figure 8C was classified as having an asymmetric
surround in the SAT, yet the SMT map was best fitted with a
bilaterally symmetric surround model. Even in these instances of
disparate classifications there are elements of agreement between
the two tests: although the surround map in Figure 8C had two
inhibitory peaks, that in the lower right was markedly weaker and
the SAT revealed a small, secondary maximum in exactly the
position corresponding to the weaker SMT peak.
Cerebral Cortex Oct/Nov 1997, V 7 N 7 673
The SMT maps shown in Figures 3, 5, 6 and 8 illustrate the
quality of the fits made by the surround models to the surround
topography. In general, the fits were quite satisfactory, even if
the models employed were relatively simple. Given the generally
high quality of the fits, we used the position of the maximum
inhibition indicated by the models to compare the angular
position of the maximum surround inhibition in SAT and SMT.
There was an excellent correlation between the OPA and the
SMT angle (Fig. 12). Not only was there a strong correlation (r =
0.92, P < 10–6), but the regression line had a slope close to 1 and
passed through the origin. This strong correlation verifies the
impression given by a comparison of the SAT and the SMT maps
in Figures 3, 5 and 8.
While there is a high degree of correspondence between the
angular positions of the maximum inhibitions in the SAT and
SMT, the match is not as good for the amount of inhibition.
Although the average maximum inhibitions in the SAT and SMT
were very similar (median values = 75 and 73% respectively), the
correlation between the maximum inhibitions in these tests was
only 0.43 (Spearman rank, P < 0.002). This is probably due to the
fact that the center stimulus in SMT could be close to optimal in
some cells but was much less so in others. Also, the surround
inf luences may be recruited to different degrees in the SAT and
SMT because of the different sizes of peripheral stimuli used in
these tests.
Heterogeneity of Surround Inf luence: Overlap with ERF
or Genuine?
The SMT results allow us to test the supposition that the
surround heterogeneity ref lects different amounts of overlap
between the peripheral stimuli and the ERF, in an even more
stringent manner. Here we have a direct measurement of the
responses evoked by each peripheral stimulus in the P2D test,
and since many more positions were tested, the correlation
between excitatory potential arising from the portions of the
ERF encroached upon by the peripheral stimuli and the levels of
inhibitory interaction observed can be also calculated for many
more points. The distribution of the correlation coefficients
(Spearman rank) for the 51 neurons tested with the SMT is given
in Figure 13. It is obvious that in most cells there was no
correlation at all between degree of excitation of the ERF and the
level of inhibitory interactions in the SMT. The median
correlation coefficient was a mere 0.01 and only two neurons
had significant negative correlations. Thus in most cells, the
spatial distribution of surround inf luences did not ref lect the
overlap of the peripheral stimuli with the ERF.
Radial Distribution of Surround Inf luence
The SAT test gives no resolution in the radial direction, and
provides only information about the angular distribution of the
surround inf luence. The SMT, however, explored a fairly wide
area surrounding the CRF using a relatively small surround
stimulus probe. It can thus provide us with information about
the angular as well as the radial distribution of the antagonistic
surrounds of MT/V5 cells. One of the most important aspects of
the radial distribution is how far the most effectual part of the
surround is located from the ERF center, since this indicates the
spatial range over which the surround mechanism operates. We
measured the distance, expressed in units of the ERF radius,
from the RF center to the portion eliciting the strongest
inhibition in the SMT, as identified by the models fitted to the
SMT maps. This was possible in all three surround models. The
distribution of these relative surround distances is shown in
674 Spatial Distribution of Antagonistic Surround • Xiao et al.
Figure 14A. The median value is 1.5 ERF radii. In nearly
two-thirds of the cells, the radial surround distance was <2 radial
ERF units. In the examples shown in Figures 3, 5, 7 and 8, the
maximum inhibition is indeed relatively close to the RF center
(see Appendix). The relative surround distance measured
directly from the SMT matches rather well the radial distance of
the surround maximum extracted from modeling the summation
curve (the radial offset as defined by Raiguel et al., 1995).
Indeed the median relative surround distance was 1.5 in the
present study compared to a median radial offset of 1.44 in
Raiguel et al. (1995).
Given the similarity between surround distance as measured
by the SMT and surround distance extracted from the summation
curve at the population level, we wondered whether this
relationship would hold for individual neurons. We derived from
the summation curve the radius at which the suppression falls to
50% of its peak value (the inhibition extent as defined in Raiguel
et al., 1995). This corresponds relatively well to the surround
distance defined from modeling the SMT data. As shown in
Figure 14B there was a consistent relationship between the two
distances (r = 0.73, P < 10–6) in most cells (38/42) for which data
were available. The four neurons which failed to show this
relationship in fact exhibited larger surround distances in the
summation test than anticipated from the position of the
maximum inhibition in SMT. These neurons had weaker
inhibitory regions lying at the edge of the area explored with our
SMT test, in addition to the main inhibitory regions fitted by the
surround models. Thus these surrounds had weak, distal tails,
explaining how surround inf luences continued to summate up
to the largest diameters in the summation test.
Discussion
Heterogeneity of the Antagonistic Surround
The main finding of the present study is the heterogeneity that is
present in the antagonistic surround of MT/V5 neurons. In nearly
80% of the MT/V5 neurons, the angular distribution of the
surround inf luence was not uniform, as one would expect if the
surround really encircled the CRF. This proportion is based on
the Rayleigh test which tests the angular distribution of SAT data
for departures from a uniform distribution. The results from the
SAT test are supported by those of the SMT test which generally
detected a heterogeneous surround in the same neurons where
the SAT indicated surround heterogeneity (Table 3). These
results are the more significant insofar as in most neurons we
explored the entire antagonistic surround. Indeed, in 73% of the
neurons, we had positive evidence from the summation curve
that the surround completely recruited by the largest stimuli.
Only in 4% of the neurons was there definitive evidence to the
contrary. Thus in most neurons the peripheral SAT stimuli
probed virtually the entire radial extent of the antagonistic
surround. Their outer edge generally extended beyond the outer
border of the surround and their inner edge came as close as 1°
from the inner border of the surround, indicated by the optimal
diameter in the summation curve.
It could be argued that surround heterogeneity might arise
entirely as a result of overlap between the peripheral SAT or SMT
stimuli and the excitatory RF. Different degrees of overlap could
produce different excitation levels which could then explain the
varying degrees of inhibitory interaction observed in the SAT or
SMT. In the vast majority of neurons, there was no overlap
between the peripheral stimuli of these tests and the CRF. There
was, however, some degree of overlap in most neurons between
the peripheral stimuli and the ERF. Hence, we need to exclude
the possibility that overlap is the source of the observed
heterogeneities in the surround inf luence. For the SMT data we
could do so by comparing the excitatory effect of a given grid
position in the P2D with its inhibitory interaction with the
center stimulus in SMT. In most (95%) of the neurons there was
no significant relationship between excitatory inf luence and the
degree of inhibitory interaction (Fig. 13). The SAT data point
towards exactly the same conclusions. Although in this case the
activation by the peripheral SAT stimuli could only be estimated
from the P2D, it also failed to correlate with the strength of
inhibitory interactions in most (91%) neurons. The absence of
such correlations does not support the view that surround
inhibitory inf luences are distributed around the CRF equally and
that the surround effect is unequal simply because they must
contend against unequal amounts of excitation. Thus it must be
the inhibitory surround inf luences themselves that are unequally
distributed around the CRF in the majority of MT/V5 neurons.
The tendency for the two surround regions to f lank the
elongated ERF in bilaterally symmetric surround neurons then
simply indicates that some degree of interaction exists between
the mechanisms specifying the ERF and the surround angular
distributions.
At first glance our findings appear to be in disagreement with
those of Tanaka et al. (1986), who compared the inhibitory
effects of two pairs of 90°-wide wedges positioned in the
surround either along the axis of the preferred direction of
motion or orthogonal to it. They observed no difference in the
degree of surround inhibition between the two orientations
using this broad surround stimulation. The relatively small size of
some of the surround regions and the absence of any consistent
relationship between the position of maximum surround
inf luence and the preferred direction of motion may, in part,
explain these negative findings. An additional factor might be
the observation (Xiao et al., 1995) that, in some neurons, weak
surround inf luences on opposite sides of the CRF might
summate comparatively well to produce a reasonably uniform
inhibition for opposite pairs of surround stimuli.
Three Groups of Antagonistic Surrounds
We divided the neurons into three broad categories: those
with asymmetric surrounds, those with bilaterally symmetric
surrounds and those with circularly symmetric surrounds.
Although this division may be to some extent arbitrary there are
indeed many neurons in which the surround inf luence is
concentrated in one or two restricted regions. It is also clear that
many intermediate cases occur as is shown by Figures 7 and 8.
Despite the fact that the angular surround distributions form
a continuum, the three groups have a number of distinct
properties; since their average angular distributions are
statistically dissimilar, they display differences in their laminar
distributions and surround strength, and relationships between
ERF orientation and positions of inhibitory regions differ.
The distinction between asymmetric surrounds and bilaterally
symmetric surrounds is reminiscent of the inhibitory end-zones
in the end-stopped neurons of primary visual cortex. Initially it
was argued that all neurons had two inhibitory end-zones but
that these could be of unequal strength (Orban et al., 1979). Yet
further work in the monkey clearly suggests that some V1
neurons have but a single end-zone (Peterhans, et al., 1987). The
analogy between end-zones and antagonistic surrounds is
appropriate here, since both regions are silent and are revealed
only through their interaction with the discharge regions or CRF
(Allman et al., 1985a,b). There is further similarity with the
simple cells in primary cortex insofar as the asymmetric and
bilaterally symmetric surrounds resemble S2 and S3 neurons
(Jones and Palmer, 1987) respectively. The major difference is
that the antagonistic regions in simple cells coincide almost
perfectly with excitatory inputs for the opposite contrast. This
is, of course, not possible in MT/V5, since all MT/V5 neurons
prefer the same direction of motion over the entire extent of the
ERF (Raiguel et al., 1995).
Angular and Radial Distribution of Surround Inf luences
The SAT and SMT tests both provide information about the
angular distribution of surround inf luences, but the SMT also
provides information about the radial distribution of these
inf luences. These results have some direct bearing on our efforts
to model the surround (Raiguel et al., 1995). That model
assumed that surround inf luences were uniformly distributed
around the CRF but stressed that this assumption was not
critical. In that publication we described the average surround as
a suppressive annulus whose maximum was located ∼1.5 times
the ERF radius from the RF center. In view of the present results
we have to add that in most neurons there is a marked degree of
modulation within that annulus, and that in half the neurons, the
effect reaches a maximum at a single point and gradually falls to
zero toward the opposite side of the CRF. In another quarter of
the neurons, the inf luence is near-maximum at two points on
the annulus at opposite sides of the CRF, and weaker in between.
The SMT data show that the maximum surround inf luences were
located at a distance from the RF center averaging 1.5 times the
ERF radius. This is in relatively good agreement with the radial
offset of 1.44 ERF radii obtained in the modeling study (Raiguel
et al., 1995). Furthermore for most (38/42) individual neurons
the relative surround distance measured in the SMT correlated
with the extent of the surround extracted from the summation
test. In conclusion, in most (90%) neurons the antagonistic
surround is located relatively close to the ERF and can be
estimated rather well by modeling the summation curve.
It should be noted that since the CRF is much smaller than the
ERF, the surround would appear much larger if the measurement
unit were the CRF radius rather than the ERF radius as used in
our studies (present study and Raiguel et al., 1995). Hence there
is not necessarily any contradiction between our findings and
earlier reports that the surround extended far beyond the CRF
(Allman et al., 1985a,b; Tanaka et al., 1986), since the standard
used to measure the distance was the smaller CRF.
Functional Considerations
In most previous studies of MT/V5 properties, only the CRF of
MT/V5 neurons has been stimulated, yet when the CRF alone is
stimulated, the cell population appears relatively homogeneous:
all have a fairly large RF responding to light and dark contours
and are direction selective (Maunsell et al. 1983; Albright, 1984;
Lagae et al. 1993). Seen from this vantage point, these neurons
differ only with respect to their speed–response curves (Lagae et
al., 1993; Orban, 1997) which change with eccentricity, and in
their ability to resolve the aperture problem, meaning that some
neurons are pattern direction selective (Movshon et al., 1985)
while others are not. Taking the antagonistic surround into
consideration reveals a much richer pattern of diversity. At the
most fundamental level, some neurons have an antagonistic
surround while others lack surrounds, a characteristic related to
laminar position (Raiguel et al., 1995). Furthermore, the present
study has shown that there are three different types of surround
Cerebral Cortex Oct/Nov 1997, V 7 N 7 675
distributions, and this is again linked to laminar position. Thus
considering the surround in addition to the CRF yields a much
richer diversity of MT/V5 response patterns. The more that test
stimuli can yield a diversity of response patterns, the closer they
come to revealing the function of the area. For example, moving
light and dark bars, which may be interesting stimuli with which
to investigate V1, are much less suitable for studying area MT/V5.
Large moving random dot patterns stimulating the surround as
well as the CRF are capable of revealing more about the
functional diversity of MT/V5 neurons.
Initial studies (Allman et al., 1985a,b; Tanaka et al., 1986;
Born and Tootell, 1992), which assumed the surround was
circularly symmetric, proposed the idea that the role of the
antagonistic surround is figure–ground segregation. Amongst
other functions, this might be used to direct attention or to
preprocess signals for the extraction of kinetic boundary
orientations (Orban and Gulyás, 1988). It also has been
suggested that the surround could restrict processing to that of
the images of (small) objects as opposed to processing entire
scenes or surroundings (Tanaka et al., 1986). These roles should
now be relegated to the relatively small proportion of MT/V5
neurons with truly circular symmetric surrounds. The other two
types of surrounds in effect take the first and second directional
derivatives of the velocity distribution over the retina. The
first-order derivative, represented neurally by the asymmetric
surround neurons, could be used to extract information about
surface orientation in depth. Direct experimental evidence for
this view has been obtained by our group (Orban et al., 1996;
Xiao et al., 1997). Indeed, we have shown that MT/V5 neurons
are selective for the speed gradient direction corresponding to
the direction of tilt in depth specified by motion. The secondorder derivative, represented by the bilaterally symmetric
surround neurons, could be used to extract information about
surface curvature as specified by motion. These two functions
are not incompatible with the surround also having the more
general function of restricting processing to images of small
objects. Indeed in some cells the tuning for speed gradients was
apparent only for relatively small stimuli, and no response at all
was obtained for large stimuli, even when these were tilted in
depth (Xiao et al., 1997). Such neurons may correspond to those
described above as having distal surround tails. Finally it is worth
mentioning that second-order derivatives of the velocity
gradient, whether directional or not, can also be used to extract
heading direction. Thus bilaterally symmetric or circularly
symmetric surround neurons in area MT/V5 could contribute to
the functionality usually ascribed to MST neurons (Duffy and
Wurtz, 1995; Bradley et al., 1996).
In conclusion, the antagonistic surround of MT/V5 neurons is
an integral part of their RF and we would be in a far better
position to appreciate the function of neurons in area MT/V5 by
stimulating both surround and CRF, rather than by masking the
surround as has generally been done in the past.
Notes
The technical assistance of P. Kayenbergh, G. Meulemans and G.
Vanparrijs is gratefully acknowledged. We would also like to express our
gratitude to Janssens Pharmaceutica for their generosity in supplying the
Sufentanil used in these experiments. This work was supported by grants
from the Regional Ministry of Education (GOA 95–6), the CEC (Esprit
BR A Insight II), and the National Research Council (FGWO 9.0022.85)
Address correspondence to G.A. Orban, Laboratorium voor Neuro- en
Psychof ysiologie, Katholieke Universiteit Leuven, School of Medicine,
Campus Gasthuisberg, Herestraat 49, B-3000 Leuven, Belgium. E-mail
[email protected].
676 Spatial Distribution of Antagonistic Surround • Xiao et al.
References
Albright TD (1984) Direction and orientation selectivity of neurons in
visual area MT of the macaque. J Neurophysiol 52:1106–1130
Allman J, Miezin F, McGuinness E (1985a) Stimulus-specific responses
from beyond the classical receptive field: neurophysiological
mechanisms for local-glial comparison in visual neurons. Annu Rev
Neurosci 8:407–430.
Allman J, Miezin F, McGuinness E (1985b) Direction- and velocity-specific
resonses from beyond the classical receptive field in the middle
temporal visual area (MT). Perception 14:105–126.
Batschelet E (1981) Circular statistics in biology. London: Academic Press.
Bishop PO, Coombs JS, Henry GH (1971) Responses of visual contours:
spatio-temporal aspects of excitation in the receptive fields of simple
striate neurons. J Physiol 219:625–657.
Born RT, Tootell RBH (1992) Segregation of global and local motion
processing in the primate middle temporal visual area. Nature
357:497–499.
Buracas GT, Albright TD (1996) Contribution of area MT to perception of
three-dimensional shape: a comutational study. Vis Res 36:869–888.
Bradley DC, Maxwell M, Andersen R A, Banks MS, Shenoy KV (1996)
Mechanisms of heading perception in primate visual cortex. Science
273:1544–1547.
Desimone R, Ungerleider LG (1986) Multiple visual areas in the caudal
superior temporal sulcus of the macaque. J Comp Neurol
218:164–189.
Desimone R, Schein SJ, Moran J, Ungerleider LG (1985) Contour, color,
and shape analysis beyond the striate cortex. Vis Res 25:441–452.
Droulez J, Cornilleau-Pérès B (1990) Visual perception of surface
curvature. The species variation and its physiological implications.
Biol Cybern 62:211–224.
Duffy CJ, Wurtz RH (1995) Response of monkey MST neurons to optic
f low stimuli with shifted centers of motion. J Neurosci 15:5192–5208.
Ghose GM, Roe AW, Ts’o DY (1994) Features of functional organization
within primate V4. Soc Neurosci Abstr 20:840.
Jones JP, Palmer LA (1987) The two-dimensional spatial structure of
simple receptive fields in cat striate cortex. J Neurophysiol
58:1187–1211.
Koenderinck JJ, Van Doorn AJ (1992) Second order optic f low. J Opt Soc
Am 9:530–538.
Knierim JJ, Van Essen DC (1992) Neuronal responses to static texture
patterns in area V1 of the alert macaque monkey. J Neurophysiol
67:961–980.
Lagae L, Gulyás B, Raiguel S, Orban GA (1989) Laminar analysis of motion
information processing in macauqe V5. Brain Res 496:361–367.
Lagae L, Maes H, Raiguel S, Orban GA (1994) Responses of macaque STS
neurons to optic f low componenets: a comparison of areas MT and
MST. J Neurophysiol 71:1597–1626.
Lagae L, Raiguel S, Orban GA (1993) Speed and direction selectivity of
macaque middle temporal neurons. J Neurophysiol 69:19–39.
Maunsell JHR, Van Essen DC (1983) Functional properties of neurons in
middle temporal visual area of the macaque monkey. I. Selectivity for
stimulus direction, speed, and orientation. J Neurophysiol
49:1127–1147.
Movshon JA, Adelson EH, Gizzi MS, Newsome WT (1985) The analysis of
moving visual patterns. In Pattern recognition mechanisms (Gatass R,
Gross C, eds), pp 117–151. Vatican City: Pontifica Academia
Scientiarium.
Orban GA (1997) Visual processing in macaque area MT/V5 and its
satellites (MSTd and MSTv). In: Cerebral cortex, Vol 14, Extrastriate
cortex (Kaas J, Rockland K, eds). New York: Plenum Press.
Orban GA, Gulyás B, (1988) Image segregation by motion: cortical
mechanisms and implementation in neural networks. In: Neural
computers (Eckmiller R, v d Malburg Ch, eds), NATO ASI Series Vol
F41, pp 149–158.
Orban GA, Kato H, Bishop PO (1979) Dimensions and properties of
end-zone inhibitory areas in receptive fields of hypercomplex cells in
cat striate cortex. J Neurophysiol 42:833–849.
Orban GA, Gulyás B, Vogels R (1987), Inf luence of a moving textured
background on direction selectivity of cat striate neurons. J
Neurophysiol 57:1792–1812.
Orban GA, Xiao K-K, Marcar V, Raiguel S, (1996) Selectivity of macaque
MT neurons for direction of tilt in depth. Invest Opthalmol Vis Sci
37:S485.
Peterhans E, von der Heydt R (1987) The role of end-stopped receptive
fields in contour perception. In: New frontiers in brain research:
procededings of the 15th Göttingen neurobiology conference (Elsner
N, Creutzfeldt O, eds), p 29. Stuttgart: Thieme Verlag.
Raiguel S, Van Hulle MM, Xiao D-K, Marcar VL, Orban GA (1995) Shape
and spatial distribution of receptive fields and antagonistic motion
surrounds in the middle temporal area (V5) of the macaque. Eur J
Neurosci 7:2064–2082.
Schein SJ, Desimone R, (1987) Visual properties of neurons in area V4 of
the macaque: sensitivity to stimulus form. J Neurophysiol 57:835–868.
Sillito AM, Grieve KL, Jones HE, Cudeiro J, Davis J (1995) Visual cortical
mechanisms detecting focal orientation discontinuities. Nature
387:492–496.
Tanaka K, Hikosaka K, Saito H, Yukie M, Fukada Y, Iwai E (1986) Analysis
of local and wide-field movements in the superior temporal visual
areas of the macaque monkey. J Neurosci 6:134–144.
Van Essen JHR, Maunsell DC, Bixby JL (1981) The middle temporal visual
area in the Macaque: myeloarchitecture, connections, functional
properties and topographic organization. J Comp Neurol 199:
293–326.
Vogels R, Orban GA (1994) Activity of inferior temporal neurons during
orientation discrimination with successively presented gratings.
J Neurophysiol 4:1428–1451.
Xiao D-K, Marcar VL, Raiguel SE, Orban GA (1994) Does the surround
really surround the classical receptive field of macaque MT cells? Soc
Neurosci Abstr 20:773.
Xiao D-K, Raiguel S, Marcar V, Koenderink J, Orban GA (1995) Spatial
heterogeneity of inhibitory surrounds in the middle temporal visual
area. Proc Natl Acad Sci USA 92:11303–11306.
Xiao D-K, Marcar VL, Raiguel SE, Orban GA (1997) Selectivity of macaque
MT/V5 neurones for surface orientation in depth specified by motion.
Eur J Neurosci 9:956–964.
Zeki SM (1983) Color coding in the cerebral cortex: the reaction of cells
in monkey visual cortex to wavelengths and colors. J Neurosci
9:741–765.
Appendix
Test data for example cells
,
Cell number
7916
7805
6604
6615
7720
6316
8212
7921
Ecc
Layer
CRF diameter (deg)
6.4
IV
3.2
11.9
V
4.0
31.2
II
3.5
6.8
IIIb
2.2
19.8
IIIb
3.3
0.8
IIIc
4.0
10.0
IV
6.7
12.0
VI
5.1
P2D
50% ERF diameter (deg)
25% ERF diameter (deg)
ERF oblateness
ERF orientation (deg)
8.0
11.3
15.1
125
13.0
18.4
43.2
90
8.2
11.6
45.0
173
5.2
7.4
55.6
108
9.1
12.9
22.9
85
6.8
9.6
10.0
9.7
13.7
14.0
6.1
8.6
9.7
SUM
Optimum diameter (deg)
Radial extent (deg)
Max stimulus inhibition (%)
3.2
5
100
6.4
6.6
100
6.4
7.8
97
3.2
3
98
9.6
9.6
68
6.4
11
100
6.4
9
68
9.6
7.2
55
SAT
USI
BSI
OPA (deg)
Full-screen inhibition (%)
Max position inhibition (%)
0.36
0.12
112
100
100
0.19
0.44
25
100
100
0.02
0.07
0.13
0.05
87
67
100
90
0.11
0.16
123
74
75
0.37
0.18
156
100
67
0.24
0.15
118
71
52
0.35
0.58
68
57
43
SMT
Surround distance (deg)
Fit type
Angle (deg)
Maximum inhibition (%)
5.8
asym
109
88
9.5
bilat
48
82
5.8
circ
6.3
circ
55
79
11
bilat
140
89
6.1
bilat
160
88
9.3
asym
98
69
5.8
bilat
56
50
,
,
Cerebral Cortex Oct/Nov 1997, V 7 N 7 677