© 2001 Nature Publishing Group http://neurosci.nature.com
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
The spatial transformation of color
in the primary visual cortex of the
macaque monkey
Elizabeth N. Johnson, Michael J. Hawken and Robert Shapley
Center for Neural Science, New York University, 4 Washington Place, New York, New York 10003, USA
Correspondence should be addressed to E.N.J. ([email protected])
Perceptually, color is used to discriminate objects by hue and to identify color boundaries. The
primate retina and the lateral geniculate nucleus (LGN) have cell populations sensitive to color modulation, but the role of the primary visual cortex (V1) in color signal processing is uncertain. We reevaluated color processing in V1 by studying single-neuron responses to luminance and to
equiluminant color patterns equated for cone contrast. Many neurons respond robustly to both
equiluminant color and luminance modulation (color-luminance cells). Also, there are neurons that
prefer luminance (luminance cells), and a few neurons that prefer color (color cells). Surprisingly,
most color-luminance cells are spatial-frequency tuned, with approximately equal selectivity for
chromatic and achromatic patterns. Therefore, V1 retains the color sensitivity provided by the LGN,
and adds spatial selectivity for color boundaries.
Color perception requires elaborate neural computations. The
brain constructs a color signal to recover the reflective properties of
the surfaces it looks at, independent of the vagaries of illumination and the colors of the surroundings. However, it is not known
how this computation is done in the brain. For instance, it is still
under investigation whether the cerebral cortex calculates the color
of a surface in a strictly modular fashion1, independent of other
stimulus properties, or whether the calculation of color is linked,
inextricably, with scene organization and spatial structure2,3.
What is the responsiveness of neurons in V1 to color stimuli? A
number of studies were directed at answering this question4–7, but
there is still substantial disagreement. A pioneering study of
macaque V1 reported that few cells were sensitive to chromatic
stimuli and that most preferred achromatic bars or edges8. A later
study also found few cells in macaque V1 that responded to equiluminant color stimuli as well as they did to luminance7. Conversely,
other investigators claimed that a significant fraction of V1 cells
were responsive to stimuli defined purely by color differences with
their backgrounds5,9. Consistent with these latter studies on monkey cortex, a functional magnetic resonance imaging (fMRI) study
indicates a fairly high sensitivity to equiluminant chromatic stimuli
in human V1 (ref. 10). In our experiments on macaque V1, we used
equiluminant and luminance stimuli that were equated in their
cone contrast; they were roughly equivalent in their power to drive
the magnocellular and the parvocellular input to the cortex. We
found that many neurons were as responsive to equiluminant color
stimuli as they were to luminance-defined stimuli.
Another major question is the spatial specificity of the
response of V1 cells to color. Threshold psychophysics of color
detection reveals that for equiluminant color stimuli, the largest
targets, or the grating patterns of the lowest spatial frequencies,
are most detectable11–13. This has led to the expectation that color
perception is mediated by neuronal mechanisms that are large
in spatial dimension, or that prefer very low spatial frequency.
nature neuroscience • volume 4 no 4 • april 2001
Indeed, neurons in the macaque LGN are spatially low pass to
equiluminant patterns13,14. However, there are psychophysical
experiments that indicate the presence of color channels that are
highly tuned for spatial frequency15. Furthermore, perceptual
experiments on color induction and the importance of color
boundaries for color appearance16,17 imply that color mechanisms are spatially selective. The second goal of our experiments
was to study the spatial selectivity of V1 neurons in response to
achromatic luminance and chromatic equiluminant stimuli. Most
V1 neurons we identified as being responsive to color were highly selective for spatial pattern, with approximately equal selectivity for chromatic and achromatic stimuli. These color-selective
neurons in V1 spatially transform the neural image delivered to
them by the LGN. These results indicate that V1 neurons could
have a major role in the visual perception of color, and that in
V1, the neuronal analysis of color is linked to the analysis of form.
RESULTS
Spatial frequency response functions for 167 cells in 25 monkeys
were measured by stimulating each cell with drifting sinusoidal
gratings modulated in black/white luminance or red/green equiluminant contrast, as well as with gratings modulated in the L-,
M- and S-cone isolating directions. Cone-isolating stimuli selectively modulate one class of cones while keeping the responses
of the other two constant. It is thereby possible to measure the
relative cone input and the spatial properties of each type of cone
input for each cell. Each grating was equated approximately for
cone contrast. The cells were classified into three groups based
on the ratio of their peak responses to luminance and equiluminant gratings, termed the sensitivity index.
Sensitivity index =
responsemax (equiluminance)
response max (luminance)
409
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
F1 amplitude (spikes/s)
60
40
Luminance
Equiluminance
30
40
20
20
20
10
0
b
e
c
0
0
d
100
Response (spikes/s)
f
40
60
80
40
20
40
20
0
.3
1
3
Fig. 1. Spatial frequency responses for six V1 neurons
to black/white luminance and red/green equiluminance. Response magnitude is the first harmonic (F1)
amplitude for simple cells and the mean rate for complex cells. (a, b) Spatial frequency tuning for cells that
respond predominantly to luminance. (a) A simple
cell from layer 6. (b) A complex cell from layer 4Cα.
(c, d) Spatial frequency tuning for cells that respond
comparably to both luminance and equiluminance.
(c) A simple cell from layer 2/3. (d) A complex cell
from layer 4B. (e, f) Spatial frequency tuning for cells
that respond preferentially to equiluminance. (e) A
simple cell from layer 5. (f) A complex cell from
layer 2/3. The luminance and equiluminance stimuli
were equated for contrast. The horizontal dashed
lines indicate two standard deviations above and
below the mean spontaneous firing rate. The smooth
curves were fit to the data using a difference of
Gaussians described in Methods.
0
0
.1
.1
.3
1
3
.1
.3
1
3
Spatial frequency
(cycles per degree)
The peak response was estimated from a difference of Gaussians fit
to each of the spatial tuning functions. Some cells showed little
or no response to equiluminance, but responded well to luminance (Fig. 1a and b), and thus had low sensitivity indices
(index < 0.5; luminance group, Fig. 2, n = 100). Other cells gave
comparable responses to equiluminant and luminance gratings
(Fig. 1c and d), having sensitivity indices of about one
(0.5 ≤ index ≤ 2; color-luminance group, Fig. 2, n = 48). The last
group of cells gave large responses to equiluminant gratings and
little or no response to luminance gratings, yielding sensitivity
indexes greater than 2 (Fig. 1e and f; color group, Fig. 2, n = 19).
Although the cells are distributed continuously in this sensitivity
index, we put the cells into groups to analyze the spatial tuning
properties of the cells that gave comparable responses to both
luminance and equiluminant stimuli.
More than half the neurons sampled (60%) gave preferential
responses to luminance, and thus had a low index (Fig. 2). Very
few of these low-index cells were low-pass in spatial tuning (5%;
5/100; see Methods). Conversely, the cells that responded primarily to chromatic equiluminance (high-index cells) gave mostly low-pass spatial tuning responses (74%; 14/19). In terms of
spatial selectivity, these neurons were similar to parvocellular
neurons in the LGN13,14. A few of these high-index cells were
selective for spatial frequency (Fig. 1f). The majority of the population of neurons responding to both luminance and equiluminance, with a sensitivity index approximately equal to one,
was spatial frequency selective (83%; 40/48). Few of these colorluminance cells gave low-pass responses to either color (2/48) or
luminance (2/48), while giving a spatially selective response to
Fig. 2. Ratio of equiluminance response to luminance response (sensitivity ratio) in simple and complex cells across cortical layers. Cells with
ratios less than 0.5 were classified as luminance cells, cells with ratios
greater than 2 were classified as color cells, and cells with ratios between
0.5 and 2 (between the dashed lines) were classified as cells responding
to both luminance and equiluminance (color-luminance cells). Large symbols are neurons with spatially low-pass tuning to equiluminance.
Cortical layering is based on histological reconstruction of the electrode
penetration and is represented here as the normalized cortical depth
along with the anatomical label18,19,44. For some neurons, cortical layering could not be determined. These data points are not shown.
410
the other stimulus type, and a few gave low-pass responses to
both color and luminance (8%; 4/48). The color-luminance neurons have the greatest potential for carrying both spatially and
chromatically opponent information.
In addition to analyzing the electrophysiological responses of
the neurons, we studied their anatomical location. In macaque V1,
functional anatomy is seen most readily in the grouping of cells into
either distinct columns or layers. A large number of connections in
primate V1 spread laterally18. Many of these horizontal connections are confined to a given layer, providing strong evidence for a
functional organization of the visual cortex into specific layers. In the
current study, cells were assigned a cortical depth and layer by histological reconstruction of the electrode track (see Methods)19. The
largest fraction of color-luminance cells was found in layers 2/3
(Fig. 3a). Cells that responded preferentially to color were found in
layers 2/3, 5 and 6. A preponderance of cells responded preferentially to luminance in layer 4B, and the majority of cells in layer 4Cα
and layer 6 were dominated by luminance (Fig. 3b).
Colorluminance
Luminance
0
Color
Simple
Complex
n = 167
2/3
Normalized cortical depth (%)
© 2001 Nature Publishing Group http://neurosci.nature.com
a
4A
50
4B
4Cα
4Cβ
5
6
100
0
0.015
0.06
0.25
1
4
16
64
Sensitivity index
nature neuroscience • volume 4 no 4 • april 2001
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
Percent luminance cells
30
Luminance cells (n = 100)
b
100
80
25
2/3 (n = 28)
60
40
20
20
100
10
80
4B (n = 32)
60
5
40
20
2/3 4A 4B 4Cα 4Cβ 5
6 Unknown
0
100
30
80
Color cells (n = 19)
4Cα (n = 23)
60
25
40
20
20
15
100
0
Percent cells
Percent color cells
Fig. 3. Laminar distribution of neurons in each classification.
(a) Percentage of neurons in each classification in each cortical layer. (b) Percentage of neurons in each layer in each classification. The percentage of neurons in layer 4A is not
shown in (b) because only a small number of cells were studied in this layer (n = 3); all were classified as luminance cells.
0
15
0
10
5
0
2/3 4A 4B 4Cα 4Cβ 5
6 Unknown
80
4Cβ (n = 11)
60
40
20
0
color-luminance group (1.95 ± 0.70 octaves, n = 87),
although these bandwidth distributions were not significantly different (luminance bandwidth versus
color-luminance luminance bandwidth, p > 0.1; luminance bandwidth versus color-luminance equiluminance bandwidth, p > 0.1). Moreover, within each
color-luminance cell, the spatial frequency preferences
for the achromatic and the color stimuli were not significantly different (Student’s paired t-test, p > 0.1),
nor were the sf bandwidths significantly different within these cells (p > 0.1). For each neuron, as an index of
the difference between its spatial frequency preferences for color and for luminance, we calculated the
following ratio.
100
Percent color-luminance cells
© 2001 Nature Publishing Group http://neurosci.nature.com
a
30
80
Color-luminance cells (n = 48)
25
20
5 (n = 18)
(preferred luminance sf – preferred equiluminance sf)
(preferred luminance sf + preferred equiluminance sf)
60
40
20
0
The distribution of this index shows that the preferred
spatial frequency for color and luminance was
80
10
not different; the mean of the index’s distribution was
60
–0.01 ± 0.2, with a very small skewness of 0.09. The
40
5
bandwidths for color and for luminance were com20
pared by calculating directly the difference between
0
0
2/3 4A 4B 4Cα 4Cβ 5
6 Unknown
the luminance and equiluminance bandwidths. The
100
Cortical layer
Unknown (n = 20)
mean of the distribution of bandwidth differences
80
60
was 0.14 ± 0.75, with a distribution skew of –0.2.
40
These analyses of the similarities of spatial properties
20
for color and luminance stimuli suggest that V1 is
0
capable of discriminating the spatial properties of
Lum Color Color-lum
color as well as luminance patterns with approximately similar precision. In contrast, the high-index
cells that prefer chromatic stimuli exhibited mostly low-pass tunWe characterized the pattern responses of all cells by meaing to chromatic gratings. This suggests that color-preferring
suring their response as a function of spatial frequency. Spatial
cells in V1 may have a role in color perception, but they are not
frequency curves were fit empirically with a difference of Gauslikely to be the substrate for color boundary discrimination.
sians function (see Methods, Figs. 1 and 4) to capture the main
Most of the color-luminance neurons maintain their spatial
features of the curves: the spatial frequency preference and bandfrequency tuning characteristics for L- and M-cone-isolating
width. The spatial frequency responses were measured and fit
stimuli (Fig. 6a and b). Most of these neurons do not respond to
for all five color directions (black/white luminance, red/green
S-cone-isolating stimuli (Fig. 6c). Excluding cells with low-pass
equiluminance, L-, M- and S-cone isolating directions; Fig. 4).
tuning and cells with indeterminate bandwidth, mean sf bandFirst, we considered the spatial selectivity of the color-lumiwidth for luminance (2.05 ± 0.70 octaves, n = 35) was not signance class. Most of the 48 color-luminance neurons had spatial
nificantly different from mean sf bandwidth for any of the
frequency preferences and bandwidths that were similar for color
cone-isolating stimuli (mean L-cone-isolating sf bandwidth,
and luminance patterns (Fig. 5a and b). Excluding cells with low2.41 ± 1.05 octaves, n = 30, Student’s t-test, 0.05 < p < 0.1, Fig. 6a;
pass tuning, the mean preferred spatial frequency (sf) for lumimean M-cone-isolating sf bandwidth, 2.37 ± 1.05 octaves,
nance (mean ± s.d., 2.56 ± 1.26 cycles per degree,
n = 28, p > 0.1, Fig. 6b; mean S-cone-isolating sf bandwidth,
n = 42) was not significantly different from the mean preferred sf
2.10 ± 1.15 octaves, n = 7; p > 0.1, Fig. 6c). Within each colorfor equiluminance (2.73 ± 1.56 cycles per degree, n = 42; Student’s
luminance cell, the spatial frequency bandwidth in the achrot-test, p > 0.1). Similarly, excluding cells with low-pass tuning and
matic direction was not significantly different from the
cells with indeterminate bandwidth, the mean sf bandwidth for
bandwidth in each cone-isolating direction (Student’s paired
luminance (2.05 ± 0.70 octaves, n = 35) was not significantly dift-test, p > 0.1 for L-, M- and S-cone-isolating directions).
ferent from the mean sf bandwidth for equiluminance
Preferences for cone-isolating stimuli also resembled lumi(2.09 ± 0.94 octaves, n = 37; p > 0.1). Cells classified as preferring
nance preferences. Excluding cells with low-pass tuning, the mean
luminance had bandwidths marginally narrower than those in the
15
nature neuroscience • volume 4 no 4 • april 2001
100
6 (n = 32)
411
© 2001 Nature Publishing Group http://neurosci.nature.com
Luminance
Equiluminance
M-cone
L-cone
Fig. 4. Spatial frequency response to luminance gratings,
equiluminant gratings, L-cone, M-cone and S-cone-isolating gratings for a simple cell in layer 4Cα. The equiluminance/luminance response ratio was 0.58 for this neuron.
The smooth curves were fit to the data with a difference of
Gaussians as described in Methods.
S-cone
40
30
20
10
0
0.1
0.3
1
3
0.1
0.3
1
3
0.1
0.3
1
3
0.1
0.3
1
3
0.1
0.3
1
3
Spatial frequency
(cycles per degree)
Fig. 5. Spatial properties of all simple and complex cells that respond to
both luminance gratings and equiluminant gratings (n = 48). (a) Spatial
frequency (SF) preference for luminance versus equiluminance. (b)
Spatial frequency bandwidth (BW, in octaves) for luminance versus equiluminance. Cells with indeterminate bandwidth are not plotted (n = 8).
412
stimuli, that these neurons demonstrate the classic characteristics of color opponency—subtraction of L- and M-cone inputs.
Twelve of the fourteen of the cells that exhibited chromatic opponency were also spatially selective for the equiluminant and coneisolating stimuli. Thus, the majority of the simple
color-luminance neurons were double-opponent cells.
To determine that the color-luminance neurons were sensitive to photometric equiluminance, and were not behaving
as slightly miscalibrated photometers, we performed a colorexchange experiment on each cell (see Methods). We examined color directions with slightly different balances of red and
green around equiluminance. If a cell behaves as a slightly miscalibrated photometer, it will give a response to photometric
equiluminance, but it will have a null response at a red/green
ratio slightly displaced from photometric equiluminance. The
color exchange results indicated that the small number of sim-
a
Preferred equiluminance SF (cycles per degree)
preferred sf for luminance (2.56 ± 1.26 cycles per degree, n = 42)
was not significantly different from the mean preferred sf for
L-cone-isolating stimuli (2.39 ± 1.37 cycles per degree, n = 37;
Student’s t-test, p > 0.1) or from the mean preferred sf for Mcone-isolating stimuli (2.6 ± 1.26 cycles per degree, n = 40;
p > 0.1). Only a few color-luminance cells gave a measurable
response to S-cone isolating stimuli; in such cells, the mean preferred sf for S-cone isolating stimuli was significantly lower
(1.60 ± 0.76 cycles per degree, n = 9) than the mean preferred sf
for luminance (p < 0.05). The spatial frequency preference in
the luminance direction and the sf preference in L-cone and in
M-cone directions were not significantly different within colorluminance cells (Student’s paired t-test, 0.05 < p < 0.1 when comparing the preferred sf to luminance to the preferred sf in the
L-cone-isolating direction; p > 0.1 when comparing the preferred
sf to luminance to the sf preference in the M-cone-isolating direction). The color-luminance neurons that responded to S-coneisolating stimuli have significantly different spatial frequency
preferences for S-cone stimuli than for luminance (paired t-test,
p < 0.05). These data suggest that most color-luminance neurons
maintain their spatial frequency tuning characteristics for achromatic stimuli, red/green equiluminant stimuli and (at least in the
L- and M-cone-isolating directions) to stimuli that are chromatic,
but not necessarily equiluminant. This strengthens the notion
that these neurons are indeed spatially and chromatically opponent; that is, they are double-opponent cells.
To establish further that the neurons that we classified as
responsive to both color and luminance were capable of spatial
discrimination in the chromatic domain, we determined their
chromatic opponency and orientation selectivity. Chromatic
opponency in simple cells was established by analyzing the Land M-cone phase difference (Fig. 7) at the peaks of the L- and
M-cone isolating spatial frequency responses (Fig. 4). Cells showing red/green chromatic opponency will have phase differences
between L- and M-cone responses that cluster around 180°
(Fig. 7a), as seen in cortical neurons that respond only to
red/green equiluminance (sensitivity index > 2). Cells without
chromatic opponency have phase differences for L- and M-cone
isolating stimuli close to 0° (Fig. 7b) because the cone signals add
for these neurons. In the color-luminance class, 14/20 simple cells
exhibited signs of chromatic opponency, with phase differences of
90–180° between responses from each cone input (Fig. 7c). Many
of these cells had approximately equal amplitude responses to Land M-cone-isolating stimuli, which were out of phase with each
other. It is important to emphasize that it is because these neurons
give both substantial responses to red/green equiluminance, and
also responses of opposite sign to the L- and M-cone-isolating
3
1
Simple
Complex
low-pass
low-pass
1
3
Preferred luminance SF (cycles per degree)
b
low-pass
Equiluminance BW (octaves)
© 2001 Nature Publishing Group http://neurosci.nature.com
F1 amplitude (spikes/s)
articles
3
1
Simple
Complex
1
3
low-pass
Luminance BW (octaves)
nature neuroscience • volume 4 no 4 • april 2001
© 2001 Nature Publishing Group http://neurosci.nature.com
DISCUSSION
Color-luminance cells make up a large majority of cells in V1
that respond to color and a significant fraction of the total cells in
V1. Such cells have been observed previously in layers 2/3, but
their role in color vision was obscured by the use of stimuli that
were higher in luminance contrast than color contrast7. An earlier study also reported the finding of what we call color-luminance cells, but these cells were not characterized fully5.
nature neuroscience • volume 4 no 4 • april 2001
no response
low-pass
L-cone isolating BW (octaves)
ple color-luminance cells that had phase differences between
the L- and M-cone responses of less than 90° were cells that
gave response nulls slightly offset from photometric equiluminance (Fig. 7c). However, the simple cells that showed evidence of chromatic opponency from the phase difference
analysis did not null in the color-exchange experiment. A number of complex cells gave spatially selective responses to both
luminance and color (26/28). Although complex cells cannot be
analyzed for phase differences, we did examine their responses in the color-exchange experiment. The color-exchange
results indicate that 6/28 complex color-luminance neurons
had response nulls slightly offset from photometric equiluminance. Therefore, the majority (22/28; 79%) exhibited signs of
chromatic opponency. It is likely that the mechanisms underlying the processing of cone inputs in these complex cells are
similar to those of the color-opponent simple cells.
Orientation selectivity in cortical neurons has primarily been
associated with neurons responsive to luminance-defined stimuli, and only non-oriented neurons have been thought to produce
significant responses to purely chromatic stimuli4,7. To determine
the selectivity for orientation in the three groups of neurons in
our sample, we measured orientation selectivity to high-contrast
black/white luminance stimuli, and calculated circular variance
from these measurements (see Methods). As expected, many cells
responding preferentially to luminance gratings are highly selective for orientation. Overall, the orientation circular variances for
the cells responding preferentially to luminance had a bimodal
distribution, with peaks centered around 0.2 (sharply tuned) and
0.7 (broadly tuned), with a population mean of 0.45. This suggests that although many luminance-responsive cells are welltuned for orientation, there are many that are weakly selective.
The color-luminance neurons also showed a high-degree of orientation selectivity to high-contrast luminance, with a mean of
0.40. This suggests that many of the color-luminance neurons are
at least as selective for orientation as the luminance-preferring
population. The cells that responded preferentially to equiluminant color were mostly poorly tuned for orientation (mean, 0.72).
Our preliminary investigations of orientation selectivity using
equiluminant chromatic gratings suggest that color-luminance
neurons show approximately equal orientation selectivity to both
chromatic and luminance gratings. Similar results were reported
previously in V1 (ref. 20; S. Elfar and R.L. De Valois, Invest. Ophthal. Vis. Sci. Supp. 32, 2866, 1991).
a
3
L-cone
1
b
Simple
Complex
1
low-pass
3
no response
low-pass
c
M-cone isolating BW (octaves)
Fig. 6. Spatial frequency bandwidths of all simple and complex colorluminance cells for luminance gratings and cone-isolating gratings.
(a) Spatial frequency bandwidth (in octaves) for luminance versus
L-cone-isolating stimuli (n = 37). (b) Spatial frequency bandwidth for
luminance versus M-cone-isolating stimuli (n = 35). (c) Spatial frequency
bandwidth for luminance vs. S-cone isolating stimuli (n = 41). Cells with
indeterminate bandwidth are not plotted. Cells with no measurable
response to stimuli in a particular cone-isolating direction are shown in
the box on the top of each plot.
3
M-cone
1
Simple
Complex
1
low-pass
3
no response
low-pass
S-cone isolating BW (octaves)
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
3
S-cone
1
Simple
Complex
1
3
low-pass
Luminance BW (octaves)
Color-luminance cells generally give spatially bandpass
responses to patterns modulated in equiluminant color, and they
often show similar preference and selectivity for both black/white
luminance and color. It is these spatial characteristics that bear
upon the issue of double-opponency in the cortex: cells that
receive antagonistic cone inputs that are spatially tuned. As indicated by this criterion, most of the neurons in the color-luminance population are double-opponent cells.
Double-opponent cells could underlie some of the color and
luminance interactions observed psychophysically, such as color
discrimination21, the spatial frequency selectivity of simultaneous
masking with luminance and equiluminant colored stimuli22,23,
and chromatic induction24–26. Visual neurophysiologists disagree
upon the incidence of any kind of double-opponent neuron4,5.
The interpretation of the results of these earlier neurophysiological studies is unclear, primarily because luminance and chromatic stimuli were not equated for cone contrast. The use of
high-contrast luminance gratings and chromatic gratings of
413
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
15
a
Color cells
10
Fig. 7. Histograms of the L- and M-cone phase differences from the Land M-cone-the population. (a) The phase differences for neurons
responding predominantly to luminance (luminance cells). (b) The phase
differences for neurons responding predominantly to equiluminance
(color cells). (c) The phase difference for neurons responding comparably
to both chromatic equiluminance and luminance (color-luminance cells).
0
20
40
60
80 100 120 140 160 180
15
b
Luminance cells
10
5
0
0
c
20
40
60
80 100 120 140 160 180
15
Color-luminance cells
10
5
0
0
20
40
60
80 100 120 140 160 180
Phase difference
much lower effective contrast has led people to overestimate the
relative weight of the luminance component of the response, and
to underestimate chromatic responsiveness. Previous studies of
the chromatic properties in the cortex focused on neurons that
responded preferentially to color, not investigating the spatiochromatic properties of those cells that produced similar responses to chromatically and achromatically modulated stimuli (with
one conspicuous exception5). Our results indicate that there is
a substantial spatial transformation of chromatic signals in the
color-luminance cells.
A major difference between the color-responsive cells in V1
and in LGN is in the spatial selectivity of the V1 neurons for colored patterns. Neurons in the parvocellular layers of the macaque
LGN often demonstrate color opponency, but almost always
exhibit low-pass spatial selectivity for equiluminant patterns13,14,27. This is because the cone-opponent mechanisms in
the LGN are low pass, and equiluminant stimuli cause these lowpass mechanisms’ responses to add synergistically. The neural
network in V1 that drives the color-luminance cells transforms
the geniculate color signal spatially, to achieve spatial selectivity
for color (Fig. 8). To get the spatially selective curve for V1, there
Fig. 8. Responses of a LGN parvocellular neuron and a V1 simple
color-luminance cell to equiluminant red/green gratings of different
spatial frequencies.
414
must be spatial antagonism (subtractive or divisive) for each of
the cone mechanisms for equiluminant stimuli.
It is significant that the color-luminance cells that we classify
as double opponent give robust responses to both equiluminant
color and luminance contrast. It was suggested that neurons previously considered double opponent4 were actually modified
Type II cells6. Modified Type II cells have a central region that is
color opponent, whereas the suppressive surround is not color
selective. The modified Type II surround does not exhibit coloropponent responses of opposite sign to the center. Such modified Type II cells would produce weak responses to any spatially
extended stimulus because of their strong suppressive surrounds.
Thus, we would not be able to study them with our set of stimuli
(on a large bright background), and they would not be included in our database of V1 neurons. The perceptual function of
modified Type II cells is unclear, as these neurons respond robustly only to small spots of colored light on a dark background. Both
modified Type II and true double-opponent cells, if they exist,
would show color-opponent responses to small spots. However,
modified Type II responses are suppressed by large spots of any
color, and respond poorly to spatially extended spatial patterns
of equiluminant color6. If all neurons thought to be double-opponent color cells in previous studies4 were in fact modified Type
II cells6, then such neurons were not included in our analysis. In
our experience, neurons that have suppressive surrounds are present in V1, but are encountered relatively infrequently. Most of
the color-luminance cells that we found to be double opponent
gave robust, spatially selective responses to both equiluminant
color and luminance contrast, and to spatial patterns on a large
bright background, distinguishing them from modified Type II
cells. The double-opponent cells we describe need not have a concentric organization, as in the classical description4,28. Instead,
many were oriented and might have, for instance, a side-by-side
arrangement of chromatically opponent subunits. This further
differentiates them from modified Type II cells. Color-luminance
neurons are quite prevalent, especially in layers 2 and 3.
LGN
V1
50
F1 amplitude (spikes/s)
0
Number of cells
© 2001 Nature Publishing Group http://neurosci.nature.com
5
40
30
20
10
0
0.1
0.3
1
3
Spatial frequency
(cycles per degree)
nature neuroscience • volume 4 no 4 • april 2001
© 2001 Nature Publishing Group http://neurosci.nature.com
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
Finding the locus in V1 cortex of color-responsive neurons has
been delayed somewhat by the concentration in previous studies
on color-only cells. Previous studies reported the presence of double-opponent cells in V1 in layer 4 (ref. 29) and the cytochrome
oxidase blobs of layers 2/3 (ref. 4). The laminar distribution of
the color-luminance neurons indicates that these cells are found in
all cortical layers, but are concentrated especially in superficial
layers 2/3. Cells in layers 2/3 are important in the output of visual information in V1 to many extrastriate areas30–33. Previous data
suggest that color-selective cells project to many visual cortical
areas, and that the chromatic and achromatic tuning characteristics are similar for a large proportion of V2 and V3 cells34,35.
Hence, the cells in V1 carrying spatial information about color
and luminance probably project to many visual cortical areas.
It has been suggested that color information is conveyed predominantly via cells that have non-oriented4 and spatially lowpass7 chromatic responses. Most (14/19) neurons in our V1
sample that responded preferentially to chromatic equiluminance
are low-pass, non-oriented units. However, these cells make up
only 8% of all the neurons in our study and only 21% of the
color-responsive neurons we sampled. Although color-only cells
may be important in color perception, it is unlikely that they are
sufficient to account for all aspects of color vision. This is consistent with the idea that the color contrast sensitivity curve11,12
is an envelope of a multiplicity of spatial mechanisms10,15,23.
Color-luminance cells may be responsible for chromatic processing that is different from the cells that give preferential
responses to color, as they show evidence of chromatic opponency, often demonstrate a high degree of orientation selectivity,
and have spatial tuning to both color and luminance. Color-luminance neurons may respond to cues for form, like boundaries or
features, and they may take signals from color or black and white
as needed to define the form cue. They may also provide the
needed signals for color contrast induction because they respond
at color boundaries, but not in the interior of colored regions,
because of their spatial selectivity.
METHODS
We recorded extracellular responses from 167 neurons in the parafoveal
primary visual cortex of anesthetized (sufentanil citrate, 6 µg/kg/h) and
paralyzed (pancuronium bromide, 0.1 mg/kg/h or vecuronium bromide,
0.1 mg/kg/h) adult Old-World monkeys (Macaca fascicularis). All procedures conformed to the guidelines approved by the New York University
Animal Welfare Committee. We recorded single units as described previously36. Small electrolytic marking lesions (2–3 µA for 3 s, electrode tip
negative) were made through each penetration, to reconstruct the recording sites with respect to the laminar boundaries of the cortex (as described
in ref. 19). Although the sections were stained for the mitochondrial enzyme
cytochrome oxidase, it was not possible to locate cells in the blob/interblob
regions due to the length of the recording session (4–5 days), as cytochrome
oxidase activity decreases due to the duration of acute experiments.
Visual stimuli were generated on a Silicon Graphics O2 computer and
displayed on a Sony Multiscan 17seII color monitor measuring 31.4 cm
wide and 23.5 cm high. The refresh rate of the monitor was 100 Hz, with
a mean luminance of 53 cd/m2. The chromaticity of the background was
x = 0.288, y = 0.294. The stimuli were viewed at a distance of 115 cm.
Each cell was characterized to determine the receptive field’s optimal
parameters for orientation, temporal frequency, area and contrast using
sinusoidal luminance gratings. The optimized values from these measurements were used in the determination of the spatial frequency tuning
for luminance, red/green equiluminance and the three cone-isolating
directions. The red/green equiluminant gratings were produced by modulating the red and green guns of the CRT in antiphase with modulation
depths calibrated to be equal and opposite in luminance. The monitor
calibrations for luminance were based on the human spectral sensitivity
nature neuroscience • volume 4 no 4 • april 2001
function (Vλ), and were determined photometrically. All stimuli used
in these experiments were of the same mean luminance as the surround.
Stimuli for the three cone-isolating directions (L-, M- and S-cone) were
produced by appropriately adjusting the modulation of the three CRT
guns to null out the responses of two of the three cone types37.
In the spatial frequency tuning experiments, each type of grating was
approximately equated for cone contrast, as follows. Cone excitations
were calculated as the dot product of the cone absorption fundamentals38
and the spectral energy distribution of the CRT gun primaries measured
with a Photo Research spectroradiometer. Cone contrast was calculated as
the modulation of the guns for each cone divided by the mean excitation
for each cone. For the equiluminant stimuli, L-cone contrast was 0.04
and M-contrast was –0.096. A chromatically opponent mechanism would
respond to the difference between these contrasts, so the effective equiluminant cone contrast would be approximately 0.14. Although the maximum luminance modulation attainable is 1.0, we tried to equate the
luminance and chromatic equiluminance stimuli in terms of cone contrast, using a luminance modulation of 0.2. The use of a somewhat higher effective luminance than equiluminant contrast could have caused us to
overestimate the luminance cell population a small amount. For the coneisolating stimuli, the L-cone contrast was 0.13, the M-cone contrast was
0.15, and the S-cone contrast was 0.24.
Spatial tuning was measured in all color directions with drifting sinusoidal gratings. Each stimulus was presented for 4 seconds on a background of mean luminance (53 cd/m2) followed by a blank of mean
luminance of the same duration to determine the spontaneous firing rate
and to avoid response adaptation. Spatial frequencies from full-field modulation to approximately 10 cycles per degree were presented in equal logarithmic intervals. To try to avoid chromatic aberration, we recorded in the
parafovea (roughly 2–5° eccentric), where the spatial frequency tuning is
limited to intermediate to low spatial frequencies. We believe the effects of
chromatic aberration on our classification system are negligible because
of the spatial frequency range and the low contrast of the stimuli.
The responses were compiled and averaged relative to the temporal period of the grating to form post-stimulus time histograms. These
histograms were Fourier analyzed to calculate the mean response rate
(DC) as well as the amplitude and phase of the fundamental stimulus
frequency (F1). The cells were classified as simple or complex according to the ratio of the mean to first harmonic response39. Cells that
did not give a response of at least 10 spikes per second above the mean
spontaneous rate to either luminance or equiluminant chromatic gratings were excluded from the analysis.
To make quantitative comparisons, we fit the spatial frequency
responses with a difference of Gaussians of the following form.
2
2
R(f) = R 0 + Kce −((f – µc )/2σc ) − Ks e −((f − µs )/2σs )
Here, R0 is estimated from the mean spontaneous activity of the neuron, Kc, Ks,, σc, µc, σs and µs were optimized to make the best MSE fit
to the data, and f denotes spatial frequency. For simple cells, R0 = 0. From
this fit, we were able to extract the optimal spatial frequency and spatial
tuning bandwidth from each tuning curve. Spatial tuning bandwidth is
defined as the distance in octaves between the highest and lowest frequencies to which a cell was half as sensitive as it was to the preferred
frequency. If the best-fitting curve did not have enough low-frequency
attenuation to cause the response to drop to one-half the peak amplitude, the cell was classified as being low-pass in spatial tuning. Student’s
t-tests were done to examine whether the distributions of spatial frequency preferences and the distributions of spatial bandwidths differed
for high-index, low-index and color-luminance cells. The significance
level in all tests was p < 0.05.
In the color-exchange experiment, the red gun contrast was fixed at
1.0, and the green gun contrast varied from 0 to –1.0. The green and red
modulation were 180° out of phase. The stimuli were drifting at the optimal orientation, spatial frequency and temporal frequency as determined
by the initial receptive field characterization. The methods for colorexchange are described in detail elsewhere40.
415
© 2001 Nature Publishing Group http://neurosci.nature.com
articles
© 2001 Nature Publishing Group http://neurosci.nature.com
Orientation tuning was determined for each cell with high-contrast
black/white, drifting grating stimuli of the optimal spatial and temporal
frequency. Orientation was varied in 15° or 20° steps through 360°. Orientation responses for the two directions of drift were combined, and
circular variance was determined from these response measurements.
Circular variance measures the orientation selectivity based on all the
orientations measured, and it is defined as V = 1 – R, in which R is
the resultant.
R=
Σ k r k exp{i2πθk /180}
Σ k rk
Here, θk represents equally spaced orientation angles spanning 0° to 360°,
and rk represents the spike rate at each orientation. For complex cells,
the spontaneous rate was subtracted from the mean spike rate, and for
simple cells, the spike rate was measured as the amplitude of the first harmonic response. Cells with very sharp orientation tuning are mapped to
values of V close to 0, and those with broad orientation tuning are
mapped to values close to one20,41–43.
ACKNOWLEDGEMENTS
We thank D. Ringach, M. Sceniak, I. Mareschal, J.A. Henrie and F. Mechler for
helping with the physiology experiments. L. Smith helped in the histological
reconstruction and during physiology experiments. We also thank R.C. Reid for
comments on the manuscript. This work was supported by National Institutes of
Health grant EY01472 and EY8300, National Institute of Mental Health
predoctoral grant MH12430-01 and Core Grant for Vision Research P30-EY13079.
RECEIVED 24 OCTOBER 2000; ACCEPTED 23 FEBRUARY 2001
1. Zeki, S. A Vision of the Brain (Blackwell Science, Oxford, 1993).
2. Wallach, H. Ueber visuell wahrgenommene Bewegungsrichtung. Psych.
Forsch. 20, 325–380 (1935). Translated into English in Wüerger, S., Shapley,
R. & Rubin, N. On the visually perceived direction of motion. Perception 25,
1317–1368 (1996).
3. Lennie, P. in Color Vision: From Genes to Perception (eds. Gegenfurtner, K. R.
& Sharpe, L. T.) 235–247 (Cambridge Univ. Press, Cambridge, UK, 1999).
4. Livingstone, M. & Hubel, D. H. Anatomy and physiology of a color system in
the primate visual cortex. J. Neurosci. 4, 309–356 (1984).
5. Thorell, L. G., De Valois, R. L. & Albrecht, D. G. Spatial mapping of monkey V1
cells with pure color and luminance stimuli. Vision Res. 24, 751–769 (1984).
6. Ts’o, D. Y. & Gilbert, C. D. The organization of chromatic and spatial
interactions in the primate striate cortex. J. Neurosci. 8, 1712–1727 (1988).
7. Lennie, P., Krauskopf, J. & Sclar, G. Chromatic mechanisms in striate cortex
of macaque. J. Neurosci. 10, 649–669 (1990).
8. Hubel, D. H. & Wiesel, T. N. Receptive fields and functional architecture of
monkey striate cortex. J. Physiol. (Lond.) 195, 215–243 (1968).
9. Dow, B. M. Functional classes of cells and their laminar distribution in
monkey visual cortex. J. Neurophysiol. 37, 927–946 (1974).
10. Engel, S. A., Zhang, X. & Wandell, B. A. Color tuning in human visual cortex
measured using functional magnetic resonance imaging. Nature 388, 68–71
(1997).
11. van der Horst, G. J. C. & Bouman, M. A. Spatiotemporal chromaticity
discrimination. J. Opt. Soc. Am. 59, 1482–1488 (1969).
12. Mullen, K. T. The contrast sensitivity of human colour vision to red-green
and blue-yellow chromatic gratings. J. Physiol. (Lond.) 359, 381–400 (1985).
13. De Valois, R. L. & De Valois, K. K. Spatial Vision (eds. Broadbent, D. E. et al.)
218 (Oxford Univ. Press, New York, 1988).
14. Kaplan, E., Shapley, R. M. & Purpura, K. Color and luminance contrast as
tools for probing the primate retina. Neurosci. Res. Suppl. 8, S151–S165
(1988).
416
15. Bradley, A., Switkes, E. & De Valois, K. Orientation and spatial frequency
selectivity of adaptation to color and luminance gratings. Vision Res. 28,
841–856 (1988).
16. Krauskopf, J. Effect of retinal image stabilization on the appearance of
heterochromatic targets. J. Opt. Soc. Am. 53, 741–744 (1963).
17. Yarbus, A. L. Eye Movements and Vision (Plenum, New York, 1967).
18. Callaway, E. M. Local circuits in primary visual cortex of the macaque
monkey. Annu. Rev. Neurosci. 21, 47–74 (1998).
19. Hawken, M. J., Parker, A. J. & Lund, J. S. Laminar organization and contrast
sensitivity of direction-selective cells in the striate cortex of the Old World
monkey. J. Neurosci. 8, 3541–3548 (1988).
20. Leventhal, A. G., Thompson, K. G., Liu, D., Zhou, Y. & Ault, S.
J. Concomitant sensitivity to orientation, direction, and color of cells in layers
2, 3, and 4 of monkey striate cortex. J. Neurosci. 15, 1808–1818 (1995).
21. Cole, G. R., Stromeyer, C. F. I. & Kronauer, R. E. Visual interactions with
luminance and chromatic stimuli. J. Opt. Soc. Am. A 7, 128–140 (1990).
22. De Valois, K. K. & Switkes, E. Simultaneous masking interactions between
chromatic and luminance gratings. J. Opt. Soc. Am. 73, 11–18 (1983).
23. Switkes, E., Bradley, A. & De Valois, K. K. Contrast dependence and
mechanisms of masking interactions among chromatic and luminance
gratings. J. Opt. Soc. Am. A 5, 1149–1162 (1988).
24. Shevell, S. K. & Wei, J. Chromatic induction: border contrast or adaptation to
surrounding light. Vision Res. 38, 1561–1566 (1998).
25. Krauskopf, J., Williams, D. R., Mandler, M. B. & Brown, A. Higher order color
mechanisms. Vision Res. 26, 23–32 (1986).
26. Jameson, D. & Hurvich, L. M. Perceived color and its dependence on focal,
surrounding, and preceding stimulus variables. J. Opt. Soc. Am. 49, 890–898
(1959).
27. Reid, R. C. & Shapley, R. M. Spatial structure of cone inputs to receptive fields
in primate lateral geniculate nucleus. Nature 356, 716–718 (1992).
28. Michael, C. R. Color vision mechanisms in monkey striate cortex: dualopponent cells with concentric receptive fields. J. Neurophysiol. 41, 572–588
(1978).
29. Gouras, P. Opponent-color cells in different layers of foveal striate cortex.
J. Physiol. (Lond.) 238, 583–602 (1974).
30. Van Essen, D. C., Newsome, W. T., Maunsell, J. H. & Bixby, J. L. The
projections from striate cortex (V1) to areas V2 and V3 in the macaque
monkey: asymmetries, areal boundaries, and patchy connections. J. Comp.
Neurol. 244, 451–480 (1986).
31. DeYoe, E. A. & Van Essen, D. C. Concurrent processing streams in monkey
visual cortex. Trends Neurosci. 11, 219–226 (1988).
32. Levitt, J. B., Lund, J. S. & Yoshioka, T. Anatomical substrates for early stages in
cortical processing of visual information in the macaque monkey. Behav.
Brain Res. 76, 5–19 (1996).
33. Boyd, J. D. & Casagrande, V. A. Relationships between cytochrome oxidase
(CO) blobs in primate primary visual cortex (V1) and the distribution of
neurons projecting to the middle temporal area (MT). J. Comp. Neurol. 409,
573–591 (1999).
34. Kiper, D. C., Fenstemaker, S. B. & Gegenfurtner, K. R. Chromatic properties
of neurons in macaque area V2. Vis. Neurosci. 14, 1061–1072 (1997).
35. Gegenfurtner, K. R., Kiper, D. C. & Levitt, J. B. Functional properties of
neurons in macaque area V3. J. Neurophysiol. 77, 1906–1923 (1997).
36. Sceniak, M. P., Ringach, D. L., Hawken, M. J. & Shapley, R. Contrast’s effect
on spatial summation by macaque V1 neurons. Nat. Neurosci. 2, 733–739
(1999).
37. Estevez, O. & Spekreijse, H. The “silent substitution” method in visual
research. Vision Res. 22, 681–691 (1982).
38. Stockman, A. & Sharpe, L. T. Spectral sensitivities of the middle- and longwavelength sensitive cones derived from measurements in observers of
known genotype. Vision Res. 40, 1711–1737 (2000).
39. Skottun, B. C. et al. Classifying simple and complex cells on the basis of
response modulation. Vision Res. 31, 1079–1086 (1991).
40. Shapley, R. M. & Hawken, M. J. in Color Vision: From Genes to Perception (eds.
Gegenfurtner, K. R. & Sharpe, L. T.) 221–234 (Cambridge Univ. Press,
Cambridge, 1999).
41. Mardia, K. V. Statistics of Directional Data (Academic, London, 1972).
42. Batschelet, E. Circular Statistics in Biology (Academic, London, 1981).
43. Levick, W. R. & Thibos, L. N. Analysis of orientation bias in cat retina.
J. Physiol. (Lond.) 329, 243–261 (1982).
44. Lund, J. S., Hendrickson, A. E., Ogren, M. P. & Tobin, E. A. Anatomical
organization of primate visual cortex area V1. J. Comp. Neurol. 202, 19–45 (1981).
nature neuroscience • volume 4 no 4 • april 2001