The literature on orientation and color columns in monkey visual cortex is reviewed. The orientation column model most consistent with existing data is one containing ‘stripes’ of alternating positive and negative orientation ‘singularities’ (cytochrome oxidase blobs) which run along the centers of ocular dominance (OD) columns, with horizontal and vertical orientations alternating at interblob centers. Evidence is summarized suggesting that color is mapped continuously across the monkey’s primary visual cortex, with the ends of the spectrum located at ‘red’ and ‘blue’ cytochrome oxidase blobs and extra-spectral purple located between adjacent red and blue blobs in the same OD column. In the orientation column model, the ‘linear zones’ of Obermayer and Blasdel have the appearance of the lines on a pumpkin. A pinwheel model of color columns, consistent with existing data, includes spectral and extra-spectral colors as spokes. Spectral iso-color lines run across iso-orientation lines in linear zones, while extra-spectral iso-color lines occupy the ‘saddle points’ of Obermayer and Blasdel. The color column model accounts for closure of the perceptual color circle, as proposed by Isaac Newton in 1704, but does not account for color opponency.
Lines and borders in the visual world can be given numerical designations ranging from 0 to 180°, based on their orientation with respect to a standard meridian. The horizontal meridian is most often used as the standard, so that a horizontal line is said to have an orientation of zero, with vertical corresponding to 90° and the two diagonals corresponding to 45 and 135°. Other (‘oblique’) line orientations are designated by number only (e.g. 17°, 22.5°).
In the 1960s Hubel and Wiesel (1962,1968) made the discovery that the seemingly abstract concept ‘line orientation’ was represented as a continuous variable in the primary visual cortex (striate cortex, area V1) of cats and monkeys. A recording microelectrode traversing the cortex parallel or nearly parallel to the surface encountered cells whose preferred orientation, designated numerically, changed linearly in relation to horizontal distance travelled by the electrode. Hubel and Wiesel (1974,1977) illustrated this phenomenon in the form of graphs of preferred line orientation versus horizontal distance through the cortical tissue (Fig. 1).
In the course of studying orientation specificity, Hubel and Wiesel (1962,1968) discovered the principle of ocular dominance. They noted that some cells in primary visual cortex received input almost exclusively from one or the other eye (ocular dominance groups 1 and 7); other cells received roughly equal inputs from both eyes (ocular dominance group 4); intermediate groups showed somewhat stronger input from one eye than the other (ocular dominance groups 2, 3, 5 and 6). Contralateral eye dominance was designated by the numbers 1–3, ipsilateral eye dominance by the numbers 5–7.
Using a reduced silver method, LeVay et al. (LeVay et al., 1975) demonstrated that the primary visual cortex in monkeys contained an orderly ‘map’ of right and left eye dominance regions or ‘columns’ (Fig. 2). Physiological studies (Hubel and Wiesel, 1968) suggested that the right/left eye segregation extended from the cortical surface to white matter, which was confirmed using a 2-deoxyglucose autoradiographic method (Kennedy et al., 1976; Hubel et al., 1978). The term ‘column’, derived from earlier work by Mountcastle (Mountcastle, 1957) in the somatosensory cortex, carried with it the suggestion that ‘columnar organization’ might be a basic property of the cerebral cortex in general. Hubel and Wiesel (Hubel and Wiesel, 1962, 1963, 1968) also described the basic orientation unit as a ‘column’, though the linearity of the orientation change with horizontal distance (Fig. 1) was suggestive of a continuous gradient.
Orientation Column Model
Once the orderly pattern of ocular dominance columns was known, it became of interest to establish the geometrical relationship between the orientation and ocular dominance column networks. Since ocular dominance columns were mostly parallel stripes, it was natural to imagine orientation columns as stripes running orthogonal to ocular dominance stripes. Hubel and Wiesel (Hubel and Wiesel, 1972) proposed this arrangement in a model, which Hubel (Hubel, 1988) subsequently referred to as the ‘ice cube’ model (Fig. 3), since it had the appearance, when viewed from above, of an ice cube tray. The ice cube model was logical but speculative, and Hubel and Wiesel issued disclaimers about it, indicating that biological variability would, in all likelihood, cause deviations from the idealized ‘ice cube’ form.
A few years after the publication of the ice cube model, Braitenberg and Braitenberg (Braitenberg and Braitenberg, 1979) proposed a ‘centric’ model (Fig. 4), involving circular arrays of orientation columns, characterized by periodic orientation ‘singularities’ (Swindale, 1982), where columns for all orientations came together in a single point. In the vicinity of a singularity the various orientations were arrayed as they actually appear in the visual world, with vertical and horizontal running orthogonal to one another and the two diagonal orientations also running orthogonally to one another, at 45° angles to horizontal and vertical. Blasdel and Salama (Blasdel and Salama, 1986) noted the ‘pinwheel’ appearance of the Braitenberg and Braitenberg (Braitenberg and Braitenberg, 1979) model.
Cytochrome Oxidase Blobs and Orientation Columns
The pinwheel model did not receive much attention until the discovery in the early 1980s of cytochrome oxidase blobs (Fig. 5) in the monkey visual cortex (Hendrickson et al., 1981; Horton and Hubel, 1981; Humphrey and Hendrickson, 1983; Horton, 1984), following Wong-Riley’s (Wong-Riley, 1979) initial report in a study of cat cortex. Cells in the cytochrome oxidase blobs lacked orientation selectivity (Horton and Hubel, 1981; Hubel and Livingstone, 1981; Livingstone and Hubel, 1984; Blasdel and Salama, 1986), leading to speculation that the blobs might be the orientation singularities postulated by Braitenberg and Braitenberg (Braitenberg and Braitenberg, 1979). Several new circular orientation column models appeared in rapid succession (Dow and Bauer, 1984; Horton, 1984; Gotz, 1987, 1988; Baxter and Dow, 1989) [for a review see (Erwin et al., 1995)].
Baxter and Dow (Baxter and Dow, 1989) described four possible pinwheel or centric models (Fig. 6), with orientation singularities centered on cytochrome oxidase blobs. Singularities could be of index ½ or 1, depending on whether they encompassed 180 or 360° of orientation rotation. Singularities could be either positive or negative, depending on whether orientation rotated clockwise or counterclockwise with clockwise movement around the center. The models were labelled E or A, depending on whether positive singularities occurred in every blob (E type) or in alternate blobs (A type). The E1 model corresponded to Braitenberg and Braitenberg’s (Braitenberg and Braitenberg, 1979) pinwheel model, converted into a rectangular array [Horton’s fig. 49 (Horton, 1984)]. The A1 ‘checkerboard’ model had been proposed by Dow and Bauer (Dow and Bauer, 1984). The A½ model had been proposed by Gotz (Gotz, 1987, 1988).
The voltage-sensitive dye studies of Blasdel and colleagues (Blasdel and Salama, 1986; Blasdel, 1992; Obermayer and Blasdel, 1993) showed singularities to be of index ½, thereby eliminating the E1 and A1 models. The findings of Blasdel and colleagues were confirmed by Bartfeld and Grinvald (Bartfeld and Grinvald, 1992), using a different optical imaging method that does not require voltage-sensitive dyes (Grinvald et al., 1986).
The other two models in Figure 6, the E½ and the A½, can also be ruled out by optical imaging studies, which showed that orientation singularities are mostly located in the middle of ocular dominance stripes (Bartfeld and Grinvald, 1992; Obermayer and Blasdel, 1993) and that adjacent singularities within the same ocular dominance column tend to have the same sign, with ‘saddle points’ between them, while adjacent singularities across an ocular dominance column border tend to have opposite signs, with ‘linear zones’ between them (Obermayer and Blasdel, 1993). ‘Linear zones’ in this context are two-dimensional patches, 0.5–1.0 mm across, within which iso-orientation lines remain roughly parallel to one another (Fig. 7). ‘Saddle points’ in this context are two-dimensional patches within which orientation preference remains approximately constant (Fig. 7).
The optical imaging studies suggest orientation column models with alternating positive and negative singularity ‘stripes’ running down the centers of ocular dominance columns. There are two such models, as shown in Figure 8A,B. Comparing Figure 8A and B, one notes that interblob centers alternate between horizontal and vertical orientations in Figure 8A and between the two diagonal orientations in Figure 8B. This results in apparent orientation anisotropies, with horizontal and vertical predominating in Figure 8A and the two diagonals predominating in Figure 8B. Is there evidence to support either anisotropy?
The Oblique Effect
The ‘oblique effect’, increased visual acuity for horizontal and vertical lines, has been recognized since the 1960s (Campbell and Kulikowski, 1966; Campbell et al., 1966; Mitchell et al., 1967; Appelle, 1972; Berkley et al., 1975; Mustillo et al., 1988; Saarinen and Levi, 1995; Reisbeck and Gegenfurtner, 1998). The oblique effect is present in monkeys (Mansfield and Ronner, 1978; Bauer et al., 1979; Boltz et al., 1979; Williams et al., 1981), cats (see LeVay and Nelson, 1991) and ferrets (Coppola et al., 1998), as well as humans, and applies to stereoacuity (Mustillo et al., 1988) and vernier acuity (Saarinen and Levi, 1995), as well as iso-luminant (Reisbeck and Gegenfurtner, 1998) and luminant (i.e. black/white) grating acuity.
The oblique effect is not due to the optics of the eye (Campbell et al., 1966; Mitchell et al., 1967). It is consistent with some single unit studies (Mansfield, 1974; DeValois et al., 1982; LeVay and Nelson, 1991) and two brain imaging studies (Coppola et al., 1998; Furmanski and Engel, 2000) indicating more cells and more striate cortical tissue devoted to horizontal and vertical than other orientations. The effect is most prominent in central vision (Mansfield, 1974; Berkley et al., 1975).
The oblique effect suggests that the model in Figure 8A is preferable to the model in Figure 8B. While there are studies that do not find the anisotropy for horizontal and vertical orientations in striate cortex (see LeVay and Nelson, 1991), there are no studies reporting an anisotropy for diagonals.
In his Opticks (Newton, 1704; cited in MacAdam, 1970) Isaac Newton presented two crucial concepts for color vision: first, that white light can be broken down by refraction through a lens into colors of decreasing wavelength in the order red, orange, yellow, green, blue, violet; second, that the long and short wavelength ends of the color ‘spectrum’ can be combined to create a color circle, with the non-spectral color purple at the interface between red and violet. Figure 9 shows the electro-magnetic spectrum, including the zone between 400 and 700 nm, which is ‘visible’ to the human eye (Carpenter, 1984). Figure 10 shows Newton’s color circle.
Thomas Young (Young, 1802; MacAdam, 1970) suggested that three retinal receptors sensitive to different portions of the visible spectrum would be adequate to subserve color vision and Hermann von Helmholtz (von Helmholtz, 1866; MacAdam, 1970) provided psychophysical support for what came to be known as the Young/Helmholtz trichromatic theory.
Ewald Hering (Hering, 1920; Hurvich and Jameson, 1964) proposed, on the basis of his own subjective impressions, that there appear to be four primary colors, red, yellow, green and blue, which function as two sets of opponent pairs. Hering displayed his ‘primary’ colors in a circular fashion, with red opposite green and yellow opposite blue (Fig. 11). As he noted, one can have a yellowish red (i.e. orange) or a bluish red (i.e. purple), but not a greenish red (or a reddish green). Hering also postulated a third opponent system, white and black, to account for luminance contrast effects. Hurvich and Jameson (Hurvich and Jameson, 1957) developed Hering’s ideas into a quantitative psychophysical system (Hurvich, 1981), providing for color opponency what Helmholtz had provided for trichromacy.
For some time, trichromacy and color opponency were viewed as competing theories, though it now seems evident that both are correct. There are clearly three types of cones (Smith and Pokorny, 1975; Schnapf et al., 1988; Nathans et al., 1992) and their signals are transformed in the brain into color opponent mechanisms (Boynton, 1979; Hurvich, 1981; Vautin and Dow, 1985). Color vision in macaque monkeys is biologically and perceptually similar to human color vision (DeValois et al., 1974a; Sandell et al., 1979; Harosi, 1982).
Newton’s color circle (Fig. 10) closes the spectrum; Hering’s color circle (Fig. 11) introduces orthogonal opponent axes. Figure 12 combines Newton’s and Hering’s color circles, depicting purple (at the left) as reddish blue, orange as yellowish red, lime as greenish yellow and aqua as bluish green. The named colors of Figure 12 are illustrated in Figure 13.
If line orientation, represented as angles from 0 to 180°, is mapped as a continuous variable in the monkey visual cortex (Hubel and Wiesel, 1974, 1977), might not the colors in Figures 12 and 13 be similarly mapped onto the monkey brain? To go from three cones in the retina to Figures 12 and 13, the brain must accomplish two tasks. It must join the two ends of the spectrum to form purple and it must create color opponency.
Cells receiving excitatory input from both the long wavelength-sensitive (L) and short wavelength-sensitive (S) cones, with apparent inhibitory input from the middle wavelength-sensitive (M) cones, have been described in the primary visual cortex of macaque monkeys (Gouras, 1970, 1974; Dow, 1974; Thorell et al., 1984; Vautin and Dow, 1985; Dow and Vautin, 1987; Yoshioka et al., 1996). DeValois and colleagues (Cottaris and DeValois, 1998; DeValois et al., 2000b) have reported an expansion of the blue (i.e. short wavelength-sensitive, ‘S’) cone system at the visual cortical level. A possible role for this expansion is closure of the color circle.
DeValois et al. (DeValois et al., 1966) presented evidence suggesting that color opponency in the macaque monkey begins at the level of the lateral geniculate nucleus, the first way-station on the visual pathway from retina to cortex. Wiesel and Hubel (Wiesel and Hubel, 1966), using small (and large) spots as test stimuli, showed that the apparent color opponency of most lateral geniculate ‘color’ cells (i.e. type I cells) was center/ surround opponency, similar to what had been reported in retinal ganglion cells (Kuffler, 1953).
Color opponency in the monkey visual cortex, using small spots and narrow bars as test stimuli to avoid encroachment on receptive field surrounds, was first reported by Hubel and Wiesel (Hubel and Wiesel, 1968), and subsequently confirmed by Dow and Gouras (Dow and Gouras, 1973; Dow, 1974; Gouras 1974). Michael (Michael, 1981) also reported finding color opponency in monkey visual cortex and presented data suggesting separate columnar systems for the opponent colors red and green, but not for yellow and blue.
Cytochrome Oxidase Blobs and Color Columns
The discovery of cytochrome oxidase blobs (Fig. 5) and the fact that cells in blobs lacked orientation selectivity raised the possibility that blob cells might be color selective, especially in view of earlier reports indicating partial dissociation of orientation and color processing in macaque monkey visual cortex (Dow, 1974; Baizer et al., 1977; Zeki, 1978).
Livingstone and Hubel (Livingstone and Hubel, 1984) proposed that the cytochrome oxidase blobs constituted a system of non-oriented color columns interdigitated between mostly achromatic orientation columns. Their proposal implied that all colors would be represented within each blob ‘column’, though they did not present evidence on this issue.
Ts’o and Gilbert (Ts’o and Gilbert, 1988) conducted a study of blob columns and concluded that there were two types, one processing the opponent colors red and green, the other processing the opponent colors yellow and blue. They did not specify how cells with opposite color preferences would be situated within a single column.
Dow (Dow, 1974) noted the presence of non-oriented color cells in the upper layers of monkey visual cortex. Dow and Vautin (Dow and Vautin, 1987) found two types of non-oriented upper layer zones, one containing mostly red cells, the other containing mostly blue cells. Yoshioka and Dow (Yoshioka and Dow, 1996) documented the non-oriented upper layer zones as cytochrome oxidase blobs, proposing that there were two kinds of cytochrome oxidase blobs, red blobs and blue blobs.
Dow and Vautin (Dow and Vautin, 1987) noted the existence of many oriented color cells with preferences for colors toward the middle of the spectrum, such as orange, yellow, lime, green and aqua, and Yoshioka and Dow (Yoshioka and Dow, 1996) showed that these oriented mid-spectral color cells were located in interblob regions. The results of the two studies led to the (initially) counterintuitive proposal that mid-spectral colors (i.e. yellow, lime, green) are more closely associated with orientation selectivity than end-spectral colors (i.e. red, purple, blue).
Psychophysical data support a mid-spectral/orientation association. Visual acuity is greater for mid-spectral than for end-spectral lines and gratings (Walls, 1943; Riggs, 1965; LeGrand, 1967; Kelton et al., 1978; Mullen, 1987). The yellow macular pigment, which selectively filters out blue light, assists in this process, along with the scarcity of blue cones in the foveal region of the retina (Wald, 1967; Walls, 1967; Williams et al., 1981). Another contributing factor involves chromatic aberration and accommodation (Walls, 1943, 1967; Hartridge, 1947; LeGrand, 1967; Kruger et al., 1993). In order to bring spectral light rays to a focus at the back of the retina, the lens must adjust its shape, a process known as accommodation. It is not possible to accommodate such that long and short wavelengths are both brought to a sharp focus on the retina. For standard viewing purposes, the accommodation mechanism selects the best compromise, which turns out to be optimal focus for middle wavelengths. Thus, red, blue and their mixture purple are generally defocused and associated with a certain degree of blurring. Mid-spectral images (yellow, lime, green) are better focused and associated with higher levels of visual acuity. Mid-spectral images are also typically associated with higher levels of luminance (Fig. 9) and, on that basis, increased visual acuity (DeValois et al., 1974b; Yoshioka and Dow, 1996; Yoshioka et al., 1996).
2-Deoxyglucose studies of Tootell et al. (Tootell et al., 1988a) showed a striking tendency for red and blue stimuli to label blobs more strongly than either yellow or green stimuli, supporting the concept of red and blue processing in blobs. Other studies (Lennie et al., 1990; Leventhal et al., 1995; Johnson et al., 2001) did not find color cells restricted to blob regions, supporting the concept of color processing as a distributed function across the V1 cortical surface. Studies by Tootell et al. (Tootell et al., 1988b), Silverman et al. (Silverman et al., 1989) and Edwards et al. (Edwards et al., 1995) indicated gradual rather than abrupt changes in contrast sensitivity and spatial frequency responses with increasing distance from cytochrome oxidase blob centers, supporting the notion of smooth gradients rather than distinct functional compartments in the macaque monkey’s primary visual cortex. The possibility that ‘spatial frequency’ may be represented as a continuous variable within the striate cortical tissue (Silverman et al., 1989), while beyond the scope of the present discussion, is compatible with the model being presented here.
The results summarized above suggest that color may be mapped in continuous fashion across the surface of area V1, analogous to the continuous mapping of orientation. Cytochrome oxidase blobs, a discontinuity in the orientation map, serve an integral role in a continuous color map. A continuous color map has its own discontinuity, namely the interblob centers, which are achromatic (Dow and Vautin, 1987; Yoshioka and Dow, 1996).
Color Column Model
Having selected an orientation column model, we now select a color column model. Evidence from several sources (Dow and Vautin, 1987; Tootell et al., 1988a; Yoshioka and Dow, 1996) indicates that blobs come in two types, containing predominantly red cells or blue cells. We do not know the proportions of red blobs and blue blobs and we do not know how red and blue blobs are distributed over the cortical surface.
Based on perceptual considerations (Hurvich, 1981; DeValois and DeValois, 1993; DeValois et al., 2000a), roughly equal numbers of red and blue blobs would appear likely (see also Dow and Vautin, 1987). The 2-deoxyglucose data of Tootell et al. (Tootell et al., 1988a) do not indicate a preponderance of either red or blue blobs. Vautin and Dow (Vautin and Dow, 1985) found roughly equal numbers of ‘red’ and ‘blue’ cells, using luminance matched stimuli; Yoshioka et al. (Yoshioka et al., 1996), in the same laboratory, found fewer blue than red cells, using dimmer blue stimuli. Livingstone and Hubel (Livingstone and Hubel, 1984) reported a preponderance of red cells in blobs, but their color testing was not systematic, often involving broad band filters without apparent luminance calibration, and they may have missed some blue cells. Ts’o and Gilbert (Ts’o and Gilbert, 1988) reported twice as many red cells as blue cells in blobs (30 versus 15%), but they also used broad band filters and may have missed some blue cells. Optimal activation of blue cells requires narrow band filters to avoid inhibition from the middle wavelength-sensitive (M) cone system (Dow, 1974; Gouras, 1974).
The most likely place for the short wavelength-sensitive (S) cone system expansion (see above) reported by DeValois and colleagues in monkey striate cortex (Cottaris and DeValois, 1998; DeValois et al., 2000b) is in the non-oriented blob regions, where the S cone system appears to be concentrated in striate cortex (Dow, 1974; Dow and Vautin, 1987; Tootell et al., 1988a; Ts’o and Gilbert, 1988; Yoshioka and Dow, 1996) [for a comprehensive review see (Hendry and Reid, 2000)].
There are three possible non-random arrangements of equal numbers of two types of blobs in a given small subregion of cortical tissue, namely alternating or stripes in two orientations (Fig. 14). Option B in Figure 14 would involve a right versus left eye color bias, which is clearly unacceptable. Option C in Figure 14 would have iso-color lines (e.g. blue–blue, red–red) running orthogonal to ocular dominance stripes, which seems undesirable, given that iso-orientation lines appear to run orthogonal to ocular dominance stripes, at least in ‘linear zones’ (Obermayer and Blasdel, 1993). For optimal matching of orientations and colors, one would like iso-color lines to run orthogonal to iso-orientation lines, i.e. parallel to ocular dominance stripes in ‘linear zones’. By exclusion, option A in Figure 14 is the most likely of the three.
Figure 15A,B illustrates the orientation/color column model. The two figures are superimposable, though it is easier to describe them (and view them) separately. Figure 15A shows the orientation column model, which has a ‘pumpkin’ appearance and is based on Obermayer and Blasdel’s (Obermayer and Blasdel, 1993) data (Fig. 7). The pumpkin lines or barrel ‘staves’, which also have the appearance of meridian lines, are the ‘linear zones’ of Obermayer and Blasdel (Obermayer and Blasdel, 1993). The staves cross ocular dominance (OD) column boundaries (vertical bar) at 90° angles. The spaces between pumpkins are the ‘saddle points’ of Obermayer and Blasdel (Obermayer and Blasdel, 1993). Orientations associated with individual staves are indicated. ‘Singularities’ (cytochrome oxidase blobs) are shown as empty circles where staves converge. The model does not include the ‘fractures’ of Obermayer and Blasdel (Obermayer and Blasdel, 1993).
The model suggests that the ‘oblique effect’ (see above) is related to the sizes of the ‘saddle points’ versus the areas bounded by individual pumpkin staves. Separation between staves is adjustable; with smaller stave separation the saddle points become larger.
Figure 15A predicts that saddle points should be associated with either vertical or horizontal orientation and that diagonal orientations should be represented in the middle of linear zones. Figure 3 of Obermayer and Blasdel (Obermayer and Blasdel, 1993) shows several saddle points, which appear to be associated with either horizontal or vertical orientations, and one linear zone, with a left oblique orientation in the middle.
The ‘pinwheel’ color model is illustrated in Figure 15B. The interblob centers, indicated as empty circles, are achromatic ‘singularities’, with colors arranged as spokes. Spectral colors (see Fig. 12), labeled ROYL (red, orange, yellow, lime) and BAGL (blue, aqua, green, lime), occupy the two lateral quadrants of each color circle. Extra-spectral colors, labeled B, pB, bP, P (blue, purplish blue, bluish purple, purple) and R, pR, rP, P (red, purplish red, reddish purple, purple), occupy the upper and lower quadrants of each circle. Spectral colors occupy the (binocular) regions near OD column boundaries, with mid-spectral lime at the actual boundary. Extra-spectral colors occupy the middle (monocular) portions of each OD column. As mentioned above, mid-spectral colors are more suitable for precise binocularity mechanisms, due to the higher visual acuity associated with them. According to the model, the middle portions of OD columns combine inputs from the two ends of the spectrum, while the border zones combine inputs from the two eyes. The notion of closing the ends of the spectrum to create extra-spectral colors was originally formulated by Isaac Newton (Newton, 1704; MacAdam, 1970).
Interblob center regions are the likely sites for processing of ‘luminance’, including the achromatic ‘colors’ white, gray and black (Dow and Vautin, 1987; Yoshioka and Dow, 1996; Yoshioka et al., 1996), the particular achromatic color depending on relative luminance in comparison to nearby interblob centers, i.e. a given interblob center would function as ‘white’ if local interblob firing rates were lower, ‘black’ if local interblob firing rates were higher and ‘gray’ if local interblob firing rates were the same.
Superimposition of Figure 15A,B indicates that spectral iso-color lines run roughly orthogonal to iso-orientation lines in ‘linear zones’ (Obermayer and Blasdel, 1993), which is appealing for the purposes of information storage and retrieval. Orthogonality would permit each color to be separately matched with all orientations and each orientation to be separately matched with all colors. The positioning of the lines could be adjusted so as to optimize orthogonality.
The association of vertical and horizontal orientations with color singularities (i.e. achromacy) in Figure 15 is a prediction of the model and may be related to the ‘oblique effect’ mentioned earlier. This particular association should be testable using either psychophysical or biological techniques.
Color Opponency Model
A problem with the color model of Figure 15B is that it does not account for color opponency. The opponent colors red/green and yellow/blue are not located on opposite sides of the circle, as in the color circles of Figures 11–13. The color ‘singularities’ of Figure 15B are of index 1, which is why the model looks like a pinwheel (Braitenberg and Braitenberg, 1979; Erwin et al., 1995). To optimize color opponency, one would prefer them to be of index ½.
Color opponency is present in selected cells of area V1 (Hubel and Wiesel, 1968; Dow and Gouras, 1973; Dow, 1974; Gouras, 1974; Michael 1981; Livingstone and Hubel, 1984; Vautin and Dow, 1985; Ts’o and Gilbert, 1988), but full perceptual color opponency may be deferred to a higher level of visual processing. Area V2, on the basis of its known anatomical and functional organization (Hubel and Livingstone, 1987; Tootell and Hamilton, 1989; Yoshioka and Dow, 1996; Kiper et al., 1997; Ts’o et al., 2001), appears to offer a better geometry for color opponency (B.M. Dow, in preparation).
A particular problem for color opponency in V1, according to the present scheme, is that two of the primary colors, red and blue, are represented by mostly non-oriented and monocular cells, while the other two primary colors, green and yellow, are represented by mostly oriented and binocular cells. In V2, with some rearranging, this apparent incompatibility may be corrected. The recent study of Ts’o et al. (Ts’o et al., 2001), showing interactions between ‘color’ and ‘disparity’ zones in area V2, provides insight into the rearranging process. Incidentally, it should be noted that the end-spectral/mid-spectral, non-oriented/oriented distinction reported by our laboratory (Dow and Vautin, 1987; Yoshioka and Dow, 1996; Yoshioka et al., 1996) is based on awake monkey recordings with full binocularity (and active fixation). Results obtained in anesthetized monkeys with monocular testing (or uncorrected binocular disparity) might be different.
Summary and Conclusions
Following a review of the existing literature, an idealized model or map of the macaque monkey’s primary visual cortex has been constructed. The model incorporates ocular dominance columns, orientation columns and color columns. The vagaries of genetics and developmental insults, etc., will introduce irregularities into the system, such that the actual map in any given monkey will differ in its details from the one presented here. This becomes apparent when one examines Obermayer and Blasdel’s (Obermayer and Blasdel, 1993) reconstructions from voltage-sensitive dye data (Fig. 7). There is, likewise, individual variability in ocular dominance maps (LeVay et al., 1975, 1985) and retinotopic maps (Van Essen et al., 1984; Dow et al., 1985) obtained from different monkeys. The issue is not whether a given monkey’s orientation/color map exactly reproduces the model presented here, but whether the model illustrates the basic organizing principles involved in the mapping process.
The model suggests that orientation singularities coincide with cytochrome oxidase blobs. Optical imaging researchers (Bartfeld and Grinvald, 1992; Blasdel, 1992) have pointed out that this is not always the case. However, it should be noted that orientation singularities are identified in living tissue, while cytochrome oxidase blob locations are determined post-mortem, following tissue perfusion and fixation. Uneven tissue shrinkage could account for some of the discrepancies. A second possible source of error is pressure on the living brain tissue from the optical imaging apparatus (and a recording microelectrode, if one is used). A third possible source of error is from vascular pulsations, which may cause movement of the tissue during imaging. The bottom line is that one is dealing here with a living, pulsating, chemically sensitive tissue. It is inevitable that there will be deviations from perfect matching of pre-mortem physiology and post-mortem anatomy.
The model is testable. Blasdel and colleagues (Blasdel and Salama 1986; Blasdel, 1992; Obermayer and Blasdel, 1993) have used line orientations to look at voltage-sensitive dye activation patterns in monkey striate cortex. Presumably their studies could be repeated using colored gratings, similar to the stimuli used by Tootell et al. (Tootell et al., 1988a) in their 2-deoxyglucose studies. Functional magnetic resonance imaging (fMRI) techniques, such as have been used extensively in humans, may be adaptable to monkeys as well (Logothetis et al., 1999, 2001) and fMRI resolution may now permit visualization of cortical columns (Grinvald et al., 2000; Kim et al., 2000).
Combining microelectrode recording with optical imaging (Ts’o et al., 2001) can be particularly valuable, with optical imaging results being used to guide microelectrode placement and microelectrode recording results being used to verify optical imaging data. Arrays of multiple electrodes (Eckhorn et al., 1988) may be useful in columnar mapping studies. An advantage of microelectrode recordings, whether individual or multiple, is that one can sample different depths within the same column and document actual columnar architecture (Hubel and Wiesel, 1968; Dow and Vautin, 1987; Eckhorn et al., 1988; Ts’o and Gilbert, 1988; Yoshioka and Dow, 1996; Ts’o et al., 2001), including the possibility of columnar differences in the deeper layers of striate cortex (Bauer et al., 1980, 1983). Optical reflectance methods sample only the upper layers of cortex.
I thank Dr Seunghyun Yoo for assistance with figures and several anonymous Cerebral Cortex reviewers for their helpful suggestions.