We have shown in the accompanying paper that optical imaging of macaque striate cortex reveals patches that are preferentially activated by equiluminant chromatic gratings compared with luminance gratings. These imaged color patches are highly correlated, although not always in one-to-one correspondence, with the cytochrome-oxidase (CO) blobs. In the present study, we have investigated the electrophysiological properties of neurons in the imaged color patches and the CO blobs. Our results indicate that individual blobs tend to contain cells of only one type of color opponency: either red/green or blue/yellow. Individual imaged color patches, however, can bridge blobs of similar opponency or differing opponency. When imaged color patches contain two blobs of differing opponency, the cells in the bridge region exhibit mixed color properties that are not opponent along the two cardinal color axes (either red/green or blue/yellow). Two blobs within a single imaged color patch receive input from the same eye or from different eyes. In the latter case, the bridge region between blobs contains binocular cells that are color selective. Because the cells recorded in imaged color patches were more color selective and unoriented than cells outside of color patches, color properties appear to be organized in a clustered and segregated fashion in primate V1.
Although many early studies described the existence of color-selective cells in the macaque striate cortex (see Dow 1974; Dow and Gouras 1973; Dow and Vautin 1987; Gouras 1974; Gouras and Kruger 1979; Hubel and Wiesel 1968; Poggio et al. 1975; Thorell et al. 1984), a turning point in our understanding of cortical color processing came with the work of Livingstone and Hubel (1984). Their study provided a description of a functional organization for color processing in striate cortex (V1) by correlating single-unit recordings with cytochrome-oxidase (CO) histology. Livingstone and Hubel reported that patches of high cytochrome-oxidase activity (blobs) (Wong-Riley 1979) contained a concentration of unoriented, monocular, color-selective cells, compared with the surrounding interblob regions that stain more lightly in the histology. They also showed that most color-opponent cells in the CO blobs of V1 have either center/surround receptive field properties, similar to those found in parvocellular lateral geniculate nucleus (LGN) (Wiesel and Hubel 1966), or center-only receptive fields, more rarely seen in the LGN.
Additional color cell types beyond the LGN-like cells have also been found in the CO blobs of V1. Livingstone and Hubel (1984) described double opponent cells in V1 that showed chromatic opponencies of the opposite sign for the centers versus surrounds (Daw 1968; Hubel and Wiesel 1968). Ts'o and Gilbert observed both blue/yellow as well as red/green color opponency in the blobs, and described another type of color cell receptive field, named modified type II cells, that show either red/green or blue/yellow opponency in the centers and broadband suppressive surrounds (Ts'o and Gilbert 1988). Ts'o and Gilbert also reported color-selective oriented cells that were often found near the blob/interblob borders. In the same study they demonstrated that CO blob color cells were clustered according to their color opponency such that individual CO blobs seemed dedicated to processing one particular color opponency, red/green or blue/yellow. They also found that red/green CO blobs outnumbered blue/yellow CO blobs by a ratio of about 3:1 (Ts'o and Gilbert 1988).
Several subsequent studies, however, have called into question the relationship between the distribution of V1 color cells and the CO blobs. Lennie et al. (1990), using different methodology, concluded that there was not a correlation of color-selective cells within CO blobs (see also Leventhal et al. 1995). Part of the discrepancy may be resolved by the findings of Livingstone and Hubel (1984) andTs'o and Gilbert (1988) of unoriented but broadband, non–color-selective type III cells within many CO blobs. The prevalence of such non–color-selective blob cells might explain the results of Lennie et al., since these cells would dilute any estimate of the proportion of color cells in the blobs. The presence of color-selective oriented cells outside of the CO blobs might also contribute to a dilution of the estimate. In addition, Lennie and colleagues studied very few cells in each compartment (they recorded 6 CO blob cells). In attempting to resolve these disparate studies, it would be helpful to be able to target electrode penetrations to the CO blobs in vivo, during the experimental session.
Optical imaging has provided a means to explore architecture in visual cortex by providing functional maps in vivo (Bartfeld and Grinvald 1992; Blasdel 1992a,b; Blasdel and Salama 1986; Bonhoeffer and Grinvald 1991;Grinvald et al. 1988; Lieke et al. 1989;Ts'o et al. 1990). In the previous paper we have demonstrated maps that reveal that color-selective cells tend to cluster in patches that are closely related to the CO blobs. In the present study we use these maps to guide the investigation of the receptive field properties of cells in and around the CO blobs and color patches with single- and multi-unit recordings.
Eleven monkeys (Macaca fascicularis) weighing 3.0–6.35 kg were prepared as described in detail in the companion paper. Animals were anesthetized initially with ketamine (20 mg/kg im), maintained with sodium thiopental (20 mg/kg iv), and paralyzed with vecuronium bromide (0.1 mg · kg−1· h−1 iv). Artificial respiration was provided and end-tidal CO2 monitored and maintain at 4%. The animal's electroencephalogram (EEG), electrocardiogram (EKG), and temperature were also monitored. For optical imaging, a stainless steel optical chamber was mounted around a 1-cm2craniotomy and filled with silicone oil to reduce cortical movement. For unit recording, the craniotomy was covered with a layer of agarose (3% in physiological saline) to stabilize the cortex.
Stimuli and data collection
Single- and multi-unit extracellular activity was recorded with glass-coated tungsten electrodes (Alan Ainsworth, London). The maps obtained during optical imaging were used to guide electrode placement, and the locations of recordings were marked on a vasculature map corresponding to the region of functional imaging. All cells were recorded from superficial layers (within 500 μm from the pin surface).
Action potentials were monitored over a loudspeaker for qualitative characterization of responses, or saved on a computer for quantitative analysis (see following text). For qualitative characterization of receptive fields, a hand-held projector was used to stimulate receptive fields either with broadband light or with equal energy interference filters spaced every 30 nm from 450–630 nm presented on a dark monitor with white paper, placed in front of the animal.
Cells were first tested with an oriented white stimulus to assess orientation selectivity. Orientation selectivity was classified asA through D using the scale of Livingstone and Hubel (1984) where A indicates sharp tuning, andD indicates unoriented responses. Then we tested responses to large and small spots, both white and colored, using the interference filters. Orientation preference was also tested with different colors. In general, most unoriented cells had a choppy and/or unreliable response to oriented bars that were longer than the receptive field center. Responses were gauged for both the onset and offset of the stimulus and were tested with stationary and moving stimuli.
Ocular dominance was also tested using the Hubel and Wiesel (1962) scale of 1–7, where 1 refers to a purely contralateral response, 7 refers to a purely ipsilateral response, and 2–5 indicate varying degrees of binocularity. To assess monocularity rather than eye dominance, we have averaged 1's with 7's, 2's with 6's, and 3's with 5's and remade the scale as 1–4 in Figs. 3-7.
Cells were categorized as color-opponent if they exhibited anon response to one range of wavelengths (for example, 600–630 nm would be called red-on) and an offresponse to another range of wavelengths (for example, 510–540 nm would be called green-off). As is shown in the results (Figs. 11-13), this method of hand characterization was in good agreement with the quantitative evaluation of cells. The majority of color-opponent cells showed classical red/green or blue/yellow opponency.
In addition to classical color-opponent cells, a small population of neurons had mixed color properties; that is, they could not be characterized either as purely red/green or blue/yellow. Other authors have also found neurons in V1 that received mixed color input (Cottaris and De Valois 1998; De Monasterio and Schein 1982; Lennie et al. 1990). These mixed color cells in most cases are predominantly red/green, but with a weaker blue/yellow component. In general, these cells are difficult to distinguish from red/green opponent cells when tested only with narrowband monochromatic stimuli. There are two exceptions, however. First, we observed a small number of cells that in all respects were similar to red-on/green-off cells, but also gave distinct on responses at 450 nm. We also observed cells that exhibited green-on/red-off responses that had strong blue-on responses as well. These latter cells were categorized as mixed cells if they showed equally strong responses to both 450 nm (blue) and 510 nm (green), but theiron responses to 480 nm were stronger than their responses to both 450 and 510 nm. This categorization is similar to that ofDe Monasterio and Schein (1982). They measured the peak responses of the surround to be near 480 nm for LGN cells with red-on centers and blue + green-off surrounds along with a broader half-bandwidth compared with red-on/green- off cells. Furthermore, red/green opponent neurons in both the LGN and striate cortex never show a peak as low as 480 nm when tested with a narrowband stimulus (De Valois 1965; Dow 1974; Dow and Gouras 1973; Wiesel and Hubel 1966). The peak always drops from around 510 nm for a cell with M-cone input, whereas a cell with S-cone input has a peak around 420 or 450 nm and shows a relatively reduced response to 480 nm.
For the quantitative comparison with cell characterizations made qualitatively, cone-isolating sinewave gratings and achromatic luminance sinewave gratings were used during collection of single-cell responses. To isolate the responses due to each of the three cone classes, the silent substitution technique was used. That is, a direction in color space was chosen so that modulation along that direction would be visible to only one of the three cone classes (Estevez and Spekreijse 1982; the cone spectra were from human psychophysical data: Smith and Pokorny 1972). A spectroradiometer that measured luminous energy at 2-nm increments between 430 and 690 nm (Photoresearch, Spectrascan, PR 703A) was used to calibrate the color stimuli. The maximum obtainable cone contrasts for these stimuli were 16% for the L cones, 19% for the M cones, and 80% for the S cones from a common white point.
The optimal spatial and temporal frequencies were first determined subjectively with achromatic drifting gratings, and these values were used for all subsequent stimuli. First, a coarse orientation tuning curve was obtained in steps of 45°. Next, a near-optimal orientation was used to test the responses to L, M, and S cone-isolating gratings (see above), as well as a luminance grating at 25% contrast. Single units were differentiated by the spike sorting software on the Discovery acquisition software package (Datawave Technologies, Longmont, CO). Maximum amplitude, spike height, and occasionally spike width were used to sort the waveforms of single spikes. Peristimulus time histograms (PSTHs) were calculated time-locked to the periodic sinusoidal stimulus. The orientation tuning curves shown are calculated from the mean firing rates for each stimulus type.
Following electrode recordings, several lesions were made within the region of optical imaging, and their locations were marked on the corresponding vasculature map. Recording sites were generally not lesioned since we found this to interfere with staining for cytochrome oxidase. Instead, lesions at the periphery of the recorded regions were used to align the histology to the optical images. Most lesions were made with 4 μA, electrode negative, constant current, held for 4 s. These lesions were easily seen in several adjacent 30-μm histological sections. Since electrode recordings were also marked on the vasculature map, we were able to align both the images and the electrophysiology with staining for CO. The tissue was prepared as described in the accompanying paper (Landisman and Ts'o 2002).
We investigated the color-selective patches seen in optical images (see previous paper, Landisman and Ts'o 2002) with perpendicular electrode penetrations confined to the superficial layers (layers II/III). We also recorded from color-selective regions using only ocular dominance (OD) maps as approximate guides of CO blob locations (Ts'o et al. 1990). These approaches allowed us to assess properties of cells within imaged color patches and to compare these properties to those of cells within and around CO blobs.
The results are organized into four sections. First, we present a single case to illustrate the relationships between optically imaged color maps, OD, CO blobs, and electrophysiological recordings (Figs. 1and 2). Second, we tabulate these relationships, both for the single case illustrated in Figs. 1 and 2 (Figs. 3 and 4) and also for a larger population of recordings from different cases (Figs. 5-7). Third, we examine the finer structure of color patches that encompass pairs of blobs (Figs. 8-10). Finally, we present a quantitative analysis of color responses, using cone-isolating grating stimuli (seemethods), as a validation of our qualitative characterization of receptive fields in the rest of the study (Figs.11-13).
Color selectivity in color patches and CO blobs
Optical images of color selectivity were good indicators of the location of color-opponent cells, as illustrated in Fig.1. As in all of the cases reported in this study, the color selectivity map was generated by subtracting responses to luminance stimuli for both eyes from responses to red/green stimuli for both eyes (Landisman and Ts'o 2002). Dark patches show regions that were more responsive to a red/green stimulus than to a luminance stimulus. In this case, the electrophysiology confirmed that neurons in the dark patches were color selective. Moreover, these patches contained predominantly unoriented cells, as previous studies found in the CO blobs (Livingstone and Hubel 1984; Ts'o and Gilbert 1988). The comparison of the electrophysiological properties to the map of OD (Fig. 1 C) reveals that color selectivity can fall both within the centers of OD columns (where CO blobs reside) as well as on OD borders, indicating that color selectivity is not restricted to the CO blobs. This finding is supported more directly by the data in Fig.2, which show a direct comparison of the color map and electrophysiology to staining for CO blobs.
Most single penetrations inside a color patch revealed one type of color opponency. In the data from the case shown in Fig. 1, only 1/17 color-selective penetrations contained cells of both opponency types (i.e., red/green and blue/yellow). Pooling data from 91 penetrations containing red/green or blue/yellow opponent cells (from a total of 21 experiments in 11 animals), only 7.7% (7/91) of the penetrations contained both classes of opponency. Each penetration considered had from two to six cells. Within this same pool, most penetrations showed red/green opponency (57/91, or 62.6%), and many fewer showed exclusively blue/yellow opponency (27/91, or 29.7%). For the case shown in Fig. 1, 10 penetrations were exclusively red/green, and 5 penetrations were exclusively blue/yellow.
Neighboring penetrations within the same color patch, however, often showed different color selectivity (see following text, Figs. 9 and10). Of 27 color patches with multiple penetrations (ranging from 2 to 10 penetrations, with an average of 4 per patch), 12, or 44%, showed mixed color opponency. The majority of color patches showed exclusively red/green opponency (13/27, or 48%). Very few showed exclusively blue/yellow opponency (2/27, or 7%). One example of a color patch with mixed opponency can be seen in the top left corner of Fig.1 B, which shows a color patch containing a penetration of red/green-selective cells and a penetration containing blue/yellow-selective cells. Penetrations with only type III cells (broadband, unoriented cells) or broadband oriented cells usually did not lie within the color-selective patches. However, many type III cells were observed in penetrations with other cells showing either red/green or blue/yellow opponency. This figure also illustrates the prevalence of red/green color selectivity over blue/yellow selectivity seen in the pooled data, confirming the findings of Ts'o and Gilbert (1988).
In Fig. 2 we compare the data from Fig. 1 with the corresponding CO staining. Although a slight majority of unoriented color-selective penetrations fell within the blobs or near their edges, there were many such penetrations outside the blobs. These penetrations with color-selective cells were virtually all inside imaged color patches (see Fig. 3). As noted in the previous paper (Landisman and Ts'o 2002), there was significant overlap between the blobs and color patches (data from 5 different cases show that 23/29 CO blob penetration were inside color patches, or 79%). Conversely, most of the penetrations outside of the CO blobs were also outside of the color patches (45/63 where n = 63 penetrations outside of CO blobs, 45 outside of color patches, 18 inside of color patches). However, the two populations are not identical: some patches corresponded to a single CO blob, but some overlapped pairs of blobs. Most of the color-selective penetrations, however, fell both within the imaged color patches and within the CO blobs.
Summaries of relationship between receptive field properties and cortical compartments
The electrophysiological properties for the case illustrated in Figs. 1 and 2 are summarized in Figs. 3 and4, which show the OD, orientation preference, and color selectivity of all the cells with respect to their location inside or outside of the imaged color patches (Fig. 3) and inside or outside of the CO blobs (Fig. 4). The monocularity of cells in the color-selective patches was not significantly different from that seen in the “interpatch” regions, but the orientation tuning and color selectivity were very different (Fig. 3). A greater proportion of cells were unoriented (categories C and D) within the patches than outside of the patches. Cells outside of the patches had a fairly even distribution of orientation selectivity. As seen in Fig. 1, most of the color-selective cells were recorded within the optically imaged color patches. However, within the color patches a similar number of cells were color selective as were not. This result mirrors the finding in the CO blobs of many unoriented broadband cells interspersed with blob color-selective cells (Livingstone and Hubel 1984; Ts'o and Gilbert 1988).
The similarity of the properties of the color patches to those of the CO blobs can be seen by comparing the graphs in Fig. 3 with those in Fig. 4. As was true for the color patches, the blobs had a greater number of unoriented and color-selective cells than the interblob regions, but the CO blobs also contained a large number of broadband cells compared with color-selective cells (like the color patches). The distributions of OD for all the populations shown in this case (for both Figs. 3 and 4) were biased toward monocular cells (category 1, see legends). This bias is due to the fact that penetrations were largely targeted to fall along the centers of OD columns.
The data in Figs. 1-4 are from a single illustrative case. To show that results for the color patches and blobs were consistent across animals, we analyzed data from five additional experimental cases (Figs. 5-7). These data do not include the cells from the case shown in Figs. 1-4. The trends indicated in Fig.5 are consistent with those seen for the color patches in Figs. 1 and 3. As in the individual case (Fig. 3), there were many broadband cells as well as color-selective cells within the color patches, and there were few color-selective cells outside of the color patches. Most of the broadband cells within color patches were type III cells (broadband, center/surround). In the five pooled cases, there is a greater proportion of color to broadband cells inside the color patches, and the disparity between red/green- and blue/yellow-selective cells was more prominent: in the individual case there were only 50% more red/green cells (Fig. 3), whereas in the population data there were almost four times as many red/green cells as blue/yellow cells (Fig. 5). The difference in these samples is due exclusively to a disparity in the number of cells recorded at each penetration since the individual case and pooled data have the same ratio (2:1) of red/green penetrations to blue/yellow penetrations. These samples are consistent with the findings ofTs'o and Gilbert (1988), who suggested that red/green blobs outnumber blue/yellow blobs by as much as 3:1. The “mixed” color cells, whose spectral tuning could not be explained by a classic red/green or blue/yellow opponency (see methods), were quite rare.
Cells inside the color patches differed from cells outside the patches in both OD and orientation tuning (Fig. 5). Outside of color patches, there was a fairly even distribution of OD. By comparison, cells within the color patches were more likely to exhibit monocularity, although more binocular cells were found than would be expected within the CO blobs (Livingstone and Hubel 1984) (also see Fig.6). Orientation tuning outside of color patches favored more sharply tuned cells (categories A and B) with very few unoriented cells (C and D), whereas the reverse was true within color patches. These trends were less clear in the analysis of the single case in Fig. 3.
Like the plots for color patches in the individual case (Fig. 3) versus the population (Fig. 5), the plots for CO blobs are extremely similar for the individual case (Fig. 4) versus the larger population (Fig. 6). Again, there was less of a bias toward monocular sampling in the larger population than in the individual case, seen mostly for the cells falling outside of the CO blobs. Also, there was a slightly greater proportion of color to broadband cells inside the CO blobs than was shown in the individual case.
Graphs of properties inside and outside of monocularity centers (n = 155 cells) indicate that there is no significant difference in the response properties of CO blobs and monocularity centers (Fig. 6 vs. Fig. 7). The bias toward monocularity as well as color selectivity inside CO blobs and inside imaged monocularity centers indicates that these are a very similar if not identical population of cells, as suggested byTs'o et al. (1990). It is worth noting that the mixed color cells seen in the imaged color patches (Figs. 3 and 5) were never found in the CO blobs (Figs. 4 and 6) or in the imaged monocularity centers (Fig. 7).
The greater number of cells with sharp orientation tuning as well as more binocular cells outside of CO blobs compared with inside CO blobs (Figs. 4 and 6) might suggest that interblob cells lying between OD columns are more sharply tuned than interblob cells along an OD column. To test this, we plotted monocularity versus orientation tuning for cells outside of the CO blobs by counting the number of cells for each category (data not shown). A χ2 analysis for the independence of two variables did not indicate any significant difference (P > 0.1). Thus orientation selectivity and OD selectivity are independent for cells in the interblob regions.
Examples of color patch microstructure
We have demonstrated that imaged color patches contain a predominance of unoriented, color-opponent cells. Previous work has shown that CO blobs also contain such cells (Livingstone and Hubel 1984, see Figs. 2, 4, and 6), and that individual CO blobs are dominated by either red/green opponent cells or blue/yellow opponent cells (Ts'o and Gilbert 1988). Color patches and blobs are not identical in our study, however, as shown in the companion paper (Landisman and Ts'o 2002). In particular, a single color patch often spans two blobs.
Here, we examine the electrophysiological properties of neurons within such extended color-selective regions. We targeted such regions by using OD maps to define presumptive blobs (Ts'o et al. 1990) and then by making multiple electrode penetrations along a line connecting two adjacent blobs. In most cases, the interblob cells tended not to be color selective. We show three examples in which cells between two blobs were color opponent: two from adjacent blobs with different OD and one from adjacent blobs with the same OD.
In the first example (Fig. 8), we show recordings from an extended color-selective region whose cells share the same color opponency (red/green) and lack of orientation selectivity, but have different OD. The area for recording was targeted with the optically imaged OD map, which showed adjacent left eye and right eye monocular regions, or monocularity centers (which correspond to CO blobs) (Ts'o et al. 1990) (Fig. 8, Aand B). In this case, we made six penetrations that showed unoriented color-opponent properties: two in each of the strongly monocular regions and two nearer the OD border (Fig. 8, Band C). Note that the color preference within the color-selective region in Fig. 8 remains the same (red/green opponent), but OD shifts from monocular for one eye, to binocular, to monocular for the other eye. The borders of this color region are delineated by penetrations with broadband oriented cells (Fig. 8, C andD).
In addition to color patches that encompassed two presumptive blobs with matching color opponency, we also found examples of color patches with areas of differing color opponency and differing eye preference, bridged by a binocular region that showed mixed color properties. In Fig. 9, an example of a color patch is shown. Here, the two monocularity centers have different color opponency (either red/green or blue/yellow), and the associated bridge between the two monocular regions has mixed color properties that are not classically opponent. Blue-yellow opponent cells on the right were largely monocular for the right eye and probably correspond to the first of two CO blobs (Fig. 9, C–E). Red/green–selective cells on the left (in the 2nd presumptive blob) received monocular input from the other eye, and the bridge connecting the two monocular blobs contained binocular cells.
The binocular cells located within the bridge region showed nonclassical color opponency. As discussed in methods, the mixed cells in this example showed strong responses to 480 nm (green-blue) and slightly weaker responses to 450 nm (blue) and to 510 nm (green). They had opposite-signed response to 600 nm (orange) that was greater than the response to 630 nm (red) or 570 nm (yellow). Their color axis thus appeared to be green-blue/orange. The corresponding region in the color-selective optical image was one large color patch that, when compared with the OD map, spanned two monocularity centers (CO blobs).
Although most color patches that we observed spanned two blobs in different OD columns, we found one case in which a color-selective region bridged two blobs with the same OD. In this example (Fig.10) we made a series of evenly spaced perpendicular penetrations along the center of one OD column (the penetrations were 100 mm apart). As seen in the OD maps (Fig. 10,B and D) as well as in the CO histology (Fig. 10,C and E), this line of penetrations spanned two blobs. From left to right, the sequence of penetrations changed receptive field properties several times: 1) first we observed a cluster of broadband oriented cells, 2) then a region (the 1st CO blob) with blue/yellow opponent neurons,3) then a bridge with neurons exhibiting mixed color properties, 4) then a region with red/green opponent neurons (the 2nd CO blob), and 5) finally another group of orientation-selective neurons. The mixed bridge cells were both red- (630 nm) and blue- (450 nm) on and green- (540 nm)off.
Quantitative versus qualitative characterization of cells
To validate our qualitative receptive field mapping, we tested 25 cells with achromatic and cone-isolating gratings following our usual characterizations with hand-held stimuli. Of these cells, 22 were used for comparison with the qualitative data (data from 3 cells were thrown out, due to nonlinear or weak responses to grating stimulation). Agreement of characterization of color properties was observed in 91% of the cells (20/22), and 9% were found to have been inaccurately characterized for color (2/22). More specifically, most of each major cell type had been correctly characterized: 9/9 broadband, 6/7 red/green, and 5/6 blue/yellow cells were accurately labeled during hand characterization. Almost all of the quantitatively characterized cells that came from cases with optical images of color selectivity showed proper correspondence to their imaged color maps (12/13; 9 color-opponent cells were inside color patches: 5/6 red/green cells were accurately characterized and 3/3 blue/yellow cells were accurately characterized; 4 broadband oriented cells fell outside color patches and were accurately characterized).
In Fig. 1 B, an example of the correspondence of quantitatively analyzed cells with optical images is shown. Five cells, for which response characteristics were quantitatively analyzed following hand characterization, are marked with asterisks and numbers. All five of the marked penetrations fell within the dark regions of the color selectivity map and had color selective cells (as determined both by hand characterization and confirmed by quantitative characterization). Penetration 1 contained both oriented broadband cells (one of which was quantitatively characterized) and unoriented blue/yellow-selective cells (one of which was also quantitatively analyzed).
Three examples of cells for which the hand characterizations corresponded with the quantitative characterizations are shown in Figs.11-13. We chose for this description one of each major cell type (broadband, red/green, and blue/yellow). In the first example, a cell was characterized by hand as broadband with a preferred orientation of 60° and moderate orientation tuning (Fig.11). In the A–D scale of Livingstone and Hubel (1984), this cell was classified as B, which corresponds to a sharpness of approximately 40 to 60°. The polar plot (Fig. 11, bottom right) obtained with an achromatic grating also showed moderate orientation tuning, with a peak near 45° (a grating at 60° was not used). The four PSTHs, obtained during stimulation with either luminance or cone-isolating drifting gratings, confirmed that the cell was broadband. In particular, the L and M cone responses were in phase with each other and therefore were not antagonistic (they peaked at the same time in the PSTHs). Similarly, the luminance response, which should be equal to the sum of the cone-isolating responses (assuming linear summation), peaks at the same point. Consistent with the qualitative observation that almost no blue (450 nm) responses were evoked in this cell, the S-cone response was small and not well correlated to the stimulus. Another cell recorded at this site was also broadband and was more sharply orientation tuned (data not shown).
In the next example (Fig. 12) a cell was qualitatively characterized as a red/green opponent modified type II cell: it had a strong green-on center, a weaker red-off center, and a broadband suppressive surround. The PSTHs obtained with L- and M-cone isolating stimulation show responses that peaked 180° out of phase and were therefore opponent. The S-cone responses were very weak and not easily distinguished from baseline noise. Using quantitative measures, this cell showed very slight orientation tuning (Fig. 12, polar plot) that was not noted in the hand characterization, but the degree of tuning was fairly loose: the ratio of firing rates for the peak orientation versus other orientations is about 1.5 to 1.
Finally, the last cell shown (Fig. 13) was hand characterized as an unoriented blue/yellow cell. As would therefore be expected, the PSTHs of the L-cone and M-cone responses had the opposite phase of the S-cone PSTH. Most importantly, the strength of the S-cone response is much greater than for the broadband and red/green cells shown in Figs. 11 and 12. As proposed in the previous paper, the varying balance of L to M cone input in blue/yellow cells may explain why these cells are more responsive to the red/green stimulus than to the luminance stimulus during optical imaging. A second cell was recorded from the same location and showed the same signature: a strong S-cone response, and an opposite-signed L-cone response. The luminance response for the second cell was dominated by the S-cone response. Both cells showed very poor orientation tuning (Fig. 13, polar plot).
As there was such clear agreement between quantitatively and qualitatively determined response characteristics in the 22 cells used for comparison, we are confident that our results using hand characterization of color selectivity are comparable with those achieved by quantitative characterization.
Layout of color-selective properties with respect to color patches
In agreement with some previous studies (Livingstone and Hubel 1984, Tootell et al. 1988b, andTs'o and Gilbert 1988; but also see Lennie et al. 1990; Leventhal et al. 1995), we have shown that color selectivity is organized in a segregated fashion in V1. As shown in the companion paper (Landisman and Ts'o 2002), optical imaging reveals structures loosely related to the CO blobs, which we term color patches. These color patches are preferentially activated by chromatic stimuli relative to achromatic stimuli. Here, we have verified that these regions contain a preponderance of cells whose receptive fields are unoriented and color selective.
Furthermore, we have demonstrated that color selectivity is segregated into regions of red/green and blue/yellow opponency, based on electrophysiological recordings. Very few single penetrations revealed both types of color opponency (<8%), and red/green opponent cells were much more prevalent than blue/yellow opponent cells, as previously reported (Ts'o and Gilbert 1988). In fact, the majority of imaged color patches showed exclusively red/green opponent neurons (48%) compared with a modest 7% of patches containing exclusively blue/yellow opponent neurons. The remainder of color patches contained both red/green and blue/yellow opponent neurons as well as color cells with mixed color properties. These patches, however, still maintained segregated regions of red/green and blue/yellow selectivity within single electrophysiological penetrations. As first noted by Livingstone and Hubel (1984), most color-selective penetrations also contain a large number of type III cells (as many as 50% within a single penetration). These cells are unoriented but show no color selectivity.
Most penetrations within the CO blobs fell inside imaged color patches; however, a large number of color-selective penetrations were outside of the CO blobs but inside the color patches (29%). Thus color selectivity is not restricted to the CO blobs as histologically defined. These results are analogous to the findings in visual area V2 where the staining of the CO stripes has been shown to be only an approximate marker for the actual functional subcompartments (Roe and Ts'o 1995; Ts'o et al. 2001).
Color patch and CO blob receptive field properties
Within both imaged color patches and CO blobs there are large numbers of both broadband and color-selective cells, with the color-selective cells outnumbering the broadband cells by approximately 50%. Most of the cells in color patches and CO blobs, however, have unoriented center-surround receptive fields.
In general, blobs and color patches appear to have very similar receptive field properties with a few small exceptions. Color patches contain more binocular cells and the rare, mixed color cells not seen in CO blobs. The distribution of OD and orientation selectivity outside of the color patches is fairly even, but there are very few color-selective cells outside of the imaged color patches. For these reasons, we conclude that the color patches are a good indicator of regions of color-selective processing in V1.
Color-selective bridges between blobs
Our detailed electrophysiological survey of color patches further reveals that individual patches can contain cells all of one type of color opponency or can contain both red/green and blue/yellow cells. In the latter case, the two cell types are not randomly distributed, but instead are segregated into two regions each with only one type of opponency, which correspond to two different blobs within the same patch. The cells between the islands of red/green and blue/yellow opponency have unoriented, mixed color properties. These cells are very rare in V1 compared with other noted cell types.
The major functional differences between the blobs and the imaged color patches are that the patches contain these cells with mixed color properties (not along either the red/green or blue/yellow color axes) and/or binocular cells. Previous studies have reported neurons in V1 with mixed color responses (Cottaris and De Valois 1998;De Monasterio and Schein 1982; Lennie et al. 1990). Our data add to this evidence that these mixed color cells fall between blobs of differing opponency and also that binocular color cells fall between blobs of differing OD.
Cells in mixed bridges that fall between two CO blobs of differing color opponency and differing eye preference are particularly interesting. The binocular, mixed-color cells encountered could only emerge from direct or indirect input from four different cell types; that is, both red/green and blue/yellow from each eye. These inputs could either come from four different types of CO blobs with similar retinotopy and/or from layer 4. If a mixed bridge received input only from the two blobs that it appeared to connect, it would be impossible for the responses of each eye to have sensitivities to a combination of red/green and blue/yellow. For more simple bridges, either between blobs of the same color specificity or the same OD, this problem would not arise.
A summary and extension of our main findings is presented in Fig. 14. The model shows the layout of color-selective properties in relation to the OD columns and CO blobs. The model includes CO blobs with only one type of color opponency (red/green or blue/yellow). It should be noted that this model does not include broadband cells (mostly type III) within color opponent blobs, which we also observed (and see Livingstone and Hubel 1984). The model does show many different combinations of blob pairings, including blobs of the same color opponency but different eye preference, and blobs of different color opponency with and without different eye preference. We have provided evidence of 1) a color-selective region spanning two red/green opponent blobs with different eye preference (Fig. 8), 2) a color patch spanning two blobs of different opponency and different eye preference (Fig. 9), and 3) a color-selective region spanning two blobs of different opponency but the same eye preference (Fig. 10). We also observed color patches spanning two blue/yellow blobs of different eye preference as well as large color patches spanning three red/green blobs (data not shown). Finally, this model includes the clustering of blue/yellow CO blobs as observed by Ts'o and Gilbert (1988).
In conclusion, we have shown that the organization of the color system in the superficial layers of V1 is more complex than previously described. Patches of color selectivity often extend beyond individual blobs, and these patches are not always homogeneous for one type of color opponency. Furthermore, as suggested by other studies, the mixing of the two color opponent channels, red/green and blue/yellow, apparently begins in V1.
We thank S. Zagorski, C. Lorusso, L. Hinderstein, and A. Meyer for technical assistance and J. Maunsell and E. Kaplan for help with organization and writing.
This work was supported by National Institutes of Health Grants GM-07524-15 and EY-08240, Office of Naval Research Grant N00014-91-J-1865, the Whitaker Foundation, the McKnight Foundation, and Helen Hay Whitney Foundation support to C. E. Landisman.
Present address of D. Y. Ts'o: Neurosurgery/Physiology IHP4111, SUNY Health Science Center, 750 East Adams St., Syracuse, NY 13210.
Present address and address for reprint requests: C. E. Landisman, Dept. of Neuroscience, Box 1953, Brown University, Providence, RI 02912 (E-mail:).
- Copyright © 2002 The American Physiological Society