Cortical maps and feedback connections are ubiquitous features of the visual cerebral cortex. The role of the feedback connections, however, is unclear. This study was aimed at revealing possible organizational relationships between the feedback projections from area V2 and the functional maps of orientation and retinotopy in area V1. Optical imaging of intrinsic signals was combined with cytochrome oxidase histochemistry and connectional anatomy in owl monkeys. Tracer injections were administered at orientation-selective domains in regions of pale and thick cytochrome oxidase stripes adjacent to the border between these stripes. The feedback projections from V2 were found to be more diffuse than the intrinsic horizontal connections within V1, but they nevertheless demonstrated clustering. The clusters of feedback axons projected preferentially to interblob cytochrome oxidase regions. The distribution of preferred orientations of the recipient domains in V1 was broad but appeared biased toward values similar to the preferred orientation of the projecting cells in V2. The global spatial distribution of the feedback projections in V1 was anisotropic. The major axis of anisotropy was systematically parallel to a retinotopic axis in V1 corresponding to the preferred orientation of the cells of origin in V2. We conclude that the feedback connections from V2 to V1 might play a role in enhancing the response in V1 to collinear contour elements.
- cerebral cortex
- visual cortex
- striate cortex
- feedback connections
- cortical maps
- orientation map
- retinotopic map
- cytochrome oxidase
Feedback connections (FbkCon) constitute a subset of corticocortical fibers, defined by specific cortical layers of cell origin and axonal termination (Rockland and Pandya, 1979; Maunsell and Van Essen, 1983; Felleman and Van Essen, 1991). This study aimed at establishing whether an organizational relationship exists between the feedback projections (FbkPrj) from V2 and the maps for orientation and retinotopy in V1.
Previous studies demonstrated a nonrandom relationship between the intrinsic horizontal connections (HrzCon) (Fisken et al., 1975; Creutzfeldt et al., 1977; Gilbert and Wiesel, 1979; Rockland and Lund, 1982) and the functional maps within V1. The HrzCon selectively interconnect the systems of cytochrome oxidase (CO)-stained blobs or interblobs (Livingstone and Hubel, 1984b; Yabuta and Callaway, 1998). Clusters formed by the HrzCon preferentially interconnect columns of similar ocular dominance in macaque V1 (Malach et al., 1993; Yoshioka et al., 1996) and of similar preferred orientations in the cat (Gilbert and Wiesel, 1989; Kisvarday et al., 1997), tree shrew (Bosking et al., 1997), and macaque (Malach et al., 1993).
The HrzCon are spatially organized in a nonrandom way in relation to the retinotopic map in V1 of the tree shrew (Fitzpatrick, 1996; Bosking et al., 1997), the cat (Schmidt et al., 1997), and the squirrel monkey (Sincich and Blasdel, 2001). In these species, labeled axons extend farther along an axis that corresponds to the preferred orientation at the injection site. Similarly, feedforward connections from layer 4 to layer 2/3 in tree shrew V1 demonstrate orientation-specific axial bias (Mooser et al., 2004).
The spatial extent, orientation selectivity, and axial specificity demonstrated by the HrzCon in V1 imply that those connections might mediate geometrically specific interactions between neurons with non-overlapping classical receptive fields (RFs) (Crook et al., 2002; Stettler et al., 2002). It is unclear, however, whether the HrzCon are the exclusive neural substrates that mediate those phenomena (Angelucci et al., 2002; Crook et al., 2002).
Whereas a clear relationship exists between the HrzCon and functional maps, only limited data are available for similar relationships of the connections between V1 and V2. The feedforward and the FbkCon are segregated into two separate pathways, one between blobs in V1 and thin CO stripes in V2 (Livingstone and Hubel, 1984a; Roe and Ts'o, 1999) and the other between interblobs in V1 and thick and pale CO bands in V2 (Sincich and Horton, 2002). Feedforward axon terminals in cat area 18 are confined to regions of orientation specificity similar to that of their parent cells in area 17 (Gilbert and Wiesel, 1989).
To date, no study has addressed the question of retinotopic axial specificity of the cortical FbkPrj. The possibility of orientation specificity of these projections is a matter of debate. Although Angelucci et al. (2002) predicted such specificity, Stettler et al. (2002) ruled it out. Moreover, although clustering of the FbkPrj is a prerequisite of orientation specificity, the issue of clustering is controversial. Evidence exists for both clustering (Wong-Riley, 1979a; Livingstone and Hubel, 1984a; Rockland and Virga, 1989) and diffuse projections (Rockland and Pandya, 1979; Maunsell and Van Essen, 1983; Ungerleider and Desimone, 1986). Angelucci et al. (2002) demonstrated FbkCon that terminate in clusters, whereas Stettler et al. (2002) doubted whether patchiness exists for the projections from V2 to V1.
The present study aimed at investigating possible organizational relationships of the FbkPrj from V2 to the CO-dense structures and to the orientation and retinotopic maps in V1. Optical imaging of intrinsic signals (Grinvald et al., 1986) was combined with CO histochemistry and connectional anatomy in owl monkeys.a The findings indicate that these FbkCon might play a role in enhancing the response in V1 to collinear visual features.
Parts of this work have been published previously in abstract form (Shmuel et al., 1998).
Materials and Methods
The procedures used for preparing and maintaining the animals and for tracer injections and histology were described by Malonek et al. (1994) and Malach et al. (1993, 1994), respectively. The methods of optical imaging and analysis of functional maps were described by Shmuel and Grinvald (1996) and Bonhoeffer and Grinvald (1996). In the following, these procedures are presented in outline, whereas new methodological aspects are described in detail.
Data on FbkPrj were obtained from six adult owl monkeys (Aotus nancimai) weighing between 0.8 and 1.1 kg. Data on the HrzCon within V1 were obtained from an additional monkey. A craniotomy was performed above areas V1 and V2, and the dura was removed. The imaged area extended from 1 to 16 mm away from the midline. These regions of V1 and V2 represent the lower half of the visual hemifield near the vertical meridian, between ∼1 and ∼15° (V1) and ∼1 and ∼20° (V2) from the area centralis (Allman and Kaas, 1971, 1974). A keratometer was used to fit the corneas with zero-power contact lenses. The eyes were focused on a tangent screen at a distance of 28.5-57 cm using external lenses. The areae centrales were projected onto the screen with the aid of a reversible ophthalmoscope and a fundus camera.
We used a visual stimulator (STIM software; Kaare Christian, Rockefeller University, New York, NY) run on an IBM personal computer equipped with a graphics board (SGT; Number Nine Computer Corporation, Lexington, MA). The stimuli were displayed on a CRT screen in a 60 Hz non-interlaced mode. They subtended an angle of 25-50° in the visual field contralateral to the imaged hemisphere. To obtain retinotopic maps, monocular stimulation of the eye contralateral to the imaged hemisphere was used. To obtain orientation maps in V1 and V2, the monkeys were stimulated binocularly with high-contrast rectangularwave gratings with a spatial frequency of 0.5-1.0 cycles/degree and a duty cycle of 33%. The set of stimuli included four differently oriented gratings, spanning the orientation spectrum at a resolution of 45°. The gratings moved alternately in the two directions orthogonal to their orientation at a temporal frequency of 5 Hz.
Functional data acquisition
We used a differential video data acquisition system, IMAGER 2001 (Optical Imaging, Mountainside, NJ). Each data acquisition cycle started with an 8 s interstimulus interval, during which a stationary pattern of the stimulus was displayed. The stimulus then drifted for 2.5 s while images of the cortex were taken.
Analysis of functional maps
Data were analyzed using self-written procedures in MATLAB (The MathWorks, Natick, MA).
Orientation single-condition maps. To obtain an orientation singlecondition map, the mean cortical image obtained while stimulating with a specific orientation was divided by the mean “cocktail blank.” Highpass filtering was applied by subtracting the convolution of the result with an isotropic Gaussian with SD of 500 μm.
Differential maps. Differential orientation maps were obtained by subtracting two single-condition maps corresponding to orthogonal orientations.
Vectorial analysis. Four different orientation single-condition maps were summed vectorially on a pixel-by-pixel basis. The polar map presents the angle of the resulting vector by means of hue and its magnitude by brightness.
Maps of SD. To estimate the pixelwise reliability of either a differential map or a polar map, we computed the corresponding map of SD across the imaging session. For consistent comparison of the mapping signal-to-noise ratios across different experiments, pixelwise confidence intervals of the computed differential orientation maps were computed based on t statistics. Data from cortical locations in which the interval corresponding to a confidence coefficient of 0.95 was larger than the amplitude of differential maps were excluded from the computations of quantitative histograms.
Biotinylated dextran amine injections targeted at optically identified functional domains
Functional maps were obtained from the vicinity of the V1/V2 border, which could be identified during the experiment according to the different layouts of the orientation domains: continuous in V1 versus at intervals in V2. The exact location of the border was verified after the experiment using CO histology (Fig. 1A).
A clearly defined orientation-selective domain in V2 was selected for biotinylated dextran amine (BDA) injection. The center of this domain was marked on the functional map and transferred to the image of the cortical vasculature. The fine pattern of superficial blood vessels was viewed through an operating microscope and used as a reference to target a micropipette to the selected injection site. The pipette (8-12 μm tip diameter) was moved using a micromanipulator (MO 103; Narishige, Tokyo, Japan) and was advanced perpendicular to the local pial surface to a cortical depth of 600-800 μm, estimated by the distance traversed after first touching the cortex. The tracer was iontophoretically ejected (tip positive).
After the injections, the monkeys were maintained for at least 28 h under continuous monitoring of physiological state and level of anesthesia. They were then deeply anesthetized and perfused transcardially with PBS, followed by light fixation with 2.5% (by volume) paraformaldehyde and 0.5% glutaraldehyde in PBS containing 5% sucrose. The brains were removed, and the posterior half of the brain was flattened and postfixed. The tissue was sectioned parallel to the cortical surface at intervals of 50 μm (60 μm for monkey M1). The pattern of the superficial pial blood vessels was preserved at the uppermost section because of the care taken in flattening and sectioning of the cortical tissue.
Two in every three sections were processed for BDA (Fig. 1B), using the procedure described by Malach et al. (1993) for processing biocytin. The interleaved third sections were processed for CO histochemistry (Fig. 1A), using the procedure described by Wong-Riley (1979b).
Anatomical data analysis
The stained sections were viewed at low power with a Wild M5 macroscope equipped with light- and dark-field illumination, a camera lucida, a 35 mm camera, and a color video camera attached to a computer. Axon terminals were verified by higher-power viewing using a Universal Zeiss (Oberkochen, Germany) microscope. Digital images of the histological sections were obtained using an Olympus Optical (Tokyo, Japan) SZX12 microscope.
The tissue sections were aligned with the optically obtained images using the pattern of the superficial pial vasculature (Malach et al., 1993, 1994). To ensure exact registrations and to avoid local misalignments after global registrations, the injection site in V2 and the projections within V1 were aligned separately based on the local vasculature pattern (supplemental Fig. 1, available at www.jneurosci.org as supplemental material). We applied only linear transformations (expansion, translation, and rotation of the pattern of the local vessels); more complex distortions were not apparent within the small areas that were aligned. Deeper sections were aligned with the superficial map by using the profiles of the radial vessels descending perpendicularly relative to the cortex. To minimize inaccuracy attributable to small movements of the brain during the imaging session, the optically obtained detailed image of the cortical vasculature (“green” image) was first aligned to the map of the SD of the differential functional maps obtained optically. This map reflected the superficial blood vessels, because their response was more variable than the response from the gray matter. Thus, the map of SD could be used as an indicator of the exact locations of the vessels during the imaging session. The pattern of superficial vasculature from the first section was then aligned to the green image. Last, deeper sections were aligned sequentially from top to bottom. Each section was aligned to its neighboring upper section.
The data were subjected to quantitative analysis in each section that contained either axonal projections or retrogradely labeled cell bodies in V1. To that end, we used both low and high magnification. Axonal clusters were delineated as regions showing a clear increase in axonal density, while excluding axons of passage and fibers seen to be emanating from cells labeled in V1 (Fig. 1E) (also see Fig. 6). The delineation of axon terminal clusters was based on the following criteria: (1) a clear increase in axonal density, estimated at low magnification as denser labeling than that of the surrounding tissue; (2) emergence of clusters from the convergence of more than one axon; (3) bouton terminals and at least one bifurcation within the delineated zone (viewed under high magnification) in each of the converging axons (we could thus exclude axons of passage); and (4) the few axons seen to be emanating from cells stained in V1 were excluded from the analysis. Patch borders were drawn using a drawing tube. Each cluster was delineated by a curve that encircled the axonal components composing the cluster. This curve was positioned ∼20 μm from the closest axonal component. Patches were classified into five categories according to their axonal density. The classification was based on the mean intensity across each patch in the gray level-transformed dark-field image of the section. Before the classification the gray level-transformed image was high-pass filtered (by subtracting the convolution of the image with a Gaussian with SD of 200 μm from the image).
The slice-specific injection area was defined using two different methods. In method I, the effective-uptake zone was defined as the region of densely labeled, saturated staining around the injection site. In method II, labeled cell bodies that could be individually identified at high magnification were marked. The cells at the edge of the densely labeled saturated region around the injection site defined the anchor points for the curve of delineation (Fig. 1D). These cells were the most distant from the center of the injection radially but were still spatially adjacent to the densely labeled group of cells at the center of the injection, with no space in between. The region defined by connecting the individual cells represented the effective-uptake zone obtained by the use of method II. The two methods gave almost identical delineations. The quantitative data based on delineation of the effective-uptake zone presented throughout the paper were obtained by method I. Figure 2 presents the slice-specific diameter of the effective-uptake zone in V2 for all six monkeys. Each histological section is represented in a box. The boxes are organized in columns, one column per monkey. The vertical dimension represents cortical depth. The effective-uptake zone labeled V2 columns spanning layers 1-6 and had a diameter of 425 ± 130 μm (mean ± SD; n = 6; computed across the supragranular layers as described below).
In analyzing orientation preference and axial specificity, we considered the effective-uptake zones and axonal projections restricted to the supragranular layers (for the monkey-specific supragranular histological sections, see Fig. 7). To compute the orientation preference associated with the injection area, the spatial maps of the slice-specific effective-uptake zone were combined using a logical OR operator applied pixelwise across the slices. The resulting region was superimposed on the functional maps, and the distribution of preferred orientation values at the injection site was quantitatively analyzed using image-processing software (NIH Image and MATLAB).
To compute the orientation preference of the feedback axon terminals in V1, we computed the weighted mean orientation over the pixels that overlapped the delineation of these terminals within the supragranular layers. To compute the weighted mean, each pixel was assigned a coefficient equal to the class number of the cluster to which the pixel belonged.
The histological patterns were analyzed and aligned with the optically obtained superficial vasculature by a person other than the one who analyzed the functional maps and their relationship to the FbkPrj.
Overlap of feedback axon terminals and cytochrome oxidase blobs
Blob regions were defined by filtering and thresholding the digitized CO images. High-pass filtering was applied by subtracting the convolution of the image with an isotropic Gaussian with SD of 1000 μm from the image. This was followed by low-pass filtering (convolution with an isotropic Gaussian with SD of 50 μm) and application of a threshold. In cases in which the results of thresholding were not homogeneous over V1, blob regions were delineated manually according to the local CO contrast.
To quantify the overlap between the FbkPrj and the CO-blob regions, we calculated the area of the convex hull encircling the clusters of the FbkPrj in V1 [A(conv hull)]. Similarly, we computed the area occupied by the FbkPrj within the supragranular layers [A(FbkPrj)] and the area occupied by blobs within the convex hull [A(blobs)]. The fraction of cortical surface occupied by blobs was computed as A(blobs)/A(convex hull). The ratio of feedback clusters projecting to blob regions was computed as A(FbkPrj AND blobs)/A(FbkPrj), where “FbkPrj AND blobs” stands for the regions of overlap of the FbkPrj and blobs. This procedure ensured that the fraction of feedback axonal clusters that overlapped blob regions could be compared in an unbiased manner with the fraction of blob regions in V1 independently of the threshold we chose to define these blob regions, as long as the threshold was not extremely high or low.
Anisotropy of feedback axon terminals
To quantify the anisotropy in spatial distribution, we first computed the center of mass of each patch. The set of axonal clusters was reduced to the set of the corresponding coordinates of the centers of mass (one data point represented each cluster in each slice). For the computation that followed, see the legend of Figure 12.
To study the relationship of the FbkPrj from V2 to the functional maps in V1, we applied targeted tracer injections to owl monkeys. First, using optical imaging of intrinsic signals, we obtained orientation maps from the vicinity of the dorsal V1/V2 border. We then injected BDA into selected orientation preference domains in V2, in regions representing visual field eccentricities of 5-10°.
Layering and clustering of the feedback projections
Figure 3, A and B, presents histological patterns of the FbkPrj obtained from two monkeys. To relate the different histological sections to cortical layers, we used the sections that were processed for CO and referred to the features of CO histology as a function of cortical layers in V1: the blob pattern in the layers 2-3, dense staining in V1 and high contrast of the V1/V2 border within layer 4, and the blob pattern in layers 5-6 [we use Haessler's definition of layers; see Casagrande and Kaas (1994)]. Diffuse axonal labeling was found in layer 1. Clusters of FbkPrj were found in layers 2, 3A, and 5/6. In contrast, the intrinsic HrzCon obtained from a control injection within V1 projected to layers 2-3, 4, and 5-6. They were particularly strong in layer 3, including the lower part of this layer. Figure 4 depicts the HrzCon in the middle layers of V1 at cortical depths of 475-875 μm, in which no FbkPrj were detected.
The projections from V2 to V1 showed uneven distribution parallel to the cortex in both the supragranular and the infragranular layers (Fig. 3). To better demonstrate the clustering of axonal FbkPrj in the supragranular layers, Figure 5 presents magnified versions of the two top sections of Figure 3. Fine clusters of axons were detected with diameters in the range of 200-400 μm, separated by zones free of terminals. The clusters, however, were not as dense as those of the intrinsic connections from the control injection within V1 (Fig. 4).
The two data sets presented in Figures 3 and 5 differed with respect to staining of cell bodies in V1. The data set presented in Figures 3A and 5A was obtained from a monkey in which axonal projections as well as cell bodies were stained in V1. In contrast, no cell bodies were stained in V1 of the monkey whose data are presented in B of both figures. Nevertheless, the features of layering and clustering of the axonal terminals in V1 were similar in both monkeys.
Figure 6 demonstrates the delineation of clusters of the FbkPrj, using the magnified patterns presented in Figure 5 (for a similar demonstration, see Fig. 1E; for the criteria used to define the clusters, see Materials and Methods).
Figure 7 presents the layering of the clusters of axonal FbkPrj (in orange) and the cell bodies (in blue, + symbols) stained in V1 from all six monkeys. Each histological section is represented in a box. The boxes are organized in columns, one column per monkey. The vertical dimension represents cortical depth, whereas the horizontal arrangement of the columns is according to the distance of the injection from the V1/V2 border. The horizontal dark lines at cortical depths of 700-1100 μm represent the approximate range of layer 4, determined according to CO histochemistry. Clusters of axonal projections were encountered at cortical depths of 100-300 μm from the surface and 0-450 μm from the bottom edge of layer 4, corresponding to layers 2, 3A, and 5/6. Diffuse axonal projections were seen at cortical depths of 0-100 μm, corresponding to layer 1. Monkey M6 was exceptional, with additional clusters between 50 and 100 μm and weak axonal projections deeper than 300 μm. Note that the injection site in this monkey was the closest to the V1/V2 border, where FbkCon tend to project via the gray matter and to have features similar to the HrzCon (Salin and Bullier, 1995) (in the following computations, the clusters considered for this monkey were restricted to those detected at its densely stained sections, 50-250 μm below the pial surface). No axonal projections were encountered within layer 4 in any of the monkeys. Cell bodies were stained mainly at cortical depths of 200-800 μm. No cells were identified in V1 of monkeys M4 and M6.
Having found clusters of the FbkPrj, we went on to determine whether these clusters were related to the CO-dense structures and to the orientation map in V1.
The injection sites and feedback axon terminals in relation to cytochrome oxidase-dense structures
Table 1 summarizes the CO stripe in V2 into which each of the injections was administered. Figure 8 presents the superposition of the effective-uptake zones and feedback axon terminals from two cases. Two additional cases are described in Figures 9 and 10. All six effective-uptake zones overlapped a pale CO stripe in a region adjacent to a thick stripe. In four cases (M1, M2, M3, and M5) (Figs. 10C, 8C, 9A, 8A, respectively) they partially overlapped a thick stripe. Five of the uptake zones overlapped a pale stripe medial to a thick stripe (Xu et al., 2004). The sixth (from M4) was in a pale stripe lateral to a thick stripe.
The largest fraction of the feedback axon terminals in each case overlapped with interblob regions in V1 (Figs. 8, 9, 10). In all monkeys, the fraction of FbkPrj that overlapped interblob regions was larger than the fraction of interblob regions in V1 (Table 1). The corresponding mean fraction of FbkPrj over the six data sets was larger than the mean fraction of interblob regions in V1 (p < 0.02; one-tailed t test; n = 6).
Orientation bias of the feedback projections
Consistent with previous reports (Malach et al., 1994; Xu et al., 2004), orientation-selective domains in V2 were organized so that highly selective regions were centered on thick CO stripes, and regions of broad orientation selectivity or no detected selectivity were centered on thin CO stripes. However, the orientation domains appeared to ignore borders between thick and pale stripes. The map in V1 contained regions where orientation was organized radially (pinwheels) and linearly (linear zones). As reported in tree shrews (Bosking et al., 1997), cats (Shmuel and Grinvald, 2000), and owl monkeys (Xu et al., 2004), linear zones were observed along the V1/V2 border, where elongated isoorientation domains tended to intersect the border at right angles.
To establish whether the FbkPrj from V2 were related to the orientation maps in V1, we first examined the periodicity of the two domains. The autocorrelation functions of the FbkCon from all six monkeys demonstrated peaks in addition to the expected peak centered on the origin. The monkey-specific mean cycle of the projections was comparable with the corresponding cycle of the orientation map (Table 2).
To further examine the relationship of the FbkPrj to the orientation maps, the anatomical patterns were aligned with the corresponding functional maps. Figure 9A depicts the CO histology from M3. Figure 9B presents a differential orientation map for the left-oblique and right-oblique orientations from the same animal. The two images (Fig. 9A,B) are aligned. The delineations of the injection site (in red) and the FbkPrj (in orange) are superimposed. The core of the injected tracer encircles a dark zone (i.e., a zone with preference for the left-oblique orientation). The axonal FbkPrj tended to innervate dark zones in the orientation map of V1, with similar preference for the left-oblique orientation. To quantify the phenomenon, we computed histograms of the differential orientation values corresponding to the injection site (Fig. 9C, red) and the feedback clusters (in orange). Both histograms were skewed toward negative values of differential orientation, corresponding to preference for the left-oblique orientation.
The histograms in Figure 9D are the average normalized histograms across all six monkeys. For each monkey, the orientations (vertical vs horizontal or left-oblique vs right-oblique) to be compared were chosen according to the dominant orientation within the injection uptake zone. Negative differential orientation values were assigned to the monkey-specific dominant orientation. The average histogram of differential orientation values obtained from the injection points was, as expected, skewed toward negative values (p < 0.0001; Pearson's χ2 goodness of fit). The average histogram obtained from the axonal FbkPrj was also skewed toward negative values (p < 0.005, using the same statistical test). We concluded that the axonal FbkPrj terminate preferentially within regions in V1 with preferred orientation similar to that of their cells of origin in V2.
Figure 10 presents a typical case. Figure 10A depicts a differential map of vertical versus horizontal orientations obtained from M1. Both the core of the tracer injection and the FbkPrj occupied mainly dark zones on the map, exhibiting a preference for vertical orientation. To further evaluate the relationship of the FbkPrj to the orientation map in V1, we superimposed the projections onto a polar representation of the map (Fig. 10B). The core of the injection site occupied a region with preference for vertical orientation (in green) and a smaller region with preference for the right oblique (in yellow). A fraction of the FbkPrj overlapped regions with preference for the left oblique (in blue). The FbkPrj tended to innervate zones with similar orientations, although not exclusively.
To examine whether the spatial relationship of the FbkPrj to the orientation maps was random, we compared the goodness of fit of the histograms of the corresponding orientation values to a uniform distribution with that of a unimodal model (Table 3). For each monkey, the unimodal distribution fitted better; the mean error corresponding to a uniform model was more than double the error for a unimodal model. Table 4 presents the monkey-specific means of the unimodal distributions fitted to the histograms of orientation values corresponding to the injection site and FbkPrj. The mean orientation corresponding to the FbkPrj was correlated to that obtained from the injection site (linear regression; p < 0.06; slope of 0.98; n = 6).
Figure 10D presents the mean histograms of orientation values obtained from all six monkeys. Before the average histogram was computed, the histograms obtained from individual monkeys were aligned according to the mode of preferred orientation at the injection site. As expected from the alignment procedure, the mean histogram obtained from the injection sites peaked at zero. The mode of the histogram of orientation preference of the feedback clusters coincided with the mode corresponding to the injection site. The values of both histograms decreased with increasing orientation difference from the preferred orientation at the injection site. The distribution of orientation values corresponding to the FbkPrj was broader than that of the values obtained from the injection sites. Average portions of 84.8% of the effective-uptake zones and 60.6% of the FbkPrj overlapped orientation domains within ±45° of the mean preferred orientation of the cells of origin (differing significantly from an even distribution; p < 0.001 and p < 0.05 for the uptake zones and FbkPrj, respectively; Pearson's χ2 goodness of fit). Overall, the FbkPrj preferentially innervated orientation domains with preferred orientation similar to that of their cells of origin; this preference, however, was not exclusive, and the distribution of corresponding orientation values was broad.
Retinotopy of owl monkey V1
To determine whether the FbkPrj are related to the retinotopic map in V1, we first obtained retinotopic maps from two monkeys. Using the set of stimuli and procedure described in Figure 11C, we imaged for horizontal and vertical strips in the visual space. The anatomical study took place in the vicinity of the V1/V2 border, at an eccentricity of 5-10°. In that region, horizontal strips in the visual space mapped to bands in V1 that were approximately orthogonal to the V1/V2 border (Fig. 11A) and vertical strips mapped to bands approximately parallel to that border (Fig. 11B). Using the conventions of medial up, lateral down, and anterior to the left, the right oblique in the visual space corresponds to the orange axis, and the left oblique corresponds to the brown axis (Fig. 11D).
The retinotopy maps obtained from the two monkeys at the region of interest were almost indistinguishable (for the results from M1, see supplemental Fig. 2, available at www.jneurosci.org as supplemental material). We could therefore construct a model for retinotopy of that region (Fig. 11D). The respective mappings of the horizontal and vertical retinotopic axes were not exactly orthogonal and parallel to the V1/V2 border. An average deviation of 12° was found in the vicinity of the border, at an eccentricity of 5-10° (Fig. 11D) [for similar deviations in owl and squirrel monkeys, see Allman and Kaas (1971) and Blasdel and Campbell (2001), respectively]. Within the same region, the mean ratio of magnification factors that corresponded to the vertical and horizontal domains obtained from the two monkeys was 1.15, comparable with the value of 1.1 reported in New World monkeys (Blasdel and Campbell, 2001; Sincich and Blasdel, 2001). This ratio implies almost isotropic mapping of the visual space, as illustrated in Figure 11D. Having obtained this information, we could address the question regarding the relationship of the FbkPrj to the retinotopic map in V1.
Axial specificity of the feedback projections
The average (across injections) distance between the most distant patches of the FbkPrj was 2.82 ± 0.39 mm (mean ± SD; n = 6). Based on the retinotopic map at eccentricities of 5-10° (Fig. 11), this distance corresponded to 4.63 ± 0.64° in the visual field. This extent of FbkPrj is comparable with the upper range of the distance between clusters of HrzCon within V1 of New World monkeys (3.0 mm measured end to end) (Fig. 4) (see also Sincich and Blasdel, 2001).
To characterize the relationship of the FbkPrj from V2 to the map of visual space in V1, we first computed the vectorial sum of preferred orientations at the area occupied by the effective tracer-uptake zone. The polar plot in the left panel of Figure 12A shows the mean preferred orientation (73.9° in this instance; red bar) and the tuning (dark curve) of preference found at the injection site in V2. The middle panel presents the spatial distribution of the FbkPrj in orange. The dark dotted line marks the V1/V2 border, used here as a reference to relate the spatial distribution of the axonal projections to the retinotopic map. The red bar represents the V1 mapping of a retinotopic axis that corresponded to the mean preferred orientation at the injection site (73.9° in A). The spatial distribution of the FbkPrj was anisotropic, with the major axis of anisotropy (open double-arrowed bar) approximately parallel to a retinotopic axis that corresponded to the preferred orientation at the injection site in V2. The right panel presents a measure of anisotropy as a function of retinotopic axis (for the definition of the anisotropy measure, see the figure legend). The dark and bright green lines mark the 0.02 and 0.10 p values, respectively, for testing whether the measured anisotropy could be obtained from an isotropic distribution. The maximal value of anisotropy in Figure 12A (2.29) was obtained at a retinotopic axis of 59° (orange arrow), ∼15° away from the mean preferred orientation at the injection site (red bar). The anisotropy value obtained at the retinotopic axis corresponding to the preferred orientation at the injection site (73.9°) was 2.01, twice the value expected from an assumed isotropic distribution of the FbkPrj. Figure 12, B, C, E, and F present four more instances of axial specificity of the FbkPrj, for approximate axes of 25, 15, 0, and -50°, respectively. Similar to the phenomenon demonstrated in Figure 12A, the spatial distributions of these FbkPrj were anisotropic, with the major axis of anisotropy corresponding to the mean preferred orientation at the effective tracer-uptake zone. The mean preferred orientations at the injection sites demonstrated in B and C (∼25 and ∼15°, respectively) were approximately orthogonal to the one presented in F (approximately -50°). Consistent with the concept of axial specificity, the major axes of anisotropy of the FbkPrj in B and C were approximately orthogonal to the one in F.
Table 5 summarizes the findings on axial specificity of the FbkPrj. In Figure 12, the difference between the retinotopic axis in which maximal anisotropy was obtained and the retinotopic axis corresponding to the preferred orientation at the injection site was, in order from A to F, 14.9, -16.7, -11.8, -55.6, -10.3, and 17.1, respectively (14.2 ± 2.9°; mean ± SD of the absolute difference, excluding the data from Fig. 12D). The retinotopic axis in which maximal anisotropy was obtained was correlated with the preferred orientation within the effective tracer uptake zone (r2 = 0.705, slope of 1.01, p < 0.037, n = 6, including all the data; r2 = 0.890, slope of 1.04, p < 0.015, n = 5, excluding the data in Fig. 12D; linear regression).
The data presented in Figure 12D were exceptional, because the retinotopic axis of maximal anisotropy differed in this case by as much as 55° from the mean preferred orientation at the corresponding injection site. In addition, for the case presented in Figure 12E, the computed anisotropy at the retinotopic axis corresponding to the preferred orientation at the injection site (1.52) was smaller than in the other cases. Careful examination revealed, however, that, in both of these instances, (1) the mean preferred orientation at the two injection sites (9.4 and 1.7°) was close to 0°, (2) of all data sets, these injection sites were the closest (510 and 420 μm, respectively) to the V1/V2 border, and (3) no space was left between the most anterior FbkPrj and the V1/V2 border, as if the FbkCon would be able to spread more anteriorly if the V1/V2 border was not there (compare with all other data sets, especially with Fig. 12A,F). Based on the retinotopic map of areas V1 and V2 and items 1-3 above, if the concept of axial specificity is valid, we could expect some of the FbkPrj from these two injections to fall in the region of the representation of the vertical meridian, which in part falls within the ipsilateral representation of the other hemisphere. Our interpretation of the data is that most probably part of the FbkPrj crossed the corpus callosum and formed additional patches in V1 of the other hemisphere. Had we taken the “missing” clusters into account, we would probably have obtained more pronounced anisotropies along the retinotopic axis that corresponded to the mean preferred orientations (∼0°) within the two corresponding injection sites. We believe that, with this plausible interpretation, all six data sets are consistent with the concept of axial specificity of the FbkPrj.
Differentiating feedback projections from intrinsic connections
Although cells in V1 were retrogradely labeled in some of our experiments, the stained axonal fibers in V1 are mainly of FbkPrj rather than collaterals of V1 neurons, as indicated by the following. First, the layering pattern of the axonal projections (Fig. 7) followed the “rules” expected from FbkCon (Rockland and Pandya, 1979) and the findings of previous studies (Rockland and Virga, 1989). Second, the layering pattern of axonal projections resulting from the injections in V2 differed from that of the control injection within V1. Third, the vast majority of cell bodies were identified at cortical depths lower than the feedback axon terminals within the supragranular layers (Fig. 7). If these axon terminals had originated from the cells stained in V1, we would have expected additional axonal clusters of HrzCon at the cortical depths at which the cells were identified. Fourth, in two of the monkeys injected in V2, we found axonal clusters in V1 with not even a single stained cell body. This result indicates that, using the methods we applied, the anterograde transport from V2 cells to their axonal projections in V1 was faster and more effective than the retrograde transport from axons in V2 to their parent cells in V1, let alone to the even more distant axon collaterals of the parent cells within V1 (additional reasons to support our interpretation are included in the supplementary material, available at www.jneurosci.org). We conclude that the vast majority of labeled axonal clusters in V1 are FbkPrj originating from cells within the injection sites in V2.
Clustering of the feedback projections
Consistent with our findings, continuous labeling can often be found in macaque V1 layer 1 after injections in extrastriate areas (Rockland and Pandya, 1979; Ungerleider and Desimone, 1986). In other layers, the FbkPrj from V2 are diffuse relative to the HrzCon, but they nevertheless demonstrate clustering (Figs. 3, 5). These results are consistent with previous reports on clustering of the projections from macaque V2 to the supragranular layers of V1 (Wong-Riley, 1979a; Livingstone and Hubel, 1984a; Rockland and Virga, 1989; Angelucci et al., 2002).
In contrast, other studies reported diffuse FbkPrj. The finding by Stettler et al. (2002) that the FbkPrj show less clustering than the HrzCon is consistent with our data. Their conclusion that the FbkPrj lack clustering might reflect this comparison rather than an absolute measure. Maunsell and Van Essen (1983) and Ungerleider and Desimone (1986) reported that continuous labeling can be found in layer 4B after injection in area MT. Those connections, however, were from areas that are distantly related to V1, both in terms of cortical space and the hierarchy of visual areas.
Spatial relationship between the feedback projections and cytochrome oxidase-dense structures
Clusters of FbkPrj from regions overlapping pale and thick stripes in V2 project preferentially to interblobs in owl monkey V1. Similarly, specific connections run from V2 interstripes to interblobs in macaque V1 (Livingstone and Hubel, 1984a). Based on physiological and anatomical specificity, it was proposed that color is processed in the blobs/thin-stripes system in the macaque (Livingstone and Hubel, 1984a). Interestingly, the FbkPrj from interstripes project preferentially to interblob regions in both species, despite the difference in color vision between them. Moreover, this pattern of connectivity prevails in owl monkeys, although orientation specificity and eye dominance do not vary significantly across functional CO compartments in V1 (O'Keefe et al., 1998).
Although the injections were not confined to the centers of either pale or thick stripes, they were all targeted to orientation-selective domains, which appear to ignore borders between these stripes (Malach et al., 1994; Xu et al., 2004). This feature of orientation mapping might explain the consistent preference of the FbkPrj for termination in interblob regions across all injections in our study.
Orientation bias of the feedback projections
Approximately 60% of the feedback clusters innervated orientation domains within ±45° of the preferred orientation of the cells of origin (Fig. 10D). In comparison, ∼70% of the HrzCon in macaque V1 connect neurons whose difference in preferred orientation is within ±45° (Malach et al., 1993). In tree shrew V1, 57.6% of the HrzCon innervate sites whose difference in preferred orientation is within ±35° (Bosking et al., 1997). Thus, the degree of specificity obtained here for clusters of FbkPrj appeared lower than those reported for the HrzCon in macaque and tree shrew V1.
Although the mean preferred orientation was biased toward values similar to those within the injection sites in V2, there was some variability over the data sets (Table 4). The absolute deviation between the mean preferred orientation of the cells of origin and that of the FbkPrj was 29.5 ± 17.6° (mean ± SD). The observed variability and deviations might reflect the extent of orientation preference of the FbkPrj. Alternatively, they could be the result of residual errors in aligning the histological data with the cortical images, and the relatively short cycle of orientation maps in owl monkeys (Table 2).
Stettler et al. (2002) questioned the possibility that the FbkCon from V2 to macaque V1 show orientation preference. The difference between their findings and ours might reflect differences between species or in methodology. A recent abstract (Angelucci et al., 2003) on a study in marmosets reported orientation preference of the FbkPrj from V2 to V1 similar to that we found in owl monkeys.
Axial specificity of the feedback projections
The spatial distribution of the FbkPrj within V1 is anisotropic, with the principal axis of elongation corresponding to the preferred orientation of the V2 cells of origin. A recent abstract (Angelucci et al., 2003) on a study in marmosets supports our finding of axial specificity of the FbkPrj. In V1 retinotopic coordinates, therefore, the FbkPrj from V2 project farthest along axes representing collinear contours in visual space.
We are unaware of any previous demonstration of anisotropy in the visual space or axial specificity of the cortical FbkCon. Similar phenomena have been reported, however, for the HrzCon in tree shrews (Fitzpatrick, 1996; Bosking et al., 1997) and New World monkeys (Sincich and Blasdel, 2001). The average quantitative anisotropy in tree shrews is 1.97 (Bosking et al., 1997, their Fig. 11), whereas the corresponding value in New World monkeys is 1.7 compared with 1.70 ± 0.19 obtained here for the FbkCon (Fig. 12, Table 5). Comparable elongations (ratios of 2.1 and 2.4) were found for the FbkPrj from areas 17 and 18 to cat dorsal LGN (Murphy et al., 1999) (these elongations, however, are either retinotopically parallel or orthogonal to the preferred orientation of the cortical parent cells).
Previous studies in macaques indicated elongation of the HrzCon within V1. In most cases, these elongations are perpendicular to the ocular dominance columns, with an aspect ratio of 1.69:1 (Malach et al., 1993), 1.78:1 (Yoshioka et al., 1996), or 1.56:1 (Angelucci et al., 2002). All of these authors interpreted the elongations as a consequence of the difference in magnification between the vertical and the horizontal domains (ratio of ∼1.6), possibly resulting from duplication of the retinotopic map by the two sets of ocular dominance columns along the representation of the vertical meridian (Hubel and Wiesel, 1977; Van Essen et al., 1984; Tootell et al., 1988; Blasdel and Campbell, 2001).
In the present work, we confined the anatomical study to a region in which the map of the visual space was reproducible and predictable. The corresponding average ratio of magnifications in owl monkeys is 1.15 (Fig. 11). Thus, the anisotropies of FbkPrj found here, in the range of 1.5-2.3, cannot be accounted for by different magnifications of the retinotopic axes. Moreover, three of the five instances of axial specificity were demonstrated along a retinotopic axis closer to horizontal than to vertical (Fig. 12). Thus, the axial specificity shown here for the FbkCon is not an epi-phenomenon of the different magnifications along the major retinotopic axes.
Possible functional correlates of the specificity of the feedback projections
The elongated, orientation-biased FbkPrj from V2 spread farther than the dendritic arborizations and the minimum response fields and are likely to extend to the low-contrast spatial summation fields of neurons in V1 (Angelucci et al., 2002). However, because the extent of the clustered FbkPrj from V2 is comparable with that of the HrzCon in V1, these specific FbkPrj cannot account spatially for the long-range suppressive center-surround interactions demonstrated by V1 neurons (Levitt and Lund, 2002). In addition, although the feedback signal, depending on the context, could mediate suppressive effects (Ramsden et al., 2001), the feedback pathways to V1 carry mainly excitatory input (Shao and Burkhalter, 1996) projected preferentially to pyramidal cells (Johnson and Burkhalter, 1996). These FbkPrj link cells organized approximately linearly within the cortical manifold, with non-overlapping RFs and with an orientation bias (Figs. 9, 10) aligned to the axis of elongation (Fig. 12). Thus, these FbkPrj might play a role in mediating contour integration as well as facilitation across approximately collinear segments along short distances beyond the classical RFs of neurons in V1.
Consistent with this model, inactivation of V2 reduces the response of neurons in V1 to a visual stimulus within the RF (Sandell and Schiller, 1982; Mignard and Malpeli, 1991). Similarly, diminution of the responses is the most frequent effect of V2 inactivation on neurons in V1 stimulated with a bar at their preferred orientation (Bullier et al., 2001). Overall, the phenomena described here are ideally suited for collinear facilitation and for highlighting coherent contours within a visual image.
↵a These nocturnal simians have a rod-dominated retina and a single cone type; thus, they lack color vision. Despite these differences from diurnal primates, the CO cytoarchitecture of area V1 in owl monkeys is characterized by a mosaic of blobs and interblobs, and area V2 is characterized by a regular system of thick, pale, and thin CO stripes (Tootell et al., 1985). V1 and V2 in owl monkeys are visuotopically organized (Allman and Kaas, 1971, 1974). For a review of the visual system of owl monkeys, see Casagrande and Kaas (1994).
This study was supported by grants from the Goldsmith, Glasberg, and Minerva Foundations, by the Israel Science Foundation, by M. M. Enoch (A.G., R.M., and S.U.), and by a long-term fellowship from the European Molecular Biology Organization (A.S.). We thank L. Sincich and the anonymous reviewers for their helpful comments on a previous version of this manuscript, A. Schüz for helpful discussions, and S. R. Smith for scientific editing. R. Eilam, S. Leytus, and D. Ettner provided excellent support. This study is dedicated to the memory of Chaipi Wijenbergen, whose innovation and support made it possible.
Correspondence should be addressed to Dr. Amir Shmuel, Max-Planck Institute for Biological Cybernetics, Spemannstraße 38, 72076 Tuebingen, Germany. E-mail:.
Copyright © 2005 Society for Neuroscience 0270-6474/05/252117-15$15.00/0
↵* R.M. and A.G. contributed equally to this work.