The directionality of corticocortical projections is classified as feedforward (going from a lower to higher hierarchical levels), feedback (interconnecting descending levels), and lateral (interconnecting equivalent levels). Directionality is determined by the combined criteria of the laminar patterns of the axon terminals as well as the cells of origins and has been used to construct models of the visual system, which reveals a strict hierarchical organization (Felleman and Van Essen, 1991; Hilgetag et al., 1996a). However, these models are indeterminate partly because we have no indication of the distance separating adjacent levels. Here we have attempted to determine a graded parameter describing the anatomical relationship of interconnected areas. We have investigated whether the precise percentage of labeled supragranular layer neurons (SLN%) in each afferent area after injection in either visual areas V1 or V4 determines its hierarchical position in the Felleman and Van Essen (1991) model. This shows that pathway directionality in the Felleman and Van Essen model is characterized by a range of SLN% values. The one exception is the projection of the frontal eye field to area V4, which resembles a feedforward projection. Individual areal differences in SLN% values are highly significant, and the number of hierarchical steps separating a target area from a source area is found to be tightly correlated to SLN%. The present results show that the hierarchical rank of each afferent area is reliably indicated by SLN%, and therefore this constitutes a graded parameter that is related to hierarchical distance.
Rostral directed projections allow outflow of activity away from striate cortex [visual area V1] and are thought of as feedforward (FF) pathways. These projections originate largely from supragranular layers, target layer 4, and contrast with the reciprocal projections that originate in infragranular layers, terminate outside of layer 4, and are thought of as feedback (FB) pathways (Kuypers et al., 1965; Cragg, 1969; Spatz et al., 1970; Tigges et al., 1973, 1981; Lund et al., 1975; Kaas and Lin, 1977; Spatz, 1977;Wong-Riley, 1978; Van Essen and Zeki, 1978; Rockland and Pandya, 1979;Wall et al., 1982; Maunsell and Van Essen, 1983; Weller et al., 1984;Kennedy and Bullier, 1985; Weller and Kaas, 1985; Barbas, 1986, 1995;Andersen et al., 1990; Boussaoud et al., 1990; Morel and Bullier, 1990;Baizer et al., 1991; Colby and Duhamel, 1991; Webster et al., 1991,1994; Sousa et al., 1991; Distler et al., 1993; Nakamura et al., 1993;Barone et al., 1995; Barbas and Rempel-Clower, 1997; Felleman et al., 1997; Gattass et al., 1997).
The laminar patterns of corticocortical connections have been used to propose a hierarchical ranking of primate cortical areas in different sensory systems (Fitzpatrick and Imig, 1980; Friedman, 1983; Maunsell and Van Essen, 1983; Barbas, 1986; Pons and Kaas, 1986; Ungerleider and Desimone, 1986; Colby et al., 1988; Boussaoud et al., 1990; Van Essen et al., 1990; Felleman and Van Essen, 1991; Webster et al., 1991;Young, 1992; Distler et al., 1993; Webster et al., 1994; Hilgetag et al., 1996a; Barbas and Rempel-Clower, 1997; Felleman et al., 1997b; Kaas et al., 1999).
These models are constructed by categorizing connections between huge numbers of pairs of areas and ascribing each area to a hierarchical level in an optimal configuration. However, the number of possible configurations is enormous because of lack of criteria defining the distance separating levels (Hilgetag et al., 1996a,b). Here we have sought to overcome this problem by defining a single quantitative parameter of connectivity that will allocate areas to graded levels. Such a parameter could be provided by the proportion of supragranular layer neurons (SLN%) participating in FF and FB pathways (Kennedy and Bullier, 1985; Barbas, 1986; Barone et al., 1995;Rockland, 1997; Batardière et al., 1998a).
Quantitative techniques were used to investigate SLN% in individual areas projecting to visual areas V1 and V4 in macaque. Area V1, the primary visual area, is interconnected to areas in both the ventral and dorsal stream. Area V4 is a higher order visual area in the ventral stream and receives FF, lateral, and FB projections from higher order areas in the dorsal stream and ventral stream.
The SLN% alone successfully ranks cortical areas and is tightly correlated to the number of steps separating areas. The present findings, showing a graded parameter tightly correlated to hierarchical rank, indicate the existence of a hierarchical distance rule (Kennedy and Bullier, 1985; Rockland, 1997), which will allow an improved exploration of the organizational constraints of the interareal relationships in the visual system.
MATERIALS AND METHODS
Twenty-two retrograde tracer experiments were performed on 12 cynomolgus monkeys (Macaca fascicularis; Table1).
Anesthesia and surgery. After premedication with atropine (1.25 mg, i.m.) and dexamethasone (4 mg, i.m.), monkeys were prepared for surgery under ketamine hydrochloride (20 mg/kg, i.m.) and chlorpromazine (2 mg/kg, i.m.). After intubation, anesthesia was continued with halothane in a N2O/O2 (70:30) mixture. Heart rate was monitored, and artificial respiration was adjusted to maintain the end-tidal CO2 at 4.5–6%. The rectal temperature was maintained at 37°C. All the procedures used follow the National and European regulations concerning animal experiments and have been approved by the authorized national and veterinary agencies.
Injection of retrograde tracers. Single injections of retrograde fluorescent tracers [fast blue (FsB) and diamidino yellow (DY), 3% in H2O] were made by means of Hamilton syringes. In three cases [M56 right hemisphere (RH) FsB, M56 RHDY, M73 left hemisphere (LH)] multiple injections were made in area V1 (Table 1). In the remaining 19 cases, injections of tracers spanned 1–5 mm and were made in a stereotypical manner. Injections were made in area V4 according to the definition of Desimone and Ungerleider (1986). They were centered on the prelunate gyrus between the lunate sulcus, the infero-occipital sulcus, and the superior temporal sulcus in area V4 containing the representation of the central visual field (Maguire and Baizer, 1984; Gattass et al., 1988). Injections aimed at area V1 were made on the operculum in area V1 containing the representation of the central visual field (Hubel and Wiesel 1974; Van Essen et al., 1984). When an FsB and a DY injection were made side-by-side, they were separated by ∼3 mm. Elsewhere we have characterized the uptake zone of FsB and DY tracers (Bullier et al., 1984; Kennedy and Bullier, 1985). Examination of sections at regular intervals throughout the injection site makes it possible to determine those restricted to gray matter. All the injections but one was restricted to the cortical gray matter. The one injection that involved the white matter (BB135) returned similar values to the other injections, which confirms previous results (Barone et al., 1995;Batardière et al., 1998a). After injections, bone flaps were closed, and the scalp was stitched back into position.
Histological processing. After a survival period of 10–13 d, animals were deeply anesthetized before being perfused transcardially with 200 ml of 2.7% saline, 1–3 l of 8% paraformaldehyde/0.5% glutaraldehyde mixture in phosphate buffer (0.1m, pH 7.4), 0.5 l of 10% sucrose, 0.5 l of 20% sucrose, and 1 l of 30% sucrose in phosphate buffer (0.1m, pH 7.4). Brains were immediately removed, blocked, and horizontal or parasagittal sections (Table 1) were cut on a freezing microtome (section thickness, 40 μm). One section in three was mounted in saline onto gelatinized slides. Sections at regular intervals were reacted for cytochrome oxidase (Silverman and Tootell, 1987) and acetylcholinesterase activity (Hardy et al., 1976; Mesulam and Geula, 1994).
Examination of material. Sections were observed in UV light with oil-immersion objectives using a Leitz fluorescent microscope equipped with a D-filter set (355–425 nm). The properties and description of neurons labeled with FsB and DY are described by Keizer et al. (1983). Neurons labeled by DY exhibit a yellow nucleus, whereas neurons labeled by FsB exhibit a blue coloration in their cytoplasm. Labeled cells are observed in both infragranular and supragranular layers. An x–y plotter electronically coupled to the microscope stage was used to trace out sections and to record the position of labeled neurons. After observation, sections were counterstained with cresyl violet and projected onto charts of labeled neurons to relate the position of labeled neurons to histological borders.
Areal and laminar distribution of labeled neurons. The areal extent of a population of retrogradely labeled neurons in a cortical area after injection in the target area is referred to as a projection zone (Fig. 1 A–C). The proportion of supragranular layer neurons falls off from a peak in the center of the projection zone to minimal values in the periphery (Fig.1 D) (Meissirel et al., 1991; Barone et al., 1995;Batardière et al., 1998a). This, coupled with the curvature of the cortex, necessitates estimating the relative proportions of large numbers of labeled neurons in supragranular and infragranular layers by counting neurons at regular intervals throughout each projection zone in each of the visual areas studied (Barone et al., 1995;Batardière et al., 1998a). The laminar distribution for each projection zone is expressed as the SLN% with respect to the overall population of infragranular and supragranular labeled neurons.
Criteria for the location of cortical areas. Injections of tracers in areas V1 and V4 leads to dense labeling of an extensive region of extrastriate cortex in the parietal and temporal regions (Zeki, 1978; Maunsell and Van Essen, 1983; Kennedy and Bullier, 1985;Yukie and Iwai, 1985; Perkel et al., 1986; Tanaka et al., 1990;Felleman and Van Essen, 1991; Baizer et al., 1991; Krubitzer and Kaas, 1993; Shipp and Zeki, 1995). Label in extrastriate cortex was observed in different known visual areas: V1, V2, V3, V3A, V4, middle temporal area (MT), fundus superior temporal area (FST), temporal occipital area (TEO), temporal area (TE), lateral intraparietal area (LIP), and TH–TF, as well as frontal eye field (FEF) (Fig.2). Multiple criteria were used to allocate labeled neurons to one of these 12 areas. It was important to optimize the criteria used to distinguish different cortical areas to be able to count neurons throughout a maximum extent of the projection zones in individual areas. Some architectonic limits were obtained using Nissl staining, cytochrome oxidase, or acetylcholinesterase histochemistry, but a major criteria is the pattern of labeling itself. Because the injection sites involved cortex containing the representation of the central visual field, cortical areas that share borders where the far periphery of the visual field is represented show a discontinuous pattern of labeling. This gap in the labeling provides an important indication of the limits of the cortical area (Fig.3).
Area V1 is located in the posterior part of the brain, and the limits with area V2 are easily identified using cresyl violet staining. The location of extrastriate areas is shown in Figure 2. Area V2 is located in the posterior bank of the lunate sulcus (Van Essen and Zeki, 1978;Gattass et al., 1981) where it can be identified with cytochrome oxidase and acetylcholinesterase histochemistry (Tootell et al., 1983;Livingstone and Hubel, 1984; Barone et al., 1994).
Area V3 is located laterally in the fundus of the lunate sulcus and more medially in the posterior part of the anectant gyrus, whereas area V3A is located anterior to V3 in the anterior bank of the lunate sulcus (LS) (Zeki, 1971, 1978; Van Essen et al., 1986; Gattass et al., 1988;Felleman et al., 1997b). Mediodorsal injections in the perifoveal representation of area V1 led to restricted labeling in the anectant gyrus corresponding to area V3, as defined by Felleman et al. (1997b). This region corresponds to a subregion of the dorsomedial visual area (DM) following the definition of Beck and Kaas (1999). Labeling in area V3 is separated from labeling in area V2, which is located more medially. After injection in area V1 no labeled cells are found in area V3A in the anterior bank of the LS in agreement with previous observations (Van Essen et al., 1986; Felleman et al., 1997b). After injections in area V4, labeling was restricted to area V3A. In most cases there is a gap between the labeling in area V2 and V3A, and going from area V2 to V3A, there is a distinct increase in the density of labeling in the infragranular layers of area V3A (Fig.3 A).
Area MT is located in the posterior bank of the superior temporal sulcus (STS) and stretches from the fundus to approximately halfway up the sulcus (Zeki, 1974; Desimone and Gross, 1979; Van Essen et al., 1981; Maunsell and Van Essen, 1983; Ungerleider and Desimone, 1986). After V1 injection, the labeling in the dorsal part of the STS is well isolated from the labeling in the prelunate gyrus. This gap is more or less pronounced in cases of V4 injections, which induce a strong labeling in the adjacent area V4t (Fig. 3) (Desimone and Ungerleider, 1986; Gattass et al., 1988; Felleman and Van Essen, 1991).
After V1 and V4 injections, labeling is found in the visual motion area FST in the floor of the STS, which is anterior and ventral to area MT (Seltzer and Pandya, 1978; Desimone and Ungerleider, 1986; Ungerleider and Desimone, 1986; Boussaoud et al., 1990).
Labeling in the posterior and lateral bank of the intraparietal cortex is isolated from labeling in other areas and corresponds to the LIP (Andersen et al., 1990; Blatt et al., 1990; Boussaoud et al., 1990;Baizer et al., 1991; Colby and Duhamel, 1991; Shipp and Zeki, 1995;Colby et al., 1996).
The major input to area V4 from higher order areas is from the visual areas in the temporal lobe. Only scattered labeled cells were observed in the temporal lobe after injection in V1. Area TEO is located on the temporal lobe between the inferior occipital sulcus and the superior temporal sulcus (Iwai and Mishkin, 1969; Baizer et al., 1991; Boussaoud et al., 1991; Distler et al., 1993). Labeling is discontinuous between V4 and TEO. Anterior and ventral to TEO in the inferior temporal cortex is the temporal area TE (see also Van Essen et al., 1990; Webster et al., 1991, 1994).
In the ventral region of the temporal lobe in the parahippocampal cortex is the lateral cortical area TF and the medial area TH. These cortical areas are located medial to the rhinal fissure and posterior to the perirhinal cortex (Amaral et al., 1987; Suzuki and Amaral, 1994a,b; Suzuki, 1996). Anteriorly and medially, labeling in TF/TH shows a gap with labeling in the ventral part of areas TE at the level of the rhinal fissure (Fig. 3 C).
In the frontal cortex, labeled neurons are found systematically in the anterior bank of the arcuate sulcus, which is known to house the FEF (area 8) (Bruce and Goldberg, 1985; Huerta et al., 1987; Stanton et al., 1989).
Statistical tests. A multinomial ANOVA (Woodward et al., 1990) was used to test the hypothesis that the SLN% is equal across visual areas. Infragranular and supragranular layers were treated as within-subject factors in the analysis. By testing proportions, the problem of the variation in total number of cells is removed. The analysis does, however, incorporate the total numbers of labeled cells in the estimates of variance for each proportion, so that proportions based on small total numbers have less precision than those based on larger numbers. When a significant difference between areas was observed, the multinomial ANOVA allows us to do planned comparisons to identify the areas that violate the null hypothesis. To test the relationship between SLN% and the number of levels that separate two interconnected areas, we used the nonparametric Spearman rank correlation test.
Injections of areas V1 and V4 were performed in a stereotypical manner, they were large and spanned 1–5 mm, a method that reduces the possibility that patchiness of the target area (DeYoe et al., 1994) will influence the SLN values obtained (Scannell et al., 2000). Injections in areas V1 and V4 led to dense retrograde labeling throughout a large extent of extrastriate cortex. The criteria used to allocate neurons to individual areas is given in Materials and Methods. Retrogradely labeled neurons in the thalamus after injections in areas V1 and V4 are confined to the relevant thalamic nuclei, in the lateral geniculate nucleus and the lateral pulvinar (Kennedy and Bullier 1985; Tanaka et al., 1990; Baleydier and Morel, 1992; Shipp and Zeki, 1995).
Laminar distributions: fluctuation within the projection zone
For each cortical area and in each animal, numbers of neurons in each laminar compartment were computed for one in every three or four sections (see Materials and Methods). The regional extent of the area that contains the labeled neurons is referred as the projection zone. A representative two-dimensional reconstruction of a projection zone is shown in Figure 1 A, which illustrates the spatial distribution of labeling in area V2 after injection in area V1. At the periphery of the projecting zone, the number of labeled neurons is low and increases to reach a maximum at the center. Density profiles (Fig.1 B,C) provide a one-dimensional reconstruction of the projection zone and constitute a graphic representation of labeling, which make it possible to ensure that the appropriate sampling frequency and choice of sections have been used for each projection zone in each area. Density profiles for all projection zones were prepared that show numbers of neurons per slide going through the projection zone where the x-axis is distance in millimeters. Here and elsewhere (Barone et al., 1995;Batardière et al., 1998a) SLN% is found to vary throughout the projection zone, including the core regions (Fig.1 D). A major factor that contributes to this phenomena is the curvature of the cortex with respect to the plane of section, which results in single sections providing an uneven sampling of individual layers. Calculating SLN% at regular intervals across the projection serves to overcome this problem of cortical curvature. A representative density profile for each cortical area is shown in the right-hand side of Figures4-9. Examination of the density profiles show that peak percentages are located toward the centers of projection zones and fall off to zero in the periphery (Figs. 4-8). The density profiles illustrate the problems associated with estimating the laminar distribution. For example, because of cortical curvature and changing plane of section, peak levels of supragranular and infragranular layers frequently do not coincide (e.g., Fig. 4, top left-hand density curve; Fig. 6, three of the four density curves for MT and for LIP).
Changes of plane of section of the cortical compartments also contribute to the variability of the number of neurons in each compartment even on neighboring sections (Batardière et al., 1998a). For example, in area V2 after injection in area V1 in the case of M71 LH FsB (Fig. 4, top) individual sections in the central region show a SLN% range of 12–67%. When the SLN% is calculated throughout the projection zone, this injection returns a value of 50%. Similar results are obtained after the V4 injections (Fig. 1 D). For example, in BB119 the DY injection in area V4 (Fig. 4, bottom) leads to SLN% in the central region of the projection zone of area V3A ranging between 34 and 64%, whereas global values are 47%. These observations again demonstrate the necessity to calculate global values of SLN% throughout the projection zone to characterize the laminar distribution for a single injection (Kennedy et al., 1989; Meissirel et al., 1991; Barone et al., 1995; Batardière et al., 1998a).
Variation of the SLN% across the projection zones could also mean that individual projection zones are hybrid for FF, FB, and lateral connections, as suggested elsewhere (Ungerleider and Desimone, 1986;Andersen et al., 1990). The possibility that projection zones are hybrid reinforces the need to calculate the laminar distribution of connectivity after large injections and computing SLN% throughout the whole of the projection zone, thereby obtaining a global value for the connection being studied.
Analysis of the effect of threshold values shows that projection zone structures are reasonably robust. Figure 1 E shows the effect of taking a 10 or 20% threshold of the maximum labeling on an individual projection zone in area MT after injection in area V4. Collectively, a 10% cutoff for all projection zones pooled leads to a 3% drop in the numbers of neurons counted and 0.4% change in SLN% value.
Examination of the density profiles for projections zones in individual cortical areas makes it possible to select representative charts of labeled neurons from the center of the projection zones (Figs. 4-8,left-hand side).
Reliability of allocation of labeled neurons to individual cortical areas
In the present study we have used the pattern of labeling as a major criteria for the allocation of neurons to individual areas. Here we shall examine the reliability of this method. Injections of areas V1 and V4 were performed in a stereotypical fashion and concerned the representation of the central visual field. This means that one can expect a gap of labeling between adjacent areas when they share a border representing the peripheral visual field. In the present study, injection of area V1 led to weak or no label in area V3A, and injection of area V4 gave weak label in area V3, confirming previous findings (Van Essen et al., 1986; Felleman et al., 1997b). This means that injections in area V1 led to extrinsic label in areas V2, V3, V4, MT, FST, LIP, TE, TEO, TH–TF, and FEF. Injections in area V4 led to labeling in areas V2, V3A, V4t, MT, FST, LIP, TE, TEO, TH–TF, and FEF. Areas that share a border that represents the central visual field and where labeling can be continuous are V4t–V5 and V3A–V4 (Van Essen and Zeki 1978; Gattass et al., 1988; Boussaoud et al., 1991). In the present study, we do not attempt to distinguish label in V4 and V4t. This leaves the possibility of imperfect separation of labeling between V3A–V4 and V4t–MT. This does not influence our main conclusions in the present study for the following reasons. First, because labeling was continuous in V4t–MT after one of seven injections, we can therefore discount an effect of continuous labeling on estimates of SLN% in area MT. Second, although labeling in V3A–V4 was continuous, this concerned a minority (three of seven) of injections. Quantitative analysis of the V3A projection zones, as illustrated in Figure 1, shows that a 20% cutoff for this area leads to a 2% change in SLN% accompanied by a 1.6 mm shift of the V3A border. This would put the areal border well outside of the region that is commonly thought to be V4. In other words, if we were to repeat the analysis using a highly conservative location of V3A that would exclude V4t neurons, we would obtain very similar SLN% values for V3A to those obtained using the full projection zone as in the present study.
Distributions in areas sharing FB and FF projections to both target areas
After injections in areas V1 and V4 labeled neurons are found in similar densities in supragranular layers of area V2. This contrasts with the densities in infragranular layers, which are relatively high after injections in area V1 and virtually absent after injection in area V4 (Figs. 3 A, 4).
Quantitative results for different injections are reported in Table 1. Altogether labeling was successfully analyzed in area V2 for 11 injections in area V1 and returned a mean value of 48% SLN (range, 35–55%). Seven projection zones were analyzed in area V2 resulting from injection in area V4 and gave a mean SLN value of 93% (range, 89–99%).
In area V3, injections in area V1 led to peak densities in infragranular layers, and four injections gave a mean SLN value of 9%. Injections in V4 led to peak densities in area V3A in supragranular layers (Fig. 4), and the seven injections gave a mean SLN value of 60% (Table 1).
Figure 4 shows that those cortical areas (i.e., areas V2 and V3/V3A) that have FB projections to area V1 and FF projections to area V4 show complementary patterns of labeled neurons so that FF projections are characterized by high SLN% and FB by low SLN%.
Distributions in ventral stream areas
Areas in the temporal lobe are known to possess FB projections to both areas V1 and V4 (Morel and Bullier 1990; Distler et al., 1993;DeYoe et al., 1994; Felleman et al., 1997a), and the present results show that the density of these projections is considerably higher to area V4 than to area V1 (Fig. 5). Whereas the FB projections to area V1 from both TEO and TE are exclusively from infragranular layers, the FB projections to area V4 show large SLN% values. In TEO and TE, four projection zones from V1 injections were analyzed and returned a mean SLN% inferior to 0.5%, whereas the seven projection zones after V4 injections returned a mean SLN% of 40% in TEO and 26% in TE (Table 1).
Labeled neurons are present in the parahippocampal areas after injections in areas V1, confirming results of others (Rockland and Van Hoesen, 1994). However, these neurons were extremely sparse and exclusively located in infragranular layers (Table 1). Labeling in TH–TF was reasonably strong after injection in area V4, and labeled neurons were almost exclusively located in the infragranular layers (Fig. 5, bottom).
Distributions in FB and lateral projections from dorsal stream areas
Areas MT and FST both belong to the dorsal stream and are known to project to areas V1 and V4 (Maunsell and Van Essen, 1983; Kennedy and Bullier, 1985; Perkel et al., 1986; Ungerleider and Desimone, 1986;Tanaka et al., 1990; DeYoe et al., 1994; Rockland and Van Hoesen, 1994;Hof et al., 1996). MT is thought to have a FB projection to area V1 and a lateral projection (i.e., connecting areas on equivalent hierarchical levels) to area V4 (Maunsell and Van Essen 1983; Ungerleider and Desimone, 1986). The density and laminar distribution of the labeled neurons in MT are seen to differ to a large extent according to their target (Fig. 6, top). The FB projection from MT to area V1 showed low densities with the majority of labeled neurons principally in the infragranular layers, whereas the densities of labeled neurons projecting to area V4 were considerably higher and equally distributed in supragranular and infragranular layers. In MT, nine projection zones from V1 injections were analyzed and returned a mean SLN% value of 5%, whereas the seven projection zones after V4 injections returned a mean SLN% value of 47% (Table 1).
Area FST possesses FB projections to both areas V1 and V4 (Boussaoud et al., 1990; Tanaka et al., 1990; Felleman et al., 1997a). The density of the FB projection to area V4 was higher than that for the projection to V1, and whereas all the neurons projecting to area V1 were located in infragranular layers, some labeled neurons were located in supragranular layers after injection in area V4 (Fig. 6,middle). In FST, the seven projection zones after V4 injections returned a mean SLN of 14% (Table 1).
LIP possesses FB projections to V4 (Cavada and Goldman-Rakic, 1989;Andersen et al., 1990; Blatt et al., 1990; Tanaka et al., 1990; Shipp and Zeki, 1995). The present results show that there is a hitherto undescribed projection to area V1 that is however extremely sparse and originated almost exclusively from the infragranular layers (Table 1). The projection from LIP to area V4 was considerably stronger than the V1 projection, and a non-negligible proportion of neurons was found in the supragranular layers (Fig. 6, bottom). In LIP the seven projection zones after V4 injections returned a mean SLN% value of 27% (Table 1).
Distributions in the FEF
Several authors have described projections from the FEF to area V4 (Huerta et al., 1987; Schall et al., 1995; Shipp and Zeki, 1995;Stanton et al., 1995; Bullier et al., 1996). In the present study, one of the three cases in which multiple injections were made in area V1 was examined for labeled neurons in FEF and showed a very sparse projection (six labeled neurons) entirely originating from the infragranular layers (Table 1). This result, along with the finding in LIP (see above), confirms that anterior cortical areas that project to area V4 also project to area V1. For the projection of FEF to area V4, the present results show that although it is a weak projection, it is at least as strong as the projection from FST to V4. Surprisingly, after all seven injections in area V4, the projection to area V4 from FEF largely arises from the supragranular layers (mean SLN% value, 73%; Fig. 7).
Reciprocal projections between areas V1 and V4
It has been suggested that reciprocity is a universal feature of corticocortical pathways (Rockland and Pandya, 1979), but in the visual system ∼20% of the connections are unidirectional (Felleman and Van Essen, 1991; Salin and Bullier, 1995). We have been able to examine this issue directly for connections linking areas V1 and V4. The top part of Figure 8 shows typical labeling in area V4 after injection in area V1. This gives a range of numbers of neurons per section in the order of 10–60 (Table 1). This contrasts with the results in area V1 after injection in area V4. Of the seven projection zones analyzed in area V1, that illustrated by the density profile in Figure 8 has the highest number of neurons, the average being less than nine neurons per projection zone, which is <0.1% of the intensity of the V4 to V1 projection (Table 1).
Distribution of intrinsic labeled neurons
Labeled neurons within the area injected showed very high densities and typically extended further in area V4 than in area V1 (Fig. 9), confirming the results of others (Yoshioka et al., 1992). The four projection zones analyzed returned a mean SLN of 61% in area V1 and in 64% in area V4 (Table 1). It could be that part of the labeled neurons located close to the injection sites are interneurons. Because extrinsic connections are uniquely from pyramidal neurons, we wanted to restrict our measurement to the pyramidal population of intrinsic neurons. By deleting from the counts the labeled cells at a distance of <250 μm from the pick-up zone, we excluded a large proportion of the inhibitory intrinsic neurons (Tanigawa et al., 1998). Counts of the total population of labeled cells were reduced by 6% (in area V1) and 12% (in area V4), but the laminar distribution remained stable (paired t test; p > 0.05) in both areas (V1, SLN%: 63 vs 61%; V4, SLN%: 65 vs 64%).
SLN% characterizes individual projections
The SLN% values in the projections to areas V1 and V4 for individual injections are shown in Table 1, and a summary of the means are provided in Figure 10. Statistical analysis (Table 2) reveals that SLN% values vary across areas for V1 (multinomial ANOVA, p< 0.001) and V4 afferents (p < 0.001). In all animals in which a double injection was performed, the analysis did not reveal statistical variability because of the type of dye used (FB vs DY; all cases p > 0.05). Furthermore, in cases of V4 injections in which complete data from all the projecting areas were obtained, the multinomial ANOVA did not show statistical variations across subjects (χ2 = 4.75;p = 0.09). Comparisons of the SLN% values for adjacent hierarchical levels (as determined by the scheme of Felleman and Van Essen, 1991) showed that of the total 46 comparisons, 41 were statistically significant (Table 2). This analysis shows that each projection is characterized by a specific SLN% value.
Hence, injections in area V1 showed a significant stepwise decrease in SLN% in the sequence V2–V3. Beyond V4 there was a significant stepwise decrease in percentages in the dorsal stream going from MT to FST and in the ventral stream in the sequence V4–TEO–TE—TH–TF. However, our results did not show a significant decrease in SLN% going from V3 to V4 despite the fact that it has been reported that V3 has a FF type projection to area V4 (Felleman et al., 1997a).
Injection in area V4 showed a significant stepwise decrease in SLN% values in the ascending pathway in the sequence V1–V2–V3A. In the feedback pathways, there was a significant stepwise decrease in the percentages in the dorsal pathways in the sequence MT–LIP–FST and in the ventral pathway in the sequence TEO–TE–TH–TF.
These results suggest that in the case of FB connections the greater the projection distance, the more it involves cells located in the infragranular layers. For FF projections the converse is true so that there is a proportional increase in SLN% with increasing distance.
SLN% and hierarchical organization of the visual system
We have used the Felleman and Van Essen (1991) hierarchical model of the visual system shown in Figure11 A and related it to the SLN% of V1 and V4 afferents. For each connection we have calculated the number of hierarchical steps separating the interconnected areas (Fig. 11 B). For example, areas V1, V2, and V4 are on hierarchical levels 1, 2, and 5, respectively. In the case of the FB projection from V2 to V1, the projection crosses one level in a positive direction (difference of levels: 2 − 1 = +1). The FF projection from V2 to V4 crosses three levels, but the difference is negative (difference level: 2 − 5 = −3). In this way numbers of hierarchical steps ranges from −4 (FF pathway going from V1 to V4) to +9 (FB pathway going from TH–TF to V1).
The relationship between the SLN% values and number of hierarchical steps is highly consistent across the 20 corticocortical pathways projecting to areas V1 and V4 (Fig. 11 C). The correlation factor for the pooled SLN% values is high (r 2 = 0.77) and statistically significant (Spearman; Rho = −0.89; p = 0.0001). The correlation is not generated by one of the two sets of afferents because the same analysis for V1 and V4 independently generates only a slightly lower correlation factor (V1,r 2 = 0.69; V4,r 2 = 0.70), which remains highly significant (Spearman; V1: Rho = −0.87; p= 0.004; V4, Rho = −0.73; p = 0.02).
Figure 11 C shows that the pooled values of SLN% for individual projections correlate remarkably well with numbers of hierarchical steps. Figure12 A shows the correlation of SLN% values with hierarchy for individual V4 injections. For individual cases, variations are small both for the slopes (Fig. 12 B; mean, −9.41 ± 0.36) and the correlation factors (Fig. 12 C; mean, 0.66 ± 0.03), and are statistically significant in all cases (Spearman; all casesp < 0.05). For individual injections in both areas V1 and V4 we found that pairwise comparison of hierarchical relationship shows a 77% (range, 71–90%) fit with the Felleman and Van Essen (1991) model. This fit increases to 82% (range, 78–90%) if FEF is not included. These results are highly significant because they show that hierarchical relations can be inferred from the results of a single injection, provided the SLN% values are accurately determined for all areas projecting to the target area.
Malcolm Young's group has claimed that the precise ordering of the monkey visual areas cannot be specified exactly, mostly because of incertitude about the hierarchical levels of areas located at higher stages (Hilgetag et al., 1996a). However they have proposed an optimal peak hierarchy (Hilgetag et al., 1996b) that overlaps with a large number of solutions. Using this model we found that the laminar pattern of projections to V1 and V4 is highly correlated to the hierarchical organization of the visual system (r 2 = 0.61; Spearman, Rho = −0.83; p = 0.0001).
Hierarchy suggested by areal differences of SLN%
We have used the SLN% values to modify the Felleman and Van Essen (1991) model. Statistical tests (see Materials and Methods; Table 2) differentiated areas according to their SLN% after V1 or V4 injections and consequently placed them on distinct hierarchical levels (Fig.13). Injections in area V4 values place areas V1 and V2 on successive levels. Area V3A (60%) is on a level immediately below area V4. Injections in V1 place V1, V2, and V3 on successive levels but fail to separate areas V3 and V4. This would suggest that the hierarchical distance separating area V3 and V4 is small, and we therefore placed V3 on a level just below V4. Injections in V1 show that the SLN% in MT is lower than in V4, suggesting that MT is on a higher level than V4. V4 injections suggest that MT and TEO are on the same level. V4 injections place LIP below FST on the same level as TE. TH–TF have minimum SLN% values and are placed at the highest level. We left one empty level below areas TH–TF to eventually accommodate areas (such as area 7a and the polysensory areas of the superior temporal sulcus) that are not interconnected with areas V1 and V4 but project to the parahippocampal cortex and are thought to occupy a higher rank than FST (Felleman and Van Essen, 1991).
The model that we have derived from the SLN% does not make pairwise comparisons of connections between areas, which is the basis of theFelleman and Van Essen (1991) model. Instead rank is directly derived from SLN%, so that the injections in V1 show that MT is further than V3, which is further than V2. In one sense our handmade approach is a hybrid of the Felleman and Van Essen (1991) model because we have retained discrete levels. It is expected that a mathematical approach will generate a graded distance values between areas and consequently provide a better determinacy of the hierarchical status of the visual areas. The correlation for the Felleman and Van Essen model (1991) (r 2 = 0.77) is lower than that proposed above (r 2 = 0.87; Spearman, Rho = −0.91; p = 0.0001), indicating that we have successfully optimized hierarchical levels with SLN% values.
We shall first review in detail how areal relationships suggested by SLN% values compare to the hierarchical relationships reported in the literature. We shall briefly discuss the relevance of hierarchical schemes for understanding visual processing before concluding on the potential of SLN% to constrain models of visual cortex.
The projections from areas V1, V2, and V3A to area V4 have been shown to confirm to a FF sequence (DeYoe and Van Essen, 1985;Livingstone and Hubel, 1987; Tanaka et al., 1990; Nakamura et al., 1993; DeYoe et al., 1994; Barone et al., 1995; Felleman et al., 1997b). After injections in area V4, maximum SLN% values are located in areas V1 (100%) and V2 (93%) and significantly less in area V3A (60%). Similarly, V2 and V3 send FB projections to area V1, and again the SLN% in V2 is higher compared to V3 (47 vs 9%). These results place areas V3 and V3A on a higher level than area V2 in agreement with other reports (Felleman and Van Essen, 1991; Felleman et al., 1997b;Gattass et al., 1997).
Area V3 is reported to have FF projections to area V4 (Felleman et al., 1997b). We find that V1 injections return similar SLN% values in V3 and V4 and we have shifted area V3 to a level immediately below V4 and on the same level as V3A. Although it has been questioned whether V3 and V3A are distinct areas (Krubitzer and Kaas, 1993), it is claimed, on the basis of qualitative data that V3A and V3 exchange FF and FB projections (Felleman et al., 1997b). Our data suggests only a very small separation of these two areas so we place them on the same level.
MT is reported to have a lateral projection to V4 (Maunsell and Van Essen, 1983; Ungerleider and Desimone, 1986). Although after V4 injection, area MT has SLN% values midway between FF and FB, we have placed this area one level above V4 because (1) V4 and MT have significantly different SLN% values after injection in V1 and (2) MT SLN% are significantly different from those of areas V3A and LIP (Table 2), which are reportedly on the hierarchical level below and above MT.
The descending pathways
The Felleman and Van Essen (1991) model places TH–TF, TE, and TEO on levels 10, 9, and 7. Recent studies support this sequence (Webster et al., 1991; Distler et al., 1993). Our SLN% values (1, 26, and 40%) suggest that these areas should be on levels 10, 7, and 6 (Fig. 13). These ventral areas send a sparse but consistent FB projection to area V1 (this study; Rockland and Van Hoesen, 1994). Area TEO has significantly higher SLN% values compared to areas TE and TH—TF, which is compatible with the levels we have allocated to these areas.
Injections of tracers in FST led to labeled cells in LIP in both supragranular and infragranular layers, whereas the terminals are located in all layers (Boussaoud et al., 1990). This has led Felleman and Van Essen (1991) to classify FST and LIP as sharing lateral connections and to place these two areas on the seventh level. FST is reported to exchange FF and FB projections with V4 and MT (Boussaoud et al., 1990). Together these results therefore support the suggestion ofFelleman and Van Essen (1991) that areas FST and LIP reside on a level above areas V4 and MT. However, our results return SLN of 27% for LIP and 14% for FST, suggesting that these two areas might be on different levels.
To conclude, the Felleman and Van Essen (1991) model contradicts six of the relations suggested by the SLN%, which are V3 = V3A; V4 < MT; MT = TEO; LIP < FST; TEO < FST; LIP = TE (FF is >, FB is <, and lateral is =). Three of the relations suggested by the SLN% are compatible with the optimal hierarchy proposed by Hilgetag et al. (1996b). There are three categories of disagreements between the two models. First, in the Felleman and Van Essen (1991) model two of these relations were lateral (MT = V4 and LIP = FST) (Maunsell and Van Essen, 1983; Ungerleider and Desimone, 1986; Boussaoud et al., 1990). Although we have placed these areas on separate levels based on statistical differences of SLN%, the distance suggested is small and therefore could be smaller than the nonparametric value of a level in the Felleman and Van Essen (1991)model. Second, some of the differences of the two models concern connections that have been reported as weak and/or inconsistent (MT–TEO; TEO–FST) (Morel and Bullier, 1990; Distler et al., 1993). Third, our suggestion that LIP = TE is in agreement with Webster et al. (1994). Finally, in our model V3 = V3A, where as in theFelleman and Van Essen (1991) model V3 < V3A. Once again this could be attributable to the SLN% model detecting small differences that are not easily detected by the laminar analysis used in theFelleman and Van Essen (1991) model, which relies largely on qualitative data.
Relationship of visual areas with FEF
It is known that area V4 projects to FEF in an FF manner (Barbas and Mesulam, 1981). However, little is known about the reciprocal projection. Despite the absence of any quantitative data, Felleman and Van Essen (1991) have allocated FEF to the eighth level, and accordingly it should possess a strong FB connection to area V4. However, our results do not show projections from FEF to area V4 to be FB projections, because they show a SLN% of 72% indicative of FF. Alternatively we cannot exclude the possibility that anatomical characterization of FF and FB connections do not extend to the frontal cortex (Webster et al., 1994).
A number of other anatomical studies question putting FEF at the top of the hierarchical series. In the ventral stream, injections of retrograde and anterograde tracers in TEO revealed lateral connections with FEF (Distler et al., 1993; Webster et al., 1994; Schall et al., 1995). Likewise, injections of anterograde tracers in TE suggested that this structure exhibits FB projections to FEF, whereas retrograde tracers injected in TE suggest a lateral connectivity with FEF (Webster et al., 1994). Anterograde and retrograde tracer investigations of the relationship of FEF with the dorsal stream failed to give conclusive results with respect to FF and FB classification schemes (Boussaoud et al., 1990; Schall et al., 1995).
There is physiological evidence in favor of FEF not having a purely FB relationship with extrastriate visual areas. Many neurons in FEF respond to visual stimulation with shorter latencies than a large proportion of neurons in more caudal extrastriate visual areas (Bullier and Nowak, 1995; Thompson et al., 1996; Nowak and Bullier, 1997;Schmolesky et al., 1998). One possible cortical route for such early visual responses in FEF is via V1 projections to MT (Maunsell and Van Essen, 1983; Ungerleider and Desimone, 1986).
Physiological significance of hierarchical organizations
Our results suggest that the relative proportion of supragranular and infragranular layer neurons is a general defining feature, at least in the visual system, of FB and FF pathways. Differences in the SLN% in different pathways could be of functional significance, given the experimental evidence suggesting that pyramidal neurons in upper and lower layers have different physiological properties (Lagae et al., 1989; Douglas and Martin, 1991; Nowak et al., 1995; Raiguel et al., 1995; Ahmed et al., 1998), histochemical features (Hof et al., 1996,1997), and topographical relationships (Barbas, 1995; Barone et al., 1995). These different properties could contribute to shaping the function of FF in the construction of receptive fields (Zeki 1993;Bullier et al., 1994; Hubel 1995; Vanduffel et al., 1997) and FB pathways in visual imagery (Ishai and Sagi, 1995; Miyashita 1995) and figure ground discriminations (Zipser et al., 1996; Hupé et al., 1998; Lamme et al., 1998).
Hierarchical distance and the organization of visual areas
The analysis of topology and patterns of laminar connectivity converge to indicate a hierarchical organization of the cortical visual system (Felleman and Van Essen, 1991; Young, 1992). Hierarchy in theFelleman and Van Essen (1991) model is derived from the pairwise analysis of the laminar patters of interareal connectivity in which each connection is defined as an FF, FB, and lateral connection largely on the basis of nonquantitative data. Mathematical modeling of the same database confirms the hierarchical nature of the organization but importantly indicates that there are huge number of possible solutions (Hilgetag et al., 1996a). This indeterminate nature of the proposed organization in part stems from the absence of an indication of the distance separating levels in the hierarchical scheme. The present study shows that the SLN% for a set of areas leads to an interareal ranking, which largely fits with the Felleman and Van Essen (1991)model. This suggests that SLN% reflects an underlying functional principle and further provides an indication of the relative hierarchical distance separating areas. A distance rule of hierarchical relationships derived from SLN% values might be a universal feature of the organization of the cortex, as revealed by the laminar organization of afferents to the frontal lobe (Barbas, 1986) and to the somatosensory system in the monkey (Batardière et al., 1998b). Future modeling of an extended quantitative database of this type holds the promise of providing a determinate model of the organization of visual cortex. The present database based on laminar location of parent cell bodies does not exclude the possibility that quantitative analysis of the laminar location of axon terminals (Barbas and Rempel-Clower, 1997) might be equally important and possibly complementary.
Our analysis of the hierarchical relationships of the visual areas is based on the nomenclature provided in the review paper of Felleman and Van Essen (1991). However, the definition and the integrity of some of these visual areas is still under debate (Kaas, 1996) and will have to be taken into account in future models of the visual cortex. Furthermore, heterogeneity of individual areas, such as a differential representation of the central and peripheral fields, might be accompanied by changes in connectivity with eccentricity (Perkel et al., 1986; Baizer et al., 1991; Stepniewska and Kaas, 1996; Gattass et al., 1997), which in turn could lead to differences in the hierarchical relationships of cortex subserving the central and peripheral visual fields (Falchier et al., 2000).
This work was supported by the European Community, Quality of Life and Management of Living Resources (QLG3-1999-01064).
Correspondence should be addressed to Dr. Henry Kennedy, Cerveau et Vision Unité 371, Institut National de la Santé, et de la Recherche Médicale, 18 Avenue du Doyen Lépine, 69675 Bron Cedex, France. E-mail:.