Structure and function of corticocortical connections

General Characteristics

Corticocortical connections between functional areas are made almost exclusively by pyramidal neurons. A few exceptions have been noted: in the cat, some spiny stellate cells in area 17 send projections to area 18 (Meyer and Albus, 1981). It has been reported that in the rat, a small contingent of GABAergic smooth stellate cells send connections to neighboring cortical areas (McDonald and Burkhalter, 1993). Interhemispheric connections have also been found to arise from a few presumably inhibitory interneurons in the rat (Hughes and Peters, 1992a, 1992b) and in the cat (Peters et al., 1990). Despite this anatomical evidence, no monosynaptic inhibitory synaptic potential has ever been reported in electrophysiological studies of connections between cortical areas of the same or opposite hemisphere. It can therefore be concluded that interarea and interhemispheric corticocortical connections are massively excitatory. This does not mean that these connections have a net excitatory effect on the responses of the target area because they contact pyramidal cells as well as inhibitory interneurons. Therefore, the net effect of corticocortical connections is a mixture of excitatory and inhibitory influences.

It is known that within the local network of horizontal connections, the densest projections are to immediate neighbors. This is demonstrated by placing small injections of anterograde or retrograde tracers in a given site and examining the local distribution of labeled axons and neurons. The higher density of local connections is due to the branching pattern of axons that arborize more profusely near the main axon trunk.1

In a similar fashion, the densest interarea connections tend to be with neighboring cortical areas. For example, the strongest connections of area V2 are with neighboring areas V1 and V4 in the monkey (Fig. 33.5A). Similarly, in the cat, the strongest connections of area 17 are with neighboring areas 18 and 19. There are, however, a few examples of adjacent areas that are not interconnected, like the retrosplenial visual area in the monkey that has no connections with area V1, although it is surrounded by it on its caudal and lateral borders. Another exception to the rule of preferential connections with neighboring areas is found in the relationship between visual areas of the occipital, parietal, and temporal lobes with areas of the frontal lobe surrounding the frontal eye field area (FEF in Fig. 33.5A). These frontal areas, in addition to local connections in the frontal lobe, exchange long-distance connections with parietal, occipital, and inferotemporal cortex. Similarly, areas of the occipital and parietal cortex interconnect with adjacent areas and exchange long-distance connections with the frontal cortex (Jones and Powell, 1970; Schall et al., 1995). This dual set of connections (to neighboring areas and to frontal cortex) probably corresponds to the different functional roles played by posterior and frontal regions in vision.2

Different Types of Corticocortical Connections

Intrahemispheric corticocortical connections are often divided into two classes: feedforward and feedback. The anatomical differences between feedforward and feedback were first decribed by Rockland and Pandya (1979), who noted that some connections (forward-going) tend to originate in neurons located in supragranular layers (layers 2 and 3) and terminate around layer 4, whereas reciprocal connections (backward-directed) are predominantly made by neurons in infragranular layers (layers 5 and 6) and project outside layer 4 (Figs. 33.1 and 33.4). This was later formalized by Maunsell and Van Essen (1983) and Felleman and Van Essen (1991), who used the anatomical differences between feedforward and feedback connections3 to construct the hierarchy of cortical areas (see below).

Although feedforward and feedback connections are usually presented in their archetypal form4 (Fig. 33.1), there are many variations on the theme: feedforward connections can originate from neurons in infragranular layers (Fig. 33.4A), and their terminals often extend into the supragranular layers. Terminals of feedback connections usually avoid layer 4, but they often originate from supra- as well as infragranular layers (Fig. 33.4B). Quantitative estimates of the proportions of source neurons in infra- versus supragranular layers reveal that there is a continuum in the organization of feedback and feedforward connections instead of two separate homogeneous populations (Barone et al., 2000). Although this has not been quantified, it appears that a similar continuum is found when the laminar distribution of axonal arborization is considered. As argued earlier (Salin and Bullier, 1995), the archetypal organization of feedback connections (neurons located only in infragranular layers and providing exclusive input to layers 1 and 2) is found only for the relatively rare connections between areas that are distant on the cortical surface [such as inferotemporal (IT) cortex and V1; Fig. 33.5A). In contrast, feedback connections between neighboring areas, which are extremely numerous, do not follow the archetypal model. They originate from neurons in supra- as well as infragranular layers and terminate in all layers except the lower portion of layer 4 (Kennedy and Bullier, 1985; Kennedy et al., 1989).

Interhemispheric connections have morphological characteristics that place them in the feedforward group for the laminar position of the source neurons (in layers 2 and 3) and in the feedforward or feedback group for the distributions of the axon terminals. In fact, the laminar organization of axon terminals in interhemispheric connections tends to follow that of intrahemispheric connections. In general, a given area connects to the same areas in both the same and the opposite cortical hemipheres, and the laminar distributions of the terminals are similar for inter- and intrahemispheric connections (Kennedy et al., 1991).

Retinotopic Organization

As mentioned above, one of the major organizing principles of the visual cortical areas is that each area contains a retinotopic representation of the contralateral visual hemifield. In addition, receptive field sizes vary greatly from one area to another. Receptive field size is smallest in area V1 and increases progressively as one moves from V1 to V2, V4, and TEO to reach receptive fields that cover almost the entire visual field in some neurons of IT cortex. Similarly, receptive fields increase from V1 to V2, MT, MST and areas of the parietal cortex (Fig. 33.2). In addition, all anatomical studies show an important degree of convergence and divergence in corticocortical connections. Typically, a cortical zone a few hundred microns wide projects to and receives from a region that is usually a few millimeters wide but can cover up to 15 mm on the cortical surface of a connected area (see the review in Salin and Bullier, 1995).

Figure 33.2..  

Receptive field center size as a function of eccentricity in the visual field for areas of the dorsal stream. Note how the receptive field size increases with distance from area V1. (From Rosa, 1997.)

This large degree of convergence and divergence, and the great variation in receptive field sizes among cortical areas, have important consequences for the retinotopic organization of corticocortical connections. It is usually assumed that all connections in the visual system are retinotopically organized, meaning that the receptive field centers of the afferent neurons are included in the receptive field center of the recipient cell. This appears to be the case for thalamocortical connections (Reid and Alonso, 1995; Tanaka, 1983) and feedforward connections, as shown by mapping studies in the cat (Price et al., 1994; Sherk and Ombrellaro, 1988) and inactivation studies in the monkey (Girard and Bullier, 1989; Girard et al., 1991a, 1991b) (Fig. 33.3).

Figure 33.3..  

Retinotopic organization of feedforward, feedback, and horizontal connections. The left part of the figure schematically represents areas V1 and V2 seen from above. Triangles correspond to neuron cell bodies. The direction of the connection is indicated by the arrows on the simplified axons. The right part of the figure represents the right lower visual field of the animal (FP, fixation point; HM, horizontal meridian; VM, vertical meridian). For the feedforward connections, the large square represents the receptive field (RF) center of the V2 neuron receiving convergent information from the V1 neurons that have the small black squares as RF centers. The combination of RF centers of the afferent V1 neurons makes up the RF center of the target V2 neuron: feedforward connections are visuotopically organized. For the feedback connections, the small black square represents the RF center of the V1 neuron receiving convergent information from V2 neurons with the large open squares as RF centers. The RF centers of V2 neurons cover a larger region of visual field than that covered by the RF center of their target neuron in area V1: the feedback connections are only loosely retinotopic and can be used to mix information from distant regions of the visual field. For the horizontal connections, the gray square represents the RF center of the V1 neuron receiving convergent horizontal connections from neighboring neurons with RF centers indicated by the black squares. The combination of the horizontal afferents covers a larger part of the visual field than the RF center of the target neuron: horizontal connections are loosely retinotopic.

Such is not the case for feedback connections because of the large degree of convergence and the larger receptive field centers of neurons in areas distant from area V1 (Fig. 33.2). Earlier mapping studies showed that the organization of feedback connections is compatible with the rule that neurons tend to interconnect if their receptive field centers overlap at least partially (Salin et al., 1992). As a consequence, the extent of the visual field concerned by the feedback connections to a column of cortex corresponds to the sum of twice the average diameter of the receptive field center in the source area and the average receptive field center in the recipient area (Fig. 33.3). The progressive increase in receptive field center size with distance from V1 (Fig. 33.2) means that feedback connections concern larger and larger regions of visual field as one moves from adjacent to more distant areas. Thus, feedback from area V2 to area V1 comes from neurons coding for a limited region of visual field (a few degrees; Fig. 33.3). On the other hand, feedback projections from IT neurons (Kennedy and Bullier, 1985; Rockland and Hoesen, 1994), which have very large receptive fields, enable neurons in area V1 to be influenced by information concerning very large portions of the visual field. Feedback connections thus provide a sort of tiling of visual space, with further influences coming from progressively further areas.

In a similar way, but on a much smaller scale, the horizontal connections link together neurons with neighboring and partially overlapping receptive fields. The region of the visual field concerned by the horizontal connections corresponds approximately to the point image, that is, the extent of visual field covered by the receptive fields of neurons that are activated by a light point in the visual field5 (Angelucci et al., 2000). Thus, the extent of visual field concerned by the local connections is much smaller than that corresponding to the feedback connections from some of the most distant source areas (Fig. 33.3).

Interhemispheric connections, like horizontal connections, also interconnect neurons with overlapping and partially overlapping receptive fields, particularly around the vertical meridian. In addition, some interhemispheric connections appear to link together neurons with separate receptive fields that sometimes correspond to mirror images in both visual hemifields (Houzel and Milleret, 1999; Innocenti, 1986; Kennedy et al., 1991).

Patchy Organization and Axonal Bifurcation

Neuroanatomical tracing studies show that neurons retrogradely filled by a small deposit of retrograde tracer tend to be grouped in small patches a few hundred microns wide and separated by 500 to 1000 µm on the cortical surface, depending on the connection and the species (Fig. 33.4A). Results of anterograde tracer studies also demonstrate that terminal axons labeled by a small amount of tracer placed in a given cortical area are grouped together in small patches (Fig. 33.4C). This was originally discovered for horizontal connections (Rockland and Lund, 1982), and it was found later that extrinsic corticocortical connections also tend to be similarly organized in most species, with the possible exception of the mouse (Braitenberg and Schüz, 1991).

Figure 33.4..  

Patchy organization of corticocortical connections. A, distribution of neurons in area 17 sending feedforward connections to areas 18 and 19 of the cat visual cortex (small dots indicate neurons projecting to area 18, large dots neurons projecting to area 19, stars projecting to both areas). Note the predominance of neurons in the supragranular layers (2 and 3), the rare neurons projecting to both areas (stars). B, distribution of neurons sending feedback connections from area MT to area V1 in the macaque monkey. Note the predominance and continuous nonpatchy distribution of feedback neurons in the infragranular layers. C, detail of B: retrogradely labeled feedback neurons in layers 5 and 6 (white dots) and labeling of feedforward terminal axons in layer 4 (white cloud). Note the contrast between the small patch of feedforward axon terminals and the more extensive and nonpatchy distribution of feedback neurons (A modified from Bullier et al., 1984; C modified from Kennedy et al., 1989.)

Patchy arborization is usually observed for horizontal (Gilbert and Wiesel, 1983; Rockland and Lund, 1982) and feedforward connections (Bullier et al., 1984; DeYoe and Van Essen, 1985; Shipp and Zeki, 1985; Symonds and Rosenquist, 1984). Coupled injections of different retrograde tracers in different cortical areas produce mostly nonoverlapping patches of labeled cells (Bullier et al., 1984; DeYoe and Van Essen, 1985; Shipp and Zeki, 1985), with very few double-labeled cells in the regions of overlap (Fig. 33.4A). This suggests that a given cortical area sends feedforward connections to several other areas through a system of neuronal populations organized as interdigitating neuron patches. It is likely that these populations share common functional properties and are interconnected by patchy horizontal connections. Although less frequently demonstrated because of the small number of studies using coupled anterograde tracers compared to studies using retrograde tracers, patches of terminals axons also appear to segregate at least partially in the target area for feedforward and interhemispheric connections (Goldman-Rakic and Schwartz, 1982; Morel and Bullier, 1990).

It is likely that the patchy organization and lack of axonal bifurcation in feedforward connections are the marks of the functional specificity of such connections. This is suggested by the observation that the V1 neurons projecting to area MT belong to a specific type with homogeneous properties (Movshon and Newsome, 1996). This result is comparable to that of an earlier work by Henry and his collaborators, which also demonstrated the specific functional properties of neurons projecting from area 17 to the PMLS in the cat (Henry et al., 1978). Similarly, the patchy distribution of feedforward terminals presumably corresponds to the convergence of axon terminals of neurons with some common properties. This is suggested by the elegant experiments of Sherk in the LS area (which is slightly larger than the PMLS) (Sherk, 1990). Using a neurotoxin, she killed the neurons in a small region of that area and recorded from what presumably correspond to the terminals of afferent axons. She found that axon terminals group together according to direction selectivity, thus suggesting that neurons with the same optimal direction tend to terminate in common patches. The neurons innervated by this axon group presumably inherit the property of direction selectivity transmitted by the converging feedforward axons.

The prevalence of patchy organization is more variable in studies of feedback connections. In general, when relatively extensive injections of retrograde tracers are placed in a given area, neurons in extrastriate areas do not group themselves in well-defined patches, as in the case of feedforward connections (Fig. 33.4B; Kennedy and Bullier, 1985; Perkel et al., 1986). On the other hand, more localized injections of retrograde tracers in the supragranular layers produce patchy distributions of labeled cells in supragranular layers (Angelucci et al., 2002; Salin et al., 1995; Shipp and Grant, 1991). Similarly, injections of anterograde tracers produce patchy distributions of terminals, particularly in the supragranular layers (Henry et al., 1991; Salin et al., 1995; Wong-Riley, 1979). A continuous distribution of anterograde labeling has been reported on other occasions, but this may be due to the large size of the injection (Maunsell and Van Essen, 1983; Ungerleider et al., 1983).

There appears to be greater variability in the patchy distribution of feedback than in feedforward. The reason for this difference may be that different categories of feedback connections, concerning different sets of layers, show variable degrees of patchiness. Indeed, direct comparisons of the distributions of labeled cells or axon terminals in different layers for feedback connections show that patchy distributions exist in the connections between supragranular layers, whereas connections from infragranular layers appear to be less segregated (Fig. 33.4B), and afferent terminals in lamina 1 always project in a diffuse manner (Henry et al., 1991; Salin et al., 1992; Shipp and Grant, 1991). Similar differences between the topographic organizations of terminals in different laminae are also observed when individual axonal arbors are traced in the target area, as demonstrated by the work of Rockland and her colleagues (Rockland and Drash, 1996; Rockland et al., 1994).

It is interesting that the laminar differences observed for the patchy character of feedback connections are echoed by similar differences in the pattern of axonal bifurcation. Thus, in feedback connections, the proportion of neurons sending bifurcating axons to two cortical areas is higher in infragranular than in supragranular layers (Bullier and Kennedy, 1987). Also, the results of Rockland and her associates demonstrate that some feedback axons have long axonal collaterals in layer 1 that arborize extensively over at least two cortical areas, whereas terminals in layers 2 and 3 are restricted to one cortical area (Rockland et al., 1994).

This difference in organization across layers suggests that for a given set of feedback connections between two areas, different roles are played by different subsets of connections corresponding to different laminar distributions in the source and target areas. Thus, the variability of laminar distribution among feedback connections between areas at different distances on the cortical surface may correspond to different functional roles played by distant feedback connections (e.g., TE to V1) compared to those between adjacent areas (e.g., V2 to V1).

Assuming that the patchy and nonbifurcating nature of feedforward connections reflects the necessity to organize inputs according to specific properties, the more diffuse character of the feedback connections to layer 1 suggests that it plays a more general role, such as controlling the contrast gain or membrane potential of a large population of target neurons. Such a diffuse role cannot be extended to all feedback connections, however, because feedback connections to layers 2 and 3, with their patchy organization, probably play a very specific role in the processing of visual information.

What differentiates neurons with axonal bifurcation to several cortical areas from neurons that project to only one cortical area? This question is particularly interesting for the feedback connections from the infragranular layers that contain a sizable proportion of axonal bifurcation (Bullier and Kennedy, 1987). It is possible that such bifurcation concerns preferentially axons with fast conduction velocity. As suggested by modeling studies (Murre and Sturdy, 1995), axon size is under strong constraints in the brains of large mammals. Projection of thick axons to several areas by way of bifurcation is one way of limiting their number. Indeed, there are many examples of thick axons that bifurcate: Y cells in the cat LGN that have the largest axons send bifurcations to areas 17 and 18 (Bullier and Kennedy, 1987), and Meynert cells in layer 6 of area V1 send bifurcating axons to at least area MT, and to the superior colliculus (Fries et al., 1985). In functional terms, the bifurcation of thick axons to several cortical areas is an efficient way to coactivate several cortical areas rapidly and simultaneously.

Synaptic Transmission

Given the important anatomical differences between feedforward and feedback connections reviewed above, it is likely that differences also exist in the characteristics of synaptic transmission for these two types of connections. Reports on electron microscopic (EM) studies of all types of corticocortical connections all agree that they make excitatory synapses on their target neurons (Anderson et al., 1998; Gonchar and Burkhalter, 1999; Johnson and Burkhalter, 1996; Lowenstein and Somogyi, 1991). Possible differences between feedforward and feedback connections are therefore likely to be reflected in the excitatory or inhibitory type of neuron that is contacted. The morphology of feedforward connections from V1 to V5 was recently investigated by Anderson and his collaborators (1998). Terminal boutons formed asymmetric (presumably excitatory) synapses and tended to contact preferentially spiny neurons (excitatory, mostly pyramidal), but also terminated on smooth, presumably inhibitory, cells in 20% of the cases. Very similar proportions were reported by Lowenstein and Somogyi (1991) in their study of the feedforward projection from area 17 to PMLS in the cat, an area that has been considered homologous to area MT of primates (Payne, 1993). In a study of feedforward connections between visual cortical areas in the rat, Johnson and Burkhalter (1996) reported a smaller proportion of contacts onto synaptic shafts (10%).

Less is known concerning the synaptic organization of feedback connections. The early results of Jonhnson and Burkhalter suggested that feedback connections contact more specifically spines of pyramidal cells (98% of the cases) and rarely terminate on dendritic shafts (Johnson and Burkhalter, 1996). However, a more recent study by the same laboratory found similar proportions of terminals on parvalbumin-rich GABAergic interneurons (10%) in both feedforward and feedback connections in the rat visual system (Gonchar and Burkhalter, 1999). Differences between feedforward and feedback were found at the site of axonal contact to GABAergic parvalbumin-rich interneurons, with feedback connections terminating on distal parts of the dendrites, whereas feedforward contacted dendritic regions closer to the cell body (Gonchar and Burkhalter, 1999). This finding is in keeping with functional data from the same group showing that feedback connections have mostly an excitatory influence, whereas electrical stimulation of intrinsic and feedforward connections tends to recruit inhibitory circuits at relatively low stimulus intensities (Shao and Burkhalter, 1996). The latter results, however, should be treated with caution, because electrical stimulation acts exclusively on axonal branches (Nowak and Bullier, 1998a, 1998b). Therefore, stimulation in one area activates the efferent axons orthodromically, as well as the afferent axons antidromically, and it is impossible to differentiate the synaptic potentials evoked by direct orthodromic activation from those evoked by recurrent collaterals of antidromically activated axons. This confusion probably explains why the laminar pattern of activation elicited in a given cortical area by electrical stimulation in another area does not always fit with that predicted from the laminar distribution of axon terminals (Domenici et al., 1995; Nowak et al., 1997).

The low proportion of terminals on dendritic shafts reported for feedback connections by Burkhalter and his colleagues contrasts with the results of an earlier EM study of the feedback connections between areas 18 and 17 in the cat (Fisken et al., 1975). In that study, the authors concluded that more than 30% of the terminals of feedback connections were located on dendritic shafts. Whether this discrepancy is related to a species difference or whether there is indeed a strong feedback projection to dendritic shafts remains to be determined by further studies.

In conclusion, much remains to be done to understand the differences between the synaptic organizations of feedforward and feedback connections. It appears that both types send excitatory connections, but whether they target different proportions of inhibitory neurons has not been clearly established. Using agonists and antagonists, it will also be interesting to determine the proportions of different receptor channels contacted by these two types of connections. As mentioned above, the task is complicated by the limits of the method of electrical stimulation and by the heterogeneity of feedback connections.