The pathways from lateral geniculate nucleus to primary visual cortex

The retino-geniculo-cortical pathways in the macaque monkey are composed of at least three distinct parallel systems, the M, P, and K pathways, which originate in the retina and remain segregated until they reach V1 (see, Chapter 30 in this volume). Each of these systems seems to be specialized for the detection of a unique set of visual cues. At the other end of the visual hierarchy are dozens of extrastriate cortical areas, each of which is also specialized for the detection of particular attributes of visual scenes (cf. Desimone and Ungerleider, 1989; Felleman and Van Essen, 1991). It is believed that the extrastriate visual areas also compose two more or less parallel systems. One system is specialized for analysis of spatial relationships and involves visual areas in parietal cortex. The other is believed to be responsible for object identification and involves areas in temporal cortex. Since virtually all of the retinal input relayed through the LGN to primate cortical areas converges on area V1 (Benevento and Standage, 1982; Bullier and Kennedy, 1983), V1 provides a unique interface between the M, P, and K pathways and extrastriate cortex. It also provides a unique opportunity for computations requiring early integration of information from more than one stream.

Our understanding of circuits underlying the contributions of the M, P, and K pathways to extrastriate cortical areas is based largely on studies of the flow of information into and out of various functional compartments in V1. These functional compartments include laminar (i.e., layers 4B, 4Cα, and 4Cβ) and columnar (i.e., blob, interblob) divisions. But there are also different cell types within each compartment which are anatomically and/or connectionally distinct. Over the past several decades, considerable progress has been made in relating the functional characteristics of parallel pathways with the functional architecture of V1 and, in turn, the functional differences between neurons providing output to extrastriate areas (see Callaway, 1998b, and below for review). But there is still much to be learned about the functional and anatomical heterogeneities within each pathway and how these are related.

The locations in V1 which receive the combined input from the M, P, and K cells of the LGN correlate closely with the pattern revealed by staining for the mitochondrial enzyme cytochrome oxidase (CO) (Livingstone and Hubel, 1982). The most dense LGN input is to layer 4C, but there is also input to layer 1, blobs in layer 2/3, layer 4A, and layer 6. Each of the M, P, and K pathways has a unique relationship to these afferent termination zones.

The M pathway

Neurons in the most ventral M layers of the LGN project most strongly to the upper division of layer 4C, to layer 4Cα, and secondarily to layer 6 (Hendrickson et al., 1978; Hubel and Wiesel, 1972). Anatomical reconstructions of individual afferent axons show that they are heterogeneous. Most innervate the entire depth of layer 4Cα, but a subset selectively targets the upper portion of 4Cα (Blasdel and Lund, 1983). These termination zones closely mimic the laminar organization of the dendritic arbors of two types of layer 4Cα spiny stellate neurons (Yabuta and Callaway, 1998a; see also below). Cells with dendrites in upper 4Cα, where they would contact both types of M afferent, project axons to layer 4B and selectively to CO blobs in layer 3. Cells with narrowly stratified dendrites in lower 4Cα, where they would contact only one type of M afferent, project axons densely and selectively to CO interblobs in layer 3. These observations suggest that layer 4B and blobs are influenced by both types of M input, while interblobs are influenced exclusively by the M input that spans the entire depth of layer 4Cα (see Fig. 42.1).

Figure 42.1..  

Schematic of the pathways from magno- (M) and parvocellular (P) LGN, through V1, to extrastriate cortical areas. Two anatomical types of M afferent terminate in either upper layer 4Cα of V1 (Ma illustrated in black) or the entire depth of 4Cα. This second type of M afferent (Mb) thus provides the only input to the middle of layer 4C (illustrated by the white box). P afferents terminate in layer 4Cβ. Layer 2/3 CO interblobs receive input from layers 4Cβ and middle 4C and thus get input from P and Mb but not Ma recipient neurons. Layer 2/3 blobs receive input from neurons throughout the depth of layer 4C and thus get input from Ma, Mb, and P recipient neurons. Layer 4B spiny stellate neurons receive input only from layer 4Cα and thus only from M recipient neurons, while layer 4B pyramids receive input from both layer 4Cα and layer 4Cβ and thus also receive input from P recipient neurons. Interblob neurons, in turn, project to thick stripes and interstripes of V2. Blob neurons project to V2 thin stripes. Layer 4B spiny stellates project to area MT. Layer 4B pyramids project to areas V2 and V3, with the V2 compartment that is targeted being dependent on their position relative to the blobs (see text).

M cells could potentially vary from each other with respect to numerous functional attributes, and these variations might be systematically related to the anatomically defined termination zones of M afferents (upper versus lower layer 4Cα). We have hypothesized that the functional differences between M afferents that terminate in upper 4Cα versus those that terminate throughout 4Cα might be related to their linearity (Yabuta and Callaway, 1998a). Some cells in the M layers of the LGN respond to high spatial frequency counterphase gratings with frequency-doubled responses, irrespective of spatial phase (Kaplan and Shapley, 1982; Chapter 30 in this volume). In this respect they are similar to Y cells, which in the cat's visual system arborize preferentially in upper layer 4 of area 17 (Humphrey et al., 1985; Shapley and Perry, 1986). Most M cells in the monkey, however, respond more linearly and in this respect are similar to the cat's X cells, which arborize throughout the depth of layer 4.

Although there may be functional and anatomical similarities between cat X or Y cells and subsets of LGN M cells in the monkey, it is unclear whether these pathways are truly analogous. For example, the parasol cells of the primate retina, which give rise to the M pathway (Leventhal et al., 1981), are most similar to the cat's α-retinal ganglion cells (RGCs) and not the β-RGCs which give rise to the X pathway (Leventhal et al., 1985). Furthermore, it is not clear whether there are discrete classes of linear and nonlinear M cells or whether frequency-doubled cells simply reflect extreme nonlinearity (Levitt et al., 2001). Thus, differences in linearity are just one of many possible functional differences that may distinguish M afferents in upper versus lower 4Cα.

P and K pathways

Neurons in the dorsal LGN layers include not only P cells but also K cells. K cells are located primarily in the intercalated zones between the P layers (and M layers) but are also found scattered within the P layers (Hendry and Yoshioka, 1994). This arrangement has made it difficult to ascertain which layers of V1 receive input from P versus K cells or which of the functionally distinct neuron types recorded in the dorsal LGN correspond to P or K cells. For example, early studies describing anterograde labeling following tracer injections (or lesion-induced degeneration) in the dorsal layers of the LGN reported label in superficial layers (layers 1, 2/3, and 4A) as well as layers 4Cβ and 6 (Hendrickson et al., 1978; Hubel and Wiesel, 1972). The superficial labeling was originally believed to arise from the P pathway. But we now know that most, if not all, of the label in the superficial layers originated from K cells, not P cells (see below). Similarly, recordings from LGN neurons in and around the P layers were all attributed to the P pathway before it was appreciated that there is a distinct K pathway. And even with this realization, recordings made in the LGN cannot distinguish P cells from the K cells that are scattered within the P layers.

Despite uncertainty about the precise anatomy and physiology of the K pathway, the available evidence strongly suggests that the great majority of LGN neurons which project axons to layer 4Cβ are P cells, while only K cells expressing Ca⁺⁺/Calmodulin-dependent Protein Kinase II (αCAMKII) or calbindin (Hendry and Yoshioka, 1994) project to more superficial layers (4A and layer 2/3 blobs). There are three main pieces of evidence which support this configuration: (1) Reconstructions of individual LGN afferents fail to reveal any axons which project densely both to layer 4Cβ and to superficial layers (Blasdel and Lund, 1983; Freund et al., 1989)—these inputs therefore appear to arise from anatomically distinct populations. (2) The K pathway originates from RGCs with the smallest-diameter axons and these axons innervate the intercalated K layers (Conley and Fitzpatrick, 1989). (3) LGN neurons retrogradely labeled from superficial layers of V1 include cells in the intercalated K layers, along with scattered cells in the P layers, and these cells stain for αCAMKII or calbindin (Hendry and Yoshioka, 1994). Although these findings are strongly suggestive, further studies are required to test more exhaustively the possibility that some afferents might innervate both layer 4Cβ and superficial layers or that some αCAMKII–expressing cells might connect to layer 4Cβ. There may also be cells projecting to layer 4A that do not express αCAMKII (Hendry and Yoshioka, 1994).

Functional Properties of M, P, and K Cells

Relative to P or K cells, M cells have excellent contrast sensitivity, prefer lower spatial frequencies and higher temporal frequencies, and lack cone-opponent receptive fields (see Chapter 30 in this volume for review). The most distinctive property of cells recorded in the dorsal layers of the LGN (P and K layers) is that, unlike M cells, they have cone-opponent receptive fields (e.g., Wiesel and Hubel, 1966). These cells are tuned along the cardinal red-green and blue-yellow color axes (Derrington et al., 1984; De Valois et al., 2000). Thus, they either receive input from long-wavelength sensitive (L) cones opposed to middle-wavelength sensitive (M) cones (red-green color opponency) or they have short-wavelength sensitive (S) cone input opposed by the L + M cones (blue-yellow color opponency).

The great majority of cells recorded in and around the P layers have red-green color opponency (Wiesel and Hubel, 1966). Since K cells are only a minority of the cells in this region, it follows that P cells must include cells with red-green opponency. But since there are no direct observations linking these cell types with receptive-field properties, it is possible that the P pathway might also include cells with blue-yellow color opponency.

There is also compelling evidence that a substantial portion of the K pathway neurons have blue-ON/yellow-OFF receptive fields (see Hendry and Reid, 2000, for review). Blue-ON receptive fields originate with bistratified retinal ganglion cells which connect to intercalated layers K3 and K4 of the LGN. Neurons recorded in the K3 and K4 intercalated layers also have blue-ON receptive fields (see Hendry and Reid, 2000, and personal observation). Neurons with blue-OFF/yellow-ON receptive fields have also been recorded in the dorsal layers of the LGN, but they have not been linked to the P or K pathway (De Valois et al., 2000, and personal observation). It is also not known whether the K pathway includes cells with red-green color opponency.

Projections to Extrastriate Cortical Areas

The next section of this chapter will describe the cell types which receive direct input from the M, P, and K pathways and how they distribute this information to V1's extrastriate cortical projection neurons. In this context, it is therefore also helpful to review briefly the organization of V1 neurons that project to various extrastriate cortical areas and to functionally distinct compartments in area V2.

As a general rule, neurons which project to dorsal visual areas are found in layer 4B, while those that project to ventral visual areas are found in layer 2/3 (see Felleman and Van Essen, 1991, for review). For example, layer 4B neurons provide direct input to areas V3 and MT, while layer 2/3 neurons connect to area V4. But detailed connectivity between areas V1 and V2 has only very recently been studied in the macaque monkey (Sincich and Horton, 2002). Although it was previously appreciated that neurons throughout the cortical depth, from layers 2 to 4B, contain neurons connecting to area V2, their relationships to CO stripe compartments in V2 and to blobs and interblobs in V1 had not been investigated. Because each of the CO stripe compartments in V2 connects to a unique set of cortical areas, this is particularly relevant to the question of how each type of V1 cortical projection neuron distributes information to dorsal and ventral visual areas. In particular, the V2 thick stripes connect to areas V3 and MT, while thin stripes and interstripes connect to V4 (for review see De Yoe and Van Essen, 1988; Zeki and Shipp, 1988).

Input to the CO compartments in V2 is most closely related to the CO organization in V1, not to V1's laminar organization. Layer 2 to 4B neurons in (and under) interblob regions of V1 connect to V2 thick stripes and interstripes, while those in (and under) blob regions connect to V2 thin stripes (Sincich and Horton, 2002). Thus, although it had long been thought that layer 4B provided the only input to V2 thick stripes, it is now clear that layer 2/3 and 4A interblob neurons also provide input to the thick stripes (Sincich and Horton, 2002). It had also been thought that V2 thin stripes receive input only from layer 2/3 blob neurons, but the new results indicate that the layer 4A and 4B cells under blobs also connect to thin stripes (Sincich and Horton, 2002). It is noteworthy that the input from layer 4B to V2 tends to arise from pyramidal neurons, while the direct input to MT from V1 arises predominantly from layer 4B spiny stellate neurons (Shipp and Zeki, 1989; see further below).