The geniculo-cortical loop

Anatomical Considerations

A common feature of the projections transferring retinal information to higher levels is that numerically the feedforward synapses constitute a relatively small component of the total excitatory synaptic input converging on the cells that relay the input to the next level. Thus, at each level in the system, the transfer of information is potentially subject to influences from both lateral interactions in the local horizontal network and feedback from higher levels in the system. In the LGN of cat and primates the retinal afferents comprise about 10% of the input to relay cells, while the corticofugal feedback connections to the relay cells represent 30% of the input (Sherman, 2001; Wilson, 1993). As well as the relay cells, the cortical feedback connections target intrinsic inhibitory interneurons in the LGN and inhibitory neurons in the thalamic reticular nucleus/perigeniculate nucleus. The main details of the circuitry in the LGN are summarized in Figure 37.1. In the cat the projection to the LGN, which derives primarily from areas 17 and 18, is dense, and individual corticofugal axons' arborizations have a central core projection of approximately 180–1080 µ with a sparse scattering of long-range axons which spread over 500–2000 µ (Murphy et al., 2000; Murphy and Sillito, 1996; Robson, 1984). This needs to be placed in the perspective of the fact that the average spread of the retinal X axonal arborizations is 150 µ and that of the Y axons is 375 µ (Bowling and Michael, 1984). Thus, even within their central core, individual corticofugal axons innervate an area of the LGN that extends significantly beyond their own location in retinotopic space. These points and the circuitry in the LGN are summarized in Figure 37.1. This, together with their longer-range connections, means they can influence inputs that may lie outside their own classical receptive field. There is also good evidence favoring the view that the projection comprises axons with several diameters and conduction velocities. Individual axons, both coarse and fine, seem to innervate both perigeniculate nucleus (PGN) and LGN (Murphy and Sillito, 1996; Murphy et al., 2000). Similarly, estimates of the conduction velocity from latency measurements provide support for two groups of corticofugal afferents (Grieve and Sillito, 1995b; Tsumoto et al., 1978) projecting to the LGN and PGN. The fastest pathway allows for a very rapid influence of the corticofugal system on the developing response and would enable the loop from LGN cell response to cortex and back within approximately 3–5 msec. Even the slower group would enable a loop within 4–10 msec. The organization of layer 6 and its interactions with the LGN in the primate are particularly interesting because there is a clear segregation of cells in relation to the incoming channels. Cells projecting to the parvocellular layers of the LGN lie in the upper part of layer 6, while the lower part contains cells projecting to both magnocellular and parvocellular laminae (Fitzpatrick et al., 1994). Additionally, the intercalated cells in the LGN that send axons to layer 1 and the cytochrome oxidase–rich blobs (Hendry and Yoshioka, 1994) seem to receive feedback from a distinct group of cells in the deepest part of layer 6 (Fitzpatrick et al., 1994). Cells in the middle of layer 6 do not project to the LGN but may project to the claustrum.

Figure 37.1..  

Schematic summary of the geniculo-cortical loop. Shaded ovals denote the typical retinotopic extent (at 5 degrees of eccentricity from the area centralis) of the axonal arborizations of X and Y retinal ganglion cell axons and corticofugal axons. Two dimensions are given for feedback axons: the smaller dimension depicts the typical spread for the dense central core projection, while the larger dimension depicts the overall coverage of the axonal arborization. Percentages to the right of the drawing summarize the contribution of synaptic contacts from retinal axons, corticofugal feedback axons, and inhibitory interneurons synapsing on LGN relay cells, as well as the contributions from geniculate axons, layer 6 collaterals, and layer 4 connections synapsing on spiny stellate cells in layer 4 of the cortex.

An interesting feature of the organization of feedback connections from the visual cortex to the LGN in the cat is that the most dense component of the terminal arborizations of the corticofugal axons in the LGN exhibits marked asymmetry. These arborizations are extended either parallel or perpendicular to the axis of the orientation preference of the parent cell in layer 6 of the visual cortex (Murphy et al., 1999), as shown in Figure 37.2. This anatomical asymmetry strongly suggests that the feedback might be organized to influence the processing of information relevant to the extraction of the orientation of contours in the visual cortex. We discuss this further in relation to functional data below.

Figure 37.2..  

Asymmetric termination pattern of corticofugal axons links to parent cell response properties. A–E, Experimental paradigm. Receptive field properties of layer 6 cortical cells were mapped prior to labeling with biocytin (A). The receptive fields (B) were superimposed on the geniculate retinotopic map (C). Single axons were reconstructed from serial sections and the bouton distribution was quantified in three dimensions, then analyzed with respect to their relation to the geniculate representation of visual space (D, E). F, G, Two main patterns of connectivity were observed. Feedback axons either innervated LGN cells lying along a line corresponding to the preferred orientation of the parent layer 6 cell (F) or orthogonal to (hence lying along the preferred axis of stimulus movement) the parent cell orientation preference (G). (Data taken from Murphy et al., 1999.)

Firing Patterns

A key question is whether under normal circumstances layer 6 cells in the visual cortex directly drive LGN relay cells or whether they simply modulate their response to the retinal input. The synaptic input from the corticofugal axons, although extensive, as discussed above, targets the distal portion of the relay cell dendrites via ionotropic and metabotropic receptors (Sherman, 2001; Von Krosigk et al., 1999) while the retinal axons target the proximal dendrites via ionotropic receptors. While it is clear that under certain nonphysiological circumstances (e.g., electrical stimulation of layer 6 cells at 30–50 Hz) the feedback pathway effects are optimized and can drive relay cells directly (Lindstrom and Wróbel, 1990), the balance of the evidence favors a feedback-driven modulation of the probability that relay cells will respond to their retinal input (Sherman and Guillery, 1998). Certainly for visual stimulation, several investigations support the view that virtually all the spikes fired by X- and Y-type relay cells in the cat can be linked to spikes in the retinal ganglion cells providing the input (Cleland and Lee, 1985; Mastronarde, 1987). This point needs further experimental verification with stimuli that will strongly activate layer 6 cells in the visual cortex. Additionally, the suggestion that burst mode firing in relay cells, originally associated with synchronized firing of relay cells and slow-wave sleep, also occurs in the waking state (Guido and Weyand, 1995; Sherman, 2001) adds another dimension of complexity (Rowe and Fischer, 2001). Burst mode firing occurs when relay cells have been hyperpolarized for periods of 100 msec or more and follows the deinactivation of a voltage- and time-dependent calcium current (I_T). Under these conditions, when the cell receives a suprathreshold depolarizing input, there is a calcium influx generating a low-threshold depolarizing spike that then activates a burst of conventional spikes. The size of the low-threshold depolarizing spike and the number of conventional spikes in the burst depend on the degree of hyperpolarization of the membrane, not the magnitude of the suprathreshold activating input. A sustained depolarization of relay cells for 100 msec or more inactivates I_T, and the cell switches to tonic mode firing associated with a linear transmission of sensory information and in this sense associated more conventionally with the waking state.

We suggest that the key to the reported presence of burst firing mode in the waking state is the strength of the hyperpolarization that can follow from the visual input. An example could be a group of OFF-center X cells with receptive fields lying within a bright, featureless, and thus uniform field. This would produce a visually driven hyperpolarization of their membrane potential. The salient components of an image are the loci where a change occurs; thus, in this sense, the hypothetical group of cells would have nothing to signal. On the other hand, the appearance of a moving dark contour over their receptive field would provoke a depolarizing input and a low-threshold calcium spike driving a burst of action potentials.

It has been suggested that burst firing in thalamic cells in the waking state serves as a “wake-up” call to the cortical cells receiving the input (Sherman, 2001). However, in the sense that the cortex is already in the waking state, this may be a slightly misleading description. Rather, the burst of action potentials in the input to the cortex would provide a high-security signal to focus the circuitry on the new feature driving attention in the cortical mechanism. One suggested role for the corticofugal feedback system is that it may contribute to selective attention via its influence on geniculate firing patterns. Because the feedback axons influence LGN relay cells both directly via synapses involving ionotropic and metabotropic receptors (Sherman, 2001; Von Krosigk et al., 1999), and indirectly via an input to inhibitory interneurons, they have the capacity to exert a complex pattern of control over relay cells that could switch the behavior of LGN cells between tonic and burst modes. Recent work has questioned whether the patterned activity that occurs normally in the visual cortex during visual stimulation might switch the behavior of LGN cells via the feedback system (Wang et al., submitted). While recording simultaneously in cortex and LGN, focal iontophoretic application of a GABA_B receptor antagonist, CGP 55845 (see below), was used to produce local relief of the GABA_B inhibition of layer 6 cells (Fig. 37.3A). This reversibly enhanced the gain of their visually driven responses without affecting their spontaneous firing rate, making it possible to isolate the effects of this change in a controlled fashion. This change in cortical visual responses led to a statistically significant shift in the ratio of burst to tonic firing for 68% of LGN cells tested. Of these, 43% showed a shift from tonic to bursting (e.g., Fig. 37.3B) and 25% from bursting to tonic firing (Fig. 37.3C). These effects did not follow from the drug's application causing a state-dependent shift in the state of the cortex because simultaneously recorded LGN cells showed opposite direction shifts in firing. Thus, the data indicate that a focal change in the visual response magnitude of layer 6 cells can produce a clear switch in the firing pattern of LGN cells. In some cases, it moves them toward the tonic firing pattern and the faithful relay of their visual input; in others, it moves them to burst firing and a response mode suggested to underlie early signal detection (Sherman, 2001). With the complex input derived from stimuli in the natural visual world, the selective adjustment of the transfer properties of the LGN provides a means of alerting the system to salient change, while at the same time optimizing its capacity to relay accurate information about what has already engaged the system.

Figure 37.3..  

Effect of corticofugal feedback on LGN cell firing patterns. A, Experimental paradigm. Simultaneous recordings were obtained from a layer 6 cell and an LGN cell. Responses to visual stimulation were compared before, during, and after focal iontophoretic application of the GABA_B antagonist CGP in layer 6. Spike trains of LGN cell responses during periods of visual stimulation were divided into periods of burst and tonic activity. The first action potential in a burst showed a preceding silent period of at least 100 msec followed by a second spike with an interspike interval ≤4 msec. Any subsequent action potentials with preceding interspike intervals ≤4 msec were also considered to be part of a burst. All other spikes were regarded as tonic. B,C, The peristimulus time histograms (PSTHs) show the responses of two LGN cells to an optimal diameter flashing spot located over the classical receptive field (CRF) before, during, and after focal enhancement of layer 6. For both cells, there was a clear change in firing pattern during focal enhancement of layer 6. Burst spikes are colored red; tonic spikes are coloured black. PSTHs are plotted with 5 msec bins. (Data summarized from Wang et al., submitted.) (See color plate 20.)

Spatial Interactions

The corticofugal system also exerts a subtle but clear influence on the spatial properties of the LGN relay cell field. The fundamental effect seems to be an enhancement of the strength of the inhibitory surround in the presence of moving stimuli, so cells are more strongly patch-suppressed (and end-stopped) and the excitatory discharge zone for a moving stimulus is more focused (Cudeiro and Sillito, 1996; Jones et al., 2000; Murphy and Sillito, 1987; Sillito et al., 1993). Thus, the records in Figure 37.4A give an example of the length tuning of simultaneously recorded LGN cells, one with corticofugal feedback and one without. The loss of surround suppression in the absence of feedback is clear. The effect over the group of LGN cells tested is summarized by the block histogram in Figure 37.4B. These effects occurred without changing the degree of suppression associated with a stationary flashing stimulus of varying diameter. The enhancement of the inhibitory surround for moving stimuli also seems to lead to an increased sensitivity to orientation contrast, direction contrast, and temporal/phase contrast between center and surround mechanisms (Cudeiro and Sillito, 1996; Sillito and Jones, 1997; Sillito et al., 1993, 1999). The examples in Figure 37.5A–D illustrate the way LGN cells are influenced by direction (A), orientation (B, F), and temporal frequency (D) contrast in the presence of feedback, while the examples in Figure 37.5C and 37.5G show how the sensitivity to orientation contrast between a central and a surround stimulus is influenced by corticofugal feedback. In the absence of feedback, the sensitivity to the difference in orientation is greatly reduced, as summarized by the population histograms in Figure 37.5E. We stress that this reflects a sensitivity to orientation contrast and is not influenced by the absolute orientation of the stimulus. This point is underlined by the surface plot in Figure 37.5F, which shows the response of an LGN cell when the orientations of a central and a surrounding stimulus are varied independently in a random and interleaved fashion. Those instances in which the orientations of the inner and outer stimuli are the same lie along a diagonal from top right to bottom left. This diagonal is marked by a trough of low response levels. Thus, the cell signals orientation contrast. These effects all seem to follow from an increased resolution of the receptive field center-surround mechanism for a discontinuity in moving stimuli and serve to highlight the points of change in the input to the cortex. The influence of the feedback in this sense might broadly be described as enhancing fine-scale segmentation.

Figure 37.4..  

Effect of corticofugal feedback on LGN cell length tuning. A,B, Length tuning curves for two simultaneously recorded LGN cells. One was recorded from an LGN with corticofugal feedback (A), the other (B) from the other side of the brain from an LGN without feedback. The responses evoked by optimal and larger than optimal stationary spots of light flashed over the receptive field are superimposed on each graph (denoted by the filled spots). C,D, Bar histograms summarizing the distribution of end-inhibition with (C) and without (D) feedback. Cells are grouped into 10 categories for degree of end-inhibition. Category 10 contains cells showing total suppression of response at longer bar lengths, while those in category 1 show no suppression. (Observations summarized from Murphy and Sillito, 1987.)

Figure 37.5..  

Corticofugal feedback enhances local segmentation effects in LGN cells. A, LGN cell response to direction contrast. The bar histogram shows the cell's response (spikes/second + 1 SE) to three stimulus conditions (shown diagrammatically by the stimulus icons above). These comprised a circular patch of grating overlying the receptive field (RF) center, the introduction of a surrounding field of identical grating drifting in the same direction of motion, and the presence of a surrounding grating with its direction of drift reversed. Drift direction is denoted by white arrowheads. B, LGN cell response to orientation contrast. Stimulus details as for A, but the last record shows the effect of changing the orientation of the surround grating. C, LGN cell response to orientation contrast in the absence of feedback. Stimulus details as for B. D, LGN cell response to temporal/phase contrast. Stimulus details as for A, but the last record shows the effect of changing the drift rate of the surrounding grating. E, Influence of feedback on orientation contrast. The histograms plot the percentage change in response observed between iso-oriented and orthogonally oriented surround configurations (normalized with respect to the center-only response) in the presence (upper panel) and absence (lower panel) of corticofugal feedback. The mean increase in response magnitude for the switch from iso- to cross-oriented surround was 24.3% in the presence of feedback, which was reduced to 5.6% without feedback. F, LGN cells are sensitive to orientation contrast. The surface plot shows the response of an LGN cell (in the presence of feedback) to varying the orientation of an inner patch of grating in the presence of an outer patch of grating also of varying orientation. The diagonal running from bottom left to top right represents all those points where the orientation of the center and surround stimuli were the same over a complete sequence of absolute orientations. Response magnitude is shown by the height and shading of the contour. G, Reduction in orientation contrast in the absence of feedback. Stimulus details as for F. Note the absence of the pronounced diagonal trough. (Observations summarized from Cudeiro and Sillito, 1996; Sillito and Jones, 1997; Sillito et al., 1993, 1999.)

The other side of the coin to segmentation is integration. The input from common elements belonging to, say, a moving contour needs to be integrated by a mechanism detecting the presence of the contour. At the most basic level, simple cells in the visual cortex integrate the inputs from LGN cells to detect the presence of contours with a particular orientation (Fig. 37.6). An appropriately oriented bar would coactivate the converging inputs illustrated in Figure 37.6 and synchronize their firing. The degree of synchronization of the inputs has a strong bearing on their ability to influence the simple cell. Indeed, evidence shows supralinear enhancement of transmission from heterosynaptic geniculate inputs to layer 4 simple cells in the visual cortex for spikes occurring within ∼5 msec of each other (Usrey et al., 2000). It is thus very interesting that the feedback-enhanced center-surround antagonism influences the stimulus-driven synchronization (Andolina et al., submitted) of the discharges of LGN cells when they are precisely coactivated by a moving contour. We suggest in Figure 37.7 that the enhanced surround antagonism “focuses” the effective spatial extent of the receptive field center of LGN cells, leading to greater precision in firing when the two inputs are precisely coactivated. The evidence for this is summarized in Figure 37.6A–D. The experiments involved recordings in the cat LGN A laminae, using electrode assemblies configured to sample three cells of varying separation (Fig. 37.6A) in the presence and absence of feedback (Andolina et al., submitted). A grating patch was centered over one of the fields and was used to record the responses of the three cells at a range of orientations that included the angles linking each cell pair. Raw cross-correlograms (which show the stimulus-linked information that the second-order neuron sees) were then computed. These reflect the synchronicity of the inputs as “seen” by a theoretical simple cell for each pair at each orientation. Figure 37.6B shows representative results for a pair of X cells. A 5 msec window centered at zero lag (red) indicates those spikes that might generate supralinear enhancement of transmission. Note that shifts in orientation as small as 2° to either side of the angle linking the fields changed the count in this integration window. Converting the data to a surface representation of cross-correlogram time against grating orientation (Fig. 37.6C) reveals that the peak of the cross-correlogram shifted systematically throughout the time domain as orientation was varied. Interestingly, in the absence of feedback, this shift was less pronounced (Fig. 37.6D). As would be expected from this result, orientation tuning curves constructed from the synchronized spikes within the 5 msec window were much broader in the absence of feedback (see Fig. 37.6E, F, mean half width at half height 4.4 degrees with feedback and 13.17 degrees without feedback). Essentially this means that feedback greatly enhances the sensitivity to orientation in the stimulus-driven synchronization of the firing of LGN cells and thus the input to simple cells. In this sense, the organization is linked to contour and possibly motion integration.

Figure 37.6..  

Corticofugal feedback influences stimulus-linked synchronization between LGN cell pairs. A, Experimental paradigm. The responses of pairs of LGN cells to drifting gratings of varying orientation were recorded, and raw cross-correlograms were constructed for each stimulus orientation. B, Raw cross-correlograms recorded from a pair of LGN cells for three stimulus orientations. The shaded bar centered at time 0 indicates a 5 msec time span and highlights the correlated events occurring within the supralinear integration window of a cortical layer 4 cell. C,D, Surface representation showing the cross-correlation data (y-axis) versus orientation (x-axis) for LGN cells recorded in the presence (C) and absence (D) of feedback. The color scale represents the number of correlated events from low (dark blue) to high (dark red). E, Schematic example of an orientation tuning curve derived from the number of correlated events occurring in a 5 msec integration window. F, Orientation tuning curves for two LGN cells plotting the number of events in a 5 msec integration window recorded in the presence (black line) and absence (red line) of feedback. (Observations summarized from Andolina et al., submitted.) (See color plate 21.)

Figure 37.7..  

Diagram summarizing the view that feedback-linked enhanced surround antagonism leads to greater synchronization of LGN cell firing to coherent contours and thus increases the probability of firing in cortical layer 4 cells.