Summary This review considers the influential notion of a canonical (cortical) microcircuit in light of recent theories about neuronal processing. Specifically, we conciliate quantitative studies of microcircuitry and the functional logic of neuronal computations. We revisit the established idea that message passing among hierarchical cortical areas implements a form of Bayesian inference – paying careful attention to the implications for intrinsic connections among neuronal populations. By deriving canonical forms for these computations, one can associate specific neuronal populations with specific computational roles. This analysis discloses a remarkable correspondence between the microcircuitry of the cortical column and the connectivity implied by predictive coding. Furthermore, it provides some intuitive insights into the functional asymmetries between feedforward and feedback connections and the characteristic frequencies over which they operate.

Simple cells within layer IV of the cat primary visual cortex are selective for lines of a specific orientation. It has been proposed that their receptive-field properties are established by the pattern of connections that they receive from the lateral geniculate nucleus (LGN) of the thalamus. Thalamic inputs, however, represent only a small proportion of the synapses made onto simple cells, and others have argued that corticocortical connections are likely to be important in shaping simple-cell response properties. Here we describe a mechanism that might be involved in selectively strengthening the effect of thalamic inputs. We show that neighbouring geniculate neurons with overlapping receptive fields of the same type (on-centre or off-centre) often fire spikes that are synchronized to within 1 millisecond. Moreover, these neurons often project to a common cortical target neuron where synchronous spikes are more effective in evoking a postsynaptic response. We propose that precisely correlated firing within a group of geniculate neurons could serve to reinforce the thalamic input to cortical simple cells.

Hundreds of thalamic axons ramify within a column of cat visual cortex; yet each layer 4 neuron receives input from only a fraction of them. We have examined the specificity of these connections by recording simultaneously from layer 4 simple cells and cells in the lateral geniculate nucleus with spatially overlapping receptive fields (n ϭ 221 cell pairs). Because of the precise retinotopic organization of visual cortex, the geniculate axons and simple-cell dendrites of these cell pairs should have overlapped within layer 4. Nevertheless, monosynaptic connections were identified in only 33% of all cases, as estimated by cross-correlation analysis. The visual responses of monosynaptically connected geniculate cells and simple cells were closely related. The probability of connection was greatest when a geniculate center overlapped a strong simple-cell subregion of the same sign (ON or OFF) near the center of the subregion. This probability was further increased when the time courses of the visual responses were similar. In addition, the connections were strongest when the simple-cell subregion and the geniculate center were matched in position, sign, and size. The rules of connectivity between geniculate afferents and simple cells resemble those found for retinal afferents to geniculate cells. The connections along the retinogeniculocortical pathway, therefore, show a precision that goes beyond simple retinotopy to include many other response properties, such as receptive-field sign, timing, subregion strength, and size. This specificity in wiring emphasizes the need for developmental mechanisms (presumably correlation-based) that can select among afferents that differ only slightly in their response properties. Key words: visual cortex; simple cell; thalamus; thalamocortical; LGN; correlated firingAlthough separated by a single synapse, geniculate cells and cortical simple cells have very different response properties. Geniculate cells have receptive fields with a circular center and a concentric, antagonistic surround. Simple cells have receptive fields with elongated, parallel subregions. According to the original hypothesis of Hubel and Wiesel (1962), simple receptive fields are constructed from the convergence of geniculate inputs with receptive fields aligned in visual space. This hypothesis has received experimental support for simple cells in layer 4 of cat visual cortex Alonso, 1995, 1996;Ferster et al., 1996;Chung and Ferster, 1998). Specifically, if the receptive-field center of a geniculate cell overlaps a simple-cell subregion of the same sign (ON or OFF), then there is a high probability that the simple cell and the geniculate cell will be connected. Otherwise, the probability of finding a connection is almost zero (Reid and Alonso, 1995).The position and sign of receptive fields, however, may not be the only relevant factors in determining connectivity. Differences in response timing (Cleland et al., 1971a;Hoffmann et al., 1972; Mastronarde, 1987a,b;Humphrey and Weller, 1988;Wolfe and Palmer, 199...

Retinal ganglion cells and their target neurons in the principal layers of the lateral geniculate nucleus (LGN) of the thalamus have very similar, center-surround receptive fields. Although some geniculate neurons are dominated by a single retinal afferent, others receive both strong and weak inputs from several retinal afferents. In the present study, experiments were performed in the cat that examined the specificity and strength of monosynaptic connections between retinal ganglion cells and their target neurons. The responses of 205 pairs of retinal ganglion cells and geniculate neurons with overlapping receptive-field centers or surrounds were studied. Receptive fields were mapped quantitatively using a white-noise stimulus; connectivity was assessed by cross-correlating the retinal and geniculate spike trains. Of the 205 pairs, 12 were determined to have monosynaptic connections. Both the likelihood that cells were connected and the strength of connections increased with increasing similarity between retinal and geniculate receptive fields. Connections were never found between cells with <50% spatial overlap between their centers. The results suggest that although geniculate neurons often receive input from several retinal afferents, these multiple afferents represent a select subset of the retinal ganglion cells with overlapping receptive-field centers.