presence of feedback loops to V1 from numerous areas along the visual cortical pathway allows the analysis of complex properties of the visual image, such as surface segmentation, to be referred back to earlier stages.
Anatomical evidence shows that at the early cortical levels the substrate exists for providing cells with input from relatively large parts of the visual field. From physiological studies, it now appears that the response properties of cells are modulated by stimuli lying outside of the “classical” RF. The structure and specificity of RFs and of cortical functional architecture are increasingly seen as context dependent. This may represent the cellular mechanism of perceptual studies showing that the visual system is capable of linking contours and surfaces in a process of perceptual fill-in (2–6), that our perception of the attributes of local features is influenced by context (7–14), and that simple contours can be picked out of a noisy background (15–18). The perceptual phenomena obey rules that are consonant with the patterns of connectivity in primary visual cortex, supporting the idea that a major component of the process of spatial integration occurs there. One can characterize a host of processes, referred to as intermediate-level vision, that occur between the discrimination of simple visual attributes, such as orientation, and the identification of complex objects. The neural mechanisms may be found at earlier stages than previously believed.
Our initial evidence of a cellular substrate for spatial integration was a pattern of long-range horizontal connections formed by the axons of pyramidal cells in V1 (19–23). The extensive plexus of horizontal connections was revealed in experiments which mapped the intrinsic cortical circuit and related cortical connections to RF properties by labeling the full axonal arbor of functionally characterized cells (Fig. 1). The findings were quite surprising, since they seemed to violate the principles of RF structure and cortical topography. Because of the extensive spread of the horizontal connections, their cellular targets are capable of integrating input over an area of cortex that represents an extent of visual field roughly an order of magnitude larger in area than the cells' own RFs. This finding contrasted with the belief that all the connections in the cortex are vertical, between cells with overlapping RFs and similar orientation preference, with relatively little lateral transfer of information. In effect it posed a contradiction in the definition of the RF, in that it suggested that cells should be sensitive to stimuli lying outside of the RF. The resolution to this conflict lies in the way the RF is defined, which is highly dependent on the stimulus used. When one uses a simple stimulus, such as a single short line segment of the appropriate orientation, one can activate a cell to suprathreshold levels over a very limited area, the classical RF. If, in addition to a line lying within the classical RF, one places additional stimuli outside the RF, the response of the cell changes. The nature of the change depends on the precise geometric relationship between the stimuli lying within and outside the RF, and it correlates well with the influence of context on the perception of local features.
The rules governing contextual influences are mirrored by the registration between the long-range horizontal connections and cortical functional architecture. The first relationship is the extent of visual space that is represented by the area of cortex over which these connections spread. The horizontal connections spread laterally up to 6–8 mm (19,20,25). As was shown by Hubel and Wiesel (26), a distance of two hypercolumn diameters (a hypercolumn is defined as a complete set of orientation columns), or roughly 1.5 mm, is the minimal cortical distance between cells with nonoverlapping RFs. The distance covered by the longest-range horizontal connections, therefore, separates cells with RFs that are several RF diameters apart. The second principle of organization of the horizontal connections is revealed by the clustered nature of their axon collaterals (Fig. 1). These clusters are separated by 0.5–1 mm, approximating the width of an individual hypercolumn. The relationship between the clusters and the columnar functional architecture was shown in several ways. Crosscorrelation analysis, which is a statistical technique relating the time of occurrence of action potentials in pairs of neurons and which provides a measure of effective connection strength between the neurons, showed that cells with correlated firing had similar orientation preference (27,28). Correlated firing was found even for cell pairs with nonoverlapping RFs, as would be expected for the distances spanned by the horizontal connections. The registration between the clustering of the horizontal connections and orientation columns was revealed anatomically by labeling the orientation columns by 2-deoxyglucose autoradiography and marking the horizontal connections with extracellularly applied tracers (25). This showed that the horizontal connections ran between columns of similar orientation specificity.
Visualization of Lateral Cortical Interactions with Optical Recording. More recently we have used optical recording to obtain a functional measure of the extent and specificity of the horizontal connections. Optical recording reveals in vivo the pattern of activity, projected onto the cortical surface, elicited by a given stimulus. It relies on changes in surface reflectance, referred to as the “intrinsic signals,” that are linked to metabolic changes resulting from local neural activity. Optical recordings of this signal have been used to visualize the cortical functional architecture, in particular the arrangement of orientation columns (29–32). We used optical recording to