Shunting inhibition, a silent step in visual cortical computation

Abstract

Brain computation, in the early visual system, is often considered as a hierarchical process in which features extracted in a given sensory relay are not present in previous stages of integration. In particular, orientation preference and its fine tuning selectivity are functional properties shared by most cortical cells and they are not observed at the preceding geniculate stage. A classical problem is identifying the mechanisms and circuitry underlying these computations. Several organizational principles have been proposed, giving different weights to the feedforward thalamocortical drive or to intracortical recurrent architectures. Within this context, an important issue is whether intracortical inhibition is fundamental for the genesis of stimulus selectivity, or rather normalizes spike response tuning with respect to other features such as stimulus strength or contrast, without influencing the selectivity bias and preference expressed in the excitatory input alone. We review here experimental observations concerning the presence or absence of inhibitory input evoked by non-preferred orientation/directions. Intracellular current clamp and voltage clamp recordings are analyzed in the light of new methods allowing us (1) to increase the visibility of inhibitory input, and (2) to continuously measure the visually evoked dynamics of input conductances. We conclude that there exists a diversity of synaptic input combinations generating the same profile of spike-based orientation selectivity, and that this diversity most likely reflects anatomical non-homogeneities in input sampling provided by the local context of the columnar and lateral intracortical network in which the considered cortical cell is embedded.

Introduction

A prevailing concept in the role of thalamocortical pathways in sensory processing is the dominant influence of feedforward connectivity. In the case of the mammalian visual system, it is well established that topographic maps and the organization of visual receptive fields in ON and OFF discharge zones result from the strong imprint of the feedforward input ([27] review in [24]). Hierarchical models of visual processing assume that the spatial convergence of afferents from one relay to the next determines the functional architecture of the target structure. For example, these models predict that the unique selectivity of the cortical cell firing to oriented contours found in primary visual cortex is obtained by the precise combination of inputs arising from geniculate cells of the same type (ON or OFF) whose receptive fields are offset and aligned along a given axis in the visual field. Electrophysiological correlates have been reported at least in the thalamic input recipient layer in ferret and cat visual primary cortex [11], [37].

Such a view, which requires extreme precision in the wiring of extrinsic thalamic afferents to primary sensory areas, may also apply to interareal connections at a higher level of cortical processing. Recent evidence, often interpreted as supporting the serial feedforward nature of processing up to perceptual decision centers, comes from latency measurements of scalp event-related potentials in frontal cortical areas. Changes in the electrical activity correlated with complex cognitive tasks, such as deciding on the animal vs. non-animal identity of briefly flashed images, can be detected as early as 150 ms following the image onset in the human and monkey observers [45]. This type of evidence seems to indicate, at least for certain perceptual tasks, a strong timing constraint that limits the involvement of lateral and recurrent cortical processing, given the number of synapses to be serially crossed along the cortico-cortical pathway. According to this schema, no time is left for the short- and long-distance intrinsic recurrent connections nor for the feedback from higher areas, except but to amplify or tune the functional bias set by the feedforward connectivity.

A fundamental issue in terms of computation is to understand whether this apparent feedforward dominance faithfully reflects the amount of intracortical processing, or if more complex spatio-temporal non-linear interactions between feedforward and intracortical synaptic events have to be considered. We will review both models and electrophysiological data, which argue for an important role of local intracortical inhibitory processing, the effect of which is often hidden at the level of single cell spiking behaviour.