Feedforward and feedback processing
Traditionally the visual system has been viewed as a feedforward hierarchy: early visual areas are sensitive to simple features in the scene due to the spatial arrangement of their inputs and higher visual areas are tuned for more complex combinations of features or entire objects. It is now clear that this description is incorrect. Responses in early visual areas are modulated by cognitive factors such as attention and task demands and they are sensitive to global contextual factors such as figure-ground organization.
Our research has focused on understanding the mechanisms by which these modulations occur and how such changes in activity in early visual areas correlate with the behavior of animals performing cognitive tasks. We have proposed that early visual responses in primary visual cortex (V1) are mainly driven by feed-forward processing and are not sensitive to contextual information, whereas later responses (>100ms after stimulus onset) are affected by feedback from higher visual areas. We have shown that late-period activity correlates with the perception of animals on figure-ground tasks and can predict the responses of animals in curve-tracing tasks suggesting that feedback makes more high-level forms of knowledge accessible to V1 neurons.
The activity of cells in V1 is affected by context. We record spiking activity from cells when their receptive field (green region) is placed on a figure (blue) or a background (purple). The initial response of the cell is driven by feed-forward input and is sensitive to the orientation of the line-elements inside the receptive field. This early response cannot discriminate between figure and background. After 100ms the cell’s activity increases when its RF is on the figure and is suppressed when it falls on the background. This later response is thought to be due to feedback projections from higher visual areas.
Interactions between lower and higher areas of visual cortex
Ongoing research in our group is aimed at examining how attentional feedback from higher visual areas onto lower visual areas enables visual perception. We primarily study this in humans and non-human primates. However, the precise mechanisms by which feed-back and feed-forward connections modulate neural activity are currently not fully understood. While lacking the organizational sophistication of primate visual cortex, several recent studies have systematically mapped the global topography of visual cortical areas in mice. This detailed parcellation of mouse visual cortex has delineated nine know areas, and revealing two additional areas previously not described as being visuotopically mapped in mice.
Using such topographic maps and defined area boundaries, several features of map organization have been characterized. These advances have important implications for the mouse as a model for vision research. Additionally, with the development of GCaMP transgenic mice, we now have unprecedented new ways of monitoring and understanding brain function. GCaMP, the genetically encoded green fluorescent protein-based calcium indicator, provides a powerful tool for detecting calcium transients induced by stimulus-evoked action potentials at the level of individual cells and synapses in awake, behaving animals. This allows for the monitoring of activity within specific subregions of the brain that are engaged or activated by a particular cognitive task.
Hence, by utilizing GCaMP expressing mice and training them to perform cognitive tasks that are as analogous as possible to those used in humans and non-human primates, we are currently developing a novel approach of imaging large regions of the mouse cortex using widefield transcranial imaging techniques. By doing so, we hope to unravel some of the complex structural and functional interactions of neural networks that underlie visual perception, particularly in relation to feed-back and feed-forward connections.
Role of cortical layers
Neocortex consists of six layers with a remarkable degree of similarity across different cortical areas. Cells in the same layer have stereotypical patterns of connectivity, both to cells within the same layer and to cells in different layers. Furthermore connections to other cortical and sub-cortical areas originate and terminate in particular layers. This has lead to the idea of a standard cortical micro-circuit which implements similar calculations in different areas, but is flexible enough to be applied to multiple tasks.
Feedforward connections primarily target layer 4, driving strong responses which spread through the cortical micro-circuit and are typically tuned for the orientation of small texture elements. Feedback projections to V1 target layers 1,2 and 5 and lead to modulations of activity particularly in layers 5/6 and 2/3. Feedback labels the square figure (shown below) with enhanced activity, helping to perceptually segment the square from its background.