Circuit mechanisms for the construction of invariant representations

About This Grant

Project Summary Visual processing transforms sensory inputs into signals used to guide behavior. These transformations occur over multiple stages of the cortical hierarchy to generate representations that are increasingly invariant to nuisance features such as size, angle and lighting. Such invariance is fundamental for scene segmentation and object recognition. Thus, understanding the cellular and circuit mechanisms underlying these transformations will provide key insights into cortical function. In this project, we will dissect invariant coding of motion direction to the specific components of the stimulus. For instance, when multiple drifting gratings are superimposed to form a plaid, the perceived direction is in the pattern direction, rather than that of the individual components. “Pattern cells” that encode this direction have been identified in the visual system of primates, carnivores and most recently rodents, suggesting that these cells are important for the perception of motion direction. The predominant model proposes that pattern cells are constructed through a series of sequential transformations: direction selectivity is established first, followed by spatial invariance and finally motion invariance. Our preliminary data from mice and marmosets argue against this clean hierarchical transformation, revealing the presence of spatially dependent pattern cells in V1 of mice and marmosets. Here we will test the hypothesis that the foundation for pattern motion is first laid by spatial inhomogeneities in the feedforward pathway and that the major computation is to generate a spatially invariant representation in the higher visual areas (HVAs). In Aim 1 we will examine how thalamic drive supports pattern selectivity in V1 of rodents and primates. We will use extracellular recordings to compare how spatial receptive fields create pattern selectivity and then use intracellular recordings in conjunction with optogenetics to determine how thalamic inputs sculpt these representations. These data will be used to build a receptive field model of cortical processing that explains cellular responses to plaids in V1. In Aim 2, we will determine how motion and spatial invariance increase from V1 to the HVAs in marmosets and mice (areas MT and AL). We will interrogate this transformation by using pre- and post-synaptic imaging approaches to test the hypothesis that spatially variant signals in V1 converge onto individual neurons in the higher areas to generate invariance. In Aim 3, we will investigate the contribution of local inhibitory circuits to pattern selectivity. We will use two-photon imaging to assess the plaid direction selectivity of inhibitory interneurons, and whole-cell recordings to determine how inhibitory inputs are integrated. Finally, these data will be used to elaborate our computational model to determine the role of recurrent circuits in the generation of pattern selectivity. Together, these experiments will provide a mechanistic explanation for the construction of an invariant cortical representation, from the level of synaptic inputs onto molecularly defined cell types within and across cortical areas in two species.