The vertebrate retina translates visual images into electrical signals in the optic nerve, initiating the basis of all visual perception. This process is accomplished by dozens of diverse types of interneurons, each of which comprises a population of many thousands of cells. Each of these populations cover the visual field, acting together to process different aspects of visual images. Although many informative studies of retinal neural function have used single cell recordings, understanding the coordinated actions of many cells requires the recording and analysis of cell populations. This proposal focuses on amacrine cells, a diverse population of inhibitory interneurons. In particular we study wide-field amacrine cells, a prominent class of cells that make long distance connections across the retina, acting to combine visual signals from distant locations in the image. We have little information assigning computations to specific cells of this type. Using genetically identified populations of wide-field amacrine cells in the mouse retina, we will record neural activity from these populations optically, along with simultaneously recording electrically from populations of retinal ganglion cells. Neural responses to complex stimuli including natural scenes will be interpreted using advanced computational models. The primary goals of these studies are to 1) perform the first population scale measurements of sparse wide-field amacrine cells, in particular to measure how their selectivity for visual features varies dynamically during natural scenes, 2) Analyze the neural code of these cells under natural scenes using state-of-the-art computational models that can capture retinal responses to arbitrarily complex stimuli, 3) Test the hypothesis that sparse wide-field amacrine cells perform similar computations on different channels of information, acting to remove correlations from the ganglion cell population during natural scenes. These results will have immediate applicability to the emerging field of retinal prostheses, as is used to treat prevalent diseases such as age-related macular degeneration and retinitis pigmentosa by replacing the function of the damaged retina with a high resolution electronic circuit. Measurements of the retinal neural code and the computations that are performed will be directly useful for incorporation into retinal prosthesis systems.
Baccus, Stephen A. The retina is a complex network of many cell types, including the most diverse but poorly understood class of cells, inhibitory neurons. By understanding how inhibitory neurons represent visual information, we can begin to address how these cells and their connections degenerate during retinal diseases, an essential step in designing treatments for these diseases. Furthermore, by capturing the responses of these neurons under natural visual stimuli with accurate computational models, this research will be immediately applicable to electronic retinal prosthesis systems that aim to restore vision in cases of photoreceptor degeneration.
| Maheswaranathan, Niru; Kastner, David B; Baccus, Stephen A et al. (2018) Inferring hidden structure in multilayered neural circuits. PLoS Comput Biol 14:e1006291 |
| McIntosh, Lane T; Maheswaranathan, Niru; Nayebi, Aran et al. (2016) Deep Learning Models of the Retinal Response to Natural Scenes. Adv Neural Inf Process Syst 29:1369-1377 |
| Jadzinsky, Pablo D; Baccus, Stephen A (2015) Synchronized amplification of local information transmission by peripheral retinal input. Elife 4: |
