Brandon R. Nanfito^1,2,3, Kristina J. Nielsen^1,2,3; ¹Johns Hopkins School of Medicine, ²Zanvyl Krieger Mind/Brain Institute, ³Kavli Neuroscience Discovery Institute

Sensory systems must explicitly signal information from stimulus inputs, which are usually composed of multiple feature dimensions (e.g. the orientation and direction of drifting gratings). Traditionally, tuning functions are used to quantify how informative single neurons are about a particular stimulus parameter. However, the response pattern from a population of tuned neurons is necessary to disambiguate stimulus identity, particularly in the face of higher dimensional stimuli. In adult animals, the population response has been used to evaluate how well an area distinguishes between stimuli across multiple feature dimensions. Here, we extend this approach to evaluate and compare the function of primary visual cortex (V1) and higher motion area PMLS in developing ferrets. Single neuron responses to gratings drifting in twelve different directions were recorded using extracellular electrophysiology in animals ranging from post-natal day 30 (around eye opening) to 52 (when orientation and direction tuning has matured). We then used principal component analysis to capture any low-dimensional structure in the population response, followed by linear discriminant analysis on the full dimensional data to further evaluate what information (e.g. orientation vs. direction) can be read out at each stage of development. Our results indicate that around eye opening, both areas represent orientation but not direction. In adult animals, however, they have distinct representational geometries. Adult PMLS responses maximally separate gratings moving in opposing directions despite having a common orientation. Thus, PMLS is optimized for encoding direction at the expense of orientation because the representations of each orientation is split across an axis of motion. V1 however balances separating different directions with clustering similar orientations. This difference in functional specialization appears just a few days after eye opening. The decoding analysis demonstrates this comes at the expense of retrieving orientation information in PMLS, whereas V1 maximizes performance in both feature dimensions.

Acknowledgements: This work was supported by the NIH (1R01EY035807) and the Kavli NDI