Synergistic geniculate and cortical dynamics facilitate a decorrelated spatial frequency code in the early visual system

Abstract

Sensory stimuli are encoded by the neuronal firing patterns they evoke in the brain. This neural code becomes less correlated as information ascends through the visual system. In the primary visual cortex (V1), a spatial frequency (SF) tuning shift from coarse to fine features occurs alongside a reduction in correlations between stimulus representations. Our previous study suggested that this decorrelation is facilitated by coarse-to-fine processing in V1. However, there is evidence that coarse-to-fine processing emerges in the upstream dorsal lateral geniculate nucleus (dLGN), and it is unknown whether correlations between stimulus representations also decrease in this brain region. Therefore, the extent to which decorrelation is inherited from dLGN, is driven by local circuit dynamics in V1, or is the result of synergy between these areas is unknown. In this study, we compared extracellular neuronal activity recorded from dLGN and V1 of mice (of either sex) in response to sinusoidal gratings of different SFs. Despite also exhibiting coarse-to-fine processing, dLGN did not exhibit decorrelation in contrast to V1, suggesting that decorrelation emerges following a cortical transformation. In V1, many units exhibited a delayed shift to suppression that interacted with coarse-to-fine shifts on a time course coinciding with the decorrelation. Our results are therefore consistent with decorrelation emerging in V1 from a synergy between response properties in both dLGN and V1. These results demonstrate that geniculocortical dynamics enable discrimination between rich visual details and highlight the importance of cross-regional synergy to sensory processing.

Significance Statement The visual system encodes large amounts of information about the physical world with limited resources. Encoding is more efficient when the responses of neurons are less correlated with each other, but the mechanisms that achieve decorrelation in the early visual system remain largely unknown. Here, we show evidence that response properties in dLGN and V1 lead to population-wide fluctuations in the balance of enhancement and suppression that dynamically reduce correlations between low and high SFs. These results suggest that SF information is encoded most efficiently during a population-wide transition from enhancement to suppression and highlight the importance of geniculocortical circuits in encoding rich visual information.