Adaptive spatiotemporal filtering of social electrosensory signals by cerebellar feedback

Sensory feedback can suppress predictable input, yet most research on this mechanism used simplified stimuli that do not capture the spatial and temporal structure of natural signals. We examined how cerebellum-like feedback, in brown ghost knifefish, shapes the encoding of realistic social electrosensory signals. These naturalistic signals replicated the spatial structure of social signals, spanned the range of typical AM beat frequencies, and incorporated movement-related contrast envelopes. Single-unit recordings were obtained from electrosensory lateral line lobe pyramidal cells while a conspecific-like signal was delivered. Responses were compared before and after pharmacological blockade of descending cerebellar feedback. As expected, feedback strongly attenuated responses to spatially-realistic low-frequency beats but did not alter beat-cycle coding at higher frequencies. However, feedback reduced the neural representation of movement-related contrast envelopes across all beat frequencies. Using information-theoretic tools, we showed that feedback strength adapts dynamically to the stimulus contrast over the past few hundred ms. Furthermore, we examined how the spatial coding by the pyramidal cell population is influenced by feedback. We show that feedback enhances localization cues. This improvement arose because feedback reduces weak background responses while not affecting the strongest responses, thereby increasing spatial contrast. These findings show that the same descending pathway jointly filters temporal and spatial features of the full range of natural signals, adapting to recent input statistics. This process exemplifies a general mechanism by which a predictive feedback can improve the efficiency of natural sensory representations.