Nonlinear receptive fields evoke redundant retinal coding of natural scenes

Area of Science:

• Neuroscience
• Vision Science
• Computational Neuroscience

Background:

• The efficient coding hypothesis suggests the retina minimizes redundancy in natural scenes by discarding spatiotemporal correlations.
• Ganglion cells, the retina's output neurons, are predicted to exhibit decorrelation and redundancy reduction under natural visual input.
• The impact of gaze shifts, a dominant factor in natural vision, on these retinal coding principles remains unclear.

Purpose of the Study:

• To investigate whether retinal ganglion cell activity maintains decorrelation and redundancy reduction during natural gaze shifts.
• To determine if gaze patterns influence the coding efficiency of retinal output neurons in marmoset and mouse models.

Main Methods:

• Recorded spiking responses of distinct retinal ganglion cell types in marmoset and mouse.
• Utilized natural stimuli with species-specific gaze patterns to analyze neural activity.
• Employed model-based analyses to investigate the mechanisms underlying observed response correlations.

Main Results:

• Species-specific gaze patterns induced correlated spiking responses within and across different ganglion cell types.
• These correlated responses disrupted predicted redundancy reduction, signaling periods of high spatial contrast during fixation.
• Nonlinear pooling of ganglion cell inputs explained the observed response correlations.

Conclusions:

• Retinal processing of natural gaze shifts deviates from the predictions of the efficient coding hypothesis.
• Correlated activity in ganglion cells plays a role in signaling salient visual information during natural viewing.
• Findings highlight cell-type-specific adaptations in retinal coding under dynamic visual conditions.