How nonlinear processing shapes natural stimulus encoding in the retina
by Dimokratis Karamanlis
Date of Examination:2022-02-23
Date of issue:2022-05-31
Advisor:Prof. Dr. Tim Gollisch
Referee:Prof. Dr. Tim Gollisch
Referee:Prof. Dr. Alexander Gail
Files in this item
This file will be freely accessible after 22.02.2023.
EnglishUnderstanding natural vision is one of the fundamental goals of sensory neuroscience. The only part of the visual system that may currently be amenable to such a complete understanding is the vertebrate retina, where there is increasing convergence of physiological and anatomical evidence that explain the transformation from visual stimulus to neural response. A first-order approximation of retinal processing assumes that the retina acts as a parallel stream of linear filters, interfacing between the photoreceptors and the axons of retinal ganglion cells, which form the optic nerve. Computational models based on linear filtering only partially capture the retinal output under naturalistic stimulation. The shortcomings of this linear picture of the retina come as no surprise in the light of the multiple examples of nonlinear processing within the retinal circuit. The prime example is associated with signal transduction between bipolar and ganglion cells, which can be highly nonlinear. However, these nonlinearities are mostly studied in isolation with targeted artificial visual stimuli, and their significance for natural stimuli is still unclear. Using multielectrode-array recordings from the isolated mouse retina, I investigated the necessity of nonlinear processing for natural image encoding and found differential sensitivity of ganglion cells to natural spatial structure. I then showed that this sensitivity is a property of nonlinearities acting over the receptive field center, and I established that different nonlinearity types can be found among different types of retinal ganglion cells. I developed models that can leverage nonlinear properties of the retinal circuit and tested their predictions with responses to natural scene stimuli. This modeling approach yielded complete nonlinear receptive field descriptions of major cell types in the mouse retina and could capture the retinal output under stimuli with naturalistic temporal dynamics. Together, I showed that nonlinear retinal processing is highly relevant for the encoding of natural scenes, extends functional descriptions of neuronal types beyond linear receptive fields, and can reduce the gap in predicting the retinal output to natural visual inputs.
Keywords: retina; receptive field; retinal ganglion cell; nonlinear spatial integration; natural stimuli; multielectrode arrays; encoding models