Testing efficient coding in realistic models of the retinal network

点击翻译按钮获取中文摘要

This project aims at developing a detailed model of the excitatory pathway of the retina and to test if it follows an efficient coding strategy of visual information. For this, we will first acquire experimental data to decompose the different steps of retinal processing. We will then use these data to build models of the retinal excitatory pathway that can predict and explain how complex inputs, i.e. natural images, are processed at the frontend of vision. Finally, we will use these models to test the hypothesis that the organization of this pathway is compatible with the principles of efficient coding. The retina’s excitatory pathway consists of three steps: First, photoreceptors transduce light into electrical activity and transmit the signal via specialized glutamatergic (“ribbon”) synapses to bipolar cells (BCs). Next, BCs pool from several photoreceptors and relay their signal again via another ribbon synapse to retinal ganglion cells (RGCs). Many studies have characterized the processing along this pathway, however, how upstream processing steps shape a RGC’s response properties when presenting natural images are far from understood. Here, we will record the responses of RGCs to natural images and use novel tools to investigate the contribution of BCs. First, to characterize the excitatory input impinging on RGCs and the resulting postsynaptic potentials, we will record BC output and RGC dendritic voltage using two-photon (2P) imaging with genetically-encoded glutamate and voltage sensors, respectively, while showing natural images to the photoreceptors. Second, to study how the BC output is integrated at the RGC level to generate spike trains, we will combine advanced 2P digital holography with optogenetics to selectively stimulate individual BCs while recording the impact of this stimulation on the RGC spiking using multielectrode arrays (MEAs). Here, the stimulation patterns will reproduce how BCs respond to flashed natural images. Next, we will construct a model that integrates these data: BC output, RGC dendritic voltage, and RGC spiking in response to natural images, and RGC responses to holographic BC stimulation. Integrating these heterogeneous data – consisting of synaptic output, postsynaptic voltage and spikes, as well as different modes of spatio-temporal stimulation – in a single model is a novel challenge. However, we expect that building and testing such a model will give unprecedented insight into how natural images are processed by the retinal excitatory pathway. Finally, having an accurate model of this pathway, we will be able to test quantitatively if its organization is compatible with efficient coding principles. For this, we will take advantage of novel methods to test if complex, non-linear models are optimizing information transmission.