Decoding time-dependent nonlinearities underlying responses to natural stimuli

About This Grant

Summary and expected impact Our understanding of retinal processing, built from decades of work with artificial stimuli, fails to account for responses to natural images and movies. Current mechanistic models fail to generalize across stimuli, while deep learning models with better generalization lack interpretability. This proposal will help close this gap by investigating time-dependent nonlinearities with particular emphasis on bipolar cell synapses—critical sites that exhibit strong synaptic plasticity and serve as convergence points for parallel pathways. We hypothesize that current models fail because they do not capture how diverse time-dependent mechanisms at bipolar synapses and other circuit locations work together during natural viewing. Aim 1: Standard retinal models trained on artificial stimuli fail to engage the time-dependent nonlinearities that dominate responses to natural stimuli. We will develop stimuli that strongly engage time-dependent nonlinearities specifically at bipolar cell output synapses, probing synaptic depression and recovery dynamics. We will then develop dynamic models incorporating time-varying nonlinearities, including our validated biophysical photoreceptor model coupled with synaptic plasticity models, to isolate and quantify bipolar synaptic contributions relative to other adaptive mechanisms. Aim 2: We will determine how time-dependent nonlinearities differ across retinal pathways. First, we will characterize time-dependent nonlinearities in rod versus cone bipolar synaptic outputs at mesopic light levels. Second, we will systematically map time-dependent nonlinearities in commonly encountered RGC types in both primate (midget, parasol) and mouse (alpha) retina, measuring spike outputs, excitatory inputs, and inhibitory inputs to determine where nonlinearities arise and how pathway interactions create emergent properties. This work will advance understanding of retinal computation by creating interpretable models that incorporate the diverse time-dependent nonlinearities at bipolar synapses and other circuit locations. The resulting models will aid in the design of retinal prostheses that must function under natural viewing conditions and provide insights into visual processing deficits.