Efficient representations of fast dynamic stimuli in populations of ON- and OFF-cells

Jan Benda – Universität Tübingen
Jan Grewe – Universität Tübingen
Benjamin Lindner – HU Berlin

Splitting information processing into separate ON/OFF pathways is a common design pattern in sensory systems. This allows to encode increases and decreases of stimulus amplitudes with lower average firing rates and consequently is energetically more efficient compared to dense codes typically found at the level of receptor neurons. The reduced firing rates, however, pose a severe problem for encoding dynamic stimuli with frequency components much higher than typical firing rates. For the electrosensory system of electric fish we know about the relevance of such fast stimuli in the context of courtship behavior. How is this possible despite intrinsic noise and limited number of neurons? Here we would like to address the question about the optimal design of a population of ON/OFF neurons for encoding high-frequency stimuli. To this end we combine electrophysiological experiments on the electrosensory system of weakly electric fish with analytical and numerical modeling approaches that allow to gain general insights into this fundamental problem. Theoretically, we will consider models of two populations of integrate-and-fire neurons, representing the ON/OFF neurons. We will first consider populations of uncoupled neurons that are driven by a common (global) broadband stimulus and quantify the information transfer in a weighted sum of the two populations' output. We ask for the optimal ratio of ON/OFF cells and the optimal distribution of firing rates for efficient information transmission. With the electrosensory system in mind, we then add (i) amplitude modulations of a carrier signal as stimulus and (ii) inputs that arrive through another layer of cells, the electroreceptors or P-units that are modelled by stochastic and adapting leaky integrate-and-fire neurons with a burst mechanism. On the experimental side, we extracellularly record populations of ON/OFF pyramidal neurons in the hindbrain of electric fish using silicon probes and analyze the encoding of fast amplitude modulations and beats of the fish's electric field. We will test the model predictions with respect to the optimal distribution of ON/OFF cells and their firing rates. In a second step we explore the effects of the delayed feedback originating from the Pd nucleus that provides a diffusive feedback of the population’s output activity on high-frequency coding. On the theory side we include a feedback term to the model. In the experiments we block the feedback pathway pharmacologically. In this way we can both experimentally and theoretically assess the role of this feedback pathway on processing of fast signals.