Neural encoding of the rod photoreceptor response

Abstract

Abstract: : Purpose: Understanding the neural code is a central requirement for advanced technologies that aim to interface and communicate with the CNS. Most of what is known about the neural code is within the context of spiking CNS neurons. Very little is known about the statistical properties of the neural code of photoreceptors. In the present work a computational model is developed that describes the encoding of stepped light stimuli by rod photoreceptors. Methods: The conceptual foundation of the present work is a theoretical model of the dynamics of light adaptation (1) which describes the dynamic light adapting rod response in terms of summating elemental photic signals: u_j(i) = ω_j ε_DA(i,I_b) + (1–ω_j) ε_LA(i,I_b), where ω_j is a weighting parameter that determines the peak amplitude and shape of the elemental response u_j(i) at times j, and ε_DA(i,I_b) and ε_LA(i,I_b) are the derived single photon responses in the dark and maximally light–adapted conditions (see (1) and (2)). From this, a statistical model based on Bayes' rule was developed, adapted from similar statistical encoding/decoding models characterizing the neural code of CNS spiking neurons (3). Implementations of the present model were done in Matlab (Mathworks, Natick, MA). Results: The dynamic rod response to the onset of stepped backgrounds of light (e.g. intensities on the order of I_b = 1.2 and 0.4 sc cd m^–2) was determined as it evolved towards steady state without a priori knowledge of the response, in effect representing the neural encoding of the input stimuli. Inputs S(t) were assumed to occur with probability P[S(t)], while intermediate light adapting single photon responses u_j(max_i,int_i) (i.e. the single photon rod response as it evolves towards steady state) were assumed to be taken from the probability distribution P[u_j(max_j,int_j)]. The kinetics of u_j(max_i,int_i) are defined in terms of the responses' peak amplitude (max_j) and integration time (int_j), the full time course of the responses being computed using an optimization algorithm. Conclusions: The neural encoding of rod photoreceptor responses to stepped background light inputs were modeled in terms of dynamic single photon responses who's evolution towards steady state are defined within the context of Bayesian statistics. The structure of the present model may provide a framework for modeling and understanding how rods encode and represent dynamic light stimuli, thereby complimenting our understanding of neural encoding/decoding by spiking neurons. (1) Silva and Pepperberg (in press) IEEE Trans. Biomed. Eng., January 2004. (2) Silva et. al. (2001) J. Physiol. 543:203. (3) Rieke et. al.(1997) Spikes: Characterizing the neural code. MIT Press.

Keywords: computational modeling • photoreceptors • signal transduction