MSTd Neurons Encode Nonlinear Combinations of Retinal and Extra-retinal Signals

Abstract

AbstractNeuronal activity in the dorsal Medial Superior Temporal area (MSTd) is assumed to depend on both retinal and extra-retinal input. Most of the neurons show activity when presented with a moving large-field visual stimulus. In addition, many MSTd neurons are activated during smooth pursuit eye movements even in the absence of retinal input. However, the interaction between the retinal and extra-retinal input is not yet fully understood.Here we present novel insights regarding the tuning of MSTd neurons for combinations of different input variables using an information-theoretic approach. Neuronal tuning functions can be expressed by the conditional probability of observing a spike given any combination of input variables. However, accurately determining such probabilistic tuning functions from experimental data poses several challenges such as determining the neuronal latencies and finding the combination of input variables which is most related to neuronal activity. Our approach solves these issues by maximizing the mutual information between the probability distributions of spike occurrence and input variables.We analyzed the dependence of MSTd neuronal activity in monkeys on various retinal and extra-retinal signals during presentation of a large-field visual stimulus moving randomly with quasi equally distributed frequencies (white noise). Across the population, neuronal activity depended on different combinations of retinal and extra-retinal input. The interrelation between the input variables exhibited in many cases strong non-linear characteristics. These findings support the hypothesis that MSTd uses a basis function representation for encoding various retinal and extra-retinal signals.