Adaptive coding for auditory spatial cues

Adaptation of sensory neurons to the prevailing environment is thought to underlie improved coding of relevant stimulus features. Here, we report that neurons in the inferior colliculus (IC) of anesthetized guinea pigs adapt to statistical distributions of interaural time differences (ITDs - one of the binaural cues for sound-source localization). Each distribution contained all of the ITDs in the range of ±330 µs, but with a different high probability range (HPR) of ITDs in each from which ITDs were chosen 80% of the time. The center (mean) or the width (variance) of the HPR was altered systematically to assess the sensitivity of each neuron to the underlying distribution of ITDs. Neural rate-vs.-ITD functions shifted to accommodate the change in the underlying distribution when the mean of the HPR, but not the variance, was altered. Assuming a population of similarly tuned binaural neurons with Poisson spike statistics inputting to an IC neuron, we used an established model of synaptic depression to compute the IC postsynaptic current. By including a stochastic leaky integrate-and-fire mechanism in the IC spike generator, the model predicts the changes in neural gain and the shapes of rate-ITD functions as well as the shifts in Fisher information, a measure of coding accuracy, observed in the physiological data.