Abstract
Data from a variety of experimental preparations reveal that when a multisensory neuron is unable to integrate a pair of cross-modal inputs, it "defaults" to a response less than or equal to its largest unisensory component response -- not a sum of those responses (i.e., the additive prediction). Furthermore, when the neuron does integrate its cross-modal inputs, its response often exceeds the sum of its component responses. Additive predictions prove to be consistently biased in predicting a multisensory response in several ways: (a) response begins earlier than predicted, (b) early portions are stronger than predicted, and (c) later portions are weaker than predicted. Thus, additive models fail to capture the dynamic features of multisensory integration and underrate its ability to enhance incoming information. Here we illustrate a simple, biologically-inspired model based on input current summation rather than the more familiar response summation. Unisensory responses for a given neuron are transformed into input currents which are summed and combined with an inhibitory factor and then transformed back into a predicted response. Using a minimal set of parameters (i.e., membrane time constant, noise, and an inhibitory scaling factor), this model accurately matches the entire temporal profile of the multisensory response for nearly all neurons. Even when all model parameters are fixed, the models predictive power far exceeds that of the additive model. This is the first model using the actual unisensory responses as inputs that can accurately predict the temporal evolution of actual multisensory responses. Supported by NIH grants EY016716 and NS036916
Meeting abstract presented at VSS 2014