Bridging the gap between theories of sensory cue integration and the physiology of multisensory neurons

Integration of multiple sensory cues is crucial for adaptive behaviour, but understanding its neural basis has been hampered by a divergence in the literature into two largely non-overlapping conceptual approaches. One approach seeks to predict and quantify cue integration behaviour in a rigorous psychophysical framework, whereas the other has focused on the empirical principles that govern multisensory integration by neurons, including its anatomical and developmental origins.Tasked with reconciling these camps are the computational theorists who strive to create models of multisensory processing that are both biologically realistic and capable of explaining psychophysical performance. Progress has been made on this front, an example of which is the theory of probabilistic population coding. However, several obstacles remain, such as accounting for key differences between model predictions and neurophysiological data and a lack of neuronal recordings from behaving animals performing psychophysical cue integration tasks.A recent set of studies on the integration of visual and vestibular cues for self-motion perception draws from both sets of approaches and has attempted to narrow the conceptual gap between them. This work uses macaque monkeys trained to report their heading direction in a virtual-reality apparatus, while researchers monitor single-unit activity in a multisensory region of the extrastriate visual cortex (the dorsal medial superior temporal area (MSTd)).The emerging picture is one in which computations performed by multisensory neurons in area MSTd can be linked to psychophysical performance and may also help to explain the empirical principles that have driven the field for two decades. Divisive normalization, a common property of neuronal circuits, may be one of the key computations that give rise to the desired behaviour.Further work will be needed to develop more complete models that connect the neurobiological details of multisensory integration with the probabilistic computations underlying cue integration performance and perception in general.