Dendritic low pass filtering shapes midbrain neural responses to behaviorally relevant stimuli

Neurons are equipped with powerful analog processing units – dendrites. However, the role linear and nonlinear dendritic properties play in shaping a neuron’s output to filter behaviorally relevant information and subsequently drive behavior, is poorly understood. Here, we address this question using wide-field neurons of the mouse superior colliculus; a genetically targetable midbrain cell-type that receives direct input from the retina and mediates innate defensive and hunting behaviors. To understand how wide-field neurons combine their inputs, we measured their output: the responses to behaviorally relevant visual stimuli in dendrites and cell-bodies, as well as their inputs: from the retina and collicular inhibitory interneurons, using a combination of viral transsynaptic circuit tracing, electrophysiology and two-photon calcium imaging. We found that wide-field neuron cell-bodies, known to preferably respond to stimuli mimicking approaching and cruising predators, also show strong responses to a “receding prey stimulus” and cluster by contrast preference into three types: On, Off and On-Off. Spatially confined signals measured in wide-field neuron dendrites in response to large stationary stimuli, revealed diverse signals clustering in distinct collicular layers. Corresponding to these diverse signals, we identified ten types of retinal ganglion cells providing ten parallel information streams to wide-field neurons. To determine which combination of inputs and dendritic filtering properties are necessary to reproduce wide-field neuron responses, we tested an array of linear and nonlinear models. This led to three discoveries: First, linear mapping of retinal and local inputs onto wide-field neuron dendrites revealed a layered input organization, but could not accurately reconstruct cell-body responses. Second, contributions of nonlinear dendritic summation were minor. Third, the accurate reconstruction of cell-body responses to approaching, receding and dimming stimuli mandates strong low pass filtering along the extensive dendritic arbor. The precise input-output mapping allowed to model the dendritic properties necessary to reproduce physiological neural responses.