347 — Single neurons may encode simultaneous stimuli by switching between activity patterns

Caruso et al (10.1038/s41467-018-05121-8)

Read on 02 August 2018
#neuroscience  #signal-processing  #audio  #3D  #stimuli  #multiplexing  #neurons  #monkey  #primate  #inferior-colliculus  #localization 

In most neuroscience research, we make the assumption that a neuron acts a certain way in the absence of a stimulus, and then in the presence of a stimulus, it acts in a different way. This is of course a simplified model, but it helps us make sense of the complex ways in which stimuli and neural response are connected.

But we know well that neurons can have complex responses to single stimuli, or to multiple stimuli. Even the visual system exhibits resting behavior that would make a one-stimulus-one-response purist doubt himself (#48, #241, #128). What is the mechanism that enables the brain to handle multiple stimuli simultaneously? One possible mechanism is that the brain delegates different pure stimuli to entirely different neurons. But that would require a vast number of very finely tuned neurons — a model we know to be false. The alternative is that individual neurons can respond to more than one stimulus within a very short timeframe. And that is what this paper explains.

Armed with single-unit recordings from the inferior colliculus of a monkey (a region associated with audition and sound localization in 3D), the researchers compared the spiking response of individual neurons during sound localization of one single sound and two simultaneous sounds.

When listening to two simultaneous sounds, the monkey was still able to locate the origins of each sound in 3D. When given sound $A$, a neuron might exhibit response $R_A$; and in response to sound $B$, that same neuron might exhibit response $R_B$. When both $A$ and $B$ are played, the neuron must in some way capture both stimuli: How does it do this? One possibility is that it averages the responses $R_A$ and $R_B$; but instead, the authors found that the neuron fluctuated between $R_A$ and $R_B$ states, performing the spike train associated with $A$ for a little bit and then switching to that for $B$ (and then repeating). This is interesting because it means that individual neurons can have different outputs from the same inputs varying only with time: It also begs the question of how neurons are able to synchronize this multiplexing across a population.

This is also a great illustration of how neurons are fundamentally different from computational neural networks: Neural networks are deterministic and each “neuron” returns the same response given the same inputs, whereas here, neurons’ responses are time- as well as input-dependent.