Odor Representations in Olfactory Cortex: Distributed Rate Coding and Decorrelated Population Activity

From AcaWiki
Jump to: navigation, search

Citation: Keiji Miura, Zachary F. Mainen, Naoshige Uchida (2012/06/21) Odor Representations in Olfactory Cortex: Distributed Rate Coding and Decorrelated Population Activity. Neuron (RSS)
DOI (original publisher): 10.1016/j.neuron.2012.04.021
Semantic Scholar (metadata): 10.1016/j.neuron.2012.04.021
Sci-Hub (fulltext): 10.1016/j.neuron.2012.04.021
Internet Archive Scholar (search for fulltext): Odor Representations in Olfactory Cortex: Distributed Rate Coding and Decorrelated Population Activity
Download: http://www.sciencedirect.com/science/article/pii/S0896627312003893
Tagged: Neuroscience (RSS) olfaction (RSS), coding (RSS)


This paper addresses the question of how neuronal spikes in the olfactory cortex guide sensory decisions in the rat. It starts from the observation that a single sniff generally suffices for precise odor discrimination, yet the mechanism involved and the sources of noise in behavioral responses are not fully understood. As the coding of odor information in the olfactory cortex is less well-known than that in the olfactory bulb, the authors measure the output of ensembles of 5-15 neurons in the olfactory cortex of 8 trained rats, and analyze the statistics of their responses to different odors.

Goals and Methods

Rodents often engage in active sampling, taking rapid sniffs while exploring. This suggests that each sniff suffices to give a snapshot of odors. This experiment aims to understand better what happens to encode sensory information over the timescale of a single sniff, and how it is transformed on the way to the brain to influence behavior.

Neural ensemble activity is measured in the anterior piriform cortex [aPC] of rats performing odor categorization, and principal components of the activity are identified. These cortex ensembles show very low noise correlation, suggesting that it has a very well-defined reprsentation of odor identity. An optimal theoretical linear decoder tested against these inputs could discriminate odors as well as mice are observed to, with only 100 neurons. However the rat aPC has 10,000 times as many; possible implications of this variance are discussed as motivation for further study.

Six odors were tested and an ensemble of neurons in the aPC recorded. Both latency and peak timing of aPC responses were measured, as were spike counts.

A number of theoretical decoders were used to map the observed latencies and spike counts to odor classification.

Results and Analysis

Each sniff was seen to trigger a transient spike linked closely to the onset of inhalation. Odor stimulation stimulated a broad ensemble of neurons, with fairly little directional selectivity. 45% of all neurons were stimulated by one of the odors, but each odor triggered only 16%; so the neurons were fairly selective.

The theoretical decoder that worked significantly better than all others relied on spike counts, and was hardly improved with the addition of the other data. So this seems to be the most likely candidate of those tested to related to actual encoding odor information.

Information provided by spike counts could account for the rapid discrimination of odors. On the other hand, this decoder worked much more effectively than the rodents in practice, so some other part of the whole system is missing from the model. Choice probability analysis suggests individual aPC neurons are very weakly correlated if at all. This was true regardless of the distance between neurons or the similarity of their odor tuning.

Further study is needed to understand a few points:

  • Why might the olfactory bulb and cortex use different odor coding strategies? The codex is much larger and could afford to use a more widely-distributed code, which could also enable memory creation.
  • How does olfactory information switch from temporal coding to rate coding?
  • What limits behavioral accuracy, which seems to be lower than a good linear decoder might do with only 100 neurons?
  • Can second and further sniffs not significantly improve decoding? This remains controversial. Rate information seems to peak 100ms from the onset of the first sniff. But in this experiment little improvement in decoding was observed with multiple sniffs.

Theoretical and Practical Relevance

This paper suggests that odors generate characteristic transient bursts immediately after a sniff, and that the identity of an odor is more closely related to properties of those bursts such as the burst spike count than to spikes over an entire sniff cycle or to their timing. It also suggests the olfactory bulb and cortex code odors differently, and have a higher theoretical accuracy in odor discrimination than has been observed in rat behavior.