The piriform cortex in the mammalian brain is the largest cortical region that receives direct sensory input from the olfactory bulb as well as complex top-down inputs from higher brain regions. It is located in the ventral part of the brain, right behind the olfactory bulb and above the eyeball. While this heterogeneous brain region is composed of many subregions, the overall roles assigned to the piriform cortex are the integration of elementary odor information and the association of this representation with other sensory and task variables (such as spatial location or context). Most frequently, the piriform cortex is seen as encoding odor identity in distributed ensembles of neurons. Whether and how this encoding evolves during learning and experience is not well characterized or understood.
There is some evidence that the similarity of neural representations (the particular patterns or combinations of neurons activated by a given stimulus) of individual pure odorants could be shaped by passive experience and that active learning of how to tell a small number of mixtures apart could lead to changes in these representations. Yet, how mice and the piriform cortex deal with a large number of mixtures with a huge diversity of compositional similarity is not known.
Can mice distinguish a particular mixture, such as lemon, from numerous other mixtures? Will the neural representation of lemon be modified as mice learn, and if so, how? When mice are forced to place all the other mixtures into a non-lemon category, will their representation also become grouped such that individual identity becomes fuzzier? When a new smell that is sort of like lemon (say lime) is introduced, will mice classify it as lemon or as “other”? Will the neural representation of lime start off being similar to lemon, and become very different when mice are taught to categorize it as “other”?
We designed a new task where mice are trained to distinguish a specific odor mixture (“target”) from hundreds of other non-target mixtures. We then recorded spiking activity in the posterior piriform cortex (pPC) in behaving mice to investigate how the invariant target mixture and the highly varying non-target mixtures are represented and how the representation changes as mice continue learning. Once mice became experts at this task, we introduced ambiguous mixtures, which had components from both the target and non-target mixtures, to see how mice classified them.
When we tested the mice with these mixtures, our findings included three main ideas:
(1) With learning, the target odor mixture becomes over-represented, with more pPC neurons selective for the target odor mixture than the other odors repeated in the task.
(2) With continued “over-training” in expert mice, the pPC representation continues to evolve, and becomes more robust and selective, even without any obvious benefit in behavior.
(3) However, this enhanced selectivity benefits behavior during the later perturbation with more ambiguous “probe” mixtures.