Neuron, doi:10.1016/j.neuron.2013.08.026

Credit: ELSEVIER

Most natural odors are mixtures that can be perceived as wholes (for instance, 'grapefruit') or in categories with degrees of refinement ('citrusy'). Humans can usually distinguish up to three components of a mixture, but insects and rodents frequently do better. Because of odorant mixing, the process of olfaction involves several object recognition problems. These include discrimination and generalization (shared features define a new group). Shen et al. used locusts to study how odors are represented and computed by different parts of its olfactory system. The authors tested the responses to various mixtures of two monomolecular odors by 168 antennal lobe projection neurons (PNs) and the responses to mixtures of up to eight diverse monomolecular odors by 174 PNs and 209 mushroom body Kenyon cells (KCs, the direct targets of PNs and the site for associative memory) in 61 animals. Although PNs tended to respond strongly to many of the odors tested, KC responses were much sparser and showed higher odor specificity, with many KCs signaling the presence of single odor components in mixture. The responses of both populations could be decoded over short time windows to perform odor identification, categorization and generalization. The collective results suggest that dense PN odor representations are reformatted by the KCs into sparse ones to better effect discrimination and generalization further downstream.