New insights into how the olfactory system represents odours — at least in locusts — come from work described by Perez-Orive et al. in Science. They find that representations become increasingly sparse and specific as they are processed by the early stages of the olfactory system, and that correlations in firing are crucial for this effect.

In insects, odours are detected by receptor neurons that project to the antennal lobe. Here, projection neurons (PNs) carry the signal on to the mushroom body, which has been implicated in olfactory learning in insects. In the mushroom body, PNs form synapses on Kenyon cells (KCs). Each PN contacts hundreds of KCs, and each KC receives inputs from many fewer PNs (an estimated 10–20).

The authors recorded the responses of PNs and KCs to various odours. They found that individual PNs responded strongly to a variety of odours, and that their responses had complex temporal patterns, often outlasting the stimulus, that were odour- and neuron-specific. Odour-evoked PN responses also showed global oscillations at around 20–30 Hz. KC responses, by contrast, were much more selective: each KC responded to very few odours, and their responses were brief (on average, just two action potentials).

What gives rise to the marked sparsening of odour representations between the PNs and the KCs? Recordings from KCs revealed odour-evoked subthreshold potentials with odour-specific timing. Much of this activity was inhibitory, and Perez-Orive et al. conclude that the likely source of this inhibition is a group of lateral horn interneurons that contain GABA (γ-aminobutyric acid). The lateral horn receives inputs from PNs and projects directly to the mushroom body. Lateral horn interneurons fire vigorously in response to odours, but their firing is in the opposite half of the PN oscillation cycle from the excitatory PN inputs to the KCs.

Because lateral horn interneurons fire in response to a wide range of odours, it is likely that KCs receive reliable inhibitory inputs from these interneurons during one half of each oscillation cycle, and more selective excitatory inputs from PNs during the other half. The authors propose that KCs act as coincidence detectors, firing only when excited by multiple PNs. Inhibitory potentials antagonize mistimed PN action potentials, so that only time-locked PN spikes at the right point in the oscillation cycle can excite a KC. Direct stimulation of PNs and recording from KCs suggested that KC responses are further sharpened by active conductances.

These results highlight the potential for a combination of oscillatory synchronization and both neuronal and circuit properties to shape and filter neuronal information. Such mechanisms are proposed to be important in other systems, and it is becoming clear that they must be taken into account when interpreting recordings of neuronal activity.