Encoding a temporally structured stimulus with a temporally structured neural representation Stacey L Brown, Joby Joseph & Mark Stopfer Sensory neural systems use spatiotemporal coding mechanisms to represent stimuli. These time-varying response patterns sometimes outlast the stimulus. Can the temporal structure of a stimulus interfere with, or even disrupt, the spatiotemporal structure of the neural representation? We investigated this potential confound in the locust olfactory system. When odors were presented in trains of nearly overlapping pulses, responses of first-order interneurons (projection neurons) changed reliably, and often markedly, with pulse position as responses to one pulse interfered with subsequent responses. However, using the responses of an ensemble of projection neurons, we could accurately classify the odorants as well as characterize the temporal properties of the stimulus. Further, we found that second-order follower neurons showed firing patterns consistent with the information in the projection-neuron ensemble. Thus, ensemble-based spatiotemporal coding could disambiguate complex and potentially confounding temporally structured sensory stimuli and thereby provide an invariant response to a stimulus presented in various ways. Visual 1–4 , auditory 5–7 , tactile 8–10 , and olfactory 11–15 senses all use temporal coding mechanisms 16 for which the timing, rather than just the rate, of action potentials is important for the neural representation of the stimulus 17,18 . If a neural response pattern significantly outlasts the stimulus, new stimuli might arrive before the response to a previous stimulus is complete 19 . This situation is particularly interesting in olfaction; in principal neurons, odor pulses can elicit response patterns that endure beyond the pulse’s offset 11,14 and, further, natural odor plumes can have repeated, rapid and nearly overlapping encounters with olfactory receptors 20 . Given these potential confounds, how can neural systems use spatiotemporal representations to encode and decode temporally structured stimuli? In the locust, projection neurons in the antennal lobe—the analogs of vertebrate mitral cells 21 —respond to odors with complex spiking patterns that consist of epochs of excitation, inhibition and quiescence. These patterns vary with odor identity and concentration, and can greatly outlast the odor receptor neurons’ encounter with the odor 11,14 . The antennal lobe’s network of excitatory projection neurons and inhibitory local neurons generates these firing patterns when driven by input from olfactory receptor neurons 22,23 . These spiking patterns, distributed broadly across the projection neuron population, are parsed into brief segments by odor-evoked oscillations and are read by downstream follower neurons (the Kenyon cells) which receive con- vergent input from many projection neurons 15,24 (R.A. Jortner and G. Laurent, Soc. Neurosci. Abstr. 412, 21, 2004). The oscillating projec- tion neuron ensemble, through feed-forward inhibition, generates brief integration windows in the Kenyon cells 24 . Kenyon cells respond to odors with very sparse spiking, often demonstrating great specificity with respect to odors and even particular concentrations of odors 25,26 . Here we examine how the locust olfactory system responds to very short (100-ms) repeated odor pulses; the timing was chosen to approx- imate that observed in natural odor plumes 20 . In a turbulent environ- ment, odor is carried in an intermittent fashion in the form of filaments of odor-laden air that vary in concentration, duration and frequency with which they appear. These factors are modulated by wind speed, amount of turbulence, delivery mechanism and distance from the odor source 20,27 . We selected stimulus parameters on the basis of odor plume measurements made outdoors, in which filaments were observed to encounter a sensor in a series of bursts that averaged about 100 ms in duration and arrived, on average, at approximately 500-ms intervals 20 . We made intracellular recordings from projection neurons and extracellular tetrode recordings from projection neurons and Kenyon cells in adult locusts. We delivered brief (100-ms) odor pulses in trains of 3 or 10 pulses; inter-pulse intervals ranged from 500 to 1,250 ms but were constant within a train. For each pulse pattern, we delivered blocks of 10 trials (15–30 s inter-trial interval), with the blocks given in random order. We used a variety of odorants and concentrations (see Methods). Our results showed that ensemble-based coding mechanisms can disambiguate complex and potentially confounding temporally struc- tured sensory stimuli, thus providing an invariant response to a stimulus presented in different ways while preserving information about the stimulus timing. RESULTS Projection neuron responses vary with odor pulse pattern For most odor–projection neuron combinations, the number of spikes elicited by odor pulses changed reliably, and sometimes substantially, with pulse position, because lengthy responses to one pulse interfered Received 15 August; accepted 19 September; published online 16 October 2005; doi:10.1038/nn1559 National Institute of Child Health and Human Development, US National Institutes of Health, Building 35, Room 3A-102, Bethesda, Maryland 20892, USA. Correspondence should be addressed to M.S. (stopferm@mail.nih.gov). 1568 VOLUME 8 [ NUMBER 11 [ NOVEMBER 2005 NATURE NEUROSCIENCE ARTICLES © 2005 Nature Publishing Group http://www.nature.com/natureneuroscience