| All animals must use patterns of olfactory receptor neuron (ORN) activity to compute appropriate behavioral responses to odors, but the form of this computation is unknown. In both flies and mammals, each ORN expresses one type of odorant receptor that determines its odor response profile, and all ORNs expressing the same receptor project to the same compartment in the brain, or glomerulus. These glomeruli constitute parallel processing channels that relay olfactory information to other areas, and are often co-activated by odors. One popular hypothesis holds that activity in each glomerulus contributes to a central representation of valence, or pleasantness, by a fixed linear weight. Alternatively, non-linear interactions between specific glomeruli could confer increased selectivity or sensitivity for ethologically important odors. We investigate these alternative models by optogenetically activating olfactory glomeruli in freely walking Drosophila, and use video tracking to analyze the behavioral responses of thousands of individual flies. Flies respond to optogenetic fictive odors, but require wind to orient their walking responses. Some glomeruli produce robust attraction when activated individually. Combining stimulation of pairs of glomeruli produces unpredictable results: some pairs sum to produce attraction greater than that elicited by either component, but others produce the same behavior as the more attractive component alone. Surprisingly, we find no reliably repulsive glomeruli, but some potently reduce attraction in combinations. Based on which glomeruli summate, we develop a simple model that establishes a lower bound on the dimensionality of internal olfactory representations. Although we provide evidence that the brain combines signals from glomeruli using several pools with overlapping but distinct sets of inputs, detailed computational analysis of walking trajectories suggests that flies respond to these stimuli by graded recruitment of a single behavioral program. |
