Behavioural Brain Research 87 (1997) 1 – 14 Review article A computational model of the response of honey bee antennal lobe circuitry to odor mixtures: overshadowing, blocking and unblocking can arise from lateral inhibition Christiane Linster a, *, Brian H. Smith b a Department of Psychology, Harard Uniersity, 33, Kirkland Street, Cambridge, MA 02138, USA b Department of Entomology, 1735 Neil Ae., Ohio State Uniersity, Columbus, OH 43210 -1220, USA Received 26 July 1996; accepted 3 December 1996 Abstract Recent studies of learning about elements of odorant mixtures in honey bees have identified several types of interactions between mixture components, such as overshadowing and blocking. The latter phenomenon in particular indicates at least a limited ability of subjects to identify the most salient element of a binary mixture. Here we show that the circuitry in the antennal lobes, the first neuropil in which synaptic interaction affects sensory processing, could give rise to both effects given the incorporation of modifiable synapses onto inhibitory circuitry. The neural model of the antennal lobe that we present incorporates identified cell types and includes a biologically realistic modulatory neuron with which modifiable Hebb-like synaptic interactions take place. A learning rule that incorporates modifiable connections from output (projection) neurons onto the modulatory neuron is sufficient to account for behavioral results on generalization and overshadowing. A second type of excitatory connection from the modulatory neuron onto local inhibitory interneurons is necessary to reproduce behavioral results from blocking and unblocking. We suggest that the neural representations of odor mixtures in the antennal lobe can be modified by previous exposure to one of the mixture components. These results provide testable hypotheses that will guide future behavioral and physiological analyses. © 1997 Elsevier Science B.V. Keywords: Olfaction; Mixtures; Honey bee; Antennal lobe; Associative learning; Overshadowing; Blocking; Computational model 1. Introduction Odor mixtures can present peculiar problems to an animal, particularly in regard to feature segmentation, because the exact composition of a mixture of odorants and/or the background in which it is presented can change across short spatial and temporal scales. For example, a foraging honey bee, having just received reinforcement at a particular flower, must make a deci- sion whether or not the odor of the next flower signals similar reinforcement even though the mixture may vary considerably from flower to flower [30]. Feature segmentation, that is, perception of the elements of the mixture, is made difficult by the tuning characteristics of sensory cells in peripheral olfactory systems [41]. Sensory cells for non-pheromonal odorants are typi- cally broadly tuned to respond to many different odor- ants, as are cells involved in detecting floral odorants in bees [13]. Given the highly nonlinear interactions that arise as a result of the transduction pathways [1,6,43], it Abbreiations: LN, local inhibitory interneuron; O1 and O2 refer to two odorants in a mixture; PN, projection (or output) neuron; R pure and R mix , refer to the responses of the VUM neuron to pure odors and mixtures; VUM, refers to the representation in the model of the ventral unpaired medial cell of maxillary neuromere 1 (i.e. VUMmx1 [14]). * Corresponding author. Tel.: +1 617 4962555; fax: +1 617 4953738; e-mail: linster@berg.harvard.edu. 0166-4328/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved.