Bees acquire route-based memories but not cognitive maps in a familiar landscape

Recent experiments suggested that forager honey bees, Apis mellifera, can store 'cognitive maps' of a landscape, i.e. internal representations that allow them to take novel short cuts while flying between sites that they had visited separately from the nest. Replication of these experiments revealed that the behaviour predicted by the map hypothesis is better explained by the hypothesis that bees use landmarks associated with specific routes travelled previously. As flexible as the bees' orientation is, there remains no evidence that the underlying spatial memory encodes the kind of geometrical information learned by humans and other vertebrates. A recent study (Gould 1986) concluded that honey bees, Apis mellifera, can navigate using 'landmark-based mental maps' (Gould & Towne 1987), representations in memory of the geometrical relationships among different locations in a familiar landscape. This finding would be extremely important if true, because it would imply the existence of vertebrate-like internal representations (e.g. Menzel 1973, 1978; Gallistel 1989) in an insect. However, the original study failed to exclude alternative explanations for the behavioural result that was taken as evidence for maps (Dyer & Seeley 1989), and the result itself has proved hard to verify (Menzel 1989; Menzel et al. 1990; Wehner & Menzel 1990). I have therefore replicated the original experiments to investigate whether the orientation predicted by the map hypothesis could be observed when bees are tested in a different landscape from that used by Gould, and, if so, whether the underlying memory is a cognitive map or a less abstract representation of the landscape. In his original experiments, Gould (1986) captured bees leaving their hive for a familiar feeding site A and displaced them to an alternative site B (e.g. Fig. 1). On release at site B, most of the bees headed straight for site A, rather than homeward or in the compass direction they were about to fly on leaving the hive. Gould concluded that a bee could set the course to the food by using a mental map that registered the relative positions of site A and site B, as learned during previous flights. The underlying supposition is that a bee could compute and fly the path to site A even if it had never before flown that route and could not directly see landmarks learned on other routes previously flown to site A. Hence, taken at face value, Gould's results suggest that bees store in memory both the visual appearance of a given familiar site in the landscape and geometrical information that they can use to determine their position relative to any other familiar site (Cartwright & Collett 1987; Gallistel 1989). There are at least two simple explanations for Gould's results that do not involve map-like internal representations (Dyer & Seeley 1989). First, bees at a release site might have aimed towards the food using the same panorama of major landmarks (e.g. pastures, woodlots, hills) used previously when travelling from the hive. This would mean flying at a new angle relative to these cues, but such flexibility is already known for bees finding a familiar feeding site amidst small landmarks close to the food (Cartwright & Collett 1983). Second, perhaps the route from the release site to the food was not novel to the bees. Both explanations are consistent with the suggestion that insects learn about landscapes as if storing sequences of 'snapshots' of landmarks seen along familiar routes (Wehner 1981). Neither explanation requires that geometrical relationships among sites or among routes be represented in memory along with the snapshots. To investigate these hypotheses, I repeated Gould's (1986) basic experiment in a landscape in 0003-3472/91/020239 + 08 $03.00/0 9 1991 The Association for the Study of Animal Behaviour