Honeybee cognition: From numbers to extraction of regularities

Insects are not mere reflex machines. Instead, they adapt their behaviour flexibly to changing environmental contingencies. Among the insects, honeybees (Apis mellifera) possess an impressive repertoire of cognitive abilities, despite their limited number of neurons. Thanks to the standardization of behavioral, neurobiological, neuroimaging, and genetic methods, bees became a widely used invertebrate model in research. Importantly, the study of their capacities allows us to integrate evidence from an invertebrate species into broader scientific frameworks - often based on vertebrate studies - supporting a deeper understanding of the evolution of certain cognitive mechanisms and their universality. Honeybees can process different information from their environment, such as the numerousness of an array or the relationships – both perceptual and abstract – between objects. Once identified, such relationships allow bees to form distinct categories to which they will refer to implement adaptive choices. An ongoing debate focused on whether numerical abilities in bees are supported by a unified neural mechanism – as for vertebrates - or if multiple segregated mechanisms are involved. Additionally, there is interest in further expanding our knowledge about the extent of bees’ categorization capacities in different contexts. This thesis aims to address these questions, providing evidence that can shed light on the neural organization and limits of honeybees’ cognitive abilities, as well as on potential similarities or differences with other species. In the first two studies, the existence of a general mechanism for the estimation of quantity in honeybees was investigated. Specifically, I addressed the issue of whether bees’ numerical abilities are supported by a general magnitude mechanism that estimates continuous (e.g., space, time, size) and discrete (i.e., number) quantities. In the first study, we investigated the bees' ability to transfer learning from numerical to size dimension. Using appetitive-aversive conditioning, independent groups of free-flying foragers were trained to discriminate between larger and smaller visual numerousness (i.e., 2 vs. 4, 2 vs. 3, 4 vs. 8, 4 vs. 6; 0.5 or 0.67 ratio difference). We then tested the bee's generalization ability with a comparison between stimuli with different sizes and identical numerosity (e.g., 4 larger elements vs. 4 smaller elements). Honeybees spontaneously chose the congruent size with respect to their training. No effect of numerical contrast and ratio difference experienced was found as bees previously reinforced toward the larger numerosity, chose the relatively larger size, and vice versa. These results demonstrated the ability of this insect species to make a transfer from the numerical to the size dimension. Given the possibility of asymmetric relationships between magnitudes, we sought to explore whether honeybees possess the capacity to make the reverse transfer as well, from a continuous (size) to a discrete (number) dimension. Similar to the previous study, free-flying foragers were trained to discriminate between relatively larger vs. smaller squares or diamonds. Their generalization ability over novel shapes (i.e., circles) and novel dimensions (i.e., number) was subsequently tested. Our results confirmed the ability of bees to transfer size discrimination to novel shapes. Moreover, when presented with a 4 vs. 8 elements comparison, bees spontaneously selected the congruent numerosity with respect to their training (i.e., bees trained to select the smaller/larger size, selected the smaller/larger numerosity, respectively). To check for any perceptual cue involvement in bees’ decision-making, different continuous variables covarying with numerosity were controlled for (i.e., total area, contour length, stimulus size, convex hull). Subsequent analyses also revealed no role of spatial frequency in the bees’ behavior. The results revealed a bee’s capacity to transfer between numerical and size dimensions, suggesting the universality of the magnitudes coding mechanism and highlighting the presence of a unified circuit supporting discrete and continuous quantity processing. The second aim of this thesis was to enlarge our knowledge of the ability of bees to spontaneously encode regularities from the physical world. To this purpose, I tested bees' ability to extrapolate the structure of temporally defined odor sequences. In a series of six experiments, the spontaneous and trained ability of bee foragers to learn, memorize, and generalize an odor sequence composed of three distinct odors was tested. A proboscis extension response (PER) conditioning paradigm was employed (i.e., absolute, differential, and generalization). The first two experiments investigated honeybees’ ability to learn an arbitrary odor sequence. Bees were trained to respond to a specific sequence of three odors and then tested for their spontaneous ability to generalize their response to novel sequences with a similar structure but composed of novel odors and to reject novel configurations although composed of familiar odors. The role of a particular odor position in the sequence, the odor-reward temporal closeness, and their possible effects on memory were also investigated in the third experiment. The fourth and fifth experiments aimed to understand the effect of differential conditioning on bees’ learning ability. Lastly, we determined whether a conditioning procedure favouring a generalization strategy could lead to the spontaneous encoding of the internal sequence structure. In general, the results highlighted an early tendency of bees to encode the single odor properties, instead of learning the entire sequence structure, together with a significantly increased response towards the novel odor configurations composed of familiar odors. No effect of the odor’s position or temporal closeness with the reward was apparent. During absolute and differential conditioning, bees likely employed two strategies to memorize the dyad of the first and second elements of the sequence, together with a more general response to novelty. However, the use of a transfer paradigm potentially revealed a weak spontaneous generalization over similar structures one hour after the training, irrespective of the single-element properties. Overall, these results shed light on the strategies employed by bees to solve an odor abstraction task, highlighting the crucial role of the type of conditioning to let them emerge. Altogether, the thesis provides new evidence on honeybees’ cognition. The findings have implications not only for the study of bees’ behavior but also for broader investigations into the universal development of basic cognitive mechanisms and the convergent evolution of similar abilities in small and large brains.