Learning and memory mediate interactions between plants and their pollinators, as animals associate various features of floral displays with rewards received in exchange for the transfer of pollen. Most of what we know about the cognitive ecology of pollination involves research on generalist bees (Apis, Bombus) that are amenable to learning experiments in lab settings. Such controlled conditions are required to demonstrate that learning has occurred, by collecting repeated measures of individual behavior. Two common approaches involve free-flying behavioral assays (requiring beekeeping and generating low sample sizes) or conditioning of the proboscis extension response (PER, requiring specialized equipment and training). Due to these constraints, the majority of research on bee cognition concerns only three commercial species (Apis mellifera, Bombus impatiens and B. terrestris), with little-to-no extant data on individual-, population- or species-level variation in cognitive traits. We describe a field-based learning assay that permits non-lethal assessment of floral preferences, learning speed, and memory performance. We refined this novel technique (Free-Moving PER), and tested it in a field study of individual, population, and species-level variation in bumble bee learning performance at sites in the Sierra Nevada mountains. This is thought to represent the first dataset on associative learning in wild bumble bees.
Within days, the new assay (FMPER) generated sample sizes in field-caught bees that would take weeks to achieve using traditional methods. Bumble bees (B. impatiens) could be readily trained to respond to a rewarded stimulus (e.g. yellow) and to discriminate between two color stimuli (e.g. yellow rewarded, blue unrewarded). Bees retained these associations in long-term memory, and, consistent with learning theory, showed higher performance when the cost of an error was increased (i.e. when the unrewarded color was paired with quinine rather than water). When we compared learning performance among male and female B. vosnesenskii and B. bifarius, we found no sex differences, although species-level differences existed. Likewise, B. vosnesenskii tested across sites showed little variation in performance. We describe how this technique might be expanded to study population- and foraging-associated cognitive differences, and applied in situ to understand whether bees can discriminate among floral traits of interest (e.g. among colors of co-flowering plant species). We highlight FMPER’s potential to bring insights from the substantial body of literature on pollinator learning and memory into a field setting, with the goal of estimating ecologically-relevant measures of cognitive performance and their potential consequences for floral phenotypes.