Despite their small brains, many insects form long-term memories of the spatial distribution of resources. To support this, some species display ‘orientation’ flights to increase capture of landscape cues around novel resources. The role of orientation behaviour in spatial learning has been broadly explored in Hymenoptera, which are well known to navigate between their nests and floral resources. Here, we describe exaggerated flight behaviours expressed in the foraging context in Heliconiini butterflies, resembling orientation behaviours in Hymenoptera. Heliconiini butterflies are of particular interest as they include the genus Heliconius, which is reported to have a greater capacity for spatial memory in association with a novel dietary strategy of pollen feeding, including formation of stable foraging routes. We hypothesised that these flight behaviours we observe may provide a strategy to memorize the location of floral resources. We compared Heliconius and non-Heliconius butterflies to evaluate this ability in pollen and non-pollen-feeding Heliconiini species. We characterized three behavioural patterns that are directed towards a new floral resource: circle flights, hovering, and feeding bouts. Although all Heliconiini studied displayed these patterns, Heliconius spent more time performing these flight patterns, in particular hovering before landing. We suggest that these behaviours could provide the context in which individuals acquired sensory information to guide future foraging events, and our data indicate greater emphasis on this process during memory formation. Our findings of behaviours reminiscent of orientation flights in Heliconiini butterflies provide new avenues of research on how spatial learning and memory have converged between Lepidoptera and Hymenoptera.

Behavioural experiments were performed in an experimental cage with a group of 6 butterflies of a single species each time. We used two species of Heliconius, H. erato and H. melpomene, and, for comparison purposes, two closely related species that do not feed on pollen, Dryas iulia and Dryadula phaetusa. Behavioural video footage was analysed using BORIS software (Behavioural Observation Research Interactive Software, Friard and Gamba 2016).

Experiments were performed in an experimental cage (3 m x 3 m x 2 m, Figure S1) with a group of 6 butterflies of a single species each time. Individuals were marked and sexed to allow individual identification. The food resource available in the cage consisted of three pots of flowering plants: Stachytarpheta mutabilis and/or Pentas lanceolata, sparsely positioned in the cage (Figure S1). These plant species are well known to attract butterflies, are regularly used to feed Heliconiini butterflies, and provide both nectar and pollen (Estrada and Jiggins, 2002). The butterflies had two days to become habituated in the new cage and could freely feed from the flowers.

After the cage habituation period, a new floral resource was placed in the centre of the cage each morning (~ 9:00 AM), when all species are motivated to feed (Dell’Aglio et al., 2024). The experimental floral resource consisted of inflorescences of Lantana camara freshly collected and put in a 50ml Falcon tube with water. Two action cameras (GoPro©) were positioned to film approaches to the flowers from the side and from below (Figure S1). To capture the first interactions between the butterflies and the resource, we recorded the first 40 minutes after this resource was introduced. Subsequently, the cameras and the experimental floral resource were removed from the cage. This procedure was repeated over three consecutive days.

Behavioural video footage was analysed using BORIS software (Behavioural Observation Research Interactive Software, Friard and Gamba 2016). The behaviours recorded and described in the manuscript were defined during previous observations in the wild. Using the videos recorded from the side of the floral resource, we recorded: i) time until first feeding (time to find the new resource), ii) duration and number of feeding attempts, and iii) duration and number of ‘hovering’ and ‘circular flights’ behaviours around the flowers (see Table 1 for full description). To calculate the average duration for which a behaviour was expressed, we used the total duration of the behaviour (seconds) divided by the number of occurrences of the behaviour. To calculate the number of hovering flights per feeding attempt, we used the total number of hovering occurrences divided by the total number of feeding occurrences. Feeding bouts were annotated using tabular events data from BORIS, in which periods of consecutive feeding occurrences (the maximum time between feeding events to be included in the same bout was stipulated to be 3 minutes) were grouped per individual. Using the videos captured from below the floral resource, we recorded the diameter (maximum observable distance from the individual take off) and duration of the circular flights. We note that the field of view of the camera was less than the total diameter of the cage, and as a result, the diameters of the largest circular flights could not be quantified, which may impact our later analyses of how this trait changes with time.

To explore the importance of floral odours to these behaviours, we added a final trial for each species with naïve butterflies, repeating the same methodology, but using an artificial flower (orange foam sheet with a 1 ml Eppendorf tube for solution attached in the centre) instead of Lantana flowers. In this experiment, all real flowers were removed from the cage and only the new artificial flower was positioned in the centre of the cage. On the 1^st and 2^nd day, the artificial flower contained a sugar-water solution (20% sugar and 80% water), and on the 3^rd day, a clean, previously unused artificial flower was empty. During this trial, we recorded the same behaviours as listed above. Naïve butterflies are not strongly attracted to artificial flowers, and typically must be trained to use them. Therefore, we expected feeding attempts from this experiment to be low in number, as we focused on innate behaviours of untrained individuals. As such, quantitative data from this experiment are limited. However, it allowed us to assess whether visual cues associated with the feeder would stimulate flight behaviours around the potential food source, or if olfactory cues were essential to initiate them.

We used linear mixed-effect models (GLMM) implemented with the ‘lme4’ package in R (Bates et al., 2015), transforming data to a Log-normal distribution to fit the normality assumption. We included day and species as fixed factors, and individuals and group (butterflies tested on the same day) as random factors. Significance of fixed factors was determined through analysis of deviance test (Type II Wald Χ² test) and followed by Tukey's post hoc tests in R (R Core Team, 2020). Count data was analysed using generalized linear model (GLM) with the Poisson family, followed by analysis of deviance tests. To test for differences between pollen-feeding and non-pollen-feeding Heliconiini, some analyses were calculated grouping the species in “Heliconius” (H. erato and H. melpomene) and “Non-Heliconius” (D. iulia and D. phaetusa). 28 individuals who did not interact with the flowers were removed from the data set. Therefore, results are based on a final dataset of 14 H. erato (8 females, 6 males), 19 H. melpomene (6 females, 13 males), 15 D. iulia (6 females, 9 males) and 15 D. phaetusa (5 females, 10 males), all of which interacted with the flowers on all three days of the experiment (a total of 63 out of 91 individuals). None of the behaviours observed were influenced by sex (all p > 0.05), so this factor was removed from the analysis.