Projects

One of the most fascinating behaviours recently discovered in ants is wound care, in which injured ants are cared for by their nestmates. This behaviour has been shown to increase the survival of injured ants. While it has been described in several species of ants, the factors that regulate an individual’s decision to provide wound care were unknown. Using the FORT system, we looked at how individual and social factors interact to regulate wound care in Camponotus fellah. We combined detailed behavioural observations with high-resolution spatial data to show that caregivers are ants transitioning between the nurse and forager roles in the colony. Intriguingly, their decision to provide care depends on their social and spatial affinity with the injured ants before the injury occurs.

In this project we looked at how the amount of brood affects the behaviour and social network structure of workers in an ant colony. We performed experiments with Camponotus floridanus and compared colonies with different amounts of brood.

We are still analysing the data we collected, so check back later for updates.

Classical models of division of labour in social insects assume that workers differ in their response thresholds to task-related stimuli, and that these differences are sufficient to generate division of labour in the colony. Recent empirical work shows that response thresholds alone cannot explain the observed patterns of division of labour, and that other response parameters, like response probability and response intensity, also play an important role. I have developed a framework that formalises these three response parameters and their hierarchical relationship, and am using it to explore how these parameters interact to regulate division of labour in social insect colonies.

I am now working on simulations to explore how these parameters can interact to generate different patterns of division of labour.

Biologging is rapidly growing as a viable technology for studying animal behaviour in the wild, but analysing the data collected from these devices is still a bottleneck in many studies. Most biological studies use standard machine learning models, like random forests, to infer behaviour from accelerometer data, but these models are not designed to account for the temporal structure of the data and so require a lot of manual feature engineering. In this project, we use deep learning models to infer behaviour from accelerometer data in a more automated way, and compare their performance to standard machine learning models. We are particularly focused on foundational tabular models (TabPFN) and state-of-the-art time series models (ROCKETs), and on how key decisions during the pre-processing of these large datasets influence model performance on a per-behaviour basis.

In this project we are looking at how male courtship in Drosophila melanogaster changes with infection status. We use SLEAP for pose estimation of several individual flies from video data, followed by manually extracting features from the pose and then training a hierarchical deep learning tabular classifier to quantify behaviour from these videos.

We have developed an end-to-end pipeline that quantifies male courtship with high accuracy and are now analysing the data we collected. Check back later for updates.

Vervet monkeys are strongly matrilineal, and the social relationships individuals form early in life strongly influence their fitness as adults, but we still don’t know how these relationships form in the first place. In this project we used multi-layer network analysis to compare the ontogeny of social relationships across different behavioural contexts (e.g. grooming, proximity) in a population of wild vervet monkeys in South Africa. We find that as juveniles age, the overlap between their social interaction and proximity networks increases, whereas both networks become less similar to their kinship networks. This change is driven by juveniles’ increasing independence from their mothers.

In a project spearheaded by Patrick Kohl from the University of Würzburg, we looked at whether honey bee populations show dance dialects (differences in the slope between waggle dance duration and the distance of the food source indicated). We found that the species differ in this slope, and that it is inversely correlated with their foraging range (A. cerana has the steepest slope and the shortest foraging range, and A. dorsata the shallowest slope and the largest foraging range in Bangalore). Dance dialects therefore represent an adaptation to food availability in the environment.

The signals produced by foragers in the waggle dance vary among the extant honey bee species (A. florea produces a visual but not an acoustic signal, A. cerana does the reverse, while A. dorsata produces both). However, nothing was known about how the receivers of these signals, the dance followers, respond to these differences. We found that dance follower behaviour is conserved across all three species throughout the waggle dance, and is similar to follower behaviour in A. mellifera. This indicates that the mechanism by which spatial information is transmitted in the waggle dance is likely to be conserved across the genus Apis.

The effect of optic flow on waggle dance behaviour had previously been explored in A. mellifera, using foragers trained in an artificial tunnel. My collaborators and I looked at the effect of natural variation in the optic flow of foraging environments on the waggle dance activity of A. florea and A. cerana. We found that, as in A. mellifera, higher optic flow led to a more rapid increase in waggle phase duration with distance, but only in A. florea and not in A. cerana.

In a second paper, we looked at the dance calibration curves of different A. cerana lineages, and performed a common garden experiment in which we transported colonies from the Himalayas to Bangalore to see how the change in environment affects the dance calibration curve.

A. dorsata is unique in that it not only forages at night but also performs waggle dances at night, providing insights into how bees navigate and communicate in low light. While several studies have looked at nocturnal dances in A. dorsata, they have all focused on colony-level responses. My collaborators and I looked at how individual foragers adjust their dances during the night, in the absence of the visual cues available during the day. Most foragers already adjust their dance to the expected position of the sun at sunrise, and we identified different strategies individual foragers use to perform this update.

In this project, we looked at how recruitment activity to the same food source varies within a group of foragers. We found that foragers showed strong differences, consistent over days, in their probability of dancing and the intensity of their dances for the same food source. When we removed some foragers from the group, only the more active individuals changed their total recruitment activity, and they did so by increasing their probability of dancing, not its intensity. Since removing foragers potentially changes social interactions in the hive, the probability of dancing is associated more with the ‘social’ perception of the food reward than dance intensity is.

To explore the mechanisms underlying inter-individual variation in recruitment activity in honey bees, we looked at sucrose response thresholds and whole-brain transcript levels of specific genes in individual foragers. Sucrose responsiveness is linked to the ‘behavioural state’ an individual is in; dance intensity showed a small positive correlation with sucrose responsiveness, but dance probability did not. We also found that the foraging gene showed a strong negative correlation with the total recruitment activity of individual foragers, adding a new dimension to the role this gene plays in food-related motor behaviours in insects.

One exciting, but largely unexplored, area of research is the role that consistent individual variation plays within social insect groups. Building on an excellent agent-based model by Schürch and Grüter, we introduced variation in the probability and intensity of dancing among the agents, capturing the individual variation in total recruitment seen in our empirical data. We then compared the food influx of colonies with and without inter-agent variation to determine how this variation affects colony fitness under different environmental conditions.