Optimizing multivariate behavioural syndrome models in locusts using automated video tracking

Locusts exhibit a behavioural syndrome known as ‘behavioural phase polyphenism’, in which a number of behavioural traits change markedly in response to local population density. ‘Solitarious’ phase individuals, which are typical of low-density populations, change within hours from being relatively sedentary and repelled by other locusts to congregating actively with conspecifics and becoming more active (the ‘gregarious’ phase). In wild populations, this behavioural plasticity can lead to the emergence of mass marching bands of nymphs and winged adult swarms. Much of our understanding of behavioural phase transition comes from laboratory experiments, which routinely employ an arena-based assay to measure a suite of behavioural variables encompassing aspects of activity, movement pattern and responses towards a stimulus group of other locusts. Individuals are then quantitatively phenotyped along a linear scale from solitarious to gregarious, by entering their returned measurements for several behavioural characters into a logistic regression model. Recently, automated video tracking has enabled multiple experimenters to use a single behavioural model, rather than each having to construct their own. Here, we have taken advantage of another powerful feature of automated tracking systems: the opportunity to use stored data to conduct a rigorous optimization process, which both ensures that the derived statistical model encapsulates the multidimensional nature of locust behavioural phase to best effect, and also provides a new understanding of the relationship between different behaviours.