An Electrophysiological Investigation into Selective Attention in the Dragonfly

Abstract

All animals, including humans, must contend with distracting sensory input in order to achieve behavioural success. Selective attention, stimulus tracking, and prediction are fundamental computations that drive goal-directed behaviour across species and tasks, but the study of how these computations are implemented in neuronal architecture has been largely focussed on vertebrates. However, there is mounting evidence that despite small brains, insects are capable of complex computations and exhibit behaviour and performance that surpasses even the most advanced modern robotics and artificial intelligence. Adult dragonflies (Insecta: Anisoptera) are predatory pursuit specialists that intercept prey and conspecifics (Territorial rivals, mating partners) mid-air with high success rates, by flying along interception trajectories based on predictive internal models. Additionally, adult dragonflies are immune to the ‘confusion effect,’ a reduction in predatory capture success experienced by vertebrate predators when hunting targets amidst a swarm. Previously, we have identified a system of ‘Small-Target Motion Detector’ neurons in the dragonfly (Hemicordulia sp.) optic lobe that are thought to underlie target-pursuit behaviours. In particular, one well-characterised STMD termed ‘Centrifugal Small-Target Motion Detector 1’ (CSTMD1) readily exhibits both ‘selective attention’ for a single target within a pair of rival targets and ‘predictive gain modulation’ that enhances the neuronal response to targets following a predicted trajectory. In comparison to selective attention observed in vertebrate studies, CSTMD1 exhibits absolute encoding of the selected target (rather than weighted, or relative encoding) and responds as if the distractor did not exist. Such robust stimulus representation could be critical for rapid pursuits in dynamic environments and avoid motor-control errors associated with relative stimulus representation. In order to understand how the dragonfly achieves such behavioural success, we have recorded intracellular electrophysiological spiking activity from CSTMD in vivo during the presentation of small moving targets. We show that target selection in CSTMD1 is able to ‘lock-on’ to a selected target, even when challenged by an abrupt-onset, highly salient distractor. Intriguingly, CSTMD1 is also able to dynamically switch between targets. In order to achieve such a fine balance between resistance to distraction and flexibility, we show that dragonfly attention system utilises Preattentive enhancement of targets and inhibition of return in combination to ‘gatekeep’ access to attentional competition mechanisms, providing a filter for transient and inconsistent stimuli but allowing novel, coherent stimuli likely to represent a target of interest (prey, conspecific, or predator) to capture attention. We further place the neurobiological properties of target selection in the dragonfly STMD system into behavioural and ecological context by investigating the exogenous stimulus properties that drive target selection and switching, and additionally show target selection and tracking in swarmlike conditions that resemble real-life feeding conditions encountered by dragonfly behaving in the wild. Understanding how dragonflies achieve such remarkable behavioural success with such comparatively limited computational architecture has the potential to inform the development of bioinspired computer vision, artificial intelligence, and Neurobotics platforms, such as self-driving vehicles or search-and-rescue drones, as well as illuminate fundamental mechanisms of neuronal processing and representation for stimulus selection, target-tracking, and prediction.