Because the ability to mechanically process plant material is fundamental to all herbivory, our first aim was to understand how both the morphology and physiology of the biting-chewing apparatus of insects interact with the mechanical and structural properties of the plant to determine feeding performance. To this end, we developed experimental tools to directly quantify (i) the 3D architecture of the muscles responsible for force generation during feeding; and (ii) the maximum bite forces insects can generate at different mandible opening angles. Our experiments demonstrate that leaf-cutter ants are extraordinarily specialised to generate large bite forces; the size-specific force they generate is the highest ever recorded. This specialisation is manifested in several remarkable adaptations, such as an indirect attachment of muscle fibres to the mandible tendon in order to optimise volume occupancy of muscle in the head capsule, or changes in head capsule shape with size to accommodate a disproportionate increase in the amount of muscle. It also brings with it ecological advantages: leaf-cutter ants can forage on the fast majority of tropical plant leaves without having hefty investment into extremely large workers. Our methodological approaches and biomechanical modelling of the bite performance is general, and can be used to assess head functional morphology and bite performance across the insect tree of life.
Next, we quantified the force required to cut "pseudoleaves" with ant mandibles. Surprisingly, this force is substantially smaller than for sharp scalpel blades, and indeed we demonstrated that leaf-cutter ant mandibles are close to "ideally sharp", ie they cut some sheets with the minimally possible force. As a direct consequence, the effect of mandible size on cutting force is weak. However, because the mandibles are so sharp, they are also extremely susceptible to wear, as demonstrated by the observation that worn mandibles require twice as much effort to cut with than pristine mandibles of the same size. The effect of wear far dominates the effect of size, and is thus a main focus of follow-up work.
Because we are ultimately interested in understanding how the mechanical aspects of plant feeding may influence behaviour and evolution, we also worked on tools which enable us to analyse behaviour in various experimental conditions and with large numbers of individuals. For example, we are interested in how ant colonies “assign” workers of different sizes to forage on food sources of different mechanical properties. Traditionally, addressing such a question would involve tedious and time consuming manual labour – individual ant workers are extracted from foraging sites and weighed by hand. To overcome the limitations and potential for bias associated with this approach, we made use of recent advances in computer vision and machine learning, and taught a computer to perform these tasks for us: We designed and built a photogrammetry platform to generate photorealistic 3D models of insects, and then placed these models into various environments, using 3D graphics engines developed for the design of computer games. By generating tens of thousands of these images, and associating with them relevant information on the number, position and size of the imaged animals, we trained deep neural networks to detect, track and size-estimate ants, so enabling us to collect large amounts of data on ant foraging behaviour.
