Sexual selection has driven the evolution of weaponry for males to fight rivals to gain access to females. Although weapons are predicted to increase males’ reproductive success, they are also expected to incur costs and may impair functional activities, including foraging. Using feeding assays, we tested whether the enlarged mandibles of Auckland tree wētā (Hemideina thoracica) impact feeding activity (the total volume of biomass consumed, bite rate, and number of foraging visits) and foraging behaviour (time spent moving, feeding, or stationary). We predicted that increased mandible length in male wētā would hinder their feeding rates. However, we found that wētā with longer heads fed at a faster rate and spent less time foraging than wētā with smaller heads, regardless of sex. Contrary to expectations that weapons impede functional activities, our results demonstrate that exaggerated traits can improve feeding performance and may offer benefits other than increased mating success.

Ethics

Ethical approval was not required from a regulatory body because invertebrate research does not necessitate a permit in New Zealand. However, we followed best practise guidelines outlined by ABS to minimise stress to the study animals whereby specimens were always handled briefly and under nocturnal conditions when wētā are naturally active. Gloves were used when handling wētā from different reserves to minimise transfer of disease and, at the conclusion of all experimental work, all specimens were released to their respective reserves as required by our collection permits.

Wētā collection and husbandry

Sixty-six Auckland tree wētā were collected from Sanctuary Mountain Maungatautari and Lake Rotopiko in the Waikato Region, New Zealand, in November and December 2019. Individuals were located by searching visually at night and transported to the laboratory. Using electronic callipers (Jobmate), we measured the head length (i.e. distance from the vertex to the tip of the left mandible) of each wētā and used it as a proxy for mandible length.  Each individual was weighed using an analytical digital balance (Sartorius BP221 S). Wētā were co-housed in 2 L plastic containers (15 x 15 x 8 cm H) with individuals from the same locality in either male/female or female/female pairs; otherwise wētā were housed individually. Each container held leaf litter and piece of PVC pipe for refuge. Plastic lids were covered with 1 ml gauze mesh to allow airflow, while moistened dental gauze provided a source of water. Wētā were fed twice weekly on a diet of fresh native plant leaves (including Melicytus ramiflorus, Kunzea ericoides, Myrsine australis, and Coprosma lucida) and supplemented occasionally with dried mealworms.

All containers were kept in a temperature-controlled room (18-20°C) on a 9:15 h dark/light cycle and regularly misted with water. Wētā were allowed to acclimatize to captivity for at least one week before experimental trials commenced. Seventeen wētā died before feeding trials began, and three wētā were excluded due to their reluctance to forage within the trial, leaving a total sample size of n = 45 (24 male, 21 female).

Foraging trials

We created foraging chambers by dividing a glass aquarium with corflute panels to form three independent arenas (each measuring 22 length x 23 width x 23 height cm). Each chamber was lined with paper towels and contained a water dish with a PVC refuge.

To ensure wētā were not satiated, a 24-hour starvation period was implemented the day before each trial. On the following morning at 0900 hrs, the subjects were transferred into a randomly allocated chamber and left undisturbed for the remainder of the day to minimise disruption before trials began.

On the evening of each trial, we used a circular hole-punch to create C. lucida leaf discs that were a standardised size (25 mm diameter; mean weight ± standard deviation = 0.5 g ± 0.03). Pilot trials determined that allocating three discs per wētā ensured enough food was available. Leaf discs were weighed before each trial (hereafter, pre-trial weight).

As wētā typically emerge within one hour of sunset, at approximately 2000 hrs we added three leaf discs to each chamber, then vacated the room and recorded wētā behaviour using video cameras (Sony FDR-AX53). One camera was set to view all three chambers simultaneously (to record wētā behaviour), while the remaining three cameras were zoomed to only view one leaf disc per chamber (to determine bite rate). We filmed in night vision mode using infra-red lighting for approximately 10 hours over-night until 0800 hrs the following day.

At the conclusion of each trial, wētā were returned to their housing containers and provided with fresh food. We removed any remaining leaf discs from the experimental chambers and re-weighed them (hereafter, post-trial weight). The apparatus was cleaned using mild detergent and left to dry before repeating trials with new wētā. Each wētā was only used once in a trial.

Data collection

We calculated the total volume of biomass that each wētā consumed by subtracting post-trial weight from pre-trial weight. We did not account for weight loss due to desiccation because it was more ecologically appropriate to use wet weight to represent the total volume of food consumed rather than the amount of carbon (i.e. using dry mass). Water within the leaf would also be essential for foraging because wētā typically obtain water from the vegetation they consume and is therefore likely directly relevant to the total volume consumed.

To calculate bite rate, we waited until feeding first commenced and then counted the number of mandibular chews in a 30 s period. Wētā feed intermittently and the 20 s period ensured that wētā were likely to feed continuously for the 30 second observation period. The total number of bites was doubled to generate bite rate per minute.

We quantified the proportion of time spent feeding, moving, sedentary and within the refuge by scoring the behaviours observed on video using Solomon Coder. Wētā were considered to be feeding if they were manipulating the leaf disc with their mandibles or chewing while within 2 cm of the leaf disc. Wētā were coded as moving if they were mobile for more than 2 s and sedentary if they were stationary for longer than 2 s. Individuals were considered to be within their refuge if more than half of their body was within the artificial refuge and stationery for longer than 5 s. We also counted the number of visits a wētā made to a leaf disc (i.e. foraging episodes). Finally, we determined the feeding rate of each wētā by dividing the total volume of biomass consumed by the duration that an individual spent feeding.

Data analysis

We fitted a series of multiple linear regression models to the data with log head length, log body mass and sex as explanatory variables in R. Bite rate was modelled with a Poisson generalized linear model. While rates are typically modelled using a Poisson distribution, the data for feeding rate was non-integer and therefore not compatible with Poisson modelling. In addition, the residual plots indicated that using a Gaussian distributed model was better suited to the feeding rate data. We estimated variance inflation factors (VIFs) using the vif function in the ‘car’ package and found minimal collinearity between predictor variables (i.e. VIF < 3). With the exception of bite rate, we transformed the response variables (log or log+1) to meet model assumptions.

To assess model fit, we constructed a range of models varying from a null model (intercept only) to a full model with 2-way interactions (head length*sex; mass*sex). The final model was selected by using the ‘AICcmodavg’ package, where we selected the model with the lowest Akaike’s information criterion (AIC) and removed models above the threshold of ΔAIC > 2. If the models tied (ΔAIC < 2), we performed maximum likelihood model averaging on a reduced set of candidate models, using the model.avg function from the ‘MuMIN’ package. We considered model averaging as an appropriate method of reducing variance while still considering the collective contributions of multiple models. As the average accounts for the strengths and weaknesses of each individual model, using model averaging was evaluated as a robust way of capturing the underlying patterns in the data. From the subset of models, we generated mean estimates for the relevant parameters, their unconditional standard errors, and the relative importance (Akaike weight sum) of each variable in the average model. Summary statistics were calculated as mean ± standard error.