Variation in sexual signals and defensive strategies elicits receiver-dependent shifts in attractiveness

Sexual selection often favors the evolution of conspicuous mating displays. Emitting such overt displays carries the risk of interception by eavesdropping enemies, i.e., predators, parasitoids, and parasites that exploit communication systems to find and attack their signaling victims. Yet, many signalers respond to variation in perceived eavesdropper risk, protecting themselves through risk-dependent inducible defenses to mitigate potential costs. Given that signalers are embedded in communication networks in which they interact with other signalers, target receivers, and multiple eavesdropping enemies, here we investigate how variation in signaling and defensive strategies impacted by an eavesdropping enemy (frog-biting midges; Diptera: Corethrellidae) affects other receivers in a communication network. Ultimately, we aim to determine if and to what extent effects that cascade throughout the network shape relative fitness among chorusing males. Using female choice experiments with túngara frogs (Engystomops pustulosus) and predation experiments with eavesdropping, fringe-lipped bats (Trachops cirrhosus), we show that variation in the call elaboration and defensive strategies of competing males shape the relative fitness of males. Defensive strategies targeting eavesdropping frog-biting midges indirectly shift a male’s relative attractiveness to females and predatory bats, though the mechanisms and impacts are context- and receiver-specific. These findings showcase how the frequency-dependent effects of micropredation can dynamically shape variation in secondary sexual characteristics and thus influence the mechanisms driving sexual selection.

Methods

Study site

This research was performed at the Smithsonian Tropical Research Institute (Gamboa, Panamá; 9°07.0’N, 79°41.9’W). Frogs and bats were collected and tested within 2 km of STRI facilities.

Experimental stimuli

To represent the variation in calling behaviors from swatting frogs we created ripple and airborne treatment playback stimuli based on previously recorded samples from our study population (described in (Leavell et al. 2022)). The range of calls used captures natural variation in male calling behavior under diverse micropredator densities and intensities of male-male competition. Following (Ryan and Rand 2003), we z-transformed the values for call rates, swat rates, and total chucks (all recorded over 50 continuous calls, per male). We then used Euclidian distances to select the samples that represented the range of trait space: the sample closest to the mean (i.e., x = 0, y = 0, z = 0; where x = call rate, y = swat rate, z = total chucks), eight samples at approximately ± 1 SD, and eight samples at extreme values (Figures S1-2). The experimental design was tailored to increase statistical power given the distinct ecological contexts in which bats and female frogs make decisions. Given that frog-eating bats consume about half their body weight per night (~15 grams) and readily participate in many foraging trials per night, they can be tested multiple times over the course of a night (Hemingway et al. 2020). Thus, all 17 playback files were used in the bat foraging experiment (43.0 ± 13.9 trials/bat; mean ± s.d.; n = 5 bats). Conversely, female túngara frogs typically respond to fewer mate choice trials before losing their motivation to perform phonotaxis (5.2 ± 2.6 trials/female; mean ± s.d.; n = 32 females). To maximize the range of male traits played back to females, while minimizing the experimental trials presented to them, we therefore focused on the eight extreme samples in the female choice experiment.

Each experiment was a two-choice preference test in response to simultaneous playback of airborne calls with call-induced ripples and swat-induced ripples. We constructed 1-minute audio files in Adobe Audition that captured the call rate and average complexity for these samples, as well as audio files that captured the rate of ripple production based on the sample’s swat rate. For samples that featured many ornamental “chucks”, we appended either one or two chucks to calls in a randomized order to reach the appropriate number of chucks for our 1-minute playback file. Audio files were looped for experiments. In a trial, a treatment playback from the multivariate dataset (consisting of airborne calls with call-induced ripples and swat-induced ripples) was played concurrent with an average call from the population (i.e., 0.5 calls/second and 1 chuck/call (Ryan 1985), with accompanying call ripples, but no swat ripples). Swat audio files were always positioned between two calls. For 1 swat, the middle of the swat audio file was positioned at the middle point between the end of the 1st call and the start of the 2nd. For 2 swats, the middle of the 1st swat was positioned at the middle point, as was the start of the 2nd swat. For 3 swats, the end of the 1st swat, the middle of the 2nd swat, and the start of the 3rd swat occurred at the middle point of the calls.

Female choice

In August 2022, we collected mated (i.e. in amplexus with a male) female túngara frogs after sundown, between 19:30 – 01:00 hrs. This research was conducted in accordance with a Smithsonian Tropical Research Institute institutional animal care and use protocol SI-22042 and Panamanian legal and ethical regulations (Ministerio de Ambiente permit ARG-097-2022). Following protocols established by the American Society of Ichthyologists and Herpetologists (https://asih.org/animal-care-guidelines), we toe-clipped all frogs to prevent pseudoreplication and released each frog with its mate on the same night and at the individual’s point of capture. We conducted phonotaxis experiments in a dark, acoustically-isolated chamber (2.8 m x 2.8 m x 2.8 m; 25.0 – 25.7 degrees Celcius). Inside the chamber was a wood-framed, plastic-lined pool (2.23 m x 0.60 m x 0.07 m) filled with dechlorinated tap water to a depth of 1.5 cm. To produce airborne playback, two small speakers (Fostex FE103En) in custom built cabinets were positioned ~ 1 m across from and facing each other. All speakers were amplified through Pyle PTA4 amplifiers. The airborne playback speakers sat slightly above the water, supported by 2 cm PVC “tee” fittings that rested in the water. The mid-point between the speakers, which was also the mid-point of the chamber’s opposing walls, was the center of the arena and the female’s entry point at the beginning of each trial.

To produce ripples, we drove custom-built loudspeakers (as described in (Halfwerk et al. 2016) placed outside the arena. A vinyl tube was fitted to the opening of each loudspeaker on one end and ran along a PVC frame (97 cm x 81 cm x 33 cm) to which it was secured. At the other end of the tube was a nozzle that rested adjacent to its associated airborne speaker and perpendicular to the water surface such that a small meniscus formed around the nozzle. Following the methods in (Leavell et al. 2023), we calibrated call and swat ripple playbacks before and after experiments using the digital output of a digital laser vibrometer (LDV; Polytec PDV-100; Velocity = 20 mm/s, Low Pass = 22 kHz, High Pass = none), focused on a reflective marker floating on the water surface. Calibrations were recorded using a Marantz audio recorder (PMD661 MKII; 48 kHz sample rate, 24 bit). All calibration recordings were subsequently processed using custom code to derive velocity and frequency measurements. Though there was some variation across playbacks, the vast majority fell within the natural range of call and swat ripple measurements described in Leavell et al. 2023 (see supplementary R script).

To observe female behavior, we illuminated the room with infrared lights (Sima Model SL-100IR) and recorded video with a Sony Handycam (FDR-AX33) in night-shot mode. At the start of a trial, we broadcast the simultaneous playback of male stimuli. Females were allowed to emerge from the end of a PVC tube at the center of the arena. We considered a choice was made when the female paused within 10 cm of a speaker (front or back) or moved without stopping within one body-length of the nozzle. We interpreted female preference as a proxy for relative male fitness. We also measured the time it took from emergence into the arena until a female made a choice, which we refer to as “latency to choose”. Before and after testing a female, we ensured that the maximum intensity of the whine call component was set to 82 dB sound pressure level (SPL; peak, C-weighted, fast; re. 20 µPa) at the female’s point of release. This intensity matches natural call intensities at 1.0 m from a calling male. SPL was measured with a Brüel & Kjær 2238 Mediator Sound Level Meter (peak, C weighting, fast).

Bat foraging

From January to March 2019, we captured and tested wild-caught Trachops cirrhosus within Soberanía National Park, Panamá. To ensure no individual was re-used in experiments, each bat was marked with a passive integrated transponder (PIT tag, 12 mm, ~0.1 g and ~0.3% of body mass; Biomark, Boise, ID, U.S.A.). Following testing, all bats were released at their point of capture. This research was conducted in accordance with a Smithsonian Tropical Research Institute institutional animal care and use protocol 2017-0102-2020-A8 and Panamanian legal and ethical regulations (Ministerio de Ambiente permit SE/AP-13-18).

Calibration of the playback system occurred as in the female choice experiments. Given that these bats hunt from above, speakers were placed ~1 m apart, facing up, over the center of neighboring pools of water in an outdoor flight cage (5×5×2.5 m). Water depth was maintained at ~ 3.0 cm. To present the multimodal stimuli as if they were coming from a single male, the ripple speaker and vinyl tube setup used in the female choice experiment was arranged such that the nozzle was nearly under the speaker. Following previous studies with this bat species (Halfwerk et al. 2014), we first trained bats (n = 5) to hunt at two small feeding platforms (i.e. acoustically transparent mesh atop speakers’ diaphragms). As in the experiments with female frogs, speakers were positioned 1 m from each other over the pools of water. For training, we used a mix of the playback stimuli as well as recordings of the mating calls of other local frog species. Bats learned to begin foraging from a corner roost positioned equidistant (4 m) from each speaker. Once bats were trained, we began the experiment. We presented a food reward (fish) on both speakers at a 50% reward rate (randomized across trials, but with no more than four rewarded trials sequentially). To control for potential use of echolocation or chemosensory cues to discriminate between rewarded versus non-rewarded speakers, we used size- and shape-matched stones covered with fish “juice” when a speaker was not rewarded.

Statistical analysis

All statistical analyses were performed in R version 4.0.3 (R Core Team 2020). For our analyses, we formulated Bayesian regression models using the brms package version 2.18.0 (Bürkner 2017). For female and bat preference experiments, we fit mixed logistic regression models of choice between treatment and control playbacks, with call rate, total chucks and swat rate as z-transformed fixed effects and the individual receiver as a random intercept. We then examined the individual and interactive effects of call rate, swat rate and total chucks on mate choice in female túngara frogs and prey choice in predatory bats. For further inference of female behavior, we fit generalized linear mixed models of female latency to choose a male. We compared models with lognormal, shifted lognormal, Poisson and negative binomial error structures. All latency models had the same fixed and random effects as previously described. Models were compared by assessing expected log pointwise predictive densities with the loo package version 2.4.1 (Vehtari et al. 2017).

For all models we used weakly informative priors to ensure that draws from the prior could be from any hypothetical dataset (Gabry et al. 2019). In this way, these priors include probability mass around potential, but not implausible extreme values. We ran 4 chains with 3000 iterations each and discarded the first 500, resulting in 10,000 posterior samples. To ensure model convergence, we visually examined traceplots for good mixing of chains and confirmed that < 1.1 for all parameters. We performed posterior predictive checks to assess model fit with the bayesplot package version 1.8.1 (Gabry and Mahr 2021). Strength of evidence for effects was determined following the framework established in (Muff et al. 2021) and (Kruschke and Liddell 2018), using posterior probability distributions degree of overlap with values of interest. See supplementary materials for details specific to each model.