Predator stimulus, habitat structure, and body size jointly shape antipredator strategy in Neotropical anurans

GENERAL INFORMATION 1. Paper Citation Authors (2026). Predator stimulus, habitat structure, and body size jointly shape antipredator strategy in Neotropical anurans: a Bayesian phylogenetic analysis. Behavioral Ecology [DOI: to be added on acceptance]. 2. Brief abstract Predation is a multi-stage process in which prey can break the chain of capture at different points by deploying distinct defenses, from pre-detection crypsis to post-contact active flight. We tested whether antipredator strategy in anurans — a clade combining a deeply conserved jumping body plan with striking microhabitat diversification — has co-diversified with morphology across the radiation, by exposing 89 individuals from 17 Neotropical species (seven families, six microhabitat categories) to a fully factorial combination of three arenas of contrasting structural complexity and two predator stimuli (visual approach and tactile contact). Trial-level data on immobility versus flight and conditional jump distance were analysed with a Bayesian phylogenetic hurdle log-normal model that jointly estimates both response components with phylogenetic random effects on each. Predator touch reduced the probability of immobility from 94% to 15% while increasing conditional jump distance by ~26%. Habitat complexity increased immobility and decreased conditional distance. Larger body size strongly increased immobility probability with only a weak effect on conditional distance. Both response components exhibited substantial phylogenetic signal (λ_μ ≈ 0.80; λ_hu ≈ 0.71), suggesting co-diversification of antipredator behaviour with locomotor morphology across the anuran microhabitat radiation. 3. Originators ==to be fulfilled upon acceptance== 4. Contact information ==to be fulfilled upon acceptance== [Corresponding author name] [Institution] [Full institutional address] Email: [email] ORCID: [orcid id] 5. Date of data collection Behavioural trials were conducted within two weeks of the animals being captured. Individuals were collected from the field between 30 November 2013 and 19 December 2015. 6. Geographic location(s) of data collection ==to be fulfilled upon acceptance== 7. Information about funding sources that supported the collection and curation of the data ==to be fulfilled upon acceptance== ACCESS INFORMATION 1. Licenses/restrictions placed on the data CC-BY 4.0 Attribution. 2. Data derived from other sources The species-level time-calibrated phylogeny (filogenia_jessyca_newNames.tre) was pruned from the comprehensive frog phylogeny published by Portik et al. (2023). When citing this data archive, please also cite the original phylogeny source: Portik, D. M., Streicher, J. W., & Wiens, J. J. (2023). Frog phylogeny: a time-calibrated, species-level tree based on hundreds of loci and 5,242 species. Molecular Phylogenetics and Evolution, 188, 107907. DATA FILES AND VARIABLES 1. data.csv Raw trial-level dataset. Comma-separated values; 534 rows (one row per behavioural trial) × 7 columns. Each row corresponds to a single antipredator trial of one individual in one arena × stimulus condition. 89 individuals were tested across the 17 species. Each individual was tested under multiple combinations of arena × stimulus (~6 trials per individual on average). Especie: species name in Genus_species format with underscore (e.g., Boana_faber). Tip labels in the phylogeny file match this format exactly. 17 species total. individuo: unique individual identifier (alphanumeric). 89 individuals total. tipo_arena: arena type. Three levels: bush (structurally complex bush vegetation arena), leaf_litter (semi-structured leaf-litter arena), empty (open arena without cover). tipo_de_estimulo: predator stimulus type. Two levels: approach (slow visual approach by a snake stimulus simulating pre-contact predation stages) and touch (tactile contact by the snake stimulus simulating physical capture attempt). SVL: snout–vent length of the individual, in millimetres. Continuous variable measured prior to trials with digital calipers (precision 0.01 mm). resposta_antipred: antipredator response. Two levels: immobile (individual remained motionless throughout the trial) or jump (individual escaped via a saltatorial jump). distancia_saltada: jump distance, in centimetres. Continuous non-negative variable. Coded as 0 when resposta_antipred = immobile (i.e., the zero indicates absence of escape jump, not missing data). Measured as the straight-line distance from the individual's starting position to its landing position. 2. filogenia_jessyca_newNames.tre Time-calibrated ultrametric species-level phylogeny in Newick format. 17 tips, with tip labels matching the Especiecolumn of data.csv exactly. Branch lengths are in units of millions of years. The tree was pruned from the comprehensive species-level frog tree of Portik et al. (2023) using ape::drop.tip(). Four terminals (listed in the manuscript Methods) were placed by congener substitution where direct molecular sampling was unavailable. Format: Newick (.tre) Tips: 17 Type: ultrametric, time-calibrated Branch length units: millions of years Loaded with: ape::read.tree("filogenia_jessyca_newNames.tre") CODE SCRIPTS AND WORKFLOW Two Quarto (.qmd) files reproduce all analyses and figures reported in the manuscript. The recommended workflow is to run antipredator_v2.qmd first (which fits and saves the M1 model) and then fig1_production.qmd (which loads the saved model to produce the empirical phylomorphospace figure). 1. antipredator_v2.qmd Main reproducible analysis pipeline (~1,750 lines). Reads data.csv and filogenia_jessyca_newNames.tre, performs the full analysis reported in the manuscript, and writes m1_full.rds to the working directory. Sections: Data import, cleaning, and exploratory analysis Phylogeny pruning and construction of the phylogenetic covariance matrix Model M1 specification: bf(distance ~ arena * stimulus + log_svl_c + (1|gr(species, cov=A)) + (1|ind_id), hu ~ arena * stimulus + log_svl_c + (1|gr(species, cov=A)) + (1|ind_id), family = hurdle_lognormal()) Model M2 (alternative additive specification) for model comparison MCMC sampling via cmdstanr backend (4 chains × 6,000 iterations, warmup 2,000, thin 2) Convergence diagnostics: trace plots, density overlays, Rhat, ESS, posterior-predictive checks Prior sensitivity analysis via priorsense Model comparison via leave-one-out cross-validation (loo) with Pareto-k diagnostics Marginal effects, variance partitioning, Pagel's λ estimation Phylomorphospace and clade-level random-effect plots Effect size analysis: probability of direction (pd) and region of practical equivalence (ROPE) via bayestestR Session information Expected runtime: 15–25 minutes on a MacBook Pro M1 (mostly MCMC sampling). After the first render, chunk-level caching makes subsequent renders nearly instantaneous. 2. fig1_production.qmd Standalone script that produces the empirical phylomorphospace figure presented in the Results section. Reads m1_full.rdsand filogenia_jessyca_newNames.tre, extracts the posterior coordinates of each species on both response components, and composes a ggplot figure with clade-coloured points, 95% bivariate credible regions (via stat_ellipse), species labels (via ggrepel), and a mini-phylogeny inset (via ggtree and patchwork). Expected runtime: ~30 seconds. 3. antipredator_v2.html Pre-rendered HTML output of antipredator_v2.qmd, included so reviewers and readers can inspect all results, figures, and diagnostics without installing the software stack required for re-running the analyses. 4. m1_full.rds The fitted M1 brms model, serialized as RDS (~26 MB). Loaded with m1 <- readRDS("m1_full.rds"). Included to allow rapid reproduction of post-hoc analyses, figures, and tables without re-running the ~15–25 minute MCMC sampling. SOFTWARE VERSIONS All analyses were conducted on macOS Tahoe 26.4.1 with R 4.5.x and CmdStan 2.35.x. The full sessionInfo() output is included at the end of antipredator_v2.html. R version 4.5.x loaded packages: brms: 2.21.x cmdstanr: 0.X.X posterior: 1.6.x tidybayes: 3.0.x bayesplot: 1.11.x loo: 2.7.x priorsense: 1.0.x bayestestR: 0.13.x tidyverse: 2.0.x patchwork: 1.2.x ggrepel: 0.9.x cowplot: 1.1.x ape: 5.8 phytools: 2.3.x ggtree (Bioconductor): 3.14.x CmdStan 2.35.x Installed via cmdstanr::install_cmdstan(). C++ compilation handled by Apple Clang on macOS. Quarto 1.4.x Required to render the .qmd files. Install from quarto.org. REFERENCES Bürkner, P.-C. (2017). brms: An R package for Bayesian multilevel models using Stan. Journal of Statistical Software, 80(1), 1–28. https://doi.org/10.18637/jss.v080.i01 Kruschke, J. K. (2021). Bayesian Analysis Reporting Guidelines. Nature Human Behaviour, 5, 1282–1291. https://doi.org/10.1038/s41562-021-01177-7 Portik, D. M., Streicher, J. W., & Wiens, J. J. (2023). Frog phylogeny: a time-calibrated, species-level tree based on hundreds of loci and 5,242 species. Molecular Phylogenetics and Evolution, 188, 107907. https://doi.org/10.1016/j.ympev.2023.107907 Whitlock, M. C. (2011). Data archiving in ecology and evolution: best practices. Trends in Ecology and Evolution, 26(2), 61–65. https://doi.org/10.1016/j.tree.2010.11.006