Infanticide is widespread across the animal kingdom, but the physiological drivers of infanticide versus care or neglect are relatively unexplored. Here, we identified salient environmental and physiological antecedents of infanticide in the mimic poison frog (Ranitomeya imitator), a biparental amphibian in which female parents feed their tadpoles unfertilized eggs. Specifically, we explored potential environmental cues influencing infant-directed behavior by evaluating changes in the frequency of food provisioning and tadpole mortality after either cross-fostering tadpoles between family units or displacing tadpoles within the terraria of their parents. We found that changes in offspring location reduce care and increase infanticide. Specifically, parents fed their displaced offspring less and, in some instances, tadpole mortality increased. We also investigated whether care and infanticide were related to changes in steroid hormone concentrations in an unfamiliar setting. Infanticide of fertilized eggs and hatchlings in the new territory included cannibalism and was associated with lower testosterone concentrations, but not with changes in corticosterone. Overall, our results support earlier findings that familiarity with offspring location drives parental investment in poison frogs, while indicating an association between low androgen levels and infanticidal behavior in an amphibian.

Animal husbandry

All R. imitator eggs and tadpoles in this study were captive-bred in our colony. Adults were purchased from Indoor Ecosystems between 2019-2021 (Whitehouse, Ohio) or Ruffing’s Ranitomeya between 2021-2023 (Tiffin, Ohio, USA). Breeding pairs were housed in 30 x 30 x 45 cm terraria containing sphagnum moss substrate, driftwood, live plants, egg deposition sites, and film canisters filled with water for tadpole deposition. Terraria were automatically misted ten times daily, and frogs were fed live Drosophila fruit flies dusted with Repashy Calcium Plus (Oceanside, CA, USA) three times per week and springtails once a week. All procedures in this study were approved by the Stanford University Animal Care and Use Committee (Protocol #34242).

Offspring recognition assays

Tadpoles used in this study were observed in home or foster terraria as described (all with the same dimensions and substrates). Ranitomeya imitator pairs used for this study were required to have previously successfully raised a tadpole within 31 days of the experiment. Locations of tadpole deposition were marked on the outside of the terrarium with the date and time to ensure consistency in placing tadpole canisters in the same location throughout the duration of the study. Tadpoles were randomly placed into three experimental conditions (Fig. 1A): same-morph cross-foster (n = 16 across 6 terraria), different-morph cross-foster (n = 12 across 5 terraria), or intra-tank displacement (n = 18 across 7 terraria). Control tadpoles (n = 26 across 11 terraria) served as a measure of parental investment under normal conditions.

We observed tadpoles in each trial for 14 days and measured trophic egg deposits as a proxy for parental investment. We randomized treatments across tadpoles, using individuals below Gosner stage 32 (Gosner, 1960) to avoid counting feeding patterns that resulted from changes in investments as tadpoles neared metamorphosis. We collected data for every tadpole under controlled conditions before randomly applying treatments. Sampling multiple (one to four) tadpoles per terrarium allowed us to capture focal pair variation in parental care.

Tadpoles were always transported in their respective canisters to retain any tadpole-specific olfactory cues already present in the water and to minimize stress. Tadpoles displaced within their parents’ terraria (“intra-tank displacement”) were moved at a minimum of 5 cm and a maximum of 30 cm, depending on availability on the terrarium floor. We purposely selected a minimum distance greater than 2 cm, as this was the minimum distance for indirect offspring discrimination identified in other poison frogs (Stynoski, 2009). Tadpoles were moved either to a laterally opposite corner of the terrarium or elsewhere along the terrarium edge. When undergoing this manipulation, tadpoles were always moved to a location not previously occupied by a canister.

"Takeover” behavioral assays

We designed an assay to identify hormonal changes in infanticidal versus non-infanticidal parents during a simulated territory takeover based on Ringler et al. (2017). We selected paired adults that raised at least one offspring in their terrarium within 31 days of the takeover trial as a proxy for parental experience and reproductive status. Frogs from the same reproductive pair were put into separate takeover trials at the same time to mitigate any risk of mate absence influencing behavior directed toward offspring. The takeover terrarium had the same dimensions, with elements of the environment (vegetation, replacement of dead leaves and substrate, fresh water) disrupted and replaced to introduce unfamiliarity. Rearrangements and cleaning occurred between trials. Takeover terraria always housed at least two tadpole canisters, one upright and filled with water and the second empty and sideways, to remove bias for water availability and shelter, respectively. At the beginning of each trial, frogs were provided with a dish of springtails available ad libitum for the length of the trial and were provided with fruit flies on the same schedule as their home tanks.

While baseline hormone collection of the adult occurred, we placed a single, unrelated fertilized late-stage egg to hatchling (Gosner stage 19-22) on a leaf on the terrarium floor. The leaf was positioned below a Wyze Cam v3 to record behavior using previously described methods (Goolsby et al., 2023). Frogs (n = 12 females, 12 males) were introduced to the terrarium and recorded for a maximum of seven days. Infanticidal behavior was characterized as either consumption of the fertilized egg or hatchling (Supplementary Video 1) or as repetitive physical disturbances to the egg, where the snout repeatedly dug at the jelly-like casing in what we interpreted as an attempt at egg cannibalism impeded by the encasement (Supplementary Video 2).

Steroid hormone collection and processing

We collected hormones immediately before displacing frogs to the new tank (“baseline”), 24 ± 3 hours later (“move”), and following infanticide or tadpole transport, or after seven days if no such behavior occurred (“final”), with hormone collections occurring at approximately the same time of day or immediately following infanticide (Fig. 2A-B). Three subjects demonstrated infanticide on the day after displacement, so hormone measurements counted for both “move” and “final” sampling events. Baseline collection always occurred between 11:30 AM and 2:30 PM to avoid confounding effects of circadian physiology on hormone concentrations. We ultimately analyzed data from 59 samples across 22 individuals.

As shown in previous studies with poison frogs and other amphibians (Gabor et al., 2013; Baugh et al., 2018; Baugh and Gray-Gailliard, 2021; Rodríguez et al., 2022; Love et al., 2023), water-borne testosterone and corticosterone were collected as a non-invasive measurement that reflects circulating levels of both hormones. Frogs were individually moved to a Petri dish containing 40 mL of distilled water treated with reverse osmosis conditioner to prevent osmotic stress (Josh’s Frogs RO R/x, Osowo, MI, USA). After 60 minutes, we measured body length (snout-vent length) using a digital caliper (Shahe Measuring Tools, Amazon) and body mass using a Maxus precision pocket scale (sensitivity 200 x 0.01 g, Amazon). We also measured mass at each hormone sampling event, then averaged to account for variations related to body fluctuations.

We pre-extracted steroid hormones using Sep-Pak C18 cartridges (Waters, Milford, MA, USA). The cartridges were conditioned using 2 mL of 100% ethanol, followed by 2 mL of Milli-Q water treated with reverse osmosis conditioner. Water samples were pushed through the column at a rate of approx. 10mL/minute, and then eluted with 4 mL of 100% ethanol. Ethanolic extracts were stored in glass vials at 4 ℃ until nitrogen evaporation. One day prior to hormone quantification, the 4 mL of eluted hormone samples were divided into 2 mL new glass tubes for separate corticosterone and testosterone analyses from the same frog. The 2 mL tubes were placed in a 37 ℃ water bath and evaporated with gentle N2 gas flow. After evaporation, hormone samples were resuspended in 250 μl of assay buffer specific to the corresponding ENZO kit and stored at 4 ℃ overnight until analysis the following day.

Steroid hormone quantification via enzyme-linked immunosorbent assays (ELISA)

After overnight incubation at 4 ℃, samples were warmed to room temperature, and corticosterone and testosterone concentrations were determined using commercially available ELISA kits (ENZO Life Sciences, Farmingdale, NY; Corticosterone: cat. no. ADI-900-097, antibody: donkey anti-sheep IgG, sensitivity: 27.0 pg/mL; Testosterone: cat. no. ADI-900-065, antibody: goat anti-mouse IgG, sensitivity: 5.67 pg/mL) according to the manufacturer's instructions. We chose to analyze corticosterone rather than cortisol concentrations because corticosterone is suspected to be the main adrenocorticotropic hormone (ACTH)-responsive glucocorticoid for R. imitator (Cockrem, 2013; Westrick et al., 2023). Samples were vortexed, and 100 µL of the samples were added in duplicate in individual wells of microtiter plates. Following manufacturer instructions, plates were read at 405 nm, with correction between 570 and 590 nm, using a microplate reader (Synergy H1, BioTek Instruments, Winooski, VT, USA). Hormone concentrations were calculated using a four-parameter logistic curve in the software Gen5 (version 3.05, BioTek Instruments, Winooski, VT, USA). Samples with duplicate measurements that yielded a coefficient of variation (CV) > 30% were excluded from later analyses (corticosterone: 7 of 59 samples, 11.9%; testosterone: 2 of 59 samples, 3.4%). Before removing these duplicates, intra-assay CV values for corticosterone and testosterone were 14.0% and 10.0%. After removal, intra-assay CV values were 8.4% and 9.4%. Inter-assay CV values were 11.8% and 13.7%.

Statistics

We cleaned and analyzed all data using RStudio (R version 4.2.3, R Core Team, Boston, MA). We implemented all statistical analyses using a generalized linear mixed model (GLMM) framework. For each analysis, we used the Akaike Information Criterion corrected for small sample sizes (AICc; Burnham and Anderson, 2004) to choose between alternative model fits. We evaluated standard model diagnostics and tested for outliers and quantile deviations using the ‘DHARMa’ package (version 0.4.6, Hartig, 2022). All models in this manuscript passed all diagnostic tests. To evaluate the significance of main effects, we performed a Type III analysis of variance with Satterthwaite’s method. We evaluated pairwise comparisons and accounted for multiple comparisons using the Tukey method in the ‘emmeans’ package (version 1.9.0, Lenth, 2023). We estimated effect sizes in multi-termed modeling using the ‘effectsize’ package (version 0.8.6, Ben-Shachar et al., 2020) and with pairwise effect size comparisons using ‘emmeans’. Figures were created using the package ‘ggplot2’ (version 3.4.4, Wickham, 2016) and assembled in Adobe Illustrator (Adobe Illustrator 2023).

Offspring manipulations

 We analyzed the number of eggs deposited for each tadpole among the four experimental conditions (Fig. 1A). To test whether egg deposition changes due to cross-fostering and intra-tank displacement, we used a GLMM with a zero-inflation component implemented using the ‘glmmTMB’ package (version 1.1.8, Brooks et al., 2017). We chose a Poisson distribution based on best fit, with eggs fed as a dependent variable, experiment type and number of siblings within a terrarium as independent variables, and focal terrarium as a random effect.

Takeover trials

To determine how testosterone, corticosterone, and testosterone: corticosterone ratios change relative to infanticidal behavior, we implemented a GLMM using the ‘lmerTest’ package (3.1-3, Kuznetsova et al., 2017). We set the outcome (“infanticide” vs “non-infanticide”), sampling event (“baseline” vs. “move” vs. “final”), sex, and their interactions as fixed effects and individual frog identity, average individual mass, volume of collection water, and body length (snout-vent length) as random effects. We selected this method rather than linearly normalizing by mass, body length, or water collection volume following correlation analyses (Figs. S1-2). We estimated average individual mass as the average body mass across sampling events, as this was necessary to account for intra-individual differences in feeding. Since mass was not collected for one individual frog, we used mean imputation to estimate the single missing value as the averaged individual mass across the entire study. To improve interpretation of regression coefficients, we removed insignificant interaction terms using stepwise backward regression.

We ran separate models for testosterone and corticosterone. We were additionally interested in ratios of testosterone to corticosterone, results for which are available in the supplement (Figs. S3-4; Tables S1-3). We selected appropriate data transformations (natural log, square root, inverse, no transformation) for modeling each dependent variable using AICc. Using AICc and diagnostic criteria, we selected a log transformation for corticosterone and an inverse transformation for testosterone. Ratios were modeled using the untransformed testosterone and corticosterone values to compute an initial ratio value, which was then log-transformed. Because corticosterone, testosterone, and ratio values of samples exceeding the CV threshold of 30% are deemed unreliable, we treated these values as missing data. All of our final models passed all diagnostic tests.