The Mamoní Valley Preserve is located in Central Panama and is managed as part of the larger Mamoní Valley Conservation area, which has 1,500 ha of old-growth forest and 3,500 ha of watersheds that are undergoing restoration efforts following degradation from cattle and agriculture (Geoversity, 2020). A. limosus were collected here as a source population for the captive assurance colony at the Panama Amphibian Rescue and Conservation Project following the first Bd detection in the adjacent Chagres National Park in 2009 (Lewis et al., 2019). We selected a first-order perennial stream tributary to the Mamoní River for its accessibility, deep pools separated by riffles, and waterfalls in a mature secondary rainforest that does not have extant Atelopus populations. This site was about 5 km SE from the Madroño River where a small wild population of A. limosus persists.

The aim of this study was to evaluate the movement and survival of 60 frogs that were held in the breeding facility and released directly into the wild (hard release) compared with 23 frogs that had been held for 30 days prior to release in mesocosms established in the wild (soft release). We used animals aged approximately 1 year derived from two clutches from founders collected at the Madroño River. ARC-Alim-F1-11 were descended from a male founder collected in 2010 and a female collected in 2011. ARC-Alim-F1-13 were derived from a male collected in 2010 and a female reared in captivity in 2013. The overall sex ratio was approximately even (37 female : 46 male). Frogs were individually marked with visible alphanumeric tags in accordance with the manufacturer’s instructions [VI alpha; Visible Implant (VI) Tags standard size (1.2 mm × 2.7 mm), Northwest Marine Technology] (Heard et al., 2008; Pittman et al., 2008; Osbourn et al., 2011). We used a sterilized (alcohol-flamed) injector needle to insert the tag under the skin of each frog in the tibial region of the left hind leg rather than the dorsum, because it was less pigmented. We individually photographed the frogs’ dorsal and ventral sides as a backup photographic identification reference. This proved necessary as the sub-cutaneous tags were difficult to read upon recapture, and we do not recommend them for future Atelopus studies. Soft-released animals were held individually in outdoor mesocosms (76 cm × 76 cm × 46 cm polythene mesh with 6 mm gaps) for 1 month prior to release (Estrada et al., 2022). We established three consecutive 80-m stream transects and released 7–8 “soft-release” and 20 “hard-release” A. limosus on each transect, placing one randomly selected frog every 5 m on either side of the stream at the transition from the dry to the rainy season on 17 May 2017. We conducted recapture surveys for the A. limosus and swabbed the existing frog community for Bd to estimate prevalence by conducting nocturnal and diurnal searches of the stream transects on weeks 4 and 8.

Frogs were swabbed for Bd following a standard protocol using an MW113 swab five times on each foot, 10 times on the belly, and 10 times on each thigh for a total of 40 passes (Kriger et al., 2006; Hyatt et al., 2007). Swabs were placed into 1.5-mL microcentrifuge tubes and stored in a –20°C freezer until processing. Swabs were extracted using Applied Biosystems PrepMan™ Ultra (Waltham, USA) following the manufacturer’s protocol. For Bd detection, we used a real-time PCR assay following a standard protocol for Bd (Boyle et al., 2004) with some slight modifications. We used a Roche LightCycler^® 96 System and 20 μL reactions consisting of 5 μL of sample and 15 μL of a master mix (Kriger et al., 2006; Hyatt et al., 2007; Estrada et al., 2022). We tested the samples in triplicate, along with five Bd-negative and one or two Bd-positive controls on every plate (Boyle et al., 2004; Kriger and Hero, 2007). Samples with two or three positive reactions were considered Bd positive. A few samples were repeated to confirm results. We quantified the infection load of Bd-positive samples based on the estimated number of zoospore equivalents using standards prepared from the JEL 423 Bd strain (Boyle et al., 2004; Estrada et al., 2022). We averaged the quantified load values and multiplied them by 100 to compensate for the extraction and dilution of the samples.

We equipped 17 female A. limosus with radio transmitters prior to release (nine soft-release and eight hard-release frogs). We used some of the smallest commercially available 0.31 g LB-2X transmitters with a 21-day battery life (Holohil Systems Ltd.) and attached transmitters to the largest females, which weighed approximately 3 g or more, to approximate the 10% radio transmitter-to-body-mass ratio traditionally observed in herpetological studies (Heyer et al., 1994; Altobelli et al., 2022). Surgical silicone tubing was used as a waistband with a fine cotton thread to mount the radio transmitters on the amphibians (Figure 1) (Bartelt and Peterson, 2000; Rowley and Alford, 2007; Pašukonis et al., 2014). The organic cotton thread eventually degrades and breaks, acting as an auto-release mechanism (Rowley and Alford, 2007; Pašukonis et al., 2014). We used a Biotracker radio receiver (Lotek, Ontario, Canada) to track A. limosus equipped with radio transmitters after release once per day during weekdays (total of 31 sessions). When an individual was located, we recorded coordinates using a Garmin GPS in WGS84. Fluorescent flagging tape with the individual’s identification number was placed on the nearest vegetation as a starting point for the next radio-tracking period. Once a week we handled radio-tracked frogs to examine the condition of their skin where the silicone tubing contacted the frog, weighed and measured SVL, and swabbed them for Bd. The standard battery life of the transmitters is 21 days, with some variation, so we replaced radio transmitters every 18 days for a total radio-tracking duration of 54 days, after which the harness was removed and the frog released. If abrasion lesions were noted during the weekly swabbing and physical evaluation (n = 5 animals), the frog was removed from the study, the harness was removed, and the frog was placed in a plastic mouse cage and fed with wingless fruit flies. Abrasions were externally treated daily with silver sulfadiazine cream for 1 week, then frogs were released without any transmitter. Any non-radio-tagged A. limosus encountered during these radio-tracking surveys were identified or photographed and swabbed for incorporation into the mark–recapture estimates. The entire site was traversed during each radio telemetry survey, representing a continuous unconstrained search for all released animals. We also conducted weekly dedicated stream transect surveys to find non-radio-tracked animals, and the animals found were identified, photographed, and swabbed for mark–recapture analysis.

We estimated treatment-specific (hard vs. soft) survival probabilities by fitting a Cormack–Jolly–Seber model to individual encounter history data (Cormack, 1964; Jolly, 1965; Seber, 1965). We fit this model in a hierarchical Bayesian framework by specifying (1) a sub-model for the true state of each individual over time (alive or dead) and (2) a sub-model for the observation data (observed or not) (Royle, 2008; Kéry and Schaub, 2011). Our dataset integrated both radio-tagged individuals and VI-tagged individuals recaptured by chance in each of 31 radio-tracking survey occasions and four dedicated stream recapture surveys in the 58-day study period (total of 291 recaptures). We estimated standardized 30-day survival probabilities for both the hard- and soft-release treatment groups. Field observations indicate that radio-tracked individuals may have poorer survivorship than VI-tagged animals due to transmitter lesions, but for the purposes of this analysis we assumed survival did not differ between radio-tracked and VI-tagged individuals in order to estimate general survivorship for hard- and soft-released groups. We accounted for different detection probabilities associated with radio telemetry and visual resighting of individuals marked with visual implant tags but assumed, due to similarly low numbers of recaptures, that the detection probability of VI-tagged individuals did not differ between dedicated stream surveys and incidental observations during radio telemetry surveys. In cases of known mortality events (n = 7 animals), this additional information was incorporated by supplying the alive/dead state of the individual as observed data, rather than treating the state as a parameter to be estimated for all surveys in which the individual was known to be dead. The lone exception was in the case of a single accidental mortality from a frog being stepped on, where observation data were “right-censored” by treating data as missing for all surveys after the mortality occurred to avoid biasing survival estimates. Observation data were also right-censored for four individuals from the date when radio transmitter signal was lost, with data treated as missing for all survey dates after the last detection of each individual.

We fit the model using the JAGS software run using the jagsUI package in the R program (Kellner, 2019; Plummer, 2020; R Development Core Team, 2021). We assigned uniform (0,1) priors for both detection and monthly survival probabilities. As is commonplace for most Bayesian analyses, JAGS performs model fitting using Markov chain Monte Carlo (MCMC) sampling by simulating probability distributions that are too complex to calculate mathematically. Probability distributions for model parameters were obtained by running three independent MCMC chains for 500,000 iterations each, discarding the first 250,000 iterations from each chain as “burn-in”, and thinning the remainder at a rate of 1 : 20 to obtain 37,500 samples from the joint posterior distribution (code provided in Supplementary Material 2).

The individual fates of all radio-tracked animals are summarized in Table 2. Four of the frogs (25%) experienced abrasions from the radio transmitter and were removed from the experiment. All the abrasions were detected during the third consecutive transmitter placement (between days 36 and 54). Four of the transmitters (25%) failed at some point during the tracking. Two of the frogs died from predation and one likely died from Bd. Only two were released into the wild following their final recapture at the end of the 58-day tracking period (Table 2).

We estimated per-survey detection probabilities of 0.024 (95% CRI: 0.014 to 0.04) for visual resighting of marked individuals and 0.88 (95% CRI: 0.84 to 0.92) for radio tagged individuals. Our low recapture rate of marked individuals led to a small sample size and considerable uncertainty in survival estimates. Monthly survival probability was estimated at 0.31 (95% CRI: 0.13 to 0.57) for individuals in the hard-release treatment and 0.46 (95% CRI: 0.23 to 0.72) for individuals in the soft-release treatment. The estimated difference in monthly survival (soft – hard) was 0.15 (95% CRI: –0.15 to 0.44), which equated to a posterior probability of 0.85 that individuals in the soft-release treatment had higher survival rates (Figure 4).

After 1 month in mesocosms, about 25% of the frogs were Bd positive (Table 3, Estrada et al., 2022). This prevalence of Bd was similar to that observed in the stream frog community (24%–29%; Table 4) during our study. Smaller species of amphibians such as Dendrobatids or recent metamorphs of larger species were observed inside the mesocosms. Contaminated leaflitter potentially served as an additional source of infection. None of the hard-released animals were Bd positive at the time of release or upon subsequent recapture. One radio-tracked animal was found deceased with high Bd loads (6,950 zoospore equivalents) and two animals with transmitter-related abrasions died with high Bd loads (6,653 and 181,400 zoospore equivalents), with Bd likely contributing to the poor health and skin condition of these animals.