Body condition, skin bacterial communities and disease status: insights from the first release trial of the limosa harlequin frog, Atelopus limosus

Abstract

Many endangered amphibian species survive in captive breeding facilities, but there have been few attempts to reintroduce captive-born individuals to rebuild wild populations. We conducted a soft-release trial of limosa harlequin frogs, Atelopus limosus, which are highly susceptible to the amphibian chytrid fungus Batrachochytrium dendrobatidis (Bd), to understand changes associated with the transition from captivity to the wild. Specifically, we assessed changes in body condition, skin-associated bacterial communities and disease status after release. Frogs were housed individually in field mesocosms and monitored for 27 days. Body condition did not significantly change in the mesocosms, and was similar to, or higher than, that of wild conspecifics at day 27. The skin bacteria of captive-born frogs, based on 16S rRNA gene amplicons, became similar to that of wild conspecifics after 27 days in mesocosms. Prevalence of Bd in wild conspecifics was 13–27%, and 15% of the A. limosus in mesocosms became infected with Bd, but no mortality of infected frogs was observed. We conclude that mesocosms are suitable for systematically and repeatedly monitoring amphibians during release trials, and that body condition, the skin microbiome, and Bd status can all change within one month of placement of captive-born individuals back into the wild.

1. Background

Human activities are accelerating the loss of biodiversity [1,2]. As a result, in addition to in situ conservation, ex situ approaches, such as captive breeding programmes (CBPs), have become common for assuring endangered species survival when threats cannot be mitigated in natural habitats [3–5]. However, the establishment of assurance colonies under captive conditions is difficult, largely owing to the maintenance and management of genetic diversity of poorly known species [6]. Additionally, several aspects associated with captive environments can facilitate adaptation to captivity with potential negative effects on individual fitness, which could worsen over successive generations in captivity [7]. The chances of survival after release depend on whether captive-born individuals can adapt to the natural environment, and whether they are able to withstand emergent factors, such as infectious disease [8]. Despite these challenges, CBPs are integral for the recovery of endangered species; 15% of threatened vertebrate species are currently kept in captive facilities around the world [9]. Ultimately, the reintroduction of animals from captivity to their natural habitat is a primary goal of CBPs. However, attempted releases of captive-reared animals have had mixed results [10,11], while others remain unpublished, and initial release methods for many species are determined through a process of trial and error.

The rate of release success of captive-born vertebrates can increase with an appropriate assessment focused on conservation outcomes at the population, species and ecosystem level [3,10]. However, for species fighting infectious diseases, success rates are still poor [11,12]. Amphibians are facing disease-related population declines and extinctions worldwide. The emergence of chytridiomycosis, an infectious disease caused by the fungal pathogens Batrachochytrium dendrobatidis (Bd) [13] and Batrachochytrium salamandrivorans (Bsal) [14], is recognized as a major threat to amphibians [15,16]. Adaptive management frameworks for amphibians that are threatened by Bd include interventions to prevent further extinctions [17,18]. For instance, efforts have focused on reducing Bd in the environment [19] and infections on the amphibian hosts [20,21], while also improving the capacity of populations to persist despite increased disease-related mortality [18]. In areas where Bd is enzootic, additional pressures, such as climate change and habitat degradation, can compromise already struggling populations [17]. Since threats faced by amphibians are complex and often synergistic, mitigating threats in the wild remains a major challenge for reintroduction programmes.

Besides chytrid infections, post-release survival of captive-born amphibians could be influenced by factors associated with captivity, such as changes in behaviour, body condition and skin-associated microbial communities. Hence, as body condition is used as a metric to assess overall health [22] and host-associated microbial communities may be associated with protection against pathogens [23–25], changes in these variables pre- and post-release may be used as indicators of health and survival probability. In addition, while skin-associated microbial communities differ between captive and wild individuals [26–29], it is unknown whether differences will be maintained after individuals are released to the natural habitat.

Harlequin frogs (genus Atelopus) have experienced dramatic population declines and local extinctions throughout their distributional range in Central and South America [30]. Most Atelopus species are endemic, have relatively small distribution ranges and are highly susceptible to Bd infection, which has prompted their prioritization in CBPs [31]. We assessed changes in body condition, skin bacterial community and Bd infection status in captive-born limosa harlequin frogs, Atelopus limosus, during a soft-release trial using field enclosures (mesocosms) in a site where the species thrived prior to Bd arrival. We compared these data with surveys of persisting wild A. limosus to understand the amphibian–Bd–microbiome interactions during the transition from captive to wild conditions.

2. Methods

(a) . Field conditions and experimental design

Mamoní includes several streams and rivers flowing south into the Pacific Ocean and is located in the Chepo district within the province of Panamá, Panama. Two sections of a stream (approx. 500 m total) with suitable A. limosus habitat were selected for our trial. This site experienced a Bd outbreak in 2011–2012 that resulted in a drastic decline of the local amphibian community, including the A. limosus population (R. Ibáñez 2011–2012, unpublished data).

Mesocosms (76 × 76× 46 cm) were built from a non-toxic, pliable, yet semi-rigid polyethylene mesh used in commercial aquaculture. In each mesocosm, we placed rocks, palm leaves, branches and leaf litter from the local environment to provide cover and habitat structure that would encourage natural behaviour and movement (electronic supplementary material, figure S1). In April 2017, we placed 30 captive-born A. limosus (15 females and 15 males; rearing conditions in electronic supplementary material, methods) in individual mesocosms distributed along the stream edge. No supplemental food was provided, as the mesh allowed small arthropod prey to enter, while reducing mortality risk from larger predators. We did not conduct diet studies, but observed faecal pellets and potential prey items (e.g. ants, termites, mites and springtails) in the mesocosms, suggesting that natural feeding was occurring. Frogs were monitored for body condition, bacterial community and Bd status on a weekly basis across sampling days: 0 (=captivity), 1, 7, 14, 21 and 27, after placement in the mesocosms. After day 27 sampling, all remaining frogs were transferred back to the Panama Amphibian Rescue and Conservation project (PARC) for a follow-up study.

(b) . Field sampling and laboratory processing

(i) . Body condition

We recorded body mass (g), snout–vent length (SVL, mm) and signs of infection or injuries on a weekly basis (from day 0 to day 27). We assessed the change in body condition throughout the soft-release trial using the ‘scaled mass index’ (SMI) following [32]. For each individual, this method predicts a standardized estimation of mass based on a specific size (L₀) and uses as a scaling exponent (b_SMA) the slope from a standardized major axis (SMA) regression between log₁₀(mass) and log₁₀(SVL). The SMA regression was conducted using the function lmodel2 from the lmodel2 package [33]. In addition, for comparative purposes, we calculated the SMI of wild adult frogs that were surveyed at the study site in 2011–2012 during initial rescue efforts (total = 63; 10 females and 53 males) and during the present study (total = 5; one female and four males). Because of the size sexual dimorphism (SVL mean ± s.d.: females = 35.64 ± 5.99 mm; males = 27.40 ± 1.05 mm), we calculated the SMI for each sex separately.

(ii) . Bacterial community and Batrachochytrium dendrobatidis status

We swabbed the skin of the experimental captive-born A. limosus while still in captivity (day 0) and then on days 1, 7, 14, 21 and 27, and of the wild ones we encountered during the experiment (N = 8), with a sterile rayon swab to collect a sample of the bacterial community and to assess Bd infection status, as described by Belden et al. [34]. Wild individuals were additionally photographed dorsally and ventrally for individual identification purposes and released immediately after sampling. At two time points (day 2 and day 30), we surveyed and swabbed the broader amphibian community at the study site by conducting nocturnal surveys of the 500 m stream transect to assess the incidence of Bd at the community level (N = 126, 75 on day 2 and 51 on day 30). Swabs were placed in 1.5 ml sterile microcentrifuge tubes and kept frozen at the field station until they could be moved to −80°C.

(iii) . DNA extraction and sequence processing for skin bacterial communities

DNA from A. limosus swabs to characterize skin bacterial communities and to detect and quantify Bd infections was extracted with the Qiagen DNeasy blood and tissue kit (Valencia, CA), following the manufacturer's protocol with a lysozyme pre-treatment. 16S rRNA gene amplicon sequencing of the bacterial V4 region was used to assess bacterial communities [35]. Sample preparation for sequencing followed Estrada et al. [36].

Samples for this study were part of two Illumina MiSeq runs (single end, 250 bp) at Harvard's Dana Farber Cancer Institute genomics facility. Demultiplexing of sequences was completed in QIIME 2 [37]. Samples from each run were then quality-filtered, denoised and chimera-checked with DADA2 [38], with all reads less than 251 bp removed, and any reads with Phred quality scores less than 20 removed. Following that, sequence data from the two runs were merged into a single feature table, as were the representative sequences for each run. With these merged files, an operational taxonomic unit (OTU) table was created with all the samples based on 97% de novo clustering using VSEARCH [39]. We further filtered out any OTUs that had fewer than 0.01% of the total reads [40]. Taxonomy was assigned to the remaining OTUs with a machine-learning-based method using the pre-trained SILVA database classifier [41,42]. OTUs assigned as chloroplast or mitochondria were then filtered out, leaving only bacterial OTUs. We examined a rarefaction plot to determine the appropriate rarefaction depth for the dataset and chose to rarefy at 10 000 reads per sample. This resulted in the loss of 5 samples from the initial dataset that had lower read counts. The final OTU table contained 513 OTUs across a total of 109 samples (101 from captive-born A. limosus, 8 from wild frogs), and this was the table used for microbiome analyses. A randomized axelerated maximum likelihood (RAxML) phylogenetic tree of the OTUs was created using a GTRCAT substitution model and 100 bootstrap replicates, with a midpoint root added so it could be used in phylogenetic analyses.

(iv) . Batrachochytrium dendrobatidis detection and quantification

We used quantitative TaqMan PCR [43,44] to quantify Bd prevalence and infection load in both the Atelopus and the amphibian community samples (details in electronic supplementary material, information).

(c) . Statistical analysis

(i) . Body condition estimates

For body condition analyses, we used 20 captive-born frogs (10 females and 10 males; electronic supplementary material, table S1) for which we had measurements from all sampling times. As the data met assumptions of normality (Shapiro–Wilk normality test), we fitted a linear mixed model (LMM; lmer function, package lme4) with SMI as the response variable, time (sampling day: 0, 7, 14, 21 and 27) as a predictor and frog ID as a random factor, allowing for a random intercept [45]. We visualized residual plots against model predictions to confirm the appropriateness of the error distribution, and the significance of the predictor was determined (Anova function, Type II sums of squares, package car) [46]. In addition, we complemented our assessment of the change in frog body condition during the mesocosm trial by conducting a paired t-test comparing SMI means between captive-born individuals before (day 0) and at the end (day 27) of the soft-release trial. Lastly, to determine whether captive-born individuals reached an SMI comparable to that of wild conspecifics at the end of the soft-release (day 27), we conducted Mann–Whitney U-tests for unpaired data. We used the non-parametric test in this case because the data from wild individuals were non-normal. t-tests and Mann–Whitney U-tests were conducted using the package stats.

(ii) . Change in skin bacterial communities after release

We compared skin bacterial communities of A. limosus before (day 0) and after (day 27) soft-release into the mesocosms with those on wild individuals from the extant population using Mann–Whitney U-tests. We computed the α-diversity metrics OTU richness, the Shannon index (considers both richness and evenness) and Faith's phylogenetic diversity with QIIME 2. These tests were implemented using the package stats.

We used permutational multi-variate analysis of variance (PERMANOVA, adonis function, package vegan, [47]) to analyse variation in bacterial community composition and structure before (day 0) and after (day 27) soft-release into mesocosms, and to compare mesocosm individuals with wild frogs. To visualize the overall change in the bacterial communities throughout the soft-release, we used principal coordinate analysis (PCoA) based on Bray–Curtis and Jaccard distance matrices, which included all the samples from frogs in captivity (day 0), mesocosms (day 27) and wild conspecifics. We also examined these patterns based on the phylogenetic β-diversity distance metrics (weighted and unweighted unifrac). To assess the changes in variation among individuals before and after the soft-release trial and how they compare with the variation among wild individuals, we compared the multi-variate homogeneity of group dispersion (i.e. distance of samples to cluster centroid) using the different dissimilarity metrics. We calculated dispersion within each group (captivity, day 27 and wild) using the function betadisper (package vegan), and conducted statistical comparisons by applying a permutation-based test (999 permutations) using the function permutest (package vegan), allowing for pairwise comparisons.

The ordination results suggested differences in bacterial community structure of the captive (day 0) frogs versus the mesocosm day 27 and wild frogs. We identified key bacterial taxa responsible for the observed difference between frogs in captivity (day 0) versus the other two groups using ANCOM2 (an extension of the analysis of composition of microbiomes, ANCOM), which accounts for the compositional nature of the data [48]. To implement ANCOM2, we followed the R pipeline developed by Lin in 2019 [49] and used the default parameters. ANCOM2 computes the proportion of OTU log-ratios (pairwise comparisons) with significant differences between groups and uses a cut-off value (proportion ≥ 0.7) above which an OTU is considered differentially abundant. For our analysis, OTUs with W statistics greater than or equal to168 (number of significant pairwise comparisons) were considered differentially abundant between groups (electronic supplementary material, figure S2). To more clearly visualize these results, we plotted the relative abundances of the 85 OTUs that were identified as differentially abundant in a heatmap generated using the R function heatmap.2 in the package gplots [50].

For a subset of 12 frogs, we also tracked the changes in the skin bacterial communities through time (days 0, 1, 7, 14, 21 and 27). For this, we focused on changes in OTU richness and Faith's phylogenetic diversity over those time points, using an LMM (lmer function, package lme4) based on normality assumptions. We fitted the LMM by including OTU richness or Faith's phylogenetic diversity as the response variable, time (sampling day: 0, 7, 14, 21 and 27) as a predictor and frogID as a random factor, allowing for a random intercept [45]. Residual plots against model predictions confirmed the appropriateness of the error distribution, and the significance of the predictor was determined (Anova function, Type II sums of squares, package car). Exploratory models including time as a random slope to assess individual-level variation resulted in overfitted models owing to a lack of statistical power to include complex random-effect structures. We examined variation in community structure across time points for these 12 individuals using a PERMANOVA (adonis function, package vegan, [47]) and used a PCoA based on Bray–Curtis dissimilarities to visualize the variation among frogs and across time points. We also examined this pattern based on the phylogenetically based weighted unifrac distances.

(iii) . Batrachochytrium dendrobatidis infection status

We quantified Bd zoospores in a total of 262 swabs: 136 from A. limosus (including mesocosm-housed individuals; 12 Bd-positive samples from four individual frogs, table 1) and 126 from frogs of the amphibian community, including eight wild A. limosus. We qualitatively compared log-transformed Bd infection loads of the 12 Bd-positive A. limosus samples with those of the broader community (N = 16 species, 126 total individuals; electronic supplementary material, table S2). As only four captive-born A. limosus tested positive for Bd infection after placement in mesocosms, trends we see in Bd results should be interpreted with caution.

Table 1.

Bd prevalence and infection intensity of captive-born A. limosus throughout the soft-release trial (i.e. sampling days). Bd prevalence represents the proportion of Bd-positive (Bd+) individuals out of the total number of individuals sampled. Bd infection estimates (mean and range) are presented in zoospore equivalents (z.e.).

sampling day	total frogs sampled	number Bd+	prevalence (%)	mean Bd intensity (z.e.)	range of Bd load
0	30	0	0	0	0
1	30	0	0	0	0
7	28	0	0	0	0
14	28	2	7	15.5	2–29
21	28	2	7	280.5	70–491
27	26	4	15	5838.50	5–23 100

3. Results

(a) . Change in Atelopus limosus body condition during the trial differed between sexes

There were different trends between females and males in the change in SMI throughout the soft-release trial; the SMI decreased in females (LMM: X² = 20.13, p < 0.0001, parameter estimate = −0.015, s.e. = 0.003; figure 1), but did not change substantially in males (LMM: X² = 0.67, p = 0.41, parameter estimate = −0.002, s.e. = 0.003; figure 1). The same pattern was detected for the direct comparisons between day 0 and day 27, with a change seen in females (t = 3.80, d.f. = 9, p = 0.004), but not in males (t = 1.70, d.f. = 9, p = 0.12). While SMIs of captive-born females at the end of the trial (day 27) were lower than those of wild females, this was a not statistically significant difference (W = 40, p = 0.31; figure 1). By contrast, SMIs of captive-born males were higher than those of wild males at the end of the trial (W = 557, p < 0.0001; figure 1).

Figure 1.

Change in SMI of soft-released frogs throughout the experiment (captivity = day 0 (grey points); in mesocosms = days 1–27 (gold points)), and the comparison with wild conspecifics (green points). Points represent the SMI of individual frogs. Inverted triangles and bars represent the arithmetic mean and bootstrap 95% confidence intervals, respectively. (Online version in colour.)

(b) . The skin bacterial community on captive-born Atelopus limosus reverts to wild condition

Our dataset contained 513 unique bacterial OTUs, which included 136 bacterial families from 13 phyla. OTUs associated with amphibian skin predominantly belonged to the phyla Proteobacteria, Bacteroidetes, Actinobacteria, Firmicutes and Verrucomicrobia. The 35 bacterial families with the highest relative abundance (greater than 0.01) represented 91% of the bacterial community associated with frogs in captivity, in mesocosms and in the wild (electronic supplementary material, figure S3). Changes in bacterial community structure on captive-born A. limosus after placement into mesocosms were mainly driven by changes in the bacterial families with the highest relative abundances, such as Sphingobacteriaceae (day 0: 47% – day 27: 17%). Wild frogs were enriched with Intrasporangiaceae (9%).

Bacterial α-diversity, in terms of both OTU richness and the Shannon Index, increased after soft-release into mesocosms and became similar to that of wild individuals by day 27 (figure 2a,b, day 27 – wild: OTU richness W = 103.5, p = 0.90; Shannon index W = 109, p = 0.73). By contrast, and as expected, frogs in captivity (day 0) had lower α-diversity than wild frogs (figure 2a,b, day 0 – wild: OTU richness W = 69, p = 0.070; Shannon index W = 64, p = 0.045;). For example, mean bacterial OTU richness of captive-born frogs (day 0) was about 79% of that observed in wild frogs. The results were the same for Faith's phylogenetic diversity (day 27 – wild: W = 101, p = 0.98; day 0 – wild: W = 70, p = 0.076; data not shown).

Figure 2.

Alpha-diversity as represented by OTU richness (a) and the Shannon index (b). PCoA showing dissimilarities based on Bray–Curtis (c) and Jaccard (d). In the ordinations, each point represents the bacterial community structure on the skin of one frog. All figures show data from frogs in captivity (N = 30) sampled on day 0 and in mesocosms (N = 26) sampled on day 27. Frogs in the wild (N = 8) were sampled, when encountered, throughout the present study. (Online version in colour.)

There was a significant shift in skin bacterial community composition and structure on captive-born A. limosus after soft-release into mesocosms, ultimately overlapping the community found on wild conspecifics (Bray–Curtis: pseudo-F = 6.988, R² = 0.376, p < 0.001, figure 2c; Jaccard: pseudo-F = 8.699, R² = 0.441, p < 0.001, figure 2d). The patterns were the same when considering phylogenetic relatedness with weighted and unweighted unifrac matrices (weighted unifrac pseudo-F = 11.486, R² = 0.277, p < 0.001; unweighted unifrac pseudo-F = 15.14, R² = 0.335, p < 0.001; data not shown). The changes in bacterial community structure after the soft-release trial also included a significant increase in community dispersion, which by day 27 matched that observed in wild individuals. This pattern was consistent across all the dissimilarity metrics (Bray–Curtis F = 14.03, p = 0.001; Jaccard F = 12.83, p = 0.001; unweighted unifrac F = 15.19, p = 0.001; weighted unifrac F = 3.12, p = 0.038; all metrics: captivity–day 27 and captivity–wild: permuted p < 0.05; day 27–wild: permuted p > 0.05).

With the ANCOM2 analysis, we identified 85 bacterial OTUs that were differentially abundant between captivity (day 0) and the other two groups (figure 3; electronic supplementary material, figure S2 and table S3). Differentially abundant OTUs that were associated with captive conditions at day 0 included members of the families Sphingomonadaceae, Rhodobacteraceae (genus Paracoccus) and Moraxellaceae (Acinetobacter), which persisted at low relative abundances during the soft-release trial. There were also differentially abundant OTUs that increased in their relative abundance after frogs were placed in mesocosms, including members of the families Staphylococcaceae (Staphylococcus), Xanthomonadaceae (Stenotrophomonas), Enterobacteriaceae and Sanguibacteraceae (Sanguibacter). Only a few differentially abundant OTUs, however, presented high relative abundances solely on the skin of wild frogs (e.g. Intrasporangiaceae).

Figure 3.

Heatmap showing the relative abundances (log-transformed) of the differentially abundant OTUs identified with ANCOM2. Differentially abundant OTUs were determined based on a cut-off of 0.7 (proportion of significant pairwise comparisons). These OTUs best defined the bacterial community diversity of A. limosus before (day 0) and after (day 27) placement in mesocosms and of wild conspecifics. Rows represent differentially abundant OTUs and columns are A. limosus individuals. OTU taxonomy is presented at the family level, except for three OTUs whose family was not available in the SILVA database, followed by the first three letters of the phylum. Aci: Acidobacteria; Act: Actinobacteria; Bac: Bacteroidetes; Fir: Firmicutes; Pla: Planctomycetes; Pro: Proteobacteria; Ver: Verrucomicrobia. (Online version in colour.)

For the subset of 12 individuals on which we tracked skin bacterial communities over time, we observed a sudden drop in OTU richness immediately after placement into mesocosms, followed by an overall increase after one week of exposure to natural conditions (LMM: X² = 11.58, p < 0.001, parameter estimate = 1.76, s.e. = 0.52; figure 4a). This result was consistent when considering Faith's phylogenetic diversity (LMM: X² = 19.54, p < 0.001, parameter estimate = 0.24, s.e. = 0.05; data not shown). Sampling day also had a significant effect on bacterial community structure (Bray–Curtis pseudo-F = 7.662, R² = 0.386, p < 0.001, figure 4b; weighted unifrac pseudo-F = 9.679, R²= 0.442, p < 0.001, data not shown).

Figure 4.

OTU richness and community structure of the skin bacterial community of 12 captive-born A. limosus before (day 0) and throughout (days 1–27) the soft-release trial. OTU richness (a) of individuals over time; the black line represents predicted OTU richness, with the respective 95% confidence interval, based on an LMM. Each point represents an individual richness value, colour-coded by individual frogs. Triangles mark frogs that were Bd+ at that time point. Bacterial community structure (b) on individuals over time is represented by a PCoA plot based on Bray–Curtis dissimilarities. Each point represents the community structure on an individual, with the larger points representing the cluster centroid of each time point (day 0–day 27). Both figures show data from a subset of 12 experimental frogs sampled at each time point after placement in the mesocosms.

(c) . Batrachochytrium dendrobatidis infection pattern was similar across reintroduced and wild frogs

All A. limosus tested negative for Bd infection while in captivity. After one month of exposure to natural conditions in mesocosms, 15% (4/26) of captive-born individuals were Bd-positive, and the mean infection intensity of positive individuals by day 27 was 5838 zoospore equivalents (z.e.) (range: 5–528 500 z.e., table 1). Infection was first detected on two individuals on day 14. By the end of the release trial the overall prevalence of Bd and infection intensity in captive-born frogs was within the ranges observed in the wild frog community (13–27% prevalence with a range of 12–491 600 z.e.; electronic supplementary material, table S2).

Of the 30 frogs, 4 (13%) died during the mesocosm trial; one was potentially lost to army ant predation. The other three deceased frogs were Bd-negative and the cause of death was unknown.

4. Discussion

Few studies have attempted to quantify changes associated with the reintroduction of captive-born frogs, and yet, these data are critical for moving forward with plans to get animals back into nature from CBPs. We used terrestrial mesocosms and a soft-release design to track changes in body mass, skin-associated bacterial communities, and infection status on limosa harlequin frogs at a site where the species used to thrive prior to Bd arrival. Captive-born frogs were exposed to natural environmental conditions, including diet and substrates characteristic of their natural habitat. By housing frogs in mesocosms, we had the ability to reliably recapture individuals at multiple timepoints and use these observations to inform future release trials.

In other amphibian releases (alpine tree frog in Australia [51]; Chinese giant salamanders in China [29]), both increased and decreased body mass have been observed without a significant effect on survival. In our study, after 27 days in mesocosms, survivorship was high (87%). We found that female A. limosus body condition (based on SMI) decreased after placement into mesocosms, but it was not a significant change, and it did not drop below the SMI of wild counterparts. Male body condition did not change from captivity to the mesocosms, and their SMI remained higher than that of wild counterparts at the end of our trial, perhaps because in some instances, mesocosms could increase foraging time because frogs are protected from many predators. As wild individuals are adapted to local habitat conditions, we suggest that a criterion for natural or ‘ideal’ body condition after release should approximate that of wild conspecifics, which our trial seemed to meet over the 27 days. However, we did not monitor actual food intake of the frogs, and while we did observe the presence of invertebrate prey in the mesocosms, the decline in female SMI could have occurred because they did not have adequate food resources available, which is another possible outcome of being placed in a mesocosm. Regardless, monitoring body condition in release trials may inform animal husbandry adjustments prior to release, point to the need of assistance (i.e. feeding) during acclimatization and help determine whether body condition in captivity relates to post-release survival.

Changes in skin-associated bacterial communities were observed soon after placing captive-born A. limosus into mesocosms, and by the end of the trial, skin communities were indistinguishable from those of wild conspecifics from the extant population. Our results are consistent with those that have found reduced bacterial community diversity under captive conditions [52–54]. Captive A. limosus had 79% of the OTU richness of extant wild frogs, similar to the pattern observed in a prior study on captive-bred Panamanian golden frogs, A. zeteki, which after 8 years in ex situ CBPs, shared 70% of their bacterial taxa with wild frogs [26]. Less diverse bacterial communities in captivity are attributed to artificial and sterile environments, less diverse diets, and the routine veterinary use of antibiotics and antifungals in captive collections [55].

The effects of captivity on host-associated microbiomes have been described for a wide range of vertebrate species, but the bacterial community shift post-release has been less explored. Here, we provide evidence that there is a rapid shift of the skin bacterial communities from a captive-type to a wild-type after placing captive-bred frogs in the natural environment. Given the known importance of environmental reservoirs for maintenance of the amphibian skin microbial community [53,56], we assume this rapid shift is due to contact with the soil and other surfaces in the mesocosms, along with the simultaneous loss of exposure to laboratory-specific microbes. Interestingly, we found extensive variation in individual responses (e.g. in OTU richness, we see large fluxes for individuals over the sampling time points). This has been seen in other mesocosm studies examining the changes in amphibian skin microbial communities of individuals over time [57]. In some cases, this may be due to rainfall events or other environmental factors or disturbances that either directly impact the hosts or alter the environmental source pools of bacteria [36,57]. Despite the shorter-term variations, the bacterial communities on the frogs in our mesocosms arrived at a composition similar to that of wild conspecifics after approximately a month. This suggests that while environmental source pools are likely important, host factors, such as innate immune responses, are also likely important in determining the final composition of the skin bacterial communities.

While this shift back to a wild-type skin bacterial community seems to happen quickly, it is not clear if this should be viewed as a positive or negative event in terms of future Atelopus conservation. While we tend to applaud the ‘natural state’, Atelopus with their natural skin bacterial communities have suffered devastating population losses in the face of Bd [30,58,59]. Therefore, it is not clear that the natural skin bacterial communities of Atelopus can provide host disease defence, as suggested in some other amphibian species [60]. Prior studies have manipulated the skin microbiome of Atelopus in laboratory trials to try to prevent Bd infection [61,62], but in general, the manipulation of bacterial skin communities via probiotic treatments was unsuccessful. Our observation of a rapid shift in the skin microbiome of subadults placed in natural conditions suggests that laboratory-based manipulations of these bacterial communities designed to provide Bd protection to individuals released in Bd-endemic areas may not be maintained under natural conditions. Thus, other approaches, such as laboratory-based selection for Bd-resistance, may be more valuable to pursue for long-term Atelopus conservation and reintroduction. Indeed, a recent study suggested that some persisting populations of Atelopus have evolved anti-Bd skin secretions that have allowed survivors to persist and perhaps even recover from the disease [63].

Captive breeding facilities keep pathogen-free populations with strict biosecurity protocols to reduce disease risk [64,65]. While our released animals were all Bd-negative at the beginning of the trial, we detected the first Bd-positive individuals after two weeks in the mesocosms. In total, four captive-born frogs (15% prevalence) were infected with Bd over the 28 days of the trial. This infection timeframe is consistent with laboratory Bd-infection trials of Atelopus that have detected Bd-positive samples one week after experimental inoculation, with a rapid increase in infection intensity over time [66–68]. Though our trial spanned a short period of time, our surveys of wild frogs at the site also indicate that Bd is present in the environment, and capable of infecting susceptible species and naive captive-born individuals soon after exposure to their natural habitat. We do not know the exact route of exposure for our four mesocosm frogs that became infected. Given our set-up, the most likely scenarios may have been rain events that transported zoospores in the environment (the mesocosms were not directly in contact with the stream) or contact with other amphibians that were small enough to enter the mesocosms.

In the broader amphibian community, Bd prevalence estimates ranged from 13 to 27% over the course of our trial. Infection intensity during that time varied across amphibian species at the study site. Specifically, we found higher Bd infection intensity in Colostethus panamansis, Incilius coniferus and Pristimantis cruentus; some of these species have suffered population declines at higher elevations [58]. Within this context, the co-occurrence of an Atelopus species with other members of the community harbouring a relatively high Bd prevalence have been found previously, including in Panama. For instance, at a mid-elevation site (approx. 500 m) in Panama, Perez et al. [69] detected high Bd prevalence on C. panamansis (24%), Rhaebo haematiticus (42%), Silverstoneia flotator (33%) and Smilisca spp. (100%), while Atelopus varius still persisted. Overall, despite our inability to detect changes in community Bd prevalence during the release trial, our results suggest that amphibians in this site are surviving with Bd in an enzootic infection stage, and that some species are more susceptible to infection than others. Understanding disease risk before, during and after reintroductions is critical to achieve long-term conservation goals not only for Atelopus species, but also for other declining Neotropical amphibians. While we have a limited ability to make strong inferences from a handful of infected soft-released individuals and their skin-associated bacterial communities or body condition, repeated observations with appropriate monitoring of control sites may help us to understand the disease consequences of releasing susceptible captive animals into the wild.

Ultimately, many Neotropical amphibian reintroduction efforts may face a high risk of failure, mainly because Bd is now enzootic in many regions [19], and virulence has not attenuated [63]. Additional challenges, such as climate change, pollution and habitat degradation, place additional pressure on these populations [17]. However, some species that were thought to have experienced local extinctions are rebounding [69,70], and communities are showing signs of recovery. For captive frogs, reintroduction efforts require adaptive management approaches that use novel field, molecular, reproductive and geospatial tools to facilitate the reestablishment of self-sustaining wild populations [10,71,72]. Moreover, human dimensions have been recently identified as a key component for amphibian reintroduction success [73]. The goal of this release trial was not to restore a wild population of A. limosus, but to better understand some of the challenges faced by frogs as they transition from captivity back into the wild, and to inform adaptive management initiatives. Mesocosms, and this soft-release approach, allowed us to effectively recapture animals and observe disease and microbiome shifts during this transition. Extended monitoring of captive and released individuals may help to further study disease dynamics in captive-bred animals and provide further insights into captive-to-wild translocations.

Ethics

This study was conducted under approval of the Institutional Animal Care and Use Committees of Virginia Tech (11-105-BIOL) and the Smithsonian Tropical Research Institute (2013-0401-2016-A3). As individuals were reared and housed at the Panama Amphibian Rescue and Conservation (PARC) Project prior to soft-release, their implementation and research committee established that the proposal met all animal welfare requirements. Fieldwork was completed under permit SE/A-47-12, granted by the Panama Ministry of Environment (MiAmbiente).

Data accessibility

Raw DNA sequences: NCBI, http://www.ncbi.nlm.nih.gov/bioproject/721972. Atelopus limosus mesocosm data, wild conspecific data and broader community data: Virginia Tech data repository, https://doi.org/10.7294/20057252 [74].

Electronic supplementary material is available online [75].

Authors' contributions

A.E.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, visualization, writing—original draft, writing—review and editing; D.M.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, validation, visualization, writing—review and editing; B.G.: conceptualization, methodology, project administration, resources, supervision, validation, writing—review and editing; R.I.: formal analysis, methodology, project administration, resources, supervision, validation, writing—review and editing; L.K.B.: conceptualization, data curation, funding acquisition, investigation, methodology, project administration, resources, software, supervision, validation, visualization, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed herein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by a Smithsonian Tropical Research Institute short-term fellowship (to A.E. and D.M.) and the National Science Foundation (grant no. DEB-1136640 to L.K.B.).