We collected 40 metamorphosed adult newts (alpine newts, Ichthyosaura alpestris, ALP: 10 males and 10 females; palmate newts, Lissotriton helveticus, PAL: 10 males and 10 females) in the same pond in the Larzac plateau, France, on May 28, 2018. The pond is permanent, has a natural substrate, is highly vegetated and is surrounded by agricultural lands (crops) and a few trees. All animals were immediately transported to the laboratory in individual containers. Upon arrival in the animal housing facilities, the newts were randomly placed in intraspecific heterosexual pairs into 135 l aquaria (60 × 60 cm, 50 cm water depth; Figure 1A) filled with dechlorinated water (aerated for 48 h to evaporate chlorine) and oxygenated with air pumps. Photoperiod (14 h light, 325 lux) and temperature (mean ± SE = 16.5 ± 0.7°C) reflected environmental conditions from the capture location. Partial water changes, oxygen concentration measures and titration of reactive nitrogen forms were conducted regularly to keep the water quality constant. Each pair of newts was fed every 2 days at 18:00, with 400 mg of defrosted bloodworms (Ocean Nutrition, Dartmouth, Canada). The food was frozen to reduce chances of transmission of exogenous bacteria and zoonotic agents reported in live bloodworms (Rouf and Rigney, 1993; Broza and Halpern, 2001; Moore et al., 2003).

After 4 months of this first aquatic phase, the newts were placed in individual plastic containers (23 × 15 × 6 cm; Figure 1C) filled with a cork substratum and kept in a dark, refrigerated, incubator gradually set to 4°C to create a 3-months long terrestrial overwintering period. The newts were not fed in order to comply with their natural metabolism during this phase, but were sprayed 3 times with dechlorinated tap-water (30 cm distance from the newts) every 2 days to keep their skin moist. The incubator’s temperature was progressively increased before the start of the second aquatic phase, which was conducted in the exact same conditions as the first one, and lasted 3 months. The same pairs of individuals from the first aquatic phase were reunited (Figure 1B). Co-housing the newts during their aquatic phases enabled them to express mating behavior, which naturally occurs in the wild and is likely necessary to their welfare. Nevertheless, our study species can overwinter alone in the wild, so we separated them during the overwintering phase.

The equipment used for capture, transport and maintenance of the newts was thoroughly washed and disinfected before and after use with a 3% VIRKON solution (Van Rooij et al., 2017).

The skin microbiota of the newts was sampled upon capture in the field, and monthly throughout captivity (11 sampling events per newt; Figure 1A). Samples collection consisted of a non-invasive skin swabbing. Each individual was held with a new pair of nitrile gloves and was gently rubbed with two sterile swabs (MW100 rayon tipped dry swab, MWE, Corsham, UK) as follows: 10 strikes back-and-forth on the ventrum, five on each side of the tail, five on each side of the back, five rolls on each hand and foot. The swabs were preserved dry, on ice upon collection in the field, at -25°C at the field station, and at-80°C after being transported to the laboratory, until further processing. One swab was used for the microbiota analysis and the other was kept for long-term archiving in our laboratory.

Despite our biosecurity measures, exogenous bacteria could be introduced through the water poured into the aquaria or through the cork used as substrate in the overwintering containers. To identify these environmental bacteria, swabs were monthly stirred 20 times in the water of control aquaria during aquatic phases, and rubbed 10 times on the cork of overwintering control containers during the terrestrial phase. These control aquaria (n = 3) and containers (n = 3) did not contain any newts but were maintained in the same conditions as the ones that did.

DNA was extracted from the swabs using the DNeasy PowerSoil Pro kit (QIAGEN, Hilden, Germany), following the manufacturer’s instructions and including non-template controls (NTCs). Following the protocol outlined in Harrison et al. (2019), library preparation (using 515F and 806R primers) and community amplicon sequencing of the hypervariable V4 region of the 16S ribosomal RNA gene (~254 bp) were conducted on a MiSeq system (Illumina, San Diego, California, USA), at a depth of 30,000 reads. Demultiplexed sequences were processed using DADA2 v.1.8 (Callahan et al., 2016). Forward and reverse reads were truncated at decreasing quality (respectively 240 and 150 bp), and chimeric Amplicon Sequence Variants (ASVs) were removed by reconstruction against more abundant parent ASVs. Taxonomy was assigned to representative sequences using a naive Bayesian classifier implemented in QIIME 2, trained against EMPO 3 “animal surface” habitat-specific taxonomic weights (Kaehler et al., 2019). Assignments were accepted above a 0.7 confidence threshold. To identify symbiotic phylotypes with known inhibitory activity against Bd, representative sequences were aligned to the Antifungal Isolates Database (Woodhams et al., 2015) in QIIME 2.

Preprocessing of the sequences was carried out using the R package phyloseq (McMurdie and Holmes, 2013). Only bacterial sequences were kept, and 13 contaminant ASVs identified from NTCs were removed using the R package decontam (Davis et al., 2018). To address the uneven depth of coverage, all samples were normalized by rarefaction without replacement (Cameron et al., 2021) at 22140 reads. ASVs with no taxonomic affiliation at the phylum level, and spurious ASVs making up less than 0.005% of the total reads, were filtered out from the data (Bokulich et al., 2013). The final dataset was comprised of 12,255 ASVs, across a total of 436 newt samples.

Differences in alpha (within-sample diversity) and beta (among-samples dissimilarity) diversity between species and phases were investigated using statistical tests of similar structure (described below). Alpha diversity was quantified using Chao1 (estimated ASV richness) and Shannon (estimated ASV evenness) indices. Factors potentially influencing these indices were included in linear and linear mixed models (described below), and tested through analyses of variance (ANOVA). In cases where the residuals of the models did not meet the assumptions of normality and homoscedasticity associated with ANOVAs, a log-transformation of the response variable was successful at resolving these assumptions. Pairwise contrasts between phases were tested using estimated marginal means. Beta diversity was quantified using the weighted Unifrac distance (phylogenetic distance weighted by species abundance information) to investigate differences in community structure among samples (Lozupone et al., 2011). Permutational multivariate analyses of variance (PERMANOVAs) implemented using the adonis function (n = 9,999 permutations) were used to test similar models as for the alpha diversity. Pairwise differences between phases were tested using a pairwise adonis test. Differences in within-group variation in community structure (i.e., differences in compositional variance of microbiota) were investigated using betadisper tests. To identify bacterial taxa responsible for the observed differences in community structure among samples, differential abundance analyses were completed on unrarefied data (436 newt samples, 14,185 taxa) using DEseq2 (Love et al., 2014).

Initial differences in microbiota diversity between wild samples were investigated using a model that included species, sex, their interaction and individual snout-vent length (SVL) as fixed effects. For both alpha and beta diversity indices, this model showed that sex and SVL (mean ± SE = 4.19 ± 0.08 cm) had no effect on the diversity of the microbiota, therefore these variables were not included in subsequent models for parsimony purposes. A second model restricted on data from the two first sampling events (i.e., in the wild and after 1 month ex-situ) (Figure 1A) was built to investigate short-term changes in diversity over the transition from the wild to captivity. It included species, time of sampling, and their interaction as fixed effects, and individual identity as a random intercept. Lastly, the effect of phase-shifts on the microbiota was investigated using the full dataset, through a model including species, phase and their interaction as fixed effects. That model included individual identity, aquarium identity, as well as month of sampling as random effects. Estimates associated with the covariates in all models were deemed significant if associated with p-values below a 0.05 threshold. If interaction terms were not statistically significant, models were rebuilt without them (Beck and Bliwise, 2014).

All analyses were conducted in the R environment v.4.1.0 (R Core Team, 2022). Shared ASVs between species and phases were visualized using Venn diagrams created in the R package ggVennDiagram (Gao et al., 2021). Variation in beta diversity was visualized using Principal Coordinates Analyses (PCoA), built using the R package vegan (Oksanen et al., 2020). Other graphical representations were plotted using the R packages ggplot2 (Wickham, 2016) and ggpubr (Kassambara, 2019). All data and code are publicly available at Figshare repository.¹

In both alpine and palmate newts, the skin microbiota of wild individuals was dominated by Proteobacteria, Verrucomicrobiota and Bacteroidota (Figure 2). The microbiota of palmate newts was more diverse (Shannon, F_(1,34) = 5.42, p = 0.026) (Figure 3A), comprised more ASVs (Supplementary Figure S1) and tended to be richer than that of alpine newts (Chao1, F_(1,34) = 2.84, p = 0.101) (Figure 3B). It also comprised more Bd-inhibitory phylotypes (Figure 4; Supplementary Figure S1). In both species, Bd-inhibitory phylotypes belonged to Proteobacteria, Actinobacteriota, Bacteroidota or Firmicutes phyla (Figure 4). The alpha diversity of the microbiota of wild newts was not significantly determined by their sex (Chao1, F_(1,34) = 0.34, p = 0.563; Shannon, F_(1,34) = 0.12 p = 0.735), nor by their SVL (Chao1, F_(1,34) = 0.16 p = 0.696; Shannon, F_(1,34) = 0.87, p = 0.356).

In the wild, the beta diversity of skin bacterial communities differed between species (F_(1,20) = 19.06, p < 0.001) and sexes (F_(1,20) = 12.03, p < 0.001), which explained 29 and 18% of the variation in community structure, respectively (Supplementary Figure S2), but was not significantly affected by the SVL of the newts (F_(1,20) = 0.95, p = 0.570). The compositional variance of microbiota did not significantly differ between species (F_(1,36) = 0.75, p = 0.397) nor between sexes (F_(1,36) = 0.16, p = 0.697). However, three ASVs, of which two Comamonadaceae and one Bacteroidales, had significantly greater relative abundance in palmate newts compared to alpine newts. One of these Comamonadaceae phylotypes has known Bd-inhibitory activity according to the Antifungal Isolates Database (Supplementary Table 1).

One month after their transfer into captivity, the alpha diversity of the newts’ skin microbiota had significantly decreased (Chao1, F_(1,39) = 288.04, p < 0.001; Shannon, F_(1,39) = 12.41, p = 0.001). Differences between newt species in evenness of their bacterial communities persisted (Shannon, F_(1,39) = 5.20, p = 0.028) while their richness remained comparable (Chao1, F_(1,38) = 3.66, p = 0.063) (Figure 3). The transfer into captivity also led to a 10-fold decrease in the total number of ASVs in both newt populations despite the acquisition of new taxa; after 1 month ex-situ, these new taxa outnumbered the ASVs retained from the wild microbiota (Supplementary Figure S3). The majority of these new bacteria in the microbiota of captive newts were absent from the water filling their aquaria, and most of the exogenous phylotypes introduced through this medium did not colonize their skin microbiota (Supplementary Figure S4A). More Bd-inhibitory phylotypes were lost than acquired during the transfer (ALP, 58 lost vs. 12 new; PAL, 74 lost vs. 8 new), with only 9 and 11 protective phylotypes retained in the populations of alpine and palmate newts after 1 month ex-situ, respectively (Supplementary Figure S3). Consequently, the species differences in relative abundance of Bd-inhibitory phylotypes observed in the wild (ALP, 1.4%; PAL, 10%) decreased in captivity (ALP, 3.4%; PAL, 4.0%) (Figure 4).

The beta diversity of the newts’ skin bacterial communities significantly changed after their transfer into captivity (F_(1,36) = 22.66, p < 0.001), and was differentially affected by that transfer among newt species (F_(1,36) = 3.19, p = 0.034). While 19% of the variation in community structure among samples from the wild and the first month in captivity was explained by this environmental change, 18% was still due to significant differences between species (F_(1,36) = 21.53, p < 0.001) (Figure 5A). The structure of the microbiota was not significantly different between individuals (F_(38,36) = 0.87, p = 0.770) and its compositional variance was homogenous across species (F_(1,76) = 0.08, p = 0.776) and between wild and captive newts (F_(1,76) = 0.81, p = 0.373). The transfer into captivity led to rapid changes in the relative abundance of ASVs belonging to 12 and 13 different bacterial phyla, in alpine and palmate newts, respectively. In both species, the relative abundance of Proteobacteria increased while the proportion of Bacteroidota and Verrucomicrobiota decreased (Figure 2). More specifically in alpine newts, 151 ASVs (of which eight Bd-inhibitory taxa) had a significantly greater relative abundance in wild individuals, while 44 (of which two Bd-inhibitory taxa) were more abundant after 1 month in captivity. In palmate newts, 218 phylotypes (of which 16 Bd-inhibitory taxa) had greater relative abundance in the wild, and 49 (of which three Bd-inhibitory taxa) were more abundant in captivity (Figure 6A). Overall, 117 of the decreasing ASVs and 20 of the increasing phylotypes were common to both species, and were similarly affected by this transition from the wild to captivity in this relatively short timeframe (Supplementary Tables S2, S3).

The comparison of samples from the wild and after 10 months of captivity revealed an even stronger effect of long-term captivity, with significant changes in the relative abundance of 189 and 269 ASVS, across 12 and 14 phyla in the microbiota of alpine and palmate newts, respectively. Overall, 175 phylotypes decreased in abundance in the microbiota of alpine newts, while 14 increased. In palmate newts, 253 phylotypes were significantly less abundant after 10 months in captivity, while 16 had increased in abundance. All of these increased phylotypes remain untested against Bd. Among ASVs that significantly decreased in abundance, 16 and 10 had known Bd-inhibitory activity in alpine and palmate newts, respectively (Figure 6B; Supplementary Tables S11, S12).

Phase-shifts caused significant changes in the alpha diversity of microbial communities (Chao1, F_(3,7) = 19.55, p < 0.001; Shannon, F_(3,7) = 8.03, p = 0.011); it was generally high in the wild, decreased through captive aquatic phases, and increased during the overwintering (Figure 3). Overall, the diversity of the microbiota was not significantly different between host species (Chao1, F_(1,18) = 0.52, p = 0.482; Shannon, F_(1,23) = 1.04, p = 0.318). Microbiota were richer in wild individuals compared to any captive period (Supplementary Table S4), but their evenness was not uniformly affected by phase-shifts between the two species (Shannon, F_(3,403) = 9.05, p < 0.001) (Supplementary Table S5).

The beta diversity of bacterial communities significantly differed between aquatic and terrestrial phases, with phase-shifts explaining over 51% of the variation in community structure among samples (F_(3,390) = 179.85, p < 0.001) (Figure 5B). Across the 10 months of captivity, the beta diversity of the newts’ microbiota also significantly differed between species (explaining 3.9% of the variation; F_(1,390) = 41.43, p < 0.001), between individuals (F_(38,390) = 1.32, p = 0.014), and was differentially affected by phase-shifts among species (F_(3,390) = 9.70, p < 0.001). In both alpine and palmate newts, each phase was associated with distinct community structures (Supplementary Table S6) although Proteobacteria remained dominant throughout the 10 months of monitoring (Figure 2). Moreover, the interindividual variance in microbiota structure was not homogenous among phases and was significantly higher during the overwintering phase (F_(3,432) = 10.90, p < 0.001). In both species, the initially large proportion of Verrucomicrobiota in samples from the first aquatic phase decreased suddenly over the transition to the overwintering period. Conversely, Bacteroidota rapidly increased in proportion, as well as Actinobacteriota and Patescibacteria after a longer latency. These two latter phyla were present in negligible abundance on the cork from the overwintering containers (Figure 2), and only seven of the new phylotypes that colonized the newts’ microbiota during the overwintering may have come from that material (Supplementary Figure S4B). More precisely, differential abundance analyses revealed that the transition from the first aquatic phase to the overwintering period was coupled with significant changes in the abundance of 193 ASVs across 13 bacterial phyla in alpine newts, and of 191 ASVs across 14 different bacterial phyla in palmate newts. In both species, the number of phylotypes that significantly increased and decreased in abundance over this phase shift was similar. However, twice as many Bd-inhibitory phylotypes increased than decreased in abundance over this shift (Figure 7A; Supplementary Tables 7–8), and overwintering was therefore associated with higher abundance of Bd-inhibitory ASVs in both species (Figure 4). During the transition from the overwintering to the second aquatic phase, 107 and 108 phylotypes significantly decreased in abundance, while 37 and 48 ASVs increased in abundance in alpine and palmate newts, respectively. Bd-inhibitory phylotypes were affected similarly, and therefore decreased in abundance over that phase-shift, although this transition was more marked in alpine than palmate newts (Figure 4). The same phyla as in the shift from the first aquatic phase to the overwintering phase were involved, with the exception of an ASV from the Nitrospirota phylum (Figure 7B; Supplementary Tables S9, S10).

The dominant phyla identified within our samples are typically present in amphibian skin microbiota (Kueneman et al., 2014; Bletz et al., 2016; García-Sánchez et al., 2022). Despite being phylogenetically close (Zhang et al., 2008; Recuero et al., 2014) and inhabiting the same pond in the wild, alpine and palmate newts had distinct skin bacterial communities. Species-specific variation in microbiota composition is often reported in cohabiting amphibians (Kueneman et al., 2014; Walke et al., 2014) and the natural differences in community structure between species could explain their different susceptibility to skin diseases. Indeed, previous research suggests that palmate newts may be tolerant to Bd, while this pathogen can be lethal to alpine newts (Cheatsazan et al., 2013; Miaud et al., 2016). Our results show that in the wild, palmate newts had a larger number of different bacterial phylotypes, a higher alpha diversity, and a higher proportion of Bd-inhibitory bacteria than alpine newts. Moreover, two Comamonadaceae phylotypes were significantly more abundant in palmate newts; several isolates in this family inhibit Bd in-vitro (Walke et al., 2015) and one of the ASVs identified here has known Bd-inhibitory activity (Woodhams et al., 2015). Differences in abundance of these phylotypes should be tested in other populations to confirm whether they explain the distinct susceptibilities of alpine and palmate newts to Bd.

Despite initial differences in composition and diversity of their bacterial communities in the wild, both host species were similarly affected by their transfer into captivity. After only 1 month ex-situ, the newts had a different microbiota structure than in the wild. Studies comparing wild and captive-reared amphibians (Becker et al., 2014; Sabino-Pinto et al., 2016; Kueneman et al., 2016b) or other vertebrates (Dallas and Warne, 2022) similarly report distinct microbial communities between individuals living in-and ex-situ. The diversity of the microbiota of both alpine and palmate newts strongly decreased upon arrival in captivity, corroborating findings from shorter-term captivity studies (Bates et al., 2019) and comparisons of wild amphibians to conspecific individuals maintained ex-situ for several generations (Sabino-Pinto et al., 2016; Passos et al., 2018). Interestingly, the proportion of Bd-inhibitory phylotypes was not specifically affected by captivity and while a few new protective bacteria were acquired ex-situ, they did not become significantly abundant over the 10 months of captivity. Similar results were reported in red-backed salamanders, which conserve constant proportions of the symbiont Janthinobacterium lividum when transferred into captivity (Loudon et al., 2014). Conversely, the proportion of Proteobacteria in the microbiota of the newts strongly increased after only 1 month in captivity. Elevated abundance of this ubiquitous bacterial phylum (Kersters et al., 2006) is reported in other studies comparing captive amphibians to wild conspecifics (Bataille et al., 2016; Bates et al., 2019).

Overall, little of the wild microbiota was retained through the transfer into captivity, and the microbial community of captive newts comprised a reduced number of different phylotypes than when they were in the wild. This corroborates findings that the microbiota of amphibians maintained in a biosecure collection for several generations was less rich than that of wild individuals (Passos et al., 2018). Interestingly, most phylotypes present in the microbiota of the newts after 1 month in captivity were new and of unknown origin, despite our biosecurity protocol. Although the water filling their aquaria may have introduced up to 83 exogenous bacteria, most phylotypes present in the water did not colonize their microbiota. Indeed, despite being largely assembled from environmental reservoirs of bacteria in the wild (Walke et al., 2014), the amphibian skin microbiota may not select for most microbes present ex-situ. This is supported by findings that the microbiota of African clawed frogs reared under standard and sterile husbandry protocols have similar diversity indices (Piccinni et al., 2021). However, it should be noted that swab-sampling of the water and cork as we did is too restrictive to fully identify environmental microbiota, so the comparison with the microbiota of the newts is indicative rather than comprehensive.

Seasonal shifts in the structure of skin microbiota are reported in wild amphibians (Longo et al., 2015; Tong et al., 2020; Douglas et al., 2021) but to our knowledge, they have never been explored under artificial settings ex-situ. We found that although alpine and palmate newts maintained distinguishable skin bacterial communities throughout the 10 months of monitoring, they were similarly affected by the transfer into captivity and the artificial phase-shifts. Their alpha diversity, which had strongly diminished upon arrival in captivity, significantly increased during overwintering. In the wild, this phase has been associated with various effects on the alpha diversity of skin microbiota depending on host species (Longo et al., 2015; Tong et al., 2020; Douglas et al., 2021). Despite reestablishing similar levels of alpha diversity as in the wild, overwintering was associated with a very different microbial community structure and a much higher compositional variance than other phases. The segregation of microbiota structures was stronger between phases than it was between newt species, suggesting a similar and strong effect of artificial phase-shifts on the beta diversity of both species. Indeed, terrestrial and freshwater ecosystems, occupied during overwintering and active phases respectively, are generally associated with distinct microbial environments (Thompson et al., 2017). Moreover, seasonal variation in skin structure and feeding activity in amphibians can indirectly affect their microbiota, as diet (Antwis et al., 2014), sloughing (Meyer et al., 2012) and skin morphology (Kueneman et al., 2014) are known to influence their skin bacterial communities. Lastly, interactions among microbes also likely participated in shaping microbiota throughout phase-shifts, although this could not be further explored in this study.

Interestingly, while a few studies comparing different species in-and ex-situ have reported an among-individual uniformization of the microbiota in captivity (Becker et al., 2014; Hernández-Gómez et al., 2019; Edenborough et al., 2020), an individual signature in the structure of the microbiota persisted throughout phase-shifts in our experimental newts. Despite the significant decrease in alpha diversity and in abundance of many phylotypes over the 10 months of monitoring, the heterogeneity of bacterial communities between individuals was not reduced by captivity. Moreover, the variation in composition of the microbiota of captive newts was hardly influenced by exogenous bacteria in their ex-situ environment. For example, only seven of the phylotypes acquired in the overwintering phase were common to the cork substrate filling their containers. Moreover, changes in the relative abundance of major phyla in the microbiota of the newts throughout phase-shifts revealed a strong selection for bacterial taxa associated with each phase. We observed an increase in proportion of Bacteroidota, Actinobacteriota and of antifungal ASVs during the overwintering terrestrial phase, which are also reported in wild individuals during this period of elevated prevalence of Bd (Tong et al., 2020; Douglas et al., 2021; Le Sage et al., 2021).

Several authors encouraged the integration of microbiota to conservation plans developed for their hosts (Trevelline et al., 2019; West et al., 2019; Carthey et al., 2020), and this approach would be particularly relevant for amphibians given the essential role of their bacterial symbionts against emerging diseases (Vredenburg et al., 2011; Rebollar et al., 2020). Indeed, microbiota richness is associated with Bd inhibition (Piovia-Scott et al., 2017), and generally, highly diverse microbiota seem beneficial to the health of their host (Longo et al., 2015; Bates et al., 2018; Harrison et al., 2019). Yet, our results show that current management strategies such as ex-situ conservation and artificial habitat-shifts alter amphibian skin bacterial communities, which could consequently be detrimental to their host and reduce the success of reintroduction efforts. Functional analyses and transcriptomics should now be implemented to characterize the implications of these shifts in microbiota structure for their amphibian hosts.

Although bioaugmentation with probiotics can successfully limit diversity loss in microbiota of captive animals (Bletz et al., 2013; Woodhams et al., 2016), this strategy does not perpetuate exact replica of natural bacterial communities, and thus cannot prevent the elimination of phylotypes associated with specific metabolic pathways, which could affect their host functional resilience or put them at higher risk of infection by pathogens if reintroduced in the wild (Dallas and Warne, 2022). “Rewilding” the microbiota of captive animals by placing them in outdoor mesocosms before reintroduction in the wild may be a more promising solution, but should be investigated further (Kueneman et al., 2022). Considering the accelerating decline rate of amphibians (McCallum, 2007), it is critical to integrate microbiota to amphibian applied conservation and to continue to develop methods to maintain wild microbiota in captivity. More generally, the effect of conservation approaches on the microbiota of endangered species should be closely investigated, as they can be considered as an overlooked form of anthropogenic disturbance.