Changes to Skin Microbiome Community Dynamics Support Resistance to Enzootic and Epizootic Strains of Batrachochytrium dendrobatidis (Bd) in the Spotted Salamander (Ambystoma maculatum)

ABSTRACT

Global spread of animal pathogens has contributed to species declines and extinctions. In regions where a particular disease is enzootic, pathogen inhibition may arise through protection provided by host‐associated microbiomes. Amphibians skin microbiomes can inhibit growth of the fungal pathogen Batrachochytrium dendrobatidis (Bd), preventing emergence of disease through a range of microbe‐mediated antifungal mechanisms, allowing hosts to resist Bd infection. However, it remains unclear how skin microbiomes may shift in community composition or structure following infection by different Bd strain types. We assessed infection dynamics of Bd‐resistant amphibians ( Ambystoma maculatum ) following experimental exposure to enzootic and epizootic strains of Bd‐GPL and tracking pathogen load and bacterial skin microbiome community responses from exposure through to recovery, using 16S rRNA metabarcoding. We found that microbiome communities shifted post‐exposure, with increasing diversity, dominance, abundance and total proportion of known Bd‐inhibitory microbes, indicating microbial rescue effects during infection. We also observed lower intra‐host variation in diversity during recovery, indicating a shared functional response across the host population and broadly indicative of microbial community resilience. Salamanders exposed to enzootic Bd had greater pathogen loads over time and demonstrated more prolonged community changes and more putatively protective microbiomes, whereas epizootic Bd infection was more rapidly cleared following temporary increase in inhibitory microbes. Collectively, these results indicate that skin microbiomes may offer a crucial barrier to fungal disease in Bd‐resistant amphibians, with exposure to pathogens inducing changes in microbial community structure that benefit hosts, possibly driven by localized coevolutionary changes in infection dynamics. Our work illustrates how complex host‐pathogen interactions are mediated by skin microbiomes through changes in microbial community dynamics that favour pathogen resistant microbes.

1. Introduction

The global spread of animal pathogens has dramatically influenced our understanding of the impacts of disease on populations, communities and ecosystems (Cunningham et al. 2017). As disease epizootics progress to enzootic stages, understanding emerging mechanisms of pathogen resistance is essential to provide deeper insight into infection dynamics in susceptible populations (Savage and Zamudio 2016). Host‐associated microbiomes, including skin and gut microbiomes, are an important component of pathogen resistance (Williams et al. 2018), and there is increasing recognition that microbial interactions during infection can serve a protective role for the host (Jin Song et al. 2019). Broadly, microbiomes are often indicative of host health (Bravo et al. 2022), and provide essential functions to host development (McFall‐Ngai et al. 2013), physiology (Kohl and Carey 2016) and immunology (Woodhams et al. 2023). Host‐associated microbiomes can determine how organisms respond to infection though microbial rescue, by recruitment or compositional changes in microbe communities that drive beneficial resistance functions and limit the impact of pathogens on the host (Mueller et al. 2020). Protective microbiome responses can be especially important during resistance to infection, when anti‐pathogen microbes become more abundant and increase production of pathogen‐inhibitory secondary metabolites (Siomko et al. 2023). The host microbiome can thus act as analogous to acquired immunity (Woodhams et al. 2023), by responding protectively against the pathogen and developing a resilient community that is increasingly responsive to future infection by the same or related pathogens (Zorgani and Das 2024). It follows that shifts in community structure and composition should reflect how microbiomes respond differently to infection (Costello et al. 2012), and if resistance to infection has emerged in disease enzootic regions, host‐associated microbiomes should respond with pathogen resistant community changes when host‐pathogen dynamics are locally adapted.

Microbial rescue effects during infection are usually indicated by increased prevalence, abundance or proportion of pathogen‐inhibitory microbes post‐infection (Mueller et al. 2020). In addition, a protective microbial response can also involve changes to community composition and structure between microbiomes during phases of infection, disease and recovery (Woodhams et al. 2023). For example, infection may disrupt microbiome communities by changing patterns of abundance of different genera through shifts in interspecific interactions, resulting in niche displacement, dominance, decline or even local extirpation (Costello et al. 2012). Disease may further disturb microbial communities to the point of major instability, resulting in microbial dysbiosis, where essential host‐related microbiome functions like immunity are limited or lost altogether (Petersen and Round 2014). In a given population of hosts subject to infection, microbiome instability and dysregulation should encourage greater dissimilarity between host microbiomes, increasing microbial dispersion between communities (Zaneveld et al. 2017). However, if microbiomes respond similarly to infection, community composition and structure can converge across hosts and reach a new equilibrium of disease resilience during recovery (Philippot et al. 2021). When protective microbial changes during infection mediate recovery by recruitment of the same beneficial microbes across hosts, microbiomes should become more similar between hosts, decreasing microbial dispersion (Woodhams et al. 2023). These protective responses may drive microbiome community dynamics when host and pathogen are locally adapted; for example, indicators of microbial resistance (Jiménez et al. 2022) and resilience (Ellison et al. 2021) to local pathogen strains arise in disease enzootic regions. However, communities resistant to one type of organism may remain vulnerable to related, but different organisms (Philippot et al. 2021), and epizootic pathogen outbreaks can be so disruptive that the microbiome fails to recover, resulting in a state of dysbiosis (Wynne et al. 2020). Therefore, understanding the range of host‐associated microbiome plasticity across different pathogen strains, from disturbance to resistance, is crucial for assessing the impacts of disease outbreaks, and determining if microbiomes are prominent mechanisms of resistance when host‐pathogen dynamics are locally adapted.

Batrachoytridum dendrobatidis (Bd) is a fungal pathogen (Longcore et al. 1999) with broad, worldwide distribution and has contributed to decline and extinction of a variety of amphibians (Scheele et al. 2019). Bd can occur as epizootic or enzootic (Lips 2016), with some populations or geographic regions known to be experiencing recovery from past epizootic outbreaks, implying emerging localized mechanisms of disease resistance or infection tolerance (Scheele et al. 2019). One mechanism of Bd resistance is the amphibian skin microbiome, where production of anti‐fungal secondary metabolites (Becker et al. 2009; Martin et al. 2019), anti‐fungal enzymes (Martínez‐Ugalde et al. 2024) and competitive exclusion of the pathogen (McLaren and Callahan 2020; Jones et al. 2024) can inhibit Bd growth. Resistance to Bd can arise when skin microbes inhibit Bd infection through collective action (Loudon et al. 2014; Woodhams et al. 2014), in union with the host's own innate immunity (Myers et al. 2012), and these components collectively comprise resistance functions of the amphibian skin mucosome (Bates et al. 2022). Thousands of Bd‐inhibitory skin bacteria have been identified in amphibian hosts (Woodhams et al. 2015), and profiling for these microbes has informed how microbiome composition can shape disease resistance, including community richness (Piovia‐Scott et al. 2017), total proportion (Chen et al. 2022) and community dominance (Walke et al. 2017) of resistant microbes. The microbiome community dynamics of amphibian hosts in natural populations provide evidence of ecological rescue effects during Bd infection (Mueller et al. 2020) and microbiome community resilience to disease (Woodhams et al. 2023). For example, in Bd‐enzootic amphibian populations there often is a higher proportion of Bd inhibitory microbes on amphibian skin compared to epizootic populations (Lam et al. 2010), and these host populations tend to have less variation in microbial community composition and structure (Ellison et al. 2019). These dynamics may indicate that host microbiomes respond functionally to past infections as a component of organismal plasticity, although such microbial community traits may also simply reflect phylogeographic variation (Osborne et al. 2024). Importantly, experimental work is needed to further elucidate the extent that amphibian host microbiomes shift functionally in response to Bd infection, and more broadly, if localized microbiome dynamics can explain disease resistance in natural populations. In contrast to resistance, susceptible amphibians experience disruption of the skin microbiome by Bd infection during disease epizootics (Jani and Briggs 2014; Becker et al. 2019), which are often driven by highly virulent strains (Piovia‐Scott et al. 2015). However, even resistant hosts who are asymptomatic can experience microbial dysbiosis during infection (Wasimuddin et al. 2018), but this has not been investigated in Bd‐resistant amphibians.

We assessed bacterial skin microbiome responses of spotted salamanders ( Ambystoma maculatum ) following experimental exposure to Bd strains of enzootic and epizootic origin. Ambystoma salamanders are a robust model for emerging Bd resistance in enzootic regions and are asymptomatic during inoculation experiments (Gahl et al. 2012), despite testing positive for Bd in the wild (Ouellet et al. 2005). The mucosomes of Ambystoma maculatum exhibit strong disease‐inhibitory responses (Barnhart et al. 2020), including known production of anti‐microbial peptides (AMPs) during infection (Pereira and Woodley 2021), but their microbial responses to Bd strain variation remain unknown. We predicted that skin microbes with known Bd‐inhibitory function will be more abundant and dominant in salamander microbiomes following Bd exposure, and that variation in microbiome responses should be strain‐specific, with greater protective microbial response following exposure to a local Bd strain and resistance to the pathogen. Accordingly, the microbiome community should experience more profound disruption following exposure to foreign Bd, and pathogen loads should be greater. We also predicted that Bd exposure would elicit increased variation between host microbiomes, but if recovery occurs, such variation should have diminished, supporting microbiome community resilience to infection.

2. Materials and Methods

2.1. Animal Collection and Husbandry

During May 2019, we collected ten A. maculatum egg masses from a pond near Buckhorn, ON Canada (44°34′54.8″N 78°19′27.6″W). We reared larvae in 80 L outdoor mesocosms through to metamorphosis and fed animals daily using brine shrimp (Artemia sp.). Mesocosms were supplied with 10% pH balanced Holtfreter's solution, and water changes were conducted once per week. Upon transformation, animals were transferred to a terrestrial environment pending the start of the experiment (approx. 60 days after metamorphosis). During the experiment, salamanders were housed in individual polypropylene enclosures (17 cm x 11 cm × 6 cm) and kept at 19°C with 12:12‐h light: dark cycles. Substrate consisted of unbleached paper towels moistened with pH balanced Holtfreter's solution. Salamanders received daily substrate changes, daily hydration, and were fed vitamin dusted domestic crickets ad lib twice per week. Salamanders were confirmed to be Bd‐negative with qPCR (see methods below) before the start of the experiment and were allowed to acclimate in individual containers for 2 weeks before inoculation. Salamanders from each egg mass were randomly assigned to one of three treatments: local Bd‐exposed, foreign Bd‐exposed and a sham‐exposed control group, with 28 salamanders per treatment group.

2.2. Bd Strain Culture and Inoculation

We exposed salamanders to two strains of the global panzootic lineage of Batrachochytrium dendrobatidis (Bd‐GPL) to document microbiome responses from exposure through to recovery. The local strain (JEL 261) was from Montreal, Canada (45.534327, −73.150072) and isolated from juvenile Lithobates catesbeianus in 1999 (Schloegel et al. 2012); the foreign strain (JEL 423) was isolated from Hylomantis lemur in El Cope, Panama (8.667146, −80.621410) during a Bd epizootic in 2004 (Lips et al. 2006). JEL 423 is known to have high virulence (DiRenzo et al. 2014), including in North American amphibian species (Gahl et al. 2012). These Bd strains represent sub‐populations of Bd‐GPL (local strain: Eastern North America, foreign strain: Panama; F_st = 0.202; Marshall et al. 2019). Stock cultures were prepared from thawed cryopreserved isolates from the CZEUM n.d. collection (Collection of Zoosporic Eufungi at the University of Michigan (CZEUM), Ann Arbor, Michigan, USA). Bd was cultured in 1% tryptone broth following standard protocols, and pure zoospores were harvested by flooding Bd cultured agar plates with sterilized diH₂0 to prepare Bd inoculum. We determined zoospore concentration with the Countess Automated Cell Counter (Invitrogen), ensuring an equal concentration of zoospores between strain inoculums. Before inoculation, we confirmed the viability of Bd by observing living and motile zoospores under the microscope. We inoculated salamanders by pipetting 1 × 10⁶ zoospores directly onto the ventral surface of each salamander, followed by a 24‐h bath exposure of the same pipetted inoculum in 50 mL of Holtfreter's solution. This volume ensured the ventral portion of salamanders remained submerged throughout the 24‐h bath exposure, and this dosage of Bd is known to cause mortality in susceptible species (Carey et al. 2006). Controls were sham inoculated with the same volume of sterilized diH₂0 and placed in a bath of Holtfreter's solution. After 24 h, salamanders were transferred back to their individual containers. We collected pathogen load and microbiome samples by swabbing salamanders with sterilized cotton‐tipped swabs and preserving swabs in 95% ethanol. We swabbed the ventral side and each limb ten times (Brem et al. 2007): (i) Post‐acclimation (Day 0); (ii) Post‐exposure (Day 1); (iii) Infection (Days 5,10, 20); and (iv) Recovery (Day 30). For the present study we sequenced salamander microbiomes from Day 0 (post‐acclimation), Day 5 (infection), and Day 30 (recovery). We selected a subset of 15 salamanders per group for microbiome analysis, prioritizing sequencing the microbiomes of salamanders who were positive for Bd infection, and then randomly selecting additional microbiomes. To account for temporal variation in microbiomes over the course of the experiment, we sampled 15 randomly selected untreated controls at Day 0 and Day 30. We also measured mass and snout‐vent length (SVL) of salamanders at Day 0, 10, 20 and 30. Salamanders were euthanized with MS‐222 after Day 30.

To add further ecological context to our experimental results, we also sampled the skin microbiomes of 18 wild adult A. maculatum salamanders found on land around the egg collection pond during spring breeding. All animal care procedures were approved and conducted in accordance with Trent University Animal Care Committee protocol #25635. Bd culturing and inoculation were conducted under Biosafety work permit #29065, and all Bd wastewater was disposed of by autoclaving.

2.3. Pathogen Load Quantification and 16S rRNA Amplicon Sequencing

DNA was extracted from preserved swabs with the QIAGEN DNeasy blood and tissue kit, with modified incubation protocols improving extraction yield (Adamowicz et al. 2014). We quantified Bd load using Taq‐man qPCR assay (Boyle et al. 2004), with samples run in triplicate using a 9‐point standard curve on each plate (0.1–10⁷ ITS copies) and gBlock gene fragments (IDT). Library preparation and sequencing were conducted at the Centre for the Analysis of Genome Evolution and Function (CAGEF), University of Toronto. We amplified the V4 hypervariable region of the 16S rRNA gene, using the primer pair 515F and 806R (Caporaso et al. 2012). Each sample's amplification was conducted in triplicate, and we confirmed amplification with electrophoresis. We quantified pooled triplicates using PicoGreen and standardized concentrations before final pooling. We purified the library with Ampure XP beads and sequenced using the Illumina MiSeq (V2 chemistry kit: 150 bp × 2). We also sequenced negative controls, including sterile swab extractions and PCR template‐free negative controls. We also included a single species positive control ( Pseudomonas aeruginosa ) and a mock community positive control (Zymo Microbial Community DNA Standard D6305).

2.4. Pathogen Load, Body Mass and Body Length Analysis

We assessed differences in pathogen load over time and between strains using a zero‐inflated generalized linear mixed model (GLMM) with a Tweedie distribution and a log link function (Jorgensen 1997). The GLMM model used strain and time as fixed effects and individual as a random effect to account for repeat sampling, with Day 1 and foreign strain as reference conditions. We also compared the rate of change in mass and SVL with linear mixed‐effects models (LMM) with time as a fixed effect and individual as a random effect.

2.5. Microbiome Sequence Processing and Analysis

We examined sequence quality with FastQC and MultiQC (Andrews 2010; Ewels et al. 2016), and used cut adapt (default settings) to remove sequencing errors (Martin 2011). We used vsearch‐fastq_mergepairs (Rognes et al. 2016) on default setting to quality trim and assemble paired‐end sequences. We then processed assembled sequences in Qiime2 version 2023.2 (Bolyen et al. 2019) with the quality‐filter function. We clustered quality filtered sequences into Amplicon Sequence Variants (ASVs) using the deblur pipeline of Qiime2, removing global singletons and chimeric sequences (File S1). We used classify‐hybrid‐vsearch‐sklearn in Qiime2 to assign ASV taxonomy (File S2), using the average ReadyToWear trained Silva database version 138.1 (Bokulich et al. 2018; Kaehler et al. 2019). We removed ASVs with an abundance < 0.01% and removed ASVs identified as chloroplast or mitochondria. We generated a phylogenetic tree of ASVs using the SEPP function in Qiime2. We conducted microbial composition and abundance analyses in R 4.3.2, using phyloseq (version 1.46; McMurdie and Holmes 2013) and used phyloseq to import and merge our ASV abundance table, sample metadata, microbial phylogeny and taxonomy. We filtered microbiomes with decontam (version 1.22; Davis et al. 2018) to remove environmental microbial contamination based on prevalence in negative controls (threshold = 0.5), and the frequency of ASVs whose abundancies are inversely correlated with total DNA concentration (threshold = 0.1). Furthermore, we removed all ASVs that were identified as Bd‐inhibitory and were present in negative extraction or PCR controls, so that assigning inhibitory function was only done on microbes found exclusively on salamanders. We assessed the completeness of community sampling on our final filtered microbiomes with Good's coverage estimator (Good 1953) as well as by producing rarefaction curves for each of our treatment groups using the rarecurve function from the R package Vegan (Oksanen et al. 2015). We assessed salamander microbiome function by identifying putative Bd‐inhibitory microbial symbionts using the Antifungal Isolates Database (version 2023.2; Woodhams et al. 2015). This database consists of approximately 3000 culturable microbial isolates with known Bd‐inhibitory function. We used a strict curation of the database, only identifying ASVs which exhibited inhibitory function against Bd. We used vsearch cluster‐features‐closed‐reference in qiime2 to query our filtered microbiomes against the known inhibitory sequence database, at 100% sequence similarity.

We normalized ASV abundance for community composition and structure analyses using cumulative sum scaling (CSS) with metagenomeSeq (version 1.43; Paulson et al. 2013), and compared microbiome communities by measuring beta diversity and community dispersion. We measured beta diversity and dispersion with weighted and unweighted Unifrac distances (Lozupone and Knight 2005). We used the resulting distance matrix to visualize microbial community differences using non‐metric multi‐dimensional scaling (NMDS) ordination, and tested differences in community composition (unweighted) and structure (weighted) at three time points, before infection (Day 0), during infection (Day 5), during recovery (Day 30), using permutational multivariate analysis of variance (PERMANOVA; Anderson 2017) on the distance matrix with time as a predictor for 9999 permutations, using the adonis function in Vegan (Oksanen et al. 2015). We assessed dispersion with BETADISP on the distance matrix, and used permutest to assess differences across time (Oksanen et al. 2015). For both PERMANOVA and BETADISP analyses we accounted for multiple pairwise comparisons across time (see Ellison et al. 2021) using the False Discovery Rate (FDR) method with Benjamini‐Hochberg correction (p < 0.05; Benjamini and Hochberg 1995). We ran similar analyses for beta diversity and dispersion in the control group, comparing between Day 1 and Day 30. We determined dominant taxa in microbial communities using > 1% of the mean total abundance as a threshold (see Campbell et al. 2011; Walke et al. 2017).

Rarefication in alpha diversity assessment of microbial communities can be controversial (Hong et al. 2022), with distinct trade‐offs between sequencing‐depth biases (Schloss 2024) and Type 1 error (McMurdie and Holmes 2014). We determined if rarefication was appropriate by including sequencing depth as a covariate in alpha diversity analyses (see Weiss et al. 2017; Muletz‐Wolz et al. 2021). Sequencing depth did not influence statistical outcomes in alpha diversity over time (Table S1), thereby justifying use of non‐rarefied data for alpha diversity estimates. We estimated differences in alpha diversity using observed species richness and the Shannon index across strain treatments and over time via a LMM, with strain and time as fixed effects and individual as a random effect.

We estimated differential abundance in ASVs between before infection and recovery to assess which ASVs respond to Bd infection, using a consensus approach (Nearing et al. 2022) with DESeq2 (version 1.42.1; Love et al. 2014) and ANCOM‐BC (version 2.4; Lin and Peddada 2020). Both DESeq2 and ANCOM‐BC are recommended for microbiome analysis (Calgaro et al. 2020; Nearing et al. 2022) and should balance sensitivity for detection with DESeq2, combined with the more conservative but lower‐power analysis of ANCOM‐BC (Nearing et al. 2022). For all analyses, we used microbiomes before infection as the reference condition, and used non‐rarefied data. We ran DESeq2 with the phyloseq_to_deseq2 function, using default setting. For size‐factor estimation we set to poscounts to account for ASVs that are missing in at least one sample. We ran ANCOM‐BC by making a TSE summarized experiment object from phyloseq and using the ancombc2 function in R, with default settings. We used a fold change cut‐off of > 2, and to control the false discovery rate we used Benjamini‐Hochberg adjusted p‐values for all analyses (p < 0.05). Following Siomko et al. (2023), we also estimated changes to proportional abundance of Bd‐inhibitory ASVs over time using a GLMM with a binomial probability distribution (inhibitory and non‐inhibitory) and logit link function, including time as a fixed effect and individual as a random effect to account for repeat sampling. We included an observation‐level random effect in all models to address overdispersion (Harrison 2015).

3. Results

3.1. Bd Infection Dynamics

All salamanders tested positive for Bd after the inoculation experiment, although infection loads were markedly low (Figure 1). Excluding Day 1, exposure to local Bd yielded higher pathogen loads over time compared to the foreign strain at all time points (all p < 0.001; Figure 1, Table S2). A majority of animals infected with the local strain (89.2%, n = 25) were still infected by Day 5 and mean infection loads increased through Day 10 (53.6% infected, n = 15) and Day 20 (39.3% infected, n = 11). By Day 30, 25% (n = 7) of salamanders exposed to the local strain remained infected, whereas salamanders exposed to the foreign strain rapidly reduced infection and pathogen load immediately from Day 1 to Day 5 and then remained at negligible levels for the remainder of the experiment (Figure 1). Most (71.4%, n = 20) salamanders exposed to the foreign strain cleared the infection by Day 5 and by Day 20, infection prevalence declined to 7.1% (n = 2), with only 3.5% (n = 1) remaining infected by Day 30 (Figure 1). All salamanders were asymptomatic during infection and generally did not grow (mass or SVL) slower than the control salamanders, although the rate of growth in mass for local‐exposed salamanders was marginally less than in the controls (Figure S1; Table S3).

FIGURE 1.

Mean pathogen load through time for Ambystoma maculatum exposed to 1 × 10⁶ zoospores of local and foreign strains of Bd (n = 28). Pathogen load is in units of ITS copy number from Bd qPCR. Shaded areas represent standard error. Percent values indicate the prevalence of individuals infected at each time point, within each strain.

3.2. Microbiome Community Composition and Structure

Skin microbiome sequencing produced a total of 12,280,088 paired end reads from our sequence library of 146 samples (n = 45 local, n = 45 foreign, n = 30 control, n = 18 wild, n = 4 positive control, n = 4 negative control). Excluding negative controls, sequencing averaged 44,341 reads per sample library (range: 32,744–155,077). After clustering sequences into ASVs and following all quality filtering steps to remove environmental contamination, we retained 2,098,292 sequence reads (17.1%) averaging 15,205 per sample (range: 2,019–53,413). Negative extraction and PCR controls retained an average of 201 reads (range: 8–590), and mock community controls were consistent with expected sequencing accuracy. For our microbiomes from lab reared salamanders, ASV clustering produced a total of 14,249 ASVs after environmental contamination filtering, with a mean of 269 ASVs per microbiome (range: 48–583). We retained 40 ASVs found in negative controls that were not identified as environmental contaminants but removed 25 Bd‐inhibitory ASVs from analysis for their presence in negative controls. Good's coverage scores ranged from 98.2% to 99%, and rarefaction curves reached asymptotes across all treatment groups (Figure S2), indicating sequencing depth was sufficient to capture community compositions (Table S4). Proteobacteria (Pseudomonadota), Bacteroidota and Verrucomicrobiota composed the most abundant bacterial phyla in salamander skin microbiomes (Figures S3–S5). Across the duration of the experiment, we identified a total of 129 putative Bd‐inhibitory ASVs in salamander microbiomes, with a mean of 17 per microbiome (range: 3–41). Both Bd‐inhibitory and non‐inhibitory microbes responded to Bd exposure across periods of infection and recovery.

Bd infection influenced community structure and composition of salamander microbiomes. Here, we focus on changes to community structure and have included compositional changes in supplemental material (Tables S5 and S6; Figures S6 and S7). Beta diversity was altered by Bd infection, and community structure differed between all treatments and sampling periods (Figure 2A,C; Table 1). Microbial dispersion also varied with Bd infection, with an overall pattern of infection initially increasing dispersion but dispersion decreasing by recovery (Figure 2B,D). These patterns were prominent in the local strain, where Bd infection increased dispersion, dispersion then decreased during recovery and microbiomes were less dispersed and more similar in structure during recovery than before infection (Table 1; Figure 2B,D). The foreign strain exhibited similar patterns of changing dispersion; however, dispersion only marginally increased during infection, and although dispersion decreased during recovery, it was not less in recovery than before infection (Table 1; Figure 2B,D). In the control group, community composition and structure did change over time without infection (Table 1; Figure 2E), however, these changes did not cause variation in structural dispersion (Table 1: Figure 2F).

FIGURE 2.

Changes in beta diversity (A, C, E) and dispersion (B, D, F) of A. maculatum microbiomes (n = 15) before infection (Day 0), during infection (Day 5) and during recovery (Day 30) in local Bd (A, B), foreign Bd (C, D), and controls (E, F), using NMDS ordination with weighted Unifrac distance.

TABLE 1.

Summary statistics for beta diversity (PERMANOVA) and microbial dispersion (BETADISPR) using weighted Unifrac distances.

Beta diversity (PERMANOVA)
Full models
Local Bd	R 2 = 0.19, F = 5.15, p < 0.001
Foreign Bd	R 2 = 0.18, F = 4.76, p < 0.001
Control	R 2 = 0.42, F = 20.6, p < 0.001

Pairwise comparisons (FDR corrected p‐values)
	Before vs. Infection	Infection vs. Recovery	Before vs. Recovery
Local Bd	R 2 = 0.07, F = 2.35, p = 0.009	R 2 = 0.14, F = 4.78, p < 0.001	R2 0.26=, F = 10.14 p < 0.001
Foreign Bd	R 2 = 0.08, F = 2.47, p = 0.010	R 2 = 0.11, F = 3.78, p < 0.001	R2 = 0.23, F = 8.72 p < 0.001

Dispersion (BETADISPR)
Full models
Local Bd	F = 12.01, p < 0.001
Foreign Bd	F = 3.69, p = 0.031
Control	F = 0.091, p = 0.755

Pairwise comparisons (FDR corrected p‐values)
	Before vs. Infection	Infection vs. Recovery	Before vs. Recovery
Local Bd	F = 6.09, p = 0.017	F = 19.26, p = 0.003	F = 10.78, p = 0.010
Foreign Bd	F = 3.78, p = 0.078	F = 6.22, p = 0.072	F = 0.62, p = 0.443

Note: For all pairwise comparisons, p‐values were corrected for False Discovery Rate (FDR) with Benjamini‐Hochberg correction.

3.3. Alpha Diversity

Because changes in observed species richness produced similar results as changes in the Shannon index (Figure S8; Table S7) we present alpha diversity as measured by the Shannon index. Microbiome alpha diversity was largely consistent from before infection to recovery for both Bd strains (Figure 3A). Irrespective of Bd strain, microbiome alpha diversity did not change during the transition to infection (all p > 0.88), whereas during recovery alpha diversity was marginally greater in the foreign strain treatment (t(56) = 1.99, p = 0.051). However, when only considering the assemblage of Bd‐inhibitory microbes, alpha diversity was influenced by Bd strain. Although unaltered by infection in the local strain (t(56) = 0.61, p = 0.545), alpha diversity of inhibitory ASVs increased during infection (t(56) = 2.07, p = 0.042) and recovery (t(56) = 2.29, p = 0.026; Figure 3B) in the foreign strain. Furthermore, alpha diversity of inhibitory ASVs was greater in the foreign strain during recovery than in the local strain (t(56) = 2.53, p = 0.014; Figure 3B). Control group alpha diversity was consistent through time (t(28) = −0.88, p = 0.385; Figure 3A), although alpha diversity of inhibitory microbes declined over time (t(28) = −2.34, p = 0.03; Figure 3B).

FIGURE 3.

Alpha diversity measured by the Shannon Index of A. maculatum microbiomes (n = 15) from before infection, during infection and recovery in local and foreign Bd strains, with controls sampled from Day 0 to Day 30. Panel A displays ASVs of the full filtered and non‐rarefied microbiomes, and panel B displays only the assemblage of putative Bd‐inhibitory ASVs. Asterisks display significant differences in alpha diversity.

3.4. Dominance

Dominance of microbiome community structure was influenced by Bd infection, with dominant microbial taxa shifting from before infection to recovery (Tables 2 and 3). Few microbes remained dominant in either Bd strain treatment, with shared dominance being primarily between before and during infection and novel dominance arising during recovery (Tables 2 and 3). While both strain treatments had a single dominant Bd inhibitory taxa before infection (Caulobacter), additional inhibitory microbes became dominant during infection and recovery (Tables 2 and 3). However, responses differed across Bd strains, with most Bd inhibitory dominance occurring during recovery in the local strain (Table 2) and during infection in the foreign strain (Table 3). In the local stain treatment, no additional inhibitory microbes became dominant during infection and Caulobacter was the only known inhibitory microbe that remained dominant through to recovery. However, during recovery the local strain elicited a shift in inhibitory community dominance, with Acinetobacter and bacteria from family Comamonadaceae (genus unknown; Table 2). In the foreign strain, Caulobacter retained dominance during infection only, and most newly dominant inhibitory taxa arose during infection, including Pseudomonas, Acinetobacter and family Comamonadaceae (genus unknown; Table 3). However, by recovery a single inhibitory taxa (Acinetobacter) remained dominant. Notably, Acinetobacter and Comamonadaceae are the same dominant ASV's observed during recovery in the local strain. None of the Bd‐inhibitory ASV's that became dominant in the treatment groups were dominant in controls at the end of the experiment, with the exception of Caulobacter (Table 4).

TABLE 2.

Dominant ASVs (> 1% total mean abundance) in local Bd salamander microbiomes from before infection, during infection and recovery.

Local Bd
Before infection			During infection			Recovery
ASV Genus	%	n	ASV Genus	%	n	ASV Genus	%	n
—			—			Acinetobacter	1.9	8
Caulobacter	9.3	14	Caulobacter	6.2	11	Caulobacter	17.7	9
Cellvibrio‐2	16.2	8	—			—
Cellvibrio‐3	2.4	1	—			—
—			—			Comamonadaceae‐F1	6.3	7
—			—			Comamonadaceae‐F3	5.6	7
Cytophaga	5.4	5	Cytophaga	16	5	—
Devosia‐1	5.3	14	Devosia‐1	7.4	15	Devosia‐1	4.1	11
—			—			Devosia‐3	4.2	6
Flavobacterium‐1	2.9	9	Flavobacterium‐1	1.6	7	—
Flavobacterium‐2	6.8	10	Flavobacterium‐2	4.8	9	Flavobacterium‐2	2.3	7
Flavobacterium‐3	1.7	3	Flavobacterium‐3	5.8	3	—
—			—			Flavobacterium‐4	1.9	11
—			Opitutaceae‐F1	2.6	2	—
—			Opitutaceae‐F2	1.7	3	—
—			—			Phenylobacterium	2.8	7
Prosthecobacter‐1	3.2	8	—			—
—			Prosthecobacter‐2	1.4	5	Prosthecobacter‐2	1.1	7
Sphingobacteriaceae‐F	1.2	4	Sphingobacteriaceae‐F	1.3	5	—	—
—			—			Spirosomaceae‐F	1.0	1
—			—			Taibaiella‐2	3.1	1
—			—			Tepidisphaera	3.3	10
—			—			Variovorax	2.6	10

Note: Each row represents the presence or absence of dominance of the same ASV over time. ASVs are delineated to the level of genus, except in the case genus is unknown, and ASVs are at the level of family (‐F). Multiple species in a single genus or family are numbered, and all taxonomic designations are the same between strain treatments. Precent (%) indicates the average proportional abundance of each dominant ASV, and n indicates the number of microbiomes in which each ASV displayed dominance (total n = 15). Bd‐inhibitory ASVs are in bold.

TABLE 3.

Dominant ASVs (> 1% total mean abundance) in foreign Bd salamander microbiomes from before infection, during infection and recovery.

Foreign Bd
Before infection			During infection			Recovery
ASV Genus	%	n	ASV Genus	%	n	ASV Genus	%	n
—			—			Abditibacterium	1.2	11
—			Acinetobacter	1.6	9	Acinetobacter	3.9	9
—			Allorhizobium‐1	1.2	10	Allorhizobium‐1	1.2	10
—			—			Allorhizobium‐2	3.9	10
Caulobacter	6.9	13	Caulobacter	4.5	6	—
Cellvibrio‐2	2.2	2	—			—
Cytophaga	3.3	7	—			—
—			Comamonadaceae‐F1	1.5	1	—
—			Comamonadaceae‐F2	1.2	5	Comamonadaceae‐F2	1.2	11
—			Comamonadaceae‐F3	1.8	3	—
—			—			Comamonadaceae‐F4	3.9	9
Devosia‐1	3.4	13	Devosia‐1	5.9	15	—
Devosia‐2	3.1	10	—			—
Flavobacterium‐2	5.6	12	Flavobacterium‐2	6.1	11	Flavobacterium‐2	1.2	12
Flavobacterium‐3	4.4	3	Flavobacterium‐3	4.8	3	—
—			—			Niveispirillum‐1	3.9	3
—			—			Niveispirillum‐2	1.2	4
Prosthecobacter‐1	3.4	13	—			—
—			Pseudomonas	1.3	10	—
Sphingobacteriaceae‐F	2.4	6	Sphingobacteriaceae‐F	1.2	4	—
—			Sphingomonas	5.0	2	—
Taibaiella‐1	1.0	4	Taibaiella‐1	1.5	2	—

Note: Each row represents the presence or absence of dominance of the same ASV over time. ASVs are delineated to the level of genus, except in the case genus is unknown, and ASVs are at the level of family (‐F). Multiple species in a single genus or family are numbered, and all taxonomic designations are the same between strain treatments. Percent (%) indicates the average proportional abundance of each dominant ASV, and n indicates the number of microbiomes in which each ASV displayed dominance (total n = 15). Bd‐inhibitory ASVs are in bold.

TABLE 4.

Dominant ASVs (> 1% total mean abundance) in control salamander microbiomes from before infection, during infection and recovery.

Control
Day 0			Day 30
ASV Genus	%	n	ASV Genus	%	n
Acinetobacter	11.4	15	—	—	—
—	—	—	Caulobacter	10.3	8
—	—	—	Cellvibrio‐1	2.6	3
—		—	Cellvibrio‐2	1.8	2
Chryseobacterium	4.8	8	—	—	—
—	—	—	Devosia	7.1	15
Comamonadaceae‐F	1.2	4	—	—	—
Dyadobacter‐1	1.6	6	—	—	—
Dyadobacter‐2	1.0	8	—	—	—
—	—	—	Emticicia‐1	2.7	5
—	—	—	Emticicia‐2	1.4	3
Flavobacterium‐1	20.8	14	—	—	—
—	—	—	Flavobacterium‐2	9.2	8
Fluviicola	1.3	2	—	—	—
—	—	—	Kapabacteriales	2.2	5
—	—	—	Moheibacter	1.2	5
—	—	—	Niveispirillum	13.0	11
Odoribacter	1.1	2
—	—	—	Opitutaceae‐F	2.4	5
Pedobacter‐1	1.5	8	—	—	—
Pedobacter‐2	1.7	8	—	—	—
Pedobacter‐3	2.5	7	—	—	—
Polynucleobacter	2.5	10	—	—	—
Pseudomonas‐1	7.7	13	—	—	—
Pseudomonas‐2	1.2	4	—	—	—
Pseudomonas‐3	1.2	11	—	—	—
Rosenbergiella	2.3	9	—	—	—
—	—	—	Sphingomonas	6.4	8
Taibaiella‐1	1.3	4	—	—	—
—	—	—	Tahibacter	1.5	2

Note: Each row represents the presence or absence of dominance of the same ASV over time. ASVs are delineated to the level of genus, except in the case genus is unknown, and ASVs are at the level of family (‐F). Multiple species in a single genus or family are numbered, and all taxonomic designations are the same between strain treatments. Precent (%) indicates the average proportional abundance of each dominant ASV, and n indicates the number of microbiomes in which each ASV displayed dominance (total n = 15). Bd‐inhibitory ASVs are in bold.

3.5. Differential Abundance of ASVs During Infection

Individual microbial abundance across different treatments responded to Bd infection either positively or negatively (Table 5; see Table S8 for ASV taxonomy). We also highlight the total proportion of inhibitory and non‐inhibitory ASVs in microbiomes to emphasize community‐level responses to infection (Figure 4; Table S9). For example, despite local Bd treatment showing fewer ASVs increasing in abundance from before infection to recovery, these changes resulted in a greater proportion of inhibitory taxa by recovery (z = 3.02, p = 0.002; Figure 4A,B). In contrast, more inhibitory ASVs increased in the foreign treatment, but the proportion of inhibitory taxa had only a small peak during infection and did not increase by recovery (z = −0.34, p = 0.73; Figure 4C,D).

TABLE 5.

Changes in ASV differential abundance from the consensus results of DESeq2 and ANCOM‐BC from before infection to recovery in Bd‐strain treatments and control at the level of genus.

	Local Bd		Foreign Bd		Control
Bd‐Inhibitory ASVs	+4	−0	+8	−0	+3	−17
Non‐Inhibitory ASVs	+31	−8	+25	−12	+44	−21

Shared differentially abundant ASVs
	Bd treatments		Local and control		Foreign and control		All treatments
Bd‐Inhibitory ASVs	+2	−0	+1	−0	+1	−0	+1	−0
Non‐Inhibitory ASVs	+4	−2	+10	−4	+7	−6	+2	−1

Note: ASVs are divided by Bd‐inhibitory and non‐inhibitory, and (+) indicates the number of ASVs which have increased in abundance; (−) indicates the number of ASVs which have decreased in abundance.

FIGURE 4.

Changes in differential abundance (A, C, E) and proportionality (B, D, F) of Bd‐inhibitory and non‐inhibitory ASVs from before infection (Day 0) to recovery (Day 30) in local (A, B) and foreign (C, D) strains of Bd as well as changes overtime in controls (E, F). Differential abundance is displayed as consensus identified DESeq2 fold‐changes of ASVs increasing or decreasing by Bd recovery, and proportionality is displayed as predicted proportions (+/− 95% confidence interval) of ASV sequence read abundance of inhibitory and non‐inhibitory ASVs across time. Percent values in panels B, D and F display the predicted proportions Bd‐inhibitory ASVs at each timepoint.

Changes in microbial abundance were largely strain‐specific, with few shared patterns between strains or the control treatment (Table 5). The majority of ASVs responding to infection were members of the phylum Proteobacteria, including all of the Bd‐inhibitory ASVs which increased in abundance in foreign and local exposed salamanders (Figure 5). Notably, no Bd‐inhibitory ASVs decreased in abundance from infection to recovery and decrease of inhibitory ASVs only occurred in controls (Table 5). Furthermore, inhibitory ASVs of the phylum Actinobacteriota also decreased in controls, in addition to Proteobacteria (Figure S9). Known Bd‐inhibitory ASVs, Pseudomonas and Neorhizobium, increased in both local and foreign strain treatments; however, Neorhizobium also increased in controls. Remaining patterns of increasing inhibitory ASVs were specific to each strain treatment and included inhibitory ASVs in the local (Klebsiella, Comamonas) and foreign strain (Bosea, Stenotrophomonas, Sphingobium, two species from genus Brevundimonas, and an additional species from Pseudomonas). In the controls, changes in ASV abundance over time exhibited opposite patterns from Bd treatments, with two Pseudomonas species decreasing in controls despite increasing in treated animals. Two known inhibitory ASVs (Variovorax and Caulobacter) uniquely increased in abundance among controls, and notably seventeen inhibitory ASVs became less abundant with time, whereas no known Bd‐inhibitory ASV decreased in abundance in the treatments (Figure 4; Table S7). All inhibitory ASVs which decreased in abundance in controls were present in local and foreign exposed microbiomes, even if their abundance was unchanged by infection. Apart from Pseudomonas, inhibitory ASVs with high proportional abundance or dominance in controls at the outset of the experiment did not increase in abundance in the treatment groups when exposed to Bd (Table 4; Table S8).

FIGURE 5.

Changes in differential abundance of Bd‐inhibitory and non‐inhibitory ASVs from before infection to recovery in local (A) and foreign (B) strains of Bd at the level of Phylum. Differential abundance is displayed as consensus identified DESeq2 fold‐changes of ASVs increasing or decreasing by Bd recovery. Bd‐inhibitory ASVs are indicated by asterisks.

3.6. Comparisons to Wild Salamander Microbiomes

We identified 15,348 ASVs in wild salamander microbiomes and found that wild microbiomes were also dominated by Proteobacteria, and Bacteroidota, included Verrucomicrobiota at a lower abundance than captive microbiomes, and Actinobacteriota at a greater abundance. We found that species richness was greater than in experimental animals (mean ASVs: 1403 per individual, range: 596–2583), however, inhibitory richness was more comparable between wild and lab animals, with 124 ASVs known to be Bd‐inhibitory and a mean of 38 ASVs per microbiome (range: 25–60). Wild salamander microbiomes had a mean proportional Bd‐inhibitory abundance of 11.3%. Roughly 54% of Bd‐inhibitory microbes detected during our experiment were also present on wild salamanders, and despite compositional dissimilarity, several known Bd‐inhibitory taxa were present in both groups. For example, despite low similarity in dominance of inhibitory communities between the wild and the lab, all dominant Bd‐inhibitory ASVs in the lab were found in wild salamander microbiomes, and of the known Bd‐inhibitory ASVs which increased in abundance, two ASVs in the local strain (Neorhizobium and Klebsiella) and six in the foreign strain (Neorhizobium, Bosea, both Brevundimonas, Pseudomonas and Stenotrophomonas) were also present in wild microbiomes. We also observed 56% of wild A. maculatum were positive for Bd, although at very low pathogen loads (mean copy‐number = 1.2, range: 0.25–6.6).

4. Discussion

Our experiments show that A. maculatum , a salamander known for its resistance to chytridiomycosis (Barnhart et al. 2020; Barnhart‐McCarty et al. 2024), experienced putatively protective community changes in its skin bacterial microbiome during experimental infection with enzootic and epizootic Bd. Notably, salamanders exposed to a local Bd strain generated the most protective and resilient microbiome communities. However, contrary to our prediction, foreign Bd‐exposed microbiomes also exhibited short‐term protective community changes and stronger resistance to infection and did not exhibit dysbiosis. We found that hosts shared Bd‐inhibitory bacterial taxa across both strain treatments but differed in the timing and specific constitution of their post‐treatment microbiomes. Our results provide new insight showing that changes to microbiome community dynamics may be driven by localized co‐evolutionary processes. Anti‐pathogen host‐associated microbiomes may represent a common mechanism of improved resistance to infection and disease (Mieog et al. 2009; Lemieux‐Labonté et al. 2017; Hill et al. 2018), and our work strengthens the emerging understanding of host‐associated microbiome dynamics by demonstrating that changing skin microbiome communities are an important component of organismal plasticity during infection.

Pathogens that have a widespread distribution may exhibit different phenotypes across their geographic range (Forsythe et al. 2018; Parker et al. 2023), and these phenotypes may influence infectivity and virulence (Fisher et al. 2009). Typically, local host populations experience lower pathogen virulence when infected with enzootic lineages, leading to lower levels of disease and mortality (Belasen et al. 2022). These disease dynamics arise when hosts evolve resistance mechanisms that limit infection (Roy and Kirchner 2000), and/or when pathogens evolve lower virulence to enhance survival and transmission (Gandon et al. 2002). These complementary forces can drive coevolutionary relationships, allowing hosts and pathogens to co‐exist (Best and Kerr 2000; Buckingham and Ashby 2022). For example, enzootic Bd can exhibit reduced virulence though smaller zoosporangium size (Lambertini et al. 2016) and slower growth or reproduction (Stevenson et al. 2013), which may be less disruptive to the amphibian skin and therefore allow pathogens to evade the host's corresponding immune response (Grogan et al. 2023). Accordingly, we observed more persistent low‐load Bd infections in salamanders exposed to the local strain, suggesting that resistant amphibians can better withstand local pathogens (see also Fu and Waldman 2019). While this response may be mediated by lower virulence, it is notable that salamanders exposed to the local strain still exhibited infection resistance, and correspondingly we observed emergence of anti‐Bd microbiomes over the course of the experiment. Infection may drive these microbiome community responses through changes in pathogen‐microbe competition or else via host‐microbe immune interactions, both of which promote host resistance (McLaren and Callahan 2020; Woodhams et al. 2023).

The induced higher total proportion and dominance of Bd‐inhibitory microbes following exposure to local Bd align with expected community‐level changes in pathogen‐inhibitory microbes in enzootic populations of Bd and related pathogens (e.g., Lam et al. 2010; Lemieux‐Labonté et al. 2017). Increased abundance of resistant microbes (i.e., total proportion and dominance) should amplify production of anti‐fungal metabolites (Walke et al. 2017), therefore, local‐exposed skin microbiomes should have increased resistance functions even despite greater pathogen abundance (see Jiménez et al. 2022). Further, emergence of strong resilience to local Bd during recovery reflects microbiome community dynamics in disease‐enzootic populations (e.g., Ellison et al. 2019; Gao et al. 2021), where a trend toward reduced community dispersion suggests a shared protective response across the host population, as a feature of locally‐adapted host‐pathogen systems. We propose that when enzootic pathogens have lower virulence (e.g., Fisher et al. 2009; Lambertini et al. 2016) and evade more rapid resistance mechanisms, skin microbiome defences may importantly contribute to infection regulation. Our results suggest that resistance via the skin microbiome may be particularly important to infection dynamics in Bd‐enzootic regions (Nava‐González et al. 2021), but these mechanisms may also broadly regulate fungal‐infection resistance when locally adapted host‐pathogen systems are not strongly driven by acquired immunity (Field et al. 2015; Lemieux‐Labonté et al. 2017). Indeed, clearing infection though such resistance strategies (Wilber et al. 2024) may drive host‐pathogen dynamics that are frequently observed in enzootic pathogen regions, where resistant hosts exhibit periodic clearance and re‐infection (Briggs et al. 2010).

In disease‐susceptible populations, virulent pathogens can cause disturbance to host‐associated microbiomes (Wasimuddin et al. 2018; Wynne et al. 2020). For example, Bd‐susceptible amphibians do not experience protective microbial responses during lethal infections (Becker et al. 2019), and skin microbiomes of susceptible species can be disrupted when infections are worse in epizootic populations (Jani et al. 2017). However, our work highlights putatively protective changes in the skin microbiome during the early phases of infection by an epizootic strain in resistant hosts who were asymptomatic to infection. Further, although microbiome resilience was comparatively weaker following exposure to the foreign strain, community dispersion did not increase into the recovery phase, indicating lack of a prolonged disturbance or dysbiosis (Zaneveld et al. 2017). We also observed unexpectedly rapid decline in infection following exposure to the foreign strain, suggesting resistance to epizootic Bd was stronger than to enzootic Bd. Likewise, following near total recovery, most protective microbiome responses faded, except for inhibitory diversity which remained high. Therefore, at later stages of infection there appeared to be a lessened pressure for a microbial community domination by Bd‐inhibitory bacteria, allowing the microbiome to return to a reduced inhibitory composition (see Le Sage et al. 2021). Rapid and effective pathogen resistance may require robust skin defences following exposure (Paludan et al. 2021), and such constitutive defences may be more effective when virulent pathogens are not adapted to evade the host's immune response (McDonald et al. 2023). Further, it is understood that strongly‐resistant amphibians often rely on a variety of constitutive defences from the skin (Grogan et al. 2023) rather than via inducible immunity per se (Poorten and Rosenblum 2016), implying that the rapid resistance we observed in A. maculatum when exposed to the foreign Bd pathogen is likely mediated by constitutive mechanisms working in tandem with changes to the microbiome. These additional defences include skin anti‐microbial peptides (AMPs, Jiménez et al. 2022), which are secreted from granular glands when infection disrupts the amphibian skin (Rollins‐Smith and Conlon 2005) and can be especially effective when A. maculatum is exposed to different pathogens (Barnhart et al. 2020; Pereira and Woodley 2021). Regardless, the higher inhibitory dominance and diversity following foreign Bd exposure implies that microbiome community changes were integral to host infection resistance. Microbiome diversity may prevent or limit pathogen invasion by reducing available niche space (Spragge et al. 2023). However, we note that increases in inhibitory diversity were greater than increases in overall bacterial diversity, and microbiome diversity did not increase following local Bd exposure. Therefore, we contend that targeted increases in the diversity of inhibitory bacteria could signal interactions between the skin microbiome and host‐specific responses to more virulent pathogens. Indeed, AMP production may enhance the growth of Bd‐inhibitory bacteria (Flechas et al. 2019; Woodhams et al. 2020), and these interactions may potentially drive the observed increase in inhibitory diversity. We suggest that during rapid resistance to foreign pathogen exposure, protective changes to the microbiome may be an integrative and commensurate component of constitutive resistance.

Amphibian skin microbiomes are often characterized by the presence of bacterial phyla Actinobacteriota, Bacteroidota, Firmicutes, Proteobacteria and Verrucomicrobiota (Kueneman et al. 2014; Mutnale et al. 2021; Woodhams et al. 2023), and these taxonomic groups can play important functional roles promoting host skin homeostasis and defence (Rebollar et al. 2018). Our work demonstrates ASVs from these groups were the most respondent to infection through changes in abundance, with the exception of Actinobacteriota, remaining unchanged by infection, but varying in abundance overtime when salamanders were unexposed to Bd. Bacterial taxa from Proteobacteria were the most responsive to infection, primary increasing in abundance, and members of this phylum are known to produce enzymes capable of degrading chitin, a key structural component of fungal cell walls (Martínez‐Ugalde et al. 2024), as well as a range of anti‐fungal secondary metabolites (Woodhams et al. 2018; Wax et al. 2023). Thus, broader taxonomic changes in response to infection may have supported functional shifts in pathogen resistance beyond ASVs identified as inhibitory. Bacterial ASVs known to inhibit Bd were also responsive to infection, and several genera shared patterns of increasing dominance or abundance across both strain treatments, including Acinetobacter, Comamonadaceae and Pseudomonas, all being common symbionts of salamander skin (Goodwin et al. 2022; Jiménez et al. 2022; Brooks et al. 2023). In particular, bacterial species from genus, Acinetobacter and Pseudomonas are known to be important and widespread mediators of infection resistance (Becker et al. 2015; Lemieux‐Labonté et al. 2017; Rebollar et al. 2018). Strain‐specific responses were observed in several known inhibitory genera that increased in abundance following treatment: In the local‐Bd treatment, Klebsiella (Jervis et al. 2021) and Comamonas (Bates et al. 2022); in the foreign‐Bd treatment, Stenotrophomonas (Muletz‐Wolz et al. 2017; Antwis and Harrison 2018) and two species of Brevundimonas (Bates et al. 2022). Note that these taxa, as well as all dominant Bd‐inhibitory taxa, were also present in skin microbiomes from wild A. maculatum captured locally, suggesting these bacterial sequence variants represent true symbionts or commensals of the spotted salamander. Previous microbiome research on wild A. maculatum revealed a low abundance of Bd‐inhibitory microbes and limited anti‐fungal function of skin microbiomes when salamanders are uninfected with Bd (Urrutia‐Carter et al. 2025). However, our work in Bd‐positive populations demonstrates that Bd‐inhibitory ASVs found on wild A. maculatum proliferate during experimental exposure to Bd. We infer these contrasting findings in A. maculatum are consistent because they support plasticity of the skin microbiome (Kolodny and Schulenburg 2020), where anti‐pathogen community compositions and functions emerge only in response to the stressor of infection, and are absent or diminished when not induced. Importantly, we acknowledge that our methods provide only an estimation of microbiome function, and that we have not measured Bd‐inhibition directly. Future work should prioritize more direct estimations of microbiome function at a community level by measuring gene expression of the host and microbiome (Westermann and Vogel 2021), skin microbiome metabolite profiles (Bates et al. 2022), and the production of AMPs by the host during infection (Jiménez et al. 2022). These combined measurements may better reveal mechanisms of resistance (González‐Serrano et al. 2025) as well as important interactions between microbes (Rebollar et al. 2016), microbe‐pathogen interactions (Torres‐Sánchez and Longo 2022) and host‐mediation of microbiome composition and structure though immune factors (Kubinak et al. 2015).

We housed salamanders in isolated soil‐free enclosures, and therefore our experimental design provides evidence that protective changes to microbiome communities can occur independently through intra‐host mechanisms and are not wholly reliant on microbial exchange between individuals or the environment (Woodhams et al. 2023). Our experiment also demonstrates that microbiomes from Bd‐resistant hosts can generate anti‐pathogen responses naturally within skin microbiomes and without prophylaxis (Siomko et al. 2023) or probiotics (Bletz et al. 2013). Changes to control microbiomes in our experiment also provide important context for the relevance of different Bd‐inhibitory taxa. Many putative inhibitory taxa found on controls were unresponsive to Bd‐treatments, and even high inhibitory compositions at the outset of the experiment in controls decreased overtime. Bd‐inhibitory members in non‐infected microbiomes may indicate the multifunctionality of these taxa (González‐Serrano et al. 2025), and these results demonstrate that inoculation experiments are useful for discerning which putative Bd‐inhibitory taxa actually respond to Bd infection. Finally, by tracking changes to community dispersion between hosts, we observed temporary dysregulation of the microbiomes during infection (Zaneveld et al. 2017), but that by recovery, microbiome resilience (i.e., ‘ecological resilience’, see Philippot et al. 2021; Woodhams et al. 2023) emerged together with microbiome community‐level shifts in key anti‐pathogen bacteria. Therefore, convergence of microbiome communities during recovery may represent restored homeostasis and regulation (Zaneveld et al. 2017) while also reflecting shared community changes from common recruitment of beneficial microbes. Taken together, we infer that emergence of microbiome community resilience alongside changes supporting pathogen inhibitory functions show that these interactions likely confer mutual benefits between microbes and host (McLaren and Callahan 2020). We conclude our experiments demonstrate that these mechanisms act against infection as a component of organismal plasticity (Kolodny and Schulenburg 2020). Microbial rescue effects during our experiment show that a set of inhibitory microbes can respond ubiquitously during infection by different pathogen strains. However, the numerous strain‐specific responses suggest that microbiome plasticity is predominantly sensitive to the conditions of infection, driven by host‐pathogen adaptation. In disease enzootic regions, host‐associated microbiomes may therefore become important mediators of improved resistance to infection.

Author Contributions

T.W.C.: Study design, data collection, data analysis, original draft and writing. C.J.K.: review and editing. D.L.: review and editing, D.L.M.: review and editing.

Funding

Funding for this research was provided by a Natural Sciences and Engineering Council Canada Discovery Grant to DLM.

Ethics Statement

All animal care procedures were approved and conducted in accordance with Trent University Animal Care Committee protocol #25635. Bd culturing and inoculation was conducted under Biosafety work permit #29065, and all Bd wastewater was disposed of by autoclaving.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

File S1: Representative sequences for each ASV. ASVs were resolved using deblur in QIIME2 version 2023.2, following quality filtering and removal of global singletons and chimeric sequences.

File S2: ASV taxonomic assignments. Taxonomy was assigned with classify‐hybrid‐vsearch‐sklearn in Qiime2 using the average ReadyToWear trained Silva database version 138.1.

Table S1: LMM model results for changes in alpha diversity overtime. We assessed differences in the Shannon index across treatments using time and strain as fixed effects and individual as a random effect. Model included show changes in alpha diversity for the full bacterial microbiome as well as the assemblage of only inhibitory ASVs, with both local (261) and foreign (423) as reference conditions. We ran the Control LMM with time as a fixed effect and individual as a random effect. An additional model was included with read count as a fixed effect, demonstrating this model does not change statistical significance. p‐values and degrees of freedom were generated using the Satterthwaite approximation.

Table S2: GLMM model results for differences in pathogen load overtime. We used a zero‐inflated Tweedie distribution GLMM including strain and time as fixed effects and individual as a random effect, with Day 1 and foreign strain as reference conditions. Residual degrees of freedom for all fixed effects = 266.

Figure S1: Rates of growth over time for salamander mass (A) and SVL (B). Rates of change are slopes from LMM, including time as a numeric fixed effect representing days, with 95% confidence intervals.

Table S3: LMM model results for Mass and SVL. Time and treatment were included as fixed effects and individual was included as a random effect to account for repeat sampling. p‐values and degrees of freedom were generated using the Satterthwaite approximation, and pairwise comparisons were adjusted using the Tukey method for multiple testing. Degrees of freedom for all fixed effects = 246. In both models control salamanders and Day 0 are set as the references.

Figure S2: Rarefaction curves of species richness for salamander microbiomes by treatment (n = 15). Curves were generated for each sample by randomly subsampling sequencing reads and calculation number of observed ASVs at increasing sequencing depth, using a step size of 1000 reads.

Table S4: Depth of sequencing and coverage of microbiome community for each sample using Good's coverage estimator. Read counts are generated from the total bacterial microbiome after filtering of environmental contamination. In the sample codes, HP = Day 0, H5 = Day 5, H30 = Day 30.

Figure S3: Changes in taxonomic composition at the level of phylum over time for local‐exposed salamander microbiomes. Stacked bar plots indicate mean relative abundance of each phylum across salamander microbiomes (n = 15). The top ten most abundant phyla are depicted, and reaming low‐abundance phyla have been relegated to ‘Other’.

Figure S4: Changes in taxonomic composition at the level of phylum over time for foreign‐exposed salamander microbiomes. Stacked bar plots indicate mean relative abundance of each phylum across salamander microbiomes (n = 15). The top ten most abundant phyla are depicted, and reaming low‐abundance phyla have been relegated to ‘Other’.

Figure S5: Changes in taxonomic composition at the level of phylum over time for control salamander microbiomes. Stacked bar plots indicate mean relative abundance of each phylum across salamander microbiomes (n = 15). The top ten most abundant phyla are depicted, and reaming low‐abundance phyla have been relegated to ‘Other’.

Figure S6: Changes in Beta diversity (A,C) and dispersion (B,D) of A. maculatum microbiomes (n = 15) before infection, during infection and during recovery in local (A,B) and foreign (C,D) Bd strains, using NMDS ordination with unweighted Unifrac distance.

Table S5: Summary statistics for beta diversity (PERMANOVA), and microbial dispersion (BETADISPR) using unweighted Unifrac distances.

Figure S7: Changes in beta diversity (A) and dispersion (B) of A. maculatum microbiomes (n = 15) from Day 0 to Day 30 in control salamanders, using NMDS ordination with unweighted Unifrac distance.

Table S6: Control microbiomes summary statistics for beta diversity (PERMANOVA), and microbial dispersion (BETADISPR) using unweighted Unifrac distances.

Figure S8: Observed species richness of A. maculatum microbiomes (n = 15) from before infection, during infection and recovery in local and foreign Bd strains, with controls sampled from Day 0 to Day 30. Panel A displays ASVs of the full filtered and non‐rarefied microbiomes, and panel B display only the assemblage of putative Bd‐inhibitory ASVs. Asterisks display significant differences in alpha diversity.

Table S7: LMM model results for changes in observed species richness overtime. We assessed differences in species richness across treatments using time and strain as fixed effects and individual as a random effect. Model included show changes in species richness for the full bacterial microbiome as well as the assemblage of only inhibitory ASVs, with foreign (423) as reference conditions. We ran the Control LMM with time as a fixed effect and individual as a random effect.

Table S8: Changes in bacterial ASV differential abundance from the consensus results of DESeq2 and ANCOM‐BC from before infection to recovery. Fold‐change values are displayed from DESeq2. ASVs are delineated to the level of genus, except in the case genus is unknown, and ASVs are at the level of family (_F) or order (_O).

Table S9: GLMM model results comparing the proportional abundance of inhibitory ASVs overtime using a binomial probability distribution and logit link function, including time as a fixed effect and individual as a random effect to account for repeat sampling, and an observation‐level random effect in all models to address overdispersion. Residual degrees of freedom = 40.

Figure S9: Changes in differential abundance of Bd‐inhibitory and non‐inhibitory ASVs from before infection to recovery in control microbiomes at the level of Phylum. Differential abundance is displayed as consensus identified DESeq2 fold‐changes of ASVs increasing or decreasing by Bd recovery. Bd‐inhibitory ASVs are indicated by asterisks.

Data Availability Statement

16S rRNA sequence data has been archived to the GenBank Sequence Read Archive database, BioProject # PRJNA1354488.