Differential skin-bacteriome-mediated defense against chytridiomycosis in two neotropical frog species

Abstract

Symbiotic microbial communities have been implicated in host resistance to pathogens, but their effects are rarely demonstrated experimentally in wildlife. This study tested how skin-associated bacterial communities (i.e., bacteriomes) influence infection by the chytrid fungus Batrachochytrium dendrobatidis (Bd) in two tropical frog species, Haddadus binotatus and Ischnocnema henselii, which differ in susceptibility to Bd. Using a 2x2 factorial experimental design, frogs of both species were assigned to treatments crossing Bd exposure and antibiotic-mediated bacteriome suppression. In parallel, we cultured 786 bacterial isolates from frog skin and assayed their ability to inhibit Bd in vitro, generating a functional database of Bd-inhibitory symbionts. Haddadus binotatus with an unsuppressed skin bacteriome and exposed to Bd showed no reduction in survival relative to Bd-unexposed controls, consistent with the lack of Bd infection previously observed in wild populations. In contrast, bacteriome suppression increased mortality and infection intensity under Bd exposure. Bd infection intensity in H. binotatus also decreased with the proportion of Bd-inhibitory sequence reads in the bacteriome, and a significant interaction between antibiotic treatment and Bd exposure affecting survival was detected, consistent with bacteriome-mediated protection in this species. In contrast, I. henselii experienced lower survival under Bd exposure in general. Species-specific log-rank tests revealed that bacteriome suppression significantly increased mortality under Bd exposure in H. binotatus but not in I. henselii, where survival was reduced under Bd exposure regardless of bacteriome state, suggesting fundamentally different defense strategies between species. Functional attributes of microbial communities, rather than diversity alone, appear to be key to disease outcomes. This work advances understanding of host–microbe–pathogen interactions and highlights microbiome function as a critical axis of wildlife disease defense.

Supplementary Information

The online version contains supplementary material available at 10.1186/s42523-026-00593-2.

Introduction

Early foundational studies in disease ecology theorized pathogen dynamics considering host populations as uniform and homogenous [1, 2]. Since then, the knowledge of pathogen dynamics has evolved and now considers multiple sources of heterogeneity, such as individual variation and species-specific adaptations, as influencers of host-pathogen interactions [3–5]. The chytrid fungus Batrachochytrium dendrobatidis (Bd) is a host-generalist skin pathogen of amphibians, with research indicating that hosts have multiple defense strategies [6]. For example, adaptations include behavioral avoidance of infection [7], tolerance to infection through genetic adaptations [8, 9], and mechanisms limiting pathogen growth and persistence, including the recruitment of pathogen-antagonist microbes [6]. In amphibians, one key player in these defenses is the skin microbiome: a diverse community of symbiotic microbes that can act as a first line of defense against pathogens such as Bd, working individually or in concert with the host immune system [10].

Taxonomic and functional diversity of the microbiome can often inhibit pathogen colonization, aligning with principles of competitive exclusion and invasion biology. For example, a more diverse microbial community can slow or inhibit the establishment of invading pathogens [11, 12]. A higher diversity of microbes on amphibian skin could also increase the production of Bd-inhibitory metabolites and promote synergistic interactions among distinct bacterial taxa [13–15]. Recent findings indicate that diverse communities on the skin may act as a direct barrier to Bd in addition to regulating the host immune system [16]. This action is thought to be achieved through the production of metabolites and signaling molecules that can modulate immune responses [17]. Additionally, repeated exposure to pathogens and their metabolic products has been linked to shifts in the amphibian skin microbiome, with changes linked to acquired resistance [18]. The role of the skin microbiome in amphibian defense is particularly evident when considering the functional specificity of bacterial assemblages (hereafter bacteriomes) in combating pathogens, such as the production of Bd-inhibitory metabolites [19].

Compounds produced by bacteria, such as violacein and prodigiosin, have been shown to inhibit chytrid fungi in vitro and in vivo [20]. This was first evidenced by Harris et al. [21] who demonstrated in vitro inhibition of Bd by bacterial isolates from the skin of salamanders. After the standardization of challenge assays for inferring Bd-inhibitory function of bacterial metabolites by Bell et al., amphibian disease ecologists joined forces to create an open-source functional database of Bd-inhibitory bacteria based on 16 S rRNA amplicon sequences [22, 23].

Brazil’s Atlantic Forest, a global biodiversity hotspot, experienced multiple amphibian declines during the late 1970s, which have since been mainly attributed to chytridiomycosis [24, 25]. More recent field studies have reported enzootic infection patterns across diverse taxa, with well-sampled species persisting with intermediate Bd prevalence and infection loads [26–29]. One striking exception is the leaf-litter frog Haddadus binotatus, a common forest-associated Atlantic Forest endemic. Despite extensive sampling across multiple Atlantic Forest landscapes, H. binotatus is rarely found infected with Bd, in stark contrast to Ischnocnema henselii [30], a closely related species (both within the superfamily Brachycephaloidea) that has similar reproductive ecology and habitat use [31, 32]. In a recent comprehensive field survey across 40 sites spanning over 450 km of Atlantic Forest [30], I. henselii had the highest reported Bd prevalence of all six species in the study (34.8%), while no H. binotatus tested positive despite the high sample size (0%, n = 126) and geographic range of sampling. Haddadus binotatus also exhibited the highest overall and core skin-associated bacterial diversity and possessed a unique bacteriome network structure, suggesting a more functionally active skin bacteriome that could contribute to disease defense [30]. This pair of closely related, ecologically similar species that differ in Bd infection dynamics offers a unique opportunity to explore bacteriome-mediated defense mechanisms against a globally distributed pathogen of high conservation concern.

Here, we experimentally tested how skin bacteriome diversity, composition, and predicted Bd-inhibitory function influenced disease susceptibility in H. binotatus and I. henselii under controlled laboratory conditions. We used antibiotics to suppress the bacteriome of both focal host species to test whether the inferred Bd resistance of H. binotatus reported in the previous field-based study [30] is mediated by the skin-associated bacteriome. Using a 2x2 factorial design crossing Bd exposure and antibiotic-treatment, we predicted that: (i) H. binotatus would only develop high Bd infection loads (and possibly show signs of chytridiomycosis) in Bd-exposed individuals treated with antibiotics; (ii) conversely, for I. henselii, Bd-infection loads would be unaffected by antibiotic treatment; (iii) the bacteriome of both amphibian species would vary in their degree of Bd-inhibitory function. During the experiment, we performed a series of in vitro bacterial isolation and Bd challenge assays, generating a comprehensive dataset on Bd-inhibitory capacity of 786 bacterial isolates from individuals frogs used in our experiment. We then mapped bacterial reads from swabs taken throughout the experiment to our functional isolate database to estimate the Bd-inhibitory function of the bacteriome. We predicted that (iv) H. binotatus would have a higher richness and proportion of bacterial reads associated with Bd-inhibitory functions compared to I. henselii. Finally, we expected that (v) Bd-inhibitory function, as estimated from our in vitro challenges, would be a stronger predictor of Bd susceptibility in H. binotatus than in I. henselii. While most studies have focused on the inhibitory activity of individual bacterial isolates in vitro, our experimental design evaluates the bacteriome as an integrated defensive system in vivo. Our framework expands the understanding of microbially mediated protection, revealing how community-level bacterial function contributes to amphibian resistance against Bd.

Experimental procedures

Animal collection

We collected study animals from three Atlantic Forest fragments in the municipality of São Luís do Paraitinga, in the state of São Paulo, Brazil (Fig. 1). Sampling took place over a three-day field campaign during the rainy, amphibian breeding season of 2023. We collected 40 individuals of Haddadus binotatus and 40 Ischnocnema henselii. We captured each individual frog with disposable nitrile gloves, rinsed the skin surface with autoclaved distilled water to remove transient microorganisms, and collected two skin swab samples. Swabs were taken from the ventral abdomen and inner thighs using 10 strokes per swab, following Boyle et al. (2007). The swabs across the experiment were taken sequentially from the same body regions: one for cryopreservation in 20% glycerol solution for downstream bacterial culturing, and one swab preserved dry for molecular analyses [33]. We kept vials on ice in the field and then placed them in a -80 °C freezer until processing. Frog transport, housing, and acclimation details are available in the Supplementary Methods.

Fig. 1.

Map showing collection sites for each individual in the three sampled fragments of Brazil’s Atlantic Forest

Experimental treatments and monitoring

Prior to treatment assignment, individuals of each species were stratified separately to ensure balanced representation across groups. For H. binotatus, stratification was based on body size and sex, and individuals were randomly assigned to treatments within these categories. For I. henselii, sex determination was not possible through external morphology; therefore, stratification was based on body size alone. This resulted in comparable group compositions across treatments, with mean SVL ranging from 39.4 to 41.4 mm and mean body mass from 5.1 to 6.0 g among H. binotatus individuals.

In the laboratory, we treated 20 individuals of each focal amphibian species with antibiotics to suppress their bacteriomes, following the protocol from Holden et al. [34] with minor modifications (see Supplementary Methods). Individual frogs treated with antibiotics (antibiotic+ treatment; n = 20 per species) were housed overnight (14 h) in sterile plastic containers with enough antibiotic solution to completely cover their legs. Individual frogs assigned to control groups (antibiotic- treatment; n = 20 per species) were placed in containers with sterile artificial pond water only, covering their legs for the same amount of time. Following this treatment, we left experimental animals undisturbed for 36 h to minimize stress and then swabbed them to quantify baseline skin-associated bacterial communities before the onset of the experiment.

Thirty-six hours after administering either antibiotic baths or control treatments, we exposed half of the study animals to a Bd inoculum and exposed the rest to a sterile control inoculum (see Supplementary Methods). Each frog in the Bd+ treatment was exposed to approximately 4.2 × 10⁶ Bd zoospores/mL (1 × 10⁷ zoospores total) in a 10 mL bath. Individuals were maintained in contact with the Bd or control solution in sterile plastic containers for 45 min before being returned to their terraria. This exposure level was selected to provide a controlled and sufficiently strong infection challenge to test host resistance under standardized conditions, rather than to directly mimic natural exposure levels, as is common in experimental Bd infection studies [35–37]. This created a 2⋅2 fully factorial design with four treatment groups with equal sampling size (n = 10) for each of our two focal species: antibiotic+ / Bd+, antibiotic+ / Bd-, antibiotic- / Bd+, and antibiotic- / Bd-.

Eight days after the experimental pathogen exposure, we swabbed all individuals using new gloves for each animal. We monitored frogs daily and if a dead frog was detected, we immediately swabbed and removed the individual from the experiment. Forty days following Bd exposure, we concluded the experiment and collected a final swab from the remaining animals. Immediately after the last swabbing event, we humanely euthanized the study animals using a bath of tricaine methane sulfonate (MS-222, 2 g/L, pH 7). All specimens were deposited at the Museu de Diversidade Biologica (MDBio), at UNICAMP in São Paulo, Brazil.

Molecular methods and bioinformatics

We extracted DNA from swabs using GMax Mini Genomic DNA Extraction Kits (IBI Scientific) following standard protocols with the addition of a lysis step using a lysozyme solution to improve extraction of gram-positive bacterial DNA (see Supplementary Methods). Next, we added the standard amount of Proteinase K and incubated the swabs overnight at 60 °C.

To identify bacteria in our samples, we used PCR targeting the full-length 16 S rRNA region. DNA concentration and quality were assessed using a NanoDrop spectrophotometer (A260/A280 ratio). Extraction blanks (n = 4) and PCR no-template controls (n = 2 per plate) were included in all sequencing and qPCR runs. We then sequenced the samples using PacBio HiFi long-read sequencing with the Kinnex 16 S rRNA kit [38] at Penn State University Genomics Core Facility and received demultiplexed bacterial sequences. Amplicon Sequence Variants (ASVs) detected in extraction blanks were removed from all downstream analyses. Contamination levels were low (median < 0.01% of reads per plate), and removal of blank-associated ASVs did not qualitatively change our results. Details of sequence processing are available on the Supplementary Methods.

For analyses of alpha diversity, we calculated ASV richness for each sample. For beta diversity, we calculated unweighted UniFrac (UU) distances between samples and performed principal coordinates analysis (PCoA), using the first and second axes (PCo1 and PCo2) to summarize variation in community composition for downstream analyses. To quantify within-group variability in community composition, we calculated multivariate dispersion (betadisper function in the R package vegan [39]) based on UU distance matrices, with host species and treatment considered separately. Dispersion reflects the distance of individual samples to their group centroid in multivariate space and is used here as a proxy for increased stochasticity or instability in community structure [40], rather than as a direct measure of compositional differences among groups.

We then used TaqMan qPCR assay with Bd-specific primers to quantify Bd on swab samples [33] (see Supplementary Methods). We performed the qPCRs on an Applied Biosystems QuantStudio™ 3 system to estimate Bd infection loads (measured in ITS-gene copy number). We added a sevenfold dilution series of Bd plasmid standards (Pisces Molecular, Boulder, CO, USA) in each run. We transformed Bd loads to correct for non-normal residual distributions in downstream statistical analyses using the formula log10(ITS gene copies + 1). Each sample was run in singlicate. Each qPCR plate included a sevenfold dilution series of Bd-specific plasmid standards (Pisces Molecular) as positive controls, no-template controls as negative controls, and synthetic internal positive controls (IPCs) to assess potential PCR inhibition.

Isolate culturing and challenge assays

We isolated bacteria from glycerol-preserved swabs collected in field sampling to axenic culture using standard microbiological techniques [41]. To perform challenge assays against Bd, we extracted the metabolites from bacterial supernatants (see Supplementary Methods). We then conducted challenge assays in 96-well microplates, adapting the standardized method from Bell et al. [23]. Experimental wells contained 50 µL of Bd zoospores at a concentration of 2 × 10⁶ mL^− 1 and 50 µL of a bacterial cell-free extract. The positive control wells contained 50 µL Bd and 50 µL 1% Tryptone (T1%) liquid media. We included two types of negative controls: a (media-only) control, which consisted of wells containing 100 µL replicates of T1% alone, and a heat-killed control of 50 µL of heat-killed Bd zoospores (80 °C for 30 min) at a concentration of 2 × 10⁶ mL^− 1 in 50 µL of T1%. We added nutrient-depleted controls containing 50 µL of Bd zoospores at a concentration of 2 × 10⁶ mL^− 1 and 50 µL of artificial pond water instead of T1%. Each plate contained 16 bacterial supernatant samples in triplicate (experimental wells), and three wells of each control type. We incubated assay plates at 21 °C and recorded optical density of each well at 492 nM on days 0, 3, 5 and 8 using a Thermoscientific Varioskan Lux microplate reader, which was sufficient time for the growth curve to reach a plateau [42, 43]. To calculate Bd growth in each well we followed the established protocol described by Bell et al. [23]. Isolates were considered Bd-inhibitory when the average Bd growth in their extract experimental wells were equal or smaller than 50% of the growth observed in the positive control (Bd in T1%) in the respective assay plate. Sequencing of isolates and database generation protocols are available on the Supplementary Methods.

The in vivo experiment used isolate CLFT 278 (Bd-GPL/Bd-Brazil lineage), which is locally abundant and therefore appropriate for testing host responses in an ecologically relevant host–pathogen context. The in vitro inhibition assays used isolate CLFT 041 (Bd-GPL), a widely used reference strain in functional inhibition studies conducted over the years and available in the expanding AmphiBac reference database [22], enabling comparison with the existing reference database and the literature [23]. These choices reflect complementary goals: a standardized resistance challenge in vivo and comparability across functional assays in vitro.

Statistical analyses

We analyzed amphibian survival using Cox proportional hazards models implemented in R (package survival [44]). To test the effects of bacteriome disruption and pathogen exposure on host survival, we fitted models including species identity, antibiotic treatment, Bd exposure, and their interaction. Baseline bacteriome function was incorporated as the proportion of Bd-inhibitory sequence reads measured at the field sampling timepoint. Because some treatment groups exhibited near-complete survival or mortality, we avoided fitting highly parameterized species-specific interaction models and instead based inference on global models including frog species as a fixed effect. To evaluate whether infection dynamics predicted host survival, we fitted time-varying Cox proportional hazards models using a start–stop framework, in which Bd load (log10-transformed ITS copies + 1) was included as a time-varying covariate. Individual identity (frog ID) was included as a clustering term to account for repeated measures. To further interpret treatment-specific effects within host species, we conducted planned log-rank tests comparing survival between Bd-exposed frogs with and without antibiotic treatment, separately for each species.

We analyzed the effects of antibiotic-induced bacteriome suppression on Bd infection dynamics and bacterial community structure in H. binotatus and I. henselii. For each species, we used Gaussian generalized linear mixed-effects models (GLMMs) fitted with the glmmTMB [45] package in R. Models included treatment (Bd-only, antibiotic-only, co-inoculation, control), the ratio of Bd-inhibitory sequence reads, and sampling day as fixed effects. We also included an interaction between treatment and the ratio of Bd-inhibitory sequence reads and random intercepts for individual frogs to account for repeated measures. Model predictions were visualized with ggeffects [46] and ggplot2 [47]. We ran parallel GLMMs to test the effects of antibiotic treatment, Bd load, and their interaction on bacteriome community composition (unweighted UniFrac dispersion).

To compare bacterial ASV diversity metrics, proportion of Bd-inhibitory sequence reads, and the richness of Bd-inhibitory ASVs from field samples, we used Wilcoxon tests performed in JMP. Differentially abundant ASV taxa were identified using ANCOM-BC2 in R. Analyses were performed at the genus level after agglomerating ASVs. For field samples, models included host species as the predictor, and for experimental samples, treatment (Bd exposure and antibiotic treatment). P-values were adjusted using the Holm method, and taxa with adjusted q < 0.05 were considered statistically significant. To test links between taxonomic differences and antifungal function, representative ASV sequences (PacBio full-length 16 S) were compared to a curated database of cultured bacterial isolates (Sanger 16 S) with known Bd-inhibitory activity using local pairwise alignment. Because direct sequence matches were limited, functional interpretation was based primarily on sample-level metrics derived from read matching to the isolate database. We tested whether bacterial genus identity explained differences in the probability of Bd inhibition among isolates using logistic regression. Genera with fewer than ten isolates were collapsed into a single “other” category to avoid overparameterization. We fitted logistic regression models in R with genus (collapsed as necessary) as the predictor of Bd inhibition status. To visualize genus-specific probabilities of Bd inhibition, we generated predicted probabilities from the fitted models and plotted them for each genus using ggplot2 in R.

Results

Survival analysis

We observed strong treatment-dependent differences in survival between host species. In Haddadus binotatus (Fig. 2A), survival remained high under Bd exposure alone, but declined sharply when Bd exposure was combined with antibiotic treatment. In contrast, Ischnocnema henselii exhibited reduced survival under Bd exposure independently of bacteriome suppression with antibiotics, although mortality was highest in the combined Bd and antibiotic treatment.

Fig. 2.

Kaplan-Meier survival curves (A) showing a drop in survival for Batrachochytrium dendrobatidis (Bd)-exposed individuals with suppressed bacteriome (Bd infection + antibiotic treatment, purple) for Haddadus binotatus; and (B) for all Bd-exposed Ischnocnema henselii independently of bacteriome suppression

Cox proportional hazards models revealed significant differences in survival between host species, with I. henselii exhibiting higher mortality rate than H. binotatus (HR = 2.848, p = 0.012). We also detected a statistically significant interaction between antibiotic treatment and Bd exposure (HR = 11.290, p = 0.048), indicating that bacteriome disruption may be linked to reduced survival when individual frogs are exposed to Bd. Time-varying Cox models further revealed that higher Bd loads predicted a reduction in frog survival (HR = 2.212, p < 0.001), and frogs with an unsuppressed bacteriome (antibiotic-negative treatment) showed higher survival (HR = 0.372, p = 0.006). Planned log-rank tests comparing Bd-exposed frogs with and without antibiotic treatment revealed that bacteriome suppression significantly increased mortality in H. binotatus (χ² = 5.500, df = 1, p = 0.019) but not in I. henselii (χ² = 0.500, df = 1, p = 0.474), where mortality was high under Bd exposure regardless of bacteriome suppression.

Disentangling drivers of Bd infection intensity

At the time of capture in the field, we found only two Bd-infected individuals of H. binotatus out of 40 sampled individuals. For H. binotatus, mixed-effects models revealed that Bd infection intensity was not affected by antibiotic treatment (β = −0.082, t₃₈ = −0.18, p = 0.861), but was significantly associated with estimated bacteriome function. Specifically, higher proportions of putative Bd-inhibitory sequence reads were associated with lower Bd infection intensity (β = −0.97, t₁₀₂ = −2.32, p = 0.022; Fig. 3C). This indicates that Bd-inhibitory bacteriome function, rather than antibiotic treatment per se, predicts infection intensity in this species. In I. henselii, Bd infection intensity was not explained by antibiotic treatment (β = 0.20, t₃₇ = 0.32, p = 0.751). The proportion of Bd-inhibitory sequence reads showed a non-significant association with infection intensity (β = −0.62, t₉₂ = −1.76, p = 0.082; Fig. 3D), indicating weaker or inconsistent functional effects of the skin bacteriome in this species.

Fig. 3.

Batrachochytrium dendrobatidis (Bd) infection intensity and estimated Bd-inhibitory bacterial function over time and between treatments in Haddadus binotatus and Ischnocnema henselii. Panels show predicted infection intensity (log-transformed) over (a, b) sampling day and (c, d) Bd-inhibitory function, separated by species (H. binotatus a, c; I. henselii b, d). Points represent raw data jittered for clarity; colored lines and ribbons depict model predictions and 95% confidence intervals by treatment group

Bacterial isolation and culturing

Using glycerol-preserved swabs collected at the time of capture in the field, we cultured and isolated a total of 786 bacterial isolates. From the H. binotatus individuals in the experiment, we cultured 374 bacterial isolates from 139 species, 62 genera, 27 families, 13 orders, 7 classes and 4 phyla. Among the 412 isolates cultured from I. henselii individuals in the experiment, we identified 244 species, 87 genera, 35 families, 17 orders, 10 classes and 6 phyla (Fig. S2). Based on our screening of bacterial isolates for Bd inhibition, at the time of capture in the field, H. binotatus harbored a higher proportion of Bd-inhibitory isolates (43.2%) compared to I. henselii (36.3%), although this difference was not statistically significant (t = 0.421, n = 786, p = 0.673). The probability of bacterial isolates inhibiting Bd growth varied significantly across bacterial genera (likelihood ratio test: χ² = 44.509, df = 13, p < 0.001). Among genera represented by ten or more isolates, predicted probabilities of Bd inhibition ranged from 0.20 in Sphingobacterium (n = 14) to 0.93 in Stenotrophomonas (n = 34). Remarkably, Stenotrophomonas exhibited the highest association with Bd inhibition, with a logistic regression coefficient (β) of 2.56 ± 0.89 (p = 0.004), corresponding to an odds ratio of approximately 13, indicating a strong effect of this genus on antifungal activity in vitro. Overall, bacterial genus explained a significant portion of the variance in Bd-inhibitory potential, suggesting phylogenetic structuring of antifungal activity across the cultured skin bacteriome of both frog species (Fig. S3).

Bacteriome community structure and diversity

Prior to the onset of the experiment, H. binotatus and I. henselii skin bacteriomes showed similar patterns of total bacteriome richness (S = 1026, Z=-0772, p = 0.440; Fig. 4A) and community composition (PERMANOVA: F = 1.093, R² = 0.016, n = 70, p = 0.242; Fig. 4B). The proportion of Bd-inhibitory sequence reads per field sample also did not vary among the two focal host species (S = 992, Z = -1.207, p = 0.227; Fig. 4C). Interestingly, H. binotatus showed a higher richness of Bd-inhibitory ASVs based on our reference database from in vitro challenges (S = 952.5, Z = -1.708, p = 0.043; Fig. 4D).

Fig. 4.

Haddadus binotatus and Ischnocnema henselii have similar (A) richness of bacterial ASVs and (B) composition (represented as the first two principal coordinates axes generated through UU distance matrices) in skin bacteriomes from samples collected at the time of capture in the field. Comparison of Batrachochytrium dendrobatidis (Bd)-inhibitory bacterial community (C) richness and (D) proportion of Bd-inhibitory sequence reads between H. binotatus and I. henselii in field-collected samples. Asterisk indicates a statistically significant (p < 0.050) difference in richness of Bd-inhibitory ASVs between the two focal host species at the moment of capture in the field

Bacteriome dispersion responded differently to Bd infection intensity and antibiotic treatment between species. In H. binotatus, dispersion decreased significantly with increasing Bd infection intensity (β = −0.018, t = − 3.77, p < 0.001). Although antibiotic treatment alone had no significant effect on dispersion (β = −0.013, t = − 0.76, p = 0.451), we detected a strong positive interaction between antibiotic treatment and Bd infection intensity (β = 0.026, t = 3.39, p = 0.001), indicating that bacteriome suppression altered the relationship between infection intensity and bacterial community dispersion. In I. henselii, dispersion also declined with increasing Bd infection intensity (β = −0.015, t = − 2.82, p = 0.006). In contrast to H. binotatus, antibiotic treatment significantly reduced dispersion overall (β = −0.045, t = − 2.07, p = 0.046), but no interaction between Bd infection intensity and antibiotics was detected (β = 0.007, t = 1.04, p = 0.299).

ANCOM-BC2 identified multiple genera that differed significantly between host species in the field and across experimental treatments. In the field, most differentially abundant taxa were enriched in I. henselii, including named genera such as Cutibacterium, Flavobacterium, and Sphingobacterium, as well as unclassified or candidate genera (e.g., JC017, OMJO01). In contrast, fewer taxa were enriched in H. binotatus, including Xiphinematobacter, Phenylobacterium, Actinoallomurus, and Moraxella (Fig. 5).

Fig. 5.

Differentially abundant bacterial genera between host species in the field. ANCOM-BC2 identified bacterial genera enriched in Ischnocnema henselii (positive log-fold change) and Haddadus binotatus (negative log-fold change). Bars represent log-fold changes in relative abundance between host species. Unclassified or candidate genera are shown using database-assigned identifiers (e.g., JC017, OMJO01)

Antibiotic treatment produced strong shifts in bacteriome composition, with enrichment of Chitinophaga and reduction of several genera, including Herbiconiux, Pseudescherichia, Aeromicrobium, and Luteolibacter. The combined Bd + antibiotic treatment further altered community composition, with enrichment of Achromobacter and Brucella, and reductions in taxa including Dyella, Herbiconiux and Luteolibacter.

Discussion

In wildlife, microbiome diversity and composition are often used as proxies for host health and disease resistance [48, 49]. In amphibians, low bacteriome diversity has been associated with higher risk of species endangerment [50], whereas the Bd-inhibitory function of symbiotic microbial communities is considered as a mechanism of disease defense [16, 43, 51]. In this study, we present experimental evidence linking the skin bacteriome to defense against chytridiomycosis in tropical amphibians, with the strength of bacteriome-mediated protection differing markedly between two ecologically similar species. Our experiment tested both the individual and interactive effects of Bd infection and bacteriome suppression on frog survival. Suppressing the skin bacteriome in Bd-exposed H. binotatus led to higher hazard ratios, mortality, and Bd infection intensity, whereas Bd exposure alone had no detectable effects. The magnitude of this interaction indicates that bacteriome suppression reduced survival by more than an order of magnitude. Crucially, antibiotic treatment alone, in the absence of Bd exposure, did not reduce survival in either species, indicating the lethal effect is contingent on pathogen exposure and more consistent with bacteriome-mediated than antibiotic-mediated pathology. The proportion of Bd-inhibitory sequence reads was negatively associated with Bd infection intensity in H. binotatus, suggesting functional bacteriome composition limits pathogen infection intensity in this species. Conversely, antibiotic treatment did not independently modulate survival in I. henselii, and the proportion of Bd-inhibitory reads showed only a non-significant trend toward reduced infection intensity, indicating weaker or less consistent bacteriome-mediated protection. Although we did not examine interactions among host-derived responses, antibiotics, and the bacteriome, our findings align with bacteriome-mediated defense in H. binotatus. Our findings may help guide future studies aimed at pinpointing important microbial taxa, their Bd-inhibitory compounds, and indicators of bacteriome stability useful for disease mitigation.

In a recent field-based study, members of our team did not detect Bd-infected H. binotatus among 126 individuals surveyed across 40 Atlantic Forest sites [30]. Similarly, Brito de Assis et al. [52] detected no Bd-infected H. binotatus among 66 sampled individuals in a separate geographic study. Given these findings, we initially considered two possibilities: either H. binotatus exhibits strong resistance to Bd infection, or it is highly susceptible, with rapid mortality preventing detection. While we cannot confirm that their skin bacteriomes remained fully undisturbed under laboratory conditions, the presence of Bd-infected individuals surviving initial exposure and maintaining lower Bd infection intensity argues in favor of resistance as a disease defense mechanism in this species. Most H. binotatus individuals in our study were not infected with Bd, with only two testing positive at the time of capture in the wild. Infected frogs with unsuppressed bacteriomes developed lower Bd infection intensities and had higher survival than antibiotic-treated individuals, consistent with the significant negative relationship between Bd-inhibitory read proportion and Bd infection intensity. Considering that uninfected, antibiotic-treated frogs showed no reduction in survival, our results point to the bacteriome as a plausible defense mechanism in H. binotatus. Additional explanations to the low Bd occurrence in wild populations of H. binotatus [30] are also plausible, including low natural exposure due to behavioral avoidance of Bd, habitat-mediated differences in encounter rates, or microhabitat use that reduces contact with infectious zoospores. However, the inclusion of the closely related species I. henselii in the experiment, a species that shares a direct-developing life history and occurs in overlapping habitats [32], provides a relevant comparative framework that may help disentangle exposure-related effects from intrinsic differences in host susceptibility.

For I. henselii, we observed low survival after Bd exposure in the laboratory experiment, indicating high Bd susceptibility. In contrast to H. binotatus, high Bd prevalence has been consistently observed in I. henselii across multiple field surveys spanning different habitats and climatic conditions [30, 53]. High prevalence has been associated with tolerance mechanisms in other amphibian species [54, 55]. In our experiment, I. henselii showed a marked drop in survival starting 12 days after Bd exposure, contrasting with the steep drop in survival observed in bacteriome-suppressed H. binotatus starting 7 days after Bd exposure, suggesting different defense mechanisms operate in each species. Critically, survival did not differ between Bd-exposed I. henselii with and without antibiotic treatment, indicating that bacteriome suppression did not further increase mortality in this species. Bd exposure alone was sufficient to drive high mortality regardless of bacteriome state. This contrasts sharply with H. binotatus, where bacteriome suppression under Bd exposure significantly increased mortality, consistent with a species that relies more critically on its skin bacteriome for protection. The apparently higher mortality in the control group relative to the antibiotic-only group in I. henselii is unexpected and attributed to stochastic mortality, which is supported by the lack of a significant main effect of antibiotic treatment on survival in our Cox models.

The effects of metabolites secreted by symbiotic skin bacteria on Bd growth are well documented in vitro [22, 23, 34, 56], and in vivo studies have also found correlations between the proportion of Bd-inhibitory read and amphibian population persistence [29, 57, 58]. Besides the direct effects of metabolites on pathogen growth, interactions between bacterial taxa can ultimately lead to secretion of new and more potent metabolites, reinforcing the importance of diversity in functional microbial communities [56]. Certain bacteria known for producing Bd-inhibitory metabolites also synthesize volatile organic compounds, which may further contribute to the suppression of pathogen growth [59]. Taken together, mounting evidence supports the idea that the presence of a few functional taxa, along with overall community structure, can be strong determinants of microbially mediated defenses in amphibians, even when the overall community composition remains relatively similar [60, 61], as observed in our study. Our functional dataset, based on Bd challenge assays using bacteria cultured from the same individuals frogs used in our experiment, provides a more reliable basis for assessing anti-pathogen function than broad, non-specific reference databases. We also matched high-quality PacBio full-16 S rRNA reads to our functional database using a 100% similarity cut-off, which is more precise than matching based on Illumina 16 S rRNAreads from the V4 region alone. Although different Bd isolates were used across experiments due to logistical constraints, both belong to the Bd-GPL lineage, suggesting that the functional patterns observed are unlikely to be driven by isolate identity.

While both focal host species showed similar overall bacteriome composition and richness at the moment of sampling in the wild, H. binotatus exhibited greater Bd-inhibitory bacterial richness at the time of capture. The richness of Bd-inhibitory bacteria has been associated with a reduced capacity for Bd colonization [55, 56], consistent with the observed increased infection intensity in antibiotic-treated H. binotatus. ANCOM-BC2 identified Achromobacter and Brucella as enriched specifically under combined Bd + antibiotic treatment, while Chitinophaga was enriched under antibiotic treatment alone. In a context of synergy between global change stressors and disease, higher functional taxon diversity has been linked to more robust microbiome-mediated defenses [57–60]. This could occur through functional redundancy, where a lost taxon is replaced by a functionally equivalent one [53, 61]. The statistically significant effect of Bd-inhibitory read proportion on infection intensity in H. binotatus reinforces this idea, while the absence of this effect in I. henselii suggests that functional buffering through the bacteriome is stronger in H. binotatus than in this closely related species.

Bacterial communities changed during the experiment. Bacteriome dispersion showed divergent patterns between host species and treatments. In a laboratory environment, aseptic conditions are expected to homogenize host-associated microbial communities [62]. That is the pattern we observed for both amphibian species, as higher infection intensity was associated with lower bacteriome dispersion. However, this relationship reversed in Bd-infected H. binotatus treated with antibiotics, where dispersion increased with Bd infection intensity (as indicated by the statistically significant interaction term), consistent with the Anna Karenina principle [34]. In contrast, I. henselii showed a different pattern: antibiotic treatment was associated with reduced dispersion overall, with no significant interaction between Bd infection intensity and antibiotic treatment. This contrasts with the infection-dependent stochasticity seen in H. binotatus. Although increased dispersion alone is not synonymous with dysbiosis, we use it here as a proxy for an altered microbiome state detrimental to host health [60, 63–65]. In agreement with this prediction, our results show a strong link between bacteriome suppression, higher bacteriome dispersion, and increased hazard ratios and infection loads in H. binotatus.

It is important to acknowledge that antibiotics could have effects beyond suppressing the host-associated bacteriome. Antibiotics like those used in this study have been linked to increased oxidative stress in amphibians [66], and have been found to affect keratinization, cornification, and cardiac muscle development in tadpoles [67]. Mechanisms of host-derived immunity, such as the production of antimicrobial peptides, could also play a role in disease resistance and be affected by antibiotics. This was outside the scope of this study. Importantly, the time-varying Cox model showed that antibiotic treatment was associated with reduced survival independently of Bd load, which may partially reflect these host-derived effects. We interpret our results as consistent with, rather than conclusive proof of, bacteriome-mediated defense, and encourage future studies that explicitly partition host immune and microbiome contributions to pathogen resistance.

Conclusions

We (1) investigated mechanisms of bacteriome-mediated disease defenses proposed in a recent field study focused on a tropical amphibian species, revealing that the strength of this protection differs markedly between two ecologically similar species [30], (2) found evidence supporting bacteriome-mediated defenses against chytridiomycosis in this system, and (3) developed a comprehensive functional database based on in vitro challenge assays. Integrating assessments of host innate and adaptive immunity with microbiome function, including the effects of microbiome suppression, could improve our understanding of the mechanisms driving disease defense in amphibians and other vertebrates.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Experimental design: LKS and CGB. Fieldwork: RM. Experiment setup and running: LKS, JRE, MBT, ACZ, LFT, MP and CGB. Microbial culturing and assays: LKS and ABA. Molecular lab: LKS; Data analysis: LKS, SMB and CGB; First draft writing: LKS. All authors contributed to the writing process.

Funding

This research was funded by The National Science Foundation (DBI-2120084; DEB-2413542), the São Paulo Research Foundation (FAPESP #2022/11096-8; #2020/02994-7; #2022/07125-2; #2025/20215-9), and the National Council for Scientific and Technological Development (CNPq #302834/2020-6; #152052/2024-0).

Data availability

Amplicon sequence data is available under NCBI (PRJNA1354894), isolates sequences are available in the AmphiBac database (https:/github.com/AmphiBac). All data is also openly available at request to the correspondent author (mailto: laurakauerschuck@gmail.com). Code and metadata are available as supplementary material.

Declarations

Ethical approval

This study, including animal husbandry, experimental procedures and euthanasia were conducted under approval of both The Pennsylvania State University’s Institutional Animal Care and Use Committee (PROTO202302538) and the State University of Campinas Commitee of Ethics and Animal Use (CEUA #6341-1/2023). Field collections of amphibians were authorized by SISBio (89980-1) and assessment of microorganisms was authorized by SISGen (AACFE69). This experiment is compliant to ARRIVE guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.