Significance
Habitat fragmentation is a major driver of biodiversity loss, yet its effects on host-associated microbiomes and disease resistance remain poorly understood. We show that spatial separation between critical habitats, such as natural forests and aquatic breeding sites, impairs amphibians’ ability to recruit protective skin bacteria that defend against the deadly chytrid fungus Batrachochytrium dendrobatidis. This disruption of multihabitat connectivity reduces microbiome antifungal function, leading to increased disease risk in amphibians. Our findings reveal a mechanistic link between environmental disturbance, microbiome health, and pathogen susceptibility, showing that conserving spatial habitat connectivity is vital for preserving wildlife’s natural defenses and helping to prevent population declines and reduce extinction risk. This work advances our perspective on how landscape change reshapes wildlife disease dynamics.
Keywords: landscape connectivity, acquired resistance, microbiota, microbial filtering
Abstract
Disruption of habitat connectivity alters host movement patterns and pathogen exposure in wildlife. Changes in exposure dynamics have led to increased research interest in host-associated microbial communities (i.e., microbiomes), particularly in how repeated encounters with pathogens may drive microbial filtering processes that favor the assembly of pathogen-inhibiting microbiomes, a concept known as the adaptive microbiome principle. Understanding how habitat connectivity and pathogen exposure shape adaptive microbiomes remains a key frontier in disease ecology. For widely distributed waterborne pathogens such as the amphibian chytrid fungus Batrachochytrium dendrobatidis (Bd), sustained host exposure in contiguous terrestrial–aquatic habitats may promote selection of microbiomes with enhanced antifungal properties. In contrast, under conditions of high habitat split, where key habitats such as forests and water bodies are spatially disconnected, limited exposure to Bd during the pre-tbreeding and overwintering seasons may hinder the selection of Bd-inhibitory microbiomes that are critical for host protection during the subsequent breeding season, when pathogen exposure in water bodies is at its peak. Our results demonstrate that habitat split may limit recruitment of putative Bd-inhibitory skin bacteria, while Bd loads increase with habitat split for certain amphibian species. Results from Joint Species Distribution Models also indicate that habitat split is a key driver of reduced skin bacterial diversity, even after accounting for biotic and abiotic metrics. Our study provides evidence that spatial connectivity among natural habitats is essential for maintaining multiple levels of biodiversity, from host species to their associated functional microbiomes, highlighting a critical link between environmental disturbance, microbial defenses, and disease dynamics.
Wildlife faces a multitude of threats, including anthropogenic habitat loss and disease emergence, both of which reduce the capacity of landscapes to sustain vertebrate biodiversity (1, 2). Habitat loss often leads to habitat fragmentation and habitat split, two related but distinct forms of landscape disruption. In tropical forest environments, fragmentation is defined as the fractionation of large, continuous forested areas into smaller isolated forest patches, whereas habitat split is the disruption of spatial connectivity among the multiple classes of natural environments that many organisms require to complete their life cycles (e.g., spatial separation between remnants of natural forests and aquatic breeding sites) (3). Both habitat split and fragmentation impact vertebrate movement, aggregation (4), pathogen transmission, and exposure to environmental microbes (5). Although disruption of habitat connectivity can reduce pathogen transmission and disease (6), it may also reduce host immunity to pathogens by limiting regular exposure that occurs in contiguous environments (7). Thus, to analyze the relationship between environmental disturbance and vertebrate health, we must understand how changes in natural habitats impact host immunity components, including the recruitment of microbial symbionts and host exposure to pathogens.
A growing body of literature links host microbial symbionts to immunological defenses in vertebrates. Symbiotic microbial communities and their metabolic products, also called the microbiome, are known to provide protection against pathogens (8–11). While pathogenic infections may sometimes lead to detrimental imbalances of the microbiome, termed dysbiosis (12–14), sustained pathogen pressure can also alter microbial interactions which, coupled with host-derived responses, may result in more highly “adapted” microbiomes enriched with pathogen-inhibiting microbial taxa (15). Previous pathogen exposure can assemble microbial communities that are adapted for subsequent pathogen encounters and are able to more rapidly respondto the invading pathogen (16). Coined as the adaptive microbiome principle, this “microbiome memory” from previous pathogen exposure could facilitate enhanced host resistance and recovery following pathogen invasion (17).
Maintaining an “adapted microbiome” may require both regular, low-load pathogen exposures and access to beneficial microbial reservoirs throughout natural landscapes. Fragmentation and habitat split, however, could restrict or limit the development and/or maintenance of adaptive microbiomes (7). Thus, this barrier to microbiome-mediated defense deserves attention as a mechanism underlying observed increases in wildlife disease risk in disturbed and fragmented landscapes (18–21). Providing some evidence for this hypothesis, microbial communities in animal populations from heavily disturbed landscapes often differ in diversity and structure, and they may exhibit reduced antipathogen functionality compared to those in natural, undisturbed habitats (22–25).
Amphibians are an ideal model to test how habitat split and fragmentation shape host-microbiome responses to disease because most species seasonally aggregate near water, which influences rates of exposure to the globally distributed and catastrophic waterborne pathogen Batrachochytrium dendrobatidis (Bd). Indeed, both Bd outbreaks and habitat split have been individually linked to amphibian population declines and extinctions, mainly in biodiverse tropical regions (3, 26, 27). Recent findings indicate that disease could be acting synergistically with habitat split and fragmentation to explain the observed population declines in both tropical and temperate amphibians (7, 28). While open-canopy habitats generally reduce Bd growth and persistence by raising temperatures above the pathogen’s thermal optimum (29–31), recent studies reveal that tropical amphibians in small Atlantic Forest fragments have higher Bd infection loads than those in continuous forests (28). Thus, habitat split could drive increased infection in amphibians that undertake seasonal breeding migrations by concentrating naive hosts that lack acquired immunity.
In landscapes where natural terrestrial and aquatic environments are connected, amphibians may experience consistent encounters with water bodies year-round (32, 33). This could facilitate resistance to waterborne Bd through repeated low-level pathogen exposures (34), possibly promoting the persistence of Bd-inhibitory microbes and resilient skin microbial communities. Consistent with this hypothesis, recent experimental findings show that repeated exposure to Bd metabolites increases the estimated Bd-inhibitory function of the amphibian skin bacterial community (35). While this estimate of Bd-inhibitory function is based on genotypic similarity to known Bd-inhibitory isolates in vitro, several studies indicate that in vitro antifungal properties strongly correlate with Bd-inhibitory function in vivo (36, 37). In amphibian populations coexisting with Bd in an enzootic stage, individuals with subclinical infections tend to harbor distinct skin bacterial communities enriched with Bd-inhibitory taxa (38, 39) when compared to those experiencing Bd-induced epizootics driving population declines (40) and those naïve to the pathogen (39). This suggests that Bd exposure may shape skin microbiomes in ways that influence their ability to inhibit Bd growth, indicating that skin microbiomes that are capable of inhibiting Bd could be selected for over periods of pathogen exposure; a likely scenario in continuous natural habitats (Fig. 1A).
Fig. 1.
Hypothetical amphibian skin microbiome responses under the contrasting scenarios of (A) connected natural habitats and (B) split natural habitats.
Conversely, in landscapes occupied by humans, land use often leads to habitat split for amphibians (i.e., spatial separation between remnants of natural forests and aquatic breeding sites) (41). Especially in amphibian species with biphasic life-cycles (i.e., species that have an aquatic larval phase and a terrestrial postmetamorphic phase), restricted movement could increase disease risk within forest fragments through limited exposure to Bd during the nonbreeding season, particularly when water bodies are located outside forest fragments (26). Such isolation then precludes individuals from “priming” their immune system against pathogens such as Bd, as well as from recruiting and maintaining an “adaptive microbiome” with Bd-inhibitory function (7, 15, 42) (Fig. 1B). The skin-associated bacterial communities of amphibians found in disturbed and fragmented habitats have shown distinct structures (43–47) and increased dispersion (45, 47), which in some cases indicates a state of dysbiosis linked to disease (48, 49). Within the tropics, for instance, individuals of the treefrog Boana faber translocated from continuous forests to small, isolated forest fragments showed that even exposure to forest edges drives significant changes in the composition of skin bacterial communities and a reduction in putative Bd-inhibitory function (46).
In the Neotropics, habitat split affecting amphibians is a widespread phenomenon, particularly in regions of intermediate topographic complexity (41). In Brazil’s Atlantic Forest, our focal study area, habitat disturbances, including agriculture, livestock farming, industry, and urban settlements, are concentrated in valleys. “Dry” forest fragments without permanent water bodies are often left on steeper slopes and hilltops (41, 50). This historical forest fragmentation process therefore produced a large proportion of small upland forest fragments without perennial water bodies like ponds or streams. In such landscapes, amphibians are often forced to migrate between upland forest fragments and aquatic habitats through disturbed environments to complete their life cycle or to seek shelter from dry conditions (26). Previous results highlight that habitat split is linked to population declines and the historical loss of amphibian species diversity in Brazil’s Atlantic Forest (3, 26).
Our overarching goal is to test whether microbiome-mediated pathogen defenses in multiple species of amphibians are altered by habitat split in fragmented tropical and subtropical landscapes. Our central hypothesis is that repeated exposures to a waterborne pathogen in continuous natural habitats could result in an amplified pathogen-inhibiting capacity within the amphibian skin microbiome, estimated using an expanding pathogen-specific reference dataset based on assays of fungal growth when challenged with bacterial metabolites (51) (Fig. 1). If multihabitat connectivity is an important driver of host microbiome function, we expect host-associated bacteria to respond strongly to the degree of habitat split while controlling for other local- and landscape-scale metrics of environmental disturbance. We predict that the most abundant amphibian skin-associated bacteria known to inhibit Bd in vitro will show reduced relative abundance in amphibians sampled from areas with high habitat split. As a result, we expect that Bd infection loads will increase in areas of high habitat split for multiple amphibian species. Unraveling the role of multihabitat connectivity on host microbiome assembly and function could enhance our understanding of adaptive microbiomes and acquired disease resistance in increasingly threatened wildlife.
Methods
Sampling Design, Spatial Metrics, and Study Species.
Our field work took place in a section of Brazil’s Atlantic Forest located within the state of São Paulo, Brazil (locations shown in SI Appendix, Fig. S1). Our sampling design comprised eight landscapes (defined as circular areas of 10-km radius each; SI Appendix, Fig. S1) spread across the study region, which varied in their degree of habitat loss (ranging from 13.55 to 81.32% natural land cover remaining). Within each landscape, we sampled in five areas of either forest fragments or continuous forestenvironment, for a total of 40 sampling sites. Landscapes varied in their degree of habitat disturbance, though were relatively similar in terms of geographic, temperature, elevation (latitude), topography, and rainfall patterns, which allowed us to control for variation in geoclimatic conditions. Our sampling took place within the amphibian breeding season (from September 2018 to January 2019). To account for temporal variation, we sampled each of the 40 sampling sites twice during this period (at the beginning and toward the end of the breeding season).
Landscape and sampling site-level spatial data (e.g., hydrological and land use/land cover) were obtained from the repository of the Fundação Brasileira Para o Desenvolvimento Sustentável (Brazilian Foundation for Sustainable Development). Land cover layers had a 5 × 5 m resolution and were generated using the RapidEye imagery at a scale of 1:10,000, for the year 2017 (Fundação Brasileira Para o Desenvolvimento Sustentável, 2020). Land use/land cover data included natural habitats and anthropogenic habitat classes such as agriculture, silviculture, and urbanization, in addition to drainage networks comprising streams, lakes, and swamps. Within the context of our study, we calculated the following metrics using the extracted data: 1) “natural land cover” at the landscape level (N = 8)—defined as the percent of natural forest within the 10-km circular radius that defines each landscape; 2) “habitat split” at the sampling site level (N = 40)—represented by the average minimum distance (at north, south, east, and west) between the edge of the natural habitat and the nearest drainage network for each sampling site, which included marshes, streams, and artificial ponds; 3) and “density of forest edge” at the sampling site level (250 m radius), as a proxy for edge effect (N = 40)—calculated as the sum of the lengths (m) of all edge segments of the respective sampling site, divided by the total landscape area (m2) and multiplied by 10,000 in order to transform the ratio to hectares (meters/hectares). Variation in habitat split among forest fragments of similar sizes and edge densities allowed us to tease apart the effects of forest fragmentation from habitat split.
To test the potential effect of habitat split on amphibian skin bacterial communities, we focused on four amphibian species that exhibit different development modes: aquatic larvae or direct development of embryos within eggs in terrestrial environments. Amphibian species with distinct developmental modes differ in the degree of water requirements for breeding (33) and, consequently, exposure to Bd’s waterborne zoospores (52). For instance, Atlantic Forest frogs with an aquatic larval stage may need to migrate across a matrix of disturbed habitat to reach their natural reproductive habitat, whereas those with direct terrestrial development may avoid dispersing outside remnants of natural forest because they are able to complete their life cycle in the forest leaf litter. For the species with aquatic larvae, we focused our sampling on the treefrog B. faber (Hylidae) and the forest toad Rhinella ornata (Bufonidae); and for those with direct development, we sampled the terrestrial-breeding species Haddadus binotatus (Craugastoridae) and Ischnocnema henselii (Brachycephalidae). In a previous study by our group using this study dataset, Martins et al. (53) reported that these four species showed different Bd infection prevalence across the landscapes during our study, with I. henselii showing the highest Bd prevalence (34.8%, n = 204) and H. binotatus with no detectable Bd infections (0%, n = 126), while the other two species showed intermediate levels of Bd infection prevalence [R. ornata: 5.4% (n = 112) and B. faber: 13.8% (n = 356)]. Refer to SI Appendix, Fig. S2 and Table S1 for detailed sample sizes by species across all study sites.
During surveys, we used the sampling technique “free and unrestricted search” to detect and sample individuals of the focal study species. This method is considered an efficient approach for sampling a large number of individuals (54). We captured each frog using a new pair of nitrile gloves. Upon capture, we rinsed their skin with distilled water to remove transient bacteria and collected samples of skin bacterial communities by swabbing their skin (using swabs MW113, Medical Wire Equipment) following Hyatt et al. (55). Skin swabs were then individually stored in sterile 1.5 mL microcentrifuge tubes, kept on ice during fieldwork, moved to a freezer at the field research facility, then stored at −20ºC in the laboratory. All frogs were released at the point of capture after swabbing.
Bd Quantification.
We extracted Bd and bacterial DNA from skin swabs using the DNeasy Blood and Tissue kit (Qiagen, Valencia, CA), following the manufacturer’s protocol but extending the incubation period at 56 °C to overnight to increase DNA yield. We tested for Bd and quantified infection intensity using a standard probe-based qPCR assay, which targets the ITS1 region (primers ITS1-3 and 5.8S rRNA) (56). To run qPCR assays for Bd, we diluted template DNA 1:10 with nuclease-free water and ran each sample in duplicate. We determined Bd infection intensities using standard curves derived from synthetic standards ranging from 102 to 106 copies (Integrated DNA Technologies) (57). Thus, Bd infection estimates correspond to ITS copy numbers. Sample duplicates with consistent Bd amplification were considered positive; inconsistent replicates were repeated a third time. Infection intensity estimates for each sample represent the average of the two positive duplicate runs. Negative controls were included in the assays to rule out contamination.
Skin Bacterial Community Characterization.
We characterized frogs’ skin bacterial communities by amplifying and sequencing the V4 region of the 16S rRNA gene of the bacterial genome. Bacterial DNA was processed using the universal primers 515F and 806R with a dual index strategy (58, 59). PCRs for each sample included 12.2 μL of ultra-pure water (Invitrogen), 4 μL of 5× Phire reaction buffer (Finnzyme, Espoo, Finland), 0.4 μL of 2.5 nM dNTPs (Invitrogen), 0.4 μL of Phire Hot Start II (Finnzyme, Espoo, Finland), 0.5 μL of each primer, and 2 μL of template DNA, for a total volume of 10 μL in duplicate reactions. PCR conditions consisted of an initial denaturation step of 3 min at 98 °C, followed by 38 cycles of 5 s at 98 °C, 5 s at 50 °C, and 15 s at 72 °C, and a final extension of 3 min at 72 °C. Samples were run in duplicate, and each 96-well plate included a negative control containing nuclease-free water in the place of template DNA. PCR products were then pooled and visualized in a 1% agarose gel to confirm amplification and the absence of contaminants in negative controls. Last, we pooled equal volumes of PCR amplicons of all samples to make a composite sample, which we purified and quantified using a QIAquick Gel Extraction kit (Qiagen, Valencia, CA) and a Qubit 2.0 fluorometer (dsDNA Broad-Range Assay kit; ThermoFisher Scientific, San Jose, CA), respectively. The pooled library was sequenced on an Illumina MiSeq (2 × 250 bp) sequencer. Metafiles and 16S rRNA amplicon sequences were deposited in the NCBI Sequence Read Archive database (https://www.ncbi.nlm.nih.gov/sra, under accession number: PRJNA1375872).
We processed demultiplexed sequences from Martins et al. (53) using the Quantitative Insights into Microbial Ecology 2 (QIIME2 v. 2019.1) (60). We filtered low-quality reads using the default parameters of the q-score plugin and clustered sequences into suboperational taxonomic units (sOTUs) using the Deblur method (61). We used forward reads for compatibility with Deblur and trimmed sequences to 150 bp to maximize read quality. To build a phylogenetic tree of bacterial sOTUs, we aligned sequences using the MAFFT method (62), masking poorly aligned regions, and built a tree using FastTree2 with default parameters (63). To assign taxonomy to sOTUs, we used the scikit-learn naïve Bayes classifier and the Greengenes 13.8 reference sequence database (64–66). We removed sOTUs identified as mitochondria and chloroplasts, and eight sOTUs identified as differentially abundant in the PCR controls, as they are likely to be potential contaminants. In addition, we removed sOTUs with fewer than 0.005% (438 sequences) of the total number of sequences (8,767,188) in the dataset (67). Due to substantial variation in sequencing depth across samples, we rarefied our feature table to 2,000 sequences per sample. For this manuscript, we excluded samples from the species Bokermannohyla circumdata and B. hylax due to small sample sizes across landscapes. This resulted in a final rarefied feature table of 586 frog samples, a total of 1,438 sOTUs, and 1,172,000 sequences.
To test for changes in the Bd-inhibitory potential of host-associated bacteria in our study system, we generated a metric of putative Bd-inhibitory function for each frog’s skin bacterial community using the AmphiBac reference database (51). This database consists of 16S rRNA gene sequences from bacterial isolates that have been cultured and tested against Bd using in vitro challenge assays to assess the inhibitory effect of their secondary metabolites (68). Assays like this, which use single bacterial isolates, likely provide conservative estimates of Bd inhibition. Interactions among microbes often stimulate the production of additional antimicrobial compounds, meaning isolates grown alone are likely to produce fewer inhibitory compounds than they would in coculture (69). Other community-level mechanisms (e.g., biofilm formation, nutrient competition, niche exclusion) may also contribute to the suppression of Bd growth and infection processes (70).
To date, the AmphiBac database contains 7,874 tested isolates, including 1,046 isolates from Atlantic Forest frogs, 793 of which were isolated from our focal amphibian species. Using this resource, we predicted the Bd-inhibitory potential of NGS-sequenced microbiome samples through 16S rRNA gene matching. For each sample, we computed a closed-reference metric for putative function that accounts for differences in the amount of database representation across samples. Specifically, we used the unrarefied feature table to calculate the pre-rarefaction ratio of Bd-inhibitory reads (those that matched AmphibBac_InhibitoryStrict_2025.1 at 100% identity) to the total number of reads matching the full AmphiBac database of tested isolates (AmphiBac_FullDatabase_2025.1-tested at 100% identity). This approach ensures that predictions were limited to isolates with experimentally verified functional information. We repeated this procedure using the Brazil-specific subset of AmphiBac (AmphiBac-2025.1-Braziltested). The unrarefied feature table contained 5,996,578 reads (mean number of sOTUs per species: B. faber—182, H. binotatus—353, I. henselii—314, R. ornata—267). On average, 67.8% of 16S rRNA reads from each sample matched the full AmphiBac database (Fig. 2), and 39.4% matched the Brazil subset. For more details on taxonomic representation (SI Appendix, Fig. S3), phylogenetic coverage (SI Appendix, Fig. S4), and variation in diversity among frog species in terms of matched inhibitory reads (SI Appendix, Fig. S5), please see SI Appendix.
Fig. 2.
Representation of cultured bacteria (read counts) by host species and status for the Bd-challenge assays.
Statistical Analyses.
Overview.
Considering that habitat split has been previously linked to increases in Bd infection loads in our study system (7), we tested for this link in the present dataset and took into account potential differences in the effect of habitat split across frog host species. Then, we focused on measuring the association between habitat split and the putative Bd-inhibitory function and structure of the skin bacterial communities of our focal amphibian species. Different statistical models for Bd infection load and putative Bd-inhibitory function were used to ensure model fit while maintaining our ability to address the research objectives of our study.
Testing the link between habitat split and Bd infection load.
To test the relationship between habitat split and Bd infection load in our four focal amphibian species, we used a Zero-Inflated Negative Binomial model with a log-link function to account for excess zeros and overdispersion in the infection data. Bd loads were rounded to the nearest integer using the R function ceiling so that samples with low but detectable loads were retained as positive. The model included Bd load as a response variable and the following variables as predictors: natural land cover (%), density of forest edge (ha), habitat split (split distance in m), Julian date (a metric of time), amphibian species, and an interaction term between habitat split and amphibian species. This interaction term aimed to include potential differences in the effect of habitat split on frog species infection loads, as we expect it to depend on species life history. Prior to model fitting, we standardized the predictor variables associated with sampling site- and landscape-level features, such as habitat split, density of forest edge, and natural land cover by centering and scaling their values to z-scores (mean = 0 and variance = 1; function “scale”). We fitted the model using the package glmmTMB (71) and the argument “ziformula” to specify a zero-inflated model. To assess model fit and error distribution, we visualized residual plots against model predictions using the R package DHARMa (72). We also tested for multicollinearity among model predictors using the function check_collinearity from the R package performance (73), which indicated collinearity values below a commonly used Variance Inflation Factor (VIF) threshold (VIF < 10; observed VIF values ranged from 1.81 to 6.04) (74).
Testing the link between habitat split and putative bacterial Bd-inhibitory function.
We tested the effect of habitat split on the proportion of putative Bd-inhibitory bacteria by fitting a Generalized Linear Mixed Model (GLMM) with a binomial distribution and logit link function. Given that our response variable referred to the proportion of “successes” (i.e., proportion of matched Bd-inhibitory sequence reads), we used the argument “weights” to specify the total number of “trials” at the sample level, which referred to the total number of matched reads to the full AmphiBac database. The model included the proportion of Bd-inhibitory sequence reads in each amphibian skin swab as a response variable, and the following variables as predictors: density of forest edge (ha), habitat split (m), Bd infection occurrence, amphibian species, and an interaction term between habitat split and amphibian species (as for the Bd infection model). Prior to model fitting, we standardized the predictor variables as described above. In addition, to objectively test the effect of habitat split and improve model fit, the model included landscape ID (within which sampling sites are nested) and Julian date (a metric of time) as random factors to account for interlandscape variation in natural land cover, spatial autocorrelation, and temporal variation. We fitted the model using the function glmer from the package lme4 (75). We evaluated model fit by assessing residual plots against model predictions using the qqmath function from the package lattice (76). Last, we calculated summary statistics and P-values of each fixed factor by comparing nested models using Likelihood Ratio Tests (77).
Testing for the association between habitat split and changes in skin bacterial communities.
We tested how frog skin bacterial communities correlate with landscape and sampling site-level covariates, habitat-split, time, and Bd infection occurrence by fitting Joint Species Distribution Models (JSDMs) using the Hierarchical Modeling of Species Communities (HMSC) framework (78). Overall, this type of model follows the structure of a multivariate hierarchical generalized mixed effect model that uses Bayesian inference (79). We built a JSDM for each amphibian host species and, to improve model fitting, these models included only the sOTUs present in at least 10% of the samples and with a relative abundance above 0.005% in at least one sample (80). Hence, the number of sOTUs in each model varied among amphibian hosts and ranged from 340 in R. ornata to 510 in H. binotatus (SI Appendix, Table S2). To account for the excess of zeros (i.e., sOTU absences) in the community data, we applied a hurdle modeling approach for each amphibian species. This approach consisted of fitting a binomial model (with a probit link function) to sOTU presence–absence data and a log-normal model to sOTU abundance data (scaled to zero mean and unit variance) conditional on presence (i.e., zeros are considered missing data and thus ignored) (80, 81).
The structure of the JSDMs consisted of a response matrix Y composed of either sOTU presence–absence data or sOTU abundances, and a set of predictors defined in a matrix X. The model predictors included natural land cover (%), density of forest edge (ha), habitat split (m), Julian date (a metric of time), and Bd infection occurrence. Prior to model fitting, we standardized the predictor variables (except for Julian date and Bd occurrence) by centering and scaling their values to z-scores. To reduce the potential for biased model inference and prediction, we included a spatial random effect at the sampling site level by including their respective coordinates in a matrix S. Spatial random effects in JSDMs can model both spatial variation as well as potential sOTU co-occurrence patterns through residual species to species correlation matrices after accounting for their response to model predictors (78, 79). We built the model with the package Hmsc (82).
We fitted each JSDM using the default prior distribution from the package Hmsc and four Markov Chain Monte Carlo (MCMC) chains to sample the posterior distribution. Each MCMC chain ran for 150,000 iterations, out of which the first 50,000 were discarded as burn-in, and the rest thinned by 100 to yield 1,000 posterior samples—for a total of 4,000 posterior samples. We fitted the JSDMs using the function sampleMcmc, also from the package Hmsc. To ensure posterior samples are an accurate approximation of the posterior distribution, we quantitatively assessed the convergence of MCMC chains by computing the mean effective sample size (Hmsc package function effectiveSize) and potential scale reduction factor (function gelman.diag) across the parameters assessing sOTU responses to model predictors (79). We assessed model fit in terms of explanatory power of JSDMs by calculating the Area Under the Curve (AUC) measure for the presence–absence models and the coefficient of determination R2 for the abundance models (79).
Results
We found that higher levels of habitat split predicted a lower proportion of putative Bd-inhibitory bacterial reads in all frog species (LRT χ2 = 87,813, logit scale parameter estimate = −1.437, P < 0.0001; Fig. 3A). Habitat split had a stronger effect on microbiome Bd-inhibitory function than density of forest edge (LRT χ2 = 592, logit scale parameter estimate = −0.164, P < 0.0001). The slope of the relationship between proportion of Bd-inhibitory bacterial reads and habitat split also varied among frog species (LRT interaction term χ2 = 84,013, P < 0.0001); it was the highest in H. binotatus and R. ornata—the two species showing the lowest Bd prevalence and infection loads (logit scale parameter estimate = −2.302 and −2.649, respectively), and the lowest in B. faber and I. henselii (−1.437 and −1.464, respectively). The proportion of putative Bd-inhibitory bacterial reads also differed among frog species (LRT χ2 = 211,563, P < 0.0001) and ranged from an average of 0.580 (SD ± 0.298) in I. henselii to 0.780 (±0.181) in B. faber and 0.774 (±0.135) in R. ornata, with an intermediate estimate of 0.739 (±0.124) in H. binotatus. Individual amphibians carrying Bd infections at the moment of capture also showed a significantly lower proportion of Bd-inhibitory bacterial reads than uninfected individuals (LRT χ2 = 121, logit scale parameter estimate = −0.050, P < 0.0001). We also detected a similar relationship between predicted Bd-inhibitory function and habitat split, and its interaction with amphibian host species, using the database consisting of only bacterial isolates from Brazil’s Atlantic Forest frogs (SI Appendix, Fig. S6).
Fig. 3.
Results from the most parsimonious GLMMs, controlling for temporal and interlandscape variation, supporting our hypothesis (described in text and in Fig. 1) that habitat split reduces predicted Bd-inhibitory function in the amphibian skin microbiome (A), which could be linked to higher Bd infections in some of the focal amphibian species (B). (A) is based on 100% matches to the Full AmphiBac database, which encompasses 7,874 bacterial isolates tested against Bd. The “proportion of matched Bd-inhibitory sequence reads” refers to the number of sample reads matching the Strict Inhibitory AmphiBac database divided by the number of sample reads matching the Full AmphiBac database (both matched at 100% similarity). H. binotatus is not shown in (B) because no individuals were infected with Bd.
Among the species included in the model, we found that high habitat split was linked to an increase in Bd infection loads in I. henselii (log scale parameter estimate = 1.293, z = 2.122, P = 0.033; Fig. 3B and SI Appendix, Fig. S7), the species with the observed highest prevalence and average infection loads. A similar, though nonsignificant positive trend was observed for R. ornata (log scale parameter estimate = 0.995, z = 1.517, P = 0.129), but not for B. faber (log scale parameter estimate = −0.050, P = 0.928; Fig. 3B). In addition, we detected differences in Bd infection loads among host species (χ2 = 54.60, P < 0.001) and an association with natural land cover (χ2 = 4.67, log scale parameter estimate = 1.34, P = 0.03). Last, we did not detect a relationship between forest edge (χ2 = 1.60, log scale parameter estimate = −0.58, P = 0.20) and Bd infection loads, and a marginally nonsignificant effect of Julian date (χ2 = 2.94, log scale parameter estimate = −0.001, P = 0.09).
All JSDMs testing the effect of Bd infection occurrence, time, landscape- and sampling site-level factors on sOTU occurrence probability (binomial model) and abundance (log-normal model) showed adequate convergence. The estimated mean effective sample size for each model was similar to the actual number of posterior samples (4,000; SI Appendix, Table S2), and potential scale reduction factors for each model approached one (SI Appendix, Table S2). Hence, model convergence assessment for predictors (Beta parameters—a proxy for environmental filtering) and spatial random effect (Omega parameter—representing unmeasured environmental predictors or residual species associations) indicated little to no autocorrelation among consecutive samples and consistent results in each MCMC chain, indicating that posterior samples are representative of the true posterior distribution (79). Explanatory power of presence–absence models based on AUC values ranged from 0.74 to 0.79 (SI Appendix, Table S2), indicating a relatively high power to discriminate sOTU presences from absences—models with AUC values above 0.7 are considered reliable and informative (79, 83, 84). As for abundance models conditional on presence, explanatory power ranged from R2 values of 0.18 to 0.36 (SI Appendix, Table S2).
Partition of explained variance among components of the models indicated that the spatial random effect (i.e., sampling site), habitat split, and time (Julian date) were the three most important factors driving changes in sOTU probability of occurrence (Fig. 4) and abundance (SI Appendix, Table S2). However, the relative contribution of these factors to the explained variance varied across frog species and the type of model. In terms of sOTU probability of occurrence (a proxy for changes in bacterial community composition), habitat split accounted for the second largest amount of explained variation in R. ornata and H. binotatus after sampling site, and explained a comparable amount to natural land cover in I. henselii (also after sampling site; Fig. 4). As for B. faber, time was the factor accounting for the largest amount of explained variation followed by sampling site (Fig. 4). In terms of sOTU abundance (a proxy for changes in community structure), variance partitioning showed marked differences between the amphibian species that migrate across the matrix of disturbed habitat to reach their reproductive sites (B. faber and R. ornata) and those that remain within forest fragments throughout the breeding season (H. binotatus and I. henselii). For R. ornata, habitat split was the factor that accounted for the second largest amount of explained variation (14.6%; after the 48.8% explained by sampling site) in sOTU abundances; and third in B. faber, in which it accounted for a 18.7% of the explained variation, following natural land cover (23.4%) and sampling site (24.5%). In contrast, for H. binotatus and I. henselii, variation across sampling sites accounted for the largest amount of explained variance in sOTU abundance, suggesting either a spatial signal, strong co-occurrence patterns among sOTUs, and/or an effect of unmeasured variables (79).
Fig. 4.
Variance partitioning among JSDMs predictor variables in each frog species based on the binomial (presence–absence) models. Each bar represents, at the sOTU level, the proportion of explained variance by each model predictor: random effect (sampling site), Bd infection occurrence, natural land cover, habitat split, and density of forest edge. Text boxes in figures show mean values (across sOTUs) of explained variance for each predictor.
Considering bacterial responses to habitat split based on the binomial models (i.e., changes in community composition), we found that at least 30% (with ≥85% posterior probability) of all sOTUs from each frog species were influenced by this factor (SI Appendix, Table S2). Interestingly, for species that migrate across the matrix of disturbed habitat (B. faber and R. ornata), the probability of occurrence of most of the responding sOTUs decreased with habitat split (SI Appendix, Fig. S8 and Table S2). On the other hand, in species that remain in forest fragments, the probability of occurrence of sOTUs tended to increase with habitat split (SI Appendix, Fig. S8 and Table S2). In addition, when considering both the most abundant and core sOTUs (present in at least 90% of samples) of the focal frog species, which were identified by Martins et al. (53) using the same 16S rRNA sequences, we determined similar responses to habitat split (SI Appendix, Table S3).
Our results also indicated higher temporal turnover in skin bacterial community composition relative to that in sOTU abundance, as a higher proportion of sOTUs responded to Julian date in the presence–absence models compared to those in the abundance models (SI Appendix, Table S2). Indeed, Julian date accounted for the largest portion of the explained variance in sOTU probability of occurrence in B. faber, and third in the other three species (Fig. 4). The proportion of sOTUs responding to time ranged between 0.40 and 0.73, and while most sOTUs responded negatively in B. faber (0.51; SI Appendix, Table S2), this pattern was less clear in the other frog species. Thus, B. faber appears to experience a consistent temporal reduction in sOTU richness, whereas the other species may have experienced a change in sOTU community composition rather than a reduction in richness.
In terms of infection by Bd, I. henselii, the species with the highest prevalence of infected individuals sampled (34%), had the largest proportion of sOTUs responding to Bd in terms of probability of occurrence (0.35 in I. henselii—in contrast to a maximum proportion of 0.14 in the other species; SI Appendix, Table S2) and changes in relative abundance (0.13 in I. henselii—in contrast to a maximum of 0.06 in the other species). All sOTUs in I. henselii (or nearly all in the case of changes in abundance) responded positively to Bd infection (SI Appendix, Fig. S8).
Discussion
Habitat split in fragmented environments could increase disease risk in species with complex life cycles through several mechanisms that alter host innate and adaptive immunity [reviewed in Becker et al. (7)]. Our study provides evidence that environmental disturbances resulting in habitat split may also disrupt the amphibian skin microbiome, increasing Bd infection loads in forest-specialist amphibian species. This highlights a critical link between habitat connectivity, symbiotic microbial defenses, and disease dynamics. In line with our hypothesis, we found that the predicted Bd-inhibitory function of the host skin microbiome decreased in areas of high habitat split, while Bd infection loads increased with the degree of habitat split. These results indicate that synergistic effects of land use change and disease may stem from alterations in the host-associated microbiome.
In Brazil’s Atlantic Forest, greater spatial connectivity of natural forest habitats is associated with more diverse and predictable skin microbial communities of amphibian populations (43). This connectivity could thus increase functional redundancy in the microbiome and lead to greater stability of symbiotic microbial communities. Host exposure to varying microclimatic conditions and environmental sources of microbes in fragmented landscapes may therefore impact host-associated microbial community assembly, potentially through changes in the relative contribution of deterministic and stochastic filtering mechanisms, such as those of metacommunity theory (85–88). Our findings, combined with the recent studies highlighted above, indicate that deterministic microbiome filtering could occur in response to Bd exposure concurrently as habitat loss is likely reducing the pool of environmental microbes (43, 89) and lowering the potential for microbial rescue through more stochastic assembly processes (90, 91). Indeed, in tropical systems like ours, host-associated microbial community composition appears divergent between pristine and fragmented forest habitats (43, 46, 47). In addition, animals engaging in seasonal or breeding migratory behaviors traversing disturbed environments may experience high rates of pathogen transmission and shifts in microbial community composition (46, 47, 92). Our study shows that not only fragmentation, but also the degree of split between forest fragments and aquatic breeding sites may lead to differences in composition and putative antipathogen function of symbiotic skin bacteria. While we did not track individual movements or sample aquatic microbial pools in this study, previous telemetry results and known Bd dynamics in the Atlantic Forest indicate several plausible pathways through which habitat disturbance could shape microbiome assembly and likely contribute to residual variation not captured by our design (49). In these focal landscapes, habitat split has been shown to be tightly coupled with amphibian movement constraints and longer, costly breeding migrations through disturbed environments, with documented demographic consequences (3, 26), supporting its use here as an ecologically relevant proxy for migration-related exposure rather than a direct measure of individual movement. Thus, the strength and consistency of the spatial patterns we report demonstrate that habitat split itself is a powerful ecological filter on microbiome function.
Our results indicate that among frog species, habitat split explained much of the variance in bacterial community composition, with proportions higher or comparable to those accounted for by natural land cover measured at the landscape-scale. Yet, the fraction of variance explained by habitat split and the type of response by skin-associated bacterial communities varied between amphibian species with different developmental modes (species with aquatic larvae vs. those with direct terrestrial development). B. faber and R. ornata, two migratory species with aquatic larval development, showed the greatest (and second greatest after habitat split) amount of variance in skin bacterial composition explained by time, followed by natural land cover and habitat split. Both species also exhibited a similar decline in sOTU occurrence probability with increasing habitat split, unlike the two focal species with terrestrial direct development. This result builds on previous experimental work showing that amphibians using less structurally complex substrates harbored skin bacterial communities that were less diverse, distinctly structured, and experienced higher pathogen-induced mortality compared to those exposed to natural substrates (93). The fact that time was a primary explanatory variable for the two migratory species is unsurprising, as it likely highlights how habitat split diminishes skin bacterial diversity during seasonal breeding migrations across more homogeneous, disturbed environments (e.g., agricultural lands, silvopasture systems, and pastures). This process could lead to reduced functional redundancy and lower resilience of host-associated microbiomes (94, 95).
We found that B. faber was the only focal species exhibiting a neutral relationship between habitat split and Bd infection loads, in contrast to the positive slopes observed for I. henselii and R. ornata, the latter being a species with low baseline Bd loads that may reflect its ecology and dry, rough, and granular skin, which can limit the amount of infected tissue sampled by swabbing. B. faber is the largest, most vagile, and habitat-generalist among our focal host species, and is frequently sampled in disturbed environments and therefore potentially less prone to movement-related physiological strain and genetic erosion (32, 46, 96). Considering habitat use during the overwintering season, B. faber is also the only study species capable of vertical movement into the canopy. For instance, previous work tracking the movement of this species found individuals occupying epiphytic bromeliads in 28% of all captures (46). Tank bromeliads can hold water year-round, allowing arboreal species to access a wet microhabitat in a dry forest without moving outside of the forest edge. Hence, B. faber could be escaping the direct influence of habitat split on Bd by occupying tank bromeliads, allowing for the continuous exposure to Bd reservoirs (97) that could enhance the development of acquired resistance (34). However, these aquatic microcosms are highly simplified compared to lentic and lotic water bodies and thus may not allow for the persistence of diverse and functional microbial communities. Together, these mechanisms, which can dysregulate both innate and adaptive immunity, likely contribute to the pattern we observed beyond microbiome shifts alone, as they may help explain the observed neutral association between habitat split and Bd infection in B. faber.
Landscape-level habitat disturbances have been shown to impair both innate and adaptive immunity in amphibians, with local populations experiencing, for example, a reduction in MHC diversity (96) and elevated chronic stress (46); both of which may act independently or in synergy with microbiome shifts (7). Importantly, disruptions to spatial connectivity that affect multiple layers of amphibian immune responses are also likely to impact entire food webs by altering predator–prey dynamics (98), with potential implications for amphibian–Bd interactions. For instance, breeding habitat characteristics (including those resulting from disturbances), such as degree of exposure to ultraviolet radiation, zooplankton density (putative Bd predators), and host life stage, may modulate Bd infection rates through direct and indirect pathways (99). Additionally, higher pathogen loads may increase frogs’ vulnerability to predation, especially in open environments where they are more conspicuous and prone to dehydration; factors that could selectively eliminate heavily infected individuals (100) and help explain observed population declines associated with habitat split (3). Conversely, if natural frog predators decline in less connected, disturbed environments, heavily infected individuals may survive longer and shed pathogens over extended periods, potentially reinforcing Bd transmission cycles in areas of high habitat split. Because environmental disturbances such as habitat split can influence multiple host physiological and ecological processes that may modulate microbiome function (46), we interpret our results as reflecting the net ecological effects of habitat split on microbiome defenses rather than the action of any single mechanism.
In summary, isolation in forest fragments and habitat split may elevate disease risk in amphibians by limiting the assembly of a healthy host-associated microbiome and, at the same time, contributing to declines in immune function through stress, genetic erosion, and other factors (46, 101–103). These impacts are also widespread across a variety of taxa; for instance, habitat split affects movement patterns in migratory seabirds, caribou, and dragonflies, likely influencing pathogen exposure and the recruitment of mutualistic microbes (20, 104–108). Similarly, reduced connectivity between freshwater and marine environments in migratory salmon likely alters pathogen exposure and acquired resistance to viral infections, which impacts the host-associated microbiome (108, 109). Understanding the variation in microbial filtering and diversity of microbial interactions among amphibians and other vertebrates is crucial for predicting how habitat split, and other global change stressors, affect the stability and function of host-associated microbial communities.
Our research extends the boundaries of vertebrate ecology by exploring the relationships between host microbiome assembly, processes at individual host and population scales, and their consequences for disease resistance and maintenance of biological diversity. These findings help define the ecological processes driving the epidemiology of environmentally transmitted microorganisms, including both pathogens and beneficial symbionts. Our focus is broadly applicable to understanding how multiple scales of habitat disturbances and disruption of spatial connectivity shape an increasingly recognized component of host defense. Identifying the underlying mechanisms that promote the recruitment and selection of host microbiomes with pathogen-suppressive activity amid varying spatial patterns of multihabitat connectivity is an important future step toward implementing habitat restoration strategies aimed at safeguarding increasingly threatened global biodiversity.
Ethics Statement.
Sampling procedures were approved by the Universidade Federal de São Carlos Animal Ethics Committee (protocol 2464230518). Our research was conducted under the permits: SISBIO: 63917; COTEC: 260108‐005.880/2018; and SISGEN: A663126, AC2F7B2, AC88222.
Supplementary Material
Appendix 01 (PDF)
Acknowledgments
We thank Dr. Raúl A. González-Pech for the technical support with submitting statistical models to the supercomputer. C.G.B. thanks The NSF (IOS‐1947681, DEB-2227340, DEB-2413542, BII-2120084) for funding. C.F.B.H. thanks São Paulo Research Foundation (FAPESP) for financial support (proc #2021/10639-5) and the Brazilian National Council for Scientific and Technological Development (CNPq) for a research fellowship. D.C.W. thanks the NSF (IOS-1845634 and BII-2120084) for funding.
Author contributions
C.G.B. designed research; D.M., R.A.M., P.R.P., S.B., W.J.N., L.K.S., S.E.G., M.L.L., P.J.K., D.C.W., M.C.B., V.A.S.-P., and C.F.B.H. performed research; D.M. analyzed data; and D.M., S.B., W.J.N., and C.G.B. wrote the paper.
Competing interests
The authors declare no competing interest.
Footnotes
This article is a PNAS Direct Submission.
Contributor Information
Daniel Medina, Email: dmedina@fieldstudies.org.
C. Guilherme Becker, Email: guibecker@psu.edu.
Data, Materials, and Software Availability
Sequences, metafile, sOTU table, and R-script data have been deposited in Zenodo (https://doi.org/10.5281/zenodo.19325102) (110) and NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra, under accession number PRJNA1375872) (111).
Supporting Information
References
- 1.Daszak P., Cunningham A. A., Hyatt A. D., Emerging infectious diseases of wildlife–threats to biodiversity and human health. Science 287, 443–449 (2000). [DOI] [PubMed] [Google Scholar]
- 2.IUCN, The IUCN red list of threatened species (2022). https://www.iucnredlist.org. Accessed 26 July 2022.
- 3.Becker C. G., Fonseca C. R., Haddad C. F. B., Batista R. F., Prado P. I., Habitat split and the global decline of amphibians. Science 318, 1775–1777 (2007). [DOI] [PubMed] [Google Scholar]
- 4.Pope S. E., Fahrig L., Merriam H. G., Landscape complementation and metapopulation effects on leopard frog populations. Ecology 81, 2498–2508 (2000). [Google Scholar]
- 5.Teitelbaum C. S., Huang S., Hall R. J., Altizer S., Migratory behaviour predicts greater parasite diversity in ungulates. Proc. R. Soc. B Biol. Sci. 285, 20180089 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Hess G., Disease in metapopulation models: Implications for conservation. Ecology 77, 1617–1632 (1996). [Google Scholar]
- 7.Becker C. G., et al. , Habitat split as a driver of disease in amphibians. Biol. Rev. 98, 727–746 (2023). [DOI] [PubMed] [Google Scholar]
- 8.Harris R. N., et al. , Skin microbes on frogs prevent morbidity and mortality caused by a lethal skin fungus. ISME J. 3, 818–824 (2009). [DOI] [PubMed] [Google Scholar]
- 9.Libertucci J., Young V. B., The role of the microbiota in infectious diseases. Nat. Microbiol. 4, 35–45 (2019). [DOI] [PubMed] [Google Scholar]
- 10.Woodhams D. C., et al. , Probiotics modulate a novel amphibian skin defense peptide that is antifungal and facilitates growth of antifungal bacteria. Microb. Ecol. 79, 192–202 (2020). [DOI] [PubMed] [Google Scholar]
- 11.Woodhams D. C., et al. , Symbiotic bacteria contribute to innate immune defenses of the threatened mountain yellow-legged frog, Rana muscosa. Biol. Conserv. 138, 390–398 (2007). [Google Scholar]
- 12.Petersen C., Round J. L., Defining dysbiosis and its influence on host immunity and disease. Cell. Microbiol. 16, 1024–1033 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Rosado P. M., et al. , Marine probiotics: Increasing coral resistance to bleaching through microbiome manipulation. ISME J. 13, 921–936 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Singh V., et al. , Dysregulated microbial fermentation of soluble fiber induces cholestatic liver cancer. Cell 175, 679–694.e22 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Jin Song S., et al. , Engineering the microbiome for animal health and conservation. Exp. Biol. Med. 244, 494–504 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Zilber-Rosenberg I., Rosenberg E., Role of microorganisms in the evolution of animals and plants: The hologenome theory of evolution. FEMS Microbiol. Rev. 32, 723–735 (2008). [DOI] [PubMed] [Google Scholar]
- 17.Woodhams D. C., McCartney J., Walke J. B., Whetstone R., The adaptive microbiome hypothesis and immune interactions in amphibian mucus. Dev. Comp. Immunol. 145, 104690 (2023), 10.1016/j.dci.2023.104690. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Brearley G., et al. , Wildlife disease prevalence in human-modified landscapes. Biol. Rev. Camb. Philos. Soc. 88, 427–442 (2013). [DOI] [PubMed] [Google Scholar]
- 19.Diuk-Wasser M. A., VanAcker M. C., Fernandez M. P., Impact of land use changes and habitat fragmentation on the eco-epidemiology of tick-borne diseases. J. Med. Entomol. 58, 1546–1564 (2021). [DOI] [PubMed] [Google Scholar]
- 20.Mony C., Uroy L., Khalfallah F., Haddad N., Vandenkoornhuyse P., Landscape connectivity for the invisibles. Ecography 2022, e06041 (2022). [Google Scholar]
- 21.Rulli M. C., Santini M., Hayman D. T. S., D’Odorico P., The nexus between forest fragmentation in Africa and Ebola virus disease outbreaks. Sci. Rep. 7, 41613 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Amato K. R., et al. , Habitat degradation impacts black howler monkey (Alouatta pigra) gastrointestinal microbiomes. ISME J. 7, 1344–1353 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Bahrndorff S., Alemu T., Alemneh T., Lund Nielsen J., The microbiome of animals: Implications for conservation biology. Int. J. Genomics 2016, 5304028 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Chang C.-W., Huang B.-H., Lin S.-M., Huang C.-L., Liao P.-C., Changes of diet and dominant intestinal microbes in farmland frogs. BMC Microbiol. 16, 33 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Teyssier A., et al. , Inside the guts of the city: Urban-induced alterations of the gut microbiota in a wild passerine. Sci. Total Environ. 612, 1276–1286 (2018). [DOI] [PubMed] [Google Scholar]
- 26.Becker C. G., Fonseca C. R., Haddad C. F. B., Prado P. I., Habitat split as a cause of local population declines of amphibians with aquatic larvae. Conserv. Biol. 24, 287–294 (2010). [DOI] [PubMed] [Google Scholar]
- 27.Scheele B. C., et al. , Amphibian fungal panzootic causes catastrophic and ongoing loss of biodiversity. Science 363, 1459–1463 (2019). [DOI] [PubMed] [Google Scholar]
- 28.Belasen A. M., Russell I. D., Zamudio K. R., Bletz M. C., Endemic lineages of Batrachochytrium dendrobatidis are associated with reduced chytridiomycosis-induced mortality in amphibians: Evidence from a meta-analysis of experimental infection studies. Front. Vet. Sci. 9, 756686 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Becker C. G., Rodriguez D., Longo A. V., Talaba A. L., Zamudio K. R., Disease risk in temperate amphibian populations is higher at closed-canopy sites. PLoS One 7, e48205 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Becker C. G., Zamudio K. R., Tropical amphibian populations experience higher disease risk in natural habitats. Proc. Natl. Acad. Sci. U.S.A. 108, 9893–9898 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Rohr J. R., Raffel T. R., Linking global climate and temperature variability to widespread amphibian declines putatively caused by disease. Proc. Natl. Acad. Sci. U.S.A. 107, 8269–8274 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Haddad C. F. B., et al. , Guide to the Amphibians of the Atlantic Forest: Diversity and Biology (Anolis Books, 2013). [Google Scholar]
- 33.Haddad C. F. B., Prado C. P. A., Reproductive modes in frogs and their unexpected diversity in the Atlantic Forest of Brazil. BioScience 55, 207–217 (2005). [Google Scholar]
- 34.McMahon T. A., et al. , Amphibians acquire resistance to live and dead fungus overcoming fungal immunosuppression. Nature 511, 224–227 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Siomko S. A., et al. , Selection of an anti-pathogen skin microbiome following prophylaxis treatment in an amphibian model system. Philos. Trans. R. Soc. Lond. B Biol. Sci. 378, 20220126 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Becker M. H., et al. , Composition of symbiotic bacteria predicts survival in Panamanian golden frogs infected with a lethal fungus. Proc. R. Soc. B Biol. Sci. 282, 20142881 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Muletz-Wolz C. R., Fleischer R. C., Lips K. R., Fungal disease and temperature alter skin microbiome structure in an experimental salamander system. Mol. Ecol. 28, 2917–2931 (2019). [DOI] [PubMed] [Google Scholar]
- 38.Loyau A., Bouchali R., Sentenac H., Schmeller D. S., The commensal skin microbiome of amphibian mountain populations and its association with the pathogen Batrachochytrium dendrobatidis. Environ. Microbiol. 26, e16699 (2024). [DOI] [PubMed] [Google Scholar]
- 39.Rebollar E. A., et al. , Skin bacterial diversity of Panamanian frogs is associated with host susceptibility and presence of Batrachochytrium dendrobatidis. ISME J. 10, 1682–1695 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Jani A. J., Knapp R. A., Briggs C. J., Epidemic and endemic pathogen dynamics correspond to distinct host population microbiomes at a landscape scale. Proc. R. Soc. B Biol. Sci. 284, 20170944 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Viana V. M., Tabanez A. A. J., Batista J. L., “Restoration and management of fragmented landscapes” in Tropical Forest Remnants: Ecology, Management and Conservation of Fragmented Communities, Laurance W. F., Bierregaard R. O., Eds. (University of Chicago Press, 1997), pp. 347–365. [Google Scholar]
- 42.Mueller E. A., Wisnoski N. I., Peralta A. L., Lennon J. T., Microbial rescue effects: How microbiomes can save hosts from extinction. Funct. Ecol. 34, 2055–2064 (2020). [Google Scholar]
- 43.Becker C. G., Longo A. V., Haddad C. F. B., Zamudio K. R., Land cover and forest connectivity alter the interactions among host, pathogen and skin microbiome. Proc. R. Soc. B Biol. Sci. 284, 20170582 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Hughey M. C., et al. , Skin bacterial microbiome of a generalist Puerto Rican frog varies along elevation and land use gradients. PeerJ 2017, e3688 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Jiménez R. R., Alvarado G., Sandoval J., Sommer S., Habitat disturbance influences the skin microbiome of a rediscovered neotropical-montane frog. BMC Microbiol. 20, 292 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Neely W. J., et al. , Linking microbiome and stress hormone responses in wild tropical treefrogs across continuous and fragmented forests. Commun. Biol. 6, 1–15 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Neely W. J., et al. , Habitat disturbance linked with host microbiome dispersion and Bd dynamics in temperate amphibians. Microb. Ecol. 84, 901–910 (2022). [DOI] [PubMed] [Google Scholar]
- 48.Ahmed H. I., Herrera M., Liew Y. J., Aranda M., Long-term temperature stress in the coral model Aiptasia supports the “Anna Karenina Principle” for bacterial microbiomes. Front. Microbiol. 10, 975 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Zaneveld J. R., McMinds R., Thurber R. V., Stress and stability: Applying the Anna Karenina principle to animal microbiomes. Nat. Microbiol. 2, 17121 (2017). [DOI] [PubMed] [Google Scholar]
- 50.Galindo-Leal C., Camara I. D. B., The Atlantic Forest of South America: Biodiversity status, threats, and outlook. Electron. Green J. 1, 953–955 (2003). [Google Scholar]
- 51.Woodhams D. C., et al. , Antifungal isolates database of amphibian skin-associated bacteria and function against emerging fungal pathogens. Ecology 96, 595–595 (2015). [Google Scholar]
- 52.Mesquita A. F. C., et al. , Low resistance to chytridiomycosis in direct-developing amphibians. Sci. Rep. 7, 16605 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.Martins R. A., et al. , Signatures of functional bacteriome structure in a tropical direct-developing amphibian species. Anim. Microbiome 4, 40 (2022), 10.1186/s42523-022-00188-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Castro-Herrera F., Rueda J., Cortez C., “Técnicas para el inventario y muestreo de anfibios: Una compilación” in Manuales Para La Conservación, Técnicas De Inventario Y Monitoreo Para Los Anfibios De La Región Tropical Andina, Angulo A., Rueda-Almonacid J. V., Rodríguez-Mahecha J. V., La Marca E., Eds. (Conservación Internacional, 2006), pp. 135–172. [Google Scholar]
- 55.Hyatt A. D., et al. , Diagnostic assays and sampling protocols for the detection of Batrachochytrium dendrobatidis. Dis. Aquat. Organ. 73, 175–192 (2007). [DOI] [PubMed] [Google Scholar]
- 56.Boyle D. G., Boyle D. B., Olsen V., Morgan J. A. T., Hyatt A. D., Rapid quantitative detection of chytridiomycosis (Batrachochytrium dendrobatidis) in amphibian samples using real-time Taqman PCR assay. Dis. Aquat. Organ. 60, 141–148 (2004). [DOI] [PubMed] [Google Scholar]
- 57.Rebollar E. A., Woodhams D. C., LaBumbard B., Kielgast J., Harris R. N., Prevalence and pathogen load estimates for the fungus Batrachochytrium dendrobatidis are impacted by ITS DNA copy number variation. Dis. Aquat. Organ. 123, 213–226 (2017). [DOI] [PubMed] [Google Scholar]
- 58.Caporaso J. G., et al. , Ultra-high-throughput microbial community analysis on the Illumina HiSeq and MiSeq platforms. ISME J. 6, 1621–1624 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 59.Kozich J. J., Westcott S. L., Baxter N. T., Highlander S. K., Schloss P. D., Development of a dual-index sequencing strategy and curation pipeline for analyzing amplicon sequence data on the MiSeq Illumina sequencing platform. Appl. Environ. Microbiol. 79, 5112–5120 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 60.Bolyen E., et al. , Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol. 37, 852–857 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61.Amir A., et al. , Deblur rapidly resolves single-nucleotide community sequence patterns. mSystems 2, 759 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 62.Katoh K., Misawa K., Kuma K., Miyata T., MAFFT: A novel method for rapid multiple sequence alignment based on fast fourier transform. Nucleic Acids Res. 30, 3059–3066 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 63.Price M. N., Dehal P. S., Arkin A. P., FastTree 2—Approximately maximum-likelihood trees for large alignments. PLoS One 5, e9490 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 64.Bokulich N. A., et al. , Optimizing taxonomic classification of marker-gene amplicon sequences with QIIME 2’s q2-feature-classifier plugin. Microbiome 6, 90 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 65.DeSantis T. Z., et al. , Greengenes, a chimera-checked 16S rRNA gene database and workbench compatible with ARB. Appl. Environ. Microbiol. 72, 5069–5072 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66.McDonald D., et al. , An improved Greengenes taxonomy with explicit ranks for ecological and evolutionary analyses of bacteria and archaea. ISME J. 6, 610–618 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 67.Bokulich N. A., et al. , Quality-filtering vastly improves diversity estimates from Illumina amplicon sequencing. Nat. Methods 10, 57–59 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 68.Bell S. C., Alford R. A., Garland S., Padilla G., Thomas A. D., Screening bacterial metabolites for inhibitory effects against Batrachochytrium dendrobatidis using a spectrophotometric assay. Dis. Aquat. Organ. 103, 77–85 (2013). [DOI] [PubMed] [Google Scholar]
- 69.Piovia-Scott J., et al. , Greater species richness of bacterial skin symbionts better suppresses the amphibian fungal pathogen Batrachochytrium dendrobatidis. Microb. Ecol. 74, 217–226 (2017). [DOI] [PubMed] [Google Scholar]
- 70.Chen M. Y., Alexiev A., McKenzie V. J., Bacterial biofilm thickness and fungal inhibitory bacterial richness both prevent establishment of the amphibian fungal pathogen Batrachochytrium dendrobatidis. Appl. Environ. Microbiol. 88, e01604-21 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 71.Brooks M. E., et al. , glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. R J. 9, 378–400 (2017). [Google Scholar]
- 72.Hartig F., Lohse L., DHARMa: Residual diagnostics for hierarchical (multi-level/mixed) regression models (2022). 10.32614/CRAN.package.DHARMa. Deposited 8 September 2022. [DOI]
- 73.Lüdecke D., Ben-Shachar M. S., Patil I., Waggoner P., Makowski D., performance: An R package for assessment, comparison and testing of statistical models. J. Open Source Softw. 6, 3139 (2021). [Google Scholar]
- 74.Dormann C. F., et al. , Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography 36, 27–46 (2013). [Google Scholar]
- 75.Bates D., Mächler M., Bolker B., Walker S., Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015). [Google Scholar]
- 76.Sarkar D., Lattice: Multivariate Data Visualization with R (Springer, ed. 1, 2008). [Google Scholar]
- 77.Zuur A., Ieno E. N., Walker N., Saveliev A. A., Smith G. M., Mixed Effects Models and Extensions in Ecology with R (Springer, Softcover reprint of hardcover 1st ed. 2009 edition, 2011). [Google Scholar]
- 78.Ovaskainen O., et al. , How to make more out of community data? A conceptual framework and its implementation as models and software. Ecol. Lett. 20, 561–576 (2017). [DOI] [PubMed] [Google Scholar]
- 79.Ovaskainen O., Abrego N., Joint Species Distribution Modelling: With Applications in R (Cambridge University Press, ed. 1, 2020). [Google Scholar]
- 80.Odriozola I., et al. , Fungal communities are important determinants of bacterial community composition in deadwood. mSystems 6, 1–12 (2021), 10.1128/mSystems.01017-20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 81.Abrego N., et al. , Fungal communities decline with urbanization more in air than in soil. ISME J. 14, 2806–2815 (2020), 10.1038/s41396-020-0732-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 82.Tikhonov G., et al. , Joint species distribution modelling with the R-package Hmsc. Methods Ecol. Evol. 11, 442–447 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 83.Manel S., Williams H. C., Ormerod S. J., Evaluating presence–absence models in ecology: The need to account for prevalence. J. Appl. Ecol. 38, 921–931 (2001). [Google Scholar]
- 84.Swets J. A., Measuring the accuracy of diagnostic systems. Science 240, 1285–1293 (1988). [DOI] [PubMed] [Google Scholar]
- 85.Araujo G., Montoya J. M., Thomas T., Webster N. S., Lurgi M., A mechanistic framework for complex microbe-host symbioses. Trends Microbiol. 33, 96–111 (2025). [DOI] [PubMed] [Google Scholar]
- 86.Buttimer S., Hernández-Gómez O., Rosenblum E. B., Skin bacterial metacommunities of San Francisco Bay Area salamanders are structured by host genus and habitat quality. FEMS Microbiol. Ecol. 97, fiab162 (2021). [DOI] [PubMed] [Google Scholar]
- 87.Leibold M. A., et al. , The metacommunity concept: A framework for multi-scale community ecology. Ecol. Lett. 7, 601–613 (2004). [Google Scholar]
- 88.Mihaljevic J. R., Linking metacommunity theory and symbiont evolutionary ecology. Trends Ecol. Evol. 27, 323–329 (2012). [DOI] [PubMed] [Google Scholar]
- 89.Muletz C. R., Myers J. M., Domangue R. J., Herrick J. B., Harris R. N., Soil bioaugmentation with amphibian cutaneous bacteria protects amphibian hosts from infection by Batrachochytrium dendrobatidis. Biol. Conserv. 152, 119–126 (2012). [Google Scholar]
- 90.Stegen J. C., Lin X., Fredrickson J. K., Konopka A. E., Estimating and mapping ecological processes influencing microbial community assembly. Front. Microbiol. 6, 370 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 91.Zhou J., Ning D., Stochastic community assembly: Does it matter in microbial ecology? Microbiol. Mol. Biol. Rev. 81, e00002-17 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 92.Tracey J. A., Bevins S. N., VandeWoude S., Crooks K. R., An agent-based movement model to assess the impact of landscape fragmentation on disease transmission. Ecosphere 5, 1–24 (2014). [Google Scholar]
- 93.Harrison X. A., et al. , Diversity-stability dynamics of the amphibian skin microbiome and susceptibility to a lethal viral pathogen. Front. Microbiol. 10, 2883 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 94.Biggs C. R., et al. , Does functional redundancy affect ecological stability and resilience? A review and meta-analysis. Ecosphere 11, e03184 (2020). [Google Scholar]
- 95.Tilman D., Downing J. A., Biodiversity and stability in grasslands. Nature 367, 363–365 (1994). [Google Scholar]
- 96.Belasen A. M., Bletz M. C., Leite D. D. S., Toledo L. F., James T. Y., Long-term habitat fragmentation is associated with reduced MHC IIB diversity and increased infections in amphibian hosts. Front. Ecol. Evol. 6, 236 (2019), 10.3389/fevo.2018.00236. [DOI] [Google Scholar]
- 97.Ruano-Fajardo G., Toledo L., Mott T., Jumping into a trap: High prevalence of chytrid fungus in the preferred microhabitats of a bromeliad-specialist frog. Dis. Aquat. Organ. 121, 223–232 (2016). [DOI] [PubMed] [Google Scholar]
- 98.Pires M. M., Benchimol M., Cruz L. R., Peres C. A., Terrestrial food web complexity in Amazonian forests decays with habitat loss. Curr. Biol. 33, 389–396.e3 (2023). [DOI] [PubMed] [Google Scholar]
- 99.Hite J. L., Bosch J., Fernández-Beaskoetxea S., Medina D., Hall S. R., Joint effects of habitat, zooplankton, host stage structure and diversity on amphibian chytrid. Proc. R. Soc. B Biol. Sci. 283, 20160832–20160910 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 100.Packer C., Holt R. D., Hudson P. J., Lafferty K. D., Dobson A. P., Keeping the herds healthy and alert: Implications of predator control for infectious disease. Ecol. Lett. 6, 797–802 (2003). [Google Scholar]
- 101.Belasen A. M., et al. , Habitat fragmentation in the Brazilian Atlantic Forest is associated with erosion of frog immunogenetic diversity and increased fungal infections. Immunogenetics 74, 431–441 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 102.Bernardo-Cravo A. P., Schmeller D. S., Chatzinotas A., Vredenburg V. T., Loyau A., Environmental factors and host microbiomes shape host-pathogen dynamics. Trends Parasitol. 36, 616–633 (2020), 10.1016/j.pt.2020.04.010. [DOI] [PubMed] [Google Scholar]
- 103.Rebollar E. A., Martínez-Ugalde E., Orta A. H., The amphibian skin microbiome and its protective role against chytridiomycosis. Herpetologica 76, 167–212 (2020). [Google Scholar]
- 104.Altizer S., Bartel R., Han B. A., Animal migration and infectious disease risk. Science 331, 296–302 (2011). [DOI] [PubMed] [Google Scholar]
- 105.Becker D. J., et al. , Macroimmunology: The drivers and consequences of spatial patterns in wildlife immune defence. J. Anim. Ecol. 89, 972–995 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 106.Dingle H., Migration: The Biology of Life on the Move (Oxford University Press, ed. 2, 2014). [Google Scholar]
- 107.Lennox R. J., et al. , Conservation physiology of animal migration. Conserv. Physiol. 4, cov072 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 108.Reid K. M., et al. , Salmonid alphavirus infection causes skin dysbiosis in Atlantic salmon (Salmo salar L.) post-smolts. PLoS One 12, e0172856 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 109.Twardek W. M., Evidence of a hydraulically challenging reach serving as a barrier for the upstream migration of infection-burdened adult steelhead. Conserv. Physiol. 7, coz023 (2019), 10.1093/conphys/coz023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 110.Medina D., Linking frogs skin bacteria diversity, putative anti-Bd function and habitat split [Data set]. Zenodo. 10.5281/zenodo.19325102. Deposited 29 March 2026. [DOI]
- 111.Schuck L. K., Connecting habitat frogs. NCBI. https://www.ncbi.nlm.nih.gov/bioproject/1375872. Deposited 5 December 2025.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix 01 (PDF)
Data Availability Statement
Sequences, metafile, sOTU table, and R-script data have been deposited in Zenodo (https://doi.org/10.5281/zenodo.19325102) (110) and NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra, under accession number PRJNA1375872) (111).




