Arthropod–bacteria interactions influence assembly of aquatic host microbiome and pathogen defense

Abstract

The host-associated microbiome is vital to host immunity and pathogen defense. In aquatic ecosystems, organisms may interact with environmental bacteria to influence the pool of potential symbionts, but the effects of these interactions on host microbiome assembly and pathogen resistance are unresolved. We used replicated bromeliad microecosystems to test for indirect effects of arthropod–bacteria interactions on host microbiome assembly and pathogen burden, using tadpoles and the fungal amphibian pathogen Batrachochytrium dendrobatidis as a model host–pathogen system. Arthropods influenced host microbiome assembly by altering the pool of environmental bacteria, with arthropod–bacteria interactions specifically reducing host colonization by transient bacteria and promoting antimicrobial components of aquatic bacterial communities. Arthropods also reduced fungal zoospores in the environment, but fungal infection burdens in tadpoles corresponded most closely with arthropod-mediated patterns in microbiome assembly. This result indicates that the cascading effects of arthropods on the maintenance of a protective host microbiome may be more strongly linked to host health than negative effects of arthropods on pools of pathogenic zoospores. Our work reveals tight links between healthy ecosystem dynamics and the functioning of host microbiomes, suggesting that ecosystem disturbances such as loss of arthropods may have downstream effects on host-associated microbial pathogen defenses and host fitness.

1. Introduction

Animals carry complex communities of symbiotic microorganisms that influence host–pathogen interactions [1,2]. These microbial assemblages (the microbiome) comprise a mix of organisms that range from transient to permanent and from beneficial to detrimental [2]. The host microbiome may positively influence host health by outcompeting invading pathogens, producing pathogen-inhibiting metabolites and regulating host immune functions [1,3]. Alternatively, imbalances in the host microbiome (dysbiosis) may increase susceptibility to disease [4,5]. For example, the presence of a distinct gut microbiota protected bumblebees against intestinal parasites, whereas increased diversity of transient taxa in the bee microbiome predisposed hosts to parasite infection [6,7]. Thus, understanding the formation, maintenance and functions of host microbiomes is critical for predicting host disease outcomes.

Assembly of the host microbiome may be influenced by vertical and horizontal transmission, the chemistry of host tissues, innate and adaptive immunity and environmental source pools [8]. Environmental bacterial reservoirs may be particularly important to the microbiomes of aquatic hosts because the whole organism is in continual contact with bacteria in the water column [9]. Biotic community interactions are important drivers of environmental bacterial community structure [10,11]. For example, invertebrate detritivores are closely associated with environmental bacteria in aquatic ecosystems, and together they play a pivotal role in ecosystem functions such as detritus processing and nutrient cycling, with invertebrates both consuming and providing nutrients to bacterial communities [10,12]. However, the cascading effects of these biotic interactions on host microbiome structure and pathogen resistance are not well understood.

Tank bromeliads are ideal model systems for investigating cascading effects of aquatic community dynamics on host-associated bacterial assemblages and pathogen defense. Tank bromeliads are naturally occurring, replicable microecosystems known for the rich and distinctive biotic communities that inhabit their water-holding leaf bases [13,14]. Even relatively small bromeliads can support fully functional microecosystems, with resident organisms forming complex food webs [12,15]. Bacterial communities of bromeliads are just beginning to be described, but recent studies have documented high taxonomic diversity and have recognized these organisms as fundamental to carbon, nutrient and energy cycles in tropical forests [16,17]. An abundant arthropod fauna of bromeliads consists primarily of crustaceans, aquatic hemipterans, worms, leeches and larval and adult insects occupying functional roles ranging from herbivores and detritivores to top predators [18]. Bromeliads also function as obligate habitats and refuges for many amphibian species [19,20]. Amphibians may complete their whole life cycle in bromeliads, with their trophic position changing from omnivorous at the tadpole stage to carnivorous after metamorphosis [21,22].

Amphibians host complex internal (gut) and external (skin) bacterial communities [23,24]. Assembly of the amphibian microbiome begins at hatching [25]. At metamorphosis, shifts in the immune system, diet and skin structure may lead to changes in bacterial community structure [26]. Throughout development, the microbiome of amphibians may contribute to pathogen defense [26]. For example, bacterial assemblies on amphibians may inhibit the fungal amphibian pathogen Batrachochytrium dendrobatidis (Bd) [27,28]. Bd infects the mouthparts of tadpoles and the skin of post-metamorphic frogs [29,30]. Infections build exponentially as swimming zoospores encyst on the host surface and transfer the fungus inside host cells, where zoosporangia develop and eventually release new generations of zoospores to the environment [31,32]. In general, host damage from Bd is dose-dependent [33]. Bd is usually sublethal to tadpoles but infections may spike to lethal levels immediately after metamorphosis and sublethal fitness effects of infection, such as reduced growth of tadpoles, can increase mortality risk in later life stages [34]. Thus, healthy functioning of the microbiome at the tadpole stage may be critical in maintaining good body condition and regulating pathogen burdens carried through metamorphosis [26].

The composition of amphibian bacterial assemblages may be influenced by environmental factors such as water source and environmental bacterial reservoirs [35,36], with subsequent effects on host resistance to Bd infection [37–39]. Conversely, Bd infections may disturb the host microbiome [40]. Trophic interactions such as consumption of Bd zoospores by aquatic organisms have been shown to alter Bd infection dynamics [41–43], but the ways in which aquatic community dynamics may shape the amphibian microbiome and pathogen defense are unclear. We investigated the indirect effects of aquatic community dynamics (arthropod–bacteria interactions) on host bacterial composition and fungal pathogen burden in bromeliad-associated tadpoles. In a controlled laboratory experiment, we stocked nursery-raised bromeliads with ‘natural’ water (water collected from wild bromeliads that contained live bacteria and microeukaryotes), tadpoles and a Bd inoculum. We then compared environmental bacterial assemblages, host bacterial assemblages and Bd infection intensity between bromeliads with and without communities of aquatic arthropod detritivores. To provide an alternate bacterial reservoir for comparative purposes, we replicated this design with bromeliads stocked with ‘pre-sterilized’ water (bromeliad water that we autoclaved prior to stocking bromeliads). We predicted that in addition to trophic interactions among aquatic organisms and Bd (arthropod–Bd, microeukaryote–Bd, arthropod–microeukaryote–Bd), arthropods could also indirectly influence the severity of Bd infections through a cascade of effects in which (i) arthropod–bacteria interactions influence environmental bacterial communities, (ii) environmental bacterial communities influence the host microbiome and (iii) the host microbiome influences host resistance to Bd. Our results suggest that the biotic interactions forming the basis for healthy ecosystem function also support beneficial host bacterial symbioses that are essential to pathogen defense.

2. Material and methods

(a). Experimental procedures

We obtained potted Neoregelia compacta tank bromeliads from a nursery in Campinas, São Paulo, Brazil. To remove organisms and detritus, bromeliads were rinsed and air-dried. We obtained one-week-old tadpoles (Gosner's stage 25 [44]) of the bromeliad-specialist treefrog Ololygon alcatraz (Anura, Hylidae) from a Bd-free ex situ conservation breeding programme at Fundação Parque Zoológico de São Paulo, SP, Brazil. Using young, captive-bred amphibians ensured that initial host microbiomes were homogeneous and that microbiomes were actively assembling. For the Bd inoculum, we used the isolate CLFT 196 (Global Panzootic Lineage-2), which originated from a tadpole collected in São Paulo state, Brazil [45]. To obtain Bd zoospores for the inoculum, we added a growing liquid Bd culture to Petri plates containing 1% tryptone agar. Plates were incubated at 17°C for 9 days leading up to the experiment. We then added distilled water to the plates to form a zoospore suspension, collected the suspension and calculated the concentration of zoospores with a haemocytometer.

We used a 2 × 2 fully factorial experimental design with water treatment (natural or pre-sterilized) crossed with aquatic arthropod detritivores (present or absent). We used 12 replicate bromeliads for each of the four treatment combinations. We arranged the bromeliads in a blocked design, with 12 spatial blocks each containing one bromeliad from each treatment combination. The location of each bromeliad within each block was determined randomly. We collected water and arthropods for the experiment from wild bromeliads in a forest fragment in Campinas, São Paulo, Brazil (lat: −22.8316°, long: −46.9651°), containing vegetation representative of Brazil's Atlantic Forest. The water was collected with pipettes, filtered with 0.125-mm geological sieves and homogenized. For the pre-sterilized water treatment, we autoclaved the wild-collected water. We used the three most abundant aquatic detritivore species and selected the number of each species to be added to each bromeliad based on their relative abundances in the field. The three species were the microcrustacean Elpidium bromeliarum (Limnocytheridae; collector; n = 10), larvae of the mosquito Culex sp. (Culicidae; filter-feeder; n = 5) and larvae of the biting midge Forcipomyia (Phytohelea) sp. (Ceratopogonidae; collector; n = 4), which collectively feed on fine and coarse particulate matter, algae, microbes including fungi and bacteria, and smaller invertebrates throughout the water column [18,46,47]. Water and arthropods were stored at 20°C overnight prior to stocking bromeliads.

On day 0, we added the water treatment (100 ml of natural or pre-sterilized water), the arthropod treatment (present or absent), one randomly assigned O. alcatraz tadpole and a standardized zoospore inoculum (1.5 ml of the zoospore suspension at 3.22 × 10⁶ zoospores per ml) to each bromeliad. The experiment was performed under controlled laboratory conditions (20°C on a 12 h day/night cycle). We fed tadpoles every 3 d with dried and autoclaved Spirulina (10 mg). The Spirulina also served as a detritus source for the bromeliad community; the bromeliad species we used typically do not accumulate large fragments of leaf litter in the wild. On day 13, we quantified microeukaryotes (ciliates, flagellates and rotifers) in each bromeliad containing natural water. For each bromeliad, we collected 1 ml of water, fixed and stained the sample with glycerol and Rose Bengal, counted microeukaryotes in five 20-µl subsamples under a compound microscope and averaged the five counts. On day 17, we sampled environmental bacterial communities and Bd zoospore densities in each bromeliad by swabbing the submerged, inner leaf surfaces with a sterile swab. We also measured water pH (PCD-650, Oakton Instruments, Vernon Hills, IL, USA) and chlorophyll a concentration (Aquafluore, Turner Designs, San Jose, CA, USA) in each bromeliad. We then measured tadpole body mass as an estimate of sublethal fitness effects of Bd infection and euthanized tadpoles with an overdose of MS-222. Tadpoles and swabs were stored at −20°C until further processing.

Because bromeliads were not sterilized before the experiment and were held under open-air laboratory conditions during the experiment, we expected that bromeliads carried residual bacteria and that bromeliads would be colonized by ambient bacteria from the laboratory, but the differences between the natural and pre-sterilized water created divergent environmental bacterial reservoirs that were distinguishable throughout the experiment. Specifically, environmental bacterial communities in bromeliads containing natural water had higher phylogenetic bacterial diversity (figure 1a; electronic supplementary material, table S1), higher operational taxonomic unit (OTU) richness (table 1 and electronic supplementary material, table S1) and lower proportions of putatively Bd-inhibitory taxa (electronic supplementary material, table S1) than pre-sterilized water (see below). The composition of environmental bacterial communities also differed between natural and pre-sterilized water (figure 2; electronic supplementary material, table S1; see below).

Figure 1.

Characteristics of the aquatic environment (a,b) and tadpole microbiomes (c,d) including average of Faith's phylogenetic bacterial diversity (a,c) and average logged Bd zoospore genomic equivalents detected (b,d). Error bars represent s.e. (Online version in colour.)

Table 1.

For bromeliad microecosystems containing tadpoles exposed to Batrachochytrium dendrobatidis, a water treatment (natural or pre-sterilized), and an arthropod treatment (present or absent) for 17 days: bacterial OTUs detected in the aquatic environment (total 1237 OTUs from 21 phyla) and in tadpole microbiomes (total 860 OTUs from 16 phyla). Nine environmental samples were excluded after rarefaction and six environmental samples were excluded due to contamination. Six tadpoles could not be recovered from bromeliads at the end of the experiment.

	water treatment	arthropod treatment	sample size	total OTUs detected	mean OTUs per bromeliad ± s.e.
environment	natural	present	11	859	255 ± 13
		absent	8	731	255 ± 15
	pre-sterilized	present	8	630	186 ± 20
		absent	6	431	141 ± 8
tadpoles	natural	present	10	332	86 ± 7
		absent	12	573	122 ± 13
	pre-sterilized	present	10	279	64 ± 7
		absent	10	283	64 ± 8

Figure 2.

Bacterial community composition of the aquatic environment (enclosed by dashed line) and tadpole microbiomes, calculated using principal coordinates analysis based on unweighted UniFrac distances. Small circles indicate data points; large circles indicate group centroids. (Online version in colour.)

(b). Bacterial sequencing and pathogen quantification

We sequenced the V4 region of the 16S rRNA gene of bacterial genomes to characterize OTUs and quantified Bd from swabs (environmental samples) and whole tadpoles. Because tadpoles were small and delicate, we were unable to sample internal and external bacterial communities separately, so our samples represent bacteria from whole tadpole microbiomes (skin, mouthparts and gut combined). To extract DNA, we used the DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA, USA), following the manufacturer's protocol with minor modifications to increase DNA yield, including increasing the incubation time in the lysis stage to 4 h for tadpoles and overnight for environmental samples. To quantify Bd zoospore genomic equivalents (ZGE), we used Taqman quantitative PCR assays, following standard protocols [48]. To amplify and sequence bacterial DNA, we followed the Earth Microbiome Project 16S Illumina Amplicon Protocol, which targets the V4 region of the 16S rRNA gene using universal primers 515F and 806R, using a dual index approach [49–51]. We amplified samples in duplicate using Phire^® Hot Start II DNA Polymerase (Finnzyme, Espoo, Finland) and combined duplicate PCR products. We visualized amplicons in 1% agarose gel to confirm consistent gel band strength and then pooled samples in equal volumes to generate the amplicon library. We purified the library using the QIAquick Gel Extraction Kit (Qiagen, Valencia, CA, USA) and measured DNA concentration of the purified sample using the Qubit 2.0 fluorometer with the dsDNA Broad-Range Assay Kit (ThermoFisher Scientific, San Jose, CA, USA). The library was sequenced using an Illumina MiSeq (2 × 250 bp) sequencer at the Tufts University Core Facility (TUCF Genomics), Boston, Massachusetts, USA. Sequences were deposited in the NCBI Sequence Read Archive (accession number PRJNA533309).

We analysed bacterial sequences using Quantitative Insights into Microbial Ecology (QIIME) v. 1.9.0 [52]. Only forward reads, trimmed to 150 base pairs, were used for the analyses. We used standard quality control settings to filter low-quality reads and split sequences. We used the deblur workflow [53] to cluster sequences into sub-operational taxonomic units (sOTUs) and assigned taxonomy with RDP [54] using the Greengenes 13.8 database from May 2013 [55,56]. We discarded sequences identified as chloroplasts or mitochondria and sOTUs with less than 10 reads detected across the dataset. We rarefied samples to 10 000 reads (rarefaction curves provided in electronic supplementary material, figure S1). None of the PCR controls were retained after filtering and rarefaction. We aligned sequences with PyNAST [57] and constructed a phylogenetic tree using FastTree with QIIME default parameters [58]. For analyses of alpha diversity, we calculated the total number of observed OTUs for each sample and Faith's phylogenetic diversity [59], a widely used diversity index that incorporates phylogenetic relatedness among bacterial taxa, which should be a proxy for ecological diversity [23,38,60]. For analyses of bacterial community composition, we calculated unweighted and weighted UniFrac distances between samples. We also used a database of bacterial sequences with known functional activity against Bd [61] to identify putatively inhibitory OTUs (hereafter ‘Bd-inhibitory’) in environmental samples. We matched these OTUs to our OTU table at 97% sequence similarity. In addition, we used Jaccard distances between samples to calculate bacterial dispersion (a measure of microbiome stability among individuals within a population; betadisper function in vegan package in Program R) as an index of dysbiosis (decrease in microbiome stability) of the tadpole microbiome [62–66]. Final sample sizes are provided in table 1.

(c). Statistical analyses

We tested for main and interactive effects of water treatment (natural or pre-sterilized) and arthropod treatment (present or absent) on a number of individual response variables. We used general linear models (GLMs) with normal probability distribution and identity link function for models with the following response variables: phylogenetic diversity of environmental bacteria, phylogenetic diversity of tadpole microbiomes, OTU richness of environmental bacteria, OTU richness of tadpole microbiomes, dispersion of tadpole microbiomes, tadpole Bd burden, tadpole body mass, density of environmental zoospores, density of aquatic microeukaryotes (natural water only), chlorophyll a concentration, water turbidity and water pH. For models with community composition of environmental bacteria or tadpole microbiomes as the response variable, we performed permutation multivariate analysis of variance (PERMANOVA) on weighted and unweighted UniFrac distances using the adonis function from the vegan package in Program R [63,64]. To visualize differences in bacterial community composition, we performed principal coordinate analyses on unweighted UniFrac distances in QIIME and plotted values for the first two principal coordinate axes with centroids using the pca2d function from the phytools package in R [64,67]. To determine the bacterial taxa driving the observed differences in composition between treatments, we used linear discriminant analysis (LDA) effect size (LEfSe) method [68] on the galaxy platform (http://huttenhower.sph.harvard.edu/galaxy/), with the four treatment combinations as the class variable and using default parameters. For the model with proportion of Bd-inhibitory taxa in environmental bacterial communities as the response variable, we used a GLM with a binomial probability distribution and logit link function. We also performed Pearson's correlations to test for correlations between tadpole Bd burden and phylogenetic diversity of bacteria in both natural and pre-sterilized water. We used JMP software [69] for all GLMs and correlations.

We used path analyses to investigate ecological connections between arthropods, environmental communities and Bd infection-related outcomes of the experiment (tadpole infection burden and body mass). We performed separate path analyses for bromeliads with natural and pre-sterilized water as an alternative to including direct measurements of environmental and host microbiota because microbiota data for nine samples were excluded after rarefaction and this would have compromised our statistical power. We included all ecologically relevant paths among arthropods (presence or absence), chlorophyll a level, microeukaryote density, environmental Bd density, tadpole infection burden and tadpole body mass. For each water treatment, we first ran full models with all paths and then reduced the models using backward selection, by sequentially removing unsupported variables to improve model fit. From a correlation matrix, estimates of standardized path coefficients, with their associated standard errors, were derived by maximum Wishart likelihood (500 iterations), allowing for identification of significant paths. Latent (unmeasured) variables were estimated for each response variable in the model. We used a modified Akaike information criterion (AIC) as a measurement of model fit. We used RAMONA scripts in SYSTAT to run the path analyses.

3. Results

(a). Environmental communities

Arthropods influenced the species composition (PERMANOVA for unweighted UniFrac: F_1,29 = 2.639; R² = 0.070; p = 0.001; figure 2; electronic supplementary material, figure S2) and relative abundance of taxa (PERMANOVA for weighted UniFrac: F_1,29 = 4.090; R² = 0.116; p = 0.002) in environmental bacterial communities (electronic supplementary material, table S1). Specifically, environmental communities in natural water contained higher abundances of Cellulomonadaceae (Actinobacteria), Mycobacteriaceae (Actinobacteria) and Acidobacteriales (Acidobacteria) when arthropods were absent and higher abundances of Frimbriimonas (Armatimonadetes) when arthropods were present (LEfSe; electronic supplementary material, figure S2 and table S2). When arthropods were present, environmental bacterial communities also contained higher proportions of Bd-inhibitory taxa (GLM: β = 0.145; χ² = 3.822; p = 0.051; electronic supplementary material, table S1) and bromeliads contained lower densities of environmental Bd zoospores across both water types (GLM: β = −0.799; t = −2.76; p = 0.011; figure 1b; electronic supplementary material, table S1). Bromeliads contained lower densities of microeukaryotes (natural water only; GLM: F_1,16 = 5.099; R² = 0.242; β = −2014.75; SE = 892.233; t = −2.26; p = 0.038) and had lower levels of water turbidity (GLM: β = −2.796; t = −2.89; p = 0.008; electronic supplementary material, table S1) when arthropods were present compared to absent. We did not detect effects of water treatment or arthropod treatment on water pH (electronic supplementary material, table S1).

(b). Host microbial communities

Tadpole microbiome structure largely depended on the interaction between water treatment and arthropod treatment. Specifically, tadpole microbiomes had lower phylogenetic bacterial diversity (GLM interaction term: β = −2.770; t = −2.07; p = 0.045; figure 1c) and lower dispersion (GLM interaction term: β = −0.044; t = −2.39; p = 0.022) when arthropods were present, but only in natural water (electronic supplementary material, table S1). We observed a similar pattern in OTU richness of tadpole microbiomes, with relatively low and consistent OTU richness in pre-sterilized water, the highest OTU richness in natural water without arthropods, and substantially reduced OTU richness in natural water with arthropods (table 1; electronic supplementary material, table S1). Similarly, we found an interactive effect of water treatment and arthropod treatment on composition of tadpole microbiomes (PERMANOVA interaction term for unweighted UniFrac: F_1,38 = 3.140; R² = 0.065; p = 0.001; PERMANOVA interaction term for weighted UniFrac: F_1,38 = 9.224; R² = 0.174; p = 0.001; electronic supplementary material, table S1). Microbiome composition was similar among tadpoles in pre-sterilized water across arthropod treatments (in figure 2, light grey and dark grey clusters overlap), the most divergent among tadpoles in natural water without arthropods (in figure 2, dark green cluster varies from others along both the first and second principal coordinate axes) and significantly less divergent among tadpoles in natural water with arthropods (in figure 2, light green cluster aligns with pre-sterilized water on first principal coordinate axis). Differences in community composition of tadpole microbiomes in natural water were driven by higher abundances of Acidobacteria and Armatimonadetes as well as other taxa when arthropods were absent and higher abundances of Firmicutes and Verrucomicrobiae (Verrucomicrobia) when arthropods were present (LEfSe; electronic supplementary material, figure S2 and table S2).

(c). Host fitness

Tadpole Bd burdens depended on water treatment (GLM: β = 0.437; t = 3.15; p = 0.003; figure 1d; electronic supplementary material, table S1), with relatively low and consistent Bd burdens in pre-sterilized water (arthropods absent: mean ± s.e. = 76 ± 31 g.e.; present: mean ± s.e. = 49 ± 9 g.e.), the highest Bd burdens when arthropods were absent in natural water (mean ± s.e. = 186 ± 48 g.e.) and substantially more moderate Bd burdens when arthropods were present in natural water (mean ± s.e. = 111 ± 31 g.e.). Overall, we found positive trendlines in correlations between phylogenetic diversity of host bacterial communities and Bd infection burden in tadpoles, although this relationship was only statistically significant for tadpoles in natural water (natural: r = 0.61, n = 22, p = 0.002; pre-sterilized: r = 0.25, n = 20, p = 0.287). Tadpole body mass at the end of the experiment was higher in pre-sterilized water (mean ± s.e. = 24.8 ± 2.1 mg) than in natural water (mean ± s.e. = 17.4 ± 1.4 mg; GLM: β = −0.0074; t = −2.94; p = 0.006), even though the natural water had higher levels of chlorophyll a, which we used as a proxy for abundance of algae, a tadpole food resource (GLM: β = −63.5076; t = −2.29; p = 0.028, electronic supplementary material, table S1).

(d). Path analysis

In both path models, we observed a strong negative effect of arthropods on environmental Bd density (figure 3). In bromeliads containing natural water, environmental Bd density was a positive predictor of Bd infection burden and infection burden was in turn a negative predictor of tadpole body mass (figure 3). By contrast, in bromeliads containing pre-sterilized water, environmental Bd density was a negative predictor of Bd infection burden and we did not detect a significant association between infection burden and tadpole body mass (figure 3). Removing chlorophyll a and microeukaryote density improved model fit (natural water: full model AIC = 1.812, reduced model AIC = 0.815, 95% CI for reduced model = 0.850, 1.300; pre-sterilized water: full model AIC = 1.362, reduced model AIC = 0.855, 95% CI for reduced model = 0.895, 1.360).

Figure 3.

Path analyses showing a cascade of effects from arthropods to density of Bd zoospores in the environment to Bd infection burden in tadpoles to tadpole body mass. We conducted separate path analyses for bromeliad microecosystems containing natural (a) and pre-sterilized water (b). We first ran full models with all ecologically relevant paths and then reduced the models using backward selection, by sequentially removing unsupported variables to improve model fit. Numbers are standardized path coefficients for the reduced, best-fit models (*p < 0.05). Latent (unmeasured) variables are represented by ‘u’. Unsupported paths shown in grey were removed from the final model to improve model fit.

4. Discussion

We demonstrated that biotic interactions in an aquatic microecosystem influenced the community structure of host microbiomes and host health. Aquatic arthropod detritivores influenced the overall composition of environmental bacteria, with interactions between arthropods and environmental bacteria specifically enhancing the component of the environmental bacterial reservoir that could be pathogen-inhibiting if recruited by the host. Potential mechanisms by which arthropods could influence environmental bacteria include selective feeding, altering movement or composition of particulate matter in the aquatic environment, facilitating certain bacterial taxa through nutrient inputs such as faeces, or seeding the environment with members of their own bacterial microbiome [10,12,70].

In the presence of arthropods, the aquatic microenvironment of bromeliads contained lower densities of Bd zoospores and had lower levels of water turbidity, suggesting that arthropods consumed zoospores or that organic matter processing by arthropods reduced zoospore survival or infectivity [12,43]. Regardless of the specific mechanism, arthropods regulated the environmental zoospore source pool in the bromeliad microenvironment. However, water treatment had stronger effects on the build-up of Bd infections in tadpoles than environmental zoospore densities. Our path analysis for natural water revealed that environmental Bd density was a positive predictor of Bd infection burden in tadpoles, suggesting an efficient cycle of reinfection in which zoospore discharge (transfer of zoospores from tadpoles to water) occurred at a similar pace to zoospore encystment (transfer of zoospores from water to tadpoles). These conditions appeared to promote Bd infections severe enough to negatively influence tadpole fitness (lower body mass) despite higher levels of algal food resources in natural water. By contrast, our path analysis for pre-sterilized water revealed that environmental Bd density was a negative predictor of tadpole infection burden, suggesting a less efficient cycle of reinfection in which zoospore discharge outpaced zoospore encystment or vice versa. These conditions appeared to hamper infection build-up and prevent negative fitness effects of infection in tadpoles. Thus, while arthropods consistently reduced environmental zoospore densities across water treatments, this did not appear to influence the pace of the Bd infection process in tadpoles as strongly as water treatment.

We observed consistently low Bd burdens in tadpoles in pre-sterilized water regardless of the presence or absence of arthropods, the highest Bd burdens in tadpoles in natural water without arthropods and substantially more moderate Bd burdens in natural water with arthropods. This trend matches our observations of the effects of arthropods on the phylogenetic diversity, dispersion and composition of tadpole microbiomes, with little effect of arthropods in pre-sterilized water compared with a pronounced moderating effect of arthropods on the diversity, dispersion and compositional distinctiveness of the microbiome in natural water. By comparison, the trends in our Bd burden data contrast with the effects of arthropods on environmental zoospore density, which was similar in both natural and pre-sterilized water. Based on these findings, we suggest that patterns in Bd burdens were driven more strongly by arthropod activity that reduced recruitment of diverse bacteria by tadpoles and less strongly by arthropod activity that reduced environmental Bd zoospores. While we were unable to detect a significant effect of arthropods on tadpole Bd burdens in our GLM, we view the correspondence between our Bd burden data and effects of arthropods on microbiome structure as convincing considering the high individual variation that is characteristic of host Bd infection burdens under experimental conditions (e.g. [71]). We also consider these trends in our Bd burden data, while non-significant, to be ecologically relevant given that even small changes in rates of exponential pathogen population growth at the larval stage may be critical in regulating pathogen burdens carried through metamorphosis, at which point the disease chytridiomycosis may develop if pathogen loads eventually exceed a critical threshold [33,72].

Thus, we observed the combination of a uniquely enhancing effect of natural water on the diversity of the tadpole microbiome and a uniquely facilitative effect of natural water on the Bd reinfection cycle, with these effects weakened in the presence of arthropods. The structure of the tadpole microbiome may not necessarily be related to the Bd reinfection cycle in the host if factors external to the host such as water quality (e.g. acidity, suspended organic matter) or density of predatory microeukaryotes influence the survival of free-swimming zoospores [41–43]. However, we did not detect differences in water quality (pH and turbidity) between natural and pre-sterilized water, and microeukaryotes either did not appear to influence environmental Bd density (based on our path analysis for natural water) or were considered unlikely to colonize the pre-sterilized water treatments during the short duration of the experiment. Moreover, we detected no difference in densities of environmental Bd zoospores between natural and pre-sterilized water, and no evidence that arthropods preferentially consumed zoospores in natural water more than in pre-sterilized water, which collectively suggests that arthropods interacted primarily with environmental bacteria and that differences in Bd dynamics between water treatments depended on within-host factors such as the microbiome rather than zoospore survival in the environment.

Our findings are consistent with a scenario in natural water in which environmental bacteria contributed to host microbial dysbiosis, which facilitated Bd infections, with an ameliorating effect of arthropods on host dysbiosis and Bd infection through interactions with environmental bacteria. The high environmental bacterial diversity in natural water could have contributed to host dysbiosis by overwhelming the host microbiome with transient taxa or increasing the chances of opportunistic bacterial infection, with this dysbiosis predisposing hosts to Bd infection. By contrast, our findings indicate that tadpoles in pre-sterilized water were faced with less potentially hazardous environmental bacterial communities and were able to maintain balanced and stable microbiomes and consistently more robust defenses against Bd irrespective of the presence or absence of arthropods.

Combined, our findings suggest that aquatic host species may face considerable challenges in maintaining stable and balanced symbiotic bacterial communities in microbially rich aquatic habitats. Many previous studies have reported greater pathogen-suppressing functions of more diverse bacterial communities, potentially through dominance or complementarity effects [73–75]. By contrast, we found what appear to be detrimental cascading effects of high environmental bacterial diversity on pathogen resistance. The environmental bacterial diversity in our study could be unusually high given that we homogenized water from multiple bromeliads to create our natural water treatment. Even so, we suggest that high environmental bacterial diversity may not increase the odds of pathogen-inhibiting taxa occurring or being selected by hosts but rather to the contrary, may contribute taxa that threaten the balance of the host microbiome, potentially predisposing the host to disease. At the same time, we observed clear evidence of arthropod–bacteria interactions in the environment and cascading effects on host colonization that appeared to help maintain pathogen-inhibiting functions of the host microbiome. Thus, our study suggests that arthropod–bacteria interactions promoted host-associated bacterial communities with optimal richness and composition for protective functioning [27,38,76].

Previous studies have demonstrated the strong trophic interdependence between detritivores and bacteria in aquatic ecosystems [10,12], but our study is novel in suggesting that the interactions of detritivores with environmental bacteria may specifically reduce host colonization by transient taxa and promote antimicrobial components of aquatic bacterial communities that translate to the host microbiome and host health in the face of invading pathogens. Our work reveals tight links between the healthy functioning of aquatic ecosystems and assembly of a beneficial host microbiome, suggesting that disturbances to ecosystems, such as loss of detritivores, may have important cascading effects on the protective functions of host-associated bacteria [77]. Useful extensions to our study would be to assess the effects of aquatic community dynamics on host core bacteria (i.e. key microbes present on all or most host individuals) and on Bd-inhibitory bacterial taxa recruited to the tadpole mouthparts. Future studies should also investigate whether this pronounced relationship between ecosystem functioning and integrity of the host microbiome scales up to disease systems in larger aquatic habitats such as streams and ponds and translates to different host–pathogen systems.

Ethics

The experiment was performed in accordance with the guidelines of the Institutional Animal Care and Use Committee at the Universidade Estadual de Campinas (CEUA), SP, Brazil.

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material.

Authors' contributions

M.F.K.-B. and C.G.B. conceived of the study. M.R.P., C.S.L., G.H.M., M.F.K.-B., L.P.R., F.R. and C.G.B. carried out the study. S.E.G., M.L.L., W.J.N., and C.G.B. carried out the molecular laboratory work. S.E.G., M.L.L., M.C.B., D.C.W. and C.G.B. analysed the data. G.Q.R., C.F.B.H., L.F.T. and C.G.B. coordinated the study and participated in interpreting the data. S.E.G. drafted the manuscript. All authors critically revised the manuscript. All authors gave final approval for publication and agree to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

São Paulo Research Foundation (FAPESP) provided grants to L.F.T. (no. 2014/23388-7) and C.F.B.H. (no. 2013/50741-7 and no. 2014/50342-8) and research fellowships to M.L.L. (no. 2017/26162-8), L.P.R. (no. 2016/03344-0), and M.R.P. (CAPES no. 001). National Council of Technological and Scientific Development (CNPq) provided research fellowships to C.G.B. (no. 312895/2014-3), L.F.T. (no. 300896/2016-6) and C.F.B.H. (no. 302518/2013-4).