Serratia marcescens shapes cutaneous bacterial communities and influences survival of an amphibian host

Abstract

Ongoing investigations into the interactions between microbial communities and their associated hosts are changing how emerging diseases are perceived and ameliorated. Of the numerous host–microbiome–disease systems of study, the emergence of chytridiomycosis (caused by Batrachochytrium dendrobatidis, hereafter Bd) has been implicated in ongoing declines and extinction events of amphibians worldwide. Interestingly, there has been differential survival among amphibians in resisting Bd infection and subsequent disease. One factor thought to contribute to this resistance is the host-associated cutaneous microbiota. This has raised the possibility of using genetically modified probiotics to restructure the host-associated microbiota for desired anti-fungal outcomes. Here, we use a previously described strain of Serratia marcescens (Sm) for the manipulation of amphibian cutaneous microbiota. Sm was genetically altered to have a dysfunctional pathway for the production of the extracellular metabolite prodigiosin. This genetically altered strain (Δpig) and the functional prodigiosin producing strain (wild-type, WT) were compared for their microbial community and anti-Bd effects both in vitro and in vivo. In vitro, Bd growth was significantly repressed in the presence of prodigiosin. In vivo, the inoculation of both Sm strains was shown to significantly influence amphibian microbiota diversity with the Δpig-Sm treatment showing increasing alpha diversity, and the WT-Sm having no temporal effect on diversity. Differences were also seen in host mortality with Δpig-Sm treatments exhibiting significantly decreased survival probability when compared with WT-Sm in the presence of Bd. These results are an important proof-of-concept for linking the use of genetically modified probiotic bacteria to host microbial community structure and disease outcomes, which in the future may provide a way to ameliorate disease and address critical frontiers in disease and microbial ecology.

1. Background

Microbial communities are being increasingly recognized for their profound role in the health of their associated hosts. While recent work has proved fruitful in understanding the nature of these relationships and their significance to host survival, many aspects of host–microbiota ecology remain largely unknown [1,2]. This includes many uncertainties regarding specific mechanisms pertaining to the ability and extent to which microbial communities can be manipulated for the amelioration of host disease [3,4]. Previous work has demonstrated that the introduction of new bacteria to a given microbial community can have dramatic effects on host survival when facing an infection [5]. However, little is known on how such probiotic applications affect the overall host-associated microbial community structure over time. Additionally, an intriguing yet understudied possibility in treating disease is the genetic manipulation of probiotic bacteria for functional enhancement of the host-associated microbial community structure. The development of genetic engineering tools for the alteration of bacterial community constituents has been recently explored in host–microbiome systems such as bees and mice [6,7]. However, the development of such tools for use in predictive microbiota restructuring as a treatment for host disease remains largely unclear. Altering the microbial community structure, function and host interaction through genetic alteration may, therefore, prove to be an important component in the future treatment and prevention of disease in various host systems.

One disease system of immense interest, for both its contributions to new conservation tools and basic knowledge in disease ecology and epidemiology, is the amphibian-Batrachochytrium dendrobatidis (hereafter Bd) disease system. Bd is a fungal pathogen of amphibians belonging to the Chytridiomycota [8]. This fungal pathogen has had a devastating impact on susceptible amphibian populations worldwide resulting, in part, in severe population declines as well as extinction events [9]. However, not all amphibian species and populations are equally affected with some populations showing resilience to initial exposure and/or increased survival to re-exposure as post-epizootic relic populations [10]. The developmental stage of individuals has also been shown to be a potential factor in determining survival. Critical factors include a lack of a fully developed immune response in recent metamorphs and long-term immunosuppression effects from early life-history exposure to Bd [11,12]. Surprisingly, in many cases, this pathogen has shown little reduction in its virulence. Rather, the emergence of resistant relic populations is likely caused by changes in host defences [13]. These defences include variation within the major histocompatibility complexes (MHC) of the amphibian adaptive immune response [14,15], cutaneous secretion of host-generated skin anti-microbial peptides [16,17], the host-associated cutaneous microbiota [18] or some combination thereof. While host produced immune defences such as MHC and innate defences such as secreted peptides are normally implicated in host disease responses, the potential role for the host-associated microbiota in aiding in the amelioration of disease remains a possible explanatory factor for host resistance to disease and the emergence of post-epizootic relic populations. Evidence supporting this idea would also lend credence to the efficacy of manipulating the host's microbial community for disease amelioration through probiotic application based on mechanisms of ecological competition or exclusion. Alternatively, predictive artificial restructuring of the host-associated microbial community with genetically altered bacteria would represent a novel method of precision intervention in protecting those amphibian species still facing the likely possibilities of decline and extinction.

Of the many bacteria continually being identified for their anti-Bd properties [19], Serratia species have been recognized for their ability to inhibit the growth of Bd [20] and have also been shown to be a common constituent of the amphibian microbial community in both Anura and Caudata [21]. Among the intriguing properties of Serratia marcescens (Sm) are the production of extracellular chitinases and glucanases, toxin delivery through the type-VI secretion system, and the production of the known anti-microbial secondary metabolite prodigiosin [22,23]. In this study, we examined the effect of having a functional prodigiosin biosynthesis pathway on microbial community structure and host disease outcome by the introduction of either wild-type (WT) Sm or a strain with an impaired prodigiosin biosynthesis pathway (ΔpigM-Sm). Metabolite-minus strains may be used to modify the microbial community so as to induce emergent community effects that are anti-Bd. While the loss of prodigiosin production was shown to reduce anti-Bd activity in vitro, the effect could be the opposite in a community setting. Altering microbial communities to have anti-Bd properties could, therefore, be completed through the use of either gain-of-function or knockout/knockdown strains. In using the later approach, we sought to determine the effects of a single gene of one probiotic bacteria on the overall host–microbiota system when challenged with a pathogen. We accordingly hypothesized that the application of WT- and ΔpigM-Sm would have significant differential effects on the overall microbial community structure as well as on Acris blanchardi (Blanchard's Cricket frog) metamorph responses when challenged with Bd. A. blanchardi was used in this study as it is known to be infected by Bd [24] and may, therefore, be both an important carrier of Bd and also susceptible to emerging strains of Bd directly or indirectly through effects with Bd as an important cofactor. The utilization of recent metamorphs also allowed for the study of impacts of host-associated microbial communities at a life-history stage with an underdeveloped acquired immune response [11]. Additionally, synergistic effects of Bd and other effectors such as pollutants have been argued as a mechanism for recent declines of A. blanchardi [25–27].

We also hypothesized that the differences in microbial community structure and disease-associated host characteristics would be influenced by time. Specifically, we examined the effect of time on microbial community diversity between treatments. While treatments included the application of Sm and Bd, we also examined the effects of washing the skin of the host to reduce transient cutaneous microbiota before sampling. This was performed as washing has been a previously reported sampling method but with temporal effects that are unclear [28]. Therefore, we tested whether there was a positive correlation between length of time, treatment conditions and experimental measurements of A. blanchardi (mortality, mass, Bd presence/absence, bacterial community diversity and Sm abundance).

The results herein provide important insights into the possibility of genetic manipulation as a component of probiotic treatment options in the amelioration of chytridiomycosis in an amphibian system and, more broadly, in other host–microbiome–disease systems.

2. Material and methods

(a). Animal capture and husbandry

All A. blanchardi were metamorphs captured in Nebraska along the Missouri river and corresponding backwaters (59-mile segment of the Missouri National Recreational River). Specimens were collected under an approved Nebraska Game and Parks Commission Scientific and Educational Permit (no. 1006). Experimental work followed an approved University of South Dakota IACUC protocol. All individuals were captured by net or hand while gloved, with gloves being changed between individual frog handling. Each individual frog was put into a separate plastic bag and kept in a cooler for transportation to the University of South Dakota animal holding facility the same day as capture. At the animal holding facility, each individual was swabbed for Bd detection by qPCR. Each swab consisted of three strokes ventral and three strokes dorsal for each individual. Swabs were kept in a freezer at −20°C until further use.

Once in animal holding, each individual was kept in a plastic container containing Coco Husk (Exo Terra, Rolf C. Hagen Corp., Mansfield, MA, USA) that had been sterilized by autoclaving (15 m, 121°C). Each housing unit was sprayed with autoclaved spring water daily and each individual was fed two pinhead crickets three times per week (six pinhead crickets per week; Lazy H, Labelle, FL, USA). Light/dark cycles were set to 42.00′ N latitude to mimic natural environmental light conditions. The temperature was kept at 21.5–22°C based on daily temperature logs.

All animals alive at the end of the experiment were euthanized by topical benzocaine overdose. A subset of the experimental animals were preserved and deposited at the Biodiversity Center Herpetology Collection at The University of Texas at Austin (TNHC 106207–106238).

(b). Growth assays

After construction and verification of the ΔpigM:kan^R Sm (hereafter ΔpigM-Sm) strain (electronic supplementary material, Methods), in vitro growth curves and challenge assays were conducted. The growth assay was completed in triplicate (n = 3 per strain) and with broth only (LB-Miller) controls. All growth was conducted at 21°C with measurements being taken every hour for 24 h with a spectrophotometer (OD = 600). All samples were normalized for the same starting cell concentration.

The challenge assays were conducted as per Bell et al. [29]. Briefly, 96-well plates were inoculated with varying concentrations of Sm supernatant (from centrifuged culture, 0.22 µm filter-sterilized). Starting growth values for beginning Bd concentration were normalized for all wells. Dilution cultures for Bd only were also made and analysed to control for nutrient limiting effects as per Bell et al. [29]. All growth was conducted at 21°C over the course of 8 days.

(c). Treatments and application

Twelve treatment conditions composing a 2 × 2 × 3 factorial design (electronic supplementary material, table S2) were used in the in vivo microbiota manipulation experiment. These included two levels in the Bd factor (live Bd, no-Bd plus culture media), two levels in the washing protocol (washed, unwashed) and three levels in the bacterial treatment (no-Sm, WT-Sm and ΔpigM-Sm; methods detailing construction and verification of ΔpigM-Sm are available in the electronic supplementary material, Supplementary methods). After initial capture, all individuals were exposed to either the control (LB broth only), WT-Sm (at approx. 35 × 10⁶ cells ml⁻¹ based on plating serial dilutions) or ΔpigM-Sm (at approx. 12 × 10⁶ cells ml⁻¹ based on plating serial dilutions). All exposure was done by inserting the individual frog in a 50.0 ml falcon tube containing 10.0 ml of one of the three treatments for 30 min. All individuals were then held for 3 days before Bd exposure for the respective bacteria to incorporate and stabilize as members of the cutaneous microbiota. On day 4, individuals were exposed to either the control (only tryptone–gelatin-hydrolysate–lactose (TGhL) broth) or Bd in TGhL broth (approx. 1.125 × 10⁸ zoospores ml⁻¹), which was considered day 1 for the in vivo experiment (day 1 for all results herein correspond to this convention, with the exception of Bd day 1 which was taken on the day of A. blanchardi collection prior to Sm exposure). The washing treatments were applied prior to Sm inoculation. The washed treatment consisted of pipetting sterilized spring water (3.0 ml ventral and 3.0 ml dorsal) onto each frog. The unwashed treatments had no water wash. All Bd exposure was completed by inserting the individual frog in a 50.0 ml falcon tube containing 5.0 ml of one of the two treatments for 30 min. The number of individuals in each treatment was set at n = 10 at the beginning of Sm inoculation, however, some individuals died during the Sm incorporation period shortly after capture and were not included in any statistical analysis. The n for each treatment as reported in electronic supplementary material, table S2 reflects the exclusion of these individuals.

(d). DNA extraction and purification

All DNA extraction was conducted using the Qiagen (Hilden, Germany) DNeasy Blood and Tissue kit. Overnight digestion for all samples was performed using standard protocols for tissue with a proteinase K digestion at 56°C overnight. After extraction, DNA was cleaned and concentrated using the ZR-9 Genomic DNA Clean and Concentrator-5 kit following standard protocols (Zymo Research, Irvine, CA, USA).

(e). qPCR

Samples were extracted and purified as described above prior to qPCR. qPCR was used to detect Bd zoospore presence using primers and a TaqMan probe as per Boyle et al. [30]. All samples were also analysed with controls lacking DNA and a standard curve with order of magnitude multiples of one copy of the Bd ITS gene. The Bd swab was in addition to the weekly swabbing for 16S rRNA gene analysis, and was taken after the 16S swab (same methodology), and was used for presence/absence detection. All swabs used had DNA extraction and purification performed as described above and subject to enhanced clean-up as per Mosher et al. [31]. For the purposes of statistical analysis in the mixed-effects models, a sample was reported as a positive when at least two of the three replicates had detectable amplification and mean detection was greater than or equal to 0.1 zoospore (9.8 copies). A 0.1 zoospore threshold was set as per previous work [32] so as to account for potential loss in the sample during the clean-up step. A genomic zoospore equivalency of 98 was used for a detection threshold based on the number of ITS1-5.8S genes in Bd strain JEL423 as previously reported (98.2 copies zoospore⁻¹; [33]).

(f). Library preparation and next-generation sequencing

DNA from each sample swab was quantified on a Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA) using Qubit™ dsDNA HS Assay Kit (quantitation range: 0.2–100 ng). Up to 15 ng of DNA was used to prepare a library for next-generation sequencing using a modified dual indexing protocol proposed by Illumina (Illumina 16S Metagenomic Sequencing Protocol (15044223 Rev. B)). The first round of amplification used primers targeting the V4 region of the 16S rRNA gene. Illumina overhang adapters were added to 515F and 806R primers (V4_515F 5′-GTG YCA GCM GCC GCG GTA A-3′ and V4_806R 5′-GGA CTA CHV GGG TWT CTA AT-3′ or V4_515F 5′-TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG GTG YCA GCM GCC GCG GTA A-3′ and V4_806R 5′-GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GGG ACT ACH VGG GTW TCT AAT-3′; 515F and 806R base primer indicated in last two sequences). Primary amplification was carried out in duplicate for each sample using a 2× KAPA HiFi HotStart Ready Mix (KAPA Biosystems, Wilmington, MA, USA) and the following thermocycling protocol: initial denaturation at 95°C for 3 min, followed by 25 cycles of 98°C for 20 s, 55°C for 15 s and 72°C for 30 s, and a final extension at 72°C for 5 min on a Veriti Thermal Cycler (Thermo Fisher Scientific, Waltham, MA, USA). All of the reactions were then purified with Agencourt AMPure XP beads (Beckman Coulter, Brea, CA, USA) prior to indexing. Secondary amplification of each sample was completed using a 2× KAPA HiFi HotStart Ready Mix (KAPA Biosystems, Wilmington, MA, USA) and a combination of two unique Nextera XT Index primers (N7xx and S5xx). Amplification was completed on a Veriti Thermal Cycler (Thermo Fisher Scientific, Waltham, MA, USA) using the following thermocycling protocol: initial denaturation at 95°C for 3 min, followed by eight cycles of 98°C for 20 s, 55°C for 15 s and 72°C for 30 s, and a final extension at 72°C for 5 min. The libraries were purified with Agencourt AMPure XP beads (Beckman Coulter, Brea, CA, USA) and quantified on a Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA), using Qubit™ dsDNA HS Assay Kit (quantitation range: 0.2–100 ng), normalized, and pulled together. The final library was gel re-purified using the Wizard^® SV Gel and PCR Clean-Up System (Promega Corporation, Madison, WI, USA). Paired-end sequencing was performed on an Illumina MiSeq instrument using MiSeq Reagent Kit v3 for 600 cycles (Illumina Inc., San Diego, CA, USA). All library preparation and sequencing was completed in WestCore at Black Hills State University (Spearfish, SD, USA).

(g). Sequence data and statistical analyses

Initial processing and analysis of raw sequencing data were completed using Mothur (v.1.39.5; [34]). Briefly, the standard MiSeq SOP was followed starting with generating contigs from paired-end reads, clean-up steps (including screening, filtering and chimaera removal using UCHIME), alignment to the Silva database [1.32], and generation of operational taxonomic units (OTUs) for statistical analysis ([35]; corresponding webpage accessed 29 January 2018).

All statistical analyses were performed in R v.3.4.3 [36]. Analyses of the in vitro experiments were completed with linear models and repeated measures ANOVA. Estimated marginal means (EMMs) and Tukey's pairwise tests were used to compare multiple treatments where necessary. This was completed using the emmeans package [37].

Analysis of the in vivo experimental results, with the dataset including missing data points and random variables, necessitated linear mixed-effects models. Package lme4 [38] was used to create linear mixed-effects models of the relationship between the measured response variable and the predictor variables: treatment, day and individual. As fixed effects, we entered a day–treatment interaction term into the model. Individual frogs were included in the model as a random effect. The response variable used between models consisted of either inverse Simpson diversity, mass (g), Bd presence/absence or Serratia abundance. Visual inspection of residual plots in all models did not reveal any obvious deviations from normality or homoscedasticity (electronic supplementary material, figure S5–S8). Model selection was carried out by likelihood ratio tests of the full model with the day–treatment interaction effect in question against the null model without the day–treatment interaction effect in question. Additional post hoc testing with EMMs comparisons and Tukey's pairwise tests between treatments and between days was completed using the emmeans package [36].

Analysis of beta-diversity between treatment populations was done using Bray–Curtis dissimilarities and the adonis function in the R package vegan [39]. Post hoc pairwise comparisons adjusted for false discovery using the Benjamini–Hochberg (BH) procedure were used to determine between treatment significance at differing time points. Pairwise comparisons used the adonis.pairwise function. Adjusted p-values were obtained from the p.adjust function which uses a conservative implementation of BH methods for false discovery correction with defined controls instead of asymptotic estimations. Based on this implementation, we, therefore, allowed α = 0.1 for adjusted p-value assessment of significance.

All scripts and models used for the analysis of results presented in this paper have been deposited in GitHub (https://github.com/kvasir7/Acris_microbiome) and are publicly accessible. All 16S rRNA gene sequences have been deposited in the NCBI Sequence Read Archive (BioProject: PRJNA435631).

3. Results

To initiate a study on the role of Sm-produced prodigiosin in modulating host–microbiome dynamics and disease outcome, the ΔpigM-Sm mutant construct was verified for expected genotype and phenotype (electronic supplementary material, figure S1–S3; Supplementary methods). After verification of the prodigiosin deficient strain, the anti-Bd effects of both WT-Sm and ΔpigM-Sm were also compared in vitro using a cell-free supernatant assay ([29]; figure 1). Results indicated significantly depressed growth of Bd when grown in the presence of the WT supernatant (visibly red) when compared with the mutant-Sm supernatant (not visibly red, lacking prodigiosin; p = 0.014, α = 0.05) and the Bd only control (p < 0.0001, α = 0.05). The extracted supernatant was introduced at t = 0, with results based on this initial inoculation with prodigiosin-containing supernatant as opposed to multiple additions throughout the course of the experiment.

Figure 1.

Growth curves of B. dendrobatidis (Bd; GPL-Panama). Bd only (black), Bd challenged with pigMsupernatant (red) and Bd challenged with WT-Sm supernatant (blue) are all represented. Supernatant was added at t = 0. Error bars represent 1 s.d. N ≥ 6 replicates per treatment group are represented. (Online version in colour.)

With significant effects of prodigiosin production on Bd growth confirmed, we then conducted the in vivo (with frogs) portion of the study in which the different Sm strains were tested in wash/no-wash and Bd/no-Bd treatment combinations. In conducting this experiment mass, Bd presence/absence, microbial community (inverse Simpson) diversity and Sm abundance were all measured as response variables. The interaction between treatment and time was significant in predicting mass, bacterial community inverse Simpson diversity, Bd presence/absence and Sm abundances (p < 0.05, α = 0.05). Post hoc tests on the models were completed using EMMs. All EMM pairwise comparison results are given (electronic supplementary material, tables S5–S8).

After the model was determined, zoospore detection, mass, inverse Simpson diversity and Sm abundance were all examined as response variables in separate models. In analysing data in the zoospore model, the presence of zoospores as detected by qPCR was not significantly different between treatment groups at the same times based on EMM comparisons (electronic supplementary material, figure S7-B; p > 0.05, α = 0.05; see electronic supplementary material, table 5 for estimates ± s.e.). A small amount of Bd detection was recorded before the introduction of Bd (JEL 423), indicating a low level of Bd prevalence in the environment from which individuals were collected. However, Bd infection qualitatively increased one week after exposure to Bd JEL 423, including in individuals with no-Bd detection. Bd was absent from all live individuals three weeks post-Bd (JEL 423) exposure (electronic supplementary material, figure S7-A).

Pairwise comparisons using the mass model indicated a small but significant (p < 0.05, α = 0.05; see electronic supplementary material, table S6 for all estimates ± s.e.) decrease in mass for frogs within all treatment groups from days 1 to 35, with the only exception being treatment 2 (no-Sm control + Bd, unwashed). Mass differences between treatment groups on the same day were also examined with no significant differences (p > 0.05, α = 0.05; see electronic supplementary material, table S6 for estimates ± s.e.) found in all pairwise treatment comparisons for each day.

Alpha diversity using the inverse Simpson diversity metric was calculated for each microbiota sample at each time point collected. As there were no significant alpha-diversity differences seen in Bd versus no-Bd exposure between any of the treatments, these treatments were pooled for further analysis of diversity. In comparing inverse Simpson diversity change over time, a significant increase in diversity between day 1 and day 29 was seen in both unwashed and washed ΔpigM-Sm treatments (p = 0.033 and p < 0.0001, respectively, α = 0.05; see electronic supplementary material, table S7 for estimates ± s.e.) as well as in the washed control treatment (p = 0.001, α = 0.05). No significant change in inverse Simpson diversity was observed in the unwashed control or WT-Sm treatments. Additionally, Serratia became a major component (greater than 10.0% for at least one time point) of the host's microbiota after inoculation with both WT and mutant ΔpigM-Sm (figure 2a). For 16S rRNA gene sequencing, summary statistics indicate a mean sequence depth of 55 403 reads per sample (min = 11 535; max = 91 809; electronic supplementary material, figure S4). Comparisons of Sm abundances between treatments and time were performed using EMMs from a linear mixed-effect model using the same model procedure as reported above with interaction effects included (electronic supplementary material, table S8, figure S8).

Figure 2.

(a) Bar graphs showing relative abundance (%) of bacterial genera comprising greater than 10.0% of the microbial community at a minimum of one time point. Figures are divided by treatment (given on top) and week (number corresponds to first day of week). (b) Principal coordinates analysis (PCoA) comparing between sample community structure among individual communities. Treatment factors and time are indicated for each panel with each dot representing an individual sample. (Online version in colour.)

Between treatment (sample population) beta-diversity was also examined using principal coordinates analysis (PCoA) and comparisons from Bray–Curtis dissimilarities calculated using the phyloseq and vegan packages in R [39,40] The PCoA showed clustering among bacteria treatment groups at week 1 and 2 in Bd-unwashed and no-Bd-unwashed treatments (figure 2b). This was followed by a PERMANOVA in which the model included time, bacteria, wash and Bd (infection) treatments as interacting terms added sequentially, respectively. Ordering was based on a priori results from the ordination and also model comparison in which factors were compared with and without interactions. When factors were examined individually, only bacteria exhibited both significance in the PERMANOVA and also non-significance in group dispersion testing (pseudo-F_2,252 = 7.73; p = 0.001, α = 0.05; dispersion test: F_2,252 = 1.27, p = 0.27, α = 0.05) indicating differences among bacteria treatments. However, interaction effects revealed significance among two- and three-factor interactions involving all factors (electronic supplementary material, tables S9-A and S9-B). Of note, the time had both significant group dispersions and two- and three-way interaction effects indicating that there are likely differences in both location and dispersion. However, significant effects were not seen in the four-way interaction including time (pseudo-F_2,252 = 1.34; p = 0.17, α = 0.05). All PERMANOVA test statistics and group dispersion tests by factor are given (electronic supplementary material, tables S9-A and S9-B). As significant differences were observed among treatments, post hoc tests were subsequently conducted for all treatment group-time pairwise combinations using the pairwise.adonis function ([41]; electronic supplementary material, table S9). Results from this analysis indicated significant differences between Sm and no-Sm (control) and ΔpigM-Sm and no-Sm (control) treatments at varying time points and among Bd versus no-Bd and washed versus unwashed treatments (39.20 > F(model)_1,1–10 > 3.15, p < 0.10, α = 0.10; electronic supplementary material, table S9-C).

Survival curve analysis was conducted using the Kalbfleisch–Prentice and Tsiatis–Link–Breslow estimates. These estimates indicated a significant difference between bacteria treatments in the unwashed + Bd treatment group (figure 3, p(global) < 0.001, α = 0.05, d.f. = 2) but not in the washed + Bd treatment group or in either of the no-Bd treatment groups. Follow-up pairwise log-rank comparisons indicated that the ΔpigM-Sm treatment showed a significantly decreased survival probability compared to the WT-Sm treatment (p = 0.00026, α = 0.05, d.f. = 1). However, significant differences were not seen in comparing the other bacteria treatment combinations in the unwashed + Bd group (p > 0.05, α = 0.05, d.f. = 1). In testing other pairwise comparisons with single factor differences, a significantly higher survival rate was seen in the unwashed + Bd + no-Sm treatment when compared with the washed + Bd + no-Sm treatment (p = 0.0071, α = 0.05, d.f. = 1). Additional pairwise comparisons with two-factor differences were completed also resulting in significant differences when various combinations of two factors were tested (see electronic supplementary material, table S4).

Figure 3.

Survival plots with panels faceted by treatment conditions Bd/no-Bd and washed/unwashed. Curves in each sub-plot are colour-coded by bacterial treatment: mutant-Sm, WT-Sm or no-Sm (given in figure legend). Global log-rank tests indicate a significant difference between Bd-unwashed curves (p < 0.001). Follow-up pairwise log-rank comparison values are given in electronic supplementary material, table S4. (Online version in colour.)

4. Discussion

The use of genetically engineered bacterial strains for treating or reducing infections of sites that are normally polymicrobial has not been systematically investigated. We hypothesized that a prodigiosin producing strain of Sm would limit or eliminate Bd infections in an amphibian system. The reported effects of prodigiosin on in vitro Bd growth show that initial exposure to prodigiosin-containing bacterial supernatant inhibits growth. Decreasing differences in the growth also support the idea that metabolites such as prodigiosin are used or degraded over time and lose their anti-fungal and anti-bacterial effects if they are not continually produced by bacteria. The results of this study also indicate that inoculation of A. blanchardi with the WT- or mutant-Sm strains alters host-associated bacterial community diversity and differentially alters survival in the presence of Bd. Specifically, the addition of a recombinant strain of Sm having a dysfunctional prodigiosin biosynthesis capability was correlated with both an increase in bacterial community diversity over time and also a decrease in host survival probability. This indicates that specific functions of a single member species of a bacterial community can have dramatic effects on host-associated bacterial community diversity, which may not necessarily be beneficial for survival. Additionally, we have also shown that there were differential effects in the survival probability of varying treatment groups when examining the synergistic effects of multiple factors, although the causes for these changes are less clear.

The relatively low detection of Bd genomic zoospore equivalents indicates A. blanchardi possesses some level of intrinsic resistance to Bd infection. This is consistent with current and historical reports of Bd prevalence in the Midwestern United States [25] and is consistent with our goal to examine the efficacy of microbial therapies on populations that are not Bd naive but potentially susceptible to more virulent strains of Bd to which such populations have not previously been exposed. Bd strain JEL 423, which is part of the global panzootic lineage (GPL), is known for its high virulence in both the wild and in laboratory experiments [42]. The A. blanchardi frogs used in this study were nonetheless only partially susceptible to JEL 423 exposure. Whether this is due to previous exposure to Bd in the source environment enabling an adaptive immune response, factors of laboratory conditions per se, or developmental stage at capture is at this point not well understood. The total absence of Bd after three weeks is also intriguing and likely due to all frogs either clearing Bd or dying. Whether mortality was specifically caused by Bd infection, latent effects from pre-collection environmental factors, or some combination thereof is not clear. Nevertheless, it is clear that Bd was a significant factor among certain treatment conditions in the reported study and either caused or was a significant factor in the observed differential mortality. Continual seeding and exposure to Bd from environmental sources may also be important in maintaining infection in some frog species (e.g. A. blanchardi).

In measuring the mass of the experimental animals, we detected small but significant decreases in mass in most treatments over time. However, we found no significant differences between treatments. The overall decrease is thought to be an artefact of experimental conditions, as it was seen in all treatments. At this time it is unclear as to what the cause of this systemic effect may be. The resulting mortality was also higher than expected. However, this may be due to the use of recent metamorphs, which are known to experience high-mortality rates in natural settings, with possible carry-over of pre-collection environmental effects. While various precautions were taken in our husbandry efforts, we also acknowledge the possibility of systemic mortality due to captivity conditions.

Bacterial community diversity was seen to significantly increase over time in ΔpigM-Sm and the washed control treatment. This suggests that the prodigiosin deficient strain of Sm promotes bacterial community diversity over the WT-Sm strain, presumably due to inhibitory effects of a functional prodigiosin pathway on bacterial growth (i.e. WT-Sm killing other bacteria species and reducing diversity). This is consistent with the known anti-bacterial and anti-fungal properties of prodigiosin producing Sm [43]. The significant increase in diversity in the washed control treatment, when compared with the unwashed treatment also indicates that washing may promote community diversity. The reason for this is unclear but may be due to recolonization from within the body. For example, it is possible that the gut microbiota is more stable in captivity when compared with the cutaneous microbiota, which is known to change over time in captive anurans [44]. That recolonization from a more diverse gut microbiota could result in the increased diversity of the cutaneous microbiota is, therefore, a possibility. Whether such an increase in diversity due to colonization would be sustained temporally or would decrease due to the establishment of dominant species is at this point not well understood. Alternatively, washing may remove a dominant, less adherent species from the skin and allow other gut or probiotic derived species to fill the resulting unoccupied niche. Niche availability has been recently recognized as an important factor in successful probiotic application [45] lending support to this possibility. This could also suggest that colonization by previously unrepresented species in the cutaneous community occurs when such an invading microbial species (bacteria or fungi) is able to displace other microbes while also having the adherence mechanisms to resist removal.

The observed increase in alpha diversity among the Sm treatments was not, however, associated with a decrease in mortality. The results indicate that while the application of the ΔpigM-Sm probiotic increased community diversity, it was also associated with an increase in host mortality compared to the WT-Sm. This could be explained by the possibility that while prodigiosin is anti-Bd, it is also inhibitory towards other bacteria and therefore not associated with an increase in community diversity as seen with the WT-Sm. Additionally, a lack of prodigiosin production in the ΔpigM-Sm treatments could explain both an increase in community diversity and also a decrease in effectiveness against Bd and resulting increase in host mortality. The ΔpigM-Sm probiotic addition, while diversity promoting, might also repress or eliminate key anti­-Bd bacteria in the host-associated microbial community. A lack of prodigiosin production likely affects Sm niche occupation thereby changing which bacteria it is competing with and influencing on the amphibian skin. The bacteria that are in competition with Sm under no prodigiosin conditions could be repressed or eliminated. These same bacteria could also have greater anti-Bd activity than the probiotic or new and more diverse community resulting in a net loss of anti-Bd activity for the host-associated microbiota. While seemingly counterintuitive, this would nevertheless indicate the non-trivial and subtle nature of any probiotic-based method for disease amelioration. More broadly, future predictive microbial restructuring work in any organism will need to address such unforeseen outcomes.

While the results of our experiments are important in showing the potential for genetic level manipulation of host-microbiota and subsequent disease outcomes, doing so in a predictive manner is perhaps the more challenging future goal for large-scale manipulation in a wild or laboratory setting. Understanding the effects of specific secondary metabolites and protein products on host–microbiota interactions remains an open and difficult challenge. Developing the ability to predictively up- and down-regulate the production of protein products or metabolites of introduced bacteria in vivo is also an important component for establishing safety mechanisms to deal with unintended effects to the host and ecosystem [46] when using genetic recombineering based treatment options. Other investigations centred on altering host-associated bacterial community dynamics, such as current work with bacteriophage therapies [47], present similar issues. We also recognize the importance of bioengineering tools such as genetic switches for continued efforts in this direction [3]. While such methodologies are not without their challenges, they offer potentially transformative options in an era of emerging diseases with limited treatment options. These results, therefore, present important insights for the possibility of microbiota restructuring via the introduction of recombinant bacteria as a way to ameliorate host disease and also for developing an appropriate ecological and evolutionary-based understanding of host disease and host–microbiota interactions.

Ethics

Experiments reported herein were compliant with all institutional ethical standards and permitting procedures. All Acris blanchardi used in this study were collected under an approved Nebraska Game and Parks Commission Scientific and Educational Permit (1006). All aspects of experimental work with live Acris blanchardi followed an approved University of South Dakota IACUC protocol (17-13).

Data accessibility

All scripts and models used for the analysis of results presented in this paper have been deposited in GitHub (https://github.com/kvasir7/Acris_microbiome) and are publicly accessible. All 16S rRNA gene sequences have been deposited in the NCBI Sequence Read Archive (BioProject: PRJNA435631). Raw experimental data have been provided as electronic supplementary material.

Authors' contributions

J.D.M. was involved in all experiments presented, data analysis and manuscript preparation; S.P.O. was involved in all in vitro experiments presented and also in manuscript preparation; E.S. was involved in all in vivo experiments and manuscript preparation; J.L.K. was involved in all aspects of experimental design, data analysis and manuscript preparation. All authors have given approval for publication and agree to be held responsible for all results reported herein.

Competing interests

We declare we have no competing interests.

Funding

Funding for sequencing was provided in part by an Institutional Development Award from the National Institute of General Medicine Sciences (NIGMS) of the National Institutes of Health (NIH; award number: P20GM103443). Computations supporting this project were performed on the Lawrence High-Performance Computing Cluster at the University of South Dakota, which is funded by the South Dakota Board of Regents and the National Science Foundation (NSF; award number: 1626516). The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent official views of NIGMS, NIH or NSF.