Triclosan-Selected Host-Associated Microbiota Perform Xenobiotic Biotransformations in Larval Zebrafish

Abstract

Microbiota regulate important physiologic processes during early host development. They also biotransform xenobiotics and serve as key intermediaries for chemical exposure. Antimicrobial agents in the environment may disrupt these complex interactions and alter key metabolic functions provided by host-associated microbiota. To examine the role of microbiota in xenobiotic metabolism, we exposed zebrafish larvae to the antimicrobial agent triclosan. Conventionally colonized (CC), microbe-free axenic (AX), or axenic colonized on day 1 (AC1) zebrafish were exposed to 0.16–0.30 μM triclosan or vehicle on days 1, 6, 7, 8, and 9 days post fertilization (dpf). After 6 and 10dpf, host-associated microbial community structure and putative function were assessed by 16S rRNA gene sequencing. At 10dpf, triclosan exposure selected for bacterial taxa, including Rheinheimera. Triclosan-selected microbes were predicted to be enriched in pathways related to mechanisms of antibiotic resistance, sulfonation, oxidative stress, and drug metabolism. Furthermore, at 10dpf, colonized zebrafish contained 2.5–3 times more triclosan relative to AX larvae. Nontargeted chemical analysis revealed that, relative to AX larvae, both cohorts of colonized larvae showed elevations in 23 chemical features, including parent triclosan and putative triclosan sulfate. Taken together, these data suggest that triclosan exposure selects for microbes that harbor the capacity to biotransform triclosan into chemical metabolites with unknown toxicity profiles. More broadly, these data support the concept that microbiota modify the toxicokinetics of xenobiotic exposure.

Microbes provide an important interface between the human body and its environment. In the gut, there are over 100-fold more unique microbial genes than human genes (Qin et al., 2010). Collectively, these microbes have a much larger metabolic repertoire compared to human cells (Koppel et al., 2017). They metabolize not only dietary compounds, such as fiber, but also pharmaceutical compounds and environmental chemicals (Spanogiannopoulos et al., 2016). Whereas host metabolism evolved to favor the catabolism and excretion of many xenobiotics, resident microbes often metabolize compounds in a way that supports microbial growth (Koppel et al., 2017). In so doing, microbiota can modify the toxicity profile of xenobiotics by performing biotransformations that affect chemical potency in the host. Examples include the prodrug sulfasalazine which is bioactivated by Lactobacillus and inhibited by antibiotics (Peppercorn and Goldman, 1972) and the industrial chemical melamine, which is transformed by gut microbes into the renal toxicant cyanuric acid (Ingelfinger, 2008; Zheng et al., 2013). Despite increasing evidence linking the gut microbiome with xenobiotic toxicity and efficacy, the complex interactions between microbiota metabolism and environmental chemicals remain poorly understood (Claus et al., 2016).

Triclosan is an antimicrobial chemical that has become ubiquitous in the environment (Yueh and Tukey, 2016). It is readily absorbed by the skin and gastrointestinal tract and metabolized to glucuronide and sulfate conjugates (Fang et al., 2010). Survey studies have identified triclosan in >75% of human urine, plasma, and breastmilk samples analyzed (Allmyr et al., 2006; Etzel et al., 2017; Li et al., 2015). In 2016, triclosan was banned in over-the-counter wash products by the U.S. Food and Drug Administration due to potential health risks, including endocrine disruption and antibiotic resistance (U.S. FDA 2016). However, it remains widely used in various household products, such as toothpaste, textiles, and toys. The degradation products of triclosan include methyltriclosan, dioxins, chlorophenols, and chloroform, several of which are routinely detected in aquatic systems (Dann and Hontela, 2011; Perez et al., 2013; Yueh and Tukey, 2016). Importantly, triclosan has been found to disrupt the gut microbiota in humans (Ribado et al., 2017), mice (Gao et al., 2017), and fish (Gaulke et al., 2016; Narrowe et al., 2015). Disruption to the gut microbiome has been associated with negative health outcomes such as obesity, neurological disorders, and cancer (Cho and Blaser, 2012; Lozupone et al., 2012), though notably there have been no direct links between triclosan-mediated impacts on the microbiome and adverse health effects.

Zebrafish (Danio rerio) are a valuable vertebrate model in which to study the role of the host microbiome on xenobiotic toxicity. Zebrafish are widely used in toxicology due to their small size, rapid development, optical transparency, external embryonic development, and fully defined genome (Howe et al., 2013). In addition, they contain a complex metacommunity of intestinal microbes that varies based on developmental stage (Burns et al., 2016; Roeselers et al., 2011; Stephens et al., 2016), much like that of humans (Lozupone et al., 2012). We previously reported that axenic (microbe-free) zebrafish or larvae developmentally exposed to antibiotics show a hyperactive behavioral phenotype with temporal dependency (Phelps et al., 2017), consistent with axenic mice models (Diaz Heijtz et al., 2011). Together, these studies provide further evidence that microbiota are critical for normal brain development and behavior (Diaz Heijtz et al., 2011; Phelps et al., 2017).

Although several studies have shown that environmentally relevant exposure to triclosan can restructure the gut microbiome (Gaulke et al., 2016; Narrowe et al., 2015; Ribado et al., 2017), the specific role of host and microbiota in responding to triclosan is unclear. To disentangle this relationship, here we employed a unique 3-cohort system in which zebrafish larvae were either conventionally colonized (CC), axenic then colonized at 1-day post fertilization (dpf), or axenic through the duration of the experiment. After exposing larvae to triclosan through early development, we assessed impacts on the microbial community structure and associated putative function. In addition, we measured triclosan and screened for the presence of triclosan metabolites. We predicted changes to the microbiome consistent with chemical exposure, and that microbially colonized larvae would have lower levels of triclosan and its metabolites. Our results suggest that triclosan exposure selects for microbes that both increase whole-body tissue concentrations of parent compound and harbor the capacity to biotransform triclosan into chemical metabolites with unknown toxicity profiles. More broadly, our data support the concept that the microbiome has a role in the toxicokinetics of xenobiotic exposure.

MATERIALS AND METHODS

Zebrafish husbandry.

All procedures involving zebrafish were approved by the Institutional Animal Care and Use Committee at the U.S. EPA National Health and Environmental Effects Research Laboratory. All methods were carried out in accordance with the relevant guidelines and regulations. A mixed wildtype adult zebrafish line (D. rerio) was generated and maintained as previously described (Phelps et al., 2017). Briefly, AB or Tupfel long fin wildtype strains were mixed into an unspecified EPA line 1 time per year, beginning in 2011. Zebrafish adults were housed in 6l tanks at an approximate density of 8 fish/l. Adults were fed Gemma Micro 300 (Skretting) once daily and shell free E-Z Egg (Brine Shrimp Direct) twice daily Mondays through Fridays and once daily on weekends. Zebrafish were maintained on a 14:10 light cycle at 28.5°C. To collect embryos, adults were bred every 2–3weeks by placing 60–100 adults in 10l angled static breeding tanks overnight. At 8 AM, adults were transferred to new bottom tanks containing treated reverse osmosis water (60 mg/l sodium bicarbonate and 0.4g/l Crystal Sea Bioassay Formula Marine Mix) and embryos were collected 40–60min later.

Axenic derivation and conventionalization.

Axenic (AX) zebrafish larvae were generated as previously described (Phelps et al., 2017). CC controls were generated by collecting embryos at 0dpf that were not been treated with antibiotics, povidone-iodine solution, or bleach. As a control for the derivation process, a subset of AX embryos was conventionalized with microbiota harvested from a standard aquaculture facility at 1dpf (axenic zebrafish colonized on day 1 or AC1) by modifying our previously published conventionalization process (Phelps et al., 2017). Briefly, fish facility water was syringe-filtered using a sterile 5μm filter to remove debris. In addition, a portion of 5 μm filtered fish facility water was syringe-filtered using a sterile 0.2 μm filter to generate microbe-free fish facility water. At 1dpf, CC and AC1 flasks received an 80% media change consisting of 10ml of microbe-containing 5 μm filtered fish facility water and 10ml of sterile 10% Hanks’ Balanced Salt Solution (HBSS). In comparison, AX flasks received 10ml of sterile 0.2 μm filtered fish facility water and 10ml of sterile 10% HBSS. All flasks were housed statically through 6dpf. From 6 to 9dpf, all flasks underwent a daily 80% media change and received 75 kilogray gamma-irradiated Gemma Micro 75 at a final concentration of 0.04% (vol/vol) to eliminate microbial contributions from the food source. Sterility testing was performed by visual inspection of inoculated and aerobically and anaerobically incubated tryptic soy agar (TSA) plates (samples collected on 1 and 6dpf), and nutrient broth, brain heart infusion, and sabouraud dextrose tubes (samples collected on 10 dpf) as previously described (Phelps et al., 2017). If samples from AX flasks showed microbial growth that was confirmed by 16S rRNA gene sequencing, those flasks were excluded from the study. If samples from AC1 flasks showed growth prior to conventionalization on day 1 that was confirmed via metagenomic sequencing, samples from contaminated flasks were excluded from the study.

Chemical exposures.

All exposures were performed in T25 tissue culture flasks containing 25ml of media incubated at 26°C on a 14:10h light:dark cycle. Triclosan (CAS: 3380-34-5; Lot: LRAA9502) was purchased from Sigma-Aldrich. Fifty millimolar stocks were prepared in anhydrous dimethyl sulfoxide (DMSO), aliquoted, and stored at −80°C. Each aliquot was thawed once, used, then discarded; 1000× working solutions were prepared in brown glass vials and stored in the dark for the duration of each 10-day experiment. For the range-finding experiment, CC embryos were exposed to 0.05, 0.09, 0.16, 0.30, or 1.04 μM triclosan or 0.1% DMSO on 1, 6, 7, 8, and 9 dpf (15 embryos per flask, n = 3 flasks). Eighty percent media renewals were conducted daily. Embryos were assessed on 6 and 10 dpf for malformations and death. EC₅₀ values were obtained from zebrafish developmental toxicity assay data by performing nonlinear regression using the log(agonist) versus normalized response—Variable slope equation $[\text{Y}=100/(1+{10}^{^}\left(\right({\text{LogEC}}_{50}-\text{X})*\text{HillSlope})\left)\right]$ (GraphPad Prism 7) (Catron et al., 2019). For the metagenomic sequencing and analytical chemistry experiment, CC, AX, or AC1 embryos or larvae were exposed to 0.16 or 0.3 μM triclosan or 0.1% DMSO on 1, 6, 7, 8, and 9dpf in concert with daily 80% media renewals with sterile 10% HBSS (Figure 1). To plan for potential microbial contamination of AC1 or AX groups and subsequent loss of samples, there were 4 replicate flasks of CC embryos, 5 replicate flasks of AC1 embryos, and 6 replicate flasks of AX embryos per treatment group (ie, 0.16 or 0.30 μM triclosan or 0.1% DMSO) included in the experiment. Each flask contained 40 embryos. On days 6 and 10, 8 larvae per flask were pooled to comprise a single biological replicate for metagenomic sequencing and another 8 larvae per flask were pooled for analytical chemistry. Each flask was a biological replicate. In addition to flasks containing zebrafish, blank fish-free flasks exposed to 0.16 or 0.30 μM triclosan or 0.1% DMSO were also generated (n = 4 flasks/concentration) to serve as quality controls for analytical chemistry analyses. Finally, media samples were collected from each flask prior to and directly following exposures conducted on 1 and 6dpf and before sample collection on 10dpf. The number of biological replicates that passed sterility requirements and were included in the study are shown in Supplementary Table 4 and listed in each figure legend.

Figure 1.

Generation of a zebrafish experimental system with different colonization statuses. To test whether host-associated microbes modify triclosan toxicokinetics and/or toxicodynamics, axenic (AX), conventionally colonized (CC), and axenic larvae colonized on day 1 (AC1) were developmentally exposed to triclosan on 1, 6, 7, 8, and 9 dpf. At 6 and 10 dpf, microbiota community structure and triclosan uptake and biotransformation were assessed. NTA = non-targeted analytical chemistry.

DNA sequencing of the 16S rRNA gene.

On 6 and 10dpf, zebrafish samples underwent DNA isolation, 16S rRNA gene amplification, library preparation, and DNA sequencing on an Illumina MiSeq instrument as previously described (Phelps et al., 2017). Positive and negative PCR control reactions were run with every 30 samples and sequenced to assess sequencing error and potential PCR contamination. Positive controls consisted of a mixture of equal concentrations of genomic DNA as described previously (Phelps et al., 2017). Negative controls consisted of 10 mM Tris-HCl at pH 8.5, which was used to dilute DNA extracts.

Analysis of 16S rRNA gene sequences.

MiSeq paired-end sequences were trimmed at a length of 250 bp and quality filtered at an expected error of less than 0.5% using USEARCH v7 (Edgar, 2013). Reads were analyzed using the QIIME 1.9.0 software package (Caporaso et al., 2010, 2011). The total number of sequence reads for each sample can be found in Supplementary Table 1. A cutoff of 500 reads per sample was used for downstream analyses. All samples met this criterion. A closed-reference operational taxonomic unit (OTU) table was generated within QIIME-1.9.0 with the Greengenes 13_8 reference database. After quality filtering, 713 sequences were included in the OTU table (Supplementary Table 2). Alpha and beta diversity analyses were performed using PRIMER 7 software (Primer-E v7.0.11) for total number of species (S), Margalef’s species richness (d), Pielou’s evenness index (J’), Simpson index (1 – λ’), Shannon diversity index (H’), Bray-Curtis similarities, nonmetric multidimensional scaling (NMDS), and heatmap analyses as previously described (Catron et al., 2019; Phelps et al., 2017). Species richness was also estimated using the iNEXT version 2.0.12 R package with default settings, q = 0, and endpoint = 15000 (Hsieh et al., 2016). Differences between observed and expected species richness were visualized using rarefaction curve analysis. Treatment differences in estimated species richness were analyzed using Kruskal-Wallis nonparametric test (p < .05) followed by Dunn’s pairwise comparisons (p < .05). For comparisons of Bray-Curtis similarity scores, a permutational multivariate analysis of variance (PERMANOVA) with Monte Carlo pairwise comparisons (p < .05) was used. For pairwise comparisons of bacterial taxa, we used a 2-sided White’s nonparametric t test (White et al., 2009) with a Storey false discovery rate (FDR) multi-test correction within STAMP v.2, with a bootstrapped 95% confidence interval (100 bootstrap replicates; corrected q-value < 0.05) that returned 1–11 significantly different taxa (Figs. 4D, 5D, and 5E). Stacked bar plots were generated using OTUs that contributed at least 1% to any of the sample totals. The KEGG orthology classification scheme was used for functional annotations (Kanehisa and Goto, 2000). Kruskal-Wallis and pairwise Wilcoxon tests were conducted (α = .05) using triclosan as the main categorical variable. Linear discriminant analysis scores > 2.0 were used to assess functional enrichment. PICRUSt (phylogenetic investigation of communities by reconstruction of unobserved states) functional predictions were generated as previously described (Catron et al., 2019). We report the functional predictions in which there were no significant differences between the DMSO control groups and in which the p-values for both CC and AC1 were less than .05.

Figure 4.

Microbiota community structure was disrupted in axenic larvae colonized on day 1 (AC1) triclosan (TCS)-exposed zebrafish at 6 dpf. 16S rRNA gene sequencing was performed on 6 dpf conventionally colonized (CC) and AC1 larvae developmentally exposed 0.16 or 0.30 μM triclosan or 0.1% DMSO. A, Nonmetric multidimensional scaling (NMDS) plot. B, Heatmap showing nonhierarchical clustering representing relative abundance of family-level taxonomy. C, Stacked bar plot reflecting genus-level taxa (n = 4–5). Each biological replicate was comprised of a pool of 8 larvae obtained from unique flasks. D, Extended error bar plots showing significant changes in genus-level taxa following exposure to 0.30 μM TCS in AC1 larvae. There were no significant changes in CC larvae. White’s nonparametric t test was used to identify significant (p ≤ .01) pairwise comparisons. n = 4 biological replicates with 10 larvae per replicate.

Figure 5.

Microbiota community structure was disrupted in conventionally colonized (CC) and axenic colonized on day 1 (AC1) triclosan (TCS)-exposed zebrafish at 10 dpf. 16S rRNA gene sequencing was performed on 10 dpf CC and AC1 larvae developmentally exposed 0.16 or 0.30 μM triclosan or 0.1% DMSO. A, Nonmetric multidimensional scaling (NMDS) plot. B, Heatmap showing nonhierarchical clustering representing relative abundance of family-level taxonomy. C, Stacked bar plot reflecting genus-level taxa (n = 4–5). Each biological replicate was comprised of a pool of 8 larvae obtained from unique flasks. Extended error bar plots showing significant changes in genus-level taxa following exposure to 0.30 μM TCS in (D) AC1 and (E) CC larvae. White’s nonparametric t test was used to identify significant (p ≤ .01) pairwise comparisons. n = 4 biological replicates with 10 larvae per replicate.

Analytical chemistry.

Stock solutions of triclosan and ¹³C₁₂ triclosan (Cambridge Isotope Laboratories) were prepared in methanol and stored at −20°C. Intermediate standards were prepared daily in acetonitrile. Matrix-matched calibration standards, blanks, and quality control samples were prepared by spiking exposure media or tissue homogenate. At 6 or 10dpf, pools of 8 larvae were anesthetized by rapid cooling, flash frozen in liquid nitrogen, and stored at −80°C. Zebrafish tissue samples were homogenized in 500 μl acetonitrile containing ¹³C₁₂ triclosan as an internal standard with approximately 150 mg 1.0 mm Dia zirconia/silica beads using a FastPrep 24 homogenizer. The homogenate was centrifuged for 15 min at 14000 rpm at 4°C. For analysis, 100 μl of supernatant was diluted with 100 μl of HPLC grade water. Media samples were prepared for analysis by dilution of 100 μl of flask media with 100 μl of acetonitrile containing ¹³C₁₂ triclosan. Targeted and nontargeted analytical methods were conducted by liquid chromatography-mass spectrometry (LC-MS) on an Agilent 1100 Series and MSD-TOF time of flight mass spectrometer equipped with electrospray ionization (ESI) source that was operated in negative ionization mode for targeted analytical chemistry. Negative ESI was used for targeted analysis, whereas positive and negative ESI were used for nontargeted analysis. LC separation was achieved using Kinetex EVO (50 for exposure media or 100 for zebrafish tissue) × 2.1 mm, C₁₈, 2.6 μm, 100 Å LC column (Phenomenex) with gradient elution at a flow rate of 300 μl/min using 0.4 mM ammonium formate in methanol and water. Source conditions were optimized for the [M–H]⁻ ion of triclosan at m/z 286.9435. ESI source conditions were as follows: nebulizer (psi) 30; gas flow (l/min) 10.0; gas temp (°C) 350; fragmentor (V) −80; and VCap (V) −1500. Integration, calibration, and quantitation were performed using Mass Hunter B.04.00. The method range in tissue was 1.73–173pmol/larva with a 0.3 pmol/larva method detection limit (MDL). Ninety-nine percent recovery was observed for samples spiked at 17.3 pmol/larva and matrix effects were observed to be less than 10%. The method range in exposure media was 1.52 nm to 3.45 μm with a 0.8nm MDL with matrix effects observed to be less than 10%. Batch results were accepted based on the following criteria: calibration curve based a minimum of 6 standards with a correlation coefficient of > 0.99 and calibration standards accuracy tolerance < 20% (< 30% at the lower limit of quantitation); quality control sample accuracy tolerance < 20% precision expressed as %relative standard deviation < 20% with > 75% of all individual quality control standards satisfying accuracy criteria; and blank response < MDL.

Targeted analytical data analysis.

Triclosan concentrations were measured across 88 total samples, collected from 18 unique groups (3 colonization statuses × Doses × days), with 3–6 replicate samples per group. Individual samples with failed sterility tests were excluded from statistical analyses. Therefore, a total of 78 tissue and media samples were analyzed (Supplementary Table 4). For the tissue samples, multiple linear regression models (SAS Proc Mixed) were used to identify significant predictors of triclosan (pmol/larva). Samples that were < LOD (0.28 pmol/larva; n = 11) were replaced with the value LOD/sqrt(2). A square-root transformation was applied to all measured and imputed values to satisfy modeling assumptions related to normality and homoscedasticity. Separate models were used to evaluate measurements at 6 and 10dpf. In each model, consideration was given to fixed effects for “Status” (CC, AC1, and AX), “Dose” (DMSO control, 0.16 μM triclosan, and 0.30 μM triclosan), and their interaction (“Status × Dose”). Fixed effects were retained in the model at a significance level of p ≤ .05 using backwards stepwise elimination. Pairwise comparisons across groups were evaluated using differences of least squares means and Bonferroni-adjusted p-values (p < .05). The media samples were analyzed as above, with the exception that the DMSO controls were not included in the analyses, and a square-root transformation was not applied.

Nontargeted analytical chemistry data analysis.

Zebrafish cellular lysate used for targeted triclosan analysis (n = 78) was also analyzed using a nontargeted approach to screen for metabolites associated with triclosan exposure and microbial colonization status. Raw data were processed using Agilent Profinder vendor software (v.8.00) for molecular feature extraction and integration using default settings with modifications detailed in Supplementary Table 3. Chemical features were tentatively assigned based on formula hits against the EPA Comptox Chemicals Dashboard (https://comptox.epa.gov/dashboard/; last accessed September 9, 2019) and an Agilent PCDL of the METLIN database (doi: 10.1097/01.ftd.0000179845.53213.39). Chemical features and assignments were filtered using an in-house data processing script in R (available upon request) using analytical criteria as follows and described in Figure 2. First, replicate features were removed as duplicates if multiple species were detected within a 10 ppm mass window, within a 0.05 retention time window, and with at least 1 matching intensity on a per-sample basis. Second, poorly reproducible features were removed if they were detected in < 66% of sample replicates and/or had analytical replicate coefficient of variation (CV) values < 0.2. Based on the prevalence of sparse low intensity signals, features were removed if the mean intensity was < 100 000 in every sample group. Chemical assignments were also rejected if the PCDL match scoring function was < 90. Triclosan was only detected in 10 dpf samples. Therefore, 10 dpf data were selected for feature identification via 3 within colonization status comparisons (0.30 μM triclosan vs 0.1% DMSO in CC, AX, or AC1 cohorts) to identify chemical features induced following exposure to triclosan. In addition, 3 between colonization status comparisons (CC—0.30 μM triclosan vs AX—0.30 μM triclosan; AC1–0.30 μM triclosan vs AX—0.30 μM triclosan; and CC—0.30 μM triclosan vs AC1–0.30 μM triclosan) were made to identify chemical features that are dependent on the presence of host-associated microbiota. Using a Mann-Whitney U test, features of interest were selected using threshold cutoffs (Fold-Change > 2, Q < 0.05).

Figure 2.

Host-associated microbiota biotransform triclosan into triclosan sulfate. Zebrafish cellular lysate used for targeted triclosan analysis was analyzed by a non-targeted screening approach to identify metabolites associated with triclosan exposure and microbial colonization status. Nontargeted analysis was performed on samples obtained from axenic (AX), axenic colonized on day 1 (AC1), and conventionally colonized (CC) larvae developmentally exposed 0.16 or 0.30 μM triclosan or 0.1% DMSO (n = 3–5). Each biological replicate was comprised of a pool of 8 larvae. A, Biotransformation scheme. SULT, sulfotransferase; UGT, UDP-glucuronosyltransferase; p450, cytochrome p450; MeT, methyltransferase. B, Nontargeted analysis workflow describing chemical feature filtering criteria. Between-group comparisons were performed on 10dpf larvae exposed to 0.30 μM triclosan. Features of interest were selected using threshold cutoffs (Fold-Change > 2, Q < 0.05).

Data availability.

The datasets generated during the current study are available by searching for the manuscript title at https://catalog.data.gov, last accessed September 9, 2019.

RESULTS

A Multi-Colonization Status Experimental System Was Used to Test Whether Microbiota Modify the Toxicokinetics of Triclosan Exposure

In order to determine whether host-associated microbiota modify the toxicokinetics of xenobiotic exposures, we developed and characterized an experimental system comprised of CC, microbe-free axenic (AX), and AC1 zebrafish larvae (Phelps et al., 2017; Figure 1). To determine concentrations of triclosan that did not result in malformations or lethality, a range-finding experiment was performed in CC zebrafish exposed to 0.5–1.04 μM triclosan (quarter-log spacing) or 0.1% DMSO on 1, 6, 7, 8, and 9 dpf. Developmental toxicity, shown as percent abnormal larvae, was visually assessed at 6 dpf (Figure 3A) and 10 dpf (Figure 3B). Curve-fitting revealed the half-maximal concentrations (EC₅₀) for developmental toxicity to be 0.82 μM at 6 dpf and 0.35 μM at 10 dpf. The no observable effect concentration at 10 dpf was determined to be 0.30 μM triclosan. Therefore, 0.30 and 0.16 μM (quarter-log spacing) were used as test concentrations for the remainder of the study.

Figure 3.

Characterization of a zebrafish experimental system with different colonization statuses. At (A) 6 days post fertilization (dpf) and (B) 10 dpf, developmental toxicity was assessed. Asterisk indicates significant difference from 0.1% DMSO control. Significance determined by one-way ANOVA with a Tukey multiple comparison post hoc test (*p < .0001). Nonlinear regression was performed with Hill slope curve-fitting for half-maximal EC₅₀ value determinations. n = 3 replicate flasks with 15 larvae per flask.

At 6 dpf, Colonization Status Was the Main Determinant of Host-Associated Microbiota Community Structure in Zebrafish Exposed to Triclosan

To assess the effect of triclosan exposure on microbial community structure, the 16S rRNA gene was sequenced in 6 dpf zebrafish developmentally exposed to 0.16 or 0.30 μM triclosan or 0.1% DMSO on 1 dpf. NMDS revealed community separation by colonization status in the first dimension and by dose along the second dimension, with a greater separation for AC1 than CC (Figure 4A). In support of this, unsupervised hierarchical clustering showed that the major discriminating factor was colonization status (ie, AC1 vs CC), not exposure group (ie, DMSO vs triclosan; Figure 4B). Genus-level taxonomy revealed marked differences in AC1 and CC cohorts, regardless of exposure group (Figure 4C). A White’s nonparametric t test with a Storey FDR multiple test correction demonstrated that in the AC1 cohort, Methylobacterium was decreased following exposure to 0.30 μM triclosan, although the detected difference only reflected a 0.2% change in taxon abundance (Figure 4D). There were no significant genus-level differences in the CC larvae exposed to 0.30 μM triclosan, relative to the DMSO-exposed control.

Overall, Bray-Curtis percent similarity scores were significantly different when comparing within exposure groups (PERMANOVA p = .015). However, there were no significant genus-level differences in the CC larvae exposed to 0.30 μM triclosan, relative to the DMSO-exposed control (Supplementary Figure 1A). Between exposure group comparisons also resulted in significantly different Bray-Curtis percent similarity scores (PERMANOVA p < .0001), with pairwise comparisons suggesting changes in community diversity associated with triclosan exposure in the AC1 cohort (Supplementary Figure 1C). Alpha diversity metrics showed no differences in CC or AC1 larvae exposed to triclosan at 6 dpf (Supplementary Figs. 2A–E).

At 10 dpf, Triclosan Effects on Microbial Communities Converged in Colonized Cohorts

Microbiota community structure was also assessed at 10 dpf in larvae exposed to 0.16 or 0.30 μM triclosan or 0.1% DMSO at 1, 6, 7, 8, and 9 dpf. NMDS analysis showed that the primary effect was dose (separation along the first dimension) and the secondary effect was colonization status (separation along the second dimension; Figure 5A). Compared to community structure at 6 dpf, there was a strong convergence noted (50% similarity) in CC and AC1 cohorts exposed to 0.16 or 0.30 μM triclosan. Further in contrast to results at 6 dpf, where colonization status was the primary driver in unsupervised clustering (Figure 4B), results at 10 dpf showed exposure group (0.1% DMSO vs triclosan) to be the major discriminating variable in unbiased clustering of taxonomic data (Figure 5B). No concentration-specific differences in community structure were observed in CC or AC1 larvae exposed to increasing concentrations of triclosan (Figs. 5A–C). Qualitative observations of genus-level taxonomy in the stacked bar plot showed that although differences in baseline structure in CC and AC1 0.1% DMSO controls were present, triclosan exposure universally selected for an expansion of Rheinheimera (blue) and, to a lesser extent, Pseudomonas (green) in CC and AC1 cohorts (Figure 4C). This was supported by statistical analysis of taxonomic profiles (White’s nonparametric t test with a Storey FDR multiple test correction), in which we found that exposure to 0.30 μM triclosan resulted in increased Rheinheimera in both the CC and AC1 cohorts (Figs. 5D and 5E) and Pseudomonas in the AC1 cohort (Figure 5D). In the AC1 cohort, triclosan exposure also resulted in a significant decrease in Chitinimonas, Acidovorax, Fluviicola, Sphingopyxis, and Agrobacterium, and a significant increase in Vogesella, Variovorax, Stenotrophomonas, and Pseudoxanthomonas (Figure 5D). In the CC cohort, exposure to triclosan also resulted in a decrease in Acidovorax (Figure 5E).

In contrast to 6 dpf (Supplementary Figure 1A), within group Bray-Curtis pairwise percent similarity scores were significantly elevated in triclosan-exposed groups at 10 dpf, relative to the 0.1% DMSO control, which might suggest reduced diversity (Supplementary Figure 1B). This notion is further supported by the observance of reduced percent similarity in between group comparisons when community structure in triclosan-exposed flasks was directly compared to 0.1% DMSO control flasks (Supplementary Figure 1D). In measures of alpha diversity, relative to 0.1% DMSO controls, there was a significant reduction in species richness in AC1 larvae exposed to triclosan but no other significant differences in CC or AC1 larvae (Supplementary Figs. 3A–E). Further analysis of observed versus estimated species richness suggested that maximum sequencing coverage was not obtained (Supplementary Figure 4A). However, estimated species richness was also unaffected following exposure to 0.16 or 0.3 μM triclosan, relative to the DMSO control (Supplementary Figure 4B).

Triclosan-Selected Microbes Differed in Their Functional Profile

To identify putative triclosan-induced changes in global functional profiles of zebrafish microbial communities, PICRUSt was applied to predict metagenome function using 16S rRNA gene sequencing data (Figure 6A). Linear discriminant analysis effect size (LefSe) analysis showed significant enrichment or repression of 20 Level 3 KEGG predictions upon triclosan exposure (Figure 6 and Supplementary Figure 4). Enriched pathways included biofilm formation (CC, p = .023; AC1, p < .01), sulfur relay system (CC, p = .024; AC1, p < .01), alanine, aspartate, and glutamate metabolism (CC, p = .025; AC1, p < .01), monobactam biosynthesis (CC, p = .023; AC1, p < .01), biosynthesis of unsaturated fatty acids (CC, p = .018; AC1, p < .01), and drug metabolism other enzymes (CC, p = .023; AC1, p < .01; Figure 6B). In other pathways, we also observed repressed functions following triclosan exposure at 10 dpf, including quorum sensing (CC, p = .024; AC1, p = .042), dioxin degradation (CC, p = .024; AC1, p < .01), xylene degradation (CC, p = .035; AC1, p < .01), and atrazine degradation (CC, p = .031; AC1, p < .01; Figure 6C). Additional pathways with significant enriched and repressed changes in function are shown in Supplementary Figure 4.

Figure 6.

Developmental exposure to triclosan altered predicted KEGG pathways. A, Relative abundances of all Level 2 pathways are depicted for each concentration in circular barplots. Triclosan exposure (B) enriched and (C) depressed select Level 3 KEGG pathways shown as boxplots. Within each individual boxplot, different letters indicate significant differences (Kruskal-Wallis test following by Dunn’s pairwise comparisons, p < .05). 0 μM = DMSO vehicle control. n = 4 biological replicates with 10 larvae per replicate.

Colonized Zebrafish Contained 2–3× More Triclosan Than Microbe-Free AX Larvae at 10 dpf

To assess the effect of triclosan-selected microbial colonization on triclosan uptake, targeted analytical chemistry was performed at 6 and 10 dpf in larvae exposed to 0.16 or 0.30 μM triclosan or 0.1% DMSO on 1, 6, 7, 8, and 9 dpf. At 6 dpf, there was a significant effect of dose (p < .0001), but no effect of colonization status or interaction between the 2 terms (Figure 7A). In comparison, at 10 dpf, there was a significant main effect of triclosan dose (p < .0001), colonization status (p < .0001) and the interaction between dose and status (p = .0081) in which colonized zebrafish contain roughly 2–3 times more triclosan than their microbe-free counterparts (Figure 7B). Triclosan concentrations were also quantified in flask media samples at 1 and 6 dpf before and after dosing and at the end of the study, at 10 dpf. At 1 dpf before dosing, triclosan levels were undetectable in all exposure groups. At all subsequent time points, there was significantly more triclosan detected in fish-free flasks, relative to AX, CC, and AC1 flasks, indicating that the presence of zebrafish leads to increased chemical uptake (Supplementary Figs. 5B–D). Directly after dosing on 1 and 6 dpf, there were roughly similar levels of triclosan in empty, AX, CC, and AC1 empty flasks (Supplementary Figs. 5A and 5C). At 6 dpf before dosing and 10 dpf, elevated levels of triclosan were measured in colonized flasks (ie, CC and AC1) relative to AX flasks (Supplementary Figs. 5B and 5D).

Figure 7.

Colonized zebrafish contain higher concentrations of parent triclosan. Targeted mass spectrometry was performed (A) 6 dpf and (B) 10 dpf axenic (AX), axenic colonized on day 1 (AC1), and conventionally colonized (CC) larvae developmentally exposed 0.16 or 0.30 μM triclosan or 0.1% DMSO (n = 3–5). Each biological replicate was comprised of a pool of 8 larvae. Overall effects of status, dose, and the interaction were determined using a fixed effects model. Significance between groups was determined using the differences of the least square means. Different letters indicate statistical significance (p < .05).

Host-Associated Microbes Biotransformed Triclosan

Nontargeted analysis was performed to determine whether triclosan-selected microbiota biotransform triclosan. Triclosan may be sulfated, glucuronidated, and/or methylated by both microbes and the host (Figure 2A) (Fang et al., 2010). A total of 1434 features obtained by nontargeted analysis in positive and negative ionization modes were assessed according to the chemical feature filtering strategy depicted in Figure 2B. Importantly, features were retained if a > 2-fold-change difference was detected between control and treated groups within each colonization status (eg, AX 0.1% DMSO vs AX 0.30 μM triclosan). Features were also selected if between colonization status comparisons for the highest exposure group (eg, AX 0.30 μM triclosan vs CC 0.30 μM triclosan) were > 2-fold different. Using this strategy, the parent compound was detected in 10 dpf, but not 6 dpf, samples. Detailed analyses therefore focused on the 10 dpf dataset where 78 chemical features of interest were selected using threshold cutoffs (Fold-change > 2, Q < 0.05). As predicted, CC, AX, and AC1 zebrafish exposed to triclosan revealed changes in global chemical metabolism profiles when making comparisons to colonization status-specific 0.1% DMSO controls (Supplementary Figs. 6A–C). Between colonization status comparisons showed that most features were overrepresented in colonized cohorts (CC or AC1) relative to microbe-free AX larvae (Red dots in Figs. 8A and 8B). Also of interest, a comparison of triclosan-exposed colonized cohorts (ie, CC 0.30 μM triclosan vs AC1 0.30 μM triclosan) resulted in only a small number of significantly different chemical features (Figure 8C), suggesting that the derivation on 0 dpf and conventionalization on 1 dpf had only a modest effect on chemical metabolism at 10 dpf.

Figure 8.

Host-associated microbiota biotransform triclosan into triclosan sulfate. Zebrafish cellular lysate used for targeted triclosan analysis was analyzed by a non-targeted screening approach to identify metabolites associated with triclosan exposure and microbial colonization status. Nontargeted analysis was performed on samples obtained from axenic (AX), axenic colonized on day 1 (AC1), and conventionally colonized (CC) larvae developmentally exposed 0.16 or 0.30 μM triclosan or 0.1% DMSO (n = 3–5). Each biological replicate was comprised of a pool of 8 larvae. Volcano plots depicting (A) CC versus AX, (B) AC1 versus AX, or (C) CC versus AC1 comparisons are shown. Dotted lines indicate threshold cutoffs (Fold-Change > 2, Q < 0.05). Between-group comparisons were performed on 10 dpf larvae exposed to 0.30 μM triclosan. Features of interest were selected using threshold cutoffs (Fold-Change > 2, Q < 0.05). Nontargeted analysis data depicting the relative abundance of (D) triclosan and (E) triclosan sulfate feature counts are shown. The overall effects of status, dose, and the interaction of status and dose were determined using a fixed effects model.

A comparison of chemical features that were similarly abundant (based on identity and directionality) in CC and AC1 cohorts exposed to triclosan showed that, relative to AX larvae, most chemical features (23/24) were overrepresented in both colonized cohorts at roughly the same magnitude (Table 1). Similar to targeted chemistry data shown in Figure 7B, nontargeted analysis showed elevated levels of parent triclosan (C₁₂H₇Cl₃O₂) in colonized larvae relative to microbe-free axenic larvae (green dots in Figs. 8A–D; Table 2). The single chlorinated chemical feature that was detected by nontargeted analysis was predicted (Level 3; Schymanski et al., 2014) to be triclosan sulfate (C₁₂H₇Cl₃O₅S), a reaction product catalyzed by sulfotransferases (Figure 2A). In line with parent nontargeted analysis data (Figure 7A), higher levels of triclosan sulfate were detected in colonized zebrafish, relative to AX larvae (light purple dots in Figs. 8A–C and 8E; Table 3).

Table 1.

Concordant Nontargeted Analysis Chemical Features in CC and AC1 Zebrafish Larvae Exposed to 0.30 μM Triclosan (TCS), Relative to Microbe-Free AX Larvae Exposed to 0.30 μM Triclosan

	CC 0.30 μM TCS vs AX 0.30 μM TCS		AC1 0.30 μM TCS vs. AX 0.30 μM TCS		CC 0.30 μM TCS vs AC1 0.30 μM TCS
Compound (Chemical Feature)	Mann-Whitney p-Value	Log2(Fold-Change)	Mann-Whitney p-Value	Log2(Fold-Change)	Mann-Whitney p-Value	Log2(Fold-Change)
C11H16O2	.021	4.832	.014	4.701	Na	Na
C12H7Cl3O2	.021	2.415	.014	1.957	0.050	1.234
C12H7Cl3O5S	.021	4.512	.014	1.578	0.050	2.860
C6H18O3Si3	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
C17H18N4	.021	4.162	.014	3.316	Na	Na
C19H43NO4P	.021	5.902	.014	5.001	Na	Na
C25H38O4	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
C25H26N4O2	.014	≥ 2 × 1016	.011	< 2 × 1016	Na	Na
C31H43NO4	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
157.1465@9.705995	.021	2.655	.014	2.457	Na	Na
182.0731@8.065999	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
220.1104@11.032001	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
284.2167@12.234006	.021	3.125	.014	3.198	Na	Na
294.2554@12.38701	.021	5.031	.027	3.799	Na	Na
294.3061@11.733001	.047	≥ 2 × 1016	.032	≥ 2 × 1016	Na	Na
330.3412@12.901996	.047	≤ 2 × 1016	.029	≤ 2 × 1016	Na	Na
358.3101@13.817006	.021	5.684	.014	6.574	Na	Na
521.6332@12.490998	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
549.3783@13.253999	.047	≥ 2 × 1016	.032	≥ 2 × 1016	Na	Na
567.4351@12.112	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
569.3486@12.298003	.047	≥ 2 × 1016	.032	≥ 2 × 1016	Na	Na
589.32@12.115998	.014	≥ 2 × 1016	.011	≥ 2 × 1016	Na	Na
624.6512@12.6849985	.018	3.799	.013	4.064	Na	Na
681.0613@13.628999	.014	≥ 2 × 1016	.032	≥ 2 × 1016	Na	Na

Chemical features (rows) were compared across all colonization statuses (e.g., CC 0.30 μM triclosan vs AX 0.30 μM triclosan). Abbreviations: AC1, axenic colonized on day 1; AX, axenic; CC, conventionally colonized; Na, not applicable. Colors represent colonization status.

Table 2.

Nontargeted Analysis Fixed Effects Model Results for Triclosan (C₁₂H₇Cl₃O₂)

Effect	Status		Triclosan (μM)		Estimate	Standard Error	DF	t Value	Pr > |t|	Adjusted p
Status	CC	AX			0.905	0.122	34	7.42	< .0001	< .0001
Status	AC	AX			0.605	0.116	34	5.22	< .0001	< .0001
Status	CC	AC			−0.300	0.115	34	−2.6	.014	.041
Dose			0.30	0	1.292	0.119	34	10.8	< .0001	< .0001
Dose			0.16	0	0.884	0.118	34	7.52	< .0001	< .0001
Dose			0.30	0.16	0.407	0.115	34	3.55	.001	.004

Overall effects of status and dose on levels of triclosan were determined using a fixed effects model. Significance between groups was determined using the differences of the least square means with a Bonferroni multiple test correction. Abbreviations: AC1, axenic colonized on day 1; AX, axenic; CC, conventionally colonized; DF, degrees of freedom. Colors represent colonization status.

Table 3.

Nontargeted Analysis Fixed Effects Model Results for Triclosan Sulfate (C₁₂H₇C_l3O₅S)

Effect	Status		Triclosan (μM)		Estimate	Standard Error	DF	t Value	Pr > |t|	Adjusted p
Status	CC	AX			1.368	0.181	23	7.55	< .0001	< .0001
Status	AC1	AX			0.510	0.171	23	2.97	.0068	.0203
Status	CC	AC1			−0.858	0.177	23	−4.86	< .0001	.0002
Dose			0.30	0.16	0.415	0.144	23	2.89	.0083	.0083

Overall effects of status and dose on levels of triclosan sulfate were determined using a fixed effects model. Significance between groups was determined using the differences of the least square means with a Bonferroni multiple test correction. Abbreviations: AC1, axenic colonized on day 1; AX, axenic; CC, conventionally colonized; DF, degrees of freedom. Colors represent colonization status.

DISCUSSION

Here, we used an established larval zebrafish model and unique 3-cohort design to investigate the role of host-associated microbiota in triclosan metabolism. We found that exposure to environmentally-relevant levels of triclosan (Kolpin et al., 2002; Yueh and Tukey, 2016) during early development selects for microbes that likely function to both increase tissue concentrations of the compound and biotransform it into metabolites with unknown toxicity. Triclosan-selected microbes were enriched for unique functions, several of which may relate to the mechanism conferring resistance. Enriched functions are related to pathways of sulfonation, oxidative stress, and drug metabolism. Several putative microbial functions were also depressed in triclosan-exposed larvae, including the degradation of dioxin, xylene, and atrazine, signal transduction pathways, and cell-to-cell communication. Further, we found that relative to axenic (ie, microbe-free) larvae, chemical profiles were dramatically altered in both cohorts of colonized larvae.

We found that compared to axenic larvae, those colonized with microbes had higher levels of triclosan metabolites, and specifically, putatively more triclosan sulfate. We suggest this provides evidence that host-associated microbes are responsible for the observed biotransformation of triclosan. Triclosan is metabolized through conjugation reactions to glucuronide, methyl, and sulfate conjugates. When exposure levels are below 1 μM, as in our study, sulfonation is expected to be the major metabolic pathway (Yueh and Tukey, 2016). Based on our results, microbiota appear to drive the sulfonation reaction. In support of this hypothesis, we found a significant enrichment in the sulfur relay system in triclosan-exposed larvae. In other studies, gut microbes thiolate the triclosan oxygen atom, which can then be subsequently oxidized to the sulfonic acid form (Numata et al., 2006; van Bergen et al., 2014), a phenomenon that may also be occurring in our system. The possibility remains that axenic zebrafish may contain a compromised xenobiotic metabolizing system that may account for observed differences in triclosan biotransformation. However, in support of microbiota-mediated xenobiotic transformation, microbial communities in sludge treated with triclosan generate triclosan sulfate (Chen et al., 2015). Also, in catfish, triclosan sulfonation occurs at a higher rate in the intestine, whereas triclosan glucuronidation occurs at a higher rate in the liver (James et al., 2012). These studies collectively suggest that environmental microbes and host-associated microbes in multiple fish species generate triclosan sulfate; however, future work should aim to further characterize the specific role of microbiota in the metabolism of triclosan and other environmental chemicals.

The microbiome serves a central role in host health, in part by providing resistance to colonization by pathogenic species (Baumler and Sperandio, 2016). Perturbations through environmental exposures, such as antibacterial products, can change the composition of bacterial taxa in the gut and result in increased host susceptibility to opportunistic pathogens (Modi et al., 2014). Antibiotics, or antimicrobial agents that target bacteria, can enrich the presence of resistance genes, increase the mobilization of genetic elements, and open niches to pathogenic intrusion (Modi et al., 2014). For example, in humans, antibiotic exposure can lead to an expansion of the diarrhea-causing Clostridium difficile (Leffler and Lamont, 2015). Similarly, exposure to triclosan has been shown to promote nasal susceptibility to Staphylococcus aureus, a common opportunistic bacterial pathogen (Syed et al., 2014). Here, we found that triclosan exposure increased relative abundance of 2 genera of aerobic Gram-negative bacteria, Rheinheimera and Pseudomonas. Pseudomonas species can be important human pathogens and strongly resistant to most antibiotics (Stover et al., 2000). Rheinheimera is less studied but commonly identified in aquatic systems and in soil (Presta et al., 2017; Sheu et al., 2018).

Triclosan functions as an antibacterial agent by inhibiting function of the gene fabI, which encodes the enzyme enoylacyl-carrier protein reductase (ENR) (Chuanchuen et al., 2001; McMurry et al., 1998). ENR is a vital enzyme in the biosynthesis of bacterial type II fatty acids (McMurry et al., 1998). Several mechanisms conferring resistance to triclosan have been identified, including mutations in fabI, Fab protein overexpression, active chemical efflux, and expression of alternate Fab proteins, as in the case of Pseudomonas (Chuanchuen et al., 2001). Based on microbiota structural data, significantly enriched and repressed bacterial functions were predicted for triclosan-selected microbes. Of particular interest are the putative bacterial functions enriched in triclosan-exposed zebrafish which might provide context for our data showing elevated levels of parent triclosan and triclosan sulfate in colonized larvae relative to axenic zebrafish. Here, we observed enrichment in several pathways that may relate to triclosan-dependent selection mechanisms. It is important to note that triclosan-selected microbes likely include taxa that exhibit xenobiotic resistance and/or microbes that either succumb to the chemical at a slower rate or exhibit swifter repopulation kinetics. Regardless, changes in structural data provide important context for the observed differences in detected levels of triclosan and biotransformation products. The first enriched pathway of note in triclosan-exposed zebrafish was monobactam synthesis. Monobactams are beta-lactam antibiotics that target type II fatty acid biosynthesis, the same mechanism through which triclosan acts. In addition, the monobactam biosynthesis pathway includes alanine, aspartate, and glutamate metabolism (Kanehisa and Goto, 2000), which was also enriched in triclosan-exposed larvae. More work is needed to understand whether specific monobactam-producing strains of Pseudomonas use this strategy to increase competitiveness in complex communities exposed to triclosan. The second triclosan-enriched pathway potentially related to triclosan selection is the biosynthesis of unsaturated fatty acids. Long-chain unsaturated fatty acids have antibacterial properties that function by inhibiting fatty acid biosynthesis (Zheng et al., 2005). Indeed, we observed enrichment in the biosynthesis of unsaturated fatty acids following triclosan exposure in both cohorts of colonized zebrafish. Some strains of Rheinheimera are also known to produce high levels of these unsaturated fatty acids (Yoon et al., 2007). This raises the possibility that the enrichment of triclosan observed in colonized zebrafish in the current study may be related, in part, to its capacity for unsaturated fatty acid biosynthesis. Lastly, biofilm formation was enriched, which functions to protect bacterial cells from stress, including antimicrobial exposure (Mah and O’Toole, 2001). Importantly, these imputed gene functions are predictions, useful for generating hypotheses, but limited in that they do not provide direct evidence of microbial community function. It should also be noted that PICRUSt imputed functional predictions required a closed-reference OTU table. Because unassigned taxa were discarded prior to analysis, there may be additional chemical-dependent changes in putative microbial functions that were not captured in the current analysis.

To put these functional predictions in context, we measured concentrations of triclosan and screened for triclosan metabolites. Contrary to our initial hypothesis, we observed increased levels of triclosan and triclosan sulfate in larvae that were colonized with microbes. One explanation for this novel observation is that triclosan-selected bacteria in the gut may be functioning as a localized sink that increases and maintains chemical levels. There is a physical difference between colonized larvae, which have a distended gut, and axenic larvae, which have an empty gastrointestinal tract (Phelps et al., 2017). We speculate that the presence of microbes may change the flow kinetics of triclosan through the gastrointestinal tract and function to increase up-take in the host, possibly through the increased surface area provided by triclosan-selected microbes. Three other potential sinks within this closed system are the bacterial communities in the skin, water, and biofilm in the flask. The microbiota in these sinks may metabolize the chemical or transport it through uptake or efflux proteins (Blanco et al., 2016; Catron et al., 2019), resulting in increased complexity of the exposure system relative to axenic cultures. Furthermore, sampling on 6 and 10 dpf (ie, “before dosing”) revealed increased levels of triclosan in media samples from flasks containing colonized larvae. There are several possibilities that may explain this novel finding. First, triclosan-selected microbes, indifferent to the action of triclosan, may allow the chemical to accumulate, resulting in its detection where microbes are present, including the media and host compartments. Second, in the absence of microbial sinks (ie, host, media, and flask biofilm), triclosan will likely be metabolized by the fish and removed from the closed system (James et al., 2012). Although other studies have found increased levels of chemical metabolites in colonized relative to axenic animals (Zimmermann et al., 2019), our finding of higher levels of the parent compound in both colonized larvae and media samples appears novel and may be specific to characteristics of the zebrafish experimental system.

Our findings support previous studies that examined the effects of xenobiotic exposure on the fish microbiome. Juvenile fathead minnows (Pimephales promelas) exposed to environmentally relevant levels of triclosan (100–1000 ng/l), as well as adult zebrafish exposed to higher levels (approximately 35 μg per day), both showed community disruption and decreased bacterial diversity following chemical exposure (Gaulke et al., 2016; Narrowe et al., 2015). An increase in Pseudomonas appears to be robust in both studies, as well as our own, despite differences in study design and lifestage of the animals. In a related study from our lab, we previously identified the effects from exposure to bisphenols on microbiota in zebrafish larvae (Catron et al., 2019). There are several parallels to note with the present study. First, we found that bisphenol A exposure also selected for both Rheinheimera and Pseudomonas and reduced the abundance of Chitinimonas (Catron et al., 2019). Furthermore, chemical exposure led to a predicted enrichment in the drug metabolism pathway (Catron et al., 2019). Examining whether there are long-term phenotypic effects that arise from chemical disruption to the microbiome during early development will be an important next step.

Humans are routinely exposed to antimicrobial agents that can potentially disrupt microbiota. Such changes can not only modify chemical toxicity via microbial metabolism, but also increase the transfer of antibiotic-resistant genes (Khan et al., 2016, 2018; Modi et al., 2014). Interactions between antibiotics and other xenobiotics can further exacerbate these effects. For example, coexposure of triclosan and bisphenol A results in increased chemical levels in certain tissues (Pollock et al., 2014). A recent report from the National Academies of Sciences, Engineering, and Medicine noted the need for risk assessment approaches to include an evaluation of the interaction between environmental chemicals and the microbiome (National Academies of Sciences, Engineering, and Medicine, 2018). Our study further highlights this need. We found that microbes increased levels of triclosan and biotransformed it into chemical metabolites of unknown toxicity and that triclosan exposure selected for microbes with different putative functional profiles. Together, these findings suggest that selection of triclosan-selected microbes can affect biotransformation of the parent compound. The microbiome plays a key intermediary role between the environment and the host, and shifts in its functional profile, as observed here, may have important consequences for health.