Assessing the Role of the Gut Microbiome in Methylmercury Demethylation and Elimination in Humans and Gnotobiotic Mice

Abstract

The risk of methylmercury (MeHg) toxicity following ingestion of contaminated foodstuffs (e.g., fish) is directly related to the kinetics of MeHg elimination among individuals. Yet, the factors driving the wide range of inter-individual variability in MeHg elimination within a population are poorly understood. Here, we investigated the relationship between MeHg elimination, gut microbiome demethylation activity, and gut microbiome composition using a coordinated human clinical trial and gnotobiotic mouse modeling approach together with metagenomic sequence analysis. We first observed MeHg elimination half-lives (t_1/2) ranging from 28 to 90 days across 27 volunteers. Subsequently, we found that ingestion of a prebiotic-induced changes in the gut microbiome and mixed effects (increased, decrease, and no effect) on elimination in these same individuals. Nonetheless, elimination rates were found to correlate with MeHg demethylation activity in cultured stool samples. In mice, attempts to remove the microbiome via generation of germ-free (GF) animals or through antibiotic (Abx) treatment both diminished MeHg demethylation to a similar extent. While both conditions substantially slowed elimination, Abx treatment resulted in significantly slower elimination than the GF condition, a indicating an additional role for host-derived factors in supporting elimination. Human fecal microbiomes transplanted to GF mice restored elimination rates to that seen in control mice. Metagenomic sequence analysis of human fecal DNA did not identify genes encoding protein typically involved in demethylation (e.g. merB, organomercury lyase). However, the abundance of several anaerobic taxa, notably Alistipes onderdonkii, were positively correlated with MeHg elimination. Surprisingly, mono-colonization of GF free mice with A. onderdonkii did not restore MeHg elimination to control levels. Collectively, our findings indicate the human gut microbiome uses a non-conventional pathway of demethylation to increase MeHg elimination that relies on yet to be resolved functions encoded by the gut microbes and the host. (supported by NIEHS R01 ES030940, P30 ES001247, T32 207026. Clinical Trial NCT04060212, prospectively registered 10/1/2019)

Introduction

Mercury is a toxic, non-essential metal common to terrestrial and aquatic ecosystems world-wide (Driscoll et al. 2013). Mercury occurs naturally in elemental (Hg⁰) and inorganic (Hg(II)) forms but has become significantly elevated in many aquatic ecosystems over the last two centuries due to anthropogenic activities, such as combustion of fossil fuels and through mining activities. Hg(II) can be converted to a more highly toxic species, methylmercury (MeHg), via the activity of diverse anaerobic Bacteria and Archaea living in anoxic soils, aquatic sediments, and oxic subsurface ocean waters (Gilmour et al. 2011; Gilmour CC 2013; Podar et al. 2015). As a result, MeHg has become a ubiquitous environmental contaminant and significant concern for human health due to its bioaccumulation in aquatic food chains, in particular in dietary fish and seafood (Oken and Bellinger 2008).

MeHg disrupts neurodevelopment and neuronal function, acting primarily through its high affinity for thiol- and selenium- containing compounds and proteins (NRC 2000), leading to functional disruption (e.g., channel proteins) and altered redox responses by glutathione and other antioxidants (Han et al. 2017; Stringary 2008). Despite well documented toxic effects, disease risk following MeHg exposure remains uncertain due to variable outcomes seen across exposed populations (Davidson et al. 1999; Davidson et al. 2006; Davidson et al. 2008; Debes et al. 2006; Grandjean et al. 1997; Jacobson 2001). Compared to other environmental toxicants, like arsenic, MeHg is eliminated slowly with an average half-life (t_1/2) of ~50 days. However, individual elimination rates vary widely with t_1/2 estimates ranging from <30 to >120 days (reviewed in (Rand and Caito 2019)). Since MeHg t_1/2 is a direct determinant of body burden with regular MeHg exposure (e.g. weekly fish consumption), inter-individual variation in MeHg elimination is a subject of increasing scientific study (Caito et al. 2018; Rand et al. 2016; Yaginuma-Sakurai et al. 2012). The mechanisms controlling MeHg metabolism and elimination in mammals and humans remain incompletely understood. Postprandially, MeHg is readily absorbed into the circulation system. Existing evidence supports a model where the majority of MeHg in the body is then eliminated via the liver and biliary transport to the intestine where two fates are followed: efficient reabsorption of MeHg to the circulation system via the intestine (Norseth 1973) or demethylation to give Hg(II) that is poorly reabsorbed and readily excreted via feces (Ballatori and Clarkson 1983; Ishihara 2000; Norseth and Clarkson 1970). Speciation of Hg from exposed humans typically shows >90% of total mercury (tHg) is excreted in stool and present as Hg(II), implicating that demethylation is required for elimination (Ishihara 2000). This model is consistent with variable proportions of Hg(II) and MeHg observed in human stool samples (30–100% Hg(II)) (Caito et al. 2018; Rand et al. 2016; Yaginuma-Sakurai et al. 2012) and may contribute to the overall inter-individual variability in MeHg elimination(Caito et al. 2018; Rand et al. 2016).

Results, predominantly from rodent exposure models, suggest that the gut microbiome is responsible, at least in part, for demethylation of MeHg (Rowland et al. 1978; Rowland et al. 1984). For example, Rowland et al. (Rowland 1988; Rowland et al. 1977) showed that the rate of MeHg elimination slowed substantially and Hg in target peripheral tissues was retained longer when mice were exposed to MeHg with a preceding co-exposure of antibiotics. The hypothesis that the gut microbiome contributes to MeHg elimination is further supported by a recent study showing that MeHg elimination slowed significantly in two human subjects taking antibiotics (Caito et al. 2018).

A well characterized mechanism for MeHg demethylation in microorganisms centers on mer-encoded functionalities, namely organomercurial lyase (MerB) and mercuric reductase (MerA), which carry out reductive demethylation of MeHg yielding methane and Hg(II) (activity of MerB), the latter of which is reduced to Hg⁰ (activity of MerA) that can volatilize out of the cell and environment (Barkay et al. 2003). Microbial species encoding the mer locus are commonly isolated from samples of fresh or saltwater sediments (Barkay et al. 2003). Evidence for the presence and activity of mer-operon containing bacteria in the human gut remains inconsistent (Liebert et al. 1997; Perez-Valdespino et al. 2013; Rothenberg et al. 2016), with the most recent bioinformatic analyses suggesting mer-encoded functionalities are unlikely to significantly contribute to Hg transformations this environment (Christakis et al. 2021). Alternative mechanisms of MeHg demethylation via oxidative pathways have been attributed to anaerobic bacteria associated with the mammalian gut (reviewed in (Barkay and Gu 2022)). Notably, Kozak and Forsberg (Kozak and Forsberg 1979) showed that the rumen microbiome of cows has robust demethylation activity, while at the same time shows no activity to methylate Hg(II). Several strains of anaerobic bacteria such as Desulfovibrio desulfuricans, which is closely related to strains found in the rumen and human gut microbiome, have been shown to demethylate MeHg (Baldi et al. 1993; Gilmour et al. 2011; Graham et al. 2012; Kozak and Forsberg 1979). However, such strains do encode MerAB (Christakis et al. 2021). Interestingly, Rowland et al. reported that cultures of Streptococci, Staphylococci, Lactobacilli, Bifidobacteria, Bacteroides, and Esherichia coli, many of which were isolated from human stool samples, also demethylated MeHg (Rowland et al. 1975; Rowland et al. 1978). Further, the genomes of representatives of many of these genera also do not encode MerAB (Christakis et al. 2021). While the demethylation activities in the studies above have not been corroborated, they strongly suggest that the gut microbiome plays a role in the demethylation of MeHg in the human gut (Yang et al. 2022).

Here, we report an investigation of the relationship between MeHg elimination, demethylation activity, and microbial taxonomic and functional diversity in the gut microbiome by conducting a human clinical Trial in concert with gnotobiotic mouse experiments and metagenomic sequence analyses. The MeHg kinetics in volunteers following a prescribed tuna fish eating protocol were examined across two Trials whereby a control condition (Trial 1) was compared with the effects of prebiotic administration in a subsequent trial (Trial 2). Kinetics of MeHg demethylation derived from Hg measurements in hair samples were correlated with microbiome demethylation activity in vitro determined using cultured stool samples. In parallel, we compared MeHg t_1/2 of conventional mice (intact microbiome) with that of germ-free counterparts with and without fecal microbiome transplantation (FMT) from human donors. We further tested the influence of antibiotics on MeHg t_1/2 in mice as an alternative means to alter the microbiome. Finally, using shotgun metagenomic sequencing of participant stool samples we identified microbial taxa and their inferred functionalities that correlated with MeHg elimination kinetics. Our results provide evidence that the human gut microbiome plays a significant role in MeHg elimination and demethylation, however, the results also point to a non-conventional microbial pathway (possibly oxidative in nature) that promotes MeHg demethylation and elimination, possibly in combination with host-specific responses.

Materials and Methods

Study Design.

The MerMES study design (Rand et al. 2014) was used here to determine an individual’s MeHg elimination rate that in turn could be correlated with gut microbiome characteristics. The overall study design is summarized in Figure 1. Each participant underwent two Trials, the first without any intervention, and the second with a prebiotic supplementation intended to shift the gut microbiome characteristics. This latter intervention strategy was an attempt to determine the effects of shifting microbiome characteristics on MeHg elimination rate. Volunteers were recruited from the University of Rochester Medical Center community and were enrolled as participants if they were between 18 and 80 years of age and in good general health on a self-reported basis. Volunteers were excluded if they had known allergies to fish, were pregnant or lactating, used hair dyes or treatments, had known gastrointestinal or renal disorders, or had used systemic antibiotics within the last two months. Participant characteristics including sex, height, weight, ethnicity and prior fish-eating habits were also collected upon consenting and enrolling. Study procedures were reviewed and approved by the University of Rochester Research Subjects Review Board and consent was obtained from each subject.

Fig 1. MerMES clinical study design.

A. Scheduled fish meal consumption 7 days apart and schematic of sampling and analysis of hair and stool. B. The timeline of Trial 1 and 2 fish meal ingestion, elimination period, sampling and prebiotic intervention (Trial 2).

Fish meal consumption.

For each trial, participants consumed three fish meals at 7-day intervals. Fish meals consisted of wild caught yellowfin tuna steaks, ranging from 2 to 8 ounces, that were prepared by and purchased from a local supermarket. Steaks were cut from fish of similar age and size, weighed, individually wrapped and frozen by the market staff. In addition, a small sample (approximately 0.5 oz) from each of three body regions (nape, body, and tail) of each fish was obtained for Hg analysis. Mercury levels in the fish were determined using a Perkin Elmer NexION 2000 ICP-mass spectrometer (see below). Mercury levels in each of the five fish used across the study varied widely, ranging from 0.19 to 2.0 ppm, nonetheless they showed consistent levels (<20 % variation) across the body three body regions (nape, body, and tail) of each fish. All Hg values were within the range of those reported previously for yellowfin tuna by the FDA (FDA 1991–2010). This analysis was completed prior to preparation and distribution of the fish steaks and was used to calculate appropriate portion sizes in order to: 1) keep peak MeHg blood levels below the EPA Reference Dose of 5.8 μg/L and 2) to achieve a similar proportion of MeHg dose relative to body weight across all of the participants. Subjects were provided with three portions of frozen fish steaks in a cooler for transport home. Preparation of fish for eating was at the discretion of each subject, since previous studies showed that cooking methods have no appreciable effect on MeHg content (Sherlock et al. 1984). Using a calculation of retention after intermittent dose (Aaberg et al. 1969) and individual body weights of participants, portion sizes were estimated for each subject such that consumption of three fish meals over 14 days would not exceed the EPA reference dose. After the third fish meal, participants were instructed to refrain from fish or seafood consumption for a period of 60 days.

Stool samples were collected at approximately 14 days after the last fish meal and hair samples were collected upon completion of the trial (60 days post last fish meal). Mercury analyses were limited to hair and feces since only a small amount of MeHg is eliminated via other routes (e.g., <5% via urine in first 70 days (Smith et al. 1994)). Hair samples were analyzed for longitudinal Hg levels to determine elimination rates (see below) and stool samples were used as a source of gut microbes for culture and for DNA isolation. For Trial 2, subjects were asked to take a daily supplement of a commercially available prebiotic (PreBiotin^™, Jackson GI Medical) seven days prior to their first fish meal. Prebiotin^™ consists of an oligosaccharide enriched inulin. Individually packaged portions (4 g) of the Prebiotin^™ Prebiotic Fiber Stick Pac were provided to participants. This study targeted a dose of 8 g/day based on published effects of inulin on gut microbiota in the range of 4 to 16g/day dosages (Holscher et al. 2015; Salazar et al. 2015). Participants were instructed to consume the 8 g of prebiotic a day, as one dose or two 4g doses at different times.

Hair and stool sampling.

Hair samples were taken at approximately 60 days after the final fish meal essentially as previously described (Caito et al. 2018). Briefly, hairs were manually extracted using a brisk pull of a tuft of 4–8 strands from the crown region of the scalp using a gloved hand resulting in abstraction of follicle/root remnants on the hair shaft. No additional steps were taken to cleanse the hair prior to analysis, as previous studies show that washing and chelating steps are ineffective for removal of potentially contaminating hair surface Hg from exogenous exposures (Ali Aldroobi et al. 2013; Li et al. 2008). Single hair strands were aligned linearly and adhered to double-sided clear tape (Scotch brand) previously adhered to a clean glass microscope slide. Intervals of 1 cm were marked on the tape adjacent to the hair strand as reference points to orient the hair for laser ablation.

Stool samples were collected at home by the participant using a Fisherbrand Commode Specimen Collection System (Fisher Scientific, Cat. #02-544-208) approximately 30 days after consuming their last fish meal. Upon collection, the participant immediately transferred a small portion of the stool directly into a DNA Genotek OMNIgene•GUT tube following the manufacturer’s instructions (DNAgenotek Ottawa, CA). This sample was utilized for isolation of DNA for metagenomic analyses. A second large portion of the stool sample was placed into a prelabeled 50ml conical tube and capped. These two samples were returned to the lab and subsequently frozen at −20°C.

Longitudinal analysis of Hg in hair.

Spatially-resolved analysis of Hg in hair was performed with laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at the Dartmouth Trace Element Analysis Core Facility essentially as described previously (Rand et al. 2016). Laser ablation was carried out with a NWR213 laser ablation unit (ESL, Bozeman, Montana) equipped with a programmable X-Y stage for setting coordinates for ablation of a series of spots along the hair shaft. Hg levels spanning the 60-day elimination period, the 14-day fish meal period and a period prior to the fish consumption were captured by LA-ICP-MS of 90 consecutive ablation spots at 333 μm intervals (equivalent to ~ 1-day growth month (Harkey 1993; Myers and Hamilton 1951)) over a 3.0 cm segment of the hair closest to the root. Elemental detection was performed with an Agilent 7900 ICP-MS (Santa Clara, California) monitoring ²⁰²Hg and ³⁴S isotopes in no-gas mode. The ³⁴S isotope was used as an internal standard for the relative mass of hair captured with each ablation plume as previously described (Legrand et al. 2007). Analysis parameters were: 55 μm laser spot size with 3 second dwell time, 10 Hz laser frequency and 40% laser power, equating to a fluence of 4.00 J/cm². With this method we typically achieve detection limits for Hg of 10–50 ng/g depending on laser spot size and instrument sensitivity. Data were collected as time resolved intensity of ²⁰²Hg and ³⁴S. Hg values were expressed as a ratio of the peak areas of the transient signals for ²⁰²Hg and ³⁴S multiplied by a factor of 1,000 since the signal intensity for ²⁰²Hg was approximately 3 orders of magnitude lower than ³⁴S, consistent with the low abundance of Hg relative to S in hair. As we previously reported (Rand 2014), absolute Hg quantification using this method was not necessary since the goal of this study was to determine a kinetic parameter of Hg elimination, which can be derived from the change in relative concentration of Hg over time.

MeHg elimination rate determination.

Elimination rates were determined essentially as described previously (Caito et al. 2018). Briefly, data were plotted in terms of Hg:S versus time (days). Growth rate of each individual hair was calculated by determining the distance between the Hg:S spikes that correspond to the first and last fish meal and then dividing by the number of days between these meals. Elimination data spanned the interval beginning 2 spots (approximately 48 h) after the last fish meal to the margin of root material at the follicle end of the hair. Elimination rates were calculated based on a one-compartment model with the assumption of a first-order process according to:

$${\text{C}}_{\text{t}}={\text{C}}_0\cdot {\text{e}}^{-\text{kel}\cdot \text{t}}$$

where C_t is the concentration of Hg (i.e., Hg:S) at time t and C₀ is the concentration of Hg at t = 0. The elimination rate constant (k_el [days⁻¹]) was determined from the slope of a linear fit to a plot of ln(Hg:S) versus time created using Prism (Graphpad Software, Inc. La Jolla, California). As previously reported (Rand et al. 2016), we assumed a baseline of zero Hg:S for rate calculations. Elimination rates were used to determine a half-life (t_1/2 = 0.693/k_el) in units of days.

MeHg biotransformation in fecal cultures.

Using an anaerobic glove box, a subsample of feces weighing 0.03 g was inoculated into 3 ml of Brain Heart Infusion (BHI) liquid media (Anaerobe Systems Morgan Hill, CA) containing 1 ppm MeHg in 15 ml polypropylene culture tubes. Each experimental culture was set up in duplicate, with one of the two culture tubes undergoing heat inactivation and serving as a control. BHI media was chosen as it is a non-selective medium suitable for the general growth of anaerobic microorganisms and other fastidious microorganisms. Immediately after inoculation cultures were gently shaken to break up particulate feces and 300 μl was collected for measurement of MeHg concentration at t=0. Culture tubes were capped tightly and removed from the anaerobic chamber. Heat inactivated controls were placed into a water bath at 100°C for 1 h. Cultures were then placed in a shaking incubator set at 37°C with 200 rpm. Subsamples of 300 μl were then collected under anaerobic conditions at 48 and 96 h. The amount of total Hg remaining at each time point was then assessed by thermal decomposition and atomic absorption using a DMA-80 Direct Mercury Analyzer (Milestone SRL). Values were determined on a wet weight basis and expressed in ppm (μg/g). The reduced level of tHg occurring with time was interpreted to reflect overall biotransformation including the rate limiting demethylation step followed by reduction resulting in evaporative loss of Hg from the reaction volume.

Mouse husbandry and experimentation.

All mouse experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at Montana State University (MSU). Mouse experiments were carried out in an Association for Assessment and Accreditation of Laboratory Animal Care (AALAC)-accredited Animal Resource Center (ARC) at MSU. Germ-free, gnotobiotic, and humanized mice were maintained on Lab Diet’s specialty formulated, autoclavable animal diet 5013. Conventional mice were maintained on Lab Diet 5053 (identical percentage of calories from fat, protein, and carbohydrate) that was not autoclaved. Mice had free access to food and reverse osmosis sterilized water.

Conventional mice handling, MeHg elimination, stool collection and antibiotic treatment.

C57Bl/6J conventional mice were ordered from Jackson labs for all conventional mouse experiments. Mice were acclimatized for at least 5 days prior to the start of experiments. All murine experiments are summarized in Figure S1. Methylmercury chloride (MeHg, Sigma-Aldrich) was prepared by dissolving solid MeHg in DMSO and diluting with PBS followed by sterile filtration with DMSO compatible syringe filters (nylon, 0.4 μM). Conventional mice were exposed to a single dose of 5 mg per kg of body weight (5mg/kg) MeHg by oral gavage. Fecal pellets were collected on days 1,7,14, and 21 post-gavage. Blood was collected on days 1,7, and 14 by submandibular bleed. Mice were humanely euthanized on day 21 and a terminal bleed was performed from the portal vein. Antibiotic treated conventional mice were given filter sterilized water containing 2 mg/mL each of streptomycin sulfate, neomycin sulfate and bacitracin at least 5 days before mercury gavage and throughout the remainder of the experiment as previously described (Rowland 1988). Antibiotic water was made fresh approximately once a week and sterile filtered.

Germ-free mice handling, MeHg elimination, stool collection and antibiotic treatment.

C57Bl/6J germ-free mice were transferred from a breeding isolator to a sterile experimental isolator (Park Bio Inc.) before the start of each experiment and checked for sterility. Sterility methods included PCR amplification of 16S rRNA gene subunits, Gram stains, and cultivation on agar plates or liquid culture in YCFAC media both aerobically and anaerobically. All materials were autoclaved and alcide aerosol was used to introduce all in-going items into the sterile chamber through an outer airlock chamber. Isolators were kept at positive pressure to reduce risk of outside contamination. One fecal pellet from each cage was collected before and after each experiment and extracted DNA was amplified by PCR of the 16S rRNA v3-v4 variable region to confirm sterility. After confirming absence of amplifiable 16S rRNA genes, GF mice were orally gavaged with 5 mg/kg MeHg. Blood and fecal pellets were collected on days 1, 7,14, and 21 post-gavage. An additional treatment included antibiotic treatment of GF mice using the same regimen as for antibiotic treated conventional mice (above) starting with triple-antibiotic water one week prior to the MeHg dosage by oral gavage.

Gnotobiotic and Humanized mice.

C57Bl/6J germ-free mice were transferred into a sterilized experimental isolator and confirmed to be free of PCR-amplifiable 16S rRNA genes, per the assay described above. A culture of A. onderdonkii (German Collection of Microorganisms and Cell Cultures, DSM 19147) was prepared by inoculating Yeast Casitone Fatty Acids (YCFA) pre-reduced broth in an anaerobic chamber (COY Systems) and cultured statically for 72 hours at 37°C in the dark. An aliquot of 100 μL of culture was used to colonize germ-free mice via oral gavage. Following gavage, mice were acclimatized for one week and A. onderdonkii colonization was confirmed using a species-specific PCR from fecal pellet DNA as described above. The presence of A. onderdonkii in the murine gut was validated by PCR amplification of murine fecal pellet DNA using primers specific for the transcriptional regulator encoding gene, araC, in A. onderdonkii (AP019734). Primers were designed using the NCBI primer designer targeting unique regions. The forward and reverse primers were: 5’-TCAGCATCGTACAGGCCATC-3’ and 5’-ACGACTTCACGGAACCCTTC-3’, respectively. Specificity of the primers for araC, in A. onderdonkii was confirmed. Mice were then dosed with MeHg 5 mg/kg body weight as described for conventional mice. Colonization was confirmed again at the end of the experiment (day 21 post-gavage).

To humanize GF mice, a slurry was prepared under anaerobic conditions using frozen fecal samples from two participants from Trial 1 that demonstrated different elimination rates (S27 and S36, Table S1). Approximately 90 mg of stool was transferred with sterile toothpicks into separate culture tubes and homogenized in 5 mL of sterile water. C57Bl/6J germ-free mice were transferred from breeding isolators in HEPA-filter-topped cages to a biosafety cabinet where they were given slurry from the two human donors by oral gavage. Following one week of acclimatization in the biosafety cabinet, mice were transferred to a HEPA-filtered ventilation rack and dosed with 5 mg/kg MeHg as described for conventional mice. Blood and fecal pellets were collected as described above on days 1, 7, 14, and 21 post-mercury exposure.

MeHg Elimination Rate Determinations in Mice.

To assess MeHg elimination rates for each mouse, the total mercury (tHg) in mouse blood at each of the collected timepoints was determined. Collected blood samples were subject to acid digestion and tHg quantification using ICP-MS methods. Briefly, 75–200 μL of blood were acid hydrolyzed with 500 μL of concentrated nitric acid (69%) with heating at 95°C for 1h. Samples were cooled at room temperature before diluting to 10 mL with de-ionized water. This diluted sample was then analyzed by ICP-MS (Perkin Elmer NexION 2000) relative to standardized solutions of Hg ranging from 3.90 to 500 ng/mL MeHg. Instrument settings were 1600W power in standard mode for measuring the Hg²⁰² isotope. The limit of detection (LOD) for Hg in LC ICP-MS experiments was 0.1 ng/ml (ppb) and the limit of quantification (LOQ) is 10x the LOD, 1 ng/ml (ppb).

Sex as a biological variable was beyond the scope of this study and we therefore collapsed sexes to achieve greater statistical significance in blood Hg kinetic data and stool Hg(II) analyses. Elimination rates were calculated based on a one-compartment model with the assumption of a first-order process as indicated above for human elimination rates. Half-life values for all mice were log transformed to test for normality and equality of variance for one-way ANOVA statistics. Multiple comparisons tests of tHg t_1/2 in mouse cohorts were performed using Tukey-Kramer’s HSD. Significance threshold was set to p <0.05.

MeHg Demethylation Determinations in Mice.

Demethylation was determined via speciation of tHg in stool samples collected at day 7 by HPLC-ICP-MS methods as described in Narukawa et al. (Narukawa et al. 2018). Briefly, a single fecal pellet from each mouse was weighed and dispersed in 300 μL of a digestion buffer solution containing 3.25 mM glutathione (GSH), 32.3 mM HCl, and 20 mM KCl that was heated at 65°C for 10 mins to soften the pellet. Pellets were then homogenized using an electric hand-held microfuge tube mortar and shaken to equilibrate overnight at room temperature. Digested samples were subsequently filtered through a 0.45 μm cellulose acetate filter (Corning, Salt Lake City Utah) and 50 μl of filtrate was subject to analysis using a Perkin Elmer NexION 2000 ICP-MS with prior separation using a 150 mm Agilent Eclipse XBD C18 HPLC column using a pore size of 5 μm. The mobile phase used for all experiments was 0.1% cysteine and 0.1% HCl run isocratically with a flow rate of 1 mL/min. Instrument calibrations were performed prior to analysis using standards of both MeHg and Hg(II) at the concentration ranging from 3.9 to 500 ng/mL. Peak volumes of MeHg and Hg(II) generated from all samples were resolved with the Empower software (Perkin Elmer) and compared against those of the calibration curve to determine Hg concentration. ICP-MS Instrument conditions for all runs were as stated above for elimination rate analyses. One-Way ANOVA statistics were performed on %Hg(II) for day 7 fecal pellets (n=5–7) from each MeHg exposed mouse cohort in GraphPad Prism 9 (v. 9.4.1). Multiple comparisons tests were run using Tukey-Kramer’s HSD correction. Significance threshold was set to p<0.05.

Stool DNA extraction and sequencing.

DNA was extracted from duplicate stool sample aliquots from each participant using the Qiagen PowerSoil Pro Kit (Qiagen) following the manufacturer’s instructions. Equal volumes of duplicate extracts were pooled and quantified fluorometrically via the high sensitivity Qubit assay (Thermo Fisher Scientific). Each pooled sample was subjected to paired-end shotgun metagenomic sequencing on the Illumina NovaSeq6000 platform; 2 × 150 bp paired-end reads, at the University of Wisconsin Biotechnology Center and with the Illumina regular fragment (300 bp) method of library preparation. Reads were quality-filtered, trimmed, and cleaved of Illumina-specific sequencing adapters using TrimGalore v0.6.5 (Kreuger 2023).

Metagenomic Assembly and Binning.

Metagenomic analysis followed pipelines that we have previously described (Colman et al. 2022). Trimmed sequence reads were down-sampled and normalized using the bbnorm module of bbmap (version 38.96), specifying a target read depth of 100 and a minimum read depth of 5 to reduce read redundancy. The down-sampled reads were then assembled with metaSPAdes (version 3.15.4) specifying default parameters, with the exception of specifying the -meta option that optimizes the assembly for metagenomic sequence data (Bankevich et al. 2012). Assembled contigs were indexed and subjected to read mapping using bowtie2 (version 2.4.5) (Langmead and Salzberg 2012) to determine contig read depth. Binning of contigs into metagenome assembled genomes (MAGs), was conducted using three binning algorithms, including metaBAT v.2, MaxBin v.2, and CONCOCT v.1.1.0, followed by refinement of bin sets based on estimated contamination and completeness (via CheckM v.1.1.3 estimates), using MetaWRAP (version 1.3) (Uritskiy et al. 2018). MAG abundance information was also calculated using MetaWRAP. Protein identification and annotation was performed of coding sequences was conducted using PROKKA (version 1.14.5) (Seemann 2014), with default parameters. The MAGs were taxonomically classified using the Genome Taxonomy Database Toolkit (GTDB-tk). GTDBTK generates taxonomic placements by aligning MAGs against the bac120 and ar122 bacterial and archaeal genome databases, respectively, and verifies this placement by generating pairwise average amino acid identity (AAI) and average nucleotide identity (ANI) calculations between the MAG and the database genome.

Identification of MerAB in gut metagenomes.

A BLASTp-based approach was used to determine if canonical mercury detoxification mechanisms, i.e., MerAB, were encoded in gut metagenomes (Camacho et al. 2009). Both binned and unbinned metagenome contigs were subjected to BLASTp using MerA from the bacterial transposon Tn501 (Stanisich et al. 1977) and the archaeal taxon Saccharolobus solfataricus P2 (Schelert et al. 2004) as queries. Homologs of MerB were identified by using the homologs from Pseudodesulfovibrio mercurii (WP_202945505.1), and Bacillus cereus (WP_167316880.1) as queries. An e-value cut-off of 10e⁻³⁰, a query coverage cut-off of 50%, and a percent identity cut-off of 50% were specified.

Community taxonomic analysis and relationships to Hg demethylation rate.

Overview.

Three independent approaches were used to process and organize community metagenomic data prior to performing statistical analyses aimed at identifying potential relationships between community members and MeHg elimination kinetics. Briefly, the first approach was assembly- and bin-independent. Here, metagenomic reads were used to generate abundance weighted taxonomic information for each community metagenome, allowing for relationships with MeHg elimination kinetics to be statistically examined. In the second approach, metagenomic sequence reads were subjected to assembly, binning, and read mapping to generate abundance-weighted metagenome assembled genomes (MAGs) that could be correlated with MeHg elimination kinetic data. These MAGs were taxonomically classified, allowing for MeHg elimination kinetics to be related to individual populations in the sample. In the third approach, proteins encoded on assembled contigs with lengths of >1000 bp were subjected to cluster analysis and statistical analysis, and for a subset of clusters that significantly correlated with MeHg elimination rate, taxonomic evaluation. This allowed for comparison to the outputs of the first two approaches. Details of the methods used in each of these approaches are reported below.

Approach 1: Assembly- and bin-independent determination of community taxonomic composition.

Kraken v2 (Wood and Salzberg 2014) was used to evaluate the taxonomic composition of quality-filtered and trimmed metagenome reads. Kraken generates k-mers from short reads and aligns them to the NCBI genomic reference database to infer taxonomic composition whilst avoiding artifacts and biases that may be introduced during the binning process. The output from Kraken was then compiled at the highest taxonomic rank possible and this was used to generate a table reflecting the relative abundance of taxa among subject stool communities for downstream statistical analyses.

Two statistical tests were performed to evaluate the correlation between the abundance of taxa in stool communities and participant MeHg elimination rate. Pearson regressions were used to compare the abundance of taxa in a stool community to participant MeHg elimination rate. In a second analysis, participants were assigned to a quartile based on elimination rate. Specifically, these four quartiles were assigned based on the number of standard deviations away the mean elimination rate and included i) more than one standard deviation below the mean elimination rate, ii) between the average and one standard deviation below the mean elimination rate, iii) above the average but below one standard deviation above the mean elimination rate, and iv) above one standard deviation above the mean elimination rate. ANOVAs were performed to determine which taxa had statistically significant differences in abundance between quartiles. Both statistical tests were performed using the R base package (version 3.6.1).

Approach 2: Community metagenome assembly, binning, and taxonomic analysis.

MAGs from each stool community metagenome were organized into operational taxonomic units (OTUs) to facilitate crosswise comparison of taxonomy between metagenomes and to permit downstream statistical analyses. To achieve this, fastANI (version 1.3.2) was first used to calculate the pairwise ANI for each MAG pair (Jain et al. 2018). MAGs were then clustered into 589 OTUs using mothur (version 1.42.1) specifying a >95% pairwise ANI threshold (Schloss et al. 2009), since 95% ANI consistently corresponds to species level designations (Chan et al. 2012). MAG abundances in each community, as determined above, were then used to create an abundance weighted table permitting comparisons of MAG distributions across metagenomes. Pearson correlation coefficients and ANOVAs between the relative abundances of MAGs and participant MeHg elimination rate were calculated using the R base package as described in Approach 1.

Approach 3: Community metagenome protein clustering.

Assembled (but not binned) metagenome sequence data (i.e., contigs) from approach 1 were subjected to analysis via PROKKA v1.14.5 to identify open reading frames, call protein sequences, and provide first order protein annotations (Seemann 2014). Protein sequences were then subjected to hierarchical clustering using CD-hit (Fu et al. 2012). Protein sequences were sequentially clustered at 90 percent identity, then 60 percent identity, and then 30 percent identity to minimize redundant clustering of highly similar sequences. As such, final clusters represent collections of proteins that share between 30 and 100% sequence identity. Protein clusters were then processed with the R base package to determine the inferred abundance of each protein cluster in each metagenome based on the abundance of the contig that the protein was encoded, as calculated above.

Initially, Pearson correlations were conducted between the abundance of each protein cluster (i.e., the number of proteins in a metagenome belonging to each protein cluster) and participant MeHg elimination rate. Once protein clusters that were significantly correlated (p < 0.05) with participant MeHg elimination rate were identified, they were used to identify the proportion of proteins encoded in each MAG that belonged to statistically significant clusters. The overall correlation coefficient (Pearson R) of protein clusters in each MAG that significantly correlated with elimination rate (p < 0.05) was then averaged; proteins from clusters that were not significantly correlated with elimination rate (p > 0.05) were assigned a value of zero. This average value was then abundance weighted based on abundance of each MAG within each community. This value was then divided by the number of encoded proteins in a MAG. This value was then subsequently averaged for the entire OTU. This yielded a value indicating the average proportion of significant proteins per OTU weighted by protein correlation and normalized to MAG size. R base package was used to perform these calculations.

Phylogenetic diversity analyses.

The phylogenetic diversity of MAGs within each community (generated above in approach 2) was calculated and used to determine if diversity correlated with participant MeHg elimination rate. Housekeeping genes encoded in each MAG were called using GTDBTK (Chaumeil et al. 2019), aligned using ClustalOmega (version 1.2.4) (Sievers and Higgins 2021), and alignments were concatenated. IQ-tree (version 1.6.12), specifying the LG substitution model and 1000 bootstrap replicates, was used to reconstruct the phylogeny of MAGs (Hoang et al. 2018). The topology of the phylogeny was compared to phylogenies for each metagenome generated as a component of the GTDBTK taxonomic assignment pipeline (described above). Faith’s Phylogenetic diversity for MAGs comprising each metagenome was then calculated in R using the package picante version 1.8.2 (Kembel et al. 2010). To minimize bias from rare community members, the tips of the tree were abundance weighted, using abundances inferred using methods described above in approach 2. A standardized effect size of mean pairwise distance test was performed to evaluate the degree of phylogenetic clustering in the communities (Faith 1992). Tips were shuffled for the null model, and 1000 replicates were performed. The Pearson correlation between participant MeHg elimination and Faith’s PD was evaluated with the R base package.

False discovery rate correction was performed on the pearson correlations to determine the likelihood of false positive OTUs. The R stats package (version 4.2.2) (team 2020) was used to perform false discovery rate correction specifying the Benjamini and Hochberg model (BH) which is typically preferred when there are a large number of observations (Benjamini and Hochberg 1995). Error correction was performed using only OTUs present in at least half of the metagenomes to parse down the number of observations to include those likely to impact the MeHg elimination rate.

Principle Coordinate Ordination.

A principal coordinate ordination (PCO) was constructed to evaluate if the overall structure of participant stool communities is related to participant MeHg elimination rate. First, MAGs were assigned to OTUs at 95% ANI, as described above. This served to facilitate comparison of community structure across metagenomes. MAG abundances in each community were used to create an abundance weighted table of OTUs for each metagenome. Bray Curtis dissimilarity was then calculated for these communities using the R base package. The resultant Bray Curtis dissimilarity matrix was then plotted as a PCO to visually evaluate patterns in community composition. To facilitate this comparison, communities were assigned to quartiles based on elimination rate, as described above. The communities were then overlaid by color based on their quartile assignment. A mean pairwise distance was calculated (i.e., the average distance between a participant on the pcoa and all the other participants) using that PCO, and then evaluated for the correlation between MPD (mean pairwise distance) and participant MeHg elimination rate. Additionally, MPD was calculated for each sample using the same Bray Curtis dissimilarity equation, and then the correlation between MPD and participant MeHg elimination rate was calculated.

OTU network analysis.

Network analysis was performed to identify co-occurring OTUs in participant stool communities. As with prior analyses, an abundance weighted table of OTUs for each community was created and used to define Pearson correlations for each OTU pair using the R base package. The resultant list of correlations (both positive and negative) was used as input to Cytoscape (version 3.9.1) for network construction (Shannon et al. 2013).

The Cytoscape generated network was used to visually inspect patterns in community composition and OTU co-occurrences, and to determine if these connectivity patterns qualitatively matched with MeHg elimination patterns. To achieve this, the network was first trimmed to only contain OTU’s present in at least ten communities. This reduced the number of nodes to include only the top 10% most widespread OTUs. Secondly, the size of each OTU node was scaled according to the Pearson correlation coefficient (R) relating its abundance and the participant MeHg elimination rate, where a larger R corresponded to a larger node. Negatively corelated OTU’s had their corresponding node colored red to help distinguish between positively and negatively correlated OTUs. The edges connecting the nodes, which correspond to the degree of co-occurrence between those OTUs, were colored green if they corresponded to a positive correlation and were colored red if they corresponded to a negative correlation. Finally, the node representing OTU 15, A. onderdonkii, received a green outline to help visualize connections between A. onderdonkii and other community members.

Results

Study Population.

Thirty-six volunteers from the Rochester, NY metropolitan area were recruited for the study. Six of the subjects dropped out prior to Trial 1. One subject completed the fish consumption portion and donation of stool in Trial 1 but did not return to donate hair and we were therefore unable to determine their elimination rate for MeHg. Overall, 29 subjects were retained and completed all portions of Trial 1 in this study (15 males and 14 females, Table S1). Participant ages ranged from 21–64 years with 21 of the 29 subjects between the ages of 23–30. The cohort consisted of 15 Caucasian females, 18 Caucasian males, one Asian male, one female with more than one race, and one female with an unreported race. All subjects were asked to resume and maintain their typical diets and refrain from any fish or seafood consumption during the 60-day elimination period. Two subjects, S9 and S18, reported illnesses requiring the use of antibiotics during the elimination phase of the study. However, stool sampling occurred prior to the antibiotic use and each of the subject’s hair analyses permitted elimination rate extrapolation prior to the antibiotic administration. As such, a complete dataset was obtained for these two participants. Another two participants, S1 and S7, reported discomfort due to the prebiotic and discontinued the prebiotic within the first week of Trial 2, but were nonetheless included in the Trial 2 dataset.

MeHg elimination kinetics and gut microbial demethylation activity in humans.

We established a hair and gut microbiome sample repository and a dataset to assess the relationship between MeHg elimination and demethylation from the recruited a cohort of volunteers. Among the 29 enrolled participants we were able to generate hair-derived elimination rates and t_1/2 of MeHg for 28 individuals in Trial 1 and 25 individuals in Trial 2, (Table S1A). Examples of the longitudinal profile of the Hg levels (Hg:S) in the hair that grew over the course of the trial period is shown in Figure 2 for two individuals, S9 and S13, in Trial 1. Spikes in Hg:S were seen that correlate with the day that each fish meal was consumed. The decline in Hg:S following the final Hg:S spike was used to extrapolate the elimination rate, determined from the slope of the ln(Hg:S) v. time plot (Fig. 2A, B inset). Elimination rates in both Trials showed a similar range: Trial 1 ranged from k_el = 0.023 days⁻¹ to 0.008 days⁻¹ (t_1/2 = 30 to 89 days, see Table S1A) and Trial 2 ranged from k_el = 0.028 days⁻¹ to 0.008 days⁻¹ (t_1/2 = 28 to 90 days, see Table S1A). These results suggest that a 3-fold difference in elimination rate is common among even a small cohort of individuals.

Fig 2. In vivo MeHg elimination rates and in vitro demethylation.

A,B. Longitudinal quantification of total Hg (Hg:S) in hair of subjects S13 and S9 over the course of the Trial 1. C. Loss of total mercury due to demethylation, reduction and volatilization in human stool cultures from subjects S9 and S13. D. Correlation and Pearson regression statistics of elimination rate (half-life) and percent Hg remaining in in vitro cultures after 96 hours

Stool samples collected from subjects in Trial 1were cultured anaerobically in vitro for 96 hours in rich media (brain heart infusion broth). Total Hg was quantified in cultures at 48 and 96 hours as a proxy for MeHg demethylation. A significant loss of total Hg (tHg) over 96 hours in fecal cultures was observed in all cultures compared to heat-inactivated controls, indicating this activity required viable, metabolically active bacteria. To test whether in vitro demethylation corresponded to in vivo elimination, we compared t_1/2 values to the tHg remaining in in vitro cultures which showed a significant correlation with Pearson regression statistics (Fig. 2D).

Effect of prebiotic on MeHg elimination in humans.

In Trial 2, subjects ingested a prebiotic 7 days prior to eating tuna and throughout the Trial. The profile of taxa abundance at the Phylum level obtained from metagenomic sequencing (see below) confirmed that each participant’s microbiome underwent a significant change, as seen for 21 of the individuals that had both a complete elimination dataset (see Table S1A) and metagenomic sequence data meeting cut-off criteria (see below) for each Trial (Fig 3). The prebiotic intervention showed no significant shift in the average elimination rate across the cohort between Trials (Trial 1 average k_el = 0.0155 days⁻¹ (t_1/2 = 48.5 days), Trial 2 average of k_el = 0.0152 days⁻¹ (t_1/2 = 49.1days, see Table S1A). Furthermore, the outcomes were mixed, whereby among the 21 participants, seven showed a faster elimination (shorter t_1/2, green connectors) and eight showed slower elimination (longer t_1/2, red connectors) in Trial 2 compared to Trial 1. Six subjects showed an unchanged elimination rate with prebiotic administration (Fig 3 and Table S1A).

Fig 3. Changes in human MeHg elimination rate with prebiotic exposure.

MeHg elimination rate (t_1/2) of 21 participants following the MerMES protocol (Fig 1) are shown superimposed on the percent composition of indicated taxa (phyla). “A” designates Trial 1 and “B” designates Trial 2 results for each participant. Community composition is denoted at the phyla level, with the exception of Firmicutes, which is subdivided into four monophyletic clades, following the organization of the genome taxonomy database (Version 07-RS207).

Relationship of elimination rate with participant characteristics and fish diet.

Despite the apparent lack of predictable change in MeHg elimination due to prebiotic treatment, elimination rates within each Trial and across both Trials could be used to assess correlations with subject attributes. There were no significant correlations found between elimination rate and sex, age, height, weight, body mass index (BMI) or BMI categories (normal, overweight, obese) (Data not shown). In addition, there was no significant difference in elimination rate between subjects who previously reported fish eating of any kind in their diet and those who did not. Curiously, in Trial 1, participants who reported routine eating of tuna of any type (canned or fresh) showed a significantly faster MeHg elimination rate compared to those who did not eat tuna (tuna eaters (n= 12) t_1/2 = 41.3 +/− 8.8 days, non-tuna eaters (n=16) t_1/2 = 53.5+/−16.1 days, p=0.0254, t-test). Interestingly, this relationship of tuna eating with faster elimination was lost in Trial 2, where prebiotic administration had perturbed gut microbiome composition (Fig. 3). These data suggest dietary differences can influence gut MeHg demethylating activity and elimination.

MeHg elimination kinetics and gut microbial demethylation in conventional, germ-free and antibiotic treated mice.

We next compared Hg elimination under controlled gut microbiome conditions. Conventional (CV) mice had significantly faster t_1/2 compared to both antibiotic- (Abx) treated (4.5 days vs. 12.7 days) and germ-free (GF) mice (7.1 days; Fig. 4, Table S2A). Interestingly, the Abx-treated mice eliminated significantly slower than the GF mice, suggesting a host response to Abx may influence elimination to a greater extent than the GF condition. Speciation of tHg in day 7 stool pellets of CV mice showed nearly 83% of tHg as Hg(II) (Fig 5, Table S3A). By comparison, 17% and 20% of tHg was observed as Hg(II) in stool of GF and Abx groups, respectively (Fig. 5, Table S3A). These data, collectively demonstrate the critical role of the microbiome in Hg elimination from the body.

Fig 4. MeHg elimination in mice with various manipulations of the gut microbiome.

Elimination t_1/2 (log transformed) of MeHg in conventional (CV), Antibiotic-treated Conventional (Abx), Germ-Free (GF), Alistipes onderdonkii gnotobiotic mice (A.o.) and mice humanized with two different human gut microbiomes (S27, S36). Bars indicate mean and std. dev. (absolute t_1/2 value in days above the data points). Multiple comparison testing showing significant differences between CV and all other groups are shown, * = p<0.05, ** = p<0.005. Additional comparisons are in Table S2B.

Fig 5. Total %Hg as Hg(II) in stools of mice with various manipulations of the gut microbiome.

Total Hg as Hg(II) in stools, representing demethylated MeHg, in conventional (CV), Antibiotic-treated conventional (Abx), Germ-Free (GF), Alistipes onderdonkii gnotobiotic mice and mice humanized with two different human gut microbiomes (S27, S36). Bars indicate mean and std. dev. Multiple comparison testing showing significant differences between CV and all other groups are shown, * = p<0.05, ** = p<0.005. Additional comparisons are in Table S3B.

The reduced MeHg elimination of the GF condition in mice afforded the opportunity to assess the human microbiome directly for its activity in enhancing MeHg elimination and its association with demethylation. GF mice were split into two groups and colonized with donor fecal samples from two human fecal slurries from participant S27 and S36 from Trial 1 (see Table S1A). Subsequent to a bolus oral dose of MeHg the humanized mice showed a t_1/2 of 4.2 days and 3.7 days for S27 and S36, respectively, which was not significantly different from each other or from the CV mice (Figure 4, Table S2A, B). In addition, we observed 54% and 64% of tHg as Hg(II) in pellets of the S27 and S36 humanized mice, respectively (Fig 5, Table S3A). Together, these findings demonstrate that a restoration of fast elimination in GF mice by the human gut microbiome parallels a greater demethylation of MeHg. Yet, elimination rates equivalent to that of CV mice were achieved despite the fact that %tHg as Hg(II) in pellets did not approach that of CV mice. Furthermore, participants S27 and S36 showed a more than 2-fold difference in MeHg elimination rate in their respective human conditions (Table S1A), which was not reflected in the respective humanized mice.

Stool microbiome metagenomic sequencing.

Using stool sample DNA and paired-end shotgun sequencing we assembled metagenomes from sequence reads of 27 of the 29 participants in Trial 1 (Table S4, S5). An average of 4.13 × 10⁸ base pairs were sequenced per community metagenome, with no community having less than 2.19 × 10⁸ base pairs of total sequence. Following quality filtering, assembly, and binning, a total of 2219 metagenome assembled genomes (MAGs) were identified across the 27 participant stool metagenomes, 1399 of which had estimated completeness of >90% (Table S5). As in previous analyses of the human stool samples (Falony et al. 2016; Zhernakova et al. 2016), communities tended to be dominated by Bacteroidetes and Firmicutes at the phylum level, while the most abundant OTUs were affiliated with Prevotella copri (average abundance of 3.59% of the binned community), Agathobacter rectalis (2.85%), and Ruminococcus bromii (2.25%) (Table S5). The majority of OTUs were present in no more than 3 communities and were in low abundance (average abundance of 0.09 % of the total community). Consequently, 66% of all MAGs constituted less than 0.5% of their respective communities. An exception to this were MAGs identified as A. onderdonkii which were present in 17 out of the 27 participant stool communities. MAGs affiliated with A. onderdonkii constituted, on average, 1.0% of each binned community (Table S5).

Identification of MerAB in gut metagenomes.

Assembled, but un-binned contigs were screened for homologs of MerAB homologs as indicators of the ability of the endogenous microbiome to degrade MeHg using canonical degradation pathways (Boyd 2012; Podar et al. 2015). Surprising, homologs of MerB were not detected in any community metagenome. This indicates that canonical pathways for demethylating MeHg are of little importance in human gut microbiomes.

Three MerA homologs were detected in two participant stool communities, two of which were encoded on unbinned contigs. Two of the homologs exhibit 100% sequence identity to MerA from Streptococcus pneumoniae (WP_000958943), a bacterium common to respiratory microbiomes (Dietl et al. 2021). The other MerA homolog was identified in a MAG affiliated (99.3%) with Dellaglioa algida. Yet, the MerA homolog from this MAG exhibited 100% sequence similarity to MerA from Haemophilus influenzae (WP_064083205), also a frequent member of the respiratory microbiome. While multiple other MAGs closely affiliated with D. algida were identified in participant stool communities, none of those MAGs encoded MerA. We presume this MerA to be either acquired in D. algida by a recent horizontal gene transfer or to have been incorrectly binned with D. algida contigs.

Analysis of microbiome structure in relation to elimination rate.

Metagenomes from Trial 1 only were used for downstream analyses and correlations with elimination (Table S6, S7) since these were unperturbed by prebiotic, which showed variable effects on elimination in Trial 2. In addition, data from two participants was not obtained due to insufficient hair Hg analysis for an elimination rate determination (S2) and sequence data that did not meet assembly quality cut-offs (S1). These 27 metagenomic sequences were used for assigning operational taxonomic units (OTUs) and for analyses of abundance, diversity and community structure in relation to elimination kinetics (Table S6–8).

Faith’s phylogenetic diversity (PD) of each metagenome did not significantly correlate with participant MeHg elimination rate (Pearson R = 0.219, p = 0.272) (Table S8). To further investigate connections between community structure and elimination rate, we performed a standardized effect size (SES) of mean pairwise distance (MPD-SES) to determine the degree of phylogenetic clustering in the communities relative to the null model (shuffled nodes). Of the 27 communities subjected to this analysis, 6 communities were significantly dispersed (SES of MPD vs null communities, mpd.obs.z > 0, p > 0.95), 16 communities were non-significantly dispersed (mpd.obs.z > 0, p < 0.95), and 6 were clustered, albeit non-significantly (mpd.obs.z < 0, p > 0.05). When considered together, there was no correlation between overall community phylogenetic structure and participant elimination rate.

Relationships between community composition and MeHg demethylation rate.

A Principal Coordinate Ordination (PCO) was generated using the taxonomic composition of the binned communities comprising each metagenome to visually evaluate if there are relationships between community composition and MeHg demethylation rate. To facilitate this, participants were assigned to quartiles based on MeHg elimination rate, then colored on the PCO according to their quartile. Visual inspection of the PCO demonstrated that communities do not generally cluster based on MeHg elimination rate, as indicated by the dispersion of communities in the same quartile (Figure 6). There was no significant correlation between mean pairwise distance (MPD) and participant MeHg elimination rate (p-value: 0.689, pearson correlation −0.081), indicating that overall community composition is not strongly correlated with MeHg elimination.

Fig 6. Principal component ordination (PCO) representing abundance weighted Bray-Curtis dissimilarity of OTUs across 27 metagenomes.

OTUs were generated at 95% ANI (species level) and abundance weighted. Each point represents the taxonomy of a single metagenomic community via the abundance of its OTUs. Greater distance between points indicates greater dissimilarity between metagenome taxonomy. Points are colored based on standard deviations (S.D.) away from mean elimination rate: Purple: >1 S.D. below mean, Blue: < 1 S.D. below mean, Yellow: < 1 S.D. above mean, Red: >1 S.D. above mean. The correlation between MPD (mean pairwise distance) and participant MeHg elimination rate did not reach significance (p-value is 0.689, and the pearson correlation is −0.081)

Taxonomic and functional analysis of metagenomes and relationship to participant MeHg demethylation rate.

Three independent approaches were taken to identify taxa whose distribution and abundance in the gut microbiome metagenomes positively correlate with participant MeHg demethylation (see methods). Approach 1: Assembly- and bin-independent determination of community taxonomic composition. Using assembly- and bin-independent taxonomic analysis of sequence reads and their abundance, each sequence read (k-mer) was assigned a taxonomic designation, and then the taxonomic structure of the entire community was estimated via the taxonomic designation of the component k-mers. Evaluation of quality-filtered and trimmed metagenome k-mers resulted in the identification of 11,270 taxon units across the 27 participant stool sample metagenomes. Pearson regressions of the abundances of these 11,270 taxon units, as inferred by k-mer read depth, as a function of participant MeHg elimination rate indicated that 234 were significantly correlated with MeHg elimination rate. Of these significant taxon units, the genus Alistipes was the most abundant, with 3.6% of identifiable k-mers within the 27 community metagenomes. The abundance of Alistipes was positively correlated with participant MeHg elimination rate (Pearson R = 0.444, p = 0.010). Of similarly high abundance were k-mers from the class Bacilli, which were negatively correlated with participant MeHg elimination rate (Pearson R = −0.406, p = 0.034). Approach 2: Community metagenome assembly, binning, and taxonomic analysis. Next, we used an assembly- and bin-dependent approach wherein the whole community structure was inferred based on the taxonomic assignment and inferred abundance of each component MAG. These two approaches then entailed identifying correlations between metagenome taxonomic composition and participant MeHg elimination rate. MAGs organized into operational taxonomic units (OTUs) (>95% average nucleotide identity (ANI)) were subjected to Pearson regressions to identify those that correlated with participant MeHg elimination rate. The distribution and abundances of 10 OTUs were significantly correlated with participant MeHg elimination rate. However, four of these were detected in five or fewer metagenomes. Consequently, their significance was presumed to be artifactual. Of the five remaining OTU’s, three were positively correlated with elimination rate and two negatively correlated with elimination rate (Table 1). ANOVAs were also used to evaluate the abundance of OTUs in participant groups that were generated based on their elimination rate (Table 1). One OTU, A. onderdonkii, had the most significant p-value in both statistical tests (Table 1). Additionally, the abundances of two OTUs, both affiliated with the family Lachnospiraceae, had a statistically significant (Pearson p-value < 0.02, Anova p-value = 0.01) negative correlation with elimination rate in both statistical tests. Approach 3: Community metagenome protein clustering. The third approach entailed clustering proteins encoded on assembled contigs (>1000 bp) into clusters based on sequence similarity and then comparing the distribution and abundance of individual protein clusters in community metagenomes to the participant MeHg elimination rate. Proteins encoded on assembled contigs were subjected to hierarchical clustering based on amino acid sequence similarity and were then subjected to Pearson regression to identify relationships between protein sequence frequency, as determined by read depth of contigs encoding those proteins, with participant MeHg elimination rate. Of the 734,023 protein clusters generated, 19,101 were significantly (p < 0.05) correlated with participant MeHg elimination rate. The number of significant protein clusters per OTU (calculated in Approach 2) was then calculated, weighted by the Pearson R value (negative or positive), and then normalized to MAG size (Figure S2). OTU 15, related to A. onderdonkii, had the highest average proportion of significant proteins (3.3% of the genome) that significantly correlated with elimination rate. This is consistent with the findings of the bin-independent (Approach 1) and bin-dependent (Approach 2) analyses, both of which found that the abundance of A. onderdonkii is significantly and positively correlated with participant MeHg elimination rate.

Table 1. Operational taxonomic units (OTUs) correlated with MeHg elimination rate.

OTUs with significant Pearson correlation comparing their abundance with the participant MeHg elimination rate are shown. The p-value for the Pearson correlation and the ANOVA are indicated along with the F-stat comparing OTU abundance with participant elimination bin (bins are based on standard deviations away from mean elimination rate across all participants). Also indicated is the average correlation of all proteins encoded in the MAG comprising that OTU, the average abundance of that OTU, and the taxonomic classification of that OTU.

OTU	p-value (Pearson Correlation)	Pearson Correlation	p-value (ANOVA)	F-stat (ANOVA)	Average correlation of encoded proteins	# of samples the OTU is present in (out of 27)	Classification
15	0.007	0.503	0.008	8.261	0.296	18	Alistipes onderdonkii (Bacteroidota)
62	0.019	−0.450	0.010	7.647	0.076	10	CAG-81 sp900066535 (Firmicutes, Lachnospiraceae)
25	0.028	−0.423	0.010	7.667	0.180	17	Blautia_A sp900066165 (Firmicutes, Lachnospiraceae)
121	0.045	0.399	0.136	2.374	0.226	5	Bacteroides salyersiae (Bacteroidota)
30	0.049	0.388	0.115	2.666	0.210	16	Adlercreutzia celatus (Actinobacteriota)

False discovery rate correction was used to determine the likelihood that OTU 15 (Alistipes onderdonkii) is a false positive. The q-value for OTU 15 was 0.25, indicating a 25% chance of being a false positive. However, when considering that all three statistical analyses above point towards the relevance of this OTU, it seems unlikely that Alistipes onderdonkii is truly a false positive, rather than a true positive. Moreover, the appropriateness of false discovery rate correction in microbiome studies has recently come into question, like in Xiao et al (Xiao et al. 2017), because these approaches falsely require the assumption that all OTUs in a microbiome are independent, when in reality, most phenotypic outcomes are likely due to a complicated ecological interaction between many dependent OTUs, rather than single, independent, OTUs.

Co-occurrence of A. onderdonkii and other community members.

Co-occurrence of OTUs can signal a commensal activity or function of certain species and strains of microbes. In general, the co-occurrence analysis demonstrated that the community structure did not correspond with elimination rate. There was no tendency for OTUs that positively correlated with MeHg elimination to significantly co-occur with other positively correlated OTUs. Furthermore, no positively correlated OTUs exhibited negative correlations with negatively correlated OTUs, which would be expected if those two taxa excluded each other. Rather, OTUs tended to co-occur in patterns apparently unrelated to their correlation with MeHg elimination rate. For example, the abundance of OTU 15, A. onderdonkii, was significantly correlated (p<0.05) with the abundance of 45 other OTUs. All of these correlations were positive, indicating that these OTUs tend to co-occur. However, none of these other OTUs were significantly correlated with MeHg elimination rate. Inspection of the generated co-occurrence network (Figure 7) confirms there was no clear separation of positively and negatively correlated OTUs, and the network structure suggests that OTU co-occurrence patterns are unrelated to MeHg elimination rate.

Fig 7. OTU abundance, co-occurrence, and correlation to MeHg elimination in Trial 1 metagenomes.

Nodes represent OTUs present in at least 10 of 27 metagenomes surveyed. Size of nodes depicts correlation to MeHg elimination rate where yellow indicates positive and blue indicates negative correlation. A. onderdonkii node is outlined in green. Edges represent significant Pearson correlation between OTU co-occurrence and abundance (p < 0.05). Distance is inverse of correlation (shorter distance = stronger correlation), red lines represent significant negative correlations.

Gnotobiotic A. onderdonkii colonized mice.

The significant correlation of A. onderdonkii occurrence and abundance with MeHg elimination rate, together with the lack of co-occurrence patterns suggested that A. onderdonkii may independently confer greater elimination activity to the gut. To test this, GF mice were colonized with A. onderdonkii and assessed for MeHg elimination. A. onderdonkii gnotobiotic mice showed a t_1/2 of 7.7 days, which was not significantly different than the GF mice (t_1/2 of 7.1 days, Fig 4, and Table S2A). In addition, we observed 29.7% of tHg in the Hg(II) form in pellets of A. onderdonkii mice, which was not significantly greater than Hg(II) seen in GF mice (Fig 5, Table S3A). Overall, these data suggest that while A. onderdonkii is associated with faster MeHg elimination rate in the context of the human gut microbiome, in isolation it does not restore MeHg elimination and demethylation kinetics when mono-colonized in the mouse gut. Importantly, since the levels of A. onderdonkii in the mice was not quantitatively established, the lack of correlation could point to low levels of colonization of this strain when in isolation.

Discussion

Requirement of MeHg demethylation in elimination.

Our results strongly support a mechanism whereby MeHg demethylation in the gut is necessary to achieve a maximal elimination rate of MeHg from the body. We find that, in humans, MeHg elimination rate correlates with the degree of MeHg biotransforming activity (demethylation, reduction and volatilization) in human stool cultures. This agrees with our prior finding of a correlation of elimination rate with proportion of tHg that is Hg(II) in human stools, the later being a proxy for demethylation(Caito et al. 2018). Furthermore, in mice, a drastic slowing of elimination induced with GF and Abx treatment conditions correlates with a much-reduced demethylation (lower %Hg(II)) in the stool. Despite this strong overall correlation of demethylation and MeHg elimination rate, we also observed conditions where demethylation and elimination rate are not strictly correlated. In humanized mice, elimination rates equivalent to that of CV mice are achieved where the demethylation (%Hg(II) in stool) was in the range of 54–64% Hg(II), compared to 83% Hg(II) in CV mice. This finding indicates that maximal elimination rates can be achieved with a significant fraction of the MeHg in the feces escaping demethylation. This profile is consistent with there being a “threshold” at which the extent of demethylation of MeHg in the gut (e.g., approximately 50% in stool) is sufficient for maximal elimination. Importantly, these data indicate that leveraging the demethylating characteristic of the microbiome to influence elimination rate may have a more limited application, being useful in individuals that are previously shown to have MeHg elimination rates well below normal or that have experienced a recent antibiotic treatment. Importantly, these data support the notion that the microbiome may have additional influences on gut physiology that augment MeHg elimination in a host dependent manner.

Evidence for host determinants of MeHg elimination.

Two lines of evidence indicate the host plays a role in determining MeHg elimination. First, the fact that GF mice carry out elimination at a rate that is only reduced by half of that of conventional mice indicates that host tissues can harbor mechanisms to eliminate MeHg without a bacterial contribution. It is of note that prior investigations using rodent models tested for elimination in a GF context that typically employed antibiotics to achieve this condition (Rowland 1988; Rowland et al. 1977). To our knowledge, this is the first evaluation of MeHg in a GF condition performed under sterile technique. Furthermore, when comparing GF versus Abx treatment conditions, we see that demethylation of MeHg is prevented to the same degree (i.e. 17–20 % Hg(II) in stool) yet, the elimination rate of the GF mouse is significantly faster than the Abx treated mouse. As both of these mice models are essentially devoid of gut microbes, it can be inferred that the host tissues are influencing this significant elimination difference. Alternatively, it could be that the Abx acts upon the gut tissue to enhance MeHg uptake and thereby increase MeHg retention in the enterohepatic cycling. However, this is unlikely since Abx is not seen to slow elimination in GF mice.

A second line of evidence for host-dependent elimination activity comes from the non-linear association of elimination rate with the degree of demethylation with humanized mice, as mentioned above, and seen previously in human stools (measured as %Hg(II) in the stool). Caito et al, (Caito et al. 2018) showed that several individuals exhibiting >95% demethylation can vary in elimination rate more than 2-fold (e.g. t_1/2 = 30 to 63 days), indicating factors aside from microbial demethylation play a significant role. Overall, these findings have revised our understanding of the critical factors contributing to MeHg metabolism and elimination in humans, reinforcing a central role for the microbiome, while also defining a role for host derived factors.

The nature of the microbial activity that contributes to MeHg demethylation and enhanced elimination remains unknown.

Our in-depth statistical and functional examination of MAGs failed to identify the demethylating merB homologs. This is in keeping with several studies that have failed to find merB homologs in human gut microbiomes (Guo et al. 2018; Rothenberg et al. 2016) including our prior findings that merB homologs are rare in a comprehensive search of human microbiome databases (Christakis et al. 2021). This finding brings to question the fundamental nature of MeHg demethylating activity in the human gut. One possibility is that the activity of sulfate reducing bacteria (SRBs), notably of the Desulfovibrio genus, plays a role. For example, D. desulfuricans has been shown to be resident in the rumen of cows and demethylate MeHg in defined culture conditions (Baldi et al. 1993; Gilmour et al. 2011; Graham et al. 2012). A recent study of Yang et al (Yang et al. 2022) presents evidence that higher levels of Desulfovibrio and methanogens in human gut microbiomes in three Chinese populations associate with higher levels of MeHg demethylation. In contrast, no significant correlations of elimination rate with abundance of six different Desulfovibrio taxa were seen across our dataset here (data not shown). It remains a possibility that demethylation proceeds via an abiotic mechanism involving hydrogen sulfide (H₂S) produced by SRBs that mediates the conversion of MeHg to a (CH₃Hg)₂S intermediate and ultimately to Hg(II) (Barkay and Gu 2022; Pan-Hou and Imura 1981). Alternatively, mechanism involving seleno-amino acids or selenium and glutathione mediated demethylation are also possible (Iwata et al. 1982; Khan and Wang 2010). A confirmation of demethylation mechanism will require further investigation of reaction steps in defined culture conditions in vitro.

Bioinformatics points to A. Onderdonkeii.

Our finding that community structure shows no correlation with MeHg elimination suggest that MeHg metabolism and elimination is achieved with just a few select microbial taxa, either independently or within a requisite co-existing consortia. By evaluating the distribution of MAG OTUs (bin-dependent approach), protein clusters, and taxonomy of reads (bin-independent approach) of participant stool microbiomes in relation to participant MeHg demethylation rate, we found that the abundance of A. onderdonkeii is significantly correlated with faster MeHg elimination. Yet, our attempts to monocolonize GF mice with A. onderdonkeii neither restored MeHg demethylation nor returned elimination rate to normal. While this result suggests that A. onderdonkeii itself does not directly demethylate MeHg, further experiments are required to validate that A. onderdonkeii reaches sufficient abundance in a monocolonized status. It remains a possibility that A. onderdonkeii may function in supporting a consortium colonization of a few or several bacterial species that are required to achieve demethylation in the gut. Alternatively, A. onderdonkeii may serve no functional role, but its abundance may simply be a byproduct of favorable conditions that support abundance of the consortia that do promote demethylation. In this regard, it is of note that associations of A. onderdonkeii abundance with elimination rate were not apparent across samples in Trial 2, where microbiomes were undergoing perturbation with prebiotic. Additional analyses will be required to establish whether or not A. onderdonkeii can support a functional role in MeHg metabolism in the context of a more complex microbial mixture, or alternatively serve as a useful biomarker of a favorable gut composition for MeHg metabolism.

Towards an intervention to enhance MeHg metabolism and elimination in people.

Our attempt to induce favorable changes in the gut microbiome with prebiotic to enhance MeHg elimination gave variable results. We interpret this variable effect as being consistent with our metagenomic analyses that indicate MeHg elimination is not related to bacterial community structure itself. However, it is interesting to note that a subtle but significant correlation of elimination rate was seen among participants who did or did not eat any tuna, whereby tuna fish eaters showed faster MeHg elimination. With this finding, it is tempting to speculate that consistent exposure of the human gut microbiome to low levels of MeHg (e.g. in the tuna) can condition the gut composition by enriching it for MeHg metabolizing taxa. Alternatively, a favorable microbiome may be conditioned from other nutrients that are elevated in tuna, including selenium (Yamashita et al. 2010), which may also play a direct role in MeHg metabolism via conjugation. Nonetheless, these results suggest that enhanced metabolism and elimination of MeHg from humans can be achieved through the diet, yet may require a more complex nutritional profile than a simple pre- or probiotic supplement. Future studies are warranted to characterize the MeHg kinetics in habitual tuna eaters to investigate a potential compensatory effect of the gut microbiome that is chronically exposed to MeHg.

A new perspective on microbial determinants of MeHg metabolism in humans.

Speciation of tHg excreted in stool on day 7 in Abx and GF mice shows two important properties: 1) the host’s ability to demethylate MeHg without bacteria present and 2) the host’s ability to excrete Hg as MeHg. The current understanding is that demethylation occurs through microbially-mediated mechanisms, however, the GF and Abx mouse excretes approximately 20% Hg(II) in stool demonstrating that demethylation is occurring, albeit at a much-reduced level. It stands to reason that GF and Abx mice both have unidentified host-mediated mechanisms of demethylation activity. One possibility is that MeHg demethylation is occurring intracellularly at the level of host tissues, for example, in the gut epithelial cells or in the liver. Yet another possible contributor to MeHg elimination is the sloughing off of gut epithelial cells as a course of normal gut physiology. This latter process may be responsible for MeHg excretion without demethylation. The extent to which these host-derived mechanisms actually rely on the microbiome will require more investigation, exploiting the system of complementary in vivo and in vitro models that we have developed here.

Limitations of the study.

One limitation of this study is the small number of participants and human microbiome samples evaluated. Despite the wide range in elimination rates observed, the statistical power for making associations is limited by the few numbers of participant exhibiting the high and low rates of elimination. Another possible confounding aspect of the study concerns the interaction of omega-3 fatty acids (FAs) from the fish with the microbiome. Evidence points to an ability of omega-3 FAs to influence gut microbial composition (Menni et al. 2017; Zhuang et al. 2020). While all participants received the same fish, those who did not previously have fish in their diet may have experienced more substantial and acute microbiome changes than the regular fish eaters, which could in turn affect their MeHg kinetics. Also, our analysis of humanized GF mice is limited to two samples, and warrants a broader evaluation. While an association of sex with variability in elimination rates is not seen in the relatively limited number of humans evaluated so far, sex dependent effects on the microbiome manipulation in mice stands to be investigated in greater detail.

Summary and Conclusions.

An evaluation of the relationship of MeHg elimination rate and demethylation as a consequence of manipulations of the gut microbiome in both humans and mice have concluded that the gut microbiome harbors a potent demethylation activity independent of the host, that in turn contributes greatly to the overall elimination of MeHg from the body. Nonetheless, MeHg elimination can occur in the absence of gut microbes, albeit at a much-reduced rate and reduced level of demethylation. The microbial species required for demethylation in the gut remain to be identified, however, are likely to share an environment that favors abundance of A. onderdonkeii. Overall, these findings highlight the ability of gut microbiome to strongly influence MeHg kinetics, which furthermore, may be enhanced by diet and vary in conjunction with host derived factors.

Data availability.

The datasets generated are available in the NCBI SRA repository, bioproject: PRJNA954827. (web link pending)