A low dose of beryllium exposure influences gut microbiota and promotes metabolic syndrome.

Elements, including essential minerals and metals, play an important role in human biological functions. Some studies have suggested that exposure to certain elements such as arsenic and cadmium can disturb gut microbiota and result in metabolic disorders. Despite considerable evidence that exposure to toxic heavy metals is significantly associated with host metabolic disorder, few studies have investigated the relationships between fecal elements, gut microbiota, and host metabolism. In order to gain a deeper understanding of the impact of various fecal elements on gut microbiota and metabolic disorder, we have to investigate fecal elements from human stools for association study and demonstrate that certain elements have the causal effect on metabolic disease and changes in gut microbiota. We analyzed 28 elements from 304 feces of human twins and evaluated the effects of fecal elements on both gut microbiota and metabolic disorder. Fecal beryllium content was found significantly correlated with biomarkers of metabolic disease and metabolic disease-related gut microbiota such as Akkermansia and Bifidobacterium . In vitro human fecal culture showed marked reduction of species evenness and Bifidobacterium abundance after beryllium treatment. Notably, in mice fed a high-fat diet, 30 ppb of beryllium exposure resulted in significant body weight gain and increased plasma biomarkers for metabolic disorder with an altered microbial community. Beryllium exposure also affected cecal short chain fatty acid profiles, colonic function, and inflammation. Our findings indicate that low doses of beryllium, almost similar to the current criteria for beryllium in drinking water, perturb the gut microbiota and worsen metabolic disorders, MetS-related high-density lipoprotein cholesterol (HDL), total cholesterol (tCholesterol), low-density lipoprotein cholesterol (LDL), body mass index (BMI), triglyceride, uric acid, aspartate transaminase (AST), alanine transaminase (ALT), waist circumference (Waist), fasting blood sugar (FBS), systolic and diastolic blood pressure (SBP and DBP), high sensitivity C-reactive protein (hsCRP), and fasting blood insulin (FBInsulin), using Spearman’s rank correlation test (Fig. 1a). When fecal elements were sorted based on their correlation with MetS, two elements, Be and Ca, showed the strongest correlation compared with that of other elements. Single MetS-related biomarkers such as BMI, AST, ALT, Waist, hsCRP, and FBInsulin showed a significant positive correlation with Be. In contrast, Ca showed a significant negative correlation with BMI, Waist, FBS, and SBP biomarkers.

environment. We also observed that energy metabolism, such as methane, pyruvate, arginine, and proline metabolism, was enriched in the low-Be group compared with the high group. Thus, we found that Be status in the gut may influence not only microbial composition but also subsequent metabolic function of gut microbiota.
Be induces gut microbiome changes in in vitro cultured human feces.
We next carried out in vitro anaerobic colonic culture using human feces to analyze the impact of Be exposure on the human microbiome and to evaluate the direct relationship between Be and the gut microbiota while excluding the host effect. After 24 h of in vitro cultivation, the microbial composition remained identical to the original human fecal samples except for a slight reduction in Bacteroides (Fig. 3a). Notably, Be treatment markedly altered microbial composition. The Be treatment groups were easily distinguished from the Be untreated group, suggesting that exposure to Be directly alters the microbial community (Fig. 3b). In particular, there was a significant reduction in Coprococcus, Bifidobacterium, Ruminococcus, and Roseburia, whereas Bilophila significantly increased in the Be exposure groups compared with the control group ( Fig. 3c).
Be promotes MetS in high-fat diet (HFD)-fed mice.
To determine the influence of Be exposure on host metabolism and the microbial community, we performed in vivo mouse experiments. We first examined body weight gain (%) and feed intake over 7 weeks. In the normal diet (ND) group, there was no significant increase in body weight gain after 30 ppb of Be exposure and feed intake remained constant (Fig. 4a). However, in the HFD group, 30 ppb of Be exposure resulted in modest but significant gains in overall weight, indicating that Be may accelerate HFD-induced obesity (Fig. 3b). Although 3 ppb of Be did not lead to a significant weight gain over 7 weeks, the weight gain was similar to that of the weight gain observed within the first 2 weeks after treatment with 30 ppb Be. Such an increase in body weight gain was linked to the upsurge of feed consumption during early development. There was no significant difference in the intake of water between all groups, regardless of ND or HFD (Additional file 5: Figure S3). The theoretical average Be exposure was 43.84 ng/week, 300.50 ng/week, and 337.47 ng/week in HFD-Be3, ND-Be30, and HFD-Be30 groups, respectively. Next, we investigated the possible causal relationship between Be exposure and host adiposity. In the ND group, MetS-related biomarkers such as plasma concentrations of triglycerides (TGs), glucose, insulin, leptin, and adiponectin did not show any significant differences after Be exposure, whereas epididymal white adipose tissue (eWAT) weight was significantly higher than in control after exposure to 30 ppb Be (Additional file 6: Figure S4). For the HFD group however and similar to the results of body weight gain, Be exposure worsened host adiposity (Fig. 4c). Plasma levels of TGs, glucose, and insulin were also markedly increased in HFD-Be30. The adipokine leptin also increased after 30 ppb Be exposure, whereas adiponectin significantly decreased in both HFD-Be3 and HFD-Be30 groups compared with control. We were unable to observe a significant difference in eWAT weight across the three HFD groups, possibly because the eWAT was almost saturated due to the 7-week HFD. Taken together, these findings indicate that long-term exposure to Be may promote metabolic disorders in HFD mice.
Be alters the composition of gut microbiota in HFD mice.
To determine whether the metabolic disorder developed after Be exposure is associated with changes in microbial composition, we analyzed the cecal microbiome. In β-diversity analysis, HFD-Be3 and HFD-Be30 showed marked microbial community changes compared with that of the control group, and this clear microbial shift pattern was not seen in ND groups ( Fig. 5a and Additional file 7: Figure S5). To further examine the changes in the gut microbial community of the HFD group due to Be exposure, we performed fecal microbiome analysis using mice fecal samples collected over 5 weeks (Fig. 5b). Starting at week 0 week, prior to Be treatment, three HFD groups were gathered in one cluster.
However, by week 5, the Be exposure groups moved to another cluster. Remarkably, these changes in microbial clusters occurred within a week. We investigated α-diversity, such as Chao1 richness and the Simpson and Shannon diversity indices, to examine whether the reduction in microbial diversity concurrent with a high Be level observed in human fecal samples (Additional file 3: Figure S2) can be reproduced by the in vivo mice study ( Fig. 5c). Indeed, similar to the human fecal analysis results, Simpson and Shannon diversity indices were significantly decreased in the HFD-Be groups. To assess the variation in gut microbiota between groups and identify features differentially abundant in the HFD-Con and HFD-Be groups, we implemented the random forest machine-learning algorithm (Fig. 5d, e) and found significant decreases in relative abundance of Allobaculum and Akkermansia and increasing trends in Lactobacillus and Oscillospira.
Next, we evaluated whether Be exposure affects the production of SCFAs-the major microorganism-derived metabolites (Fig. 6a). Total SCFA content in the cecum was not significantly different between the groups; however, there were significant changes in the ratio of acetate, propionate, and butyrate, which are the three major SCFAs. Acetate levels increased after Be exposure in the HFD group, while propionate and butyrate levels decreased significantly. Such a result was reasonable because Akkermansia, Coprococcus, and Bifidobacterium, which were reduced after exposure to Be, are the representative propionate and butyrate producers and have a strong butyrogenic effect, respectively. Be exposure also affected colonic mRNA expression in HFD-fed mice (Fig. 6b); Be exposure has markedly increased the expression of the mucin (Muc)2 gene, which is associated with gel-forming mucin production, whereas occludin (Occl) and zonula occludens (Zo)-1 expression-associated with intestinal cell integrity-was not significantly affected.
Notably, there was a significant decrease in peptide YY (PYY), the anorexigenic hormone.
Moreover, the inflammation-related genes, including inducible nitric oxide synthase (Nos2) and interleukin (IL)1β, were highly expressed in the Be groups, whereas IL-10 expression, which is associated with inflammation regulation, was inhibited. We further compared the ratio of Gram-positive (G+) to Gram-negative (G-) bacteria between the groups to determine whether the metabolic abnormalities were induced by low-grade inflammation with metabolic endotoxemia. Although the HFD groups showed a significant reduction in the G+/G-ratio compared with that of the ND groups, there was no clear decrease in the G+/G-ratio in HFD-Be groups based on Be concentration (Additional file 8: Figure S6). We then measured plasma lipopolysaccharide (LPS) levels to assess whether LPS from Gbacteria spilled into systemic circulation and caused metabolic endotoxemia (Fig. 6c) and found that an increase in Be exposure led to a minimal increase in plasma LPS levels that was not significant. Nevertheless, plasma LPS levels in the HFD groups were higher than in the ND groups, and HFD-Be3 and HFD-Be30 showed significant increases compared with that of the ND-Con group, indicating that the combination of HFD and Be exposure may lead to more severe endotoxemia. To examine whether the combined increase in Be and LPS, which may have resulted from mucosal barrier disintegration by HFD, is linked to the induction of inflammatory response, we performed in vitro cell experiments using mouse macrophage-like RAW264.7 cells. We carried out Be treatment with or without LPS to assess the combinatory effects of Be and LPS (Fig. 6d) and found that Be + LPS significantly increased nitrite production compared with that of the LPS only treatment group. Be treatments alone did not increase nitrite production.

Discussion
In this study, we conducted metallomic analysis to measure fecal element content and performed NMR-based fecal metabolite analysis of human stool samples. Our results underscore the fact that convergence studies using omics data from excreted human feces may be important for innovative data analysis with systemic perspectives that can help in the acquisition of more information regarding the gut environment.
It has been recently revealed that some heavy metals such as As, Cd, and Pb may lead to a shift in the gut microbial community and worsen metabolic disorders even at relatively low concentrations [7, 9,11]. Although not significant, As, Cd and Pb levels have shown a relatively strong positive correlation with MetS in our studies (Fig. 1a). Moreover, As showed a negative relationship with Odoribacter and Parabacteroides abundance, which were abundant in the healthy group compared with the MetS group in our previous study [14] (Fig. 1b). In contrast, Zn, which has been recently linked to improving diabetes via insulin resistance and blood sugar reduction [5], showed the opposite results in the correlation analysis of MetS and gut microbiota (Fig. 1a, b). These findings indicate that the metallomic analysis results are reliable. Furthermore, Be and Ca were the elements highly associated with MetS (Fig. 1a). According to recent studies, Ca supplementation leads to a significant decrease in body weight and adiposity, along with significant increases in Bifidobacterium, Bacteroides, and Akkermansia abundance [17]; this is consistent with our findings on human fecal samples (Fig. 1b). Although the association between Ca and metabolic disorders has already been reported in some studies [17][18][19], few have reported on the risk of Be on MetS. We thus evaluated the relationship between Be, the human gut microbiome, and MetS status using an in vivo mice model and found that a low dose of Be exposure can influence gut microbial changes and worsen MetS. This finding is of great significance as it reveals the risk of a low dose of Be on MetS.
Studies on the health risks of Be have reported that Be exposure is associated with several adverse health outcomes including lung cancer as well as acute and chronic Be disease [20][21][22]. Because Be toxicity has mostly been reported with regards to the respiratory tract, legal regulations and studies on Be have focused on exposure through inhalation. In this work, we utilized 3 ppb and 30 ppb Be in drinking water to evaluate the effect of low doses of Be exposure via oral intake on the metabolic homeostasis of HFD mice. The daily Be intake level of mice exposed to 30 ppb of Be water was equivalent to ~ 1.7 µg/kg/day estimated from the consumption of drinking water, which closely corresponded to the tolerable daily intake (TDI) of 2 µg/kg/day and was much lower than the commonly known no observed adverse effect level (NOAEL) of Be, which is 0.1 mg/kg/day. Such a low dose of Be was still found to affect body weight gain and worsen MetS symptoms in HFD mice (Fig. 4). In particular, analysis of the gut microbiome and colonic mRNA expression revealed that exposure to both 30 ppb and 3 ppb Be can disturb the microbial community and expression of MetS-related genes in HFD mice ( Fig. 5 and Fig. 6). The current drinking water standard for Be established in 1992 by the Environmental Protection Agency (USEPA, 1992) is 4 ppb, and a recent WHO report determined that it is not necessary to set a formal guideline value for Be in drinking water because Be is rarely found in drinking water at concentrations that warrant health concerns [23,24]. However, according to one study on Be exposure, the estimated total daily Be intake of the general US population was 423 ng, with the largest contributions being from drinking water (300 ng/day) and food (120 ng/day), with smaller contributions from air and dust (2.8 ng/day) [23]. Thus, our results emphasize that more reliable and extensive investigations are required to analyze the occurrence, intake, and toxicity of Be from not only air but also from food and drinking water, and that the current water standard for Be may need to be revised.
Other studies have also evaluated the risk of Be through oral intake [25,26], yet unlike our findings, they reported less than 10% body weight loss in animal studies. Perhaps this is because the experimental conditions were different from our study; for example, the studies focused on the cancer risk of Be, which was therefore tested at a concentration of more than 5 ppm Be, which is approximately 150 times higher than the maximum concentration (30 ppb) we used. Moreover, they evaluated Be risk under ND conditions, whereas we found that a low dose of Be exposure increased risk under HFD conditions.
Notably, other studies found that a low dose of Be exposure increased body weight and feed intake during the first 30 days [27,28]. This is consistent with our results where increased feed intake was observed during the beginning stages of the experiment ( Fig. 3b). Furthermore, our cell experiments demonstrated the combinatory effects of Be and LPS on the increase in inflammation, suggesting that low doses of Be may be toxic under certain conditions (Fig. 6d). There is a related study that showed how co-treatment with LPS and 100 µM Be sulfate (equivalent to 900 ppb Be) significantly increased IL-1β and decreased IL-10 compared with that of the LPS only treatment [29]. These findings are consistent with our findings, and even we could obtain a similar result from the low dose of Be treatment that was about 30 times less than the test. Taking all these findings into consideration, the risk of Be exposure may become more severe when mucosal barrier functions are incomplete, such as under HFD conditions or during early developmental stages.
HFDs increase Gram-negative bacteria abundance and mucosal barrier disintegration, which is followed by LPS spilling into the systemic circulation, causing metabolic endotoxemia with low-grade inflammation [30]. In our study, pro-inflammatory markers such as Nos2 and IL-1β were highly expressed in HFD-Be groups, whereas the expression of IL-10, which is an anti-inflammatory cytokine associated with inflammation regulation, was significantly reduced (Fig. 6b), indicating that colonic inflammation became worse. In addition, HFD-Be groups showed a significant increase in plasma LPS levels compared with that of the ND-Con group (Fig. 6c). Thus, metabolic endotoxemia and intestinal inflammation markedly increased due to the combination of HFD and Be. What is unusual is that the hyper-expression of the Muc2 gene, which is associated with the release of gelforming mucin, was observed (Fig. 6b). According to another study, if pro-inflammatory cytokine levels are increased, then Muc2, Muc1, and Muc4 expression increases in goblet cells, after which the continuous Muc1 and Muc4 expression finally promote cancer development [31]. Given that Be is one of the most toxic metals, the hyper-expression of Muc2 may result from the compensatory host defense response against HFD and Be exposure to prevent mucosal barrier impairment. Moreover, HFDs may not only cause disintegration of the mucus barrier but they can also induce a slow transit through the gut. Consequently, this increases the duration of time toxic elements such as Be remain in the gastrointestinal tract, which would have a higher probability of worsening the gut environment. Therefore, HFD-Be groups showed a more pronounced effect on the altered gut microbiota and worsened MetS compared with that of the ND-Be group.
Be exposure led to significant changes in the gut microbiota. It is noteworthy that our human study, in vitro fecal culture, and in vivo mouse experiments showed Be-dependent reduction in the abundance of Akkermansia and Bifidobacterium, which have been heavily studied for their effects on MetS suppression [15,16]. In particular, Be exposure significantly decreased the production of propionate and butyrate as well as expression of the anorexigenic PYY (Fig. 6a). Because butyrate and propionate are reported to be predominantly anti-obesogenic, inducing the secretion of anorexigenic hormones such as PYY and glucagon-like peptide (GLP)-1 for appetite regulation [32], and as Akkermansia and Bifidobacterium are known as propionate-producing bacteria [33], these changes appear reasonable. Moreover, acetate ratios were markedly increased in the human Behigh group and the HFD-Be mice groups with a temporary increase in feed consumption during early mouse development ( Fig. 4b and Fig. 6a). As the majority of acetate is absorbed into the body where it acts as a substrate for hepatic and adipocyte lipogenesis, acetate is believed to have more obesogenic potential than do propionate and butyrate [32]. In addition, a recent study has shown that high levels of plasma acetate originating from gut microbial dysbiosis can increase glucose-induced insulin secretion to promote the release of ghrelin-an orexigenic hormone-leading to metabolic disease through parasympathetic activation [34]. Thus, Be exposure may lead to worsened MetS concurrent with an increase in appetite resulting from the alteration in gut microbiota.

Conclusions
Overall, our study indicated that fecal elements are highly associated with gut microbiota and MetS. Exposure to Be may lead to changes in gut microbial composition and worsen MetS even at very low concentrations. Our study is significant for several reasons as analysis of fecal elements and metabolites can provide information on environmental factors, including human exposure levels to various elements and chemicals. Moreover, certain gut microbiota were strongly correlated with fecal elements, demonstrating a need for follow-up research on these elements to further elucidate the interaction mechanisms between elements, gut microbiota, and diseases. In particular, we found that a low dose of Be-almost identical to the current global standard levels for drinking water-can be an important cause of MetS by disturbing gut microbiota and inducing inflammation responses in mice fed HFD. Nevertheless, our study has some inherent limitations when conducting cross-sectional studies. Despite these limitations, certain elements showed significant associations with MetS and gut microbiota. Longitudinal studies will be required for specific cohorts along with large-scale element analysis of additional human cohort for further verification. Moreover, further research is warranted to elucidate the disturbance mechanism of Be on gut microbiota and host metabolism.

Human subjects
This study was approved by the Institutional Review Board of the Seoul National University (IRB No. 144-2011-07-11) and was performed according to the Helsinki Declaration.
Written informed consent was obtained from each participant. A total of 304 individuals from participants enrolled in the Healthy Twin Study in South Korea [35] were selected for this study. Fecal samples from participants were collected at home and immediately frozen in a home freezer, followed by transfer to clinics and storage at -80℃ until further analysis. All participants filled in questionnaires covering lifestyle, medication, disease history, biochemical tests, and anthropometrical measurements. The gut microbiome data were acquired from our previous study (accession number: ERPO10289) [14]. The demographic characteristics of the study subjects are listed in Additional file 9: Table S3.

Measurement of MetS components and definition of MetS
Measurements of waist circumference, blood pressure, triglyceride, HDL cholesterol, and FBS have been previously published [14]. MetS was defined following the revised National

In vitro batch culture of human fecal microbiota
In vitro colonic fermentation was performed according to Long's method with minor modification [36]. Briefly, 8 g/L NaCl, 1.15 g/L Na 2 HPO 4 , 0.5 g/L L-cysteine, 0.2 g/L KCl, and 0.2 g/L KH 2 PO 4 were dissolved in distilled water and autoclaved for making PBS medium. Fecal samples were obtained from three healthy donors (age 20-30; mean BMI 22.3) who had taken no antibiotics or prebiotics for three months prior to the study.
Written informed consent was obtained from donors, and the study was approved by the institutional review board of the Korea Institute of Science and Technology (IRB No. 2015-003). Fecal slurry (10% w/v) was prepared by diluting and suspending the fecal samples with PBS medium in an anaerobic chamber (Coy Laboratory Products Inc., Ann Arbor, MI).
The cultivation was started with 5% fecal inoculum by adding 0.9 mL of 10% fecal slurry into 0.9 mL PBS medium (total volume, 20 mL) using a 96-deep well plate. Culturing was (4) ND with water (n = 6), and (5) ND with 30 ppb of Be water (n = 6). Feed and water intake were recorded once a week for each cage and the data were used for the calculation of average intake per mouse per week. The theoretical weekly Be exposure was calculated by multiplying the added Be concentration by the average weekly water intake. Body weight of each mouse was measured once a week and stool samples were collected once a week and immediately stored at -80 ℃ for further analysis. Animals were euthanized by CO 2 inhalation at the beginning of the light cycle and after 16 h of food deprivation. Blood samples were collected by cardiac puncture in microtubes containing Naive Bayes classifier was trained using the 16S rRNA region (V3-V4), the primer set and read length used (319F/806R, 469 bp), and the Greengenes 99% reference set (v13.8) [39]. This trained feature classifier was then used to assign taxonomy to each read using the default settings in QIIME. Microbial composition at a certain level as well as α-and βdiversity were analyzed using MicrobiomeAnalyst [40]. Non-metric multidimensional scaling (NMDS) plots were generated from a Bray-Curtis distance matrix, and a principal coordinate analysis (PCoA) plot was generated using unweighted Unifrac distances to visually represent microbiota compositional differences among groups. Random forest, a supervised learning method for the classification of human microbiome data [41], was used to select subsets of taxa (genus level) that are highly discriminative of the type of community from Be-exposed mice. We measured feature importance as the mean decrease in model accuracy when that feature's values were permuted randomly using 500 trees and seven repetitions.

Microbial functional analysis
Phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt) was used to infer putative functional metagenomes from 16S rRNA gene sequence profiles [42]. As the tool adapts OTUs with Greengene IDs, OTUs were picked with closed reference against the May 2013 Greengenes database. The relative abundance of each functional pathway was obtained for each sample, and non-microbial functional pathways belonging to the "Organismal Systems" and "Human Diseases" categories were excluded from downstream analysis. To determine metabolic features that were differentially abundant between each element status (low and high level), linear discriminant analysis effect size (LEfSe) was applied under the condition α = 0.05, with an LDA score of at least two [43].

NMR-based metabolomic analysis
For NMR-based metabolomic analysis, samples were prepared according to Lamichhane's method with minor modifications [44]. Briefly, human fecal samples (~200 mg) were mixed with 1000 μL DDW, vortexed for 30 s and homogenized with a tissue homogenizer for 5 min. After centrifugation (14,000 × g, 4℃) for 10 min, 300 μL of supernatant was mixed with 60 μL deuterium oxide (D 2 O) containing 0.025 mg/mL 3-(trimethylsilyl) propionic acid-d4 sodium salt, 60 μL of 1 mM imidazole, 60 μL of 2 mM NaN3, and 120 μL of 0.5 M KH 2 PO 4 . The mixtures were vortexed for 1 min and centrifuged at 14,000 × g for 10 min. The clear supernatant was then transferred to a 5 mm NMR tube (Wilmad-LabGlass, Vineland, NJ) for NMR analysis. All 1 H-NMR spectra were acquired using a Varian 500 MHz NMR system (Varian, Palo Alto, CA) equipped with a cold flow-probe. 1 H-NMR spectra were collected at 25 °C using the water presaturation pulse sequence. Spectra were collected with 64 transients using a 4 s acquisition time and a 2 s recycle delay.

SCFA measurement
Cecal SCFAs content was determined by gas chromatography. Cecal contents (~80 mg) were homogenized in 500 μL deionized water, after which the samples were acidified with 50 μL 50% sulfuric acid, followed by vortexing at room temperature for 5 min. After centrifugation at 14,000 × g for 10 min, 400 μL of the supernatant was transferred to a new tube, and 40 μL internal standard (1% 2-methyl pentanoic acid) and 400 μL anhydrous ethyl ether were added. The tube was vortexed for 1 min and then centrifuged at 14,000 × g for 10 min. The upper ether layer was used for further analysis. Volatile Free Acid Mix

Determination of NO production
The nitrite concentration in the culture medium was measured as an indicator of NO production using the Griess reaction. After incubation with test samples for 48 h, the supernatant from each well (50 μL) was transferred to a fresh 96-well plate, after which 25 μL of 1% sulfanilamide and 25 μL of 0.1% naphthyl-ethylenediamine in 5% HCl was added.
After 10 min of incubation at room temperature, the absorbance of each well was measured at 540 nm using a Synergy HT microplate reader (Biotek, Winooski, VT).
Relative nitrite production was calculated relative to the LPS only treatment group.

Statistical and subsequent bioinformatics analysis
Statistical analysis of all grouped data was performed using R software or GraphPad Prism 7 (GraphPad Software Inc., La Jolla, CA). Significance was determined using a two-tailed Student's t test, Mann-Whitney test, or one-way ANOVA corrected for multiple comparisons with a Sidak test compared with the control group. Microbial data processing and multivariate statistical analysis were performed using MicrobiomeAnalyst. Permutational multivariate analysis of variance (PERMANOVA) was performed to test the association between microbiome composition and Be exposure based on NMDS. Association of fecal elements with MetS-related clinical biomarkers and gut microbiota was assessed by Spearman's rank correlation analysis. A correlation heatmap was generated using the R package "Pheatmap." P-values were adjusted for multiple testing with the Benjamini-Hochberg method. Multivariate analysis using a multivariate association with linear models (MaAsLin) was performed to identify significant associations of microbial abundances with metabolic status or fecal element status [46]. Age, sex, and smoking This project was designed by K.H.C. and J.S.Y. Element analysis was performed by J.S.Y. and Be-treated mice (n = 6/group). Data represent the mean ± SD. Significance was calculated using unpaired two-tailed Student's t test. **P < 0.01. Additional file 7: Figure S5. NMDS plot of the cecal microbiota of Be-exposed ND-fed mice.
Bray-Curtis distance matrix was calculated from the genus-level relative abundance data.
Significance was determined using PERMANOVA.
Additional file 8: Figure S6. The in vivo mouse ratios of Gram-positive/Gram-negative bacteria. Box plots show the median (horizontal line), mean (cross) and IQR; whiskers represent the minimum and maximum values. Significance was determined using twotailed Student's t test or one-way ANOVA corrected for multiple comparisons with a Sidak test vs. ND-Con or HFD-Con groups. *P < 0.05; NS, no significant difference. Additional file 9: Table S3. Characteristics of the human study population. BP, blood pressure; DZ, dizygotic; F, female; FBS, fasting blood sugar; HDL, high-density lipoprotein cholesterol; M, male; MetS, metabolic syndrome; MZ, monozygotic; Non-twin, parents or siblings of twin pairs; waist, waist circumference. Values are mean ± SD or n (%).
a Abnormal values for at least three of the following, waist, BP, triglyceride, HDL, and FBS. Figure 1 Associations between fecal elements with MetS risk and human gut microbiota. a biomarkers. Fecal elements measured from human stools (n = 304) are arranged in increasing order of correlation with MetS. Asterisks represent significant associations at FDR adjusted P values of < 0.2. b Spearman's rank correlation between fecal elements and gut microbiota (genus level). The filtered set of taxa excluding low-abundance taxa (< 0.1% of mean relative abundance) was used for correlation analysis. Asterisks represent significant associations at FDR adjusted P values of < 0.01. The microbiota marked in blue and red denote taxa significantly enriched in the healthy group and MetS group, respectively. in our previous study [14].  PCoA score plot with 95% confidence ellipse based on unweighted UniFrac metrics was analyzed to investigate the β-diversity of the community. c Relative abundances for five discriminative taxa, Coprococcus, Bifidobacterium, Microbial perturbation in mice fed HFD and exposed to Be. a Non-metric multidimensional scaling (NMDS) plot of the cecal microbiota of Be-exposed HFDfed mice. b NMDS plot of the fecal microbiota of Be exposure over time in HFDfed mice. Stools were collected at three time points (0, 1, and 5 weeks) and analyzed for 16S rRNA gene sequencing. Bray-Curtis distance matrix calculated from the genus-level relative abundance data was used. Significance was determined using PERMANOVA. c Diversity analysis of the cecal microbial community in Be-exposed HFD-fed mice. Chao1 species richness estimator as well as Simpson and Shannon index values for microbial evenness were calculated to investigate the α-diversity of each group. Box plots show median (horizontal line), mean (cross), and IQR, while whiskers represent the minimum and maximum values. Significance was determined using unpaired two-tailed Student's t test or one-way ANOVA corrected for multiple comparisons with a Sidak test vs. control group. d Feature importance scores for the ten most predictive genera in the random forest classifier. Feature importance was measured as the mean decrease in model accuracy when that feature's values were permuted randomly. e Relative abundance of two discriminative taxa, Allobaculum and Akkermansia.

Figure 6
Effect of Be exposure on cecal SCFAs and colonic function. a Cecal SCFA profiles of mice fed an ND or HFD and exposed to Be. b mRNA levels of the mucin production (Muc2 and Muc3), tight junction (Occl and Zo-1), appetite suppression (PYY and GLP-1), and inflammation-related genes in proximal colons of Beexposed HFD-fed mice. c Plasma LPS levels of mice fed an ND or HFD and exposed to Be. Box plots show median (horizontal line), mean (cross), and IQR, while whiskers represent the minimum and maximum values. Significance was