Fmo5 ^−/− mice, generated on a C57BL/6J background (Gonzalez Malagon et al., 2015), and wild-type C57BL/6J mice were bred at University College London and housed with free access to food and water, as described previously (Gonzalez Malagon et al., 2015). Urine, blood and tissue samples were collected between 9:00 a.m. and 12 noon, as described (Gonzalez Malagon et al., 2015; Veeravalli et al., 2017). All procedures were carried out under appropriate Home Office Licences in accordance with the UK Animal Scientific Procedures Act and with local ethics committee approval (Animal Welfare and Ethical Review Body). Details of animal numbers, timing, doses etc are provided in the Figure captions.

Stomach contents were prepared for metabolite analysis as previously described (Varshavi et al., 2020). All NMR experiments and data analyses were conducted at 600 MHz for ¹H NMR as previously described (Varshavi et al., 2018), including data acquisition and processing.

Briefly, for urine, all samples were collected onto an ice-cooled surface and then frozen on solid CO₂ and stored at 193 K until transported on solid CO₂ for analysis. 50 μl of thawed urine from each mouse was mixed with 25 μl of phosphate buffer (81/19 (v/v) 0.6 M K₂HPO₄/0.6 M NaH₂PO₄ in 100% ²H₂O, pH 7.4, containing 0.5 mM sodium 3-(trimethylsilyl)-2,2′,3,3′-tetradeuteropropionate (TSP), as NMR shift reference, and 9 mM sodium azide, as an anti-microbial. Buffered samples were centrifuged at 13,000 g for 5 min at 4°C to remove any suspended particles. 60 μl of supernatant was then transferred into new 1.7 mm outer diameter NMR tubes (Norell, S-1.7-500-1) using an accurate, long-needle electronic syringe (SGE eVol XR). Samples were loaded into 96-well plates in a Bruker SampleJet sample changer (Bruker BioSpin, Germany) where the temperature was maintained at 277 K. Each sample was automatically pre-heated to 300.0 K prior to data acquisition.

¹H NMR spectra were recorded on a Bruker Avance III spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany) operating at 600.44 MHz and at a temperature of 300.0 K. A standard one-dimensional (1D) NOESY presaturation pulse sequence, Bruker code noesygppr1d, with gradient pulses (RD-90°-t1-90°-tm-90°-acquire) was acquired with water suppression applied during the relaxation delay of 2 s, with a mixing time of 100 ms and a 90° pulse of 11.2 μs. For each spectrum, 8 dummy scans were used to establish spin-state equilibrium, then 256 free induction decay transients were collected into 65,536 data points with a spectral width of 20 ppm.

NMR spectra were processed using the software TopSpin 3.2 (Bruker BioSpin, UK). Prior to applying Fourier transformation, free induction decays were multiplied by an exponential function corresponding to a line broadening of 0.3 Hz. The 1D ¹H NMR spectra were manually phased, baseline corrected and referenced to the chemical shift of TSP (0.0 ppm).

2D NMR methods such as J-resolved, correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), HSQC and HMBC were acquired to assist with spectral interpretation and metabolite identification. Representative experimental data acquisition and data processing parameters for all these experiments on the urine of a week 60 FMO5 knockout mouse are given in Supplementary Table S1. The key experiments for metabolite identification were 2D ¹H, ¹H COSY, 2D ¹H, ¹³C HSQC and 2D ¹H, ¹³C HMBC experiments which enabled proton-to-proton correlations over 2 to 6 bonds, proton to carbon-13 correlations over 1 bond and proton to carbon-13 correlations over 2 to 3 bonds respectively, to be determined.

Metabolite identification was accomplished according to published guidelines (Sumner et al., 2007; Everett, 2015; Dona et al., 2016). The 1D and 2D NMR spectra shown in the Figures were processed using MNova 11.0 (Mestrelab Research S.L.). Metabolites were quantified by NMR signal area measurements as described previously (Veeravalli et al., 2020).

Urine was spiked with authentic meso-2,3-butanediol and 2R,3R-butanediol (Sigma-Aldrich), and analysed by NMR, as described above. High-resolution ultra-performance liquid chromatography–mass spectrometry (UPLC-MS) was conducted on urine samples from 30-week-old Fmo5 ^−/− and wild-type mice, essentially as described previously (Varshavi et al., 2018), with the following modifications. Urine samples (60 μl) were centrifuged (13,000g for 10 min at 4°C) to remove particulates. Supernatant (50 μl) was diluted with 100 μl of water, centrifuged at 13,000g for 5 min at 4°C then transferred into a vial before injection into the UPLC-TOF-MS (Waters, Manchester, UK). UPLC was in isocratic mode with an eluent of 0.1% formic acid, 10% acetonitrile. The flow rate was 0.4 ml/min, injection volume was 1μl, the MS source temperature was 130°C and the scan time was 0.5 s. To attempt to distinguish between the two enantiomeric forms of 2,3-butanediol, urine was spiked with the enantiomeric NMR shift reagents sodium [(R)-1,2-diaminopropane-N,N,N′,N′-tetraacetato] samarate (III), hydrate and sodium [(S)-1,2-diaminopropane-N,N,N′,N′-tetraacetato] samarate (III), hydrate (Tokyo Chemical Industry UK Ltd.) and analysed by 600 MHz ¹H NMR.

Mice were treated by oral dosing of ampicillin (1 g/l) and neomycin (0.5 g/l) in their drinking water for 14 days. Antibiotic-containing water was replenished twice weekly, and animals were given free access to the standard chow diet, as described previously (Scott et al., 2017).

Wild-type mice were dosed orally in their drinking water with 2,3-butanediol in a 4:1:1 ratio of meso-2,3-butanediol (Sigma-Aldrich):2R,3R-butanediol (Alfa-Aesar):2S,3S-butanediol (Fischer Scientific). Dosing regimes and animal numbers are given in the text and figure legends.

Plasma was isolated from freshly collected blood samples, as described (Hough et al., 2002), and biochemical analyses carried out at the Medical Research Council Harwell Institute (Harwell, Oxfordshire, UK). Insulin was measured by enzyme-linked immunosorbent assay, as described previously (Scott et al., 2017).

Fresh fecal lysates were prepared, as described (Willing et al., 2011), from donor Fmo5 ^−/− male mice (n = 3), which were 16-weeks old at the start of the experiment. The recipient group were wild-type, male mice treated with a single injection of streptomycin (2 mg/g body weight) 1 day before the first fecal transplant dose (Willing et al., 2011). Recipient mice were treated with fecal lysate via oral gavage on days 1, 4, and 7. An age-matched control group was treated via oral gavage with phosphate-buffered saline on days 1, 4, and 7. At the start of the experiment recipient and control wild-type mice were 9-weeks old. Urine was collected 1 day after each gavage treatment and then at 13, 15, and 18 weeks of age. Blood was collected at 18 weeks of age.

Stomach contents were isolated after dissection of the intestinal tract. Stomach bacteria were identified and quantified, via analysis of 16S ribosomal RNA gene sequences, as described previously (Scott et al., 2017).

Data are expressed as means ± SEM. Statistical significance was determined using an unpaired two-tailed t test and GraphPad Prism (version 9.1.0; GraphPad Software, Inc., La Jolla, CA) or by false-discovery-rate (FDR)-controlled ANOVA as described previously (Varshavi et al., 2018).

Previously we have reported an NMR-based analysis comparing the urinary metabolite profiles of Fmo5 ^−/− and wild-type male mice as they age (Varshavi et al., 2018). Here we extend this study to investigate discriminating features between Fmo5 ^−/− and wild-type mouse urine. Unbiased, unsupervised, multivariate principal component analysis of the ¹H NMR spectra of urine from Fmo5 ^−/− mice showed it to be distinct from the urine of age-matched wild-type mice at all timepoints studied: 15, 30, 45 and 60 weeks of age. Unusual methyl proton signals at ca. 1.146 and 1.149 ppm (Figure 1A and Supplementary Figure S1) were present in the ¹H NMR spectra of urine of all Fmo5 ^−/− mice and absent from the spectra of urine of all wild-type mice; at 60-weeks of age (Figure 1A, n = 5 KO, n = 4 WT, p = 0.00015 and p = 0.0002) and also at 15-weeks (Supplementary Figure S1, n = 4 KO, n = 4 WT, p = 0.0002 and p = 0.0003), 30-weeks (n = 4 KO, n = 4 WT, p = 0.000008 and p = 0.0002), 30-weeks (Set 2 repeat study, n = 5 KO, n = 5 WT, p = 0.000004 and p = 0.000002) and 45-weeks of age (n = 5 KO, n = 4 WT, p = 0.00006 and p = 0.00001, all values for the signals at ca. 1.146 and 1.149 ppm respectively and all false discovery rate controlled with an FDR = 0.1).

The metabolites responsible for the unique signals in Fmo5 ^−/− mouse urine were identified by a series of 2D NMR spectra. A 2D ¹H J-resolved NMR spectrum (Supplementary Figure S2) showed that the pseudo-doublets (Figure 1A) were in fact complex, second order, multiplet signals, with the signal at 1.146 ppm having at least six transitions and that at 1.149 ppm having at least four transitions, which is unusual for methyl group resonances. A 2D ¹H COSY (correlated spectroscopy) NMR spectrum of Fmo5 ^−/− mouse urine (Supplementary Figure S3) showed clear vicinal coupling connectivities from the methyl signals at 1.146 and 1.149 ppm to methyne hydrogens, resonating at ca. 3.734 and 3.630 ppm respectively. These data indicated that the signals were due to the meso- (Human Metabolite Database, HMDB03156) and enantiomeric (HMDB33007) isomers of the small diol, 2,3-butanediol. The identity of the metabolites was confirmed by 2D ¹H HSQC (heteronuclear single quantum coherence) and HMBC (heteronuclear multiple bond correlation) NMR spectra, with spectral data matching those of reference materials (Supplementary Table S2). These identifications were verified by spiking authentic reference standards into the urine sample of an Fmo5 ^−/− mouse (Figure 1B).

Orthogonal confirmation of the identification of 2,3-butanediol was achieved by high-resolution UPLC electrospray MS experiments. Authentic meso- and 2R,3R-butanediol eluted at retention times of ca. 1.0 and 1.1 min respectively. The positive-ion electrospray mass spectra of the two isomers were dominated by [M + H-H₂O]⁺ peaks at m/z 73.0646 (theoretical 73.0648, taking into account the mass of the electron; mass error 2.7 ppm). The single-ion chromatogram for m/z 73.0650 showed peaks at 1.0 and 1.1 min in urine from a 30-week-old Fmo5 ^−/− male mouse, confirming the presence of both the meso- and enantiomeric-isomers of 2,3-butanediol. No such signals were present in the urine of a 30-week-old wild-type mouse, confirming the absence of 2,3-butanediol isomers.

To determine which isomer or isomers of enantiomeric-2,3-butanediol were present, experiments were conducted with both the R and S enantiomers of the aqueous-soluble, chiral chemical shift reagent [1,2-diaminopropane-N,N,N′,N′-tetraacetato] samarate (III). Unfortunately, no significant enantiomer-specific chemical shift changes were observed in the ¹H NMR spectra when these reagents were added to authentic 2R,3R- and 2S,3S-butanediol. Thus, the chirality of the enantiomeric 2,3-butanediol observed in the urine of Fmo5 ^−/− mice was not determined.

Quantification of the NMR signals for the 2,3-butanediol isomers in the urine of 15-, 30-, 45- and 60-week-old male Fmo5 ^−/− mice showed that the meso form was always present in a greater amount than the enantiomeric form(s) (Figure 1C) and the ratio of meso to enantiomeric form(s) increased with age (Figure 1D).

Fmo5 ^−/− mice have lower plasma concentrations of total and HDL cholesterol than their wild-type counterparts (Gonzalez Malagon et al., 2015). The gut microbiota is known to influence cholesterol metabolism of the host animal (Kriaa et al., 2019). To determine whether the gut microbiota contributed to the lower plasma cholesterol of Fmo5 ^−/− mice we investigated the effect of antibiotic treatment. Treatment of Fmo5 ^−/− mice with antibiotics increased the plasma concentrations of both total cholesterol, from 2.50 ± 0.18 mmol/l to 3.58 ± 0.17 mmol/l (p < 0.01), and HDL cholesterol, from 1.37 ± 0.13 mmol/l to 2.52 ± 0.12 mmol/l (p < 0.001), in each case reaching levels similar to those in wild-type mice (Figures 2A,B). In contrast, antibiotic treatment of wild-type mice had no effect on plasma concentrations of either total cholesterol or HDL cholesterol (Figures 2A,B). The results indicate that the lower plasma cholesterol concentration of Fmo5 ^−/− mice was influenced by the gut microbiome.

To determine whether the 2,3-butanediol detected in Fmo5 ^−/− mouse urine was of microbial origin, urine from antibiotic-treated Fmo5 ^−/− mice was analysed by ¹H NMR. Antibiotic treatment decreased the concentrations of the meso and enantiomeric isomers by about 85% and 80% respectively. The number of bacteria detected in feces of antibiotic-treated Fmo5 ^−/− and wild-type mice was substantially depleted and the small population that remained was dominated by Firmicutes (94%) in both sets of animals. Although 2,3-butanediol was not fully eliminated by the antibiotic regime used, the results indicate that both the meso and enantiomeric forms are largely of microbial origin.

Based on the finding that antibiotic treatment of Fmo5 ^−/− mice increased plasma concentrations of total and HDL cholesterol, and concomitantly reduced the concentrations of 2,3-butanediol, we hypothesized that 2,3-butanediol affects plasma cholesterol concentration. To test this, we investigated the effect of 2,3-butanediol treatment of wild-type mice on their plasma cholesterol concentration. Quantification of 2,3-butanediol isomers in the urine of Fmo5 ^−/− mice (Figure 1D) indicated that a reasonable approximation for dosing wild-type mice would be a 2:1 mix of the meso:enantiomeric forms. Because the proportion of the R,R and S,S enantiomers of 2,3-butanediol in Fmo5 ^−/− mouse urine is not known, we used a 4:1:1 mix of meso:R,R:S,S isomers. We aimed to dose wild-type mice with an amount that would result in their excreting 2,3-butanediol in urine in concentrations similar to those excreted by Fmo5 ^−/− mice. To determine an appropriate dose, wild-type mice were treated with a range of concentrations of the 4:1:1 mix of 2,3-butanediol isomers in their drinking water and the 2,3-butanediol excreted in urine quantified. The dose received by each animal was calculated on the basis of the average daily water intake of the mice. Mice dosed with the equivalent of 250 mg/kg/d had a urinary concentration of 2,3-butanediol in the range excreted by Fmo5 ^−/− mice.

Wild-type mice, aged 13 weeks, were randomly placed into four cohorts and their plasma cholesterol measured. There was no significant difference in plasma concentrations of either total or HDL cholesterol among the four cohorts (Figures 3A,B). Cohorts were then treated with 2,3-butanediol for 4 weeks at doses of 60, 250 or 600 mg/kg/d or were untreated. 2,3-butanediol was present in the urine of all treated mice, but was absent from urine of untreated animals (Supplementary Table S3). No differences were detected by ANOVA, with an FDR of 0.1, in the abundance of any other metabolites in the urine of treated and untreated mice. However, taurine levels were statistically significantly, inversely correlated with 2,3-butanediol levels in the urines of the treated male mice (data not shown).

Over the 4-week period, from age 13–17 weeks, in untreated mice the plasma concentration of total cholesterol increased from 2.75 ± 0.15 mmol/l to 3.92 ± 0.30 mmol/l (p < 0.01), an increase of 43% (Figure 3A), and that of HDL cholesterol from 1.96 ± 0.11 mmol/l to 2.71 ± 0.18 mmol/l (p < 0.01), an increase of 38% (Figure 3B). Similar age-related increases in plasma total cholesterol and HDL cholesterol have been reported (Gonzalez Malagon et al., 2015). After 4 weeks of treatment with 2,3-butanediol at a dose of 250 mg/kg/d, plasma concentrations of both total cholesterol (2.98 ± 0.21 mmol/l) and HDL cholesterol (2.08 ± 0.09 mmol/l) were significantly lower (p < 0.05) than those of mice that were untreated for 4 weeks, and not significantly different from the concentrations at the start of the treatment (Figures 3A,B). Therefore, 2,3-butanediol, at a dose of 250 mg/kg/d, prevented age-related increases in plasma concentrations of total and HDL cholesterol. However, treatment of mice with 2,3-butanediol at doses of 60 or 600 mg/kg/d did not prevent age-related increases in plasma concentrations of either total or HDL cholesterol, which reached levels that were not significantly different from those of mice untreated for 4 weeks (Figures 3A,B).

We also investigated the effect of 2,3-butanediol treatment on the plasma concentrations of non-esterified fatty acids (NEFA) and triglycerides. At the start of the experiment (13-week-old mice) there were no significant differences in NEFA or triglyceride concentrations among the four cohorts. After 4 weeks, NEFA and triglyceride concentrations of untreated mice were not significantly different from those at the start of the experiment. Treatment of mice for 4 weeks with 2,3-butanediol at a dose of either 250 or 600 mg/kg/d significantly decreased plasma concentrations of both NEFA and triglycerides, whereas a dose of 60 mg/kg/d had no significant effect (Figures 3C,D).

2,3-butanediol treatment at doses of 60, 250 or 600 ng/kg/d had no effect on the plasma concentration of glucose (Figure 3E). After treatment of mice with 250 mg/kg/d, the plasma concentration of insulin (1.93 ± 0.51 ng/ml) was not significantly different from that of untreated mice (1.75 ± 0.19 ng/ml). The effect on plasma insulin concentration of doses of 60 and 600 mg/kg/d was not determined.

Treatment of mice with 2,3-butanediol had no effect on food intake or body weight. However, the EWAT:body weight ratio was significantly decreased in mice dosed with 2,3-butanediol at 250 mg/kg/d (Figure 3F). Although the EWAT:body weight ratio was lower following a 2,3-butanediol dose of 600 mg/kg/d, it was not significantly different from that of untreated mice (Figure 3F).

Having established that a relatively short-term (4-week) treatment with 2,3-butanediol prevented an age-related increase in plasma cholesterol and decreased the EWAT:body weight ratio, we investigated whether these effects could be maintained or enhanced over a longer-term treatment and whether they could be sustained after withdrawal of 2,3-butanediol. Eight-week-old male wild-type mice were treated with the same 4:1:1 mix of meso:R,R:S,S 2,3-butanediol isomers at a dose of 200 mg/kg/d. At 22 weeks of age, that is, after 14 weeks of treatment, 2,3-butanediol was withdrawn from the drinking water of half the animals, which were maintained for a further 6 weeks on water with no added 2,3-butanediol. This cohort was termed “washout”. Analysis of urine of the washout cohort, collected 1 week after withdrawal of treatment, confirmed the absence of 2,3-butanediol. The remaining animals continued to be treated with 2,3-butanediol for a further 6 weeks. This cohort was termed “throughout”. Both cohorts were aged 28 weeks at the end of the experiment.

At 28 weeks, the plasma total cholesterol concentrations of the throughout (3.07 ± 0.18 mmol/l) and washout (2.67 ± 0.09 mmol/l) cohorts were significantly lower than that of age-matched untreated mice (3.78 ± 0.09 mmol/l) (Figure 4A) and were similar to those of age-matched Fmo5 ^−/− mice (2.88 ± 0.12 mmol/l) (Gonzalez Malagon et al., 2015). The plasma cholesterol concentration of the throughout cohort was similar to that of eight-week-old mice at the start of the experiment (3.22 ± 0.06 mmol/l); however, that of the washout cohort was lower (Figure 4A). In contrast, the plasma cholesterol concentration of 28-week-old untreated mice was significantly higher than that of eight-week-old mice (Figure 4A).

At the end of the treatment regime, the body weight and weight of EWAT was determined for each of the cohorts. The EWAT:body weight ratios of both the throughout (0.023 ± 0.003) and washout (0.018 ± 0.003) cohorts were significantly less than that of age-matched, untreated mice (0.036 ± 0.002) (Figure 4B) and were similar to that of age-matched Fmo5 ^−/− mice (Gonzalez Malagon et al., 2015). Therefore, as with short-term treatment, long-term treatment with 2,3-butanediol prevented an age-related increase in plasma cholesterol concentration and decreased the EWAT:body weight ratio, and both effects were sustained after withdrawal of 2,3-butanediol, i.e., in the throughout and washout treatment regimes. However, in contrast to short-term treatment with 2,3-butanediol, long-term treatment had no effect on plasma concentrations of NEFA or triglygerides.

During the course of the treatment, from eight- to 28-weeks of age, the body weights of both the throughout and washout male cohorts were similar to those of age-matched untreated animals (Figure 4C), as was their food intake (data not shown). This is in contrast to Fmo5 ^−/− mice, which exhibit a reduction in weight gain from about 20-weeks of age, and by 30-weeks of age weigh ∼10% less than wild-type mice, despite greater intake of food (Gonzalez Malagon et al., 2015). As was the case with short-term treatment, long-term treatment with 2,3-butanediol had no effect on the plasma concentrations of glucose or insulin.

In treated animals, the plasma concentrations of alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, creatine kinase and lactate dehydrogenase were within the normal range for these enzymes (Hough et al., 2002), indicating that long-term treatment of mice with 2,3-butanediol had no overt deleterious effects on organ or tissue function.

By 20 weeks of age, in comparison with wild-type mice, male Fmo5 ^−/− mice have a lower plasma cholesterol concentration and begin to exhibit a reduction in body weight gain (Gonzalez Malagon et al., 2015). The composition of the gut microbiota of Fmo5 ^−/− mice differs from that of wild-type mice and the differences become more pronounced as the mice age (Scott et al., 2017). To explore the possibility that gut bacteria present in Fmo5 ^−/− mice could affect plasma cholesterol concentration or body weight gain, we carried out fecal transplants from Fmo5 ^−/− mice to wild-type mice.

No differences were observed in the plasma concentrations of cholesterol, HDL cholesterol, triglycerides, NEFA, glucose, ketone bodies or insulin between the transplanted mice and control animals (Supplementary Table S4). Analysis of urine of the transplanted mice detected no 2,3-butanediol. After fecal transplantation from Fmo5 ^−/− mice, wild-type mice surprisingly gained significantly more weight than age-matched non-transplanted mice (Figure 5A). The increased gain in body weight of the transplanted mice was accompanied by an increase in EWAT:body weight ratio, but this was not statistically significant (Figure 5B).

The results of the fecal transplant suggested that bacteria present in the large intestine of Fmo5 ^−/− mice do not affect plasma cholesterol concentration, but, unexpectedly, appear to be involved in promoting weight gain. This raised the question of the location of bacteria that produce 2,3-butanediol, and the possibility that 2,3-butanediol is produced by bacteria higher up the digestive tract, in either the stomach or small intestine.

Figure 6 shows the composition of the microbiota in the stomach of wild-type and Fmo5 ^−/− mice. In wild-type mice, the stomach microbiota was dominated by the phylum Firmicutes (77%), with Bacteroidetes representing only 2%, whereas in Fmo5 ^−/− mice, the phyla Firmicutes and Bacteroidetes were present in almost equal abundance (Firmicutes 44%, Bacteroidetes 40%), and there was a higher proportion of Proteobacteria in Fmo5 ^−/− mice (8%) than in wild-type mice (2%) (Figure 6A). At the genus level, Allobaculum (phylum, Firmicutes) dominates in wild-type mice (43%), whereas in Fmo5 ^−/− mice the dominant genus was Lactobacillus (phylum, Firmicutes), at 21% (Figure 6B). The genus Lactobacillus is known to produce 2,3-butanediol (Ledingham and Neish, 1954) and NMR analysis detected signals close to those expected for both isomers of the metabolite at low levels in the stomach contents of Fmo5 ^−/− mice (Supplementary Figure S3), but not in the contents of duodenum, jejunum, ileum, cecum or colon.