The gut microbiota modulates host energy and lipid metabolism in mice^[S]

Animals

Male GF Swiss Webster mice (aged 12–14 weeks) were maintained in flexible plastic film isolators under a strict 12-h-light cycle (lights on at 06:00 h). Sterility was routinely confirmed by culturing and PCR analysis from feces using universal primers amplifying the 16S rRNA gene. Age-matched male CONV-R Swiss Webster mice were transferred to identical isolators at weaning. Both groups of mice were fed an autoclaved chow diet (Labdiet, St. Louis, MO) ad libitum unless otherwise stated. To produce conventionalized (CONV-D) mice, we conventionalized 12-week-old GF mice with gut microbiota from Swiss Webster donor mice as previously described (16). The study protocols were approved by the University of Gothenburg Animal Studies Committee.

Blood was collected from the vena cava under deep isoflurane anesthesia after a 4 h fast, unless otherwise stated, and the mice were subsequently euthanized by cervical dislocation. The liver and epididymal white adipose tissues were immediately removed and snap frozen in liquid nitrogen.

Metabolomic analyses using GC coupled to time-of-flight MS platform

Serum samples (30 µl) were combined with 10 µl of an internal standard, labeled palmitic acid (16:0-16,16,16d₃; 500 mg/l), and 400 µl of methanol, vortexed for 2 min, and incubated for 30 min at room temperature. The supernatant was separated by centrifugation at 5,590 g for 5 min at room temperature. The sample was dried under constant flow of nitrogen. Twenty-five microliters of 2% methoxyamine hydrochloride in pyridine was added to the dried sample and incubated at 45°C for 1 h and then derivatized with 25 µl of N-methyl-N-(trimethylsilyl)-trifluoroacetamide by incubating at 45°C for 1 h. Five microliters of retention index standard mixture with five alkanes (400 mg/l) was added to the metabolite mixture. Sample order for analysis was established by randomization. The samples were analyzed on a Leco Pegasus 4D GC coupled to time-of-flight MS (GCxGC-ToF/MS) mass spectrometer with Agilent technologies 6890N GC and Combi PAL autosampler.

The metabolites were identified using an in-house reference compound library and by searching the reference mass spectral library. Mass spectra from the GCxGC-TOF/MS analysis were searched against the Palisade Complete Mass Spectral Library, 600K Edition (Palisade Mass Spectrometry, Ithaca, NY), which includes all spectra available from the NIST 2002 and Wiley registry collections and 150,000 other spectra. The matches to reference spectra are based on a weighted dot product of the two spectra, with higher m/z peaks having more weight than the lower. A similarity value is assigned between 0 and 999, with 999 being a perfect match and 750 generally considered as a reasonable match. We used the conservative cutoff criterion of 850 for identification.

Lipidomic analyses using ultra performance liquid chromatography/MS platform

Serum (10 µl) and liver samples (5–10 mg) were diluted with 0.9% NaCl (10 µl for serum, 50 µl for liver) and adipose tissue samples (5–10 mg) were diluted with 200 µl PBS buffer. All samples were spiked with an internal standard (10 µl for serum and liver, 20 µl for adipose) (24). The samples were subsequently extracted with chloroform-methanol (2:1) solvent (100 µl for serum, 200 µl for liver, 400 µl with 40 µl of PBS buffer for adipose), homogenized with a glass rod (serum) or a Retsch homogenizer (Mixer MILL type MM301) for 2 min at 25 Hz (liver) or 20 Hz (adipose) at 4°C by adding two zirconium oxide grinding balls, vortexed (1 min for serum, 2 min for liver and adipose), incubated at room temperature (1 h for serum and adipose, 30 min for liver), and centrifuged at 5,590 g for 3 min. From the separated lower phase, an aliquot (60 µl for serum, 100 µl for liver, 200 µl for adipose) was mixed with a labeled standard mixture (three stable isotope-labeled reference compounds; 10 μl for serum and liver, 20 µl for adipose) and 0.5–1.0 μl injection was used for LC/MS analysis. Sample order for analysis was established by randomization. Lipid extracts were analyzed on a Q-ToF Premier mass spectrometer (Waters) combined with an Acquity ultra performance liquid chromatography/MS (for specific settings, see Supplementary Methods).

Modeling

To include the correlation structure of lipidomics data into the analysis and therefore explore possible associations between different lipid molecular species, we applied the partial least squares discriminant analysis (PLS/DA) (25, 26) using the SIMPLS algorithm to calculate the model (27). PLS/DA is a common approach to multivariate metabolomics data analysis (28, 29). PLS/DA analysis maximizes the product of variance matrix of measured variables (e.g., lipidomic profile data) and correlation of measured data with properties of interest (e.g., CONV-R vs. GF mice). PLS/DA makes latent variables of original matrix X (predictor variable, e.g., lipidomics data) and matrix Y (response variable; e.g., mouse groups). Latent variables are formed as a linear combination of all the original variables in X in such a way that most of the association with Y variables can be explained together with the variation in X. Contiguous-blocks cross-validation method and Q² scores were used to develop the models (30). Q² indicates how accurately the data, either classed or nonclassed, can be predicted, and this term is more relevant to supervised pattern recognition processes. Q² scores over 0.08 indicate a model that is better than chance, whereas a score between 0.7 and 1.0 demonstrates a highly robust trend. For each model built, the loading scores and the variable importance on projection (VIP) parameters were examined, in conjunction with the original data, to identify which metabolites contributed most to clusterings or a trend observed in the data. Loading scores describe the correlation between the original variables and the new component variables, whereas VIP parameters are essentially a measure of the degree to which a particular variable explains the Y variance (class membership). PLS/DA analyses were performed using Matlab, version 7.5 (Mathworks, Natick, MA) and PLS Toolbox, version 4.2, of the Matlab package (Eigenvector Research, Wenatchee, WA).

Measurements of serum lipids

Total serum triglycerides and cholesterol were analyzed according to the manufacturer's protocols (Thermo Electron, Grenoble, France). The lipid distribution in plasma lipoprotein fractions was assessed by fast protein liquid chromatography gel filtration with a Superose 6 HR 10/30 column (Pharmacia, Uppsala, Sweden). Each fraction was subsequently analyzed for tri­glyceride or cholesterol content as above. Plasma lipoproteins were also separated by agarose gel electrophoresis (31) and the chylomicron staining was quantified by densitometry.

Hepatic VLDL production

To determine the VLDL production rate, lipolysis was blocked by injecting mice with 12.5 mg of Triton WR1339 (10% solution in saline) intravenously after an overnight fast. Blood samples were drawn from the tail vein at 0, 10, 30, 60, and 90 min after injection. Triglycerides were measured using an enzymatic color­imetric assay (Thermo Electron, Grenoble, France) according to the manufacturer's protocol.

Administration of lipid bolus

Mice were gavaged with 400 µl heavy whipping cream (36% fat, Arla, Sweden) after an overnight fast and euthanized 1 or 4 h after the lipid bolus. Serum was collected and triglycerides were measured as above.

Statistics

Student's t-test was applied to test for pairwise differences between the means, single factor ANOVA was applied to test for differences among means, and Dunn-Sidák multiple comparison procedure was applied to identify which means were significantly different, using the multcompare function of the MATLAB Statistical Toolbox. P values < 0.05 were considered as statistically significant. False Discovery Rate or the expected proportion of false discoveries among the rejected hypotheses was estimated using the method by Benjamini et al. (32). The adjusted P values (q-values) were calculated with the function “p.adjust” using the R statistical software (http://www.r-project.org/).

The gut microbiota affects the serum metabolome

We performed MS-based metabolic profiling of serum from CONV-R and GF Swiss Webster mice after a 4 h fast by using two-dimensional GC coupled to time-of-flight MS (GCxGC-ToF/MS) and identified 185 metabolites. Modeling and clustering by PLS/DA revealed that serum metabolite profiles clustered according to colonization status (Fig. 1A, corresponding VIP values are listed in supplementary Table I, and all 29 metabolites that are significantly altered in CONV-R compared with GF mice are listed in supplementary Table II).

As expected, we observed increased levels of the microbially derived metabolites and metabolites involved in xenobiotic metabolism in serum from CONV-R mice (Fig. 1B). 3-Hydroxyphenylpropionic acid, a product of catechin metabolism (33), and hydrocinnamic acid (or benzenepropanoic acid), which is produced by clostridium species (34), were elevated in CONV-R mice. In addition, we observed increases in rhamnose, a component of the outer cell membrane of acid-fast bacteria in the mycobacterium genus (35) (Fig. 1B). These findings demonstrate that the host serum metabolite profile reflects microbial metabolism in the intestine. Serum levels of glucuronic acid, which is associated with phase II (conjugation) metabolism of xenobiotic compounds (36), were also increased in CONV-R mice compared with CONV-R mice (Fig. 1B). This increase is consistent with increased ex­­pression of Ugt2b38 in the liver (R. Hezaveh, C. Reigstad, P. Gopalacharyulu, M. Oresic, F. Bäckhed, unpublished data).

The gut microbiota also promoted increases in the serum levels of pyruvic acid and the tricarboxylic acid metabolites citric acid, fumaric acid, and mails acid (all metabolites involved in energy metabolism) and reduced serum levels of urea and the urea cycle metabolite l-ornithine (Fig. 1B). Serum levels of several essential cellular building blocks, including sugars, amino acids, and fatty acids, were also modified. In particular, serum from CONV-R mice had decreased levels of long-chain fatty acids [the saturated fatty acid palmitic acid (16:0) and the unsaturated essential fatty acid linoleic acid (18:2n-6)] and cholesterol, and increased levels of the dietary phytosterol campesterol and glucose (Fig. 1B). We also identified increased serum levels of the monoamine ne­urotransmitters dopamine and tyramine and of trans- 2-aminomethylcyclopropanecarboxylic acid, a cyclopro­pane analog of γ-aminobutyric acid, in CONV-R mice (Fig. 1B).

The gut microbiota affects the serum lipidome

To investigate in further detail how the gut microbiota affects the serum lipidome, we performed high-resolution lipidomics of serum from CONV-R and GF mice after a 4 h fast using ultra performance liquid chromatography/MS. This technology allows several hundred lipid species to be simultaneously and accurately analyzed (24). We identified and quantified 333 lipids in serum from CONV-R and GF mice, and PLS/DA modeling and clustering revealed significant differences in the global serum lipid profiles between CONV-R and GF mice (Fig. 2A; corresponding VIP values are listed in supplementary Table III, and all significantly altered lipid species are listed in supplementary Table IV).

We found that CONV-R mice had elevated serum levels of two cholesteryl ester and three sphinghomyelin species (Fig. 2B, C; supplementary Table IV). A more dramatic effect of the gut microbiota was observed for phosphatidylcholines and triglycerides: serum levels of 18 phosphatidylcholine species (including three of the most abundant) were increased in CONV-R mice while three species were reduced; and serum levels of nine triglyceride species (including two of the five most abundant) were reduced (Fig. 2D, E; supplementary Table IV).

To investigate whether a shorter microbial colonization could induce similar changes in the serum lipidome, we colonized 12-week-old GF mice for 2 weeks (CONV-D) and analyzed their serum lipids on the same platform as above. We found that the reductions in triglyceride levels in CONV-D mice compared with GF counterparts were similar to those observed in CONV-R mice (supplementary Table IV). In contrast, CONV-D mice did not exhibit increased serum levels of cholesteryl esters, sphingomyelins, or phosphatidylcholines (supplementary Table IV). In fact, two of the three reduced phosphatidylcholines in CONV-R mice were also reduced in CONV-D mice (supplementary Table IV).

The gut microbiota affects the adipose lipidome

We also explored the effect of the gut microbiota on the adipose lipidome by analyzing epididymal adipose tissue using the same high-resolution lipidomics technique and mice as above. PLS/DA modeling and clustering revealed significant differences in the global adipose lipid profiles between CONV-R and GF mice after a 4 h fast (Fig. 3A; corresponding VIP values are listed in supplementary Table V and significantly altered lipid species are listed in supplementary Table VI).

In contrast to the differences observed in the serum lipidome, the only significant change in the phosphatidylcholine species was a decrease in one of the most abundant species in adipose tissue from CONV-R mice (Fig. 3B) and there were no alterations in the cholesteryl ester and sphingomyelin species (data not shown). Of all the lipid species measured in the adipose tissue, the gut microbiota had its greatest effect on the triglycerides: in further contrast to the serum lipidome, adipose levels of 13 triglyceride species (including one of the most abundant) were increased in CONV-R mice and only two were reduced (Fig. 3C; supplementary Table VI). As expected the increased triglyceride levels correlated with increased adipose mass and elevated levels of leptin (10.4 ± 1.9 vs. 22.0 ± 1.3 ng/ml, P = 0.003, n = 4/group) and insulin (1.3 ± 0.3 vs. 3.8 ± 0.9 ng/ml, P = 0.03, n = 4/group) in CONV-R mice compared with GF counterparts, which is agreement with a previous study (16).

The gut microbiota affects the liver lipidome

We next used the same high-resolution lipidomics technique and mice as above to analyze the effect of the gut microbiota on the liver lipidome. PLS/DA modeling and clustering revealed significant differences in the global liver lipid profiles between CONV-R and GF mice after a 4 h fast (Fig. 4A, corresponding VIP values are listed in supplementary Table VII and significantly altered lipid species are listed in supplementary Table SVIII).

We observed many more microbiota-related changes in the liver lipidome compared with the serum and adipose tissue (supplementary Table VIII). In contrast to the serum lipidome, we observed decreased levels of two of the three most abundant cholesteryl ester species in CONV-R mice (Fig. 4B). We also observed increases in four of the shorter phosphatidylcholines (≤36C) and decreases in the longer (≥38C) phosphatidylcholines in CONV-R mice (supplementary Table VIII). Furthermore, 95 triglyceride species (including four of the most abundant) increased in CONV-R mice and only two decreased (Fig. 4C; supplementary Table VIII).

In contrast to serum triglycerides, the liver triglycerides only trended toward being increased in CONV-D mice following a 2 week colonization of GF mice (supplementary Table VIII), suggesting that liver triglycerides responded slower than serum triglycerides to colonization.

Levels of chylomicrons but not VLDL are reduced in serum of CONV-R mice

Triglycerides in serum exist mainly packaged in the form of lipoproteins: chylomicrons, which transport dietary triglycerides from the intestine, and VLDL, which transport hepatic triglycerides from the liver to adipose tissue (37). We therefore investigated if a specific effect of the gut microbiota on serum chylomicron levels could explain why we observed reductions in triglyceride species in serum but increases in triglyceride species in adipose tissue and liver from CONV-R mice.

Agarose gel electrophoresis followed by lipid staining of the chylomicron region of agarose gels demonstrated 40% lower chylomicron levels in CONV-R mice compared with GF mice (Fig. 5A, B). In contrast, fast protein liquid chromatography followed by enzymatic measurements of tri­glyceride content in each fraction showed that there were no differences in VLDL-triglyceride levels between CONV-R and GF mice (Fig. 5C). By using Triton WR1339 to inhibit peripheral clearance of lipoproteins, mainly VLDL after an overnight fast, we also showed that the VLDL production rate was higher in CONV-R mice compared with GF mice (Fig. 5E, F).

Because our serum lipidomics data (Fig. 2B) indicated that CONV-R mice had elevated cholesteryl esters, we investigated whether the gut microbiota affected HDL and/or LDL cholesterol. Determining cholesterol levels in the fractions obtained in Fig. 2B suggested that CONV-R mice have slightly elevated LDL cholesterol, while HDL cholesterol was unaltered (Fig. 5D).

Triglyceride absorption from the gut is not reduced in CONV-R mice

The reduced levels of chylomicrons observed in CONV-R mice after a 4 h fast could be caused by decreased triglyceride absorption from the gut and/or increased triglyc­eride clearance. To directly test the first possibility, we administered a bolus of heavy whipping cream (36% tri­glycerides) intragastrically to CONV-R and GF mice after an overnight fast. We did not find any significant differences in serum triglycerides at baseline or after lipid bolus administration (Fig. 6A). However, analysis of agarose gels showed that chylomicron levels in serum from CONV-R mice were reduced 4 h after the lipid bolus, but not after 1 h (Fig. 6B). Together, these results are consistent with increased triglyceride clearance but not with decreased tri­glyceride absorption from the gut in the presence of gut microbiota.

In addition, we performed lipidomics at 1 and 4 h after the lipid bolus to investigate whether the gut microbiota affected the absorption/metabolism of specific lipid species under these conditions. We observed increases in two of the five most abundant phosphatidylcholines in CONV-R mice serum 1 h and 4 h after a lipid bolus and decreases in phosphatidylcholine (36:2) in CONV-R mice serum 1 h after a lipid bolus (supplementary Figs. IB and IIB). The levels of three of the five most abundant sphingomyelins in serum after 4 h and one sphingomyelin after 1 h were elevated in CONV-R mice, and sphingomyelins (d18:1/ 22:0) were reduced in CONV-R mice serum both 1 h and 4 h after a lipid bolus (supplementary Fig. IB and IIB). All significantly altered lipid species are listed in supplementary Tables IX and X. Our data thus suggest that the gut microbiota does not affect triglyceride absorption; however, other lipid species are modulated by a gut microbiota following lipid administration.