Diet-Induced Alterations of Host Cholesterol Metabolism Are Likely To Affect the Gut Microbiota Composition in Hamsters

The fecal samples analyzed were obtained in a previous study that determined the effect of dietary PSE on hamsters' lipid metabolism (19). The handling of animals, feed composition, plant sterol composition of the diets, sample collection, and metabolic analysis were described in that report. Briefly, Bio-F1B male Syrian hamsters (Bio Breeders, Watertown, MA) were individually caged and randomly assigned to four dietary treatments throughout a 4-week period, of which three were included in the present study: a modified AIN-93 M diet containing no PSE (C) and diets containing 5% (wt/wt) plant sterols esterified with fatty acids from beef tallow (BT) or stearic acid (SA). The final energy distribution of each diet was 36% fat, 35% carbohydrate, and 29% protein. The animals were housed in a facility with controlled atmosphere (25°C) under 12-h light/dark cycles and had access to food and water ad libitum. The animals were euthanized by CO₂ asphyxiation after 4 weeks of dietary intervention. Blood was collected by cardiac puncture, and plasma was obtained by centrifugation (2,000 × g for 30 min at 4°C). Total and HDL cholesterol were enzymatically quantified in the plasma samples. Livers were excised and immediately frozen in liquid nitrogen. Total cholesterol, triglycerides, free cholesterol, esterified cholesterol, and phospholipids were measured in the livers. Cholesterol absorption was quantified in fecal samples collected at week 3 with radiolabeled sterols as previously described (29). The complete fecal output was collected during week 4 and stored frozen (−80°C). Fecal concentration of neutral sterols, bile acids, cholesterol, dihydrocholesterol, coprostan-3-one, and coprostan-3-ol was determined as previously described (19).

The gut microbial composition was determined in fecal samples of hamsters fed a control diet (C) (n = 7) or plant sterols esterified with SA (n = 9) and BT (n = 6). DNA was extracted using a standard method that combined enzymatic and mechanical cell lysis with phenol-chloroform extractions (15). The fecal microbial community was characterized by massive parallel sequencing of the V3 region of the 16S rRNA. PCRs were performed with the forward primer (A-338F) 5′-gcctccctcgcgccatcagACTCCTACGGGAGGCAGCAG-3′ and the reverse primer (B-518R) 5′-gccttgccagcccgctcagNNNNNNNNATTACCGCGGCTGCTGG-3′ (with the A and B adaptors indicated in lowercase and an 8-nucleotide barcode shown as Ns), and products were sequenced using the Roche Genome Sequencer GS-FLX (454 Life Sciences) as described previously (15). The sequence data set is available upon request.

The sequences obtained were subjected to quality control using the QIIME pipeline (30). Sequences <150 or >350 bp in length were removed, as were sequences containing one or more ambiguous nucleotides or mismatches to the primer or barcode, an average quality score below 25, and homopolymer runs longer than 6 bp. Chimera removal was performed using the BLAST Fragments algorithm in QIIME. An average of 1,700 sequences per sample was obtained after quality control. Taxonomical characterization of the sequences was done with the Classifier tool from the Ribosomal Database Project (RDP) (31), which classified the sequences from the phylum to the genus level. Additionally, operational taxonomic units (OTU) were determined using a 97% sequence similarity cutoff to characterize the microbiota at a lower taxonomic level roughly equivalent to bacterial species. OTUs were generated by aligning the quality-controlled sequences with the Infernal Alignment algorithm of RDP, followed by clustering with the Complete Linkage Clustering tool of RDP. The exact abundance of OTUs determined to be significantly affected by PSE consumption or associated with host physiological parameters was determined using BLASTn as described previously (32). Briefly, 5 representative sequences of the selected OTUs were taxonomically assigned and aligned by ClustalW within their respective phyla. Phylogenetic trees were constructed for each phylum with the neighbor-joining algorithm, and distance matrices were generated (MEGA 4.0) (33). Sequences that clustered in the tree with >97% similarity were combined into a single OTU, and consensus sequences were generated within these clusters. The consensus sequences were used to assign all sequences in the entire sequence set to respective OTUs by aligning them in BioEdit (34) with the BLASTn algorithm (>97% similarity and at least 95% overlap) against a local database composed of all of the quality-controlled sequences. UniFrac analysis was performed using the QIIME pipeline to investigate the beta diversity of microbial communities (30).

Representative bacterial strains originating from the mammalian gastrointestinal tract were selected to test for antibacterial activity of cholesterol, cholesteryl-linoleate, and the plant sterols β-sitosterol and stigmastanol when incorporated into micelles containing bile salts and lecithin. The bacteria and growth media used were Bifidobacterium longum subsp. infantis ATCC 15697^T (in MRS medium supplemented with 0.5 mg/liter l-cysteine), Lactobacillus reuteri Lpuph-1 (in MRS medium), Eggerthella lenta ATCC 25559^T (in MRS medium), Slackia heliotrinireducens ATCC 29202^T (in MRS medium), Collinsella intestinalis ATCC 13228^T (in peptone yeast glucose [PYG] medium), two human isolates of Collinsella aerofaciens, KD-D8-5 and IM-D3-18 (in PYG medium), and Clostridium histolyticum ATCC 19401^T (in PYG medium). All media were prereduced for 24 h, and bacterial inocula for inhibition tests were prepared as follows under anaerobic conditions at 37°C (Bactron IV anaerobic chamber; Shel Laboratory). Bacterial cultures started from one single colony were grown for 24 h, transferred to fresh media (1% inoculum), and then grown for another 16 h.

Micelles containing cholesterol (>98% purity; Sigma), cholesteryl-linoleate (>99% purity; Sigma), or plant sterols (stigmastanol and β-sitosterol, synthesized as described in reference 21) were prepared as described by Brown et al. (21), with minor modifications. Briefly, 48 μl of a solution containing either cholesterol, cholesteryl-linoleate, stigmastanol, or β-sitosterol (161 mg/ml in chloroform) was mixed with 41 μl of a lecithin solution (221.3 mg/ml in chloroform), and chloroform was removed under a stream of nitrogen. The resulting mixture of lecithin and cholesterol/sterols was dissolved in a solution containing 5.4 mg/ml of the bile salt sodium taurocholate (>97% purity; Sigma) in distilled water. The solutions were sonicated for 3 to 6 min with 30% amplitude using a Branson 450 Sonifier (Danbury, CT) and afterwards filter sterilized with 0.45-μm pore filters (Fisherbrand; Fisher Scientific). To test for the antibacterial capacity of cholesterol, cholesteryl-linoleate, or the plant sterols, 5 μl of a 1:10 dilution of the bacterial inocula was transferred into a solution made of 500 μl of prereduced micelle suspensions and 500 μl of the appropriate prereduced double-concentrated (2×) media. Given the antibacterial activity of bile acids, micelle suspensions containing bile acids but no sterols were inoculated and used as controls. Cultures were incubated anaerobically at 37°C for 12 h (mid-log phase), and spectrophotometric optical density (OD) quantifications were performed at 600 nm (BioMate3; Thermo Scientific). Three biological replicates were conducted for the experiments. The concentration of cholesterol in these experiments (386 μg/ml media) is approximately double the concentration of the average cholesterol levels measured in the fecal samples of hamsters consuming the control diet (170 ± 111 μg/g) but still around 20 times lower than the fecal cholesterol levels present in SA-treated hamsters (4,303 ± 1,930 μg/g). Unfortunately, generation of micelles containing higher amounts of cholesterol, which would have better reflected the in vivo conditions, was not possible due to the formation of a precipitate.

Results were expressed as means ± standard deviations (SD) unless otherwise stated. The impact of dietary treatments on the abundance of individual bacterial taxa was analyzed by one-way analyses of variance (ANOVA) and Tukey's post hoc tests. Correlations between bacterial groups and host physiological measurements were assessed with Pearson's correlations and nonlinear regressions. A nonlinear four-parameter sigmoidal inhibitory model that describes dose-response relationships between bacteria and inhibitory compounds was used to fit the data (27, 28). The model is represented by the equation Y = Y_o + (Y_max × X^slope)/(EC₅₀^slope + X^slope), where X represents the metabolic parameter, Y the abundance of the bacterial taxon in the fecal sample, Y_o and Y_max are the minimum and maximum effects, respectively, EC₅₀ is the concentration where 50% of the maximum effect is measured, and the slope is the sigmoidicity coefficient.

The statistical analyses were performed using GraphPad Prism version 5.0 (GraphPad Software). P < 0.05 in the ANOVAs and correlation coefficients of r > 0.60 (in absolute values) were considered significant. In addition to the analyses described above, partial correlations were performed with SAS/STAT software to evaluate the associations between metabolic markers and abundance of bacterial populations when excluding the effect of diet using linear regressions. To decrease the number of variables in this analysis and increase the power of the test, only a few bacterial taxa (Coriobacteriaceae, Erysipelotrichaceae, unclassified Erysipelotrichaceae, and OTU1) and metabolic markers (whole-body cholesterol synthesis, fecal cholesterol excretion, fecal neutral sterols, and fecal bile acid) that were determined to be pertinent based on the previous analyses were selected to conduct the partial correlation statistics.

454 Pyrosequencing was used to characterize the fecal microbiota of hamsters fed a control diet or a diet supplemented with 5% plant sterols esterified with stearic acid (SA) or beef tallow (BT). The inclusion of PSE in the diet, in particular SA, had extensive effects on the fecal microbiota composition. UniFrac analysis revealed that the fecal microbial communities of hamsters fed the SA diet clustered separately from those of the control group (Fig. 1A). Analysis of Shannon diversity coefficients revealed that dietary addition of SA reduced the diversity of the fecal microbiota compared to that of the control (P < 0.05) (Fig. 1B).

Dietary PSE significantly altered the gut microbiota at all taxonomic levels (Table 1). PSE had a dramatic effect on the abundance of the phylum Actinobacteria, and especially on the family Coriobacteriaceae, which declined 10- and 4-fold with SA or BT intake, respectively. Three OTUs belonging to this family were significantly reduced through PSE (Table 1; also see Fig. S1 in the supplemental material). SA induced a significant increase in the phylum Firmicutes which was, to a large degree, due to a bloom of a single OTU (OTU16) within the family Eubacteriaceae (Table 1). It is possible that this increase was due to the ability of this organism to utilize the additional cholesterol that became available as a growth substrate in the gut of PSE-treated hamsters. However, the presence of OTU16 is strictly linked to the absence of the OTUs whose abundance was reduced by PSE (data not shown), suggesting that the bloom was caused by OTU16 expanding into niches that became vacant. Although total Firmicutes levels showed an increase with SA, five OTUs belonging to the family Erysipelotrichaceae did not follow this general trend and instead showed a dramatic reduction (Table 1; also see Fig. S2).

The feeding of PSE induced extensive changes in the lipid and cholesterol metabolism of hamsters (19), and many of the individual metabolites were highly interrelated (Fig. 2A). One of the goals of the present study was to investigate whether the lipidemic effects were associated with shifts in the gut microbiota. The correlation analysis revealed that host cholesterol metabolism was extensively interlinked with the bacterial community (Fig. 2). The tightest associations were observed within the phylum Actinobacteria and especially the family Coriobacteriaceae, which showed remarkable associations with cholesterol absorption (r = 0.75, P < 0.0001), whole-body cholesterol synthesis (r = −0.75, P < 0.0001), fecal biliary cholesterol excretion (r = −0.75, P < 0.0001), liver free cholesterol (r = 0.73, P < 0.0001), plasma non-HDL cholesterol (r = 0.68, P = 0.0005), and liver weight (r = 0.82, P < 0.0001), among others. Similar high correlations were observed for the OTUs belonging to the Coriobacteriaceae and Erysipelotrichaceae families that were affected by PSE administration. Total Firmicutes and unclassified Erysipelotrichaceae also displayed significant associations with host lipidemic markers, although they were less significant than associations detected for Coriobacteriaceae and the individual OTUs (Fig. 2).

One mechanism through which gut bacteria could lower host cholesterol levels is the transformation of cholesterol to coprostan-3-one, coprostan-3-ol, and dihydrocholesterol, which are, to a large degree, excreted (29). However, fecal concentrations of these cholesterol derivatives were not affected by PSE intake, and their fecal concentrations did not correlate with any of the host lipid metabolism markers tested (Fig. 2A). In addition, the correlation analysis did not provide evidence for an association between coprostan-3-one, coprostan-3-ol, and dihydrocholesterol and specific bacterial members of the gut microbiota (Fig. 2). These findings suggest that PSE-induced shifts of the gut microbiota did not impact bacterial cholesterol metabolites in the gastrointestinal lumen, and the bacterial action on the cholesterol pool was not responsible for the improvements observed in host lipid metabolism.

Upon visual inspection of the correlation plots, the associations of bacterial taxa with host cholesterol synthesis and excretion appeared to follow an exponential relationship. In addition, these associations appeared to resemble dose-response curves observed in microbial antibiotic assays. We therefore tested how bacterial proportions and cholesterol synthesis/excretion fitted a four-parameter sigmoidal model for bacterial inhibition used to describe dose-response relationships between bacteria and inhibitory compounds (27, 28). This analysis revealed that the relationships between cholesterol synthesis, fecal biliary cholesterol excretion, and fecal cholesterol with several bacterial taxa showed an almost perfect fit to the inhibitory model (Fig. 3). For example, the associations between Coriobacteriaceae and host cholesterol synthesis, fecal biliary cholesterol excretion, and fecal cholesterol excretion fitted the model with regression coefficients of −0.93, −0.87, and −0.91, respectively. The model was an even better fit for the data obtained for OTU1, which is related to Eggerthella lenta. Interestingly, the fecal proportions of OTU13, which was classified as a member of the Erysipelotrichaceae family, also showed a tight association with cholesterol synthesis and excretion with an excellent fit to the model. In total, four OTUs, two belonging to the Coriobacteriaceae (OTUs 1 and 4) and two belonging to the Erysipelotrichaceae (OTUs 13 and 15), displayed a significant fit to the four-parameter sigmoidal model (r = −0.94, −0.71, −0.73, and −0.68, respectively). The fact that the abundance of taxonomically distinct bacterial taxa displayed associations with fecal cholesterol concentrations that fit an inhibition model suggests an antibacterial role of the excreted cholesterol. In contrast, bile acid concentrations showed no negative associations with the fecal proportions of Coriobacteriaceae- or Erysipelotrichaceae-related phylotypes that were affected by the feeding of PSE (see Fig. S3 in the supplemental material).

Partial correlation analysis revealed that the negative associations between the abundance of Erysipelotrichaceae with whole-body cholesterol synthesis and concentrations of fecal neutral sterols (P = 0.033 and 0.034, respectively) were independent of dietary PSE. This analysis suggested that the links between fecal proportions of Erysipelotrichaceae and Coriobacteriaceae with fecal cholesterol also are independent of PSE intake (P = 0.076 and 0.069, respectively). Importantly, the abundance of the family Coriobacteriaceae (P = 0.0726, r = −0.7122) and OTUs 1 (P = 0.0244, r = −0.82) and 13 (P = 0.0032, r = −0.92) also showed significant negative associations with fecal cholesterol excretion in hamsters on the control diet alone (see Fig. S4 in the supplemental material). These findings suggest that the associations detected between host cholesterol metabolism and the microbiota are not solely caused by PSE but also by a direct association of the host phenotype with the bacterial taxa.

To gain insight into the functional interactions between bacteria and cholesterol metabolism, we determined the capability of cholesterol, cholesteryl-linoleate, and the plant sterols stigmastanol and β-sitosterol to inhibit bacterial growth when included in model bile mixed micelles containing a bile salt and lecithin, which are natural components of micelles in the gut (35). We used plant sterols for these experiments as PSE get hydrolyzed in the gut, and the unesterified sterol is the compound that is incorporated into intestinal micelles (20). We tested for the inhibition of seven strains of gut bacteria, including strains belonging to the Coriobacteriaceae, as members of this bacterial family were most affected in our animal experiments. No statistically significant levels of inhibition were detected for any of the treatments. However, out of the seven strains tested, Lactobacillus reuteri, Clostridium histolyticum, and Collinsella aerofaciens showed a modest degree of inhibition. All three showed reduced growth in the presence of micelles containing cholesteryl-linoleate compared to control micelles containing only lecithin and bile acid (Fig. 4). Growth of Lactobacillus reuteri also appeared to be inhibited by micelles containing cholesterol.