Cholesterol Metabolism by Uncultured Human Gut Bacteria Influences Host Cholesterol Level

Abstract

The human microbiome encodes extensive metabolic capabilities, but our understanding of the mechanisms linking gut microbes to human metabolism remains limited. Here, we focus on the conversion of cholesterol to the poorly absorbed sterol coprostanol by the gut microbiota to develop a framework for the identification of functional enzymes and microbes. By integrating paired metagenomics and metabolomics data from existing cohorts with biochemical knowledge and experimentation, we predict and validate a group of microbial cholesterol dehydrogenases that contribute to coprostanol formation. These enzymes are encoded by ismA genes in a clade of uncultured microorganisms, which are prevalent in geographically diverse human cohorts. Individuals harboring coprostanol-forming microbes have significantly lower fecal cholesterol levels and lower serum total cholesterol with effects comparable to those attributed to variations in lipid homeostasis genes. Thus, cholesterol metabolism by these microbes may play important roles in reducing intestinal and serum cholesterol concentrations, directly impacting human health.

Keywords: Clostridium cluster IV; Framingham Heart Study; cholesterol; coprostanol; hydroxysteroid dehydrogenase; metabolomics; metagenomic species; metagenomics; microbial dark matter; microbiome.

Graphical abstract

Figure 1

Levels of Serum Cholesterol Are Important for Human Health and Can Be Modulated by a Variety of Factors, Including the Potential Metabolism of Cholesterol by the Gut Microbiota Intestinal cholesterol levels are influenced by both dietary and host-derived cholesterol. Intervention by changes in diet or use of statins both affect levels of intestinal cholesterol, while the use of ezetimibe blocks uptake of intestinal cholesterol. Gut microbial metabolism of cholesterol may also serve to reduce cholesterol absorption in the intestine, resulting in lower serum cholesterol levels. The proposed pathway for microbial conversion of cholesterol (1) to coprostanol (4) in the microbiota involves the intermediates cholestenone (2) and coprostanone (3).

Figure 2

Integrative Analysis of Metagenomes, Metabolomes, Isolate Genomes, and Enzymatic Functions Reveals Candidate Bacterial Genes Involved in Cholesterol Metabolism in the Human Gut Microbiome Human gut microbiome genes from a de novo assembled gene catalog, after additional clustering step into groups of homologous proteins (at least 50% aa identity), were correlated with coprostanol detection in paired metagenomic-metabolomic samples and further prioritized by incorporating information from relevant microorganisms and enzymes. (A) Scores of specificity and sensitivity in relation to presence of coprostanol were calculated for each cluster of homologous proteins, and their density is represented through hexagonal bin plot; 8.6% of protein clusters are found with greater than 50% specificity and sensitivity to coprostanol detection. (B) Proteins encoded by gut microbes of interest (implicated in coprostanol formation in the literature) were used to query the clusters of homologous proteins. Clusters containing proteins with >50% aa identity to proteins found within a specified organism were used to generate a smoothed trend line (see Figure S1E for the location of species matched clusters). According to the location of these trend lines, E. coprostanoligenes matching clusters are more specifically associated with coprostanol formation than clusters from other microbes. (C) Clusters of homologous proteins were queried with characterized enzymes known to either catalyze the oxidation of cholesterol to cholestenone: cholesterol oxidases (PF09129), AcmA from S. denitrificans, and Rv1106c from M. tuberculosis or enzymes that can perform very similar chemical transformations (HSDs: RUMGNA_00694, Elen_1325, Elen_0198, and KGH18088). USEARCH ublast (Edgar, 2010) analysis was performed with inclusive cutoffs (>25% aa identity and 50% coverage). (D) Combining the evidence from (A)–(C), 4 putative HSDs in E. coprostanoligenes were identified, 3 of which (ECOP170, ECOP726, and ECOP442) had high specificity with regard to the presence of coprostanol (>0.9), albeit with greatly varying sensitivity. All four enzymes were chosen for further biochemical validation. All panels based on dataset 1 analysis; see Figures S1A–S1D for dataset 2 analysis. See also Figure S1 and Table S1.

Figure 3

Uncharacterized 3β-Hydroxysteroid Dehydrogenase Enzymes from E. coprostanoligenes and Phylogenetically Related Human-Associated Bacteria Oxidize Cholesterol to Cholestenone (A and B) (A) A 3β-hydroxysteroid dehydrogenase (3β-HSD) enzyme from E. coprostanoligenes, ECOP170, converts cholesterol (1) to cholestenone (2) and (B) converts coprostanol (4) to coprostanone (3) in the presence of a mixture of 100 μM of NAD⁺ and 100 μM NADP⁺. (C) ECOP170 homologs from gut bacteria heterologously expressed in E. coli convert the 3β-OH groups (blue) of cholesterol and coprostanol (gray squares) to the corresponding ketones (2 and 3, respectively) but were not able to convert primary bile acids, which have 3α-OH groups (red), to the corresponding keto bile acids (white squares). We considered the detection of any of the desired products after overnight incubation in an assay condition to be metabolism. (D) A multiple sequence alignment of the 25 human-associated cholesterol dehydrogenase homologs and ECOP170 showing the conserved active site residues S138, Y151, and K155. Mutation of any of these residues to alanine in ECOP170 completely abolishes activity (red), whereas proteins with mutations in neighboring residues retain activity (green). Cholesterol dehydrogenases highlighted in blue are confirmed biochemically to oxidize cholesterol. See also Figures S2–S4 and Tables S2 and S3.

Figure 4

Cholesterol Dehydrogenase-Encoding Gut Bacteria Are Uncultured Members of Cluster IV Clostridium and Are Prevalent Across Geographically Diverse Human Populations (A) 20 different MSPs containing ismA genes could be identified in human gut microbiome datasets. Phylogenetic tree was generated using PhyloPhlAn and includes all IsmA-encoding MSPs as well as species in the direct neighborhood or marker species for Clostridium cluster IV and cluster XIVa. (B) Ex vivo conversion of cholesterol to coprostanol by human fecal samples. Coprostanol formation occurred in 4 of the 8 samples cultured in basal cholesterol medium, with all 4 metabolizing samples containing at least one of the IsmA-encoding species identified at day 3. (C) Proportion of microbiome samples within each respective cohort that contains at least one IsmA-encoding species. IsmA-encoding species msp_0205, msp_0421, msp_0238, and msp_0196 are the most abundant across all of the populations examined. Dotted lines show species whose IsmA proteins have been shown to metabolize cholesterol in vitro. (D) Relative abundance levels of IsmA-encoding species present in the gut microbiome when stratified by disease state (HMP2: total, n = 1,581; non-IBD, n = 411, avg rel. ab. = 1.047; UC, n = 437, avg rel. ab. = 0.8613; CD, n = 733, avg rel. ab. = 0.4864; non-IBD versus UC p = 0.72; non-IBD versus CD p = 0.009; UC versus CD p = 0.009; PRISM: total, n = 154; non-IBD, n = 34, avg rel. ab. = 1.51; UC, n = 52, avg rel. ab. = 1.437; CD, n = 68, avg rel. ab. = 0.525; non-IBD versus UC p = 0.14; non-IBD versus CD p = 1.33e−6; UC versus CD p = 6.30e−4). p values were determined by a Wald test for PRISM (a linear model) and a Satterthwaite's method for HMP2 (a mixed linear model with random effect for subjects) and corrected for multiple comparisons with Benjamini-Hochberg method. The center bar represents the mean and error bars representing 95% CIs. See also Figure S5 and Table S4.

Figure 5

Fecal Coprostanol Formation Is Correlated to the Presence of Cholesterol Dehydrogenases in Gut Microbiomes (A) Two independent human cohorts with paired fecal metagenomics and metabolomics were used to investigate the association between IsmA-encoding species and coprostanol formation. The presence of IsmA-encoding bacteria in the gut microbiome is highly correlated to the presence of fecal coprostanol (detected). Odds ratios for PRISM and HMP2 cohorts are 42.73 (95% CI: 11.28; 283.54) and 28.94 (95% CI: 13.64; 61.41), respectively. (B) Stool samples from patients with ismA+ species in their microbiotas have lower stool cholesterol (1), and higher cholestenone (2) and coprostanol (4) as determined by untargeted fecal metabolomics. Each point represents an independent sample with the center bar representing the mean and error bars representing SEM (PRISM: ismA+ species negative samples n = 99, positive samples n = 55, HMP2: ismA+ species negative samples n = 302, positive samples n = 169). Analysis was performed using linear (cholesterol and cholestenone) and logistic (coprostanol) regressions for PRISM and mixed effect linear (cholesterol and cholestenone) and logistic (coprostanol) models to account for repeated measures in HMP2, including the following as covariates in all models: age, gender, antibiotic usage (yes/no) and disease status (non-IBD, CD, or UC). See STAR Methods for details. (C) Molar ratios of cholesterol (gray), cholestenone (dark blue), and coprostanol (light blue) are shown for 11 coprostanol-positive stool samples from the PRISM cohort as measured by targeted metabolomics. Across the samples, molar ratios for coprostanol vary from 0.956 in sample 8,982 to 0.074 in sample 8,573. See also Figure S6 and Table S5.

Figure 6

Meta-Analysis Reveals an Association of Total Cholesterol Levels with Encoder Status in CVD Cohorts Three studies were included in random effects meta-analysis (highlighted in orange and annotated on y axis). The changes in HDL-C, LDL-C, and TC levels (mmol/L) between encoders and non-encoders are represented for each study by the center of the square (95% CI presented by respective horizontal black lines). The combined results of the meta-analysis are represented by blue diamonds with point estimate presented by vertical diamond points and dashed line, whereas respective 95% CI is presented by the horizontal line in the diamond's center. Solid black line represents the null effect. Studies on the right of this line (i.e., positive values on the x axis) have a higher level of serum lipids in encoders compared with non-encoders; studies on the left (i.e., negative values on the x axis) have lower levels of serum lipids in encoders than in non-encoders. In a meta-analysis of three studies, we observed 0.15-mmol/L lower level of TC in encoders than in non-encoders (95% CI: −0.27, −0.03). Meta-analysis results for HDL-C and LDL-C were not statistically significant. See also Table S6 for full results.