A gut pathobiont regulates circulating glycine and host metabolism in a twin study comparing vegan and omnivorous diets

Abstract

Metabolic diseases including type 2 diabetes and obesity pose a significant global health burden. Plant-based diets, including vegan diets, are linked to favorable metabolic outcomes, yet the underlying mechanisms remain unclear. In a randomized trial involving 21 pairs of identical twins, we investigated the effects of vegan and omnivorous diets on the host metabolome, immune system, and gut microbiome. Vegan diets induced significant shifts in serum and stool metabolomes, cytokine profiles, and gut microbial composition. Despite lower dietary glycine intake, vegan diet subjects exhibited elevated serum glycine levels linked to reduced abundance of the gut pathobiont Bilophila wadsworthia. Functional studies demonstrated that B. wadsworthia metabolizes glycine via the glycine reductase pathway and modulates host glycine availability. Removing B. wadsworthia from a complex microbiota in mice elevated glycine levels and improved metabolic markers. These findings reveal a previously underappreciated mechanism by which diet regulates host metabolic status via the gut microbiota.

Figure 1.. Overview of Twins Nutrition Study and implementation of vegan and omnivorous diets

(A) Trial timeline (top) and sample collection schema (bottom). Each subject was tracked for 8 weeks. Dietary recall data, blood and stool were collected at Weeks 0, 4 and 8. (B) A CONSORT Flow Diagram of the individuals recruited and screened for this study. (C) Principal coordinate analysis performed on NDSR dietary recall data at baseline (Week 0; left panel) before participants switched to their study-prescribed diet and at study end (Week 8; right panel). Dietary recall data consisted of 165 elements that described macro- and micronutrient composition of diet. Treatment group explained a significant portion of variation at study end (P < 0.001, Adonis test) but not at baseline (P = 0.99, Adonis test). (D) Ten most important variables used to predict treatment groups in a random forest model trained on NDSR dietary recall data from Week 8 diet logs. The model had a classification accuracy of 96.8% in predicting treatment groups from dietary data.

Figure 2.. Vegan twins have distinct metabolomes compared to omnivorous twins

(A) Volcano plot of untargeted serum metabolomics showing the metabolites enriched in vegans (blue) and omnivores (red). P-values were false-discovery rate corrected with the Benjamini-Hochberg method. Horizontal dashed line represents a corrected p-value threshold of significance of 0.05. (CMPF = 3-Carboxy-4-methyl-5-propyl-2-furanpropanoic acid). (B) Volcano plot of untargeted stool metabolomics showing the metabolites enriched in vegans (blue) and omnivores (red). P-values were false-discovery rate corrected with the Benjamini-Hochberg method. Horizontal dashed line represents a corrected p-value threshold of significance of 0.05. (C) Violin plots of normalized ion intensities (arbitrary units) of a subset of serum metabolites that were significantly enriched in the vegan (blue) or omnivorous diets (red). Trend lines show linear regression fit for each diet along with the standard errors in gray. (D) Principal coordinate analysis of metabolite abundances involved in histidine metabolism in the stool metabolomics data at Week 8 (top) and of metabolite abundances of metabolites annotated as glycerophospholipids in the serum metabolomics data at Week 8 (bottom). Points are colored according to the treatment group (vegan diet = blue, omnivore diet = red). Ellipses represent 95% confidence intervals. (E) Violin plots of Euclidean distance between serum metabolomics profiles for intra-subject distance over time, between-twin distance at baseline, between-twin distance at study end and inter-subject differences at baseline. Intra-subject distances are not significantly different from between-twin distances at baseline (P = 0.21, Wilcoxon rank-sum test) but they are significantly greater than between-twin distances at study end (P = 0.00098, Wilcoxon rank-sum test). Between-twin distances at Week 0 are significantly greater than between-twin distances at Week 8 (P = 0.0038, Wilcoxon rank-sum test). Inter-subject distances are significantly greater than between-twin distances at Week 8 (P = 0.008, Wilcoxon rank-sum test). (F) Violin plots of Euclidean distance between stool metabolomics profiles for intra-subject distance over time, between-twin distance at baseline, between-twin distance at Week 8 and inter-subject differences at baseline. Intra-subject distances are not significantly different from between-twin distances at baseline (P = 0.51, Wilcoxon rank-sum test) but they are significantly greater than between-twin distances at study end (P = 2.4 × 10⁻⁵, Wilcoxon rank-sum test). Between-twin distances at Week 0 are significantly greater than between-twin distances at Week 8 (P = 0.00025, Wilcoxon rank-sum test). Inter-subject distances are significantly greater than between-twin distances at Week 8 (P < 2.2 × 10⁻¹⁶, Wilcoxon rank-sum test).

Figure 3.. Vegan twins have distinct metabolomes and immune profiles compared to omnivorous twins

(A) Violin plots of normalized expression values (arbitrary units) of the two serum cytokines that were significantly enriched in the vegan diet: stem cell factor (SCF, P_adjusted = 0.024, LME) and fractalkine (CX3CL1, P_adjusted = 0.035, LME). Trend lines show linear regression fit for each diet along with the standard errors in gray. (B) Bar plots showing the mean fold-change between twins of the fold-change over time from baseline to the end of the study. Blue bars are elevated in the vegan twins, magenta bars are elevated in the omnivorous twins. Error bars represent the standard error. Two cytokines were significantly elevated in the vegan group (red asterisks): FGF21 (P = 0.022, Student’s t-test) and CX3CL1 (P = 0.019, Student’s t-test). (C) Correlation network graph showing significant correlations between metabolites, cytokines, clinical parameters, and gut microbial species. Nodes are colored according to data type. Edges are colored according to direction and strength of correlation. P-values of all pairwise Spearman correlations were calculated and then an adjacency matrix was assembled using only statistically significant correlations after Benjamini-Hochberg p-value correction (P_adj. < 0.05). The backbone of this network was then extracted using the L-spar method to extract the most robust correlations. (D) Accuracy of leave-one-out cross-validation (LOOCV) of random forest models predicting treatment groups at baseline (orange bars) and study end (purple bars) for serum metabolomics, stool metabolomics and serum cytokines as well as dietary log data.

Figure 4.. Glycine is significantly elevated in circulation of vegan twins despite lower dietary consumption

(A) Violin plot showing levels of serum glycine (normalized ion intensity) over time in the vegan (blue) and omnivore (red) treatment groups (P_adjusted = 2.11 × 10⁻⁵, LME). Trend lines show linear regression fit for each diet along with the standard errors in gray. (B) Violin plot showing levels of glycine in diet (grams) over time in the vegan (blue) and omnivore (red) treatment groups (P = 3.05 × 10⁻⁵, LME). Trend lines show linear regression fit for each diet along with the standard errors in gray. (C) Scatter plot comparing percent dietary consumption (x-axis) and serum abundance (y-axis) between vegan and omnivore groups of the 17 proteinogenic amino acids that were measured in both the dietary log data and serum metabolomics data. All 17 amino acids had reduced consumption in vegans; circles represent non-essential amino acids, triangles represent essential amino acids. Amino acids colored in red had statistically significant differences in serum abundance between vegans and omnivores at Week 8 (glycine: P = 0.043, valine: P =0.043, Wilcoxon rank-sum tests; multiple testing p-value correction was performed using the Benjamini-Hochberg method). (D) Scatter plot comparing normalized ion intensity of glycine (x-axis; centered and scaled) with levels of fasting insulin (international units (iU) per milliliter (mL)). Linear regression model showed a statistically significant relationship between the two variables (P = 0.002, R² = 0.737).

Figure 5.. Decreased amino acid degradation in gut microbiota of vegan twins

(A) Volcano plot of KEGG orthologies significantly enriched in either vegan or omnivore diets. Genes are colored according to KEGG pathways indicated in legend. P-values were false-discovery rate corrected with the Benjamini-Hochberg method. Horizontal dashed line represents a corrected p-value threshold of significance of 0.05. (fpr = ferrodoxin-NADP+ reductase, kce = 3-keto-5-aminohexanoate cleavage enzyme, grdE = glycine reductase complex component B subunits alpha and beta). Reaction schematics below the plot demonstrate inputs and outputs of the lysine degradation and glycine reductase pathways. (B) Stacked bar chart of the relative abundance of gut microbial species that harbor the glycine reductase pathway over time and between treatment groups. (C) Multigene BLAST of glycine reductase pathway using B. wadsworthia as the search seed. (D) Scatter plots comparing the relative abundance of B. wadsworthia with normalized ion intensity of glycine(left) and relative abundance of B. wadsworthia with levels of fasting insulin (international units (iU) per milliliter (mL); right). Linear regression model showed statistically significant association between both comparisons (B. wadsworthia v. Glycine: P = 0.004; B. wadsworthia v. fasting insulin: P = 0.009).

Figure 6.. Bilophila wadsworthia consumes glycine via a glycine reductase pathway

(A) Growth curves of B. wadsworthia in liquid culture using a minimal, defined “freshwater salts” medium adding either formate, taurine, or both to the medium. Growths occurred over 48 hours. Experiments were performed in triplicate. Dots represent mean OD₆₀₀, vertical lines represent standard error. (OD₆₀₀ = optical density at 600nm wavelength). (B) Growth curves of B. wadsworthia in liquid culture using a minimal, defined “freshwater salts” medium adding varying concentrations of glycine to the medium. Growths occurred over 48 hours. Experiments were performed in triplicate. Dots represent mean OD₆₀₀, vertical lines represent standard error. (C) Growth curves of B. wadsworthia in liquid culture using a minimal, defined “freshwater salts” medium adding varying concentrations of pyruvate to the medium. Growths occurred over 48 hours. Experiments were performed in triplicate. Dots represent mean OD₆₀₀, vertical lines represent standard error. (D) Growth curves of B. wadsworthia in liquid culture using a minimal, defined “freshwater salts” medium adding glycine, pyruvate or both to the growth medium. Growths occurred over 48 hours. Experiments were performed in triplicate. Dots represent mean OD₆₀₀, vertical lines represent standard error. (E) Overview of proposed Stickland reaction between pyruvate and glycine. Red dot indicates carbon-13 used to trace stable isotopes in LC-MS experiment presented in panels F and D. (F) Stacked bar charts of concentrations of glycine (M+0), ¹³C-glycine (M+1), acetate (M+0) and ¹³C-acetate (M+1) over time in defined minimal media supplemented with pyruvate and ¹³C-glycine. Experiment performed in triplicate, error bars represent standard error of the mean. (G) Bar chart showing molar ratio of ¹³C-acetate accumulation to ¹³C-glycine depletion. Error bars represent standard error of the mean. (H) Bar charts showing relative expression fold changes of genes involved in GlyR pathway with and without glycine added to the culture media. Experiments performed in triplicate, error bars represent standard error of the mean. Statistical significance was assessed using a Student’s two-tailed t test (., P < 0.1; *, P < 0.05; **, P < 0.01)

Figure 7.. Metabolic impacts of removing B. wadsworthia from a complex, defined gut community

(A) Schematic of the experiment. Germ-free C57BL/6 mice were colonized with hCom2 (n = 9 mice) or hCom2ΔBw (n = 5 mice) and housed for 4 weeks before sacrifice. Fecal pellets were subjected to metagenomic analysis. Serum samples were subjected to metabolite profiling and a comprehensive metabolic panel. Liver samples were subjected to metabolite profiling and gene expression profiling with RT-qPCR. (B) B. wadsworthia is maintained at relative abundance according to metagenomics analysis in hCom2-colonized mice (left) while it is undetectable in hCom2ΔBw-colonized mice (right). Each point is an individual species, lines connect the same species over time; the collection of dots at each time point represents the community averaged over the mice in the group. The dots and connecting lines for B. wadsworthia (Bw) are colored red. (Inoc. = Inoculum community; N.D. = Not detectable). (C) Relative abundance comparison for each species between the hCom2 and hCom2ΔBw communities in the initial inoculum (left) and at Week 4 of the experiment (right). The dots for Bw are colored red. Pearson coefficient of correlation (ρ) between the two communities in the inoculum is 0.963 and at Week 4 is 0.958. Species with more than 10-fold difference are labeled (Ca = Collinsella aerofaciens, Bf = Bryantella formatexigens, Rt = Ruminococcus torques, Df = Dorea formicigenerans, Ba = Bacteroides sp.). (D) Volcano plot showing differentially enriched species between the hCom2 and hCom2ΔBw communities at Week 4. Horizontal dashed line represents threshold of statistical significance. Only Bw was statistically significant between the groups (q-value < 0.01, fold-change > 100). Negative log₁₀ fold changes indicate higher abundance in hCom2 community. P-values were adjusted with the Benjamini-Hochberg method to account for multiple hypothesis testing. (E) Quantification of glycine in serum (left) and liver (right) using LC-MS. Serum glycine was significantly higher in hCom2ΔBw-colonized mice (P = 0.048, Student’s t-test). (F) Body weight was significantly higher for hCom2-colonized mice after 4 weeks (P = 0.032, Student’s t-test). (G) LDL-cholesterol was significantly higher for hCom2-colonized mice after 4 weeks (P = 0.008, Student’s t-test). (H) Creatinine and alkaline phosphatase were significantly higher in hCom2-colonized mice after 4 weeks (creatinine: P = 0.0001, Student’s t-test; alkaline phosphatase: P = 0.043, Student’s t-test). (I) RT-qPCR-based profiling of liver tissue for genes involved in glutathione-dependent redox homeostasis. Gstt2 is expressed at significantly higher levels in hCom2ΔBw-colonized mice (P = 0.002, Student’s t-test). Gpx6 was expressed at higher levels in hCom2-colonized mice, although this difference did not meet the threshold of significance (P = 0.052, Student’s t-test).