Diet, Microbiome, and Inflammation Predictors of Fecal and Plasma Short-Chain Fatty Acids in Humans

Abstract

Background: Gut microbes produce short-chain fatty acids (SCFAs), which are associated with broad health benefits. However, it is not fully known how diet and/or the gut microbiome could be modulated to improve SCFA production.

Objectives: The objective of this study was to identify dietary, inflammatory, and/or microbiome predictors of SCFAs in a cohort of healthy adults.

Methods: SCFAs were measured in fecal and plasma samples from 359 healthy adults in the United States Department of Agriculture Nutritional Phenotyping Study. Habitual and recent diet was assessed using a Food Frequency Questionnaire and Automated Self-Administered 24-h Dietary Assesment Tool dietary recalls. Markers of systemic and gut inflammation were measured in fecal and plasma samples. The gut microbiome was assessed using shotgun metagenomics. Using statistics and machine learning, we determined how the abundance and composition of SCFAs varied with measures of diet, inflammation, and the gut microbiome.

Results: We show that fecal pH may be a good proxy for fecal SCFA abundance. A higher Healthy Eating Index for a habitual diet was associated with a compositional increase in fecal butyrate relative to acetate and propionate. SCFAs were associated with markers of subclinical gastrointestinal (GI) inflammation. Fecal SCFA abundance was inversely related to plasma lipopolysaccharide-binding protein. When we analyzed hierarchically organized diet and microbiome data with taxonomy-aware algorithms, we observed that diet and microbiome features were far more predictive of fecal SCFA abundances compared to plasma SCFA abundances. The top diet and microbiome predictors of fecal butyrate included potatoes and the thiamine biosynthesis pathway, respectively.

Conclusions: These results suggest that resistant starch in the form of potatoes and microbially produced thiamine provide a substrate and essential cofactor, respectively, for butyrate synthesis. Thiamine may be a rate-limiting nutrient for butyrate production in adults. Overall, these findings illustrate the complex biology underpinning SCFA production in the gut. This trial was registered at clinicaltrials.gov as NCT02367287.

Keywords: diet; gut microbiome; inflammation; machine learning; short-chain fatty acids.

FIGURE 1

An overview of SCFA abundances in a healthy United States cohort. (A) The abundances of fecal SCFAs (acetate, propionate, and butyrate) for 313 individuals. (B) The abundance of plasma SCFAs (acetate, propionate, and butyrate) for 315 individuals. (C) Partial correlation between each fecal SCFA and either BMI, age, or sex. (D) Pearson correlations between fecal and plasma SCFAs. (E) Partial correlation between fecal pH and total fecal SCFAs, adjusting for age, sex, BMI, stool weight, and stool consistency in the model. BMI, body mass index; SCFA, short-chain fatty acid.

FIGURE 2

Predicting SCFA abundances using machine learning. The mean MAE percent change between a trained machine model and a null model for fecal and plasma SCFAs using dietary, inflammation, and microbiome features. Solid horizontal lines represent the mean MAE percent change over the null model for all SCFAs of a given data type. When TaxaHFE was employed, solid lines indicate that TaxaHFE was trained on all samples prior to machine learning, and dashed lines indicate TaxaHFE was trained on only the training sample subset. ASA24, Automated Self-Administered 24-h Dietary Assessment Tool; MAE, mean absolute error; SCFA, short-chain fatty acid; TaxaHFE, Taxonomic Hierarchical Feature Engineering; HUMAnN3, HMP Unified Metabolic Analysis Network.

FIGURE 3

Top features for ML models predicting SCFA abundances from diet. SHAP beeswarm plots showing the top features (by mean absolute SHAP value) for fecal (A) acetate, (B) propionate, and (C) butyrate. BMI, body mass index; SCFA, short-chain fatty acids; SHAP, SHapley Additive explanations; ML, machine learning.

FIGURE 4

Investigating foods predictive of fecal butyrate. (A) A metacoder taxonomic plot illustrating the food taxa under L2 white potatoes and Puerto Rican starchy vegetables. The colors indicate mean abundance differences in consumption between individuals in the top and bottom tertile for the butyrate-ratio. (B) A stacked bar plot of the most abundant L3 children within the L2 white potatoes and Puerto Rican starchy vegetables. (C) A partial correlation between L2 white potatoes and Puerto Rican starchy vegetables with fecal butyrate.

FIGURE 5

Microbiome-rooted features are predictive of fecal butyrate. (A) Top directly varying (a positive correlation between feature value and SHAP value) TaxaHFE features from microbial taxonomy. These features increase in relative abundance with increasing butyrate-ratio. (B) Finer taxonomic details of the taxonomic groups are presented in (A). The color represents median log-fold differences in abundance between the top and bottom tertiles of butyrate-ratio (C) Top directly varying (a positive correlation between feature value and SHAP value) HUMANnN3 features with butyrate-ratio. (D) The top 9 species of bacteria containing the thiamine diphosphate biosynthesis pathway by mean abundance. SHAP, SHapley Additive explanations; HUMANnN3, HMP Unified Metabolic Analysis Network; TaxaHFE, Taxonomic Hierarchical Feature Engineering.