A three-country analysis of the gut microbiome indicates taxon associations with diet vary by taxon resolution and population

Abstract

Emerging research suggests that diet plays a vital role in shaping the composition and function of the gut microbiota. Although substantial efforts have been made to identify general patterns linking diet to the gut microbiome, much of this research has been concentrated on a small number of countries. Additionally, both diet and the gut microbiome have highly complex and individualized configurations, and there is growing evidence that tailoring diets to individual gut microbiota profiles may optimize the path toward improving or maintaining health and preventing disease. Using fecal metagenomic data from 1,177 individuals across three countries, we examine the relationship between diet and bacterial genera, focusing on Prevotella and Faecalibacterium, which have gained significant attention for their potential roles in human health and strong associations with dietary patterns. We find that these two genera in particular show significant associations with many aspects of diet but these associations vary in scale and direction, depending on the level of metagenomic resolution (i.e., genus level by reads and strain level by metagenome-assembled genomes) and the contextual population. These results highlight the growing importance of building metagenomic data sets that are standardized, comprehensive, and representative of diverse populations to increase our ability to tease apart the complex relationship between diet and the microbiome.

Importance: An analysis of fecal microbiome data from individuals in the United States, United Kingdom, and Mexico shows that associations with dietary components vary both by country and by level of resolution (i.e., genus and strain). Our work sheds light on why there may be conflicting reports regarding microbial associations with diet, disease, and health.

Keywords: Faecalibacterium; Prevotella; diet; human microbiome; metagenomics.

Fig 1

Study and data overview. (a) The table shows the number of samples analyzed by shotgun metagenomics from each country and the resulting sample size after filtering steps. (b) Data collected included taxonomic profiles and MAGs from the sequence data, and diet and lifestyle information from self-reported answers to questionnaires. (c) Radar plots show z-normalized values for key personal and diet-related variables, averaged by country. (d) A principal coordinates analysis plot shows robust Principal Component Analysis (PCA) distance among the microbiome samples, colored by the cohort country. Partial R2 from a Permutational Multivariate Analysis of Variance (PERMANOVA) is reported. A confusion matrix shows the classification accuracy using a random forest classifier (fivefold cross validation) on the taxonomic feature table. The mean Area Under the Curve (AUC) for the Receiver Operating Characteristic (ROC) curve is also reported.

Fig 2

Aspects of Prevotella and Faecalibacterium at different levels of resolution highlight the complexity of diet-microbe associations, each showing ties with many dietary variables, but varying in direction and strength by country. (a) Heatmap shows t-scores and P-values for testing the correlation between the nucleotide diversity in sGBs and dietary variables, pooled per individual by bacterial genus. (b) Bar plots show the number of significant correlations between each bacterial genus and dietary variables across all participants (purple), as well as within each country: US (blue), UK (orange), and Mexico (green). (c) An association map shows the dietary variables that were significantly correlated with the log ratio of each Prevotella sGB to the sum of all Prevotella (top) and each Faecalibacterium sGB to the sum of all Faecalibacterium (bottom) in the full data set and stratified by country, after accounting for the variation explained by covariates (cohort, BMI, antibiotic history, and level of well-being). Multiple comparisons were corrected using the Benjamini-Hochberg method with a 5% False Discovery Rate (FDR).