Summary
The degree to which diet reproducibly alters the human and mouse gut microbiota remains unclear. Here, we focus on the consumption of a high-fat diet (HFD), one of the most frequently studied dietary interventions in mice. We employed a subject-level meta-analysis framework for unbiased collection and analysis of publicly available 16S rRNA gene and metagenomic sequencing data from studies examining HFD in rodent models. In total, we re-analyzed 27 studies, 1101 samples, and 106 million reads mapping to 16S rRNA gene sequences. We report reproducible changes in gut microbial community structure both within and between studies, including a significant increase in the Firmicutes phylum and decrease in the Bacteroidetes phylum; however, reduced alpha diversity is not a consistent feature of HFD. Finer taxonomic analysis revealed that the strongest signal of HFD on microbiota species composition is Lactococcus spp., which we demonstrate is a common dietary contaminant through the molecular testing of dietary ingredients, culturing, microscopy, and germ-free mouse experiments. After in silico removal of Lactococcus spp., we employed machine learning to define a unique operational taxonomic unit (OTU)-based signature capable of predicting the dietary intake of mice and demonstrate that phylogenetic and gene-family transformations of this model are capable of accurately predicting human samples in controlled feeding settings (area under the receiver operator curve = 0.75 and 0.88 respectively). Together, these results demonstrate the utility of microbiome meta-analyses in identifying robust bacterial signals for mechanistic studies and creates a framework for the routine meta-analysis of microbiome studies in preclinical models.