Distinct microbial communities in the murine gut are revealed by taxonomy-independent phylogenetic random forests

Gut microbiome analysis using 16S rRNA frequently focuses on summary statistics (e.g. diversity) or single taxonomic scales (e.g. Operational Taxonomic units, OTUs). This approach risks misinterpreting the phylogenetic or abundance scales of community differences (e.g. over-emphasising the role of single strains). We therefore constructed a 16S phylogenetic tree from mouse stool and colonic mucus communities. Random forest models, of all 428,234 clades, tested community differences among niches (stool versus mucus), host ages (6 versus 18 weeks), genotypes (wildtype versus colitis prone- mdr1a -/- ) and social groups (co-housed siblings). Models discriminated all criteria except host genotype, where no community differences were found. Host social groups differed in abundant, low-level, taxa whereas intermediate phylogenetic and abundance scales distinguished ages and niches. Thus, treating evolutionary clades of microbes equivalently without reference to OTUs or taxonomy, clearly identifies whether and how gut microbial communities are distinct and provides a novel way to define functionally important bacteria.