Human gut microbiota subspecies carry implicit information for in-depth microbiome research

Microbial strains from same species can have distinct functional characteristics owing to their different gene content. As the highest resolution, strains are mainly host-specific, thus obscuring unbiased associations, and hindering deductive research. Here, we comprehensively define the human gut microbiota at consistently-annotated subspecies resolution in an unbiased, cohort-independent manner, and demonstrate that we can generalize across distinct populations worldwide while maintaining specificity and improving interstudy reproducibility. We developed panhashome, a sketching-based method for rapid subspecies quantification and identification of genes that drive the intraspecies variations, and showed that subspecies carry implicit information undetectable at species level. By meta-analysis of colorectal cancer (CRC) datasets, we identified disease-associated subspecies whose sibling subspecies or species are not. Subspecies-based machine-learning CRC diagnostic algorithm outperformed species-level methods by leveraging the unique subspecies-level information. This subspecies catalogue allows identification of genes that drive the functional differences between subspecies as fundamental step in mechanistically understanding microbiome-phenotype interactions.