Universal gene-level bimodality in natural microbial communities

Bimodality-the coexistence of two peaks in trait distributions-is common in natural ecosystems. In microbiomes, bimodality of species abundances is known. However, whether this pattern applies to community functionality remains unclear. Here, we systematically investigate the abundance distributions of individual genes in different microbiomes, from human gut to ocean, revealing widespread gene-level bimodality. The bimodal genes are enriched in niche-specific pathways, suggesting their roles in ecological adaptation of the community. Based on their abundances, we develop a framework for microbiome functional typing, offering a gene-centric alternative to the taxonomy-based paradigm. Applied to the human gut, our approach identifies eleven genes exhibiting robust bimodality across western countries. These genes are associated with diseases such as liver cirrhosis. Machine learning models leveraging these genes are predictive of these diseases, underscoring their potential as clinically relevant biomarkers. Our work provides critical insights for microbiome functional architecture and has implications for microbiome-based diagnostics.