Abstract
Plasmids are mobile genetic elements found across all domains of life. As plasmids often encode determinants of fitness, their evolution is intertwined with their hosts. However, naturally occurring plasmids remain far less understood than their hosts due to the lack of frameworks to recognize plasmids and to classify them into evolutionary groups. Here we trained a machine learning model that recognizes plasmids based on genetic architecture with state-of-the-art accuracy. We applied this model to a global collection of human gut metagenomes to identify 68,350 unique plasmids, 13,280 of which had a very high model confidence and represent more than an order of magnitude increase over the number of known plasmids that we detected in this environment. To understand the evolution of these plasmids, we developed a generalizable approach that enabled us to define 1,169 ‘plasmid systems’. Each system consists of plasmids that share a backbone sequence containing core plasmid functions, such as replication and conjugation, but vary in cargo genes that are often critical to the host, such as antibiotic resistance, amino acid biosynthesis, and tRNA modification. Members of the same system are often found in geographically distinct human populations, revealing cargo genes that likely respond to environmental selection. The ecological patterns of plasmids we observed could not be explained by microbial taxonomy. This work uncovers the tremendous diversity of plasmids and demonstrates the need to characterize them as a separate component of microbiomes distinct from their hosts.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Updated several sections for clarity and to add new results; Updated Figure 1; Updated Supplementary Figure S2; Added Supplementary Figures S12-S15; Updated Supplementary Tables S3 and S4; Added Supplementary Tables S9-S12
