Automated and Machine Learning Guided Culturomics of Personalized Gut Microbiomes

The human gut microbiome is made up of hundreds of strains, exhibiting extensive interpersonal diversity. Current approaches to build gut strain biobanks to isolate and study individual strains are labor-intensive, not scalable, and lack integration of strain phenotypes with genomic data. Here, we describe a high-throughput system that leverages phenotypic and genomic data for automated anaerobic strain isolation and cultivation. This system was used to establish an extensive and personalized collection of gut strains from 12 healthy individuals composed of 16,584 isolates and 653 high-quality whole genomes spanning 358 unique taxa, which in aggregate covered >80% of the human gut microbiome. We demonstrated that the morphology and spatial arrangement of colonies can predict taxonomic identity, species-specific growth profiles, and interspecies interactions. Comparative analysis of intra- and inter-personal strain-level genomic variations revealed patterns of gut microbiome evolution, selection, and horizontal gene transfer. Automated platforms that link taxa-level phenotypic data with genomic data will serve as a key resource to build personalized microbiome biobanks on-demand and help unlock the mechanistic basis of gut microbiota in health and disease.