Integration of system phenotypes in microbiome networks to identify candidate synthetic communities: a study of the grafted tomato rhizobiome

Abstract

Understanding factors influencing microbial interactions, and designing methods to identify key taxa, are complex challenges for achieving microbiome-based agriculture. Here we study how grafting and the choice of rootstock influence root-associated fungal communities in a grafted tomato system. We studied three tomato rootstocks (BHN589, RST-04-106 and Maxifort) grafted to a BHN589 scion and profiled the fungal communities in the endosphere and rhizosphere by sequencing the Internal Transcribed Spacer (ITS2). The data provided evidence for a rootstock effect (explaining ~2% of the total captured variation, p < 0.01) on the fungal community. Moreover, the most productive rootstock, Maxifort, supported greater fungal species richness than the other rootstocks or controls. We then constructed a phenotype-OTU network analysis (PhONA) using an integrated machine learning and network analysis approach based on sequence-based fungal Operational Taxonomic Units (OTUs) and associated tomato yield data. PhONA provides a graphical framework to select a testable and manageable number of OTUs to support microbiome-enhanced agriculture. We identified differentially abundant OTUs specific to each rootstock in both endosphere and rhizosphere compartments. Subsequent analyses using PhONA identified OTUs that were directly associated with tomato fruit yield, and others that were indirectly linked to yield through their links to these OTUs. Fungal OTUs that are directly or indirectly linked with tomato yield may represent candidates for synthetic communities to be explored in agricultural systems. IMPORTANCEThe realized benefits of microbiome analyses for plant health and disease management are often limited by the lack of methods to select manageable and testable synthetic microbiomes. We evaluated the composition and diversity of root-associated fungal communities from grafted tomatoes. We then constructed a phenotype-OTU network analysis (PhONA) using these linear and network models. By incorporating yield data in the network, PhONA identified OTUs that were directly predictive of tomato yield, and others that were indirectly linked to yield through their links to these OTUs. Follow-up functional studies of taxa associated with effective rootstocks, identified using approaches like PhONA, could support the design of synthetic fungal communities for microbiome-based crop production and disease management. The PhONA framework is flexible for incorporation of other phenotypic data and the underlying models can readily be generalized to accommodate other microbiome or other omics data.