Predicting bacterial-mediated entomopathogenicity through comparative genomics and statistical modeling

Bacterial genomes encode for vast functional diversity and have both beneficial and detrimental effects on insect hosts. However, connecting bacterial genotype to a host-associated phenotype can be experimentally time-consuming, particularly if insecticidal mechanisms are highly host-specific. To streamline the identification of virulence genes on a new host, we tested if merging existing mechanistic knowledge with in vivo tests on a small number of bacterial isolates could predict bacterial genes associated with entomopathogenesis. We used a model consisting of Drosophila melanogaster interactions with pathogenic and commensal genome-sequenced strains of Pseudomonas bacteria. We compiled a database of previously described insecticidal and biocontrol genes within the Pseudomonas genus and used comparative genomics to probe the distribution of these genes across Pseudomonas strains. We found natural variation in the presence of known insecticidal genes across the genus. We tested the insect-killing capacity of 13 Pseudomonas spp. strains against D. melanogaster and found natural variation in insecticidal activity. To identify bacterial genes associated with fly mortality, we employed two statistical models to correlate bacterial virulence with the presence of previously described insecticidal activity. To validate our predictions, we used a P. aeruginosa PAO1 transposon mutant library and identified 8 operons that are necessary for killing D. melanogaster . We show that by combining existing literature with phenotyping a small number of strains, we can identify genes necessary for insecticidal activity in a previously untested insect model. More broadly these findings illustrate a discovery pipeline for bacterial virulence mechanisms, accelerating the discovery of insect pest biocontrol mechanisms.