Predicting Bacteriophage Susceptibility from Escherichia coli genotype

The main limitation to the widespread use of bacteriophages (i.e. bacterial viruses) in health, agriculture and biotechnologies relies on the near impossibility to predict the outcome of a specific host-phage interaction. Indeed, mechanisms underlying this remarkable host specificity have been primarily studied in a handful of individual bacterium-phage systems, and an integrative theoretical framework accurately describing the relative influence of host and phage factors remains to be elucidated. Owing to its incredibly diverse pangenome, Escherichia coli displays a wide range of phenotypes and is one of key players of the gut microbiome, an environment where it is exposed to a broad range of indigenous phages known as coliphages. In addition, the unparalleled availability of biological and genomic resources for both E. coli strains and coliphages makes it the most appropriate model to attempt a systematic, genomics-informed, deciphering of phage susceptibility. The present project aims at developing a machine-learning-driven experimental workflow, which exploits the natural diversity of collections of E. coli and coliphages, high-throughput genetic screens and massive susceptibility assays to create a predictive genotype-to-phenotype model. A scalable methodology allowing massive spot assays of phage susceptibility has been developed and successfully applied to an initial set of 75 strains and 95 coliphages, yielding remarkably robust results for over 7,000 single bacterium-phage systems. Cross-comparison of these susceptibility patterns with a range of known important host and phage genetic factors determined through comparative genomics (e.g. phylogenetic relationships, predicted serotype, defense systems...) gave a first glimpse into their relative importance for successful phage infection.