A New Pipeline for Designing Phage Cocktails Based on Phage-Bacteria Infection Networks

Abstract

In recent years, the spread of antibiotic-resistant bacteria and efforts to preserve food microbiota have induced renewed interest in phage therapy. Phage cocktails, instead of a single phage, are commonly used as antibacterial agents since the hosts are unlikely to become resistant to several phages simultaneously. While the spectrum of activity might increase with cocktail complexity, excessive phages could produce side effects, such as the horizontal transfer of genes that augment the fitness of host strains, dysbiosis or high manufacturing costs. Therefore, cocktail formulation represents a compromise between achieving substantial reduction in the bacterial loads and restricting its complexity. Despite the abovementioned points, the observed bacterial load reduction does not increase significantly with the size of phage cocktails, indicating the requirement for a systematic approach to their design. In this work, the information provided by host range matrices was analyzed after building phage-bacteria infection networks (PBINs). To this end, we conducted a meta-analysis of 35 host range matrices, including recently published studies and new datasets comprising Escherichia coli strains isolated during ripening of artisanal raw milk cheese and virulent coliphages from ewes’ feces. The nestedness temperature, which reflects the host range hierarchy of the phages, was determined from bipartite host range matrices using heuristic (Nestedness Temperature Calculator) and genetic (BinMatNest) algorithms. The latter optimizes matrix packing, leading to lower temperatures, i.e., it simplifies the identification of the phages with the broadest host range. The structure of infection networks suggests that generalist phages (and not specialist phages) tend to succeed in infecting less susceptible bacteria. A new metric (Φ), which considers some properties of the host range matrices (fill, temperature, and number of bacteria), is proposed as an estimator of phage cocktail size. To identify the best candidates, agglomerative hierarchical clustering using Ward’s method was implemented. Finally, a cocktail was formulated for the biocontrol of cheese-isolated E. coli, reducing bacterial counts by five orders of magnitude.