Harnessing droplet microfluidics and morphology-based deep learning for the label-free study of polymicrobial-phage interactions

Abstract

Evaluating the impact of bacteriophages on bacterial communities is required to assess the future utility of phage therapy. Methods able to study bacterial polycultures in the presence of phages are useful to mimic evolutionary pressures found in natural environments and recapitulate complex ecological contexts. Bacteriophages can drive rapid genetic and phenotypic changes in host cells. However, the presence of other bacteria can also impact bacterial densities and community structure, and classical methods remain lengthy and resource intensive. Here, we introduce a microdroplet-based encapsulation method in which bacterial co-cultures are imaged using Z-stack brightfield microscopy. The method relies on automated droplet imaging using a novel AI-based autofocus function, coupled with morphology-based deep learning models for accurate identification of two morphologically distinct bacterial species. We show that we can monitor the relative growth dynamics of P. aeruginosa and S. aureus growing in 11 picolitre droplets for up to 24 hours. We demonstrate quantification of growth rates, bacterial densities and lysis dynamics of the two species without the need for plating. We show that a potent lytic phage of P. aeruginosa can either fully lyse the initial P. aeruginosa population or keep its density low long-term when in the presence of S. aureus.