A framework to assess antibiotic effects using time-lapse microscopy

Bacterial counts for the purposes of pharmacokinetic-pharmacodynamic (PKPD) modelling are typically described in colony-forming units (CFU). This method is limited due to the time and effort required to grow bacterial colonies. This method also counts colonies, which represent large numbers of bacteria, meaning it lacks information on the individual cell-level.

Bacteria have been shown to change shape when exposed to certain antibiotics, and shape changes such as filamentation have been linked to bacterial regrowth and increased endotoxin release, which has been associated with sepsis. These shape changes cannot be assessed with CFU counts. Time-lapse microscopy, in which images of bacteria are taken via microscope over a period of time, has the potential to solve these issues because it requires less time than CFU counts and has the ability to visualize large numbers of individual bacteria.

In this project, time-lapse microscopy is utilized to generate bacterial counts using an automated image analysis pipeline, and the morphology of each individual bacterium is identified using a machine learning model. Using this information, PKPD models are created to describe how antibiotic concentration causes bacteria to change shape and die over time. The initial dataset used in generating this pipeline was gram-negative rod-shaped bacteria exposed to beta-lactams, specifically E. coli exposed to ertapenem (EC+ETP) and P. aeruginosa exposed to meropenem (PA+MEM). Using this data, we were able to describe how healthy rod-shaped bacteria formed bulges or filaments, filaments formed bulged filaments, and bulged filaments and bulges lost their cell contents in relation to time and drug concentration. With the framework now in place, the project is being expanded to additional datasets involving different antibiotic/bacteria combinations.