Machine Learning-Based, Label-Free Motion Tracking of Leptospira spp. on Cultured Cells

Leptospira spp. swim in liquid and crawl on surfaces with two periplasmic flagella. The Leptospira motility is a crucial virulence factor, especially crawling motility, which is possibly associated with the host-dependent severity of leptospirosis. In vitro infection assay using monolayers of cultured host cells provides basic knowledge on host-pathogen interaction in leptospirosis; however, it is difficult to observe leptospires on the host cells due to their similarity in refractive index. Though fluorescent labeling could solve the problem in general, it could negatively affect the leptospiral physiology. Here, we introduce a label-free observation technique for Leptospira spp. using machine learning. The machine-learning-based "background subtraction" automatically detects bacteria moving over cultured animal cells, allowing researchers to analyze bacterial motion trajectories, velocity, and diffusivity. Assay without any labeling procedures, including genetic manipulations, removes the restriction of available serovars and strains of Leptospira.