Single-Cell Imaging in Bacterial Biofilms: Data Acquisition, Image Processing, and Quantitative Analysis

Wang, Yibo, Chemistry - Graduate School of Arts and Sciences, University of Virginia
Gahlmann, Andreas, AS-Chemistry (CHEM), University of Virginia

Bacterial biofilms are complex 3-dimentional (3D) structures with substantial spatial and temporal heterogeneity at the single-cell level. Simultaneous multi-cell tracking in 3D is thus critical for analyzing single-cell behaviors, such as motility and metabolism, as well as lineage tracing in biofilms. Due to phototoxicity and photobleaching concerns, fluorescence images are often subject to low signal-to-background ratios (SBRs). High cell density, and large relative cell movements from frame to frame add additional challenges for accurate segmentation and tracking of individual cells in living biofilms. To address these challenges, I used lattice light sheet microscopy (LLSM) to image 3D bacterial biofilms with high spatial and temporal resolution and without substantial light-induced degradation of the SBRs over time. I additionally incorporated a hermetically sealed microfluidic flow channel into the LLSM to sustain bacterial biofilm growth under precisely controllable physical and chemical conditions. To enable accurate cell segmentation in the acquired 3D movies, I trained convolutional neural networks (CNNs) to perform voxel classification and to translate 3D fluorescence images into 3D intermediate image representations that are more resistant to over- and under-segmentation errors. Using this approach, improved segmentation results are obtained even for low SBRs and/or high cell density biofilm images. In order to track individual cells, I further leveraged a separate machine learning algorithm to select cell features that facilitate linking corresponding cells between frames. We demonstrate the applications and limitations of our entire data processing pipeline by systematically evaluating tracking accuracy using both simulated and experimentally acquired time-lapse images. The combination of non-invasive imaging and machine-learning based computational image analysis pipeline provides new opportunities for investigating time-dependent phenomena in living bacterial biofilms with single-cell resolution.

PHD (Doctor of Philosophy)
Issued Date: