BCM3D 2.0: accurate segmentation of single bacterial cells in dense biofilms using computationally generated intermediate image representations

Published in Microbiology
Like

Share this post

Choose a social network to share with, or copy the shortened URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

Figure 1. (a) Experimentally acquired 3D fluorescence time-lapse images of a growing S. oneidensis biofilm with overlaid single-cell segmentation contours (single 2D slices shown for clarity).  Images were acquired every five minutes for five hours. Corresponding cells in different frames are displayed in the same color. (b) An example of 3D tracking and lineage tracing for simulated biofilm images. For clarity, spatial trajectories and lineages originating from only a single ancestor cell is displayed. The estimated graph is shown in blue and the corresponding ground truth graph is shown in red. (c) Evolution of cell volumes over time for a single cell in the biofilm. Cell division events are indicated with blue crosses. The single-cell growth rate was measured by calculating the slope of the trajectory segment between consecutive cell division events.

What is the reported novelty? 

Bacterial Cell Morphometry 3D 2.0 (BCM3D 2.0) is a machine-learning based image analysis workflow that addresses the challenge of segmenting the 3D shapes of single bacterial cells in 3D biofilm images. BCM3D 2.0 substantially improves 3D bacterial cell segmentation accuracy over the previous state-of-the-art.  More importantly, BCM3D 2.0 performs particularly well for datasets with low signal-to-background ratios and high cell densities, which are often encountered in high-resolution live-cell biofilm imaging.

Instead of training convolutional neural networks (CNNs) to perform voxel classification (as in BCM3D 1.0, Non-invasive single-cell morphometry in living bacterial biofilms, M. Zhang, et al., Nature communications 2020), we trained CNNs to translate 3D fluorescence images into intermediate 3D image representations that are, when combined appropriately later, more amenable to conventional mathematical image processing. The convolutional neural networks of BCM3D 2.0 are trained using simulated data. As such, BCM3D 2.0 can be readily retrained and applied to handle different bacterial cell shapes and high-resolution confocal and light sheet-based fluorescence imaging approaches.

When the segmentation results provided by BCM3D 2.0 are used as the input to a nearest neighbor tracking algorithm, simultaneous multi-cell tracking in 3D biofilms becomes feasible. We found however that accurate multi-cell tracking in 3D time-lapse movies is possible with a nearest neighbor tracking algorithm, only if the relative cell movement between consecutive frames is small. Depending on the type of biofilm and the bacterial species, small relative cell movement can be achieved using moderate time resolutions of 1-5 minutes. If larger cell movements occur, accurate single-cell observables, such as growth rates and cell division times, can still be extracted based on manual tracking establishing that this information is in fact contained in 3D movies with low to moderate time resolution.

Why is this new capability needed? 

BCM3D 2.0 opens the door to tracking the behaviors of single cells in bacterial biofilms in 3D space and time. Bacterial biofilms have long been recognized to feature heterogeneous microcompartments that influence the physiology of individual cells. However, quantifying time-dependent phenomena in 3D bacterial biofilms with single-cell resolution has not been possible. A key experimental bottleneck is accurate segmentation (3D shape delineation) and tracking of single cells in crowded biofilms over extended time periods. BCM3D 2.0 improves bacterial cell segmentation substantially, so that reliable single-cell measurements can now be made. Here, we demonstrate the capability to measure single cell growth rates (quantified as changes of cell volume over time) and single cell division times inside 3D biofilms.

Who will benefit from the new capabilities? 

We anticipate that the cell segmentation capabilities of BCM3D 2.0 will prove useful for any research efforts focusing on resolving cell-cell interactions, biochemical signaling between individual cells, and activation of gene regulatory networks by individual cells in 3D microbial communities. BCM3D 2.0 is publicly available on GitHub (https://github.com/GahlmannLab/BCM3D-2.0) and through the Open Science Framework (https://osf.io/m4637/). The BCM3D 2.0 training strategy can be implemented on any workstation computer utilizing GPUs. We aim to develop BCM3D into an integrated, versatile, user-friendly platform for bacterial single-cell morphometry and single-cell tracking in complex 3D biofilm and tissue environments. User feedback is greatly appreciated.

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Microbiology
Life Sciences > Biological Sciences > Microbiology

Related Collections

With collections, you can get published faster and increase your visibility.

Live biotherapeutics and medicinal microbiome products

This Collection welcomes contributions on recent developments in the utilization of live biotherapeutics and medicinal microbiome products for the treatment or prevention of diseases. In many cases, suitable drug treatments are still lacking due to the ineffectiveness of existing standards or the presence of long-term side effects. We are particularly interested in difficult-to-treat non-communicable diseases whose pathogenesis is associated with microbiota.

Publishing Model: Open Access

Deadline: Aug 02, 2024