Smart microscopy methods for quantitative microbiology

Quantitative microbiology seeks to provide insights into bacterial behaviour, growth, and interaction. Widefield fluorescence microscopy, in particular, has revolutionised this field by enabling high-throughput, non-invasive, time-lapse observations of cellular processes at the single-cell level. However, despite its power, this method is not without limitations. Optical artefacts distort quantitative measurements, obscuring the accuracy and precision of critical parameters like cell morphology, gene expression, and single-molecule counts, necessitating advanced methods to mitigate their impact. A 'smart' microscopy experiment is designed to minimise errors and biases while yielding maximum biological insights. This involves combining knowledge of the biological question with the imaging system and advanced bioimage analysis techniques. Sources of error and bias in quantitative microbiology may arise from the imaging environment and the interaction between sample geometry and image formation. These errors are minimised through adjustments to the experimental design, and any remaining errors are addressed using advanced bioimage analysis techniques. This work introduces methods for applying smart microscopy to quantitative microbiology through the development of a virtual microscopy platform called SyMBac: Synthetic Micrographs of Bacteria. SyMBac combines a biophysical model of cell growth with an optical model of the microscope to generate realistic synthetic microscopy data under various conditions. It allows for tuning and toggling of optical artefacts, enabling direct comparisons between ground truth and observations. Key sources of error include diffraction, which is dictated by the microscope’s point spread function and blurs the image by spreading light from a point source, and projection, which integrates information from multiple sample depths into a single 2D image. These artefacts, while not making bacteria unresolvable, cause significant image blurring and analytical challenges. SyMBac simulations reveal how diffraction, projection, and noise compromise measurements of cell morphology, gene expression, and single molecule counts. These artefacts hinder accurate segmentation, distort cell shapes, and misallocate photons during fluorescence imaging, preventing reliable quantification. Diffraction and projection also limit the density at which single molecules can be individually resolved, affecting detection accuracy. By quantifying these errors, we propose mitigations such as experimental changes, correction factors, and deep learning techniques. The mother machine microfluidic device emerges as a robust tool for quantitative microbiology, and our ability to simulate images of it guides its design. Deep learning segmentation models trained on synthetic data demonstrated reduced bias, greater precision, and temporal coherence, enabling the precise study of bacterial growth dynamics. Using these models, and for the first time, we quantified bacterial width regulation during stationary phase transitions with enough precision to reveal a sizer-like behaviour in the width dimension alone. Finally, we applied ultra high-throughput imaging to study rare persistence events. New tools for processing and visualising large datasets, combined with SyMBac, increased throughput by two orders of magnitude, enabling the observation of extremely rare persister events under various conditions. This scale revealed that under ampicillin treatment, persisters are predominantly slow-growing rather than dormant. We also find that a significant number of viable but non-culturable cells show signs of growth but are likely induced by antibiotics, while some cells which have VBNC like properties are actually persisters and are able to re-grow after removal of the antibiotic. These results challenge the literature’s clean delineation between susceptible, persister, and viable but non-culturable cells. This work advances the field of quantitative microbiology by providing a framework to rigorously understand and mitigate optical artefacts through the SyMBac platform. By simulating the effects of diffraction, projection, and noise on bacterial imaging, and coupling these insights with experimental design optimisations and machine learning tools, we address long-standing challenges in the accurate interpretation of microscopy data. These contributions enable new levels of precision and accuracy in bacterial measurements from standard widefield experiments while also democratising high-quality data collection due to the ubiquity of widefield imaging. We hope that the methodologies presented not only improve experimental reproducibility but also set the stage for future innovations in bioimage analysis and quantitative microbiology with microscopy.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2025

Author dc:creator

• Hardo, Georgeos

Advisor dc:contributor.advisor

• Bakshi, Somenath

• Microbiology
• Quantitative Microbiology
• Bacterial Physiology
• Single-Cell Biology
• Biophysics
• Computational Biology
• Bioimage Analysis
• Systems Biology
• Microscopy
• Widefield Microscopy
• Fluorescence Microscopy
• Phase Contrast Microscopy
• Live-Cell Imaging
• Timelapse Microscopy
• Smart Microscopy
• High-Throughput Microscopy
• Ultra High-Throughput Microscopy
• Image Formation
• Image Synthesis
• Image Generation
• Image Simulation
• Virtual Microscopy
• Machine Learning
• Deep Learning
• Image Segmentation
• Microfluidics

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/eed4df0f-5ac7-41c8-9ca6-36e8449bc879/download
• http://purl.org/NET/rdflicense/allrightsreserved

Language dc:language

eng

Author Identifier

0000-0003-0037-1293

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/384086