🦠 microbiology

Phage phenotyping by measuring plaque expansion dynamics

This study introduces a scalable, computer vision-based method that continuously tracks plaque expansion dynamics to reveal distinct kinetic phenotypes and taxonomic signatures of bacteriophages, offering a superior alternative to traditional static endpoint measurements for accelerating phage discovery and genotype-phenotype mapping.

Original authors: Costa, J., van der Steen, B. A., Mulla, Y.

Published 2026-10-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Costa, J., van der Steen, B. A., Mulla, Y.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Viruses that infect bacteria, known as bacteriophages, are the most abundant biological entities on Earth, numbering in the trillions and displaying a staggering variety of shapes and genetic codes. Scientists have long known that these tiny hunters are incredibly specific, often targeting only a single species of bacteria, and that they can kill their hosts at vastly different speeds. This speed matters deeply for medicine, particularly in the emerging field of phage therapy, where doctors use these viruses to treat bacterial infections that antibiotics can no longer cure. To choose the right virus for a patient, researchers need to know not just if a phage can infect a bacterium, but how quickly and effectively it can destroy it. For decades, the standard way to study this has been to look at the final result of an experiment: a clear, circular hole in a lawn of bacteria, called a plaque. However, measuring only the size of this hole at the end of the experiment is like judging a runner's speed only by looking at where they stopped, ignoring how fast they ran in the first half of the race versus the second. This static view misses the dynamic changes that happen as the bacteria grow and the virus spreads, leaving a gap in our understanding of how these microscopic predators actually work.

To fill this gap, a team of researchers in the Netherlands developed a new way to watch these viral battles unfold in real time. Instead of waiting until the end to measure the final size of the clear zones, they set up a system to take pictures of the bacteria and viruses every ten minutes for sixteen hours. They used a standard flatbed scanner, the kind found in many offices, to capture high-resolution images of the petri dishes as the viruses ate their way through the bacterial lawn. By combining this simple hardware with custom software that could automatically track the expanding edges of the clear spots, they turned a static snapshot into a moving movie of infection. This approach allowed them to measure exactly how fast the viruses were spreading at different stages of the bacterial growth cycle, capturing details that were previously invisible.

When the researchers applied this method to a large collection of sixty-nine different phages, all designed to infect the same type of E. coli bacteria, they discovered that every virus had its own unique rhythm. Some viruses started spreading slowly and then sped up, while others moved quickly at first and then slowed down. The team found that these patterns were not random; they were consistent for each specific type of virus. More importantly, the speed at which a virus began to spread was strongly linked to the size of its genetic material. Viruses with smaller genomes tended to expand faster than those with larger ones. This relationship held true across the board, suggesting that the physical size of the virus's genetic blueprint is a major factor in how quickly it can move through a bacterial population.

The study also revealed a surprising disconnect between how viruses behave in a liquid soup versus how they behave on a solid surface. In liquid cultures, where bacteria float freely, researchers measure success by seeing how long it takes for the entire population to collapse. The new time-lapse method showed that the speed of a virus spreading on a solid surface had almost no connection to how quickly it killed bacteria in a liquid. This means that the two environments trigger different biological responses, and relying on liquid tests alone might give a misleading picture of how a virus will perform in more complex, real-world scenarios like the human body or a biofilm. The researchers concluded that the way a virus expands on a solid surface offers a distinct and valuable window into its behavior that liquid tests simply cannot provide.

Perhaps the most striking finding was that the researchers could identify the family of a virus just by watching how it moved. Using a computer program trained to recognize patterns in the expansion speed and timing, they were able to correctly guess the family of a virus about seventy-two percent of the time. In contrast, if they only looked at the final size of the clear spot at the end of the experiment, their success rate dropped to below forty percent, which is worse than simply guessing the most common family every time. This proves that the story of how a virus spreads is far more informative than the final size of the hole it leaves behind. By capturing the full history of the infection, this new method provides a much richer, more accurate way to classify and understand these diverse viruses, offering a powerful tool for future discoveries in microbiology and medicine.

Technical Summary: Phage Phenotyping by Measuring Plaque Expansion Dynamics

Problem Statement
While metagenomic sequencing has revealed staggering genomic diversity among bacteriophages, phenotypic characterization remains a bottleneck. Current methods largely rely on binary infection outcomes (infect/non-infect) or liquid culture assays. However, many phages fail to produce measurable effects in liquid suspensions, and binary approaches miss critical variations in infection rates. Furthermore, the gold standard for quantitative phenotyping—the plaque assay—is typically limited to endpoint measurements (final plaque size). This static snapshot fails to disentangle phage performance across the sequential bacterial growth phases (lag, exponential, stationary) that occur during the assay, as plaque size is a composite metric of these changing states.

Methodology
The authors developed a cost-effective, scalable platform to continuously track plaque development on semi-solid media using consumer-grade hardware and open-source software.

Experimental Setup: A bacterial lawn of Escherichia coli MG1655 is established in a semi-solid agar layer within single-well plates. Phages from the BASEL collection are spotted at low concentrations to ensure single-plaque origins. Plates are incubated at 30°C.
Imaging: A consumer-grade flatbed scanner (Epson Perfection V600) captures 24-bit RGB images every 10 minutes for 16 hours. The system utilizes a custom software package built on the TWAIN interface and SiLA 2 standards for laboratory automation integration.
Image Processing: A custom Python package processes the time-series data. It involves manual annotation of plate perimeters and background correction areas, followed by pixel intensity normalization.
Edge Detection and Modeling:
- Radius Extraction: The algorithm identifies plaque centers and performs radial integration of pixel intensity. A Richards curve (sigmoidal function) is fitted to the radial intensity profile at each timepoint to determine the plaque edge (inflection point), reducing noise compared to direct circle detection.
- Phenomenological Model: The resulting radius-over-time profiles are fitted with a minimal phenomenological model consisting of two linear expansion phases connected by a logarithmic transition. This yields five key parameters: delay time, initial expansion rate, final expansion rate, transition time, and transition width.
Quality Control: Four mathematical filters are applied to exclude low-quality data, including checks for circle contrast, sigmoidal fit quality ( $R^2 > 0.95$ ), and overall model fit quality.
Comparative Analysis: The study compares these plaque dynamics against liquid culture collapse times and correlates expansion parameters with phage genome size and taxonomic family.

Key Contributions and Results

Dynamic Tracking: The method successfully captures plaque expansion dynamics across all bacterial growth phases, revealing that plaques exhibit a lag phase followed by linear expansion where the rate drops as bacterial growth declines.
Diversity of Kinetics: Applied to 69 phages in the BASEL collection, the study identified reproducible dynamics in 47 phages. The initial expansion rate was found to be consistent within a phage but distinct between different phages.
Decoupling from Liquid Assays: The study found no significant correlation between the initial plaque expansion rate and the bacterial collapse time in liquid culture ( $\rho=0.07$ ). This indicates that plaque expansion dynamics and liquid culture activity represent distinct physiological and kinetic processes.
Genome Size and Taxonomy: The initial expansion rate demonstrates a significant inverse correlation with viral genome size ( $\rho=-0.7$ ). However, this correlation is confounded by phage taxonomy, as genome size and expansion rates both vary significantly by family. When correcting for family, the correlation between genome size and expansion dynamics disappears.
Predictive Power: Using a Random Forest classifier, the authors demonstrated that phage taxonomic family can be predicted with 72.1% accuracy using only the five dynamic plaque expansion parameters. In contrast, conventional endpoint plaque sizes yielded only 40.4% accuracy, performing below the benchmark of guessing the most common family.

Significance and Claims
The paper claims that this time-lapse phenotyping approach yields robust kinetic data superior to end-point measurements. By moving beyond static snapshots, the method provides a scalable platform to accelerate phage discovery and genotype-phenotype mapping. The authors posit that these dynamic parameters offer a more detailed insight into phage phenotypes than current standard methods. Additionally, the open-source software and hardware integration (SiLA 2) are presented as tools to facilitate the integration of consumer-grade imaging into larger automated laboratory workflows. The authors suggest that while sequencing is cost-effective, the ability to classify phages based on dynamic phenotypes without sequencing highlights the power of this approach for optimizing phage discovery pipelines and potentially linking non-essential gene functions to specific kinetic phenotypes via explainable AI.

More like this