Science1 publisher2 min readPublished
A machine-learning tool measures bacterial invasion and DNA damage one human cell at a time
Researchers at Hungary's HUN-REN Biological Research Center built MALVINA, a machine-learning method that measures bacterial invasion, buildup and DNA damage in the same human cells. It separates infection patterns that averaged readouts make look identical, down to single cells.
The Scientist · Science desk
What happened
- MALVINA detects fluorescently labeled bacteria inside human cells and runs machine learning over the microscopy images, scoring how far each infection has gone cell by cell.
- Put against the same cells together, an invasive strain tied to inflammatory bowel disease helped an otherwise harmless lab strain get inside.
- The work was published in the journal Nature Communications.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- capability Measuring entry, buildup and DNA damage in one system lets a strain's behavior be tied to its known genome, which assays run apart cannot connect at the single-cell level.
- constraint The profiles come from human cells under a microscope, so a strain that invades a culture well is not thereby shown to cause disease in a patient.
- decision Virulence and drug testing can add invasion and DNA-damage readouts to the usual grow-or-not result, measured per cell rather than averaged across a well.
The standard question in antibiotic work is whether a bacterium grows while a drug is present: it grows or it does not, and the drug stops it or it does not [1]. That tells you nothing about what a surviving bacterium does to the cells around it.
MALVINA runs on microscopy images of fluorescently labeled bacteria sitting inside human cells, and it scores each cell separately [3]. The single-cell resolution matters. The same total bacterial count can mean many host cells each holding a few bacteria, or a few cells each packed with many, and an averaged measurement makes those two look alike [7]. MALVINA tells them apart.
"One of the strengths of the method is that we can see the details of infection cell by cell. We can distinguish whether a few bacteria enter many cells or whether large numbers accumulate in only a few cells, while also measuring the damage they cause," said Bence Bognar, a co-first author of the study [6].
On four E. coli strains the method drew distinct profiles [9]. A harmless laboratory strain, disease-associated strains that enter cells efficiently and strains making a DNA-damaging toxin each behaved differently: some invaded many cells, some built up in fewer, others caused more DNA damage [9]. A strain taken from a colorectal tumor sample invaded efficiently, accumulated once inside and induced DNA damage [10].
Bacteria rarely meet host cells alone, so the team put two strains in together. An invasive E. coli linked to inflammatory bowel disease helped an otherwise harmless lab strain get into host cells [12]. In another pairing, a genotoxic strain reduced the ability of a non-genotoxic but invasive one [13].
The measurements come from human cells under a microscope, not from patients [3]. A strain that invades a culture efficiently has not been shown here to cause disease in a person, and the panel is four E. coli strains plus a tumor isolate. The method quantifies behavior; whether that behavior predicts an outcome is a separate study.
When a strain's genome is known, MALVINA can tie its measured behavior to its genetic background, which the researchers describe as connecting genotype with virulence phenotype [14].
What to watch
- Whether the single-cell profiles predict disease or treatment response in animal models or patients, which the cell-culture work does not establish.
- Whether the method extends past E. coli to other pathogens and to larger co-infection mixtures.
- Whether the code and image pipeline are released so other labs can score their own strains the same way.