Like many projects, ours began with a carefully written research plan. But some of its most important turns came from a conversation at EMBL Heidelberg, a chance encounter at the HUN-REN Biological Research Centre (BRC) in Szeged, and a deceptively simple question from a PhD student.
During an inspiring visit to Nassos Typas’ laboratory at EMBL Heidelberg, I met a researcher using high-throughput microscopy to study how Shigella infection changes in the presence of different drugs. It made me wonder whether drugs affect not only bacterial growth, but also what pathogens actually do to human cells.
Soon after returning to Szeged, I met Szilvia Juhász in a corridor at the BRC. She showed me striking microscopy images of human colorectal epithelial cells infected with genotoxic bacteria, with DNA damage visible in the nuclei.
At the time, I was developing my EMBO Installation Grant project around a broad question: could we find new strategies against genotoxin-producing gut pathogens? Rather than searching only for compounds that kill these bacteria, I wanted to identify treatments that make them less dangerous.
As we were both building our research groups, the Juhász (HCEMM) and Lázár labs joined forces, bringing together complementary expertise in DNA repair and host-cell biology on one side, and bacterial behaviour and systems biology on the other. Why measure only bacterial invasion, we wondered, when we could also measure what happens to the human cell at the same time? Because we were both interested in genotoxic gut pathobionts associated with colorectal cancer, DNA damage was a natural host-cell readout. We decided to measure bacterial invasion and host-cell DNA damage in parallel.
The question that changed the project
We began using 96-well plates and high-content microscopy. Terézia Kovács, one of the study’s first authors, turned the idea into a working experimental system, patiently building and optimising the assay across what felt like an endless series of 96-well plates. We could already see remarkable heterogeneity at the single-cell level.
And then we averaged it all.
Thousands of individual human cells in each well were reduced to a single number.
That changed when my PhD student Bence Bognár, the study’s other first author, asked: “What exactly do we mean by bacterial invasion?” Which bacterium is more invasive: one that enters a single human cell with ten bacteria, or one that enters ten different cells with only one bacterium in each?
That question stopped us.
We realised that invasion was not one-dimensional. It had at least two components: invasion width — how broadly bacteria spread across host cells — and invasion depth — how large the bacterial burden becomes within infected cells.
The same logic applied to the host-cell response. We could measure DNA damage in bacteria-containing cells — direct genotoxicity — but also ask whether surrounding bacteria-negative cells were affected — population genotoxicity.
Using image-analysis software developed with our collaborator Péter Horváth and colleagues at Single-Cell Technologies, we trained machine-learning models to distinguish infected from non-infected cells.
Suddenly, instead of asking whether infection caused DNA damage on average, we could ask where that damage occurred and how it related to bacterial burden in the very same cell.
This was when MALVINA — Machine Learning-Based Virulence Interaction Analysis — really emerged: a microscopy-based single-cell method that measures how bacteria invade human cells and links bacterial burden to host-cell DNA damage.
From a method to a platform
We next asked whether MALVINA could do more than describe one infection.
We first compared virulence profiles across different E. coli and Klebsiella strains,
then asked whether one bacterium could change the virulence of another. We designed the measurement so that only the pathogen of interest needed to be fluorescently labelled, potentially simplifying future screens for beneficial bacteria that suppress pathogen invasion or host-cell damage.
The third application brought us back to drugs.
We worried that if a drug simply killed most extracellular bacteria during the assay, reduced invasion might only reflect fewer bacteria being available to infect cells. So we measured extracellular bacterial survival alongside the single-cell infection phenotypes.
The results surprised us.
There was no simple relationship between antimicrobial killing and invasion. Some treatments strongly increased both invasion width and invasion depth, yet reduced viable bacterial counts by roughly three orders of magnitude.
That was reassuring — and unsettling.
A drug could be highly antibacterial while reshaping virulence-related behaviour in a very different direction.
The Whiteboard Moment
Genotoxicity turned out to be harder.
With some drugs, reduced DNA damage coincided with stronger killing of the genotoxic pathogens, while others could affect human-cell DNA directly. Average DNA damage alone could no longer tell us what was causing the change.
One morning I went looking for Bence because I wanted to draw an idea on the whiteboard.
If cells containing more bacteria normally show more host-cell DNA damage, perhaps we should ask whether the relationship between bacterial burden and host-cell damage itself changes under treatment.
I drew several scenarios. A drug acting directly on human cells might increase the overall DNA damage without changing its relationship with bacterial burden. But if treatment changed how strongly bacteria damage the host cell, the relationship itself — the slope — should change.
About an hour later, Bence came back.
“Boss, aren’t these the lines you were hoping to see?”
They were.
The data contained examples strikingly similar to the scenarios we had just drawn.
That became our genotoxicity shift framework: because MALVINA measures bacterial burden and host-cell DNA damage in the same individual cells, we can model their relationship and ask how it changes under treatment.
For us, this was one of the most satisfying moments of the project — when a sketch on a whiteboard and the single-cell data suddenly met.
What MALVINA changed for us
By the end of the project, MALVINA had become more than a way to quantify bacterial invasion. It provided a framework for connecting microbial burden with host-cell response at single-cell resolution.
In this study, that response was DNA damage. But the same principle could potentially be extended to apoptosis, membrane integrity or other measurable cellular phenotypes.
MALVINA may help identify particularly invasive or genotoxic bacterial strains, uncover unexpected drug effects on bacterial virulence or host-cell toxicity that conventional bacteria-only or host-cell-only tests may miss, and support larger-scale screening of beneficial microbial strains or consortia that suppress pathogen invasion and host-cell damage.
Because killing a pathogen and disarming it are not necessarily the same thing.
A patent application has been filed for the MALVINA method, but perhaps the most important lesson from the project is simpler:
If we want to understand what makes a bacterium dangerous, averaging millions of cells is not always enough. Sometimes we need to look at infection cell by cell.