A virus that hunts bacteria

A phage is a self-replicating, self-targeting antibacterial. It finds a bacterium by recognizing a specific molecule on the cell surface — a sugar, a protein, a pilus, even an efflux pump — injects its genome, hijacks the cell’s machinery to build hundreds of copies of itself, and bursts the cell open to release them. Those copies go on to find more of the same bacterium. The dose, in other words, amplifies at the site of infection and then fades away once the target is cleared.

Because recognition is molecular and strain-level, a phage that destroys one Klebsiella strain may completely ignore another — and will ignore the hundreds of beneficial species living alongside it. No antibiotic comes close to that resolution.

Phages come in two broad lifestyles. Lytic (virulent) phages always kill: they replicate and burst the cell immediately. Temperate phages can instead integrate quietly into the bacterial genome as a “prophage,” lying dormant and sometimes carrying genes that help the bacterium — including toxin and resistance genes.

Therapeutic phage development uses strictly lytic phages, or temperate phages engineered to be obligately lytic. This is exactly why some targets are hard: C. difficile, for example, has almost no naturally lytic phages, which is why the field turns to engineered phages and phage-derived enzymes (endolysins) there.

A single phage is brittle. Bacteria mutate the receptor it targets and become resistant within days. A cocktail — several phages that attack the same pathogen through different receptors — solves this in two ways:

• Coverage. Different strains of the same species carry different surface molecules; a cocktail raises the odds that at least one phage matches the patient’s isolate.
• Resistance suppression. To escape the cocktail a bacterium must mutate several independent targets at once — vanishingly unlikely. And when escape happens, it often comes at a cost: losing the receptor can mean losing an efflux pump or a capsule, i.e. losing virulence or antibiotic resistance. Cocktails can be designed to force that trade-off.

The frontier version of this idea is “phage steering”: choosing phages precisely so the only way to resist them is to become treatable by an old antibiotic again.

Modern programs increasingly engineer phages rather than hunt for perfect natural ones. The first therapeutic use of engineered phages cured a teenager’s disseminated Mycobacterium abscessus infection in 2019. Companies now arm phages with CRISPR-Cas systems so they don’t merely lyse the cell but also shred its resistance genes — turning the phage into a tool that can disarm a population, not just kill it. Locus Biosciences, SNIPR Biome, and Eligo Bioscience are each building on this “edit, don’t just kill” logic.

The most important clinical finding of the last decade is unglamorous: phages and antibiotics often work better together. In the largest real-world cohort (100 patients, Belgium, 2024), bacterial eradication was far less likely when phages were given without concomitant antibiotics. Sub-lethal antibiotics can even drive bacteria to elongate or change shape in ways that boost phage replication.

This reframes the whole field. Phage cocktails are not here to replace antibiotics — they are here to rescue them, extend them, and reach the infections antibiotics alone can’t finish.

Phages have an unusually clean safety record — they are part of our normal biology, present in our gut, skin, and saliva by the billions. The real limitations are practical:

• Neutralizing antibodies. The immune system can learn to clear a phage over weeks, which has caused relapse in long courses — a solvable engineering and rotation problem.
• Manufacturing & titre. The PhagoBurn trial failed largely because the product lost potency in production. Dose and stability are everything.
• Strain matching. Specificity cuts both ways: you must match the phage to the patient’s isolate, which takes time the sickest patients may not have. This is the bottleneck the “phage pharmacy” idea aims to dissolve.
• Regulation. A living, adaptive cocktail does not fit the fixed-formulation drug model — the field’s central regulatory puzzle.