Engineering Phage Keys For Bacterial Locks

[SOFIA] Okay, so here's a fight we're mostly losing right now. Antibiotic-resistant bacteria kill more than a million people a year, and the drug pipeline is basically dry. And the alternative everyone keeps circling back to is a hundred years old — bacteriophages. Viruses that eat bacteria.

[DANIEL] Which is a funny thing to call new, because phage therapy predates penicillin. The Soviets ran with it for decades. So why is it interesting again now?

[SOFIA] Because we finally have the tools to treat phages like engineerable objects instead of mystery soup. That's the whole arc I want to walk through today — how we went from "here's a virus that kills this strain, we're not totally sure why" to actually designing what a phage attacks.

[DANIEL] Let's define the pieces first, because half the audience does mammalian biology and none of this is intuitive. A phage is a virus that infects bacteria. The first thing it has to do is dock — bind to some molecule on the outside of the bacterial cell. That molecule is the receptor.

[SOFIA] And the part of the phage that does the docking is the receptor-binding protein — the RBP. Think of it as the key. The receptor's the lock. If the key doesn't fit that particular lock, the phage can't get in, full stop. That's most of what "host range" means — which locks a phage can open.

[DANIEL] But docking isn't the end of it. The bacterium has interior defenses. Restriction-modification systems — enzymes that chop up foreign DNA the cell doesn't recognize. And abortive infection, Abi, where an infected cell basically kills itself to stop the phage spreading to its neighbors. So even a phage that gets in can still lose.

[SOFIA] Right, so specificity is this layered thing — the receptor out front, then the immune systems inside. And the problem for a hundred years was that we knew this in the abstract but not in detail. You'd have a phage, you'd know it killed some E. coli strains and not others, and the why was a black box. You can't engineer a black box.

[DANIEL] So the story really needs a reference set. Something clean and characterized.

[SOFIA] Which is exactly the first turning point — the BASEL collection, Harms's group, 2021. And this is the good stuff, because they didn't just grab phages, they built infrastructure. They isolated over 120 E. coli phages on a host they'd stripped of its restriction-modification systems —

[DANIEL] The K-12 MG1655 delta-RM. Which matters, because if your isolation host is chopping up incoming DNA, you only ever recover phages that happen to dodge that host's defenses. You bias your whole collection before you start. Removing RM widens the net.

[SOFIA] They settled on 68 phages, plus ten classic reference strains, and then they did the tedious, beautiful part. Every single phage — they mapped its essential receptor. Against more than fifty single-gene knockout mutants.

[DANIEL] That's the number I care about. Fifty-plus single-gene mutants means you can actually pin causation — knock out one gene, does the phage still infect. And they ran everything at three or more independent replicates. For a resource paper that's real rigor, not a convenience sample.

[SOFIA] And they found a new receptor out of it — LptD, which nobody had pegged as a terminal receptor before, used by seven small siphoviruses. Plus they quantified plating efficiency against eleven different immunity systems. Six RM, five Abi.

[DANIEL] So now you have a matrix. Phage by receptor by defense system, all measured the same way. That's the thing you can build a science on.

[SOFIA] And notice what they showed about the RBPs — swappable loci matched to seven different terminal receptors. Swappable. That word is the promise of the whole field. If the key is a module you can pop out and change —

[DANIEL] Then in principle you retarget the phage by swapping the key. In principle.

[SOFIA] Which brings us to the second beat — you can't engineer the key if you don't know its shape. Degroux and colleagues, 2023, got the first atomic-resolution structure of a phage RBP actually bound to its receptor. Phage T5's RBP, called pb5, locked onto its receptor FhuA, by cryo-EM.

[DANIEL] And the mechanism was genuinely surprising. The business end of that RBP is intrinsically disordered — floppy, no fixed structure — until it touches FhuA. Binding is what folds it.

[SOFIA] Disordered-to-ordered. It grabs the lock and only then snaps into shape. And that folding imposes a forty-five degree kink they think triggers the phage to eject its DNA into the cell. So binding isn't just "am I stuck," it's the switch that fires the whole injection.

[DANIEL] Which complicates the swappable-key story, right? If the RBP has to do conformational mechanics to trigger infection, you can't treat it as a passive Lego brick. You might swap the binding specificity and break the trigger.

[SOFIA] Exactly the tension. And meanwhile the field splits into two ways of attacking the same problem. One is computational — Klebsiella, 2024, a tool called PhageHostLearn. They took protein language model embeddings, ESM-2, fed it the receptor-binding proteins and the bacterial K-locus proteins, and trained a classifier to predict which phage hits which strain.

[DANIEL] And I went in skeptical, because in-silico host prediction is a graveyard of numbers that don't survive contact with a real isolate. But they tested on 28 carbapenem-resistant clinical isolates — actual problem bugs — and got 79.3% ROC AUC out of sample. A matching phage in the top five candidates almost ninety-four percent of the time.

[SOFIA] Which for picking a therapeutic phage is the number that matters. You don't need to be right about every interaction, you need a short list that contains a winner.

[DANIEL] It's prediction from sequence, though. It tells you the key probably fits. It doesn't tell you the phage survives the cell's interior defenses. That's still the gap.

[SOFIA] And the other prong is functional — PhageMaP, 2024. Genome-scale knockout libraries in phages themselves, T7 and Bas63, barcoded, using a Cas9-RecA method. So now you can knock out phage genes across the whole genome and ask which ones matter, under which conditions, across — I think it was 44 hosts.

[DANIEL] Which finally lets you find the anti-defense genes. And they showed those defense inhibitors are modular — you can move one from phage to phage and transfer the counter-measure. That's the flip side of BASEL. BASEL mapped the bacterial defenses; PhageMaP maps the phage's tools for beating them.

[SOFIA] And then the newest one ties the whole arc into a bow. 2025, T3 and T7, engineered to bind a nanobody as an artificial receptor.

[DANIEL] A designer lock. They put a target on the cell that doesn't exist in nature and retarget the phage onto it.

[SOFIA] And two things fell out. Whether bacteria can evolve to escape depends on how much receptor they express — so resistance isn't fixed, it's tunable. And the capsid, the head of the phage, contributes to host range independently of the RBP. Which nobody would've predicted if you thought the RBP was the whole story.

[DANIEL] That's the payoff of everything upstream. You can only discover the capsid's hidden role once you can hold the receptor constant and vary one thing at a time. That's engineering giving something back to basic biology.

[SOFIA] So the through-line: catalog the parts, solve the structures, predict from sequence, map the counter-defenses, then build a phage to your own spec. Fifteen years ago phage therapy was pick-and-pray. Now it's starting to look like design.

[DANIEL] With the honest caveat that receptor binding is one layer, and the escape and defense biology is where the field still has to prove itself. But I'd rather have the map than not.

[SOFIA] Turning viruses into tools, one lock at a time. More after the break.