Phage Engineering For Targeted Bacterial Control

[SOFIA] Okay, so today we're doing the whole life story of an idea, and it's a good one — turning viruses into tools. Specifically phages. The viruses that infect bacteria.

[DANIEL] Which are, by count, the most abundant biological entity on the planet. Something like ten to the thirty-one of them out there.

[SOFIA] And they've been quietly running bacterial ecology this whole time, deciding who lives and dies in your gut, in the ocean, everywhere. But the reason people are pouring money into this right now is antibiotic resistance. We're running out of drugs, and a phage is basically a self-replicating antibacterial that targets one species and leaves the rest of your microbiome alone.

[DANIEL] That's the promise. The problem is specificity cuts both ways. A phage is exquisitely picky. It'll kill one strain of E. coli and completely ignore the one sitting next to it.

[SOFIA] Right, and that pickiness lives in one part of the phage — the receptor-binding protein. Let me set the stage for anyone who's not in this world. A phage lands on a bacterium and has to recognize a specific molecule on the cell surface. A receptor. Could be a sugar, could be a membrane protein.

[DANIEL] And the protein doing the recognizing is the RBP — the receptor-binding protein, sitting at the tip of the tail fiber. That's the key fitting the lock. If the RBP doesn't match the receptor, no infection.

[SOFIA] So if you want phages as tools — as therapy, as diagnostics, as anything programmable — you have to understand that lock-and-key at basically every level. Which molecule is the receptor? What does the binding look like atom by atom? Can you swap the key? And that's the arc. The field has been climbing that ladder for years.

[DANIEL] And I'd argue it started somewhere that doesn't even look like phage biology. The 2019 NestLink paper.

[SOFIA] Yes! This is the sleeper. So the challenge with any binding protein — nanobodies, these little single-domain antibodies — is that when you screen a big library, you lose track of who's who. Phage display is the classic method, where you fuse your protein library to phage coat proteins and pan them against a target.

[DANIEL] It works, but it's biased. Some clones amplify better than others, and you're stuck picking colonies one at a time to figure out what you've got.

[SOFIA] So NestLink does something clever. They genetically link each binder to about thirty unique peptide barcodes — they call them flycodes — that you can read out by mass spec. So you never touch an individual clone. You just read the barcodes and rank a thousand-plus binders by off-rate in one shot. They got almost six times deeper diversity than phage display.

[DANIEL] And the reason it belongs at the root of this story — it's the mindset. Treat the phage system as a programmable readout. You're not just using phage display, you're re-engineering how information flows through it. That's the tools-not-organisms turn.

[SOFIA] Okay, this is the good stuff, because the next paper goes the complete opposite direction — instead of one clever trick, it's brute-force infrastructure. The BASEL collection, 2021.

[DANIEL] Harms's group. And I love this one because it's what the field was actually missing — a clean, characterized reference set. They isolated over a hundred and twenty E. coli phages, but the smart move was the host they used.

[SOFIA] They deleted the restriction-modification systems, right? The bacterial immune systems that chew up incoming phage DNA.

[DANIEL] A restriction-free K-12 strain. So they're not accidentally selecting only phages that dodge those defenses. They get a less biased catch. Then they narrowed to sixty-eight phages, plus ten classical references, and — this is the part that earns my respect — they mapped the essential host receptor for every single one. Against more than fifty single-gene knockout mutants.

[SOFIA] Every one. That's the lock identified for sixty-eight keys.

[DANIEL] And quantified efficiency of plating against eleven different immunity systems — six restriction-modification, five abortive infection — across seven strains, all at three or more replicates. When someone shows me the replicate count up front, I relax a little.

[SOFIA] And they found something new — LptD as a terminal receptor for seven small siphoviruses. Nobody had that. Plus they showed you could match swappable RBP regions to seven different receptors. So already, hints that the key is modular.

[DANIEL] And they deposited the whole thing at a public strain bank. So it's not one lab's private zoo. Anyone can order it and build on it. That's what turns a paper into a foundation.

[SOFIA] So now we've got the parts list. 2023 zooms all the way in — the first atomic-resolution structure of a phage protein actually gripping its receptor. T5's RBP, called pb5, bound to the FhuA receptor, by cryo-EM.

[DANIEL] And the finding is genuinely surprising. The distal half of that RBP — the business end — is intrinsically disordered before it binds. It's floppy.

[SOFIA] Which is not what you'd draw. You picture a rigid key.

[DANIEL] Right, and it folds up on contact with FhuA. Disorder-to-order. And then the binding imposes a forty-five degree kink, and they propose that kink is the trigger — the cascade that tells the phage to eject its DNA into the cell.

[SOFIA] So binding isn't just docking, it's the starting gun. That's mechanistic gold if you want to control infection. Now — 2024, two papers, and they split the road. One goes computational, one goes genetic.

[DANIEL] The computational one is PhageHostLearn, on Klebsiella. And this matters clinically — Klebsiella is a nasty multidrug-resistant pathogen.

[SOFIA] They took the RBPs and the K-locus proteins — that's the capsule sugar coat, the receptor side — ran them through ESM-2, the protein language model, to get embeddings, then trained XGBoost to predict which phage hits which strain.

[DANIEL] Eighty-two percent ROC AUC in cross-validation, and — the part that matters — seventy-nine percent on twenty-eight actual carbapenem-resistant clinical isolates. And a matching phage in the top five candidates ninety-four percent of the time.

[SOFIA] That's the whole lock-and-key idea from BASEL and the pb5 structure, turned into a prediction engine. You've internalized the biology well enough to guess the match before you run the plate.

[DANIEL] The genetic one is PhageMaP. They built barcoded knockout libraries inside the phages themselves — T7 and Bas63 — using Cas9 and RecA, and mapped which genes are essential under which conditions across forty-four hosts.

[SOFIA] So NestLink's barcoding spirit shows up again, but now inside phage genomes. And they found modular defense inhibitors — pieces you can transfer between phage genomes to beat bacterial immunity.

[DANIEL] Portable parts. That's an engineer's dream and it's exactly where BASEL's modular RBP hint was pointing.

[SOFIA] And then 2025 ties the bow. The T3-T7 evolutionary steering work. They retargeted phages to bind a nanobody as an artificial receptor —

[DANIEL] Which closes the loop back to NestLink's nanobodies, oddly enough.

[SOFIA] — and found two things. Bacterial escape from the phage depends on how much receptor the cell is expressing. And the capsid — not just the RBP — independently shapes host range.

[DANIEL] That last one is a real correction. The whole field leaned on RBP as the specificity determinant. This says the head of the phage matters too. Falsifiable, and it moves the target.

[SOFIA] So six years: from barcoding a library, to cataloguing the keys, to seeing one bind atom by atom, to predicting and rewiring them. We went from studying phages to programming them.

[DANIEL] With the honest caveat that we still don't fully know all the rules. But that's the fun part.

[SOFIA] That's the whole arc. Stick around — more after this.