Building Genomes From Scratch

[SOFIA] Okay, so here's a question that sounds like science fiction but is actually a real engineering goal: what would it take to write a genome from scratch? Not edit one gene, not swap a promoter — sit down and build the whole thing, base by base, in the organism of your choosing.

[DANIEL] And I want to be careful at the top, because "write a genome from scratch" is the destination, not where any single paper here is standing. What we're really tracing today is the toolkit you'd need to get there.

[SOFIA] Right, that's the honest framing. Nobody's booting up a designer bacterium this week. But if you line up these papers chronologically, you can see the field assembling the pieces — how do you get big DNA in, how do you build it, how do you test what it does, how do you move it around.

[DANIEL] So let's set the stakes for someone who doesn't live in this subfield. Why is writing a genome hard? It's not the chemistry of DNA synthesis — we've been able to synthesize oligos for decades.

[SOFIA] The hard part is everything around it. Think about four separate problems. One, assembly — stitching short synthesized fragments into something kilobases or megabases long without introducing scars or errors. Two, delivery — physically getting that giant construct into a living cell, especially a cell that isn't E. coli. Three, integration — putting the DNA where you want it in the chromosome. And four, function — once it's in there, does it actually do what you designed?

[DANIEL] And each of those has been a wall at different points. Let me define a couple of terms, because the methods have specific names. When you move DNA into bacteria, the word matters — chemical "transformation," or "electroporation" where you zap the membrane open, or "conjugation," which is bacteria physically mating and passing a plasmid through a bridge. Not transfection — that's for eukaryotic cells.

[SOFIA] And conjugation turns out to be the hero of our first paper, actually.

[DANIEL] It does. This is the 2022 Deinococcus radiodurans work. And Deinococcus is a wonderful problem child — it's the radiation-resistant bacterium, famously tough, and its genome is high G+C, around 62% on the megaplasmid they worked with. Chemical transformation of large constructs into it just fails.

[SOFIA] Which is the whole point. If you want to engineer non-model organisms — and that's the dream, right, not just E. coli forever — you keep hitting this delivery wall. So what they showed is that conjugation from E. coli pushes both replicating and nonreplicating plasmids into Deinococcus where chemistry couldn't.

[DANIEL] And then the move I actually found elegant — they ran it in reverse. They used a nonreplicating plasmid carrying about a kilobase of homology to a target, and captured the entire 178 kilobase megaplasmid — pulled it out as roughly a 190 kb clone propagated back in E. coli. Verified by MinION sequencing.

[SOFIA] So you can read a big piece of a weird genome out into your workhorse chassis, edit it there where all your tools work, and conjugate it back. That's the "delivery and capture" chapter.

[DANIEL] It's a nice proof that size isn't an absolute barrier if you pick the right transfer mechanism. Though I'd note — one organism, a specific megaplasmid. It's a demonstration, not a general law yet.

[SOFIA] Fair. Okay, next turning point — 2023, ORBIT in E. coli. This one's about precision at scale. So the old standard for editing the E. coli chromosome is lambda Red recombineering — you use phage proteins to swap in a piece of DNA at a homologous site.

[DANIEL] And lambda Red is workhorse-reliable but inefficient for large payloads. ORBIT's trick is to split the job. The targeting oligo carries a Bxb1 attP site — Bxb1 is a serine integrase, an enzyme from a phage that recombines two specific DNA sequences, attP and attB, very cleanly and in one direction.

[SOFIA] So lambda Red just has to install the little landing pad — the attP site — which it's good at. And then the integrase slams in the kilobase payload. And the efficiency jump is the headline: a thousandfold over lambda Red for the big insert, and it scales to thirty-thousand-member libraries.

[DANIEL] That thousandfold is the number that earns its keep. Because scaling to a 30,000-member library isn't just a bigger experiment — it changes what questions you can ask. You go from editing one site to interrogating the genome in parallel.

[SOFIA] And this is where the arc stops being about one clever trick and starts being about throughput. Which brings us to the 2024 cluster — and this is the good stuff, because now the pieces start talking to each other.

[DANIEL] Start with assembly, since that's problem one. The In- and Out-Cloning paper — de Vries and colleagues.

[SOFIA] So MoClo, modular cloning, is the Golden Gate world — you use type IIS restriction enzymes that cut outside their recognition site, which lets you leave custom overhangs and assemble parts in one pot, scarlessly. The clever bit here is using SapI, which leaves a three-nucleotide overhang.

[DANIEL] Three nucleotides — a codon. So you can define your junctions on codon boundaries and get scar-free transcription units. And "Out-Cloning" generates organism-specific acceptor plasmids on demand. That phrase — "for any organism" — is the tell. They're deliberately building the assembly layer to not care which chassis you're ultimately targeting.

[SOFIA] Which plugs right into the delivery and non-model-organism theme from Deinococcus. Then there's CAST — Banta and colleagues. CRISPR-associated transposases.

[DANIEL] And this is important mechanistically. A standard CRISPR edit makes a double-strand break and relies on the cell to repair it, which is where a lot of the mess and toxicity comes from. CASTs are RNA-guided — you program them with a guide, like CRISPR — but they insert a large payload by transposition, no double-strand break.

[SOFIA] So you get programmable targeting and big-cargo insertion without the breakage. And they used it for genome-scale knockouts, overexpression, and fusion libraries in bacteria. That's the integration-plus-throughput chapter — it builds right on the INTEGRATE line of work.

[DANIEL] And notice ORBIT and CAST are converging on the same goal from different enzymes — integrase versus transposase — both chasing clean, large, targeted insertion at library scale. That's not a clash, that's two routes up the same mountain.

[SOFIA] Then function — the Coaux-Seq paper, which I love because it closes the loop. You can write all the DNA you want, but do you know what the genes do? They barcoded roughly three-kilobase genomic fragments from eleven different bacteria, dropped them into an E. coli expression vector, and selected across twenty auxotrophic knockout backgrounds.

[DANIEL] Auxotroph meaning a strain that can't make some essential metabolite unless you hand it the right gene. So complementation — does the fragment rescue growth — becomes a functional assay, read out cheaply by barcode sequencing. First experimental validation for 53 protein functions.

[SOFIA] Including a TauE-family protein acting as a sulfate importer, when that family was known only as exporters. That's a genuine surprise falling out of a high-throughput screen.

[DANIEL] And it survives my skepticism because it's a direct functional readout — growth or no growth — not a prediction. That's the kind of result I like.

[SOFIA] And the last paper zooms all the way out — bacteria as delivery vehicles into mammalian cells. Engineered strains, lysis circuits, T4SS conjugation, outer membrane vesicles. Same conjugation machinery from the Deinococcus story, now aimed at vaccines and gene therapy.

[DANIEL] So the arc: get big DNA in, assemble it scarlessly, insert it cleanly at scale, test what it does, and deliver it where you want. Nobody's written a genome from scratch here —

[SOFIA] — but you can see every tool on the bench getting sharper. Where it's heading is integration of all five. That's the next show. For now — hands back to the studio.