Thread by @JonasKoeppel

Our story on inserting short sequences with prime editing is now online @NatureBiotech.

nature.com/articles/s4158…

So much has changed from the preprint including the discovery that 3'-flap nucleases can inhibit the insertion of long inserts. Lots to unpack 🧵

A quick recap: We wanted to understand what factors influence how well a given sequence can be inserted with prime editing. To experimentally test this, we integrated 3,604 pegRNAs into 4 target sites in three cell lines and measured their insertion rates

https://twitter.com/JonasKoeppel/status/1458765283330596870?s=20

Our cell lines differed in their mismatch repair status and we recapitulate the well-known effects of MMR in inhibiting the insertion of short sequences with our libraries.
sciencedirect.com/science/articl…
nature.com/articles/s4146…

But the prime editing process consists of many molecular steps beyond mismatch repair and we had a diverse library to interrogate them in detail. In the manuscript, we look at guide expression, epegRNAs, PE2 vs PE3, a different reverse transcriptase, and flap nucleases.

The effects of TREX1 and TREX2 overexpression were particularly interesting. Both are 3’ flap nucleases and could be able to digest the DNA flap containing the edit. Indeed, overexpression of TREX1/2 decreased the insertion efficiency, especially for longer sequences

This leads to a compelling model. Mismatch repair prevents the insertion of short sequences and flap nucleases inhibit the insertion of longer ones. Maybe transiently inhibiting TREX1/2 could increase prime editing efficiency similar to the beautiful work developing PE4/5?

This fits with our finding that structure in the reverse transcriptase template (which will become the flap) increases insertion rates. The effect is stronger for longer sequences and most when we overexpress the flap nucleases. Maybe structured flaps protect from degradation?

We would also like to thank @bsadamsom for pointing us toward the flap nucleases and for chats with Balca Mardin and Özdemirhan Serçin about MMR (at what feels like an eternity ago)

We go into a lot more features and target sites in the manuscript and @JulianeWeller put them together into a computational model 💻to predict insertion efficiencies (MinsePIE 🥧). The model is on GitHub github.com/julianeweller/… and available as a web tool elixir.ut.ee/minsepie/

Shout out to @hedipeterson, Ivan Kuzmin, and Uku Raudvere for building the web interface

Our model does well on a held-out test set and extrapolates to unseen targets. Even from different labs! @_Choi_Junhong shared data on insertions of barcodes into the TAPE-1 sequence from the DNA typewriter story. Our model performed closely to replicate correlation on these

How could this model be useful? Example: Many nucleotide combinations can encode the same amino acid tag. We predicted the efficiencies for these versions and experimentally test ones predicted to insert well or poorly. The ones predicted to insert well indeed performed better!

Side note, linking this work at least in some ways to the targeted protein degradation field, the tags we tested also include a HiBiT tag and a lenalidomide-inducible super degron from @janlabmgh Car T cell paper
science.org/doi/10.1126/sc…

Also, there are currently two exciting openings in the Parts lab @LeopoldParts at the @sangerinstitute to incept and drive projects that use large-scale genome engineering and synthetic genomics to answer questions about the human genome. I can attest to a wonderful lab!

sanger.wd3.myworkdayjobs.com/en-US/Wellcome…
sanger.wd3.myworkdayjobs.com/en-US/Wellcome…

@JulianeWeller, Elin Madli Peets, @fabiolib, @LeopoldParts #CRISPR #geneediting