What's next? 🔮 With non-toxic DNA donors that can be delivered in vivo, we're optimistic that advances in enzyme engineering can move quickly ahead ⏩. One step closer to pan-mutation editing that can be more widely applied to broader sets of patients. 🤞

Together, our study showed that this new DNA format can enable efficient and non-toxic kilobase-scale DNA insertions into the genome 🧬🎉. Congrats to @connorjtou.bsky.social & the entire team of collaborators who enabled this work - a true multi-lab effort.

The net result is that oDNAs are no longer toxic and do not induce rapid lethality in mice like dsDNA. Importantly, we were able to observe in vivo kilobase-sized gene insertion in the liver of mice for the first time without using viruses to shuttle in the DNA cargo!

Cutting a very long story short: we found that oDNAs were immune stealth when injected into mice, and no longer provoked a rapid and immune response like what we observed even at a reasonable dose of dsDNA. Losing most of one DNA strand on the DNA donor makes this DNA format evasive to cGAS/STING!

How did we solve this challenge? 💡 To circumvent toxicity, @connorjtou.bsky.social hatched 🐣 a plan to develop circular single stranded DNAs (cssDNAs) that are 'stealth' to immune sensors, and annealing of a short DNA oligo should recreate a functional dsDNA enzyme binding site - called an oDNA ⭕️

Unfortunately, most approaches to insert kb-sized sequences into cells require large dsDNA donors, resulting in toxicity in cells & mice 🐭. This toxicity is a barrier for many leading-edge DNA insertion tools since they all require dsDNA templates (except for retrotransposases that use require RNAs)

It turns out, that delivering large pieces of double-stranded DNA into cells activates our innate immune response. These immune pathways have evolved to detect 'foreign' DNA in the cytoplasm of cells, alerting that something is wrong and initiating a signaling cascade. 🚨

What was the problem? 🤔 Gene editing technologies to insert large pieces of DNA into the genome are typically comprised of 2 or more components: (1) a DNA donor molecule that encodes the correct gene, & (2) an enzyme that mediates the insertion reaction. We found this process to be quite toxic...

Inserting a 'wild-type' gene could apply to many patients, which contrasts with correcting underlying genetic mutations one DNA letter at a time. Although fixing single base mutations is possible, there are challenges for scaling these approaches in terms of time, cost, & clinical regulatory paths.

Here's a link to the job posting where you can apply in the MGB career portal: massgeneralbrigham.wd1.myworkdayjobs.com/MGBExternal/... Learn more about our research here: www.kleinstiverlab.org

Interested in gaining research experience, and curious about genome editing tool development and their application to treat human diseases? 🧬 🧫 We are hiring!! Join us at Mass General Hospital in Boston! 📣 massgeneralbrigham.wd1.myworkdayjobs.com/MGBExternal/...

An inspirational panel this morning with none other than baby KJ himself - highlighting the stories of heroes @ahrensnicklas.bsky.social & @kiranmusunuru.bsky.social ( hosted by @jasonmast.bsky.social). Crescendo: KJ's Dad on watching 🏈 with his son: "This is what we thought we would never get". ❤️👏

Congrats to @connorjtou.bsky.social on being named a 2025 #STATWunderkind. His enthusiasm in the lab is contagious and he thrives on doing really hard things. Really well deserved recognition for an innovator and emerging leader in our field! 👏 🙌 @mgbresearch.bsky.social @harvardmed.bsky.social

In vivo base editing via AAV9 or a smooth muscle cell tropic capsid AAV-PR (@casey-maguire-lab.bsky.social) substantially extended lifespan of treated MSMDS mice (+ key phenotypic changes). Notably, most AAV-ABE treated MSMDS mice died from bowel impaction, which should be avoidable in human.

@markelindsay.bsky.social and Dr. Musolino's labs created an MSMDS mouse model (that recapitulated many human phenotypes!), which we then utilized for in vivo experiments to deliver our optimized enhanced VRQR (eVRQR) ABE via a dual AAV approach. 🐁

We then performed a comprehensive off-target nomination workflow via an updated GUIDE-seq2 pipeline and the base editor-specific CHANGE-seq-BE assay, followed by rhAmpSeq-based validation of OTs (finding evidence of a small number of intergenic OTs).

To maximize on-target editing at this NGA PAM target site, we undertook a mutant allele-specific engineering approach to develop a bespoke enzyme with enhanced efficiency. Stacking activity-potentiating mutations with VRQR improved correction with minimal bystanders! 💯 ⬆️

To avoid this bystander we spatially minimized access to the M178 adenine by shifting the edit window of the ABE by 4 nt. Using either a PAM-relaxed SpG enzyme (@russelltwalton.bsky.social) or our more selective SpCas9-VRQR, this new target site dramatically minimized the M178V bystander! 🍻

By testing the impact of ACTA2 variant overexpression in human smooth muscle cells, we found that the M178V bystander edit resulted in disrupted actin polymerization; not as severely as the MSMDS R179H mutation, but this is a bystander edit that should be avoided. 🫨

When testing latest generation ABEs in an ACTA2 R179H cell line, we observed high levels of R179H correction (🎉!), but unfortunately most corrected alleles also contained a bystander edit that causes an ACTA2 M178V mutation (😩🤔). Does this bystander edit matter?

MSMDS is caused by a point mutation in the ACTA2 gene, which results in an R179H amino acid substitution - a mutation that should be correctable using adenine base editors (ABEs).

Multi-systemic smooth muscle dysfunction syndrome (MSMDS) is a rare a genetic vascular disease affecting children that Dr. Musolino sees in clinic, known to cause stroke, aortic dissection, and death. There are no therapies, motivating us to explore genome editing-based Tx for these patients.

Then, with Qin Liu @MassEyeAndEar, we utilized PAMmla to develop bespoke allele-selective Cas9 nucleases to treat a blindness causing mutation. Importantly, PAMmla reversed the biology of Cas9, now able distinguish against a canonical NGGG PAM & instead effectively target an NGTG PAM.

To enable users to predict their own bespoke Cas9 enzymes, Rachel developed an in silico directed evolution pipeline and worked with @lucapinello.bsky.social's lab to build a website that permits prediction and navigation of novel Cas9 proteins *in real time* - check it out! pammla.streamlit.app

Working with Dr. Suk See De Ravin @NIH, we demonstrated that PAMmla enzymes are highly effective as base editors to correct mutations in patient-derived primary cells while also minimizing off-targets. No need to use PAM-relaxed enzymes like SpG anymore - use more specific PAMmla proteins instead!

Because PAMmla-predicted enzymes are more selective for their PAMs, they encounter fewer off-target sites around the genome. This results in safer enzymes that minimize unwanted off-target editing.

Testing PAMmla enzymes in human cells led to encouraging results - as we'd hoped, PAM altered enzymes enabled higher levels of on-target editing for nucleases and base editors, while minimizing editing at unwanted PAMs in ways previously not possible with PAM-relaxed enzymes.

We validated that PAMmla could accurately predict the PAM profiles of Cas9 enzymes, both previously unseen before in nature or in our training data sets, and those of previously engineered SpCas9 PAM variant enzymes (developed by us or others).

@rachelsilverstein9.bsky.social then utilized machine learning to investigate all of these proteins including the rarer enzymes, training a PAM machine learning algorithm (PAMmla) that could then predict the PAM requirements of all 64 million Cas9 proteins varied at these 6 amino acids.

The selections yielded two major phenotypes: enzymes that can target 'NGG' sequences (akin to wild-type SpCas9), or those that can tolerate a relaxed 'NGN' PAM (similar to previously developed generalist enzymes). But hidden in this dataset were some more rare Cas9s enzymes..