Picture a pharmacy that has to build a brand new medicine for every single patient who walks in. Different drug, different trial, different price tag. Nobody could run that pharmacy for long. Yet that is more or less the situation for genetic medicine today.
David Liu opened a recent keynote with exactly this problem. He is the Harvard and Broad Institute chemist who gave us base editing and prime editing, so you would expect a talk about better tools. Instead, he kept returning to a more uncomfortable point. For most patients, the bottleneck is no longer the science. It is the regulatory, practical and economic system around it.
So how do you edit genes when the system can only afford so many drugs? He offered three strategies, each one a step further from “one mutation, one medicine.”
Strategy one: fix the exact mistake
Base editors and prime editors rewrite DNA without cutting both strands and without a donor template. Base editors swap one letter for another. Prime editors work more like search and replace in a word processor (Anzalone et al., 2019).
Does it work in people? The stories he told suggest yes. Alyssa Tapley was 13 when she received CAR-T cells carrying three base edits for her T-cell leukaemia. That was in 2022, and more than three years later her cancer is still undetectable. Then there is baby KJ Muldoon, who had a severe urea cycle disorder. A bespoke adenine base editor was built in about six months and given less than seven months after diagnosis. Liu said he would normally associate that kind of timeline with years, not months.
Prime editing is now in the clinic too. In chronic granulomatous disease, edited blood stem cells from two patients restored roughly 60 to 80 percent of NADPH oxidase activity in neutrophils, far above the roughly 20 percent needed for immune function.
He also showed new mouse work on alternating hemiplegia of childhood, where five ATP1A3 mutations account for more than 70 percent of cases. Prime editing delivered by AAV9 rescued seizures, movement and cognition in two mouse models, and extended their lifespan.
Impressive. But notice the catch. Every one of these is built for one mutation.
Strategy two: skip the mutation and add a healthy gene
What if you do not repair the broken copy at all? What if you simply insert a working one? The idea is decades old, but the modern twist is to do it at the gene’s own address in the genome, so its natural regulation stays intact.
Liu described two ways of doing this, and both borrow from evolution in the lab. The first, PASSIGE, uses prime editing to lay down a landing pad, then an evolved Bxb1 recombinase drops the new gene onto it (Pandey et al., 2024). The second, evoCAST, starts from a bacterial “jumping gene” system that barely works in human cells. After hundreds of rounds of phage-assisted continuous evolution, it inserts whole genes at therapeutic levels (Witte et al., 2025). Both fixes came from the same trick: if nature’s version is too weak, let evolution improve it in the lab.
Strategy three: one edit, many diseases
This was the part that got the room’s attention, and the part I find most creative.
About a third of genetic diseases involve a premature stop codon, a signal that tells the cell to stop building a protein too early. Suppressor tRNAs can read through such a signal and carry on, but as drugs they are weak and would have to be supplied for life.
Liu’s lab asked a different question. Humans carry 418 high-confidence tRNA genes, and some apparently healthy people lack entire families of them. If some are dispensable, why not use prime editing to convert one into a suppressor? The cell then makes its own suppressor tRNA for life, from a single genomic copy. They called it PERT, for prime editing-mediated readthrough of premature termination codons.
The first attempts were underwhelming. Simply swapping the anticodon gave about 5 percent of normal GFP. After screening tens of thousands of variants, the best engineered suppressor reached about 35 percent. With an optimised prime editor, about 80 percent of cells were edited in models of Batten disease, Tay-Sachs disease and Niemann-Pick disease type C1, and enzyme activity rose to 20 to 70 percent of normal. In mice, a single AAV prime editor rescued a model of Hurler syndrome, even though only about 7 percent of the gene was edited. That was enough to give near-normal levels of the missing enzyme. The work was led by Sarah Pierce and Steven Erwood and has since been published in Nature(Pierce et al., 2025).
The obvious worry
If you build a suppressor, will it also read through the cell’s normal stop codons? Liu’s answer was careful, and I think it is worth repeating carefully. The team found no off-target edits, no transcript changing more than twofold, and in whole-proteome mass spectrometry only one protein changed: the rescued GFP. None of the 4,036 human proteins that naturally end in a TAG stop showed readthrough. One possible reason is that the converted tRNA is expressed about five-fold less than the one it replaced. But he stressed this holds under the conditions tested, and he would like more sensitive methods, such as ribosome profiling, to probe it further. That honesty is exactly what I want from someone proposing a new therapy.
What I take away
I came away thinking less about any single tool and more about the question behind them. Not “can we fix this mutation?” but “how few medicines do we need to help the most people?” Mutation-specific editing, gene insertion and PERT answer that question at three different scales. And if Liu is right that the system must now evolve as fast as the science, the next breakthrough may be as much about regulation as about enzymes.
Which of the three would you bet on first? I would love to hear your thoughts in the comments.
Watch the full keynote
The figures above are the speaker’s own, as of the talk. For the full story, including the audience Q&A, watch it here: https://www.youtube.com/watch?v=Z3OM5cYCZHE
Related Content
- Gene editing delivery and virus-like particles post
- CRISPR-Cas classification post
- CRISPR diagnostics, specificity vs sensitivity post
References
Anzalone, A. V., Randolph, P. B., Davis, J. R., Sousa, A. A., Koblan, L. W., Levy, J. M., Chen, P. J., Wilson, C., Newby, G. A., Raguram, A., & Liu, D. R. (2019). Search-and-replace genome editing without double-strand breaks or donor DNA. Nature, 576, 149–157. https://doi.org/10.1038/s41586-019-1711-4
Pandey, S., Gao, X. D., Krasnow, N. A., McElroy, A., Tao, Y. A., Duby, J. E., Steinbeck, B. J., McCreary, J., Pierce, S. E., Tolar, J., Meissner, T. B., Chaikof, E. L., Osborn, M. J., & Liu, D. R. (2024). Efficient site-specific integration of large genes in mammalian cells via continuously evolved recombinases and prime editing. Nature Biomedical Engineering. https://doi.org/10.1038/s41551-024-01227-1
Pierce, S. E., Erwood, S., Oye, K., An, M., Krasnow, N., Zhang, E., Raguram, A., Seelig, D., Osborn, M. J., & Liu, D. R. (2025). Prime editing-installed suppressor tRNAs for disease-agnostic genome editing. Nature, 648(8092), 191–202. https://doi.org/10.1038/s41586-025-09732-2
Witte, I. P., Lampe, G. D., Eitzinger, S., Miller, S. M., Berríos, K. N., McElroy, A. N., King, R. T., Stringham, O. G., Gelsinger, D. R., Vo, P. L. H., Chen, A. T., Tolar, J., Osborn, M. J., Sternberg, S. H., & Liu, D. R. (2025). Programmable gene insertion in human cells with a laboratory-evolved CRISPR-associated transposase. Science, 388(6748). https://doi.org/10.1126/science.adt5199






.png)
.png)
.png)
