Cytosine is the least stable DNA base. Under some conditions, its half-life is 19 days, compared to a year or longer for the other bases. “That's what leads to this imbalance in the flavor of mutations that have peppered living systems,” says David Liu, a chemical biologist at Harvard University. The spontaneous deamination of cytosine to uracil can cause polymerases to read the former C as a T, making C•G to T•A an unusually common mutation in genomes and a recidivist offender in genetic disease.

Geneticists are always looking for effective ways to reverse point mutations. The CRISPR system can generate precise genomic changes by directing the Cas9 nuclease to cleave at a site with complementarity to a single guide RNA (sgRNA). But replacing point mutations with CRISPR requires homology-directed repair and is often inefficient. Last year, Liu's group came up with an alternative by coupling a cytidine deaminase to a version of Cas9 that nicks DNA instead of cleaving it. An sgRNA brings the resulting 'base editor' to a specific genomic location, where the deaminase converts C to U within a small Cas9-generated single-stranded DNA loop. Nicking also stimulates DNA repair, which converts the original C•G to a T•A base pair.

Tethering an adenine deaminase evolved in vitro to Cas9 enables T•A to C•G edits in the genome. Credit: Adapted from Gaudelli et al. Springer Nature (2017).

Over the past several years, the group also attempted to generate an 'A-base editor' that converts A•T base pairs to G•C using the same strategy, by fusing various adenine deaminases to Cas9. But no known deaminase worked on adenine in the context of DNA. “We realized we had to evolve our own,” says Liu.

Betting on protein evolution was risky. Postdoc Nicole Gaudelli had to flout an inveterate lab rule: do not take on a project in which step 1 is to evolve a new protein as the starting material for the rest of the project. To evolve an enzyme that accepts DNA, she chose to start with TadA, an adenine deaminase from Escherichia coli that does not need a cofactor. TadA works on single-stranded RNA, improving its chances of evolving the ability to function in the single-stranded DNA loop created by Cas9 while avoiding double-stranded off-target sites.

The researchers fused TadA to a catalytically impaired Cas9 and expressed it in E. coli equipped with defective antibiotic-resistance genes that could be restored by targeted A deamination. Antibiotic selection enriched for TadA variants that conferred a survival advantage. Iterative rounds of evolution and engineering, seven in total, were used to dramatically improve DNA adenine deaminase activity.

A key issue was the choice of bacteria for efficient screening. “It's always a risk, and the mantra in the evolution field is 'you get what you select for',” says Liu; variants could be better adapted to bacteria than mammalian cells. For example, TadA functions as a dimer; in the fifth round of evolution, the researchers guessed that evolved TadA was interacting with wild-type monomer in E. coli, and they pursued the development of a single-chain heterodimer in mammalian cells using one evolved and one original copy of TadA—a strategy that lifted the ceiling on editing efficiency improvements.

Evolving a protein takes a certain knack. “You have to make a lot of subjective decisions,” says Liu. “You consider basic biophysical principles, chemical principles, but...you end up having to make decisions in an absence of complete evidence.” In the sixth round, Gaudelli shuffled prior mutations to weed out poorly performing combinations. Amazingly, none of the mutations that she had chosen to carry forward in previous rounds dropped out, “a nice testament to how thoughtfully she had picked them,” says Liu.

Although the field is competitive, Liu gave regular updates about the lab's progress in conferences. The feedback he received encouraged him to keep working rather than release a less mature editing tool. Community interest in making a broad range of mutations, for example, drove Liu's decision to continue developing the A-base editor to shed its natural sequence preferences.

Liu's original C-base editor needed to inhibit an endogenous DNA glycosylase pathway that excises U in DNA and generated undesired byproducts, but the A-base editor faced no such competition and produces uniform edits. The published A-based editor is highly efficient and specific. It generates a ratio of desired point mutations to other modified products that is more than a thousand-fold higher than homology-directed repair. The new tool has opened many directions in therapeutics, animal models and synthetic biology; and the Liu group is set on expanding the set of DNA writing tools by working on editors that perform the remaining classes of mutation.