SYNBIOMATICA

Research

A redesigned miniature DNA cutter edited human cells far more often, without needing a motif beside its target

A GENOME editor is a protein that can be pointed at one chosen stretch of DNA and cut it, after which the cell's own repair machinery patches the break and usually garbles the sequence in the process. The hard part has rarely been the cutting. It is the aiming, and for most of the editors in laboratory use the aim is constrained by a rule that has nothing to do with the target itself. The protein will cut only where a short, specific run of DNA letters sits immediately beside the site the guide picks out, a requirement known as a protospacer adjacent motif, or PAM. Where no such run happens to lie nearby, the site is out of reach.

A preprint posted on bioRxiv on November 15th, by Qianqian Zhu, Xiaohui Zhang and colleagues at the State Key Laboratory of Common Mechanisms Research for Major Diseases, Suzhou Institute of Systems Medicine, Chinese Academy of Medical Sciences and Peking Union Medical College, takes a cutter that already ignores that rule and makes it work far better in mammalian cells. The authors report that their engineered version averaged 48.94% edits across 13 sites in the human genome, against 4.29% for the protein they started from.

The familiar editors inherit the motif rule from bacteria. CRISPR-Cas9, the system adapted from a bacterial defense against viruses, pairs a short guide RNA with a protein that cuts, and the most widely used version of that protein, from Streptococcus pyogenes, insists on the letters NGG next to the target. Smaller relatives are fussier still. The Staphylococcus aureus version, about 1,053 amino acids long, demands the sequence NNGRRT, which narrows the set of sites it can reach even as its compactness makes it easier to deliver into the body. Researchers have spent years loosening that constraint. Russell Walton, Benjamin Kleinstiver and colleagues reported in Science in 2020 a reworked Cas9 they named SpRY, which they described as near-PAMless. It works well at a wide class of motifs in human cells and less well, though still usefully, at the rest.

The alternative is to start from a protein that never had the rule. In 2025 a group including Guilhem Faure, Makoto Saito and Feng Zhang described, also in Science, a family of RNA-guided DNA-cutting systems found in bacteriophages, archaeal viruses and parasitic bacteria, which they called TIGR-Tas. Their guides are unusually small, 36 letters of RNA apiece, and the proteins they steer cut both strands of a DNA target without reference to any neighboring motif. Two of those proteins were shown to work in human cells. The more active one, from a group of bacteria called Parcubacteria and named ParTasR, is 338 amino acids long, less than a third the size of the Staphylococcus aureus Cas9 mentioned above. That matters for medicine, because the virus particles most often used to carry gene-editing machinery into a living animal can hold only a small amount of genetic material.

What they did not do is work very well. In those first human-cell tests, the best result for ParTasR was 3.6% of DNA copies carrying an edit at a gene called CXCR4, measured across six sites in a standard human kidney cell line, with three independent experiments at each site. The related protein TaTasR reached 0.8%. The edits counted in this kind of work are indels, insertions or deletions of a few letters left behind when the cell rejoins a cut end imprecisely. A rate of a few percent is enough to prove that a protein cuts inside a human cell. It is not enough to study a gene's function reliably, and it is far from enough to treat anything.

The Suzhou group set out to raise that rate by changing the protein's surface. No experimental structure of ParTasR exists, so they used AlphaFold 3 to predict the shape of the protein wrapped around its guide RNA and its DNA target, and checked the prediction against the published structure of its cousin TaTasR. DNA and RNA carry negative charge, so the authors reasoned that adding positive charge where the protein meets them would make the grip tighter. They picked out every residue near the nucleic acid that was not already positively charged, along with residues in two internal regions where similar changes have helped other editors, and swapped each one for arginine. That produced 34 single-change variants, tested one at a time against a gene called B2M in human kidney cells. Six beat the original. Combining the two best from different parts of the protein gave a variant with the changes E93R and E216R, which edited 11.41% of DNA copies at that site against 1.39% for the unchanged protein, in three independent experiments.

The second change is not to the cutter at all. When ParTasR cuts, it leaves ends that overhang, and matching overhangs can be stitched back together exactly as they were, which leaves no trace of the visit. The authors added a human enzyme called TREX2, 236 amino acids long, which chews back such overhanging ends so that the cell is more likely to rejoin them clumsily and leave an indel. Paired with the original ParTasR across five genes, TREX2 lifted the proportion of edited DNA copies from a range of 0.06% to 4.16% up to a range of 5.07% to 53.68%, again in three independent experiments per site. A different enzyme that trims from the other direction, included as a control, did nothing much. The authors then put the double-mutant protein and TREX2 into one construct, so that a cell making one makes the other, and called the result eTasR.

The headline comparison runs across 13 further genes in human kidney cells, with three independent experiments at each. Edited DNA copies ranged from 7.26% to 91.32% for eTasR and from 0.29% to 34.13% for the double mutant alone, against 0.30% to 12.43% for the original protein. Averaged over those 13 sites, that is 48.94% for eTasR and 4.29% for the starting protein, a 26.52-fold improvement, with the structural changes alone worth 2.55-fold. Against SpRY, the near-PAMless Cas9, eTasR was substantially higher at 12 of the 13 sites, while the double mutant alone came out higher at five, lower at six and about level at two. The gains survive a cruder method. In these experiments the authors sorted out the cells that had taken up the machinery before sequencing, and when they skipped that step eTasR still produced indels in 25.93% to 74.24% of DNA copies across several sites.

Two numbers in this story measure different things and should not be run together. The 3.6% from 2025 is a single best site for the unaltered protein in one experiment series. The 4.29% here is an average over 13 sites chosen by the new group. Both say the same thing about the starting material, which is that it cut human DNA rarely.

Aiming at the right place is only half of accuracy. A guide can also direct the protein to places in the genome that resemble its target, and those edits are unwanted. The authors used a prediction program to list sites across the human genome matching two of their guides within three letters, then sequenced 43 of them. Most showed no detectable editing by either eTasR or the double mutant. Seven did, six at low frequencies, and one guide with a single mismatched letter was edited heavily. The authors attribute that to the choice of guide rather than the protein, and say better guide design should help, which is a reasonable hope and not yet a result.

Because eTasR and its guide together are small enough to fit in a single virus particle, the group could test it in an animal. They chose mice bred to model hereditary tyrosinemia type I, an inherited disease of the pathway that breaks down the amino acid tyrosine, in which the liver is damaged. Such mice are kept well by a medicine called nitisinone. The group packaged eTasR together with a guide aimed at a gene called Hpd, which sits in that pathway, into one adeno-associated virus of the kind known as AAV8, a type of AAV, and injected it into a tail vein of mice aged six to eight weeks. Two weeks later they stopped the nitisinone and watched for 30 days.

Mice given the virus kept gaining weight steadily after the medicine stopped. Mice given saline instead lost more than a fifth of their body weight, and the authors collected their organs at day 18 rather than day 30. In the treated animals, 49.87% of Hpd DNA copies in the liver carried an edit, with no appreciable editing in heart, kidney, lung or spleen, measured in three mice. Liver tissue from the treated mice looked structurally normal under the microscope, with less of the Hpd protein present, and three blood markers of liver damage were in the same range as in healthy mice, while the untreated animals showed the sharp rises in two of those markers that the disease produces.

All of this comes from a preprint that has not been through peer review, and the editing numbers rest on one group's experiments. The authors are straightforward about what is unfinished. SpRY, the comparison they beat, is itself weaker than ordinary Cas9 working with its preferred NGG motif, which remains the benchmark in the laboratory and in the clinic, so matching that standard is still ahead of them. Off-target editing was detectable at a subset of the sites they looked at, and the single badly behaved guide shows that the problem has not gone away. The injection kept animals bred to have a liver disease from losing weight for a month after their medicine was withdrawn. That is a demanding test of a delivery route and of an editor, and a long way from anyone's liver.

The options for an awkward target have changed. Until now someone wanting to cut at such a spot could use a protein that ignores the motif rule and accept a low rate, or use a well-behaved Cas9 and accept that many sites are unreachable, or reach for a near-PAMless Cas9 that is too large to fit into one virus. These are preprint results in cells and in mice, so a compact editor that needs no motif beside its target now reaches high editing rates in the laboratory, and it has not been tested as a treatment in people.

Sources

  1. Zhu et al., Engineered Miniature ParTasR Enables Efficient PAMless Genome Editin
  2. Faure et al., TIGR-Tas: A family of modular RNA-guided DNA-targeting systems, Sc
  3. Walton et al., Unconstrained genome targeting with near-PAMless engineered CRISP