The Hardest Errors to Find in Gene Editing: UNCOVERseq Brings Rare Off-Target Edits into the Safety Picture

As gene editing moves toward therapeutic use, the real challenge is not only whether the correct site can be edited, but also whether extremely rare errors that may pose risks can be detected. A research team published UNCOVERseq in *Nature Communications*, presenting a cell-based method for identifying CRISPR off-target sites with the aim of incorporating low-frequency edits that could otherwise be missed into risk assessments before candidate therapies enter clinical trials.

UNCOVERseq is an improved version of GUIDE-seq. Double-stranded DNA tags are first inserted into cuts created by Cas9, followed by RNase H-dependent PCR amplification and sequencing. The team added blocking oligonucleotides to suppress uninformative products formed by adapters and tags, and adjusted the sequence alignment and statistical filtering workflow so that limited sequencing reads are more concentrated at genuine genomic junctions.

In a cross-method benchmark established using the EMX1 and FANCF guide RNAs, the researchers confirmed genuine editing sites through ultra-deep sequencing. UNCOVERseq had an analytical sensitivity of 97.6% and a precision of 78%. The former indicates that fewer genuine off-target sites were missed, while the latter means that approximately three-quarters of the sites listed as candidates were subsequently validated. The method does not directly prove that every candidate site has been edited; instead, it generates a more credible list that can be confirmed through targeted sequencing.

The researchers then screened 192 guide RNAs in HEK293-Cas9 cells, which readily amplify off-target signals, before selecting six guide RNAs spanning different levels of specificity for testing in hematopoietic stem and progenitor cells derived from three donors. The comparison included standard Cas9, high-fidelity Cas9, and adenine and cytosine base editors. The reproducibility and low-frequency detection results previously presented in a poster by the research team were also consistent with the overall direction of the formal paper.

One finding of practical importance was that the ranking of off-target sites revealed by double-stranded DNA cleavage correlated with the frequency of off-target edits produced by base editing. This suggests that developers may first use established cleavage-based detection methods to identify high-risk sites and then use targeted sequencing methods such as rhAmpSeq to confirm actual editing. However, this is only an entry point for risk screening and cannot replace validation tailored to a specific editor, cell type, and patients’ genetic differences.

The study also revealed the cost of “looking deeper.” The more low-frequency candidate sites identified, the higher the cost of subsequent confirmation. If biological replicates are omitted, more than one-quarter of the candidate sites meeting the basic prioritization criteria in the study may be missed. More importantly, the core performance comparison of 97.6% sensitivity and 78% precision was based on only two guide RNAs; not all 192 guide RNAs underwent equally comprehensive cross-method validation.

In addition, most of the authors are employed by Integrated DNA Technologies, which provides related reagents and services. The paper also explicitly states that the assay has not undergone comprehensive validation and that the product is for research use only. UNCOVERseq is therefore better regarded as a quantifiable candidate-site nomination framework with configurable process controls, rather than a standard answer capable of independently determining clinical safety. Its true value will depend on whether independent laboratories can reproduce the results and whether it can work alongside other complementary methods to form consistent regulatory evidence.