Off-target editing is one of the most discussed topics in gene therapy, and one of the most frequently misframed. The question is rarely "are there off-targets?" — with sensitive enough assays, the answer is almost always yes. The useful question is: "what is the frequency, genomic location, and functional consequence of these events, and do they pose a meaningful clinical risk?"
Frequency matters more than presence
A Cas9 off-target detected by CIRCLE-seq at a biochemical level has very different clinical relevance from the same site appearing at 0.1% allele frequency in an INDUCE-seq cell-based assay, which is different again from a site at 2% frequency in a primary haematopoietic stem cell.
Regulatory reviewers — FDA, EMA for ATMPs — are increasingly asking for quantitative data, not just ranked lists of potential off-targets. The frequency threshold that triggers concern is context-dependent: genotoxicity near a proto-oncogene at 0.01% is more concerning than an intergenic edit at 1%.
Genomic location and functional annotation
When an off-target is confirmed, the annotation questions are:
- Is it in a protein-coding gene? An intronic hit in a gene with no known cancer association is low risk. An exonic hit disrupting a tumour suppressor is not.
- Is it near a known oncogene or cancer driver? The COSMIC Cancer Gene Census is the standard reference here.
- What is the chromatin state in the target cell type? A closed-chromatin off-target in an HSC may not be expressed even if edited.
- Is the edit likely to be gain- or loss-of-function? Frameshifts and nonsense mutations have different consequences from synonymous variants.
The cell-based vs. cell-free debate
Cell-free sensitivity assays (CIRCLE-seq, GUIDE-seq in cell lysates) find every biochemically accessible site. Cell-based assays show what actually gets cut inside living cells with real chromatin structure.
For IND-enabling studies, the FDA has been explicit that cell-based data in clinically relevant cell types is expected. A beautiful CIRCLE-seq profile from HEK293 cells is not a substitute for data from CD34+ HSCs if that's your therapeutic target.
High-fidelity nucleases
One lever for reducing off-target risk is nuclease engineering. SpCas9 variants like HiFi Cas9 (R691A), eSpCas9, and Cas9-HF1 all reduce off-target editing rates with modest on-target efficiency costs. For clinical applications where regulatory risk outweighs throughput concerns, these variants are worth the trade-off.
Cas12a/Cpf1 has a different PAM preference (TTTN) and generates staggered cuts — useful in contexts where the target site precludes SpCas9. Its off-target profile is generally considered favourable compared to SpCas9, though the comparative data in primary cells is thinner.
What the data package should contain
For a typical IND-enabling bioinformatics package I'd expect:
- In silico off-target prediction (Cas-OFFinder, CRISPOR) — provides a ranked candidate list
- Cell-based DSB detection (INDUCE-seq or equivalent) in the therapeutic cell type
- Validation of top candidates by amplicon sequencing
- Functional annotation of confirmed off-targets against relevant databases
- Statistical model of clonal expansion risk where edits intersect with known driver genes
The goal is to give reviewers a quantitative, annotation-rich picture — not a reassurance that you looked.