In April 2026, FDA issued draft guidance, "Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing" (docket FDA-2026-D-1255, CBER/Office of Therapeutic Products; Federal Register notice April 15, 2026). It builds on the agency's January 2024 gene editing framework, and it gives confirmation its own section.
Off-target risk assessment has always had two jobs: nomination (finding where an editor might have cut, using cell-based, biochemical or in silico methods) and confirmation (checking, at each candidate site, whether an edit actually happened, and how often). The 2024 guidance called these "identification" and "verification". The 2026 draft renames them "Nomination" and "Confirmation", and gives confirmation its own section, with recommendations on methods, controls, depth, and reporting.
For nomination, the draft repeats the 2024 recommendation to use multiple approaches: cell-based, biochemical and in silico. Confirmation then measures the editing frequency at the nominated sites, in appropriate cell types. The draft lists "probe-based sequence enrichment, targeted sequence amplification, and/or other equivalent methods" as confirmatory testing methods. It recommends testing at all nominated sites; a subset is allowed only with scientific justification. And it says plainly that "stringent filtering criteria should be avoided to enable evaluation of editing rate at the nominated off-target edit sites". A site that comes back "no signal" because the assay couldn't amplify it isn't a clean result. It's a gap in the data.
This is where confirmation workflows can run into trouble. Multiplex PCR-based methods are well-established, but every additional primer pair in a multiplex reaction is a chance for cross-interaction, which shows up downstream as amplification bias and eventually target dropout. The larger the panel, the more likely it gets, and how much is lost depends on sequence context, panel design, and how much optimization time went into it.
Every dropped site sits in a safety dossier as "no signal", the kind of data blind spot reviewers are likely to ask about, since the draft explicitly recommends strategies to "minimize PCR amplification bias and/or primer bias".
LockSeq confirms candidate sites with gap-fill padlock probes: one probe per target, so there are no primer pairs competing in the reaction. UMI (unique molecular identifier) tagging before amplification means consensus calling suppresses stochastic errors like DNA polymerase errors and the sequencing noise that otherwise masquerades as low-frequency variants, which is what gets sensitivity down to 0.1% VAF (variant allele frequency) with high-confidence edit detection.
Dropout isn't zero with any method, and it still depends on sequence context. Sites LockSeq can't cover are flagged at panel design, before samples are run, and any site that drops out in the lab is still reported, marked as dropout. And LockSeq sits inside the guidance's own "probe-based sequence enrichment" category, named among the confirmatory methods. The guidance doesn't endorse any method, but it does name this category.
The guidance also flags a source of off-target risk that a reference genome can't see: any individual's genome carries several million variants, and a single one can create or erase an off-target site relative to a given gRNA. FDA's response on the nomination side is to ask sponsors for an in silico analysis against population variation databases, optionally stratified by genetic ancestry where disease prevalence is well characterized. On the confirmation side, LockSeq's reporting handles the same problem differently: every report classifies variants against each sample's own donor-matched, untreated control, not only against the reference genome. That separates "common" variants (present in both control and edited sample: germline or population background, unrelated to editing) from "edited-only" signal, so a donor's own genetic background doesn't get mistaken for an editing outcome, and a real edit at a variant-created site isn't masked because the reference genome never flagged it.
The draft expects off-target and translocation studies completed before the original IND is filed, with full study reports submitted alongside it. So off-target work can't wait for Phase I. It has to happen in early R&D. A confirmation method that needs multiple optimization cycles per panel, redone every time the panel changes, is the wrong tool for that timeline.
Comments on the draft closed July 14, 2026, so some of the detail is still open. The direction is clear, though: FDA wants site-by-site confirmation data, and it wants it before the IND is filed.
If you're scoping an off-target safety package and want to talk through where confirmation fits into your IND timeline, get in touch.