What FDA's new NGS guidance actually changes — and what it leaves to your judgment

Warning: Some facts and more opinions are found below. Use at your own risk!!

On April 14, 2026, FDA’s Center for Biologics Evaluation and Research (CBER) released a draft guidance on NGS-based safety assessment for genome editing therapies. It’s a companion to the January 2024 GE guidance and zooms in on a question that has dominated my off-target work for years: how do you actually use sequencing to demonstrate that your edit didn’t break something it shouldn’t have?

The guidance is a draft, comments are open until roughly mid-July, and it covers both ex vivo and in vivo products from IND through BLA. If you are a sponsor with a CRISPR/Cas9 (or base editor, or prime editor, or nuclease-of-the-month) product in or approaching the clinic, you should read it. If you’ve been through pre-IND with a genome editing product in the last few years, you will recognize a lot of it.

Here’s my take on what changed, what didn’t, and where sponsors still have to exercise judgment.

What’s actually new

Honestly? Less than the headlines suggest. Most of what’s in the document is what experienced CBER reviewers have been asking sponsors for in INTERACT and pre-IND meetings for the last several years. Anyone who has built off-target packages for a clinical-stage CRISPR program has heard most of these expectations spoken out loud.

What is new is that they’re now written down. That matters more than it sounds. A few specific reasons:

It’s harder for individual reviewers to ask for less. When recommendations live only in pre-IND minutes, you can sometimes negotiate. Once they’re in published guidance, the conversation shifts from “is this required?” to “here’s how we addressed it.”
It’s harder for sponsors to argue for less. Same coin, other side. If your CSO or your investors push back on the cost of a thorough off-target package, you now have a CBER document to point at.
It gives smaller sponsors a roadmap. Big companies with a CRISPR Therapeutics or Editas alumnus in regulatory have known what to do. Series A companies without that institutional knowledge have been guessing. Now they don’t have to.
It signals where international harmonization is heading. EMA and PEI are not bound by FDA guidance, but they do read it, and they reach similar conclusions surprisingly often. Expect overlap with what European reviewers will want.

The guidance covers, among other things: sequencing strategies (which platforms and read lengths for which questions), sample selection (how many donors, which cell types, which timepoints), analysis parameters (filtering, calling thresholds, controls), and reporting (how to present results so a reviewer can actually evaluate them).

What was already implicit

If you’ve been doing this work seriously, you’ve already been doing most of what the guidance asks for. The bones are familiar:

Computational homology search to nominate candidate off-target sites
Empirical/biochemical assays to nominate additional candidates not predicted by sequence (because Cas9 doesn’t read your in silico predictions)
Confirmatory deep sequencing of nominated sites in relevant cell types from multiple donors
Assessment of large structural changes — translocations, large deletions, chromosomal rearrangements — not just small indels at the cut site
An honest accounting of assay limits of detection

If any of those bullets are surprising, you have homework to do before your next pre-IND.

Where the guidance leaves you to make the call

This is the section I think most sponsors will actually need help with. The guidance is recommendations, not a checklist, and it intentionally leaves significant judgment in sponsor hands. A few examples:

Donor count and cell-type selection. The guidance discusses the importance of biological diversity in samples but does not prescribe an exact number of donors or which specific cell types you must use. For an autologous product, that’s relatively straightforward. For an allogeneic or in vivo product, the question of “which donors? which tissues? which timepoints?” is genuinely hard, and the right answer depends on your product, your indication, and your delivery mechanism. You will need to defend your choices.

Limit of detection vs. clinically meaningful threshold. Sequencing can detect editing events at vanishingly low frequencies. Whether a 0.01% off-target editing rate at a given site is clinically meaningful is a different question. The guidance acknowledges the tension but does not resolve it for you. Sponsors should have a defensible position on what frequency thresholds matter for their specific product and why.

How to handle “putative” off-target sites that don’t validate. Computational and biochemical methods nominate sites; targeted sequencing confirms whether those sites are actually edited in the relevant context. Plenty of nominated sites turn out to be false positives. The guidance is reasonable about this in principle. In practice, agencies sometimes ask follow-up questions about why a nominated site was deprioritized, and sponsors who have a clear, written rationale for each decision do much better than sponsors who don’t.

Loss-of-genome-integrity assessment. This is the area where I expect the most variation in sponsor responses. Large deletions, translocations, chromothripsis-like events — these require different assay approaches than small-indel detection, and the methods are evolving. The guidance is appropriately broad, but that means sponsors need to think carefully about which assays they’re using, what those assays can and cannot detect, and how they’re justifying the combination.

Bioinformatics reproducibility. The guidance signals that analysis pipelines should be well-documented, but doesn’t dictate format. If your pipeline is a tangle of one-off scripts that only your last computational biologist could rerun, you are going to have a bad time when an inspector or reviewer asks you to re-analyze with different parameters. Now is a good time to fix that.

What to do in the next 90 days

The comment period closes roughly mid-July (90 days from the April 14 publication). My suggestions for sponsors:

Read it carefully and read it twice. Especially if you have an IND in flight or filing within the next 12 months. Compare your existing package against the guidance section by section and document gaps.
Decide whether to file comments. If there are sections where the recommendations conflict with what’s technically achievable for your modality, or where you think the guidance is silent on something important, file comments. Agencies do read them. Sponsors who participate in the comment process tend to have smoother regulatory interactions.
Use this as an excuse to revisit your off-target package. Even if you’re not filing soon. The methods are evolving fast; what looked thorough in 2022 may not look thorough in 2026.
Talk to your reviewers. If you have a CBER touchpoint already (pre-IND, Type B, INTERACT), the guidance is a natural conversation opener. “We’re updating our package to align with the new draft guidance — here are the specific choices we’re making and why” is a much better conversation than “did we do enough?”

If you’re earlier in development, this is a great moment to engage with CBER through INTERACT before you’ve locked in your assay strategy. INTERACT is free, the agency is genuinely encouraging sponsors to use it, and the cost of changing course before you’ve generated GLP data is much lower than after.

A note on how this connects to your research reports

I wrote previously about how to write IND/CTA/NDA-enabling research reports. The new NGS guidance is, in a sense, the what: what to test, what to detect, how to report. The research-reports piece is the how: how to communicate that work to an agency in a way that doesn’t get you put on clinical hold.

Both matter. A perfect off-target package, badly written, lands sponsors on hold. A well-written report describing an inadequate off-target package lands sponsors on hold. You need both.

What I’m watching for in the final guidance

Things I’d like to see clarified before this becomes final:

More specific guidance on minimum donor numbers for different product modalities
Clearer framing of how to think about clinically meaningful editing thresholds
Explicit acknowledgment that loss-of-genome-integrity methods are still maturing and that sponsors should expect to use evolving methods
Some treatment of how to handle individualized N-of-1 therapies, where the standard “multiple donors, multiple lots” framework breaks down — the guidance touches on individualized therapies but the off-target framework for these is genuinely hard

If you’re a sponsor working on any of the above and you’d like a second pair of eyes on your package, your bioinformatics pipeline, or your pre-IND strategy, get in touch. Especially if you are heading to ASGCT in Boston next week — happy to grab coffee.

Andrew Kernytsky established the on- and off-target gene editing platform for the world’s first company-sponsored CRISPR clinical trial, exa-cel, while leading genomics and computational biology at CRISPR Therapeutics. Next Gen Editing provides regulatory and computational consulting for gene editing companies.