Do Epigenetic Edits Create Off-Target Methylation Changes?

They can, through three routes: sequence-similar off-target recruitment, local spreading beyond the targeted region, and effector-driven global shifts. How often this happens depends on the editor, its expression level, the guide, and the cell type.

Any specificity claim therefore needs two kinds of data: a panel of predicted off-target sites, and a genome-wide or region-wide profile. A targeted panel alone cannot support "no off-target effects".

1. Guide-directed recruitment. dCas9 binds sequences resembling the target. Because binding — not cutting — is sufficient for an effector to act, sites that would be inert in a nuclease experiment can accumulate marks here. Methylation may build up slowly at such sites, so short experiments can miss it.

2. Local spreading. Repressive effectors act over a local domain. KRAB-based repression can extend roughly a kilobase; methylation-based editors can extend further and, once established, be maintained. In element-resolution studies, spreading into a neighbouring element is the most consequential version of this problem.

3. Global shifts. High effector expression, or an effector with untethered activity, can change the mark genome-wide. The signature is a shift in global methylation or mark density relative to control, which then confounds any locus-specific comparison.

Reported off-target rates differ by an order of magnitude between studies, for reasons that are mostly methodological:

• • Assay sensitivity. WGBS at 10× coverage cannot detect a focal change at 20× depth would.
• • Editor expression level. Strong overexpression increases both untethered activity and spreading.
• • Timepoint. Methylation accumulates; a 7-day experiment underestimates what a 30-day experiment shows.
• • Cell type. Maintenance dynamics differ, so the same edit produces different persistence.
• • Definition of off-target. A change at an unannotated CpG 3 kb away is off-target, but many studies only score predicted sites.

This is why the literature should be read as a range of detection capabilities rather than a single rate.

Without the global layer, a spreading or untethered effect is invisible; without the predicted-site layer, focal guide-directed changes are invisible. Both are needed, and they answer different questions.

Two phrasings that hold up:

• • "Across 24 predicted sites and a 10-kb window flanking the target, no change exceeding 12 percentage points was detected (WGBS, mean coverage 22×)."
• • "Global mean methylation differed by 1.8% from non-targeting controls (n = 3 clones), within the assay's batch variability."

And one that does not: "No off-target effects were observed." That sentence contains no information without a detection limit and a list of what was examined.

Three situations in which specificity data should modify the conclusion rather than appear in a supplement:

• • A neighbouring promoter gained methylation and the target gene shares a pathway with it — the locus-specific mechanism is no longer clean.
• • Global methylation shifted — any locus-specific comparison now rests on a moved baseline.
• • Methylation spread across an element boundary — in element-resolution work, this means the perturbation was regional, not local.

In each case the honest move is to narrow the claim to what the perturbation actually did.

For the full design framework, including assay choice and quantification, see our longer piece on assessing off-target effects in epigenetic editing.

EDITGENE runs the specificity package alongside the edit: predicted-site panels, flanking-window methylation, global profiling, and matched non-targeting and catalytically dead controls, reported with coverage and detection limits.

Tell us the locus and the claim. Contact EDITGENE to request a project consultation or quote.

Off-target rates vary with editor, expression level, guide design, cell type, and timepoint; the routes described here are mechanisms, not frequency estimates.

Specificity claims are bounded by assay sensitivity and by the sites examined, and should be reported with both.

• • Galonska C, et al. Genome-wide tracking of dCas9-methyltransferase footprints. Nature Communications. 2018;9:597.
• • Lin L, et al. Genome-wide determination of on-target and off-target characteristics for RNA-guided DNA methylation by dCas9 methyltransferases. GigaScience. 2018;7:giy011.
• • Nuñez JK, et al. Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing. Cell. 2021;184:2503–2519.