Every genome editing event produces two categories of outcome: the intended modification at the intended genomic location, and everything else. The first category is the therapeutic goal. The second category is the regulatory problem. Understanding why the FDA has made the characterization of both categories a central requirement for gene editing IND submissions — and what that characterization actually involves — is foundational to building a defensible genome editing development program.

How Genome Editing Creates Unintended Consequences

Genome editing works by introducing a site-specific modification to the DNA sequence of a target cell. The two primary molecular pathways by which cells repair the resulting change — homology directed repair and non-homologous end joining — determine both the type of modification achieved and the type of unintended consequence that can result.

Homology directed repair uses a DNA template to copy a desired sequence into the break site with high fidelity. It is the preferred pathway for precise sequence corrections — substituting a disease-causing mutation with the functional sequence — because it produces defined, predictable edits when a donor template is provided. HDR is most active during the S and G2 phases of the cell cycle, which limits its efficiency in non-dividing cells and means that editing outcomes depend in part on the proliferative state of the target cell population.

Non-homologous end joining rejoins the two ends of a cleaved DNA molecule without a homologous template. NHEJ is the dominant repair pathway in most cell types and is active throughout the cell cycle, making it more broadly applicable than HDR. But NHEJ is imprecise: it frequently introduces small insertions or deletions — indels — at the repair site. When an indel occurs within a coding sequence, it can disrupt the reading frame of the gene, inactivating it. When it occurs near a regulatory element, it can alter gene expression in ways that are not immediately apparent. The consequences of specific indel patterns in specific genomic contexts cannot always be predicted in advance, which is why the FDA’s guidance characterizes NHEJ-based editing as carrying a risk of modifications “with unforeseen consequences.”

The distinction between HDR and NHEJ is not merely a mechanistic detail. It determines the editing strategy, the delivery requirements, the cell cycle considerations, and the characterization burden — because HDR and NHEJ produce different indel profiles, different off-target risks, and different analytical challenges.

HDR vs. NHEJ genome editing repair pathway comparison across eight dimensions. The ‘Strategic/Regulatory Implication’ column reflects FDA 2024 Genome Editing Guidance requirements and MKA analytical framework. Casgevy (exa-cel) reference reflects the first approved CRISPR therapy, which uses NHEJ-based BCL11A enhancer disruption.

What On-Target Editing Means Analytically

On-target editing refers to the intended modification at the intended genomic site. Characterizing on-target editing requires demonstrating that the editing event occurred, that it occurred at the correct location, that it produced the intended sequence change, and that it occurred at a therapeutically relevant frequency.

The last point — frequency — introduces the concept of therapeutic editing threshold. For diseases where a partial correction is sufficient for clinical benefit, the minimum percentage of edited cells or the minimum frequency of the desired edit within the target cell population must be defined and justified. This threshold is not a universal number: it depends on the indication, the patient population, the mechanism of action of the correction, and the available clinical evidence or biological rationale. For a disease like beta-thalassemia, where restoration of functional hemoglobin production above a threshold level is sufficient to eliminate transfusion dependence, the therapeutic editing threshold is clinically derived. For a disease with a more complex dose-response relationship, the threshold may be less clear and require clinical data to establish.

On-target characterization must also address unintended consequences of on-target editing — modifications at the correct genomic location that produce outcomes other than the intended change. These include large deletions around the cut site, chromosomal rearrangements involving the target locus, and integration of delivery vector sequences at the editing site. For HDR-based approaches, unintended integration of the donor template at non-target sites must also be assessed.

What Off-Target Editing Means Analytically

Off-target editing refers to unintended modifications at genomic locations other than the intended target. It occurs when the guide RNA directing the editing machinery binds to sequences elsewhere in the genome that are similar but not identical to the target sequence, and the editing enzyme introduces a break or modification at that alternative site. The frequency of off-target editing is influenced by guide RNA design, the editing enzyme being used, the concentration of editing components, and the specific genomic sequence context of the off-target site.

The FDA’s guidance is explicit about the analytical standard for off-target characterization: genome-wide analysis is recommended to reduce bias in the identification of potential off-target sites. This means that relying on computational prediction alone — identifying candidate off-target sites based on sequence similarity to the guide RNA target — is insufficient. Computational prediction misses sites that deviate from the expected pattern, and it cannot account for chromatin accessibility and other cellular context factors that influence actual off-target editing frequency in the relevant cell type.

Genome-wide empirical methods — including GUIDE-seq, CIRCLE-seq, DISCOVER-seq, and related approaches — identify actual editing events across the genome in treated cells, without the selection bias introduced by focusing only on computationally predicted sites. The FDA expects that off-target characterization use methods capable of detecting low-frequency editing events at sites that were not necessarily predicted.

For in vivo editing products, the off-target analysis must include the major cell types in which editing events are detected — not only the intended target cells. For ex vivo editing products, the final clinical product obtained from multiple donors must be evaluated, because donor-to-donor genomic sequence variation means that off-target profiles can differ between individuals. An off-target site that is efficiently edited in one donor’s cells may not be edited in another’s, or vice versa. A single-donor characterization study cannot capture this variability.

Chromosomal Integrity: Beyond Point Mutations

Off-target characterization that focuses only on small indels misses a category of unintended editing outcome that carries higher safety concern: chromosomal rearrangements. When editing machinery introduces breaks at two or more genomic locations — whether at the on-target site and one off-target site, or at multiple off-target sites — the ends generated by those breaks can be ligated to each other rather than to their correct partners. The result is a chromosomal translocation or inversion that can place oncogenes under the control of strong promoters, delete tumor suppressor sequences, or create fusion genes with oncogenic potential.

The frequency of chromosomal rearrangements in edited cell populations is generally low, but the consequence of a rearrangement involving a clinically relevant genomic locus can be severe and delayed — potentially appearing as a secondary malignancy years after therapy. The FDA’s guidance explicitly requires assessment of chromosomal abnormalities, insertions or deletions, and potential oncogenicity as part of the nonclinical safety evaluation.

Assessing chromosomal integrity requires methods beyond standard sequencing of the on-target and predicted off-target sites. Techniques including ddPCR-based translocation detection, FISH, and long-read sequencing approaches that can span rearrangement junctions are increasingly used in the field. The analytical capability to detect these events must be in place before the IND is filed, because the nonclinical safety data submitted to support first-in-human studies is expected to include chromosomal integrity assessment.

The Regulatory Implication

The FDA’s requirement for comprehensive on- and off-target characterization, including genome-wide off-target analysis and chromosomal integrity assessment, reflects a straightforward regulatory logic: genome editing introduces permanent changes to the genetic material of human cells, and those changes — intended and unintended — persist for the life of the edited cells and, in some editing strategies, the life of the patient. The characterization standard is calibrated to the permanence and scope of the intervention.

For developers, this means that off-target characterization is not a box-checking exercise that can be completed with a standard bioinformatics pipeline. It is a substantive analytical program that requires genome-wide empirical methods, multi-donor evaluation for ex vivo products, assessment in the relevant target cell type, and chromosomal integrity analysis — all conducted with sufficient sensitivity to detect low-frequency events that may not be clinically apparent until years after treatment.

Programs that build this analytical infrastructure into their development strategy from the beginning build toward regulatory submissions that the FDA can review. Programs that defer off-target characterization until it is required find that the analytical work required to support an IND or BLA is more demanding than anticipated, at exactly the stage when timelines are already under pressure.