Off-target CRISPR edits (often abbreviated as OTEs) are unintended genetic changes that occur when the Cas9 enzyme cuts DNA at sites other than the one a researcher designed it to hit. Every CRISPR experiment carries some risk of these stray cuts, and understanding how they arise, how to find them, and how to minimize them has become one of the most active areas in genome-editing research. The stakes are especially high for clinical applications, where a misplaced edit could disrupt a gene the patient needs or trigger dangerous chromosomal rearrangements.
Why Cas9 Sometimes Cuts the Wrong Spot
Cas9 finds its target by scanning the genome with a short guide RNA that base-pairs with a matching stretch of DNA. The system is remarkably precise, but no genome is simple enough for perfect discrimination. Sequences that look almost identical to the intended target can fool the enzyme. Crystal structures of Cas9 bound to genuine off-target DNA show that mismatches between the guide RNA and the wrong site are accommodated through noncanonical base pairing, where the two strands form wobbly but tolerated connections instead of the standard Watson-Crick pairs.1PubMed Central. Structural basis for Cas9 off-target activity When the off-target site has a single missing nucleotide compared to the guide, Cas9 handles it by skipping a base rather than bulging out the RNA, which is a subtler accommodation than researchers originally expected.
Where the mismatches sit along the guide matters enormously. Mismatches near the end of the guide RNA farthest from the short DNA motif Cas9 needs to land (called the PAM) tend to unravel the duplex and trigger a shape change in a flexible arm of the protein. That shape change can slow or prevent the enzyme from fully activating, but it does not always stop it. Mismatches closer to the PAM are more disruptive to binding, which is why the “seed region” near the PAM is considered the most critical stretch for specificity.
Finding Off-Target Edits
A major challenge is simply knowing where off-target cuts happened. You cannot fix what you cannot see, and early CRISPR experiments relied on predictions alone, checking a handful of genomic sites that looked similar to the target. That approach misses a lot. Modern detection falls into two broad categories: cell-free biochemical assays and methods that work inside living cells.
On the biochemical side, CIRCLE-seq was an early breakthrough. It circularizes genomic DNA, then adds the Cas9-guide complex; only cut circles linearize and get sequenced. This catches sites with very low-level cutting that cell-based methods might miss.2PubMed Central. CIRCLE-seq: a highly sensitive in vitro screen for genome-wide CRISPR-Cas9 nuclease off-targets A newer method, CROFT-seq, aims to do something similar but faster and at lower cost.3Synthetic Biology. Off-target detection of CRISPR-Cas9 nuclease in vitro with CROFT-Seq The trade-off with all biochemical assays is that they strip away the cellular context: DNA that is tightly packaged in a living cell and would be hard for Cas9 to reach is fully exposed in a test tube, so these methods tend to overestimate the number of real off-target sites.
Inside cells, GUIDE-seq remains a gold standard. It works by flooding cells with short double-stranded DNA tags that get captured wherever Cas9 makes a cut. Sequencing those tags reveals the locations of breaks genome-wide. When researchers applied GUIDE-seq to 13 different guide RNAs in human cells, they found wide variability in off-target activity and discovered off-target site characteristics that nobody had predicted computationally.4PubMed Central. GUIDE-seq enables genome-wide profiling of off-target cleavage by CRISPR-Cas nucleases Other cell-based approaches include DISCOVER-seq, which tracks where the cell’s own DNA repair machinery shows up after a cut, and BLISS, which labels breaks in situ. Each method has different sensitivity and bias, and a thorough safety assessment often combines more than one.5Frontiers in Genome Editing. Off-Target Analysis in Gene Editing and Applications for Clinical Translation of CRISPR/Cas9 in HIV-1 Therapy
Engineered High-Fidelity Cas9 Variants
Rather than just detecting off-target edits after the fact, researchers have reengineered the Cas9 protein itself to be pickier. The first widely adopted version, SpCas9-HF1, has mutations that weaken the protein’s grip on DNA at points that do not contribute to on-target recognition. With most guide RNAs tested in human cells, SpCas9-HF1 retained on-target cutting comparable to wild-type Cas9, and when guides targeted standard non-repetitive sequences, off-target events dropped to undetectable levels by genome-wide methods.6PubMed Central. High-fidelity CRISPR-Cas9 nucleases with no detectable genome-wide off-target effects
The mechanism behind these high-fidelity variants is elegant. By slowing down the rate of DNA cleavage, the engineered enzymes give off-target DNA more time to fall away from the complex before it is cut. The slowdown affects both on-target and off-target substrates, but the effect is larger for off-target DNA, which is already a weaker match.7Nature Communications. Engineered CRISPR/Cas9 enzymes improve discrimination by slowing DNA cleavage to allow release of off-target DNA Think of it like a bouncer at a door who takes longer to check IDs: legitimate guests still get in, but imposters are more likely to give up and leave.
There is a catch, though. High-throughput screens comparing high-fidelity variants HiFi and LZ3 against wild-type Cas9 found that roughly 20% of guide RNAs lost significant cutting efficiency when paired with these variants.8Nucleic Acids Research. Guide-specific loss of efficiency and off-target reduction with Cas9 variants The efficiency loss depends on the sequence in both the seed region and at positions farther from the PAM that interact with a specific domain of the protein. In practice, this means you cannot simply swap in a high-fidelity Cas9 and assume everything will work the same; each guide RNA may need to be tested with the variant to confirm adequate activity. Researchers have also engineered high-fidelity versions of smaller Cas9 proteins from other bacterial species. One variant derived from Staphylococcus auricularis showed roughly 60- to 110-fold improvement in specificity at certain sites compared to the wild-type version of that enzyme.9PubMed. Engineered Staphylococcus auricularis Cas9 with high-fidelity
Chemical Tweaks to the Guide RNA
Instead of changing the protein, you can modify the guide RNA itself. Incorporating a chemical modification called 2′-O-methyl-3′-phosphonoacetate (MP for short) at specific positions in the backbone of the guide RNA dramatically reduced off-target cutting while preserving on-target performance in clinically relevant genes.10Nucleic Acids Research. Improving CRISPR–Cas specificity with chemical modifications in single-guide RNAs These modifications work in part by stiffening the guide RNA so it is less willing to tolerate the wobble base pairs that enable off-target binding. Chemical modifications also improve the stability and longevity of guide RNAs in therapeutic settings, and can reduce immune reactions the body mounts against foreign RNA molecules.11PubMed. Chemical Modifications in RNA Interference and CRISPR/Cas Genome Editing Reagents
Delivery Format Changes the Risk
How you get CRISPR components into a cell turns out to matter almost as much as which variant you use. The three main delivery formats are DNA (a plasmid that encodes both Cas9 and the guide), mRNA (which the cell translates into Cas9 protein), and pre-assembled ribonucleoprotein (RNP), where the Cas9 protein arrives already loaded with its guide RNA and ready to cut.
RNP delivery is the fastest-acting format because no transcription or translation is needed, but it is also the most transient. The protein does its work and is degraded relatively quickly, limiting the window during which it can wander to off-target sites.12PubMed Central. Non-viral delivery systems for CRISPR/Cas9-based genome editing: challenges and opportunities By contrast, a DNA plasmid can keep producing Cas9 for hours or days, giving the enzyme more opportunities to find and cut lookalike sequences. For brain-targeted editing, self-deliverable RNPs have been developed that cross cell membranes without viral vectors, offering transient editing activity that minimizes both off-target effects and immune reactions.13Nature Communications. Engineering self-deliverable ribonucleoproteins for genome editing in the brain The general principle is straightforward: the shorter the editing machinery is active inside the cell, the fewer stray cuts it makes.
Chromatin Packaging as a Natural Shield
Inside a living cell, DNA is not floating around naked. It is wound around proteins and packed into a structure called chromatin, and how tightly that packaging is wound at any given site affects whether Cas9 can reach it. Tightly packed, silenced chromatin blocks Cas9 from both binding and cutting, and artificially opening that chromatin restores editing efficiency.14PubMed Central. The Impact of Chromatin Dynamics on Cas9-Mediated Genome Editing in Human Cells
This cuts both ways for off-target risk. A large-scale analysis using a tool called EGOLD found that off-target edits occur far more frequently in open chromatin regions than in closed ones.15Cell Discovery. Decoding the role of chromatin context in the off-target effects of CRISPR gene editing with EGOLD That means cell type matters: the same guide RNA might cause different off-target profiles in a liver cell versus a neuron, because different genes are active and accessible in each. Biochemical assays that use naked DNA will flag potential off-target sites that are locked away in chromatin in the cell type you actually care about, which is why combining in vitro and cell-based detection is so valuable.
Computational Prediction Is Getting Better, but Not Perfect
Machine learning and deep learning models now routinely predict where a given guide RNA is likely to cause off-target cuts. Early tools relied on sequence similarity rules, but modern algorithms train on experimental datasets from GUIDE-seq and similar assays.16PubMed Central. Using traditional machine learning and deep learning methods for on- and off-target prediction in CRISPR/Cas9: a review One recent framework, CCLMoff, uses a pretrained RNA language model and achieved an area-under-the-curve score above 0.98, indicating very strong discrimination between true off-target sites and safe ones.17Communications Biology. A versatile CRISPR/Cas9 system off-target prediction tool using language model Earlier deep learning approaches using convolutional neural networks also showed strong predictive performance on benchmark datasets.18Bioinformatics. Off-target predictions in CRISPR-Cas9 gene editing using deep learning
The limitation is that these models are only as good as their training data, and most training data comes from a handful of cell types and species. A model trained overwhelmingly on human cell-line experiments may not predict accurately in, say, a rice genome. The field is moving toward integrating chromatin accessibility and epigenetic features into predictions, which should improve accuracy for specific cell types, but no algorithm yet replaces experimental validation.
Base Editors and Prime Editors Have Their Own Off-Target Profiles
Base editors and prime editors were developed partly to avoid the double-stranded DNA breaks that Cas9 creates, which are the root cause of many off-target problems. But avoiding one kind of risk does not eliminate all risk. Base editors, which chemically convert one DNA letter to another without cutting both strands, have their own distinct off-target signatures. A direct comparison between the adenine base editor ABE8e and standard Cas9 found that ABE8e actually caused significantly more off-target activity, with nearly 99% of validated off-target sites unique to the base editor and not shared with the nuclease.19Nature Biotechnology. Sensitive and unbiased genome-wide profiling of base-editor-induced off-target activity using CHANGE-seq-BE This means you cannot simply use a Cas9 off-target screen and assume it covers a base editor’s risk; each tool needs its own profiling.
Base editors also cause “bystander” edits, changing nearby bases within the editing window that were not the intended target. Newer engineered variants aim to narrow this window. One recent adenine base editor called ABE8e-YA was designed to preferentially edit adenines in specific two-letter sequence motifs, achieving up to a threefold increase in efficiency at those motifs while reducing bystander edits at neighboring cytosines and minimizing DNA and RNA off-targets.20PubMed Central. A motif preferred adenine base editor with minimal bystander and off-targets editing
Prime editors, which use a modified Cas9 that nicks only one DNA strand, tend to produce fewer unwanted insertions and deletions. Researchers have further reduced these by introducing additional mutations into the nickase component. Variants with multiple combined mutations cut the frequency of unwanted insertions and deletions by roughly two- to fourfold compared to the standard prime editor.21Nature Communications. Prime editing with genuine Cas9 nickases minimizes unwanted indels
Epigenetic Editing Brings a Different Kind of Off-Target Problem
Epigenetic editors do not change the DNA sequence at all. Instead, they add or remove chemical tags on DNA or its packaging proteins to turn genes on or off. You might expect this to sidestep off-target mutation concerns entirely, but it introduces a different category of stray effects. When researchers used two different epigenetic tools (dCas9-DNMT3A and CRISPRoff) to add methylation tags at a single target site, they found significant unintended methylation gains at thousands of other sites across the genome. The off-target methylation patterns from both tools were correlated with each other, suggesting these are not random footprints but consistent bystander modifications driven by the editing machinery itself.22Nature Aging. Epigenetic editing at individual age-associated CpGs affects the genome-wide epigenetic aging landscape This is a relatively new area of concern, and the long-term consequences of widespread low-level epigenetic changes remain unclear.
Structural Chromosomal Damage
The scariest category of off-target damage is not a single misplaced point mutation but large-scale chromosomal rearrangements. When Cas9 cuts at two or more sites simultaneously, whether on-target and off-target or two off-targets, the cell can rejoin the broken ends incorrectly, producing translocations (where segments from different chromosomes fuse), large deletions, or inversions. An in vivo study in zebrafish found that CRISPR-Cas9 induced large structural variants at both on-target and off-target sites, including one 903-base-pair deletion at an off-target site that completely removed an exon from an unintended gene, and these variants were heritable across generations.23Nature Communications. CRISPR-Cas9 induces large structural variants at on-target and off-target sites in vivo that segregate across generations
Translocation analysis tells a similar story. Using a tool called CRISPECTOR, researchers found that when a guide RNA targeting the RAG2 gene was used, the fusions formed between two off-target sites actually occurred more frequently than fusions involving the on-target site itself.24Nature Communications. CRISPECTOR provides accurate estimation of genome editing translocation and off-target activity from comparative NGS data All four theoretically possible translocation configurations from two simultaneous double-strand breaks were observed in the experimental data. These findings underscore why off-target detection that looks only for small insertions or deletions at predicted sites can miss the most dangerous outcomes.
Human Genetic Variation Complicates Everything
Most guide RNA design happens against a single reference genome, but real patients have millions of individual genetic variants scattered across their DNA. A single-nucleotide variant in a patient’s genome can create a new off-target site that does not exist in the reference, or it can reduce the number of mismatches at an existing near-match site, making that off-target more potent.25PubMed Central. Human genetic variation alters CRISPR-Cas9 on- and off-targeting specificity at therapeutically implicated loci This means two patients receiving the same CRISPR therapy could have different off-target risk profiles depending on their genotypes. For clinical applications, the implication is that off-target assessment ideally should be personalized, or at least should account for common population-level variants at sites flagged as potential off-targets.
The Clinical Risk Assessment Pipeline
For CRISPR therapies heading toward patients, regulators expect a systematic evaluation of off-target risk. The emerging standard is a two-step process. First, a broad screen generates a list of potential off-target sites using high-sensitivity methods (both computational prediction and experimental assays like GUIDE-seq or CIRCLE-seq). Second, each candidate site is checked by deep sequencing in a model system that closely matches the clinical use conditions, such as the target cell type treated with the actual therapeutic construct.26PubMed. Analysis of Off-target Effects and Risk Assessment Leading from Preclinical to Clinical Trials of Gene-edited Therapeutic Products Only off-target events that are both detected by the broad screen and confirmed by targeted sequencing are carried forward into the formal risk assessment. This tiered approach balances thoroughness with feasibility, since sequencing every base of every treated cell’s genome at the depth needed to catch rare events is not yet practical at scale.
How Earlier Editing Tools Compare
CRISPR-Cas9 is not the first programmable gene-editing tool, and comparing it with its predecessors puts its off-target profile in context. Zinc finger nucleases (ZFNs), the first generation of customizable DNA cutters, can produce enormous numbers of off-target events. When researchers applied GUIDE-seq to three ZFNs targeting the same region of HPV, they found off-target counts ranging from about 290 to over 1,800 sites per enzyme.27PubMed Central. The comparison of ZFNs, TALENs, and SpCas9 by GUIDE-seq in HPV-targeted gene therapy TALENs, the second generation, showed intermediate specificity. Cas9, when well-designed, generally outperforms both, though the comparison depends heavily on the specific target site and the design quality of each tool. The point is not that Cas9 is perfect but that its off-target problem is substantially more manageable than what came before.
Off-Target Edits in Crop Plants
The risk calculus for off-target edits looks quite different in agriculture compared to medicine. Plant breeders have always introduced genetic variation, whether through chemical mutagenesis, radiation, or simple cross-breeding, and these conventional methods produce far more random mutations than a well-designed CRISPR experiment. When off-target changes from CRISPR are compared to the background variation from standard breeding or even the chemical treatment used to deliver CRISPR reagents into plant cells, the editing tool itself contributes a negligible fraction of the total mutations.28PubMed Central. Plant Genome Editing and the Relevance of Off-Target Changes
Genome-wide sequencing of CRISPR-edited plants bears this out. In one study, edited plants carried an average of about eight single-nucleotide changes and 20 small insertions or deletions, but those numbers were not statistically different from control plants that had been treated with the delivery chemical alone and never exposed to a gene editor.29Scientific Reports. Genome-wide specificity of plant genome editing by both CRISPR–Cas9 and TALEN A broader literature review came to a similar conclusion, though the authors cautioned that the finding rests on a limited amount of available data and more comprehensive studies are needed.30PubMed Central. Occurrence and Nature of Off-Target Modifications by CRISPR-Cas Genome Editing in Plants Plants also have an advantage that human patients do not: unwanted edits can be bred out over a generation or two by crossing the edited line with unedited plants and selecting offspring that kept the desired change but lost any stray mutations.
Beyond Cas9
The CRISPR toolkit keeps growing, and newer enzymes offer different trade-offs. Cas12 (also known as Cpf1) recognizes a different landing motif on DNA, creates staggered cuts instead of blunt ones, and has its own off-target profile that in some contexts is more favorable than Cas9’s. Cas13 targets RNA instead of DNA, meaning any off-target effects are temporary since RNA is naturally degraded by the cell. Each system has distinct strengths: Cas12’s requirement for a T-rich landing motif limits the number of genomic sites where it can bind, which inherently reduces the pool of possible off-targets, while Cas13’s RNA-level editing avoids permanent genomic changes entirely.31PubMed. Assessing off-target effects in CRISPR/Cas9: challenges and strategies for precision DNA editing The growing diversity of CRISPR effectors means researchers increasingly match the tool to the job, choosing the enzyme whose off-target characteristics best fit the application rather than defaulting to Cas9 for everything.