Key Takeaways
Large-payload genome engineering could enable mutation-agnostic therapies by inserting a functional gene or replacing a defective genomic region rather than correcting individual pathogenic variants.
CRISPR-associated transposases, engineered recombinases, prime editing–derived systems, and bridge recombinases offer distinct routes to gene-sized insertion, inversion, excision, and larger genomic rearrangements.
Successful large-payload editing requires characterization of integration location, orientation, copy number, cargo integrity, donor–genome junctions, off-target integration, and structural genomic changes.
Delivery and manufacturing become more complex when therapeutic systems combine donor DNA, guide RNA, editor proteins or messenger RNA, recombinases, transposases, delivery vehicles, and edited cells.
No large-payload genome-editing platform has yet emerged as the definitive therapeutic solution, and future success will depend on balancing editing efficiency with specificity, manufacturability, analytical control, and functional performance.
From Correcting Mutations to Rewriting Genomic Regions
Therapeutic genome editing has largely developed around a few core operations: correcting a nucleotide substitution, introducing a small sequence change, disrupting a gene, or modifying gene expression. These approaches can address diseases with well-defined molecular causes, but they may also tie each therapeutic design to a specific pathogenic variant, mutation class, or accessible editing window.
That model becomes more difficult for genes associated with many pathogenic variants distributed across multiple exons or regulatory regions. Correcting each variant separately could require numerous editing reagents, product configurations, and development programs. Large-payload genome engineering offers a different strategy: install a functional gene, replace a larger defective sequence, or reconstruct a genomic region in a way that could address many disease-causing variants through the same therapeutic design. Although therapeutic use of programmable genome rearrangement remains early, recombination-based genome engineering has long been used in model systems.1
This mutation-agnostic concept is among the most important potential advantages of programmable gene insertion. Rather than repairing each loss-of-function mutation individually, a developer could seek to supply an intact coding sequence at a defined genomic location. Recent work on CRISPR-associated transposases has explicitly positioned programmable gene integration as a potential route to mutation-agnostic treatment of loss-of-function diseases, while engineered recombinases have been developed with entire-gene insertion in mind.2,3
The opportunity extends beyond gene addition. Emerging systems can perform deletions, replacements, inversions, excisions, and larger rearrangements, broadening the possible unit of intervention from a single nucleotide to an exon, a coding sequence, a regulatory element, or a much larger chromosomal segment. Programmable bridge recombinases have demonstrated insertion, excision, inversion, and megabase-scale rearrangement in human cells, illustrating how genome engineering may eventually address structural abnormalities that cannot be resolved through single-base correction.4
Large-payload editing will not be appropriate for every genetic disease. A functional gene inserted outside its native context may not reproduce endogenous regulation, and a replacement strategy may not overcome dominant-negative biology, tissue-specific expression requirements, or delivery barriers. Gene size, target-cell turnover, immune responses, and the consequences of sustained editor expression will also influence feasibility. Even so, the ability to change larger genomic units opens a therapeutic design space that conventional editing cannot fully reach.
A Diversifying Toolkit for Large-Payload Genome Engineering
Several mechanistically distinct technology families are advancing toward gene-sized insertion and broader genome rewriting. They differ in how they identify a target, how they move DNA, whether they require a pre-installed recognition sequence, how many components must enter the cell, and what unintended products may accompany the desired edit.
CRISPR-associated transposase (CAST) systems, combine RNA-guided targeting with transposition machinery. In principle, this allows a programmable RNA sequence to direct the insertion of DNA cargo without relying on conventional homology-directed repair. A compact Type V-K CAST system has been engineered to insert therapeutic cargo into the human genome, including a 3.6-kilobase Factor IX construct at the AAVS1 locus. The study also examined insertion orientation, both donor–genome junctions, complete and incomplete products, and unintended donor incorporation, demonstrating that nominal integration efficiency captures only part of the resulting genomic profile.5
Because naturally occurring CAST systems evolved in bacterial contexts, translating them into efficient human-cell gene insertion often requires substantial protein engineering. Laboratory evolution has produced CAST variants with substantially greater integration activity in human cells than their starting systems. That progress supports the feasibility of adapting bacterial machinery to mammalian genome engineering, but it also shows that natural activity cannot be assumed to translate directly into therapeutic performance.2
Large serine recombinases provide another route to gene-sized insertion. These enzymes catalyze recombination between defined DNA recognition sites and can integrate multi-kilobase cargo without conventional homologous recombination. Earlier implementations often depended on a landing site already present in the genome, limiting their use at native human loci. More recent work has engineered recombinases capable of recognizing selected endogenous sites and integrating DNA cargoes of up to 12 kilobases in human cells, including stem cells, primary T cells, and nondividing cells.3
The appeal of recombinases lies partly in their ability to perform direct, site-specific integration, but their development raises its own questions. A therapeutic system must balance activity with sequence selectivity, expand the range of targetable genomic sites, and limit recombination at unintended sequences. The donor, recombinase, and any targeting or delivery components must also reach the same cell in an effective ratio and within a suitable time window.
Prime editing has generated a third family of large-insertion approaches. Prime editors were initially developed to write small sequence changes without creating a targeted double-strand break, but paired and hybrid configurations can extend that principle to larger genomic operations. Twin prime editing uses two prime-editing guide RNAs to create complementary sequence changes, which enables larger replacements, deletions, and inversions. It can also install recognition sites for a recombinase, establishing a programmed landing pad for subsequent cargo integration.6
PASTE combines prime-editing–mediated installation of a recombinase attachment site with integrase-mediated cargo insertion. The published system inserted DNA sequences as large as approximately 36 kilobases in human cell lines, primary T cells, and nondividing primary human hepatocytes. This reported capacity illustrates the potential reach of hybrid editing systems, although it should not be interpreted as evidence that integrations at the upper end of that range are already routine, efficient, or therapeutically deliverable.7
PASSIGE follows a related strategy while separating the prime editor and recombinase rather than fusing them into one protein architecture. Continuous evolution was used to improve recombinase performance, producing higher integration activity in mammalian cells. This work illustrates how the same overall editing concept can change materially through modifications to enzyme architecture, expression, component arrangement, and delivery.8
Prime Assembly represents another developing approach, using linear donor fragments with overlapping sequences to construct larger genomic insertions. Its emergence reinforces a broader point: large-payload genome engineering is becoming a field of modular design strategies rather than a single platform class. Different systems may divide the operation among targeting, sequence writing, recombination, donor assembly, and delivery in different ways.9
Bridge recombinases expand the possible scale and type of edit further. These systems use programmable bridge RNAs to specify interactions between DNA substrates, creating a potential route to insertion, excision, inversion, and large genomic rearrangement. Megabase-scale rearrangements have been demonstrated in human cells, moving the discussion beyond gene-sized cargo toward manipulation of chromosomal architecture.4
These capabilities remain experimental, and they are not equally mature. A system that performs a large rearrangement in cultured human cells has not necessarily established a viable therapeutic route, a manufacturable product, or an acceptable safety profile. The growing range of available operations nevertheless changes the central development question. It is no longer only whether a genomic sequence can be altered but whether the resulting genomic product can be precisely defined and consistently controlled.
Four Routes to Large-Scale Genome Writing
CRISPR-Associated Transposases
CAST systems combine RNA-guided targeting with transposition machinery to insert DNA cargo at programmed genomic sites. Their principal development questions include activity in human cells, multicomponent delivery, integration orientation, cargo completeness, and off-target insertion.
Engineered Recombinases
Large serine recombinases catalyze site-specific integration of multi-kilobase DNA cargo. Current engineering efforts seek to improve efficiency, expand the range of endogenous target sites, and reduce recombination at unintended sequences.
Prime Editing–Derived Systems
Twin prime editing, PASTE, PASSIGE, and Prime Assembly use prime editing alone or in combination with recombinases, integrases, and engineered donors. These strategies distribute the editing process across several programmed steps, enabling larger insertions while increasing component and delivery complexity.
Bridge Recombinases
Bridge recombinases use programmable bridge RNAs to specify DNA recombination. They have demonstrated insertion, excision, inversion, and megabase-scale rearrangement in human cells, although their therapeutic and manufacturing development remains at an early stage.
The Product Is the Edited Genome
Large-payload integration creates a more complex product than a corrected nucleotide or small insertion. A positive signal at the target locus does not establish that every edited cell contains the same genomic outcome, that the cargo is complete, or that the surrounding chromosome remains intact.
The first analytical task is to confirm where integration occurred. Targeted integration must be distinguished from random or off-target incorporation, and developers need to determine whether cargo is present at one location or several. For systems that recognize endogenous sequences, naturally occurring variation at the target site may also influence integration efficiency or specificity.
Orientation adds another dimension. Some insertion mechanisms can generate forward and reverse products, while others may favor one direction without eliminating alternative outcomes. Both cargo–genome junctions must be examined to determine whether the insertion is complete and structurally correct. A single correctly formed junction does not establish that the opposite end of the cargo is intact.
Potential outcomes include truncated cargo, tandem insertion, rearranged donor sequences, partial integration, donor-backbone incorporation, and co-integration of unintended material. These are not theoretical concerns. CAST studies have characterized orientation, junction structure, incomplete integration, and unintended donor products, demonstrating the need to define the molecular architecture of the insert rather than reporting only the percentage of cells positive for an on-target event.5
Integrated copy number is also relevant to product definition. The objective need not be exactly one copy in every application, but the distribution should be understood and justified. Multiple copies could alter expression, increase the opportunity for rearrangement, or produce variable cellular phenotypes. In an ex vivo cell product, expansion may enrich or deplete cells with particular integration patterns. In vivo, different cells within the target tissue may acquire different numbers or configurations of the cargo.
European regulatory guidance identifies integration location, copy number, transgene structural integrity, rearrangement, off-target integration, persistence, and genomic stability as important considerations for integrating gene-therapy systems, including systems that use mobile elements or site-specific recombinases.10
Functional expression must also be connected to genomic identity. A complete insertion at the intended site may still produce inadequate, excessive, unstable, or cell type–inappropriate expression. Depending on the therapeutic design, characterization may need to address transcription, protein production, expression durability, cellular function, and preservation of relevant endogenous regulatory relationships. The desired product is not simply DNA at a target locus but a genomic change that produces a reproducible biological effect.
Seeing Beyond the Target Site
Conventional amplicon sequencing is well suited to measuring small changes near a known target, but it can miss outcomes whose junctions extend beyond the amplified region. Large deletions, complex rearrangements, translocations, chromosome-arm loss, and other structural changes may fall outside the view of short, locus-focused assays.
Studies of nuclease-mediated editing have demonstrated that DNA repair can produce large deletions and complex rearrangements extending many kilobases from a target site. Other work has detected atypical off-target structural variants through linked-read sequencing and optical genome mapping, showing that sequence similarity alone may not identify every relevant site of genomic disruption.11,12
The use of an HDR-enhancing DNA-dependent protein kinase inhibitor has also been associated with kilobase- and megabase-scale deletions, chromosome-arm loss, and translocations. Those outcomes were investigated using a combination of long-read sequencing, copy-number analysis, single-cell RNA sequencing, and unbiased translocation detection. The study underscores the value of combining methods that interrogate different scales and classes of genomic change.13
These findings do not mean that every large-payload platform will produce the same structural risks. Systems that avoid a targeted double-strand break may have different outcome profiles from nuclease-dependent approaches. Recombinases, transposases, prime editors, and hybrid systems also create distinct intermediates and depend on different cellular pathways. Analytical strategies should therefore follow the mechanism, donor structure, cell type, delivery format, and plausible failure modes of the product.
Long-read sequencing can be particularly useful for confirming full-length cargo, resolving both integration junctions, phasing complex products, and detecting rearrangements that extend beyond short amplicons. Other methods may contribute complementary information, including targeted capture, digital polymerase chain reaction, whole-genome sequencing, optical genome mapping, cytogenetic analysis, and translocation assays.
No single method provides complete coverage. High-depth targeted sequencing may detect rare local events but miss larger structures. Whole-genome methods provide broader coverage but may lack sensitivity for low-frequency outcomes. Copy-number assays may reveal gain or loss without identifying the associated junction. A risk-based analytical package may therefore require several orthogonal methods rather than reliance on one sequencing readout.
Bulk measurements can also conceal heterogeneity. A population with an acceptable average integration rate may contain minority subpopulations with markedly different genomic architectures. Single-cell and clonal approaches can help determine whether structural abnormalities are concentrated in a small group of cells and whether genomic outcomes correlate with altered expression or phenotype. Genome-Shuffle-seq has demonstrated multiplex generation, mapping, and analysis of structural variants at single-cell resolution, illustrating the level of heterogeneity that may become analytically relevant.14
Single-cell sequencing is not necessarily required for every large-payload product. Its value will depend on the product format and risk profile. It may be particularly informative for ex vivo products that contain heterogeneous edited populations, undergo substantial expansion, or could preserve rare clones with disproportionate biological consequences.
The U.S. Food and Drug Administration’s (FDA’s) 2026 draft guidance on next-generation sequencing reflects growing regulatory attention to off-target editing and loss of genome integrity in both ex vivo and in vivo genome-editing products. Because the document remains draft guidance, it should be read as the agency’s current thinking rather than a fixed universal assay prescription.15
What Must Be Confirmed After Large-Payload Integration?
A product-specific analytical strategy may need to establish:
the intended genomic location;
complete cargo sequence and structural integrity;
integration orientation;
both donor–genome junctions;
integrated copy number;
the absence of partial cargo, donor backbone, or unintended production sequences;
off-target integration;
local deletions and rearrangements;
chromosome-scale structural changes;
cell-to-cell heterogeneity; and
functional and durable expression.
The appropriate combination of sequencing, copy-number, cytogenetic, structural, and functional assays will depend on the editing mechanism, donor format, cell type, delivery strategy, and identified risks.
Delivery as a Systems-Engineering Challenge
Large-payload genome engineering often requires coordinated delivery of several distinct materials. Depending on the platform, these may include an editor protein or messenger RNA, one or more guide RNAs, a recombinase or transposase, donor DNA, accessory factors, and/or a delivery vehicle. Each component must reach the appropriate intracellular compartment in an effective amount and at a suitable time.
The challenge is therefore broader than cargo capacity. Component ratio, expression duration, intracellular stability, cellular uptake, endosomal escape, nuclear entry, and donor availability can all influence both efficiency and the distribution of editing outcomes. Prolonged editor expression may improve activity while also extending the opportunity for unintended events. Excess donor may support integration but increase residual material or off-target incorporation. Delivery conditions that maximize editing can also reduce viability or alter cell phenotype.
Ex vivo editing provides direct access to target cells and allows the use of electroporation or transfection under controlled conditions. Components can be removed after editing, and the resulting population can be characterized before administration. Developers may also enrich edited cells or expand selected populations. These advantages come with manufacturing challenges involving cell viability, functional preservation, culture conditions, chain of identity, processing time, and population heterogeneity.
In vivo delivery avoids removal and manipulation of patient cells, but it must overcome tissue targeting, extracellular degradation, cellular uptake, payload size, and coordinated delivery of multiple components. Cells that receive only part of the system cannot be removed before treatment, and an administered donor may distribute differently from the editor or targeting components. Although lipid nanoparticles and viral vectors have enabled important forms of genome-editor delivery, current evidence does not establish that in vivo delivery of complete large-payload insertion systems has been solved.
The choice among DNA, messenger RNA, protein, ribonucleoprotein complexes, plasmids, linear DNA, minicircles, viral vectors, nanoparticles, and ex vivo electroporation changes the manufacturing process as well as the biological profile. Each format brings different impurities, stability considerations, storage requirements, expression kinetics, innate immune risks, and analytical needs. A highly active editing architecture may therefore be less developable than a somewhat less efficient system that can be delivered transiently and manufactured reproducibly.
Donor DNA as a Critical Manufacturing Input
The donor sequence defines much of the intended therapeutic product. It may include the coding region, promoters, regulatory elements, recombination sites, homology sequences, and manufacturing-related backbone elements. Its design, production, purification, and testing are therefore linked directly to the composition of the resulting edited genome.
Potential donor formats include plasmid DNA, linear double-stranded DNA, single-stranded DNA in selected applications, minicircle DNA, viral-vector genomes, and overlapping donor fragments assembled during editing. These formats are not interchangeable. They differ in production method, topology, stability, cellular handling, purity profile, and propensity for unintended incorporation.
Regulatory guidance identifies sequence identity, structural integrity, concentration, purity, sterility or bioburden, endotoxin, and freedom from extraneous agents as relevant considerations for DNA vectors and templates. Additional controls may address residual host-cell DNA, host-cell proteins, RNA, production enzymes, antibiotic-resistance elements, and bacterial-backbone sequences.16,17
These attributes matter because donor-related impurities can become more than conventional process impurities. If backbone sequences, incomplete products, or residual templates are incorporated into the genome, they become part of the cellular product. Donor design should therefore minimize unnecessary sequences and support analytical discrimination among intended cargo, production backbone, partial products, and rearranged material.
Minicircle DNA represents one research-stage effort to reduce bacterial backbone and improve the purity of circular DNA used for genome editing. A cell-free production method has been developed to generate high-purity minicircles while addressing parental-plasmid contamination, incomplete recombination, endotoxin, sequence accuracy, and purification yield. This work illustrates emerging options for DNA-component production, but it does not establish minicircle DNA as a standard good manufacturing practice donor format for therapeutic large-payload editing.18
Large or structurally complex donor constructs may create additional challenges in sequence stability, propagation, yield, purification, analytical method development, storage, and shipping. The relevant specifications should reflect the donor’s relationship to the final genomic product rather than treating it only as an input that disappears after the editing reaction.
Manufacturing a Multicomponent Product
Large-payload systems can bring together materials usually associated with several manufacturing disciplines: recombinant proteins, synthetic or transcribed RNA, plasmid or linear DNA, viral or nonviral delivery vehicles, and living cells. Each component may have its own manufacturing process, control strategy, impurity profile, storage condition, release method, and supplier.
Editor proteins and recombinases may require testing for identity, purity, aggregation, biological activity, process residuals, and stability. Messenger RNA and guide RNAs may require controls for sequence, integrity, purity, capping, polyadenylation, residual template, enzymes, and functional activity. Delivery vehicles introduce further attributes related to composition, particle properties, encapsulation, potency, residual solvents, and physical stability. An ex vivo edited-cell product adds cell identity, viability, phenotype, potency, microbial safety, residual editing components, and genomic heterogeneity.
The system must also work as a coordinated whole. Component concentration, ratio, order of addition, exposure time, delivery conditions, cell density, culture conditions, and recovery period may alter both the frequency and quality of the edit. A process that raises the percentage of on-target integration could also change copy-number distribution, increase incomplete products, or select for a subpopulation with different growth characteristics.
Potency testing may therefore need to connect several levels of activity. Component-specific assays can establish that an editor, recombinase, RNA, or delivery vehicle performs its intended biochemical function, but the final product may require evidence of successful delivery, correct integration, functional expression, and a relevant cellular response. No single reporter assay will necessarily represent the complete mechanism of action.
Stability and logistics become equally complex. DNA, RNA, protein, nanoparticles, and cells may require different formulations, temperatures, shipping conditions, and in-use hold times. The clinically administered configuration must also be defined. Components may be premixed, co-packaged, assembled shortly before use, delivered sequentially, or introduced during a cell-manufacturing process. Each configuration changes the operational workflow and the associated control strategy.
These requirements create a particular role for development and manufacturing partners capable of coordinating several modalities. Relevant capabilities may include DNA, RNA, protein, vector, nanoparticle, or cell manufacturing; integrated analytical development; genomic characterization; raw-material qualification; potency-assay development; regulatory documentation; and technology transfer among specialized facilities. The ability to manage the interfaces among components may prove as important as expertise in any single component.
FDA’s final genome-editing guidance treats the editor, targeting element, donor DNA, and delivery system as interconnected parts of product development and recommends appropriate manufacturing, testing, and stability information for relevant components. EMA guidance similarly addresses combinations of proteins, RNAs, vectors, linear DNA, plasmids, and delivery systems within genome-editing products.16,19
Comparability While the Architecture Evolves
Large-payload platforms are still being optimized. A developer may change the editor sequence, guide design, recombinase or transposase, attachment site, donor configuration, promoter, component ratio, delivery vehicle, or manufacturing process in pursuit of better efficiency, specificity, or scalability.
Such changes may affect more than the percentage of cells edited. They could alter integration location, orientation, copy-number distribution, junction structure, off-target activity, cargo completeness, structural-variant profile, cellular toxicity, functional expression, or potency. A more active editor is not necessarily comparable to its predecessor if it produces a materially different distribution of genomic products.
Comparability planning should begin before the platform is considered final. Reference materials, retained samples, and analytical continuity can preserve the ability to evaluate later changes. When possible, developers may benefit from avoiding simultaneous changes to the editor, donor, delivery system, and manufacturing process, because multiple overlapping changes make it harder to attribute observed differences.
The FDA’s 2026 draft guidance on prior knowledge states that manufacturing and analytical information may be leveraged across genome-editing programs when there is sufficient similarity in molecular structure, manufacturing processes, facilities, equipment, and operating conditions. Relevant knowledge may come from a sponsor’s own platform, a contract development and manufacturing organization, a supplier, a technology provider, a master file, or published literature. The guidance does not support automatic bridging, and the relevance of prior knowledge must still be scientifically justified.20
Comparability assessments may need to address component identity and purity, biological activity, integration efficiency, insertion architecture, copy number, genomic integrity, potency, and stability. The appropriate package will depend on the nature of the change and the risks of the product. A modification that affects only a well-characterized manufacturing input may require a different evaluation from a redesigned editor with altered targeting or catalytic behavior.
What Will Determine Which Platforms Advance?
The largest reported cargo or highest nominal integration rate will not by itself identify the most viable therapeutic platform. Systems will need to balance cargo capacity, targeting flexibility, efficiency, specificity, orientation control, copy-number distribution, structural integrity, delivery feasibility, manufacturability, component stability, analytical tractability, and functional benefit.
A platform with moderate integration efficiency could be more developable if it produces a narrow, predictable range of genomic products and can be delivered transiently. A more active system may be less attractive if it creates mixed orientations, partial cargo, off-target integration, or structural abnormalities that are difficult to detect and control. Similarly, an analytically precise system may remain impractical if its complete machinery cannot reach the relevant tissue.
Different architectures may ultimately serve different applications rather than competing for one universal role. Ex vivo cell products can accommodate delivery, enrichment, and characterization strategies that may be impossible in vivo. Gene-sized insertion, regional replacement, and megabase-scale rearrangement also represent different levels of biological and manufacturing complexity. Platform selection will likely depend on the target cell, disease mechanism, required cargo size, genomic operation, delivery route, and acceptable product heterogeneity.
Large-payload genome engineering has already moved beyond a theoretical extension of CRISPR. Multiple systems can insert gene-sized cargo or perform larger genomic operations in human cells. Their therapeutic development now depends on making those changes reproducible, characterizable, manufacturable, and controllable. The future of genome engineering may be measured less by the maximum number of bases a system can move than by how completely developers can define every genomic product it creates.
References
1. Rothstein, Rodney J. “One-Step Gene Disruption in Yeast.” Methods in Enzymology. 101: 202–211 (1983).
2. Witte, Isaac P, et al. “Programmable Gene Insertion in Human Cells with a Laboratory-Evolved CRISPR-Associated Transposase.” Science. 388: eadt5199 (2025).
3. Fanton, Alison, et al. “Site-Specific DNA Insertion into the Human Genome with Engineered Recombinases.” Nature Biotechnology. 6 Nov. 2025.
4. Perry, Nicholas T, et al. “Megabase-Scale Human Genome Rearrangement with Programmable Bridge Recombinases.” Science. 391: eadz0276 (2026).
5. Liu, Jason, et al. “Integration of Therapeutic Cargo into the Human Genome with Programmable Type V-K CAST.” Nature Communications. 16: 2427 (2025).
6. Anzalone, Andrew V, et al. “Programmable Deletion, Replacement, Integration and Inversion of Large DNA Sequences with Twin Prime Editing.” Nature Biotechnology. 40: 731–740 (2022).
7. Yarnall, Matthew TN, et al. “Drag-and-Drop Genome Insertion of Large Sequences without Double-Strand DNA Cleavage Using CRISPR-Directed Integrases.” Nature Biotechnology. 41: 500–512 (2023).
8. Pandey, Smriti, et al. “Efficient Site-Specific Integration of Large Genes in Mammalian Cells via Continuously Evolved Recombinases and Prime Editing.” Nature Biomedical Engineering. 9: 22–39 (2025).
9. Liu, Bin, et al. “Prime Assembly with Linear DNA Donors Enables Large Genomic Insertions.” Nature. 29 Apr. 2026.
10. Guideline on the Quality, Non-Clinical and Clinical Aspects of Gene Therapy Medicinal Products. European Medicines Agency. Mar. 2018.
11. Kosicki, Michael, Kärt Tomberg, and Allan Bradley. “Repair of Double-Strand Breaks Induced by CRISPR–Cas9 Leads to Large Deletions and Complex Rearrangements.” Nature Biotechnology. 36: 765–771 (2018).
12. Tsai, Hsiu-Hui, et al. “Whole Genomic Analysis Reveals Atypical Non-Homologous Off-Target Large Structural Variants Induced by CRISPR–Cas9-Mediated Genome Editing.” Nature Communications. 14: 5183 (2023).
13. Cullot, Grégoire, et al. “Genome Editing with the HDR-Enhancing DNA-PKcs Inhibitor AZD7648 Causes Large-Scale Genomic Alterations.” Nature Biotechnology. 43: 1778–1782 (2025).
14. Pinglay, Sudarshan, et al. “Multiplex Generation and Single-Cell Analysis of Structural Variants in Mammalian Cells.” Science. 387: eado5978 (2025).
15. Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing: Draft Guidance for Industry. U.S. Food and Drug Administration. 14 Apr. 2026.
16. Guideline on Quality, Non-Clinical and Clinical Requirements for Investigational Advanced Therapy Medicinal Products in Clinical Trials. European Medicines Agency.. Feb. 2025.
17. Chemistry, Manufacturing, and Control (CMC) Information for Human Gene Therapy Investigational New Drug Applications (INDs): Guidance for Industry. U.S. Food and Drug Administration. Jan. 2020.
18. Oliynyk, Roman Teo, et al. “Plasmid2MC: Efficient Cell-Free Generation of High-Purity Minicircle DNA for Genome Editing in Mammalian Cells.” Communications Biology. 8: 1778 (2025).
19. Human Gene Therapy Products Incorporating Human Genome Editing: Guidance for Industry. U.S. Food and Drug Administration. Jan. 2024.
20. Leveraging Prior Knowledge in the Development of Human Gene Therapy Products Incorporating Genome Editing: Draft Guidance for Industry. U.S. Food and Drug Administration. Jun. 2026.
