{
  "id": "base-editing",
  "title": "Base Editing",
  "lang": "en",
  "source_lang": "en",
  "summary": "Base editing enables targeted nucleotide substitutions in DNA or RNA by coupling programmable sequence recognition to enzymes that chemically modify nucleobases.",
  "category": "science",
  "subcategory": "biology",
  "type": "concept",
  "aliases": [
    "nucleobase editing",
    "targeted base editing"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/base-editing/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/base-editing/",
      "json": "https://aiwiki.page/zh/base-editing.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Base editing is a family of genome-editing technologies that change individual bases in DNA without ordinarily requiring a programmed break across both DNA strands or a separate donor DNA template. Most established systems combine a programmable DNA-binding component with an enzyme that chemically modifies a nucleobase. Cellular repair or replication then converts the modified base into a stable sequence change. Related systems edit RNA, usually producing a temporary change rather than altering the genome.\n\nThe technology is important because many biologically consequential mutations involve single-base substitutions. Base editors can correct some such variants, introduce protective changes, disrupt selected genetic functions, and systematically test how sequence differences affect cells and organisms. Their practical capabilities depend on which chemical conversions an editor supports, where it can bind, which bases lie within its activity window, and how a particular cell processes the resulting intermediate.\n\nFirst demonstrated in widely adopted forms in 2016 and 2017, DNA base editing expanded the capabilities of CRISPR beyond DNA cleavage. Subsequent developments have broadened its targeting range, improved its specificity, enabled editing of mitochondrial DNA, and supported experimental therapies. Nevertheless, base editing is not synonymous with exact single-letter correction: unwanted changes within the target region, editing elsewhere, delivery barriers, and variable biological responses remain central considerations."
    },
    {
      "heading": "Origins and relationship to other editing methods",
      "text": "Conventional CRISPR genome editing uses an RNA-guided nuclease, commonly Cas9, to cut DNA. Repair through non-homologous end joining often generates small insertions or deletions, making this approach useful for disrupting a gene. Precise replacement can instead use homology-directed repair with a supplied template, but its efficiency depends strongly on cell type, cell-cycle state, and delivery of the repair material. Base editing arose from a different strategy: localize a chemical reaction to a chosen DNA sequence and exploit the cell’s own processing machinery to establish the desired substitution.\n\nA foundational report by Alexis Komor and colleagues in David Liu’s laboratory appeared in 2016. It described cytosine base editors that joined a cytidine deaminase to a catalytically impaired Cas9. The first-generation construct, BE1, established targeted cytosine deamination; BE2 added an inhibitor of uracil removal; and BE3 used a Cas9 nickase to bias processing toward the desired outcome. Also in 2016, Keiji Nishida and colleagues reported Target-AID, a related strategy using the cytidine deaminase PmCDA1. These studies demonstrated that programmable DNA recognition could direct base conversion without relying on a conventional double-strand break.\n\nIn 2017, Nicole Gaudelli and colleagues, also working in Liu’s laboratory, introduced adenine base editors. Their development required substantial protein engineering because no naturally occurring enzyme was known to provide the required efficient adenine-to-inosine conversion in DNA. Together, cytosine and adenine editors made the two classes of transition substitution accessible: exchanges within the pyrimidines, cytosine and thymine, or within the purines, adenine and guanine.\n\nBase editing differs from prime editing, introduced in 2019. Prime editors use a Cas9 nickase fused to a reverse transcriptase and a specialized guide RNA that encodes the intended sequence change. They can install all classes of base substitution and certain small insertions and deletions, whereas conventional base editors have a narrower chemical repertoire. Base editors, however, do not require a reverse-transcribed editing template, and their relative simplicity can be advantageous at compatible targets. Neither method is universally superior: efficiency, product purity, sequence context, delivery, and the intended change determine their suitability.\n\nBase editing also differs from epigenetic editing, which changes regulatory marks or recruits regulatory proteins without necessarily altering the DNA sequence. A DNA base edit changes the sequence itself and can be inherited by daughter cells. An RNA base edit usually disappears as the edited RNA is degraded and replaced. These distinctions are fundamental when assessing whether an intervention is permanent, reversible, or transmissible through reproduction."
    },
    {
      "heading": "Programmable targeting and the editing window",
      "text": "Most nuclear DNA base editors contain three functional elements: a programmable targeting module, a base-modifying enzyme, and components that influence localization or DNA processing. The targeting module is usually a modified CRISPR-associated protein. A guide RNA specifies a complementary DNA sequence, while a nearby protospacer-adjacent motif, or PAM, permits recognition by the CRISPR protein. Widely used Streptococcus pyogenes Cas9 typically recognizes an NGG PAM, where N represents any DNA base.\n\nAfter recognition, the guide RNA pairs with one DNA strand and displaces the other, forming an R-loop. This exposed single-stranded DNA provides the substrate for many deaminases, which act poorly on intact double-stranded DNA. In common Cas9 base-editor architectures, the deaminase modifies bases on the displaced, non-guide-complementary strand. A nickase version of Cas9 can then cut the opposite strand at one position. The chemical modification and this strand-specific nick serve different purposes: one creates the altered base, while the other influences which strand the cell uses during processing.\n\nEditing is concentrated within an activity window, rather than at a single invariant nucleotide. For many early Cas9 cytosine editors, substantial activity occurred around positions 4–8 of the approximately 20-nucleotide protospacer, counting from the PAM-distal end. This is a useful historical reference, not a universal rule. The window depends on the deaminase, linker, Cas protein, sequence context, and experimental conditions. Some editors have deliberately narrow windows; others edit over a broader region.\n\nSeveral editable bases may lie within that window. If the intended target is one cytosine among three, the editor may generate molecules carrying the desired conversion alone, combinations of conversions, or no conversion. Changes to neighboring bases are called bystander edits. They are on-target events in the spatial sense but may be unwanted biologically. A silent bystander substitution in one gene can be harmless, whereas another may alter an amino acid, a splice signal, or a regulatory sequence.\n\nTarget accessibility also depends on chromatin, local sequence, and protein occupancy. Engineered Cas9 variants that recognize alternative PAMs expand the number of reachable bases, while Cas12-derived systems provide different PAM preferences and targeting geometries. Relaxing PAM requirements does not remove every constraint: binding specificity, editing-window placement, enzyme sequence preferences, and delivery remain relevant. A base is therefore “targetable” only when a compatible recognition site places it in a productive editing configuration."
    },
    {
      "heading": "Cytosine base editors and repair-dependent outcomes",
      "text": "Cytosine base editors, commonly abbreviated CBEs, usually convert a C•G base pair into T•A. Their initiating reaction is deamination: removal of an amino group from cytosine produces uracil. Although uracil is a normal RNA base, it is not a canonical DNA base. When encountered as a template, however, it generally directs incorporation of adenine, allowing the original cytosine position to become thymine after further processing or replication.\n\nThe initial product is a U•G mismatch, not an immediately completed T•A substitution. Cellular DNA repair can recognize uracil as a lesion and remove it. In particular, uracil-DNA glycosylase initiates base excision repair by cleaving the bond between uracil and its sugar. The resulting abasic site may be repaired back to the original sequence or processed into other products. Many CBEs therefore contain one or more copies of a uracil glycosylase inhibitor, usually called UGI, to protect the uracil intermediate.\n\nA nick in the opposite, guanine-containing strand further biases processing. Repair synthesis can replace the guanine with adenine using the uracil-containing strand as the template. Subsequent processing or DNA replication establishes T•A. This dependence on cellular pathways explains why the same editor and target sequence can behave differently in distinct cell types. The frequency of the desired substitution reflects not only deaminase activity but also the balance among competing repair and replication processes.\n\nEarly cytosine editors frequently used rat APOBEC1, while other architectures incorporated PmCDA1 or members of the APOBEC family. These enzymes differ in their preferences for neighboring bases, activity-window widths, and propensity to act on unintended substrates. Engineering has produced editors with improved activity in previously difficult sequence contexts, narrower windows, altered preferences, or reduced collateral activity. Later BE4-type designs also improved product purity through architectural changes, including additional uracil glycosylase inhibition.\n\nCytosine editing is not chemically restricted to a single final outcome in every construct. If uracil is removed and the abasic intermediate is processed through selected repair pathways, C-to-G or other substitutions can occur. Dedicated C-to-G base editors exploit this principle rather than suppressing it. Such systems extend base editing into transversions, which exchange a pyrimidine for a purine or vice versa. Their performance is often more dependent on sequence context and repair state than that of established transition editors.\n\nReported substitution direction must be interpreted with respect to the strand being described. A C-to-T conversion on one DNA strand appears as G-to-A when read from its complement. Consequently, a cytosine editor can address either description of the same base-pair change if a suitable targeting orientation is available."
    },
    {
      "heading": "Adenine base editors and expansion of editing chemistry",
      "text": "Adenine base editors, or ABEs, generally convert an A•T pair into G•C. Their initiating reaction converts adenine to inosine, whose base is hypoxanthine. DNA polymerases generally read inosine as guanine, enabling replacement of the opposite thymine with cytosine and eventual establishment of a G•C pair. As with cytosine editing, the complementary description is equally valid: an A-to-G change on one strand is T-to-C on the other.\n\nThe central engineering challenge was to obtain an enzyme capable of efficiently deaminating adenine in DNA. The starting point was TadA, a bacterial enzyme whose natural substrate is adenine within transfer RNA. Through directed evolution, researchers altered its substrate recognition and activity. The resulting ABE7.10 architecture combined an evolved TadA component with a wild-type TadA partner and Cas9 nickase. It demonstrated that an enzyme adapted to RNA could be reconfigured for programmable DNA modification.\n\nLater variants, including ABE8-family editors, increased activity and broadened useful editing contexts. Greater catalytic activity can enable effective editing with shorter exposure or in difficult cells, but it does not automatically mean greater precision. A highly active deaminase may edit more positions within its window or act more readily on unintended substrates. Conversely, reducing activity to improve specificity can lower the desired editing yield. Editor selection therefore involves balancing efficiency, bystander editing, off-target activity, and the biological importance of each possible product.\n\nClassical ABEs often produce relatively few insertions and deletions compared with double-strand-break editing, but their outcomes are not perfectly uniform. The initiating A•T-to-I•T mismatch must still be resolved within a living cell. Local DNA structure, target binding, nicking, replication, and repair influence the final distribution. Editing can also be asymmetric between the two copies of a gene or among different cells in a population.\n\nThe repertoire of base editing has expanded beyond the original C-to-T and A-to-G reactions. Dual-function editors combine adenine- and cytosine-editing activities, allowing both classes of transition within a targeted region. Other architectures alter repair processing or engineer deaminases to support additional substitutions. These approaches are useful for generating sequence diversity, although a broader repertoire can also increase the number of mixed or unwanted products.\n\nThe term base editing consequently describes a technological principle rather than a single biochemical reaction. Established transition editors act mainly through deamination, while newer systems may combine base modification with deliberately redirected repair. Claims that a platform can install a particular substitution must be distinguished from evidence that it does so efficiently and cleanly across many sequence contexts, cell types, and delivery conditions."
    },
    {
      "heading": "RNA and mitochondrial base editing",
      "text": "RNA editing extends programmable base modification beyond permanent genomic changes. Adenosine-to-inosine RNA editing often uses enzymes from the ADAR family, whose natural roles include modifying double-stranded RNA. Because the translation machinery and many sequencing workflows interpret inosine similarly to guanosine, an A-to-I change can alter a protein-coding message without changing its DNA template. In 2017, the REPAIR platform coupled catalytically inactive Cas13 to an engineered ADAR deaminase domain, establishing one route to programmable RNA editing.\n\nOther RNA-editing strategies recruit endogenous ADAR proteins with designed RNAs rather than delivering a large enzyme fusion. Engineered systems have also enabled cytidine-to-uridine RNA editing. Their consequences can include changes to codons, splice behavior, or other RNA functions. Because RNA molecules turn over, the effect is usually transient unless the editing machinery remains present. This provides a fundamentally different duration of action from genomic editing but may require repeated administration or sustained expression for persistent effects.\n\nSpecificity problems also differ. An RNA editor may modify unintended transcripts or additional adenosines near the target. The amount and lifetime of the target messenger RNA, its secondary structure, and access by cellular proteins influence editing. A change measured across total RNA does not necessarily translate directly into the same fraction of altered protein, because transcript stability and translation efficiency can differ.\n\nEditing the genome of the mitochondrion presents a separate problem. Standard CRISPR systems depend on guide RNAs, whose reliable delivery into the mitochondrial matrix has been a major obstacle. In 2020, Beverly Mok and colleagues described DddA-derived cytosine base editors, or DdCBEs. These systems use a bacterial toxin-derived deaminase capable of acting on double-stranded DNA, combined with programmable transcription activator-like effector DNA-binding proteins.\n\nDddA is split into two portions and attached to proteins that bind nearby DNA sequences. Bringing the halves together at the target reconstitutes local activity, while mitochondrial targeting signals direct the constructs into the organelle. The original systems primarily produced C•G-to-T•A changes and exhibited sequence-context preferences. Later engineering expanded targeting options and improved specificity. TALE-linked deaminase systems have also enabled adenine editing in mitochondrial DNA.\n\nMitochondrial genomes occur in many copies per cell, so outcomes are often described through heteroplasmy, the coexistence of different mitochondrial sequence variants. Shifting the proportion of genomes carrying a variant can have a different biological effect from changing every copy. Measurements must therefore distinguish editing percentage from the number and distribution of affected mitochondrial genomes, while also assessing possible changes in nuclear DNA and unintended mitochondrial sites."
    },
    {
      "heading": "Delivery and experimental measurement",
      "text": "An effective base editor must reach the appropriate cells, enter the relevant compartment, and remain active long enough to produce the intended change. Editors can be delivered as DNA encoding the machinery, as messenger RNA together with guide RNA, or as preassembled protein–RNA complexes. These formats differ in stability, duration of expression, manufacturing requirements, and opportunities for unwanted activity. Persistent expression may increase editing but also prolong exposure to off-target reactions.\n\nViral delivery is constrained by cargo size. Adeno-associated virus, or AAV, has a packaging capacity of roughly 4.7 kilobases, whereas many Cas9-based editor expression cassettes are too large for a single conventional AAV vector. Split-editor strategies distribute components between vectors and reassemble them inside cells, sometimes using split inteins. This introduces an additional requirement: the same cell must receive both components in suitable amounts.\n\nNonviral approaches include lipid nanoparticles, which can deliver editor mRNA and guide RNA, and electroporation, which transiently permeabilizes cell membranes. Electroporation is widely used for cells edited outside the body, including immune cells and blood-forming progenitors. Lipid nanoparticles have been particularly useful for delivery to the liver. Efficient delivery to many other tissues remains difficult because anatomical barriers, uptake pathways, intracellular trafficking, and cell-specific tolerability differ.\n\nEditing performance is commonly measured by targeted DNA sequencing. Several quantities must be distinguished. Editing efficiency describes how often the intended change occurs; product purity describes the distribution of desired and undesired products; and functional correction measures the resulting biological effect. A high desired-base frequency can coexist with substantial bystander editing. Conversely, a modest editing percentage may be biologically meaningful if edited cells produce a diffusible protein or gain a selective advantage.\n\nBulk measurements can conceal differences among cells. A result showing 50% edited sequence reads might reflect monoallelic editing in most diploid cells, biallelic editing in half the cells, or a more complex mixture involving copy-number variation. Single-cell assays, clonal analysis, and phased sequencing help resolve these possibilities. In diploid or polyploid systems, linking the desired edit to nearby variants can also establish which allele was changed.\n\nReliable assessment extends beyond a short sequencing amplicon. Depending on the application, studies examine large deletions, rearrangements, transcriptome changes, protein function, viability, and long-term cell behavior. Appropriate controls separate effects of the intended edit from delivery toxicity, editor expression, and background variation. In therapeutic development, the relevant endpoint is not simply whether a nucleotide changed, but whether the resulting cell population has the intended function and an acceptable safety profile."
    },
    {
      "heading": "Research, biotechnology, and agriculture",
      "text": "Base editing provides a way to investigate causation at individual nucleotide positions. Classical gene knockout asks what happens when a gene is disrupted; base editing can ask what happens when one amino acid, splice signal, or regulatory base is changed. This resolution is valuable because many naturally occurring variants alter function subtly rather than abolishing it. Introducing such variants into otherwise comparable cells helps distinguish functional effects from associations observed in populations.\n\nPooled screens use libraries of guide RNAs to generate many targeted changes across a cell population. Cells are then assessed for a phenotype, such as altered growth, drug sensitivity, or expression of a surface protein. Sequencing identifies which guides, and ultimately which edits, are associated with the measured outcome. These experiments can map important residues in enzymes, identify resistance mutations, and probe noncoding regulatory regions. Their interpretation requires direct attention to editing windows: one guide may produce several genotypes rather than a single defined variant.\n\nBase editors can also disrupt genes without creating a programmed double-strand break. Cytosine editing can convert certain codons into premature stop codons, including CAA or CAG into TAA or TAG in a suitable orientation. Editing conserved splice donor or acceptor sequences can alter RNA splicing, while changes to an initiation codon may interfere with translation. These approaches depend on transcript structure and may not yield complete loss of function; alternative splicing, downstream initiation, or truncated proteins can complicate the outcome.\n\nIn biotechnology, targeted substitutions help investigate and improve enzymes, microbial production pathways, and cellular traits. Diversifying editors can generate a local range of sequence variants for selection rather than installing just one predetermined substitution. Such methods complement conventional mutagenesis by concentrating variation within a chosen gene or region. They do not eliminate the need to characterize linked changes or determine which alteration causes the selected phenotype.\n\nPlant researchers have used cytosine and adenine editing in crops including rice, wheat, maize, and tomato. Targets have included acetolactate synthase genes associated with herbicide response, genes influencing disease susceptibility, and pathways affecting grain quality or plant architecture. Editing outcomes depend on transformation efficiency, regeneration capacity, ploidy, and the availability of a compatible target sequence. In polyploid crops such as bread wheat, several related gene copies may need to be considered.\n\nA final edited plant can sometimes be obtained without retaining the DNA construct used to produce the edit, for example through segregation in later generations or transient delivery. That genetic status does not by itself determine agronomic performance, environmental effects, or legal classification. Those questions depend on the resulting trait, the production method, and the regulatory framework of the relevant jurisdiction."
    },
    {
      "heading": "Medical development and clinical translation",
      "text": "Medical uses fall into two broad delivery categories. In ex vivo editing, cells are collected, modified outside the body, assessed, and then administered. In in vivo editing, the machinery is delivered directly to the patient. Ex vivo workflows allow more control over the edited cell product but require cell collection and specialized manufacturing. In vivo approaches can reach cells that cannot readily be removed and replaced, yet offer less opportunity to inspect or select individual edited cells before exposure.\n\nBlood disorders provide prominent examples because hematopoietic stem cells can be collected and can generate long-lived blood-cell lineages. Experimental base-editing strategies for sickle cell disease have included changing the disease-associated hemoglobin sequence to the non-sickling Hb G-Makassar variant rather than restoring the exact common sequence. Other approaches modify regulatory elements to increase fetal hemoglobin. Similar regulatory strategies are relevant to thalassemia, although each intervention has its own editing chemistry, biological consequences, and development history.\n\nBase editing has also been used to manufacture investigational donor-derived T cells for cancer treatment. Removing selected cell-surface or signaling components can help adapt donor cells for use in recipients and support combination with other therapies. A report published in 2023 described base-edited CAR T cells used in children with relapsed T-cell acute lymphoblastic leukemia. The application illustrated the value of making several targeted disruptions while avoiding the multiple programmed double-strand breaks that can increase rearrangement risks.\n\nFor in vivo editing, the liver has been an early focus because delivery technologies can reach hepatocytes effectively. Investigational programs have targeted genes such as PCSK9, whose reduced activity lowers circulating low-density lipoprotein cholesterol. Preclinical studies established that base editing could produce durable changes in liver gene function, and early human studies have evaluated liver-directed adenine editors. Target engagement or a change in a blood biomarker is distinct from proof of long-term clinical benefit, which requires appropriate follow-up.\n\nNot every pathogenic variant is directly correctable with available base editors. A candidate change must match the editor’s chemistry and targeting constraints, and nearby substitutions must be evaluated. In some cases, altering a compensatory pathway or introducing a protective variant is more feasible than restoring the original sequence. Such strategies require evidence about the consequences of the new genotype rather than an assumption that any reduction in a disease-associated marker is sufficient.\n\nClinical evaluation must address acute delivery reactions, immune responses, unintended edits, durability, and the long-term behavior of edited cells. DNA edits may persist after the editor has disappeared, making extended monitoring important. Manufacturing consistency is also central: an edited product is a population of cells or a delivery formulation with measurable variation, not merely a molecular design. Findings in cultured cells or animal models establish particular aspects of feasibility but do not alone establish safety or effectiveness in humans."
    },
    {
      "heading": "Specificity, biological limits, and governance",
      "text": "Unintended editing arises through several mechanisms that require different tests. Guide-dependent off-target editing occurs when the targeting complex binds a partially matching sequence elsewhere in the genome and modifies accessible bases there. Guide-independent editing can occur when a deaminase acts on exposed DNA or RNA without productive recognition of the intended target. Bystander edits form a third category: they occur within the intended target region but alter bases other than the desired one.\n\nSome early cytosine and adenine editors produced substantial unintended RNA editing under particular expression conditions. Protein engineering reduced this activity in later variants. Related strategies have reduced guide-independent DNA deamination or narrowed editing windows. These improvements are editor-specific: the properties of one construct cannot be assumed for every system bearing the same broad label. Dose and exposure duration also matter, so specificity measured under prolonged laboratory expression may differ from that after transient delivery.\n\nDetection is intrinsically incomplete. Computational searches identify plausible sequence-related off-target sites, while biochemical and cellular assays nominate additional candidates. Targeted deep sequencing can measure rare changes at selected locations, but it cannot discover every possible site. Whole-genome sequencing has broad coverage yet limited sensitivity for rare variants in mixed populations and can be confounded by pre-existing mutations. Strong assessments therefore combine complementary methods and compare edited samples with suitable controls.\n\nAvoiding a programmed double-strand break reduces an important source of genome damage, but it does not make base editing damage-free. Nicking, abasic intermediates, repair processing, and occasional coincident lesions can produce insertions, deletions, or other abnormalities. Cellular stress and selection can also change which edited cells survive. The significance of an unintended change depends on its position, frequency, functional effect, and whether the affected cell can persist or expand.\n\nSomatic editing and heritable editing raise different governance questions. Somatic interventions target the treated individual, whereas editing embryos or reproductive cells can transmit changes to descendants. Embryo editing additionally faces mosaicism, in which cells within the same embryo acquire different genotypes. The existence of a method that avoids conventional double-strand breaks does not resolve uncertainties about developmental effects, intergenerational consequences, consent, or oversight.\n\nAccess, intellectual property, and manufacturing capacity also shape the technology’s use. Personalized editors may require extensive validation for very small patient populations, while ex vivo products depend on specialized facilities and coordinated clinical infrastructure. Agricultural regulation varies among jurisdictions in how it treats sequence changes, retained foreign DNA, and production methods. Across these settings, evaluation depends on the particular editor, delivery system, target, and resulting biological product rather than on the term “base editing” alone."
    }
  ],
  "links": [
    "genome-editing",
    "dna",
    "enzyme",
    "rna",
    "mutation",
    "crispr",
    "non-homologous-end-joining",
    "gene",
    "homologous-recombination",
    "protein",
    "prime-editing",
    "epigenetics",
    "amino-acid",
    "chromatin",
    "dna-repair",
    "base-excision-repair",
    "dna-replication",
    "transfer-rna",
    "directed-evolution",
    "codon",
    "messenger-rna",
    "mitochondrion",
    "lipid-nanoparticle",
    "dna-sequencing",
    "allele",
    "phenotype",
    "rna-splicing",
    "hemoglobin",
    "thalassemia",
    "t-cell",
    "genome",
    "mosaicism"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}