Base editors are a new family of programmable genome editing tools that fuse ssDNA (single-stranded DNA) modifying enzymes to catalytically inactive CRISPR-associated (Cas) endonucleases to induce highly efficient single base changes. With dozens of base editors now reported, it is apparent that these tools are highly modular; many combinations of ssDNA modifying enzymes and Cas proteins have resulted in a variety of base editors, each with its own unique properties and potential uses. In this perspective, we describe currently available base editors, highlighting their modular nature and describing the various options available for each component. Furthermore, we briefly discuss applications in synthetic biology and genome engineering where base editors have presented unique advantages over alternative techniques.
Base editors are new genome editing tools capable of introducing single nucleotide variants (SNVs) with high efficiency in a programmable manner. They are an expansion of clustered regularly interspaced short palindromic repeats (CRISPR)-derived tools, incorporating various elements of this revolutionary technology that have made it so ubiquitous in laboratory applications. CRISPR-associated (Cas) endonucleases utilize a guide RNA (gRNA) molecule to bind to a target genomic DNA sequence and introduce a double-stranded DNA break (DSB) .
For this process to occur, the target sequence of interest (defined as the protospacer) must be both complementary to the sequence of the gRNA and adjacent to a Cas protein-specific recognition sequence motif (the protospacer adjacent motif or PAM). The Cas9–gRNA complex searches the genome, finds its matching protospacer-PAM sequence, and invades the genomic DNA to form an R-loop, where the gRNA engages in Watson–Crick base pairing with one strand of the protospacer. Once bound, the Cas protein introduces a DSB to initiate the first step of traditional genome editing. DSB-reliant genome editing techniques then utilize one of two repair pathways to manipulate the genome following DSB introduction. Non-homologous end joining (NHEJ) will repair the break without requiring a template but results in stochastic nucleotide insertions or deletions (indels) at the site of the DSB, while homology directed repair (HDR) will utilize a user-provided homologous DNA template for precise genomic manipulations .
Shortcomings of DSB-reliant genome editing include the difficulty in controlling indel sequences and sizes with NHEJ, the low efficiency of HDR in post-mitotic cells, and the inherent competition between these two repair pathways often resulting in intractable mixtures of genome editing products [3,4]. Base editors were developed to provide an alternative, DSB-free mechanism for making site-specific single nucleotide changes via chemical modification of the DNA nucleobases. This perspective will cover the many iterations of base editors, focusing on their modularity and flexibility, while also highlighting some unique applications these tools are primed for.
Considering the modular nature of base editors
Base editors are composed of two key elements: a Cas protein for programmable DNA binding and a single-stranded DNA (ssDNA) modifying enzyme for targeted nucleobase alteration. The original cytidine base editor (BE1)  used the catalytically inactive Streptococcus pyogenes Cas9 (dCas9, capable of binding DNA without cleavage) and fused it via its N-terminus to the cytidine deaminase enzyme APOBEC1 from Rattus norvegicus for these roles, respectively. To edit, first dCas9 brings the base editor complex to a genomic locus of interest determined by the sequence of its gRNA and the presence of an NGG PAM. This is followed by the dCas9–gRNA invading the double-stranded target DNA and forming an R-loop, exposing a small window of single-stranded nucleotides on the opposing displaced strand. These nucleotides, no longer participating in Watson–Crick base pairing, are now suitable substrates for rAPOBEC1, which deaminates any cytidine bases within this ∼5-nt window into uracils, creating a U·G mismatch in the genomic DNA (Figure 1). Since uracil, a non-canonical DNA base, has the same base pairing properties as the canonical DNA base thymine, DNA replication and/or repair of the resulting U·G mismatch will result in an overall C·G to T·A base-pair edit when the modified uracil-containing strand is used as a template (Figure 1, blue inset). To increase editing efficiency in mammalian cells, where the DNA repair enzyme uracil DNA glycosylase (UNG) efficiently excises uracil from the U·G intermediate, two additional changes were made. First, a UNG inhibitor (UGI) peptide was fused to the C-terminus of the base editor to inhibit this repair pathway (to yield BE2, the second generation base editor ). Second, dCas9 was swapped for the nickase Cas9n (Cas9D10A), a variant that induces a nick in the DNA backbone of the G-containing strand of the U·G intermediate to favor replacement of this nicked strand by the cellular DNA replication and/or repair machinery (to yield BE3, the third generation base editor , Figure 1 orange inset). Additional studies revealed UNG to be responsible for poor product selectivity (C·G to non-T·A genome editing products) at certain genomic loci . To combat this, an additional copy of UGI and optimized linker lengths were engineered in a follow-up study to yield BE4, the fourth-generation original cytosine base editor (CBE).
Basic base editor design, editing chemistries, and repair manipulation.
Cas protein modulation
While BE2 and BE3 have been used to make single nucleotide changes in a variety of bacterial [7–9] and mammalian cell [10–12] applications, their modular natures have also served as scaffolds for further engineering efforts to tune their functionality and expand their substrate scopes. Due to the strict gRNA design criteria for base editing imposed by the precise PAM-to-target C distance requirements (Figure 2, blue inset), an analysis of the ClinVar dataset [5,13] found that only 27% of pathogenic T·A to C·G SNVs are targetable using base editors relying on NGG PAMs . Fortunately, there are a host of well-characterized Cas endonucleases, both native and engineered, with different PAM requirements. In subsequent follow-up studies, it was demonstrated that the basic base editor architectures of BE2 and BE3 are compatible with the replacement of the SpCas9 module with a variety of these homologous and engineered Cas proteins (Figure 2). Specifically, the VQR and VRER mutant variants of SpCas9 allow for base editing at NGA and NGCG PAMs, respectively, and replacement of SpCas9 with the smaller Staphylococcus aureus homolog SaCas9 further expanded PAM availability to include NNGRRT and NNNRRT, albeit with a different editing window (Figure 2, blue inset) . It was additionally demonstrated that the BE2 architecture is compatible with dCas12a, (formerly dCpf1), making T-rich PAM sites amenable to base editing, again with a modified editing window (Figure 2, blue inset) .
Modularity of base editors.
Base editors derived from xCas9  and Cas9-NG  have also been reported, yielding the most flexible editors described to date with NG PAM requirements . Additional Cas protein engineering efforts, such as enhanced specificity variants , have also been shown to be compatible with various base editor architectures, further demonstrating their modularity.
Cytidine deaminase engineering efforts
Engineering efforts have also been directed at the nucleobase modifying module of base editors. In an early study, the activation-induced cytidine deaminase (AID) enzyme from Petromyzon marinus, pmCDA1, was fused to the C-terminus of SpCas9 to create the base editor Target-AID . The combination of an alternate DNA modifying enzyme and the overall architecture not only shifted the substrate window but also altered the sequence specificity of the base editor due to the inherent sequence motif preference of the pmCDA1 enzyme (Figure 2, red inset). While fusion of a UGI domain, as done in BE2 and BE3, helped increase the product selectivity to favor C·G to T·A outcomes, other researchers have taken advantage of UGI-free AID-derived base editors to mutagenize specific genomic loci and create locally hyper-mutagenized, genetically encoded libraries . A variety of homologs of rAPOBEC1 have been repurposed as base editors as well, such as hAID , hAPOBEC3G , hAPOBEC3A [22,23], hAPOBECB , hAPOBEC3B(catalytic domain) [23,24], and hAPOBEC3H  yielding base editors with alternate sequence preferences and the ability to efficiently edit methylated cytosines (Figure 2, red inset). Within the APOBEC family of base editors, engineering efforts have been undertaken to both widen and narrow the base editing window, as well as create sequence-specific editors . Linker length and composition have been explored as means to modulate activity window (Figure 2, grey inset) [12,27]. Additionally, mutations in APOBEC1 known to lower the catalytic activity were found to narrow the editing window , creating more precise base editors. Conversely, the fusion complex BE-PLUS , designed to recruit multiple antibody-rAPOBEC1 fusions to the Cas9n–gRNA complex, exhibited an expanded editing window (Figure 2, green inset). Engineering efforts have also been made to create sequence-specific deaminases. An hAPOBEC3A editor took advantage of this homolog's native sequence preference to engineer a base editor tailored for deaminating TC motifs . These extensive engineering efforts have produced a litany of mutant cytidine deaminases, each tailored for specific targets and/or applications (Figure 2, red inset).
Additional nucleobase modifications
While engineering the cytidine deaminase produced altered editing windows and sequence preferences, the target nucleobase remained fixed. However, given the modular nature of the base editor architecture, it was proposed that alternative ssDNA modifying enzymes that performed different nucleotide modifying chemistries could be engineered into new base editors capable of introducing additional base transformations. With the repertoire of naturally occurring ssDNA modifying enzymes exhausted, a new base editor was created by repurposing a tRNA adenosine deaminase enzyme for ssDNA modification . Escherichia coli TadA is an enzyme that deaminates the adenosine at the wobble position 34 of tRNAArg2 to inosine . In the context of DNA replication, inosine is interpreted as guanine (Figure 1, blue inset), opening up the possibility of using ecTadA to make an A·T to G·C base editor (ABE). Given its native tRNA substrate, the wild-type ecTadA-derived ABE was not competent for genome editing. However, following seven rounds of directed evolution in which selective pressure was placed on mutated libraries of ecTadA to modify ssDNA, two combinations of mutations were found to result in efficient A·T to G·C base editing, with slightly different editing windows (ABE7.9 and ABE7.10). Analogous to rAPOBEC1 in CBE, the optimized ecTadA* has been appended to several engineered Cas9 variants to create ABEs with varying PAM requirements and editing windows [32–35] (Figure 2, red inset). Between the development of ABE and CBE, all transition mutations (purine to purine or pyrimidine to pyrimidine) are now possible using base editor technologies, given the proximity to an appropriate PAM (Figure 3).
Currently available transition mutations.
Enhancing the specificity of base editors
While the use of high-fidelity Cas variants has been shown to greatly reduce gRNA-dependent off-target DNA base editing , recent studies reported gRNA-independent off-target C·G to T·A mutations in highly transcribed regions when using BE3 in both mouse embryos  and rice . In contrast, gRNA-independent off-target A·T to G·C mutations were not observed following ABE treatment, leading researchers to hypothesize that the engineering used to develop ABE may impart resistance to non-specific ssDNA adenosine deamination. Future work is needed to identify cytidine deaminase mutants that reduce this non-specific activity. In addition to off-target DNA editing, BE3 and ABE have been observed to induce transcriptome-wide gRNA-independent off-target RNA editing [38–40]. Fortuitously, engineering efforts identified a variety of mutations in both rAPOBEC1 and TadA* that nearly fully abrogated off-target transcriptome-wide editing events without sacrificing on-target DNA editing efficiency (Figure 2, red inset) .
Base editor applications
Base editors are capable of introducing SNVs efficiently and site-specifically into the genome of mammalian [41–43], plant [35,44–46], and bacterial cells [7,47]. It is important to note that different codon optimizations of the base editors have been shown to have dramatic effects on editing efficiencies when adapting these tools for use in new cell types/organoids [12,48]. Given their high efficiencies, these tools have been widely applied to develop research models for genetic studies. CBEs and ABEs have been extensively used to both correct pathogenic SNVs and introduce SNVs of clinical interest in model organisms [42,49–51]. Additionally, two techniques called CRISPR-STOP  and iSTOP  were recently developed that use CBEs to introduce premature stop codons to generate knockout models. By circumventing the use of indels for gene knockout, the risk of creating an in-frame indel with a functional transcript is avoided. Additionally, base editing induces lower rates of apoptosis than traditional DSB-mediated genome editing . With 99.7% of human genes targetable via iSTOP gRNAs , these tools may allow researchers greater control over genome editing outcomes and less cytotoxicity when creating knockout models for further study.
Base editors have been particularly popular in the agricultural field as well, since traditional HDR-mediated genome engineering approaches can be inefficient in plants due to difficulty in delivering donor templates [53,54]. Additionally, many agricultural mutations of interest, such as resistance to herbicides and enhancement of grain production, are SNVs, making base editors well suited for agricultural engineering. Consequently, several base editor constructs have been codon-optimized to work in rice [35,44,45,55,56], maize , tomato , wheat , cotton , potato  as well as model organism such as Arabidopsis thaliana and Brassica napus . Like other CRISPR-edited crops , base edited crops are expected to pass USDA GMO-regulation restrictions as they do not deliver viral genetic material. Given the efficiency and ease of use of base editing, BE-modified crops may someday be available in commercial markets.
Synthetic biology applications
Beyond using base editors to introduce SNVs for clinical and agricultural purposes, they have also been repurposed for applications in the field of synthetic biology. Due to their high ratio of SNV introduction to indel formation, CBEs have been used to create genetically encoded libraries [21,41]. Specifically, an AID-derived base editor with a wide editing window was programmed to mutagenize the BCR-ABL gene, a protein with known SNVs resulting in imatinib-resistance. Under imatinib selection, both known and previously unidentified drug-resistant SNVs were identified . Another unique application of base editors is as molecular recording devices . In one such system, stimulus in the form of light, phage infection, or small molecule treatment resulted in analog intensity of gRNA expression and subsequent accumulation of the base editor signal. By stacking base editing targets such that sequential stimuli are required to induce edits, an analog system was created capable of recording the order and intensity of stimuli events. This resulted in a molecular recording device more temporally precise than preceding wild-type Cas9-based recorders.
Current base editor limitations and perspectives moving forward
Base editors, despite their relatively recent inception, have already been widely used for agricultural, biomedical, clinical, and synthetic biology research. Base editing technologies have grown and swiftly expanded in the last 3 years, and there are many exciting opportunities for additional expansions of this technology. Perhaps the most pressing hurdle to overcome is the issue of off-target DNA editing by CBEs, which we envision may be alleviated through additional cytidine deaminase engineering efforts. There is also a need for new assays capable of detecting gRNA-independent off-target DNA base editing beyond cost-prohibitive whole-genome sequencing to aid in these future engineering efforts to alleviate these issues. For now, researchers must carefully consider whether a base editing application is tolerable to off-target DNA editing. Additionally, editors capable of facilitating transversion mutations (purine to pyrimidine or vice versa, which account for four of the six possible base conversions) currently do not exist. These presently unavailable tools may be developed from additional nucleobase chemistries or cellular DNA repair manipulation strategies.
Due to the modularity of these tools, it is likely that future Cas protein engineering efforts will be compatible with base editors to create additional variants with expanded targeting scopes (via new PAM recognition) and increased fidelity. The lowered toxicity of base editor intermediates (uracil and inosine) compared with DSBs opens up the possibility of multiplexing base editing and introducing multiple point mutations throughout the genome of a given cell.
Given the substantial engineering efforts that have been placed into its modular components, and the technologies emerging from its application in basic research and as tools for synthetic and chemical biology, base editors hold much promise for the genome engineering community.
Base editors are a new, modular tool for creating SNVs without introducing double-stranded breaks.
The modularity of base editors allows for control over the precision and specificity of genomic edits, making them flexible tools for basic research, genome engineering, and genetically encoded libraries.
Future engineering efforts on individual modules of these tools are expected to widen the genome editing substrate scope with additional edits made possible.
A·T to G·C base editor
activation-induced cytidine deaminase
cytosine base editor
clustered regularly interspaced short palindromic repeats
homology directed repair
insertions and/or deletions
any canonical nucleotide
non-homologous end joining
protospacer adjacent motif
purine nucleotide A or G
single nucleotide variant
uracil DNA glycosylase
M.E. was supported by the Cell and Molecular Genetics (CMG) Training Program (NIGMS, T32GM007240-41).
A.C.K. is a consultant of Pairwise Plants and Beam Therapeutics, companies that are developing and utilizing base editing technologies.