Base Editing Explained: Single-Base CRISPR Therapeutics (2026)

Base editing CRISPR is the most precise form of genome editing in clinical use today. It converts a single DNA letter into another without ever cutting both strands of the double helix. Since the landmark 2016 and 2017 papers from David Liu’s lab at the Broad Institute, base editors have moved from proof-of-concept to human trials targeting sickle cell disease, cardiovascular risk, and aggressive leukemias.

What this covers: how cytosine and adenine base editors are built, what each chemistry can and cannot accomplish, how base editing compares to prime editing, the real limits around off-target effects and delivery, and where the technology stands in 2026 clinical programs.

Context: From Cut-and-Repair CRISPR to Precision Base Editing

Standard CRISPR-Cas9 works by creating a double-strand break (DSB) at a genomic target. The cell then repairs that break through one of two imprecise pathways. Non-homologous end joining (NHEJ) stitches the ends back together but frequently introduces small insertions or deletions — useful for knocking out a gene, but poor for correcting a point mutation. Homology-directed repair (HDR) can install a precise edit using a donor template, but it only works efficiently in dividing cells and its overall rates remain disappointingly low in most therapeutic contexts.

Those limitations define the gap that base editing was designed to fill. Roughly 58% of pathogenic human point mutations documented in ClinVar are single-base transitions — the kind of single-letter swap where one purine or pyrimidine is exchanged for its chemical partner class (C↔T or A↔G). If you could convert one base to another without triggering a break, you could fix a large fraction of inherited disease mutations with far higher efficiency and far lower genotoxic risk than HDR.

That is exactly the logic Alexis Komor, Nishant Packer, and David Liu followed when they published the first cytosine base editor in Nature in 2016. A year later, Nicole Gaudelli and colleagues extended the concept to adenine, publishing the ABE system in Nature in 2017. Neither approach existed in nature — the adenine deaminase in ABE was entirely engineered through directed evolution. Both papers represent not just incremental improvement on CRISPR but a conceptual redesign of how genome editing could work.

How Base Editors Work

The Fusion Architecture: dCas9 or Nickase + Deaminase

A base editor is a modular fusion protein. It combines three functional components:

A catalytically impaired Cas9. Early editors used a “dead” Cas9 (dCas9) with both nuclease domains inactivated. Modern editors use a nickase Cas9 (nCas9, D10A variant), which cuts only the non-edited strand. That single-strand nick biases the cell’s mismatch repair machinery toward accepting the newly edited base rather than reverting it.
A deaminase domain. This enzyme performs the actual chemistry. It acts on single-stranded DNA exposed when Cas9 opens the helix to scan for a PAM site and unwinds the non-target strand. Only the displaced non-target strand is accessible; the base-paired strand is protected by Cas9 and the RNA-DNA hybrid. The deaminase converts its substrate base within a defined editing window, typically nucleotides 4 through 8 counting from the PAM-distal end of the protospacer.
A single guide RNA (sgRNA). The sgRNA is identical in function to standard CRISPR — it base-pairs with the target strand and positions the Cas9 fusion protein precisely at the intended genomic locus.

Additional components in modern base editors include uracil DNA glycosylase inhibitors (UGI), which block the base excision repair enzyme that would otherwise remove the deaminated uracil before it can be read as thymine. Two tandem UGI domains are standard in the BE4max architecture developed by the Liu lab.

Figure 1. Schematic of base editor architecture. The nCas9–deaminase fusion protein (guided by sgRNA) unwinds target DNA and exposes a single-stranded editing window at nucleotide positions 4–8 of the protospacer. The deaminase converts C to U (CBE) or A to I (ABE) within that window. A nick in the non-edited strand biases mismatch repair toward locking in the intended edit. No double-strand break is created at any step.

Cytosine Base Editors: C → T

Cytosine base editors (CBEs) use APOBEC family deaminases, most commonly APOBEC1 or engineered variants. The reaction sequence is:

APOBEC deaminates cytosine (C) to uracil (U) in the exposed ssDNA window.
Uracil reads as thymine during DNA replication, installing a T in place of the original C.
On the complementary strand, the original G pairs with the new T, completing the C·G → T·A transition.
UGI inhibitors suppress BER removal of the U intermediate, improving editing yields.

BE4max (Komor et al., further optimized by Koblan et al., 2018) incorporates codon-optimized nCas9(D10A), two APOBEC1 domains, and two UGI domains. It achieves cytosine-to-thymine conversion efficiencies of 50–80% in human cells at targetable loci. AncBE4max, which uses an ancestral reconstruction of APOBEC1, extends PAM compatibility when paired with SpCas9-NG or other PAM-relaxed Cas9 variants.

The clinical and research value of C→T editing is substantial. Many gain-of-function disease mutations are G-to-A (i.e., C-to-T on the sense strand) transitions. Conversely, CBEs can also introduce premature stop codons — CAA, CAG, or CGA to TAA, TAG, TGA — to silence dominant-negative or oncogenic alleles. This “stop codon installation” strategy is central to Beam Therapeutics’ CAR-T programs and was employed in the base-edited T-cell therapy administered at Great Ormond Street Hospital in 2022 for pediatric T-cell acute lymphoblastic leukemia (T-ALL).

Adenine Base Editors: A → G

Adenine base editors (ABEs) present a trickier chemistry problem: no natural deaminase acts on adenine within DNA. Gaudelli and colleagues solved this by engineering TadA, a bacterial tRNA adenosine deaminase that normally acts on single-stranded RNA, to accept DNA substrates. Seven rounds of phage-assisted continuous evolution (PACE) produced the original TadA* variant capable of deaminating adenine in ssDNA.

The reaction:
– TadA* deaminates adenine (A) to inosine (I) in the exposed editing window.
– Inosine is read as guanine during replication, installing a G in place of A.
– The original T on the complementary strand is replaced by C, completing the A·T → G·C transition.

ABE7.10 (the 2017 publication version) has been substantially improved. ABE8e (2020, Richter et al.) incorporates an ABE8 TadA variant with ~590-fold higher catalytic activity and achieves A-to-G efficiencies exceeding 70% across diverse sites. ABE8.20m adds further mutations reducing RNA off-target activity. The current clinical-grade tool of choice for many programs is ABE8e or its successor variants.

The therapeutic logic for ABE is equally compelling. The sickle cell disease mutation is a single A·T → T·A transversion in the HBB gene — that specific change is beyond direct ABE correction. However, Beam Therapeutics’ BEAM-101 program takes an indirect route: it uses ABE to install a fetal hemoglobin reactivation edit at the BCL11A erythroid enhancer, mimicking a naturally occurring protective polymorphism. Correcting the precise E6V mutation in HbS would require a transversion editor, which neither CBE nor ABE provides.

Figure 2. Parallel biochemical pathways of CBE and ABE. Left: cytosine base editors deaminate C to U; with UGI inhibition and strand nick, the cell resolves the U-G mismatch to T-A. Right: adenine base editors deaminate A to I using an engineered TadA variant; the I-T mismatch resolves to G-C. Neither pathway requires a double-strand break.

Base Editing vs. Prime Editing

Base editing is not the only DSB-free precision editing approach. Prime editing, introduced by Andrew Anzalone, Perez-Pinera, and David Liu in Nature in 2019, uses a different architecture: a reverse transcriptase fused to nCas9 and guided by a specialized pegRNA. Prime editing can install all 12 types of point mutations (including transversions), small insertions up to ~44 bp, and small deletions. It is, in principle, more versatile than base editing.

In practice, the two tools occupy different niches.

Base editing advantages:
– Higher editing efficiency at most loci (often 2–5× higher than prime editing at matched targets)
– Simpler delivery payload (no pegRNA reverse transcription scaffold required)
– More extensive clinical validation — base editors entered human trials before prime editors
– Lower indel byproduct rates at the target site (typically <1%)

Prime editing advantages:
– Can make transversions (A→C, A→T, C→G, G→T, etc.) that base editors cannot
– Can install specific sequences, correcting frameshift mutations or inserting regulatory elements
– Can make precise deletions
– No bystander editing of adjacent bases within the window (only the pegRNA-specified position is changed)

For any given correction task, the choice depends on the mutation class. A C→T or A→G transition: use a base editor. A transversion, insertion, or frameshift: prime editing is the only DSB-free option. For the sickle cell HbS E6V mutation (A→T transversion), prime editing or HDR is required; base editing cannot make that specific change directly.

You can read more about prime editing architecture and its current clinical status in our detailed explainer on prime editing and its 2026 outlook.

Figure 3. Decision map: choosing between base editing and prime editing. Transition mutations (C↔T, A↔G) are well-served by CBEs and ABEs with high efficiency. Transversions, small insertions, and deletions require prime editing. Classic CRISPR (DSB-based) remains the practical choice when complete gene disruption — rather than correction — is the goal.

Trade-Offs, Limits, and Off-Target Effects

Base editors are more precise than DSB-based CRISPR, but they are not without limitations. Understanding the failure modes is essential for anyone evaluating their therapeutic use.

Bystander Edits

The editing window — roughly positions 4 through 8 of the protospacer — is a zone of activity, not a precise single-base address. If multiple cytosines (for CBE) or adenines (for ABE) fall within this window, the deaminase may modify more than one. These unintended edits at neighboring positions within the window are called bystander edits.

Bystander edits are a significant practical constraint. Some target sequences are inherently bystander-rich. Narrower editing windows have been engineered — YEbase editors (2022, Doman et al.) and SpRY-based designs with altered PAM flexibility can shift or narrow the activity window. But narrowing the window trades off some reduction in on-target efficiency. Tool selection requires careful in silico and in vitro screening for each specific target site.

Off-Target DNA Editing

All Cas9-based tools share the guide-RNA-dependent off-target landscape. Base editors face an additional class of off-target activity: guide-RNA-independent deamination at accessible cytosines or adenines anywhere in the genome. This occurs when the deaminase domain encounters ssDNA regions that arise transiently during replication or transcription, independent of where Cas9 is bound.

The magnitude of guide-RNA-independent off-target edits varies by deaminase variant. SECURE-BE3 (Doman et al., 2020) and related mutations in the APOBEC domain substantially reduce this activity. ABE8e has significantly lower DNA off-target rates than earlier ABE versions, though ABE8.20m was specifically further attenuated for RNA off-targets (see below).

Whole-genome sequencing studies in edited cells have so far shown that well-optimized base editors produce off-target DNA editing rates within a range considered acceptable for many therapeutic applications — but this remains an active area of surveillance in ongoing clinical programs, and no general clearance should be assumed.

Off-Target RNA Editing

A less-anticipated finding emerged from work by Grunewald et al. (Science, 2019) and Jin et al. (Science, 2019): CBEs cause widespread adenosine-to-inosine (A-to-I) editing in the transcriptome — in RNA, not DNA. This is a consequence of the APOBEC deaminase being an inherently RNA-active enzyme. APOBEC-family enzymes evolved to edit RNA and ssDNA in the context of antiviral defense; their RNA promiscuity does not disappear when the domain is fused to Cas9.

ABEs also exhibit RNA off-target activity, with the original TadA* showing substantial transcriptome-wide A-to-I changes. ABE8.20m was specifically engineered with mutations (V106W and others) that suppress RNA editing while preserving DNA editing activity. The degree to which transcriptome editing constitutes a real biological hazard — versus a benign background signal — is not fully resolved and depends on whether off-target RNA edits affect coding or regulatory sequences of interest.

The takeaway: RNA off-target profiling is now a standard requirement in base editor development, not an afterthought.

Delivery Challenges

Base editor components are larger than standard Cas9. The nCas9-deaminase-UGI fusion protein in BE4max is approximately 1,500 amino acids — larger than the 4.2-kb coding sequence that a single standard AAV serotype can accommodate along with regulatory elements. Packing a base editor into a single AAV capsid requires either a split-intein approach (dividing the protein across two AAV vectors that self-assemble in cells) or accepting a reduced regulatory cassette.

Clinical programs have generally moved toward lipid nanoparticle (LNP) delivery of mRNA (encoding the base editor protein) plus synthetic sgRNA, rather than AAV. LNP delivery to the liver is well-established from mRNA vaccine and siRNA precedent. Beam Therapeutics’ liver-directed programs (targeting PCSK9 and other loci) use LNP-mRNA delivery. Ex vivo editing of hematopoietic stem cells (HSCs) or T-cells avoids the systemic delivery problem entirely: cells are harvested, edited outside the body, and re-infused — the approach used in sickle cell and CAR-T programs.

In vivo delivery to muscle, lung, CNS, or eye remains harder. Split-AAV and LNP formulations optimized for non-liver tissues are active research areas but not yet routine clinical tools.

For perspective on how AI-driven protein engineering is expanding the toolkit available to address these delivery and engineering problems, see our coverage of RFdiffusion 2 and deep learning protein design and the structural prediction advances in AlphaFold 3 protein-ligand co-folding.

Where It Stands Clinically in 2026

Base editing has made the transition from academic technology to clinical asset. The programs below represent the most publicly documented as of mid-2026. Clinical data is still early; the following reflects publicly disclosed information and should be interpreted cautiously.

Cardiovascular Disease: Verve Therapeutics

Verve Therapeutics’ VERVE-101 uses an ABE8e base editor delivered via LNP to the liver. It targets the PCSK9 gene — a regulator of LDL receptor recycling — converting a single adenine to guanine in a codon that produces a loss-of-function variant. The goal is a single-dose, permanent reduction in LDL cholesterol, analogous to the cardiovascular protection seen in people who carry natural PCSK9 loss-of-function variants.

Early Phase 1b data presented publicly in 2023–2024 showed measurable, durable reductions in blood PCSK9 protein and LDL-C in the first participants. The company has continued enrolling in higher-dose cohorts. VERVE-102, using a next-generation base editor with improved liver tropism LNPs, is also in clinical evaluation. These programs represent the first in-human use of adenine base editing, and their safety and durability data are being watched closely by the entire field.

Sickle Cell Disease and Beta-Thalassemia: Beam Therapeutics

Beam’s BEAM-101 uses ABE to edit HSCs ex vivo at the BCL11A enhancer, reactivating fetal hemoglobin (HbF) production to compensate for defective adult hemoglobin. The strategy is biologically similar to the approved CRISPR-based therapy Casgevy (exa-cel, from Vertex and CRISPR Therapeutics), but uses base editing rather than a DSB. The anticipated advantage is fewer unintended chromosomal rearrangements and reduced genotoxic stress during ex vivo editing, which may improve engraftment and long-term safety. Early clinical data has not yet been fully disclosed as of mid-2026, but enrollment is ongoing.

Cancer: Base-Edited CAR-T Cells

Perhaps the most dramatic demonstration of base editing’s potential came from Great Ormond Street Hospital in London in 2022 (published in the New England Journal of Medicine, June 2023, as “Base-Edited CAR7 T Cells for Relapsed T-Cell Acute Lymphoblastic Leukemia”). Clinicians administered a base-edited allogeneic CAR-T product to a child with relapsed T-ALL who had no remaining standard treatment options. The T-cells had been multiplexed base-edited to: (1) disrupt the T-cell receptor to prevent graft-versus-host disease, (2) disrupt CD7 to prevent fratricide within the CAR-T product, (3) disrupt CD52 to enable alemtuzumab conditioning, and (4) install a CD7-targeting CAR. The patient achieved remission.

This case was a compassionate use, single-patient instance — not a trial result — and caution is warranted in extrapolating from it. But it demonstrated that multiplex base editing of primary human T-cells is technically feasible and could yield a clinically meaningful outcome. Beam Therapeutics and others have CAR-T programs in formal Phase 1 trials as of 2026.

Practical Outlook Checklist

For researchers and clinicians evaluating base editing:

Confirm your mutation class first. CBE covers C→T; ABE covers A→G. If you need a transversion, look to prime editing or HDR.
Screen the target sequence for bystander C or A residues in the editing window. Use in silico tools (BE-Hive, BaseEditing.net) before committing to a guide.
Profile both DNA and RNA off-targets using CIRCLE-seq, DISCOVER-seq, or similar orthogonal methods.
Match delivery modality to target tissue. LNP is well-characterized for liver; ex vivo is standard for blood. Non-liver in vivo delivery is still emerging.
Use latest-generation editors. ABE8e or ABE8.20m over ABE7.10; BE4max or AncBE4max over BE3. The gap in performance is substantial.
Plan for immunogenicity assessment. Bacterial-derived components (TadA, APOBEC) may trigger immune responses on repeat administration.

Frequently Asked Questions

What is the difference between base editing and regular CRISPR?
Standard CRISPR-Cas9 cuts both strands of DNA, relying on cellular repair pathways that are imprecise. Base editing uses a catalytically impaired Cas9 (no double-strand cut) fused to a deaminase enzyme that chemically converts a single base. The result is a highly precise single-letter change with far fewer unintended insertions or deletions.

Can base editing fix any point mutation?
No. Current base editors are limited to transition mutations: C→T (CBE) and A→G (ABE). Transversion mutations (e.g., A→T, C→G), insertions, and deletions require prime editing or HDR-based approaches. Roughly 30% of pathogenic single-nucleotide variants involve transversions, placing them outside the reach of base editing alone.

What is a bystander edit?
A bystander edit occurs when the deaminase modifies a base adjacent to the intended target within the editing window (approximately positions 4–8 of the protospacer). If a second C or A falls within this window, it may be unintentionally converted along with the intended base. Narrowed-window base editor variants and careful guide RNA selection reduce but do not eliminate this risk.

How does base editing compare to prime editing in clinical development?
As of 2026, base editing is significantly further ahead in the clinic. Verve’s VERVE-101 and Beam’s BEAM-101 are in human trials; base-edited CAR-T cells have been administered in compassionate use settings. Prime editing has not yet entered registered human clinical trials, though preclinical efficacy data are strong. The two technologies are complementary, not competing — mutation class dictates which tool applies.

Is base editing safe for use in humans?
“Safe” is context-dependent and the question remains open in early clinical data. Current evidence suggests that state-of-the-art base editors produce acceptably low rates of on-target indels (<1%), DNA off-target edits, and (with optimized designs like ABE8.20m) reduced RNA off-targets. Long-term safety data from ongoing trials will be critical. This article is informational only and does not constitute medical advice.

Who invented base editing?
Cytosine base editing was first reported by Alexis Komor, Yongjoo Kim, Michael Packer, John Zuris, and David Liu at the Broad Institute/Harvard in 2016 (Komor et al., Nature, 2016). Adenine base editing was developed by Nicole Gaudelli, Alexis Komor, Holly Rees, Mandana Packer, Anna Badran, Ian Bryson, and David Liu, published in Nature in 2017. The Liu lab at the Broad Institute remains the primary academic hub for base and prime editor development.

Base Editing Explained: Single-Base CRISPR Therapeutics