Beyond CRISPR Cutting: A Systematic Review of Base Editing and Prime Editing for Therapeutic Precision

3.2. Overview of Included Studies

Table 4 summarises the disease settings, target genes, editor platforms, delivery strategies, and therapeutic outcomes reported across the 16 included primary studies. The evidence shows three main patterns:

1. Base editing generally performed best for eligible single-nucleotide transitions.
2. Prime editing is more adaptable for complex substitutions, insertions, deletions.
3. Regulatory changes while therapeutic success depends on functional restoration, delivery feasibility, and safety rather than editing efficiency alone.

3.3. Engineering Advances Enhancing Editing Precision in Human Cells

3.3.1. Optimisation of Base Editor Architectures in Human Systems

Recent advances in genome editing have focused on improving the accuracy, efficiency, and applicability of base and prime editing in human cells. Gaudelli et al., [28] developed a genome-wide adenine base-editing approach for A•T to G•C conversion by fusing an evolved adenine deaminase to a catalytically impaired Cas9 nickase. In human cell lines, this system generated efficient A•T→G•C conversion with low indel formation, supporting adenine base editing as a high-efficiency platform. However, the study also highlighted persistent limitations, including narrow editing windows and bystander edits within the deaminase activity range.

Expanding these results, Koblan et al., [33] found that intracellular expression level, rather than catalytic activity itself, was the initial and foremost predictor of base-editing efficacy. Nuclear localisation signal optimisation, codon bias, and reconstruction of the ancestral deaminase domains generated BE4max, AncBE4max, and ABEmax. These engineered editors demonstrated substantially improved performance in a variety of human cell types, repairing pathogenic single-nucleotide polymorphisms with higher precision than first-generation editors. Two- to four-fold improvements in editing efficiency were reported compared to ABE7.10, with improved allele-correction frequencies exceeding 70% beyond several disease-relevant sites and reduced bystander editing. These findings extended the work of Gaudelli et al., [28] by demonstrating that editor expression, stability, and host-cell compatibility are important determinants of editing performance.

The remaining drawbacks of base editing were then reduced by Richter et al., [36] through a systematic increase in Protospacer Adjacent Motif (PAM) compatibility and catalytic kinetic optimisation. Using phage-assisted continuous evolution, the authors constructed the ABE8e variant, which had a nearly 590-fold higher deamination rate compared to ABE7.10 and was more broadly compatible with a wider range of Cas proteins. Kinetic studies demonstrated a significantly faster onset of editing, reaching near-maximum activity within 24 h, compared to several days for earlier ABEs. Notably, ABE8e was found to be able to effectively edit genomic loci, including regulatory elements, that were previously inaccessible to base editors. This broader targeting range extended the capabilities described by Koblan et al., [33]. However, increased off-target activity in early ABE8e variants necessitated further specificity-promoting modifications. Genome-wide profiling revealed an increase in off-target editing in both DNA and RNA of early ABE8e variants, thus identifying a measurable trade-off between efficiency and fidelity.

3.3.2. Optimisation of Prime Editing Systems in Human Cells

Chen et al., [27] investigated cellular factors that constrain prime editing in human cells. Using pooled CRISPR-interference screens, they identified mismatch repair as an important suppressor of prime-editing efficiency. The PE4 and PE5 systems, which incorporated transient mismatch-repair inhibition, substantially improved editing outcomes across multiple human loci. These findings show that prime-editing performance is shaped not only by editor design but also by the repair environment of the target cell.

Chen et al., [27] focused on repair-pathway modulation, whereas Habib et al., [29] evaluated prime-editing fidelity in human embryonic stem cells. Using inducible prime-editor platforms, whole-genome sequencing, and a direct comparison with base editing, Habib et al., [29] observed efficient and specific editing with few off-target variants. Genome-wide analysis detected no significant increase in mutations above background rates, supporting the safety and stability of optimised prime editing in therapeutically relevant human cells.

3.4. Expansion of Editing Capability Beyond Single-Nucleotide Variants

3.4.1. Large Insertions, Deletions, and Replacements Using Prime Editing

The ability of genome-editing tools to move beyond single-nucleotide replacement has expanded the therapeutic scope of precision editing in human cells. Anzalone et al., [23] addressed a key limitation of conventional base and prime editing by developing twin prime editing for larger insertions, deletions, inversions, and sequence replacements. This approach uses two pegRNAs to write complementary DNA flaps that can recombine into a defined edited sequence, increasing the range of variants that can be corrected.

Deep sequencing showed precise replacement efficiencies of 20-40% at tested loci and up to 5% for multi-kilobase edits. Comparative genome-wide profiling did not reveal an increase in guide-independent mutations above background. Twin prime editing therefore expanded the ability to perform larger programmed genomic changes, although it required more complex pegRNA design and coordination than canonical prime editing. Efficiency remained dependent on locus, edit size, and genomic context. The capacity for exon-scale correction extends therapeutic scope beyond the variants accessible to base editing, but it also highlights a practical trade-off between editing breadth and design and delivery complexity.

3.4.2. Regulatory and Promoter Editing for Therapeutic Gene Modulation

In an alternative approach to increased editing capability, Chalumeau et al. [26] targeted non-coding regulatory regions rather than protein-coding sequences. They achieved re-expression of foetal haemoglobin by rewriting γ-globin promoters in human haematopoietic cells through prime editing. They achieved approximately 50% accurate editing in human cell lines, and tailorable yet clinically meaningful editing in patient-derived Hematopoietic Stem and Progenitor Cells (HSPC), introducing several mutations associated with hereditary persistence of Foetal Haemoglobin (HPFH) into the HBG1/2 promoters. Interestingly, compounding regulatory edits led to additive action(s) in increased γ-globin expression compared to expression produced by single edits. No significant change in erythroid differentiation markers was observed by RNA sequencing, but levels of gamma-globin transcripts were up to 3x higher than normal. Unlike structural editing strategies, this approach to regulatory editing involves changing regulatory elements instead of directly changing coding elements. In comparison to Anzalone et al., [23], Chalumeau et al., [26] placed emphasis on transcriptional modulation instead of genomic restructuring, which enabled functional rescue through comparatively subtle molecular interventions. However, some editing was more efficient than others depending on the donor, suggesting biologic heterogeneity between primary human stem cell populations.

Regulatory editing was complemented with base editors by Koblan et al., [33]. The authors designed and performed mutations in the promoters to boost adenine and cytidine base editing expression in human cells. Editing levels of over 70% were achieved for selected promoter regions and resulted in uniform upregulation of target-gene expression across all cell types. While base editing is more convenient than the prime-editing approach used by Chalumeau et al., [26] for its ease of guide design and high efficiency in isolated regulatory substitutions, it was not possible to make the same combinations of changes in a single allele. Base editing is thus more efficient for single eligible regulatory variants while prime editing is more versatile for combinatorial regulatory rewriting.

3.5. Disease-Specific Therapeutic Applications in Human Models

3.5.1. Neuromuscular Disorders

Three related but different approaches toward correction were used in these three types of neuromuscular studies: initial correction of the DMD c.8713C>T variant, modifying the splice site(s) to favor exon skipping, and restructuring the gene at the exon level using twin primed editing [23, 25, 34]. The direct correction after RTT was below 20%, and the indel rate was under 5%, while the exon-skipping methods showed an increased level of applicability of the correction method across mutation classes, but results were variable at the level of pegRNA architecture and modulation of splice-site usage. To enable editing of multi-kilobase deletions and insertions, twin prime editing was developed, but this added new features in the editing procedure involving guide design and delivery. The comparative trend is therefore a compromise between specificity and range: direct editing has the highest mutation specificity, exon skipping can be applied to multiple mutation types, while twin prime can be applied to target a wide range of mutations.

The processes of functional evidence were also varied between these approaches. Exon skipping conditions were optimized and up to 40% of dystrophin transcripts could be represented as exon-spliced isoforms, compared to an improved correlation of sequence repair and dystrophin restoration with direct correction [25, 34]. Comprehensively, though, all strategies tested have been primarily performed in human myogenic in vitro models, where local administration of gene constructs to a small number of cells is more easily achieved than systemic administration to skeletal and cardiac muscle. Therefore, achieving broad, accurate, and safe tissue distribution, along with adequate stability, remains the common challenge for both editing approaches. The length of the target sequence in the reverse-transcriptase template also demonstrates that target sequence design is a crucial factor in the performance of the prime-editing system in a DMD model.

3.5.2. Hemoglobinopathies and Blood Disorders

In all haemoglobinopathies, two therapeutic approaches were observed in clinically relevant HSPC: direct correction of a pathogenic HBB allele and regulatory re-activation of foetal haemoglobin. To correct the mutation (HBB IVS1-110) in these CD34+ HSPC, Hardouin et al., [31] tried using SpRY-ABE8e mRNA electroporation, which resulted in high editing success, restoration of beta-globin expression, and unaltered erythroid differentiation. Chalumeau et al., [26] in contrast designed new HBG1/HBG2 promoters that were then incorporated into the genome at previously established disease-causing alleles, which resulted in an elevated level of gamma-globin, but not in the elimination of the disease-causing alleles themselves. The results demonstrate the potential that base editing possesses in situations where a suitable transition and delivery option exists. Promoter editing is a mutation-independent approach that could benefit genetically diverse patients.

Comparative evidence also makes different risks apparent. Direct HBB correction is allele-specific, restores the physiological beta-globin pathway, and is constrained by the editing window, bystander effects, and PAM availability. Regulatory editing was employed to avoid repeated design cycles for each mutation and led to additive increases in gamma-globin, yet there was variability in the efficiency between donors and combination of promoter edits [26, 31, 33]. In both strategies the maturation of erythroid progenitors was generally maintained, and the ability to produce functional haemoglobin was the most informative translational endpoint.

In comparison to solid tumour disorders, the haemoglobinopathy toolkit is more clinically developed as HSPC can be modified ex vivo, quality controlled and re-transplanted autologously, thus providing milder treatment side effects. Lineage-specific functional recovery for the IVS1-110 mutation reflects the association between efficient correction and restoration of haemoglobin production in erythroid progeny.

3.5.3. Cystic Fibrosis

In research on cystic fibrosis, the function of prime editing was tested exclusively in primary airway epithelial cells or organoids derived from patients, and the corrected sequence was directly connected to CFTR processing and chloride-channel function. Bulcaen et al. [24] found that they could correct drug-ineligible CFTR mutants and restore protein maturation, plasma-membrane localisation, and channel activity for N1303K in CFTR. Sousa et al., [22] targeted the common F508del mutation, and designed a combination of pegRNA, editor-expression and delivery optimisation to restore CFTR chloride transport. Both studies demonstrated significant functional rescue for cells with reasonable correction frequencies, where the corrected cells substantially participated in epithelial ion transport.

Both haemoglobinopathy and cystic fibrosis studies demonstrate excellent phenotype-level validation, but delivery for the CF model is more challenging in vivo, and requires repeated targeting of airway epithelia, whereas HSPC delivery, although cumbersome, is more tractable ex vivo. Bulcaen et al., [24] recovered N1303K glycosylation, membrane localisation, and channel function, whereas Sousa et al., [22] found that optimisation of the delivery system of the prime edited sequences and enhancement of the prime editor allowed functional repair of F508del. These studies all set functional restoration as a clear go/no-go benchmark for future translation, with only tissue coverage, durability and pulmonary delivery at larger scales remaining to be determined.

3.5.4. Rare Monogenic and Metabolic Disorders

Phenotype-level restoration is, especially in rare-disease models, an essential, translation-relevant endpoint. In hiPSCs, Sürün et al., [37] showed disease-relevant correction, and in patient fibroblasts, Jang et al., [32] correlated sequence correction with biochemical and cellular outcomes relevant to the disease. Therapeutic relevance was assessed by functional recovery, including human system in the reported cell models, as well as by editing frequency, which was kept below 2% in all cell models reported, further emphasizing that the assessing systems need to be suitable for the type of functional benefit sought.

3.6. Delivery Strategies for Human Genome Editing

3.6.1. Non-Viral and Transient Delivery Approaches

Non-viral delivery methods are broadly employed for therapeutic genome editing because of their ability to expose the genome only for a limited time, and their potential for decreasing the risk to the genome in the long term. In human myogenic cells, Mbakam et al., [34] evaluated plasmid-based and electroporation-based prime editing and found that the efficiency of editing was strongly affected by the delivery feasibility. These results showed that the delivery condition should be considered with the editor performance.

Although the efficiency was not high, editing frequencies were still below 10% and the indel rates were consistently below 2% placing the constructs in a favourable range of safety in relation to their inability to achieve high efficiency. By comparison, Hardouin et al., [31] explicitly tackled the effectiveness of delivery by using mRNA electroporation to transfect patient-derived CD34⁺ haematopoietic stem and progenitor cells with adenine base editors. This was employed to correct the IVS1-110 β-thalassemia mutation; the efficiency of this was around 80% without altering stem-cell characteristics, self-renewal, or erythroid differentiation. When compared to plasmid delivery, mRNA electroporation resulted in sustained expression of the editors and caused less cell stress and toxicity. The properties of these features render ex vivo editing feasible, especially in haematological diseases, where autologous stem-cell transplantation is already in use.

Corrective efficiencies of over 75% led to near normal haemoglobin production in differentiated erythroid progeny further reinforcing the functional relevance. To tackle the delivery related constraints, Nelson et al., [35] optimized the molecular cargo. The prime-editing efficiency and the stability of the RNA were increased in human cell lines and primary human fibroblasts by the addition of structured RNA to the 3′ end of the pegRNAs, with indel frequencies <1%. While not a direct comparison of delivery platforms, the study demonstrated that the improvements were around 3 to 4-fold in the case of engineered pegRNAs, suggesting that cargo architecture can partially mitigate suboptimal delivery conditions.

3.6.2. Pseudoviral and Advanced Delivery Systems

The purpose of advanced delivery systems is to overcome the efficiency and scalability issues of traditional non-viral delivery systems in specific human stem-cell populations. Halegua et al., [30] created new virus-like particles, called NanoScribes, to deliver prime editing components as ribonucleoprotein complexes. The platform was designed to minimize the delivery of exogenous DNA to ensure that nuclease activity is not sustained. The editing efficiencies had optimized fusogens and stabilized epegRNAs of 68-72% in HEK293T cells and approximately 25% in human iPSCs and haematopoietic stem cells.

When compared to plasmid delivery, delivery of the NanoScribe kept indel frequencies below 1%, and reduced the activity of bystander cells, while similar effect seen upon transient exposure. The system also caused editing in stem cells that were otherwise difficult to transfect and did not affect expression of OCT4 and SOX2 nor the clonogenic capacity, confirming maintenance of identity of the edited cells. However, these limitations of clinical translation are still complicated by the complexity of production, batch-to-batch consistency, scale-up of manufacturing, tissue targeting and regulatory characterisation. Optimised mRNA electroporation was not always superior to VLP delivery across all cell types. It would therefore be advisable to consider this as a complementary platform rather than a universal substitute. Furthermore, it will be dependent on reproducible manufacturing and efficient delivery to the target tissue.

The preservation of stemness markers and clonogenic capacity support the short-term cellular compatibility of NanoScribe delivery [30]. However, translational assessment must extend beyond editing efficiency to include reproducible manufacturing, potency testing, biodistribution, immunogenicity, long-term safety, and compatibility with disease-specific clinical workflows.

3.7. Editing Outcomes, Fidelity, and Translational Readiness

3.7.1. Editing Precision and Safety Profiles

The therapeutic potential of base and prime editing in human systems relies on the precision of editing as well as genomic safety. Gaudelli et al., [28] engineered adenine base editing by combining an evolved Adenine Deaminase with a Cas9 nickase, which can convert A•T to G•C without introducing any DSBs. Efficient editing with fewer indels was achieved using this platform with human cell lines, though its efficacy was limited by the availability of PAM, the position of the editing window and a bystander deamination window of 4-6 nucleotides.

Chen et al., [27] followed up this work to demonstrate that prime-editing efficiency can vary across human cell types and genomic targets. CRISPR-Interference (CRISPRi) screening identified mismatch repair as a key inhibitor of prime-editing efficiency. This comparison demonstrates that prime editing is programmable and is tightly coupled to cellular repair pathways.

Habib et al., [29] took fidelity assessment one step further and tested prime editing in human embryonic stem cells. Minimal rates of guide-independent off-target mutations were observed when expressing the editor under an inducer. Prime editing generated fewer bystander effects and RNA-level edits than base editing but showed lower efficiency in some human systems, which remains a major hurdle for clinical translation.

3.7.2. Translational Readiness and Clinical Applicability

Optimal delivery remains a key translational challenge as the optimal approach varies depending on the cell type, tissue access, editor size, and transient exposure requirement. Ultimately, because the functional potential of modest efficiency in prime editing in patient-derived CFTR models can produce meaningful functional rescue, Bulcaen et al., [24] noted that outcomes at the phenotype level are crucial to consider when evaluating therapeutic potential.

Similarly, Sousa and colleagues [22] showed that model-specific optimization can boost the feasibility of prime editing in human primary airway epithelial cells, highlighting the need for optimal editor design and delivery for different target systems.

Though important, delivery is still a tremendous hurdle in clinical translation, especially in human stem-cell systems. To solve this problem, Halegua et al., [30] proposed to transiently deliver prime-editing components to the cell surface with NanoScribe virus-like particles. This strategy facilitated editing, allowed the editors to be used at low toxicity, enabled transient exposure to the editors, and was more efficient in human induced pluripotent stem cells and haematopoietic stem cells. The key challenge, however, is to produce these materials in a scalable manner, to target the tissue specifically, and to be able to reproduce their manufacturing.

4. DISCUSSION

A key advancement over the standard CRISPR-Cas9 system is the emergence of base editing and prime editing, which allow for engineered modifications of DNA without the intention to create DSBs. In therapeutic genome editing, minimizing indel formation, chromosomal rearrangements, and genotoxic stress becomes an important factor, while the lack of DSB in their design can abate some negative repair outcomes [5]. The absence of intentional DSB formation does not eliminate other safety and translational concerns, including bystander editing, incomplete repair, and delivery-related toxicity.

Comparative evidence suggests that the two platforms have different strengths. The efficiency of base editing highly depends on eligible mutations like A→G or C→T substitutions including variants associated with haemoglobinopathies [7]. It has been limited by availability of PAM, editing window size, off-target deamination and bystander editing at neighboring bases [29, 32]. In addition to base editing, prime editing also supports small insertions, deletions, and substitutions without the need for a donor DNA template [38]. This is of specific relevance to CFTR correction in 2D airway epithelial and organoid models [22]. Nevertheless, prime-editing efficiency varies across tissues and cell types, with primary muscle, airway epithelial, and fibroblast cells generally presenting greater challenges than immortalised cell lines. These differences are likely associated with delivery conditions, pegRNA design, and cellular repair pathways, but not an inherent stem cell platform limitation [29, 32].

One noteworthy conclusion from these studies is that editing efficiency alone does not determine therapeutic value. Even modest levels of correction can produce clinically meaningful CFTR functional rescue in relevant airway epithelial and organoid models [22, 24]. By comparison, Duchenne muscular dystrophy has much more widespread delivery and tissue coverage needs [34, 25]. Haemoglobinopathies present a comparatively favourable translational context because HSPC can be edited ex vivo, characterised before autologous transplantation, and assessed for functional recovery [26, 31]. Thus, clinical readiness depends not simply on the percentage of edited cells but also on the proportion of cells that must be corrected, anatomical distribution, durability of functional rescue, and delivery feasibility.

The delivery aspect is thus still a formidable challenge, especially for tissues that lack ex vivo methods for delivery. While the delivery to HSPC and mRNA-based delivery are effective, it is harder to deliver these to skeletal muscle, pulmonary epithelium, and other in vivo targets [16]. All three delivery strategies, non-viral, pseudoviral, and viral face distinct challenges related to immunogenicity, editor size, tissue access, and manufacturing scalability.

Safety and regulatory considerations are also a crucial part of clinical translation. Base-editing and prime-editing technologies lessen dependence on DSB-mediated repair [38–40]. Thus, regulators will expect strong evidence for: genome integrity; off-target activity; long-term safety; and Chemistry, Manufacturing and Controls (CMC) [38, 41]. Considering the implications for future generations from germline editing, as well as other issues regarding ethics, also argues for a somatic-only approach [41].

The two technologies, base editing and prime editing, show therapeutic potential from the 16 systematically selected primary studies, depending on the type of mutation, delivery technology and human cell context (Table 5).

LIMITATIONS OF THE REVIEW

The limited scope of this review reflects the variety of human cell types, delivery methods, outcome measures, and editor architectures assessed. Few human-relevant studies met the inclusion criteria, further making quantitative meta-analysis difficult.

Further, there are still numerous studies at an early stage included under this umbrella with various endpoints reported such as editing efficiency, specificity, viability and functional restoration [32]. This variability can be a problem in determining the threshold for clinical readiness. Hopefully, in the future, more studies will report standardised safety and functional outcomes on more similar human models, which will strengthen future reviews. As the evidence base expands, transparent machine-learning methods may help integrate heterogeneous molecular, delivery, safety, and functional outcome variables; cross-domain AI studies demonstrate complementary strategies for unstructured-data extraction, prediction from historical and contextual features, and explanation of model outputs [42–44]. Such methods would require biological validation and should supplement, rather than replace, mechanistic and experimental evidence.