High-efficiency TadA Cytosine Base Editors for Precise Modeling of Human Disease Variants

Abstract

Many missense mutations identified in genetic testing are variants of uncertain significance (VUS), not yet classified as either benign or pathogenic. Systematic determination of their functional relevance is a pressing clinical need. CRISPR-mediated base editing can precisely introduce precise variants into model organisms for functional testing, but current editors face efficiency and targeting constraints. We developed TCBE-Umax, a family of TadA-derived cytosine base editors optimized for zebrafish. Engineering the TadA deaminase domain improved editing efficiency and reduced sequence context bias, expanded PAM compatibility, and minimized bystander edits and indel formation. Our editors achieved efficient biallelic editing, enabling rapid functional assessment of genetic variants in the F0 (founding) zebrafish. As a proof of concept, we evaluated 15 VUS linked to hereditary hearing loss, determining pathogenicity through phenotypic analysis. With high efficiency and versatility, TCBE-Umax base editors provide a powerful platform for studying genetic variants and disease in vivo.

Editor’s summary

A suite of TadA-derived cytosine base editors was optimized to improve efficiency, reduce sequence bias, expand PAM compatibility, and minimize artifacts, enabling precise genetic interrogation of variants of uncertain significance (VUS) in zebrafish.

Introduction

The exponential growth of clinical genomics has identified numerous genetic variants, especially single-nucleotide variants (SNVs), in human patients, but determining their functional significance remains a significant challenge, as over 40% variants are classified as variants of uncertain significance (VUS)¹. VUS represent a particular burden in clinical diagnosis and genetic counseling, as their impact on protein function and disease pathogenesis is unclear. Currently, millions of these VUS await functional characterization, creating a significant bottleneck in patient care. This uncertainty leaves clinicians unable to definitively inform patients whether specific genetic changes are responsible for their conditions. Hence, determining their functional significance is a pressing need. Zebrafish (Danio rerio) have emerged as an ideal system for the rapid functional validation of human genetic variants due to their high genetic conservation with humans, with ~80% of human disease genes having zebrafish orthologs, rapid development, optical transparency, and amenability to genetic manipulation². The ability to rapidly generate and analyze multiple genetic variants in zebrafish can provide crucial insights into variant pathogenicity and disease mechanisms³.

Our ability to study gene function and human diseases in cell systems and model organisms has been revolutionized by the advent of precise genome-editing technologies⁴. Among these, CRISPR-Cas based DNA base editors are powerful tools for introducing targeted point mutations without generating double-stranded DNA breaks or requiring donor DNA templates⁵. In particular, cytosine base editors (CBEs) enable the programmable conversion of C•G to T•A base pairs, allowing researchers to model disease-causing mutations and study gene function through precisely targeted changes⁶. Current CBEs typically utilize naturally occurring cytidine deaminases, such as APOBEC1 or AID, fused to Cas9 nickase⁷. To bind DNA, Cas9 requires a short DNA sequence, termed the protospacer adjacent motif (PAM), adjacent to the cleavage site at the locus to be targeted, and then introduces point mutations without generating double-strand breaks. While these editors can achieve efficient base editing, but face several limitations in in vivo applications, including a large size that complicates delivery, high off-target activity due to the promiscuous nature of cytidine deaminases, relatively wide editing windows that reduce precision, and quite high insertion/deletion (indel) rates⁸. These limitations are particularly problematic when modeling precise disease variants. To overcome these limitations, various new CBEs have recently been developed for use in cultured cells and plants by engineering the adenine deaminase TadA8e into a cytosine deaminase, resulting in TadA-derived CBEs (TadCBEs or Td-CBEs), which include Td-CBEmax and TadCBEd^9–11. These more versatile base editors combine higher on-target editing efficiency with substantially reduced Cas-independent DNA and RNA off-target activity compared to traditional CBEs^9–12.

Their improved properties make TadCBEs particularly attractive for use in zebrafish, where precise editing and minimal off-target effects are crucial for generating clean genetic models and accurately assessing the genetic variants. TadCBEs are smaller than traditional CBEs, which enables more efficient delivery and expression in zebrafish embryos, and facilitates high-throughput analysis of multiple variants. In our previous study, we developed the first zebrafish (z) cytosine base editor, zTadCBE, which combines high editing efficiency with low indel rates and demonstrated efficient functionality in zebrafish¹³. However, while zTadCBE does not exhibit a strong sequence-context bias, it shows reduced activity at certain AC or GC sites. More importantly, zTadCBE has limitations, including insufficient cytosine editing activity, which is crucial for evaluating genetic variants in the F0 generation. Evidently, the current activity level of zTadCBE is inadequate to fulfill this requirement¹³. Recently, next-generation TadA-derived CBEs, such as TadCBE6, have been developed with improved efficiencies; however, in vivo efficiencies have not been tested¹⁴.

Here, we developed, optimized, and characterized highly efficient and precise TadA-derived CBEs, termed TCBE-Umax, capable of generating biallelic edits in zebrafish embryos. We evaluated their editing efficiency, target site preferences, and editing precision across multiple genomic loci. To demonstrate utility for variant analysis, we applied TCBE-Umax to model clinically identified VUS in zebrafish orthologs of human disease-associated genes and established their functional impact through phenotypic analysis of the F0 (founder) generation. Therefore, this work not only establishes TadA-derived CBEs as effective tools for precise genetic manipulation in zebrafish but also advances our fundamental understanding of gene function in development and disease. It also provides a powerful platform for rapid functional assessment of human genetic variants. The potential to rapidly and systematically classify genetic variants makes the development of TadA-derived CBEs in zebrafish significant for interpreting clinical genetic data, understanding disease mechanisms, and ultimately improving patient diagnosis and care.

Results

Development of highly efficient TadA-derived CBE editors

Our goal was to develop high-efficiency next-generation TadA-derived CBE editors to enable biallelic editing in vivo. Recently, Zhang et al. employed phage-assisted evolution to develop CBE6s from a TadA-mediated dual cytosine and adenine base editor¹⁴. Through this process, they identified mutations at positions N46 and Y73 in TadA, which not only eliminated residual A•T-to-G•C editing but also enhanced C•G-to-T•A editing while broadening sequence-context compatibility¹⁴. Our previous research further demonstrated that the V82S and Q154R mutations in TadA enhance the C-to-T editing efficiency of zTadCBE in zebrafish¹³. However, whether these mutations are compatible with N46 and Y73, and whether their combination could further enhance CBE activity, remains unknown.

To address this question, we constructed four novel editors incorporating different combinations of these mutations. To optimize functionality in zebrafish, all editors underwent codon optimization and were designated as follows: TCBE-1.1 (N46I, Y73P, V82S, and Q154R), TCBE-1.2 (N46V, Y73P, V82S, and Q154R), TCBE-1.3 (N46L, Y73P, V82S, and Q154R), and TCBE-1.4 (N46C, Y73P, V82S, and Q154C), all four editors were constructed based on the TadA-derived dual editor (TadDE), similar to what was done in the previous study¹⁴ (Fig. 1a). To evaluate the performance of these newly designed enzymes, we selected seven target sites with varying sequence contexts and assessed their C-to-T editing efficiencies using high-throughput sequencing. All editors exhibited editing activity at the target cytosines, albeit with variable efficiencies (Fig. 1b). Among these, TCBE-1.2 demonstrated the highest editing efficiency, followed by TCBE-1.4, whereas TCBE-1.1 and TCBE-1.3 showed no significant improvements compared to the original base editor, zTadCBE (Fig. 1c). Regarding indel formation, apart from TCBE-1.3, the remaining three editors did not exhibit a notable increase compared to zTadCBE (Fig. 1d and 1e). Low-frequency undesired by-products (C-to-A and C-to-G conversions) were detected at a frequency of less than 6% (Fig. 1f), demonstrating high editing specificity for all editors.

Figure 1: Screening and characterization of hyper TadA-derived Cytosine base editor TCBE-Umax.

(a) Schematic representation of constructs for five TadA-derived cytosine base editors. Components include: bpNLS (bipartite nuclear localization signal), various deaminase mutations (zTadA*, TadA*−1.1 through TadA*−1.4), XTEN (32-amino acid flexible linker), nSpCas9 (SpCas9 D10A nickase), GGS linker (GGSSGGS amino acid sequence), and UGI (Uracil glycosylase inhibitor).

(b) Heat maps depicting C-to-T conversion frequencies at each position for indicated sites across five editors: zTadCBE, TCBE-1.1, TCBE-1.2, TCBE-1.3, and TCBE-1.4. Data represent the mean values from three independent biological replicates.

(c) Mean editing efficiency comparison of zTadCBE, TCBE-1.1, TCBE-1.2, TCBE-1.3, and TCBE-1.4 editors based on data from Fig. 1b. Individual data points show mean editing activity per site, with the central dotted line indicating overall mean efficiency. Statistical significance was determined by a nonparametric two-sided Wilcoxon matched pairs signed rank test (P values shown above the violin plot).

(d) Indel frequencies of the indicated sites for zTadCBE, TCBE-1.1, TCBE-1.2, TCBE-1.3, and TCBE-1.4. Data are presented as mean values ± standard deviation (SD), calculated from three biological replicates.

(e) Comprehensive indel frequency analysis for zTadCBE, TCBE-1.1, TCBE-1.2, TCBE-1.3, and TCBE-1.4. Each data point represents the mean indel frequency per site, with the central dotted line indicating the overall mean. Statistical significance was assessed by the nonparametric two-sided Wilcoxon matched pairs signed rank test (P values indicated above the violin diagram).

(f) Product purity analysis at seven endogenous genomic loci. Values represent mean ± standard deviation from three independent experiments.

(g) Editing efficiency of TCBE-Umax at the additional 25 endogenous targets across 17 genes. Data represent the mean values from three independent biological replicates.

(h) Assessment of the mean editing efficiency of zTadCBE and TCBE-Umax using 17 gRNAs targeting NGG PAMs. Each data point represents the average editing activity at a specific site. The central dotted line represents the mean of all data points. P-values are displayed at the top of the violin plot. Statistical analysis was performed using a two-tailed paired t-test (P values indicated).

(i) Base editing efficiency of TCBE-Umax systems at target C in various NCN sequence contexts. Data points represent mean editing activity per site. Statistical analysis was performed using a two-tailed paired t-test (P values indicated).

All source data are available in the Source Data file.

Given the superior activity displayed by TCBE-1.2, we named it TCBE-Umax and focused on its further characterization and optimization. To ensure that the optimized editor exhibits broad applicability across diverse genomic loci, we further characterized the editing profile of TCBE-Umax at an additional 25 endogenous zebrafish target sites with NGG protospacer-adjacent motifs (PAM), covering 17 genes of interest. All tested sites exhibited high editing efficiencies (up to 89%) within the primary editing window (positions 4–8 of the protospacer), resulting in the expected C-to-T conversions (Fig.1g). Compared to the previous tool, zTadCBE, TCBE-Umax demonstrated a more than double (~2-fold) improvement in average editing efficiency (Fig. 1h and Supplementary Fig. 1). Because traditional CBEs have strong TC sequence context bias, we also assessed the sequence-context preference of TCBE-Umax, which exhibited consistent C•G-to-T•A editing efficiency across all tested sequence contexts (Fig. 1i).

To further characterize the base specificity of TCBE-Umax in vivo, we asked whether this editor retains any residual A-to-G editing activity inherited from the TadA-derived deaminase. Our previous studies reported that zTadCBE exhibited minimal A-to-G activity (<10%) at certain sites in zebrafish¹³. While these unintended edits can be diluted over successive generations, these edits pose analytical challenges, particularly when using the F0 generation for direct phenotype assessment. We therefore compared TCBE-Umax with zTadCBE at three loci (bai2, hars-T1 and nop56) where zTadCBE mediates low A-to-G conversions. As expected, zTadCBE induced clear A-to-G substitutions at the target adenines, whereas TCBE-Umax did not generate detectable A-to-G edits above background at any of the tested sites (Supplementary Fig.2).

High editing activity can increase off-target effects, to assess this, we first examined the gRNA-dependent off-target activity of TCBE-Umax at the top three predicted sites for the p53, kcnq3-T2, and spata5l1-T2 gRNAs. High-throughput sequencing revealed that TCBE-Umax showed no detectable editing at the kcnq3-T2 and spata5l1-T2 loci, while exhibiting slightly reduced off-target effects (<10%) for the p53-targeting gRNA (Supplementary Fig. 3a). Next, we investigated whether TCBE-Umax induces gRNA-independent off-target modifications across the genome. To do this, we co-injected TCBE-Umax with its respective gRNA, along with a catalytically inactive Staphylococcus aureus-Cas9 system (dSaCas9) and gRNAs targeting R-loop sites 1–3. The dSaCas9 system facilitated the formation of R-loops, short single-stranded DNA segments that could potentially serve as substrates for TCBE-Umax (Supplementary Fig. 3b). NGS analysis showed that TCBE-Umax maintained comparable on-target editing efficiency regardless of dSaCas9 presence (Supplementary Fig. 3c). Regarding off-target activity, TCBE-Umax exhibited negligible editing at all three R-loop sites compared to the control group (Supplementary Fig. 3d), highlighting its high specificity.

Expanding the Targeting Range of TCBE-Umax

Having shown that our TCBE-Umax editors are significantly more efficient than previous TCBEs, we next set out to improve further by expanding their targeting scope. Base editors derived from Streptococcus pyogenes Cas9 (SpCas9) are constrained by both the requirement of the protospacer adjacent motif (PAM), a short 5’-NGG-3’ DNA sequence adjacent to the cleavage site, and their restricted base editing to a narrow window of 4–8 bases distal to the PAM. Engineered SpCas9 enzymes, such as SpRY and SpG, have been developed to exhibit more flexible PAM recognition and are compatible with single-base editing systems in zebrafish^{3, 15, 16}. These modified enzymes enable the recognition of atypical 5’-NNN-3’ PAM sequences, broadening the targetable genomic range of base editors.

To expand the targeting scope of TCBE-Umax-mediated cytosine base editing, we replaced the nSpCas9 in the TCBE-Umax construct with nSpRYCas9, generating TCBE-Umax-SpRY (Fig. 2a). To evaluate its C-to-T editing efficiency at NNN PAM sites in zebrafish, we selected 18 loci with atypical PAM sequences and co-injected single-guide RNAs (sgRNAs) with TCBE-Umax-SpRY-encoding mRNA into one-cell-stage embryos. Sequencing results revealed C-to-T conversions at all 18 loci, albeit with varying efficiencies (Fig. 2b). Compared to our previously developed zTadCBE-SpRY¹³, TCBE-Umax-SpRY demonstrated an average 1.7-fold increase in C-to-T editing efficiency (Fig. 2c and 2d). These findings confirm that TadA-based C-to-T base editing is compatible with nSpRYCas9, enabling efficient editing across a broad range of PAM sequences in zebrafish and significantly expanding the targeting scope of the base editor.

Figure 2: Expanding Targeting Range of different TCBE-Umax editors.

(a) Schematics of the TCBE-Umax-SpRY editor used for cytosine base editing in zebrafish.

(b) Summary of TCBE-Umax-SpRY editing efficiency across 18 NNN PAM sites in 11 genes. Heat maps depicting C-to-T conversion frequencies at each position for the indicated sites. Data represent the mean values from three independent biological replicates.

(c) Comparison of editing efficiencies between zTadCBE-SpRY and TCBE-Umax-SpRY using 18 gRNAs targeting NNN PAMs. Numbers indicate the position of the edited base within each gRNA. Data represent mean ± standard deviation (SD) from three biological replicates.

(d) Analysis of mean editing efficiency comparing zTadCBE-SpRY and TCBE-Umax-SpRY based on data from Fig. 2c. Individual data points show mean editing efficiency per site, with the overall mean indicated by a central dotted line. P values from two-tailed paired t-tests are shown above the violin plot.

(e) Comparison of editing efficiencies among TCBE-Umax-SpRY, TCBE-Umax-SpG, and TCBE-Umax-NG at five different loci with NGN PAM. Numbers indicate the editing base position within the gRNA. Values represent mean ± SD (n=3 biological replicates).

The editing efficiency comparison among TCBE-Umax-SpRY, TCBE-Umax-SpG, and TCBE-Umax-NG targeting five different loci with NGN PAM. The editing base position within the gRNA is indicated by numbers. Values are presented as mean value ± standard deviation (SD), n = 3 biological replicates. A nonparametric two-sided Wilcoxon matched pairs signed rank test was performed (with P values indicated).

All source data are available in the Source Data file.

While several Cas9 enzymes, including SpGCas9, Cas9-NG, and ScCas9, have been reported to function in zebrafish, only the SpRY enzyme has been demonstrated to be compatible with CBEs in this model organism^{15, 17, 18}. Similar to a previous report¹⁸, our experiments showed that CBE4-SpG exhibited extremely low activity in zebrafish, rendering it unsuitable for practical applications in zebrafish. To further enhance the diversity of base editing tools and expand options for editing specific loci, we constructed two additional editors, TCBE-Umax-SpG and TCBE-Umax-NG, both of which require only a single G in the PAM. Upon evaluating five such NGN PAM sites, these two editors exhibited variable activities across different loci, confirming their compatibility with the TCBE-Umax system (Fig. 2e). While their average activity was comparable to that of TCBE-Umax-SpRY (Fig. 2f), this editor demonstrated superior editing efficiency at specific loci. For instance, at the kras E484K-NGT locus, TCBE-Umax-NG achieved an editing efficiency of 53%, whereas TCBE-Umax-SpRY exhibited only 7% efficiency. Similarly, at the kras R26C-NGA locus, TCBE-Umax-SpG outperformed the other two editors (Fig. 2e). These results highlight that, in non-NGG PAM contexts, selecting different Cas9 editors can effectively maximize base editing efficiency at specific loci. We further demonstrated that all the TCBE-Umax editors discussed above achieved high germline targeting efficiency and transmission rates (Supplementary Table 1), highlighting their strong capability for precise and efficient base editing.

Characterization of Two TCBE-Umax Editors with Reduced Indel Generation and Expanded Targeting Range

While evaluating TCBE-Umax activity, we observed that approximately one-third (8/25) of the target regions exhibited overlapping peaks near the target sites in the Sanger sequencing chromatograms, suggesting the presence of indels alongside cytosine base editing (Supplementary Fig. 4). The relatively high frequency of indels compromises the precision of this tool, particularly when assessing variant pathogenicity in the F0 generation. Therefore, we set out to develop an editor that minimizes indel formation whilst maintaining high editing efficiency. CRISPR-Cas9 naturally contains multiple sites in its protein sequence where inteins can be inserted without disrupting function¹⁹. A previous study reported that the amino acid at position 1054 within the RuvC domain of Cas9 is structurally positioned on the protein surface, where it exhibits conformational flexibility and lies in proximity to the non-target DNA strand, making it potentially susceptible to deamination²⁰. Given this, we hypothesized that altering the relative position of the deaminase and Cas9 might modify the intrinsic properties of the base editor. To test this hypothesis, we inserted the deaminase at position 1054 within Cas9, without linkers, directly fusing it to Cas9 to create a new editor, which we designated as TCBE-Umax-ex1 (Fig. 3a and 3b). Initial Sanger sequencing at two sites with high indel frequencies following TCBE-Umax editing led to an unexpected discovery, TCBE-Umax-ex1 significantly reduced indel formation (Supplementary Fig. 5). To systematically and quantitatively evaluate its effect on indel generation, we selected eight loci with high indel frequencies by TCBE-Umax editing and performed high-throughput sequencing. The results demonstrated a substantial reduction in indel generation across all tested sites, with indel levels decreasing to approximately one-fourth of those observed with TCBE-Umax (Supplementary Fig. 6). In addition, analysis of by-products revealed that TCBE-Umax-ex1 also maintained a low level of undesired base conversions (Supplementary Fig. 7). Notably, on-target editing activity remained comparable or slightly improved (Fig. 3c) to TCBE-Umax. Meanwhile, TCBE-Umax-ex1 exhibited a slight shift in its editing window compared to TCBE-Umax at these eight loci. We, therefore, assessed its activity at an additional 13 target sites (Fig. 3d). The final analysis revealed that TCBE-Umax-ex1 has an editing window spanning positions 3 to 12, with peak activity concentrated between positions 4 and 10, whereas TCBE-Umax primarily edits positions 4–8 (Fig. 3e).

Figure 3: Characterization of TCBE-Umax Editors with Expanded Targeting Range and Reduced Indel Formation.

(a) Schematic representation of the TCBE-Umax-ex1 construct.

(b) AlphaFold2-predicted structure of TCBE-Umax-ex1.

(c) Comparative editing efficiency of TCBE-Umax and TCBE-Umax-ex1 at eight endogenous genomic sites in zebrafish. The heat map shows average editing percentages from three independent biological replicates.

(d) TCBE-Umax-ex1 editing efficiency across 13 endogenous targets in 13 distinct genes. Heat maps depicting C-to-T conversion frequencies at each position for indicated sites. Data represents the mean values from three independent biological replicates.

(e) Efficiency and targeting window comparison between TCBE-Umax and TCBE-Umax-ex1. Data points represent average editing activity per site. The targeting windows, measured from the 5' to the 3' termini of the targeting sites, are highlighted in green for TCBE-Umax (positions 4–10) and in blue for TCBE-Umax-ex1 (positions 3–12). Analysis is based on data from three independent experiments.

(f) Schematic representation of the TCBE-Umax-ex2 construct, designed to modify the cytosine base editing window.

(g) AlphaFold2-predicted structure of TCBE-Umax-ex2.

(h) TCBE-Umax-ex2 editing efficiency across 15 endogenous targets in 12 genes. The heat map displays average editing percentages from three independent biological replicates.

(i) Efficiency and targeting window comparison between TCBE-Umax and TCBE-Umax-ex2. Data points represent average editing activity per site. The targeting windows, measured from the 5' to the 3' terminal of the targeting site, are highlighted in green for TCBE-Umax (positions 4–10) and red for TCBE-Umax-ex2 (positions 4–16). Analysis is based on data from three independent experiments. All source data are available in the Source Data file.

While TCBE-Umax-ex1 expands the editing window, its activity remains distant from the PAM site. To overcome this limitation, we developed a second editor, TCBE-Umax-ex2 (Fig. 3f, 3g). In TCBE-Umax-ex2, the HNH domain of Cas9 was deleted. A GGS-linker was introduced to connect SpCas9 at position S793 to the N-terminus of TCBE-Umax, while an SGG-linker linked the C-terminus of TCBE-Umax to SpCas9 at position R919. To assess its editing window, we evaluated TCBE-Umax-ex2 across 15 target sites in zebrafish. (Fig. 3h). Compared to conventional TCBE-Umax, TCBE-Umax-ex2 shifted the editing window toward the PAM-proximal region of the protospacer, extending from positions 5 to 16, with peak activity at positions 13–15 (Fig. 3i). By-product analysis using high-throughput sequencing revealed that this tool is highly specific, displaying minimal levels of undesired base conversions (Supplementary Fig. 8). Further indel analysis demonstrated that TCBE-Umax-ex2 also induces very low levels of indels (Supplementary Fig. 9). Notably, TCBE-Umax-ex2 exhibited enhanced editing efficiency at specific loci compared to zTadCBE-ex2 from our previous study¹³ (Supplementary Fig. 10), further expanding its utility for precise base editing applications. Our findings demonstrate that these two TCBE-Umax editors effectively address key limitations in targeting range and reduce unwanted indel formation. Their optimized designs provide greater flexibility in selecting efficient editors for specific loci, offering more precise tools for pathogenicity assessments and broader applications in base editing, particularly at sites prone to high indel rates.

Characterization of Three TCBE-Umax Editors with Enhanced Precision.

Expanding the editing window of base editors enhances their utility in studying SNVs. However, broader editing windows also increase the risk of bystander edits, introducing confounding factors in the assessment of SNV pathogenicity. Therefore, developing base editors with narrower editing windows while maintaining high activity remains a critical challenge, particularly for clinical applications and functional testing. Several strategies have been employed to restrict the editing window of base editors. These include structure-guided engineering to modify deaminase properties, optimizing the deaminase-Cas9 linkage to enhance editing precision, and using library-assisted protein evolution to generate editors with enhanced specificity^21–23.

To assess whether specific modifications could effectively limit the editing window of TCBE-Umax in zebrafish, we developed three editors: TCBE-Umax-rest1, featuring an N108Q mutation in the deaminase domain; TCBE-Umax-rest2, with a linker deletion between the deaminase and Cas9 domains; and TCBE-Umax-rest3, incorporating three targeted mutations (N119D, N122H, G125D) known to enhance specificity. These changes were chosen based on their ability to narrow editing windows in cell culture systems, enabling us to evaluate their impact in a vertebrate in vivo model (Fig. 4a). High-throughput sequencing at eight target sites revealed that three editors effectively narrowed the editing window to varying degrees (Fig. 4b). Among them, TCBE-Umax-rest2 demonstrated the most favorable editing profile, with activity centered at C4-C7 and the lowest bystander editing efficiency at six of the eight loci (6/8). TCBE-Umax-rest1 ranked second in performance, whereas TCBE-Umax-rest3 did not exhibit a significant advantage over the other two editors (Fig. 4b). Regarding on-target efficiency, despite reduced activity, TCBE-Umax-rest2 retains ~75% of TCBE-Umax activity. TCBE-Umax-rest3 ranked second in activity. TCBE-Umax-rest1 exhibited a significant reduction in activity but retained a substantial proportion (44%) of the efficiency of the original tool (Fig. 4c). In terms of indel formation, only TCBE-Umax-rest2 showed a significant reduction, while the other two editors exhibited no notable difference from TCBE-Umax (Supplementary Fig. 11). Notably, TCBE-Umax-rest1 does not mirror the behavior of ABE-Umax-rest1, where the same N108 mutation improves precision but simultaneously elevates indel formation (Supplementary Fig. 12). In our TCBE-Umax framework, TCBE-Umax-rest1 achieves a more restricted editing window while maintaining indel levels that remain acceptable, highlighting that identical mutations can have distinct functional consequences in different base editor architectures and therefore require empirical evaluation in each context. By-product analysis revealed that TCBE-Umax-rest1, -rest2, and -rest3 each displayed noticeable C-to-G conversion activity at specific sites (hars-T4-C7, atp2b2-T3-C6 and kcnq3-T1-C6) containing the TCT motif (Supplementary Fig. 13). This observation is consistent with findings reported in previous studies²⁴. Analysis of editing window profiles indicated that TCBE-Umax-rest1 primarily targeted C5-C7 bases, whereas TCBE-Umax-rest2 exhibited a broader editing range spanning C4-C7 bases. As for TCBE-Umax-rest3, minimal activity was consistently detected at positions C8-C9 at six of the eight loci (Fig. 4d). Based on these findings, TCBE-Umax-rest2 emerges as the preferred, best-performing editor for precision editing applications, provided that C4 editing does not interfere with experimental outcomes. If C4 editing is undesirable, TCBE-Umax-rest1 presents a viable alternative. These results underscore the potential of structure-guided engineering and linker optimization in refining CBE editing specificity, offering enhanced precision for single-base editing applications in zebrafish and other model systems.

Figure 4: Characteristics of TCBE-Umax Editors as Precise Genome Editing Tools in Zebrafish.

(a) Schematic representation of TCBE-UMax editors (TCBE-Umax, TCBE-Umax-rest1, TCBE-UMax-rest2, and TCBE-UMax-rest3) designed for precise cytosine base editing. TCBE-Umax-rest1 incorporates an N108Q mutation in the deaminase domain; For TCBE-Umax-rest2, the XTEN linker between the deaminase and Cas9 domains was deleted; TCBE-Umax-rest3 harbors three rationally designed mutations (N119D, N122H, and G125D) within the deaminase.

(b) Editing efficiency comparison of TCBE-Umax, TCBE-Umax-rest1, TCBE-Umax-rest2, and TCBE-Umax-rest3 at 8 endogenous genomic sites in zebrafish. The heat map displays average editing percentages from three independent biological replicates.

(c) Comparative analysis of average C-to-T editing efficiency among TCBE-Umax, TCBE-Umax-rest1, TCBE-Umax-rest2, and TCBE-Umax-rest3 editors at the 8 target sites shown in panel b.

(d) Schematic illustration of the editing window for TCBE-Umax tools. Green highlighting indicates the main editing windows. The SpCas9 cutting site is marked by a red triangle, with the PAM sequence and its complement highlighted in light blue.

All source data are available in the Source Data file.

Functional Analysis of Candidate Disease Genes in the F0 Generation Using TCBE-Umax

CRISPR-Cas9-mediated genome editing enables the direct creation of homozygous knockout organisms without requiring multiple generations of breeding through its capacity to generate biallelic mutations^25–27. This capability is particularly significant in zebrafish research, where phenotypic analysis can be conducted directly in the F0 generation with unprecedented speed and efficiency^25–27. We aimed to investigate whether biallelic mutations in disease-causing genes can be generated using our TCBE editor and whether the resulting phenotypes can be screened in the F0 generation. To test this approach, we selected zebrafish orthologs of six human disease genes (myo7aa, hars, med4, nup160, etfa, myhz1.1), which, when mutated, cause diseases affecting multiple tissues. MYO7A, a motor protein essential for inner ear and retinal function, causes Usher syndrome type 1B and non-syndromic hearing loss when mutated²⁸. We induced a premature stop codon in zebrafish myo7aa by changing C→T (CGA (Arg) to TGA (stop)) base using TCBE-Umax, achieving 100% base conversion. We analyzed lateral line hair cell function utilizing YO-PRO-1 dye, which is taken up by mechanotransduction channels. Control animals showed normal hair cell development, while edited animals showed compromised hair cell function, demonstrated by the absence of YO-PRO-1 dye (Fig. 5a). Similarly, we generated premature stop codons using the TCBE-Umax editor in the zebrafish orthologs of HARS1, MED4, and NUP160. HARS1 encodes a tRNA synthetase crucial for protein synthesis associated with Usher syndrome type 3B and Charcot-Marie-Tooth disease type 2W²⁹. MED4, a transcriptional regulator in the mediator complex, is associated with developmental delays and congenital anomalies³⁰. NUP160, a nuclear pore protein involved in molecular transport, is associated with neurodevelopmental disorders and steroid-resistant nephrotic syndrome³¹. Mutations in these three genes showed early developmental abnormalities (Fig. 5b, 5c, and 5d). Additionally, we introduced a premature stop codon in etfa, a mitochondrial protein essential for fatty acid metabolism, which causes Multiple Acyl-CoA Dehydrogenase Deficiency³². Control and edited larvae were analyzed using whole-mount staining with Oil Red O (ORO) to assess lipid accumulation in the liver. Indeed, edited animals showed increased lipid accumulation in the liver, suggestive of steatosis and hyperlipidemia (Fig. 5e). Finally, we utilized TCBE-Umax to create a missense mutation in the myhz1.1 gene by converting C-to-T, which changes Arg (CGT) to Cys (TGT). The zebrafish gene myhz1.1 encodes a protein predicted to be part of the myosin II complex, and its human orthologs have been implicated in multiple muscle tissue diseases³³. TCBE-Umax induced the R1398C mutation with high efficiency. Indeed, edited zebrafish larvae (R1398C) showed abnormal muscle development by 5 days post fertilization (dpf) compared to control animals, demonstrated by fluorescent-conjugated phalloidin staining (Fig. 5f). In conclusion, we were able to generate both loss of function and missense mutation in the orthologs of various human disease genes and analyzed the phenotypes in injected animals (F0), thereby facilitating rapid functional analysis of candidate disease genes.

Figure 5: Functional Evaluation of Candidate Disease Genes in the F0 Generation Using TCBE-Umax.

(a) Schematic representation of the myo7aa (R570*) target locus. The target sequence is shown with the PAM site underlined. Red highlighting indicates the original nucleotide and amino acid positions, while blue highlighting shows the expected nucleotide and amino acid substitutions. The figure presents sequencing results and phenotypes of myo7aa (R570*) mutations induced by TCBE-Umax. In the Sanger sequencing chromatograms, red arrowheads indicate the expected nucleotide substitutions, while green arrowheads mark bystander base substitutions. Adjacent to the chromatograms are images of neuromast hair cells labeled with YO-PRO-1 live dye (green). The images are representative of three independent experiments. Scale bars: 10 μm.

(b) Schematic representation of the hars (Q417*) target site. The target sequence is shown with an underlined PAM region. Red highlights the original nucleotide and amino acid, while blue highlights the anticipated nucleotide and amino acid substitutions. The sequencing data and phenotypic observations from TCBE-Umax-mediated editing of hars (Q417*) are presented. A red arrowhead indicates the expected nucleotide changes. Animals edited using TCBE-Umax exhibited developmental defects during the early stages. The images are representative of three independent experiments. Scale bar: 200 μm.

(c) A schematic diagram illustrates the med4 (Q61*) target site. The diagram shows the target sequence with its PAM sequence underlined. Red highlighting marks the original nucleotide and amino acid, while blue indicates their intended modifications. The results include Sanger sequencing chromatograms and phenotype observations from TCBE-Umax editing. Red arrowheads indicate the expected nucleotide changes, and green arrowheads mark unintended bystander substitutions in the chromatograms. Animals edited using TCBE-Umax exhibited developmental abnormalities during the early stages. The images are representative of three independent experiments. Scale bar: 200 μm.

(d) Schematic Representation of the nup160 (Q288*) gene locus. The target sequence is displayed with the PAM underlined. The original nucleotide and amino acid (red) and desired changes (blue) are highlighted. Sequencing results and phenotypic effects of the TCBE-Umax-induced nup160 (Q288*) mutation are presented. A red arrowhead marks expected nucleotide substitutions. Edited animals exhibiting early-stage developmental defects were consistently observed. The images are representative of three independent experiments. Scale bar: 200 μm.

(e) Schematic Diagram of the etfa (Q256*) target locus. The target sequence is shown with the PAM underlined. The original nucleotide, amino acid (red), and expected changes (blue) are highlighted. Sequencing results and phenotypic effects of etfa (Q256*) mutations induced by TCBE-Umax are presented. A red arrowhead marks expected nucleotide substitutions. Edited animals exhibited increased lipid accumulation, a consistent finding across three experiments, scale bar: 200 μm.

(f) Schematic Diagram of the myhz1.1 (R1398C) target locus. The target sequence is displayed with the PAM underlined. Original (red) and expected (blue) nucleotide and amino acid changes are highlighted. Sequencing results and phenotypic observations of myhz1.1 (R1398C) mutations induced by TCBE-Umax are shown. The red arrowhead marks expected nucleotide substitutions, while a green arrowhead in Sanger chromatograms indicates bystander substitutions. Edited animals exhibited abnormal muscular development, as observed through fluorescent-conjugated phalloidin staining. The images are representative of three independent experiments, scale bar: 100 μm.

Rapid Functional Validation of Genetic Variants associated with genetic disorders.

The completion of genome sequencing projects has revealed countless single-nucleotide polymorphisms (SNPs), predominantly missense mutations. Given the high efficiency and specificity of TCBE-Umax editors, we investigated their potential for functional analysis of these genetic variants, particularly “variants of uncertain significance” (VUS), which are not known to be either benign or disease-causing.

Single-base editing technologies (CBE or ABE) enable precise single-nucleotide modifications; however, their use in organism-wide studies is limited by variable editing efficiency across different genetic locations, sequence context dependencies, and inherent technical constraints. We set out to utilize the TCBE-Umax system, which boasts high efficiency and minimal sequence context bias, to assess the pathogenicity of VUS in Myo7a and Cdh23, two key proteins implicated in hereditary hearing loss. Cadherin-23, encoded by the CDH23 gene, is essential for forming stereocilia tip links in sensory hair cells. Mutations in CDH23 can cause either Usher syndrome type 1D or non-syndromic deafness (DFNB12)³⁴, both of which impair mechanotransduction and hearing. Similarly, the MYO7A gene encodes Myosin VIIA, an actin-based motor protein crucial for the organization of stereocilia, intracellular trafficking, and mechanotransduction in hair cells. Mutations in MYO7A result in Usher syndrome type 1B or DFNB2, causing progressive hearing loss and vestibular dysfunction.³⁵ In zebrafish, loss-of-function mutations in either cdh23 or myo7aa disrupt hair cell function and stereocilia integrity, leading to impaired mechanosensory signaling and loss of the Acoustic Startle Response (AEBR)^{36, 37}. For these studies, we utilized the zebrafish lateral line system, which enables rapid assessment of hair cell function in an intact organism. The molecular machinery of hair cells is highly conserved between zebrafish and humans, making findings from this model system directly relevant to human hearing loss research. As described earlier, we employed the YO-PRO-1 live dye uptake assay to quantitatively measure mechanotransduction function, while the accessibility of hair cells allows direct observation of phenotypic changes (Fig. 6a).

Figure 6: Rapid Functional Evaluation of VUS Associated with Hearing Loss.

(a) The schematic representation of the zebrafish lateral line sensory system consists of clusters of mechanosensory hair cells known as neuromasts. These lateral line hair cells are functionally and molecularly similar to those in the zebrafish inner ear, making them an excellent model for studying hearing function. Functional hair cells in live zebrafish embryos can be labeled using YO-PRO-1 dye, taken up through hair cell mechanotransduction channels. A schematic of the stereocilia structure highlights key molecular components, including Cdh23 and Myo7a.

(b) Experimental workflow for the rapid assessment of VUS pathogenicity using zebrafish.

(c) Schematic diagrams of TCBE-Umax2 constructs, designed to enhance cytosine base editing efficiency.

(d) Comparison of editing efficiencies between TCBE-Umax and TCBE-Umax2 at five target sites. Fold changes are indicated above each group. Data are presented as mean ± standard deviation (SD) from three biological replicates.

(e) Summary of all 15 VUS target sites, including their editing efficiencies and observed outcomes.

(f) Yo-PRO-1 staining results corresponding to the 15 VUS target sites. Three independent experiments yielded similar results. Scale bars: 10 μm. Sites edited by TCBE-Umax2 are highlighted in red boxes.

All source data are available in the Source Data file. Panels 6a, b generated in biorender.com

We selected 15 missense mutations (C>T) in the MYO7A and CDH23 genes from the ClinVar database for functional evaluation (Supplementary Figs. 14, 15, and 16). Of these mutations, 14 were single-nucleotide polymorphisms (SNPs) classified as variants of uncertain significance (VUS), while one was considered likely pathogenic and served as our positive control. Our screening pipeline began with co-injecting TCBE-Umax mRNA and target variant-specific gRNAs into zebrafish embryos at the single-cell stage. After 2 dpf, we evaluated editing efficiency by analyzing pooled genomic DNA from 10 random embryos using PCR and Sanger sequencing, with variants showing greater than 50% efficiency advancing to further testing. At 5 dpf, we assessed morphology and performed Acoustic Startle Response tests, followed by YO-PRO-1 live staining at 6 dpf to evaluate hair cell function. Finally, we conducted single embryo genotyping for variants exhibiting phenotypic effects on groups of 5–12 embryos to validate editing outcomes and determine the correlation between editing efficiency and phenotype severity. Specifically, variants showing positive phenotypes in both assays underwent genotyping with 5 embryos per group, while cases with phenotypic variability required genotyping of 12 embryos to accurately determine variant pathogenicity through the correlation of editing efficiency with phenotypic severity (Fig. 6b).

Initial assessment of editing efficiency in pooled embryos at 2 dpf revealed that 10 of the 15 sites achieved average editing efficiency above 50% (Supplementary Table 2). Among these 10 variants, we performed acoustic response assays, and five of the variants did not respond to sound stimuli, strongly indicating pathogenicity, a finding subsequently confirmed through YO-PRO-1 staining and genotyping (Fig. 6e, 6f, Supplementary Fig. 17, and Supplementary Table 2). The remaining five high-efficiency variants underwent YO-PRO-1 staining and genotyping analysis in 12 individual embryos per group. Despite high editing efficiency (>50%, reaching 100% at some sites), these embryos exhibited staining patterns comparable to those of the controls, suggesting that these variants are likely benign (Fig. 6e, 6f, Supplementary Fig. 17, and Supplementary Table 2).

As mentioned above, in our initial screening, five variants displayed low editing efficiency (<50%), complicating accurate pathogenicity assessment. To address this limitation, we further optimized TCBE-Umax to improve its efficiency for F0-generation zebrafish studies. Recent studies exploring nCas9 engineering to enhance base editing have revealed that N1317R + A1322R mutations in Cas9 improve target DNA binding, showing enhanced activity at 5 of 6 tested sites³⁸. Based on these findings, we incorporated these mutations into the TCBE-Umax editor to create the TCBE-Umax2 editor (Fig. 6c).

TCBE-Umax2 significantly improved on-target editing efficiency across all five previously low-efficiency sites, achieving 2-to 15.5-fold increases compared to TCBE-Umax and raising all average editing efficiencies above 50% (Fig. 6d). Subsequent phenotypic and genotypic assessment of these variants revealed another two as pathogenic, exhibiting uniform loss of acoustic response and reduced YO-PRO-1-positive hair cells, while two others showed normal function, confirming them as benign (Fig. 6e, 6f, Supplementary Fig.17 and Supplementary Table 2). One variant, myo7aa-S211N, presented an intriguing case: despite consistently high editing efficiency (>50%) across all 12 tested embryos, YO-PRO-1 staining results varied significantly, some embryos showed substantial signal reduction, while others remained normal (Supplementary Table 2). This variability suggests the need for further evaluation in stable germline-transmitted zebrafish lines, highlighting that single mutations can produce varying phenotypic manifestations, as observed in human patients.

It is worth noting that TCBE-Umax2 is not a universal solution. When benchmarked across a broader panel of endogenous zebrafish loci, TCBE-Umax2 displayed more variable C-to-T editing efficiencies than TCBE-Umax and was often associated with elevated indel frequencies as well as detectable gRNA-dependent off-target edits at some predicted sites (Supplementary Fig.18–20). Nonetheless, the modular design of TCBE-Umax2, which combines a hyperactive deaminase with an activity-enhanced Cas9, allowed us to reuse the same three precision-enhancing mutations previously engineered into TCBE-Umax. The resulting variants maintained efficient C-to-T editing while restricting activity to narrow windows across representative targets (Supplementary Fig. 21).

Taken together, our optimized strategy, using TCBE-Umax as a first-line editor and TCBE-Umax2-based variants as complementary tools for refractory loci, enabled rapid functional assessment of 14 VUS in zebrafish in the F0 generation, establishing a general framework for large-scale, organismal-level evaluation of variant pathogenicity.

Strategy for selecting the appropriate TadA-based cytosine base editor (T-CBE)

We have developed several TCBE-Umax editors, each with unique characteristics (Supplementary Fig. 22), and selecting the most suitable TadA-based cytosine base editor (T-CBE) is crucial for achieving optimal results. Our selection strategy, outlined in Figure 7, begins by checking for a GG motif 6–21 base pairs downstream of the target cytosine to meet the NGG PAM requirement. If present, we prioritize tools recognizing NGG PAM sequences, favoring TCBE-Umax (editing window: 4–10 bases on the protospacer), TCBE-Umax-ex1 (window: 3–12 bases), or TCBE-Umax-ex2 (window: 4–16 bases) for maximum efficiency, with TCBE-Umax-ex1 being the top choice due to its minimal or absent indel induction. If TCBE-Umax shows suboptimal activity at a given site, TCBE-Umax2 may be considered as an alternative to improve editing efficiency. For applications prioritizing precision, TCBE-Umax-rest1 (window: 5–7) or TCBE-Umax-rest2 (window: 4–7) are recommended. Notably, TCBE-Umax-rest1 avoids editing at cytosine position 4, although it sacrifices some on-target efficiency. If no GG motif is detected, we suggest using editors with an “NGN” PAM and testing TCBE-Umax-SpRY, TCBE-Umax-SpG, and TCBE-Umax-NG to identify the most efficient editor. In the absence of an “NGN” PAM, TCBE-Umax-SpRY is the preferred option, with PAM preference ranked as NAN > NYN (where Y is C or T), and the target cytosine should ideally fall within the primary editing window of the editor to maximize effectiveness. Overall, the TCBE-Umax effectors we’ve engineered hold substantial promise for overcoming the limitations of earlier base editors, enhancing disease model development, and advancing potential gene therapy applications.

Figure 7: Approach for selecting the optimal TadA-based cytosine base editor (T-CBE).

The selection of the right tool for C-to-T editing depends on the presence of a specific DNA sequence motif, and a suitable editor can be identified as follows: Identify GG motif within 6–21 base pairs downstream of the target cytosine (red asterisks). Option 1: If a GG motif is present, (1) for high efficiency, consider TCBE-Umax (window 4–10, with the PAM defined as positions 21–23), TCBE-Umax-ex1 (window 3–12), or TCBE-Umax-ex2 (window 4–16); If TCBE-Umax shows suboptimal activity at a given site, TCBE-Umax2 may be considered as an alternative to improve editing efficiency. (2) for high precision, choose TCBE-Umax-rest1 (window 5–7) or TCBE-Umax-rest2 (window 4–7). To minimize interference at the C4 position, TCBE-Umax-rest1 is the preferred choice, albeit with a slight trade-off in on-target efficiency. Option 2: If no GG motif is found, first test NGN PAM across all three editors (TCBE-Umax-SpRY, TCBE-Umax-SpG, and TCBE-Umax-NG) and select the optimal one. If unavailable, use TCBE-Umax-SpRY, prioritizing PAM sequences in this order: NAN > NYN. (3) The edited cytosine should ideally be positioned within the main activity window to fully leverage the capabilities of each editor. The schematics, represented in different colors, illustrate various cytosine-based editing tools. The blue box represents the possible targeting window, while the magenta box indicates the main targeting window. Figure generated in biorender.com

Discussion

In this study, we engineered and optimized a series of 13 next-generation TadA-based cytosine base editors (TadCBEs) for precise genome editing in zebrafish. Our findings demonstrate that these TadCBEs exhibit significantly improved editing efficiency, broader sequence compatibility, and reduced off-target effects compared to traditional cytosine base editors. Through optimization of key mutations (N46V, Y73P, V82S, and Q154R) in the TadA deaminase domain, TCBE-Umax demonstrated enhanced base editing efficiency while minimizing unintended modifications. We successfully integrated these improved CBEs with alternative Cas9 enzymes (SpRY, SpG, and NG) to expand targetable genomic regions, thereby overcoming PAM constraints and enabling more flexible base editing applications.

Unlike the previous variant zTadCBE, TCBE-Umax did not display detectable A-to-G activity at the same sites tested for zTadCBE¹³. This improved selectivity aligns with previous findings from mammalian cell-based assays, further validating TCBE-Umax as a highly specific base editor^9–11.

A crucial aspect of our study was investigating the PAM recognition preferences of TCBE-Umax editors to expand their targeting range and improve precision. The SpRY enzyme was engineered from SpG by introducing five additional mutations, which enabled the recognition of NNN PAM³⁹. While the fact that SpRY is compatible with cytosine deaminases suggested that SpG should be compatible as well, CBE4-SpG previously showed unexpectedly low activity in zebrafish¹⁵. This discrepancy, though not fully understood, has rendered CBE4-SpG impractical for zebrafish applications. Another enzyme, SpCas9-NG, also recognizes NG PAM⁴⁰. Here, we incorporated a TadA-based deaminase derived from adenine base editor (ABE) tools into TCBE-Umax. Previous research demonstrated that ABE-SpG functions effectively in zebrafish¹⁵, leading us to hypothesize that TCBE-Umax-SpG and TCBE-Umax-NG editors would also be functional. Our results confirmed this hypothesis, as both editors showed robust activity in zebrafish. At certain loci, both editors even demonstrated superior performance compared to TCBE-Umax-SpRY, highlighting that selecting different Cas9 enzymes can optimize base editing efficiency for specific genomic targets.

Indel formation remains a significant challenge in cytosine base editing due to multiple factors, including DNA repair pathway activation, Cas9-induced nicking, and sequence context.⁶ Current models suggest that when the non-edited strand undergoes repair before the edited strand, the C-to-U conversion may be recognized as DNA damage, triggering indel formation. The spatial relationship between the nicking site and the edited base is particularly crucial for repair efficiency and fidelity. Research has shown that suboptimal positioning of the nick, either too close to or too far from the editing site, can cause DNA repair mechanisms to introduce unintended mutations⁴¹. In our study, we observed that the TCBE-Umax-ex1 editor significantly reduced indel formation at identical target sites compared to TCBE-Umax. We hypothesize that this reduction primarily stems from a shift in the editing window, which repositions the nicking site to an optimal distance from the edited base, thereby substantially reducing indel formation. However, given that chromatin architecture varies across species, additional research is needed to determine whether this phenomenon is conserved in other model organisms.

Base editing precision remains a significant challenge in genome editing. In 2022, Chen et al. showed that structure-guided engineering of ABEs (ABE8e-N108Q and ABE9) can narrow the editing window while largely preserving activity, but when we applied analogous “rest” designs to ABE-Umax in zebrafish, the resulting editors narrowed the window yet substantially reduced efficiency and increased indel formation compared with ABE-Umax system, indicating that results from mammalian cell culture studies may not directly translate to zebrafish or other model organisms. By contrast, introducing the corresponding N108Q mutation into TCBE-Umax (TCBE-Umax-rest1) caused only a moderate decrease in on-target activity while keeping indel levels low and confining C-to-T edits to a short C5-C7 window, underscoring that identical mutations can have distinct effects in different base-editor platforms and must be empirically evaluated in each system.

A key advantage of the TCBE-Umax system over existing tools is its superior editing efficiency. In our study, we successfully delivered the tool through mRNA injection into zebrafish embryos. However, previous research has shown that protein delivery of Cas9 typically achieves higher efficiency than mRNA delivery in zebrafish, primarily due to the rapid embryonic development in this model organism.⁴² This suggests that delivering TCBE-Umax as a protein could potentially yield even better results. While synthetic mRNA remains convenient for laboratory use, several alternative delivery strategies could further enhance the efficiency of single-base editing. These include using chemically modified mRNA to enhance stability and circular RNA (circRNA) to improve protein translation efficiency.^{43, 44}

Relevant to clinical diagnosis and genetic counseling, TCBE-Umax2 enabled rapid functional assessment of VUS at loci that were poorly edited by TCBE-Umax, extending our F0 zebrafish pipeline to otherwise refractory sites. However, as shown in our results, the increased activity of TCBE-Umax2 is accompanied by higher indel and gRNA-dependent off-target editing, this trade-off may reflect a mechanistic link between enhanced on-target activity and collateral editing outcomes. So, it is best regarded as a complementary tool for selected low-efficiency targets rather than a universal enhancer. More broadly, the way TCBE-Umax2 and its precision variants were assembled from a hyperactive TadA deaminase, an activity-enhanced Cas9, and window-restricting mutations, suggests a modular design strategy that could, in principle, be reused to build customized base editors in other experimental systems.

Recent studies have utilized CRISPR-based approaches to assess the variant pathogenicity of genetic variants at the cellular level and identify crucial residues in genes associated with monogenic diseases.^45–47 However, the complexity and heterogeneity of genetic mutations make it challenging to fully evaluate their pathogenic effects through cellular studies alone. In clinical practice, different missense mutations within the same gene often lead to varying disease severities, a phenomenon that can only be accurately studied through whole-organism models. This limitation becomes particularly significant when developing personalized disease models, screening drugs, and exploring therapeutic strategies. Consequently, efficient assessment of VUS at the organism level is of utmost importance, and our study provides a powerful platform for rapid functional assessment of human genetic variants.

Using AlphaMissense, a deep-learning model trained on protein structure, evolutionary conservation, and functional impact data, we evaluated 15 VUS sites⁴⁸. Our analysis revealed notable differences between computational predictions and experimental results, highlighting that predictive models cannot substitute for empirical testing. This finding highlights how point mutations behave in complex biological systems; even the most sophisticated prediction models require experimental validation.

In summary, our work establishes TCBE-Umax and its derivatives as powerful tools for precise cytosine base editing in zebrafish. However, several considerations should guide their use. First, although indels and off-target activity are markedly reduced relative to earlier editors, these outcomes are not eliminated; locus-specific chromatin states, and gRNA-dependent behaviors can still influence editing outcomes, particularly in F0 embryos with mosaicism can complicate phenotype-genotype interpretation. Second, Zebrafish codon-optimized TadA-based CBEs here may require further optimization in other model organisms. In particular, the balance between efficiency, precision, and indels can shift substantially across species. Third, although increased PAM flexibility broadens genomic accessibility, variants located in repetitive or highly structured genomic regions remain challenging. Moving forward, focus on developing Cas variants, and the position of the enzyme or linker architecture can further increase editing efficiency while reducing off-target risks. Together, these considerations highlight both the promise and current limitations of TadA-derived CBEs and outline avenues for further optimization toward more predictable, species-agnostic base-editing platforms.

Methods

Ethical Statement

All zebrafish experiments were conducted in accordance with protocol 24–18, approved by the Institutional Animal Care and Use Committee (IACUC) of Oklahoma Medical Research Facilities. All procedures adhered to the NIH Guidelines for the Care and Use of Laboratory Animals and were performed in an Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC)-certified facility. Steps were taken to minimize animal suffering and reduce the number of animals used.

Zebrafish maintenance

Adult zebrafish (Danio rerio) of the wild-type NHGRI-1⁴⁹ and TAB-5 strains were maintained in a recirculating aquaculture system (Aquaneering Inc., San Diego, CA) at 28.5 ± 0.5°C on a 14-hour light/10-hour dark cycle. Water quality parameters were monitored daily and maintained within optimal ranges: pH 7.2–7.6, conductivity 400–600 μS/cm, and dissolved oxygen >6.0 mg/L. Fish were fed three times daily with commercially available dry food (Gemma Micro 300, Skretting) and live feed. For all experiments, breeding pairs (12–15 months of age) were randomly selected from independent pools of 30 males and 30 females to minimize selection bias. The evening before breeding, male and female fish were separated by a divider in breeding tanks. The dividers were removed the following morning, at the onset of the light cycle, to enable natural spawning. Embryos were collected within 30 minutes of spawning and maintained in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄, 0.0001% methylene blue) at 28.5°C.

Plasmid Construction

The zTadCBE plasmid, developed from our previous study¹³, served as the foundation for constructing TCBE-1.1, TCBE-1.2, TCBE-1.3, and TCBE-1.4 by replacing its TadA region with synthesized, zebrafish codon-optimized fragments bearing specific mutations. To create TCBE-Umax-SpRY, TCBE-Umax-SpG, and TCBE-Umax-NG, the SpCas9 (D10A) fragment of TCBE-1.2 (TCBE-Umax) was swapped with codon-optimized zebrafish SpRYCas9 (D10A), SpGCas9 (D10A), and NG-Cas9 (D10A), respectively, using PCR-amplified fragments from the ABE-Umax-SpRY, ABE8e-SpG, and ABE8e-NG plasmids^{3, 5}. For TCBE-Umax-ex1, the TadA*-Umax monomer was inserted directly at position 1054 of SpCas9 without a linker, while TCBE-Umax-ex2 was engineered by removing the HNH domain of Cas9, connecting SpCas9 S793 to the N-terminus of TadA*-Umax with a GGS-linker, and linking the C-terminus of TadA*-Umax to SpCas9 R919 with an SGG-linker. For TCBE-Umax-rest1 and TCBE-Umax-rest3, specific mutations, N108Q for the former and N119D/N122H/G125D for the latter, were introduced into the TadA*-Umax deaminase domain, and TCBE-Umax-rest2 was generated by removing the XTEN linker between TadA*-Umax and Cas9 in TCBE-Umax. Additionally, TCBE-Umax2 was developed by incorporating N1317R and A1322R mutations into the Cas9 region of TCBE-Umax. All fragment fusions, targeted mutations, and deletions were performed using the Vazyme Mut Express II Fast Mutagenesis Kit V2 (Cellagen Technology LLC, CA, USA), and the resulting plasmids were transformed into DH5α Chemically Competent Cells (New England Biolabs, MA, USA) for amplification.

Single guide RNAs (sgRNAs) and mRNA synthesis

All single guide RNAs (sgRNAs) used in this study were chemically synthesized and modified by GenScript Inc. (Piscataway, NJ, USA) and Synthego Inc. (Redwood City, CA, USA) with MS (2′-O-methyl (M) and 2′-O-methyl 3′phosphorothioate) modifications at both termini to enhance stability and reduce immune responses. Complete target sequences and modifications for all sgRNAs are provided in Supplementary Table 3. For mRNA synthesis, template DNA was linearized using XbaI restriction enzyme (New England Biolabs, Ipswich, MA, USA) at 37°C for 2 hours, with complete linearization confirmed by agarose gel electrophoresis. In vitro transcription was performed using T3 mMESSAGE mMACHINE Kit (ThermoFisher Inc., Carlsbad, CA, USA) for all constructs. All synthesized mRNAs were purified using the Monarch RNA Cleanup Kit (New England Biolabs) as described earlier⁵⁰, with RNA integrity assessed by denaturing agarose gel electrophoresis and quality verified using Nanodrop spectrophotometry (A260/280 >1.8, A260/230 >2.0). Purified capped mRNAs and synthetic sgRNAs were dissolved in RNase-free water to prepare 2000 ng/μL stock solutions, divided into single-use aliquots (5 μL) to avoid freeze-thaw cycles, and stored at −80°C in RNase-free tubes.

Microinjection, morphological phenotyping, and imaging

For microinjection, a mixture containing sgRNA (200 ng/μL) and Cas9 mRNA (400 ng/μL) in a total volume of 2 nL was injected into zebrafish embryos at the one-cell stage using a Pneumatic PicoPump (World Precision Instruments) with borosilicate glass capillary needles⁵⁰. Embryonic development was monitored daily, and at 2–5 days post-fertilization (dpf), embryos were anesthetized with 0.016% tricaine/MS-222 (Sigma-Aldrich, St. Louis, MO, USA) and mounted in 3% methylcellulose (Sigma-Aldrich) for morphological analysis and imaging. For lateral line neuromast hair cell visualization, 5 dpf embryos were incubated with 1 mM YO-PRO^™−1 (Invitrogen, Waltham, MA, USA) for 1 hour at room temperature (23 ± 1°C) in the dark. All imaging was performed using an Olympus SZX12 stereomicroscope equipped with an Olympus DP71 color digital camera (Olympus, Tokyo, Japan) under consistent illumination and exposure settings. Following imaging, individual embryos were collected for genomic DNA extraction and genotyping to establish precise correlations between phenotype and genotype. For each experimental condition, a minimum of 50 embryos were analyzed across three independent biological replicates to ensure reproducibility.

Base editing analysis and genotyping

To quantify base editing efficiency, genomic DNA was extracted from three independent pools of randomly selected embryos (n=6 per pool) at 48 hours post-fertilization using alkaline lysis buffer (50 mM NaOH). Target regions spanning 150–300 bp around each target site were amplified using HotStart Taq-Plus DNA polymerase (Qiagen, Hilden, Germany) with locus-specific primers (sequences provided in Supplementary Table 3). PCR products were purified using the DNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA, USA) and subjected to Sanger sequencing. Sequencing chromatograms were analyzed using EditR software (v1.0.10) to quantify editing frequencies and determine editing outcomes⁵¹. For phenotype-genotype correlation studies, individual embryos were collected post-imaging or functional analysis.

Screening for Germline Transmission Events

To assess germline transmission of edited alleles, F0 founder fish were outcrossed with wild-type partners at sexual maturity (3–4 months post-fertilization). From each F0 cross, embryos were collected and divided into two independent pools (n=5 embryos per pool) at 2 days post-fertilization (dpf). Genomic DNA was extracted from each pool using alkaline lysis buffer, and the target loci were amplified using locus-specific primers (Supplementary Table 3). PCR products were analyzed by Sanger sequencing as described earlier⁵². Founder fish were classified as positive for germline transmission when consistent editing patterns were observed in embryo pools. The germline targeting efficiency was calculated as (number of positive founders/total number of screened founders) × 100%. For each positive founder, transmission rates were determined by PCR amplification and Sanger sequencing of eight individual F1 embryos, with successful transmission defined by the presence of the intended base edit. Sequence traces were analyzed using SnapGene software (v6.0) to identify precise editing outcomes, and positive F1 carriers were raised to establish stable gene-edited lines.

High-throughput sequencing (HTS) and analysis

Genomic DNA was extracted from pooled wild-type (n=10) or injected embryos (n=10) at 48 hours post-fertilization using an alkaline lysis buffer protocol (50 mM NaOH, incubated at 95°C for 20 minutes, neutralized with 1M Tris-HCl, pH 8.0). Target regions were amplified using high-fidelity Q5 polymerase (New England Biolabs) and target-specific primers (primer sequences listed in Supplementary Table 3). PCR products were purified using the Zymo DNA purification kit (Zymo Inc. USA). PCR amplicons were sequenced on an Illumina MiSeq platform using 2 × 300 bp paired-end chemistry. Sequencing data quality was assessed using FastQC (v0.11.9), and reads were analyzed using CRISPResso2 (v2.2.8) with default parameters to quantify editing outcomes⁵³. Raw sequencing data is deposited in the NCBI Sequence Read Archive.

DNA Off-target analysis

Potential off-target sites for each gRNA were comprehensively identified using two complementary computational tools: Cas-OFFinder (v3.0) and CRISPOR (v4.99). The analysis parameters were set to allow up to four mismatches in the zebrafish genome (GRCz11/danRer11), including DNA/RNA bulges. Each predicted off-target site was assigned a specificity score using the scoring algorithm of the CRISPOR tool, which considers both the number and position of mismatches relative to the PAM sequence. The top three highest-scoring potential off-target sites for each gRNA were selected based on their CFD (Cutting Frequency Determination) scores and genomic context. These selected sites were subsequently evaluated for off-target editing events using targeted amplicon sequencing on an Illumina NovaSeq 6000 platform with a minimum coverage depth of 10,000–20,000 reads per site. To measure gRNA-independent off-target editing in orthogonal R-loop assays, a mixture containing 200 ng/μL SpCas9 guide RNA, 400 ng/μL TCBE-Umax mRNA, 200 ng/μL SaCas9 guide RNA, and 300 ng/μL dSaCas9 mRNA was prepared in a total volume of 2 nL and injected into zebrafish embryos at the one-cell stage; subsequently, these selected sites were assessed for off-target editing events through targeted amplicon sequencing performed on an Illumina NovaSeq 6000 platform.

Statistics & Reproducibility

Sample sizes were not predetermined using statistical methods, but all experiments were performed with a minimum of three independent biological replicates to ensure reproducibility, with exact sample sizes detailed in figure legends and Source Data. No data points were excluded from analyses, and samples were randomly assigned to experimental groups to minimize bias. Data are presented as mean ± standard deviation (SD), and statistical analyses were conducted using GraphPad Prism version 8.0.2 (GraphPad Software, San Diego, CA, USA). Statistical significance was evaluated using two-tailed unpaired Student's t-test for comparing base editing efficiencies between different groups, while two-tailed paired Student's t-test with nonparametric Wilcoxon matched-pairs signed rank test was employed for mean editing efficiency comparisons between paired groups. For all analyses, statistical significance was defined as P < 0.05 (*), P < 0.01 (**), and P < 0.001 (***), with exact P-values reported in the figure legends and Source Data files. All raw data and statistical analyses are available in the accompanying Source Data files to ensure transparency and reproducibility of our findings.

Data Availability

NGS data are available on the National Center for Biotechnology Information Sequencing Read Archive (SRA) database under project numbers PRJNA1231105 and PRJNA1231090. All data supporting the findings of this study are available within the article and Supplementary Information files. Source data are provided in this paper.