A general theoretical framework to design base editors with reduced bystander effects

Base editors (BEs), derived from CRISPR-Cas9 systems, offer a precise approach to genome editing by performing single nucleotide substitutions without causing double-stranded DNA breaks. This precision is crucial for gene therapy applications targeting genetic diseases caused by point mutations. However, a significant challenge in base editing is the bystander effect, where the deaminase domain of the BE modifies not only the intended target base but also similar bases within the editing window (4-10 nucleotides). This reduces editing precision and can lead to undesired off-target modifications. Current strategies for mitigating bystander effects primarily rely on experimental screening of mutations guided by structural analysis. This approach can be time-consuming and inefficient. There is a need for a theoretical framework to predict the effects of mutations on BE activity and selectivity, thus accelerating the design of high-precision BEs. This paper addresses this need by developing a multi-scale theoretical approach that combines a discrete-state stochastic (chemical-kinetic) model and all-atom molecular dynamic simulations to evaluate the probabilities of editing both the target base and bystander bases. The model is parameterized using experimental data and is used to guide the design of new mutations to improve BE selectivity. The findings are then validated experimentally.

The development of CRISPR-Cas9 based genome editing tools has revolutionized biomedical research. While CRISPR-Cas9 systems can induce double-stranded DNA breaks, leading to unpredictable editing outcomes, BEs leverage nickase Cas9 (nCas9) for more precise modifications. Cytosine base editors (CBEs), for example, fuse a cytidine deaminase domain with nCas9 to convert cytosine (C) to uracil (U), subsequently converted to thymine (T) by cellular repair mechanisms. Although several BE variants have been engineered to enhance editing efficiency and purity, the challenge of bystander editing remains. Previous studies have employed rational mutagenesis guided by structural insights to improve BE selectivity, for example, the N57G mutation in A3A-BE3 greatly reduced bystander effects. However, a general theoretical framework to quantitatively predict and guide the design of mutations for enhanced BE selectivity has been lacking. Molecular dynamics (MD) simulations have been applied to study BE activity, but a comprehensive multi-scale model combining kinetic modeling and MD simulations to explicitly address bystander effects was absent prior to this study.

The authors developed a discrete-state stochastic model to describe the base editing process. This chemical-kinetic model incorporates key steps, including Cas9 binding to single-stranded DNA (ssDNA), cytidine binding to the deaminase active site, deamination to uracil, and subsequent uracil to thymine conversion. The model considers both target and bystander cytidine editing, explicitly calculating the probabilities of various editing outcomes. A crucial parameter in the model is ΔG_m, representing the binding affinity between the deaminase and ssDNA. This parameter is modulated by introducing mutations into the BE and its values are measured through all-atom MD simulations using alchemical free energy calculations. The model accounts for several kinetic parameters, including Cas9-ssDNA binding rate, deamination rate, and the impact of DNA repair on Cas9 rebinding. The first-passage probabilities method is used to solve the model and obtain probabilities for different editing outcomes (e.g., only target edited, only bystander edited, both edited). The model is parameterized by fitting it to experimental data from A3A-BE3 and its variants. The authors use MD simulations to estimate changes in binding free energy (ΔΔE) associated with different mutations in the deaminase, providing additional input to the kinetic model. The model is then used to design mutations in the A3G-BE system aimed at improving its selectivity. These mutations are introduced, and their effects are experimentally validated in various genomic loci using techniques including cell transfection, genomic DNA extraction, amplicon sequencing, and analysis using CRISPResso2. Statistical analysis is performed on the experimental results, and the findings are compared against the model's predictions. The binding free energies were obtained by performing alchemical free energy calculations via MD simulations in the Gromacs package. The binding free energy difference between target and bystander cytosines was estimated for four A3A CBE variants (A3A(S99A), A3A(Y130F), A3A(N57Q), and A3A(N57A)). This involved constructing a thermodynamic cycle and calculating the free energy change for the A3A-ssDNA complex and A3A alone due to mutations, yielding ΔΔE_t and ΔΔE_b for the target and bystander sites respectively. In the experimental validation, HEK293T, HeLa, K562, and Jurkat cell lines were used, with transfections performed using Lipofectamine 2000. Genomic DNA was extracted and amplicon sequencing was used to quantify editing efficiency. Sanger sequencing was also used to analyze the editing events.

The study's key findings include the development of a comprehensive theoretical framework that successfully predicts and explains the bystander effects observed in base editing. The model accurately reproduces experimental data obtained from A3A-BE3 and its variants. The authors demonstrate that the BE selectivity is non-monotonically dependent on ΔG_m (binding affinity), implying an optimal binding affinity for achieving maximum selectivity. By analyzing the model, they propose general design principles for creating BEs with reduced bystander effects, such as modulating the binding affinity between the deaminase and ssDNA to control the residence time of the deaminase on ssDNA. They apply this framework to design point mutations at the T218 position of A3G-BEs, resulting in variants (A3G3.14 and A3G3.15) showing improved editing selectivity at multiple genomic loci. The experimental results validate the model's predictions, demonstrating the effectiveness of the computational approach in designing high-selectivity BEs. The model successfully explains why the N57G mutation improves A3A-BE3 selectivity. The mutation increases the binding free energy (ΔΔE), leading to a higher ratio of target-to-bystander editing. Calculations show a non-monotonic effect, with optimal ΔΔE values for maximal selectivity. Beyond A3A-BE3, the model is successfully employed to optimize A3G3.1, designing T218S and T218N mutations that enhance editing selectivity experimentally. The study demonstrates a tight coupling between mutagenesis stringency (ΔΔE) and genomic site context in determining the target-to-bystander ratio. Lastly, the authors explore the impact of other parameters (γ1 and γ3) in the model, such as the on-rate of Cas9 to ssDNA and the relative probability of unbinding versus chemical transformation. Reducing γ1 is shown to amplify the effect of deaminase mutations on selectivity. Reducing γ3 does not significantly affect the maximum selectivity but shifts its location on ΔΔEm profile.

This study's findings significantly advance our understanding of base editing mechanisms and provide a powerful tool for designing more precise BEs. The developed theoretical framework provides a quantitative link between mutations in the deaminase domain and the resultant editing selectivity. The successful prediction and experimental validation of novel A3G-BE variants demonstrates the potential of this combined computational and experimental approach to accelerate base editor engineering. The non-monotonic relationship between binding affinity (ΔG_m) and selectivity underscores the importance of fine-tuning this parameter to optimize BE performance. The model's ability to accurately predict editing outcomes across multiple genomic loci and cell lines highlights its versatility and robustness. The identified general principles for BE design provide a valuable guide for future engineering efforts. The study's findings have broad implications for gene therapy, paving the way for the development of more effective and precise tools for correcting genetic defects.

This study presents a novel theoretical framework, combining kinetic modeling and molecular dynamics simulations, for designing base editors with improved selectivity. The model successfully predicts and explains the effects of mutations on both target and bystander editing. The experimental validation of the model's predictions showcases its utility in guiding the design of improved BEs. This approach significantly accelerates the development of precise gene editing tools for therapeutic applications. Future work could focus on refining the model by incorporating additional factors, such as long-range interactions and chromatin structure. Further exploration of the optimal binding affinity range across different BE systems would further enhance the design process.

The model utilizes several approximations, such as considering only local sequence context near the target base and neglecting long-range interactions within the editing window. While these approximations allow for tractability and successful explanation of existing data, they may limit the model's ability to fully capture the complexity of base editing in all scenarios. The experimental validation was performed on a specific set of genomic loci and cell lines; further investigation is necessary to determine the generality of the findings. The accuracy of the MD simulations depends on the quality of the force field used and can introduce uncertainties. The model is currently parameterized for cytosine base editors but could be extended to other base editing systems such as adenine base editors. Further, the limited number of cell lines used in the experimental validation may not fully capture the effects of cellular context and variability in editing efficiency.