AI‑Designed Minimal RNA‑Guided Nucleases: Insights from Structure and Evolution

Background

RNA‑guided nucleases such as Cas9, Cas12, Cas12f, Cas14 and TnpB are often too large to fit together with guide RNAs and regulatory elements into a single adeno‑associated virus (AAV) vector, limiting gene‑therapy applications.

Historically, new nucleases were obtained by mining microbial genomes, but natural diversity provides a trade‑off: some proteins are highly active but bulky, while others are compact but lack stability or efficiency.

Structure‑ and Evolution‑guided Design

Jennifer Doudna’s team introduced a design paradigm that combines protein‑level evolutionary patterns with three‑dimensional structural information to reconstruct minimal RNA‑guided nucleases while preserving editing activity. The work, “Structure and evolution‑guided design of minimal RNA‑guided nucleases,” was published in Science (https://www.science.org/doi/10.1126/science.aed6123).

The workflow flips the traditional discovery‑first model. First, indispensable structural elements of existing nucleases are identified; then non‑essential regions are removed or redesigned.

Figure 1: Design strategy for TnpB using the ESM‑IF1 reverse‑folding model.

ESM‑IF1 inverse‑folding generates backbone scaffolds. Because the model alone cannot guarantee preservation of nucleic‑acid‑binding sites, an evolution‑based residue‑fixing protocol was added to maximize sequence diversity while keeping essential functional residues.

Evolutionary Analysis and Leaf Splitting

Extensive alignment of TnpB, IscB and Cas families showed that only a handful of residues—those directly involved in RNA recognition and catalysis—are strictly conserved; surrounding regions vary widely in length and composition.

Consequently, TnpB was split into a DNA‑recognition REC leaf and a catalytic + RNA‑binding NUC leaf. Each leaf was independently redesigned, then recombined in pairwise tests. By tuning mutation‑threshold parameters, the researchers achieved 50‑60 % sequence divergence from the wild‑type while retaining functional compatibility.

Figure 2: Enrichment of AI‑generated TnpB variants after applying positional conservation and coupling‑strength thresholds.

High‑throughput Functional Screening

After evolutionary analysis, 1,980 REC‑NUC combinations were screened in high‑throughput functional assays. Of these, 24 % (466 variants) displayed detectable nuclease activity, and 8 % of the active variants outperformed the wild‑type ISDra2 TnpB. The dual‑constraint (conservation + coupling strength) approach yielded significantly more active variants than using a single conservation constraint.

Figure 3: Genome‑editing activity of AI‑generated TnpB variants in HEK293T cells.

Structural Validation

Cryo‑EM structures and predictive models were used to map functional regions. AI‑designed residues formed novel electrostatic and hydrogen‑bond networks at the RNA/DNA interface, remaining stable across conformations. New mutations complemented conserved core sites, collectively preserving nucleic‑acid binding.

The team identified a transient TAM‑binding intermediate conformation absent in wild‑type TnpB; this state was stabilized in the AI‑designed variants. Engineered helices retained the natural bending motion required for heteroduplex formation and nuclease activation, confirming that the inverse‑folding design preserved essential conformational dynamics.

Figure 4: Cryo‑EM structures of AI‑generated variants reveal new RNA‑DNA interface residues and conserved conformational motions.

Implications

By analyzing which structural elements are evolutionarily conserved and which can tolerate change, the researchers actively engineered novel RNA‑guided nucleases. The resulting non‑natural TnpB variants possess new nucleic‑acid contact points and high activity, expanding the design space beyond natural discovery and providing a route toward smaller, more precise, and more efficient genome‑editing tools.

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
结合蛋白质进化规律和三维结构信息，对 RNA 引导核酸酶进行系统性重构，最终设计出一系列体积更小、仍保持编辑能力的最小核酸酶。