Overview

Research Background and Core Question

Figure 1. De novo design workflow of miniature metallo ketoreductases.

Chiral alcohols are important building blocks in pharmaceuticals, fine chemicals, and functional molecules. Traditional asymmetric hydrogenation can produce these compounds efficiently, but it often relies on precious-metal catalysts, chiral ligands, and relatively complex reaction conditions. Biocatalysis provides a milder and more selective alternative. However, natural ketoreductases usually require NADPH and cofactor recycling systems, which can increase process complexity and cost.

Recent studies have shown that some metalloenzymes, such as human carbonic anhydrase II, can be repurposed for silane-mediated ketone reduction through a zinc hydride mechanism. Yet natural enzyme scaffolds were not evolved for this chemistry and often require extensive optimization.

The article therefore asks whether a small, stable, efficient, and stereoselective artificial ketoreductase can be directly designed from a theoretically defined metal active site, rather than adapted from an existing natural enzyme.

The central question of the article is whether researchers can move beyond natural enzyme scaffolds and directly design a small, stable, efficient, and stereoselective artificial ketoreductase from a theoretically defined metal active site.

What the Article Says Was Achieved

Figure 2. Computational design workflow of zinc-based metallo-ketoreductases and initial screening of catalytic reductions.

The authors built a de novo enzyme-design workflow for zinc hydride transfer. Instead of simply modifying an existing natural enzyme, they first defined the essential catalytic residues and metal-coordination environment. They then used RFDiffusionAA to generate new protein backbones and combined ProteinMPNN, LigandMPNN, AlphaFold2, RIFDock, Rosetta, and Metal3D to identify structures that could accommodate the substrate, stabilize zinc, and support hydride transfer.

The resulting family of artificial metallo-ketoreductases was named dMKR. The best initial design, dMKR50, reached 98% yield and 97% ee in asymmetric acetophenone reduction after tag optimization. Directed evolution then produced variants dMKR_V88A and dMKR_I92L, which expanded substrate scope and delivered high catalytic performance for the reduction of 3-acetyl-5-bromopyridine, with turnover numbers up to 19,000.

Core Innovations

Active-site-driven protein design

The study links a theoretical active site with modern deep-learning protein design. Instead of starting from a similar natural enzyme pocket and modifying it, the authors defined the catalytic requirements first, including zinc coordination, substrate carbonyl positioning, oxyanion stabilization, and hydride-transfer geometry. These features were then used as the design core, allowing the algorithms to identify a small protein scaffold capable of supporting the desired active site.

Implementation of a non-natural metal-hydride mechanism

The work demonstrates a non-natural reductive mechanism based on metal hydride chemistry. While natural reductive enzymes commonly rely on NADPH, this designed system uses phenylsilane as the terminal reductant. Isotope-labeling and silane-comparison experiments further support the involvement of a zinc hydride species during catalysis.

A compact enzyme with strong functional performance

The designed enzyme, dMKR, is only about 130 amino acids in length, with a molecular weight of approximately 13.8 kDa. This makes it much smaller than human carbonic anhydrase II, which is about 29 kDa. Despite its small size, dMKR shows strong stability, tolerance to organic solvents, and useful conversion across multiple substrates.

Fourth, the stereoselectivity of the designed enzymes appears partly predictable. Among 12 designs with yields above 5% and absolute ee values above 50%, 11 showed the same stereochemical outcome predicted by design, a success rate of about 91.7%. This suggests computational design may be able to produce enzymes that not only react but also predefine product configuration.

Key Experimental Results and Performance Data

Design and screening workflow

The authors built a large computational design pipeline for de novo metalloenzyme creation. They generated 8,000 protein backbones using RFDiffusionAA, selected 1,833 high-quality models, and used ProteinMPNN to produce 54,990 sequences. After AlphaFold2 prediction, docking, Rosetta optimization, and Metal3D metal-site analysis, 91 designs were selected, and 24 were experimentally tested.

Initial activity and lead identification

In the first screen, dMKR1, dMKR7, and dMKR13 showed clear activity in acetophenone reduction, each reaching more than 20% yield and 50% ee. Further screening of Family VII sequences identified stronger variants, especially dMKR49 and dMKR50. In whole-cell reactions, dMKR50 achieved 83% yield and 93% ee. After replacing the His tag with a Strep tag, performance improved to 98% yield and 97% ee, likely because the His tag interfered with zinc coordination.

Catalytic performance and stereoselectivity

Purified dMKR achieved 96% yield and 94% ee at 0.5 mol% enzyme loading. Another designed enzyme, dMKR53, produced the opposite enantiomer, S-1-phenylethanol, with 87% yield and 93% ee at 1 mol% loading. This suggests that the design workflow can generate enzymes with different stereochemical preferences.

High thermal stability

dMKR showed much higher thermal stability than human carbonic anhydrase II. Its melting temperature reached 93.8°C, compared with 58.8°C for hCAII. After incubation at 60°C for 300 minutes, dMKR still maintained about 96% yield and 94% ee. Even after treatment at 90°C for 60 minutes, it still gave 95% yield and 91% ee, while hCAII was rapidly inactivated under high-temperature conditions.

Organic-solvent tolerance

dMKR also showed strong tolerance to organic solvents. For a poorly water-soluble substrate, dMKR retained 63% yield and 97% ee in 30% 1,4-dioxane, whereas hCAII dropped sharply. In 30% DMF, dMKR gave 79% yield and 97% ee, and in 30% DMSO it reached more than 99% yield with 98% ee.

Substrate scope and regioselectivity

Under whole-cell conditions, dMKR reduced 16 ketone substrates, with yields up to more than 99% and ee values above 90% in all tested cases. The accepted substrates included aryl ketones, heteroaryl ketones, dialkyl ketones, and cyclic ketones. For 1-phenylbutane-1,3-dione, dMKR selectively reduced the internal carbonyl near the phenyl ring, while hCAII favored the terminal carbonyl. This shows that a de novo-designed pocket can create regioselectivity different from that of a natural enzyme.

Directed evolution and scale-up performance

Directed evolution further improved catalytic efficiency. dMKR_V88A increased the reduction of 2,2-dimethyloxan-4-one from 43% yield and 84% ee to 90% yield and 95% ee. dMKR_I92L reduced 3-acetyl-5-bromopyridine using only 0.0032 mol% enzyme, giving 61% yield, 94% ee, and a TON of 19,000. In a gram-scale whole-cell reaction, the same substrate was fully converted at room temperature within 5 hours, producing 4.85 g isolated product in 96% yield and 94% ee.

Kinetic validation

Kinetic analysis confirmed strong catalytic acceleration. For 3-acetyl-5-bromopyridine, dMKR_I92L showed a KM of 2.1 ± 0.2 mM, a kcat of 0.34 ± 0.01 s⁻¹, and a kcat/KM of 160 ± 20 M⁻¹ s⁻¹. Compared with the uncatalyzed reaction, the catalytic acceleration reached about 10¹⁰-fold.

Methods and Mechanistic Interpretation

The study began with a QM/MM model of zinc hydride-mediated ketone reduction in human carbonic anhydrase II. Instead of directly modifying the natural enzyme, the authors extracted the essential catalytic geometry and built new protein scaffolds around it.

The designed active site retained three key modules:

• Zinc coordination module: three histidines coordinate the Zn²⁺ center.
• Electrostatic activation module: glutamate helps stabilize the catalytic environment and activate zinc-bound water.
• Oxyanion-stabilization module: a threonine-associated region supports the carbonyl reduction intermediate.

Evidence for silane-derived hydride transfer

Mechanistic labeling experiments showed that the hydride comes from silane. When PhSiD₃ was used, deuterium incorporation at the chiral carbon was greater than 99%. In contrast, PhSiH₃ in deuterated buffer did not introduce deuterium at that position. This indicates that water or buffer is not the direct hydrogen source.

Support for a zinc-hydride pathway

Different silanes gave similar stereochemical outcomes when they could support reduction, while bulky silanes or weaker hydride donors failed to produce detectable alcohol. This supports a zinc-hydride-mediated pathway rather than direct asymmetric silane addition to the carbonyl.

Active-site validation by mutagenesis

Mutating His42, His44, or His61 strongly reduced or abolished activity, confirming the importance of the three zinc-binding histidines. Glu54 mutation reduced activity and changed product configuration from R to S, suggesting a role in substrate orientation and stereoselectivity. Thr84 mutation lowered yield but largely preserved ee, indicating that nearby backbone interactions may still help position the carbonyl group.

Metal dependence

Removing Zn²⁺ eliminated activity, and adding Zn²⁺ restored it. Native mass spectrometry showed that about 70% of dMKR was zinc-loaded, with a Zn²⁺ binding Kd of 0.87 μM. Other metals gave only partial activity, while Zn²⁺ performed best. These results confirm that dMKR is a zinc-dependent artificial metalloenzyme.

Why This Work Matters

1. This study shows a more programmable route for biocatalyst development: define the required reaction geometry, use AI-based protein design to build a suitable scaffold, and refine the catalyst through experimental screening and limited directed evolution.
2. For biomanufacturing, it suggests that enzyme engineering may move beyond reactions already found in nature. Non-natural metalloenzymes could help introduce useful organic synthesis reactions, such as metal catalysis, transfer hydrogenation, and asymmetric reduction, into biocatalytic systems.
3. The work also highlights the value of small and stable enzyme scaffolds. dMKR is compact, highly thermostable, expresses well, and can be partially purified by heat treatment, which may help reduce enzyme preparation costs and improve process robustness.

Limitations and Outlook

Although the study provides strong laboratory validation, it should not yet be viewed as proof of industrial feasibility. The current system uses phenylsilane as the terminal reductant, so cost, atom economy, byproduct handling, and scale-up still require evaluation. Catalytic efficiency and substrate scope also remain areas for improvement, especially when compared with mature industrial enzymes. Mechanistic evidence supports a zinc-hydride pathway, but key intermediates have not been directly captured. Future work will need to combine enzyme optimization with process engineering, including higher substrate loading, continuous reaction formats, product separation, reductant recycling, and cost modeling.

Key takeaways for enzyme-engineering clients

Overview of What CD ComputaBio Can Provide

AI-enabled enzyme design programs require coordination among protein structure modeling, active-site design, substrate positioning, molecular docking, molecular dynamics, mutation prioritization, and experimental validation. CD ComputaBio supports research-stage projects that need to connect computational enzyme design with practical wet-lab decisions. The following related capabilities were selected from the attached website outline because they directly match the technical decisions discussed on this page.

Reference:

1. Xu Y, Li Y, Zheng H, et al. De Novo Design of Miniature and Efficient Metallo-Ketoreductases. Journal of the American Chemical Society, 2026. https://doi.org/10.1021/jacs.6c00732

• Nature Highlights Baker Lab's New Breakthrough: Single-Protein Artificial Virus-Like Particles Reach 220 nm in Diameter
• David Baker Lab Explores the Past, Present, and Future of Protein Design
• Nature Portfolio Journal Highlights Folddisco, a 20× Faster and 4× More Storage-Efficient Tool for Large-Scale Protein Structural Motif Search
• Next-Generation Peptidomimetic Therapeutics: Backbone Modification, Molecular Engineering Strategies, and Biotechnological Advances