Hydrolytic Activity: From "Molecular Scissors" to the AI-Powered Precision Design Revolution in Enzymology

At the core of synthetic biology manufacturing lines—spanning food processing, plastic degradation, and pharmaceutical synthesis—lies a class of biocatalysts known as hydrolases, which are often metaphorically described as "molecular scissors." The human body relies on hydrolytic reactions to digest starch, proteins, and lipids; industrial green manufacturing leverages hydrolysis to degrade plastics and convert waste biomass; even intracellular DNA repair and substrate transport are fundamentally supported by hydrolytic chemistry. The quantitative metric that defines the catalytic performance of these "scissors" is hydrolytic activity.

For decades, engineering hydrolases has depended on random mutagenesis and site-directed mutagenesis, resulting in protracted R&D cycles and exceedingly low success rates. The advent of AI-powered protein design has fundamentally rewritten this paradigm—transitioning enzyme engineering from "serendipitous screening" to "on-demand precision design"—and has heralded a transformative revolution in enzymology.

Enzyme Hydrolysis Test Tube

Enzyme Hydrolysis Test Tube

I. What is Hydrolytic Activity?

Hydrolases are biocatalysts that specifically catalyze hydrolytic reactions, employing water molecules to cleave stable chemical bonds—including peptide bonds, ester bonds, and glycosidic bonds—thereby disassembling macromolecules into reusable small-molecule units. Under natural conditions, these chemical bonds are exceptionally resistant to cleavage; hydrolases drastically lower the activation energy of the reaction, enabling hydrolysis to proceed efficiently under mild conditions.

Hydrolytic activity quantitatively characterizes both the rate and efficiency of hydrolase-catalyzed reactions and serves as the primary competitive parameter for the industrial application of enzyme preparations. Higher hydrolytic activity translates directly into greater product yield per unit of input and reduced production costs.

The universally adopted unit for enzyme activity is the International Unit (IU) , defined as the amount of enzyme required to catalyze the formation of 1 μmol of product per minute under specified standard conditions. The IU is defined as the amount of enzyme that catalyzes the hydrolysis of 1 μmol of sucrose per minute under specified conditions (yielding reducing sugars, calculated as glucose). Protease activity is quantified based on the release of TCA-soluble peptide fragments, typically measured using the Folin-phenol method.

Comprehensive industrial coverage—hydrolytic activity determines production capacity.  Hydrolases are the foundational tools of green manufacturing, finding applications across a wide spectrum of industries: amylases hydrolyze starch to produce glucose syrups; proteases break down proteins to manufacture flavorings and bioactive peptides; lipases catalyze the conversion of oils and fats into fatty acids and biodiesel; PET depolymerases degrade plastic pollutants. Across all these scenarios, hydrolytic activity is the primary criterion for evaluating microbial strains and enzyme formulations, directly governing the production economics of fermentation-based cell factories.

II. Traditional Enzyme Design: A "Needle-in-a-Haystack" Effort Spanning Years

Naturally occurring hydrolases are adapted to the mild conditions of in vivo environments and are poorly suited to the harsh industrial conditions of high temperature, extreme pH, and elevated substrate concentrations. To obtain high-performance industrial enzymes, the industry has historically relied on two primary approaches—directed evolution and rational design—both of which are constrained by intractable efficiency bottlenecks.

Directed evolution: random screening on a massive scale. This approach employs artificial induction of mutagenesis to construct libraries of millions of variants, which are then individually screened for hydrolytic activity. Using the PET hydrolase IsPETase as an illustrative example, although early directed evolution efforts achieved modest improvements, the resulting variants still failed to meet the stability and catalytic efficiency required for industrial-scale production. The complete screening campaign typically extends over 3 to 5 years, with only a minute fraction of the vast mutational landscape yielding variants with genuinely enhanced properties.

Rational design: low-error-tolerance site-directed engineering. Researchers resolve the three-dimensional protein structure and rationally target mutations at active-site residues. However, the catalytic function of a hydrolase is orchestrated by tens of amino acids acting in concert, with the enzyme adopting complex conformational dynamics. A single-site mutation can readily induce cascading negative effects—such as loss of activity or compromised stability—resulting in extremely low success rates.

For a typical hydrolase comprising only 300 amino acids, the theoretical mutational space is astronomically vast. Traditional R&D, operating essentially by blind exploration, is fundamentally incompatible with the rapid iteration demands of modern biomanufacturing. This constitutes the core reason why the classical synthetic biology DBTL (Design–Build–Test–Learn) cycle has long been plagued by protracted timelines and suboptimal iteration efficiency.

DBTL Trial-and-Error VS LDBT AI Iteration Infographic

III. The AI Revolution: De Novo Design of Entirely New Hydrolases from Scratch

The year 2025 marked a milestone in AI-driven enzyme design. Science and Nature successively published landmark studies that, leveraging the iteratively advanced AI protein foundation models including PLACER, RFdiffusion, and RFdiffusion2, achieved zero-shot de novo design of hydrolases—creating catalytically active proteins that do not exist in nature without relying on any natural enzyme templates. This advancement upgraded the R&D paradigm from the traditional DBTL cycle to the AI-driven LDBT (Learn–Design–Build–Test) framework, wherein AI pre-learns from massive protein datasets and predicts structure-function relationships, dramatically compressing R&D costs and trial-and-error cycles.

February 2025, Science: AI de novo design of serine hydrolases. Serine hydrolases constitute one of the largest and most widely applied enzyme families in nature. However, traditional engineering approaches have struggled to simultaneously optimize active-site architecture and global scaffold stability, with artificial variants consistently underperforming compared to their natural counterparts.

This study employed the self-developed PLACER machine learning network—full name "Protein-Ligand Atomic Conformation Ensemble Reconstructor"—in conjunction with the RFdiffusion protein generation model to integrate analysis of protein backbone geometry, amino acid physicochemical properties, and substrate molecular structure, enabling accurate prediction of active-site atomic arrangements. Two major breakthroughs were achieved: first, the de novo construction of serine hydrolases with complete catalytic function, requiring only minimal active-site configurations for efficient ester hydrolysis; second, the generation of multiple catalytic proteins with topologies entirely distinct from any known natural serine hydrolases, demonstrating unprecedented structural novelty.

December 2025, Nature: RFdiffusion2 enables zero-shot design of metallohydrolases. The next-generation RFdiffusion2 replaced the conventional diffusion model with a flow matching framework, eliminating the requirement for pre-specified protein backbone parameters. By inputting only the target catalytic function, the AI autonomously matched amino acid stoichiometry, three-dimensional spatial arrangement, and catalytic core architecture.

Using a zinc-dependent metallohydrolase designed to degrade a fluorescent organic pollutant as the model system, and integrating quantum chemical calculations to precisely define the transition-state geometry of the hydrolytic reaction, the results were as follows: from only 96 AI-designed samples in the first round, the lead design ZETA_1 achieved a catalytic efficiency (*k*cat/KM) of 16,000 M⁻¹s⁻¹, representing an improvement of three orders of magnitude over previous artificially designed metallohydrolases. In the second round, three high-activity enzyme variants were identified, with ZETA_2 reaching a catalytic efficiency exceeding 53,000 M⁻¹s⁻¹. The complete workflow required no subsequent experimental optimization, and the catalytic performance approached that of certain natural metallohydrolases.

These two generations of results clearly delineate the AI iteration trajectory: PLACER + RFdiffusion enabled structural innovation, while RFdiffusion2 further simplified the design process and substantially enhanced catalytic activity. When coupled with automated experimental platforms, the multi-year R&D cycle can be compressed to just a few months.

IV. Industrial Translation of AI-Designed Hydrolases

Programmable Catalytic Enzyme Network

The maturation of AI design technology has transformed hydrolases from "natural endowments" into "programmable, customizable tools," with applications spanning solid waste remediation, environmental restoration, green chemistry, and food manufacturing.

1. Plastic degradation: optimizing hydrolytic activity to combat white pollution. A 2026 study published in Nature Communications confirmed that hydrolytic activity is determined not only by the catalytic center but also by distal structural elements. A surface-exposed loop structure designated SEC-loop, located approximately 25 Å from the active site, was found to physically restrict enzyme–PET substrate interactions; loop replacement substantially enhanced depolymerization efficiency. This mechanistic insight has been retroactively applied to AI model training, enabling simultaneous optimization of both the catalytic core and substrate-binding peripheral regions during protein generation, thereby systematically improving plastic-degrading enzyme performance from the design stage.

2. Broad-spectrum environmental remediation: customized metallohydrolases. Metallohydrolases, which activate water molecules via metal ions, can catalyze challenging hydrolytic reactions. Leveraging RFdiffusion2, it is now possible to design tailored hydrolases with defined substrate specificity for applications including pesticide residue degradation, industrial wastewater purification, and chemical toxin detoxification, representing a core green technology aligned with carbon neutrality objectives.

3. AI-driven design plus automated laboratory integration. Upon completion of protein sequence design by AI, the outputs are directly interfaced with automated platforms to execute strain construction, protein purification, and high-throughput hydrolytic activity screening. This integrated "dry–wet" closed-loop system fully aligns with the LDBT paradigm, delivering substantial reductions in development costs.

V. Conclusion

The evolution of hydrolase R&D has progressed through three transformative phases: natural enzyme discovery → directed evolution and rational engineering → AI-driven de novo creation of entirely new proteins.

Previously, enhancing hydrolytic activity required exhaustive wet-lab experimentation; today, with only a few hundred AI-designed samples, it is possible to generate novel enzymes with thousand-fold improvements in catalytic efficiency—even creating catalytic protein topologies that have never been observed in nature—achieving performance comparable to natural enzymes without subsequent optimization.

From PET hydrolases to amylases and proteases, the hydrolytic activity of industrial enzymes is no longer a natural endowment but a core attribute that can be precisely tuned and  . AI-powered protein design is the master key to unlocking high-performance enzyme customization, continuing to drive the replacement of high-pollution, energy-intensive traditional chemical processes with green biomanufacturing, and fortifying the technological foundation for carbon neutrality, food security, and ecological stewardship.