Enzyme Miniaturization: Revolutionizing Future Biocatalysts

Abstract

Enzyme miniaturization offers a transformative approach to overcome limitations posed by the large size of conventional enzymes in industrial, therapeutic, and diagnostic applications. However, the evolutionary optimization of enzymes for activity and stability has not inherently favored compact structures, creating challenges for modern applications requiring smaller and more efficient catalysts. In this review, we surveyed the advantages of miniature enzymes, including enhanced expressivity, folding efficiency, thermostability, and resistance to proteolysis. We described the applications of miniature enzymes as biosensors, therapeutic agents, and industrial catalysts. We highlighted strategies such as genome mining, rational design, random deletion, and de novo design for achieving enzyme miniaturization, integrating both computational and experimental techniques. By investigating these approaches, we aim to provide a framework for advancing enzyme engineering, emphasizing the unique potential of smaller enzymes to revolutionize biocatalysis, gene therapy, and biosensing technologies.

1. Introduction

Enzymes are biological catalysts that feature an active site supported by a relatively large protein scaffold (Park et al., 2006). While the active site is essential for catalytic function, the scaffold may contain structurally redundant elements (Shams et al., 2021; Zhao et al., 2023). This redundancy, though vital for adaptation to environmental changes during evolution (Björklund et al., 2006; Savino et al., 2022; Tóth-Petróczy and Tawfik, 2013), may contribute minimally to catalytic functions under the conditions typical of enzymatic applications in therapeutics and industry. Moreover, these redundant structural elements often lead to undesirable effects on enzyme function and application. For instance, they consume more cellular resources for protein expression, and potentially increase the misfolding propensity and reduce the lifetime and thermal stability of proteins (Dyson et al., 2004; Goh et al., 2004). In certain application scenarios, the size of the protein can be a critical factor. In therapeutics, large enzymes (e.g. Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) and CRISPR-associated proteins (Cas) systems) face limitations in effective delivery to target tissues and cells (Shams et al., 2021; Zhao et al., 2023). In diagnostics, large oxidoreductases used in electrochemical biosensors suffer from reduced interfacial electron transfer rates due to increased cofactor-electrode distances, compromising detection sensitivity (Kaida et al., 2019). Therefore, there is a growing demand for mining or engineering “miniature” enzymes that are smaller in size but maintain equivalent functionality.

Learning from nature, we can draw inspiration from evolutionary strategies that favor protein miniaturization, optimizing efficiency and functionality while reducing resource demands. Evolutionary studies of bacteria, archaea, and eukaryotes reveal a tendency to produce smaller proteins when specific functional constraints are absent, as demonstrated by the higher frequency of deletions compared to insertions in protein-coding genomes (Lipman et al., 2002). For instance, deletions, which can drive random genetic drift and selection pressure to produce proteins with a shorter sequence, occur eight times more frequently than insertions in Escherichia coli and approximately three times more often in humans (Lipman et al., 2002). This evolutionary trend toward smaller proteins likely reflects an optimization of resource utilization in organisms with limited energy or coding capacity. Presumably, smaller enzymes not only reduce the metabolic costs of synthesis and folding but also potentially enhance their resistance to heat and proteolytic degradation.

By leveraging nature’s evolutionary principles, we can explore strategies like genome mining to uncover miniature enzymes, identifying smaller homologs that possess the same function with their larger counterparts (Cuesta et al., 2015). For example, mining metagenomic databases leads to the discovery of clustered regularly interspaced short palindromic repeat (CRISPR)-associated nuclease Cas14, which is one-third the size of Cas9 (Harrington et al., 2018). This discovery holds significant promise for gene therapy by addressing viral vector payload constraints (Xu et al., 2021). Despite these successes, genome mining may fall short in identifying desired smaller homologs, perhaps due to a limited number of sequences for multiple sequence alignment, the target enzyme families composed solely of large proteins, or smaller homologs lacking the desired functionality (Arnold et al., 2001). This may be because evolution has not prioritized generating enzymes with compact structures unless driven by some particular harsh environmental pressures. Given the limitations of natural protein libraries, innovative strategies are needed to downsize enzymes while preserving their essential functions, which we tentatively named as “enzyme miniaturization”. Miniature enzymes have currently been created through (semi-)rational design (Ferreira et al., 2022), insertions and deletions (InDels) (Savino et al., 2022), and computationally-guided de novo enzyme design (Zanghellini, 2014). These practices have sparked considerable interest in enzyme miniaturization and have the potential to drive transformative advancements in biomanufacturing, biosensing, and therapeutic applications. Unlike these conventional enzyme engineering approaches that primarily focus on enhancing catalytic properties such as activity, stability, or selectivity, enzyme miniaturization represents a distinct strategy that aims to modify the physical properties of enzymes through substantial size reduction while preserving function, thereby addressing size-related limitations in various applications. This review aims to survey the unique advantages of miniature enzymes over their larger counterparts, and to describe existing strategies for enzyme miniaturization. This review provides insights into the opportunities and challenges for future research in enzyme miniaturization, inspiring development of new approaches for the engineering and application of next-generation enzymes.

2. Potential Intrinsic Advantages of Miniature Enzymes

This section discusses the question: what functional and synthetic benefits do small enzymes offer when engineered for contemporary industrial and therapeutic needs, particularly discussing their enhanced expression and folding profiles, thermostability, and resistance to degradation (Figure 1).

Figure 1.

Advantages of small enzymes over large enzymes in expression and folding profiles, thermostability, and resistance to degradation.

2.1. Expression and Folding Profiles

Enzymes are synthesized on ribosomes by translation from messenger RNA, after which they are released into the cellular environment and fold into a three-dimensional conformation (Kaiser and Liu, 2018). The sequence of an arbitrary enzyme must be long enough to form the three-dimensional folds essential for its function, but overly long sequences can result in higher energetic costs of synthesis (Nevers et al., 2023), slower folding (Galzitskaya et al., 2003), and increased misfolding risks (Gershenson et al., 2020).

Protein folding rates involve a broad range, spanning from microseconds to hours, with sequence length being a dominant factor influencing the folding rates (Ivankov and Finkelstein, 2004). Folding time has been demonstrated to scale exponentially with sequence length (L), though theoretical models propose varying scaling exponents ranging from L^1/2 to L^2/3 (Finkel’shteĭn and Badretdinov, 1997; Galzitskaya et al., 2003; Gutin et al., 1996; Ivankov et al., 2003; Thirumalai, 1995). A coarse-grained lattice model suggests that proteins comprising fewer than 200 amino acid residues can fold within a biologically feasible timeframe (Lin and Zewail, 2012). Proteins exceeding this length likely require additional mechanisms to expedite folding, such as co-expression with chaperones (Lin and Zewail, 2012). Interestingly, this aligns with the average lengths of natural proteins in bacteria and archaea, typically consisting of 270 and 242 residues, respectively (Nevers et al., 2023).

Smaller enzymes not only fold faster but are also more likely to adopt proper and soluble conformations essential for their catalytic functions. In contrast, larger proteins are more prone to misfolding due to a greater likelihood of becoming trapped in transient metastable states corresponding to local free energy minima (Gershenson et al., 2020). A statistical analysis of 27,267 proteins demonstrates that an increase in protein length leads to a lower probability of successful expression in a soluble form, particularly for proteins exceeding 400 residues (Goh et al., 2004). Similarly, a survey of protein features correlated with soluble expression of 95 mammalian proteins in E. coli reveals size as a significant factor: 1) Proteins of 16 to 25 kDa (average 22.8 kDa) can be expressed in a soluble form; 2) Proteins of 22.5 to 54.5 kDa (average 40.4 kDa) require fusion with solubility-enhancing tags (e.g., thioredoxin, maltose binding protein) for expression; 3) Proteins of 40.7 to 81.6 kDa (average 51.4 kDa) exhibit low soluble expression (below 1 mg/L), even when fused with solubility-enhancing tags (Dyson et al., 2004). These studies underscore the advantages of smaller enzymes, which not only fold more efficiently but are also more likely to adopt functional, soluble conformations, highlighting their potential as ideal candidates for biomanufacturing.

2.2. Thermostability

Enzyme thermostability, which refers to the ability of an enzyme to maintain its activity at elevated temperatures over extended periods (Bhalla et al., 2013), is a key property for industrial application (Haki and Rakshit, 2003). One indicator of thermostability is melting temperature (T_m), defined as the temperature at which half of the protein population is in unfolded or denatured (McGuinness et al., 2018). Although T_m itself does not correlate with protein size (Yang et al., 2019), the magnitude of the thermal shift (ΔT_m) upon substrate binding shows an inverse relationship with protein size (Watson et al., 2017). Smaller enzymes generally exhibit a more substantial ΔT_m upon binding, potentially due to their increased sensitivity to molecular interactions. Since enzymatic reactions begin with substrate binding, we hypothesize that smaller enzymes experience greater substrate-induced stabilization compared to larger enzymes. This distinction underscores the subtle but significant role of protein size in modulating enzyme stability in response to substrate binding (Watson et al., 2017).

In addition, large enzymes often contain more flexible loops and disordered terminal regions, which can compromise their thermostability. In contrast, smaller enzymes, particularly those engineered to remove these flexible regions, demonstrate enhanced thermostability due to reduced structural flexibility and increased compactness (Nezhad et al., 2022; Xu et al., 2020). For example, thermostable lipases can be constructed by truncating flexible loops and replacing residues with rigidifying ones such as proline (Xu et al., 2020). These changes stabilize the protein’s active site and improve its ability to resist high temperatures, leading to substantial increases in thermostability (Xu et al., 2020). This effect is particularly evident in small enzymes, where loop removal or rigid substitution can significantly reduce entropy in the unfolded state, making them more resilient to thermal stress.

2.3. Resistance to Degradation

Some enzymes are susceptible to proteolytic degradation, which not only compromises drug efficacy but also poses safety risks through potential adverse effects in patients (Jiskoot et al., 2012). Larger proteins exhibit faster degradation rates both in vivo and in vitro (Dice et al., 1973; Fornasiero et al., 2018), showing a size-dependent susceptibility to proteolysis. For example, a positive correlation between degradation rate and protein size was observed within the proteins in mouse brains (Fornasiero et al., 2018). Among ~3500 proteins in the brains, those degraded within 2 days comprise an average of ~570 amino acid residues, while proteins degraded within 30 days average ~260 residues (Fornasiero et al., 2018). This is likely attributed to the fact that larger proteins expose a greater surface area to the cytosol, thereby increasing the likelihood of proteolysis (Fornasiero et al., 2018). These findings highlight the inherent vulnerability of larger proteins to proteolytic degradation, emphasizing the importance of reducing protein size in designing stable enzymes for therapeutic and industrial applications.

3. Advantages of Miniature Enzymes in Biocatalytic and Biomedical Applications

Enzymes are essential in numerous industrial sectors, from fine and pharmaceutical chemistry to food and chemical production (Kirk et al., 2002). Beyond these traditional roles, enzymes are also pivotal in emerging fields like metabolic therapies (Hennigan and Lynch, 2022), genome editing (Zhang, 2019), biosensing (Kurbanoglu et al., 2020), and bioimaging (Roda et al., 2009). In the following sections, we will outline the potential advantages of smaller enzymes across a range of applications, including industrial processes, metabolic therapies, genome editing, and biosensing (Figure 2).

Figure 2.

Advantages of small enzymes over large enzymes in industrial processes, metabolic therapies, genome editing, and biosensing.

3.1. Industrial Processes

Small enzymes offer significant economic advantages in both production and application. Compared to large enzymes, their shorter sequences take less production costs (Kim et al., 2023), making them more cost-effective to manufacture at scale. Moreover, small enzymes typically feature a higher density of functional sites (Kim et al., 2023). This suggests that, for a given mass, smaller enzymes theoretically offer greater catalytic units compared to their larger counterparts. As a result, a less amount of enzyme mass is needed to achieve equivalent catalytic performance, potentially enhancing the efficiency and cost-effectiveness of industrial processes. In addition, the intrinsic expression advantages of miniature enzymes (detailed in Section 2.1) translate directly to industrial-scale production benefits. Their improved solubility (Dyson et al., 2004; Goh et al., 2004), reduced cellular energetic cost (Nevers et al., 2023), increased folding rate (Galzitskaya et al., 2003) and decreased propensity for misfolding (Gershenson et al., 2020) collectively enable higher yields and lower resource consumption during large-scale fermentation. While large, complex enzymes often achieve lower expression yields in cell culture, requiring greater volumes of cell culture and more extensive purification processing despite smaller final protein quantities (Puetz and Wurm, 2019), creating smaller-size enzymes offers a solution to these limitations (Li et al., 2024; Scott et al., 2001). These favorable characteristics substantially improve both the economics and sustainability of industrial enzyme manufacturing.

Enzyme immobilization, a technique that anchors enzymes to solid carriers, enhances their stability and enables multiple reuse cycles in industrial processes (Bernal et al., 2018; Boudrant et al., 2020). The efficiency of immobilization is strongly affected by protein size, as larger proteins face greater physical constraints when entering and diffusing into mesopores (Misson et al., 2015). This limitation is particularly evident when immobilizing enzymes in porous materials, such as metal-organic frameworks (MOFs), which contain relatively small pores on the order of a few nanometers (Bayne et al., 2013; Feng et al., 2021; Hu et al., 2021; Liang et al., 2021). Although some mesoporous materials, such as IRMOF-74-IX (Deng et al., 2012) and PCN-333 (Feng et al., 2015), feature pore diameters exceeding 5 nm that may be sufficient to accommodate certain enzymes, most MOFs have pore sizes under 2 nm, limiting their applicability to only those proteins smaller than the pores (Liang et al., 2021; Park et al., 2007). Additionally, the stringent size constraints during enzyme infiltration into MOFs can lead to protein unfolding, potentially causing partial loss of enzymatic activity in the immobilized biocomposites (Chen et al., 2012). Alternatively, a biocomposite production strategy based on encapsulation, where enzymes are incorporated during MOF synthesis, has demonstrated the ability to integrate enzymes larger than the MOF pore size; however, the tight confinement can restrict access to substrates or cofactors and impair the enzyme’s catalytic activity (Liang et al., 2021). In contrast, smaller enzymes, which can more easily diffuse into mesoporous carriers and maintain their native conformations in nanoconfinements, offer distinct advantages for enzyme immobilization and the associated biotransformations (Zadeh et al., 2015).

3.2. Metabolic Therapies

Enzymes hold significant promise for therapeutics due to their high activity and selectivity (Hennigan and Lynch, 2022; Leader et al., 2008; Porello and Cellesi, 2023). To date, over 120 therapeutic enzymes have been employed in clinical settings, with 21 specifically associated with cancer treatment (Hennigan and Lynch, 2022). The global therapeutic enzyme market size was estimated at USD 7.7 billion in 2023 and is forecast to reach USD 16.6 billion by 2030 (Industry Expert Research, March 2024). However, therapeutic enzymes face challenges such as immunogenicity, rapid clearance, and inability to target specific tissues or cells when used in human body (Hennigan and Lynch, 2022). These challenges can be mitigated by encapsulating enzymes or messenger RNAs within nanocarriers that serve as delivery vehicles (Dean et al., 2017; Zangi et al., 2023). The delivery efficiency can be enhanced by increasing the nanocarrier’s loading capacity, which is defined as the amount of enzyme cargo per unit weight of the carrier (Li et al., 2022). One of the most crucial factors affecting loading capacity is the relative size between enzyme cargo and particles (Misson et al., 2015; Zangi et al., 2023). Smaller enzyme cargoes can better accommodate within the cavity of nanocarriers, thus having a higher chance of increasing the loading capacity (Misson et al., 2015). For example, reducing mRNA size from 1,929 to 996 nucleotides decreases the proportion of empty lipid nanocarriers by over 10%. This enhanced loading capacity ultimately improves transgene expression efficiency for therapies (Li et al., 2022).

An alternative delivery method involves conjugating therapeutic enzymes with cell-penetrating peptides (CPPs) to facilitate translocation across cell membranes (Dinca et al., 2016). CPPs can transport proteins ranging in size from 30 to 150 kDa through either direct penetration or endocytosis (Chugh et al., 2010). The internalization pathways is largely affected by the size of the enzyme cargo (Tripathi et al., 2018). Small enzymes are more likely to utilize energy-independent direct penetration pathways, whereas large enzymes typically rely on energy-dependent endocytosis (Pujals et al., 2006). Due to reduced energy requirements for internalization, smaller enzymes can achieve higher cellular uptake efficiency (Tripathi et al., 2018).

For anticancer enzymes (e.g. T4 endonuclease V (Hacker et al., 2010)) that need to enter the cell nucleus to affect DNA, the size must be small enough to passively diffuse through the nuclear pore complex, which serves as the gateway to the nucleus (Winogradoff et al., 2022). Proteins exceeding a threshold of 40–60 kDa face significantly reduced nuclear pore permeation rates (Timney et al., 2016; Winogradoff et al., 2022).

3.3. Genome Editing

Of the estimated 25,000 annotated genes in the human genome, mutations in over 4,500 have been associated with disease phenotypes (www.omim.org/statistics/geneMap), with more disease-associated genetic variations being rapidly uncovered (Cox et al., 2015). Genome editing technologies based on programmable nucleases such as meganucleases (Stoddard, 2011), zinc finger nucleases (Urnov et al., 2010), transcription activator--like effector nucleases (Bogdanove and Voytas, 2011) and the CRISPR-associated nucleases (Hsu et al., 2014) are opening up the possibility for therapeutic genome editing in diseased cells and tissues (Hsu et al., 2014). Among these nucleases, CRISPR-associated nucleases, such as Cas9, have become the most widely used genome editing tools due to their advantages of simple design, low cost, and high efficiency (Xu and Li, 2020).

Nearly 70% gene therapy trials utilize viral vectors such as adeno-associated virus (AAV), adenovirus and lentivirus (Ghosh et al., 2020). Among these viral vectors, AAVs have predominantly been used for in vivo delivery of CRISPR-Cas effectors due to its high transduction efficiency, low immunogenicity, and diverse tropisms for delivering to various tissues with high specificity (Huang et al., 2022; Taha et al., 2022; Wang et al., 2019). However, AAV vectors have a limited payload capacity (~4.7 kb), presenting challenges for effectively delivering certain CRISPR-Cas effectors, such as Streptococcus pyogenes Cas9 (SpCas9, 4.1 kb, 1,368 residues), and guide RNA (Doudna, 2020; Zhang, 2019). Smaller Cas9 orthologs, such as Staphylococcus aureus Cas9 (SaCas9, 3.2 kb, 1053 residues) (Ran et al., 2015), Campylobacter jejuni Cas9 (CjCas9, 3.0 kb, 984 residues) (Kim et al., 2017), and Neisseria meningitidis Cas9 (NmCas9, 3.2 kb, 1082 residues) (Ibraheim et al., 2018) are better suited for AAV delivery. However, when these Cas9 orthologs are fused with additional elements, such as transcriptional activators (e.g. VP64-P65AD-Rta, ~1.6 kb) (Chavez et al., 2016), transcriptional repressors (e.g. Krüppel associated box domains and DNA methyltransferase, ~2.3 kb) (Nuñez et al., 2021), or other gene editing domains (e.g. reverse transcriptase, ~ 2.1 kb) (Anzalone et al., 2019), the resulting fusion proteins exceed the payload capacity of AAV vectors (Xu et al., 2021).

To overcome these delivery challenges, researchers have developed miniature CRISPR-Cas effectors through various methods: genome mining (e.g., Cas13an, 429 residues (Yoon et al., 2024); Cas14a, 530 residues (Harrington et al., 2018); Cas13d, 930 residues (Yan et al., 2018)), rational design (e.g., CasMINI, 529 residues (Xu et al., 2021); mini-RfxCas13d, 682 residues (Zhao et al., 2023); mini-SaCas9, 986 residues (Ma et al., 2018);), and random deletion (e.g., Δ4CE, 874 residues (Shams et al., 2021)), which have advanced genome-editing therapeutics. For example, the small size of Acidibacillus sulfuroxidans Cas12f1 (AsCas12f1, 422 residues) allows both the guide RNA and the Cas protein, along with necessary control elements (e.g., hU6 and CMV promoters) and tracking markers (e.g., EGFP), to fit within a single AAV delivery vehicle with a cargo size of 3.9 kb (Wu et al., 2021). This system successfully edited two target genes in human cells with up to 11.5% efficiency: the vascular endothelial growth factor A (VEGFA) gene and the programmed cell death 1 (PDCD1) gene for cancer treatment (Wu et al., 2021). This compact size of AsCas12f1 enables efficient delivery in a single viral carrier, overcoming the limitation of larger gene editing systems that require multiple delivery vehicles (Wu et al., 2021).

3.4. Biosensing

Enzymes are frequently employed as key components in the construction of biosensors for detecting target analytes in complex biological system, owing to their exceptional catalytic functions and high substrate specificity (Zhu et al., 2019). Electrochemical and optical biosensors are two main classes of enzyme-based biosensors (Zhu et al., 2019), with typical members being oxidoreductases (such as dehydrogenases) and bioluminescence enzymes (such as luciferases), respectively (Zhu et al., 2019). This section will use dehydrogenases and luciferases as examples to demonstrate how smaller enzymes can improve biosensing performance relative to their larger counterpart.

Dehydrogenases, such as glucose dehydrogenases and fructose dehydrogenases (FDH), have been immobilized on electrodes to detect specific analytes (e.g., glucose and fructose) in blood samples at ppm levels (Chen et al., 2020; Loew et al., 2017). The detection relies on an electrical signal generated by dehydrogenases that facilitate the transfer of electrons from the analyte to the electrode, using a cofactor as an electron carrier (Chen et al., 2020). However, large oxidoreductases often involve increased steric hindrance when combining with electrodes or other target proteins. This steric hindrance can compromise electron transfer or protein-protein interactions, leading to lower detection efficiency and accuracy (Adachi et al., 2020; England et al., 2016). For example, larger dehydrogenases have been observed to have lower detection efficiency as their increased size lengthens the cofactor–electrode distance, resulting in lower rate of interfacial electron transfer (Adachi et al., 2020; Kano et al., 2021; Smutok et al., 2022; Zhu et al., 2022). Previous studies indicated that the electron transfer rates decrease by 10⁴ when electron donor-acceptor distance increases from 8 to 17 Å (Lambrianou et al., 2008; Mayo et al., 1986). To address this, miniature dehydrogenases have been discovered and engineered (Hibino et al., 2017; Kaida et al., 2019; Zafar et al., 2012). One example is FDH (51 kDa) whose catalytic center of Subunit II involves three heme c moieties, namely heme 1c, 2c, and 3c. Heme 3c is situated closest to the flavin adenine dinucleotide, a critical cofactor that initiates electron transfer to the electrode (Smutok et al., 2022). It was hypothesized that bringing heme 3c closer to the electrode can enhance electron transfer efficiency, which led to the rational design of two miniature FDH variants: one with truncated heme 1c (Δ1c, 36 kDa), and the other with truncated heme 1c and 2c (Δ1c2c, 20 kDa) (Figure 2). Δ1c demonstrated a half-wave potential that was 0.05 V more negative than the native FDH (Hibino et al., 2017), whereas Δ1c2c exhibited a 0.16 V more negative potential than Δ1c (Kaida et al., 2019). Both Δ1c and Δ1c2c demonstrated an enhanced electron transfer rate, with increases of 1.3-fold and 4.1-fold, respectively (Hibino et al., 2017; Kaida et al., 2019). These enhancements suggest a design principle for developing faster electrochemical biosensors by utilizing smaller dehydrogenases (Kaida et al., 2019).

Luciferases trigger light emission through catalyzing oxidation of luciferin (Roda et al., 2009), thereby illuminating the cellular expression and functions of their fused target proteins for in vivo optical imaging (England et al., 2016). The advantage of smaller luciferases, such as NLuc, is highlighted by their enhanced stability and improved luminescence efficiency (England et al., 2016). Small luciferases like LuxSit-i (Yeh et al., 2023) are beneficial for fusion with target proteins, while larger luciferases could potentially disrupt the native functions of these proteins by physically obstructing their active sites or binding regions, thus impairing their biological activities ¹⁰⁷. Moreover, the larger counterparts may exhibit reduced bioluminescence intensity due to the increased steric hindrance between the fluorescent chromophore and the enzyme (Bhuckory et al., 2019; Dixon et al., 2016; Ohmuro-Matsuyama et al., 2022). Miniature luciferases, whether naturally discovered (Inouye et al., 2000; Markova et al., 2015; Tannous et al., 2005), experimentally engineered (Auld et al., 2018; Hall et al., 2012; Markova et al., 2012; Ohmuro-Matsuyama et al., 2022; Takenaka et al., 2008), or artificially designed (Kim et al., 2013; Yeh et al., 2023), are gaining popularity for their potential to address both of these challenges. In particular, NanoLuc luciferase (19 kDa), truncated from Oplophorus gracilirostris luciferase, demonstrates over 150-fold luminescence enhancement compared to the large-size luciferases firefly luciferase (61 kDa) and Renilla luciferase (RLuc, 36 kDa) (Hall et al., 2012). Miniature luciferases represent a promising opportunity for their integration into Bioluminescence Resonance Energy Transfer (BRET), a method widely employed for the quantitative analysis of protein-protein interactions in live cells (Dale et al., 2019; Mo et al., 2016; Stoddart et al., 2015). The 19-kDa NanoLuc luciferase, as a BRET donor, exhibits over 10-fold higher luminescence compared to the 36-kDa RLuc. This superior performance in BRET assays can be attributed to NanoLuc’s reduced steric hindrance with target proteins and its inherently enhanced luminescence profile (Mo et al., 2016).

4. Strategies for Enzyme Miniaturization

Given the superior intrinsic properties and applications of miniature enzymes, developing experimental and computational approaches to design small enzymes emerges as a crucial objective. While no approaches have been specifically tailored for enzyme miniaturization, several existing enzyme engineering strategies have yielded successful examples of miniaturized enzymes. These strategies include genome mining, rational design, random deletion, and de novo design. In this section, we will introduce these four strategies, presenting their methodologies through examples of miniaturized enzymes (Figure 3), and evaluating their respective strengths and limitations in the context of enzyme miniaturization (Table 1).

Figure 3.

Schematic overview of enzyme miniaturization strategies: genome mining, rational design, random deletion, and de novo design.

Table 1.

Summary of strategies for enzyme miniaturization.

Strategies	Features	Advantages	Limitations	Miniaturized enzyme examples
Genome mining	Screens metagenomic databases for miniature homologs using bioinformatics tools. Involves computational approaches including multiple sequence alignment, structure prediction, and structural alignment. Involves experimental approaches such as recombinant DNA techniques.	High success rate in finding functional variants due to natural evolution. High generalizability across enzyme classes.	Limited when enzyme families contain few members. Limited when families consist only of large proteins. May overlook enzymes with low sequence conservation.	SaCas9 (23% smaller than SpCas9) (Ran et al., 2015). Cas13an (60% smaller than typical Cas13) (Yoon et al., 2024). Glucuronoyl esterase (39 kDa, smaller than previously characterized variants of 42–72 kDa) (Dilokpimol et al., 2018)
Rational design	Focuses on structure-function analysis to identify and truncate redundant regions. Involves computational approaches including structural analysis and molecular modeling to guide the identification of critical functional sites and the assessment of redundant regions for potential removal. Involves experimental validation to confirm that truncations preserve activity.	Enables targeted reduction of enzyme size. Maintains critical activities. Effective for diverse enzyme classes with available structural data.	Requires understanding of structure-function relationships. Limiting application when structural and functional data is sparse. Demands resource-intensive computational modeling and experimental validation. May inadvertently impact enzyme stability.	Miniature Cas13 (retained ~70% of the wild-type enzyme’s length) (Zhao et al., 2023). Metridia luciferase (33% reduction in length) (Markova et al., 2012) picALuc (21% of the mass of firefly luciferase and 62% of the mass of the original artificial luciferase ALuc) (Ohmuro-Matsuyama et al., 2022).
Random deletion	Screens enzyme variants by creating a library of deletion mutations without requiring structure-function insights Involves gene fragmentation via nuclease digestion or polymerase chain reaction, recombinant assembly, and enzymatic assays to identify functional variants.	Does not require prior structure-function knowledge. Suitable for enzymes with limited structural data. Increases the likelihood of finding functional miniaturized forms by generating a broad library of variants. Allows exploration of deletions across a large portion of the enzyme sequence.	Screening large libraries can be inefficient. Requires high-throughput experimental function screening. May remove essential regions. Difficult to predict or control specific functional outcomes.	Miniaturized ligase (20% reduction in length) (Morelli et al., 2017). Miniaturized Cas9 (~37% reduction in length) (Shams et al., 2021).
De novo design	Creates novel enzymes with theoretical active sites (theozymes) and refines protein scaffolds to accommodate them. Utilizes advanced computational techniques, including deep learning, to design and optimize non-native protein scaffolds and active sites. Involves validating designed enzymes through experimental assays.	Enables the creation of enzymes with unique chemical functions and scaffolds not found in nature. Allows precise engineering of active sites and scaffolds to optimize desired activities and stability. Capable of designing enzymes for diverse reactions.	Requires extensive computational resources and expertise to create viable designs. Initial designs may face challenges in expression and activity. Relies heavily on existing structural and functional data. Designing novel scaffolds can lead to unpredictable outcomes.	Artificial luciferase LuxSit-i (117 residues and 13.9 kDa, smaller than any previously described luciferase) (Yeh et al., 2023). PET hydrolase (30% smaller than the industrial PET degrading enzyme LCC) (Ding et al., 2025).

4.1. Genome Mining

Metagenomes represent an enormous pool of unexplored enzymes from an estimated 10⁶−10⁸ prokaryote species that inhabit various ecosystems on earth (Cowan et al., 2005). The vast diversity of metagenomes increases the likelihood of finding miniature enzyme homologs from different species. Genome mining, which integrates predictive bioinformatic tools and recombinant DNA techniques, has turned out to be a promising strategy for identifying potential miniature enzymes through screening metagenomic databases (Harrington et al., 2018; Ran et al., 2015; Yan et al., 2018). Using Cas enzymes as an example, the search for miniature homologs typically involves three steps: (1) locating CRISPR-Cas loci with highly conserved sequence elements (e.g., cas1 gene or distinct CRISPR array) as benchmarks, (2) identifying sequences of putative Cas enzymes using probabilistic models (e.g., profile hidden Markov models) to detect remote homologs, and (3) filtering identified Cas clusters based on protein size (Chylinski et al., 2014; Shmakov et al., 2015). Employing this strategy, researchers have identified several miniature Cas enzymes, including SaCas9 (1,053 residues) (Ran et al., 2015), Cas13d (930 residues) (Yan et al., 2018), and Cas14a (530 residues) (Harrington et al., 2018), all of which are smaller than the typical SpCas9 (1,368 residues).

For enzymes with low sequence conservation, structural comparisons offer a solution, as structural cores evolve three to ten times slower than sequences (Illergård et al., 2009). The advent of atomic-accuracy prediction programs, such as AlphaFold (Jumper et al., 2021), RoseTTAFold (Baek et al., 2021) and ESMFold (Lin et al., 2023), has greatly increased the availability of protein structures. Structural alignment programs, such as DALI (Holm, 2019), TMalign (Zhang and Skolnick, 2005) and Foldseek (Van Kempen et al., 2024), can be employed to identify protein clusters with conserved structural elements. An example of this approach is the discovery of Cas13an (Yoon et al., 2024). By leveraging a Foldseek-clustered AlphaFold database (Barrio-Hernandez et al., 2023), and using representative nucleotide-binding domain within known Cas13 proteins as search queries, Yoon et al. (Yoon et al., 2024) identified Cas13an as the smallest known Cas effector to date. Comprising only 429 residues, Cas13an is on average 60% smaller than previously characterized Cas13 orthologs, which typically range from 800 to 1,400 residues in length (Yoon et al., 2024).

While CRISPR is a prominent example due to its relevance in gene therapeutics, the sequence-based and structure-based genome mining approaches can be generalized to selectively screen a wide range of miniature enzymes. For example, BLAST analysis using 15 conserved amino acid sequences identified 172 putative fungal glucuronoyl esterases (GEs) from 250 fungal genomes, with lengths ranging from 224 to 1364 residues (Dilokpimol et al., 2018). Of 16 characterized GEs, 13 from 8 phylogenetic subgroups displayed GE activity, with the smallest functional variant being 39 kDa, smaller than the previously reported minimum of 42 kDa (Dilokpimol et al., 2018). This demonstrates that genome mining effectively identifies smaller enzyme variants with preserved catalytic function.

4.2. Rational Design

Rational design, a strategy based on understanding structure-function relationships and identifying redundant regions, has been used to create miniature versions of various enzymes, such as dehydrogenases (Hibino et al., 2017; Kaida et al., 2019; Kano et al., 2021; Zhu et al., 2022), luciferases (Auld et al., 2018; Hall et al., 2012; Markova et al., 2012; Ohmuro-Matsuyama et al., 2022; Takenaka et al., 2008), esterases (Biundo et al., 2017), pullulanases (Chen et al., 2016; Wu et al., 2023), amylases (Duan et al., 2022; Shad et al., 2024), and nucleases (Ma et al., 2018; Wang et al., 2024; Zhao et al., 2023). Giving a miniaturized Cas13 as an example (Zhao et al., 2023), The process begins with identifying critical functional sites responsible for pre-crRNA processing, crRNA recognition, and target RNA cleavage based on existing structural studies. These crucial sites are retained, while the remaining structural components are categorized into secondary structural motifs. Each motif is then evaluated for potential removal based on its involvement in RNA substrate interaction. Consequently, eight redundant regions were truncated, generating miniature Cas13 variants that retained ~70% of the wild-type enzyme’s length while preserving full RNA-binding and cleavage activity (Zhao et al., 2023).

Beyond Cas13, rational design has been effectively employed to miniaturize various enzymes by carefully inspecting their structures and removing redundant regions to enhance performance. For example, a Metridia luciferase consisting of 219 residues was reduced to 147 residues through rational design (Markova et al., 2012). Multiple sequence alignments with two other luciferases revealed that the first 79 residues from the N-terminus displayed low sequence conservation, suggesting these residues were not essential for bioluminescence. Deletion of 53 and 72 residues from the N-terminus resulted in a 2.4- and 4.8-fold increase in bioluminescent activity, respectively, compared to the wild type (Markova et al., 2012).

Although rational design typically achieves a high success rate in obtaining functional variants, its generalization is constrained when experimental data are scarce, or the structure-function relationship is poorly understood (Bansal and Kundu, 2022; Savino et al., 2022; Zhao et al., 2023).

4.3. Random Deletion

Complementary to rational design approaches that demand understanding of the structure-function relationship (Bansal and Kundu, 2022; Savino et al., 2022; Vidal et al., 2023), Random deletion can identify miniature enzyme variants by screening a library of deletion mutations across various sequence regions (Savino et al., 2022), thereby free of mechanistic input. Random deletion fragments the target enzyme gene into smaller segments through nucleases digestion (Emond et al., 2020; Jones, 2005; Pikkemaat and Janssen, 2002; Shams et al., 2021) or polymerase chain reactions (Coco et al., 2001; Liu et al., 2016; Morelli et al., 2017; Pisarchik et al., 2007), and subsequently ligates these segments to construct a recombinant variant. For example, a random deletion method called in vitro transposition has been used to miniaturize a ligase consisting of 96 residues (Morelli et al., 2017). The method first inserts an artificial transposon sequence into the target gene at random positions using transposase, and then generates pairs of gene segments through polymerase chain reactions by designing primers that complement sequences within the transposon and the termini of the target gene. Eventually, these gene segments are ligated to a vector using ligase and the transposon is excised using restriction endonuclease. This method yielded a library of 9,006 mutants, with lengths ranging from 18 to 94 residues. Enzymatic assays identified a ligase variant exhibiting a 20% reduction in length while maintaining similar activity to that of the wild type (Morelli et al., 2017).

One major limitation of the in vitro transposition method lies in its low throughput capability, covering only 32% possible mutants (the total number of potential mutants is calculated as n × (n + 1) / 2, where n represents the number of nucleotides in the target gene) (Morelli et al., 2017). To address this, iterative size-exclusion and recombination (MISER) has been developed to create a library that encompasses all possible contiguous deletions through introducing restriction enzyme recognition sites between every codon throughout the gene (Shams et al., 2021). The method then employs enzymatic assays to identify the removable segments and assemble the remaining segments into recombinant variants. MISER was applied to miniaturize Cas9 and the resulting variant exhibited ~37% reduction in length while retaining approximately half of the DNA-binding activity of the wild-type Cas9 (Shams et al., 2021).

4.4. De novo Design

De novo design creates enzymes that possess novel chemical functions and/or protein scaffolds beyond nature’s repository (Huang et al., 2016; Marshall et al., 2019). Typically, enzyme de novo design starts with the creation of a theoretical active site (theozyme) for catalyzing the reaction of interest, followed by seeking an existing protein scaffold capable of accommodating the theozyme. The residues surrounding the active site are refined to maximize the stability of the active site conformation and the affinity to the transition state (Ferreira et al., 2022). This computational methodology, known as “inside-out protocol” has been used to create enzymes catalyzing Kemp elimination (Privett et al., 2012; Röthlisberger et al., 2008), Diels–Alder reaction (Siegel et al., 2010), and ester hydrolysis (Richter et al., 2012), which were experimentally validated. In the design of Kemp eliminases, 87 protein scaffolds with a diverse variety of folds, such as TIM barrels, β-propellers, jelly rolls, Rossman fold, and lipocalins, were chosen to host the potential theozyme. These scaffolds involve a wide range of lengths from 107 to 2,592 residues (Röthlisberger et al., 2008). Eight designs with lengths ranging from 251 to 371 residues showed measurable activities (Röthlisberger et al., 2008). Despite having no intention to create miniature enzymes, de novo design informs strategies to develop enzymes with a different scaffold size (Hanreich et al., 2023; Röthlisberger et al., 2008).

With deep learning emerging as a new way for building non-native protein scaffolds, creating miniature scaffolds is becoming a more feasible task than before. For example, Yeh et al. designed an artificial luciferase named LuxSit-i, which comprises merely 117 residues (14 kDa) (Yeh et al., 2023). They utilized unconstrained protein hallucination to design structures for loop and variable regions and structure-guided sequence optimization to design core regions of the enzyme (Yeh et al., 2023). This approach has generated compact luciferases, that are smaller than those achieved through rational design methods (Auld et al., 2018; Hall et al., 2012; Kim et al., 2013; Markova et al., 2012; Ohmuro-Matsuyama et al., 2022). Similarly, Ding et al. (Ding et al., 2025) created miniature polyethylene terephthalate (PET) hydrolases with molecular weights of 19 kDa, over 30% smaller than the industrial PET degradation enzyme leaf-branch compost cutinase (LCC, 28 kDa). The approach involved first extracting LCC’s catalytic motifs along with adjacent secondary structures. They then employed RF_joint (Wang et al., 2022), a deep-learning model for inpainting protein structures, to generate short sequences that could splice the extracted structures. However, initial designs showed poor expression and low activity. Aided by ProteinMPNN (Dauparas et al., 2022), a deep-learning model for protein sequence design, miniature PET hydrolases were successfully expressed and showed activity comparable to LCC (Ding et al., 2025).

In summary, enzyme miniaturization can be achieved via genome mining, rational design, random deletion, and de novo design. While all these approaches require experimental validation through recombinant DNA technologies and biological assays to test the function of designed miniaturized variants, computational techniques can accelerate the successful design of miniature enzymes. For example, genome mining benefits from bioinformatics and protein structure prediction and alignment to identify the conserved functional elements in enzyme families. Rational design benefits from structure analysis and molecular dynamics simulation to predict intermolecular interactions. De novo design is an inherently computational approach. Random deletion, despite primarily relying on experiments, can be improved in efficiency by integrating high-throughput enzyme modeling platforms like EnzyHTP (Ramírez-Palacios and Marrink, 2023; Shao et al., 2025; Shao et al., 2022), coupled with comprehensive enzyme function evaluation including catalytic activity, stability, and solubility (Goldenzweig and Fleishman, 2018; Musil et al., 2018; Nikolados et al., 2022). This function evaluation can be achieved through structure prediction (Baek et al., 2021; Jumper et al., 2021; Lin et al., 2023), molecular docking (Pagadala et al., 2017), and molecular simulation methods such as MM, QM, and QM/MM (Childers and Daggett, 2017; Jilani et al., 2021; Van der Kamp and Mulholland, 2013). To enhance the throughput and success rate of enzyme miniaturization, integrating experiments and computation is highly recommended.

5. Discussion and Outlook

Naturally occurring enzymes are versatile and efficient catalysts, honed by long-term evolution to enable the organisms to adapt optimally to their environments. Conventional engineering methods allow us to modify only a small fraction of an enzyme’s sequence, akin to how early humans shaped stones into simple tools during the Stone Age. As catalysts or drugs, enzymes should be tailored with consideration of their intrinsic chemical and physical properties and functionalities rather than being limited by evolutionary constraints. In this review, we have elucidated that small enzymes may offer considerable advantages over their larger counterparts, including improved expression levels, faster and more accurate folding, enhanced thermostability, and greater resistance to proteolysis. In a recent case of de novo designed luciferase, the smaller version demonstrated higher catalytic activity (Yeh et al., 2023). This observation suggests that miniaturization can enhance not only physical properties but also catalytic performance, such as enhanced activity, substrate binding and selectivity, which inspires future investigations. These features highlight that miniaturized enzymes are particularly well-suited for industrial applications, where cost-effective production, robustness and efficiency are critical. Additionally, as highlighted in the main text, miniaturized enzymes have the potential to drive groundbreaking advancements in metabolic and gene therapies by significantly reducing the size of therapeutic enzymes. Therefore, we envision enzyme miniaturization as a promising strategy for developing advanced enzymes with properties surpassing those of naturally occurring or conventionally mutated counterparts. This approach is poised to become an emerging focus in the field of enzyme research.

Enzyme miniaturization offers exciting opportunities for synergistic integration with other enzyme engineering approaches. Miniaturized enzymes could serve as ideal candidates for enzyme fusion techniques to create compact multi-functional chimeric enzymes with enhanced catalytic versatility and efficiency (Dixon et al., 2016; England et al., 2016; Han et al., 2023; Neugebauer et al., 2023; Xu et al., 2021). Their reduced size allows for more efficient incorporation into fusion constructs while minimizing steric hindrance between functional domains. Additionally, miniaturized enzymes could undergo directed evolution (Kubitz et al., 2022; Neugebauer et al., 2023) or mutagenesis (Han et al., 2023; Hu et al., 2024; Ohmuro-Matsuyama et al., 2023; Zhang et al., 2023) to further enhance their catalytic properties. The smaller scaffold may provide greater evolutionary plasticity, allowing for more rapid adaptation and optimization. These combinatorial approaches could yield novel biocatalysts with unprecedented functionality and performance for industrial and therapeutic applications.

Small enzymes that can perform desired catalytic functions with compact protein scaffolds serve as both ideal workhorses for various applications and valuable research models for investigating the intricate relationship between protein dynamics and catalysis. By reducing the structural redundancy of enzyme structures, researchers can gain deeper insights into how structural flexibility and motion influence enzyme functions. This practice enables the creation of minimal enzymes with high catalytic efficiency and atom economy, bridging the gap between enzyme catalysis and organocatalysis. Furthermore, miniaturization of enzymes offers great potential to discover novel protein folds, opening doors to dark areas of the protein universe that have not been untouched by natural evolution.

The field of protein design has advanced rapidly in recent years, largely due to the development of artificial intelligence. However, enzyme miniaturization remains a significant challenge. Protein folding relies on the collapse of hydrophobic residues into a tightly packed core while exposing hydrophilic residues on the surface. Shorter protein sequences complicate this process, making it more difficult to bury hydrophobic regions effectively, which increases the risk of misfolding and susceptibility to proteolysis. Consequently, even minor design flaws would significantly reduce the success rate of miniaturization. On the other hand, currently developed deep learning tools for protein structure prediction or de novo design are primarily trained on databases of standard-sized proteins rather than smaller ones. This can lead to mispredictions and unreasonable designs when applied to miniaturization tasks. To overcome these challenges, curated and well-annotated databases of small proteins, along with specialized models trained on these datasets, are essential.

Given the significances and challenges outlined above, enzyme miniaturization is inherently interdisciplinary, requiring collaboration across fields such as chemical biology, structural biology, and physical and computational biology. It holds the potential to uncover new catalytic mechanisms and protein folds, enriching our understanding of enzymology and expanding the toolkit for biocatalysis, synthetic biology and pharmaceutical industry. We anticipate rapid advancements in the design and engineering of miniature enzymes and look forward to the innovative biological solutions they will provide for addressing complex challenges across a range of applications.