Modular multi-enzyme cascades enable green and sustainable synthesis of non-canonical amino acids from glycerol

Abstract

Non-canonical amino acids (ncAAs) bearing diverse functional groups hold transformative potential in drug discovery, protein engineering, and biomaterial science. However, their industrial-scale production remains constrained by the inefficiency, high cost, and environmental burden of conventional chemical and enzymatic methods. Here, we present a modular multi-enzyme cascade platform that leverages glycerol—an abundant and sustainable byproduct of biodiesel production—as a low-cost substrate for ncAAs synthesis. Directed evolution of O-phospho-l-serine sulfhydrylase (OPSS) enhances the catalytic efficiency of C–N bond formation by 5.6-fold, enabling the efficient synthesis of triazole-functionalized ncAAs. By integrating a plug-and-play enzymatic strategy, our system enables gram- to decagram-scale production of 22 ncAAs with C–S, C–Se, and C–N side chains and can be readily scaled up to a 2 liter reaction system. Notably, water is the sole byproduct, and all products exhibit an atomic economy of >75%, highlighting the environmental compatibility of this platform. This work establishes a green, cost-effective, and industrially viable approach to expanding amino acid diversity, providing a versatile toolkit for applications in pharmaceuticals, synthetic biology, and next-generation biomaterials.

Synthetic biology enables sustainable chemical production. Here, authors develop a modular enzymatic platform to biosynthesize non-natural amino acids with C–S, C–Se, and C–N side chains from renewable substrate glycerol, achieving high efficiency and scalability.

Introduction

Amino acids are the fundamental building blocks of proteins, yet the structural diversity of natural amino acids is limited. Non-canonical amino acids (ncAAs) incorporating azido, alkenyl, nitro, and sulfur-/selenium-containing functional groups offer unique opportunities to expand the chemical and functional diversity of biomolecules¹. These ncAAs enrich the chiral complexity of natural amino acids^2–4, enhance the physicochemical and pharmacological properties of bioactive molecules^5–8, serve directly as natural medicine or drug precursors^9–13. As privileged synthetic building blocks, ncAAs provide novel reactive handles and modular structural units, enabling innovative strategies for constructing complex molecular architectures^14,15. Despite their transformative potential, the scalable and versatile production of ncAAs remains a significant challenge. While traditional chemical methods have been extensively studied, their applications are often constrained by challenges such as limited efficiency, high costs, and undesirable byproduct generation¹⁶. Given the growing demand for diverse building blocks and the limitations of conventional approaches, it has become imperative to develop catalytic strategies that are both efficient and broadly substrate-compatible. Biocatalysts, with their exceptional stereoselectivity and mild reaction conditions, emerge as ideal tools to address these challenges.

Enzymatic synthesis of ncAAs has emerged as a promising alternative, offering high stereoselectivity and milder reaction conditions^17–19. Over the past decade, engineering efforts have expanded the synthetic scope of tyrosine phenol-lyase (TPL) and tryptophan synthase (TrpB) to produce ncAAs with C–C bond side chains (Fig. 1a)^20–23. Additionally, pyridoxal 5’-phosphate (PLP)-dependent enzymes are particularly appealing for ncAAs synthesis²⁴. These enzymes catalyze the stereoselective incorporation of diverse nucleophilic substrates via an electrophilic α-aminoacrylate intermediate (Fig. 1b). For instance, PLP-dependent cysteine synthases CysM and CysK have demonstrated versatility in producing ncAAs with C–S bonds (Fig. 1c)^25,26. However, these systems face inherent limitations, including high substrate costs, narrow substrate scope in enzyme catalysis, and challenges associated with equilibrium constraints, all of which hinder their practical applications. Therefore, developing a simple, scalable, and efficient strategy for ncAAs production remains a critical goal in biocatalysis and synthetic biology.

Fig. 1. Biosynthetic strategy for the production of ncAAs via PLP-dependent enzymes.

a Biocatalytic approaches to synthesize ncAAs with C–C side chains. b The reaction mechanism by which PLP-dependent enzymes directly attach the desired side chains to the amino acid backbone for the synthesis of ncAAs. c Biocatalytic approaches to synthesize ncAAs with C–S side chains. d A substrate promiscuity enzyme is employed to design a multi-enzyme cascade pathway, starting from glycerol, to produce ncAAs with C–S, C–Se, and C–N side chains. e Bioactive molecules with a ncAA motif, shown in purple. TPL tyrosine phenol-lyase, TrpB tryptophan synthase, CysM Cysteine synthase B.

One promising solution lies in multi-enzyme cascade reactions, which leverage the synergistic cooperation of multiple enzymes to transform inexpensive precursors into high-value compounds^27–29. These systems excel in overcoming unfavorable equilibria, enabling the efficient production of target products, and offer key advantages such as modularity, scalability, and flexibility in design and assembly. Glycerol, a major byproduct of biodiesel production, has accumulated in significant quantities, posing an environmental challenge to the biofuel industry³⁰. Its conversion into high-value chemical products not only addresses this issue but also aligns with the United Nations 2030 Agenda for Sustainable Development. Multi-enzyme cascades have successfully demonstrated the synthesis of compounds such as glucosylglycerol³¹, l-aspartic acid³², and (R)-phenylethanolamine³³ from glycerol, underscoring its potential as a renewable and cost-effective feedstock.

In this work, we develop an ncAAs production system that utilizes multi-enzyme cascade reactions to generate a range of ncAAs from glycerol (Fig. 1d). By exploiting the substrate promiscuity of the key enzyme O-phospho-l-serine sulfhydrylase (OPSS) to catalyze nucleophilic substitution reactions, we synthesize ncAAs with C–S, C–Se, and C–N side chains under mild conditions. Through a combination of directed evolution and a plug-and-play strategy, we overcome the challenge of low OPSS activity, achieving ncAAs production at scales ranging from grams to tens of grams, with yields varying from good to excellent. Notably, water is the only byproduct, making the atom economy of all products highly favorable. These ncAAs demonstrate promising potential as bioactive pharmaceutical agents and drug precursors (Fig. 1e). For example, S-phenyl-l-cysteine produced and purified by this system can be easily converted into S-phenyl-l-cysteine S, S-dioxide, a potent kynureninease inhibitor, in one step. This efficient and environmentally friendly system provides a scalable solution for ncAAs production, offering immense potential for applications in biosynthesis, chemical synthesis, drug development, and biomedical research.

Results

Enzymatic nucleophilic substitution for ncAAs biosynthesis and enzyme discovery

Identifying and optimizing key catalytic elements is crucial for designing an efficient in vitro multi-enzyme cascade reaction to convert glycerol into ncAAs. The ideal catalytic elements should exhibit broad substrate compatibility, high catalytic activity, and good stability to ensure efficient performance within a complex multi-enzyme system. OPSS is a PLP-dependent transferase found in archaea³⁴. Previous studies have investigated the reaction mechanism of OPSS using X-ray crystallography and activity measurements. OPSS catalyzes product formation via a ping-pong bi-bi mechanism^35,36. The lysine residue in the active site forms an internal Schiff base with PLP. The main substrate, O-phospho-l-serine (OPS), replaces lysine in the active site, forming an external Schiff base, followed by the release of the phosphate group and the formation of an α-aminoacrylate intermediate. When a nucleophilic substrate attacks the α-aminoacrylate intermediate, the product-external Schiff base is formed. The lysine residue in the active site then reacts with the external Schiff base to regenerate the internal Schiff base and release the final product. Throughout this process, the stereochemistry at the α-carbon is retained via proton abstraction and donation occurring on the same face of the aminoacrylate intermediate, mediated by the active-site lysine residue^23,37,38. Given that the catalytic mechanism of OPSS indicates a low specificity requirement for nucleophilic attacking reagents, we hypothesize that OPSS can recognize a variety of nucleophilic reagents to synthesize the corresponding ncAAs.

To test this, we conducted in vitro enzymatic assays to compare the reactivity of CysM and CysK with three OPSSs previously identified through database mining towards non-natural nucleophiles³⁹. Allyl mercaptan (1a), potassium thiophenolate (1b), and 1,2,4-triazole (2a), which differ significantly in steric size and electron-donating groups, were used as model substrates (Fig. 2a). Activity analysis using purified enzymes indicated that CysK showed reactivity exclusively with potassium thiophenolate (Fig. 2c). In contrast, CysM and OPSSs demonstrated equivalent substrate specificity, with all three enzymes catalyzing the conversion of substrates 1a, 1b, and 2a into their respective ncAAs (Fig. 2b–d). OPSSs exhibited higher activity across all substrates, with a particularly pronounced three-order-of-magnitude superiority in its catalytic efficiency towards 2a in comparison to CysM (Supplementary Fig. 1). Notably, the pK_a of phenyl mercaptans is typically lower than that of alkyl mercaptans, which facilitates its role as a nucleophile in reactions with the α-aminoacrylate intermediate. This phenomenon was indeed observed in CysM. However, in OPSSs, both 1a and 1b exhibited similar high activity. Additionally, the three OPSSs displayed comparable activity towards azole nucleophiles, whereas AsOPSS exhibited significantly higher activity with thiol nucleophiles. In our previous findings, we noted that the OPSS-catalyzed synthesis of l-cysteine encountered product inhibition³⁹, an effect that was notably absent in the synthesis of ncAAs (Supplementary Fig. 2). These results suggested that OPSSs had considerable potential for the synthesis of ncAAs, with AsOPSS being the most effective ncAAs synthase.

Fig. 2. The identification of OPSS as a highly active class of ncAAs synthetases using various nucleophiles.

a The synthesis activity of ncAAs was compared among the cysteine synthetases CysM, CysK, and three different O-phospho-l-serine sulfhydrylases using Allyl mercaptan (1a), potassium thiophenolate (1b) and 1,2,4-triazole (2a) as nucleophiles. b−d The catalytic activity of each enzyme was evaluated by analyzing the titers of the corresponding products, S-allyl-l-cysteine, S-phenyl-l-cysteine, and 1,2,4-triazole-3-alanine, using high-performance liquid chromatography. AsOPSS OPSS from Acetobacterium sp, EbOPSS OPSS from Eubacteriaceae bacterium, CbOPSS OPSS from Clostridiales bacterium. Source data are provided as a Source Data file.

Design of modular multi-enzyme cascades for the synthesis of ncAAs from glycerol

Although OPS and several nucleophiles have been demonstrated to generate the corresponding ncAAs under OPSS catalysis, utilizing OPS as a substrate is obviously not economically feasible. To address this, a cost-effective approach was devised by designing a multi-enzyme cascade pathway that converts readily available and inexpensive glycerol into ncAAs, leveraging commercially accessible starting materials for the synthesis (Fig. 3a). The system was divided into three modules. In module I, alditol oxidase (AldO) catalyzes the oxidation of glycerol to d-glycerate, with a reduction of O₂ to H₂O₂. H₂O₂ was degraded by catalase into O₂ and H₂O, thereby safeguarding the enzymes within the pathway against the deleterious impacts of H₂O₂. Subsequently, within module II, d-glycerate underwent sequential catalytic transformations, mediated by d-glycerate-3-kinase (G3K), d-3-phosphoglycerate dehydrogenase (PGDH), and phosphoserine aminotransferase (PSAT), resulting in the conversion to OPS. The ATP required within the pathway was regenerated through coupling with polyphosphate kinase (PPK). Additionally, glutamate dehydrogenase (gluGDH) catalyzed the regeneration of NAD⁺ and l-glutamate from NADH and 2-oxoglutarate (2-og), enabling the recycling of byproducts into substrates. Finally, in module III, a plug-and-play strategy was implemented to exchange nucleophiles, thereby diversifying the synthesis of ncAAs catalyzed by OPSS. Moreover, the Gibbs free energy (ΔG′°) change of the entire pathway under physiological conditions is negative, indicating that the pathway is thermodynamically favorable for the efficient production of ncAAs (Supplementary Fig. 3).

Fig. 3. Design and construction of modular multi-enzyme cascades system for the conversion of glycerol to ncAA.

a Construction of an artificial multi-enzyme cascade reaction comprising three modules for the biosynthesis of ncAAs. b Comparison of three reported alditol oxidases and their mutants to identify the most suitable alditol oxidase. TfAldO: AldO from T. flexuosa; TfAldO^mut: TfAldO^V256L/P257I; ScAldO: AldO from Streptomyces coelicolor; ScAldO^mut: ScAldO^F278C; AbAldO: AldO from Ardenticatenaceae bacterium. c A range of PSAT, PGDH, and gluGDH from various sources was evaluated. d Time course of the biotransformation of 10 mM glycerol to 3a. Black line: d-glycerate; green line: OPS; brown line: 3a. Detailed enzyme information can be found in Supplementary Table 1. All data is presented as mean value of three independent technical experiments (n = 3) and the error bars indicate ± sd. Source data are provided as a Source Data file.

The initial investigation involved the oxidation of glycerol using various AldOs. In recent years, AldOs have attracted considerable attention for their role in sustainable glycerol upgrading^32,40–43. These enzymes employ molecular oxygen as an electron acceptor, thereby producing valuable derivatives such as d-glycerate and glyceraldehyde. The oxidation of glycerol by different AldOs was investigated at an analytical scale in the presence of catalase from E. coli (EckatG). Among all evaluated AldOs, TfAldO^mut (derived from thermophilic bacteria Thermopolyspora flexuosa) and AbAldO (derived from Ardenticatenaceae bacterium) exhibited the highest enzymatic activity (Fig. 3b). On account of its thermal stability, TfAldO^mut was selected for further analysis (Supplementary Fig. 4). With 10 mM glycerol as substrate, a 12 h analytical-scale reaction produced 6.3 mM d-glycerate and 2.2 mM glyceraldehyde (Supplementary Fig. 5). Next, the most appropriate enzymes for the assembly of module II were identified and selected. Saccharomyces cerevisiae and a select few plant species are the only known organisms that possess G3K⁴⁴. With this in mind, we selected the G3K from S. cerevisiae (ScG3K) for soluble expression in E. coli. Our module II demonstrated high efficiency in generating 3-phosphoglycerate when it was supplied with d-glycerate, appropriate amounts of ATP, polyphosphate (polyP), and PPK sourced from Sulfurovum lithotrophicum (SlPPK). To ascertain the optimal enzymes for the synthesis of OPS, a series of enzymes from different sources, including PSAT, PGDH, and gluGDH, were evaluated (Fig. 3c). PGDH from E. coli (EcPGDH), PSAT from Acinetobacter baylyi (AbPSAT), and gluGDH from Peptoniphilus asaccharolyticus (PagluGDH) were selected for their high enzymatic activity. Then, in the presence of all selected enzymes from module II, OPS was prepared in Tris-HCl buffer containing 10 mM d-glycerate. The reaction reached equilibrium after 6 h, and the yield of OPS was 6.8 mM (Supplementary Fig. 6).

Subsequently, the feasibility of the entire cascade reaction was assessed using a one-pot three-step process in vitro, with 1a serving as the model nucleophile (Fig. 3d). The multi-enzyme cascade reaction was conducted with a substrate concentration of 10 mM glycerol. Following the passage of the reaction through modules I and II, the production of d-glycerate and OPS was observed, with yields of 6.3 mM and 4.3 mM, respectively. Afterward, the purified AsOPSS and 4.3 mM 1a, in equimolar amounts to OPS, were added to the system, resulting in the formation of 4.2 mM 3a within 2 h, achieving a yield of 97.6% calculated based on the nucleophile. Drawing from the results presented, it is evident that a multi-enzyme cascade reaction has been successfully engineered to convert glycerol into ncAAs. Besides, considering that the phosphate group can be recycled by the PPK-catalysed ATP regeneration reaction, the overall reaction equation (Supplementary Table 1) shows that water is the only byproduct, from which the atom economy of 3a is calculated to be ~76.7%, and the atom economies of the remaining 21 products are also above 75%.

Identification of AsOPSS substrate spectrum and enhancement of activity

We evaluated the potential of AsOPSS to catalyze reactions with other thiol and azole analogs to expand the ncAAs library (Table 1 and Fig. 4a). AsOPSS demonstrated compatibility with a variety of structurally diverse alkyl thiols (1c-1g), aromatic thiols (1h-1n), and heteroaromatic thiols (1o-1q). It can also tolerate electron-donating substituents such as amino (1c, 1k), hydroxyl (1 d), and methoxy (1 l) groups, as well as electron-withdrawing substituents including carboxyl (1e), nitro (1 h), and bromo (1i) groups. The alkyl thiols 1c and 1e were converted into S-(2-aminoethyl)-l-cysteine (3c) and S-carboxymethyl-l-cysteine (3e), respectively. 3c is an analog of lysine that has been extensively utilized to model chemical reactions involving lysin⁴⁵, and 3e is an antioxidant and mucolytic agent utilized in the treatment of various respiratory disorders⁴⁶. AsOPSS exhibited high activity towards aromatic thiols, with the highest activity observed for 1j, reaching 256.7 U/mg. Aromatic thiols 1 h, 1i, 1k, and 1 l also displayed high activity, with the product of 1 h, S-(2-nitrophenyl)-l-cysteine (3 h), commonly utilized as a highly immunogenic hapten due to the presence of the nitrophenyl group⁴⁷. AsOPSS can also be applied to catalyze naphthalene substrates (1 m) and aromatic thiols with large substituents (1n) without additional engineering design, although there is a reduction in activity. The remaining thiols can be effectively transformed into their corresponding ncAAs, thereby enriching the structural and functional diversity of ncAAs. Moreover, AsOPSS demonstrated its capability to rapidly convert phenylselenol (1r) into S-phenyl-L-selenocysteine (3r). The chemical synthesis of chiral selenides presents significant challenges, compounded by the current lack of efficient enzymatic methodologies for constructing C-Se bonds⁴⁸. AsOPSS exhibited a remarkable specific activity of 174.9 U mg⁻¹ toward 1r, underscoring its exceptional efficacy in facilitating C-Se bond formation. This catalytic performance highlights the enzyme’s potential as a valuable biocatalytic tool for stereoselective synthesis of selenium-containing compounds.

Table 1.

Specific activities of purified AsOPSS toward nucleophiles compounds

Substrate	Specific activities(U mg-1)a
	AsOPSS	AsOPSST120Q/F144Y
Allyl mercaptan (1a)	144.7 ± 1.3a	146.2 ± 6.3
Benzenethiol (1b)	151.0 ± 1.2	89.5 ± 6.4
Cysteamine (1c)	68.7 ± 3.5	74.3 ± 2.0
2-Mercaptoethanol (1 d)	120.6 ± 4.1	115.4 ± 3.4
Mercaptoacetic acid (1e)	81.5 ± 3.4	94.8 ± 0.8
2-Methyl-2-propanethiol (1 f)	22.9 ± 1.3	10.8 ± 1.6
3-Mercapto-1,2-propanediol (1 g)	35.0 ± 2.6	15.9 ± 2.1
2-Nitrobenzenethiol (1 h)	114.0 ± 2.8	79.0 ± 4.2
2-Bromothiophenol (1i)	179.2 ± 1.7	183.2 ± 4.5
4-(Trifluoromethyl) thiophenol (1j)	256.7 ± 7.7	204.5 ± 8.1
4-Aminothiophenol (1k)	109.7 ± 6.1	70.0 ± 2.8
4-Methoxybenzenethiol (1 l)	112.0 ± 9.5	57.1 ± 4.1
1-Naphthalenethiol (1 m)	18.7 ± 1.8	16.5 ± 1.8
Triphenylmethyl mercaptan (1n)	27.1 ± 1.4	20.7 ± 2.5
4-Pyridylethylmercaptan (1o)	48.3 ± 3.5	32.9 ± 1.1
Furfuryl mercaptan (1p)	85.3 ± 3.8	66.9 ± 1.5
2-Thiophenethiol (1q)	128.2 ± 1.1	140.7 ± 3.7
Phenylselenol (1r)	174.9 ± 6.3	107.5 ± 5.0
1,2,4-Triazole (2a)	5.9 ± 0.4	23.2 ± 1.3
Tetrazole (2b)	4.8 ± 0.3	26.9 ± 0.3
1H-pyrazole-3-carbonitrile (2c)	7.8 ± 1.1	40.6 ± 1.9
5-hydroxy-pyrazol (2 d)	6.1 ± 0.3	31.8 ± 1.4

^aSpecific activity of 1 U mg⁻¹ corresponds to 1 μmol of substrate converted per minute and per mg of enzyme. All data is presented as mean value of three independent experiments (n = 3) and the error bars indicate ± sd.

Fig. 4. Substrate spectrum validation and directed evolution of AsOPSS.

a The structures of the 19 additional nucleophiles used in this study. They were employed to determine the specific activity of AsOPSS towards various nucleophiles. The reactions were conducted at an enzyme loading of 0.005 g/L, under conditions of 45 °C and pH 7.0, for a duration of 3 min. b Molecular docking analysis of the intermediate α-aminoacrylate with wild-type AsOPSS was conducted to identify potential target sites. Residues within 6 Å of the α-aminoacrylate in the active site were selected as candidate mutation sites, except for the key conserved catalytic residue K42, which may responsible for forming an internal Schiff base with PLP. The selected residues include P68, T69, S70, N72, T73, G74, R100, T120, Q143, F144, G177, T178, L212, and K213. c The distance changes between the carbon atom of the α-aminoacrylate intermediate to form the C–N bond with the nitrogen atom of the 2b during the 100 ns simulation. d. The binding mode of 2b and the α-aminoacrylate intermediate in the AsOPSS^T120Q/F144Y variant. Source data are provided as a Source Data file.

The pharmaceutical industry encompasses a multitude of reactions that form C–N bonds⁴⁹, and synthetic methods that directly convert C–H bonds into C–N bonds align with the principles of green chemistry and atom economy. Hence, the catalytic synthesis of diverse ncAAs featuring C–N side chains hold substantial importance for the pharmaceutical sector. To assess the potential of AsOPSS in facilitating C–N bond formation, three additional azole nucleophiles (2b-2d) were chosen for evaluation. The results demonstrated that all of the substrates yielded the desired products, further confirming the substrate promiscuity of OPSS. However, compared to thiol nucleophiles, AsOPSS demonstrated markedly reduced activity towards azole nucleophiles, which prompted us to pursue protein engineering of AsOPSS.

Based on the known OPSS catalytic mechanism^35,36, the α-aminoacrylate intermediate was docked into the structure model of AsOPSS predicted by AlphaFold3, leading to saturation mutagenesis of 14 residues within a 6.0 Å radius of the α-aminoacrylate (Fig. 4b). As pyrazole (2b) exhibited the lowest activity of AsOPSS when used as the nucleophile, it was selected as a model substrate for directed evolution. A total of 6720 variant clones were screened in one single-site saturation library, resulting in the identification of the variants AsOPSS^T120Q and AsOPSS^F144Y that showed higher specific activities than the wild type. The activity of the AsOPSS^F144Y increased from 4.8 U/mg to 25.0 U/mg for 2b. To further enhance the catalytic efficiency, two single-point mutations were combined to construct the double variant, AsOPSS^T120Q/F144Y. The AsOPSS^T120Q/F144Y exhibited an activity of 26.9 U/mg, representing a 5.6-fold increase compared to the wild-type. The variants were subsequently assessed using substrates 2a, 2c, and 2 d, with the double mutant AsOPSS^T120Q/F144Y demonstrating the highest activity across all azole nucleophiles tested (Supplementary Fig. 7).

To further investigate, we analyzed the binding of the α-aminoacrylate intermediate and 2b in the wild-type enzyme and the variant AsOPSS^T120Q/F144Y using 100 ns MD simulations. Trajectory analysis revealed that the variant did not affect the interaction between α-aminoacrylate and surrounding residues, but significantly altered the binding stability of 2b in the active site. Interaction force analysis showed that the mutation increased the number of hydrogen bonds formed between 2b and neighboring residues (Supplementary Fig. 8). Furthermore, we observed that the distance between 2b and the α-aminoacrylate intermediate remained within the range where nucleophilic attack could occur in the variant throughout the 100 ns simulation, whereas for the wild-type enzyme, 2b left the substrate pocket after ~28 ns (Fig. 4c). These findings indicate that the mutations stabilize the binding mode of the imidazole nucleophile, forming additional hydrogen bonds that facilitate the nucleophilic attack of the substrate on the intermediate α-aminoacrylate, thereby enhancing the catalytic activity for C–N bond formation (Fig. 4d). Nevertheless, the AsOPSS^T120Q/F144Y did not demonstrate enhanced activity towards thiols and selenols (Table 1). Consequently, the wild-type AsOPSS was selected for the synthesis of ncAAs with C–S and C–Se side-chains, whereas the AsOPSS^T120Q/F144Y was chosen for the synthesis of ncAAs with C–N side-chains.

Optimization multi-enzyme cascades for scalable production of ncAAs

To identify the optimal reaction conditions for each module, we analyzed the rate-limiting enzymes and characterized their enzymatic properties. The optimal temperature and pH for the rate-limiting enzymes TfAldO^mut, EcPGDH, and AsOPSS were used to establish the reaction conditions for modules I, II, and III, which were ascertained to be 40°C, pH 8.0; 45 °C, pH 8.0; and 60°C, pH 7.5, respectively (Supplementary Fig. 9 and Supplementary Fig. 10). In addition, it is noteworthy that in module II, the transfer of phosphate groups is catalyzed by ScG3K and SlPPK and involves magnesium ions (Mg²⁺) as their activating metal ions. Mg²⁺ can bind to phosphate groups, and the concentration of Mg²⁺ significantly affects the thermodynamics of the reaction. Therefore, a comprehensive optimization of the concentrations of Mg²⁺ and polyP in module II was conducted with 30 mM d-glycerate and 5 mM ATP (Supplementary Fig. 11). OPS was detectable across all tested concentrations of Mg²⁺, with its production rate escalating in direct proportion to the Mg²⁺ concentration. The highest yield of OPS was achieved at a Mg²⁺ concentration of 20 mM, where it peaked at 78.2% with a polyP concentration of 15 mM. However, an increase in polyP concentration resulted in the formation of a milky, viscous mixture during the initial stages of the reaction, which significantly impeded the conversion rate. Importantly, the presence of 20 mM Mg²⁺ and 15 mM polyP did not affect the production of ncAAs in module III, as illustrated in Supplementary Fig. 12.

To enhance the conversion rate, the enzyme ratios were optimized to achieve a coordinated reaction rate among the different enzymes. By optimizing the ratio of TfAldO^mut to catalase, it was determined that the optimal enzyme ratio for the combination of TfAldO^mut and catalase was 1:100 (Supplementary Fig. 13). This configuration produced 19.3 mM d-glycerate in a 12 h reaction (Fig. 5a). In module II, the optimal enzyme ratio for ScG3K: EcPGDH: AbPSAT was determined to be 10: 5: 1, with further increases in the ratios of ScG3K and EcPGDH were found to have no additional effect on OPS production (Supplementary Fig. 14). By integrating the production scale from module I, the yields of OPS reached 7.9 mM and 14.9 mM at 1 and 2 h, respectively, showcasing a remarkable enhancement in production efficiency. Furthermore, from an economic perspective, the requirement for cofactors was minimized, with 2 mM ATP, 2 mM NAD⁺, and 12 mM l-glutamate found to be adequate for the efficient synthesis of OPS (Supplementary Fig. 15).

Fig. 5. Production of ncAAs at the preparative scale.

a After the optimization of the multi-enzyme cascade system, the time course for the biosynthesis of 3a as the target product was conducted at the preparative scale. b Under optimal reaction conditions, ncAAs with C–S, C–Se, and C–N side chains were synthesized from glycerol via the multi-enzyme cascade. Yields, productions, and e.e. values are listed below. Due to the lack of D-form and racemic standards, the configurations of compounds 3o and 3 v were evaluated by comparing the optical rotations of the enzyme-catalyzed products with those of the corresponding L-form standards. All data is presented as mean value of three independent technical experiments (n = 3) and the error bars indicate ± sd. Source data are provided as a Source Data file.

A plug-and-play strategy was implemented to replace the nucleophiles and AsOPSS variants involved in module III, enabling the efficient production of multiple ncAAs through a single engineered pathway. The wild-type AsOPSS was shown to effectively catalyze the transformation of 14.9 mM OPS and equistoichiometric ratios of thiol or selenol nucleophiles, resulting in the formation of ncAAs with yields ranging from 68.4% to 98.5%. Except for compound 3 h (ee >95%) and two products (3o and 3 v) evaluated by optical rotation (Supplementary Method 1) without reported ee values, all other products exhibited ee values exceeding 99%, confirming their high stereospecificity (Fig. 5b). In the synthesis of ncAAs with C–N side chains, the low activity of wild-type AsOPSS towards azole nucleophiles necessitated the addition of a substantial amount of AsOPSS. By substituting wild-type AsOPSS with the variant AsOPSS^T120Q/F144Y, the amount of enzyme required was significantly reduced, facilitating gram-scale production of ncAAs with C–N side chains.

Finally, scale-up production was performed for selected ncAAs that exhibit pharmaceutical activity or provide opportunities for further functionalization. Using 150 mM glycerol as the substrate, various ncAA products were successfully synthesized on a 10-gram scale (Table 2 and Supplementary Fig. 16). Due to the hydrophobic nature of 3j, n and r, gradual phase separation occurred during the reaction process. Subsequent centrifugation, washing, and crystallization procedures yielded the corresponding products with high isolation efficiency. For 3a, b and e, which exhibit limited solubility in neutral aqueous solutions but demonstrate significantly reduced solubility under isoelectric point pH conditions, a pH-modulated separation and purification strategy was implemented to achieve favorable isolated yields (Supplementary Method 2). Among these productions, 3a is recognized as a pivotal bioactive constituent in garlic extracts⁵⁰, with documented biological and pharmacological activities including antioxidant effects, anticancer properties, and neuroprotective functions^51,52. 3e, as the principal constituent of the mucolytic agent Carbocisteine, has been widely employed in Asia and Europe for the management of sputum-producing respiratory disorders⁵³. Clinical studies have demonstrated that administration of Carbocisteine reduces exacerbation frequency in chronic obstructive pulmonary disease (COPD) therapy^46,54. 3j represents a fluorinated aromatic amino acid analog where fluorine substitution enhances metabolic stability and pharmacokinetic properties. The strategic incorporation of fluorine atoms, present in over 20% of FDA-approved pharmaceuticals, offers novel opportunities for developing bioactive molecules^55,56. 3n exhibits potent anticancer activity through robust inhibition of the mitosis-driving protein Eg5⁵⁷. In recent years, organoselenium compounds have emerged as a strategic molecular class in drug discovery due to their distinctive biochemical modulation capabilities⁵⁸. 3r significantly expands opportunities for identifying more efficient and cancer-selective chemopreventive and chemotherapeutic agents. Additionally, 3b serves dual roles as both a critical synthetic intermediate for the oral HIV-1 protease inhibitor Nelfinavir⁵⁹ and a precursor for the synthesis of the kynureninase inhibitor S-phenyl-l-cysteine S, S-dioxide¹³. To further functionalize 3b, isolated 3b was used as the starting material to synthesize S-phenyl-l-cysteine S, S-dioxide in a one-pot process with a conversion rate of 75% and an isolated yield of 66% (Supplementary Method 3). These results demonstrated the substantial potential of this multi-enzyme cascade system, providing a viable approach for the industrial synthesis of valuable ncAAs. In the future, in order to increase the titer of ncAAs, the yield and productivity of d-glycerate need to be further improved, which can be achieved by combining it with electrocatalysis.

Table 2.

Scalable synthesis of six non-canonical amino acids at 10-gram scale

Entry	ncAA	Yield (g L-1)	Isolated yield (%)a
1	(S-allyl-l-cysteine) 3a	9.50	68.4
2	(S-phenyl-l-cysteine) 3b	11.66	77.7
3	(S-carboxymethyl-l-cysteine) 3e	9.21	65.7
4	(S-4-(trifluoromethyl)-thiophenol-l-cysteine) 3j	14.80	74.6
5	(S-triphenylmethyl-l-cysteine) 3n	14.07	54.6
6	(Se-phenyl-l-cysteine) 3r	11.45	67.2

^aIsolated yields were determined according to the weight of the purified ncAAs and the theoretical yield. Data are from one independent experiment. Source data are provided as a Source Data file.

Discussion

In this study, we established a one-pot, three-step multienzyme cascade reaction system for the efficient biosynthesis of ncAAs from glycerol, demonstrating potential advantages in green chemistry and sustainable development. Using glycerol as the starting material and producing water as the sole by-product, the system exhibits excellent atom economy and environmental compatibility. It not only addresses the environmental challenges posed by surplus glycerol in the biodiesel industry but also achieves the green synthesis of high-value-added chemicals, aligning with the goals of sustainable development. Moreover, the modular design of this system provides relative flexibility. Under the optimal conditions for general applicability, simply replacing the nucleophiles allows for the efficient, single-pathway engineering synthesis of a wide range of ncAAs, with product yields ranging from 68.4% to 98.5%. When the system was scaled up and substrate concentration increased, it successfully achieved the anticipated production of decagram-scale products, demonstrating its scalability. This versatility enhances synthetic efficiency and product diversity, which is difficult to achieve with traditional multi-step chemical synthesis or single-enzyme catalytic systems. However, certain limitations remain, particularly in enhancing the yield and productivity of d-glycerate, which may require the integration of complementary technologies, such as electrocatalysis, for further progress.

The expanding utility of PLP-dependent enzymes in ncAAs biosynthesis stems from their unique catalytic mechanism. These enzymes facilitate β-elimination of L-amino acid substrates to generate electrophilic aminoacrylate intermediates, enabling modular incorporation of diverse nucleophiles as side chains through Michael addition. TPL and TrpB are prototypical PLP dependent enzymes that have been widely exploited for C-C side chain modification in ncAAs synthesis. However, TPL catalysis suffers from intrinsic thermodynamic constraints requiring a large excess of substrate to drive product formation⁶⁰. Likewise, TrpB’s reliance on indole derivatives restricts the structural diversity of accessible tryptophan analogs^23,61. These challenges underscore the pressing need for novel PLP‑dependent enzymes with expanded reaction scope. AsOPSS and the engineered variant exhibit remarkable catalytic efficiency and a broad substrate spectrum for ncAAs synthesis. Structural diversification of ncAAs can be conveniently achieved by leveraging abundant and cost-effective commercial nucleophiles. Moreover, the reactions catalyzed by AsOPSS exhibit significant efficiency not only for traditional nucleophilic reagents but also for aromatic reagents, including both electron-rich and electron-deficient derivatives. This highlights its ability to catalyze aromatic nucleophilic substitution reactions. For example, substrates such as 4-(trifluoromethyl)-benzenethiol and 2-nitrobenzenethiol achieved high catalytic efficiency (over 90%), and aromatic selenols also underwent notable conversion (over 70%), resulting in products with substantial functional diversity. These results underscore the remarkable capacity of OPSS to mediate aromatic substitution reactions, thereby greatly expanding the scope of enzyme-catalyzed transformations. In contrast, enzymes like CysM and CysK displayed relatively limited substrate compatibility. Through directed evolution, the AsOPSST^120Q/F144Y variant was developed, resulting in a 5.6-fold increase in activity toward azole nucleophiles, facilitating efficient C–N bond formation and broadening the range of ncAAs that could be synthesized. In comparison, many traditional enzymatic systems are restricted by substrate specificity and catalytic efficiency, limiting their ability to fulfill the diverse needs of ncAA synthesis.

Cost represents a critical determinant in industrial production. To systematically evaluate the economic feasibility for potential industrial-scale applications, we conducted a comprehensive cost analysis using 3b as a representative case study. The production cost of enzymes obtained through high-density E. coli fermentation is ~$50/kg⁶². Based on this, the production of 1 kilogram of 3b necessitates ~0.35 kg of enzymes, with an estimated cost of $ 17.5. For substrate consumption, the production of 1 kg of 3b requires 1.19 kg glycerol ($0.98), 3.95 kg polyphosphate ($1.22), 0.17 kg NAD⁺ ($19.75), 0.13 kg ATP ($8.07), 1.52 kg L-glutamate ($3.33), and 0.8 kg nucleophile 2b ($328.77). The implementation of a streamlined downstream process based on centrifugation and crystallization accounts for ~20% of the total production costs. Consequently, the total estimated production cost for 3b amounts to $ 455.54/kg. Remarkably, the current market price of 3b ($984/kg) demonstrates a multi-fold cost advantage of multi-enzyme cascade system. While commercial prices vary for other ncAAs and their corresponding nucleophiles produced in this study, all cases maintain substantial cost-benefit ratios.

Beyond economic merits, this study demonstrates a markedly reduced environmental footprint relative to conventional chemical syntheses for equivalent products. Conventional routes to 3c and  d typically relies on toxic precursors such as bromoethanol or mercaptoethanol, which pose inherent risks due to their volatility, irritancy, and ecotoxicological properties^63,64. The synthesis of 3 h conventionally requires nitrobenzene, a highly toxic aromatic compound whose severe health hazards and environmental persistence have been well-documented⁶⁵. By contrast, the enzymatic cascade system operates entirely in an aqueous medium, generates water as the sole byproduct, and eliminates dependence on heavy-metal catalysts or hazardous reagents. Furthermore, the multi-enzymatic platform achieves both high atom utilization and significantly enhanced productivity. For instance, while chemoenzymatic synthesis of 3b from high-cost tryptophan precursors attained an 81.3% yield⁶⁶, and fluorinated L-α-amino acid syntheses from economical aldehydes suffer from suboptimal volumetric productivities (<1 g/L) that hinder pharmaceutical-scale production⁶⁷, our system overcomes these limitations through optimized cascade efficiency. These features collectively highlight this study’s adherence to green chemistry principles, positioning it as a viable platform for sustainable industrial biomanufacturing.

In conclusion, this study offers a promising and scalable solution for the environmentally friendly biosynthesis of ncAAs, addressing some of the limitations of traditional chemical synthesis and single-enzyme catalysis. It provides a useful framework for the development of multi-enzyme cascade systems and the promotion of green synthesis technologies in industrial applications. Looking ahead, integrating advanced protein engineering and complementary approaches could further enhance the system’s capabilities, contributing to the broader synthesis of ncAAs and their derivatives and supporting advancements in green chemistry and biomanufacturing.

Methods

Chemicals and materials

All chemicals used in this work were obtained from Abamas-beta, Aladdin, Meryer, and Shanghai Yuanye Bio-Technology Co. The purity of reagents was >98%. The PrimeSTAR MAX DNA polymerase and restriction enzyme Dpn I were purchased from Takara (Beijing, China). The genes of cysM and cysK were amplified from the genome of E. coli. The other genes were synthesized and codon-optimized for E. coli BL21 (DE3) by Exsyn-Bio (Shanghai Saiheng BioTechnology Co.). All enzyme amino acid sequences are presented in Supplementary Data 1, and all primer sequences are provided in Supplementary Data 2.

Protein expression and purification

E. coli BL21 (DE3) harboring recombinant plasmids with the target gene for protein expression. The cells were cultured in 100 mL of LB medium with 50 mg/L kanamycin at 37°C and 220 rpm. Upon reaching an optical density at 600 nm (OD₆₀₀) of 0.6–0.8, induction of enzyme overexpression was initiated by adding 0.2 mM isopropyl-β-d-thiogalactopyranoside. After 16 h of induction at 18°C, the cells were harvested by centrifugation and resuspended in 50 mL of KPi buffer (pH 8.0) containing 10 mM imidazole and 500 mM NaCl. Subsequently, cellular lysis was achieved via high-pressure homogenization, followed by centrifugation of the lysate at 4000 g and 4°C for 20 mins. Purification of the protein from the supernatant was conducted using Ni-chelating affinity chromatography. The enzyme purity was assessed via SDS-PAGE (Supplementary Fig. 17) and quantified utilizing the BCA Protein Assay Kit from Tiangen (Beijing, China).

Enzyme activity assay

The activity of AldO was determined by measuring the production of d-glycerate and glyceraldehyde using high-performance liquid chromatography (HPLC). The reaction mixture included 100 mM glycerol, an excess of catalase, and an appropriate amount of AldO in 100 mM Tris-HCl buffer (pH 7.0). One unit of AldO activity was defined as the amount of enzyme that released 1 μmol of glyceraldehyde or 0.5 μmol of d-glycerate per minute. The effect of temperature on TfAldO^mut was evaluated in the reaction mixture (pH 7.0) at 30, 40, 50, 60, 70, and 80°C. The effect of pH was determined in the reaction mixture with pH values ranging from 5.0 to 9.0 at 40°C.

The activity of G3K was monitored by observing the generation of ADP. Routine assays were performed at 40°C, with the reaction mixture comprising 20 mM d-glycerate, 20 mM ATP, 10 mM MgCl₂, and an appropriate amount of enzyme in 100 mM Tris-HCl buffer (pH 7.0). After a 3 min incubation period, 200 µL of the reaction mixture was extracted and mixed with 100 µL of 20% TCA solution to terminate the reaction. One unit of G3K activity was defined as the amount of enzyme that released 1 μmol of ADP per minute.

The conversion of d-glycerate-3-phosphate to 3-phosphohydroxypyruvate catalyzed by PGDH is thermodynamically unfavorable for spontaneous progression. Hence, the activity of PGDH was assayed by coupling it with the downstream enzyme PSAT. The reaction mixture included 2 mM d-glycerate-3-phosphate, 5 mM NAD⁺, 5 mM l-glutamate, 100 µM PLP, 1 mM DTT, an excess of PSTA, and an appropriate amount of PGDH in 100 mM Tris-HCl buffer. One unit of PGDH activity was defined as the amount of enzyme that reduced 1 μmol of NAD⁺ per minute. The optimal temperature and pH for EcPGDH were determined to be in the ranges of 30–80°C and pH 5.0–9.0, respectively.

The activity of PSAT was assayed in 100 mM KPi buffer containing 10 mM l-glutamate, 100 µM PLP, 500 µM 3-phosphooxypyruvate, and an appropriate amount of enzyme. Assays were conducted with a 10 min incubation period and terminated by adding 20% TCA. One unit of PSTA activity was defined as the amount of enzyme that released 1 μmol of OPS per minute.

The activity of OPSS was assayed at 40°C using the reaction mixture consisting of 20 mM OPS, 100 µM PLP, 1 mM DTT, 20 mM nucleophile, and an appropriate amount of enzyme in 100 mM Tris-HCl buffer (pH 7.0). After 3 min of reaction, 20% TCA was added to terminate the reaction. One unit of OPSS activity was defined as the amount of enzyme that converted 1 μmol of substrate OPS per minute. The optimal temperature and pH for AsOPSS were determined to be in the ranges of 30–80°C and pH 5.0–9.0, respectively.

GluGDH reversibly catalyzed the regeneration of l-glutamate and NAD⁺. The activity of gluGDH was assayed by monitoring the reduction of NAD⁺. The reaction mixture comprised 10 mM 2-oxoglutarate, 10 mM NADH, 50 mM NH₄Cl, and an appropriate amount of enzyme in 100 mM Tris-HCl buffer (pH 8.0). The standard assays were performed at 45 °C and pH 8.0.

Stability measurements half-life (t_1/2)

The stability of the remaining enzymes was evaluated by measuring residual activity after incubating each enzyme under its respective optimal modular reaction conditions over varying time intervals. The inactivation kinetics of the enzyme were analyzed using a first-order kinetic model (Eq. 1), where A represents the residual enzyme activity at time t (in hours), A₀ corresponds to the initial enzyme activity, and k denotes the first-order inactivation rate constant (h⁻¹). Based on this kinetic analysis, the stability of the enzyme was quantitatively expressed through its half-life, which was defined as the time required for the residual activity to decrease to 50% of the initial activity.

$$\ln \mathrm{A}=\ln {\mathrm{A}}_0-kt$$ 1

Protein structure modeling and molecular docking

The structure of AsOPSS was predicted using AlphaFold3 in the Google Collaboratory with default parameters (https://golgi.sandbox.google.com/)^68,69. Five structures were generated for each protein, and the structure with the highest pLDDT value was selected for subsequent simulations.

Docking analysis was performed using the Glide module in Schrödinger 12.0. The three-dimensional structures of the ligand were obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/). The ligand structure was subjected to LigPrep for optimization before the docking study. The protein structure was optimized using Protein Prep Wizard in Maestro modeling interface. The prepared protein and ligand underwent docking using the Glide (Grid-Based Ligand Docking with Energetics) algorithm. A grid box of 8 × 8 × 8 Å centered on the active pocket (containing key residues 69 T, 70S, 72 N, 73 T, 120 T, 143Q, 177 G, 178 T, and 211 G) serves as a search space for finding suitable ligand binding modes. Flexible docking (SP-GLIDE) was performed and the appropriate complex was selected for further analysis based on a comparison of energy scores and geometric conformations.

MD simulation

MD simulations were employed using GROMACS 2020.3 package, with the AMBER14SB force field⁷⁰. The force field parameters for the α-aminoacrylate and 2b were sourced from the https://www.bio2byte.be/acpype/ online website (charge method = bcc, atom type = AMBER)⁷¹. Initially, energy minimization was performed using the steepest descent algorithm for 50000 steps, followed by the conjugate gradient algorithm, with a force constant of 100 kJ/mol/nm for the protein’s backbone atoms, to ensure the reasonability of overall protein conformation. Afterward, under the canonical ensemble (NVT), the system was gradually heated to 310 K over 100 ps using the V-rescale temperature coupling method. Then, the Berenson pressure coupling method was used to maintain a pressure of 1 bar for 1 ns throughout the simulation, which was repeated three times. After the above steps, a 100 ns MD simulation was performed without any positional restraints using three different replicas, respectively. The LINCS algorithm was employed to constrain only hydrogen bonds. The particle mesh Ewald method was utilized to compute long-range electrostatic interactions, and a cutoff distance of 10 Å for short-range van der Waals interactions. The time step was set to 2 fs, and the trajectory coordinates were saved every 10 ps. Finally, 1000 frames were extracted from the simulation trajectory and analyzed for protein-ligand non-covalent interactions using the PLIP tool⁷².

Construction of AsOPSS variants and high-through screening

The plasmid pET-28a, which harbors the AsOPSS sequence, was employed as the template for site-saturation mutagenesis. Primers incorporating NNK codons were designed to target the amino acids intended for mutation. The PCR products, treated with DpnI, were subsequently transformed into E. coli BL21 (DE3) cells.

The high-throughput screening was conducted based on the byproduct phosphate generated from the synthesis of ncAAs. The formation of ammonium phosphomolybdate precipitate occurs in the presence of phosphate when ammonium molybdate and ascorbic acid are combined. The rate of ncAAs production is calculated by measuring the absorbance at a wavelength of 700 nm. Each single-point mutation library was screened for 480 transformants, which were expressed at 18°C for 16 h in a self-inducing medium. The components of the self-inducing medium included 10 g/L peptone, 5 g/L yeast extract, 5 g/L glycerol, 0.5 g/L glucose, 2 g/L lactose, 2 mM MgSO₄, 5 mM Na₂SO₄, 50 mM NH₄Cl, 25 mM Na₂HPO₄, 25 mM KH₂PO₄, and 50 mg/L kanamycin. The cultured cells were harvested by centrifugation and subjected to three freeze-thaw cycles at -80°C before the reaction. The reaction mixture contained 20 mM 2b, 100 μM PLP, 20 mM OPS, and 50 μL of the freeze-thawed cells in 100 mM Tris-HCl buffer (pH 7.0). After a 30 min incubation, 48 µL of reaction termination solution was mixed with 480 µL of ammonium molybdate solution (100 mM (CH₃CO₂)₂Zn, 15 mM (NH₄)₂MoO₄) and 120 µL of ascorbic acid solution (1 g/L VC) for absorbance measurement.

Cascade reaction

For the analysis-scale reaction, the biosynthesis of ncAAs was performed at 40°C in a 1 mL volume of 100 mM Tris-HCl buffer (pH 7.0) containing 10 mM glycerol. Module I consisted of 1000 U/L TfAldO^mut and 50000 U/L EckatG, with the reaction proceeding for 12 h. Additionally, 50000 U/L of EckatG was added at the 4 h and 8 h intervals during the reaction. Following this, module II was introduced, comprising 300 U/L ScG3K, 300 U/L EcPGDH, 300 U/L AbPSAT, 500 U/L SlPPK, 500 U/L PagluGDH, along with 20 mM Mg²⁺, 10 mM polyphosphate, 5 mM ATP, 5 mM NAD⁺, and 20 mM l-glutamate. Finally, in module III, various nucleophiles in the same stoichiometric ratio as OPS generated in module II and AsOPSS were added.

For the preparative-scale reaction, 30 mM glycerol was used as the substrate and subjected sequentially to catalysis by modules I, modules II, and modules III at optimal pH and temperature conditions. Module I consisted of 2000 U/L TfAldO and 200000 U/L EckatG, with additional EckatG added during the reaction. Module II included 3000 U/L ScG3K, 1500 U/L EcPGDH, 300 U/L AbPSAT, 2000 U/L SlPPK, 2000 U/L PagluGDH, along with 20 mM Mg²⁺, 15 mM polyphosphate, 2 mM ATP, 2 mM NAD⁺, and 30 mM l-glutamate. Module III comprised various nucleophiles in the same stoichiometric ratio as OPS generated in module II and either AsOPSS or AsOPSS^T120Q/F144Y.

Analysis and Detection Methods

NMR spectra were recorded on a Bruker AV500 spectrometer (500 MHz for ¹H NMR). Mestrenova 14.0 was used for NMR analyses. HRMS data were collected on an Agilent 6546 LC-QTOF system. Mass spectrometry analysis method is shown in Supplementary Method 4. NMR and MS results are shown in Supplementary Fig. 18–83.

The concentration of glyceraldehyde and d-glycerate was determined by HPLC equipped with an Organic Acid Analysis Column (300 × 7.8 mm, Aminex HPX-87H). The column temperature was set to 65°C, with an injection volume of 20 μL and a detection wavelength of 214 nm. The mobile phase was 5 mM sulfuric acid at a flow rate of 0.6 mL/min.

The concentration of NAD⁺, NADH, ATP, and ADP were detected through HPLC equipped with a C18 column (4.6 × 150 mm, Wondasil, Shimadzu-GL). The column temperature was set to 30°C, with an injection volume of 10 μL and a detection wavelength of 214 nm. Eluent A was 10 mM ammonium acetate (0.8% triethylamine, V/V, 0.4% phosphoric acid, V/V) and Eluent B was 100% methanol. The flow rate was set at 0.6 mL/min with a ratio of 95:5 (v/v) for Eluent A and Eluent B, respectively.

The concentration of S-allyl-l-cysteine and S-phenyl-l-cysteine were detected through HPLC equipped with a C18 column (4.6 × 150 mm, Wondasil, Shimadzu-GL). The column temperature was set to 30°C, with an injection volume of 10 μL and a detection wavelength of 250 nm (S-allyl-l-cysteine) and 210 nm (S-phenyl-l-cysteine). Eluent A was 1% sulfuric acid and Eluent B was 100% methanol. The flow rate was set at 0.8 mL/min with a ratio of 95:5 (v/v) for Eluent A and Eluent B, respectively.

The concentration of OPS was determined by HPLC analysis after DNFB (2,4-dinitrofluorobenzene) derivatization. The derivatization reaction was carried out in 100 mM sodium borate buffer (pH 10.0), with the reaction solution consisting of 500 μL buffer, 200 μL sample, and 50 μL 100 mg/L DNFB and incubated at 70°C for 60 min. The column temperature was set to 33°C, with an injection volume of 10 μL and a detection wavelength of 360 nm. Eluent A was triethylammonium phosphate buffer (pH 3, 0.05 M)/acetonitrile, 74:26, v/v. Eluent B was acetonitrile. The flow rate was 0.8 mL/min with gradient elution conditions set as follows: 10 min 100% A, 4 min 16% A, and 5 min 100% A.

Chiral analysis of all ncAAs was performed following derivatization with o-phthalaldehyde (OPA) and N-acetyl-l-cysteine (NAC), with minor modifications based on the method described by Cheng et al.^73,74. Specifically, 100 µL of standard solutions or samples were mixed with 200 µL of the derivatizing reagent (10 mg/mL OPA and 10 mg/mL NAC in methanol) and 650 µL of sodium tetraborate buffer (100 mM, pH 10.0). The reaction mixture was incubated at 37°C for 2 min. Chromatographic separation was carried out using a column temperature of 40°C, an injection volume of 10 µL, and UV detection at 330 nm. The mobile phase consisted of (A) 50 mM sodium acetate buffer (pH5.6) and (B) a mixture of methanol and 50 mM sodium acetate, delivered at a flow rate of 1.0 mL/min. The A:B ratio varied depending on the compound and is detailed in the Supporting Information. Results are shown in Supplementary Fig. 84–105. GraphPad Prism 9 software was used to fit the data.

Statistics & Reproducibility

No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. All experiments were repeated at least three times to ensure reproducibility.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

S. X., Y. J., Zhi-min Li, and Zong-lin Li designed the project. S. X., S. W., and L. L. performed research. S. X. and Zong-lin Li contributed to data analysis. F. C., Zhi-min Li, and Zong-lin Li acquired funding to develop the project. Y. J., F. C., and Zhi-min Li supervised the study. S. X., U. S., F. C., and Zong-lin Li wrote the paper.

Peer review

Peer review information

Nature Communications thanks Tao Liu, Yi-Xin Huo and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data supporting the findings of this study are available within the paper and its supplementary information files. The source data generated in this study are available in the Mendeley Data repository under accession code 10.17632/g8fmhxw8c6.1. The data are publicly accessible without restriction. All enzyme amino acid sequences are presented in Supplementary Data 1, and all primer sequences are provided in Supplementary Data 2. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-63341-1.