Reprogramming product selectivity and activity in promiscuous terpene synthases via substrate conformation engineering

Abstract

Terpenoids are structurally diverse natural products with broad applications, yet their biosynthesis is often constrained by the low catalytic efficiency and poor product selectivity of terpene synthases (TPSs). Simultaneously enhancing product specificity and catalytic performance remains a major challenge in TPS engineering. Here, we elucidate the regio- and stereoselectivity of a fungal germacrene A synthase, which catalyzes the 1,10-cyclization of farnesyl diphosphate to form a central sesquiterpene intermediate. The side-chain volume of a non-catalytic residue (T169) adjacent to the substrate pocket modulates substrate-binding conformations, thereby redirecting the carbocation cascade and reshaping the product profile. Mechanism-guided pocket engineering further yielded terpene synthase variants with altered product specificity. Notably, one variant increased (+)-germacrene A production in Escherichia coli by 24.3-fold, reaching 15.8 g L⁻¹ in a bioreactor—the highest titer reported in this host to the best of our knowledge. These results provide mechanistic insight into substrate-conformation-mediated reprogramming of TPS specificity, highlighting a promising strategy for the development of high-performance terpenoid biocatalysts.

Simultaneously enhancing product specificity and catalytic performance remains a major challenge in terpene synthase engineering. Here, the authors elucidate the regio- and stereoselectivity of a fungal germacrene A synthase, which catalyses the 1,10-cyclization of farnesyl diphosphate to form a central sesquiterpene intermediate, and demonstrate that engineering substrate conformation enables simultaneous improvement of product specificity and catalytic performance.

Introduction

Terpenoids, comprising over 100,000 known compounds (dnp.chemnetbase.com), represent the largest and most structurally diverse family of natural products. Due to their diverse biological activities, many terpenoids hold significant commercial value in the pharmaceutical and fragrance industries, with notable examples including artemisinin, Taxol, and limonene^1–3. Leveraging microbial cell factories for terpenoid synthesis is a pivotal strategy for attaining sustainable production of these valuable compounds^4–8.

Despite the structural complexity of the terpenoid library, these compounds are biosynthesized from relatively simple linear, achiral C_5n isoprenoid diphosphate (Fig. 1a)⁹. Terpene synthases (TPS) catalyze the formation of terpenoid skeletons and are often referred to as terpene cyclases, reflecting the prevalence of cyclization reactions in their development. The reactions catalyzed by TPSs feature the carbocation-driven cascade cyclization, including attacks by olefins, proton and hydride transfers, methyl and methylene migrations, and final quenching via deprotonation or hydration^10–13. Throughout the multistep cyclization cascade of terpenoids, more than half of the substrate carbon atoms typically undergo changes in bonding, hybridization, and stereochemistry^10,14. As a result, the processes catalyzed by TPSs are among the most complex reactions found in nature.

Fig. 1. Biosynthetic pathway of terpenoid skeletons.

a Schematic representation of the biosynthetic pathways for terpenoid skeletons. MVA, mevalonate pathway; MEP, methylerythritol phosphate pathway; IDI, isopentenyl diphosphate isomerase; TPS, terpenoid synthases; GAP, glyceraldehyde-3-phosphate. b Chemical reaction processes involving the stereoisomers of germacrene A. GAS refers to germacrene A synthase.

TPSs are a naturally promiscuous class of enzymes capable of synthesizing a variety of secondary products from a single substrate, which facilitates the host’s adaptive responses to environmental stimuli^3,15. This product promiscuity presents challenges for targeted terpenoid biosynthesis, but it also provides valuable opportunities for the directed evolution of these enzymes in the laboratory^16–19. TPSs have become a focal point in enzyme engineering research, with numerous successful modifications aimed at altering the target product spectrum^20–24. For instance, site-specific mutations that modulate protonation and deprotonation events enable fine-tuning of carbocation cascade trajectories, thereby reshaping cyclization pathways and diversifying terpene scaffolds^16,25. Similarly, altering the participation of water molecules in cascade termination can influence the final carbon skeletons of the products²⁶. Another notable example is the engineering of monoterpene synthase HCinS, where mechanistic insights into its reaction cascade facilitated single-residue substitutions that redirected multiple carbocation intermediates into distinct end products²⁷. Beyond such active-site modifications, even residues located distally from the catalytic pocket can influence the TPS product outcome, and targeted changes at these relatively inert positions have proven effective in reprogramming enzyme specificity^28,29. This context-dependent plasticity is further compounded by the inherent conformational flexibility of substrate, which underlies their remarkable product promiscuity ³⁰. However, direct experimental evidence connecting substrate-pocket alterations to substrate-binding conformations in TPSs remains lacking. Additionally, efforts to reprogram TPS product profiles are often accompanied by diminished catalytic efficiency^27,31,32, making the concurrent enhancement of product selectivity and enzymatic activity a significant challenge.

β-Elemene is a non-cytotoxic, broad-spectrum Class II anticancer agent that is widely applied in clinical settings^33–36. However, the TPS directly responsible for its biosynthesis remains unidentified. Current bioproduction strategies instead rely on germacrene A as a key intermediate, which can thermally rearrange to β-elemene via a Cope rearrangement reaction³⁷. Germacrene A, generated via a 1,10-cyclization of farnesyl diphosphate (FPP), is a central precursor in sesquiterpene biosynthesis^38–40. Germacrene A synthase (GAS) catalyzes this cyclization, establishing a chiral center and yielding two stereoisomers. Only (+)-germacrene A can rearrange into the biologically active (–)-β-elemene enantiomer, which is the form used in clinical formulations such as elemene oral emulsions (Fig. 1b)^41,42. However, no high-resolution experimental structure of GAS is currently available, and the mechanistic basis for its regio- and stereoselectivity remains poorly understood. Consequently, the identification and engineering of highly efficient GAS variants constitute a key bottleneck in the microbial production of (+)-germacrene A, the essential precursor for (–)-β-elemene.

In this study, a fungal (+)-germacrene A synthase was used as a model system to elucidate the mechanisms underlying the regio- and stereoselectivity of GAS. During the engineering of this enzyme, we identified a non-catalytic residue that can reshape the product profile. Structural analysis combined with molecular dynamics simulations revealed that the side-chain volume at this position modulates the substrate-binding conformation. To our knowledge, our work provides direct experimental and mechanistic evidence that alterations in the substrate pocket can control substrate-binding conformations, ultimately determining TPS product specificity. Mechanism-guided pocket engineering further yielded variants with not only improved product specificity but also markedly increased catalytic efficiency, overcoming the long-standing trade-off between activity and specificity. Notably, a mutant exhibiting a 24.3-fold improvement over the wild-type enzyme was obtained, enabling (+)-germacrene A production to reach 15.8 g L⁻¹ in Escherichia coli—the highest reported in this host to the best of our knowledge. Our findings provide mechanistic insight into substrate-conformation-mediated reprogramming of TPS specificity, highlighting a promising strategy for developing high-performance terpenoid biocatalysts.

Results

Identification of the products generated by FpGAS

To investigate the role of substrate-binding conformational flexibility in the reprogramming of promiscuous TPSs, we selected FpGAS from Fusarium poae—a representative member of the promiscuous clade III defined by Bian et al.—as the model enzyme for this study (Supplementary Fig. 1)⁴³. Amino acid sequence analysis revealed that FpGAS showed the highest sequence identity (79.6%) with FgGS, a promiscuous enzyme exhibiting both di- and sesterterpene synthase activities^44,45. However, results from both in vivo and in vitro assays indicate that FpGAS lacks diterpene synthase activity (Supplementary Fig. 2), suggesting that it is unable to handle larger terpenoid substrates.

To determine the terpene synthase activity of FpGAS, we integrated the gene into an E. coli chassis harboring a preconstructed MVA pathway⁴⁶, and subsequently characterized the products generated by FpGAS. The gas chromatography-mass spectrometry (GC-MS) analysis indicated that FpGAS can produce a diverse range of sesquiterpene products, with β-elemene (1) being the predominant product, accounting for 38.1% of the total (Fig. 2a, b and Supplementary Fig. 3). The stereochemical features of this product were further confirmed on a chiral column, identifying it as (–)-β-elemene (Supplementary Fig. 4). Furthermore, NMR analysis of the fermentation products under ambient conditions revealed that (+)-germacrene A is the major product (Supplementary Fig. 5)⁴⁷, indicating that the (–)-β-elemene detected by GC originates from a thermal Cope rearrangement of (+)-germacrene A. The acyclic sesquiterpenes (E)-nerolidol (2) and β-farnesene (3) also represented a significant portion at 21.1% and 17.8%, respectively. The proportions of other products were all below 10%, including elemol (4) at 7.5%, α-farnesene (5) at 3.5%, γ-elemene (6) at 2.3%, shyobunol (7) at 1.0%, 5,10-di-epi-shyobunol (8) at 0.5%, and germacrene D-4-ol (9) at 0.2% (Fig. 2a, b and Supplementary Figs. 6–13). Additionally, several unidentified peaks accounted for 8.1% of the total product composition (a–e) (Fig. 2a, b).

Fig. 2. GC-MS analysis of products generated by FpGAS.

a GC-MS profile of products produced by FpGAS. The strain harboring the vector lacking FpGAS was used as the control. Labels are consistent across the entire manuscript. b Relative proportions of sesquiterpene compounds generated by FpGAS. c Proposed catalytic pathway of FpGAS. Source data are provided as a Source Data file.

By analyzing the carbon–hydrogen skeletons of products 1–9, we inferred their biosynthetic pathways. The formation of the three linear compounds 2, 3, and 5 occurred through hydration and deprotonation reactions that quenched the carbocation intermediates, without involving cyclization (Fig. 2c). The six monocyclic sesquiterpenes all underwent C1-C10 bond formation reactions. The intermediate deprotonated at C12 produced (+)-germacrene A, while deprotonation at C10 yielded germacrene B. Both of these substances subsequently underwent thermal Cope rearrangements, resulting in the formation of 1 and 6. The carbocation intermediate was quenched by hydration at C11, followed by rearrangement to produce 4. Furthermore, certain cyclization intermediates underwent a 1,3-hydride shift or two consecutive 1,2-hydride shifts, followed by hydration and intramolecular rearrangement, resulting in the formation of compounds 7, 8, and 9. In these cyclized products, cyclization occurred exclusively between C1 and C10, preserving an R configuration at the C10 position (Fig. 2c). The only exception is germacrene B, which underwent a loss of chirality as a result of deprotonation. This indicates that the cyclization reactions catalyzed by FpGAS demonstrate excellent regio- and stereoselectivity.

Structural basis for the product promiscuity

Attempts to crystallize the apo form of FpGAS were unsuccessful, as no suitable crystals were obtained for X-ray diffraction. Instead, the crystal structure of FpGAS in complex with Mg²⁺ and pyrophosphate (PPi) was resolved at a resolution of 2.5 Å. This indicates that Mg²⁺ and PPi play key roles in stabilizing a uniform conformation of FpGAS. As a class I TPS, FpGAS adopts an α-domain architecture comprising 11 core α-helices (A–K) (Fig. 3a). The two conserved motifs typical of class I TPSs—DDxxD and NSE/DTE—are present in FpGAS as DSVLE and NSE, respectively (Supplementary Fig. 14). In the DSVLE motif within helix C, residues D94 and E98 coordinate Mg²⁺ ions A and C, while Mg²⁺ ion B is coordinated by residues N209, S213, and E217 of the NSE motif located in helix H. In addition, PPi and water molecules occupy the remaining coordination sites of the metal ions (Fig. 3b). The binding of PPi is facilitated not only by Mg²⁺ coordination but also by polar interactions involving amino acids D94, E98, N209, S213, and Y301, as well as basic residues R165, R216, and R300, with interaction distances ranging from 2.6 to 3.2 Å (Fig. 3c).

Fig. 3. Structural analysis of the substrate-binding pocket of FpGAS.

a Overall structure of FpGAS. The 11 core α-helices are designated A to K. b The binding mode of Mg²⁺. c Key amino acid residues interact with pyrophosphate (PPi) in the active pocket. d The HEPES binds in the substrate-binding pocket. The 2mF_o-DF_c densities for HEPES are shown as blue mesh at the 0.8σ level. e Amino acids comprising the substrate-binding pocket of FpGAS. f Volume of the substrate-binding pocket of FpGAS. Dashed lines represent possible interactions, and distances are shown in angstroms.

A HEPES molecule from the crystallization buffer, characterized by well-defined electron densities, was identified within the substrate-binding pocket of FpGAS (Fig. 3d and Supplementary Fig. 15). The bottom of this pocket is predominantly lined with hydrophobic residues, including F67, A90, F91, W140, A171, F173, and V174, whereas the top is enriched with polar residues such as D94, D168, N209, and N291 (Fig. 3e). This spatial distribution closely mirrors the amphipathic nature of the native substrate FPP, which contains a polar head group and a hydrophobic tail¹². Notably, FpGAS possesses a substantially larger substrate-binding pocket, with a calculated volume of 647 ų to accumulate substrate FPP (Fig. 3f), compared to the calculated molecular volume of FPP of ~300 ų⁴⁸. This enlarged active pocket likely contributes to the observed product promiscuity. The larger pocket provides greater conformational flexibility, allowing the enzyme to accommodate diverse substrate orientations and carbocation intermediates, which in turn facilitates the formation of a broader range of terpenoid products³⁰.

A molecular switch residue modulating product distribution

Enzymes with high catalytic efficiency and strong product specificity are highly desirable for biosynthetic applications. To engineer biocatalysts with improved traits based on the promiscuous FpGAS, we employed a random mutagenesis strategy, using the yield of (+)-germacrene A as the selection criterion. Screening of 188 mutants led to the identification of one variant, designated 136#, which exhibited a 2.8-fold increase in (+)-germacrene A production (Fig. 4a, b and Supplementary Fig. 16). Sequencing of this mutant revealed two amino acid substitutions: I148V and T169A. To dissect their individual contributions, single-point mutants were constructed. The I148V mutation caused a slight reduction in (+)-germacrene A yield, while T169A alone resulted in an approximately 3-fold enhancement (Fig. 4b). Structural analysis of FpGAS showed that I148 is located on the E-helix, with its side chain facing the protein surface, whereas T169 resides on the F-helix, with its side chain oriented toward the substrate-binding pocket (Fig. 4c and Supplementary Fig. 17). Mutation at T169 likely alters the shape and hydrophobicity of the active site, directly impacting catalytic properties.

Fig. 4. T169 functions as a single-residue switch that controls the cyclization reaction in FpGAS.

a GC profiles of crude extracts obtained from FpGAS and its mutants. b Relative activity of FpGAS and its mutants in producing (+)-Germacrene A. Data are presented as mean ± standard deviation, from three independent biological replicates (n = 3). Individual data points are shown. c Structural analysis of the mutant residues in 136#. d GC analysis of the extracts from FpGAS and its mutants. The strain harboring the vector lacking FpGAS was used as the control. e Relative activity of FpGAS and its mutants. Source data are provided as a Source Data file.

To further elucidate the role of T169 in modulating FpGAS catalytic performance, we performed site-directed mutagenesis at this position. The T169S variant produced a similar yield of (+)-germacrene A compared to T169A; however, it showed a markedly higher proportion of 2, with an approximate 70% increase (Fig. 4e). This shift is likely attributable to the hydrophilic nature of the serine side chain, which may promote water molecule access to the active site and increase the probability of carbocation quenching via hydration. In contrast, the T169G mutant, featuring a smaller side chain, exhibited significantly enhanced catalytic performance, resulting in a 19-fold increase in (+)-germacrene A production and a 12.4-fold increase in overall enzymatic activity. Moreover, the proportion of (+)-germacrene A in the total product profile increased to 57% (Fig. 4d, e).

Enlargement of the side chain at position 169 markedly inhibits the cyclization activity of FpGAS, resulting in a predominance of non-cyclized products. In the T169L mutant, compound 3 became the major product, with a yield 10.8-fold higher than that of the wild type and accounting for 71.7% of the total products. Together, the two non-cyclized products, 2 and 3, comprised over 92.3% of the total (Fig. 4d, e). Similarly, the T169F variant functioned predominantly as a β-farnesene synthase, producing 90.5% β-farnesene (Fig. 4d, e). In contrast, increasing side chain size while altering hydrophilicity, as in the T169K mutation, led to complete loss of enzymatic activity (Fig. 4d). Previous work has established that class I TPSs harbor an R–D–G motif acting as a substrate-induced-fit element⁴⁹; in FpGAS, this corresponds to an R165–D168–T169 motif, and structural superposition confirms that the spatial configuration of this motif is highly conserved (Supplementary Fig. 18). Our findings further reveal that T169 functions not only as part of this induced-fit regulatory element but also as a pivotal residue that modulates product specificity in FpGAS.

Mechanism of product regulation by the switch residue

To investigate how the residue at position 169 acts as a single-residue switch regulating the cyclization reaction in FpGAS, we determined the structures of several FpGAS mutants. The complex structure of the T169G mutant with FPP was obtained by soaking crystals in a cryoprotectant solution containing the substrate. Well-defined electron density for FPP is clearly visible in this structure (Fig. 5a and Supplementary Fig. 15). The lack of substrate turnover is likely due to the constrained crystal lattice, which restricts the conformational flexibility required for catalysis. The binding modes of the magnesium ions and the PPi group closely resemble those seen in the wild-type protein (Supplementary Fig. 19). The hydrophobic portion of FPP is accommodated within a pocket rich in aromatic residues, such as F67, F91, F173, and W294 (Fig. 5b). This aromatic cage establishes a hydrophobic microenvironment that protects the carbocation intermediates from premature aqueous quenching, while simultaneously stabilizing them within the active site through cation–π interactions^50,51.

Fig. 5. Catalytic mechanism of FpGAS variants.

a Farnesyl diphosphate (FPP) bound within the substrate-binding pocket of the T169G variant. b The surrounding amino acid environment of the substrate-binding pocket in T169G. c Key amino acids involved in the catalytic reaction in T169G. d Mechanistic insights into the regio- and stereoselective catalysis of FPP by T169G. e β-Farnesene (BFA) bound in the substrate-binding pocket of T169F. f Residues involved in BFA binding in T169F. g Key catalytic amino acids in the T169F variant. h Superimposed structure of T169G and T169F, highlighting conformational differences of ligands. i Histograms of the C1-C10 distance in FpGAS and its mutants during MD simulations (0.09 Å bin width). j Histograms of the dihedral angle C10-C11-H10-C1 of FpGAS and its mutants during MD simulations (1° bin width). k Representative binding mode observed in MD simulations. l Residues at position 169 regulate the cyclization process in FpGAS. The 2mF_o-DF_c densities for ligands are shown as blue mesh at the 0.8σ level. Distances are indicated by dashed lines (units in Å).

The positively charged R165, with its guanidinium nitrogen atoms positioned 2.8 Å from the oxygen atom of the pyrophosphate group, stabilizes the negative charge on the pyrophosphate and facilitates its release from the substrate (Fig. 5c). The distances between C10 and C1 and between C11 and C1 are 3.5 Å and 4.0 Å, respectively. The shorter C10–C1 distance, together with Markovnikov’s rule, suggests that the C10 atom of the olefin preferentially attacks the C1 carbocation, leading to the formation of a C1–C10 cyclized product. C1 is positioned on the Si face of the C10-C11 double bond, with the dihedral angle (C10-C11-H10-C1) at −106.4^o, resulting in the introduction of a new chiral center at C10 with an absolute R configuration upon cyclization (Fig. 5d)⁵¹. Additionally, the negatively charged residue D94, located approximately 3.4 Å from C14, together with the neighboring diphosphate group, forms a locally anionic environment that is likely to facilitate deprotonation of the reaction intermediate^12,31,52, thereby contributing to the formation of the final product, (+)-germacrene A (Fig. 5c). Notably, a water molecule near C11 may also participate in quenching the carbocation, resulting in the formation of product 4 (Fig. 5c).

Attempts to obtain the FPP complex with T169F were unsuccessful; instead, we soaked the crystals with PPi and β-farnesene, which allowed us to resolve the structure of T169F bound to its major product. In this complex, clear electron densities were observed for both PPi and β-farnesene (Fig. 5e and Supplementary Fig. 15). β-Farnesene adopts an extended conformation within the hydrophobic pocket (Fig. 5f), with the distance between C1 and C10 measuring 4.6 Å, thereby inhibiting C1–C10 cyclization (Supplementary Fig. 20). The departing PPi group likely acts as a catalytic base, facilitating deprotonation of the intermediate³⁰. Additionally, residue Y300, with its phenolic hydroxyl oxygen positioned 2.5 Å from PPi and 3.3 Å from C13 of β-farnesene, may serve as a proton shuttle during intermediate deprotonation, promoting β-farnesene formation (Fig. 5g).

Superimposition of the T169G and T169F variant structures shows that side-chain size at position 169 strongly influences substrate-binding conformation. In T169F, adoption of the FPP conformation observed in the T169G–FPP complex would result in steric clashes with the bulky phenylalanine side chain, indicating that this binding mode is unlikely (Fig. 5h). Molecular dynamics (MD) simulations were performed to investigate FPP binding differences among the mutants, starting from the T169G–FPP complex, while FPP was docked into the WT and T169F variants based on the ligand positions in the original structures (Supplementary Figs. 21 and 22). During a 200 ns simulation of FpGAS bound to FPP, the most frequently observed C1–C10 distance was 4.7 Å, and the dihedral angle (C10–C11–H10–C1) peaked at −129° (Fig. 5i–k), indicating that cyclization occurs on the Si face of the double bond. In the FpGAS-T169G–FPP complex, the distance increased slightly to 4.9 Å, and the dihedral angle shifted to −120°, consistent with the observation that both enzymes generate the same set of products, albeit in different proportions. Conversely, in the FpGAS-T169F–FPP complex, the distance increased to 6.7 Å and the dihedral angle shifted to −23° (Fig. 5i–k), hindering cyclized product formation and favoring the production of linear products.

These findings suggest that the residue at position 169 in FpGAS governs product formation through a mechanism distinct from the commonly reported stabilization of carbocationic intermediates. By reshaping the substrate-binding pocket, it modulates substrate conformation and, in turn, governs the progression of the carbocation cascade³²—ultimately directing the production of distinct product profiles (Fig. 5l). Moreover, this mode of regulation does not appear to compromise catalytic activity (Fig. 4e), offering a potential strategy for simultaneously enhancing both specificity and catalytic efficiency in TPS engineering.

Rational design of the substrate pocket

Structural modifications to the substrate-binding pocket of FpGAS can alter the substrate binding mode, thereby redirecting the carbocation cascade and ultimately determining product specificity. Building on this insight, we engineered additional T169G based variants to further fine-tune its catalytic behavior. Expanding the substrate-binding pocket increases substrate flexibility, which often leads to a broader, more promiscuous product profile. For example, the V71A variant of FpGAS—characterized by a smaller side chain at position 71—produced products 1, 2, and 4 in proportions of 19%, 27%, and 34%, respectively. This relatively uniform distribution of multiple products indicates a reduction in catalytic specificity (Fig. 6a). Conversely, reducing the substrate-binding pocket size to restrict substrate conformational freedom tends to enhance product specificity. The A143L mutation, for instance, nearly abolished the formation of product 2 while increasing the major product 1 to 61%, thereby improving catalytic selectivity (Fig. 6b).

Fig. 6. Rational engineering of FpGAS variants.

a–c GC profiles of crude extracts from FpGAS variants with (a) expanded substrate-binding pockets, (b) constricted pockets, and (c) constricted pockets combined with altered hydrophilicity. The strain harboring the vector lacking FpGAS was used as the control. d–g structural analysis of the rationally selected mutation sites within the FpGAS-T169G–FPP active site: F67 (d), V71 (e), A143 (f), and Q287 (g).

Incorporation of polar amino acids near the active site can facilitate water access, increasing the likelihood of terminating the carbocation cascade at earlier stages. The Q287R mutant predominantly yielded the polar alcohol product 2, which accounted for 75% of total product output (Fig. 6c). However, many mutations near the substrate—such as F67L, A90L, and S295Y—abolished or greatly reduced enzymatic activity, indicating that these residues are important for maintaining proper active-site structure and function (Fig. 6a and Supplementary Figs. 23–25). In contrast, mutations at more distal positions—such as V71, A143, and Q287, which are located more than 6 Å away from the substrate—were more successful in altering the enzyme’s product profile (Fig. 6d–g). These findings highlight that TPS engineering should focus not only on residues directly involved in substrate binding and catalysis but also on more distal, second-shell residues that can significantly influence active site architecture and catalytic pathways.

(+)-Germacrene A fermentation in bioreactors

Shake flask fermentation using the FpGAS-T169G/A143L variant yielded 1.99 g L⁻¹ of product 1, representing a 24.3-fold increase compared to the wild-type enzyme (Fig.7a). This engineered strain was subsequently employed as a microbial cell factory for the production of (+)-germacrene A in a 5-L bioreactor, with glucose as the sole carbon source (Fig.7b). The fermentation was conducted in two distinct phases. During the initial phase, cell growth was prioritized to support downstream biosynthesis. After approximately 7 hours, the culture reached an OD₆₀₀ of ~25, at which point IPTG was added to induce enzyme expression and initiate product formation. In the production phase, dodecane was introduced to the culture to enable in situ extraction of the sesquiterpene product and alleviate product inhibition. After 60 hours of cultivation, (+)-germacrene A accumulated to 15.8 g L⁻¹ in the organic phase (Fig. 7c), representing the highest titer reported to date in E. coli. The average production rate reached 0.26 g L⁻¹ h⁻¹, also the highest known for this compound in this host. These findings underscore the potential of promiscuous terpene synthase engineering to enhance terpenoid biosynthesis in microbial systems.

Fig. 7. (+)-Germacrene A production using FpGAS variant in E. coli.

a Improvement of (+)-germacrene A production through enzyme engineering of FpGAS. Data are presented as mean ± standard deviation from three independent biological replicates (n = 3). Individual data points are shown. b Schematic illustration of the metabolic engineering strategy for enhancing precursor supply for (+)-germacrene A biosynthesis in E. coli. MvaE, acetyl-CoA acetyltransferase/HMG-CoA reductase; MvaS, HMG-CoA synthase; ERG12, mevalonate kinase; ERG8, phosphomevalonate kinase; ERG19, mevalonate diphosphate decarboxylase; ispA, farnesyl diphosphate synthase; IDI, isopentenyl diphosphate isomerase. c Representative fed-batch fermentation profile of the engineered E. coli strain expressing the optimized FpGAS variant for (+)-germacrene A production. Product concentrations were determined from three technical replicate measurements, and error bars represent the corresponding SD. Source data are provided as a Source Data file.

Discussion

Enzymes are essential biocatalysts in synthetic biology, enabling the selective synthesis of complex molecules under mild conditions^53–55. While evolution has yielded a vast array of enzymes, many native catalysts—particularly those involved in secondary metabolism—exhibit low efficiency and broad product promiscuity^56,57. Such flexibility supports metabolic diversity but limits precision in engineered systems. Mechanism-guided enzyme engineering is therefore key to unlocking their full biosynthetic potential^29,56. TPSs are a prime example, catalyzing carbocation-driven cyclization that often yields complex mixtures^58,59. Enhancing their product selectivity and catalytic activity is therefore critical for the targeted biosynthesis of high-value terpenoids.

As a focal point in enzyme engineering, TPSs have been investigated using a variety of strategies. Early efforts to engineer TPSs were hindered by limited structural information and an incomplete understanding of their catalytic mechanisms. As a result, initial strategies primarily relied on non-rational mutagenesis or comparative analysis of sequence-divergent plasticity regions among homologous enzymes to guide site-directed modifications^29,60. TPS engineering has progressively shifted toward rational design approaches informed by catalytic principles and theoretical modeling with advances in structural biology and mechanistic elucidation^61–64. Several single-point mutations at residues that stabilize carbocation intermediates have been shown to drastically alter product distributions⁶⁵. Additionally, modulating the spatial arrangement of active-site water molecules has enabled the formation of hydroxylated terpenoids²⁶. Reshaping the substrate-binding pocket represents another powerful strategy, allowing for expansion of substrate scope—such as the conversion of a monoterpene synthase into a sesquiterpene synthase⁶⁶—or facilitating catalysis of non-native substrates⁶⁷.

In this work, we demonstrate that the substrate-binding conformation serves as a key determinant of product outcome. Remarkably, substitution of a single amino acid is sufficient to remodel the substrate’s pose within the active pocket. Similar residue-level “switches” have been widely reported in TPSs, including those that mediate substrate-induced conformational changes of the enzyme⁴⁹, regulate deprotonation or reprotonation of carbocation intermediates²⁵, or dictate the trajectories of cascade cyclization reactions²⁷. In the present study, we identify a distinct type of “switch” that specifically modulates the substrate-binding conformation, revealing a previously unrecognized mechanism to control product selectivity and underscoring the exceptional plasticity of terpene synthases for rational engineering. However, such a substrate-conformation-guided approach is not universally applicable across all enzyme classes, as many enzymes require precise recognition of specific functional groups for catalysis. In contrast, TPSs catalyze carbocation-driven cascade cyclization, wherein product formation is highly sensitive to the initial substrate conformation. A similar principle has been observed in cytochrome P450 enzymes, which mediate a broad array of oxidative transformations via radical intermediates and likewise exhibit substrate-conformation-dependent reactivity within the active site^68,69.

TPSs are broadly distributed across plants, fungi, and bacteria. Phylogenetic analyses indicate that plant- and microbe-derived TPSs form well-separated clades^70,71. Investigations of class I TPSs have revealed species-specific preferences in how the substrate hydrocarbon chain engages with the two diphosphate oxygen atoms projecting into the active-site pocket⁷², and the FpGAS complex structure presented here corroborates this preference. Such divergence is likely to influence substrate orientation and the ensuing carbocation cascade, providing a mechanistic basis for the frequent occurrence of enantiomeric products or intermediates between plant and bacterial TPSs. Previous studies have likewise documented distinct product stereo-chemistries for canonical plant and microbial GASs, supporting functional differentiation of TPSs along evolutionary lineages⁵¹. In contrast, FpGAS represents an exception: despite its fungal origin, it produces the (+)-germacrene A. This deviation may arise from its relatively spacious active site and the fact that it is not a highly product-specific GAS. Furthermore, the pronounced conformational flexibility of TPS substrates, the exceptional reactivity of carbocation intermediates, and the multitude of possible termination pathways—each highly sensitive to subtle perturbations—collectively render the prediction of TPS product profiles and stereochemistry exceedingly challenging.

The biosynthetic pathway of the broad-spectrum antineoplastic compound (–)-β-elemene has garnered significant attention in recent years^73,74. Microbial hosts commonly employed for its production include E. coli and yeast. Metabolic engineering strategies typically involve the integration of a heterologous MVA pathway, attenuation of competing carbon fluxes⁷⁵, and overexpression of rate-limiting enzymes in precursor biosynthesis to enhance precursor availability⁷³. As the key enzyme in this pathway, GAS has been the focus of intensive efforts aimed at identifying or engineering variants with improved catalytic efficiency, representing a promising route to yield enhancement^51,76. In this study, we adopted a strategy to modulate product profiles and enzymatic activity by regulating substrate conformation. Using this approach, we identified an FpGAS mutant with a 24.3-fold increase in GAS activity relative to the wild-type enzyme. In fed-batch fermentation, this variant achieved a (+)-germacrene A titer of 15.8 g L⁻¹, surpassing the previously reported maximum titer of 7.74 g L⁻¹ in E. coli⁷⁷. While eukaryotic expression systems generally afford higher germacrene A titers, Qi Liu et al. reported a titer of 39 g L⁻¹ in Yarrowia lipolytica after 240 h of optimized bioreactor fermentation through MVA pathway rewiring and modulation of intracellular acetyl-CoA fluxes⁷⁴. Nevertheless, our engineered E. coli system exhibited a substantially shorter fermentation time and a higher production rate (0.26 g L⁻¹ h⁻¹), thus offering notable advantages over eukaryotic systems in terms of process efficiency.

In summary, our study provides direct experimental and mechanistic evidence that non-catalytic residues surrounding the TPS substrate pocket shape substrate-binding conformations, thereby directing carbocation intermediates and determining the final product. Targeting these residues to modulate substrate conformation provides a strategy for controlling product outcome and guiding the rational engineering of terpene synthases.

Methods

Plasmids, strains and culture medium

Candidate germacrene synthase gene, FpGAS, was retrieved from NCBI and synthesized by BGI Genomics (China). FpGAS gene was cloned into pET28a, while MVA pathway genes were cloned into pACYCDuet-1 or pTrcHis2B plasmids, with the resulting plasmids designated pE, pA, and pT, respectively. (Supplementary Table 1). E. coli BL21 (DE3) served as the host for protein expression and terpenoid production. Strains were typically grown in LB medium (10 g L⁻¹ tryptone, 5 g L⁻¹ yeast extract, 10 g L⁻¹ NaCl). The fermentation medium contained 20 g L⁻¹ glucose, 9.8 g L⁻¹ K₂HPO₄, 5 g L⁻¹ beef extract, 0.3 g L⁻¹ ferric ammonium citrate, 2.1 g L⁻¹ citric acid monohydrate, 0.06 g L⁻¹ MgSO₄, and 1 mL L^-1 trace element solution. The trace element solution comprised 0.37 g L⁻¹ (NH₄)₆Mo₇O₂₄·4H₂O, 0.29 g L⁻¹ ZnSO₄·7H₂O, 2.47 g L⁻¹ H₃BO₄, 0.25 g L⁻¹ CuSO₄·5H₂O, and 1.58 g L⁻¹ MnCl₂·4H₂O. Antibiotics—ampicillin (100 mg L⁻¹), kanamycin (50 mg L⁻¹), and chloramphenicol (34 mg L⁻¹)—were added as required.

Mutagenesis of FpGAS

To sample a broad sequence space without relying on prior structural assumptions, we employed random mutagenesis. FpGAS mutant libraries were constructed by error-prone PCR (er-PCR) using primers F-1 and R-1 (Supplementary Table 2) with a mutation rate of ~2.3 mutations per kb, as specified by the Diversify® PCR Random Mutagenesis Kit (Takara). Mutations were introduced across the full coding region without pre-selected sites, allowing both mutation positions and amino-acid substitutions to arise in an unbiased manner. PCR products were gel-purified and ligated into pET28a plasmids carrying ispA and IDI genes using the ClonExpress® II One Step Cloning Kit (Vazyme). Constructs were transformed into E. coli BL21 (DE3) along with pT and pA plasmids. Mutants were screened based on product yield. Site-directed mutagenesis was performed using primers in Supplementary Table 2 and confirmed by DNA sequencing.

Fermentation and product identification

E. coli BL21 (DE3) carrying pE, pA, and pT (plasmids listed in Table S1) was cultured in 50 mL fermentation medium at 37 °C, 200 rpm. At OD₆₀₀ 0.6–0.9, expression was induced with 0.1 mM IPTG, and 20% (v/v) decane was added for in situ terpenoid extraction. Cultures were incubated at 30 °C, 200 rpm for 48 hours. Terpenoids were extracted with decane and analyzed by GC-MS (Thermo TRACE GC ULTRA/TSQ QUANTUM XLS) using a DB-5MS column with a temperature program from 80 °C to 300 °C. Compounds with peak areas >1% were identified via the NIST 2017 library. Terpenoids were also quantified by GC-FID (SHIMADZU GC-2014) under the same conditions. Product percentages were calculated by peak integration using LabSolutions software; results represent the mean ± SD of three replicates. Chiral separation and identification of the product produced by FpGAS were performed following the method described in the referenced literature⁵¹. The analysis was conducted on an Agilent 1260 Infinity II LC system using a normal-phase CHIRALCEL OD-H column (5.0 μm, 250 mm × 4.6 mm i.d.). The column temperature was maintained at 40 °C. The mobile phase consisted of n-hexane (A, 95%) and methanol (B, 5%) at a flow rate of 0.5 mL/min. The injection volume was 1 μL, and detection was performed at 212 nm and 254 nm.

Fed-batch fermentations were conducted in a 5-L bioreactor charged with 2 L of fermentation medium. A 70% (w/v) glucose solution was continuously fed once the initial carbon source was depleted, as indicated by a sharp increase in dissolved oxygen (DO). After ~7-8 h of cultivation, when the culture reached an OD₆₀₀ of 25–30, 0.1 mM IPTG was added, 400 mL of dodecane was overlaid, and the temperature was shifted to 30 °C. Samples were withdrawn every 8–12 h to monitor cell density (OD₆₀₀), residual glucose, and β-farnesene production.

Protein expression and purification

E. coli BL21(DE3) harboring pET28a-FpGAS was cultured in LB medium with 100 μg/mL kanamycin at 37 °C, 200 rpm until OD₆₀₀ reached 0.8 (~3 h). Protein expression was induced with 0.2 mM IPTG, followed by overnight incubation at 16 °C, 100 rpm. Cells were harvested, resuspended in binding buffer (50 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM imidazole), lysed by ultrasonication, and clarified by centrifugation (15,000 × g, 30 min). The supernatant was loaded onto a Ni²⁺ Sepharose HP column. After washing with 40 mM imidazole-containing buffer, proteins were eluted with 250 mM imidazole. Further purification was performed by size-exclusion chromatography on a Superdex 75 column using buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl). Protein purity (>95%) was confirmed by SDS-PAGE, and concentration was determined by UV absorbance at 280 nm. Purified proteins were stored at –80 °C in storage buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl). Site-directed mutants were generated by overlapping PCR (primers listed in Table S2) and expressed/purified using the same protocol as the wild-type.

Crystallization, data collection, and structure determination

Purified FpGAS and variants were concentrated to ~24 mg/mL for crystallization. Initial crystals were obtained using commercial high-throughput screening kits and a Formulatrix NT8 robot with 96-well plates, then optimized in 24-well plates. Crystals used for X-ray data collection are listed in Supplementary Table 3. All crystals were cryoprotected by soaking in a well solution with 30% glycerol for 10 seconds and flash-frozen in liquid nitrogen. For substrate-bound structures, crystals were soaked in cryoprotectant containing FPP for 1 h before freezing.

X-ray diffraction data were collected at the BL10U2 or BL19U1 beamlines of the Shanghai Synchrotron Radiation Facility^78,79. Data indexing, integration, and scaling were conducted using XDS⁸⁰. The structures of FpGAS and its mutants were determined by molecular replacement using the PHENIX program suite and the PDB 6VYD as a search model⁴⁵, and their structures were built in COOT⁸¹. Refinement of the structure was performed using the programs COOT and PHENIX. All the final models were evaluated using the wwPDB Validation System⁸², and data collection and refinement statistics were shown in Supplementary Table 3. Protein cavity volumes were calculated using KVFinder-web⁸³, which employs a grid-based approach to detect cavities and computes their volumes as the sum of grid-defined voxels (ų). To estimate the volume of the substrate-binding pocket accommodating FPP, Mg²⁺ ions were included as part of the biomolecular surface during cavity detection. Grid spacing and probe sizes were set according to the default parameters recommended in KVFinder-web. All molecular graphics were created with PyMOL (http://www.pymol.org).

Molecular dynamics simulations

The MD simulations were performed with the GROMACS package⁸⁴. The initial coordinates of protein and ligands were assigned based on our reported crystal structure of FpGAS and its variants. The T169G–FPP complex served as the starting structure for MD simulations, while FPP was docked into the WT and T169F variants, guided by the ligand positions in the original structures. The MD simulations were performed using the Amber ff99SB force field⁸⁵, while water molecules were modeled using the TIP3P model⁸⁶. The ligand force field parameters were generated using the general AMBER force field (GAFF)⁸⁷, and the partial atomic charge of the substrates was defined by the restrained electrostatic potential (RESP) charge⁸⁸. Gaussian16 and Multiwfn were used for RESP charge calculation^89,90.

The preparation steps of MD included hydrogen minimization, solvent molecule minimization, and positional restraints of 5 kcal/ mol/ Ų applied to C_α atoms and the Mg²⁺ ions. The positional restraints were retained during heating to 300 K (in 100 ps using NVT ensemble) and pressure equilibration (in NPT ensemble, at 300 K and 1 bar, for 100 ps). Subsequently, the positional restraints were gradually removed in a series of 50 ps NPT ensemble simulations. Following equilibration, trajectories were generated through 200 ns MD production simulations and saved every 10 ps. MD simulations were performed in triplicate to ensure reproducibility of the results. Analyses were conducted using the last 100 ns of the equilibrated trajectories. The distances and dihedral angles were calculated using the GROMACS software package. Figures were generated using the PyMOL Molecular Graphic System (version 2.5.2).

Reporting summary

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

Author contributions

Conceptualization: S.D., X.M., H.Z.; Methodology: S.D., H.L., W.L., X.C., Q.W., L.W.; Investigation: S.D., H.L., W.L., X.C., T.C., R.Z.; Visualization: S.D., H.L., W.L., Y.H.; Supervision: X.M., H.Z; Writing—original draft: S.D.; Writing—review & editing: All authors reviewed and edited the manuscript.

Peer review

Peer review information

Nature Communications thanks Baofu Xu, Eliott Wenger and the other anonymous reviewer for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Key R&D Program of China (2025YFC3409200, and 2022YFC2106200), National Natural Science Foundation of China (grant numbers: U24A20478, 32471485, 32171203, and 32170084), Qingdao New Energy Shan-dong Laboratory grant (QNESL OP202308), and Taishan Scholars Program (tstp20250542).

Data availability

The structural data for the FpGAS complex with PPi-(Mg²⁺)₃, the FpGAS-T169G complex with FPP-(Mg²⁺)₃, and the FpGAS-T169F complex with PPi-(Mg²⁺)₃ and β-farnesene generated in this study have been deposited in the Protein Data Bank (PDB) under accession codes 9VSG, 9VSH, and 9VSI, respectively. Additional data generated or analyzed during this study have been deposited in Figshare and are publicly available at https://doi.org/10.6084/m9.figshare.29565428. All other data generated or analyzed during this study are included in this Article, the Supplementary Information, and the Source Data file, and are available from corresponding author(s) upon request. 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 https://doi.org/10.1038/s41467-026-77224-6.