Diverse oxidosqualene cyclases for triterpenoid biosynthesis

Abstract

Triterpenoids, a significant group of secondary metabolites, possess diverse skeletons and excellent pharmacological activities and are widely applied in cosmetics, medicine, and biopesticides. Traditional plant extraction and chemical synthesis cannot meet the needs of commercial production, and heterologous biosynthesis of triterpenoids has become a promising strategy. Oxidosqualene cyclases (OSCs) are the key enzymes that determine triterpenoid biosynthesis. However, most OSCs exhibit low heterologous expression, low catalytic activity, and promiscuous catalytic products, which impedes the heterologous biosynthesis of triterpenoids. In this study, we constructed a dataset of 348 OSCs with known functions, which is the largest centralized resource of characterized OSCs to date. The 88 triterpenoid skeletons produced by these OSCs are classified based on their structural features, 49 of which have been experimentally characterized as the main products of OSCs, and the formation mechanisms of the common skeletons are summarized. Phylogenetic analysis and motif analysis elucidated the relationships among the sequences, structures, and products of plant-derived OSCs. Phylogenetic analysis showed that plant-derived OSCs were divided into two major clades, which corresponded to the biosynthetic mechanisms of the "protosteryl/dammarenyl" dichotomy. Motif analysis showed that 5 motifs are present in plant-derived OSCs, including 3 related to the catalytic activity of the OSCs and 2 related to product specificity. Finally, we summarized the common strategies for mining OSCs to provide directions for the discovery of novel OSCs and new triterpenoid skeletons. Our study will help discover or design OSCs with specific product profiles as functional elements for applications in synthetic biology.

1. Introduction

Terpenoids represent one of the largest, most chemically, structurally and functionally diverse class of natural products derived from living organisms. QYResearch's new survey estimated that the global terpenoids market size is expected to reach USD 1047.6 million by 2029. According to the number of isoprene units, terpenoids can be grouped into monoterpenoids (C10), sesquiterpenoids (C15), diterpenoids (C20), triterpenoids (C30), tetraterpenoids (C40) and polyterpenoids (>C40) [1]. More than 80,000 terpenoids have been found to date [2], and there are triterpenoids with over 20,000 different structures identified in nature [3], displaying up to 200 different carbon skeletons [4]. Due to their structural diversity and excellent pharmacological activities, some triterpenoids have been developed into clinical drugs, skincare products or biopesticides (Fig. 1) [5]. For instance, QS-21, a natural triterpenoid saponin from the tree bark of Quillaja saponaria, is a potent vaccine adjuvant and remains the only natural triterpene clinically approved for clinical use in humans [6]. Asiatic acid, a natural pentacyclic triterpenoid, serves as principal bioactive constituent of the medicinal plant Centella asiatica and exerts wound-healing, anti-inflammatory, and antioxidant properties, which has been widely applied in skin care products [7]. Azadirachtin A is a tetranortriterpenoid derived from neem tree seeds and exhibits excellent insecticidal bioactivities against almost 600 insect pests, which is one of the prominent biopesticides commercialized [8]. However, the application potential of most triterpenoids is not fully exploited due to their insufficient purity and quantity. At present, the primary sources of triterpenoids mainly rely on plant extraction and chemical synthesis. These two production methods have many limitations for scaling up. Natural extraction method fails to meet the market demand because of the low content levels of triterpenoids in their native plants. The chemical synthesis of triterpenoids is challenging due to their structural complexities. With the development of metabolic engineering and synthetic biology, microbial synthesis is an attractive alternative for the production of triterpenoids.

Fig. 1.

Summary of triterpenoids applications.

Elucidating the biosynthetic pathways of triterpenoids is key to achieving their efficient production in microbial systems. Even though triterpenoids are chemically diverse, all biologically formed triterpenoids are derived from the isoprenoid biosynthetic pathway and can be separated into four main segments: the formation of 5-carbon building blocks, the formation of precursors 2,3-oxidosqualene formation, the synthesis of triterpenoid carbon skeletons, and the post-modification of triterpenoid scaffolds (Fig. 2) [9]. First, two universal building blocks of all terpenoids, isopentenyl pyrophosphate (IPP) and its isomer dimethylallyl pyrophosphate (DMAPP), are synthesized either through the mevalonate pathway (MVA) or the methylerythritol phosphate pathway (MEP) pathway. Subsequently, IPP and DMAPP are converted into linear C30 compound 2,3-oxidosqualene via multi-enzyme cascade reactions. Then, 2,3-oxidosqualene is cyclized by members of oxidosqualene cyclases (OSCs) to form various triterpene skeletons, such as β-amyrin, lupeol, lanosterol, etc. Finally, specific triterpene backbone is further decorated by other groups of enzymes such as cytochrome P450 monooxygenases (P450s) for oxygenation, glycosyltransferases for glycosylation or acyltransferases for acylation. With the decoding of biosynthetic pathways, there are some triterpenoids have been successfully synthesized in microbial cell factories, including asiatic acid [10], ursolic acid [11], etc. Especially, dammarane-type triterpenoids have been increased to the 10 g per liter scale [12]. However, the current microbial synthesis efficiency for other triterpenoids, for example, lupane-type, oleanane-type, and ursane-type triterpenoids, remains remarkably below industrial levels and there are certain limitations for large-scale production.

Fig. 2.

Schematic of triterpenoids biosynthetic pathways. Abbreviations not explained in the main text: GAP, glyceraldehyde-3-phosphate; DXS, 1-deoxy-d-xylulose-5-phosphate synthase; DXP, 1-deoxy-d-xylulose 5-phosphate; DXR, 1-deoxy-d-xylulose-5-phosphate reductoisomerase; MCT, 2-cmethyl-d-erythritol-4-phosphate cytidylyltransferase; CDP-ME, 4-(cytidine-5′-diphospho)-2-C-methyl-d-erythritol; CMK, 4-diphosphocytidyl-2-C-methyl-d-erythritol kinase; CDP-ME2P, 2-phospho-4-(cytidine-5′-diphospho)-2-C-methyl-d-erythritol; MDS, 2-C-methyl-d-erythritol-2,4-cyclodiphosphate synthase; MEcPP, 2-C-methyl-d-erythritol-2,4-cyclodiphosphate; HDS, 4-hydroxy-3-methylbut-2-enyl-diphosphate synthase; HMBPP, 1-hydroxy-2-methyl-2-butenyl-4-diphosphate; HDR, 4-hydroxy-3-methylbut-2-enyl diphosphate reductase; IDI, isopentenyl diphosphate isomerase; AACT, acetoacetyl-CoA thiolase; HMGS, 3-hydroxy-3-methyl glutaryl coenzyme A synthase; HMG-CoA, 3-hydroxy-3-methylglutaryl-CoA; HMGR, 3-hydroxy-3-methylglutaryl-CoA reductase; MK, mevalonate kinase; MVP, 5-phosphomevalonate; PMK, phosphomevalonate kinase; MVPP, 5-diphosphomevalonate; MVD, mevalonate 5-diphosphate decarboxylase; GPS, geranyl pyrophosphate synthase; GPP, geranyl pyrophosphate; FPS, farnesyl pyrophosphate synthase; FPP, farnesyl pyrophosphate; SQS, squalene synthase; SQE, squalene epoxidase; CYP450, cytochrome P450 monooxygenases; UGT, UDP-glycosyltransferase.

OSCs is a key enzyme in triterpenoid biosynthesis and converts 2,3-oxidosqualene into a variety of multi-ring scaffolds, which contribute to the diversity of natural triterpenoids. To date, more than 300 functionally OSCs have been cloned, purified, and biochemically characterized from many species [13]. Many OSCs have been widely expressed in microorganisms for heterologous production of triterpenoids. However, the yield of triterpenoids varies depending on OSCs obtained from different origins. For example, β-amyrin synthase from Glycyrrhiza glabra yielded more β-amyrin than other species [14]. Therefore, screening new OSCs isozymes from diverse sources will a potent strategy for improving triterpenoids production. Additionally, some OSCs can generate the products of different conformation, which impacts the yield of targeted products and causes a complicated purification process. Due to the lack of crystal structures of OSCs, it is difficult to improve product specificity through protein engineering. Machine learning and computational simulation contribute to predicting a highly accurate three-dimensional protein structure based on amino acid sequence and efficiently guiding the rational designing of OSCs. However, there is lack of a comprehensive and accurate functional database of OSCs for reference, which affects the model accuracy and leads to false prediction.

In this review, we compiled a dataset of 348 characterized OSCs from public databases and the literature. These OSCs generated 88 triterpenoid skeletons, which are categorized into 5 groups according to the number of rings in the skeletons. The triterpenoids with 1-3 rings are less than tetra- and pentacyclic triterpenes. In addition, we also reviewed the catalytic mechanisms of OSCs in catalyzing 2,3-oxidosqualene to some typical triterpenoid skeletons. Based on structure-guided phylogeny, we explained the correlation between plant-derived OSCs sequences and their catalytic products, which provides the possibility for the functional prediction of OSCs. The motifs affecting plant-derived OSCs catalytic activity and product specificity were also analyzed. Finally, we summarized the prevalent strategies for discovering OSCs genes with new triterpenoid skeletons. Overall, this study provides a consolidated overview of the sequence and functional research of OSCs and the discovery of novel OSCs.

2. Database of characterized OSCs

With the rapid development of high throughput sequencing machines, more and more OSCs have been identified from a variety of species in the past decades, which can synthesize up to 200 triterpene skeletons [4]. However, little is known about the relationship between protein structures and product diversity of OSCs. Therefore, we manually collated OSCs with biochemically characterized products in previous studies, obtaining a comprehensive set of functional OSCs. The dataset contains functional information of 348 OSCs, including gene name, amino acid sequences, GenBank, major and minor products, species origins and literatures, which is the most comprehensive and largest functional database currently reported (Supplementary File S1). These OSCs were derived from 124 plant species, 23 fungal species, 7 animal species, 5 sea cucumber species, and 2 algae species. According to the product profiles, OSCs can be divided into two groups. 230 OSCs are monofunctional synthases that can convert 2,3-oxidosqualene into a single triterpenoid product, while 118 OSCs are multifunctional enzymes that can synthesize more than one product. Especially, the enzyme baruol synthase (BARS1) from Arabidopsis thaliana produce baruol (90%) and 22 minor products (0.02−3% each) [15]. The product promiscuity is closely related to the protein structure and catalytic mechanism of OSCs.

3. Diverse structures of triterpenoid skeletons catalyzed by OSCs

Triterpenoids are synthesized through modifications of their basic skeletons. A total of 88 triterpenoid basic skeletons were identified by examining the triterpenoid compounds generated by 348 OSCs. These skeletons are roughly divided into 5 groups: 2 monocyclic, 7 bicyclic, 10 tricyclic, 36 tetracyclic, and 33 pentacyclic triterpenoid skeletons, 49 of which can be catalytically formed as the main product by OSCs (Fig. 3). Furthermore, 12 seco-triterpenoids with open-ring features were found, including seco-α-amyrin, camelliol B, camelliol A, etc. Significant differences in reaction pathways and mechanisms of various OSCs reveal the complexity and diversity of triterpenoid biosynthesis. The reaction mechanism of the OSCs catalyzing 2,3-oxidesqualene mainly includes substrate binding and folding, epoxide protonation, hydride transfer and methyl rearrangement reactions, and finally deprotonation or water capture [16,17]. In the entire catalytic process, the prefolding step of the initial substrate is crucial. Depending on the carbocation intermediates produced during the initial substrate prefolding phase, triterpene cyclization can be roughly divided into two pathways: the “chair-boat-chair” (C-B-C) conformation protosteryl cation pathway and the "chair-chair-chair" (C-C-C) conformation dammarenyl cation pathway (Fig. 4). Most tetracyclic sterols, such as lanosterol and cycloartenol, are typically generated via the protosteryl cation pathway. In contrast, the pentacyclic triterpenoid skeletons such as β-amyrin, lupeol, and tirucalla-7,24-dien-3β-ol, as well as dammarane-type tetracyclic triterpenoid skeletons, are formed via the dammarenyl cation pathway. Unlike the protosteryl/dammarenyl dichotomy catalytic pathway described above, a new cyclization pathway has recently been discovered, which produced a new pentacyclic triterpenoid, orysatinol. It possesses an unprecedented chair-(semi-chair)-chair-chair envelope (C-sC-C-C) conformation, which can switch between the "C-C-C″ conformation and the “C-B-C” conformation [18]. Orysatinol originates from a completely new cyclization pathway, that is, it comes from a tricyclic C-13 cation (Fig. 4a). This cation is formed by a bicyclic C8 cation with the "C-B″ conformation attacking the C-13 and C-14 double bonds of the side chain from below.

Fig. 3.

The triterpenoid basic skeletons produced by the characterized OSCs. Grey, blue, green, purple and cyan correspond to monocyclic, bicyclic, tricyclic, tetracyclic and pentacyclic triterpenes, respectively. An asterisk indicates the main product of the OSCs. # indicates the open-ring triterpenoids produced by OSCs.

Fig. 4.

Different carbocation intermediates and products of 2,3-oxidosqualene cyclization by OSCs. a) The “C-B-C” conformation protosteryl cation pathway: The protosteryl cation drives the formation of the protostane-type triterpene skeleton through the “C-B-C” conformation. b) and c) the “C-C-C” conformation dammarenyl cation pathway. b) The dammarane cation (C-C-C) drives the formation of the dammarane-type tetracyclic triterpene skeleton through the “C-C-C” conformation. c) The dammarane cation (C-C-C) further undergoes ring expansion to form the dammarane-type pentacyclic triterpene skeleton. The blue route represents the cyclization pathway of orysatinol, which originates from the bicyclic C-8 cation (C-B) and has an unprecedented “C-sC-C-C” conformation. The red, purple, and green routes, respectively, show the various relationships between the protostane-type tetracyclic, the dammarane-type tetracyclic, the dammarane-type pentacyclic triterpene skeletons and the carbocation intermediates. Solid arrows indicate a single step reaction, while dashed arrows indicate multiple step reactions.

In addition to using 2,3-oxidesqualene, some OSCs can also accept 22,23; 2,3-dioxidosqualene (DOS) as a substrate. In the dataset, there are 16 OSCs that can accommodate DOS, generating 9 oxygen-containing heterocyclic triterpenoid skeletons. Among them, 4 contain 22,23-epoxy ring, while 5 contain 5- or 6-membered oxygen-containing heterocycles formed by the participation of the 22,23-epoxy ring in cyclization (Fig. 5). Particularly notable is α-onocerin. It is a symmetrical tetracyclic triterpenoid that can be generated as a single product by DOS through a one-step enzymatic reaction, or through a two-step enzymatic reaction [19].

Fig. 5.

Different carbocation intermediates and products of DOS cyclization by OSCs. The green box diagram shows triterpenoids with a 22,23-epoxy ring and their cyclization pathways, while the pink box diagram shows triterpenoids with a 22,23-epoxy ring and their cyclization pathways.

4. Phylogenetic analysis of OSCs

Plant-derived OSCs (308 in total) are the most numerous in the database and exhibit a diverse range of catalytic products. In order to explore the relationship between the protein sequences and catalytic products, we performed a thorough phylogenetic analysis of these OSCs. Plant-derived OSC amino acid sequences were aligned using ClustalW with default parameters as implemented in the program MEGA version 11 [20], and all positions containing gaps and missing data were excluded. The maximum likelihood phylogenetic tree was constructed using IQTree version 3.1.2 [21]. The best-fit amino acid substitution model was automatically selected by ModelFinder [22] according to the Bayesian Information Criterion (BIC). Branch support was assessed using ultrafast bootstrap approximation [23] with 1000 replicates, together with the SH-aLRT test (1000 replicates), using the PtOSC from Phaeodactylum tricornutum CCAP 1055/1 as the outgroup. The final tree was visualized in ChiPlot (https://www.chiplot.online/) [24].

The phylogenetic analysis revealed that the OSCs' protein sequences were clustered into two obviously distinct clades: Clade I and Clade II, and the OSCs within one specific clade have the same or similar cyclization pathway (Fig. 6). The multiple OSCs in Clade I, including cycloartenol synthase and cucurbitadienol synthase, which form several protostane-type triterpenoid skeletons by the "C-B-C″ conformation protosteryl cation pathway. The OSCs in Clade II, including β-amyrin synthase and lupeol synthase, form several dammarane-type triterpenoid skeletons by the "C-C-C″ dammarenyl cation pathway. The evolution of OSCs is mainly driven by whole-genome duplication and tandem gene duplication [25]. It is generally believed that plant OSCs involved in triterpene biosynthesis are directly or indirectly derived from the ancient cycloartenol synthase gene [26]. Further phylogenetic analysis revealed that OSCs from monocotyledonous and dicotyledonous plants form separate major clusters within Clade I and Clade II, confirming three key events of gene duplication: D1, D2, and D3 (Fig. 6). The ancient duplication D1 occurred before the divergence of monocotyledons and dicotyledons, representing the ancient gene duplication of the ancestral cycloartenol synthase (ACS) and the ancestral lanosterol synthase-like (ALSL) OSCs. D2 and D3 represent the divergence of dicotyledonous and monocotyledonous OSCs, which is consistent with previous studies [26]. Clearly, the multiple duplications of the ACS gene promoted the expansion of OSC genes in monocotyledonous plants, while the multiple duplications of the ALSL gene facilitated the massive proliferation of OSC genes in dicotyledonous plants. Therefore, it can be inferred that the ability of OSCs to cyclize 2,3-oxidosqualene into derivatives with the “C-C”conformation may originate from the ancestral OSC that produces derivatives with the “C-B” conformation, and the change in substrate folding mode is the basis of this major functional differentiation event.

Fig. 6.

Phylogenetic analysis of the characterized OSCs from plants. The purple branch indicates that the OSC can produce the main product derived from the bicyclic cation with the “C-B” conformation, while the cyan branch indicates that the OSC can produce the main product derived from the bicyclic cation with the "C-C″ conformation. The green area represents OSCs from algae, gymnosperm, and ferns; the pink area represents OSCs from monocotyledons; and the blue area represents OSCs from dicotyledons. Black boxes denote proposed duplication events during plant OSCs evolution.

To better understand the evolutionary relationships of the OSCs in each clade, we performed phylogenetic analyses separately. Clade I consists of 79 OSCs from 53 species, including 31 OSCs from monocotyledonous plants, and 48 OSCs from dicotyledonous plants (Fig. 7).

Fig. 7.

Phylogenetic analysis of OSCs in clade Ⅰ. Triangle represents single function OSCs, while star represents multiple function OSCs. The blue branch indicates that the OSC can produce the main product derived from the bicyclic cation with the "C-C″ conformation, while the red branch indicates that the OSC can also produce the main product derived from the bicyclic cation with the “C-C” conformation.

Phylogenetic analysis shows that the OSCs in clade I are divided into 2 groups: Group I and Group II. Group Ⅰ is mainly composed of 48 OSCs that are derived from dicotyledonous plants, including 39 cycloartenol synthases, 8 cucurbitadienol synthases, and 1 lanosterol synthases. These OSCs in this group are almost all monofunctional enzymes, and their main products are all developed from bicyclic cations with a “C-B” conformation. Group Ⅱ consists of 31 OSCs that are derived from monocotyledonous plants. Of these, 11 can produce products with a “C-C” conformation, whereas 20 can generate products with a "C-B″ conformation. Almost all OSCs in monocotyledonous plants that can produce products with a “C-C” conformation are clustered in clade 1. This is different from the OSCs in dicotyledonous plants, where almost all OSCs that can produce “C-C” configuration products are located in clade II. This indicates that the OSCs in monocotyledonous plants are more similar to sterol synthases [27]. Notably, the multifunctional orysatinol synthase, parkeol synthase, and polypodatetraenol synthase are OsOSC7 genes from different rice subspecies. The differences in their catalytic products and the different conformations of the cations during the formation of the various triterpenoid structures indicate a high degree of difference between the OSC7 variants [28].

Clade II is a large dicotyledonous plant clade, consisting of 221 OSCs from 91 dicotyledonous plants and 2 OSCs from 2 monocotyledonous plants. More than half of the OSCs are promiscuous, capable of generating multiple triterpene skeletons, such as dammarane, lupane, usane, etc. Moreover, the main products of nearly all OSCs are derived from the bicyclic cation with the “C-C” conformation, and over 80% OSCs can catalyze the 2,3-oxidosqualene to produce a pentacyclic triterpene skeleton (Fig. 8).

Fig. 8.

Phylogenetic analysis of OSCs in clade Ⅱ. Triangle represents single function OSCs, while star represents multiple function OSCs.The blue, yellow and green branches respectively indicate Group Ⅰ, Group Ⅱ and Group Ⅲ.

Phylogenetic analysis shows that clade II is divided into 3 groups: group I, group II, and group III. Group I has three main branches (Branch A, Branch B and Branch C) and contains 70 OSCs. Branch A consists of 11 OSCs, most of which are responsible for catalyzing the 2,3-oxidosqualene to produce lanosterol. Branch B contains 29 OSCs, mainly OSCs with the function of synthesizing lupeol, among which 17 OSCs have been identified as single-function enzymes that can produce lupeol as the product, and 2 OSCs can produce lupeol as the main product. Branch C contains 30 OSCs, among which over 70% OSCs have the main products being taraxaterane-type and ursane-type triterpenoids, such as α-amyrin, baurelenol, and φ-taraxasterol. Group Ⅱ also consists of three branches (Branch a, b, and c), containing 53 OSCs. Branch a contain single-function β-amyrin synthases and lupeol synthase, as well as multifunctional OSCs that produce α-/β-amyrin and other mixed products. Branch b contains 19 OSCs, including 15 friedelin synthases and 4 isomultiflorenol synthases. Friedelin can produce friedelin as the main product, which is the precursor of celastrol, a promising bioactive compound with many encouraging preclinical applications [29]. Branch c contains 16 OSCs, among them, the OSCs capable of generating tetracyclic triterpene skeletons with the “C-C-C” conformation, such as triucallane, euphane, and baccharane, are the most numerous. Group Ⅲ contains 83 OSCs, among which over 90% of the OSCs can catalyze the 2,3-oxidosqualene to produce β-amyrin. 58 of these OSCs can produce β-amyrin as the main product, while 17 OSCs can produce β-amyrin as a minor product. Interestingly, OSCs that produce lupeol as a single product are mainly distributed in group Ⅰ, while OSCs that produce multiple products such as lupeol, α-amyrin, and β-amyrin are more scatteredly distributed in group Ⅱ and group Ⅲ. These two groups form two different lupeol synthase branches, which is consistent with previous studies [30]. In addition, within the entire clade Ⅱ, a total of 144 OSCs from 68 species can produce β-amyrin, which is the largest class of OSCs genes discovered so far, and also indicates that β-amyrin dominates the products of the five-ring triterpenoids catalyzed by OSCs.

5. Motif analysis of OSCs

OSCs catalyze one of the most complex ring-forming reactions in nature. Many characterized OSCs are multifunctional enzymes that can produce a series of cyclization products. Despite the structural diversity of the products, the protein sequences of OSCs exhibit significant conservation, where a few key amino acids could govern the product selectivity. For instance, MdOSC4 and MdOSC5 are two OSCs derived from Malus × Domestica, exhibiting up to 93% similarity in their amino acid sequences, yet they synthesize distinct triterpenes [31]. To explore the relationship between the key residues and active sites of OSCs and their catalytic products, we used the TBtools to predict the conserved motifs of 308 OSCs from plants (Fig. 9). We used ClustalW with the default parameters in MEGA version 11 [20] to align the amino acid sequences of plant-derived OSCs, manually extracting conserved motifs from each sequence, and comparing known OSC motifs across sequences (Supplementary File S2).

Fig. 9.

Conserved motifs of plant-derived OSC proteins. The numbers 1–10 are followed by the corresponding amino acid sequences of the conserved motifs.

The analysis revealed that the most conserved motif in OSCs is the specific amino acid repeat sequence, "QXXXXXW", which exists in multiple places in the squalene-hopene cyclase and OSCs sequences (motif 3, 6, and 10). This repeat sequence is a negatively charged aromatic amino acid region located on the enzyme surface, which plays a role in stabilizing or enhancing the structure and function of the protein during the 2,3-oxidosqualene cyclization reaction [32]. Significantly, nearly half of the OSCs exhibit a “DGGWG” motif (motif 3) formed by residues D624-W629 (human lanosterol synthase numbering hereafter) within the “QXXXXXW” motif at positions Q623-W629 and the conserved glycine downstream of this sequence [33]. This motif corresponds to the δ domain specific to OSCs with βγ structure, and this is the main determinant of the activity of triterpenoid synthases with the δ domain [33,34]. Another highly conserved motif is the "DCTAE" (motif 5) that is responsible for substrate binding and initiating the polycyclization of 2,3-oxidosqualene [35], where the D455 residue is an essential residue for electrophilic cyclization [36] and is present in all characterized OSCs.

In addition to the above motifs, we also analyzed the conserved motifs that affect product specificity. Most β-amyrin/cycloartenol synthases possess a conserved "VFM/VFN" triad motif (motif 4). This triad motif is conserved as "VFM" in 67 out of 85 β-amyrin synthases and conserved as "VFN" in 47 out of 52 cycloartenol synthases. Available functional evidence suggests this motif contributes to the enzyme activity. Mutants possessing this triad motif have higher yields of β-amyrin/cycloartenol compared to the wild type lacking this motif, mainly due to the stabilizing effect of these three amino acids on the substrate intermediate, with the cation-π interaction of Phe playing a major role [37], though these mechanistic insights currently rely on a limited number of studies. MXCXCR (motif 1) is also a conserved region that determines the specificity of the product, and it mainly stabilizes the tetracyclic or pentacyclic intermediates through cation-π interactions [13]. In tetracyclic OSCs, the conserved region is dominated by "MWCHCR" in the cyclasterol synthase and cucurbitadienol synthase, which has been observed in 50 cycloartenol synthases and 8 cucurbitadienol synthases. In pentacyclic OSCs, the conserved region is dominated by "MWCYCR" in β-amyrin synthase [38], covering 66 out of 85 β-amyrin synthases; the conserved region is dominated by "MXCYCR" in α-amyrin synthase [39], covering 17 out of 25 α-amyrin synthases; and the conserved region is dominated by "MLCYCR" in lupeol [38], covering 19 out of 33 lupeol synthases.

6. The discovery of OSCs

To date, more than 200 triterpene skeletons have been reported in nature [4]. Only 88 different triterpenoid skeletons were produced by 348 OSCs, indicating that there are still many OSCs with new functions that remain to be discovered. Biological sequences contain vast amounts of valuable information, but our exploration of them is only the tip of the iceberg. With the rapid development of sequencing technology, a large amount of genomic and transcriptomic data has been published. This has given rise to the establishment and improvement of various types of databases, such as NCBI (National Center for Biotechnology Information, https://static.pubmed.gov/portal/portal.fcgi/), UniprotKB (https://www.uniprot.org/uniprotkb), and InterPro (https://www.ebi.ac.uk/interpro/), which provide a basis for the mining of OSCs. Here, we summarize various triterpenoid synthases mining methods to improve understanding of these methods and discover triterpenoid synthases with new functions (Fig. 10).

Fig. 10.

Mining methods for triterpenoid synthases. a) Direct cloning. b) Multi-omics combined analysis. c) Algorithm-assisted analysis.

6.1. Direct cloning

The essence of OSCs mining lies in the analysis of gene or protein sequences. This method mainly involves comparing the sequences of characterized OSCs to obtain conserved motifs, such as "QXXXXXW" and "DCTAE". Based on this, primers are designed and employed in nested PCR to amplify or screen the cDNA libraries in order to obtain the core fragment. Then the full-length sequences of candidate OSCs genes are obtained through RACE (Rapid Amplification of cDNA Ends) technology. Finally, the candidate OSCs' functions are determined using experiments (Fig. 10a). Many OSCs have been successfully cloned and characterized through this method, such as PNX and PNY from Panax [40], β-amyrin synthase BgbAS and lupeol synthase BgLUS from Bruguiera gymnorrhiza, and multifunctional germanicol synthase RsM1 and multifunctional taraxerol synthase RsM2 from Rhizophora stylosa Griff [41]. This method is straightforward and convenient, allowing the identification of novel candidate variants with high similarity to known enzyme sequences. However, it is highly dependent on these conserved motifs. When the homology between the target OSC and the known sequence is low, these motifs may have been changed. This can lead to amplification failure, causing the candidate sequences to be omitted during the amplification stage.

6.2. Multi-omics combined analysis

The synthesis of plant secondary metabolites such as terpenoids, flavonoids, and alkaloids [42] not only shows species specificity but also varies across different tissues and developmental stages within the single plant, and is easily influenced by biotic and abiotic factors [43]. Consequently, metabolomics combined with other omics technologies can be used to collect data on plant metabolite synthesis in various conditions (e.g., tissue, developmental stage). Subsequently, by correlating metabolite changes with gene and protein expression, OSCs can be mined (Fig. 10b). Srivastava et al. [44] identified the metabolite fingerprints of Terminalia arjuna leaves and stem bark, confirming that the accumulation levels of oleanane triterpene/saponin differented in the two tissues. Based on this, they sequenced the transcriptomes of different tissues and analyzed the differentially expressed transcripts in leaf and bark tissues to elucidate functional genes closely related to triterpene synthesis. Finally, four OSCs were mined and identified in Terminalia arjuna, among which TaOSC1, TaOSC3, and TaOSC4 were characterized as β-amyrin synthase, cycloartenol synthase, and lupeol synthase, respectively. TaOSC2 was found to be a multifunctional OSC that simultaneously produces α-amyrin and β-amyrin. Sun et al. [45] used metabolomics to analyze the specific distribution of metabolites in different tissues of Aesculus chinensis, and they discovered that aescin was highly accumulated in seeds. Subsequently, they obtained the genomic information of Aesculus chinensis through Illumina and ONT sequencing and Hi-C assembly. Then, they analyzed the biosynthetic gene clusters (BGCs) and discovered 4 BGCs containing OSCs. Finally, AcOSC6 was found to be highly expressed in seeds, and the experiment characterized it as β-amyrin synthase.

The methods can not only narrow down or determine the genes associated with triterpene biosynthesis from the intersection of these omics data but also provide a more comprehensive understanding of the triterpene biosynthesis pathway. The results of different omics analyses can also be mutually verified, enhancing their reliability [46]. However, this method has limitations such as massive and complex datasets, high analytical difficulty, and significant cost, and it is frequently only applicable to mining OSC genes in a specific plant.

6.3. Algorithm-assisted analysis

With the rapid development of sequencing technology, a large amount of sequence data has been published, which provides an opportunity to directly mine OSCs in the database. However, accurately identifying OSCs with catalytic functions in the vast amount of data presents a major challenge. Here, we introduce the current methods used for mining OSCs in databases (Fig. 10c).

Basic Local Alignment Search Tool (BLAST) and Hidden Markov Models (HMM) are two classic sequence analysis algorithms, and they are the most commonly used in the OSCs mining. OSC sequences with the same function generally have a high similarity, whereas OSC sequences with different functions generally have a low similarity. To obtain OSCs with the same or similar functions, known functional OSCs are often used as probes to search in public databases or genomic data, which enables mining with the same or similar carbon skeletons. This method is a good choice for mining rare skeletons of triterpenoids and related OSCs. For example, arabinediol has an uncommon malabaricane-type 6,6,5-tricyclic triterpenoid skeleton [47]. Using arabinediol synthase AtPEN1 from Arabidopsis thaliana as the query sequence, the arabinediol synthase AmAs in Astragalus membranaceus was identified in the public database [48]. Protostane-type terpenoid skeletons are rarely found in nature, and the enzymes responsible for their biosynthesis are rarely reported. Using the unique protostane synthase AfPDS from Aspergillus fumigatus as the probe, a unique plant-derived protostadienol cyclase (AoPDS) was discovered in the public Alisma orientale transcriptome data [49]. Due to the high conservation of OSCs, more OSC sequences can also be selected as the query sequence for mining OSCs [50,51].

The HMM algorithm mainly searches based on the conserved sequence regions in the sequence, such as protein domains. The C-terminal domain of squalene-hopene cyclase (PF13243) and the N-terminal domain of squalene-hopene cyclase (PF13249) are two typical domains of OSCs. Using the HMM of these two domains as probes to mine OSCs is a common method [52]. Huang et al. [53] used the HMM of these two domains as probes to search the Perilla frutescens genome data and discovered 9 candidate OSC genes. Among them, 4 were characterized as OSCs with catalytic functions. Alternatively, multiple identified OSC protein sequences also can be used to create an HMM, which can be used as a probe to mine OSCs. Tang [54] constructed an HMM profile of Pyr4 and its homologues that is an atypical transmembrane terpene cyclase and does not have conserved domains. Using the HMM as probes, 1050 BGCs containing Pyr4 homologous proteins were extracted from 1990 genomes in the NCBI database. Then, the research focused on BGCs with lower similarity to known fungal BGCs, encoding Pyr4-like proteins, and different from polyketide synthases or PaxC-like pentenyltransferases of the core enzyme. Finally, 16 potential BGCs were found, containing one or two proteins homologous to squalene-hopene cyclases or OSCs. Three of these BGCs were selected for experimental verification, resulting in a series of novel onoceroid triterpene products. This algorithm can be used to identify distantly related sequences, avoiding the bias toward related species and invalid matches in non-conserved regions that occur during single sequence searches. It allows for more precise and comprehensive identification of homologous proteins.

Sequence Similarity Network (SSN) have gradually been applied in the field of enzyme mining. SSN analysis is a multidimensional network that uses sequence similarity to cluster related proteins and display their relationships [55]. By linking the OSCs with known and unknown functions, it is possible to quickly determine novel or potential OSCs genes. Jakob Franke team [56] constructed an SSN containing 170 OSCs with known functions and 1891 candidate OSC sequences derived from the 1000 Plant Transcriptome database. By optimizing the alignment score threshold, the SSN can accurately reflect the functional differences of OSCs. And 6 noncanonical OSC candidate genes were screened out, among which 3 can catalyze the production of rare or novel triterpene stereoisomers, including (3S,13S)-malabarica-17,21-diene-3β,14-diol, 19-epi-lupeol, and protostahopenol. These findings greatly expand the stereochemical diversity of triterpenoids and lay the foundation for the development of new triterpenoid drugs. One of the main advantages of SSN is its ability to quickly visualize relationships among thousands of protein sequences. However, SSN clusters by autonomously adjusting the alignment score threshold, which may be subjective and introduce certain deviations during the adjustment of the network.

Currently, many genomic annotation analysis tools are developed based on sequence homology, which results in approximately 40% of enzymes being incorrectly annotated as having equivalent functions to known enzymes or as hypothetical proteins or domains with unknown functions [57]. To break through this limitation, OSC mining methods based on machine learning have been gradually developed and applied. By using advanced algorithms to learn patterns and relationships from diverse data sources, including sequence information, structural properties, functional annotations, etc., the accuracy of enzyme function predictions has been gradually improved [58]. CLEAN (contrastive learning–enabled enzyme annotation) is an enzyme annotation tool. When tested on the Price-149 dataset, its performance was significantly better than BLASTp. It can fully annotate understudied enzymes, correct mislabeled enzymes, and identify promiscuous enzymes with two or more EC numbers [59]. Cucurbitacin biosynthesis is primarily regulated by the bitter (Bi) gene, which encodes an OSC capable of catalyzing the 2,3-oxidosqualene to produce cucurbitadienol, the precursor to cucurbitacin. Xu et al. [60] used CLEAN to screen Bi genes across 21 plant genomes in the database and discovered 19 Bi genes that encode cucurbitadiene synthase. Among these, As03G2784 from the Aquilaria sinensis was experimentally characterized to confirm its function as cucurbitadiene synthase. However, it is important to note that "performance superior to BLASTp" does not mean that it can accurately identify every OSC's function. Different triterpenoids can be produced by highly similar OSC sequences, and it is challenging to use sequence comparison learning representations alone to catch the few crucial residue differences that determine the products' specificity. The CLEAN method is effective in predicting cucurbitadienol OSCs, likely because the training set has relatively sufficient examples of cucurbitadienol synthases and the sequence-function mapping relationship is clearer. However, for other OSCs with diverse products or limited training data, such as β-amyrin synthase, the model may fail due to sample imbalance and blurred functional boundaries, misclassifying sequences similar but product-different OSCs as the OSCs having the same function. In addition, although machine learning has improved the accuracy of OSC function prediction, it is highly dependent on data and requires a large amount of high-quality experimentally validated data as support. Currently, the OSC experimental data is limited, and the model is prone to overconfidence or generalization bias on enzymes with little information.

Sequence homology is not completely equivalent to its function. Even highly similar sequences may have different functions, and a single amino acid change may also cause a change in enzyme function. Therefore, only relying on sequence similarity does not accurately mine enzymes with similar functions. It is generally believed that the 3D structure of an enzyme is directly related to its function [58].

In 2020, AlphaFold2 achieved a landmark result in predicting protein structures from primary sequences, and is now able to accurately predict the 3D structures of many proteins [61]. As a result, triterpenoid synthase can be mined through homology modeling and molecular docking. In 2022, Professor Liu's team [62] first discovered the chimeric class I triterpenoid synthase TvTS and MpMS from filamentous fungi through genome mining of public databases. They contain the prenyl transferase (PT) and terpene cyclase (TC) domains. The TC domain catalyzes the cyclization of HexPP to form a new triterpene skeleton. Through a thorough analysis of the terpene synthase domain of TvTS and the hexameric structure of MpMS, and combined with the mutations of key amino acids, a precise mining strategy based on AlphaFold and substrate molecule docking was established: using AlphaFold2 to batch predict the potential triterpenoid synthase protein structures, combined with substrate molecule docking. Finally, two triterpenoid synthases (CgCS and PTTC074) were successfully predicted and screened, and they can both synthesize the new triterpene colleterpenol. This method further highlights the feasibility and potential of AI-assisted deep mining of novel triterpenoid compounds.

7. Summary and prospect

In recent years, triterpenoids have attracted increasing attention owing to their structural diversity and numerous pharmacological activities. With the development of synthetic biology and bioinformatics, the biosynthetic pathways of triterpenoids have gradually become clear. As key enzymes responsible for the synthesis of various triterpene skeletons, a comprehensive understanding of OSCs is conducive to clarifying the biosynthetic mechanism of triterpenoids, promoting the discovery of triterpene skeletons, and establishing a basis for the efficient heterologous synthesis of triterpenoids. Herein, we manually collected and curated 348 characterized OSCs from 161 species. This database contains information such as species origin, catalytic products, and protein sequences. Our study will help discover or design OSCs with specific product profiles as functional elements for applications in synthetic biology.

In general, the classification of OSCs protein sequences using phylogenetic trees corresponds to the classification of their product formation mechanisms. However, there are still some differences. In clade II, many multifunctional pentacyclic triterpenoid synthases from different groups, despite having different product generation ratios, all have similar product profiles and can produce pentacyclic triterpenes such as β-amyrin, α-amyrin, taraxasterol, and lupeol. This is mainly because the reaction process catalyzed by OSCs to convert 2,3-oxidosqualene into pentacyclic triterpenes involves many complex cyclization and rearrangement steps, including the longest known (9-step) cation shuttle run (CSR) cascade rearrangement reaction. During this process, early termination of CSR can generate some pentacyclic triterpene skeletons, such as β-amyrin, isomultiflorenol, and lupeol [63]. While many OSC sequences are highly conserved, subtle protein sequence differences can also affect the product profile of OSCs. Extensive mutagenesis studies have shown that changes in a single residue can change the function of OSCs [16,49,64]. In this study, we have summarized the common conserved motifs in OSCs. Many motifs are not as well-conserved or as strict as commonly thought. For instance, ItOSC2 from Iris tectorum is a rare multifunctional α-amyrin synthase that does not have the common "MLCYCR" motif associated with α-amyrin synthases, but instead has the "FLALAR" motif [65]. This also suggests that the relationship between the OSC sequence and its function, as well as the catalytic mechanism, still needs further research. Homology modeling is a powerful tool for predicting the structure of target proteins, especially suitable for OSCs with highly homologous crystal structures [66]. The quantum mechanical/molecular mechanical (QM/MM) molecular dynamics (MD) simulations technique not only detects the reaction process of the substrate in the active pocket during the calculation of substrate-enzyme binding, but also supplements the influence of the enzyme environment. It is a commonly used tool for studying the catalytic mechanism of OSCs [67].

It is reported that there are over 200 triterpene skeletons existing in nature, while the OSCs in the database have only generated 88 triterpene skeletons, and only 49 of these skeletons can be catalytically produced as OSCs' main products. Moreover, the majority of OSCs in the database share common four-ring and five-ring triterpenoid skeleton catalytic functions, and only the remaining 12.4% produce various rare triterpenoids in a selective manner. Therefore, how to discover OSCs with new catalytic products, superior activity, and selectivity is a current hot issue. Currently, the mining and modification of OSCs has shifted from traditional methods to artificial intelligence (AI)-assisted approaches like machine learning and deep learning. Although AI has certain advantages in handling large datasets and predicting protein structures, it still cannot accurately predict protein functions. This necessitates experimental characterization of a large number of sequences in the process of mining OSCs and obtaining superior active OSCs through the "mutation-screening" strategy. The emergence of automated high-throughput screening and detection technologies has solved these problems. Through automated high-throughput equipment for massive trial-and-error, the "design-build-test-learn" closed loop can be completed more quickly, cost-effectively, and in multiple cycles, achieving efficient and automated mining of natural products [68]. Several automated high-throughput platforms have been established both domestically and internationally [69,70]. Automated high-throughput techniques have been applied in the discovery of multiple natural products, such as terpenoids [71,72]. Currently, the mainstream strategy underlying the aforementioned high-throughput screening platform still relies on heterologous expression of candidate OSC genes in microbial hosts (such as Saccharomyces cerevisiae). However, when enzymes from different sources are heterologously expressed in host cells, there may be problems such as different codon preferences, inability to fold, low stability, lack of cofactor and substrate, etc., which can result in the identification of only high-abundance products [73]. The plant transient expression system (e.g., Agrobacterium-mediated transient transformation of Nicotiana benthamiana leaves) can offer a protein folding and post-translational modification environment that is closer to the native plant cellular context and can effectively avoid problems such as codon bias commonly encountered in microbial heterologous expression [18]. The system has successfully identified the catalytic functions of several plant OSCs, including the multifunctional α-amyrin synthase MdOSC1 from Malus × domestica [74] and the α-onocerin synthase OsONS1 from Ononis spinosa [19]. It is a powerful tool for uncovering the true catalytic properties of plant OSCs and discovering low-abundance products.

Data availability statement

All functional information of database are available within Supplementary File S1.

The conserved base sequences of the plant-derived OSCs obtained through our dataset have been placed in the Supplementary File S2.

CRediT authorship contribution statement

Qiannan Hao: Writing – original draft. Shengli Wang: Writing – review & editing. Lei Zhang: Investigation. Lin Yuan: Supervision. Guangrong Zhao: Methodology. Jianjun Qiao: Supervision, Project administration, Funding acquisition. Weiguo Li: Funding acquisition. Qinggele Caiyin: Project administration.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: