Characterization and protein engineering of glycosyltransferases for the biosynthesis of diverse hepatoprotective cycloartane‐type saponins in Astragalus membranaceus

Summary

Although plant secondary metabolites are important source of new drugs, obtaining these compounds is challenging due to their high structural diversity and low abundance. The roots of Astragalus membranaceus are a popular herbal medicine worldwide. It contains a series of cycloartane‐type saponins (astragalosides) as hepatoprotective and antivirus components. However, astragalosides exhibit complex sugar substitution patterns which hindered their purification and bioactivity investigation. In this work, glycosyltransferases (GT) from A. membranaceus were studied to synthesize structurally diverse astragalosides. Three new GTs, AmGT1/5 and AmGT9, were characterized as 3‐O‐glycosyltransferase and 25‐O‐glycosyltransferase of cycloastragenol respectively. AmGT1_G146V/I variants were obtained as specific 3‐O‐xylosyltransferases by sequence alignment, molecular modelling and site‐directed mutagenesis. A combinatorial synthesis system was established using AmGT1/5/9, AmGT1_G146V/S and the reported AmGT8 and AmGT8_A394F. The system allowed the synthesis of 13 astragalosides in Astragalus root with conversion rates from 22.6% to 98.7%, covering most of the sugar‐substitution patterns for astragalosides. In addition, AmGT1 exhibited remarkable sugar donor promiscuity to use 10 different donors, and was used to synthesize three novel astragalosides and ginsenosides. Glycosylation remarkably improved the hepatoprotective and SARS‐CoV‐2 inhibition activities for triterpenoids. This is one of the first attempts to produce a series of herbal constituents via combinatorial synthesis. The results provided new biocatalytic tools for saponin biosynthesis.

Glycosylation of various bioactive astragalosides.

Introduction

Plant‐derived natural products are important source for therapeutic agents (Najmi et al., 2022; Patridge et al., 2016). As secondary metabolites, compounds in one plant usually exhibit high structural diversity with similar backbones and various substitution groups. However, obtaining these homologues are usually challenging due to their similar physicochemical properties and relatively low abundance. Usually, kilograms of plant materials are used to afford milligrams of compounds (Corsello and Garg, 2015; Guerra‐Bubb et al., 2012), which was laborious and time‐consuming. As an alternative method, biosynthesis is green and efficient to produce structurally diverse natural products (Corsello and Garg, 2015; Guerra‐Bubb et al., 2012).

Astragalosides are known as effective compounds of the herbal medicine Astragalus root (roots of Astragalus membranaceus; Ionkova et al., 2014; Su et al., 2021). They exhibit remarkable bioactivities in hepatoprotection, anti‐virus, cardiovascular protection and immunomodulatory (Chu et al., 2010; Li et al., 2018; Liang et al., 2020). Astragalosides possess a cycloartane‐type triterpenoid skeleton named cycloastragenol (CA, 1), which was modified by glucosylation at C3/C6/C25‐OH and/or xylosylation at C3‐OH (Su et al., 2021). At least 22 astragalosides have been reported from Astragalus root, containing 12 different glycosylation patterns (Table S1). They suffer from low abundance and poor UV (ultraviolet) absorption, which challenged their purification, and hindered further study on their biological activities. Therefore, it is necessary to employ biosynthetic approaches to prepare astragalosides.

Though we have identified the scaffold‐forming enzyme for astragalosides (Chen et al., 2022), the downstream steps remain unclear. Glycosylation, usually catalysed by uridine diphosphate (UDP)‐sugar‐dependent glycosyltransferases (GT), plays an important role to synthesize astragalosides (Kurze et al., 2021; Rahimi et al., 2019). Up to now, at least 67 plant triterpenoid glycosyltransferases have been reported, most of which catalyse the glycosylation of oleanane‐type and dammarane‐type triterpenoids (Table S2). For cycloartane‐type triterpenoids, only AmGT8 and its mutants reported by our group could catalyse their glycosylation (Zhang et al., 2022). Other GTs modifying the cycloartane backbones remain unknown. Meanwhile, a number of astragalosides contain xylosyl groups, while specific O‐xylosyltrasferase has rarely been reported, except for Pn3‐32‐i5 from Panax notoginseng (Wang et al., 2020), GmUGT73F4 from Glycine max (Sayama et al., 2012) and PgUGT94Q13 from Panax ginseng (Li et al., 2021). The latter two enzymes could still use UDP‐glucose (UDP‐Glc) with a lower activity. Due to the large size of the substrate binding pocket, it is difficult to modify the function of triterpenoid GTs. Usually the desired function was obtained by screening thousands of mutants (Li et al., 2020; Wei et al., 2015; Zhang et al., 2020a, 2022). Thus, the discovery and engineering of biosynthetic catalysts for astragalosides remain challenging.

In this work, we used transcriptome mining and protein engineering to obtain a series of new GTs modifying cycloartane‐type triterpenoids. AmGT5, AmGT1_G146V/I and AmGT9 were characterized as 3‐O‐glucosyltransferase, 3‐O‐xylosyltransferases and 25‐O‐glucosyltransferase of cycloastragenol respectively. These enzymes were used in combination with the 2′‐O‐glucosyltransferase AmGT8 and 6‐O‐glucosyltransferase AmGT8_A394F to produce 13 astragalosides in Astragalus root with the conversion rates ranged from 22.6% to 98.7%. They covered most of the sugar‐substitution patterns for astragalosides. This is one of the first attempts to produce a whole series of herbal constituents via combinatorial synthesis. The products were tested for inhibition activities against SARS‐CoV‐2 proteins and hepatoprotective activities against acetaminophen‐induced cell death, where several bioactive astragalosides were discovered.

Results and discussion

Chemical analysis of astragalosides in A. membranaceus

The roots of Astragalus membranaceus (Astragalus root) contain a series of cycloartane‐type saponins. Among them, 22 out of 48 possess a cycloastragenol skeleton and are named as astragalosides (Su et al., 2021). Astragalosides are the major bioactive components of Astragalus root (Ionkova et al., 2014). According to the substitution of sugar moieties, they could be divided into 12 groups (Table S1), with glucosylation at C3/C6/C25‐OH and/or xylosylation at C3‐OH. For example, astragaloside IV is one of the most abundant saponins in Astragalus root which contain a 3‐O‐xyloside and a 6‐O‐glucoside (3X,6G). In order to elucidate the structures of various astragalosides, Astragalus roots were analysed using ultra‐high performance liquid chromatography coupled with mass spectrometry (UHPLC/MS). The samples were treated with ammonia hydrolysis (13% NH₃·H₂O, 65 °C, 1 h) to degrade the downstream acylation products before analysis. In total, 14 astragalosides (2–15) were unambiguously characterized by comparing the retention time and tandem mass spectrometry (MS/MS) spectra with reference standards except for 15 (Figure 1, Figure S1–S6). They covered all 12 substitution patterns reported in Astragalus root. They also contained two extra patterns, 3GG (7) and 25G (8), which have been reported from other Astragalus plants (Barbic et al., 2010; Gulcemal et al., 2012). These results indicated that Astragalus root contains at least 14 astragalosides with different glycosylation patterns, which covered almost all natural cycloastragenol‐derived saponins identified so far. These astragalosides should be synthesized by different glycosyltransferases in A. membranaceus.

Figure 1.

Chemical analysis of astragalosides in Astragalus root. (a) Raw materials and the decoction pieces of Astragalus root. (b) LC/MS analysis of Astragalus root extracts after ammonia hydrolysis. TIC, total ion chromatogram; EIC, extracted ion chromatogram. In EIC, m/z 513.3550 ([CA + Na]⁺), m/z 645.3973 ([CA + Xyl + Na]⁺), m/z 697.4158 ([CA + Glc + HCOO]⁻), m/z 829.4581 ([CA + Glc + Xyl + HCOO]⁻), m/z 859.4687 ([CA + 2Glc + HCOO]⁻) and m/z 991.5110 ([CA + 2Glc + Xyl + HCOO]⁻) were extracted. (c) Structures of compounds 1–15. * (1–7) were identified by comparing with reference standards (Figure S1‐S6); # (8–14) were purified and identified by NMR (Appendix  S1 ); Δ (15) was identified by HR‐MS/MS (Figure S6). Glc, glucosyl group. Xyl, xylosyl group.

Candidate gene screening and functional characterization

To explore the GTs synthesizing astragalosides, candidate genes were identified from A. membranaceus transcriptome by BLASTn (basic local alignment search tool for nucleic acid, e < 10⁻⁵⁰ and ORF length > 1200 bp), using plant triterpenoid GTs (Table S2) as probes. Three GT genes were found and named as AmGT1, AmGT5 and AmGT9 (Genbank Accession Nos. ON075039, ON075040 and ON075041), together with the reported AmGT8 (Zhang et al., 2022). Phylogenetic analysis (Figure 2a) showed that AmGT1 and AmGT5 grouped together with 3‐O‐glycosyltransferases from Glycyrrhiza uralensis (Chen et al., 2019; Zhang et al., 2020a), indicating their similar function. AmGT9 clustered with 3‐O‐glycosyltransferases from Barbarea vulgaris (Erthmann et al., 2018; Rahimi et al., 2019) and 28‐O‐glycosyltransferases from Centella asiatica (Costa et al., 2017; Kim et al., 2017), which was away from legume triterpenoid GTs, suggesting a distinct function. In addition, AmGT8, reported as a 3/6/2′‐O‐glycosyltransferase from A. membranaceus, was grouped with 2′‐O‐glycosyltransferases from G. uralensis (Nomura et al., 2019) and G. max (Takagi et al., 2018) as expected. These results suggested the diversity of GTs in A. membranaceus which might contribute to the diversity of astragalosides. According to the results of real‐time quantitative polymerase chain reaction (qRT‐PCR) analysis, AmGT8 showed high expression mainly in roots, where astragalosides 2–15 were detected, indicating its potential role in the biosynthesis of these compounds. AmGT1 and AmGT9 expressed in roots, stems and leaves, while the relative expression level of AmGT5 was low (Figure S7). AmGT1/5/9 were cloned from A. membranaceus and expressed in Escherichia coli using pET28a(+) vectors. The proteins were then purified and used for functional characterization in vitro (Figure S8).

Figure 2.

Phylogenetic analysis and functional characterization of AmGT1, AmGT5 and AmGT9. (a) Phylogenetic analysis of four AmGTs with 25 previously characterized plant triterpenoid GTs (Table S2). The tree was established by MEGA6 using the Maximum Likelihood method (bootstrap = 1000) with default parameters. (b) Functions of AmGTs with cycloastragenol (CA, 1) as the substrate and UDP‐Glc and/or UDP‐Xyl as the sugar donor. (c) UHPLC/CAD chromatograms of the reaction mixtures. Products were characterized by comparing with reference standards or NMR analysis except for 16.

The biochemical characteristics of AmGT1/5/9 proteins (5 μg) were studied using CA (0.1 mM) as the substrate, and UDP‐Glc (0.5 mM) as the sugar donor. AmGT1/5/9 showed their maximum activity at pH 8.0 (50 mM Na₂HPO₄‐NaH₂PO₄) and 37 °C, and were independent of divalent cations (Figure S9). AmGT1 could catalyse CA to produce compounds 2 (conversion rate 69.2%) and 9 (30.8%) (Figure 2c). According to MS/MS analysis, 2 and 9 were identified as mono‐O‐glucoside ([M‐H + HCOOH]⁻ m/z 697.4125, [M + Na]⁺ m/z 675.4107) and di‐O‐glucoside ([M‐H + HCOOH]⁻ m/z 859.4646, [M‐H‐Glc]⁻ m/z 651.4080, [M‐H‐2Glc]⁻ m/z 489.3548) of CA respectively (Figures S2, S4). Compound 2 was characterized as CA‐3‐O‐glucoside by comparing with a reference standard (Figure S1). Compound 9 was prepared from a scaled‐up reaction and characterized by nuclear magnetic resonance (NMR) spectroscopy. The glycosylation sites were characterized at C‐3 and C‐25 according to the heteronuclear multiple bond correlation (HMBC) of C‐3/H‐1′ (δ _C 89.4/δ _H 5.02) and C‐25/H‐1″ (δ _C 79.0/δ _H 5.08) (Appendix S1). Similarly, the products of AmGT5 and AmGT9 were CA‐3‐O‐glucoside (2, 65.6%) and CA‐25‐O‐glucoside (8, 84.1%) respectively (Figure 2c). These results suggested that AmGT1/5 and AmGT9 are 3‐O‐glycosyltransferase and 25‐O‐glycosyltransferase of cycloastragenol respectively. They are the only GTs modifying cycloartane‐type triterpenoids except for AmGT8, and AmGT9 is the first GT modifying the side chain of cycloartane‐type triterpenoids.

Since several astragalosides contain O‐xylosyl groups at C‐3, we then studied the function of AmGTs using alternative sugar donors. When UDP‐xylose (UDP‐Xyl) was used, the product of AmGT1/5 was CA‐3‐O‐xyloside (3, 93.3% for AmGT1 and 63.4% for AmGT5), while the product of AmGT9 was tentatively identified as CA‐25‐O‐xyloside (16, 71.1%) (Figure S6). When a mixed sugar donor (UDP‐Glc:UDP‐Xyl:CA = 5:5:1) was used, AmGT1 could catalyse the formation of mono‐O‐glucoside 2, mono‐O‐xyloside 3, di‐O‐glycosides 9 (CA‐3‐O‐glucoside‐25‐O‐glucoside) and 11 (CA‐3‐O‐xyloside‐25‐O‐glucoside) at the same time, while AmGT5/9 could not produce any xyloside (Figure 2c). Compounds 2 and 3 were identified by comparison with reference standards (Figure S2), while 8/9/11 were identified by NMR spectroscopic analysis (Appendix S1). These results suggested that AmGT5 is a specific 3‐O‐glucosyltransferase, while AmGT1 could use both UDP‐Xyl and UDP‐Glc as the sugar donor with a similar selectivity, which is rare for triterpenoid GTs (Table S2).

Protein engineering of AmGT1 using site‐directed mutagenesis

Efficient production of 3‐O‐xylosyl astragalosides in vitro requires highly specific 3‐O‐xylosyltransferase. AmGT1 catalysed CA to produce 3‐O‐xyloside (conversion rate ~30%) and 3‐O‐glucoside (~50%) using a mixed sugar donor (Figure 2c), which is not satisfying for combinatorial biosynthesis. Thus, site‐directed mutagenesis was carried out to improve the sugar donor specificity for AmGT1. First, we aligned the protein sequence of 53 triterpene GTs to discover the key residues. Among them, 43 GTs are specific to UDP‐Glc, 5 GTs could tolerate other sugar donors besides UDP‐Glc, and 5 GTs are specific to donors other than UDP‐Glc (Figure S10). We found one residue, #146, which is conserved in GTs using UDP‐Glc as T/P/G, and distinct in GTs utilizing other donors (A/I/S) (Figure 3a). To predict the function of #146, molecular docking of AmGT1 with UDP‐Glc was conducted, using an AmGT1 structure generated by Alphafold2 (Tunyasuvunakool et al., 2021). The result showed that G146 was located close to the C6‐OH of UDP‐Glc (Figure 4a). Interestingly, the residue corresponding to G146 in AmGT1 is T145 in GgCGT, which has been reported to form a hydrogen bond with the C6‐OH of the glucosyl group (Zhang et al., 2020b). Since the difference between UDP‐Xyl and UDP‐Glc lies in C6‐OH, it was speculated that #146 may affect the selectivity of UDP‐Xyl/UDP‐Glc for AmGT1.

Figure 3.

Protein engineering of AmGT1 using site‐directed mutagenesis. (a) Protein sequence alignment of 53 triterpene GTs. GTs without * were specific to UDP‐Glc; GTs with * could tolerate other sugar donors besides UDP‐Glc (Figure S10). (b) Functional analysis of AmGT1 variants with CA as the substrate, while UDP‐Glc and/or UDP‐Xyl as the sugar donor. WT, wild type. 3G, CA‐3‐O‐glucoside. 3X, CA‐3‐O‐xyloside. 3X,25G, CA‐3‐O‐xyloside‐25‐O‐glucoside. 3G,25G, CA‐3,25‐di‐O‐glucoside. (c) Kinetic parameters of AmGT1 and its mutants. ND, product under the limit of quantitation.

Figure 4.

Molecular docking and molecular dynamics to explain the mechanism of function for #146. (a) Molecular docking of AmGT1, AmGT1_G146V and AmGT1_G146I with UDP‐Glc or UDP‐Xyl. (b) Time evolution of root mean square deviation (RMSD) of the protein backbone, UDP‐sugar donor and substrate in molecular dynamics simulations, coloured in blue, green and red lines respectively. The MM/GBSA binding free energies of substrate (CA) with AmGT1/AmGT1_G146V and UDP‐Glc/UDP‐Xyl were shown in each subgraph. (c) Time evolution of the distance between C3‐OH of CA and C1 of UDP‐Xyl/Glc in AmGT1 and AmGT1_G146V. The distance was illustrated in the right panel.

To explore its function, G146 was mutated into 16 other residues with different properties in AmGT1. Mutants G146V and G146I lost their activity to UDP‐Glc, but retained high conversion rate using UDP‐Xyl (Figure 3b, Figure S11). These variants became specific xylosyltransferases which are rare in nature (Table S2). In kinetic analysis, AmGT1_G146V showed higher affinity (K _m = 3.28 μM) with UDP‐Xyl than wild‐type AmGT1 (35.56 μM) when CA was used as the substrate (Figure 3c, Figure S12). More importantly, AmGT1_WT used UDP‐Xyl and UDP‐Glc with a similar selectivity (k _cat/K _m = 5.62E‐03 vs. 5.59E‐03), while AmGT1_G146V could only use UDP‐Xyl (k _cat/K _m = 6.10E‐03 vs. undetectable for UDP‐Glc). AmGT1_G146S/C/P/A showed high conversion rates using both UDP‐Glc and UDP‐Xyl, but UDP‐Xyl was preferred when a mixture (UDP‐Glc:UDP‐Xyl = 1:1) was used (Figure 3b). For example, G146S preferentially utilized UDP‐Xyl (conversion rate 72.2%) and retained weak UDP‐Glc activity (19.8%). The ratio of xylosides for G146C/P/A was 79.9%/66.3%/52.4%, respectively, which was higher than the wild‐type (35.3%). The activity of G146N/Q/M decreased, while the selectivity to sugar donors was unchanged. To sum up, residues with shorter side chain (V/I/S/C/P/A) enhanced the specificity to UDP‐Xyl and reduced the activity to UDP‐Glc, while the longer side chain (F/E/L/H/D/K/Y) reduced the activity to both donors. In addition, it is unexpected that mutation of G146 significantly eliminate its glycosylation activity towards C25‐OH (Figure S11). These results suggested that engineering of G146 could improve both UDP‐Xyl selectivity and regio‐selectivity of AmGT1.

To further explain the mechanism of function for residue #146, we simulated the interaction between AmGT1 and UDP‐sugar donors. According to molecular docking, #146 is close to the C6 of UDP‐Glc and C5 of UDP‐Xyl (Figure 4a). AmGT1 and its variants G146V/I could all use UDP‐Xyl since the reaction pocket could well accommodate the sugar donor, with a distance of 5.35 Å, 4.26 Å and 4.87 Å between C5 and G/V/I respectively. For UDP‐Glc, AmGT1_WT could accommodate its C6‐OH due to the smaller size of Gly, with a distance of 3.87 Å. When Gly was mutated to Val or Ile, steric bulk from long side chains (distance 1.69 Å/1.90 Å between C6 and V/I) hindered the appropriate binding of UDP‐Glc. In addition, G146K/Y completely lost their activities towards UDP‐Xyl due to their even longer side chains. The 100 ns molecular dynamics simulations showed that UDP‐Glc is more stable than UDP‐Xyl in the wild‐type AmGT1, while the substrate CA is more stable in AmGT1_G146V with UDP‐Xyl than that with UDP‐Glc (Figure 4b). We also calculated the binding‐free energies of CA with the enzymes using the molecular mechanics generalized Born surface area (MM/GBSA) method (Kollman et al., 2000). The binding free energies of CA with AmGT1_WT were −61.1 ± 6.4 and −56.6 ± 6.5 kcal/mol, in the presence of UDP‐Glc or UDP‐Xyl respectively. For AmGT1_G146V, the values were −48.6 ± 6.7 or −59.2 ± 6.5 kcal/mol when UDP‐Glc or UDP‐Xyl was used (given in Figure 4b). These results also supported that AmGT1_WT/AmGT1_G146V favoured for UDP‐Glc/UDP‐Xyl. In addition, the C3‐OH is closer to UDP‐Xyl (~4 Å) than UDP‐Glc (~5 Å) in AmGT1_G146V, leading to a larger probability for the xylosylation reaction (Figure 4c). These results indicated that steric bulk and sugar acceptor/donor binding state both contribute to the altered sugar donor specificity for #146 variants.

Combinatorial biosynthesis of astragalosides using AmGTs

Several AmGTs and their mutants could catalyse the glycosylation of CA. For combinatorial synthesis, it is necessary to elucidate the order of glycosylation at different sites. Therefore, the function of AmGTs was investigated using various CA glycosides as substrates, including 2–6, 8 and 13 (Figure 5a).

Figure 5.

Substrate promiscuity of AmGTs and the combinatorial synthesis of astragalosides. (a) Structures of substrates. (b) Conversion rates for different substrates catalysed by AmGTs. AmGT was abbreviated as GT, and AmGT8_A394F was abbreviated as GT8M. 3G/6G/25G/2′G means C3/C6/C25/C2′‐O‐glucoside. (c) Functional characterization of AmGT1 (GT1) and its variants with compounds 4, 8 and 13 as substrates and UDP‐Xyl as the sugar donor. 3X, C3‐O‐xyloside. (d) Combinatorial synthesis network of astragalosides 2–14 using AmGTs. The tetracyclic skeleton represents cycloastragenol (CA, 1). Glc, glucosyl group. Xyl, xylosyl group. (e) UHPLC/MS chromatograms for combinatorial biosynthetic reaction samples. Extracted ion chromatograms was shown with extraction of m/z 667.4052 ([CA + Xyl + HCOO]⁻), m/z 697.4158 ([CA + Glc + HCOO]⁻), m/z 829.4581 ([CA + Glc + Xyl + HCOO]⁻), m/z 859.4687 ([CA + 2Glc + HCOO]⁻) and m/z 991.5110 ([CA + 2Glc + Xyl + HCOO]⁻). Conversion rates noted were calculated using UHPLC/CAD (Figure S13). AmGT was abbreviated as GT. Peaks with red labels indicated target products in each reaction.

For compounds 2 (CA‐3‐O‐glucoside) and 3 (CA‐3‐O‐xyloside) with a C3‐OH substitution, AmGT8 and 9 could catalyse the glucosylation at C2′‐OH and C25‐OH respectively. However, the C6‐OH glucosylation activity of AmGT8_A394F decreased significantly. For compound 4 (CA‐6‐O‐glucoside), AmGT1/5 and AmGT9 could catalyse the glucosylation at C3‐OH and C25‐OH respectively. These results indicated that C6‐O‐glucosylation should occur before C3/C25‐O‐glycosylation. For compounds 5 (CA‐3‐O‐xyloside‐2′‐O‐glucoside) and 6 (CA‐3‐O‐xyloside‐6‐O‐glucoside), only AmGT9 showed high glycosylation activity at C25‐OH. When compound 8 (CA‐25‐O‐glucoside) was used, the C3‐OH glucosylation activity retained for AmGT1/5, while the activities of AmGT8 and AmGT8_A394F were reduced (Figure 5b). These results indicated that C2′/C6‐O‐glycosylation should occur before C25‐O‐glycosylation, while the order of C3/C25‐O‐glycosylation was interchangeable (Appendix S1).

Since that specific 3‐O‐xylosylation is important for astragalosides, AmGT1 and its variants were also tested using UDP‐Xyl as the sugar donor (Figure 5c). For wild‐type AmGT1, the substitution of C6‐OH (4 and 13) affected its catalytic activity, while the substitution of C25‐OH (8) did not. For AmGT1 variants, only G146S/P/A showed considerable xylosylation activity (69.3%–76.1%) to CA‐6‐O‐glucoside (4), though the K _m value of AmGT_G146S was only slightly lower than the wild‐type (Figure 3c, Figure S12). These results suggested that different AmGT1 variants should be used for the xylosylation of different substrates: AmGT1_WT/G146V/I for CA, and AmGT1_G146S/P/A for CA‐6‐O‐glucoside.

Based on the above results, we constructed the combinatorial synthesis network of astragalosides 2–14 by AmGTs (Figure 5d). Xylosylation of C3‐OH could be catalysed by AmGT1_G146V/I for compound 1, AmGT1_G146S/P/A for compound 4 and AmGT1 for compound 8. Glucosylation of C3‐OH could be catalysed by AmGT1/5, while C25‐OH by AmGT9. Meanwhile, glucosylation at C2′‐OH was achieved by AmGT8, while at C6‐OH by AmGT8_A394F. The order of glycosylation for most astragalosides should be C6‐OH first, then C3‐OH, C2′‐OH and C25‐OH. For astragalosides without C6/C2′‐OH, glycosylation order at C3‐OH and C25‐OH could be switched.

Astragalus root contains at least 14 astragalosides (2–15) with different glycosylation patterns. Among them, 2, 3, 4 and 8 are mono‐glycosides which could be synthesized by AmGT1_G146S, AmGT1_G146V, AmGT8_A394F and AmGT9 with a conversion rate up to 97.2%, 95.5%, 45% and 84.1% respectively. The di‐glucoside 7 could be produced using AmGT8 with a conversion rate of >90% (Zhang et al., 2022). For other di‐glycosides (5, 6, 9, 11–13) and tri‐glycosides (10, 14), they could be synthesized using combinatorial biosynthesis (Figure 5e). According to the biochemical characteristics, AmGT1/5/9 favoured for Na₂HPO₄‐NaH₂PO₄ buffer at pH = 8, while AmGT1_G146V/S favoured for Tris–HCl buffer at pH = 7/8 (Figure S9). Since that AmGT5 suffered from very poor activity in Tris–HCl, we used Na₂HPO₄‐NaH₂PO₄ as a compatible buffer. One‐pot, step‐by‐step reactions were established as shown in Figure 5e. The reaction mixture was analysed by UHPLC/MS to confirm the product and by ultra‐high‐performance liquid chromatography coupled with charged aerosol detection (UHPLC/CAD) to calculate the conversion rate (Figure S13). As a result, conversion rates for astragalosides 5, 6 and 9–14 were from 22.6% to 98.7%. For the reactions with conversion rates <50%, by‐products (3G from AmGT8_A394F, 25G from AmGT9) were detected which warrants further optimization. This is the first report to produce the whole series of natural products in an herb using combinatorial biosynthesis.

Structural modification of triterpenoids by AmGTs

Besides astragalosides, AmGTs could also be used to expand the structural diversity of saponins by attaching various glycosyl groups. When CA (1) or CA‐25‐O‐Glc (8) was used as the substrate, AmGT1 could utilize 10 sugar donors, including UDP‐Glc, UDP‐Xyl, UDP‐ glucosamine (UDP‐GlcN), UDP‐N‐acetylglucosamine (UDP‐GlcNAc), UDP‐glucuronic acid (UDP‐GlcA), UDP‐galactose (UDP‐Gal), UDP‐galactosamine (UDP‐GalN), UDP‐arabinose (UDP‐Ara), UDP‐rhamnose (UDP‐Rha) and thymidine diphosphate glucose (TDP‐Glc) (Figure 6a). The conversion rates were above 85% except for UDP‐GlcA and UDP‐GalN. However, when CA‐6‐O‐Glc (4) and CA‐6,25‐di‐O‐Glc (13) were used, the sugar donor promiscuity of AmGT1 decreased remarkably. CA‐3‐O‐galactoside (17) and CA‐3‐O‐acetylglucosamine (18) were purified from scaled‐up reactions using CA and AmGT1, indicating the glycosylation occurred at C3‐OH. Their structures were characterized by NMR as new compounds (Appendix  S1 ). To our best knowledge, AmGT1 was the triterpene GT with highest sugar donor promiscuity (Table S2). In contrast, AmGT5 and AmGT9 could mainly utilize UDP‐Glc and UDP‐Xyl.

Figure 6.

The sugar donor promiscuity of AmGT1/5/9. (a) Conversion rates for cycloastragenol (1), CA‐6‐O‐Glc (4), CA‐25‐O‐Glc (8) and CA‐6,25‐di‐O‐Glc (13) using different sugar donors. (b) Structures of UDP‐sugar donors. (c) Conversion rates for 20 S/R‐PPD (19/20) and 20 S/R‐PPT (21/22) using different sugar donors. (d) Structures of compounds 19–22. Asterisks in (a) and (c) indicated structures confirmed by comparing with reference standards, while arrows indicated compounds purified and characterized by NMR.

AmGTs could also be employed to synthesize ginsenosides, which are tetracyclic triterpenoid saponins similar to astragalosides, and are bioactive components from Ginseng roots (Zhao et al., 2020). Most of the ginsenosides contain two or more sugar moieties, while ginsenosides with only one sugar residue are rare. In this study, AmGTs were used to catalyse the glucosylation of protopanaxadiols (PPD, 19–20) and protopanaxatriols (PPT, 21–22) to obtain ginsenosides Rh1 and Rh2 (23–26, Figure S14). Meanwhile, both AmGT1 and AmGT8 showed good sugar donor promiscuity to use more than four different donors (Figure 6c). In total, nine catalytic products are rare ginsenosides, and 21 mono‐glycosides are unreported structures. A new compound, 3‐O‐xyloside of 20R‐PPD (27) was purified from the scaled‐up reaction using AmGT1, and was characterized by NMR (Appendix  S1 ).

AmGT1 could also catalyse the glycosylation of a variety of triterpenoids or steroids (Figure S15). Interestingly, it showed high conversion rate to compounds with 3‐β‐OH (28–31) and poor activity to 3‐α‐OH and phenolic 3‐OH (32 and 33). The product of oleanolic acid (30) was prepared and characterized as oleanolic acid‐3‐O‐glucoside (34) by NMR (Appendix  S1 ).

Biological activities of glycosylated products

As the effective components of Astragalus root, the saponin extract exhibited hepatoprotective and anti‐virus activities (Li et al., 2018; Liang et al., 2020; Sun et al., 2007; Tang et al., 2010). However, the bioactivity of various astragalosides has rarely been reported except for a few major ones such as astragaloside IV. Thus, we investigated the bioactivities of the 13 astragalosides obtained through combinatorial synthesis. Compounds 6, 9, 10, 11 and 14 showed protective effects against acetaminophen (APAP)‐induced HepG2 cell death. The cell viabilities were from 67.6% to 76.0%, which was significantly higher than their aglycone (1, 62.3%) and the model group (45.7%) (Figure 7a). Meanwhile, the inhibitory activities against key proteins of severe acute respiratory syndrome coronavirus (SARS‐CoV‐2) were evaluated in vitro (Figure 7b–e, Figure S16). The inhibition of receptor‐binding domain of the spike protein (S‐RBD) significantly increased for the tri‐glycoside 14 (IC₅₀ 15.17 μM), compared to the aglycone (1, IC₅₀ > 40 μM). In addition, the glycosylated product 34 also exhibited increased inhibitory effect against 3C‐like protease (3CL^pro, IC₅₀ 5.95 μM) and papain‐like protease (PL^pro, IC₅₀ 7.57 μM), compared to its aglycone 30. These results indicated that glycosylation could improve the pharmacological activities of triterpenoids.

Figure 7.

Biological activities of the glycosylated products. (a) Hepatoprotective activities of catalytic products against APAP‐induced injury of HepG2 cells. Silymarin (Sily) was used as the positive drug. For all groups, n = 3. *P < 0.05, **P < 0.01 compared with compound 1 (Student's t‐test). (b) Inhibitory activities of selected compounds against S‐RBD, 3CL^pro and PL^pro of SARS‐CoV‐2. For all groups, n = 3. (c–e) Dose‐dependent inhibition against S‐RBD (c), 3CL^pro (d) and PL^pro (e) of selected catalytic products. For all groups, n = 3.

Methods

Plant materials and chemicals

Two‐week‐old A. membranaceus seedlings were obtained as previously reported (Chen et al., 2022) and used for gene cloning (Figure S17a). A 3‐year‐old A. membranaceus plant was collected from Daxing'anling (Heilongjiang, China) and used for chemical analysis (Figure S17b). Reference standards 1, 5, 6, 19–26 and 28–33 were purchased from Chengdu Desite Biotechnology (China), Shanghai TargetMol Biotechnology (China), Solarbio Technology (Beijing, China) or Sigma‐Aldrich (China). Reference standards 2, 3, 4 and 7 were previously purified and identified by NMR in our group (Zhang et al., 2022). Other 12 glycosylated products (8–14, 17, 18, 27 and 34) were purified and identified by NMR in this study (Appendix  S1 ). UDP‐sugar donors were purchased from Aifly Biotechnology (China). LC/MS grade methanol and acetonitrile were purchased from Fisher Scientific (USA).

Chemical analysis of A. membranaceus

For chemical analysis, 4 mg dried powder of A. membranaceus roots was extracted by ultrasonication for 30 min in 1 mL of 75% methanol. After centrifugation, 200 μL supernatant was mixed with 200 μL of 13% NH₃·H₂O and heated at 65 °C for 1 h. After dryness, the mixture was redissolved with 1 mL of 75% methanol and centrifuged at 21 130 g for 30 min. The supernatant was analysed by LC/MS as described in ‘UHPLC/CAD and UHPLC/MS analysis’ of the Experiment.

Transcriptome mining, phylogenetic analysis and transcript level analysis

Two transcriptome datasets (NCBI Accession No. ERR706814 and SRR5343992) were used for gene screening. Plant triterpenoid GTs in Table S2 were used as templates. For phylogenetic analysis, 25 selected plant triterpenoid GTs were analysed together with AmGT1/5/9/8. The phylogenetic tree was constructed using MEGA6 Software with the Maximum Likelihood method based on ClustalW multiple alignments (Jones et al., 1992; Tamura et al., 2013). The bootstrap consensus tree inferred from 1000 replicates was used.

Molecular cloning, mutagenesis and expression of AmGTs

Total RNA was extracted from roots of 2‐week‐old A. membranaceus seedlings using the TranZol kit (Transgen Biotech, China) according to the manufacturer's instructions. The cDNA was synthesized with TransScript One‐step gDNA Removal and cDNA Synthesis SuperMix (Transgen Biotech, China). The coding regions of AmGT1, AmGT5 and AmGT9 were amplified using primers in Table S4 with TransStart FastPfu DNA Polymerase (Transgen Biotech, China) and then cloned into pET28a(+) vectors using Quick‐Change method (Bok and Keller, 2012). The mutants of AmGTs were constructed by a Fast MultiSite Mutagenesis System (Transgen Biotech, China) using the primers in Table S5.

After verification of the sequences, the recombinant plasmids were introduced into E. coli BL21(DE3) for heterologous expression. The procedures for protein expression and purification were same as reported (Chen et al., 2019). Briefly, single colonies were grown in 500 mL LB medium containing kanamycin (50 μg/mL) at 37 °C. After the OD₆₀₀ reached 0.4–0.6, the cells were induced with 0.1 mM isopropyl β‐D‐thiogalactoside at 18 °C for 18 h. The cells were then harvested by centrifugation. The recombinant proteins were purified using a Ni‐NTA column and stored in storage buffer (20 mM Tris, 500 mM NaCl, 20% glycerol, pH = 7.5).

Functional characterization, kinetic studies and combinatorial biosynthesis

The analytical reactions were carried out in a 100 μL system, containing 50 mM Na₂HPO₄‐NaH₂PO₄ buffer (pH 8.0), 0.1 mM substrate, 0.5 mM UDP‐sugar donor and 20 μg of purified enzyme. After incubation at 37 °C for 2 h, the reactions were terminated with 100 μL methanol. The mixtures were centrifuged at 21 130 g for 20 min and then analysed by UHPLC/CAD or UHPLC/MS. The conversion rates (%) were calculated by CAD peak area (A _product/A _{substrate+product} × 100%). Blank samples were prepared from reactions without sugar donors.

Kinetic studies were performed in a 100 μL system, containing 50 mM Na₂HPO₄‐NaH₂PO₄ buffer (pH 8.0), 0.33 μg of purified enzyme, 2.5 mM of the substrate and 1–200 μM of UDP‐sugar. The reactions were incubated at 37 °C for 10 min, and then terminated and analysed as described in the above paragraph. Products were quantified by LC/MS using a calibration curve. Kinetic parameters of AmGTs were calculated by fitting in the Michaelis–Menten plot.

For the combinatorial biosynthesis of di‐/tri‐glycosides, one‐pot, step‐by‐step reactions were carried out in 100 μL systems, containing 50 mM Na₂HPO₄‐NaH₂PO₄ buffer (pH 8.0) and 0.1 mM CA. For each step, 20 μg of purified enzyme and 0.5 mM UDP‐sugar donor were added. The mixture was incubated at 37 °C for 2 h before the enzyme and sugar donor were added for the next step. Finally, reactions were terminated and analysed as described in the first paragraph of this section. The conversion rates (%) were calculated by CAD peak area.

Scaled‐up reactions to prepare catalytic products

The scaled‐up reactions were performed at 37 °C overnight using a total volume of 100 mL phosphate buffer (50 mM Na₂HPO₄‐NaH₂PO₄, pH 8.0). The reaction contained 10 mg of substrate (n mol), ~20 mg sugar donor (equivalent to 2n mol) and approximate 30 mg purified AmGT. The reactions were terminated by adding 200 mL methanol, and centrifuged at 21 130 g for 20 min. The supernatants were concentrated and then redissolved in 2 mL of 50% methanol. The solution was subjected to an Agilent Zorbax SB‐C18 column (9.4 mm × 250 mm, 5 μm) and separated using a semi‐preparative HPLC/ELSD (ELSD 6000, Alltech, USA), with acetonitrile and water containing 0.03% trifluoroacetic acid (v/v) as the mobile phase. NMR spectra of the purified products were obtained on AVANCE III‐400 or AVANCE III‐600 instrument (Bruker, USA) (Appendix  S1 ).

Homology modelling, molecular docking and binding free energy calculations

The structure models of AmGT1 and its mutants were generated by Alphafold2 (Tunyasuvunakool et al., 2021). Molecular docking of protein with ligands was carried out by Autodock (Trott and Olson, 2010). Models were visualized by AutoDocktools v1.5.6 and Pymol.

The protonation states of ionizable residues were determined using PROPKA3 (implemented in Schrödinger Suite, version 2021‐3) prior to the MD simulations (Olsson et al., 2011). According to the predicted pKa values, all ASP and GLU residues were deprotonated, while LYS and ARG were all protonated. For HIS residues, HIS 22, 38, 221 and 282, and 382 were δ‐protonated, HIS 228, 232, 300, 351, 360 and 463 were ε‐protonated, and all the other HIS residues were protonated at both δ‐ and ε‐nitrogen sites. All the MD simulations were carried out using the Desmond package (Desmond Molecular Dynamics System, D. E. Shaw Research, New York) on the NVIDIA Tesla T4 equipped GPU nodes at the supercomputer centre of Linköping University. The OPLS4 force field was used for both protein and ligand atoms with the default atomic charge scheme (Lu et al., 2021). In a MD simulation, the solvent molecules were added to a 12‐Å buffering region surrounding the protein, which generated an orthorhombic box containing ~17 300 TIP3P water molecules with certain amount of counter ions Na⁺ and Cl⁻ to neutralize the system as well as to mimic the physical salt concentration of 0.15 M (Jorgensen et al., 1983). To maintain the temperature and pressure of simulations at 300.0 K and 1.0 atm, the Nose‐Hoover chain thermostat (Martyna and Klein, 1992) and Martyna‐Bobias‐Klein barostat (Martyna, 1994) were used respectively. The default procedure for minimization and equilibration implemented in Desmond was used before the 100‐ns production simulation for each protein‐ligand system. The trajectory was saved every 50 ps. Analysis of the trajectory files was completed using the Visual Molecular Dynamics (VMD) package, version 1.9.3 (Humphrey et al., 1996).

The MM/GBSA (molecular mechanics, the generalized Born model and solvent accessibility) binding free energy was calculated using the Prime module of Schrödinger Suite (version 2021‐3). A total of 400 snapshots evenly extracted from 100‐ns trajectory were used for MM/GBSA calculations. We used the OPLS4 force field to refine the residues within 8 Å of the ligand in the continuum solvation VSGB (variable dielectric surface generalized Born) (Li et al., 2011) before MM/GBSA calculations. The mean and deviation values calculated from the extracted snapshots were reported in this study.

UHPLC/CAD and UHPLC/MS analysis

UHPLC/CAD analysis was performed on a Vanquish Flex UHPLC Dual System (ThermoFisher Scientific, USA) coupled with a Charged Aerosol Detector (CAD). Samples (5 μL) were separated on an Acquity UHPLC HSS T3 column (100 mm × 2.1 mm, 1.8 μm, Waters, USA). The mobile phase consisted of water containing 0.1% formic acid (v/v, A) and acetonitrile (B). A gradient elution programme was used: 0–1 min, 30% B; 1–12 min, 30%–50% B; 12–15 min, 50%–100% B; 15–18 min, 100% B. The flow rate was 0.3 mL/min, and the column temperature was 55 °C. An automatic inverse gradient was used to eliminate the impact of changes in solvent composition during gradient elution.

UHPLC/MS analysis was performed on a Vanquish UHPLC system coupled with a Q‐Exactive quadrupole‐Orbitrap mass spectrometer equipped with a heated electrospray ionization source (ThermoFisher Scientific, USA). Samples (2 μL) were separated on an Acquity UHPLC HSS T3 column (150 mm × 2.1 mm, 1.8 μm, Waters, USA) at 50 °C. The mobile phase and gradient programme were the same as UHPLC/CAD method. The mass spectrometer was operated in the negative and positive ion modes. The MS parameters were as follows: sheath gas pressure, 45 arb; auxiliary gas pressure, 10 arb; discharge voltage, 4.5 kV; capillary temperature, 350 °C.

Hepatoprotective activities against APAP‐induced cell death

To evaluate the hepatoprotective activities against APAP‐induced cell death, HepG2 cells were treated with 20 μM compounds and 14 mM APAP for 24 h, and the cell viabilities were measured using CCK‐8 method (Kuang et al., 2017). Silymarin (20 μM) was used as the positive control.

Inhibitory activity assay against S‐RBD, 3CL^pro and PL^pro

The inhibition activity against S‐RBD was measured by ELISA as described in our previous study (Yi et al., 2022a,b). Briefly, the spike protein DRA49 (0.3 μg/mL) was incubated in the coating solution (Na₂CO₃ 1.59 mg/mL, NaHCO₃ 2.93 mg/mL, pH 9.6) on 96‐well microplates at 4 °C overnight. The plates were washed for three times, incubated at 37 °C for 40 min using wash buffer containing 2% bovine serum albumin, and washed again for three times. Candidate compounds (10 μM) or positive drug (S‐hIgG1, 5 μg/mL) were added and incubated for 1 h. ACE2 (angiotensin‐converting enzyme 2, 0.3 μg/mL) was then added and incubated for another 30 min. At last, the enzyme‐linked antibody SA‐HRP (1 : 10 000) was added. Tetramethyl benzidine was used as the chromogenic agent. Finally, the reaction was stopped by 1 M HCl and the enzyme activity was determined at OD_450nm on LabServ K3 Microplate Reader (Thermo Fisher Scientific, USA; Yi et al., 2022a,b).

The inhibition activity against 3CL^pro and PL^pro was measured by fluorescence resonance energy transfer method. For 3CL^pro, the reaction mixture contained 12.5 μg/mL protein, 3 mM substrate, 8 μM candidate compound or positive drug (GC376), in 20 mM Tris–HCl (pH 7.0) as the reaction buffer. After incubated at 25 °C for 10 min, the fluorescence was detected (emission at 490 nm and excitation at 340 nm) on a FlexStation 3 Multi‐Mode Microplate Reader (Molecular Devices, China; Dai et al., 2020; Yi et al., 2022a,b). For PL^pro, the reaction mixture contained 12.5 μg/mL protein, 5 mM substrate, 8 μM candidate compound or positive drug (GRL0617), in 50 mM HEPES (pH 7.5) as the reaction buffer. After incubated at 25 °C for 10 min, the fluorescence was detected (emission at 535 nm and excitation at 340 nm) on a FlexStation 3 Multi‐Mode Microplate Reader (Yi et al., 2022a,b).

Statistical analysis

For substrate promiscuity analysis, qRT‐PCR analysis, hepatoprotective activities and S‐RBD/3CL^pro/PL^pro inhibition assays, all measurements were in triplicate and the data were expressed as mean ± SD. The significance of difference between groups was determined by two‐tailed, paired Student's t‐tests for bioactivity assays.

Conflict of interest statement

The authors declare no conflict of interests.

Authors' contributions

XQ, MY, KC and MZ planned and designed the research. KC, MZ, BG, AH, YB, JF, RY, YY, ZW and HL performed experiments. JL, HÅ and MZ contributed to the theoretical calculation. KC, MZ, XQ and MY analysed the data. XQ and KC wrote the manuscript.

Supporting information

Table S1 Glycosylation patterns of astragalosides reported in Astragalus membranaceus and Astragalus membranaceus var. mongholicus with cycloastragenol as the skeleton. Acylation was not included.

Table S2 Plant triterpenoid glycosyltransferases reported.

Table S3 Primers for qRT‐PCR experiment.

Table S4 Primers of AmGTs for recombinant plasmid construction with pET28a(+).

Table S5 Primers used in site‐directed mutagenesis experiments.

Figure S1 LC/MS analysis of Astragalus roots after ammonia hydrolysis.

Figure S2–S6 HR‐MS and HR‐MS/MS data of compounds 1‐16.

Figure S7 Relative transcript level of AmGTs in different tissues through qRT‐PCR analysis.

Figure S8 SDS‐PAGE analysis of AmGTs characterized in this study.

Figure S9 Effects of various conditions on the activity of AmGTs.

Figure S10 Protein sequence alignment for 53 triterpene GTs.

Figure S11 The functional characterization of mutants of AmGT1 with CA as the substrate, UDP‐Glc or UDP‐Xyl as the sugar donor.

Figure S12 The kinetic parameters of AmGT1 and its mutants.

Figure S13 Combinatorial biosynthesis of di‐/tri‐ glycosides analyzed by UHPLC/CAD.

Figure S14 Production of rare ginsenosides catalyzed by AmGTs.

Figure S15 Glucosylation of other triterpenoids or steroids catalyzed by AmGT1.

Figure S16 Inhibitory activities of selected compounds against S‐RBD, 3CLpro, and PLpro of SARS‐CoV‐2.

Figure S17 Plant materials used in this study.

Figure S18–S24 Glycosylation of compounds 1–6 and 8 catalyzed by AmGTs with UDP‐Glc as the sugar donor.

Figure S25 Glycosylation of compounds 4/8/13 catalyzed by AmGT1 and its mutants with UDP‐Xyl as the sugar donor.

Figure S26–S33 Glycosylation of compound 1 catalyzed by AmGTs with different sugar donors.

Figure S34–S36 Glycosylation of compounds 8, 4, and 13 catalyzed by AmGT1 with different sugar donors.

Figure S37–S40 Glycosylation of compounds 19–22 catalyzed by AmGTs with UDPGlc as the sugar donor.

Figure S41–S44 Glycosylation of compounds 19–22 catalyzed by AmGTs with UDPXyl as the sugar donor.

Figure S45–S48 Glycosylation of compounds 19–22 catalyzed by AmGTs with TDPGlc as the sugar donor.

Figure S49–S52 Glycosylation of compounds 19–22 catalyzed by AmGTs with UDPGlcN as the sugar donor.

Figure S53–S56 Glycosylation of compounds 19–22 catalyzed by AmGTs with UDPGlcNAc as the sugar donor.

Figure S57–S60 Glycosylation of compounds 19–22 catalyzed by AmGTs with UDPAra as the sugar.