Glycosyltransferases: Mining, engineering and applications in biosynthesis of glycosylated plant natural products

Abstract

UDP-Glycosyltransferases (UGTs) catalyze the transfer of nucleotide-activated sugars to specific acceptors, among which the GT1 family enzymes are well-known for their function in biosynthesis of natural product glycosides. Elucidating GT function represents necessary step in metabolic engineering of aglycone glycosylation to produce drug leads, cosmetics, nutrients and sweeteners. In this review, we systematically summarize the phylogenetic distribution and catalytic diversity of plant GTs. We also discuss recent progress in the identification of novel GT candidates for synthesis of plant natural products (PNPs) using multi-omics technology and deep learning predicted models. We also highlight recent advances in rational design and directed evolution engineering strategies for new or improved GT functions. Finally, we cover recent breakthroughs in the application of GTs for microbial biosynthesis of some representative glycosylated PNPs, including flavonoid glycosides (fisetin 3-O-glycosides, astragalin, scutellarein 7-O-glucoside), terpenoid glycosides (rebaudioside A, ginsenosides) and polyketide glycosides (salidroside, polydatin).

1. Introduction

More than 200,000 natural products are known to be synthesized by plants (hereafter, plant natural products; PNPs), collectively constituting a large, structural diverse library of compounds with widely varying biological activities [1]. A considerable proportion of PNPs are glycosylated with diverse sugar moieties attached to the aglycones, thus greatly increasing the variety and complexity of structures. These glycosylated PNPs have been used as medicines, sweeteners, nutrients, cosmetics, and health products, consequently drawing considerable research attention to their biosynthesis and modification [2].

The biosynthesis of glycosylated compounds involves multiple complex biological processes that are orchestrated by many enzymatic systems in plants. Glycosyltransferases (GTs) (EC 2.4.x.y) are crucial for the biosynthesis of glycosylated PNPs and commonly perform the final step in their biosynthesis pathway, mediating the regio- and stereo-specific glycosidic bond formation via transfer of nucleotide-diphosphate-activated sugar moieties to a variety of biomolecules [3]. GTs involved in the biosynthesis of glycosylated PNPs are universally classified into the GT1 family in the Carbohydrate Active Enzyme database (CAZy, http://www.cazy.org/). Currently, nearly 30,000 GTs are included in the GT1 family, and an increasing number of GT sequences are reported from different organisms with the development of the high-throughput sequencing and advances in deep-learning-based analysis [[4], [5], [6]]. However, only 1% of GTs have been functionally characterized, which has limited the clarification of glycosylated PNPs biosynthetic pathways and validation of enzymatic mechanisms [7].

Using enzymatic catalysis or metabolic engineering methods, scientists can produce glycosylated PNPs at industrial scale, thus providing a reliable and scalable alternative to conventional production methods based on extraction from natural resources [[8], [9], [10], [11]]. However, wild-type GTs are often accompanied by unfavorable properties for the synthesis of target products, such as low expression levels and low catalytic activity. Thus, these obstacles persist as limiting factors in glycosylated PNP biosynthesis which require protein engineering through rational design and/or directed evolution [[12], [13], [14]].

In recent years, the rapid development of synthetic biology has accelerated the metabolic engineering of complex synthetic pathways and increased efficiency during production of limited natural resources [15,16]. Systematic optimization of redirecting metabolic fluxes in microbial hosts to produce desired products through the combined heterologous expression of plant pathways and enzymes makes it possible to construct new and efficient PNP biosynthetic routes [17]. Significant progress has been made in producing glycosylated PNPs via different hosts using synthetic biology approaches [8]. Once a microbial factory has been developed that can produce even low quantities of a target product, strategies are then employed to engineer that strain for production of industrial-scale titers [18,19].

Recently, the structure-function relationship and glycosylation mechanisms of terpenoid GTs were summarized [3,20]. The evolution and substrates coverage of GT1 family were also overviewed to pave the way for the future exploration of GT proteins [12,21]. However, there is still a lack of comprehensive review on the catalytic diversity of plant GTs in the synthesis of glycosylated PNPs. Moreover, considerable progress has been made in recent years in the mining, engineering and applications of GTs. A timely systematic review will help in constructing synthetic pathways of glycosylated PNPs. In this review, we focus on recent advances in the mining, engineering, and applications of GTs in glycosylated PNP biosynthesis. We systematically summarize the phylogenetic distribution and catalytic diversity of the characterized GTs from plant sources, as well as advanced methods for identification of candidate GTs for glycosylated PNP synthesis. In addition, we also discuss current and ongoing efforts to engineer GTs for new or improved functions by rational design and directed evolution. Finally, this review also covers recent progress in GT applications in the microbial biosynthesis of glycosylated PNPs.

2. GT diversity in glycosylated PNP synthetic pathways

In plants, GTs can glycosylate almost all major classes of secondary metabolites, such as phenylpropanoids (flavanoids, coumarins, lignans, etc.) [[22], [23], [24]], alkaloids (indole alkaloids, steroidal alkaloids, cytokinins, etc.) [[25], [26], [27]], terpenoids (monoterpenoids, diterpenoids, triterpenoids, etc.) [[28], [29], [30]], and polyketides (phenol polyketides, polycyclic aromatic polyketides, etc.) (Fig. 1) [[31], [32], [33], [34]].

Fig. 1.

Representative substrates of functionally characterized GTs in the synthesis of glycosylated PNPs. Chemical groups that are typically glycosylated were colored red.

In the present review, we phylogenetically summarized 303 characterized plant GTs (215 were collected in CAZy database, and 88 were collected from recent studies), and analyzed the diversity of their natural substrates and different glycosidic bonds they catalyze (Fig. 2). Phylogenetic trees were generated using the neighbor-joining method with the poisson model applied and using a bootstrap replication of 1000 using MEGAX64. The trees were then imported into the online software iTOL (https://itol.embl.de/) for further optimization. The 303 selected plant GTs could be phylogenetically classified into four major clades. Among the four clades, GTs associated with phenylpropanoid glycoside synthesis comprise the largest family, followed by those that synthesize terpenoid glycosides and polyketide glycosides, while GTs involved in alkaloid glycosides production are relatively rare.

Fig. 2.

Phylogenetic tree analysis of the characterized plant GTs from the GT1 family. The substrate specificities of the characterized plant GTs were depicted by different colors in peripheral circle. The glycosidic bond catalytic by the characterized plant GTs were depicted by different colors in interlayer circle. The characterized plant GTs with co-crystal structures reported were highlighted by red lines. GTs mentioned in the text were marked and numbered in red.

The different clades showed no strict catalytic boundaries for four classes of substrates (including phenylpropanoids, alkaloids, terpenoids, and polyketides). At the clade level, GTs in clades I-III collectively provide broad substrate selectivity, typically accepting almost all of the compound classes listed in Fig. 1. By contrast, clade IV GTs have a relatively narrow substrate spectrum, e.g., with obvious preference for phenylpropanoids or alkaloids.

Occasionally, functional characterizations reveal that evolutionarily close GTs may glycosylate disparate chemical aglycones. For instance, UGT84B1 accepts alkaloids [35], while UGT84B2 (78.3% similarity with UGT84B1) transforms phenylpropanoids (1 and 2 in Fig. 2, respectively) [36]. The reasons for this phenomenon may be attributable to either incomplete characterization of catalytic function, or that specific mutations in key residues, often outside of the active site region, have altered their substrate spectrum [37].

Moreover, despite the diversity in substrates at the clade level, plant GTs have high substrate specificity at the individual level, that they frequently will only accommodate one class of acceptors (Fig. 2). However, some GTs show remarkable promiscuity, showing the ability to catalyze different classes of compounds. This flexibility is key for the success of many glycosylation engineering strategies aimed at facilitating the production of diverse glycosylated PNPs. In particular, UGT71G1 (Medicago truncatula) [36,38], GuGT14, GuGT33 (Glycyrrhiza uralensis) [39] and UGT73AE1 (Carthamus tinctorius) [40] have been shown to catalyze glycosylation of phenylpropanoids (coumarins, coumarones, flavones, isoflavones, and others) and terpenoids (glycyrrhetinic acid and glycyrrhizic acid) (3–6 in Fig. 2, respectively), while InGTase1 (Ipomoea nil) [41] can catalyze phenylpropanoids and alkaloids (7 in Fig. 2), PaGT2 (Phytolacca americana) [42] can catalyze phenylpropanoids and polyketides (9 in Fig. 2), and SaGT4A (Solanum aculeatissimum) [43] can catalyze terpenoids and alkaloids (10 in Fig. 2).

In addition to handling diverse substrates, GTs can also mediate sugar moiety transfer to a diversity of acceptor atoms, O-, C-, N- and even S-atoms in PNP aglycons [3]. O-glycosides are the most common natural products of glycosylation, while the C-, N- and S-glycosides are relatively rare (Fig. 2) [21,44]. The first triterpene arabinosyl-O-GT (UGT99D1, 11 in Fig. 2) was recently discovered and characterized from Avena strigosa. This enzyme selectively adds l-arabinoside to the triterpene scaffold at the C-3 position, a modification critical for disease resistance [45]. Another O-GT (UGT84B1, 1 in Fig. 2) from Arabidopsis thaliana can transfer a glucosyl moiety to the –COOH of indole-3-acetic acid and phenylacetic acid, two primary natural auxins. In addition, 11 O-GTs from Glycyrrhiza uralensis, including isoflavone 7-O-GTs, flavonol 3-O-GTs, and promiscuous O-GTs catalyzing flavones, chalcones, and triterpenoids have been characterized [39].

C-glycoside secondary metabolites and bioactive molecules are widely distributed in plants, and are metabolically more stable than O-, N-, S-glycosides [46,47]. Interestingly, the vast majority of plant C-GTs are found in Clade I, which may indicate their evolutionary conservation and unique catalytic function for C-glycoside synthesis (Fig. 2). In particular, Ye et al. reported a di-C-glycosyltransferase GgCGT (12 in Fig. 2) from Glycyrrhiza glabra, which catalyzes a two-step di-C-glycosylation of flopropione-containing substrates with conversion rates of >98%. GgCGT is the first di-C-GT with a crystal structure containing a sugar acceptor [48]. Recently, TcCGT1 (13 in Fig. 2) from A. thaliana have been identified as an 8-C-GT, which can efficiently and regio-specifically catalyze the 8-C-glycosylation of 36 flavones and other flavonoids. This broad substrate promiscuity of TcCGT1 is enabled by a spacious binding pocket and provides a basis for efficient directed biosynthesis of valuable and diverse bioactive flavonoid C-glycosides [49].

In contrast, few examples of S- and N-GTs from plants have been described in the literature. Historically, glucosinolates were the first identified S-glycosides, which were found in cruciferous vegetables. UGT74B1 (14 in Fig. 2) from A. thaliana was reported to efficiently glycosylate thiohydroximate in glucosinolate biosynthesis [50].

A few plant GTs were shown to perform multiple catalytic functions for different glycosides. For example, a bifunctional maize glycosyltransferase (UGT708A6, 15 in Fig. 2) can produce both C- and O-glycosylated flavonoids, a property not previously described for any other GTs [51]. Additionally, PlGT7 (Pueraria lobata) [52], UGT73AE1 (Carthamus tinctorius) [40], RyUGT3A (Rubia yunnanensis), and RyUGT12 (Rubia yunnanensis) [53] (16, 6, 17, 18 in Fig. 2, respectively) can even transfer various sugars to three different nucleophilic groups (OH, NH₂, and SH) of diverse compounds, thus producing O-, N-, and S-glycosides.

3. Mining GTs for biosynthesis of glycosylated PNPs

The discovery of GTs with novel functions is a necessity for advancing their practical applications. Recently, the development of high-throughput sequencing and deep learning analysis have enabled considerable advances in enzyme mining, leading to progress in discovery of novel GTs for glycosylated PNP production (Fig. 3) [[54], [55], [56], [57], [58]].

Fig. 3.

The workflow for GT mining and application in biosynthesis of glycosylated PNPs.

Genomic analysis of whole genome sequence data can provide a multi-level resource for mining novel enzymes that can be greatly informative for researchers seeking to identify genes involved in glycosylated PNP synthesis [[59], [60], [61]]. Huang et al. demonstrated that plant GTs harbor a highly conserved 44 amino acid C-terminus motif (the plant secondary product glycosyltransferase box, PSPG box), which has been proposed to serve as a nucleotide diphosphate sugar binding site [62,63]. A recent genome-wide analysis of soybean by Rehman et al. identified 149 putative UGTs based on the PSPG box [64]. Similar approaches were used for GT identification in the genomes of chickpea, cotton, maize, and flax [[65], [66], [67], [68]]. In addition, genomic data can also provide insights into glycosylated PNP biosynthetic pathways evolution, regulation, and production [69]. Despite the exponential increase in genomic sequence data, the use of glycosylated PNPs for engineering biosynthetic pathways is still restricted to experimentally characterized GTs [70]. For example, to identify the 2′-O-GT (P2′GT) responsible for phloretin production, Zhou et al. performed genome-wide analysis in domesticated apple (Malus x domestica Borkh), which identified two P2′GTs (MdUGT88F1 and MdUGT88F4) that were validated by in vitro activity assays and relative expression analysis (Table 1) [4].

Table 1.

The recent five years characterized GTs identified by high-throughput sequencing strategies.

Enzyme Name	Accession No.	Taxonomy	Systems biology tool	Reference
UGT75D1	AAB58497.1	Arabidopsis thaliana	Genomic	[25]
UGT71C3	AAF82195.1	Arabidopsis thaliana	Genomic	[71]
UGT1	QGI57841.1	Atropa belladonna	Genomics	[72]
MdUGT88F1	ARV88476.1	Malus domestica	Genomics	[4]
UGT84A23	ANN02875.1	Punica granatum	Genomics	[73]
MdPh-4-OGT	AAX16493.1	Malus domestica Brokh	Genomics	[74]
DcUCGalT1	AKI23632.1	Daucus carota	Genomics	[75]
GuCGTa, GuCGTb	QLF98865.1, QLF98866.1	Glycyrrhiza uralensis	Genomic	[76]
UGT72AD1, UGT72AH1, UGT72Z2	AP009657.1, AOG18241.1, AKK25344.1	Lotus japonicus	Genomic	[77]
MdP2GT	AMA68117.1	Malus domestica	Genomic	[78]
MdUGT88F, MdUGT88F4	ARV88476.1	Malus domestica	Genomic	[4]
MdPh-4-OGT	AAX16493.1	Malus x domestica Brokh	Genomic	[79]
AgUCGalT1	AXU98426.1	Apium graveolens	Transcriptomic	[80]
UGT71B5	ANM66102.1	Arabidopsis thaliana	Transcriptomic	[81]
UGT76C1, UGT76C2	BAB10792.1, BAB10791.1	Arabidopsis thaliana	Transcriptomic	[82]
UGT76E12	AAK82559.1	Arabidopsis thaliana	Transcriptomic	[83]
UGT85A1	AAF18537.1	Arabidopsis thaliana	Trancsriptomic	[83]
AeCGTb, AeCGTa	QLF98868.1, QLF98867.1	Arisaema erubescens	Transcriptomic	[84]
CsUGT73A20, CsUGT75L12, CsUGT78A14, CsUGT78A15	ALO19886.1, ALO19892.1 ALO19888.1, ALO19889.1	Camellia sinensis	Transcriptomic	[85]
UGT84A57	BBI55602.1	Eutrema japonicum	Transcriptomic	[86]
GgCGT	QGL05036.1	Glycyrrhiza glabra	Transcriptomic	[87]
Pn1-31, Pn3-29, Pn3-31, Pn3-32	QOJ43864.1, QOJ43865.1, QOJ43866.1, QOJ43868.1	Panax notoginseng	Transcriptomic	[88]
RyUGT3A	QSB46663.1	Rubia yunnanensis	Transcriptomic	[53]
RsUGT75L20, RsUGT75T4	AWU66063.1, AWU66062.1	Rubus suavissimus	Transcriptomic	[76]
UGT76E2	BAA97493.1	Arabidopsis thaliana	Transcriptomic	[89]
TcCGT1	QCZ42162.1	Trollius chinensis	Transcriptomic	[49]
GuGT1	QDM38894.1	Glycyrrhiza uralensis	Transcriptomic	[39]
ApUFGT1	QDA11331.1	Andrographis paniculata	Transcriptomic	[90]
CcUGT3	QDH43895.1	Crocosmia	Transcriptomic	[91]
Sb3GT1	QBL54224.1	Scutellaria baicalensis	Transcriptomic	[92]
PgUGT95B2	AZB52139.1	Punica granatum	Transcriptomic	[93]
OcUGT1	AWD73588.1	Ornithogalum caudatum	Transcriptomic	[94]
Pq3-O-UGT1	ALE15279.1	Panax quinquefolius	Transcriptomic	[95]
Pq3-O-UGT2	ALE15280.1	Panax quinquefolius	Transcriptomic	[96]
UGT72B3	AAF97321.1	Arabidopsis thaliana	Transcriptomic	[97]
UGT73F17	AXS75258.1	Glycyrrhiza uralensis	Transcriptomic	[98]
EpPF3RT	MG264429.1	Epimedium pseudowushanense	Transcriptomic	[99]
UGT76E11	CAB62337.1	Arabidopsis thaliana	Transcriptomic	[100]
PgUGT71A27	A0A0A7HB61.1	Panax ginseng	Transcriptomic	[101]
UGT75B2	AAF79732.1	Arabidopsis thaliana	Genomic & Trancsriptomic	[102]
ChCGT, CuCGT, FcCGT	BBA18064.1, BA18063.1, BBA18062.1	Citrus hanaju	Genomic & Trancsriptomic	[103]
CtUGT1	MW629113	Cistanche tubulosa	Genomic and transcriptomic	[104]
GmSGT2	BAI99584.1	Glycine max	Genomic & Transcriptomic	[105]
UGT71K3	XP_004294260.1	Fragaria ananassa	Genomics & Transcriptomic	[106]
Pq-PPT-6/20-O-UGT, Pq-PPT-6-O-UGT	QEV87497.1, QEV87498.1	Panax quinquefolius	Genomic and Transcriptomic	[107]
PsCGTa, PsCGTb	QLF98871.1, QLF98872.1	Pistia stratiotes	Genomic and Transcriptomic	[76]
SbGT30, SbGT34 SbGT56	AMK52071.1,AMK52072.1 AMK52073.1	Scutellaria baicalensis	Genomic & Trancsriptomic	[108]
UGT90F1, UGT73B26, UDPG1	MF417497.1, MF417498.1, HQ259620.1	Siraitia grosvenorii	Genomic and Transcriptomic	[109]
ZmCGTb	QLF98873.1	Zea mays	Genomic and Transcriptomic	[76]
UGT99D1	AZQ26921.1	Avena strigosa	Transcriptomic & Protemic	[45]
UGT74H5, UGT74H6	ACD03250.1, ACD03261.1	Avena strigosa	Transcriptomic & Protemic	[110]
UGT73AD1	ALD84259.1	Centella asiatica	Transcriptomic and Proteomic	[35]
GmSSAT1	XP_003532274.1	Glycine max	Transcriptomic & Proteomic	[111]

In addition to genomic data, whole transcriptome sequencing (RNA-seq) can provide the cDNA sequence and tissue-specific expression levels of specific genes not available through genomic data [112]. Thus, many studies opt for this approach to mine for GTs relevant to the synthesis of glycosylated PNPs [113,114]. For example. Murukarthick et al. carried out transcriptome sequencing on four root samples type from Panax ginseng including whole roots of from one-year-old plants, and the main root bodies, rhizomes, and lateral roots of six-year-old plants. This analysis ultimately identified 189 GT-derived transcripts involved in ginsenoside biosynthesis [115]. Similarly, a comprehensive analysis of the transcriptome landscape of three genotypes of Stevia (SR-1, SR-2, and SR-3) revealed 143 total GT unigenes, some of which were determined to contribute to steviol glycoside biosynthesis [116].

Indeed, comparative transcriptome analysis of different organs or tissues can provide more information than genomic analysis that is relevant to screening novel candidate GTs [117]. For instance, Fan et al. showed that the anthraquinone glycosides accumulated to higher levels in roots of R. yunnanensis than in the stems or leaves [118]. Based on this finding, Yi et al. compared the transcriptomes of R. yunnanensis roots, stems and leaves and identified 32 novel candidate GT genes with high expression in root tissue from 499 putative GTs found in transcriptomic data [53]. These findings largely guided subsequent screening of candidate GTs for glycosylated PNP biosynthesis.

Taking advantage of advances in LC-MS/MS technology, a recent study developed a proteomics workflow to identify candidate GTs involved in glycosylated PNP biosynthesis [119]. For instance, Suliman et al. used tandem mass spectrometry (LC-MS/MS) to identify proteins extracted from the Golgi-enriched fractions of wheat endosperm. This analysis revealed 1135 proteins in the wheat endosperm, which identified 64 GTs by searching mass data against four databases, including UniProt, Gene Index Databases Wheat release 12.0, an in-house glycosyltransferase databank, and a contaminant database (i.e., keratins and trypsin) [120].

Moreover, comprehensive gene mining has realized through the integration of genomic, transcriptomics, and proteomics data. As an example, in depth metabolic fingerprinting and LC-MS profiling of various parts of Asparagus racemosus led to the identification of a significant number of steroidal saponins exclusively present in roots. Transcriptome sequencing from three different tissues led to the identification of 321 different genes involved in saponin biosynthesis [6]. Similarly, two homologous GTs, itUGT1 and itUGT2 (86% similarity) have been identified based on peptide mass fingerprinting and previously described transcriptomics data of Indigofera tinctoria leaves [121].

A large number of GT genes have been obtained by using the above gene mining methods, however, there are still some challenges in narrowing down the scope of target GTs mining for specific PNPs glycosylation owing to the labor-intensive function characterization of individual enzymes [20]. Therefore, development of efficient and flexible enzyme function detection methods is crucial for obtaining comprehensive enzyme functional data. Some progress has been made recently, such as the development of a high-throughput screening method by using mass spectrometry [122]. Another publication has recently highlighted the potential for a fluorescence-based method to universally monitor the activity of GTs by detecting the nucleotides generated in a biochemical reaction [123].

In addition, the development of artificial intelligence provides a unique avenue for the identification of target enzymes through rapidly expanding sequence databases. Recent advances in deep learning models for feature extraction and patter recognition for sequence classification and functional prediction of enzymes in large datasets have also facilitated the discovery of novel GTs [124,125]. Yang et al. developed a chemical-bioinformatic model for functional prediction of uncharacterized GT1 family GTs based on sequence data of 54 GTs of A. thaliana and structural information of 91 candidate substrates. The model successfully identified novel substrates for GTs and enabled functional annotation of GTs from other sources including alfalfa, oats, and bacteria. Using enzyme sequences that did not rely on experimental data, this analysis provided meaningful biological insights that guided subsequent directed evolution and mechanistic studies of GT enzymes [126].

4. Engineering of GTs for biosynthesis of glycosylated PNPs

Wild-type GTs often lack properties desirable for the synthesis of glycosylated PNPs, such as high activity, high stereo- and regio-selectivity, and minimal undesirable activities and promiscuity towards valuable unnatural substrate [12]. However, structure-based rational design and directed evolution can efficiently improve the properties of GTs for industrial applications.

As of October 20, 2021, a total of 26 GT1 family GTs involved PNP glycosylation with solved crystal structures (among nearly 30,000 total GT sequences) are available in the CAZy database (Table 2). Increasing the availability of GT structures, especially in the presence of both donor and acceptor analogues, is essential for the continued success of guided rational engineering of GT mutants with improved or altered functions [13,14]. As an alternative strategy for GT engineering, directed evolution may offer the potential for altering GT specificity and/or fine-tuning the activity of rational GT chimeras [12,127,128]. Here, we summarize several recently reported protein engineering strategies to increase the catalytic activity, broaden the substrate spectrum, and alter the regioselectivity of GTs.

Table 2.

The structure-based rational design and directed evolution of GTs that involved in the biosynthesis of glycosylated plant natural products.

Protein name	Organism	Genbank	PDB code	Representative schematic reaction	Key residues	Engineering	Reference
YjiC	Bacillus subtilis	NP_389104.1	7BOV		●Ser277 is critical for Nucleoside Diphosphate (NDP) recognition ●Glu317, Gln318, Ser128 and Ser129 are crucial for glycosyl moiety recognition	●V108A increase 5-fold for UDP glycosylation activity and improve 35% for pterostilbene glycosylation under the existence UDPG. ●L320A improve 2-fold for Kcat/KM while add 65% activity for ADPG.	[129]
UGT72B1	Arabidopsis thaliana	CAB80916.1	2VCE		●His19 is positioned to act as a Brønsted base ●Gln389 and Glu388 interact with the glucose moiety of donors ●Glu83, Ile86, Leu118, Phe119, Phe148, Leu183, and Leu197 are predominant in the acceptor binding	–	[130]
UGT74F2	Arabidopsis thaliana	AAB64024.1	5U6M		●His18 shows a central role in catalysis ●Tyr180 is important for ligand recognition or binding ●Met274 could be crucial for orientation of the salicylic acid	–	[131]
UGT89C1	Arabidopsis thaliana	AAF80123.1	6IJ7		●Asp356, His357, Pro147 and Ile148 are key residues for sugar donor recognition and specificity for UDP-β-l-rhamnose. ●His21 is a key residue as the catalytic base and the only catalytic residue involved in catalysis.	●H357Q exhibited activity with both UDP-β-l-rhamnose and UDP-glucose	[132]
UGT78K6	Clitoria ternatea	BAF49297.1	3WC4		●Pro78, Asp181 and Asp367 are involved in the acceptor binding. ●Asn137 could participate in the recognition of the glucose moiety. ●His17 is the key catalytic residue.	–	[133,134]
UGT708C1	Fagopyrum esculentum	BAP90360.1	6LLG		●Asp382, Gln383, Thr151 and Thr150 play important role in the recognition of sugar moiety. ●Phe130, Tyr102 and Phe198 bound and stabilize the acceptor. ●Arg280 and Asp96 play important roles in the catalytic activity.	–	[135]
GgCGT	Glycyrrhiza glabra	QGL05036.1	6L5P		●R285, T145, D390 and Q391 determine the sugar donor preference. ●The flopropione unit is the minimum required unit for the di-C-glycosylation due to the interactions of its 2’-/6′-OH with H27. ●The spacious substrate-binding tunnel near G389 is critical for the di-C-glycosylation activity and the broad substrate promiscuity.	●G389K mutation could switch di- to mono-C-glycosylation	[48]
UGT73P12	Glycyrrhiza uralensis	BBN60799.1	7C2X		●Arg32 is the essential residue to provide high specificity for UDP-glucuronic acid.	–	[136]
LpCGTa	Landoltia punctata	QLF98869.1	6LG1		–	–	[76]
LpCGTb	Landoltia punctata	QLF98870.1	6LFN		–	–	[76]
SbCGTa	Scutellaria baicalensis	QLF98861.1	6LG0		●His24 is critical to initiate the catalytic reaction through deprotonation of the substrate.	–	[76]
SbCGTb	Scutellaria baicalensi	QLF98862.1	6LFZ		●His23 is critical to initiate the catalytic reaction through deprotonation of the substrate.	●R94 M/I143 M/V144T/T145S/H194D/G275T/P374Q mutant had switched the function of SbCGTb to SbCGTa	[76]
UGT708A6	Zea mays	ACF81582.1	6LF6		–	–	[76]
UGT71G1	Medicago truncatula	AAW56092.1	2ACV		●His22 is the catalytic base. ●Asp121 is a key residue that may assist deprotonation of the acceptor. ●Glu381 is the key residue in recognition of the sugar donor.	–	[2]
UGT85H2	Medicago truncatula	ABE87250.1	2PQ6		●His21 and Asp125 are essential for catalytic activity.	–	[137]
UGT78G1	Medicago truncatula	ABI94025.1	3HBF		●Glu192 is the key residue for the reverse reaction. ●His26 act as the catalytic residue. ●Asp124 also plays an essential role in catalysis.	–	[138]
Os79	Oryza sativa	BAF14158.1	5TMB		●His27 activate the trichothecene O3 hydroxyl for nucleophilic attack at C1’ of the UDP-glucose donor. ●Thr291 plays a critical role in catalysis as a catalytic acid or to position the UDP moiety during the nucleophilic attack.	–	[139]
PtUGT1	Persicaria tinctoria	BBB06426.1	5NLM		●E88 could play a major role in indoxyl specificity and turnover. ●H26 is expected to be the Brønsted base. ●D122 is believed to balance the charge on the catalytic histidine.	–	[140]
PaGT2	Phytolacca americana	BAG71125.1	6JEL		●His18 and His81 are recognized as the catalytic residues.	●C142A and C142F mutants formed resveratrol 3-O-β-glucoside and resveratrol 4′-O-β-glucoside, respectively, with high regioselectivity.	[42]
PaGT3	Phytolacca americana	BAG71127.1	6LZX		●His20 is the active-site residue. ●Trp417 and Arg419 are actively participate in the formation of the acceptor-binding pocket.	–	[141]
UGT51	Saccharomyces cerevisiae	AAB67475.1	5GL5		●Asp752 serve as a catalytic base. ●Met851 is important for UGT51 activity. ●Gln1094, Asp1093 and Ser1072 make several critical interactions with the glucose moiety of donor.	●A mutant M7_1 (S81A/L82A/V84A/K92A/E96K/S129A/N172D) presented an ∼1800-fold activity improvement toward an unnatural substrate protopanaxadiol.	[142,143]
UGT74AC1	Siraitia grosvenorii	AEM42999.1	6L8W		●His18 is the general base abstracts a proton from the 3-hydroxyl group of mogrol. ●Asp111 stabilizes the catalytic conformation and balance the charge.	●Mutant M4 (T79Y/R28H/L48 M/L109I) showed ∼200-fold higher activity than WT. ●Mutant M5 (T79Y/R28H/L48 M/L109I/S15A) showed ∼3.8-fold higher than M4.	[144]
UGT74AC2	Siraitia grosvenorii	AXK92493.1	7BV3		●The uracil ring forms hydrogen bonds and parallel π-stacking interactions with A353 and W352, respectively, and the ribose ring interacts with the enzyme through hydrogen bonds with E378 and Q355, while the α-phosphate forms hydrogen bonds with H370, N374 and S375 ●The acceptor binding pocket is constructed by 13 residues, where P12, L43 and V91 are located at the entrance, and V190, M196 as well as L200 are situated at the bottom	●Mutant G11Y is found that shows 75% selectivity and >99% conversion towards silybin A-3,7-O-diglucoside. ●Three variants show enhanced regioselectivity toward silybin A-7-O-glucoside, P12Y (81% selectivity and 68% conversion), L200W (92% selectivity and 61% conversion) and Y145W (89% selectivity and 75% conversion)	[145]
UGT76G1	Stevia rebaudiana	AAR06912.1	6INF		●His25 is the general base, which deprotonates the 3-hydroxyl of the accepting glucose A to activate it as a nucleophile. ●Asp124 plays an important catalytic role in relaying protons off and on His25.	–	[146]
TcCGT1	Trollius chinensis	QCZ42162.1	6JTD		●H24 acts to stabilize both the deprotonated substrate and the product sugar, though it is not indispensable for the glycosylation activity. ●E396 plays an important role to stabilize and orient the UDP–Glc sugar.	●I94E and G284K switch C- to O-glycosylation.	[49]
VvGT1	Vitis vinifera	AAB81683.1	2C1X		●Asp374, Gln375 and Thr141 are key players in sugar recognition.	–	[147]
UGTPg45	Panax ginseng	AKA44586.1	–		–	●A mutant UGTPg45-HV with two missense mutations (Q222H and A322V) gave 70% increase of ginsenoside Rh2 yield	[148]

Wild-type GTs often show low catalytic activity when expressed in heterogeneously engineered hosts, which in many cases can restrict the industrial-scale production of glycosylated PNPs. Recently, to address the inefficiency of the key ginseng GT, our group selected fifteen mutation hotspots in the receptor binding site of Saccharomyces cerevisiae UGT51 based on its crystal structure in complex with UDP-glucose. One mutant (S81A/L82A/V84A/K92A/E96K/S129A/N172D) showed an ∼1,800-fold increase catalytic activity in the conversion of protopanaxadiol (PPD) to ginsenoside Rh2 in vitro (Table 2) [143]. Similarly, Liu et al. engineered the GT Yjic from Bacillus subtilis for Rh2 synthesis. Using a semi-rational design that included structure-guided alanine scanning and saturation mutations, mutant M315F was found to efficiently synthesize Rh2 (∼99%) and block the further glycosylation of C12–OH [149].

An in vivo directed evolution strategy in which mutations were directly introduced into the chassis was also recently developed in an effort to enhance catalytic activity. The mutants were screened based on in vivo yield of target products, resulting mutated bioparts with improved enzymatic characteristics and performance in compatibility with chassis. Using this method, the poor performance of UGTPg45 in catalyzing the conversion of its unnatural substrate, PPD to ginsenoside Rh2, was improved. As a result, the UGT mutant UGTPg45-HV was acquired which carried two missense mutations (Q222H and A322V) that conferred a 70% increase in ginsenoside Rh2 yield [148]. Recently, our group developed an ultrahigh-throughput dual-channel microfluidic droplet screening system and a fluorescence-activated cell sorting system that both enabled the high-throughput screening (>10⁷ mutants/hour) of mutants, which can contribute to the effective engineering of high activity GTs [150,151].

Recently, the promiscuous substrate specificity of GTs has inspired studies exploring how to effectively design biocatalysts for efficient and directed biosynthesis of bioactive glycosides. Structure-guided mutagenesis was conducted to alter the catalytic specificity of C-/O-glycosylation by TcCGT1. As a C-glycosyltransferase (CGT) from the medicinal plant Trollius Chinensis, TcCGT1 can catalyze the 8-C-glycosylation of 36 different flavonoids and the O-glycosylation of diverse phenolics. The spacious binding pocket characterized using its crystal structure in complex with uridine diphosphate explains its substrate promiscuity, with the substrate binding pose determining its C- or O-glycosylation activity. Site-directed mutagenesis at two residues (I94E and G284K) enabled the conversion from C- to O-glycosylation [49]. Other studies have also successfully switched sugar donor preference and acceptor substrates using either single- or multiple-point mutations not exclusively located within the binding site (Table 2) [48,76,132]. Based on the successful GT engineering cases, the specificity for the donor substrate, especially the nucleotide residue, is largely determined by the highly conserved PSPG motif in the C-terminus. Besides, mutations in the residues involved in substrate recognition may change the preference of glycosylation.

Regioselectivity is also a typical problem that must be considered in the synthesis of structurally diverse glycosides [152]. Recent studies of plant GTs have shown that point mutations can alter regioselectivity due to flexibility in the GT substrate binding pocket [153]. Fan et al. demonstrated the successful switching of regioselectivity by UGTBL1 from Bacillus licheniformis to yield polydatin (resveratrol 3-O-β-glucoside) instead of resveratrol 4′-O-β-glucoside polydatin, a compound used to relieve the toxic side effects of cisplatin and treat acute severe hemorrhagic shock. To this end, a 3D model of UGTBL1 was constructed, and residue Ile62 was found to significantly influence its regioselectivity. Mutation I62G ultimately led to the switch in regioselectivity from 4′-OH to 3-OH of resveratrol, with a roughly sevenfold increase in the formation of the preferred polydatin over that of 4′-O-glucoside compared to wild type [154]. Recently, Sun and co-workers tuned a newly identified GT from Siraitia grosvenorii (UGT74AC2) to serve as the catalyst of targeted regioselective glycosylation of the polyhydroxy substrate silybin and derivatives. Three single-site mutants (P12Y, L200W and Y145W) showed 94%, >99%, and >99% selectivity on the 3-OH, 7-OH and 3,7-O-diglycoside of the substrates, respectively, compared with that of wild type, which produced a 22%:39%:39% product mixture [132].

5. GT applications in glycosylated PNP biosynthesis

Emerging synthetic biology strategies are rapidly expanding the application of GTs in the synthesis of glycosylated PNPs, or their desirable precursors, in microbial fermentation systems [[155], [156], [157], [158]]. Here, we discuss some breakthrough of GT-based biosynthesis of glycosylated PNPs at industrial scale in model hosts like Escherichia coli and S. cerevisiae (Table 3). These representative products include flavonoid glycosides (fisetin 3-O-glycosides, astragalin, scutellarein 7-O-glucoside), terpenoid glycosides (rebaudioside A, ginsenosides), and polyketide glycosides (salidroside, polydatin).

Table 3.

The recent glycosylated plant natural products synthesis by microbial sources.

Compound	Microbial sources	GTs	Titer	Reference
Cyanidin 3-O-glucoside	E. Coli	3 GT	0.35 g/L	[159]
Luteolin-7-O-glucuronide, Quercetin-3-O-glucuronide, Quercetin 3-O-galactoside	E. Coli	AmUGT10, VvUGT, PhUGT	0.30, 0.69, 0.28 g/L	[160]
Fisetin 3-O-glucoside, Fisetin 3-O-rhamnoside	E. Coli	UGT78K1, ArGt-3	0.39, 0.34 g/L	[161]
Quercetin 3-O-galactoside, Quercetin 3-O-rhamnoside	E. Coli	RhaGT	0.94, 1.12 g/L	[162]
Scutellarein 7-O-glucoside	S. cerevisiae	SbGT34	1.20 g/L	[108]
Kaempferol 3-O-glucoside	E. Coli	AtUGT78D2	3.60 g/L	[163]
Rebaudioside A	S. cerevisiae	UGT76G1	1.16 g/L	[164]
Tyrosol glucoside	E. Coli	UGT72B14	6.7 mg/L	[165]
Ginsenoside Rh1, Ginsenoside F1	S. cerevisiae	UGTPg1, UGTPg100	0.10, 0.04 g/L	[166]
Ginsenoside Rh2, Ginsenoside Rg3	S. cerevisiae	UGTPg29, UGTPg45	0.02, 0.05 g/L	[30]
Ginsenoside Rh2	S. cerevisiae	UGT51	0.30 g/L	[143]
3β,12β-Di-O-Glc-PPD, -PPT, DM	S. cerevisiae	UGT109A1	9.05, 4.57, 11.5 mg/L	[167]
3β-O-Glc-DM	S. cerevisiae	UGT74AE2	5.60 g/L	[168]
Ginsenoside Rh2, PPD, DM	S. cerevisiae	UGTPn50	2.25, 9.05, 8.09 g/L	[148]
Ginsenoside Compound K	Yarrowia lipolytica	UGTPg1	0.16 g/L	[169]
Ginsenoside Compound K	S. cerevisia	UGTPg1	5.74 g/L	[170]
Ginsenoside Rg1, Notoginsenoside R1, Notoginsenoside R2	S. cerevisia	PgUGT71A53, PgUGT94Q13, PgUGT71A54	1.95, 1.62, 1.25 g/L	[171]
Crocetin	E. coli	YjiC, YdhE, YojK	4.42 mg/L	[172]
Kaempferol, astragalin	E. coli	AtUGT78D2	1.18, 1.74 g/L	[173]
Geranyl glucoside	E. coli	VvGT14a	0.93 g/L	[174]
Glycyrrhizin and Glycyrrhetic acid 3-O-mono-β-D-glucuronide	S. cerevisiae	UGT1A1	5.98, 2.31 mg/L	[175]
Polydatin	S. cerevisiae	PcR3GAT	0.55 g/L	[176]

Fisetin glycoside is a medicinally important flavonoid glycoside produced by various plants that has been reported to exhibit diverse medicinal effects such as prevention of cardiovascular diseases, antioxidant activity, anti-diabetic activity, and anticancer activity [[177], [178], [179]]. The regiospecific GT (UGT78K1) from Glycine max or ArGt-3 from A. thaliana were introduced into E. coli BL21 (DE3), along with the respective UDP-glucose and TDP-rhamnose biosynthetic genes from different bacterial sources, in order to achieve the bioconversion of fisetin. Approximately 1.18 g of fisetin 3-O-glucoside and 1.03 g of fisetin 3-O-rhamnoside were produced in a 3 L bioreactor [161].

Similar to this accomplishment, the GT (AtUGT78D2) from A. thaliana and a highly efficient UDP-glucose synthesis pathway were introduced into E. coli BL21 (DE3) to produce another flavonoid glycoside, astragalin, which resulted in maximal astragalin production of 1.74 g/L [173]. Other than E. coli, S. cerevisiae has also been used for flavonoid glycoside biosynthesis. Successful deletion of glucosidases in S. cerevisiae in conjunction with overexpression of the flavonoid GT SbGT34 from Scutellaria baicalensis enabled production of the medicinal compound scutellarein 7-O-glucoside. The feasibility of scaling in vivo glycosylation was demonstrated by large-scale production of 1.20 g/L scutellarein 7-O-glucoside by optimization of the appropriate fermentation conditions [108].

As the sweetest terpenoid glycoside from Stevia rebaudiana, rebaudioside A is commercially significant as a natural sweetener used in the food and beverage industry [180,181]. To produce rebaudioside A in yeast, UGT76G1 from S. rebaudiana was overexpressed under the control of the PGK1 promoter. At the same time, the Nocardia farcinica phosphoglucomutase gene, nfa44530, which participates in UDP-glucose synthesis, and the E. coli K12 glucose-1-phosphate-1 uridylyltransferase gene, galU, were also co-expressed in the recombinant S. cerevisiae. By optimizing the availability of UDP-glucose, rebaudioside A production reached 1.16 g/L [164].

As a group of glycosylated triterpenes found in Panax species, ginsenosides are synthesized from 2,3-oxidosqualene through the universal precursors, dimethylallyl diphosphate (DMAPP) and isopentenyl diphosphate (IPP) [182]. An approach for ginsenoside production was recently developed for S. cerevisiae. Following the chromosomal integration of a PPD biosynthetic pathway in yeast, UGTPg45 and UGTPg29 from P. ginseng were introduced into the recombinant cells, enabling production of the Rh2 and Rg3 ginsenosides. However, Rh2 production was relatively low (16.95 mg/L in shaken flasks) for commercialization due to the poor performance of the GT (UGTPg45) [148]. Based on these findings, our group built an efficient ginsenoside Rh2 biosynthetic cell factory by repurposing an inherently promiscuous GT (UGT51 mutant) from S. cerevisiae with an 1800-fold increase in catalytic efficiency over wild type. This strain harboring the engineered GT could produce 0.3 g/L of ginsenoside Rh2 in a 5 L fed-batch fermentation system [143]. Notably, Zhou's group successfully constructed yeast strains that could produce ginsenoside CK, ginsenoside Rg1, notoginsenoside R1, and notoginsenoside R2 by introducing a group of GTs including UGTPg1, PgUGT71A53, PgUGT94Q13, and PgUGT71A54. De novo production of these ginsenosides reached 5.74, 1.95, 1.62, and 1.25 g/L, respectively [170,171].

As one of the major polyketide glycosides in Rhodiola, salidroside (the 8-O-β-glucoside of tyrosol) has been purported to confer adaptogenic and ergogenic effects [183]. Xue et al. reported salidroside production through the expression of Rhodiola UGT72B14 in E. coli. Codon optimization resulted in significantly enhanced salidroside accumulation, reaching 6.7 mg/L, a 3.2-fold increase over that of wild-type GT [170].

Polydatin is a well-known pharmaceutical polyketide glycoside that provides anticancer, antiaging, and anti-inflammatory effects [[184], [185], [186]]. Recently, Liu et al. explored the development of a microbial chassis for polydatin production that could potentially replace plant extraction in future systems. This work also identified a key enzyme for polydatin biosynthesis, resveratrol GT, PcR3GAT. Polydatin production thus reached 0.54 g/L through the incorporation of a resveratrol biosynthesis module, UDP-glucose supply module, and GT expression module and subsequent optimization of fermentation conditions [176].

6. Conclusion and future perspectives

Glycosylation is one of the most important physiological and biochemical reactions in nature, given its crucial roles in a multitude of essential processes. This intrinsic importance has attracted long-standing and wide research attention into the characteristics of GTs to facilitate their applications in glycosylation reactions for metabolic engineering of natural product biosynthesis. While several GTs have been found to be suitable for altering glycosylation patterns, generally low catalytic activity and stringent substrate specificity remains a limiting factor in the diversification of PNPs for industrial fermentation systems. The data mining and engineering GTs are still the most promising approaches for discovering and developing novel enzymes with well-defined characteristics. High-throughput sequencing has enabled efforts to comprehensively profile the genomes, transcriptomes, and proteomes of plant species with important medicinal, industrial, or scientific applications. Furthermore, this sequencing data corroborate with functional data that have fed artificial intelligence-based computational approaches to GT discovery.

A series of discoveries have been made recently, using principles of synthetic biology, in the field of PNP glycoside biosynthesis catalyzed by GTs. However, there remains some challenges to engineering industrial hosts [187]. The metabolic engineering needed to produce a particular PNP glycoside relies on the biosynthetic routes of PNP-precursors, however these details are frequently unavailable or incomplete. In this case, candidate pathway design, enzyme selection, and pathway testing all bring different challenges [8]. More efforts should be made to reveal the complexity of natural PNP pathways.

Another obstacle present in GT application in the synthesis of PNP glycosides is that when introducing GTs into a heterologous host, they may function sub-optimally or not at all for reasons that include low expression or activity, improper folding, and mislocalization. Therefore, modifying the function of GTs need for glycosylated PNP synthesis, and tuning the biosynthetic systems to improve the yields of these natural products, is an increasing research priority, given the urgent need for affordable, effective drugs that are inefficiently produced in nature. Since traditional GT activity assays are not suitable for rapid detection, a central goal of future research on GTs is to develop universal high-throughput detection methods, such as qualitative mass spectrometry-based assays or fluorescence-based assays for rapid screening of target GTs. In addition, further structural elucidation of GTs will help to increase our understanding of the catalytic mechanism of these enzymes. The crystal structures validated by experimental methods, or predicted by high accurate artificial intelligence methods, e.g. AlphaFold2 [188] and RoseTTAFold [189], will accelerate the elucidation of the structure-function relationship. In particular, this involves identifying which hot domains or motifs of the protein affects its structure and the associated enzymatic activity, which will further guide the rational design of GTs.

Moving forward, synthetic biology-driven metabolic engineering of different desirable GT characteristics combine with relevant components of synthesis pathways will reinvigorate current efforts to increase the diversity of PNP glycosides used in industrial production for medicine, functional foods, and cosmetics.

CRediT authorship contribution statement

Bo He: Conceptualization, Writing – original draft, Data curation, Validation. Xue Bai: Writing – original draft, Data curation. Yumeng Tan: Investigation, Data curation. Wentao Xie: Investigation, Data curation. Yan Feng: Supervision. Guang-Yu Yang: Funding acquisition, Supervision, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no conflict of interests.