Biosynthesis and metabolic engineering of natural sweeteners

Abstract

Natural sweeteners have attracted widespread attention because they are eco-friendly, healthy, low in calories, and tasty. The demand for natural sweeteners is increasing together with the popularity of green, low-carbon, sustainable development. With the development of synthetic biology, microbial cell factories have emerged as an effective method to produce large amounts of natural sweeteners. This technology has significantly progressed in recent years. This review summarizes the pathways and the enzymes related to the biosynthesis of natural sweeteners, such as mogrosides, steviol glycosides, glycyrrhizin, glycyrrhetinic acid, phlorizin, trilobatin, erythritol, sorbitol, mannitol, thaumatin, monellin, and brazzein. Moreover, it focuses on the research about the microbial production of these natural sweeteners using synthetic biology methods, aiming to provide a reference for future research on the production of natural sweeteners.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13568-025-01864-y.

Introduction

Excessive intake of sugar is a common problem in people’s daily diet. Long-term consumption of high-sugar diets can cause diabetes, cardiovascular diseases, and oral diseases (Mora and Dando 2021; Bilal et al. 2022). The Food and Agriculture Organization (FAO) reported that sugar intake has been steadily increasing in developing countries, with a projected increase of 32 tons in 2028 compared with its approximate value of 203 tons in 2008 (Castro-Muñoz et al. 2022). Meanwhile, increasing health awareness has led to consumers demanding low-calorie and safer sugar substitutes, such as sweeteners. Sweeteners are additives that give a sweet taste to food and are classified into natural and artificial sweeteners (Saraiva et al. 2020; Nicholas Chua et al. 2023). Currently, sweeteners are widely used as important additives in food and beverages to satisfy people's demand for a sweet taste (Farag et al. 2022; Jeong et al. 2024). However, the safety and long-term health effects of artificial sweeteners remain controversial (Iizuka et al. 2022; Saraiva et al. 2023).

Artificial sweeteners have become a new type of pollutant because their widespread use has generated large amounts of industrial wastewater containing high concentrations of artificial sweeteners (Zeng et al. 2024), which markedly affect the gut microbial composition of earthworms (Lin et al. 2024) and cause anxiety in fish (Colín-García et al. 2023). In addition to environmental pollution, long-term overconsumption of artificial sweeteners can seriously affect health. For instance, excessive consumption of the artificial sweetener acesulfame potassium increases the risk of central precocious puberty (Wu et al. 2024). Chronic overconsumption of artificial sweeteners is also associated with metabolic syndrome, obesity (Pearlman et al. 2017; Yan et al. 2022), and a significant increase in the incidence of cardiovascular disease (Mossavar-Rahmani et al. 2019; Debras et al. 2022).

Consequently, natural sweeteners have attracted great attention because they are environment-friendly, healthy, low-calorie, and tasty. Natural sweeteners are naturally occurring compounds, mostly secondary metabolites of plants or microorganisms, that possess a strong sweet taste (Pomon et al. 2023; Chai et al. 2024). Currently, the commonly used sugar-based natural sweeteners can be categorized as follows: terpenoids (mogrosides and steviol glycosides, etc.) (Biswas et al. 2024; Zhang et al. 2023), flavonoids (phlorizin and trilobatin, etc.) (Ortiz et al. 2022), polyols (erythritol, sorbitol, mannitol, etc.) (Gauthier et al. 2024), and sweet proteins (thaumatin, monellin, brazzein, etc.) (Novik et al. 2023). The safety, low-sugar, and low-calorie characteristics of natural sweeteners make them beneficial in the management of hyperglycemia and obesity (Mejia and Pearlman 2019; Wölnerhanssen et al. 2020; Almiron-Roig et al. 2023). For instance, xylitol is a good sucrose substitute for diabetic patients (Xu et al. 2019). Additionally, natural sweeteners have better pharmacological effects. Mogrosides can mitigate wet lung, cough, and ovalbumin-induced lung inflammation (Liu et al. 2021; Dou et al. 2022). Steviol glycosides (SGs) can alleviate lipid metabolism abnormalities in patients with obesity and diabetes (Kurek et al. 2023). Trilobatin has been demonstrated to be useful in treating liver diseases (Zhang et al. 2022b; Hou et al. 2023). Thus, natural sweeteners are expected to have great market potential in the future.

Large-scale extraction of natural sweeteners from plants and their chemical synthesis are complex, expensive, low-yielding, and environment-unfriendly. Additionally, some sweeteners have complex structures, require stringent reaction conditions, and are prone to produce unwanted byproducts (Xu et al. 2019). Enhancing the yield of natural sweeteners is highly important if their future commercial demands are to be met (Regnat et al. 2018). Toward that end, microbial cell factory synthesis of natural sweeteners has become popular because it exhibits high product conversion, high specificity, and environmental friendliness (Qu et al. 2023). In recent years, the rapid development of genetic programming tools has enabled target genes to be easily and quickly transferred into target microorganisms. The advent of omics technologies, such as genomics, proteomics, and transcriptomics, and the continuous advancement of biosynthesis technologies have made the biosynthesis of natural sweeteners possible. They have also paved the way for the reconstruction of biosynthetic pathways in microbial cells to produce downstream products with more complex structures. Currently, natural sweeteners have been biosynthesized in different microbial chassis, including Saccharomyces cerevisiae, Escherichia coli, and Yarrowia lipolytica (Gold et al. 2018; Joseph et al. 2019).

This study reviews the biosynthetic pathways of natural sweeteners, such as mogrosides, SGs, glycyrrhizin, glycyrrhetinic acid, phlorizin, trilobatin, erythritol, sorbitol, mannitol, thaumatin, monellin, and brazzein, as well as the related enzymes. It also reviews the latest research on the biosynthesis of 2′-fucosyllactose (2′-FL), which has been described as a natural sugar substitute in several reports (Xu et al. 2022c; Li et al. 2023a). Moreover, it summarizes the microbial biosynthesis of these natural sweeteners in different microorganism chassis. This review will provide a reference for further research on the biosynthesis of natural sweeteners.

Biosynthetic pathways of natural sweeteners

Terpenoids

The terpenoid natural sweeteners mainly include SGs, mogroside, glycyrrhizin (GL) and glycyrrhetinic acid (GA), and the biosynthesis of all such natural sweeteners begins with isopentenyl pyrophosphate (IPP) and dimethylallyl diphosphate (DMAPP), which can be generated through the methylerythritol phosphate (MEP) or the mevalonate (MVA) pathway. Pyruvate and glyceraldehyde 3-phosphate, and acetyl-CoA are the initial precursors of MEP and MVA pathways, respectively, and both the precursors are catalyzed by successive steps in these two pathways to form IPP and DMAPP. Then different numbers of the two five-carbon units are condensed to produce the substrates of terpenoids, such as diterpenes and triterpenes.

SGs

The leaves of Stevia rebaudiana contain more than 30 different SGs, of which stevioside and rebaudioside A (Reb A) are the most abundant. Together, they are primarily responsible for the sweet taste of stevia (Arumugam et al. 2020). Stevioside is approximately 270–300 times sweeter than sucrose and has proven to be a valuable natural sweetener due to its chemical stability and good sweet taste (Arumugam et al. 2020).

In S. rebaudiana, catalyzed by farnesyl pyrophosphate synthase (FPS) and geranylgeranyl pyrophosphate (GGPP) synthase, IPP and DMAPP sequentially condense to generate farnesyl pyrophosphate (FPP) and GGPP. ent-Copalyl pyrophosphate (ent-CPP) synthase catalyzes the cyclization of GGPP into ent-CPP, which undergoes an electrocyclic reaction catalyzed by ent-kaurene synthase (ent-KS) to form ent-kaurene. The sequential action of two cytochrome P450 (CYP450) enzymes, ent-kaurene oxidase (ent-KO) and ent-kaurenoic acid 13-hydroxylase (ent-KAH), lead to the generation of ent-kaurenoic acid and, finally, steviol, which is produced when ent-KAH hydroxylates ent-kaurenoic acid at the C13 position. Various SGs can be generated via the action of glycosyltransferases (GTs) (Ceunen and Geuns 2013). Kaurene can be generated in S. cerevisiae after the transfer of genes encoding GGPPS7 and tCPPS5 from S. rebaudiana, and AtKS5 form Arabidopsis thaliana. Subsequent transfer of genes encoding KO75 (SrCYP701 from S. rebaudiana), KAH82 (SrCYP72 from S. rebaudiana), and nicotinamide adenine dinucleotide phosphate hydrogen (NADPH)-dependent CYP450 reductase (SrCPR1) from S. rebaudiana resulted in the efficient production of steviol (Gold et al. 2018). UGT73C1, a UDP-dependent glycosyltransferase (UGT) from A. thaliana, specifically binds steviol to generate the diterpenoid glucoside steviolmonoside (Zhou et al. 2018).

Rubusoside (Rub), also a highly sweet diterpenoid saponin and high-intensity sweetener extracted from the leaves of Rubus suavissimus and Angelica keiskei, is a trace component of SGs that is present in S. rebaudiana. In S. rebaudiana, UGT85C2 preferentially catalyzes the 13-O-glucosylation of steviol to generate steviolmonoside, which is further glycosylated at the C19 position by UGT74G1 to generate Rub (Mao et al. 2021). UGT73E1 can also convert steviolmonoside into Rub (Li et al. 2018). In the meanwhile, RsUGT75L20 and AkUGT75L21 from R. suavissimus and A. keiskei, respectively, can convert steviol into steviol-19-O-β-D-glucopyranoside, which can further be converted to Rub by RsUGT85A57 and AkUGT85A58 (Sun et al. 2018).

The biosynthesis of steviolbioside, stevioside, Reb A and Reb E is based on steviolmonoside. UGT91D2, encoding steviol-13-monoglucoside-1,2-glucosyltransferase, can convert steviolmonoside to steviolbioside. Multiple UGT91D2 sequences from different Stevia varieties were cloned and expressed, and the variant UGT91D2w showed the highest activity in converting steviolmonoside to steviolbioside (Wang et al. 2016). Then three UGT family members (UGT85C2, UGT74G1, and UGT76G1) from S. rebaudiana are involved in the biosynthesis of stevioside and Reb A. First, UGT85C2 glycosylates steviol at C13 to generate steviolmonoside. The C2 site on the 13-O-glucose of steviolmonoside is further glycosylated to generate steviolbioside, which is glycosylated at C19 by UGT74G1 to generate stevioside. Afterwards, UGT76G1 glycosylates the 13-O-glucose of stevioside at the C3 site to produce Reb A (Brandle, and Telmer 2007). And UGTSL2, a UGT from Solanum lycopersicum, produces Reb E by catalyzing the formation of a β1-2 glycosidic bond at the C19 position of the glycosyl group of stevioside (Chen et al. 2021).

Reb D and Reb M, which are sweeter than Reb A, have very low yields in their natural state. OsUGT91C1, a UGT from Oryza sativa, was modified by semi-rational design to efficiently catalyze β1-2 glycosylation at positions C13 site and C19 site. Furthermore, the addition of β1-6 glucose was eliminated, which allowed for greater yields of Reb D and Reb M. OsUGT91C1 catalyzes the β1-2 glycosylation of Reb A to generate Reb D, which further undergoes β1-3 glycosylation catalyzed by UGT76G1 to generate Reb M (Lee et al. 2019b; Zhang et al. 2021a). UGTSL2 from S. lycopersicum can also facilitate the formation of a β1-2 glycosidic bond at the C19 site of Reb A to generate Reb D (Chen et al. 2018).

Mogroside

Siraitia grosvenorii is a widely used natural sweetener and traditional medicine for treating respiratory diseases (Zhao et al. 2022). Mogrosides are a class of cucurbitane-type triterpenoid saponins, and some of them are used as nature sweeteners, such as Mogroside V. They are isolated from S. grosvenorii and are sweeter than stevioside (Qiao et al. 2019a, b). Mogrosides can taste sweet only if they have an α-hydroxyl at the C11 site and a mini-mum of three glycosyls. Compounds without a hydroxyl or with a keto-group at the C11 position lose their sweet taste and turn bitter. Siamenoside I (SI) is considered to be the sweetest mogroside despite having only four glycosyls: one at the C3 position and a tri-saccharide chain at the C24 position (Cicek 2020).

The biosynthesis of mogrosides begins with squalene epoxidation catalyzed by squalene epoxidase (SQE), first generating 2,3-oxidosqualene. Then, it is continued epoxidation by SQE to generate 2,3;22,23-diepoxysqualene. This intermediate undergoes cyclization catalyzed by cucurbitadienol synthase (CS) to produce 24,25-epoxycucurbitadienol. Next, a tetra-hydroxylated cucurbitane (mogrol) is generated through continuous hydroxylation by epoxide hydrolase (EPH) and CYP450. Another synthetic pathway also exists for mogrol. SgCbQ, a cucurbitadienol synthase from S. grosvenorii, can catalyze the conversion of 2,3-oxidosqualene to cucurbitadienol (Dai et al. 2015), which is further catalyzed by CYP87D18 to produce 11-hydroxyl cucurbitadienol. Then mogrol is generated under the action of other CYP450s (Zhang et al. 2016). Using homology modeling and site-directed mutagenesis studies, researchers engineered CYP87D20 variants that can effectively hydroxylate cucurbitadienol at C11 site to produce 11-hydroxyl cucurbitadienol, which participates in mogrol synthesis (Li et al. 2019; Zhang et al. 2021c). SgCPR1 and SgCPR2 from S. grosvenorii may be involved in mogroside biosynthesis and act as redox partners for different enzymes, such as SQE, EPH, and CYP450 (Zhao et al. 2018). Finally, a series of mogrosides is generated through glycosylation modification executed by GTs.

Mogroside I-A1 (MI-A1) is produced by glycosylation of mogrol at C24 catalyzed by UGT720-269-1 in S. grosvenorii, and further glycosylation at C3 (relative to MI-A1) can generate MII-E. UGT94-289-3 can glycosylate MII-E to produce MIIIX. Additionally, it can further glycosylate MIIIX to generate mogroside MIV-A and SI, respectively. Furthermore, it continues to add glycosyl group to the glycosyl chain at C24 site of MIV-A and C3 position of SI to generate MV (Itkin et al. 2016). UGT94-289-3 can also sustain catalytic glycosylation with MII-E as a substrate, ultimately resulting in the formation of SI, MIV, MV, and MVI (Cui et al. 2023). For the biosynthesis of MI-E, a new UGT, UGT74AC1, was identified in S. grosvenorii, and in vitro enzyme activity analysis revealed that UGT74AC1 can glycosylate mogrol at its C3 hydroxyl l site to form MI-E (Dai et al. 2015).

Besides, MV can be efficiently hydrolyzed via β-glucosidases to produce higher-value mogrosides. For instance, β-glucosidases from S. cerevisiae can selectively convert MV to MIII-E. Mogroside biotransformation primarily involves the hydrolysis of β1-6 glycosidic bonds at C3 and C24 positions of mogrosides. The β-glucosidase Exg1 hydrolyzes the β1-6 glycosidic bonds at C3 and C24 site of MV to generate SI and MIV, respectively. Exg1 further hydrolyzes the β1-6 glycosidic bonds of SI and MIV to produce MIII-E (Chiu et al. 2013). A β-glucosidase that can hydrolyze MV into SI was also screened in the brewer's yeast Dekkera bruxellensis (Wang et al. 2019a, b). An efficient β-glucosidase from the mycelia of Ganoderma lucidum can convert MV to MIII-E (Chiu et al. 2020).

GL and GA

GL and GA are triterpenoid natural sweeteners, and their biosynthesis begins with squalene. SQE catalyzes the epoxidation of squalene to 2,3-oxidosqualene, as the first oxidation reaction in the triterpene biosynthesis pathway. β-amyrin synthase (β-AS) catalyzes the cyclization of 2,3-oxidosqualene to generate β-amyrin, which is an important branch step in the biosynthesis of GA (Chen et al. 2013). Subsequently, cytochrome P450s participated in the subsequent reactions. Cytochrome P450 monooxygenase CYP88D6 catalyzes the two consecutive steps of the oxidation of β-amyrin at C11 to generate 11-oxo-β-amyrin, and CYP72A154 continues to catalyze the next three consecutive steps of oxidation of 11-oxo-β-amylin at C-30 to generate GA (Seki et al. 2008, 2011). In addition, studies have found that Uni25647 in licorice has the same catalytic function of CYP88D6 with higher oxidative activity, while another member of the CYP72A family, CYP72A63, has the same function with CYP72A154 (Zhu et al. 2018). Finally, UDP-glycosyltransferase 3 in Glycyrrhiza uralensis catalyzes its glucuronidation reaction using GA as a substrate, sequentially transferring two glucuronides to the C3 position of GA to generate GL (Xu et al. 2016).

Sweet proteins

Thaumatin

Derived from a tropical Marantaceae plant called T. daniellii, thaumatin is a natural sweetener that has substantial market demand today. Thaumatin exists in four forms: thaumatin I, thaumatin II, thaumatin A, and thaumatin B (Joseph et al. 2019). The complete amino acid sequence of thaumatin I was obtained by direct amino acid sequencing of purified thaumatin (Iyengar et al. 1979). Additionally, some studies analyzed and obtained the nucleotide sequence of thaumatin II, which differs from thaumatin I by five amino acids (Edens et al. 1982; Lee et al. 1988). Resequencing of purified thaumatin proteins led to the identification of two new variants, thaumatin A and thaumatin B, whose sequences were analyzed (Lee et al. 1988). Based on the nucleotide sequence of thaumatin II, a new sequence was cloned from a cDNA library and was compared and confirmed to be that of thaumatin I. Subsequently, a vector was developed using the cDNA of thaumatin I, and the production of this protein was realized in Pichia pastoris (Ide et al. 2007a, b).

Thaumatin mutants were engineered to identify the key residues responsible for the sweet taste of thaumatin I. Alanine substitution experiments revealed that four lysine residues (K49, K67, K106, and K163) and three arginine residues (R76, R79, and R82) play an important role in conferring a sweet taste upon thaumatin I, with K67 and R82 being particularly critical. Additionally, the sweetness threshold of the R82K mutant was higher than that of thaumatin I, corroborating that R82 can affect the sweet taste of thaumatin I (Ohta et al. 2008). The structure and properties of thaumatin II were determined at a resolution of 0.99 Å. The side-chain root means square deviations (RMSD) values of R67 and R82 were high, indicating the highly disordered nature of these residues. The flexible conformation of these two critical residues facilitates their interaction with the sweet taste receptor, which is a distinctive feature of thaumatin II with a strong sweet taste (Masuda et al. 2014).

Monellin

Monellin consists of two independent peptide chains, 50- and 42-amino-acids long (Bohak and Li 1976), whose sequences were determined using a sequenator (Frank and Zuber 1976). The correct three-dimensional structure of monellin is a prerequisite for its sweet taste. At high temperatures, the two chains in monellin, which are held together by noncovalent forces, dissociate and lose their original structure. Therefore, monellin loses its sweetness under acidic pH at temperatures above 50℃. Single-chain monellin (MNEI) can be obtained by directly linking the two chains via an amide bond or by inserting a dipeptide. For example, the C-terminus of the B-chain can be connected to the N-terminus of the A-chain via a Gly-Phe dipeptide linker (Kim et al. 1989; Picone and Temussi 2012; Kaushik and Udgaonkar 2023). MNEI is as sweet as natural monellin, while exhibiting greater thermal stability.

Brazzein

Brazzein was isolated from the fruit of Pentadiplandra brazzeana, and its amino acid sequence was obtained by sequence determination (Ming and Hellekant 1994). Comprising only 54 amino acids, brazzein is the most promising replacement for sugar (Nicholas Chua et al. 2023). Its sweet taste is mainly influenced by the residue at position 53 (Lim et al. 2016). Brazzein predominantly contains a pyroglutamic acid residue (pGlu) at its N-terminus, whereas the minor form that lacks pGlu (des-pGlu-Brazzein) is sweeter (Kazemi-Nasab and Shahpiri 2020).

Polyols

Xylitol

The biosynthesis of xylitol begins with the pentose phosphate pathway (PPP). Glucose-6-phosphate (G6P) dehydrogenase converts G6P into 6-phosphoglucono-δ-lactone; this is transformed into 6-phosphogluconate by 6-phosphogluconolactonase; this is converted into ribulose-5-phosphate by 6-phosphogluconate dehydrogenase; and finally, xylulose-5-phosphate is generated through the action of ribulose-phosphate-3-epimerase. Xylulose-5-phosphate gets dephosphorylated by xylose phosphatase to generate xylulose, which is transformed into xylose isomerase (xylA) (Yin et al. 2021). Finally, xylitol is produced under the action of an NADPH-dependent xylose reductase (xyrA) (Li et al. 2021).

Erythritol

Glucokinase converts glucose into G6P, which is converted into F6P (fructose-6-phosphate) by phosphoglucose isomerase. F6P then is converted to erythrose-4-phosphate and acetyl phosphate by F6P phosphoketolase (Rice et al. 2020). Erythrose-4-phosphate phosphatase dephosphorylates erythrose-4-phosphate to generate erythrose, which is eventually converted into erythritol by erythrose reductase (Erian and Sauer 2022). Alternatively, erythritol-4-phosphate dehydrogenase can transform erythrose-4-phosphate into erythritol-4-phosphate, which can produce erythritol when acted upon by a phosphatase (Rice et al. 2020).

Mannitol

F6P gets dephosphorylated by hexokinase to generate fructose that is finally reduced to mannitol dehydrogenase (MDH). In most prokaryotes, nicotinamide adenine dinucleotide (NAD +) serves as a cofactor for MDH, whereas in most eukaryotic fungi and yeasts, NADPH acts as a cofactor for this enzyme (Martínez-Miranda et al. 2022). Alternatively, mannitol-1-phosphate dehydrogenase (M1PDH) can convert F6P to mannitol-1-phosphate (M1P), which subsequently gets dephosphorylated by M1P phosphatase to produce mannitol (Gonçalves et al. 2019; Erian and Sauer 2022).

Flavonoids

Phlorizin (phloretin-2′-O-glucoside) and its positional isomer trilobatin (phloretin-4′-O-glucoside) are classified as dihydrochalcones (DHCs), which are natural, low-calorie sweeteners (Shang et al. 2022; Wang et al. 2023b). Phlorizin and trilobatin are produced via the phenylpropanoid pathway, one of the most widely studied specialized metabolic pathways, whose first three reactions are collectively referred to as the general phenylpropanoid pathway. These include the eventual conversion of phenylalanine (produced by the shikimate pathway) to p-coumaroyl-CoA via the catalysis of three enzymes: phenylalanine ammonia lyase (PAL), cinnamate-4-hydroxylase (C4H) and 4-coumarate CoA ligase (4CL) (Dong and Lin 2021; Lin et al. 2022b). DHCs are generated through side-chain reactions in the phenylpropanoid pathway (Wang et al. 2023b). Hydroxycinnamoyl-CoA double-bond reductase converts p-coumaroyl-CoA into dihydro-4-coumaroyl-CoA, which is acted upon by chalcone synthase (CHS) to produce phloretin (Ibdah et al. 2014; Wang et al. 2022b). Under the action of phloretin-2′-O-glycosyltransferase and phloretin-4′-O-glycosyltransferase, phloretin gets transformed into phlorizin and trilobatin, respectively (Ibdah et al. 2018). AmUGT71G10, a UGT from Astragalus membranaceus, can produce phlorizin using phloretin as a substrate (Hao et al. 2023). Additionally, phloretin glycosyltransferase 2, identified through activity-directed protein purification and differential gene expression analysis, efficiently catalyzes the 49-O-glycosylation of phloretin to generate trilobatin (Wang et al. 2020b).

Naringin dihydrochalcone (NDC) is a natural sweetener with antioxidant activity, widely used as an additive in the food, pharmaceutical, and cosmetic industries (Yang et al. 2018). A 1,2-rhamnosyltransferase (1,2Rhat) from Citrus maxima, can generate NDC from trilobatin (Eichenberger et al. 2017).

Neohesperidin dihydrochalcone (NHDC) is a strong sweetener obtained by the chemical conversion of neohesperidin. NHDC has antioxidant properties, which is highly sweet and shows good stability. It is widely used as a sweetener in animal feed and in foods such as confections, fruit juices, condiments, and jams (Wang et al. 2022a). Despite the lack of evidence for the natural production or biosynthesis of NHDC, some studies have partly described its biotransformation and biosynthetic pathway (Choi et al. 2021). CYP450 can facilitate the conversion of phloretin to 3-hydroxyphloretin, which can be methylated and glycosylated via O-methyltransferase, dihydrochalcone-4′-O-glycosyltransferase, and 1,2-rhamnosyltransferase to generate NHDC (Eichenberger et al. 2017; Ibdah et al. 2018).

2′-FL

The fucosylated trisaccharide 2′-FL is the simplest and most abundant human milk oligosaccharide (HMO) (Liu et al. 2022c). It is used as an additive in infant formula, dietary supplements, and medical foods (Bych et al. 2019). It was the first HMO to receive regulatory approval and has been approved in the United States, the European Union, Australia, and other countries (Liu et al. 2022c).

2′-FL is synthesized in vivo via a condensation reaction between lactose and guanosine 5′-diphosphate-L-fucose (GDP-L-fucose) catalyzed by α1,2-fucosyltransferase. GDP-L-fucose can either be synthesized de novo or through a salvage pathway (Zhu et al. 2022). In the salvage pathway, GDP-L-fucose is either produced from L-fucose through the action of two enzymes (fucokinase and GDP-L-fucose pyrophosphorylase) or the bifunctional enzyme that promotes phosphorylation and GDP transfer (Lin et al. 2022a; Li et al. 2023a).

The de novo pathway is relatively more complex. The mannose-6-phosphate isomerase manA converts F6P from glycolysis to D-mannose-6-phosphate; this is converted by the phosphate mannosidase manB and the mannose-1-phosphate guanylyltransferase manC into GDP-mannose; this is finally converted to GDP-L-fucose by the GDP-mannose-4,6-dehydratase Gmd and the NADPH-dependent GDP-L-fucose synthase Fcl (Lin et al. 2022a).

In E. coli, glycerin is converted to F6P via gluconeogenesis, which is then sequentially catalyzed by manA, manB, manC, gmd, and the GDP-L-fucose synthase wcaG to generate GDP-L-fucose. Subsequently, α1,2-fucosyltransferase catalyzes the reaction between GDP-L-fucose and lactose to produce 2′-FL (Sun et al. 2023a, b). In Bacillus subtilis, Pgi converts G6P to F6P, which is converted by manA into D-mannose-6-phosphate. Subsequently, 2′-FL synthesis was achieved by the heterologous transfer of manB, manC, gmd, and wcaG from E. coli and the α1,2-fucosyltransferase FutC from Helicobacter pylori into Bacillus subtilis (Yu et al. 2022).

Metabolic engineering of natural sweeteners

The microbial production of natural sweeteners in different hosts are shown in Tables S1 (Supplementary Material).

Terpenoids

SGs

For the microbial biosynthesis of Reb A, ent-kaurene biosynthesis was achieved in E. coli by introducing the heterologous MVA pathway and the ent-kaurene synthesis pathway. Genes encoding SrKO and its redox partner, SrCPR, were transferred into E. coli, and SrKO expression was enhanced by optimizing fermentation temperature and the concentration of isopropyl β-D-thiogalactoside (IPTG). As a result, kaurenoic acid was produced at a yield of 100.23 mg/L. Further, the gene encoding CYP714A2, a KAH from A. thaliana, was introduced into E. coli, and the steviol yield reached 9.47 mg/L. N-terminal modification of CYP714A2 into 17αtr29CYP714A2 boosted steviol yield to up to 15.47 mg/L. Finally, the transfer of genes encoding UGT85C2, UGT91D2w, UGT74G1, and UGT76G1 reconstructed the de novo biosynthetic pathway of SGs in E. coli (Fig. 1). Following this metabolic engineering workflow, researchers produced Reb A at a yield of 10.03 mg/L (Wang et al. 2016). The soluble enzyme is an important prerequisite for large-scale biocatalysis. However, UGT76G1 was mostly overexpressed in E. coli in the form of inclusion bodies. Therefore, UGT76G1 was fused with the N-terminal fusion partner, 30-phosphoadenosine-50-phosphatase (CysQ), to improve its specific activity and solubility, which were markedly enhanced compared with UGT76G1. Fusion with a soluble tag improves the structural stability of UGT76G1, thereby enhancing Reb A yield (Chen et al. 2017).

Fig.1.

The microbial biosynthesis of SGs. IPP, isopentenyl diphosphate; IPPI, isopentenyl pyrophosphate isomerase; MAPP, 3,3-dimethylallyl diphosphate; GPP, geranyl pyrophosphate; FPP, farnesyl pyrophosphate; GGPP, geranylgeranyl pyrophosphate; ent-CPPS, ent-copalyl pyrophosphate synthase; ent-CPP, ent-copalyl pyrophosphate; ent-KS, ent-kaurene synthase; ent-KO, ent-kaurene oxidase; Reb A, rebaudioside A; Reb D, rebaudioside D; Reb M, rebaudioside M; Reb E, Rebaudioside E; S19G, steviol 19-O-β-D-glucopyranoside; Rub, Rubusoside; EUGT11, isoenzyme of UGT91D2 from Oryza sativa

The biosynthetic pathway from steviol to Rub was previously reconstructed in S. cerevisiae by introducing the genes encoding AkUGT75L21, RsUGT85A57, SrUGT85C2, and SrUGT74G1 (Fig. 1). Further bioengineering was performed involving the enzyme sucrose synthase (SUS) and a SrUGT74G1 mutant. SUS can catalyze the formation of UDP-glucose (UDPG) and fructose from sucrose and UDP and are usually combined with UGTs to enable a cascade of glycosylation and UDPG regeneration. The catalytic efficiency of SrUGT74G1 can be enhanced by site-directed mutagenesis, and the double mutant SrUGT74G1S84A/E87A displayed a higher substrate conversion efficiency. Therefore, Rub was efficiently synthesized in S. cerevisiae using steviol as the substrate without providing additional UDPG by introducing the gene encoding SUS from Saccharum officinarum and replacing wild-type SrUGT74G1 with the double mutant SrUGT74G1S84A/E87A. After the fermentation conditions were optimized, the engineered yeast strain produced 0.45 ± 0.06 g/L steviolmonoside and 1.92 ± 0.17 g/L Rub using steviol as the substrate (Mao et al. 2021).

Using a modular engineering strategy, researchers successfully reconstructed the synthetic pathway of Rub in S. cerevisiae by dividing it into a terpenoid backbone synthesis module, a CYP450-catalyzed module, a Rub synthesis module, a UDPG synthesis module, and a Rub transfer module. Truncating the transmembrane domain of CPR1 enhanced the yield of steviol by 231.2%. Substrate transport was further optimized by fusing the short peptide tags RIDD and RIAD to the enzymes KO and KAH, respectively. Molecular docking revealed that Rub has a high affinity for ATP-binding cassette transporters, specifically PDR11, which mediates Rub export. Rub yield was further enhanced by optimizing its transport. A combination of metabolic engineering and replenishment fermentation enhanced the final yield of Rub to 1368.6 mg/L (Xu et al. 2022b). Using this high-yielding Rub strain as the chassis, the researchers further reconstructed the biosynthesis pathway of rebaudiosides and achieved their de novo synthesis. The catalytic efficiency of the key rate-limiting enzyme EUGT11 was enhanced through a combination of two metabolic engineering strategies: replacing the constitutive promoter with increasing gene expression and optimizing metabolic flow via dynamic regulation. The final yield of rebaudiosides (Reb A, Reb D, and Reb M) reached 132.7 mg/L (Xu et al. 2022b).

B. subtilis contains numerous UGT-encoding genes, which can be transferred into E. coli along with ginsenosides as precursors to synthesize the corresponding rare ginsenosides (Li et al. 2023c). The yjiC gene from B. subtilis, encoding a UGT, was transformed into E. coli along with a gene encoding A. thaliana SUS. A one-pot cascade reaction with Reb A and sucrose as the substrate and glycosyl donor, respectively, produced Reb L2 at a yield of 30.94 g/L (Yang et al. 2023).

Mogrosides

The yield of mogrosides can be enhanced by achieving the efficient and successful synthesis of cucurbitadienol, one of its precursors. The cucurbitadienol synthesis pathway was reconstructed in S. cerevisiae by introducing the gene encoding S. grosvenorii cucurbitadienol synthase (SgCS). The gene encoding a global regulatory factor Upc2 was over-expressed to optimize the yeast MVA pathway and increase the supply of IPP and DMAPP. Simultaneously, ERG7, which encodes lanosterol synthase, was knocked down. Finally, the yield of cucurbitadienol reached 63.00 mg/L via fed-batch fermentation (Qiao et al. 2019a).

The major obstacles to mogrosides production using yeast are the inefficient and un-controllable multi-step glycosylations and the absence of some GTs. Three UGTs (UGTMG1, UGTMS1-M7, and UGTMS2) can catalyze the formation of β1-6 and β1-2 bonds, adding primary and branching glycosyl groups to several mogrosides (Fig. 2). UGTMG1 efficiently catalyzes the primary glycosylation of MII-E. UGTMS1-M7, a UGTMS1 mutant, exhibits 74–400-fold higher catalytic efficiency than the wild type in sequentially converting MII-E to MIV-A. UGTMS2 also displays strong catalytic activity toward various glycosides, ranging from mogroside II to MV. The mogroside synthesis pathway was reconstructed in S. cerevisiae using different combinations of these three UGTs (Fig. 2). The engineered strain transformed with the gene encoding UGTMS1-M7 produced MIV and MV at a yield of 49.21 mg/L and a conversion rate of 62% with the exogenous addition of mogrol and glucose. When mogroside III-A (MIII-A) was used as a substrate, the yield of mogrosides IV and V was 114.45 mg/L, with a conversion rate of more than 99%. When genes encoding UGTMG1 and UGTMS1-M7 were introduced into Exg1-knockout yeast strains, mogrol could be completely consumed with the exogenous addition of mogrol and glucose, and the final cumulative yield of MI-E, MII-E, MIII-A, and MIV-A reached 79.30 mg/L (Li et al. 2022a). A total of 10.25 mg/L of MV in shake flasks cultivation was obtained in engineered yeast by the enhancement of precursors, inhibition of the competitive pathway, prevention of MV degradation, and lipid droplet compartmentalization and endoplasmic reticulum expansion (Qu et al. 2025).

Fig. 2.

The microbial biosynthesis of mogrosides. IPP, isopentenyl diphosphate; DMAPP, 3,3-dimethylallyl diphosphate; GPP, geranyl pyrophosphate; FPP, farnesyl pyrophosphate; SQE, squalene epoxidase; CS, cucurbitadienol synthase; EPH, epoxide hydrolase; 11H-Cuol, 11-hydroxyl cucurbitadienol; MI-E, mogroside IE; MI-A1, mogroside I-A1; MII-E, mogroside IIE; MIII, mogroside III; MIIIX, mogroside IIIX; SI, siamenoside I; MIV-A, mogroside IV-A; MV, mogroside V; MVI, mogroside VI

SI is the sweetest glycoside isolated from S. grosvenorii. However, its scarce availability and complex structure limit its use as a natural sweetener. GTs glycosylate MIII-E to generate SI. UGT94-289–2, a UDP-GT from S. grosvenorii, was modified using a semi-rational design into the UGT-M2 mutant, which was co-transformed into E. coli along with SUS from A. thaliana to efficiently convert MIII-E to SI without requiring exogenous UDPG (Fig. 2). The final productivity of SI was 29.78 g/L/day (Xu et al. 2022a). In addition to these metabolic engineering strategies, SI was generated for the first time by enzyme-loaded electrospun fibers using Exg1 at an average yield of 118 ± 0.08 mg/L/h per gram of fiber (Virly et al. 2020).

GL and GA

11-Oxo-β-amyrin is the important precursor of GL and GA, and to efficiently synthesize it in S. cerevisiae, different coupling forms of CYP88D6 and cytochrome P450 reductase (CPR) were constructed to optimize the oxidative activity of CYP88D6. It was found that a high expression ratio of CYP88D6: CPR can promote the electron transfer efficiency between CYP88D6 and CPR. After combining other optimization strategies, the yield of 11-oxo-β-amyrin in fed batch fermentation reached 810.6 mg/L (Sun et al. 2023a, b). In addition, in order to improve the biosynthesis of GA, a novel CYP450 redox system was constructed in yeast, and then the introduction of CYP450 Uni25647 and CYP72A63, and the screened best paired CPR GuCPR1 for G. uralensis, resulted in a GA production of 18.9 ± 2.0 mg/L in a 5 L fermenter (Zhu et al. 2018). There are also studies discovering the use of newly identified GuCYB5 from G. uralensis can significantly increase the production of GA. Then by MVA pathway optimization, the GA biosynthesis was increased, and through fed batch fermentation, the production of GA reached 8.78 mg/L (Wang et al. 2019a, b). Previously the enzyme responsible for the conversion of GA to GL was identified, and then this enzyme was transferred into engineered yeast strain with high production of GA, leading to the construction of complete biosynthetic pathway of GL. Combined with other metabolic engineering strategies, 5.98 ± 0.47 mg/L of GL was obtained by shake flask fermentation (Fig. 3) (Xu et al. 2021).

Fig. 3.

The microbial biosynthesis of glycyrrhetinic acid and glycyrrhizin. IPP, isopentenyl diphosphate; DMAPP, 3,3-dimethylallyl diphosphate; GPP, geranyl pyrophosphate; FPP, farnesyl pyrophosphate; SQE, squalene epoxidase; β-AS, β-amyrin synthase

Sweet proteins

Thaumatin

Thaumatin II was first produced in P. pastoris at a yield of 25 mg/L by cloning the thaumatin II-expressing gene into a yeast expression vector. It was indistinguishable from natural thaumatin II (Masuda et al. 2004). Thaumatin II has also been expressed in E. coli using a codon-optimized construct. After renaturing the recombinant protein from inclusion bodies using reduced/oxidized glutathione, the yield of thaumatin was approximately 40 mg/L (Daniell et al. 2000).

To produce thaumatin I, its cDNA was cloned downstream of the α-factor signal sequence and the Kex2 protease cleavage site to build an expression construct, then the construct was transformed into P. pastoris, which generated thaumatin I from methanol as a substrate at a yield of 30 mg/L (Ide et al. 2007a, b). Alternatively, a cDNA library was obtained through reverse transcription of mRNA, and the cDNA sequence of preprothaumatin was cloned. The cDNAs encoding mature thaumatin I and prothaumatin I was cloned into the PIC6α plasmid carrying the α-factor signal sequence to obtain the two vectors pPIC6a-TH and pPIC6a-proTH, respectively. The cDNAs encoding prethaumatin I and preprothaumatin I were cloned into the PIC6 plasmid, which does not carry the α-factor signal sequence, to obtain the two vectors pPIC6-preTH and pPIC6-preproTH, respectively. All these expression vectors were transformed into P. pastoris, and the fermentation products were evaluated. The yeast transformants containing pPIC6-preTH and pPIC6-preproTH produced a higher amount of thaumatin I, with the latter generating the highest yield (Ide et al. 2007a, b). Although no thaumatin I was detected during shake flask fermentation of the pPIC6a-proTH yeast transformant, its yield reached 10 mg/L in the fermenter. The yield of thaumatin I was further enhanced to 100 mg/L by increasing its gene copy number (Masuda et al. 2010).

Protein disulfide isomerase (PDI) is an intracellular protein involved in molecular chaperones (Sha et al. 2013). The yield of heterologous proteins can be enhanced by co-overexpressing PDI in yeast. Moreover, the addition of a C-terminal cysteine tag makes thaumatin easier to purify. By co-overexpressing thaumatin-FLAG-Cys (thaumatin binds to C-terminal cysteine tag) with PDI in P. pastoris, optimizing the pH of the medium during fermentation, and increasing the casamino acid content, the final yield of thaumatin reached 50.7 ± 5.8 mg/L (Healey et al. 2017). Besides genetic modifications, the effect of fermentation conditions on thaumatin yield was also investigated. Shake flask fermentation of thaumatin-producing P. pastoris revealed that the density of surviving cells and thaumatin productivity were higher at pH 6.0 at 30 ℃. Moreover, at pH 6.0, the density of surviving cells was higher at 25 ℃, while thaumatin productivity was higher at 30 ℃ (Joseph et al. 2022).

Monellin

E. coli is a commonly used microorganism for expressing recombinant proteins. Because monellin has poor thermal stability, three double mutants of MNEI (E2N/E23A, E2N/Y65R, and E23A/Y65R) were expressed in E. coli with yields of 11.09, 11.32, and 10.85 mg/L, respectively. E2 and E23 influence the sweetness and thermal stability of monellin, respectively (Zheng et al. 2018). Fermentation of E. coli BL21 (DE3) introducing the MNEI with sodium acetate as an additional carbon source yielded 180 mg/L monellin (Leone et al. 2015).

The low thermal stability of sweet proteins limits their use in the food industry (Tang et al. 2021). However, using site-directed mutagenesis and a GAPDH constitutive promoter, the yield of secreted recombinant monellin could be increased to 0.15 g/L in P. pastoris (Cai et al. 2016). The yield was further elevated to 2.62–2.71 g/L by initiating methanol induction at lower cell densities; this was 2.5–4.9-fold higher than the yield obtained by initiating methanol induction at higher cell densities (Jia et al. 2017).

A cDNA encoding a mutant monellin was expressed in S. cerevisiae, fused to the α-factor signal peptide and controlled by the GAL1 promoter, at a yield of 0.41 g/L (Chen et al. 2011). Additionally, using ribosomal DNA (rDNA)-mediated homologous recombination, researchers transformed S. cerevisiae with CUP1, a selection marker, and the gene encoding single-chain monellin to obtain a monellin yield of 675 mg/L (Liu et al. 2015).

In addition to E. coli and yeast, monellin has also been expressed at a yield of 0.29 g/L in B. subtilis using the sacB promoter and a signal peptide sequence (Chen et al. 2007). Food-grade proteins have also been produced in Aspergillus niger because of its strong protein secretion capacity and unique safety profile. HiBiT-tagged monellin was expressed at an ultra-low level in A. niger. Its yield was improved to 0.284 mg/L by increasing the monellin gene copy number, fusing the gene to the highly expressed endogenous glycosylase glaA, eliminating extracellular protease degradation, and optimizing media composition. This constituted the first report of recombinant monellin expression in A. niger (Li et al. 2023b).

Brazzein

The cDNA sequence of des-pE1M-brazzein, wherein the expressed protein lacks N-terminal methionine, was codon-optimized to increase the expression of brazzein. To facilitate the secretion of protein, the brazzein cDNA was cloned and linked the vector containing α-mating factor secretion leader sequence, which was transferred into Kluyveromyces lactis. The final yield of brazzein thus obtained was 104 mg/L (Jo et al. 2013). The use of an optimized chemically defined medium marginally improved brazzein yield to approximately 107 mg/L in K. lactis (Park et al. 2021). By studying the factors affecting the yield and purity of recombinant brazzein in K. lactis, such as pH, temperature, inducer concentration, carbon source, and protein purification strategy, approximately 170 mg/L brazzein was produced (Lee et al. 2019a).

Studies have shown that K. lactis can express recombinant brazzein. False disulfides, which affect the structure and stability of proteins, can easily form during protein expression. The formation of protein disulfide bonds requires PDI and the endoplasmic reticulum membrane-associated protein Ero1p (Frand and Kaiser 2000; Wang et al. 2015). Overexpression of PDI and Ero1p in K. lactis transferred into the des-pE1M-brazzein resulted in a 2.6-fold greater amount of the secreted recombinant protein (Yun et al. 2016).

A codon-optimized cDNA sequence of des-pGlu-brazzein, which lacks the initial residue pGlu, was ligated into a vector to obtain GPD-brazzein and transformed into S. cerevisiae. Despite a low yield of 9 mg/L, the brazzein obtained was highly sweet (Kazemi-Nasab and Shahpiri 2020).

Flavonoids

Overexpression of the phloretin pathway-related genes AtPAL2, AmC4H, ScCPR1, At4CL2, HaCHS, and ScTSC13 in S. cerevisiae strains resulted in the production of higher concentrations of phloretin (Fig. 4), with yields up to 42.7 ± 0.9 mg/L. On this basis, two UGTs that could glycosylate the C2 position of phloretin to produce phlorizin, MdUGT88A1 from Malus × domestica and PcUGT88F2 from Pyrus communis, were respectively transfected, and the resulting strains were all able to produce phlorizin, with the highest yield reaching 65 ± 7 mg/L. AtUGT73B2 from A. thaliana could glycosylate the C4 position of phloretin to produce trilobatin with a yield of 33 ± 3 mg/L. UDP-rhamnose was synthesized by overexpressing AtRHM2 from A. thaliana in S. cerevisiae. Thus, the genes, Cm1,2Rhat, AtRHM2, and AtUGT73B2, was transfected to generate NDC using trilobatin as the substrate (Fig. 4). NDC was successfully synthesized de novo in S. cerevisiae albeit at a low yield of 11.6 ± 0.7 mg/L (Eichenberger et al. 2017). The yield of trilobatin was further improved to 107.64 mg/L without adding UDPG by introducing the gene encoding phloretin-4′-O-glycosyltransferase (MdPh-4′-OGT) from M. × domestica Borkh. into E. coli (Fig. 4) (Nawade et al. 2020).

Fig. 4.

The microbial biosynthesis of flavonoid sweeteners. E4P, erythrose 4-phosphate; PAL, phenylalanine ammonia lyase; C4H, cinnamate 4-hydroxylase; 4CL, 4-coumarate CoA ligase; HCDBR, hydroxycinnamoyl-CoA double bond reductase; CHS, chalcone synthase; Cm1,2Rhat, a 1,2-rhamnosyltransferase from Citrus maxima; Ph-4′-OGT, phloretin-4′-O-glycosyltransferase; PGT2, phloretin glycosyltransferase 2; NDC, naringin dihydrochalcone; Ph-2′-OGT, phloretin-2′-O-glycosyltransferase

The biosynthetic pathway of flavonoid-7-O-disaccharide was developed in S. cerevisiae using flavonoid aglycones as substrates. The UDP-rhamnose regeneration system was developed through metabolic engineering, and UDPG synthesis was optimized. The engineered yeast strain obtained after the transfer of 7-O-glucosyltransferase (UF7GT) and Cm1,2Rhat can take (2S)-hesperetin as a substrate, which can be glycosylated at its 7-hydroxyl site. Subsequently, a rhamnose moiety was ligated via a 1,2-glycosidic bond, ultimately generating neohesperidin with a yield of 249.0 mg/L (Xiao et al. 2023). Finally, the alkaline catalysis and reductive hydrogenation of neohesperidin yielded NHDC (Frydman et al. 2005).

Polyols

Xylitol

In S. cerevisiae, the transcription factor Znf1 can activate a hidden xylose metabolic pathway, which markedly promotes xylose utilization and xylitol production (Fig. 5). Consequently, researchers engineered a yeast strain wherein the gene encoding Znf1 was overexpressed and the xylose repressor BUD21 was knocked down. Using xylose as a carbon source for fermentation, this yeast strain achieved a xylitol yield of 12.14 g/L (Songdech et al. 2022). Additionally, the yield was increased up to 13.66 ± 0.54 g/L by overexpressing XYL1 (encodes xylose reductase) and XYL2 (encodes xylitol dehydrogenase) from Candida tropicalis in S. cerevisiae, which promotes xylose utilization (Shi et al. 2022). Using microorganisms to ferment industrial wastewater containing xylose and glucose is also a feasible way to produce xylitol. Firstly, a strain of S. cerevisiae exhibiting high tolerance to waste xylose mother liquor (WXML) was screened and transfected with the gene XYL1 from Scheffersomyces stipites (Pichia). Further metabolic engineering was performed, so that the yeast strain could efficiently convert xylose to xylitol using WXML as the substrate. After optimizing the fermentation conditions, the strain, which was fermented in a medium containing WXML, delignified corncob residues, and cellulase, produced 91.0 g/L xylitol (He et al. 2021). In addition to S. cerevisiae, Y. lipolytica can also convert xylose to xylitol in their natural state.

Fig. 5.

The microbial biosynthesis of polyols and 2′-FL. M1P, Mannitol-1-phosphate; F6P, Fructose-6-phosphate; G6P, Glucose-6-phosphate; RL5P, Ribulose-5-Phosphate; X5P, Xylulose-5-phosphate; R5P, Ribose-5-phosphate; GD3P, Glyceraldehyde-3-phosphate; S7P, Sedoheptulose-7-phosphate; E4P, Erythrose-4-phosphate; EL4P, Erythritol-4-phosphate; F16B, Fructose-1.6-bisphosphate; G3P, Glycerol-3-phosphate; DHA, dihydroxyacetone; DHAP, dihydroxyacetonephosphate; M6P, Mannose-6-phosphate; M1P, Mannose-1-phosphate; GDP-M, GDP-mannose; GDP4K6D, GDP-4-keto-6-deoxymannose

As a carbon source, glycerol is more easily reduced than conventional carbohydrates, such as glucose, sucrose, and xylose, providing more NAD(P)H. Glycerol is the most preferred carbon source for Y. lipolytica, and its presence markedly inhibits xylose uptake. With both glycerol and xylose as substrates and optimized fermentation conditions, Y. lipolytica achieved a final xylitol yield of 53.2 g/L (Prabhu et al. 2020). Atmospheric and room-temperature plasma was used to mutagenize C. tropicalis to obtain a high-yielding strain with respect to xylitol. Simultaneously, a two-stage dissolved oxygen strategy was employed during fermentation. The final engineered strain produced 0.79 g xylitol per gram of xylose (Zhang et al. 2019).

In addition to engineered strains, a novel xylitol-producing yeast, Cyberlindnera dasilvae sp. nov., has been identified in nature. Six isolates of this strain were able to generate 12.61–31.79 g/L xylitol in yeast extract-peptone-xylose medium containing 5% xylose (Barros et al. 2021).

E. coli can also serve as a host for xylitol production. In E. coli, the replacement of xylB, the endogenous gene encoding xylulose kinase, with araL from B. subtilis resulted in dephosphorylation of D-xylulose-5-phosphate to D-xylulose. Simultaneous overexpression of the PPP-related genes zwf, pgl, gnd, rpe, and xylA resulted in a final xylose yield of 3.3 g/L. Subsequently, xylitol was successfully synthesized by introducing the XYL1 gene. Despite a low yield of 9.5 mg/L, this represented the first instance of xylitol synthesis using glucose as the substrate. This process not only lowered the cost of raw materials but also mitigated carbon catabolite repression, thus providing a novel idea for xylitol synthesis (Yin et al. 2021).

NADPH is a key reducing agent in biosynthesis and a useful electron donor in whole-cell biotransformation. NADPH flux for xylitol production in E. coli was increased through two-stage dynamic metabolic control. Firstly, G6P dehydrogenase activity was decreased, pyruvate-ferredoxin/flavodoxin oxidoreductase (Pfo) activity was increased, and fatty acid biosynthesis was disrupted. This mitigated the inhibition of membrane-bound transhydrogenase (PntAB), eventually depleting NADPH pools and elevating acetyl-CoA flux and NADPH production. The depletion of NADPH pools altered the expression and activity of the associated enzymes, leading to an increase in NADPH flux. The final xylitol yield of the engineered E. coli strain reached 200 g/L (Rice et al. 2020). Cyanobacteria can generate NADPH through photosynthesis, and their great potential for xylitol production can be realized through metabolic engineering. Synechococcus elongatus PCC7942 cells were engineered by introducing the xylose transporter EcXylE from E. coli and the NADPH-dependent xylose reductase CbXR from C. boidinii. This modified strain could efficiently metabolize and reduce xylose to xylitol. Increasing the cell density of the engineered strain by concentrating the quiescent cells further boosted xylitol synthesis, with yields reaching up to 33 g/L (Fan et al. 2020).

Erythritol

The biosynthesis of erythritol mainly depends on the pentose phosphate pathway (Fig. 5). Y. lipolytica has been the predominant erythritol-producing host organism (Rakicka-Pustułka et al. 2021). Y. lipolytica isolated from Poland lime honey samples produced 32.6 g/L erythritol and low amounts of byproducts (Ziuzia et al. 2023). Moreover, transketolase (TKL1) and transaldolase (TAL1) effectively promoted erythritol production in Y. lipolytica. Overexpression of glycerol kinase (GUT1), triose-phosphate isomerase (TPI1), TKL1, and TAL1 and knockdown of EYD1 which encodes an erythritol dehydrogenase, resulted in the production of approximately 40 g/L erythritol using glycerol as the substrate. The overexpression of ribose 5-phosphate isomerase (RKI1) further enhanced the yield of erythritol to 52 g/L (Zhang et al. 2021b).

Overexpression of the erythrose reductase homolog YALI0B07117g in Y. lipolytica can increase carbon flux toward erythritol synthesis. When a YALI0B07117g-overexpressing strain was cultured with glycerol as the sole carbon source, the erythritol yield increased from 41.15 ± 4.97 g/L to 59.83 ± 4.86 g/L at 72 h (Szczepańczyk et al. 2021). Additionally, knockdown of MDH2 and ArDH1, encoding MDH and abietol dehydrogenase, respectively, can further increase the accumulation of fructose and ribulose. Further knockdown of Ku70, EYD1, and URA3 and the simultaneous overexpression of GUT1, TKL1, TAL1, and erythrose reductase can effectively enhance erythritol production. Finally, ScFPS1, encoding a glycerol transporter from S. cerevisiae, was introduced into Y. lipolytica, and glycerol was used as the substrate to produce erythritol with a yield of 64.65 g/L (Huang et al. 2023).

Knockdown of Ku70, which is involved in non-homologous end-joining, can enhance the efficiency of homologous recombination and can eventually enhance erythritol production. MDH2 encodes the only active MDH in Y. lipolytica. Knockdown of Ku70, MDH2, and EYD1 in the high-erythritol-producing strain Y. lipolytica CGMCC7326 resulted in the production of 154 ± 5 g/L erythritol (Wang et al. 2020a). Moreover, integrating Cre, the recombinant enzyme, and deleting Mhy1 and Cla4 in the engineered strain altered the dimorphism of Y. lipolytica and produced a chassis strain with yeast-like morphology. The final yield of erythritol was elevated to 174.16 ± 2.71 g/L (Xu et al. 2023b). During the production of erythritol in Y. lipolytica, a large amount of heat is generated, which can hinder its production. Therefore, a heat-tolerant Y. lipolytica strain, BBE-18, which could grow normally at 35℃, was obtained through adaptive evolution. This lays the foundation for designing high-yielding heat-tolerant strains using metabolic engineering (Qiu et al. 2021).

Besides metabolic engineering strategies, erythritol production can be upregulated by co-cultivating Y. lipolytica and Trichoderma reesei using distiller grains as the substrate. After adjusting the ratio of the two strains and optimizing the fermentation conditions, 267.1 mg of erythritol was produced per gram of dry substrate (Liu et al. 2022a). Additionally, an erythritol microbial production factory was constructed in Moniliella pollinis using immobilized cell technology. M. pollinis was immobilized on cotton cloth for batch fermentation, finally producing 47.03 ± 6.16 g/L erythritol (Hijosa-Valsero et al. 2022).

Mannitol

The microbial biosynthesis of mannitol begins with fructose and glucose as carbon source (Fig. 5). Y. lipolytica is the chief mannitol-producing strain. A high-mannitol-producing Y. lipolytica strain, isolated from a Polish lime honey sample, produced 15.1 g/L mannitol by fermenting glycerol as a carbon source (Ziuzia et al. 2023). Twenty-one Y. lipolytica strains producing mannitol from glycerol were screened for the best parameters of mannitol production and the most resistance to NaCl. The optimal substrate concentration for fermentation was further explored, and the selected strains yielded up to 78.5 g/L mannitol through fed-batch fermentation. These strains can produce mannitol using raw glycerol with higher salt content (Juszczyk et al. 2023).

L. lactis also exhibits great potential for mannitol production. Mutations in the promoter of mtlA, which encodes enzyme IIBC of the mannitol phosphotransferase system, may enhance the expression of mtlA, mtlR, mtlF, and mtlD, and the C39T mutation elicited the most potent effect. Subsequent overexpression of mtlD and knockdown of the transcriptional activator mtlR in the mutant L. lactis strain resulted in the production of 10.1 g/L mannitol via two-step fermentation (Xiao et al. 2021).

Two lactic acid bacteria strains, Leuconostoc mesenteroides SKP 88 and L. citreum SKP 92, which convert fructose to mannitol, were isolated and characterized from Pa kimchi. After optimizing the fermentation conditions in Shine Muscat juice with glucose as the carbon source and fructose as the precursor of mannitol, L. mesenteroides SKP 88 achieved a mannitol yield of 41.6 g/L, while L. citreum SKP 92 achieved a yield of merely 23.4 g/L. Additionally, L. mesenteroides SKP 88 executed fermentation in yogurt with a mannitol yield of 15.13 g/L (Kang et al. 2023).

2′-FL

Currently, microbial cell factory biosynthesis has become the dominant industrial method for producing 2′-FL, and the biosynthetic pathway was shown in Fig. 5. In 2023, the National Health Commission of the People's Republic of China reviewed and approved the safety assessment of 2′-FL, which marked the approval of the first synthetic biology-prepared food additive. The strain approved for producing 2′-FL is an E. coli strain. Therefore, we first review the progress of research on the production of 2′-FL using metabolically engineered E. coli.

The GDP-L-fucose synthesis pathway was constructed in E. coli BL21 (DE3) by over-expressing manA, manC, manB, gmd, and Fcl. Subsequently, FutC was heterologously expressed to produce 2′-FL. Overexpression of the transcription factor RcsA may promote the expression of genes related to GDP-L-fucose synthesis. Consequently, knockdown of lon, which encodes an ATP-dependent intracellular protease, mitigating its effect on RcsA. Knockdown of wcaJ, encoding a UDPG lipid carrier transferase in the colanic acid synthesis pathway, further enhanced the supply of GDP-L-fucose. Additionally, knockdown of lacZ, which encodes an endogenous β-galactosidase, and overexpression of lacY, which encodes a lactose permease, enhanced lactose utilization. Combined with the optimization of the NADPH regeneration pathway as well as of the culture conditions, the highest 2′-FL yield achieved by the engineered E. coli strain was 9.12 g/L (Huang et al. 2017). In addition to these metabolic engineering strategies, the catalytic efficiency of the enzyme can be enhanced by increasing the copy number of FutC and fusing a thioredoxin A tag to the N-terminus of FutC. Combined with the knockdown of the LacI repressor to improve lactose utilization and knockdown of glutathione reductase to improve the intracellular oxidative environment, the engineered E. coli strain ultimately generated 10.3 g/L of 2′-FL (Lin et al. 2022a).

2′-FL was also produced by developing the GDP-L-fucose synthesis pathway and transferring another α1,2-fucosyltransferase, FucT2 from H. pylori in E. coli BL21 (DE3). To further enhance its production, ribosome-binding site 29 (RBS29) was used to regulate the expression of the upstream enzymes gmd, wcaG, and fucT2. Additionally, RIAD and RIDD were fused to gmd and wcaG. Subsequently, manC was aggregated into the gmd-wcaG enzyme complex via the orthogonal protein interaction motif (PDZ-PDZlig) to obtain the triple enzyme complex manC-gmd-wcaG. The complex is expressed in the final engineered strain, which produced 25.1 g/L of 2′-FL through fed-batch fermentation (Chen et al. 2023a).

The PTSGlc and NPTSGlc transport systems are two modes of glucose endotrans-formation in E. coli. Knockdown of genes related to the PTSGlc system can enhance the NPTSGlc system, which effectively improves the glucose endotransformation efficiency. Additionally, the biosynthesis pathway of lactose was developed in E. coli by overexpressing a β-1,4-galactosyltransferase (NmlgtB) from Neisseria meningitidis. Undergoing other metabolic engineering strategies to reduce the generation of harmful byproducts and further optimize the synthesis of 2′-FL, the final strain produced 40.44 g/L of 2′-FL through fed-batch fermentation, with a conversion rate of 0.63 g per gram of glucose (Li et al. 2022c).

Taking E. coli MG1655 as the chassis strain, researchers selected manA, manB, gmd, and wcaG to be overexpressed through the genome and manC and futC to be overex-pressed using plasmids by analyzing the catalytic activities and functions of enzymes of the 2′-FL synthesis pathway. manA and manB are the key rate-limiting enzymes in 2′-FL synthesis. Consequently, overexpression of manA and manB through a new plasmid was transferred into the engineered E. coli. Additionally, overexpression of fructose-1,6-bisphosphatase to enhance gluconeogenesis, and knockdown of the 2′-FL branching pathway genes (wcaJ and nudD), further facilitated the accumulation of GDP-L-fucose. To improve the tolerance of the strain, RpoC in the production strain was mutated through adaptive laboratory evolution. Other metabolic engineering strategies were applied to further enhance lactose utilization, transport efficiency, and 2′-FL efflux. Finally, fermentation was carried out with glycerol as the primary carbon source, and the yield of 2′-FL reached 61.06 ± 1.93 g/L, with a production efficiency of 1.70 g/L/h (Sun et al. 2023a, b). Furthermore, based on the futC from H. pylori ATCC 26695, multilevel combination strategies, such as fine-tuning gene expression levels through RBS, N-terminal fusion of prokaryotic translation initiation factor IF2 (InfB) and double-copy gene expression, have greatly improved catalytic activity and solubility of futC. Combining these modifications with the classical metabolic engineering strategy, the final yield of the E. coli strain reached 64.62 g/L (Li et al. 2022b).

The transcription factors RcsA and RcsB can both positively regulate the expression of manC, manB, gmad, gmd, and wcaG (Ni et al. 2020). Firstly, the synthetic pathway for synthesizing 2′-FL was developed in E. coli BL21 (DE3). Subsequently, the supply of GDP-L-fucose was increased by both increasing production and reducing consumption. For boosting production, RcsA and RcsB were overexpressed, their repressors clpYQ and lon were knocked down, and zwf and guanosine inosine kinase (Gsk) were also overexpressed. For reducing consumption, nudD and nudK were knocked down, which de-creased GDP-mannose hydrolysis. And wcaJ was also knocked down. In terms of transport and utilization of lactose, the efficiency was enhanced by knocking out the lactose metabolism-related genes lacZ and lacA and overexpressing lacY. Additionally, SUMO-tagged proteins were fused to the N-terminus of futC to improve its solubility expression, while the sugar transporter protein setA was overexpressed to promote the exocytosis of 2′-FL. The knockdown of the isocitrate lyase regulator (iclR) attenuated acetate formation. Finally, fed-batch fermentation with glycerol and glucose as substrates resulted in 2′-FL yields of 121.9 g/L and 111.56 g/L, respectively (Wang et al. 2023a). Utilizing E. coli's native α1,2-fucosyltransferase for 2′-FL production is also a feasible strategy. The de novo pathway genes for GDP-L-fucose, the α1,2-fucosyltransferase WbgL from E. coli O128, and rcsA and rcsB were introduced via plasmids in engineered strains deficient in lacZ and wcaJ. Simultaneously, an additional WbgL gene was integrated into the genome of the engineered bacteria, and a strong promoter (PJ23119) was used to replace the original promoters of the gene clusters manC-manB and gmd-wcaG. The highest 2′-FL yield obtained after these modifications was 79.23 g/L (Liu et al. 2022b).

The yield of 2′-FL in E. coli is often limited by the insolubility and instability of the key rate-limiting enzyme α1,2-fucosyltransferase. A mutant library of FucT2 was developed by targeted evolution to enhance its expression, combining error-prone PCR and in vitro homologous recombination of DNA to fuse KanR to the C-terminus of FucT2. The strain harboring the FucT2 mutant ultimately generated 0.31 g/L of 2′-FL, which was a 1.72-fold higher yield than the control strain (Shin et al. 2022). Therefore, this strategy can be combined with other metabolic engineering strategies to enhance the 2′-FL yield.

Besides modifying α1,2-fucosyltransferases through directed evolution, more efficient enzymes can be explored for 2′-FL production. BKHT, an α1,2-fucosyltransferase from Helicobacter sp. 11S02629-2, was also found to not generate the byproducts (difucosyllactose and 3′-FL). A combination of the traditional metabolic engineering strategy and the introduction of BKHT in E. coli enhanced the fed-batch fermentation yield of 2′-FL to 94.7 g/L, with a production efficiency of 1.14 g/L/h (Zhu et al. 2023). Similarly, SAMT, a novel α1,2-fucosyltransferase from Azospirillum lipoferum, can efficiently produce 2′-FL without generating difucosyllactose and 3′-FL. Metabolically engineered E. coli expressing SAMT produced 112.56 g/L of 2′-FL through fed-batch fermentation, with a production efficiency of 1.10 g/L/h (Chen et al. 2023b).

B. subtilis is a model microorganism widely used in the production of nutraceuticals and pharmaceuticals (Liu et al. 2019). It is also often utilized for 2′-FL production. Firstly, a GDP-L-fucose-generating strain was developed, and a salvage pathway was established in B. subtilis by introducing codon-optimized fkp from Bacteroides fragilis, which encodes L-fucokinase/GDP-L-fucose pyrophosphorylase, Secondly, overexpression of glcP, which encodes a sugar transporter, enhanced fucose uptake by the strain. Finally, the 2′-FL synthesis pathway was developed by the heterologous introduction of futC. To further enhance the yield of 2′-FL, LAC12, which encodes a lactose permease from K. lactis, was subsequently introduced to enhance lactose utilization. Simultaneously, yesZ, which en-codes a β-galactosidase, was knocked down to block lactose degradation. Then, the expression of cofactor guanosine 5′-triphosphate (GTP) regeneration module genes was appropriately adjusted to increase the synthesis of GDP-L-fucose and 2′-FL. The final 2′-FL yield in this case reached 5.01 g/L (Deng et al. 2019).

The biosynthetic pathway of 2′-FL is divided into three functional modules: 2′-FL synthesis, GTP supply, and competitive pathway. A thermal-sensitive switch based on the thermistor CI857 and clustered regularly interspaced short palindromic repeats interference (CRISPRi) controlled the expression and repression of each module. And researchers use temperature-responsive genetic circuits to conduct multi-module orderly control of 2′-FL production. The final yield of 2′-FL synthesized using these strategies reached 28.2 g/L (Yu et al. 2022).

The biosynthetic pathway of 2′-FL was reconstructed in B. subtilis by transferring six genes, including manA, manB, manC, gmd, wcaG, and futC. However, the yield was only 120 mg/L. Therefore, xylose uptake of the strain was enhanced via heterologous expression of LacY, followed by knockdown of ganA and yesZ, which encode two β-galactosidases. The xylose metabolic pathway was optimized by overexpressing xylA and xylulokinase (xylB). Finally, the 2′-FL yield was elevated to 31.2 g/L by replenishment fermentation using glycerol and xylose as substrates (Ji et al. 2022). The yield of 2′-FL could be elevated to 88.3 g/L by increasing the regeneration of the GTP, enhancing the supply of the precursor mannose-6-phosphate, using sucrose and lactose as substrates, and optimizing the fermentation conditions (Zhang et al. 2022a).

S. cerevisiae has also shown great potential as a microbial cell factory for 2′-FL. An auto inducible gene expression system was developed by reconnecting the pheromone reaction pathway of S. cerevisiae. The α pheromone of K. lactis (Kα) was used to replace the α pheromone of S. cerevisiae (Sα). The expression of the transcriptional activator Gal4p was controlled by a pheromone-responsive α promoter (Pfus1). Subsequently, the quorum sensing-mediated enhanced gene expression system based on the development of Kα was combined with the synthesis of 2′-FL. Simultaneously, the promoter of the synthetic pathway gene from GDP-mannose to 2′-FL was replaced with a lactose-inducible promoter to regulate the production phase by adjusting the time point of lactose addition. Moreover, the strains can only start production after attaining high cell densities. Knockdown of the galactose metabolism genes GAL1, GAL7, and GAL10 further enhanced the yield of 2′-FL. Through fed-batch fermentation, the 2′-FL yield of the engineered S. cerevisiae strain reached 32.05 g/L (Xu et al. 2023a).

Future prospective

Natural sweeteners have attracted widespread attention because they are eco-friendly, healthy, low in calories, and good in taste. As the concept of green, low-carbon, and sustainable development becomes increasingly popular, the demand for natural sweeteners will keep increasing. With the development of synthetic biology, the use of microbial cell factories has emerged as effective production methods to meet the demands for natural sweeteners. These methods have greatly progressed in recent years. Most natural sweeteners can now be synthesized de novo, with erythritol and mannitol being produced at the industrial scale. However, in the late phase of the production process, the separation and extraction of a single product remains a big challenge and most likely contributes to environmental pollution. Therefore, exploring more eco-friendly and efficient separation and purification methods is necessary.

Moreover, the yields of most natural sweeteners synthesized de novo are not high enough to meet the industrial requirements. The catalytic efficiency of the enzyme is a critical limiting factor for the yield. For instance, the catalytic efficiency of UGT is crucial to the production of terpenoid and flavonoid natural sweeteners. Accordingly, exploring different UGTs and improving their catalytic efficiency are key avenues in the quest to enhance the large-scale production of terpenoid and flavonoid natural sweeteners using microbial cell factories. High-throughput sequencing and multi-omics technologies can be harnessed to explore plant and microbiome data deeply with the objective of discovering more efficient UGTs. Meanwhile, the consumption of natural sweeteners by the heterologous microorganisms may be another potential reason for the limited yield, and to solve this, the selection of suitable microbial host would be an efficient strategy.

CRISPR/Cas9, as an emerging gene editing technology, can perform rapid, efficient, and precise gene editing, regulating the genomes of various organisms and tissues. Currently, it has become an important tool in synthetic biology and metabolic engineering. In recent years, new breakthroughs have been regularly made in the creation of efficient microbial cell factories for producing specific chemicals, fuels, and pharmaceuticals using CRISPR technology. Therefore, CRISPR/Cas9 can easily be applied to the biosynthesis of natural sweeteners, which will potentially enhance the efficiency of developing microbial cell factories.

Artificial intelligence, big data, and other technologies create new possibilities for scientific development. Therefore, when they are applied to the biosynthetic system of natural sweeteners, they facilitate large-scale analysis of biosynthetic pathways, construction of component libraries, high-throughput assembly and optimization, and debugging of man-made systems, thereby realizing the efficient total synthesis of natural sweeteners and their components.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

SGs

Steviol glycosides

2′-FL

2′-Fucosyllactose

Reb A

Rebaudioside A

Reb D

Rebaudioside D

Reb M

Rebaudioside M

IPP

Isopentenyl pyrophosphate

DMAPP

Dimethylallyl diphosphate

MVA

Mevalonate

FPS

Farnesyl pyrophosphate synthase

GGPP

Geranylgeranyl pyrophosphate

FPP

Farnesyl pyrophosphate

ent-CPP

ent-copalyl pyrophosphate

ent-KS

ent-kaurene synthase

ent-KO

ent-kaurene oxidase

ent-KAH

ent-kaurenoic acid 13-hydroxylase

NADPH

Nicotinamide adenine dinucleotide phosphate

Rub

Rubusoside

SI

Siamenoside I

SQE

Squalene epoxidase

CS

Cucurbitadienol synthase

EPH

Epoxide hydrolase

MI-A1

Mogroside I-A1

MII-E

Mogroside II-E

MIIIX

Mogroside IIIX

MIV-A

Mogroside IV-A

MV

Mogroside V

MIV

Mogroside IV

MI-E

Mogroside IE

GA

Glycyrrhetinic acid

GL

Glycyrrhizin

pGlu

Pyroglutamic acid residue

F6P

Fructose-6-phosphate

MDH

Mannitol-by-mannitol dehydrogenase

M1PDH

Mannitol-1-phosphate dehydrogenase

M1P

Mannitol-1-phosphate

DHCs

Dihydrochalcones

PAL

Phenylalanine ammonia lyase

C4H

Cinnamate-4-hydroxylase

4CL

4-Coumarate CoA ligase

NDC

Naringin dihydrochalcone

NHDC

Neohesperidin dihydrochalcone

SUS

Sucrose synthase

xylB

Xylulokinase

Author contributions

Conceptualization, G.W. and T.A.; Investigation, B.F., X.L., Y.L., M.L., T.Y., Y.Q. and B.L.; Writing—original draft preparation, B.F., X.L., Y.L., G.W. and T.A.; Writing—review and editing, G.W. and T.A.; Supervision, G.W. and T.A.; Funding acquisition, G.W. and T.A. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Introduction and Cultivation Project for Young Creative Talents of Higher Education of Shandong Province and Shandong Provincial Natural Science Foundation (No. ZR2021QC097 and No. ZR2021QH323).

Declarations

Competing interests

The authors declare that there is no conflict of interests.