De novo biosynthesis of bioactive isoflavonoids by engineered yeast cell factories

Phase I—Establishing the biosynthesis of scaffold isoflavone DEIN

In plants, the general phenylpropanoid pathway uses the aromatic amino acid (AAA) L-phenylalanine as a precursor for the biosynthesis of isoflavonoids as well as other flavonoids²⁴. The initial steps engage phenylalanine ammonia lyase (PAL), cinnamic acid 4-hydroxylase (C4H), and 4-coumarate-coenzyme A ligase (4CL), resulting in the conversion of L-phenylalanine to p-coumaroyl thioester. Subsequently, the chalcone precursors, naringenin chalcone (NCO) and deoxychalcone isoliquiritigenin (ISOLIG), are synthesized from the condensation of p-coumaroyl CoA and three molecules of malonyl-CoA by chalcone synthase (CHS) alone or with the co-action of NADPH-dependent chalcone reductase (CHR), respectively²⁵. Chalcone isomerase (CHI) is responsible for the further isomerization of chalcone to flavanone²⁶. While naringenin (NAG) acts as the shared structural core in isoflavone GEIN and flavonoids pathways, the flavanone liquiritigenin (LIG) is used for the biosynthesis of isoflavone DEIN. The efficient generation of LIG represents therefore the first step towards developing a yeast platform for producing DEIN.

To facilitate the screening of biosynthetic enzymes for LIG production, we used a yeast platform strain (QL11) that has previously been reported to produce a moderate level of p-coumaric acid (p-HCA) (exceeding 300 mg L⁻¹) from glucose without notable growth deficit²⁷. The plant candidate genes have been selected according to their source and enzymatic specificity/activity. We first evaluated the combinations of candidate CHS, CHR, and CHI homologs, alongside the well-characterized At4CL1 from Arabidopsis thaliana, for the biosynthesis of LIG (Fig. 2a). Specifically, three CHS-coding genes, including leguminous GmCHS8 (Glycine max) and PlCHS (Pueraria lobate) as well as non-leguminous RsCHS (Rhododendron simsii), were selected (Supplementary Fig. 3a). CHR activity has been mostly demonstrated in leguminous species²⁸; thus GmCHR5, PlCHR, and MsCHR (Medicago sativa) were screened (Supplementary Fig. 3a). Plant CHIs can be categorized into distinct isoform groups according to their evolutionary path and enzymatic profiles. Whereas type I CHIs, common to all vascular plants, convert only NCO to NAG, legume-specific type II CHIs are capable of yielding both NAG and LIG²⁶. Correspondingly, type II CHI-coding genes PlCHI1 and GmCHI1B2 were evaluated, together with a type I CHI-coding gene PsCHI1 (Paeonia suffruticosa) being used as a control for enzymatic activity. All biosynthetic genes were chromosomally integrated and transcriptionally controlled by strong constitutive promoters. Co-overexpression of At4CL1, GmCHR5, GmCHS8, and GmCHI1B2 resulted in the best LIG production at a level of 9.8 mg L⁻¹ (strain C09) among all resultant strains (C01−C11, Fig. 2b). In addition, strains C01, C04−05 and C10 generated no detectable amounts of isoflavonoid intermediates, which could be attributed to the narrow substrate specificity of type I PsCHI1 (strain C01) and low enzymatic activity of PlCHR in yeast (strains C04-05 and C10), respectively (Fig. 2b and Supplementary Fig. 3b). Subsequently, individual gene replacement in strain C09 allowed the efficient screening of additional 4CL, CHS, and CHI variants. Five plant 4CL-coding genes, including Pl4CL1, Gm4CL3, At4CL2, Ph4CL1 (Petunia hybrida), Pc4CL2 (Petroselinum crispum) and a mutant At4CL1m (I250L, N404K, I461V)²⁹, were overexpressed to generate LIG (strains C12−C17) (Supplementary Fig. 3a); however, none of these genes outperformed the wild-type At4CL1 (C09) (Fig. 2c). Also, no improvement on LIG production was observed for strains expressing alternative variants of CHS (strains C18−C20) or CHI (strains C21−C22) (Fig. 2c and Supplementary Fig. 3c). Strain C09 was therefore chosen as the starting strain for further engineering to produce DEIN.

The entry point enzyme in the isoflavonoid biosynthetic pathway is 2-hydroxyisoflavanone synthase (2-HIS), which belongs to the cytochrome P450 family and catalyzes the intramolecular aryl migration of the B-ring yielding the intermediate 2-hydroxyisoflavanones²⁵. Subsequently, dehydration of the resultant intermediate products, catalyzed by 2-hydroxyisoflavanone dehydratase (HID), gives rise to corresponding isoflavones³⁰ (Fig. 2a). The 2-HIS and HID-coding genes were mainly identified in legumes that have been confirmed to produce isoflavonoids²⁵. To identify efficient biosynthetic enzymes for DEIN formation, a group of leguminous 2-HIS and HID homologs were screened. Specifically, five 2-HIS-coding genes, including Pl2-HIS, Gm2-HIS1, Mt2-HIS1 (Medicago truncatula), Tp2-HIS (Trifolium pretense), and Ge2-HIS (Glycyrrhiza echinata), and three HID-coding genes, including PlHID, GmHID, and GeHID, were combined and overexpressed in strain C09 (Fig. 2d). While most engineered strains generated detectable amounts of DEIN, strain C28, harboring the gene combination of Ge2-HIS and GmHID, accumulated the highest level of DEIN to 0.9 mg L⁻¹ (Fig. 2d). These results show the feasibility to build de novo biosynthesis of DEIN in yeast by harnessing the diversity of plant pathway enzymes.

Phase I—Engineering the redox partner of the key enzyme P450 2-HIS

Though strain C28 produced DEIN, only a low titer was obtained with a concomitant buildup of the biosynthetic intermediate LIG (Supplementary Fig. 4), suggesting inefficiencies in product formation in the later stage of the DEIN pathway. We, therefore, moved on to engineering the activity of Ge2-HIS, considering that the P450-mediated reactions are irreversible and often rate-limiting³¹ whereas HID is believed to facilitate the spontaneous dehydration of 2-hydroxyisoflavanones^16,30.

The redox partner (RP) is an integral part of canonical P450 systems that shuttles the electrons derived from NAD(P)H to the heme iron-center to enable oxygen cleavage and substrate monooxygenation³². The endoplasmic reticulum (ER)-anchored plant P450s recruit a single RP protein, the membrane-attached flavin adenine dinucleotide (FAD)/flavin mononucleotide (FMN)-containing cytochrome P450 reductase (CPR), to transfer electrons³³ (Fig. 3a). Co-expression of the cognate CPRs is a common practice for reconstituting P450-involved plant biosynthetic pathways in yeast, such as the production of terpenoid¹² and alkaloid¹¹, which has been believed to permit more efficient P450-CPR coupling compared to the native yeast CPR³³. In addition, plant CPRs exhibit a certain degree of versatility, due to the high level of conservation represented by the amino acid residues that mediate P450 interactions³³. Accordingly, in our DEIN-producing strains, Ge2-HIS might receive electrons from AtATR2, a CPR homolog of A. thaliana introduced for optimizing the activity of P450 AtC4H to produce p-HCA (Fig. 3a). However, the limiting Ge2-HIS activity evidenced by a low DEIN titer of strain C28 motivated us to exploit alternative plant CPRs, including GmCPR1 and CrCPR2 (Catharanthus roseus) (Fig. 3b). The selected CPR-coding genes were chromosomally integrated in combination with a second copy of Ge2-HIS to strain C28. Using this approach, DEIN production of both GmCRP1-expressing strain C34 and CrCRP2-expressing strain C35 was significantly enhanced to a titer of 5.9 and 9.9 mg L⁻¹, respectively, accounting for a 284 and 544% increase compared with that of strain C33 only harboring a second copy of Ge2-HIS (Fig. 3c). This strongly indicates that there may be an improved coupling of the alternative CPRs to Ge2-HIS.

Among the diverse range of P450-dependent electron transport systems, an intramolecular electron transfer mechanism has been employed by the so-called self-sufficient P450s, in which the RP domain is naturally fused with the catalytic domain³². This unique group of P450s exhibit many advantages, including higher electron transfer efficiency, improved substrate turnover, as well as no need for a separate RP, and therefore are of great interest for biotechnological applications³⁴. To mimic this intramolecular electron transfer, surrogate RPs derived from prominent self-sufficient bacterial P450s, including (1) BM3R, the eukaryotic-like reductase domain of P450BM3 from Bacillus megaterium³⁵; (2) RhFRED, the FMN/Fe₂S₂-containing reductase domain of P450RhF from Rhodococcus sp. strain NCIMB 9784³⁶; and (3) RhF-fdx, a hybrid reductase by substituting Fe₂S₂ domain of RhFRED with ferredoxin (Fdx) from spinach³⁷, were C-terminally fused to Ge2-HIS and investigated for DEIN production (Fig. 3b). These chimeras were co-expressed with an additional copy of Ge2-HIS. While strain C36 harboring Ge2-HIS/BM3R fusion had a 156% increase in DEIN production, compared with the control strain C33 (Fig. 3c), the introduction of RhFRED and RhF-fdx decreased the biosynthesis of DEIN of corresponding engineered strains C37 and C38 (Fig. 3c).

Phase I—Identifying potential metabolic factors improving 2-HIS activity

Besides pairing with an appropriate RP, the attainment of high P450 catalytic efficiency is challenged by (I) the intracellular level of heme for the assembly of holoenzymes, (II) the ER microenvironment to accommodate functional membrane proteins and (III) the availability of redox cofactors. Next, we moved to uncover potential bottlenecks regarding these factors in limiting the biosynthesis of DEIN (Fig. 4a).

The production of active P450s requires sufficient incorporation of cofactor heme, which may deplete the intracellular pool of heme and thereby incur a cellular stress response that in turn damages the net enzymatic activity³⁸. To mitigate this potential adverse effect on the activity of Ge2-HIS, we tested different approaches to regulate heme metabolism of yeast (Fig. 4a-I). Co-overexpression of two rate-limiting enzymes in yeast heme biosynthesis, encoded by genes HEM2 and HEM3³⁹, slightly enhanced the production of DEIN to 9.5 mg L⁻¹ (strain C39, Fig. 4b). In addition, a previous study illustrated that inactivation of the transcriptional repressor Rox1 could render an elevated cellular heme level⁴⁰, resulting from the derepression of the heme biosynthetic gene HEM13. We, therefore, deleted ROX1 in strain C35, yielding a DEIN titer of 12.8 mg L⁻¹ by the resultant strain C40 (Fig. 4b), a 46% increase compared with that of the parental strain. Besides reinforcing the biosynthetic pathway, reducing degradation of heme also contributes to its intracellular accumulation and improves the P450s activity⁴¹. Accordingly, upon the deletion of HMX1, which encodes heme oxygenase responsible for heme degradation, the production of DEIN of the resultant strain C41 (10.6 mg L⁻¹) was increased by 21% relative to strain C35 (Fig. 4b).

Most plant-derived P450s and CPRs are independently tethered onto the ER via hydrophobic transmembrane anchors⁴². Modulating the biogenesis and size of the ER has previously been shown to enhance P450-involved biosynthesis of terpenoids in S. cerevisiae^43,44, a result which is likely due to a higher protein folding capacity enabled by ER expansion. To evaluate the possible beneficial effect of ER expansion for DEIN biosynthesis, we therefore elevated the intracellular level of phospholipids for ER assembly by implementing (1) the deletion of PAH1-encoded phosphatidate phosphatase that competes for the phospholipid precursor^45,46; (2) the deletion of the transcription factor Opi1 and (3) overexpression of the transcription factor Ino2 that negatively and positively control the expression of UAS_INO-containing phospholipid biosynthetic genes, respectively⁴⁷ (Fig. 4a-II). A significantly enhanced DEIN generation was observed for the OPI1 deletion strain C43 (10.8 mg L⁻¹) and the INO2-overexpressing strain C44 (11.3 mg L⁻¹), representing a 20 and 26% increase relative to strain C35 (Fig. 4c). Moreover, strain C46 harboring PAH1 deletion and INO2 overexpression also had a 20% increase in DEIN formation (10.8 mg L⁻¹) compared with that of strain C35, whereas other combinatorial modifications led to a reduction in DEIN titers of strains C45 and C47 (Fig. 4c). This difference could be attributed to the remarkably impaired cell growth of the latter two strains (Supplementary Fig. 5).

Redox cofactors NAD(P)H, the ultimate electron source in cellular metabolism, are indispensable for the catalytic cycle of plant P450s³³. Lack of NAD(P)H could reduce the P450 activity due to inefficient electron transfer. A recent report indicated that improved cellular NADPH level could enhance the P450-mediated protopanaxadiol production⁴⁸. Thus, we decided to reroute the redox metabolism to fuel the activity of Ge2-HIS. In the first strategy, genetic modifications engaged to boost the direct generation of NADPH were devised and individually implemented, including (M1a) overexpression of the transcriptional factor Stb5 that activates the expression of genes involved in the pentose phosphate pathway (PPP)⁴⁹, the major source of NADPH for anabolic processes in yeast; (M1b) overexpression of the ALD6-encoded cytoplasmic NADP⁺-dependent aldehyde dehydrogenase that converts acetaldehyde to acetate; (M2) introduction of E. coli pntAB genes encoding a membrane-bound transhydrogenase capable of reducing NADP⁺ at the expense of NADH⁵⁰; and (M3) overexpression of yeast YEF1-encoded ATP-NADH kinase that directly phosphorylates NADH to NADPH⁵¹, resulting in an elevated concentration of the phosphorylated form of this cofactor without impact on the NADPH/NAPD⁺ ratio (Fig. 4a-III). The resultant strains C49 (M1b) and C51 (M3) produced 9.9 and 9.7 mg L⁻¹ of DEIN, representing a 14% and 11% increase, respectively, compared with the parental strain C35 (Fig. 4d). Moreover, combined overexpression of NAD⁺ kinase (NADK), the sole enzyme leading to de novo NADP⁺ biosynthesis, with an NADP⁺ reducing enzyme was shown to improve NADPH-consumed bacterial isobutanol production⁵². We, therefore, evaluated this strategy via further co-expressing a prokaryotic NADK-coding gene EcyfjB (M4) (Fig. 4a-III). The reported synergistic effect was most evident for strain C52, containing (M1a) and (M4), which had a 17% increase in DEIN titer relative to strain C48 (Fig. 4d).

Phase II—Gene amplification and engineering of substrate trafficking improve DEIN biosynthesis

In screening phase I, through performing combinatorial gene screening in parallel with multiple genetic modifications, we achieved substantial de novo DEIN biosynthesis and identified important metabolic factors affecting its overproduction in yeast. However, the resultant strains exhibited two major unfavorable phenotypes, including a large amount of non-consumed precursor p-HCA (Supplementary Fig. 6) and the formation of several metabolic intermediates and byproducts (Supplementary Fig. 7). This may result from (1) metabolic imbalance between upstream p-HCA producing and the downstream pathways, (2) insufficient activity and (3) substrate promiscuity of some of the plant enzymes, and (4) inefficient cytosolic substrate transfer. We therefore next aimed to improve the production of DEIN via relieving these potential metabolic barriers.

To reduce the metabolic loss due to an excessive supply of p-HCA in background strain QL11, we instead turned to reconstructing the DEIN biosynthesis in a “clean” background without an engineered AAA pathway. With this strain, we used the galactose-induced dynamic transcription mechanism⁵³, which is mediated by GAL promoters (GALps) and has been successfully used for high-level production of value-added chemicals^12,27, to enhance the expression of DEIN pathway genes. A previously described strain QL179²⁷, derived from IMX581 by disrupting galactose utilization genes GAL1/10/7, was thus selected to confer galactose as the gratuitous inducer to activate GALps (Fig. 5a). As expected, simultaneous introduction of the GALps-controlled p-HCA and LIG pathways greatly boosted LIG titer to 37.6 mg L⁻¹ in strain I01 (Fig. 5b), a 284% increase relative to strain C09, in which a constitutive gene expression pattern was used. Interestingly, a markedly low level of p-HCA was detected in strain I01, indicating that to a certain extent the metabolic flux between the redesigned p-HCA producing and consuming pathways was balanced (Supplementary Fig. 8).

The spatial organization of catalytic enzymes is increasingly recognized to be critical for optimizing the metabolic flow through multi-step biosynthetic pathways⁵⁴. Coordinating the physical distance of adjacent enzymes, mediated by peptide linkers or synthetic scaffolds, could elevate local concentrations of enzymes and metabolites thereby speeding up the turnover of intermediates, minimizing metabolic crosstalk, and improving reaction flux⁵⁵. To test this, we implemented a synthetic fusion enzyme strategy to optimize the substrate trafficking through the LIG pathway. Specifically, two distinct oligopeptide linkers possessing flexible (GGGS, L1) or rigid (VDEAAAKSGR, L2) conformations⁵⁶ were applied to fuse the coding sequences of neighboring enzymes (Fig. 5a). Fused AtC4H (E1) with At4CL1 (E2) in both tandem orientations together with other biosynthetic genes were chromosomally integrated into strain QL179, creating strains I02−I05 (Fig. 5b). A significantly enhanced LIG production was observed for these strains; and fusion enzyme in the E1-L1-E2 orientation provided the highest LIG titer of 77.7 mg L⁻¹ (strain I02), representing a 107% improvement compared with strain I01 (Fig. 5b). This result indicates that the physical fusion of two enzymes responsible for the formation and consumption of p-HCA, respectively, could greatly drive the metabolic flux towards LIG biosynthesis. Encouraged by this, we further introduced fusion enzymes consisting of GmCHS8 (E3)-Linker-GmCHR5 (E4) or GmCHR5 (E4)-Linker-GmCHI1B2 (E5) to strain I02 (Fig. 5a). In contrast, here we found the titer of LIG to be decreased in the resultant strains I06-I13 (Supplementary Fig. 9), which may be attributed to a reduced CHR activity of fusion enzymes, considering the lower variation we observed for byproduct formation. We, therefore, selected strain I02 as the platform for incorporating the best downstream DEIN-forming biosynthetic and auxiliary enzymes (Ge2-HIS, GmHID, and CrCPR2). The DEIN titer of resultant strain I14 (20.7 mg L⁻¹, Fig. 5c) doubled relative to strain C35 harboring the constitutive promoter-driven DEIN pathway (9.9 mg L⁻¹, Fig. 3c).

Phase II—Combinatorial strategies to increase DEIN production

Improving the expression of biosynthetic genes and the cellular substrate transfer greatly enhanced the DEIN titer of strain I14. However, we also observed considerable accumulation of both intermediates (15.8 mg L⁻¹ of ISOLIG and 42.3 mg L⁻¹ of LIG, Fig. 5c) as well as byproducts (10.0 mg L⁻¹ of NAG and 1.3 mg L⁻¹ of GEIN, Fig. 5c), showing a need for strengthening the later stage of DEIN biosynthesis. To solve this, we first aimed to improve the activity of Ge2-HIS by combining effective P450-centered genetic targets identified in phase I engineering (Fig. 4a). Expectedly, the removal of heme degradation by disrupting HMX1 gene resulted in a 19% increase in DEIN titer of strain I15 (23.3 mg L⁻¹) compared with that of strain I14 (Fig. 6a), whereas ROX1 deletion negatively affected DEIN production (strain I16, Fig. 6a), this potentially being caused by the resulting loss of its regulatory role in stress resistance of S. cerevisiae⁴⁰. Subsequently, the deletion of OPI1 or overexpression of INO2 genes was individually carried out to stimulate ER expansion in strain I15; however, both resultant strains gave a lower DEIN titer (Supplementary Fig. 10a). While compromised cell growth associated with these strains (Supplementary Fig. 10b) could have weakened their DEIN generation, a shortage of intracellular heme may also be limiting the functional P450 folding and thereby blunting the effect of ER adjustment. Previous studies showed that feeding 5-aminolevulinic acid (5-ALA), the direct precursor of heme biosynthesis, could significantly increase the cellular heme level of yeast³⁸. Indeed, we found exogenous supplementation of 1 mM 5-ALA resulted in 45% (34.3 mg L⁻¹, strain I15 + A), 65% (17.3 mg L⁻¹, strain I17 + A), and 42% (27.1 mg L⁻¹, strain I18 + A), respectively, further increases in DEIN production for the strains tested (Fig. 6b and Supplementary Fig. 10a). ER-targeting modifications however exhibited no beneficial effects on DEIN production, which could be ascribed to the distinct engineering context of strains C35 and I15, implying a need for fine-tuning the interplay between ER biogenesis and P450 anchoring. Thus, strain I15 was subject to the integration of NADPH generation systems. Among selected targets, co-overexpression of native STB5 and bacterial EcyfjB genes (M1a + M4) led to the highest DEIN titer of 40.2 mg L⁻¹, a 12% improvement relative to strain I15 (strain I21, Supplementary Fig. 11).

Based on established results of cofactor refinement, we speculated that the availability of biosynthetic enzymes could emerge as a limiting factor for the conversion of LIG to DEIN. Especially, previous reports indicated that the 2-HIS enzyme in microsomal preparation from soybean cells is labile⁵⁷ and the catalytic characteristics of 2-HIS have evolved by sacrificing protein stability⁵⁸. We, therefore, introduced extra copies of the best DEIN-forming gene combination, Ge2-HIS with GmHID, to strain I21. Interestingly, while there was a 17% increase in DEIN production in strain I24 containing the second copy of selected genes, the introduction of the third copy of this gene combination further enhanced DEIN production to 53.5 mg L⁻¹ (strain I25), representing a 38% increase compared with strain I21 (Fig. 6c).

Compared with batch (glucose excess) cultivations, yeast cells grown under glucose-limited cultivation are known to have a higher biomass yield and an enhanced PPP flux⁵⁹, the latter being anticipated to favor AAA biosynthesis by increasing the availability of the precursor erythrose 4-phosphate. We, therefore, grew DEIN-producing strains under a mimicked glucose-limited fed-batch cultivation by using FeedBeads (FB) (Supplementary Fig. 12), a slow-release system for glucose⁶⁰. Expectedly, under FB conditions, strain I25 produced 62.1 mg L⁻¹ of DEIN, representing an 18% increase relative to the same strain under batch conditions (Fig. 6d). Moreover, the application of this FB strategy led to observable growth improvements and a striking increase in byproduct formation of strain I25 (Supplementary Fig. 13). These results agree also with our previous work wherein significant improvements on cellular biomass formation and p-HCA production could be achieved by growing yeast cells under glucose-limited conditions²⁷.

For the biosynthesis of one molecule of DEIN, one molecule of p-coumaroyl-CoA and three molecules of malonyl-CoA are consumed (Fig. 6e). Following our optimization of metabolic flux using the p-HCA pathway and reinforcement of the DEIN biosynthetic pathway, we speculated that the supply of malonyl-CoA had become the next limiting factor in DEIN production. In S. cerevisiae, the majority of cytosolic malonyl-CoA pool is invested in the synthesis of fatty acids (FAs), which are essential for multiple cellular functions and cell growth⁶¹. The FAS complex, composed of Fas1 and Fas2, is responsible for FAs generation in yeast with the FAS1 gene product known to impose positive autoregulation on FAS2 expression to coordinate the activity of the FAS complex⁶². Hence, we set out to fine-tune the expression of the FAS1 gene to divert malonyl-CoA towards DEIN biosynthesis (Fig. 6e). A group of yeast promoters, exhibiting differential transcriptional activities in response to glucose⁶³ (Supplementary Table 1), were used to substitute the native FAS1 promoter. Among seven evaluated promoters, replacement with BGL2p brought about the greatest DEIN titer of 76.3 mg L⁻¹ (strain I27), a 20% increase compared with strain I25 (Fig. 6f). Additionally, the production of intermediates and byproducts was also notably elevated (Supplementary Fig. 14), further reflecting that promoter replacement of FAS1 has boosted the overall metabolic flux towards isoflavonoids.

The galactose-induced transcriptional response (the GAL induction) of S. cerevisiae initiates with the association of the galactose sensor Gal3 with the regulatory inhibitor Gal80, leading to dissociation of the latter from the transcription activator Gal4, thereby allowing rapid expression of GAL genes⁵³. Constitutive GAL3 mutants (GAL3^c) have been demonstrated to confer galactose-independent activation of Gal4 ⁶⁴. This trait was recently engineered to build a positive feedback genetic circuit in which expressed Gal3^c provokes greater expression of Gal3^c and thereby enhances GAL induction⁶⁵. We speculated that DEIN production may benefit from overexpression of such a Gal3^c mutant as a result of further induction of the GALps-controlled biosynthetic pathway. However, when expressed from a high-copy vector under the control of GAL10p, the introduction of constitutive Gal3^S509P mutant led to a significant decrease in both DEIN and GEIN titers (Fig. 6g and Supplementary Fig. 15). On the other hand, by deleting gene ELP3, encoding a histone acetyltransferase that is part of elongator and RNAPII holoenzyme⁶⁶, a final DEIN titer of 85.4 mg L⁻¹ was achieved in the resultant strain I34 (Fig. 6g), representing a 12% improvement relative to strain I27. The production of GEIN was also slightly increased to 33.7 mg L⁻¹ (Fig. 6g and Supplementary Fig. 15). These results also show to be consistent with a published study wherein ELP3 deletion was found to enhance the GAL1p-mediated beta-galactosidase activity in the presence of galactose⁶⁷. The high-level accumulation of DEIN could exert cellular toxicity in S. cerevisiae and thereby impede the further improvement of its titer. We, therefore, evaluated the growth profiles of the background strain IMX581 under different concentrations of DEIN within its solubility limit. The results revealed that yeast could tolerate up to 150 mg L⁻¹ of DEIN without significant loss of growth capacity (Supplementary Fig. 16). Hence, it is reasonable to assume that the production of DEIN is non-toxic to yeast at the levels produced here.

Phase III—Production of DEIN-derived glucosides

Glycosylation represents a prevalent tailoring modification of plant flavonoids that modulates their biochemical properties, including solubility, stability, and toxicity⁶⁸. In soybean, enzymatic 7-O-glucosylation of DEIN leads to the biosynthesis of DIN⁶⁹, one of the key ingredients found in soybean-derived functional foods and nutraceuticals⁷⁰. Moreover, puerarin (PIN), an 8-C-glucoside of DEIN, is ascribed as the major bioactive chemical of P. lobate roots extract, which has long been used in Chinese traditional medicine for the prevention of cardiovascular diseases⁷¹. Recent studies also show that PIN exhibits diverse pharmacological properties including antioxidant, anticancer, vasodilation, and neuroprotection-related activity⁷². With the establishment of efficient DEIN-producing yeast platform during reconstruction phase II (Fig. 6g), we explored its application potential in the production of PIN and DIN.

The biosynthesis of flavonoid glycosides is mediated by UDP-sugar-glycosyltransferases (UGTs), which catalyze the formation of O−C or C−C bond linkages between the glycosyl group from uridine diphosphate (UDP)-activated donor sugars and the acceptor molecules^1,73. While a soybean isoflavone 7-O-glucosyltransferase exhibiting broad substrate scope was first described over 10 years ago⁶⁹, only recently Funaki et al.⁷⁴ revealed that its homolog GmUGT4 enables highly specific 7-O-glucosylation of isoflavones. On the other hand, the complete PIN pathway was fully elucidated when Wang et al.⁷¹ successfully cloned and functionally characterized a P. lobata glucosyltransferase, encoded by PlUGT43, which displays strict in vitro 8-C-glucosylation activity towards isoflavones and enables PIN production in PlUGT43-expressed soybean hairy roots. We, therefore, tested the feasibility of using these UGTs in generating DEIN glucosides (Fig. 7a). Different copies of PlUGT43 and GmUGT4 under the control of constitutive promoters were integrated into the basic DEIN producer C28, but the resultant yeast strains (E01−E03 for PIN and E04−E06 for DIN, Supplementary Fig. 2) generated no detectable level of glycosides for HPLC analysis. However, through further analysis with high-resolution LC-MS, we validated that strains E03 and E06 could generate trace amount of PIN and DIN, respectively (Fig. 7b and Supplementary Fig. 17), demonstrating that both UGTs were functional in yeast.

Besides the selection of active UGTs, the supply of glycosyl group donor UDP-glucose also plays a pivotal role in regulating glucoside production. With the efficient DEIN producer I34 in hand, we moved to enhance its capacity for biosynthesizing UDP-glucose. In S. cerevisiae, metabolic enzymes phosphoglucomutase (encoded by PGM1 and PGM2) and UDP-glucose pyrophosphorylase (encoded by UGP1) catalyze the formation of UDP-glucose branching from glucose-6-phosphate (Supplementary Fig. 18a). Through chromosomally integrated expression of UGP1 with PGM1 or PGM2 in strain I34, strains E07 and E08 were created. Additionally, to ensure adequate UGTs activity, two multi-copy plasmids, harboring genes PlUGT43 (pQC229) and GmUGT4 (pQC230) under the control of GAL1p, were constructed and individually introduced into the high-level producers of DEIN (strains I34, E07, and E08). In doing so, we found the resultant strains E09 and E10, derived from the I34 background, to produce 45.2 mg L⁻¹ of PIN and 73.2 mg L⁻¹ of DIN, respectively (Fig. 7c). Interestingly, compared with strain E10, the PlUGT43-expressing strain E09 still accumulated a considerable amount of DEIN (28.9 mg L⁻¹, Fig. 7c). This discrepancy may be attributed to the insufficient activity of PlUGT43, whose determined kinetic parameters for DEIN (K_cat = 0.35 s⁻¹, K_m = 32.8 µM, and K_cat/K_m = 1.1 × 10⁴ M⁻¹ s⁻¹)⁷¹ show to be considerably less optimal compared to GmUGT4 (K_cat = 5.89 s⁻¹, K_m = 20.3 µM, and K_cat/K_m = 2.91 × 10⁵ M⁻¹ s⁻¹)⁷⁴. Furthermore, the conversion of GEIN to 25.9 mg L⁻¹ of C-glycoside genistein 8-C-glucoside (G8G) and 26.5 mg L⁻¹ of O-glycoside genistin (GIN) was observed for strains E09 and E10 (Supplementary Fig. 18b and c), respectively, since the selected UGTs exhibit comparable glycosyltransferase activity towards GEIN^71,74. Moreover, the overexpression of UDP-glucose-forming genes resulted in full consumption of DEIN and enhanced PIN production to 72.8 mg L⁻¹ in E07-derived strain E11 and 65.4 mg L⁻¹ in E08-derived strain E12, representing a 61% and 45% increase respectively compared with strain E09 (Fig. 7c). On the other hand, such modifications resulted in no significant increase in the production of DIN (Fig. 7c) and byproduct glucosides (Supplementary Fig. 18b and c), reflecting a shortage of precursor isoflavones. Similarly, we analyzed the growth-inhibitory effects of the two glucosides on strain IMX581. Compared with their aglycon DEIN, an increased level of PIN (500 mg L⁻¹) and DIN (250 mg L⁻¹) can be tolerated by yeast to retain normal cell growth (Supplementary Fig. 19); both concentrations are much higher than the best titer achieved for the two glucosides in our study. Specifically, supplementation of DIN improved growth of yeast, which could result from the uptake of DIN and then release of glucose catalyzed by native glucosidases, such as the steryl-beta-glucosidase Egh1p⁷⁵.