Escherichia coli strain TransT1 (TransGen, Beijing, China) was used for plasmid maintenance and propagation. Genes involved in the biosynthesis of GAMG (Supplementary Table S1) were cloned into the multiple cloning sites (BamhI/XhoI; NotI) of the pESC series plasmids or pUMI-22, using a seamless cloning kit (TransGen Biotech). The Benchling CRISPR tool (https://benchling.com/crispr/) was used to design gRNAs. PCR products were obtained by gene amplification (KOD Neo Plus; TOYBO, Tokyo, Japan) and gel purification (GENE GEL JET; Thermofisher Scientific, Shanghai, China). All assembled plasmids and the gRNA plasmids were sequenced by TSINGKE Biological Technology (Hangzhou, China). All the plasmids and primers used in this study were listed in Supplementary Tables S2, S3 respectively. Oligos used to construct gRNAs were listed in Supplementary Table S4.

The CRISPR/Cas9-mediate genome editing technology was used for the construction of engineered yeast strains (Lian et al., 2018). Yeast strains were transformed using the classical LiAC/ssDNA/PEG protocol, and the transformants were selected on the appropriate agar plates. DNA donors were amplified from the constructed plasmids with 40 bp-homology arms to the yeast genomic DNA integration sites. All yeast strains used in this study were summarized in Supplementary Table S5.

E. coli recombinant strains were cultured in Luria-broth medium (OXIOD Biotech, London, United Kingdom) with 100 μg/ml ampicillin at 37 C. Yeast strains were routinely cultivated in yeast peptone dextrose (YPD) medium. Recombinant yeast strains were selected on synthetic complete medium (SED) consisting of 0.17% YNB (BD Diagnostics), 0.1% mono-sodium glutamate, and 0.06% appropriate amino acid drop out mix (MP Biomedicals, Solon, OH, United States), supplemented with 2% dextrose and 200 μg/ml G418 (Sangon Biotech, Shanghai, China). Positive yeast transformants were precultured in YPD for 24 h and then inoculated into 20 ml YPD medium in a 250 ml shake flask. After growing at 30°C and 250 rpm for 24 h, the yeast strains were transferred to a new 250 ml shake flask containing YP medium with the appropriate carbon source (glucose, galactose, or glycerol). GAMG shake flask fermentation was performed at 30°C and 250 rpm for 6 days.

For product quantification, yeast cells were harvested from 20 ml culture by centrifugation at 6,000 rpm, washed twice, and vacuum freeze-dried at to a constant weight. Then dry cell pellets were transferred into a 2 ml tube with 0.8 ml of an acetone: methanol mixture (1:1), and the resuspended yeast cells were crushed three times with glass beads. The samples were centrifuged at 12,000 rpm for 1 min, and the supernatants filtered with a 0.22 μm syringe filter were analyzed by LC-MS. The standard of chemicals were purchased from Yuanye Bio-tech Co., Ltd (Shanghai, China) and Chengdu DeSiTe Biological Technology Co., Ltd (Chengdu, China).

GAMG and GA were analyzed by LC-ESI-MS (SHIMADZU, Tokyo, Japan). 2 μL filtrate was injected and separated using a HyPURITY™ C18 HPLC (150 mm × 4.6 mm, 3 μm, Thermofisher Scientific) column with an ESI ion source equipped with a triple quadrupole mass analyzer. The mobile phase comprised of 0.1% (v/v) formic acid solution (A) and acetonitrile (B) at a flow rate of 0.2 ml/min. The gradient elution program was as follows: 0–1 min, 2% B; 1–10 min, 2.0%–80% B; 10–15 min, 80%–90% B; 15–18 min, 90%–98% B; 18–35 min, 98%–2% B; 35–40 min, 2% B. The column temperature was set at 30°C. GAMG and GA were monitored using multiple-reaction monitoring (MRM) in the negative ion mode with the following parameters: GAMG with a collision energy of −22 eV and an m/z transition from 647.35 to 453.25; GA with a collision energy of −40 eV and an m/z transition from 471.35 to 351.05.

To verify the feasibility of the designed pathway, the glycosylation module (Module III) was firstly tested. The synthetic codon optimized GuUGT73F15 and AtUDH (encoding UDP-glucose dehydrogenase from Arabidopsis thaliana to catalyze the generation of UDP-GlcA from UDP-Glc) were consecutively cloned into the pESC-LEU plasmid. The engineered strains CI04-1-1, harboring GuUGT73F15 and AtUDH expression cassettes, was cultivated in YP-Galactose medium supplemented with GA, and the supernatant as well as cell pellets were analyzed by LC-MS. A new compound (peak 1), with the same retention time and MS spectrum as those of the GAMG standard, was detected in the samples (cell pellets) only when GA was supplemented (Supplementary Figure S1). Moreover, another new compound (peak 3) [m/z 631.6] was also identified as GA-monoglucoside by MS spectra. These were consistent with the result of Chen et al. (2019) that GuUGT73F15 was able to accept UDP-GlcA and UDP-Glc as the sugar donor for the glycosylation of GA. Although GuUGT73F15 was able to accept UDP-Glc as one of the sugar donor, it also could produce a proportion of GAMG when UDP-Glc appeared in vivo. Therefore, GuUGT73F15 and AtUDH were used in subsequent studies.

After the proof-of-concept verification, the previously constructed strain BY4741-C-04, where each gene of the mevalonate (MVA) pathway was overexpressed to enhance the accumulation of isopentenyl pyrophosphate (IPP), was used as the parent strain for GAMG production (Wang C. et al., 2019) (Supplementary Table S5). Further introduction of GAMG biosynthetic pathway genes (Figure 2A), including ERG1, the β-amyrin synthase encoding gene (βAS), two cytochrome P450 enzyme (CYP) encoding genes (CYP88D6 and CYP72A63), the CYP reductase encoding gene (GuCPR1/CrCYB5), as well as ERG20 and ERG9, into the chromosome of CI04-1-1 (GA1, Supplementary Table S5), GA and GAMG was accumulated to 2.4 μg/L and 0.02 μg/L, respectively. Although de novo biosynthesis of GAMG was achieved, the production level was extremely low, indicating that the GA module (Module I) and the UDP-GlcA module (Module II) should be further optimized to redirect the metabolic fluxes towards GAMG biosynthesis.

Subsequently, the GA module (Module I) was optimized to enhance the production of GA and GAMG. Previous studies found the two CYPs (GuCYP88D6 and MtCYP72A63) to be rate-limiting for GA biosynthesis. Compared with the two CYPs used for the construction of GA1, CYPs with higher catalytic efficiency have been reported, such as UNI25647 with the C-11 hydroxylation activity (Zhu et al., 2017) and CYP72A63 mutant (T388S/W205A) with C-30 hydroxylation activity (Brown et al., 2015; Sun et al., 2020). Therefore, CYP88D6 and CYP72A63 in GA1 was replaced by UNI25647 and mutCYP72A63 to improve the production of GAMG in yeast. The replacement of CYP88D6 with UNI25647 resulted in the construction of strain GA8, whose production of GA and GAMG was increased approximately 9.8- and 154- fold to 25.76 μg/L and 3.17 μg/L, respectively. GA10 was constructed by integrating another copy of UNI25647 expression cassette into GA8, and the production of GA and GAMG were further improved by 1.68- and 1.94-fold, respectively. The replacement of CYP72A63 in strain GA8 and GA10 by its mutant (T388S/W205A) resulted in the construction of strain GA9 and GA11, whose production of GA and GAMG were even higher (Sun et al., 2020). Compared with the parent strain GA1, the production of GA and GAMG in GA11 were increased by 31- and 419-fold, respectively, via molecular engineering of the two CYP encoding genes (Zhu et al., 2017; Sun et al., 2020).

Besides molecular engineering of CYPs, the microenvironment for CYP folding and catalysis could be engineered for optimal performance as well. Considering that most plant CYPs were endoplasmic reticulum (ER) membrane localized and required NADPH for electron transfer (Hausjell et al., 2018), ER expansion and enhanced NADPH supply were commonly employed to improve the enzymatic activities of CYPs. Previous studies demonstrated that the deletion of OPI1 enabled ER expansion and then provided more space for the folding of CYPs (Arendt et al., 2017; Kim et al., 2019). Additionally, NADPH supply could be increased by introducing GAPN (NADP⁺-dependent glyceraldehyde-3-phosphate dehydrogenase from Streptococcus mutans) (Chen et al., 2014) and overexpressing ZWF1 (Kwak et al., 2019). Thus, strain GA13 was constructed by inserting the GAPN and ZWF1 expression cassettes into the OPI1 locus of GA11. The production of GA and GAMG in GA13 was 1.4- and 1.7-fold higher than those of GA11 (Figure 2B, 105.78 μg/L and 14.24 μg/L), indicating that engineering of the CYP microenvironment was beneficial for improving the production of GA and GAMG in yeast. Although the production of GA and GAMG were dramatically increased, the conversion rate (conversion rate was calculated by the molar ratio of GA to GAMG) of GA to GA was only ∼13%, indicating that the glycosylation of GA to GAMG has potential to be further improved.

Then, the UDP-GlcA module (Module II) was explored for the optimization of the GAMG biosynthetic pathway, including the enhanced supply of UDP-GlcA and inhibition of GAMG hydrolysis (Figure 3A). The supply of UDP-Glc could be enhanced by overexpressing the endogenous PGM1, PGM2, and UGP1 (Zhuang et al., 2017; Wang et al., 2019b; Wang et al., 2020). Therefore, these three genes together with AtUDH were integrated into strain GA13, resulting in the construction of strain GA14. Unfortunately, the production of GAMG was not significantly increased (Figure 3B), we speculate that overexpressing the endogenous PGM1, PGM2, and UGP1 is more benefit for the glucose promoter direct genes than the genes are driven by galactose promoters. In addition, EGH1 (ergosteryl-β-glucosidase) was reported to hydrolyze ergosteryl-β-glucoside and those with similar structures (Watanabe et al., 2015). For example, the deletion of EGH1 prevented the degradation Rh2, and accordingly enhanced Rh2 accumulation in yeast (Zhuang et al., 2017). Considering the similar structure of GAMG and Rh2, the deletion of EGH1 was expected to prevent the degradation of GAMG. Hence, GA14 was further engineered by the deletion of EGH1 (GA15). The production of GAMG in GA15 was increased to 19.63 μg/L (Figure 3B). Overall, the conversion rate of GA to GAMG was improved to 20.22% via the enhancement of UDP-GlcA supply and prevention of GAMG degradation.

Finally, the effect of carbon sources on GAMG production was investigated using the highest GAMG-producing strain GA15(Supplementary Figure S2). Considering the disruption of GAL80 in all the engineered strains, heterologous genes under the control of galactose-inducible promoters could be induced even without the presence of galactose. Thus, glucose, a much cheaper carbon source, was evaluated for GAMG fermentation. Interestingly, the replacement of carbon source in the fermentation medium from galactose to glucose resulted in more than 2-fold improvement in the production of GAMG (Figure 4), probably due to better supply of UDP-GlcA and accordingly enhanced GA glycosylation. In addition, a previous study demonstrated that the supplementation of glycerol into the fermentation medium increased noscapine production substantially, probably due to the generation of the essential cofactor NADPH and enhanced stability of the subcellular membrane localized proteins (Gekko and Serge, 1981; Crowe et al., 1984; Crowe et al., 1988; Sato et al., 1996; Costenoble et al., 2000; Bolen, 2001; Diamant et al., 2001; Meng et al., 2001; Perlmutter and David, 2002; Li et al., 2018). Therefore, glycerol at different concentrations (0, 2, 5, 8, and 10%) was supplemented to investigate GAMG production. As shown in Figure 4, 2% glycerol supplementation led to the highest production of GAMG (92.00 μg/L). Moreover, the conversion yield of GA to GAMG was improved to about 56%. However, higher glycerol concentration resulted in impaired GA and GAMG production, which was consistent with the previous study that glycerol concentration should be maintained at an appropriate level for optimal production of noscapine (Li et al., 2018).