CRISPRi-Guided Metabolic Flux Engineering for Enhanced Protopanaxadiol Production in Saccharomyces cerevisiae

Protopanaxadiol (PPD), an aglycon found in several dammarene-type ginsenosides, has high potency as a pharmaceutical. Nevertheless, application of these ginsenosides has been limited because of the high production cost due to the rare content of PPD in Panax ginseng and a long cultivation time (4–6 years). For the biological mass production of the PPD, de novo biosynthetic pathways for PPD were introduced in Saccharomyces cerevisiae and the metabolic flux toward the target molecule was restructured to avoid competition for carbon sources between native metabolic pathways and de novo biosynthetic pathways producing PPD in S. cerevisiae. Here, we report a CRISPRi (clustered regularly interspaced short palindromic repeats interference)-based customized metabolic flux system which downregulates the lanosterol (a competing metabolite of dammarenediol-II (DD-II)) synthase in S. cerevisiae. With the CRISPRi-mediated suppression of lanosterol synthase and diversion of lanosterol to DD-II and PPD in S. cerevisiae, we increased PPD production 14.4-fold in shake-flask fermentation and 5.7-fold in a long-term batch-fed fermentation.

In yeast, however, 2,3-oxidosqualene is natively converted to lanosterol, catalyzed by lanosterol synthase, and then used for biosynthesis of sterols such as ergosterol [18]. Thus, in yeast, the PPD biosynthetic pathway competes with the native sterol biosynthetic pathway for 2,3-oxidosqualene. Since deletion of the ERG7 gene, which encodes lanosterol synthase in yeast, is lethal, it is not possible to completely block the sterol biosynthetic pathway [18]. Rather, it is necessary to quantitatively control ERG7 expression (i.e., partial suppression) in order to enhance the PPD biosynthetic pathway. For example, in Saccharomyces cerevisiae, mass production of heterologous terpenes, which are synthesized from MVA, requires partial suppression of the metabolic flux of native sterol biosynthesis, in order to increase the metabolic flux of the target terpenes; this is achieved via replacement of the native promoter with the methionine-repressible MET3 promoter, the copper-repressible CTR3 promoter, or a constitutively weak promoter, or via fusion of the target enzyme to a degradation tag [19][20][21][22][23][24][25][26]. However, this strategy has several limitations. First, adding methionine increases production costs because it is consumed by yeast. Second, adding copper can result in toxicity in yeast. Third, it is difficult to quantitatively control the metabolic flux by replacing the native promoter with a constitutively weak promoter; further, selection of a suitable promoters is labor-intensive and time-consuming [27][28][29][30].
To address this, we developed a CRISPRi-based PPD biosynthesis approach for yeast. We constructed a PPD-producing yeast strain and engineered it to respond to increasing levels of the precursors. This approach alters cellular carbon flux and partially suppresses competing metabolic pathways to improve PPD production in yeast.

Construction of the PPD-Producing Yeast Strain
Because dammarenediol-II synthase (DS), protopanaxadiol synthase (PPDS), and cytochrome P450 reductase (CPR) are necessary to construct the PPD-producing yeast strain, we first constructed PgDS and PgPPDS expression cassettes that are controlled by the GPD promoter and a PgCPR expression cassette that is controlled by the PGK1 promoter ( Figure 1); these three cassettes were then integrated into the CEN.PK2-1D strain, producing the PPD-A1 strain (Table 1). PgCPR1 together catalyze the hydroxylation of DD-II at the C-12 position, resulting in PPD biosynthesis. PPD then undergoes glycosylation catalyzed by several UDP-glucosyltransferases, yielding ginsenosides [13]. In yeast, however, 2,3-oxidosqualene is natively converted to lanosterol, catalyzed by lanosterol synthase, and then used for biosynthesis of sterols such as ergosterol [18]. Thus, in yeast, the PPD biosynthetic pathway competes with the native sterol biosynthetic pathway for 2,3-oxidosqualene. Since deletion of the ERG7 gene, which encodes lanosterol synthase in yeast, is lethal, it is not possible to completely block the sterol biosynthetic pathway [18]. Rather, it is necessary to quantitatively control ERG7 expression (i.e., partial suppression) in order to enhance the PPD biosynthetic pathway. For example, in Saccharomyces cerevisiae, mass production of heterologous terpenes, which are synthesized from MVA, requires partial suppression of the metabolic flux of native sterol biosynthesis, in order to increase the metabolic flux of the target terpenes; this is achieved via replacement of the native promoter with the methionine-repressible MET3 promoter, the copper-repressible CTR3 promoter, or a constitutively weak promoter, or via fusion of the target enzyme to a degradation tag [19][20][21][22][23][24][25][26]. However, this strategy has several limitations. First, adding methionine increases production costs because it is consumed by yeast. Second, adding copper can result in toxicity in yeast. Third, it is difficult to quantitatively control the metabolic flux by replacing the native promoter with a constitutively weak promoter; further, selection of a suitable promoters is labor-intensive and time-consuming [27][28][29][30].
To address this, we developed a CRISPRi-based PPD biosynthesis approach for yeast. We constructed a PPD-producing yeast strain and engineered it to respond to increasing levels of the precursors. This approach alters cellular carbon flux and partially suppresses competing metabolic pathways to improve PPD production in yeast.

Enhancing the MVA Pathway and Squalene Monooxygenase Stability Upregulated the Triterpene Biosynthetic Pathway
Enhancing MVA-pathway metabolic flux is an efficient way to improve triterpene production in yeast [7,11,[31][32][33]. We used tHMGR1-overexpression to improve PPD production ( Figure 1). We constructed the tHMGR1 expression cassette controlled by the GPD promoter and integrated it into the PPD-A1 strain, producing the PPD-A2 strain (Table 1), which was then subjected to shake-flask fermentation. However, the PPD production of PPD-A2 was not better than that of PPD-A1, although PPD-A2 produced substantially more squalene, at 564.4 mg/L ( Figure 2). PPD-A2 produced more lanosterol, at 38.6 mg/L, but did not differ from PPD-A1 in 2,3-oxidosqualene and ergosterol production. Therefore, tHMGR1-overexpression effectively increased the production of squalene and lanosterol, but not of PPD.
To convert more of the accumulated squalene into 2,3-oxidosqualene, it is necessary to overexpress ERG1. ERG1 is a key regulator of sterol homeostasis [18,34]. In yeast, excess lanosterol accumulation leads to ERG1 degradation via the ER-associated protein degradation (ERAD) pathway [34] found that ERG1 was a target of the ERAD-associated protein ubiquitin ligase Doa10, whereas an ERG1 derivate (K278R/K284R/K311R/K360R) was stabilized from ERAD-mediated protein degradation. Here, we chose to overexpress this ERG1-derivate (ERG1m hereafter), rather than the native ERG1, in the PPD-A2 strain. To do this, we constructed the ERG1m expression cassette controlled by the TEF1 promoter and integrated it into the PPD-A2 strain, producing the PPD-A3 strain. Following shake-flask fermentation of PPD-A3, squalene production was indeed lower in PPD-A3, at 96.1 mg/L, than in PPD-A2 (Figure 2), and PPD production was higher, at 2.2 mg/L. Notably, lanosterol production was also higher in PPD-A3, at 265.8 mg/L, whereas that of ergosterol was similar, relative to that of PPD-A2. Therefore, enhancing the MVA pathway via tHMGR1 overexpression and increasing ERG1 stability via ERG1m overexpression improved the triterpene biosynthetic pathway and particularly, lanosterol and PPD production.

CRISPRi-Mediated ERG7 Suppression Improved PPD Production
Although PPD-A3 produced more PPD than PPD-A2, lanosterol showed greater fold change than PPD from PPD-A2 to PPD-A3, suggesting that, in PPD-A3, most of the 2,3oxidosqualene was converted to lanosterol rather than DD-II and PPD. To further improve PPD production in engineered yeast strains, it is necessary to efficiently suppress lanosterol biosynthesis by quantitative suppression of ERG7 expression. Partial suppression of ERG7 expression using the methionine-suppressible MET3 promoter or antisense ERG7 fragment has been reported [23,31,35]. However, MET3 promoter-mediated ERG7 suppression requires extra methionine to continuously suppress ERG7 expression. Further, antisense ERG7 fragment-mediated ERG7 suppression did not quantitatively suppress ERG7 expression. In addition, the long antisense RNA involved might be unstable. To address these limitations, we used the CRISPRi system to constitutively and quantitatively suppress ERG7 expression. We first designed five ERG7 promoter-targeting sgRNAs (sgRNA1-5) using an online tool (https://lp2.github.io/yeast-crispri/, accessed 1 December 2017) ( Figure 3). We then synthesized five SNR52pro-sgRNA-SUP4ter cassettes and cloned them into the dCas9-expressing plasmid pTDH3-dCas9-Mxi1. To co-express the dCas9 and sgRNA in yeast, we constructed five dCas9-sgRNA cassettes and integrated them into the PPD-A3 strain, producing strains PPD-A3-sgRNA1-5. To determine whether the dCas9-sgRNA cassettes suppress ERG7 expression and improve PPD production in yeast, we performed shake-flask fermentation of strains PPD-A3 and PPD-A3-sgRNA1-5 in YPD medium for 48 h (Figures 3 and 4). We first analyzed the ERG7 expression levels of these strains using quantitative RT-PCR ( Figure 3). Relative to the expression in the control strain (PPD-A3), the ERG7 expression levels of strains PPD-A3-sgRNA1-5 were decreased 0.73-, 0.58-, 0.24-, 0.40-, and 0.66-fold, respectively. PPD-A3-sgRNA3 exhibited the most effective ERG7 suppression. To exclude the possibility that dCas9 expression differed among strains PPD-A3-sgRNA1-5, we analyzed their dCas9 expression using quantitative RT-PCR (Supplementary Figure S1); dCas9 expression was similar among these strains, suggesting that their different ERG7 expression resulted from different sgRNA efficiencies.   The cell growth of PPD-A3-sgRNA1-5 was lower than that of PPD-A3 (Figure 4A  and Supplementary Table S4). In particular, PPD-A3-sgRNA3 exhibited severe growth retardation, suggesting that excessive ERG7 suppression might damage cell viability. To determine whether ERG7 suppression by dCas9-sgRNAs resulted in the suppression of lanosterol biosynthesis and improvement of PPD production, we analyzed lanosterol and PPD production after shake-flask fermentation, using the PPD-A3 and PPD-A3-sgRNA1-5 strains ( Figure 4B). Lanosterol production in PPD-A3 was 258.6 mg/L, and was lower, at 198.2, 165.9, 44.8, 152.7, and 200.8 mg/L, respectively, in PPD-A3-sgRNA1-5. Notably, the order of the suppression efficiency of lanosterol biosynthesis in strains PPD-A3-sgRNA1-5 was the same as that for ERG7 expression, suggesting that dCas9-sgRNA could quantitatively suppress both ERG7 expression and lanosterol biosynthesis. PPD production by   The cell growth of PPD-A3-sgRNA1-5 was lower than that of PPD-A3 (Figure 4A  and Supplementary Table S4). In particular, PPD-A3-sgRNA3 exhibited severe growth retardation, suggesting that excessive ERG7 suppression might damage cell viability. To determine whether ERG7 suppression by dCas9-sgRNAs resulted in the suppression of lanosterol biosynthesis and improvement of PPD production, we analyzed lanosterol and PPD production after shake-flask fermentation, using the PPD-A3 and PPD-A3-sgRNA1-5 strains ( Figure 4B). Lanosterol production in PPD-A3 was 258.6 mg/L, and was lower, at 198.2, 165.9, 44.8, 152.7, and 200.8 mg/L, respectively, in PPD-A3-sgRNA1-5. Notably, the order of the suppression efficiency of lanosterol biosynthesis in strains PPD-A3-sgRNA1-5 was the same as that for ERG7 expression, suggesting that dCas9-sgRNA could quantitatively suppress both ERG7 expression and lanosterol biosynthesis. PPD production by The cell growth of PPD-A3-sgRNA1-5 was lower than that of PPD-A3 ( Figure 4A and Supplementary Table S4). In particular, PPD-A3-sgRNA3 exhibited severe growth retardation, suggesting that excessive ERG7 suppression might damage cell viability. To determine whether ERG7 suppression by dCas9-sgRNAs resulted in the suppression of lanosterol biosynthesis and improvement of PPD production, we analyzed lanosterol and PPD production after shake-flask fermentation, using the PPD-A3 and PPD-A3-sgRNA1-5 strains ( Figure 4B). Lanosterol production in PPD-A3 was 258.6 mg/L, and was lower, at 198.2, 165.9, 44.8, 152.7, and 200.8 mg/L, respectively, in PPD-A3-sgRNA1-5. Notably, the order of the suppression efficiency of lanosterol biosynthesis in strains PPD-A3-sgRNA1-5 was the same as that for ERG7 expression, suggesting that dCas9-sgRNA could quanti-tatively suppress both ERG7 expression and lanosterol biosynthesis. PPD production by PPD-A3 was 1.9 mg/L, and was higher, at 11.2, 23.0, 17.8, 27.6, and 11.1 mg/L, respectively, in PPD-A3-sgRNA1-5. Notably, PPD-A3-sgRNA3, which exhibited the strongest ERG7 suppression, did not exhibit the highest PPD production. Rather, PPD-A3-sgRNA4 exhibited the highest PPD production, suggesting that excessive ERG7 suppression might disadvantage the biosynthetic pathway of the secondary metabolite. This indicates that our method efficiently suppressed ERG7 expression, thereby improving PPD production in the engineered yeast strain.

PPD Production via Batch-Fed Fermentation
Although several studies have assessed the use of CRISPRi to regulate metabolic flux in yeast, most of them have been limited to a few days in duration [36][37][38][39]. To determine whether dCas9-sgRNA-mediated ERG7 suppression was effective over longer periods, we performed batch-fed fermentation, using the PPD-A3 and PPD-A3-sgRNA4 strains, in a 5-L bioreactor for 216 h; the strains exhibited similar cell growth profiles ( Figure 5A). Lanosterol production by PPD-A3 reached 541.5 mg/L at 120 h and maintained this until 216 h ( Figure 5B). For PPD-A3-sgRNA4, however, lanosterol production reached 210 mg/L at 120 h, then decreased to 140 mg/L at 216 h ( Figure 5B). At 216 h, PPD production by PPD-A3 and PPD-A3-sgRNA4 was 52.1 mg/L and 294.5 mg/L, respectively ( Figure 5C). This indicates that dCas9-sgRNA-mediated ERG7 suppression over longer periods effectively suppressed lanosterol biosynthesis and improved PPD production in the engineered strain. PPD-A3 was 1.9 mg/L, and was higher, at 11.2, 23.0, 17.8, 27.6, and 11.1 mg/L, respectively, in PPD-A3-sgRNA1-5. Notably, PPD-A3-sgRNA3, which exhibited the strongest ERG7 suppression, did not exhibit the highest PPD production. Rather, PPD-A3-sgRNA4 exhibited the highest PPD production, suggesting that excessive ERG7 suppression might disadvantage the biosynthetic pathway of the secondary metabolite. This indicates that our method efficiently suppressed ERG7 expression, thereby improving PPD production in the engineered yeast strain.

PPD Production via Batch-Fed Fermentation
Although several studies have assessed the use of CRISPRi to regulate metabolic flux in yeast, most of them have been limited to a few days in duration [36][37][38][39]. To determine whether dCas9-sgRNA-mediated ERG7 suppression was effective over longer periods, we performed batch-fed fermentation, using the PPD-A3 and PPD-A3-sgRNA4 strains, in a 5-L bioreactor for 216 h; the strains exhibited similar cell growth profiles ( Figure 5A). Lanosterol production by PPD-A3 reached 541.5 mg/L at 120 h and maintained this until 216 h ( Figure 5B). For PPD-A3-sgRNA4, however, lanosterol production reached 210 mg/L at 120 h, then decreased to 140 mg/L at 216 h ( Figure 5B). At 216 h, PPD production by PPD-A3 and PPD-A3-sgRNA4 was 52.1 mg/L and 294.5 mg/L, respectively ( Figure 5C). This indicates that dCas9-sgRNA-mediated ERG7 suppression over longer periods effectively suppressed lanosterol biosynthesis and improved PPD production in the engineered strain.

Discussion
PPD has important pharmacological properties, including anti-cancer activity. To improve production of this high-value-added product, we metabolically engineered yeast strains to biosynthesize it more efficiently. We first constructed a PPD-producing yeast strain in which tHMGR1 was overexpressed to enhance the MVA pathway. Next, we overexpressed the ERG1m gene in this engineered strain, to enhance squalene conversion to 2,3-oxidosqualene. In spite of this, PPD production remained low, because the PPD and native ergosterol biosynthetic pathways compete for 2,3-oxidosqualene. To overcome this

Discussion
PPD has important pharmacological properties, including anti-cancer activity. To improve production of this high-value-added product, we metabolically engineered yeast strains to biosynthesize it more efficiently. We first constructed a PPD-producing yeast strain in which tHMGR1 was overexpressed to enhance the MVA pathway. Next, we overexpressed the ERG1m gene in this engineered strain, to enhance squalene conversion to 2,3-oxidosqualene. In spite of this, PPD production remained low, because the PPD and native ergosterol biosynthetic pathways compete for 2,3-oxidosqualene. To overcome this competition, we established a CRISPRi-based metabolic engineering strategy, and achieved stable, sustained, high-level PPD production in S. cerevisiae.
In S. cerevisiae, HMGR1 and ERG1 are key regulatory enzymes in the sterol biosynthetic pathway [34,40]. They are regulated by negative feedback at transcriptional and protein levels, to maintain ergosterol homeostasis. When we overexpressed tHMGR1 in the PPD-producing yeast strain, squalene accumulated, whereas 2,3-oxidosqualene did not accumulate, suggesting that ERG1 is also a rate-limiting enzyme in the sterol biosynthetic pathway. When we further overexpressed the ERG1m gene in the PPD-producing strain and most of the squalene was converted to lanosterol and PPD. However, lanosterol production far exceeded PPD production, suggesting that most of the 2,3-oxidosqualene were converted to lanosterol rather than DD-II and PPD. We propose a possible reason for thisthe enzymatic activity of ERG7 might be stronger than that of PgDS. In humans, lanosterol synthase is a monotopic endoplasmic reticulum (ER)-membrane protein comprising two (α/α) barrel domains connected by loops, and three smaller β-structures [41]. The active site cavity is located in the center of the enzyme, and 2,3-oxidosqualene enters this active site cavity. Although PgDS also functions at the ER, it does not contain a transmembrane domain [42]. Thus, ERG7 might have a competitive advantage over PgDS in substrate binding. To overcome this competition, we decided to suppress the ERG7 expression in engineered yeast strains.
To suppress ERG7 expression, we applied CRISPRi-guided regulation of ERG7 expression, using dCas9 and five ERG7 promoter-targeting sgRNAs (sgRNA1-5). Since the stable expression of dCas9 and sgRNA is important for constitutive suppression of target gene expression, we integrated dCas9-sgRNA cassettes into chromosomes of the PPD-A3 strain. dCas9 expression was similar among these strains, whereas ERG7 expression differed substantially. The efficiency of the sgRNAs that we constructed for ERG7 suppression was ranked (in decreasing order), as follows: sgRNA3, sgRNA4, sgRNA2, sgRNA5, and sgRNA1. Notably, the dCas9-sgRNA3 cassette retarded cell growth, suggesting that there might be an ERG7 expression threshold that affects cell viability. It is possible, by using different sgRNAs targeting the same gene, to reveal differences in efficiency [37][38][39]42]. It is likely that sgRNA3 was the most effective at suppressing ERG7 expression because its target region is closest to the TATA-box in the ERG7 promoter. Indeed, sgRNA1 and sgRNA5, for which the target regions are distant from the TATA-box in the ERG7 promoter, were less effective at suppressing ERG7 expression. These findings indicate that the sgRNA target site is a key aspect in the quantitative regulation of gene expression. Further, with the exception of ERG7 suppression by sgRNA3, the relative differences in the efficiency of the constructed sgRNAs were consistent with those in PPD production. This indicates that CRISPRi-guided suppression of ERG7 expression suppressed lanosterol biosynthesis and improved PPD production.
Notably, CRISPRi-guided suppression of ERG7 resulted in the accumulation of squalene and 2,3-oxidosqualene, but not ergosterol. Yeast cells not only regulate sterol biosynthesis but also convert excessive sterols into steryl esters, which are then stored in lipid droplets or secreted into the extracellular matrix [18]. Considering that overexpression of tHMGR1 and EGR1m enhanced sterol biosynthesis by inhibiting negative feedback, accumulation of squalene and 2,3-oxidosqualene indicates that CRISPRi-guided ERG7 suppression is a powerful tool for regulating metabolic flux. To further enhance the conversion of 2,3-oxidosqualene to DD-II and PPD in yeast, it might be helpful to increase PgDS, PgPPDS, and PgCPR expression.
The stability of the dCas9-sgRNA complex, and the accuracy of sgRNA-targeting, are important factors for achieving stable and efficient regulation of target gene expression over longer periods, when using CRISPRi to bioengineer yeast strains. Most previous studies on this have been conducted over a few days, possibly because of difficulties in maintaining dCas9-sgRNA complex stability [37][38][39]42]. Here, we performed batch-fed fermentation over 9 days, using the PPD-A3 and PPD-A3-sgRNA4 strains, to confirm the effect of CRISPRi-guided ERG7 suppression over longer periods. At 216 h, lanosterol production was lower, but PPD production was higher in the PPD-A3-sgRNA4 strain than in the unmodified PPD-producing (control) strain. This is the first study to demonstrate the efficacy of CRISPRi-guided regulation of metabolic flux for longer periods. These findings may help in the metabolic engineering of industrial strains to improve production of target molecules.

Strains and Medium
Saccharomyces cerevisiae CEN.PK2-1D was obtained from EUROSCARF (http://www. euroscarf.de/) and used as a parent strain for all engineered yeast strains. The yeast strains are listed in Table 1. Constructed yeast strains were grown in YPD medium (10 g/L yeast extract, 20 g/L peptone, and 50 g/L glucose) or SD medium (6.7 g/L yeast nitrogen base and 50 g/L glucose) at 26 • C, with shaking at 220 rpm, lacking histidine, tryptophan, leucine, and uracil where appropriate.
Escherichia coli DH5α (Enzynomics, Republic of Korea) was used for transformation and plasmid amplification. Escherichia coli cells were grown in Luria-Bertani (LB) medium (10 g/L tryptone, 5 g/L yeast extract, and 10 g/L NaCl) with 100 mg/L ampicillin at 37 • C, with shaking at 200 rpm.
Strain PPD-A2 was constructed by integrating the tHMG1 gene downstream of the TCB2 site of PPD-A1. The [GPDpro-tHMG1-CYC1ter]-URA3 cassette, with the homologous recombination region of the partial TCB2 site, was amplified from the tHMG1-carrying pRS426GPD vector, using the primer set TCB2-Integ-F/TCB2-Integ-R. This integration cassette was transformed into strain PPD-A1, followed by selection on an SD/-URA plate. Strains were verified by diagnostic PCR, and colonies containing the integration cassette were cultivated at 30 • C in SD/-URA medium, producing the PPD-A2 strain.
Strain PPD-A3 was constructed by integrating the ERG1m gene into the trp1-289 site of PPD-A2. The [TEF1pro-ERG1m-CYC1ter]-TRP1 cassette, with the homologous recombination region of the partial TRP1 gene, was amplified from the ERG1m-carrying pRS424TEF1 vector, using the primer set TRP1-Integ-F/TRP1-Integ-R. This integration cassette was transformed into strain PPD-A2, followed by selection on an SD/-URA/-TRP plate. Strains were verified by diagnostic PCR, and colonies containing integration cassette were cultivated at 30 • C in SD/-URA/-TRP medium, producing the PPD-A3 strain.

Yeast Cultivation and Batch-Fed Fermentation
For shake-flask fermentation, YPD medium was used to cultivate the yeast strains. First, 1 mL of stock cells in 25% glycerol was inoculated into a 250 mL baffled flask (TriForest, Irvine, CA, USA) containing 29 mL of YPD medium and cultivated at 26 • C, with shaking at 220 rpm, to an optical density at 600 nm (OD600) of approximately 6.0 which was measured by a GENESYS 20 visible spectrophotometer (Thermo Scientific, Waltham, MA, USA). Then, 1 mL of seed culture was inoculated into a 250 mL baffled flask containing 29 mL of YPD medium and cultivated at 26 • C, with shaking at 220 rpm, for 48 h. Flask-fermentation results are presented as means with standard deviation based on biological triplicates.
Strains PPD-A3 and PPD-A3-sgRNA4 were used for the production of PPD via batchfed fermentation in a stirred glass tank 5 L bioreactor (CNS, Daejeon, Korea), with an initial working volume of 2 L YPD medium. Seed culture was prepared in two steps. First, 1 mL of stock cells in 25% glycerol was inoculated into a 250 mL baffled flask containing 29 mL of YPD medium, followed by cultivation at 26 • C and shaking at 220 rpm, to an OD600 of approximately 6.0. Then, 12 mL of the first seed culture was inoculated into two 1 L baffled flasks (Duran) containing 138 mL of YPD medium, and cultivated at 26 • C with shaking at 220 rpm, to an OD600 of approximately 7.0. Then, 300 mL of the second seed culture was inoculated into the 5 L bioreactor containing 1.7 L of YPD medium. Fermentation was carried out at 26 • C with shaking at 300 rpm and air flow at 4 L/min. pH was controlled at 5.5 by automatic addition of 15% ammonium hydroxide (v/v). To minimize foaming in the bioreactor, 10% Antifoam 204 (v/v) was used. After the initial glucose was completely consumed, a solution containing 500 g/L glucose, 18.7 g/L KH 2 PO 4 , 6.5 g/L K 2 SO 4 , 0.53 g/L Na 2 SO 4 , 9.75 g/L MgSO 4 ·7H 2 O, 10 g/L histidine, 10 g/L leucine, 10 mL/L of trace metal solution, and 12 mL/L of vitamin solution [23], was added to the bioreactor, and the dissolved oxygen level was increased to above 50%. The glucose concentration in the bioreactor was maintained below 10 g/L.

Metabolite Extraction and HPLC Analysis
Yeast cells at a density equivalent to an OD600 of 40 were collected from the flaskfermentation and batch-fed fermentation processes into 2 mL Safe-Lock Eppendorf tubes, and centrifuged at 13,000× g for 5 min. The supernatant was discarded, and the collected cells were resuspended in 1 mL of a methanol-acetone mixture (1:1 v/v). The cells were then lysed using an MM400 homogenizer (Retsch, Germany), according to the manufacturer's protocol. Samples were then centrifuged at 13,000× g for 10 min. The supernatant (30 µL) was then injected into an Agilent 1260 Infinity II HPLC system (Agilent, Santa Clara, CA, USA), with UV detection at 203 nm. Chromatographic separation was conducted using a Prodigy 5 µm ODS-2 LC column (4.6 mm × 150 mm, Phenomenex, Torrance, CA, USA). The mobile phase consisted of water (A) and acetonitrile (B), using a gradient program of 32-65% B at 0-8 min, 65-90% B at 8-12 min, 90% B at 12-20 min, 90-100% B at 20-30 min, 100% B at 30-65 min, 100-32% B at 65-66 min, and 32% B at 66-70 min. The solvent flow rate was 1.0 mL/min and the column temperature was set to 30 • C. Analytical grade squalene, 2,3-oxidosqualene, lanosterol, and ergosterol were purchased from Sigma-Aldrich (Burlington, MA, USA). Dammarenediol-II and PPD were purchased from ChemFaces (Wuhan, China).

RNA Extraction and qRT-PCR
Yeast cells at a density equivalent to an OD600 of 20 were collected from the flaskfermentation process into a 2 mL Safe-Lock Eppendorf tube and centrifuged at 13,000× g for 5 min. Total RNA was extracted using the RNeasy Mini Kit (Qiagen, Germany), and cDNA was obtained using M-MLV Reverse Transcriptase (Thermo Scientific, Waltham, MA, USA), both according to the manufacturers' protocol. The relative mRNA level of each gene was determined using real-time PCR (Bio-Rad, Hercules, CA, USA), using specific primer sets (Supplementary Table S3), with ACTIN1 as the reference gene.