For the development of all engineered strains, S. cerevisiae 3HP-F was utilized as the initial strain (Zhang et al., 2019). Jinkairui Biotechnology Co., Ltd. (Wuhan, China) codon optimized and synthesized the DNA fragments encoding SmCPS (copalyl diphosphate synthase) (GenBank: EU003997.1), SmKSL (kaurene synthase-like enzyme) (GenBank: EF635966.2), CYP76AH24 (11-hydroxy-ferruginol synthase) (GenBank: KT157044.1), CYP76AK6 (CA synthase) (GenBank: KT157045.1), GuCPR (cytochrome P450 reductase) (GenBank: QCZ35624.1) and AtCPR (GenBank: BT008426.1), and cloned them into the plasmid pUC57. GENEWIZ (Beijing, China) codon optimized and synthesized the DNA fragment encoding SmCPR (GenBank: CBX24555). Yeast transformants were screened on synthetic medium lacking specific amino acids. Standard culture was conducted in yeast extract peptone dextrose medium (YPD), which comprised 2% peptone, 1% yeast extract, and 2% glucose. GENEWIZ (Beijing, China) synthesized the primers. TIANGEN (Beijing, China) provided the mini plasmid extraction and DNA gel mini purification kits. S. cerevisiae W303-1a genomic DNA was used to amplify the promoter, terminator, optional marker, and other native sequences.

We used the lithium acetate method to transform S. cerevisiae as described previously (Gietz and Schiestl, 2007). The expression cassette containing the coding DNA sequence, the promoter and terminator was constructed using fusion polymerase chain reaction (PCR). The expression cassettes and the expression fragments are shown in Supplementary Table S1 and Supplementary Figure S1. The primers, coding DNA sequences, promoter and terminator sequences used in the construction of strains are shown in Supplementary Tables S2–S4.

A single colony was inoculated into a tube containing 3 ml of YPD medium and cultivated for 16–18 h at 220 rpm at 30°C. The culture was then inoculated into a shake flask with 30 ml of YPD medium to an initial OD₆₀₀ of 0.05, then cultivated for 96 h under the same condition. All the experiments in conical flasks were done in triplicate. A conventional spectrophotometer (Oppler, 752 N, China) was used to determine the OD₆₀₀.

Miltiradiene, ferruginol, and CA were extracted using n-hexane from lysed cells and the supernatant of the fermentation broth. The fermentation broth was centrifuged at 10000 g for 10 min. The cell precipitate and supernatant were separated into two centrifuge tubes, after which n-hexane was added to both, and additional glass beads were added to the cell precipitate. Both tubes were then shaken vigorously in a vortex shaker for 40 min and centrifuged. The upper organic phases were aspirated with a syringe and combined. Ferruginol and CA standards were purchased from Solarbio (Beijing, China).

Miltiradiene was identified by gas chromatography-mass spectrometry. A Shimadzu GC-MS-TQ8030 apparatus with a GC column HP-5ms (Agilent Technologies, 30 m × 0.250 mm × 0.25 μm) was used to analyze the samples (1 μL). The temperature gradient was as follows: injection temperature, 250°C; 5 min at 150°C, ramp at 5°C/min to 250°C, then hold for 5 min. The spectra were scanned between 30 and 550 m/z with an ion source temperature of 260°C.

Ferruginol and CA were identified by liquid chromatography-mass spectrometry (LC-MS). High-performance liquid chromatography (HPLC) with an Elite P230II high-pressure pump system was used to quantify ferruginol and CA. A Grace Apollo C18 column (4.6 × 250 mm, 5 mm) was used for chromatographic separation. The detection ultraviolet is 230 nm. The LC-MS analysis was carried out using an Agilent Zorbax SB Aq column (100 mm × 2.1 mm × 3 μm) with Surveyor LC System (Thermo Finnigan, San Jose, CA, United States). Negative ionization was used to evaluate the samples. The elution conditions were as follows: the injection volume was 30 μL; the eluent was 40:60 water: acetonitrile; the column temperature was 30°C; the flow rate was 1 ml/min; scanning mode: first-level full scan. For compound identification, we compared the mass spectra and retention times with authentic standards.

A bioanalyzer (SBA-40C, Shandong Academy of Sciences, China) and an Aminex HPX-87H column (Bio-Rad, United States) were used to measure glucose and ethanol concentrations. The flow rate was 0.6 ml/min; the eluent was 5 mM H₂SO₄; the column temperature was 65°C.

The internal standard for miltiradiene quantitation was 1-eicosene. A calibration curve with an R² coefficient of more than 0.99 (Supplementary Figure S6) was used to quantify the ferruginol and CA in LC analysis. The highest and lowest deviations from three different cultivations were represented by error bars. Univariate analysis (t-test) was used to assess the statistical significance (p-value).

A single colony was inoculated into a tube containing 3 ml of YPD medium and cultivated for 16–18 h at 220 rpm at 30°C. The culture was used to inoculate another tube with 3 ml of YPD medium to 0.2 initial OD₆₀₀, then cultivated at 220 rpm and 30°C for 48 h for fluorescence measurements using Infinite 200 PRO Multimode Microplate Reader (TECAN, Switzerland). All the experiments were done in triplicate.

The number of integrated heterologous expression cassettes was determined using absolute qPCR with Arg4 as an internal reference gene (Abad et al., 2010). We used the primers listed in Supplementary Table S2 to quantify the open reading frames of CYP76AH24, CYP76AK6, CYP76AH1, SmCPR, and ARG4 genes to generate standard curves. Genomic DNA was extracted from distinct colonies and three repeated qPCR assays were conducted using the LightCycler 480 System with TaqMan probe qPCR TB Green Premix Ex Tap II (Vazyme, China). Detection based on the TaqMan probe was described by Zhang et al. (2014).

The preserved engineered S. cerevisiae strains were streaked onto a YPD plate and activated to obtain a single colony. This single colony was cultured in a test tube containing 3 ml of YPD at 220 rpm and 30°C for 16–18 h, and the resulting seed culture was transferred to 200 ml of YPD in a shake flask and cultured for 16–18 h under the same condition. The secondary seed culture was inoculated 2 L of YPD medium in a 5 L fermenter at a volume ratio of 1:10. The automatic addition of 2.5 M sulfuric acid and 5 M ammonia kept the pH constant at 5.5. The airflow was 2 vvm, the fermentation temperature was 30°C, and the rotation speed was varied between 200 and 600 rpm to maintain the dissolved oxygen (DO) above 35% of the atmospheric value.

The conditions for two-stage fed-batch fermentation were the same as the batch fermentation. When the initial glucose was completely spent, feeding was initiated with a solution containing 500 g/L glucose, 5.12 g/L MgSO₄·7H₂O, 0.28 g/L Na₂SO₄, 3.5 g/L K₂SO₄, 9 g/L KH₂PO₄, 0.6 g/L uracil, 0.5 g/L adenine, 1.2 g/L lysine, 12 ml vitamin solution and 10 ml trace element solution. As previously reported, the vitamin and trace element solution was prepared (Zhou et al., 2010). When glucose consumption was complete and the ethanol produced in the process of metabolism was consumed, feeding with ethanol (95%, V/V) was initiated and the ethanol concentration in the fermenter was kept at 1–6 g/L in addition to 350 ml of feeding solution components other than glucose was added. The highest and lowest deviations from three different cultivations were represented by error bars.

The codon-optimized miltiradiene synthesis genes from Salvia miltiorrhiza (SmCPS and SmKSL) were integrated into S. cerevisiae 3HP-F to form the WM1 strain. The strong promoters P _PGK1 and P _TDH3 were selected to express these two genes, respectively. Based on published mass spectra, the new peak (retention time (RT) = 17.82 min) was identified as miltiradiene (Hu et al., 2020) (Figure 2). We compared the peak areas of the internal standard 1-eicosene and miltiradiene to quantify a titer of 0.18 mg/L for miltiradiene after 4 days of shake flask culture.

Plant diterpenes are synthesized in plastids, and diterpene synthases usually contain an N-terminal plastidic transit peptide (Bohlmann et al., 1998). However, the transit peptide was reported to hinder the heterologous expression of diterpene synthases (Hu et al., 2020). Here, we identified the transit peptide of miltiradiene synthases using the ChloroP online tool (https://www.cbs.dtu.dk/services/ChloroP/) to construct the truncated variants tSmCPS and tSmKSL. Supplementary Table S3 shows the plastidic transit peptide coding sequences. The strong promoters P _PGK1 and P _TDH3 were selected to co-express tSmCPS and tSmKSL from the ura3 locus of the strain 3HP-F to form WM2, which produced 1.27 mg/L miltiradiene, representing a 7.06-fold increase over the original strain (Figure 3). Jiang et al. (2017) removed the plastidic transit peptide from the N-terminus of geraniol synthase and fused a red fluorescent protein to its C-terminus to characterize the expression level of the correctly folded protein using fluorescence measurements. To investigate the expression of tSmCPS and tSmKSL, we fused the red fluorescent protein mCherry to the C-terminus of SmCPS, tSmCPS, SmKSL, and tSmKSL and expressed these four fusion proteins in S. cerevisiae. The relative fluorescence (RFU) of tSmCPS and tSmKSL was respectively 2.1 and 2.5 times higher than that of SmCPS and SmKSL, indicating that the expression of both enzymes was enhanced (Supplementary Figure S2).

Fusion proteins can boost the effective local concentration of substrates and enzymes, improving the product yield. Linker length has a crucial effect on the function of fusion proteins (Haga et al., 2013). For example, extremely long linkers make the fusion proteins prone to degradation and affect the product yield, while extremely short linkers affect the spatial conformation of the protein, which causes it to fail to play its original catalytic role. We fused tSmCPS and tSmKSL using five flexible linkers to obtain strains WM3a-e (Figure 3). The GSTSSGSSG linker had the best effect, and the titer of miltiradiene reached 10.98 mg/L, 8.65 times higher than strain WM2, co-expressing tSmCPS and tSmKSL (Supplementary Figure S3). Overexpression of the BTS1-GGGS-ERG20, a fusion of two critical enzymes of the MVA pathway, can considerably increase miltiradiene production (Zhou et al., 2012). In addition the ERG20 ^(F96C) mutant can considerably enhance GGPP accumulation while having no apparent effect on FPP production (Ignea et al., 2015). Accordingly, BTS1-GGGS-ERG20 ^F96Cp was overexpressed to create the WM4 strain, which reached a miltiradiene titer of 172.77 mg/L (Supplementary Figure S3).

P450s are frequently utilized in the synthesis of drugs, vitamins, and spices (Ignea et al., 2015). However, few plant P450s exhibited high activity, and have been optimized using protein engineering, metabolic engineering, redox chaperone engineering, and substrate engineering approaches to increase terpene production (Xiao et al., 2019). Thus, choosing an appropriate functional cytochrome P450 reductase (CPR) is essential for maximizing the redox coupling efficiency.

Three distinct CPR encoding genes were chosen for co-expression with P450s CYP76AH24 and CYP76AK6 (from S. miltiorrhiza) in WM4, resulting in strains WCA1a (AtCPR from Arabidopsis thaliana), WCA1b (GuCPR from Glycyrrhiza uralensis), and WCA1c (SmCPR from S. miltiorrhiza). The processed samples were analyzed using LC–MS after 4 days of culture. As shown in Supplementary Figure S4, CA and ferruginol were identified as new peaks by LC–MS analysis. WCA1c containing SmCPR produced 52.5 μg/L of CA, 70% more than WCA1a containing AtCPR and 30% more than strain WCA1b containing GuCPR (Figure 4A). A high coupling efficiency and electron transfer compatibility were reported in the homologous CYP-CPR reconstituted system, which boosted the monooxygenase activity (Dietrich et al., 2005; Jensen and Møller, 2010), and was consistent with our findings.

The conversion rate of the precursor miltiradiene was low, and may be attributed to the low efficiency of spontaneous oxidation of miltiradiene into abietatriene. By contrast, CYP76AH1 can reportedly directly oxidize miltiradiene to produce ferruginol (Guo et al., 2013). To further improve the CA titer, we overexpressed CYP76AH1 to develop the WCA2 strain. The ferruginol and CA titers reached 0.97 and 0.81 mg/L, indicating that CYP76AH expression could increase the conversion of miltiradiene into ferruginol (Figure 4B).

Cytochrome B5 (Cytb5) acts as an electron transporter in many biological oxidation reactions (Zhang et al., 2007). It appears to act as a specific electron donor when involved in catalysis together with NADPH-cytochrome B5 reductase or NADPH-CPR (Gilep et al., 2001). Therefore, we overexpressed Cytb5 from S. pomifera to develop the WCA3 strain with a CA titer of 2.62 mg/L, which was 3.21 times higher than in WCA2. Furthermore, the ferruginol titer also increased 4.27 times, reaching 4.15 mg/L (Figure 4B).

The fusion of CPR with a compatible P450 enzyme could reportedly result in evolutionary advantages in terms of catalytic efficiency (McLean et al., 2007), and a fusion of the P450s CYP3A4, NADPH-CPR, and Cytb5 showed high activity (Inui et al., 2007). Therefore, we fused CYP76AH1, SmCPR, and SpCytb5 to explore the optimal fusion strategy for CA production. In our previous study, protopanaxadiol synthase and a truncated variant of ATR1 without its N-terminal transmembrane region were used to construct a fusion protein with higher activity (Zhao et al., 2016). Here, we used the TMHMM server to predict the transmembrane regions of SmCPR and SpCytb5, as shown in Supplementary Table S3. First, the truncated SmCPR was fused with CYP76AH1 and co-expressed with native SpCytb5 to construct the WCA4a strain. However, the results showed a sharp decline in CA production relative to the WCA3 strain (Figures 5A,E). We then fused the truncated SpCytb5 with CYP76AH1 and SmCPR, respectively, and then co-expressed it with native SmCPR or CYP76AH1, resulting in the strains WCA4b and WCA4c. The results showed that the CA titer of WCA4b and WCA4c was improved compared with WCA3 (Figures 5B,C,E), whereby the product titer of the best strain WCA4c reached 3.17 mg/L (Figure 5C), indicating that the fusion of truncated SpCytb5 and SmCPR improved the electron transport efficiency. Finally, we fused the truncated SmCPR, truncated SpCytb5, and CYP76AH1 to construct the WCA4d strain. However, its CA titer was unsatisfactory (Figure 5D). According to these results, we found that after excising the transmembrane region of SmCPR protein, the CA titer was significantly reduced regardless of the fusion method (Figure 5). In addition, we detected carnosol, a spontaneous oxidation product of CA, in WCA4c, which is in line with earlier research (Scheler et al., 2016). Because the product titer of strain WCA4c showed the largest increase, we chose the coexpression of SmCPR∼t28SpCytb5 fusion protein and CYP76AH1 for further experiments.

To further improve CA production, we chose a multicopy site to integrate the expression cassettes encoding the CA synthesis pathway. First, WCA4c was used as the chassis strain, and the SmCPR∼t28SpCytb5 fusion gene and CYP76AH1 were co-expressed from the δ site to obtain the WCA5 strain with CA and ferruginol titers of 4.30 and 12.74 mg/L, respectively. Then, we integrated the CYP76AH24 and CYP76AK6 genes into the ribosomal DNA site, resulting in the WCA6 strain with the highest CA titer of 6.20 mg/L. The gene copy number of strain WCA6 is shown in Supplementary Figure S5.

The normal aerobic metabolism of S. cerevisiae is accompanied by the generation of reactive oxygen species (Temple et al., 2005). In the heterologous synthesis of terpene products in microorganisms, P450s and their reductases are sometimes poorly coupled, resulting in the generation of reactive oxygen species (Paddon et al., 2013). In the CA production strain, we expressed three heterologous P450s, which may have caused excessive accumulation of reactive oxygen species. H₂O₂ is accumulated due to ATP synthesis and spontaneous or enzymatic hydrolysis of O₂ during respiratory metabolism, which can cause damage to the cell (Temple et al., 2005). Catalase is a primary H₂O₂ scavenging enzyme essential for preserving intracellular homeostasis of reactive oxygen species. In S. cerevisiae, the ScCTA1 catalase is localized to peroxisomes and mitochondria, where it degrades H₂O₂ produced during aerobic respiration and β-oxidation (Dzanaeva et al., 2020). The second ScCTT1 catalase is localized in the cytoplasm and responds to oxidative stress caused by H₂O₂ accumulation (Martins et al., 2019).

As shown in Figure 6A, we expressed ScCTA1 from the MET17 locus of WCA6 to construct the WCA7a strain. Overexpression of ScCTA1 increased CA production to 7.57 mg/L and ferruginol production to 11.38 mg/L. Next, we expressed ScCTT1 from the MET17 locus of WCA6 to construct the WCA7b strain, which exhibited increases of CA and ferruginol production to 8.84 mg/L and 14.64 mg/L. However, when the two genes were simultaneously overexpressed, CA production was reduced. Therefore, the WCA7b strain was selected for subsequent experiments.

The subcellular organelles of eukaryotes have different functions, providing a unique physicochemical environment and critical functions for cell survival. For instance, the Endoplasmic Reticulum (ER) offers a microenvironment designed for rapid and precise protein processing (Malhotra et al., 2008). Accordingly, the ER volume is a crucial determinant of the protein folding capacity of cells (Schuck et al., 2009). The ER membrane is mainly constituted by phospholipids, and the transcription factor INO2 can activate the expression of related genes to promote phospholipid synthesis, thus increasing the area of the ER membrane. In a recent study, overexpression of INO2 in yeast expanded the ER area, leading to a 71-fold increase in squalene production (Kim et al., 2019). As shown in Figure 6B, the INO2 overexpression in strain WCA8 boosted the CA titer by 29%, reaching 11.42 mg/L, and the ferruginol titer was 20.25 mg/L.

Heme is the main cofactor of P450s, and modifying the endogenous heme synthesis pathway was reported as a viable approach to enhance the activity of P450s (Michener et al., 2012). We overexpressed the HEM3 (heme synthase) gene in the WCA9 strain, which increased the CA titer by 50% to 17.12 mg/L. To increase the NADPH supply, we overexpressed the NADH kinase gene (POS5) (Miyagi et al., 2009) in WCA10, which increased the CA titer by 20%–20.54 mg/L, whereby the ferruginol titer also increased to 33.86 mg/L (Figure 6B). The transcription factor HAC1 can activate the transcription of proteins folding-related genes, activating the unfolded protein response (UPR) in the ER (Khan and Schroeder, 2008; Jonikas et al., 2009; Qu et al., 2020). To reduce the pressure caused by the expansion of the ER, we overexpressed the transcription factor HAC1 in WCA11, which increased the CA and ferruginol titers to 24.65 and 36.29 mg/L, corresponding to yields of 1.23 and 1.81 mg/g, respectively (Figure 6B).

We performed batch and fed-batch fermentation of the best strain WCA11 in a 5-L bioreactor to confirm the shake-flasks results. The residual glucose concentration, cell growth, ethanol concentration, and CA titer were evaluated during batch fermentation. Glucose was rapidly consumed within 12 h at the beginning of the fermentation (Figure 7A). At the same time, ethanol accumulation reached up to 11.93 g/L, after which ethanol was the carbon source and consumed after 48 h. CA production was detected every 12 h from the beginning of the fermentation, and reached a peak of 31.04 mg/L at 96 h. The titer of ferruginol and miltiradiene reached 54.80 and 212.45 mg/L, respectively. The final OD₆₀₀ of the strain reached 29.23 (Figure 7A).

As shown in Figure 7B, the OD₆₀₀ of WCA11 was 3.33-fold higher in fed-batch fermentation than in batch fermentation. The CA titer also increased considerably, reaching 75.18 mg/L, representing a 142.2% increase compared with batch fermentation, which is the highest CA titer reported in yeast to date, to our knowledge, (Table 1). The ferruginol and miltiradiene titers also increased to 276.58 and 1,543.27 mg/L, respectively. The two-stage feeding strategy therefore showed good results (Figure 7B).