Tailored biosynthesis of gibberellin plant hormones in yeast

The application of small amounts of natural plant growth hormones, such as gibberellins (GAs), can increase the productivity and quality of many vegetable and fruit crops. However, gibberellin growth hormones usage is limited by the high cost of their production, which is currently based on fermentation of a natural fungal producer Fusarium fujikuroi that produces a mix of several GAs. We explored the potential of the oleaginous yeast Yarrowia lipolytica to produce specific profiles of GAs. Firstly, the production of the GA-precursor ent-kaurenoic acid (KA) at 3.75 mg/L was achieved by expression of biosynthetic enzymes from the plant Arabidopsis thaliana and upregulation of the mevalonate (MVA) pathway. We then built a GA4-producing strain by extending the GA-biosynthetic pathway and upregulating the MVApathway further, resulting in 17.29 mg/L GA4. Additional expression of the F. fujikoroi GA-biosynthetic enzymes resulted in the production of GA7 (trace amounts) and GA3 (2.93 mg/L). Lastly, through protein engineering and the expression of additional KA-biosynthetic genes, we increased the GA3-production 4.4-fold resulting in 12.81 mg/L. The developed system presents a promising resource for the recombinant production of specific gibberellins, identifying bottlenecks in GA biosynthesis, and discovering new GA biosynthetic genes. Classification: Biological Sciences, Applied Biological Sciences.


Introduction
Gibberellins (GAs) are diterpene phytohormones involved in plant developmental processes like seed germination, pollen maturation, stem elongation, flower formation, and fruit development (Kato, 1955;Lang, 1956;Harrington et al., 1957;Kahn et al., 1957;Wittwer et al., 1957). Of the at least 126 known gibberellins, some are bioactive while others are inert side-products or biosynthetic intermediates (MacMillan, 2001). Bioactive gibberellins include GA 3 (gibberellic acid), GA 4 , and GA 7 ; they are used in commercial agricultural products, either pure or as mixtures (Camara et al., 2018). Spraying with gibberellins increases the size of seedless grapes, blueberries, and pears (Casanova et al., 2009;Ito et al., 2016;Zang et al., 2016), as well as the salt stress tolerance of soybeans, maize, and sugarcane (Tuna et al., 2008;Hamayun et al., 2010;Shomeili et al., 2011). Valent Biosciences, a commercial provider of GA-products, lists more than 40 plant species that can be enhanced by gibberellins (Rademacher, 2015). Gibberellins are also applied in beer brewing to enhance the malting process (MacLeod and Millar, 1962). According to some sources, the estimated global market size for gibberellins is USD 548.9 million as of 2016 (Gibberellins Market | Industry Analysis Report, 2016).
While several fungi species like Sphaceloma manihoticola or Phaeosphaeria sp. and bacteria species like Rhizobium phaseoli or Azospirillum lipoferum can produce gibberellins, the industrially used microbe for gibberellin production is the fungus Fusarium fujikuroi (formerly Gibberella fujikuroi) (Rademacher and Graebe, 1979;Atzorn et al., 1988;Bottini et al., 1989;Sassa et al., 1994;Shi et al., 2017). Submerged fermentation is most common as solid-state fermentation is complicated to scale-up (Camara et al., 2018). Titers of 3.9 g/L GA 3 for an eight-day submerged fermentation with fungal mycelium immobilized on calcium-polygalacturonate beads have been reported in the literature (Escamilla S et al., 2000). While GA 3 from submerged fermentation can be recovered with column-based adsorption or liquid-liquid extraction, it is more challenging to separate GA 4 from GA 7 due to their high structural similarity (Rademacher, 2015;Camara et al., 2018). Since GA 4 degrades faster than GA 3 and GA 7 , it can be more useful for some applications, yet most GA 4 products contain at least some GA 7 (Rademacher, 2015).
While literature describes mutated strains of F. fujikuroi that are able to produce predominately GA 7 at 234 mg/L or solely GA 4 at 420 mg/L during ten days of fermentation, it is unclear whether these strains can be used industrially (Lale and Gadre, 2010;Albermann et al., 2013a). Although genome data and transformation procedures are available for F. fujikuroi, the molecular toolkit for genome modification is still limited compared to more conventional host organisms (Wiemann et al., 2013;Hwang and Ahn, 2016;Bashyal et al., 2017;Shi et al., 2019). Furthermore, commercial efforts to produce pure GA 4 with S. manihoticola have failed (Rademacher, 2015). Moving the production into a heterologous host with more molecular tools available and established procedures for industrial-scale production could enable improved control of relative levels of bioactive GAs, better titers, and reduced production costs.
This study demonstrates the production of bioactive gibberellins in Y. lipolytica by optimizing ent-kaurenoic acid production and establishing heterologous pathways combining enzymes originating from both fungi and plants for the complete biosynthesis of bioactive gibberellins GA 3 , GA 4 , and GA 7 .
Escherichia coli strain DH5α was used for DNA manipulations. Cells were cultivated at 37 • C on Lysogeny Broth (LB) medium supplemented with 100 mg/L ampicillin for plasmid selection.
The chemicals were obtained, if not indicated otherwise, from Sigma-Aldrich. Nourseothricin was purchased from Jena BioScience GmbH (Germany).
Vectors pCfB6607 and pCfB6408 were constructed via point mutation of parental plasmids using primers indicated in Supplementary  Table S4 and following indications of QuickChange II Site-Directed Mutagenesis Kit (Agilent Technologies).

Yeast strain construction
All strains used in this study are listed in Supplementary Table S7. The plasmids for targeted integration were NotI-linearized and transformed into Y. lipolytica cells using an adapted protocol based on the published lithium acetate transformation protocol . In detail, the parental strain was spread over a YPD agar and grown for 24 h at 30 • C prior to transformation. Cells were scraped from the plate and washed with water. Subsequently, a certain amount of cells (volume corresponding to OD 600 9.2 per transformation) was resuspended in 100 μL of a transformation mix containing 43.8% poly-ethylene glycol (PEG 3350), 100 mM dithiothreitol, 0.25 g/L single-stranded DNA and 0.1 M LiAc. DNA was then added (a minimum of 500 ng). After a heat-shock at 39 • C for 1 h, cells were pelleted for 5 min at 3000 rpm. The transformation mix was removed, and cell pellets were resuspended in 1 mL YPD and allowed to recover for 2 h at 30 • C, 250 rpm agitation. Finally, cells were pelleted for 5 min at 3000 rpm, resuspended in 100 μL sterile water, and plated on selective medium.
Plates were incubated at 30 • C for 2-5 days until colonies appeared.
All strains were verified by PCR using insert-specific oligos in combination with oligos specific to the regions outside the different recombination sites. The CreA-loxP-mediated selection marker loop-out was performed as described previously . The elimination of the selection marker was verified by PCR in addition to phenotypic tests.

Cultivation
For KA-quantification, single colonies were inoculated from fresh plates into 3 mL YPD in 24-well plates with air-penetrable lids (EnzyScreen, NL) and grown for 24 h at 30 • C and 300 rpm agitation as pre-cultures. For cultivation, the required volume of the pre-culture was transferred to 50 ml YP+8%D in 250 mL baffled shake flasks for an initial optical density at 600 nm (OD 600 ) of 1.0. The shake flask cultures were cultivated for 72 h at 30 • C with 300 rpm agitation.
For GA-quantification, single colonies were inoculated from fresh plates into 2.5 or 3 mL YPD in 24-well plates with air-penetrable lids and grown for 16-24 h at 30 • C and 300 rpm agitation as pre-cultures. For cultivation, the required volume of the pre-culture was transferred to 2.5 or 3 mL YP+8%D in 24-well plates for an initial optical density at 600 nm (OD 600 ) of 0.1. The plates were cultivated for 72 h at 30 • C with 300 rpm agitation. By the end of the cultivation, either cell dry weight, OD 600 , or both were measured. OD 600 was measured using a Nano-Photometer (Implen GmbH, Germany). All cultivations were done in biological triplicates.
For the feeding experiments with GA 9 , GA 4 or GA 7 , pre-cultures and cultivation were done under the same conditions as for GAquantification. During the cultivation step, 100 mg/L of either GA 9 , GA 4 or GA 7 was added to the media (3 or 2.5 ml of YP+8% glucose). The stock solutions of GA 9 , GA 4, and GA 7 were 5 mg/mL in absolute ethanol, which yielded a final ethanol concentration of 2% in the media.
For growth profiling of selected strains, pre-culture preparations from single colonies were inoculated from fresh plates into 2.5 YPD in 24-well plates with air-penetrable lids and grown for ~16 h at 30 • C and 300 rpm agitation. Then the required volume of pre-culture was transferred to 1 mL of YP+8%D in 24-roundwell plates (Enzyscreen, CR1424f) with air-penetrable lids. The plates were cultivated, and their growth continuously measured in a Growth Profiler 960 (Enzyscreen) at 30 • C with 225 rpm agitation for 72 h. All growth profiling experiments were conducted in biological triplicates.

Analytical methods
For the KA-quantification, GA 12 -detection, GA-analysis of ST6513 and ST6514, and feeding experiments, the supernatant was sampled directly in the case of cell-free media or extracted by the following protocol. 300 μL of cultivation broth was transferred into a 2 ml microtube (Sarstedt). 500 μL of 0.5-0.75 mm acid-washed glass beads and 1.2 mL of acetonitrile were added to each tube. Thereafter, the cells were disrupted with a Precellys®24 homogenizer (Bertin Corp.) four times at 5500 rpm for 10 s, with samples being kept on ice in between rounds of breaking. The samples were then shaken for 10 min at room temperature. Lastly, the samples were centrifuged, and the supernatant fractions were sampled for analysis.
For quantification of GA 9 , GA 4 , GA 7 , and GA 3 for all other strains, 1 mL of broth were transferred to a 2 mL microtube. The samples were centrifuged, and the supernatant was extracted for analysis. Samples were diluted in water prior to analysis.
Furthermore, to investigate the presence of GA 9 , GA 4 , GA 7 , and GA 3 in ST8580 cell pellets, 1 mL of broth was transferred 2 mL microtube. The cells were pelleted by centrifugation, and the supernatant was removed. The cell pellets were washed twice by resuspension in 1 mL of water, centrifugation, and removal of the liquid phase. Thereafter, 500 μl of 0.5-0.75 mm acid-washed glass beads and 1 mL of acetonitrile were added to each tube. Subsequently, the cells were disrupted with a Pre-cellys®24 homogenizer four times at 5500 rpm for 10 s, with samples being kept on ice in between rounds of breaking. The samples were then shaken for 10 min in a DVX-2500 Multi-tube vortexer at room temperature. Lastly, the tubes were centrifuged, and the acetonitrile fractions were sampled for analysis.
The quantitative LC-MS data for the feeding experiments, KAquantification, GA-quantification for ST6513 and ST6514, and GA 12 mass confirmation can be seen in the supplementary methods.
The LC-MS/MS analysis of GAs for all other strains was performed on a Vanquish Duo UHPLC binary system (Thermo Fisher Scientific, USA) coupled to IDX-Orbitrap Mass Spectrometer (Thermo Fisher Scientific, USA). The chromatographic separation was achieved using a Waters ACQUITY BEH C18 (10 cm × 2.1 mm, 1.7 μm) equipped with an ACQUITY BEH C18 guard column kept at 40 C and using a flow rate of 0.35 mL/min. The mobile phases consisted of MilliQ© water + O.1% formic acid (A) and acetonitrile + 0.1% formic acid (B). The initial composition was 2%B held for 0.8 min, followed by a linear gradient till 5% in 3.3 min, and afterward, 100%B was reached in 10 min and held for 1 min before going back to initial conditions. Re-equilibration time was 2.7 min. The MS measurement was done in negative-heated electrospray ionization (HESI) mode with a voltage of 2500 V acquiring in full MS/MS spectra (Data dependent Acquisition-driven MS/MS) in the mass range of 70-1000 Da. The authentic GA 3 (012 2503), GA 4 (012 2532), GA 7 (012 2543), and GA 9 (012 2663) standards were acquired from OlChemIm s.r.o. (Czech Republic) and the detection limits of all GAs were 0.05 mg/L.

Establishing the biosynthetic pathway to ent-kaurenoic acid
We began with evaluating plant and fungal biosynthetic pathways for ent-kaurenoic acid production in Y. lipolytica. All genes that were not native to Y. lipolytica were codon-optimized. To produce the GAprecursor ent-kaurenoic acid via the plant pathway, we expressed the genes encoding copalyl diphosphate synthase (AtCPS), ent-kaurene synthase (AtKS), ent-kaurene oxidase (AtKO), NADPH cytochrome P450 reductase (AtATR2) from A. thaliana, and the native Y. lipolytica cytochrome b5 (YlCyb5) to generate strain ST6687 (Fig. 2). The genes encoding AtCPS, AtKS, and AtKO were full-length with the native plastidial signal peptides intact. Furthermore, an MVA-pathway optimized strain ST6349 was constructed that expressed the same genes as ST6687, a truncated version of the native 3-hydroxy-3-methyl-glutaryl-coenzyme A reductase (tHMG), and the native geranylgeranyl diphosphate synthase (GGPPS), while the endogenous squalene synthase gene was downregulated by truncating its promoter down to 50 base pairs (SQS↓) (Kildegaard et al., 2017). The HMG-reductase (HMGp) that reduces HMG-CoA to mevalonate was previously reported as one of the flux-controlling enzymes in the MVA-pathway (Cao et al., 2016(Cao et al., , 2017Yang et al., 2016). Furthermore, in Saccharomyces cerevisiae, removing the N-terminal transmembrane domain of HMGp resulted in a non-regulated protein retaining its catalytic activity (Donald et al., 1997).
Overexpression of GGPPSp has been shown to increase the production of β-carotene in Y. lipolytica and gibberellins in F. fujikuroi (Albermann et al., 2013b;Kildegaard et al., 2017). The final modification in the MVA-optimized strains is the truncation of the native squalene promoter, which was previously shown to increase β-carotene production in Y. lipolytica (Kildegaard et al., 2017).
The MVA-optimized strain ST6349 expressing the plant KAbiosynthetic pathway produced 3.75 mg/L KA, while no KA was detected for the non-optimized strain ST6687 (Fig. 3A and B). Surprisingly, the strain ST6521 with optimized MVA-pathway and expressing fungal enzymes did not produce measurable amounts of KA.

Production of the early gibberellin intermediates
Since KA production was achieved with the plant pathway enzymes, we sought to reconstruct the pathway towards bioactive GAs with mainly plant-derived enzymes. The strain ST6439 for production of the inert intermediate GA 12 was constructed based on the ent-kaurene producer ST6295 and further expression of AtKO, A. thaliana ent-kaurenoic acid oxidase (AtKAO), YlCyb5, and AtATR2. After cultivation, a compound identified as GA 12 could be detected for ST6439 based on targeted ion mass and retention time (Fig. S1). Since no standard was available, the amount of putative GA 12 could not be quantified.

Introduction of the biosynthetic pathway for bioactive gibberellin production
As the next step, we wanted to extend the pathway from GA 12 to the biologically active gibberellin GA 4 . For this, we expressed codonoptimized versions of the A. thaliana GA C20-oxidase (AtC20ox) and GA C3-oxidase (AtC3ox) in the GA 12 -producing strain ST6439 to generate ST6513 that should be able to produce GA 4 from GA 12 via the intermediates GA 15 , GA 24, and GA 9 . However, the cultivation of ST6513 did not result in detectable amounts of GA 4 . However, we still attempted to further extend the pathway from GA 4 to GA 7 and to GA 3 by expressing fungal enzymes GfP450-3p and GfDESp generating ST6514.
While GA 9 was detected for ST6514, neither GA 4 , GA 7, nor GA 3 were detected (Table S2). This indicated that AtC20oxp was active in ST6514 and catalyzed the formation of GA 9 from GA 12 . To discern whether the three biosynthetic enzymes, AtC3oxp, GfDESp, and GfP450-3p, were active, we conducted several precursor feeding experiments. When strain ST6514 was cultivated in a medium supplemented with GA 9 , it produced detectable amounts of GA 4 , which was not the case for the control strain ST3683 (Table S2). This confirmed that AtC3oxp was active in ST6514 and that the formation of GA 4 was limited by GA 9supply. When the ST6514 was supplemented with GA 4 , the presence of GA 7 was detected; However, GA 7 was also detected when ST3683 and cell-free media were supplemented with GA 4 (Table S2 and S3). This suggests that GA 4 is either spontaneously converted to GA 7 or that the commercial GA 4 product contains small amounts of GA 7 . The manufactures (OlChemIm s.r.o.) note that the GA 4 standard we used is ≥90% pure, which indicates the presence of impurities (OlchemIm s.r.o.| Products, 2021). Furthermore, other providers note that their GA 4 standards contain at least some GA 7 (Sigma-Aldrich|Gibberellin A4, 2021). Indeed, GA 4 and GA 7 are difficult to separate during purification, which may explain such impurities (Rademacher, 2015;Camara et al., 2018). Supplementing the cultivation medium of ST6514 with GA 7 did not lead to a detectable amount of GA 3 (Table S2). In summary, the results indicate that AtC20oxP and AtC3oxp were active , while GfP450-3p was likely inactive when expressed in ST6514.

Production of GA 4 by further engineering
We decided to perform additional metabolic engineering of the yeast strain ST6349 to increase the flux towards the GA-biosynthesis. To simplify the genome editing, we integrated a codon-optimized cas9 gene from Streptococcus pyogenes into GA 12 -producing strain ST6439, resulting in strain ST8257 (Fig. 2). This enabled marker-free integration of DNA constructs . Firstly, to further boost the production of GGPP, a codon-optimized version of the Synechococcus sp. GGPP synthase (SsGGPPS7) was expressed in ST8257, giving strain ST8414. Expression of SsGGPPS7 was previously shown to substantially increase carotenoid production in Y. lipolytica (Tramontin et al., 2019). Secondly, three cytochromes P450 partners: cytochrome b5, cytochrome b5 reductase, and cytochrome P450 reductase (GfCyb5, GfCyb5Red, GfCPR) from F. fujikuroi were expressed to promote the electron transfer from NADPH to F. fujikuroi cytochromes P450 for the last steps towards GA 3 , resulting in strain ST8416. Finally, AtC20ox and AtC3ox were expressed under stronger promoters to overcome the potential rate-limitation of these steps. The genes AtC20ox and AtC3ox were expressed under the control of the promoters PrExp and PrTefintron, respectively, to generate strain ST8447 and under promoters PrTefintron and PrGpd, respectively, to generate ST8448. Both strains should be capable of producing GA 4 , albeit potentially at different levels due to the different promoter pairs used. Furthermore, the PrTefintron promoter is currently the strongest known constitutive promoter for Y. lipolytica (Tai and Stephanopoulos, 2013;Holkenbrink et al., 2018).
Cultivation of ST8448 resulted in the production of 1.20 mg/L GA 4 and trace amounts of GA 9 ( Fig. 3C and D, and S2A), whereas no GA 4 or GA 9 could be detected for ST8447. The difference between the strains may be explained by the low expression level of AtC20ox from the PrExp promoter in ST8447 . After establishing de novo production of GA 4 in yeast, we proceeded to complete the biosynthetic pathway towards GA 7 and GA 3 . For this, GfDES and GfP450-3 were expressed in both strains ST8447 and ST8448, resulting in strains ST8504 and ST8505, respectively. Cultivation of ST8504 resulted in the production of 0.77 mg/L GA 4 and trace amounts of GA 9, while neither GA 7 nor GA 3 were detected ( Fig. 3C and D, and S2A). None of these four compounds were detected in strain ST8505. This contradicts the previous results, where only the parental strain of ST8505 produced GA 4 .
Feeding of ST8504 with GA 7 did not result in detectable amounts of GA 3, which suggested that GfP450-3p was inactive in this strain (Table S3). This strain also grew very poorly when supplemented with GA 7 and 2% ethanol (3 g dry cell weight/L) compared to without (15 g dry cell weight/L). Furthermore, GA 4 was also detected when both ST3683 and cell-free media were supplemented with GA 7 , possibly due to product impurities (Table S3). 3.5. Production of GA 3 , GA 4 , and GA 7 by the introduction of fungal cytochromes P450 Since no GA 3 could be detected by expressing the plant genes AtC20ox, AtC3ox, and fungal genes GfDES and GfP450-3, we sought to reconstruct the biosynthetic pathway KA to GA 3 , GA 7 , and GA 4 with fungal enzymes.
We built on top of the KA-producing strain ST6440, into which, through several rounds of transformation, we integrated the genes encoding for Cas9, SsGGPPS, and P450 auxiliary proteins GfCyb5, GfCybRed, GfCPR, generating ST8417. Since Cas9 expression may be toxic to yeast, we investigated the growth profiles of the Cas9 expressing strain ST8258 and its parental strain ST6440, however, no reductions in the growth of ST8258 compared to ST6440 were found ( Fig. S3A and  S3B).
Furthermore, we expressed two fungal genes GfP450-1 and GfP450-2, encoding cytochromes P450 that should convert KA into GA 4 . The resulting strain ST8602 produced 17.29 mg/L GA 4 and 0.81 mg/L GA 9 ( Fig. 4A and B, and S2B). The presence of GA 9 is likely due to the oxygenation of GA 12 -aldehyde by GfP450-1p resulting in GA 12 , which is then processed into GA 9 by the action of GfP450-2p (Tudzynski et al., 2002). The titer of GA 4 in strain ST8602 expressing GfP450-1 and GfP450-2 was 14.4-fold greater than what was achieved for the analogous strain ST8448 expressing AtC20ox and AtC3ox. As the next step, the strain ST8580 was built that combined all the modifications as in ST8602 and additionally expressed GfP450-3 and GfDES. Cultivation of ST8580 resulted in the production of 2.93 mg/L GA 3 , trace amounts of GA 7 , 2.64 mg/L GA 4 , and 0.77 mg/L GA 9 (Fig. 4A-D, and S2B). The identities of the compounds were confirmed by retention time and high-resolution mass-spectrometry compared to authentic standards (Fig. S4). Furthermore, analysis of the cell fraction of ST8580 did not show detectable amounts of GA 7 and only trace amounts of GA 3 , GA 4 , and GA 9 , indicating that all GAs were secreted. Lastly, we investigated the growth profiles of the GA-producing strains ST8580 and ST8602 compared to their parental strain ST8417. There were no significant differences in growth ( Fig. S3A and S3C).

Improved GA-production by increasing early GA-pathway flux
To increase GA-production, we expressed either GfCPS/KS and GfKS or AtCPS, AtKS, and AtKO in ST8580, generating strains ST10333 and ST10332, respectively. A fused enzyme consisting of AtCPS C-terminally linked to the N-terminal of AtKS (AtCPS/KS) was expressed together with AtKO in ST8580, generating ST10331. Furthermore, truncated variants of the A. thaliana KA-biosynthetic enzymes with N-terminal plastidial targeting sequences removed (tAtCPS, tAtKS, tAtKO) were expressed under individual promoters or with tAtCPS fused to tAtKS (tAtCPS/KS) alongside the individual expression of tAtKO in ST8580 generating ST10350 and ST10334, respectively. The ChloroP software and earlier reports were used to predict the plastidial targeting sequences of AtCPSp and AtKOp based on which the 2-60 and 2-28 amino acids were removed to make tAtCPSp and tAtKOp, respectively (Emanuelsson et al., 1999;Helliwell et al., 2001b). Of note, a transmembrane region at amino acid positions 6-24 was predicted for AtKOp by the InterProScan software, whereas no transmembrane regions were predicted for pAtCPS and pAtKS (Jones et al., 2014;Blum et al., 2021). No plastidial targeting sequence for AtKSp was predicted by ChloroP; however, earlier reports confirm the presence of a plastidial targeting sequence in AtKSp and suggest that the 44 first amino acids of AtKSp are involved in the translocation (Yamaguchi et al., 1998;Helliwell et al., 2001b). ST10350 exhibited the most enhanced GA-production with 12.81 mg/L GA 3 , 16.41 mg/L GA 4 , and 4.70 mg/L GA 9 , a 4.4-, 6.2-and 6.1-fold increase over ST8580, respectively (Fig. 5). Interestingly, all strains constructed from ST8580 demonstrated improved growth rates compared to ST8580 on YP+8%D media, likely due to the restoration of uracil prototrophy (Fig. S3A and S3D).

Discussion
The gibberellins GA 4 , GA 7 , and GA 3 are plant hormones with several applications in agriculture. The current production methods for these GAs are based on fermentation of the fungus F. fujikuroi (Shi et al., 2017). We have demonstrated the production of GA 3 , GA 4 , and GA 7 in recombinant yeast Y. lipolytica by expression of biosynthetic pathway enzymes from F. fujikuroi and A. thaliana. Fig. 4. GA-production by engineered Y. lipolytica strains. A. Overview of GA-producing strains showing which GA-biosynthetic genes were expressed. B. GA 4production for ST8417, ST8602, and ST8580. C. GA 7 -production for ST8417, ST8602, and ST8580. D. GA 3 -production for ST8417, ST8602, and ST8580. TR, trace amounts. Three biological replicates were used to calculate titer averages and standard deviations for all strains.
The production of KA was established by improving the flux towards GGPP and expression of AtCPS, AtKS, and AtKO. Interestingly, AtKOp has previously been functionally characterized by expression in S. cerevisiae, demonstrating its functionality in another yeast species (Helliwell et al., 1999). Yet, the presence of cofactors, substrate, and pH-levels may differ in Y. lipolytica compared to the native host environment, which may affect the activity of heterologous enzymes. Indeed, in vitro kinetic assays of N-terminally truncated AtCPS demonstrate optimal catalytic rates at 0.1 mM of the cofactor Mg 2+ with decreasing enzymatic activity at both greater and lesser Mg 2+ -concentrations. A similar inhibitory effect is exerted by the substrate GGPP, while the optimal pH for truncated AtCPS was 7.5-8 (Prisic and Peters, 2007). Differences in such factors could also explain why the expression of GfCPS/KS and GfKO did not result in KA-production. Furthermore, there exists different algorithms for codon-optimization and it has been shown that codon-optimization can sometimes result in decreased expression or impaired folding of heterologous proteins (Fath et al., 2011;Komar A. A., 2019;Saito et al., 2019;Fu et al., 2020).
It would be interesting for future research to investigate the activity of heterologous KA-biosynthetic genes from various sources and with DNA sequences based on different codon-optimization algorithms in Y. lipolytica. Notably, the first two enzymes of KA-biosynthesis, AtCPSp, and AtKSp, are located in plastids in plants, while AtKOp is located on the outer plastidial membrane, and the rest of the GA-biosynthetic enzymes are located in the ER and cytosol (Helliwell et al., 2001b;Hedden and Sponsel, 2015). In the case of AtCPSp, the protein is processed into the mature enzyme during transport into the chloroplast, likely by removal of the N-terminal plastidial targeting peptide, which may yield a better performing enzyme (Sun and Kamiya, 1994;Helliwell et al., 2001b). Indeed, the highest GA 3 production was achieved by expression of tAtCPS, tAtKS, and tAtKO, which suggests that N-terminal truncation may be a viable strategy for increasing enzyme activity.
Furthermore, the removal of plastidial targeting peptides from heterologous diterpene synthases has also been successfully used to construct yeast cell factories for the production of 13R-manoyl oxide . In fact, targeting diterpene synthases to the plant cytosol by removal of targeting peptides alongside MVA-pathway upregulation resulted in increased production of both taxadiene, 13R-manoyl oxide, and forskolin in Nicotiana benthamiana compared to the expression of nontruncated diterpene synthases and MEP-pathway upregulation (De La Peña and Sattely, 2020). Interestingly, a transmembrane domain within the truncated region was predicted for tAt-KOp, which likely renders the protein cytosolic upon removal.
Additionally, expression of fused AtCPS/KS or tAtCPS/KS together with AtKO or tAtKO, respectively, also led to improved GA-production. The fusion of the CPSp and KSp from Salvia miltiorrhiza led to enhanced production of ent-kaurene in S. cerevisiae (Su et al., 2016). Interestingly, certain mosses and liverworts have naturally occurring bifunctional CPS/KSp enzymes similar to F. fujikuroi (Hayashi et al., 2006;Kawaide et al., 2011). Therefore, enzymatic fusion could be further explored to improve GA-production in Y. lipolytica.
The further expression of AtKAOp resulted in the production of a compound putatively identified as GA 12 . This enzyme has previously been characterized and co-expressed with either AtATR1 or AtATR2 in S. cerevisiae, where co-expression with AtATR2 led to increased relative amounts of GA 12 (Helliwell et al., 2001a). This indicates that optimal redox partner pairing with cytochromes P450 like AtKAOp can lead to improved activity in heterologous systems. Although GA 12 is considered to be biologically inert, it is implicated in long-range GA signaling in A. thaliana, since GA 12 may be transported via plant vascular tissue, converted to bioactive GAs, and thereby activate GA signaling cascades in receiving tissues, resulting in decreased accumulation of growth inhibitory DELLA proteins (Regnault et al., 2015). Therefore, agricultural GA 12 -application could potentially be useful to enhance plant growth. Furthermore, GA 12 can be converted by the "Gain of Function in ABA-modulated Seed Germination 2"-enzyme (AtGas2p) to GA 12 16, 17-dihydro-16α-ol (DHGA 12 ), which can enhance seed germination and seedling development in A. thaliana . Cell factories for GA 4 -production for ST8580, ST10333, ST10332, ST10331, ST10350, and ST10334. C. GA 7 -production for ST8580, ST10333, ST10332, ST10331, ST10350, and ST10334. D. GA 3 -production for ST8580, ST10333, ST10332, ST10331, ST10350, and ST10334. TR, trace amounts. Three biological replicates were used to calculate titer averages and standard deviations for all strains. Statistical significance (two-tailed student's t-test) compared to ST8580 is represented by asterisks (*, p < 0.05. **, p < 0.001. NS, not significant). the production of GA 12 and bioactive GA 12 -derivatives may therefore become useful in the future.
While we attempted to complete the biosynthetic pathway towards GA 12 to GA 3 by expression of AtC20ox, AtC3ox, GfDES, and GfP450-3 to generate ST6514, neither GA 4 , GA 7 , nor GA 3 were produced by this strain. The feeding studies indicated that GfP450-3p was not active in ST6514. However, GA 3 was produced when GfP450-3p was expressed in ST8580. Interestingly, ST6514 expressed only AtATR2 and YlCyb5, while ST8580 expressed the fungal redox partners GfCyb5, GfCybRed, and GfCPR in addition to the prior. Therefore, the presence of specific redox partners may be necessary to support the activity of F. fujikuroi cytochromes P450. Indeed, previous research has demonstrated that the activity of heterologous cytochromes P450 in yeast depends heavily on their co-expressed redox partners. The cytochrome P450 FoCYP53A19p from Fusarium oxysporum could convert 3-hydroxybenzoic acid to 3,4dihydroxybenzoic acid when paired with an F. oxysporum CPR but not when paired with an S. cerevisiae or Candida albicans CPR (Durairaj et al., 2015). Furthermore, the activity of heterologous cytochromes P450 could be affected by codon-optimization and promoter choice, since the available constitutive promoters for Y. lipolytica vary greatly in strength .
The GA 3 -titer of 12.81 mg/L reported herein is still very low compared to the highest reported titer of 3.9 g/L for the natural producer F. fujikuroi (Escamilla S et al., 2000). However, we demonstrate that GA-production of our strains can be improved considerably by engineering the KA-biosynthetic pathway, and many existing strategies for terpenoid overproduction could be utilized for further improvements.
Indeed, Y. lipolytica is a relatively new host for recombinant terpene production, but it has already outperformed S. cerevisiae for the production of β-carotene (Larroude et al., 2018;Moser and Pichler, 2019). This is likely due to the higher acetyl-CoA flux and the lipophilic environment in Y. lipolytica, which may serve to sequester and store lipophilic terpenes (Christen and Sauer, 2011). Therefore, Y. lipolytica holds great potential as a microbial cell factory for terpenoid production.

Conclusions
Production of bioactive gibberellins GA 3 , GA 4, and GA 7 in Y. lipolytica was achieved by expressing a GA-biosynthetic pathway with both plant and fungal enzymes. The complete biosynthesis of GA 3 , GA 4, and GA 7 was done by heterologous expression of A. thaliana enzymes for KA-production and F. fujikuroi enzymes for the subsequent biosynthetic steps. We demonstrate that the GA-production could be enhanced by further engineering of the yeast chassis. This yeast-based platform holds the potential to improve the future production of bioactive GAs.

Declaration of competing interest
No conflicts of interest.

Acknowledgments
The research was funded by the Novo Nordisk Foundation (Grant agreement NNF15OC0016592, NNF20CC0035580, and NNF20OC0060809). IB and KRK acknowledge the financial support from the European Union's Horizon 2020 research and innovation programs (European Research Council,. We are grateful to Larissa Ribeiro Ramos Tramontin for the kind gift of pCfB8092 and to Dr. Mahsa Babaei for the kind gift of PR-27422 and PR-27423.