De novo production of resveratrol from glucose or ethanol by engineered Saccharomyces cerevisiae

Resveratrol is a natural antioxidant compound, used as food supplement and cosmetic ingredient. Microbial production of resveratrol has until now been achieved by supplementation of expensive substrates, p -coumaric acid or aromatic amino acids. Here we engineered the yeast Saccharomyces cerevisiae to produce resveratrol directly from glucose or ethanol via tyrosine intermediate. First we introduced the biosynthetic pathway, consisting of tyrosine ammonia-lyase from Herpetosiphon aur-antiacus , 4-coumaryl-CoA ligase from Arabidopsis thaliana and resveratrol synthase from Vitis vinifera , and obtained 2.73 7 0.05 mg L (cid:1) 1 resveratrol from glucose. Then we over-expressed feedback-insensitive alleles of ARO 4 encoding 3-deoxy-D-arabino-heptulosonate-7-phosphate and ARO 7 encoding chorismate mutase, resulting in production of 4.85 7 0.31 mg L (cid:1) 1 resveratrol from glucose as the sole carbon source. Next we improved the supply of the precursor malonyl-CoA by over-expressing a post-translational de-regulated version of the acetyl-CoA carboxylase encoding gene ACC 1; this strategy further increased resveratrol production to 6.39 7 0.03 mg L (cid:1) 1 . Subsequently, we improved the strain by performing multiple-integration of pathway genes resulting in resveratrol production of 235.57 7 7.00 mg L (cid:1) 1 . Finally, fed-batch fermentation of the ﬁ nal strain with glucose or ethanol as carbon source resulted in a resveratrol titer of 415.65 and 531.41 mg L (cid:1) 1 , respectively. & 2015 International Metabolic Engineering Society Published by Elsevier Inc. On behalf of International Metabolic Engineering Society. All rights reserved.


Introduction
Resveratrol (trans-3,5,4′-trihydroxystilbene) is a natural polyphenolic compound from the stilbene family.In plants, resveratrol plays a role as a defense compound against pathogens infection and injury.It occurs naturally in several higher plants, e.g.grapes, peanuts, blueberries, and knotweed (Mei et al., 2015).In preclinical tests resveratrol has shown a wide range of beneficial properties, i.e., antitumor, anti-inflammatory, antidiabetic, antithrombotic, and antiaging properties (Jeandet et al., 2012;Mei et al., 2015).The evidence of resveratrol effects on human health is however so far inconclusive, due to the limited number of clinical trials and small cohort sizes (Poulsen et al., 2013).Nevertheless, resveratrol has attracted much attention from pharmaceutical, food and cosmetic industries.Resveratrol is sold as over-thecounter nutritional supplement; it is included in some cosmetics products, energy drinks, and other products.The demand for resveratrol is expected to further increase in the future.The commercial resveratrol on the market is predominantly extracted from the Japanese knotweed Polygonum cuspidatum (Mei et al., 2015).The preparations range widely in purity; some can contain as low as 50% of the active ingredient.The unpurified knotweed extracts, however, additionally contain emodin which has a laxative effect (Srinivas et al., 2007).Thus, there is market demand for lower cost and high purity resveratrol.
One promising solution is biotechnological production of resveratrol by fermentation of genetically engineered microbes (Borodina and Nielsen, 2014).A number of studies have been published on microbial production of resveratrol.In all the studies, the production exclusively relied on using complex medium or supplementing the minimal medium with resveratrol precursors, p-coumaric acid, tyrosine or phenylalanine, which are expensive for industrial applications.The first study on recombinant resveratrol production described feeding p-coumaric acid to Saccharomyces cerevisiae, which expressed 4-coumaryl-CoA ligase (4CL) and resveratrol synthase (VST1) (Becker et al., 2003).Deamination of tyrosine or phenylalanine, catalyzed by tyrosine/phenylalanine ammonia lyase (TAL/PAL), is the first step of resveratrol biosynthesis pathway in plants.Transcription and enzyme activity of PAL is reduced by its product, trans-cinnamic acid (Blount et al., 2000;Bolwell et al., 1988), which means that efficient conversion of trans-cinnamic acid into p-coumaric acid is required to avoid the accumulation of toxic intermediate.The hydroxylation of transcinnamic acid into p-coumaric acid is catalyzed by cinnamic acid hydroxylase (C4H), a P450 enzyme, which requires cytochrome P450 reductase for efficient electron transfer.Trantas et al. expressed phenylalanine ammonia-lyase (PAL), cinnamate-4-hydroxylase (C4H), and cytochrome P450 reductase (CPR) in addition to 4CL and VST1 to produce resveratrol from phenylalanine (Trantas et al., 2009).In the biosynthesis of p-coumaric acid-derived compounds via phenylalanine, C4H was reported as the rate-limiting step (Trantas et al., 2009;Yan et al., 2005).A reduction of the C4H transcriptional level was observed to result in an 8-fold decrease of PAL and 2.5-fold decrease of 4CL activities in tobacco (Kumar et al., 2012).Alternatively, tyrosine ammonia-lyase (TAL) can be used to directly synthesize p-coumaric acid from tyrosine (Fig. 1), thus averting the rate-limiting hydroxylation step.Wang et al. have successfully expressed codon-optimized TAL from Rhodopseudomonas sphaeroides in yeast to obtain 1.90 mg L À 1 resveratrol, when supplemented with 12 mg L À 1 tyrosine, and 1.06 mg L À 1 resveratrol without supplementation but in complex medium, which contained tyrosine (Wang et al., 2011).In another case, Shin et al. used a tyrosine/phenylalanine ammonia-lyase from Rhodosporidium toruloides to produce 5.8 mg L À 1 resveratrol in complex medium supplemented with 2.17 g L À 1 tyrosine (Shin et al., 2012).The highest resveratrol titer reported in yeast from the literature so far is 391 mg L À 1 , obtained by supplementing complex medium with 2.46 g L À 1 p-coumaric acid and using a genetically engineered industrial Brazilian S. cerevisiae strain that overexpressed 4CL1 and STS (stilbene synthase) genes (Sydor et al., 2010).
In the present study, we aimed to achieve de novo biosynthesis of resveratrol from cheap carbon sources, glucose or ethanol, without supplementation of aromatic precursors.We chose to apply S. cerevisiae as the host, due to its safe use status in pharmaceutical biotechnology and food industry, and due to its high amenability to genetic manipulations.Here we describe step-wise metabolic engineered efforts towards obtaining an efficient cell factory for resveratrol production.

Chorismate
LB and SC drop-out media were made using pre-mixed powders from Sigma-Aldrich.The defined minimal medium used for fermentation in microtiter plates was described as before (Jensen et al., 2014b).

Plasmid and strain construction
The primers used in the study are summarized in Table 3 and the plasmids in Table 4.All the biobricks were assembled into integrative EasyClone vectors using USER cloning (Jensen et al., 2014b).The multiple integrative plasmid pCfB2068 was constructed in a different way: plasmid pCfB322 was amplified by PCR using primers Open_fw and Open_rv.Then biobricks BB293 ( o-At4CL1), BB302 ( o-P PGK1 -P TEF1 -4), and BB295 (VvVST1-4 ) were cloned into the opened pCfB322 vector following USER protocol, resulting in plasmid pCfB2067.Thereafter the plasmid pCfB2067 was PCR-amplified again by the same primers and biobricks BB429 (T o -NAT5 ), BB379 (o -HaTAL), and BB530 (P o -TDH3 ) were inserted in the same way to generate plasmid pCfB2068.
Colony PCR and sequencing was performed to confirm the correct cloning.Yeast transformations were carried out following lithium acetate protocol (Gietz and Woods, 2002).The correct genomic insertions were verified by yeast colony PCR using the primers listed in Table 3.
ST4158 and ST4160 were generated by removal of the selection markers in ST4122 and ST4159 respectively.It was performed by introducing plasmid pSH65, expressing creA gene under control of the P GAL10 promoter (Gueldener et al., 2002).Strains harboring pSH65 were grown in yeast peptone galactose medium for 12-16 h for induction and then plated on yeast peptone dextrose (YPD) agar plates.The colonies were replica-plated on SC drop-out plates to select for the colonies that have lost the markers.The loss of markers was verified by yeast colony PCR.

Determination of gene copy number by qPCR
The design of qPCR primers was conducted using the online PrimerQuest s Tool at https://eu.idtdna.com/Primerquest/Home/Index.All the primers used for qPCR in this study are listed in Supplementary Table 1.
Genomic DNA of ST4140 and ST4152 were extracted using the ZR Fungal/Bacterial DNA MiniPrep™ kit (Zymo Research) following the manufacturer's manual.The SYBR s Green qPCR MasterMix from Life Technologies was used to test for the copy number of HaTAL, At4CL1 and VvVST1.The housekeeping gene ALG9 was used as the reference.Reactions were performed in a 20 μl volume with 10 μl 2 Â SYBR Green QPCR master mix, 1 μl of each upstream and downstream primer, 0.3 μl diluted reference dye and 7.7 μl gDNA containing serially diluted gDNA template (62.5-1000 pg).A notemplate control reaction (NTC) for each gene was made by replacement of gDNA with nuclease-free PCR-grade water.
qPCR runs were performed in Stratagene Mx3005P instrument using the thermocycler program as follows: 10 min of pre-incubation at 95 °C followed by 40 amplification cycles of denaturation at 95 °C for 20 s, annealing and elongation at 60 °C for 22 s; and a single final cycle of 95 °C for 1 min, 55 °C for 30 s and 95 °C for 30 s.

Microtiter plate cultivation of yeast strains
Three biological transformants of each strain were inoculated in 0.5 mL SC drop-out liquid medium without histidine and leucine supplemented with 2% glucose in a 96-deep well microtiter plate with air-penetrable lid (EnzyScree, NL).The cultures were incubated at 30 °C with 250 rpm agitation for 24 h.50 ml of the The front part of 4CL1:: The latter part of 4CL1:: The front part of 4CL2:: The latter part of 4CL2:: seed cultures were inoculated into 0.5 mL minimal medium containing 5 mM tyrosine in a new 96-deep well plate.The minimal medium (pH 6.0) was described in (Jensen et al., 2014b).After 72 h cultivation at 30 °C with 250 rpm agitation, OD 600 was measured with a 20 times dilution in microplate reader BioTek Synergy MX (BioTek).The remaining cultures were mixed with equal volume of absolute ethanol and centrifuged at 2272 g for 30 min.The supernatants were used to analyze resveratrol concentration by HPLC.
2.6.Batch and fed-batch fermentation of yeast strains in controlled reactors Inoculum for bioreactors was prepared as following.Yeast strains from À 80 °C glycerol stock were inoculated into 5 mL of defined minimal medium with 2% glucose as above, and incubated in 13 mL tubes at 30 °C with shaking at 250 rpm for 18 h.The whole culture volume was consequently transferred into 50 mL of fresh defined minimal medium in a 500 ml baffled shake flask and incubated at 30 °C with shaking at 250 rpm for 12 h.The culture OD 600 was measured and reactors were inoculated with such a volume of inoculum that an initial OD 600 of 0.02-0.05was obtained.
Batch and fed-batch fermentations were performed in 1 L-DasGip stirrerpro s bioreactors (DasGip, Julich, Germany) with a starting volume of 0.5 L. Batch medium and medium for batch phase of fed-batch fermentation contained per 1L:30 g (batch fermentation)/40 g (fed-batch fermentation) glucose, 5 g (NH4) 2 SO 4 , 3 g KH 2 PO 4 , 0.5 g MgSO 4 Á 7H 2 O, 2 mL trace metals solution, and 1 mL vitamin solution, where the composition of trace metals and vitamin solutions was the same as in (Jensen et al., 2014b).The medium was adjusted to pH 5.0 before autoclaving in reactors at 121 °C for 20 min.Filter-sterilized glucose, trace metal and vitamins solutions were added to the medium after autoclavation.Fed-batch medium contained 10-fold higher concentration of all the nutrients in the batch phase, except for carbon source which was replaced by 200 g L À 1 glucose or 150 g L À 1 ethanol in the fed-batch medium, but was otherwise prepared in the same way.The feed was started only after residual ethanol produced from the glucose phase was completely depleted.A volumetric growth rate constant-dependent feed strategy (Villadsen et al., 2011) was adopted as described in (Scalcinati et al., 2012).The fermentations were performed at 30 °C; pH was maintained at 5.0 with automatic addition of 2 M KOH or 2 M HCl.The agitation rate was kept at 800 rpm and the air flow was set to 0.5 L per min.The dissolved oxygen concentration was above 30%

Name
Sequence (5′-3′) throughout the cultivation.The DASGIP fedbatch pro s gas analysis system, equipped with gas analyzer 1 GA4 based on zirconium dioxide and two-beam infrared sensor (DASGIP), was employed for monitoring oxygen and carbon dioxide concentration.Samples were taken at regular intervals to measure OD 600 .Supernatants were stored at À 20 °C until HPLC analysis for organic acids, glycerol, ethanol, and residual glucose.Another portion of sample was mixed with an equal volume of absolute ethanol and centrifuged at 12,000 rpm for 2 min.The supernatant was stored at À20 °C until HPLC analysis for resveratrol and p-coumaric acid.

Analytical methods
The OD 600 was measured on a Genesys 20 Spectrophotometer (Thermo Scientific).The HPLC quantification of glucose, glycerol, ethanol, succinate, and pyruvate was performed as before (Ostergaard et al., 2000).Resveratrol and p-coumaric acid were quantified on HPLC (Thermo) equipped with a Discovery HS F5 150 mm Â 2.1 mm column (particle size 3 mm).The eluent flow rate was 1.5 mL min À 1 .Linear gradient from 5% to 60% of solvent A over 0.5-9.5 min was used.Solvent A was 10 mM ammonium formate (pH 3.0, adjusted by formic acid).Solvent B was acetonitrile.Resveratrol was detected by absorbance at 304 nm with a retention time of 6.4 min and p-coumaric acid at 277 nm of 4.7 min.Resveratrol and p-coumaric acids concentrations were calculated from the standard curves, and both resveratrol and p-coumaric acid standards were purchased from Sigma-Aldrich.

Reconstruction of resveratrol biosynthetic pathway from tyrosine precursor
Our previous study showed that tyrosine ammonia-lyases from H. aurantiacus and Flavobacterium johnsoniae have high activity in yeast (Jendresen et al., 2015).By over-expressing HaTAL (ST4121) and FjTAL (ST4128) in S. cerevisiae, we obtained respectively 7.04 72.03 and 13.71 72.14 mg L À 1 of p-coumaric acid in minimal medium supplemented with 5 mM tyrosine (Fig. 2A and B).Two versions of 4-coumarate:CoA ligase from A. thaliana (At4CL1 and At4CL2) and stilbene synthase from V. vinifera (VvVST1) were then introduced under control of strong constitutive promoters in order to convert p-coumaric acid into resveratrol.Contrarily to p-coumaric acid results, the highest titers were obtained in the strains carrying HaTAL.At4CL1 consistently resulted in slightly higher resveratrol titers than At4CL2.The highest titer of 11.66 7 0.57 mg L À 1 resveratrol was obtained in ST4122, over-expressing HaTAL and VvVST1 under P TEF1 promoters and At4CL1 under P PGK1 promoter.When the direction of the double promoter was reversed, so that At4CL1 was under control of the P TEF1 promoter and VvVST1 under control of P PGK1 promoter, resveratrol titer dropped to 4.487 0.22 to mg L À 1 (ST4123).This result underlines the importance of balancing gene expression in the pathway.To reduce loss of pathway intermediates and improve turnover rates due to higher local substrate concentration, fusing related enzymes is a promising strategy (Li and Borodina, 2015).Fusion of two enzymes, At4CL1 and VvVST1, was previously reported to improve resveratrol production due to metabolic channeling  ( Zhang et al., 2006).We also attempted expressing the fusion of the two proteins, linked by Gly-Ser-Gly linker, in ST4124.This however led to a lower resveratrol titer of 4.69 70.39 mg L À 1 so we did not pursue this strategy further.

De novo biosynthesis of resveratrol from glucose
It was observed that p-coumaric acid production continued in batch cultures after the supplemented tyrosine had been consumed, thus indicating that at least a fraction of p-coumaric acid was produced de novo from glucose or ethanol (Jendresen et al., 2015).To investigate if resveratrol also can be produced directly from glucose and also to test the effect of precursor improvement, we cultivated the strain ST4135 (analogous to ST4122) with the resveratrol pathway (HaTAL, At4CL1 and VvVST1) in a minimal medium with glucose as the only carbon source.Fermentation of ST4135 resulted in 2.73 70.05 mg L À 1 resveratrol from 30 g L À 1 glucose (Fig. 3A).Interestingly, resveratrol was primarily produced during the ethanol consumption phase, which indicates that during growth on glucose the fluxes towards resveratrol precursors, tyrosine and malonyl-CoA, were low.
The aromatic amino acids are the least abundant amino acids in the S. cerevisiae cell, with intracellular concentrations of tyrosine and phenylalanine being as low as 0.5 and 0.6 mM, respectively (Braus, 1991).The biosynthesis of aromatic amino acid is strongly regulated, both at the transcriptional and posttranscriptional levels.In order to improve the flux towards tyrosine, we overexpressed feedback-inhibition resistant versions of DAHP synthase ScAro4p K229L and chorismate mutase ScAro7p G141S .This resulted in a 78%-improvement of resveratrol titer in batch cultures compared with that of the strain ST4135 harboring only the resveratrol pathway, i.e. 4.85 70.31 mg L À 1 resveratrol was obtained (Fig. 3B).
In order to further improve the flux towards acetyl-CoA, we also attempted over-expression of acetyl-CoA synthase (SeACS L641P ) from Salmonella enterica and aldehyde dehydrogenase ScAld6, in various combinations with ScAro4p K229L and ScAro7p G141S or ScAcc1p S659A, S1157A .Surprisingly no improvement was obtained (Supplementary Fig. S1).Further investigation is needed to determine the reason for the negative effect of this modification (Fig. 4)

Increasing resveratrol production via integration of multiple copies of resveratrol pathway
We hypothesized that the low activity of the resveratrol biosynthetic pathway may be limiting resveratrol biosynthesis.We therefore introduced multiple copies of the genes HaTAL, At4CL1 and VvVST1 into a strain, over-expressing ScARO4 K229L , ScARO7 G141S , and ScACC1 S659A, S1157A .The multicopy integration was achieved by using an integrative vector that targets Ty4 retrotransposons (Maury et al., submitted for publication).As the transformants were expected to integrate different copy numbers of the expression vector, we randomly screened 8 transformants to select the best producer, named ST4152.The copy number of the genes in the strain ST4152 was identified to be 8.4370.85 for HaTAL, 8.6471.60 for At4CL1, and 11.1671.23 for VvVST1 respectively by qPCR.
The strain ST4152 had a longer lag phase and lower biomass accumulation than the strain ST4140 in batch fermentation.However, the resveratrol titer in both the glucose and ethanol phases was greatly improved (Table 5).The final titer was 235.5777.00 mg L À 1 , 36-fold higher than in the parent strain ST4140.

Resveratrol production in fed-batch fermentation
As resveratrol primarily accumulated after glucose depletion in batch cultivations, we carried out fed-batch cultivations with feeding of glucose or ethanol.The batch phase was on 40 g L À 1 glucose, once all carbon sources were consumed, the carbon-limited feeding of glucose or ethanol was initiated.The strain ST4152 produced around 200 mg L À 1 in the batch phase.In the glucose feeding phase, 222.08 mg L À 1 additional resveratrol was accumulated on the feeding of 15.93 g L À 1 glucose (Fig. 5) and resulted in a final titer of 415.65 mg L À 1 resveratrol.In the ethanol feeding phase, 333.57mg L À 1 resveratrol was additionally produced from 16.63 g L À 1 ethanol resulting in a final titer of 531.41 mg L À 1 resveratrol.Small amounts of acetate and glycerol and no other by-products, such as p-coumaric acid or succinate were detected during the fed-batch process.Multiple integrative plasmid p2068 carrying HaTAL, At4CL1, and VvVST1 was integrated onto Ty4 elements of ScARO4 K229L , ScARO7 G141S and ScACC1 S659A, S1157A overexpressing strain.The strain was grown on defined minimal medium with 30 g L À 1 glucose in batch bioreactors.The displayed average values 7standard deviations were calculated from two biological replicates.

Discussion
In the previous studies resveratrol production was achieved by supplementing p-coumaric acid (Becker et al., 2003;Beekwilder et al., 2006;Shin et al., 2011;Sydor et al., 2010;Wang and Yu, 2012), which is an expensive precursor.Besides, tyrosine or phenylalanine were also used as the substrates for resveratrol production (Trantas et al., 2009;Zhang et al., 2006).Also several previous studies described resveratrol production without supplementation of precursor, but in SC-dropout (Wang et al., 2011) or Yeast extract-Peptone-Galactose (Shin et al., 2012) media, which contain tyrosine and phenylalanine.This study, for the first time, demonstrates de novo microbial production of resveratrol from cheap carbon source (glucose or ethanol) in a minimal medium.We chose TAL instead of PAL for the first step of resveratrol biosynthesis pathway (deamination from tyrosine or phenylalanine), thus by-passing the P450-dependent step.Although previous studies failed in producing resveratrol via the TAL-dependent pathway in yeast (Zhang et al., 2006), our study shows that it is possible to construct a high-level resveratrol producing yeast based on the TAL route.The resveratrol production of 530 mg L À 1 obtained in this study, to our knowledge, is the highest level reported in yeast.
This study adopted a push and pull strategy by increasing precursor supply and multiple integration of resveratrol pathway to improve resveratrol production.Aromatic amino acids were shown to have feedback inhibition on the shikimate pathway (Koopman et al., 2012;Luttik et al., 2008).Deregulation of feedback inhibition is a general way to improve production of aromatic amino acids derivatives.This has been corroborated by overexpression of aromatic amino acids-insensitive ARO4 and ARO7 alleles in several studies.Koopman et al. overexpressed feedbackinsensitive allele of ARO4 (ARO4 G226S ), which led to 2-fold increase of naringenin in S. cerevisiae (Koopman et al., 2012).Another study by Curran et al. improved muconic acid production by 50%, when overexpressing tyrosine insensitive ARO4 allele (ARO4 K229L ) in S. cerevisiae (Curran et al., 2013).By combining overexpression of the feedback insensitive alleles of ARO4 and ARO7 (ARO4 K229L and ARO7 G141S ), Luttik et al. obtained 200-fold increase of the extracellular aromatic amino acids concentration (Luttik et al., 2008).Thus, the 78% increase of resveratrol we observed upon overexpression of ARO4 K229L and ARO7 G141S is consistent with these studies.Supply of the other precursor, malonyl-CoA, catalyzed by Acc1p was a well-known limiting step in the fatty acids biosynthesis pathway due to posttranslational regulation by Snf1 protein kinase (Tehlivets et al., 2007).Shi et al. found that the phosphorylation of Acc1p by Snf1 could be diminished by sitedirected mutagenesis at Ser659 and Ser1157 positions (Shi et al., 2014).Overexpression of ACC1 S659A,S1157A in engineered S. cerevisiae resulted in 3-fold increase of fatty acid ethyl esters production and more than 3.5-fold improvement of 3-hydroxypropionic acid titer (Shi et al., 2014).The 31% improvement was obtained in this study, when we overexpressed ACC1 S659A,S1157A in the engineered strain.However, it was surprising that further overexpressing ALD6 and SeACS L641P , which was previously shown to enhance acetyl-CoA flux (Shiba et al., 2007), decreased resveratrol production.We suppose that in the strain with a single copy of resveratrol biosynthesis pathway the flux control resided mainly with the downstream of the pathway and therefore the push strategies had only a limited effect.In case of ALD6 and SeACS L641P overexpression, the negative result was possibly due to the effect of acetyl-CoA overproduction, which exhibits genome-wide regulatory effects (Shi and Tu, 2013;Zhang et al., 2013) and may also divert to other products than resveratrol especially in glucose consuming phase.In future studies it would be interesting to  investigate the effect of increased acetyl-CoA supply in a strain carrying multiple copies of resveratrol biosynthesis.High level-expression of the flux controlling enzymes in the pathway is a common solution to a low enzymatic activity problem.We chose to integrate multiple copies of the resveratrol biosynthetic genes into yeast chromosomes as a pull strategy.In comparison to the high-copy number episomal vectors, e.g., 2μ-based vectors, integration results in better strain stability.A 36-fold improvement in titer was achieved in this study when three genes, HaTAL, At4CL1 and VvVST1, were integrated in multiple copies.This finding highlights that indeed the resveratrol flux was limited by the activity of the biosynthetic enzymes.At the same time, the strain fitness decreased, manifested by a longer lag phase, lower maximum specific growth rate and lower biomass yield on glucose.This could be expected as multiple integration of several genes likely resulted in high metabolic burden for the cell.A solution to the problem could be replacement of constitutive promoters with regulated ones for control of the resveratrol biosynthesis pathway, e.g., one could use a glucose-repressed P HXT7 promoter, so the resveratrol biosynthesis is first turned on in the fed-batch phase after sufficient biomass has been accumulated in the batch phase.Moreover, the observation that resveratrol primarily accumulated in the ethanol phase of batch fermentation inspired us to use ethanol feed for fed-batch fermentation.An ethanol pulse feed strategy was also successfully applied earlier for amorphadiene production (Westfall et al., 2012).
Developing a high-producing microbial strain for industrial production of resveratrol is a serious metabolic engineering challenge.In our study, we implemented the push strategies first, followed by the pull strategy.Retrospectively, it would be better to do it the other way around.Future development of the cell factory may include optimization of activities of the resveratrol biosynthetic genes (via enzyme engineering, protein scaffolding or expression level balancing), further improvement of precursor supply, improvement of reducing co-factor (NADPH) supply, eliminating product degradation (Koopman et al., 2012), decoupling of growth and production phases, and other rational designs.Additionally genome-scale modeling or -omic data can be used to predict non-intuitive metabolic engineering strategies (Curran et al., 2013;Gold et al., 2015).

Conclusions
There is an increasing demand for resveratrol in the pharmaceutical, food and cosmetic industries.Production of resveratrol from a cheap carbon source by microbial fermentation is attractive due to short process time, reduction of production costs, reliable supply and high purity, when compared to the extraction from plant materials.In this study we constructed the resveratrol pathway via tyrosine intermediate in S. cerevisiae and for the first time demonstrated the possibility of de novo resveratrol biosynthesis from glucose.By step-wise metabolic engineering, targeted towards improvement of precursor supply and increased expression of the biosynthetic genes, we obtained an optimized strain, which produced around 0.5 g L À 1 resveratrol in fed-batch fermentation on a minimal medium.This strain represents a good basis for development of microbial resveratrol production process.

Fig. 2 .
Fig. 2. Microbial production of resveratrol from tyrosine.Strains carrying HaTAL from H. aurantiacus (A) and FjTAL from F. johnsoniae (B) together with different combinations of At4CL1 and VvVST1 were grown in minimal medium with 20 g L À 1 glucose and 5 mM tyrosine and the cultures were sampled at 72 h of cultivation.Strains ST4124, ST4127, ST4131, and ST4134 over-express fusions of At4CL1 and VvVST genes.The displayed average values 7 standard deviations were calculated from three biological replicates.

Fig. 4 .
Fig. 4. Optimization of resveratrol pathway by multiple integrations of genes.Multiple integrative plasmid p2068 carrying HaTAL, At4CL1, and VvVST1 was integrated onto Ty4 elements of ScARO4 K229L , ScARO7 G141S and ScACC1 S659A, S1157A overexpressing strain.The strain was grown on defined minimal medium with 30 g L À 1 glucose in batch bioreactors.The displayed average values 7standard deviations were calculated from two biological replicates.

Fig. 5 .
Fig.5.Fed-batch fermentation of the optimized resveratrol-producing strain ST4152.Aerobic fed-batch fermentations were carried out by feeding glucose (A) or ethanol (B) respectively following carbon source-limited feeding strategy.The cultivations were performed in duplicates (Supplementary Fig.S2); here representative graphs are shown.

Table 1
List of yeast strains used in the study.

Table 2
List of biobricks used in the study.

Table 3
List of primers used in the study.

Table 4
List of plasmids used in the study.

Table 5
Comparison of resveratrol titers and kinetic parameters obtained from different resveratrol producing strains in batch cultivations.