Metabolic engineering of yeast for fermentative production of flavonoids

1 exceed the previously reported titers in yeast. This is also the ﬁrst report of de novo biosynthesis of resokaempferol and ﬁsetin in yeast. The work demonstrates the potential of ﬂavonoid-producing yeast cell factories. (cid:1) 2017 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Flavonoids are aromatic secondary metabolites naturally synthesized by plants and fungi from aromatic amino acids L-phenylalanine and L-tyrosine.The structure of these phytochemicals is characterized by two phenolic rings and one heterocyclic ring.The main differences between flavonoids are related to the hydroxylation patterns, the position of the second aromatic ring and the saturation of the heterocyclic ring (Grotewold, 2006).Bioactive polyphenols such as flavonoids play a fundamental role in the physiology of plants; their natural functions include UV protection, reduction of oxidative damage in cells, and antibacterial effect (Cushnie and Lamb 2011).
Research on human cells showed positive properties of polyphenols in age-related diseases, cancer and cardiovascular diseases (Woelfle et al., 2010;Bulzomi et al., 2012).Moreover, some flavonoids, e.g., naringenin, have neuroprotective and antioxidant properties.Liquiritigenin has been reported as a protective agent against oxidative stress in osteoblasts, and some liquiritigenin derivatives show antitumor and antidiabetic activity (Choi, 2012;Wedick et al., 2012;Liu et al., 2012;Raza et al., 2013;Hämäläinen et al., 2007).Kaempferol, quercetin and fisetin showed anti-cancer, cardio-protective and anti-inflammatory effects (Chou et al., 2013;Chen and Chen, 2013).Experiments with rats showed that doses of 50 mg kg À1 of quercetin inhibit the migration of melanoma cells.Other experiments with humans showed that doses of 150 mg day À1 of quercetin had positive effects on the health of people with high risk of cardiovascular disease (Caltagirone et al., 2000;Mukhtar et al., 2015).Fisetin has neuroprotective, neurotrophic and anti-amyloid properties, which makes it a promising therapeutic agent for neurodegenerative disorders such as Huntington and Alzheimer diseases (Maher et al., 2011;Currais et al., 2014).
The main obstacles for large-scale production of flavonoids from plant material are the long culture periods, requirements for specific cultivation conditions and low abundance.Additionally, extraction and purification processes add cost and result in product loss and degradation (Wang et al., 2011;Routray and Orsat, 2012).The concentration of flavonoids in different varieties and sources of fruits oscillate between 30 and 4000 mg kg À1 of dry weight (Hertog et al., 1992;Paganga et al., 1999;Miean and Mohamed, 2001;Crozier et al., 1997).This means that for the production of 1 kg of flavonoids, it is required to process 0.25-33 Tons of dry weight of fruits or vegetables.
Methods for phytochemicals extraction from agro-industrial waste have been developed in the recent years, and secondary metabolites such as carotenoids and flavonoids have been extracted from this feedstock (Matharu et al., 2016;Pfaltzgraff et al., 2013;Makris et al., 2007).Extraction of polyphenols such as flavonoids from waste is promising, however there are challenges that have to be overcome in terms of cost and optimization of extraction and purification methods (Mirabella et al., 2014;Luthria, 2012).
For a sustainable supply of phytochemicals, it can be an advantage to engage genetically engineered microbial cell factories, e.g., E. coli or S. cerevisiae.Koopman et al. (2012) reported production of 108.90 mg L À1 of naringenin from glucose using an engineered yeast strain.Furthermore, feeding naringenin to engineered cells allowed production of other flavonoids, such as genistein, kaempferol and quercetin (Trantas et al., 2009).Production of naringenin in E. coli reached 29 mg L À1 using glucose as carbon source, while kaempferol and quercetin have been produced using p-coumaric acid as precursor and fisetin using L-tyrosine as a precursor (Santos et al., 2011;Leonard et al., 2006;Stahlhut et al., 2015).
This study reports the development of a novel yeast cell factory for de novo production of flavonoids from glucose.The target flavonoids included: naringenin, liquiritigenin, kaempferol, resokaempferol, quercetin, and fisetin.Intermediate compounds and by-products were quantified to characterize regulated steps in the pathways and to identify targets for future improvements of the flavonoid-producing yeast cell factories.

Plasmid construction
The plasmids with biosynthetic pathways for flavonoids production were constructed using EasyClone 2.0 integrative plasmids with auxotrophic and dominant selection markers (Stovicek et al., 2015).The vectors integrate into well-defined integration sites previously described by (Mikkelsen et al. 2012).
DNA fragments (BioBricks), encoding genes or promoters, were amplified by PCR using primers and templates as described in Table 1.Individual BioBricks were assembled into the integrative plasmids by USER cloning as described previously (Stovicek et al., 2015).Finally, the resulting vectors (Table 2) were verified by sequencing.

Strain construction
For expression of flavonoid biosynthetic pathways, in this research were employed three S. cerevisiae strains, engineered to produce p-coumaric acid (Table 3).The ST4757 strain was engineered to produce p-coumaric acid using the phenylalanine ammonia-lyase (PAL) pathway and the strains ST4069 and ST2645 produced p-coumaric acid via the tyrosine ammonialyase (TAL) pathway.To obtain flavonoid production, the strains were transformed with NotI-linearized integrative vectors carrying pathway genes via lithium acetate protocol (Gietz and Woods, 2002).The transformants were selected on synthetic drop-out medium (Sigma-Aldrich), selecting for URA3, HIS5 and LEU2 markers.For selection on acetamide, the media contained 0.17 % yeast nitrogen base without amino acids and ammonium sulfate and 6.6 g L À1 of potassium sulfate and 0.6 g L À1 acetamide.For selection of dominant markers NatMX, KanMX, BleMX or HphMX, the medium was supplemented with 100 mg L À1 nourseothricin, 200 mg L À1 G418 disulfate salt, 10 mg L À1 phleomycin or 200 mg L À1 hygromycin B respectively.Correct integration of the vectors was verified by yeast colony PCR.The yeast strains constructed in this study are listed in Table 3.

Media and cultivations
For selection of yeast transformants and for routine cultivations, synthetic complete (SC) medium as well as drop-out media (SC-Ura, SC-Leu, SC-His) and agar plates were prepared using premixed drop-out powders from Sigma-Aldrich.Synthetic fed-batch medium for S. cerevisiae M-Sc.syn-1000 (FIT) was purchased from M2P Labs GmbH (Germany).The medium was supplemented with vitamins solution (final 1% v/v) and the enzyme mix (final concentration 0.5% v/v) immediately before use.
For the quantification of production of flavonoids for each strain, five single colonies originating from independent transformants were tested.The colonies were inoculated in 0.5 ml dropout SC liquid medium without uracil, histidine, and/or leucine in 96-deep well microtiter plates with air-penetrable lid (EnzyScreen, The Netherlands).The plates were incubated at 30 °C with 250 rpm agitation at 5 cm orbit cast overnight.50 ml of the overnight cultures were used to inoculate 0.5 ml synthetic fed-batch medium in a 96-deep well plate.Fermentation was carried out for 72 h at the same conditions as before.At the end of cultivation, samples for metabolite analysis and optical density (OD) were taken.OD 600 was estimated as follows: 10 ml of fermentation broth was mixed with 190 ml water in a 96 well microtiter plate and absorbance was measured at 600 nm wavelength in microplate reader BioTek Synergy MX (BioTek).200 ml of the culture were mixed with 200 ml of absolute ethanol, the mixture was centrifuged at 2272g Table 1 Primers used in this study and list of Biobricks generated by PCR amplification, description of the templates can be found in Stahlhut et al. (2015).

PRIMERS
for 15 min and 250 ml of supernatant were analyzed for flavonoids concentration by HPLC.

Identification of fisetin
For fisetin identification, the samples were dried at room temperature under reduced pressure using a Centrifugal Vacuum Con-centrator (Savant Speed Vacs Concentrator, Thermofisher Scientific, Waltham Ma), followed by reconstitution using a 0.1% solution of formic acid in LC-MS grade acetonitrile.20 ml of each sample were then injected and analyzed on a Dionex UltiMate 3000 UHPLC (Thermo Fisher Scientific, San Jose, CA) connected to an Orbitrap Fusion Mass Spectrometer (Thermo Fisher Scientific, San Jose, CA).The UHPLC used a Supelco Discovery HS F5-3, 15 cm Â 2.1 mm, 3 lm column.Temperature was 30 °C and flow rate was 0.6 mL/min with a mobile phase of 95% 0.1 % formic acid (mobile phase A) and 5% acetonitrile with 0.1% formic acid (mobile phase B) for 2 min followed by a linear gradient to 95% mobile phase B over 12 min.This gradient was held for 2 min after which it was changed immediately to 95% mobile phase A and 5% mobile phase B and held for 6 min.The sample was passed on to the MS equipped with a heated electrospray ionization source (HESI) in a positive-ion mode with nitrogen as nebulizer gas (52 a.u.).The cone and probe temperatures were 356 °C and 420 °C, respectively.Probe gas flow was 16 a.u. and spray voltage was 3500 V. Scan range was 150-1000 Da and time between scans was 100 ms.

Flavonoids pathway design
Previously, Stahlhut et al. reported reconstruction of a biosynthetic pathway for the production of fisetin from L-tyrosine in E. coli (Fig. 1) (Stahlhut et al., 2015).This study aimed to develop a fermentation process for de novo production of a portfolio of related flavonoids from a cheap carbon source and using yeast S. cerevisiae as the host.The biosynthetic pathway towards flavonoids is depicted in Fig. 1.The precursor, p-coumaric acid, was synthesized either from L-tyrosine via tyrosine ammonia-lyase (TAL) or from L-phenylalanine by the use of phenylalanine ammonialyase (PAL).First, p-coumaric acid must be activated by a 4coumaroyl-CoA ligase 4CL, and for this a variant from P. crispum was used.Next, chalcone synthase CHS from P. hybrida and chalcone isomerase CHI from M. sativa were used to convert the resulting p-coumaroyl-CoA and three molecules of malonyl-CoA into naringenin.To obtain isoliquiritigenin, chalcone reductase CHR from A. mongholicus was additionally overexpressed.The resulting naringenin and isoliquiritigenin producing strains were further engineered towards production of respectively kaempferol and resokaempferol by overexpression of flavanone 3-hydroxylase  Lastly, two different cytochrome P450 flavonoid monooxygenases FMOs, one from F. ananassa and another from P. hybrida, were overexpressed in the kaempferol-producing strain to obtain quercetin and in the resokaempferol-producing strain to obtain fisetin.Each of the tested FMOs was fused in-frame with cytochrome P450 reductase CPR from C. roseus.

Biosynthesis of naringenin and liquiritigenin
To begin with, two platform strains, producing low or high amounts of p-coumaric acid, were explored for their capacity to make naringenin and liquiritigenin.Strain ST4069, which produced 0.24 ± 0.03 g L À1 of p-coumaric acid in the previous study, was employed as the low-producer, while strain ST2645, which reached a titer of 1.93 ± 0.26 g L À1 , was used as the high-producer (Rodriguez et al., 2015).Strain ST4069 has overexpression of a TAL gene, while strain ST2645 has additional modifications increasing the flux towards aromatic amino acids: knock-outs of ARO10 and PDC5 genes and over-expressions of ARO4 fbr and ARO7 fbr .To obtain naringenin production, 4-coumaroyl-CoA ligase 4CL, chalcone synthase CHS and chalcone isomerase CHI were overexpressed.For liquiritigenin production, chalcone reductase CHR was additionally overexpressed.The titers of naringenin and liquiritigenin were ca.3-fold higher in the strain background with optimized p-coumaric acid production, reaching 1.55 ± 0.13 mg L À1 of naringenin and 5.31 ± 0.48 mg L À1 of liquiritigenin.p-Coumaric acid accumulated in the medium, particularly in the optimized strains, where up to 8 mg L À1 of p-coumaric acid were measured.Naringenin ($1 mg L À1 ) also accumulated alongside liquiritigenin, indicating incomplete reduction by CHR (Fig. 2A-D).

Biosynthesis of kaempferol and resokaempferol
The naringenin-(strain ST5068) and liquiritigenin-(strain ST5069) producing strains were further engineered for production of respectively kaempferol and resokaempferol by overexpressing the genes encoding flavanone 3-hydroxylase F3H and flavonol synthase FLS.As a result, two new strains were obtained, strain ST5070 for the production of kaempferol and strain ST5071 for the production of resokaempferol.Strain ST5070 produced 26.57± 2.66 mg L À1 of kaempferol as well as two pathway intermediates: 10.75 ± 0.87 mg L À1 of p-coumaric acid and 0.83 ± 0.05 of naringenin (Fig. 2E).The high titer of kaempferol was surprising, considering that the parent strain produced less than 2 mg L À1 of kaempferol precursor, naringenin.The depletion of naringenin and higher capacity to produce kaempferol imply that the conversion steps of naringenin into kaempferol are very efficient.The resokaempferol producing strain (ST5071) accumulated 0.51 ± 0.03 mg L À1 of resokaempferol, 14.54 ± 1.96 mg L À1 of p-coumaric acid, 11.39 ± 0.33 mg L À1 of kaempferol, and less than 1 mg L À1 of naringenin and liquiritigenin (Fig. 2F).In the resokaempferol strain, by-product kaempferol was produced, which results from incomplete reduction of p-coumaroyl-CoA by CHR.

Biosynthesis of fisetin and quercetin
For the production of quercetin and fisetin, it is necessary to overexpress cytochrome P450 flavonoid monooxygenase (FMO) and cytochrome P450 reductase (CPR).In this study, two FMO variants were evaluated, one from F. ananassa and another from P. hybrida.Each FMO was fused in-frame to the CPR from C. roseus by a flexible glycine-serine linker (5 0 -GGGTCGAC-3 0 ).
The genes were overexpressed in the kaempferol producing strain (ST5068) to obtain 16.04 ± 0.37 mg L À1 and 20.38 ± 2.57 mg L À1 of quercetin in the strains with FMO from F. ananassa and P. hybrida respectively (Fig. 2G and I).For fisetin biosynthesis, the host strain was the strain engineered for resokaempferol production (ST5069).Upon overexpression of FMO-CPR fusions, 1.20 ± 0.19 mg L À1 and 1.65 ± 0.10 mg L À1 of fisetin was obtained respectively (Fig. 2H and J).The strains overexpressing the FMO of F. ananassa additionally accumulated kaempferol and naringenin, while the strains with FMO from P. hybrida did not secrete these intermediates, hence the latter strain (ST5074) was chosen for further work.Fisetin was detected when expressing both the FMO from F. ananassa and P. hybrida.The observed ion 287 in positive mode is congruent with the ions recognized in the fisetin standard.The spectra for the infusion of the fisetin standard gave an ion with m/z 287.0549, which corresponds to 0.35 ppm of ion with the ionic formula of C15H11O6 in positive ion mode;  this is indicative of a compound with a molecular formula of C15H10O6.

Improving fisetin production via heterologous pathway optimization
The strains designed to produce fisetin, accumulated significant amounts of quercetin.It was hypothesized that more flux could be re-directed into fisetin by exploiting the metabolic channeling effect, which had proven effective in other studies (Albertsen et al., 2011;Stahlhut et al., 2015).A fusion of CHS and CHR proteins was overexpressed in the fisetin-producing strain ST5074, resulting in ST5401.A control strain, ST5402, expressing an additional copy of non-fused CHS and CHR was constructed as well.Both strains produced higher titer of fisetin than the parental strain: strain ST5401 produced 2.11 ± 0.26 mg L À1 and strain ST5402 produced 2.29 ± 0.07 mg L À1 in comparison to 1.65 ± 0.01 mg L À1 for the parental strain ST5074 (Fig. 3A and B).As the competing byproduct, quercetin, accumulated to the same concentration as in the strain without additional copies of CHS and CHR, the ratio of the product (fisetin) to by-product (quercetin) was slightly improved.However, the fusion clearly did not provide an additional advantage in comparison to overexpression of separate enzymes, and further work is needed to optimize the linkage between the two proteins.
The strains engineered for liquiritigenin, resokaempferol and fisetin production accumulated significant amounts of byproducts.Particularly obtaining the products in the reduced branch was difficult and was always accompanied by production of nonreduced analogues.By overexpressing an extra copy of chalcone synthase CHS and chalcone reductase CHR, fisetin titer was increased from 1.65 ± 0.10 mg L À1 to 2.29 ± 0.07 mg L À1 .Clearly, further improvement of chalcone reductase is needed to obtain the reduced flavonoids.
When assembling novel heterologous pathways for production of the complex metabolites, it is important to engineer (or isolate) a microbial host with a sufficient pool of precursor metabolites for a given product and to ensure balanced enzymatic activities of the multistep metabolic pathways in order to avoid accumulation of intermediates and by-products (Venugopalan et al., 2016;Badle et al., 2014).Therefore, several engineered platform strains with different production capacities for production of flavonoids precursor p-coumaric acid were tested.In all cases, when the strains with improved flux towards p-coumaric acid via TAL route were implemented, an improvement in production of the downstream flavonoids was observed.
The accumulation of p-coumaric acid was common for all the engineered strains.The accumulation of this precursor indicates a limiting step for the conversion of p-coumaric acid into pcoumaroyl-CoA.Previous research reported the same limiting step in E. coli and yeast; they associated the accumulation of p-coumaric acid to the lack of malonyl-CoA or low activity of the 4CL for conversion of p-coumaric acid into p-coumaroyl-CoA (Stahlhut et al., 2015;Koopman et al., 2012).To overcome this limitation, other natural or engineered variants of 4CL can be tested and the supply of malonyl-CoA can be improved (Shi et al., 2014).Another possibility for improvements could be testing several different S. cerevisiae strain backgrounds, as miniature genetic and phenotypic differences might result in significant differences of the final yields of phytochemicals and biofuels (Strucko et al., 2015;Pereira et al., 2014).In higher plants, flavonoid biosynthetic enzymes are co-localized and assembled in protein complexes and the metabolic channeling effect decreases formation of by-products.In analogy, compartmentalizing or scaffolding the flavonoid biosynthetic enzymes in yeast may have a positive effect on the enzymatic activity and therefore on flavonoids production (Jeon et al., 2015;Avalos et al., 2013;Woolston et al., 2013;Dueber et al., 2009).Further studies are needed on kinetic regulation of the flux through the flavonoids pathway, on the degradation and secretion of flavonoids.

Biosynthesis of flavonoid via phenylalanine
Several studies have previously demonstrated that p-coumaric acid, the first intermediate of de novo flavonoid pathway, can be synthesized in S. cerevisiae via two routes: through tyrosine Strain ST4972 engineered for biosynthesis of fisetin using the PAL pathway.The background strain for strains ST5401 and ST5402 was strain ST5074.The strains were cultivated in FIT medium for 72 h, the mean value of extracellular concentration of compounds was calculated from five biological replicates.ammonia-lyase TAL or phenylalanine ammonia-lyase PAL (Koopman et al., 2012;Li et al., 2015).In this study, the capacity of TAL and PAL routes to supply metabolic precursors for flavonoid production was compared.For this, a strain producing p-coumaric acid via PAL route was constructed (ST4757).The strain featured overexpression of CYB5 from S. cerevisiae and PAL2, ATR2 and C4H and from Arabidopsis thaliana.The pathway for biosynthesis of fisetin was introduced into this strain as the next step.The pathway was analogous to the one described above in ST5074.Interestingly, HPLC analysis of the resulting final PAL-based strain (ST4972) revealed a different metabolic profile (Fig. 3C) as compared to the corresponding TAL-based strain.More specifically, a 5.7-fold higher accumulation of p-coumaric acid (70 mg L À1 ) was observed, while fisetin was only found in trace amounts and could not be quantified.
Surprisingly, when PAL route was applied for the production of fisetin, a much higher accumulation of p-coumaric acid was observed (69.07 ± 17.81 mg L À1 instead of 12.63 ± 0.76 mg L À1 ), the production of quercetin was lower and there was no production of fisetin.Another surprising discovery was the relatively high titer of kaempferol (26.57± 2.66 mg L À1 ), despite the parental strain produced only 1.55 ± 0.13 mg L À1 of kaempferol precursor, naringenin.This phenomenon could be explained by intracellular degradation of naringenin, its instability in the medium or lower secretion of kaempferol (Mora-Pale et al., 2013;Du et al., 2010).

Conclusions
This work presents a proof-of-concept biosynthesis of flavonoids in engineered yeast cell factories.An array of yeast strains expressing heterologous flavonoid metabolic pathways was successfully developed.The strains contained up to eight genes, where some of the strains produced significant flavonoid titers in doubledigit mg L À1 range.Moreover, the study demonstrated that flavonoids can be synthesized via two different precursor-supplying routes, via tyrosine ammonia-lyase or phenylalanine ammonialyase.The compounds quercetin and kaempferol were the ones with the highest extracellular concentrations detected in this study.This study is the first report on the biosynthesis of resokaempferol and fisetin in yeast.

Fig. 2 .
Fig. 2. Strains engineered for biosynthesis of flavonoids via TAL pathway.A. Naringenin pathway expressed in low-level p-CA strain ST4069.B. Liquiritigenin pathway expressed in low-level p-CA strain ST4069.C. Naringenin pathway expressed in high-level p-CA strain ST2645.D. Liquiritigenin pathway expressed in high-level p-CA strain ST2645.E. Kaempferol pathway expressed in high-level p-CA strain ST2645.F. Resokaempferol pathway expressed in high-level p-CA strain ST2645.G. Quercetin pathway expressed in high-level p-CA strain ST2645; flavonoid 3 0 -monooxygenase (FMO) from F. ananassa.H. Fisetin pathway expressed in high-level p-CA strain ST2645; flavonoid 3 0monooxygenase (FMO) from F. ananassa.I. Quercetin pathway expressed in high-level p-CA strain ST2645; flavonoid 3 0 -monooxygenase (FMO) from P. hybrida.J. Fisetin pathway expressed in high-level p-CA strain ST2645; flavonoid 3 0 -monooxygenase (FMO) from P. hybrida.The strains were cultivated in FIT medium for 72 h, the mean value of extracellular concentration of compounds was calculated from five biological replicates.

Fig. 3 .
Fig. 3. Strains engineered for the optimized biosynthesis of fisetin. A. ST5401, fisetin producer strain (ST5075) with an additional copy of fused CHS and CHR B. ST5402, fisetin producer strain (ST5075) with an additional copy of CHS and CHR.C.Strain ST4972 engineered for biosynthesis of fisetin using the PAL pathway.The background strain for strains ST5401 and ST5402 was strain ST5074.The strains were cultivated in FIT medium for 72 h, the mean value of extracellular concentration of compounds was calculated from five biological replicates.

Table 2
Plasmids used in this study.