Coupling gene regulatory patterns to bioprocess conditions to optimize synthetic metabolic modules for improved sesquiterpene production in yeast

Background Assembly of heterologous metabolic pathways is commonly required to generate microbial cell factories for industrial production of both commodity chemicals (including biofuels) and high-value chemicals. Promoter-mediated transcriptional regulation coordinates the expression of the individual components of these heterologous pathways. Expression patterns vary during culture as conditions change, and this can influence yeast physiology and productivity in both positive and negative ways. Well-characterized strategies are required for matching transcriptional regulation with desired output across changing culture conditions. Results Here, constitutive and inducible regulatory mechanisms were examined to optimize synthetic isoprenoid metabolic pathway modules for production of trans-nerolidol, an acyclic sesquiterpene alcohol, in yeast. The choice of regulatory system significantly affected physiological features (growth and productivity) over batch cultivation. Use of constitutive promoters resulted in poor growth during the exponential phase. Delaying expression of the assembled metabolic modules using the copper-inducible CUP1 promoter resulted in a 1.6-fold increase in the exponential-phase growth rate and a twofold increase in productivity in the post-exponential phase. However, repeated use of the CUP1 promoter in multiple expression cassettes resulted in genetic instability. A diauxie-inducible expression system, based on an engineered GAL regulatory circuit and a set of four different GAL promoters, was characterized and employed to assemble nerolidol synthetic metabolic modules. Nerolidol production was further improved by 60% to 392 mg L−1 using this approach. Various carbon source systems were investigated in batch/fed-batch cultivation to regulate induction through the GAL system; final nerolidol titres of 4–5.5 g L−1 were achieved, depending on the conditions. Conclusion Direct comparison of different transcriptional regulatory mechanisms clearly demonstrated that coupling the output strength to the fermentation stage is important to optimize the growth fitness and overall productivities of engineered cells in industrially relevant processes. Applying different well-characterized promoters with the same induction behaviour mitigates against the risks of homologous sequence-mediated genetic instability. Using these approaches, we significantly improved sesquiterpene production in yeast. Electronic supplementary material The online version of this article (doi:10.1186/s13068-017-0728-x) contains supplementary material, which is available to authorized users.


Background
Metabolic engineering and synthetic biology are now routinely used for the engineering of microorganisms for industrial production of desirable chemicals, including fuels and biochemicals [1][2][3]. In the first instance, metabolic pathway flux towards the desired product is optimized by introduction of enzymes with the best catalytic efficiency (which are often heterologous) [4][5][6]. Expression levels of these enzymes are then titrated for optimal pathway balance, in combination with other metabolic engineering strategies to redirect carbon in the metabolic network [7][8][9]. Complicating matters, the activities of synthetic pathways should be coordinated with the dynamic fermentation conditions and process stage [10]. Imbalance in pathway flux can have a negative effect on cell physiology (e.g. growth rate) and on product titre [11]. Coordination can be controlled at the transcriptional level to regulate gene expression (and hence, enzyme activity). However, there is only limited information available on the dynamic behaviour of promoters across the range of conditions that occur in industrial fermentation processes.
The budding yeast Saccharomyces cerevisiae is a common engineering platform for production of highvalue plant terpenoids. Terpenoids are a diverse class of chemicals naturally synthesized from the universal 5-carbon precursors, isopentenyl pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP) [12]. The sesquiterpene sub-class of terpenes have 15 carbon atoms and are synthesized from farnesyl pyrophosphate (FPP), which is condensed from one molecule of DMAPP and two molecules of IPP. Sesquiterpenes have broad industrial applications, including as fragrances, flavours, pharmaceuticals, solvents and fuels. A generic set of metabolic engineering approaches can be used to improve sesquiterpene production ( Fig. 1) [13][14][15][16]. First, high-level production of terpenoids requires improved flux from central carbon metabolism to the sesquiterpene precursor FPP [16][17][18][19][20][21][22][23]. This is typically achieved by augmenting the mevalonate (MVA) pathway through overexpression or heterologous expression of individual genes (including farnesyl pyrophosphate synthase, FPPS). Enhanced MVA pathway activity causes squalene accumulation [23][24][25] and it is necessary to constrain the flux-competing squalene synthase to redirect FPP flux away from sterol production and towards sesquiterpene production (Fig. 1). This can be achieved by decreasing activity of the FPP-consuming enzyme squalene synthase, either through engineered protein degradation [23] or transcriptional down-regulation [16,26,27]. These steps provide the basic principles of pathway optimization for sesquiterpene production in yeast.
An ideal microbial cell factory should simultaneously exhibit high specific production rate and high specific growth rate in a batch cultivation [28]. However, these two objectives are commonly incompatible due to the metabolic burden and/or metabolic imbalance found in the presence of engineered pathways [29][30][31]. An alternative option is to separate growth and production phases [31]. This can be achieved by induction of synthetic pathway genes upon an environmental stimulus occurring after sufficient biomass is accumulated [32].
Regulation of gene expression across a batch cultivation (expression pattern) is delicately controlled by the gene promoter, transcriptional regulatory networks and the environmental inputs (usually the cultivation conditions, including activator/repressor concentration in cultures) [33][34][35][36]. Native promoters and regulatory networks can readily be used for metabolic pathway construction and optimization if their response conditions are well characterized. Several promoters have been well characterized to achieve inducible gene expression systems in yeast; these include a copper-inducible promoter (P CUP1 , induced by high concentration of copper ion), galactoseinducible promoters (the bi-directional P GAL1 and P GAL10 promoters, de-repressed in the absence of glucose and induced when galactose is present), a sucrose-inducible promoter (P SUC1 , de-repressed in the absence of glucose and induced when sucrose is present), a high-affinity hexose transporter promoter (P HXT7 , induced when glucose levels are low) and heat shock transcriptional factor Hsf1p-mediated promoters (P SSA1 and P HSP26 ) [33,[36][37][38][39][40][41][42]. In addition, sophisticated synthetic regulatory circuits have been designed, including circuits that respond to cell density via an engineered quorum-sensing system [32] and circuits that are activated by product feedback [43]. These promoters and regulatory networks can be further explored for optimizing gene expression regulation in metabolic engineering. Trans-nerolidol is a sesquiterpene alcohol with applications as fragrance, flavour, precursor for synthetic vitamin E/K1 and others [23]. Previously, we engineered a trans-nerolidol production pathway in yeast in concert with MVA pathway augmentation and a protein-mediated flux down-regulation strategy at squalene synthase [23]. We achieved a titre of ~100 mg L −1 , but observed a decreased growth rate when using constitutive overexpression of genes. To attain high-level production of nerolidol without a growth defect, further metabolic engineering is required. In this work, we engineered transcriptional regulation module that responds to bioprocess conditions to optimize growth and production for improved nerolidol titre, in combination with metabolic pathway optimization.

Constitutive expression of genes results in decreased growth rates and constrains product titres
In our previous work, nerolidol production was improved by heterologously expressing more efficient upper MVA pathway genes from Enterococcus faecalis (EfmvaS and EfmvaE), overexpressing the yeast lower MVA pathway genes and FPP synthase and destabilizing squalene synthase (Erg9p) [23]. The resulting strain (Table 1) produced 104 ± 35 mg L −1 nerolidol over 72 h in batch cultivation on minimal medium with 20 g L −1 glucose. All of the genes, including nerolidol synthase (AcNES1) were overexpressed from plasmids using promoters with constitutive activity (P RPL4A for EfmvaS, P RPL15A for Efm-vaE, P RPL8B for ERG12, P SSB1 for ERG8, P RPL3 for MVD1, P YEF3 for IDI1, P TEF2 for ERG20 and P TEF1 for AcNES1; Tables 1 and 2). These promoters exhibit high-level activities in the exponential phase when glucose is available, but dramatically decreased activities when glucose  is depleted and cells have shifted to the ethanol growth phase [33]. Strain N6D exhibited half the specific growth rate (μ max of 0.16 ± 0.03 h −1 ) of the wild-type CEN.PK reference strain (μmax of 0.30 ± 0.01 h −1 ) [23]. We presume that this is due to the metabolic burden from high expression levels of the heterologous synthetic pathway in the exponential phase. Furthermore, the decreased expression level after the diauxic shift might lead to low productivities in the post-exponential phase. To examine productivity across the fermentation period, 24-and 72-h samples were re-analysed using a new HPLC-UV method. Consistent with the GC-MS data obtained previously [23], a titre of 125 ± 30 mg L −1 was measured at 72 h. At 24 h, the titre was 82 ± 37 mg L −1 (Fig. 2b) and the productivity was calculated to 5.7 mg g −1 biomass h −1 in the exponential phase (from 0 to 24 h) compared to only 0.18 mg g −1 biomass h −1 in the post-exponential phase (from 24 to 72 h). The decreased nerolidol production rate may be attributed to the lower promoter activity we previously reported for these constitutive promoters during the ethanol phase [33]. Indeed, the mRNA levels of the genes for nerolidol production (EfmvaS, EfmvaE, ERG12, ERG8, MVD1, IDI1, ERG20 and AcNES1) decreased 3-to 10-fold in the ethanol growth phase compared to the exponential growth phase (Fig. 2d). As noted previously [23], the mRNA levels of the reference genes ACT1 and PDC1 also decreased. Interestingly, in contrast to the classical 'housekeeping' genes ACT1 and PDC1, acetyl-CoA synthase (ACS2) exhibited similar transcriptional levels in the exponential/ethanol growth phases.

Copper-inducible expression improves nerolidol production
To avoid both the metabolic imbalances causing decreased growth rate in the exponential phase and the decreased productivities in the post-exponential phase, gene expression can be controlled using an induction system. The CUP1 promoter is responsive to copper ion concentration [38]. In the absence of additional copper, the CUP1 promoter exhibits a moderate expression level; with addition of high-concentration copper (above 100 μM), activity is induced to a level comparable to the strong TEF1 promoter in exponential phase [33]. In contrast to the TEF1 promoter, the CUP1 promoter can maintain high expression activity in the ethanol phase (in the presence of 300 μM copper; [33]). These characteristics make the CUP1 promoter an potentially useful candidate to address both of the aforementioned problems. To test the effect on sesquiterpene production, a new set of plasmids were constructed (  pPMVAu8 pRS423: P RPL4A >EfmvaS>T EFM1 -P RPL15A >EfmvaE>T EBS1 [23] pPMVAd3 pRS424: P RPL8B >ERG12>T NAT5 -P SSB1 >ERG8>T IDP1 -P RPL3 >MVD1>T PRM9 -P YEF3 >IDI1>T RPL15A [23] pJT1 pRS425: P TEF2 >ERG20>T RPL3 -P TEF1 -AcNES1-T RPL41B [23] pPMVAugw pRS423: Yeast integration plasmid; yEGFP without promoter [33] pILGH4 pILGFP3: with yEGFP removed [33] pILGB5A pILGFP3: P GAL1 -yEGFP [33]  pyrophosphate decarboxylase and IDI1-Isopentenyl diphosphate:dimethylallyl diphosphate isomerase) controlled by a TEF1 promoter and two TEF2 promoters, respectively; and finally, a nerolidol synthetic plasmid (pJT3) with ERG20 and AcNES1 controlled by two divergent CUP1 promoters (Fig. 2a). The three plasmids were co-transformed into the Erg9p-destabilized strain oH5 [23] to generate strain NC1D (Table 1). While the wild-type growth rate (μ max = 0.30 ± 0.01 h −1 ) was not fully recovered, strain NC1D exhibited a 1.6-fold faster growth rate (μ max = 0.25 ± 0.01 h −1 ) than strain N6D (Table 3), demonstrating that the metabolic imbalance in the exponential phase was partially relieved. Furthermore, higher nerolidol production was achieved, with a titre of 111 ± 26 mg L −1 at 24 h and 245 ± 8 mg L −1 at 72 h (Fig. 2b, c). This equates to production rates of 5.1 mg g −1 biomass h −1 in the period from 0 to 24 h, slightly lower than the rate in strain N6D; and 0.77 mg g −1 biomass h −1 in the post-exponential phase (from 24 to 72 h), fourfold higher than the rate in strain N6D (Table 3).  Table 3 The maximum growth rate (μ max ) and nerolidol productivities of the engineered strain in flask cultivation on 20 g L −1 glucose To verify the relationship between gene expression levels and nerolidol production, the mRNA levels in strain NC1D were also analysed. Unexpectedly, the mRNA levels for four genes from the 'lower' MVA pathway (ERG12, ERG8, MVD1 and IDI1; Fig. 2) were similar compared to those in the wild-type CEN.PK reference strain [23]. Furthermore, the overexpression cassettes could not be amplified from the genomic DNA (data not shown). The 2 μ plasmid originally containing the yeast lower MVA pathway was recovered from NC1D. Restriction mapping and DNA sequencing showed that DNA recombination occurred between pPMVAd36 and pJT3; the result of this recombination was a plasmid where the four lower MVA pathway gene expression cassettes from pPMVAd36 were replaced by the nerolidol synthase cassette from pJT3 (Additional file 1: Figure S1). This replacement pattern is consistent with two homologous recombination events occurring, one between the CUP1 promoters for ERG12 and AcNES1, and one between homologous sequences on the plasmid backbones. Despite this undesiriable recombination event, strain NC1D performed better than N6D (see above). Very low variability was observed between biological replicates for the mRNA levels ( Fig. 2e) and nerolidol production (Fig. 2b), demonstrating that the plasmid recombination and resulting strain were stable.
Despite the loss of the four lower MVA pathway genes, the upper pathway genes EfmvaS, EfmvaE and ERG20 and AcNES1 controlled by CUP1 promoters, were still present in NC1D. Consistent with our previous findings [33], the mRNA levels for these genes increased by about fourfold after copper induction, and high transcript levels were maintained in the ethanol growth phase for these genes (Fig. 2e). In the exponential phase under uninduced conditions, mRNA levels for EfmvaS and Efm-vaE were 2 and sevenfold lower ( Fig. 2e) than transcript levels driven by constitutive promoters (Fig. 2d). The 'lower' pathway was not augmented at all, considering the loss of the 'lower' MVA pathway genes during the recombination event. Together, these data support the idea that high transcription of MVA pathway genes during the exponential phase (driven by constitutive promoters) results in a metabolic imbalance that decreases growth rate and nerolidol titre. Applying the CUP1 promoter to control synthetic pathway genes improved nerolidol production; however, plasmid recombination caused by repeated usage of the CUP1 promoter is undesirable in a metabolic engineering context. Therefore, an alternative inducible expression system was developed.

Δgal80 enables auto-induction of GAL promoters over/ after diauxic shift
In a previous comparison of a set of commonly used promoters, the GAL1 promoter drove the highest expression level [33]. The GAL1 promoter and galactose-based cultivation have also been used in the initial laboratory development of several sesquiterpene-producing strains [46,47]. However, it is not feasible to use galactose as a carbon source in industrial production due to its high cost. In the core galactose regulon, Gal80p binds the transcription activator Gal4p to inhibit Gal4p-mediated transcription initiation of the GAL1 promoter in the absence of galactose; in the presence of galactose, transcriptional factor Gal3p binds with Gal80p to relieve Gal80p repression on Gal4p transcription activation (Fig. 3a) [48]. Galactoseindependent (gratuitous) activation of the GAL1 promoter can be achieved by disruption of the gal80 repressor [49]. Additionally, galactose-inducible expression driven by GAL1 promoter is co-regulated by Mig1p-mediated glucose-dependent repression: in the presence of glucose, Mig1p can bind to the GAL1, GAL3 and GAL4 promoters to inhibit gene transcription (Fig. 3a) [48,50]. Mig1pmediated repression can lead to the glucose-dependent repression of the GAL1 promoter in a gal80Δ strain [16,49,51]. Consequently, it has been shown that the GAL1 promoter can be automatically induced in a gal80Δ strain as the cells shift to the ethanol growth phase [51].
In the current study, to avoid repeated use of a single promoter for multiple genes, the promoters from the galactose metabolic genes (GAL1, GAL10, GAL2 and GAL7) and a reference TEF1 promoter were characterized in a gal80Δ strain (Tables 1, 2; Fig. 3b). GFP was used as a reporter to measure the promoter activities over an entire batch cultivation on 20 g L −1 glucose. Growth profiles were similar and showed a diauxic growth pattern in all strains (data for the gal80Δ control strain are shown in Fig. 3b; other data not shown). We showed previously that, under the same conditions, the diauxic shift pattern observed in the growth curve was coincident with glucose depletion and the start of ethanol consumption, and occurred at ~12 h [33]. Similar to our previous study in the wild-type strain [33], activity driven by the TEF promoter in the gal80Δ strain decreased dramatically during and after the diauxic shift. In contrast, the GAL promoters exhibited diauxie-inducible expression patterns: their activities started to increase at 12 h, peaking at 48 h. The strength of GAL promoters in the post-exponential phase was as follows: P GAL2 >P GAL1 >P GAL7 >P GAL10 . The auto-inducible expression pattern makes the GAL promoter in combination with gal80Δ very useful in strain development to simultaneously avoid the metabolic burden in the exponential phase and increase productivity in the ethanol growth phase.

Auto-inducible GAL promoter regulation drives efficient nerolidol production
To apply the modified galactose-inducible system (GAL promoters in gal80Δ background), a new nerolidol-producing strain, N391DA, was constructed ( Fig. 4a; Tables 1, 2). In the "upper" MVA module, genes EfmvaS (HMG-CoA synthase), ACS2 (native acetyl-CoA synthase) and EfmvaE (thiolase/HMG-CoA reductase) were controlled by P GAL1 , P GAL10 and P GAL2 , respectively. Overexpressing ACS2 has previously been shown to increase the intracellular concentration of acetyl-CoA [21], which is the precursor metabolite for the MVA pathway; hence, it was included in the 'upper' module. The three expression cassettes were integrated into the HMG2 (HMG-CoA reductase) promoter locus with the selection marker HIS3. At the same time, a P GAL7 promoter linked to a short sequence of the HMG2 gene including a K6R mutation that stabilizes Hmg2p from degradation [52,53] was introduced as a fusion with the native HMG2. As a result, HMG2 K6R was expressed from the genome under the control of P GAL7 . To construct a "lower" MVA module, ERG12 (mevalonate kinase) was controlled by P GAL2 ; the other three genes (ERG8, MVD1, IDI1) were controlled by glucose-dependent "constitutive" promoters (P TEF2 , P ADH1 and P TEF1 , respectively). This construct was integrated with the TRP1 selection marker into the PDC5 locus, which encodes a weakly expressed pyruvate decarboxylase [54]. Dysfunction of PDC5 does not cause a major change in yeast metabolism, because of complementation by isoforms PDC1 and PDC6 [55]. For the nerolidol synthesis module, P GAL1 -controlled ERG20 and P GAL2 -controlled AcNES1 were introduced on a 2μ plasmid. Finally, squalene synthase (Erg9p) was destabilized by the addition of an endoplasmic-reticulum mediated protein degradation sequence to reduce its ability to compete for FPP with nerolidol synthase [23], and gal80 was disrupted to allow diauxic induction of GAL promoters. The resulting strain N391DA was evaluated through two-phase flask cultivation.
Strain N391DA exhibited a normal exponential growth with μmax of 0.27 ± 0.01 h −1 -1.7-fold faster than the strain N6D and slightly faster than strain NC1D. Consistent with the yeast diauxic growth model, N391DA consumed glucose and produced ethanol in the exponential growth phase, and ethanol was subsequently consumed in the secondary growth phase (Fig. 4b). N391DA produced 38 mg L −1 nerolidol at 24 h, production rates of 1.4 mg g −1 biomass h −1 in the period from 0 to 24 h, fourfold lower than the rate in strain N6D. Nerolidol reached 392 ± 2 mg L −1 at 72 h, which translates to a production rate of 2.1 mg g −1 biomass h −1 in the post-exponential phase (from 24 to 72 h)-12-fold higher than the rate in strain N6D ( Table 3). The 72 h titre represents a 60% improvement relative to NC1D with copper-inducible constructs. N391DA exhibited the highest specific nerolidol production rate (r nerolidol ) between 24 and 48 h and 86% of nerolidol was produced after 24 h (Fig. 4d).
To verify the expression pattern of the genes controlled by the modified galactose-inducible system, the mRNA levels were analysed in strain N391DA (Fig. 4c). In the exponential phase, the expression levels of the reference genes ACT1 and PDA1 in N391DA were similar (twotailed t test p > 0.1) to those in strains N6D and NC1D. As observed previously, their mRNA levels decreased significantly in the ethanol growth phase; this was also the case for mRNAs levels from the genes controlled by TEF1, TEF2 and ADH1 promoters. In contrast, the genes controlled by GAL promoters exhibited increased transcriptional levels: twofold for ACS2, 11-fold for Efm-vaS, 22-fold for EfmvaE, 27-fold for HMG2, ninefold for ERG12, 26-fold for ERG20 and fourfold for AcNES1 in the ethanol growth phase, compared to in the exponential growth phase. These fold-changes are consistent with the expression pattern of the four GAL promoters as characterized above (Fig. 3).

Sucrose as an alternative carbon source for nerolidol production
Sucrose from sugar cane/sugar beet is an alternative carbon source to glucose in industrial fermentation [56][57][58].
As for the GAL genes, the invertase gene for sucrose utilization (SUC2) is under Mig1p-mediated glucose repression, which is relieved when yeast is cultivated on sucrose [59,60]. Considering this, it is reasonable to expect that the expression output from GAL promoter in gal80Δ background strain might be different on sucrose than on glucose. To investigate the effect of sucrose on nerolidol production, GAL promoter activities in gal80Δ background strains and nerolidol production for strain N391DA were characterized on sucrose. Yeast strains were pre-cultured on 40 g L −1 glucose, which minimized GAL promoter activities in gal80Δ strains (data not shown). A 6-h lag phase was exhibited after transferring to 20 g L −1 sucrose medium; this lag phase was seen in both gal80Δ strains and the GAL80 control strain ( Fig. 5a; Additional file 1: Figure S2). During the lag phase on sucrose, glucose repression on GAL promoters was relieved. GFP expression driven by the GAL1 and GAL2 promoters increased sharply during the lag phase and plateaued during exponential growth (Fig. 5a); this expression level was similar to that observed in the ethanol growth phase of glucose batch cultivation (Fig. 3b). A   Fig. 4 Characterizing strain N391DA with gal80Δ-GAL promoter constructs: a genetic modules/plasmid for the "upper" mevalonate pathway (pIM-VAu1), the "lower" mevalonate pathway (pIMVAd39T) and the nerolidol synthetic genes (pJT9R); b, d metabolic and growth profiles (N = 2); c mRNA levels (N = 3) in the exponential phase (EXP) and the ethanol growth phase (ETH, at 36 h). Two-phase flask cultivation on 20 g L −1 glucose was employed. Mean values ± standard deviations are shown further increase of ~twofold was observed after 24 h, presumably after the diauxic shift (Fig. 5a).
To characterize nerolidol production on sucrose, strain N391DA was first pre-cultured on 40 g L −1 glucose and then cultivated on 20 g L −1 sucrose. N391DA exhibited a 48-h lag phase on sucrose (Fig. 5b), dramatically longer than that for the GFP strains (Fig. 5a). The sugar profile demonstrated that strain N391DA first fermented sucrose and its hydrolysate products (glucose and fructose) into ethanol; the diauxic shift occurred by 72 h, and the strain then re-used ethanol in the post-exponential phase (Fig. 5b). In sucrose batch cultivation, the final nerolidol titre was 632 ± 57 mg L −1 , 1.6-fold higher than in glucose batch cultivations. In addition, N391DA exhibited the highest post-exponential-specific nerolidol production rate of 5 mg g −1 biomass h −1 (Fig. 5c compared to Fig. 4d).

Nerolidol production in fed-batch cultivation
To achieve high-titre nerolidol production for strain N391DA, fed-batch strategies were explored. We first explored a strategy designed to ensure that (a) fermentable sugars are catabolized through respiratory metabolism and (b) cultures are maintained under aerobic conditions. The initial feed rate was set to 1 mM glucose g −1 biomass h −1 with 600 g L −1 glucose feeding medium and then exponentially increased with a rate of 0.05 h −1 ; the feeding was switched off when dissolved oxygen (DO) was below 25% and maximum agitation and gassing were achieved, and the feeding was re-triggered when DO was above 30% (Additional file 1: Figure S3a). Two additional experiments were performed using volumetrically similar initial feed rates and feed solutions of 600 g L −1 sucrose and 400 g L −1 glucose/158 g L −1 ethanol, respectively.
In the three experiments, the respiration quotients fluctuated around 1 for glucose or sucrose feeding processes, and around 0.9 for glucose/ethanol feeding process (Additional file 1: Figure S4), demonstrating that the fermentable sugars were catabolized through respiration. All processes began in batch mode using 20 g L −1 glucose as a carbon source and proceeded through diauxie and into the ethanol growth phase until DO started in increase sharply, triggering the feed. In this batch period, the three cultures produced 406 ± 57 mg L −1 nerolidol at 30 h (Fig. 6a). In the subsequent feeding phase, the glucose/ethanol feed provided higher nerolidol production than glucose or sucrose feeding; >2 g L −1 nerolidol was achieved at 102 h for glucose/ethanol and this titre was not achieved until 150 h for glucose and sucrose feeding. The final titre for the glucose/ethanol feed was >3 g L −1 at 174 h. For all three fed-batch experiments, the specific nerolidol production rates during feeding were lower than the rate observed in the ethanol growth phase in the batch process (Fig. 6a). The C-mole yield at 102 h was 2.0 ± 0.4% in these three fed-batches.
Next, the feeding strategy was altered with the aim to maintain overflow metabolism and cycling between ethanol production and consumption (Additional file 1:  Figure S3b). After the batch phase, exponential feeding with an initial feeding rate of 3 mM glucose g −1 biomass h −1 for 600 g L −1 glucose (or volumetrically the same for 600 g L −1 sucrose) feeding medium and a specific increasing rate of 0.05 h −1 were applied. Once 50 g L −1 sugar had been fed, the feeding was paused to allow cells to consume the ethanol produced during sugar fermentation. Subsequently, 10 or 20 g L −1 sugar pulse feeding was repeatedly triggered by sharp DO increases (Additional file 1: Figure S5). The production of ethanol was confirmed by the respiration quotient being over 2 when the sugar was fed (Additional file 1: Figure S5). In the batch phase for the three batches, 404 ± 11 mg L −1 nerolidol was produced (Fig. 6b), consistent with the above results (Fig. 6a). In these three fed-batch processes (Fig. 6b), >4 g L −1 nerolidol was produced at 96 h, and the specific nerolidol production rates in the early feeding phase (t < 96 h) were noticeably higher than those in the carbon-restricted processes (Fig. 6a). At 96 h, the C-mole yield was 3.8 ± 0.1% in the two glucose-overflow fedbatches and 4.5% in the sucrose-overflowed fed-batch (Fig. 6b).

Discussion
An efficient sesquiterpene-producing yeast platform is of broad industrial interest, because it can be applied for the production of various high-value sesquiterpenes as well as FPP-derived biofuels by simply substituting a specific terpenoid production pathway (Fig. 1). There are two key principles for increasing FPP availability for sesquiterpene production in yeast: enhancing the MVA pathway to increase precursor supply, and reducing squalene synthase activity to decrease its consumption of FPP [14, 17-19, 23, 26, 27, 51, 61]. This study aimed to couple the transcriptional regulation pattern of heterologously expressed pathway genes to bioprocess conditions to minimize metabolic imbalance and optimize heterologous sesquiterpene production.
In previous studies, constitutive promoters were most often used to control the expression of heterologous genes, whereas the optimization of expression strength over the different fermentation stages has not been well investigated [7,8]. In the current study, three categories of transcription regulation patterns were applied to assemble synthetic pathways: constitutive, copper-inducible and diauxie-induced. Using reporter gene systems, we showed previously that expression outputs from constitutive promoters (including "classical" translational elongation factor promoters, P TEF1 and P TEF2 ; glycolytic promoter, P ADH1 ; ribosome biogenesis promoters, P RPL3 , P RPL15A ) dramatically decrease after the diauxic shift [33]. Our transcription data (Fig. 2d) and nerolidol production data (Fig. 2b) confirmed this. Nerolidol is not toxic and does not cause dramatic growth inhibition (Additional file 1: Figure S6), in contrast to monoterpenes [62]. But we also observed a decreased growth rate (Fig. 2c), which was consistent with metabolic burden from high-level expression in the exponential phase.
Using a copper-inducible promoter (P CUP1 ), gene expression was shifted from exponential to the ethanol growth phase (Fig. 2e). This resulted in improved growth rate during the exponential phase (Fig. 2c)

Fig. 6
Nerolidol production for strain N391DA in fed-batch cultivations: a nerolidol production (solid line) and specific production rate (r; dashed line) in carbon-source-restricted DO-triggered fed-batch cultivation with feeding carbon source of 600 g L −1 glucose (R-Glc), or 600 g L −1 sucrose (R-Suc), or 400 g L −1 glucose 158 g L −1 ethanol mixture (R-Glc/Eth); b nerolidol production (solid line) and specific production rate (r; dashed line) in carbon-source-overflowed/carbon-source-pulsing fed-batch cultivation with feeding carbon source of 600 g L −1 glucose (with 10 g L −1 glucose pulse, O10-Glc; with 20 g L −1 glucose pulse, O20-Glc) or 600 g L −1 sucrose (with 20 g L −1 sucrose pulse, O20-Suc). Vertical dashed line approximately indicated the start of feeding. Growth and process values refer to Additional file 1: Figures S4, S5. N = 1 productivities after the diauxic shift (Fig. 2b). However, the repeated use of the CUP1 promoter sequence led to undesirable homologous recombination and loss of introduced DNA fragments (Additional file 1: Figure S2). It is likely that similar losses have occurred previously when repeated promoter sequences have been used, for example, Beekwilder et al. [63] observed that almost half of colonies did not produce a carotenoid reporter product when they used the TDH3 promoter to control several carotogenic pathway genes. Genetic instability at repeat sequences has been documented in yeast [64], and provides a challenge for largescale modification/introduction of metabolic pathways due to limited availability of appropriate promoters [33].
To address this problem, we characterized four native GAL promoters (P GAL1 , P GAL10 , P GAL2 and P GAL7 ) in a gal80Δ genetic background to engineer a diauxie-inducible system [16,65], and applied them to assemble the MVA pathway modules (Figs. 3, 4a). While these promoters in gal80Δ background have been used to control the expression of the genes from the MVA pathway and amorphadiene/isoprene synthetic pathway previously [16,65], they have not been characterized in terms of behaviour throughout the fermentation nor have they been ranked. The relative strength of P GAL1 and P GAL10 has been examined previously with conflicting results. One study [41] observed P GAL10 to be stronger than P GAL1 ; in contrast, our study and two others [39,66] showed P GAL1 to be stronger than P GAL10 . It is possible that differences in experimental conditions caused this discrepancy. Using the GAL promoters, a further shift in gene expression from exponential to ethanol growth phase was achieved (Figs. 2e, 4c), which resulted in further improved growth during the exponential phase (Fig. 4b) and further improved productivities after the diauxic shift (Figs. 2b, 4d). Increased expression from the GAL promoters and improved nerolidol productivities were consistently observed when the engineered cells were cultivated on sucrose (Fig. 5). Together, these data demonstrate the importance of appropriate transcriptional control modules for optimized strain engineering. There are several genotypic differences between strains NC1D and N391DA, including promoter identity (P CUP1 vs. P GAL ), gene location (plasmid vs. genome integration) and gene expression level; and we have not examined each in detail to determine relative contribution to the phenotype.
In a previous study using the gal80Δ background, the sesquiterpene amorphadiene was produced at >40 g L −1 titre in ethanol fed-batch cultivation, whereas it reached only ~3 g L −1 in glucose fed-batch cultivation [16]. Here, nerolidol production in fed-batch cultivation was also investigated in the gal80Δ background; in contrast to the previous study, two substrate-feeding strategies, sugarrestricted and sugar-surplus, were compared (Fig. 6). In the sugar-restricted feeding processes, sugar was expected to be catabolized through respiratory metabolism; in the sugar-surplus feeding processes, ethanol was expected to accumulate in the sugar respiro-fermentative metabolism after each sugar feed and to be re-consumed when sugar was depleted. The two feeding strategies resulted in similar growth profiles (comparing OD 600 when t ≤ 102 or 96 h; Additional file 1: Figures S4, S5). However, in the carbon-overflow fed-batches, higher titre/yield/rate of nerolidol were achieved, compared to the substraterestricted fed-batches (Fig. 6). In addition to different promoter behaviours throughout the fed-batch cultivation, the improvement in nerolidol production might be due to a MVA pathway response to "overflow" metabolism, as is seen in ethanol/acetate/glycerol production through glycolysis [67][68][69]. In support of this possibility, the concentrations of intracellular pyruvate, acetyl-CoA, acetoacetyl-CoA and FPP were shown to increase as the dilution rate increased in carbon-limited chemostat processes [70,71]. The pyruvate node is a key metabolic node for regulation of carbon flux in yeast; specifically, it determines flux towards mitochondrial pyruvate dehydrogenase or the cytosolic pyruvate dehydrogenase bypass (and ultimately, acetyl-CoA, which provides substrate for the MVA pathway) [72,73]. Moreover, flux distribution into the cytosolic pyruvate dehydrogenase bypass decreases with decreased glucose feeding rate [74]. This can explain the decreased flux through the MVA pathway for nerolidol production under the carbon-limited fed-batch.

Conclusion
Here, we present an expanded modular genetic regulatory system to co-ordinate expression outputs with physiological behaviour of engineered cells under industrially relevant cultivation conditions. The gene transcriptional regulatory pattern for assembled metabolic pathways is critically important for metabolic functionality and optimal productivity in engineered strains. This principle should be considered for de novo assembly of heterogeneous metabolic pathways in the "design-build" construction cycle for industrial microbial producers. The diauxie-induction system, including the four characterized GAL promoters, is simple and efficient for developing a yeast strain for high-level sesquiterpene production. Further optimization to balance sugar and ethanol growth phases might significantly increase titres.

Two-phase flask cultivation and sampling
Nerolidol-producing strains were evaluated through two-phase flask cultivation using dodecane as a non-toxic organic extractant phase [76]. Synthetic minimal (SM) medium containing 6.7 g L −1 YNB (Sigma-Aldrich #Y0626; pH 6.0) with 20 g L −1 glucose as the carbon source was used (SM-glucose). 100 mM 2-(N-morpholino) ethanesulfonic acid (MES, Sigma-Aldrich#M8250) was used to buffer medium pH and the initial pH was adjusted to 6.0 by adding ammonium hydroxide. Strains were recovered from glycerol stocks by streaking on SM-glucose agar plates (for N6D and NC1D) or SCGHG agar plates (for N391DA) and pre-cultured in MES-buffered SM 20 g L −1 glucose (or 40 g L −1 glucose as indicated) medium to exponential phase (cell density OD 600 between 1 and 4). Pre-cultured cells were collected by centrifugation and re-suspend in fresh media before initiating two-phase cultivations. Two-phase flask cultivation was initiated by inoculating pre-cultured cells to OD 600 = 0.2 in 25 mL MES-buffered SM 20 g L −1 glucose medium (or 23 mL MES-buffered SM 20 g L −1 sucrose medium when cultures were not sampled for RNA extraction) in 250 mL flasks with solvent-resistant screw-caps [76]; 2 mL dodecane was added to extract nerolidol. Flask cultivation was performed at 30°C and 200 rpm. For all cultivations, about 3 mL culture was sampled before the end of exponential growth phase for OD measurement; meanwhile, dodecane and culture were sampled in 1:10 ratio for metabolite analysis (see below). For RNA extraction, 2 mL exponential-phase culture (OD 600 = 1-1.5) and 1 mL ethanol growth-phase culture (at 36 or 48 h) were sampled, and cells were collected and stored at −80 °C.
In batch cultivation, 130 mL batch medium and 25 ml dodecane were used initially; seed culture in exponential phase (OD 600 = 5-10) was directly inoculated to OD 600 = 0.2 in batch culture. Medium feeding was triggered when the DO increased sharply after carbon sources in batch medium were depleted. Medium feeding was programmed using the DASGIP VB.NET script feeding controller (for feeding script logic charts refer to Additional file 1: Figure S3).

Quantitation of mRNA level
Total RNA was extracted using a yeast RiboPure ™ RNA Purification Kit (Ambion #AM1926) or a TRIzol ® Plus RNA Purification Kit (Ambion #12183555). After DNase treatment, 0.1 or 0.2 μg total RNA was used for firststrand cDNA synthesis in a 10 μL reaction using Proto-Script ® II Reverse Transcriptase (NEB #M0368). The diluted cDNA was used as the template for quantitative real-time (qRT) PCR (primers are listed in Additional file 1: Table S1#29 to #40). KAPA SYBR ® FAST qPCR Kit (Kapa Biosystems#KP-KK4601) and CFX96 Touch ™ Real-Time PCR Detection System (BIO-RAD) were used in qRT-PCR. Yeast genomic or plasmid DNA was used to prepare the standard curve. C t values were analysed using CFX Manager Software (Bio-Rad Laboratories, QLD Australia). The mRNA levels (N pg −1 total RNA) were calculated by referring to the standard curve equations.

GFP fluorescence assay
To monitor GFP fluorescence over the entire batch cultivation, cells were cultivated aerobically in MES-buffered 20 mL SM-glucose (SM-sucrose) medium in 100 mL flasks. Samples were taken at indicated time points, and GFP fluorescence in single cells was analysed immediately after sampling using a flow cytometer (BD Accuri ™ C6; BD Biosciences, USA). Cultures were diluted after 12 h by 10 volume water for flow cytometer analysis. GFP fluorescence was excited by a 488 nm laser and monitored through a 530/20 nm bandpass filter (FL1.A); 10,000 events were counted per sample. The GFP fluorescence signal (FL1.A) was corrected according to the values of FSC.A (forward scatter detector) and SSC.A (side scatter detector) [33]. GFP fluorescence level was expressed as the percentage relative to the average background auto-fluorescence from exponential-phase cells of a GFP-negative reference strain GH4J3.

Metabolite analysis
Extracellular metabolites were analysed by the Metabolomics Australia Queensland Node. Trans-nerolidol was analysed using a novel HPLC method. Dodecane samples (in some cases, diluted with dodecane) were diluted in 40-fold volume of ethanol. Ethanol-diluted dodecane samples of 20 μL were injected into a Zorbax Extend C18 column (4.6 × 150 mm, 3.5 µm, Agilent PN: 763953-902) with a guard column (SecurityGuard Gemini C18, Phenomenex PN: AJO-7597). Analytes were eluted at 35 °C at 0.9 mL/min using the mixture of solvent A (high purity water, 18.2 kΩ) and solvent B (45% acetonitrile, 45% methanol and 10% water), with a linear gradient of 5-100% solvent B from 0 to 24 min, then 100% from 24 to 30 min and finally 5% from 30.1 to 35 min. Analytes of interest were monitored using a diode array detector (Agilent DAD SL, G1315C) at 196 and 202 nm wavelengths. Spectral scans were also performed on each of the compounds from 190 to 400 nm in steps of 2 nm to confirm their identity and purity. Trans-nerolidol primary pharmaceutical reference standard (93.7% purity; Sigma-Aldrich #04610590) was used to prepare the standard curve for quantification.

Physiological feature calculation
Physiological parameters were calculated as reported previously [80]: the maximum growth rates are the linear regression coefficients of the ln OD 600 versus time during the exponential growth phase; one unit of OD 600 equals 0.23 g L −1 biomass; the specific production rate of nerolidol (r nerolidol , mg g −1 biomass h −1 ) was calculated by dividing Δ nerolidol titre (mg L −1 ) with the integral of biomass (g L −1 ) in defined time (h).