Enhancing α-bisabolene titer in R. toruloides
Previous work on engineering R. toruloides for production of α-bisabolene indicated that flux through the native mevalonate pathway is relatively high in this species. A strain modified only by insertion of a heterologous α-bisabolene synthase gene (BIS) from Abies grandis under control of the native R. toruloides GAPDH (glyceraldehyde 3-phosphate dehydrogenase) promoter into WT R. toruloides achieved α-bisabolene titer of 294 mg/L in a defined medium containing 2% (w/v) glucose and 680 mg/L in a 2 L bioreactor fed with corn stover hydrolysate [6]. Since heterologous DNA was introduced into R. toruloides by Agrobacterium tumefaciens mediated transformation (ATMT), which results in DNA integration at random loci with a variable copy number, it was of interest to investigate the correlation between copy number and α-bisabolene titer for a range of PGAPDH-BIS transformants. A good correlation (R2 = 0.93, p-value = 1e-11) exists between α-bisabolene titer and PGAPDH-BIS copy number for the 20 strains examined and, since a plateau did not appear to have been reached, it seemed likely that BIS expression remained a limiting factor for α-bisabolene production (Fig. 1a). The linearity of this correlation also suggests that insertion locus may be less important than copy number in determining the level of heterologous gene expression. To test the hypothesis that the α-bisabolene titer is still limited by BIS gene expression, the highest-titer PGAPDH-BIS strain, BIS3, was selected for addition of a second expression cassette consisting of BIS under control of the native R. toruloides ANT (adenine nucleotide translocase) promoter [15]. This resulted in strain, GB2, which produced 1.5-fold more α-bisabolene than the parent strain, BIS3, and contained 6 copies of the PANT-BIS cassette in addition to the original 10 copies of the PGAPDH-BIS cassette in BIS3 (Fig. 1b).
Scale up of α-bisabolene production in lignocellulosic hydrolysate
R. toruloides has been identified as a promising host for production of renewable biofuels and bioproducts because it can efficiently utilize mixed carbon sources and tolerate potential growth inhibitors often found in lignocellulosic hydrolysates [6]. We therefore wanted to determine how the new high BIS-copy GB2 strain performs when grown on a lignocellulosic hydrolysate derived from corn-stover (DMR-EH) for α-bisabolene production [13]. Growth and productivity were measured in 2 L bioreactors and the impact of nitrogen source and pH were investigated. Two of the bioreactors (A6 and A7, respectively) were used to compare two nitrogen sources supplementing the DMR-EH base medium: complex (10 g/L yeast extract) and defined (5 g/L ammonium sulfate). The medium supplemented with defined nitrogen also contained 100 µM iron sulfate and 100 mM potassium phosphate (pH 6.0). Medium pH was maintained above 5.0 during growth in these two bioreactors, while a third culture (A8) was grown in the defined nitrogen DMR-EH medium with no pH control. Although the pH in the third bioreactor dropped significantly (~ pH 3) over the course of the run, the final α-bisabolene titer, 1.9 g/L, was close to that of the pH-controlled cultures (Fig. 2). The two pH-controlled cultures grown on defined nitrogen and complex nitrogen also reached similar α-bisabolene titers, 2.1 and 2.2 g/L, respectively. The pH in the culture supplemented with yeast extract rose slightly to around pH 6.0 during the first half of the fermentation, only dropping as low as pH 5.0 by the end of the fermentation, while the culture supplemented with ammonium sulfate required an adjustment to keep the pH above 5.0 from days 2 to 8 (Fig. 2b).
Transcriptomics and proteomics analysis of α-bisabolene producing strains
Generation of strains that produce different tiers of α-bisabolene provides an opportunity to understand how R. toruloides respond to the diversion of carbon flux toward a heterologous product and to identify potential targets for metabolic engineering. To examine this response on a systems level, wild-type R. toruloides and four strains that produced α-bisabolene at various titers (Fig. 1b) were selected for global proteomic and transcriptomic analysis. Growth and sugar consumption were monitored in SD medium containing 10 g/L glucose and samples taken at 18 and 48 hours were used for omics analysis (Fig. S1). Both protein and transcript data indicate that BIS is one of the most highly expressed genes in the highest-titer strain, GB2, harboring both PGAPDH-BIS and PANT-BIS (Fig. 3 and Table S1). Comparing the strains, BIS transcript and protein levels increase in parallel with the increase in copy number and α-bisabolene titer, surpassing expression levels of two of the strongest native genes, ANT and TEF1 (translation elongation factor 1α), in most cases (Table S1). Interestingly, expression of the first two enzymes of the mevalonate pathway, acetyl-CoA acetyltransferase (ERG10) and 3-hydroxy-3-methylglutaryl-CoA (HMG-CoA) synthase (ERG13), also increase in proportion to α-bisabolene production, suggesting that the cell induces regulation to increase metabolic flux into the mevalonate pathway to accommodate the diversion of carbon away from native mevalonate pathway products to α-bisabolene. The same is true of the native FPP synthase (ERG20), which supplies the precursor to α-bisabolene and as well as being a key control point in ergosterol biosynthesis. The expression of other pathway genes underwent more modest changes in expression, even decreasing slightly in GB2, suggesting that they may be good targets for metabolic engineering to enhance flux through the mevalonate pathway.
Enhancing 1,8-cineole production in R. toruloides by increasing GPP supply
In this study, we attempt to optimize production of both α-bisabolene and 1,8-cineole. Before attempting to optimize the mevalonate pathway for both terpenes, we first wanted to ensure that there were sufficient pools of the pyrophosphate precursors for these two products. While the α-bisabolene precursor FPP is a central metabolite in the ergosterol pathway, 1,8-cineole is made from GPP, which previous work indicates is significantly more limited in R. toruloides [2]. Similar to S. cerevisiae and other fungi, R. toruloides lacks a dedicated GPP synthase (GPPS), and FPP is made directly from the C5 precursors, IPP and DMAPP by the enzyme ERG20. Therefore, the initial strategy for engineering higher 1,8-cineole titers focused on optimizing and balancing expression levels of the two terminal enzymes: GPPS and 1,8-cineole synthase. Engineering a high monoterpene titer requires a balance between GPPS and terpene synthase activities that provides sufficient flux to accumulate the target product while avoiding growth-inhibitory levels of GPP [11, 27]. Promoters for these synthases were selected from native R. toruloides genes. Three promoters, ANT, GAPDH, and TEF1 were selected based on their relative strength, constitutive expression profiles, and utility in prior work [2, 3, 5, 15]. PANT and PGAPDH were used to express HYP3 from Hypoxylon sp. E7406B, encoding a 1,8-cineole synthase previously identified as a promising enzyme for monoterpene production in R. toruloides [2]. PTEF1 was used to drive candidate GPPSs, several of which are FPP synthases containing mutations that alter the prenyl phosphate product chain length specificity in favor of GPP. Genes were transformed into R. toruloides by ATMT in either stepwise fashion (gene stacking) or by the use of single, combined constructs. Following each transformation event, up to 40 strains were screened for 1,8-cineole production and the best strains were retested in triplicate in YPD10 medium. The highest-titer strain resulting from each transformation was selected for comparison and further strain engineering.
Various configurations of HYP3 were combined with a GPPS from A. grandis with the N-terminal plastid transit peptide removed (tAgGPPS2) or not (AgGPPS2), under control of the TEF1 promoter (Fig. 4a). Each of these strains demonstrated some improvement over the previously published 1,8-cineole titer of 35 mg/L, suggesting that GPP synthesis is limiting to some extent [2]. Combining PTEF1-AgGPPS2 with PANT-HYP3 (strain 308.14) resulted in a 1,8-cineole titer of 58 mg/L and switching the HYP3 promoter from PANT to PGAPDH did not change the titer significantly (strain 92.6-305.5, producing 66 mg/L 1,8-cineole). A comparison of strains harboring both PGAPDH-HYP3 and PANT-HYP3 with either PTEF1-tAgGPPS2 or PTEF1-AgGPPS2 indicates that the truncated GPPS (strain 92.6-309.4, 73 mg/L 1,8-cineole) performed slightly better than the full-length version (strain 92.6-308.9, 52 mg/L 1,8-cineole).
Success with engineering monoterpene production has been reported by the use of FPP synthases harboring mutations that alter the substrate-binding pocket such that prenyl phosphate chain length elongation beyond C10 (GPP) is significantly reduced [11, 28–30]. Strains were constructed to compare various mutant FPP synthases, all under control of PTEF1 and combined with PANT-HYP3, and all but one of them produced significantly more 1,8-cineole than the strain harboring the plant GPPS, PTEF1-tAgGPPS2 (Fig. 4b). Mutants of the S. cerevisiae FPP synthase (ERG20) proved to be more effective than the corresponding mutants of the native R. toruloides ERG20, while the most promising enzyme of those tested was GgFPS(N144W) from Gallus gallus (chicken). The best strain harboring PANT-HYP3 and PTEF1-GgFPS(N144W), 307.2, reached a 1,8-cineole titer of 143 mg/L.
Additional strains were constructed to test further configurations of GgFPS(N144W) and HYP3 (Fig. 4c). Translational fusions of the two genes were expressed under control of PGAPDH and although one configuration (strain 378.4, with GgFPS(N144W) N-terminal) was significantly better than the other (strain 379.14), the best 1,8-cineole titer was only 30% of that reached by strain 307.2. When PTEF1-GgFPS(N144W) was combined with HYP3 under control of PGAPDH rather than PANT (strain 92.6-304.3) the 1,8-cineole titer was surprisingly much lower, compared to that of strain 307.2. The combination of both PGAPDH-HYP3 and PANT-HYP3 with PTEF1-GgFPS(N144W) via gene stacking (strain 91.2-92.30-304.38) resulted in a 1,8-cineole titer matching that of 307.2, indicating that perhaps a plateau had been reached that might only be surpassed by increasing flux through the mevalonate pathway.
Targeted mevalonate pathway metabolic analysis
Since this is a first attempt to engineer the mevalonate pathway in R. toruloides, we supplemented the global proteomic and transcriptomic data gathered from the α-bisabolene strains with targeted metabolomic analysis of mevalonate pathway intermediates to inform an engineering strategy. Little is known about mevalonate pathway regulation in this species and, particularly as omics data showed that several pathway genes (ERG10, ERG13, ERG20) were upregulated in response to α-bisabolene synthase overexpression (Fig. 2), it was considered prudent to avoid assumptions on rate limiting steps. At the same time, mevalonate pathway engineering strategies in S. cerevisiae and other fungi typically start with overexpression of 3-hydroxy-3-methyl-glutaryl-coenzyme A reductase (HMGR) and it was of interest to see how this would impact pathway metabolite levels [31–33]. Therefore, a new strain (GB2-GPD-tHMGR) to be included in the metabolomic analysis was constructed by the introduction of a truncated HMGR from Cricetulus griseus under control of the GAPDH promoter (PGAPDH-tCgHMGR) into strain GB2. Mevalonate pathway metabolites were analyzed for four strains, R. toruloides WT, BIS3, GB2, and GB2-GPD-tHMGR, which were grown for 40 hours to late exponential phase in SD medium containing 20 g/L glucose (Fig. 5). Intracellular levels of the early and late metabolites of the mevalonate pathway, acetyl-CoA and IPP/DMAPP, respectively, did not vary greatly between the α-bisabolene-producing strains. However, the acetyl-CoA concentration was higher in the α-bisabolene producing strains, compared to WT. Mevalonate accumulated in all strains but was around 75% higher in strain GB2 and over 5-fold higher in strain GB2-GPD-tHMGR, compared to levels in R. toruloides WT. Mevalonate 5-phosphate did not accumulate in R. toruloides WT or BIS3 but was detected in GB2 and GB2-GPD-tHMGR, with a 7-fold higher concentration in the latter strain. The approximate intracellular mevalonate concentration in strain GB2-GPD-tHMGR (27 µM) was around 20-fold higher than acetyl-CoA (1.6 µM) and mevalonate 5-phosphate (1.1 µM).
The data suggests that increasing production of α-bisabolene through high-level expression of AgBIS in strain GB2 promotes flux through the mevalonate pathway leading to an increase in mevalonate and mevalonate 5-phosphate levels. Since overexpression of tCgHMGR amplifies this effect, it seems likely that HMGR is somewhat rate limiting, as is the case in many other eukaryotes [31–33], but that production of mevalonate and mevalonate 5-phosphate emerge as downstream bottlenecks. Thus, HMGR and the two enzymes that process mevalonate and mevalonate 5-phosphate, mevalonate kinase (MK) and phosphomevalonate kinase (PMK), respectively, are possible overexpression targets for increasing mevalonate pathway flux in R. toruloides.
Mevalonate pathway engineering
Orthologs of the central mevalonate genes (encoding HMGR, MK, or PMK) were selected for expression in R. toruloides based on their characteristics or success in engineering terpene production in other organisms. The promoters, PANT, PSKP1 and PDUF were selected for transgene expression based on previous characterization as very strong, strong, and medium strength (annotated as P15, P6, and P21, respectively, in that study) [15]. Prior to the combination of the candidate genes onto single constructs, they were tested individually for impact on α-bisabolene and 1,8-cineole titers by expression in strains GB2 and 307.2, respectively. Two HMGR candidates, from Delftia acidovorans (DaHMGR) and Silicibacter pomeroyi (SpHMGR), were the most impactful on 1,8-cineole titers when expressed in strain 307.2 under control of the two strongest promoters, PANT and PSKP1 (Fig. S2a). SpHMGR was previously employed to engineer the mevalonate pathway in combination with central metabolism to increase terpene yields in S. cerevisiae [4] while DaHMGR was previously employed to engineer flux through a heterologous mevalonate pathway in E. coli [34]. Both are bacterial, NADH-dependent HMGRs that lack the membrane-bound domains of eukaryotic HMGRs. Expression of heterologous MK and PMK genes, however, had a less marked impact on terpene production; perhaps not too surprising as the metabolic analysis suggests that they might only become limiting when HMGR is overexpressed (Fig. 5). Nevertheless, orthologs were down selected for expression in combinatorial constructs based on the available data. MKs, responsible for conversion of mevalonate to mevalonate-5-phosphate, are often subject to feedback inhibition by prenyl phosphates or mevalonate diphosphate but several Archaeal MKs are not subject to this control mechanism [35]. An Archaeal MK, from Methanosaeta concilii (McMK), which was previously found to be kinetically (kcat /Km) more favorable than MKs from other Archaea and S. cerevisiae, had the most favorable impact on α-bisabolene production in R. toruloides when under control of the medium-strong promoter, PSKP1 (Fig. S2b) [35]. Thus, PANT-DaHMGR, PANT-SpHMGR and PSKP1-McMK were selected for expression and the remaining promoter, PDUF, was selected to drive expression of PMKs from S. cerevisiae (ScPMK) and Streptococcus pneumoniae (SpPMK) (Fig. S2c) [36, 37].
Various combinations of the selected HMGR, MK and PMK expression cassettes were assembled onto single plasmid backbones for testing in the α-bisabolene and 1,8-cineole production strains, GB2 and 307.2, respectively. For one configuration of these genes (PANT-SpHMGR, PSKP1-McMK, and PDUF-ScPMK), three versions were built using different methods for codon optimization of the three coding sequences. Transformation of these constructs into the 1,8-cineole producing strain, 307.2, revealed that an in-house codon optimization method (expression cassette optimization, ECO), based on a combination of learnings from literature and analysis of R. toruloides RNAseq and proteomics data, worked poorly while the approaches of using the most preferred codon in R. toruloides for each amino acid (high-CAI, HC) or Genscript’s codon optimization algorithm (Genscript-optimized, GO) produced similar results (Fig. 6a). The GO method is also strongly biased towards selection of the most preferred codons and the output DNA sequences for the HC and GO methods differed by only a few percent. The success of the high-CAI codon optimization methods is surprising but has been borne out by other studies in R. toruloides [38]. Four constructs that contain HMGR, MK and PMK coding sequences, optimized using the HC method, were transformed into strain 307.2 by ATMT and the resulting strains were screened for 1,8-cineole production. Constructs pGEN-485 (PANT-SpHMGR, PSKP1-McMK, PDUF-SpPMK) and pGEN-486 (PANT-DaHMGR, PSKP1-McMK, PDUF-ScPMK) were the most successful of these, resulting in 1,8-cineole titers of over 1 g/L in strains 307.2-485.2 and 307.2-486.11 (Fig. 6b). The fact that similar titers were achieved with coupling of two different HMGR/PMK pairs suggests that pathway optimization may depend as much on achieving a balance between enzymes in the pathway as finding individual successful orthologs.
Medium and process optimization
To test whether terpene production may be limited by the lack of certain nutrients in corn stover hydrolysate, strain GB2 was grown in DMR-EH media supplemented individually with each of the components present in Yeast Nitrogen Base (Difco). Of the supplements, thiamine hydrochloride, pyridoxine hydrochloride, FeSO4, and Na2SO4 positively impacted production of α-bisabolene (Fig. S3).
Using microtiter plates with flower-shaped wells (flower plates) to enhance aeration, the highest titers were achieved in DMR-EH medium supplemented with 5 g/L (NH4)2SO4, 100 mM potassium phosphate (pH 6.0), 400 µg/L thiamine hydrochloride, 400 µg/L pyridoxine hydrochloride, 100 µg/L FeSO4, and 1 mM Na2SO4 (Fig. 7). Production of 1,8-cineole increased in both the 307.2 parent strain and the mevalonate-engineered strain 307.2-486.11, when cultured in the supplemented medium, with the latter producing 1.4 g/L. An α-bisabolene titer of 2.2 g/L was attained in strain GB2 and this improved to 2.6 g/L in the mevalonate-engineered strain GB2-485.1.
Reexamination of mevalonate pathway intermediates
Although the expression of HMGR, MK, and PMK orthologs in strain 307.2 yielded a 4-fold increase in 1,8-cineole titer, transformation of the same constructs into GB2 had a more muted impact on the α-bisabolene titer. Targeted analysis of mevalonate pathway intermediates was employed to probe this disparity in impact on sesquiterpene versus monoterpene titers (Fig. S4). Mevalonate, mevalonate 5-phosphate and mevalonate 5-diphosphate levels were all higher in the α-bisabolene producing parent strain, GB2, compared to the 1,8-cineole producing parent strain, 307.2. Upon overexpression of HMGR, MK, and PMK in these strains, intracellular levels of the same three pathway intermediates increased in both cases. However, higher levels were reached in the α-bisabolene producing strain, GB2-485.1, compared to the 1,8-cineole producing strain, 307.2-484.9. As the increase in mevalonate pathway flux translates into a larger increase in monoterpene than sesquiterpene titers, it is possible that the discrepancy is related to the balance between the C5 precursors IPP and DMAPP. Production of the monoterpene precursor GPP requires a 1:1 ratio of IPP to DMAPP, while the sesquiterpene precursor FPP requires a 2:1 ratio, suggesting IPP/DMAPP isomerase (IDI) is a possible engineering target. Overexpression of IDI has proved effective in increasing titers of sesquiterpenes and carotenoids in other organisms, with data suggesting that it becomes more important as higher IPP/DMAPP ratios are required for production of larger terpenes [39, 40]. It is also possible that sesquiterpene production is limited by FPP synthase which, unlike GPP synthase in the 1,8-cineole producing strains, has not been overexpressed in the α-bisabolene strains. However, since the quantity of α-bisabolene produced in R. toruloides is twice that for 1,8-cineole, it is entirely possible that both terpenes are limited by a common factor, such as mevalonate diphosphate decarboxylase or by a nutrient in the DMR-EH medium. These will be among the factors to be evaluated in future engineering studies.