Metabolic engineering of the pentose phosphate pathway for enhanced limonene production in the cyanobacterium Synechocystis sp. PCC 6803

Isoprenoids are diverse natural compounds, which have various applications as pharmaceuticals, fragrances, and solvents. The low yield of isoprenoids in plants makes them difficult for cost-effective production, and chemical synthesis of complex isoprenoids is impractical. Microbial production of isoprenoids has been considered as a promising approach to increase the yield. In this study, we engineered the model cyanobacterium Synechocystis sp. PCC 6803 for sustainable production of a commercially valuable isoprenoid, limonene. Limonene synthases from the plants Mentha spicata and Citrus limon were expressed in cyanobacteria for limonene production. Production of limonene was two-fold higher with limonene synthase from M. spicata than that from C. limon. To enhance isoprenoid production, computational strain design was conducted by applying the OptForce strain design algorithm on Synechocystis 6803. Based on the metabolic interventions suggested by this algorithm, genes (ribose 5-phosphate isomerase and ribulose 5-phosphate 3-epimerase) in the pentose phosphate pathway were overexpressed, and a geranyl diphosphate synthase from the plant Abies grandis was expressed to optimize the limonene biosynthetic pathway. The optimized strain produced 6.7 mg/L of limonene, a 2.3-fold improvement in productivity. Thus, this study presents a feasible strategy to engineer cyanobacteria for photosynthetic production of isoprenoids.

(gpps) from Abies grandis was expressed to optimize the limonene production pathway. The final recombinant strain led to a 2.3-fold improvement in yield, producing 6.7 mg/L of limonene in 7 days. The metabolic engineering strategies used in this study demonstrate the feasibility of increasing limonene production in Synechocystis 6803 and can be applied to phototrophic production of other high-value isoprenoids.

Results
Engineering Synechocystis 6803 for production of limonene. Limonene is a C10 cyclic isoprenoid converted from geranyl diphosphate (GPP). Due to the complex nature of carbocation rearrangement from GPP to limonene, limonene synthase produces not only limonene but also other monoterpenes such as bicyclic α-pinene and acyclic mycene 20 . To avoid the production of other unwanted byproducts, we chose limonene synthases which have the highest specificity for limonene production. Based on previous studies, limonene synthase from Citrus limon and Mentha spicata produce limonene of high purity. Expression of each of these limonene synthases in E. coli showed that the former produces 99% pure (R)-limonene 21 , and the latter generates 94% of (S)-limonene 22 . The coding sequences of lims were codon optimized for Synechocystis 6803, and the plastid targeting sequences were removed 23,24 . The truncated enzyme is known to have better catalytic activity than the native protein 25 . Genes were cloned into a pCC5.2 neutral-site-targeting plasmid and driven by the trc1O promoter for higher level expression of lims ( Fig. 2A). Expression of an enhanced yellow fluorescent protein (EYFP) from the pCC5.2 endogenous plasmid is 8 to 14 times higher than that on the chromosome 26 .
When the lims was cloned into a suicide plasmid and transformed into E. coli, we found that the gene accumulated random mutations in the E. coli host, leading to changes in amino acid residues or truncated proteins. This was presumably because the lims product is toxic to E. coli cells. To introduce a lims without mutations into Synechocystis 6803, we circumvented the E. coli cloning step by first cloning the lims into the suicide plasmid via Gibson assembly, and used the assembled product as template for PCR to amplify the lims cassette flanked by upstream and downstream homologous sequences of the neutral site in pCC5.2 26 . Subsequently, the PCR product was directly used for natural transformation into Synechocystis 6803. The lims was introduced into Synechocystis 6803 genome via double homologous recombination ( Fig. 2A). DNA sequencing results showed that the lims has no mutation in Synechocystis 6803 (data not shown). Mutants were fully segregated after re-streaking the cells several times on BG-11 plates with antibiotics.
Limonene production by engineered Synechocystis 6803 was tested by incubating cultures for 7 days. Because of the volatility of limonene, a dodecane overlay was applied on cultures to collect limonene in the organic layer. It has been reported that over 99% of limonene escapes from the cyanobacterial cultures 14 , and covering an organic overlay on cultures had little influence on growth in cyanobacteria 23 . The limonene yield by the strain expressing lims from M. spicata was two-fold higher than that by the strain expressing lims from C. limon (Fig. 2B). These results suggest that the limonene synthase from M. spicata exhibited better catalytic activity in Synechocystis 6803, and hence, the strain was used for further engineering.
Computational modeling. The iSyn731 metabolic model of Synechocystis 6803 18 was used to perform the computational strain designs using the OptForce algorithm 17 for overproduction of limonene. Based on the current understanding as reported in literature 16,27 , a connection between Calvin Benson Cycle (CBC)/PP pathway and MEP pathway (Fig. 3) was included in the iSyn731 model. By superimposing the photoautotrophic flux measurements 19 of 31 reactions of central carbon metabolism including the CBC and PP pathways of Synechocystis 6803 onto the iSyn731 model, the phenotypic space of the base strain was defined. All simulations were performed for a basis of 100 millimoles of CO 2 plus H 2 CO 3 uptake and unlimited photon supply 19 . The uptake fluxes for the remaining metabolites present in the BG11 medium was set to -1,000 and the non-growth associated ATP maintenance was set at 8.39 mmole/gDW-h. In addition, the biomass flux was fixed at the optimal value subject to the experimental flux measurements 19 . The upper bound of the fluxes of the remaining reactions was set to 1,000 mmole/gDW-h, whereas the lower bound was set to zero and -1,000 mmole/gDW-h for irreversible and reversible reactions, respectively.
Similarly, the limonene overproducing phenotype was obtained by maximizing and minimizing each flux of the metabolic model iteratively subject to the network stoichiometry, uptake and medium conditions, regulatory constraints, and overproduction target. In this work, a minimum production yield of 85% of the theoretical maximum of limonene (i.e., 15.3 mmole/gDW-h) was set as the overproduction target, while the biomass flux was constrained to be at least 10% of its theoretical maximum (i.e., 0.021 h −1 ) with the basis of 100 millimoles of carbon fixed (i.e., CO 2 plus H 2 CO 3 ). The remaining parameter values including medium conditions and regulatory constraints were the same as those in the wild-type. By contrasting the maximal range of flux variability between the wild-type strain and the over-producing strain to meet the pre-specified yield of limonene, OptForce was used to identify the minimal set of genetic interventions (i.e., deletions and up-/down-regulations). In order to first explore non-intuitive interventions, reactions from the MEP and isoprenoid biosynthesis pathways were not considered as the candidates for any form of intervention. Integer cuts were used to identify alternative optimal solutions (i.e., alternative genetic intervention choices) to achieve the minimum production yield of limonene as specified earlier. The termination criterion for the OptForce procedure was set as either meeting a production yield of at least 85% of the theoretical maximum for limonene or exceeding the maximum allowable number of reaction interventions (i.e., three). Note that the OptForce procedure works at the reaction level, which is why the set of genetic manipulations can subsequently be identified by using gene-protein-reaction (GPR) associations from the iSyn731 model. Thus, the OptForce procedure identified up-regulation of rpi and rpe as the best possible solution, which can lead up to limonene yield at 89% of its theoretical maximum (i.e., 16.02 mmole/gDW-h). By up-regulating these two genes, OptForce suggested to force more flux from the CBC/PP pathway toward MEP pathway that can ultimately increase the production yield of limonene (Fig. 3). Once the set of non-intuitive interventions was obtained, as a next step, it was logical to explore if their combination with any of the intuitive one(s) from the MEP and isoprenoid biosynthesis pathways could further improve the limonene production yield that was otherwise not possible to achieve individually (i.e., by the non-intuitive candidates or by the intuitive ones). With a target of a minimum production yield of 90% of the theoretical maximum of limonene, the OptForce procedure identified the up-regulation of gpps, rpe, and rpi that could lead the limonene production yield to 16.56 mmole/gDW.h (i.e., 92% of its theoretical maximum). Thus, the proposed interventions combined the amplification (i.e., push) of flux from the CBB/PP pathway to MEP pathway with a similar increase (i.e., pull) in the flux of the limonene synthesis. As reported in the literature 28 , this kind of push-and-pull strategy can achieve the desired level of production yield with minimal effects caused by feedback inhibition.
Genetic interventions of the PP pathway to improve limonene production. Based on the prediction of the OptForce procedure, up-regulation of rpi and rpe genes in the PP pathway increases the flux toward limonene production. To test this hypothesis, the rpi and rpe genes driven by the Synechocystis 6803 native rbcL promoter were expressed on a replicating plasmid in the limonene-producing strain, resulting in 1.3-fold increase in limonene yield (3.7 mg/L) after 7 days of cultivation (Fig. 4). Furthermore, we introduced a gene encoding a specific GPP synthase (GPPS) to optimize the limonene biosynthetic pathway. In Synechocystis 6803, formation of GPP is catalyzed by a farnesyl diphosphate (FPP) synthase, CrtE. It performs consecutive condensation of IPP with DMAPP, and only synthesizes GPP as an intermediate 29 . Although the PP pathway was engineered to stimulate the limonene yield, it is possible that the native isoprenoids pathway in Synechocystis 6803 provides insufficient GPP for limonene production since the flux is diverted toward FPP formation for pigment synthesis. In addition, OptForce also predicted an increase (i.e., from 89% to 92% of maximum theoretical limonene yield) when up-regulation of rpe and rpi was combined with the up-regulation of gpps. It was reported that the GPPS 2 from Abies grandis specifically produces GPP 30 . Expressing this specific gpps with lims, the limonene yield increased 1.4-fold (4.1 mg/L) (Fig. 4). Finally, coexpression of rpi, rpe, gpps and lims resulted in a remarkable (2.3-fold) enhancement in productivity (6.7 mg/L) (Fig. 4).
Pigment content in engineered Synechocystis 6803. Carotenoids and the phytol tail of chlorophyll, photosynthetic pigments, are derived from geranylgeranyl diphosphate (GGPP), a C20-intermediate for isoprenoid synthesis. Hence, production of limonene is expected to divert carbon flux away from pigment synthesis. To investigate the effect of limonene production on pigment content in engineered Synechocystis 6803, we extracted and quantified the chlorophyll and carotenoid contents. The chlorophyll content decreased over 30% in the gpps expression strains, whereas carotenoid levels were similar among the limonene-producing strains (Fig. 5). These results indicate that the specific GPPS diverts the carbon flux away from pigment synthesis.

Discussion
In this study, we combined metabolic engineering with model-driven strain design strategies to engineer Synechocystis 6803 for enhanced limonene production. To generate limonene-producing Synechocystis 6803, we first constructed a suicide plasmid 26 to engineer the lims into the neutral site on the pCC5.2 endogenous plasmid via double homologous recombination. This is the first time that the endogenous plasmid of Synechocystis 6803 has been used for enhanced production for the purpose of metabolic engineering. Expression of a gene on the pCC5.2 plasmid leads to higher expression level than that on the chromosome as well as the RSF1010 were overexpressed in Synechocystis 6803 to divert carbon flux toward limonene production. The Abies grandis GPPS 2 that specifically produce GPP was expressed to ensure sufficient GPP for limonene production. Ms, Mentha spicata; Ag, Abies grandis. Results were mean ± SD of three biological replicates.
self-replicating plasmid 26 . Furthermore, during the stationary phase of cell growth, the copy numbers of the endogenous plasmids (pCA2.4, pCB2.4, pCC5.2) in Synechocystis 6803 are 3 to 7 per chromosome 31 . Using the endogenous plasmid to express the lims gene driven by the constitutive promoter trc1O allows high expression level at the stationary phase, decoupling growth and production, and thus leading to higher levels of production of limonene.
The higher yield with limonene synthase from M. spicata than that from C. limon may be due to the difference in enzyme kinetics of LIMS. Unfortunately, the kinetic parameters (both K m and k cat ) are only available for the enzyme from M. spicata 25 . In addition, it may be attributed to different protein expression levels. Although the same promoter was used to control the lims from two plant species, protein expression may vary because of different mRNA sequences and codon usage. To date, the highest reported limonene productivity in cyanobacteria was achieved by engineered Synechococcus sp. PCC 7002 23 . In their study, only a lims from M. spicata was overexpressed, and the yield was over 4 mg/L in 4 days 23 . Our results also suggested that the LIMS from M. spicata performed better in limonene production (Fig. 2B). The doubling time of Synechococcus 7002 is shorter than Synechocystis 6803 32 . Thus, the higher limonene yield from Synechococcus 7002 may be due to its faster growth rate. A recent study engineered Synechococcus elongatus PCC 7942 to produce limonene, achieving a 100-fold improvement in productivity 33 . However, it should be noted that such significant increase is due to the low productivity of the original strain, which produced merely 8.5 μg/L/OD/d of limonene. The best producing strain in this study, with a lims (M. spicata) controlled by the pea plant psbA promoter, produced 2.5 mg/L limonene in 5 days 33 .
Previously, researchers have engineered Synechocystis 6803 for limonene production by overexpressing genes involved in the bottleneck steps of the MEP pathway 14 . It is known that enzymes 1-deoxy-D-xylulose-5-phosphate synthase (DXS) and isopentenyl diphosphate isomerase (IDI) catalyze the rate-limiting reactions in the MEP pathway 34,35 . With the introduction of an additional copy of endogenous dxs, idi, and gpps genes, the engineered Synechocystis 6803 produced 1.4-fold higher yield than that of the parent strain 14 . However, such improvement was less effective than the strategy used in the current study. As mentioned in the Results section, the endogenous gpps gene may not be suitable for enhancing the production of limonene. In addition, the MEP pathway is highly regulated at genetic and metabolic levels 36 . Expressing endogenous genes in the MEP pathway may be subject to native regulations, presenting a less effective engineering approach.
Instead of manipulating the MEP pathway, we took a systematic model-driven metabolic engineering approach for finding genetic interventions in order to increase the limonene production yield. As explained in the Materials and Methods section, the OptForce procedure finds the minimal interventions to reach a desired production target. To this end, we employed OptForce on our previously developed genome-scale model iSyn731 in order to 'push' more flux to MEP pathway and also to create better 'pull' for limonene synthesis (Fig. 3). From this in silico analysis, by up-regulating rpe and rpi, the metabolite pool of X5 P was found to be increased that, eventually, led to increased flux through the connection between the CBC/PP pathway and the MEP pathway. In addition, up-regulation of gpps created an improved 'pull' for limonene synthesis. Thus, the combination of this push-and-pull mechanism was proposed to be the best strategy to improve limonene yield by circumventing additional regulations (e.g. feedback inhibition). Interestingly, the same rationale could be applied to engineer cyanobacteria to produce other isoprenoid compounds.
Expression of the specific gpps modestly increased the limonene titer (Fig. 4), whereas the cellular chlorophyll content was greatly influenced (Fig. 5). Synthesis of limonene and the phytol tail of chlorophyll requires the same precursors, IPP and DMAPP. Table 1 compares the changes in chlorophyll and limonene contents between the strain with lims only and the two gpps-expressing strains. Compared to the lims-expressing strain, additional expression of gpps resulted in similar level of decrease in chlorophyll content in both strains. However, expression of rpi and rpe genes further led Our results showed that overexpressing the genes in the PP pathway led to improved limonene production, suggesting an unidentified connection between the PP pathway and isoprenoids biosynthesis (Fig. 4). Our observation is consistent with previous in vitro study using Synechocystis 6803 cell lysate 16 . However, the connection between the PP pathway and isoprenoids biosynthesis remains to be elucidated. It was first shown that in vitro isoprenoid production increased significantly by providing substrates in the PP pathway 16 , while a recent study showed that increased production of isoprenoids by PP pathway substrates does not occur through the MEP pathway 37 . By removing the terminal enzyme of the MEP pathway in Synechocystis 6803 cell lysate, isoprenoid synthesis still increased by substrates in the PP pathway 37 . Taken together, it is still unclear how the PP pathway and isoprenoid production are connected in Synechocystis 6803. While our results made a strong argument for this connection, further investigation needs to be conducted to explore the details in terms of chemical conversions and genes/enzymes associated. From the modeling context, these details sometimes do not make much of a difference if they only involve aggregating linear reaction steps.

Conclusions
In this study, we engineered the model cyanobacterium Synechocystis 6803 to produce the isoprenoid, limonene. We applied computational strain design by using the OptForce procedure to identify minimal genetic interventions for improving limonene yield. Based on the prediction, the rpi and rpe genes in the PP pathway were overexpressed, and a specific gpps was introduced to optimize the limonene biosynthetic pathway. The final engineered strain produced 6.7 mg/L of limonene, which is a 2.3-fold improvement in productivity. The approach that we demonstrated can be applied to engineer cyanobacteria to produce other valuable isoprenoids.  26 . The constructed plasmids were directly used as templates for PCR to amplify a fragment which contains the lims and a kanamycin resistance cassette flanking by upstream and downstream homologous sequences of the NSP1 (Fig. 2A). The PCR product was then purified by DNA electrophoresis, and the linear DNA was transformed into Synechocystis 6803. The rpi, rpe, and gpps genes were cloned into a broad-host-range plasmid RSF1010 harboring a spectinomycin resistance cassette 38 . All the cloning works were done by Gibson isothermal DNA assembly method 39 .

Strains construction and transformation. The lims expression cassette was transformed into
Synechocystis 6803 through homologous recombination. Cells at mid-log phase (OD 730 of 0.4 to 0.6) were incubated with 600 ng of linear DNA overnight at 30 °C in the dark. Cells were then grown on BG-11 plates supplemented with 10 μg/mL of kanamycin for selection of transformants. Colonies were patched on BG-11 plates with 20 μg/mL of kanamycin for segregation. PCR was used to verify strain segregation. For the construction of rpi, rpe, and gpps expressing strains, self-replicating plasmids (600 ng per transformation) were transformed into the strain expressing lims. Transformants were selected by BG-11 plates with 2 μg/mL of spectinomycin and 5 μg/mL of kanamycin.
Limonene production by engineered cyanobacteria. Strains were inoculated in BG-11 medium with kanamycin (10 μg/mL) and spectinomycin (4 μg/mL) to mid-log phase at 30 °C with continuous white light (50 μmoles photons m −2 s −1 ). Cells were collected by centrifugation at 7,000 x g, and washed by BG-11 medium to remove antibiotics. To test limonene production, the initial OD 730 was adjusted to 0. 34   Scientific). The oven temperature program initiated at 60 °C, and increased at 12 °C/min to 300 °C. Limonene was quantified using a (R)-limonene standard.

Identification of engineering interventions via OptForce.
We applied the OptForce algorithm 17 on the genome-scale Synechocystis 6803 model iSyn731 18 . In order to characterize the wild-type phenotype, we utilized 13 C MFA flux estimations 19 under photosynthetic condition. Below is the step-by-step procedure that we followed: Step 1: Identify the maximum biomass and limonene yields under photosynthetic condition.
Maximize v biomass or v ls Subject to Step 3: Characterize the limonene over-producing phenotype Maximize/Minimizev j ∀ j ∈ 1, ……,m Subject to

ls ls max
Step 4: Identify the MUST sets In this step, fluxes ranging from step 2 and step 3 were compared to identify three different sets: reactions to be up-regulated (MUST U ), down-regulated (MUST L ), and deleted (MUST X ).
Step ∑ ≤ of direct manipulations k # (8) Here, S ij is the stoichiometric coefficient of metabolite i in reaction j and v j is the flux value of reaction j.
Parameters v j,min and v j,max denote the minimum and maximum allowable fluxes for reaction j, respectively. V biomass and v ls represent biomass and limonene synthesis reactions under photosynthetic conditions, whereas v max biomass and v max ls represent the maximum theoretical yields of biomass and limonene under photosynthetic conditions. The minimal levels of biomass and the minimal target yield of limonene were set to be 10% of maximum biomass and 85% or 90% of maximum limonene yield, respectively. Finally, k represents the maximum number of interventions allowed.
Pigment content analysis. Cell cultures (1 mL) were collected by centrifugation at 16,000 x g for 7 min, and the supernatants were removed. To extract pigments in Synechocystis, pre-cooled methanol (1 mL) was added to the pellets, and mixed thoroughly by pipetting and vortexing. Samples were incubated at 4 °C for 20 mins, and centrifuged at 16,000 x g for 7 min. The supernatants were removed for a spectroph-otometer analysis to quantify the concentrations of carotenoids and chlorophyll. The following equations were used to calculate the pigment content: μ = . × − . × A A chlorophyll ( g/mL) (16 29 ) (8 54