Introduction of a Synthetic CO2-fixing Photorespiratory Bypass into a Cyanobacterium

Background: Photorespiration limits carbon fixation. Results: Heterologous expression and functional activity of six enzymes from the 3-hydroxypropionate bi-cycle are demonstrated in cyanobacteria. Conclusion: A synthetic CO2-fixing photorespiratory bypass can be introduced into cyanobacteria. Significance: The results lay the foundation for expressing an alternative CO2 fixation pathway in cyanobacteria, algae, and plants. Global photosynthetic productivity is limited by the enzymatic assimilation of CO2 into organic carbon compounds. Ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), the carboxylating enzyme of the Calvin-Benson cycle, poorly discriminates between CO2 and O2, leading to photorespiration and the loss of fixed carbon and nitrogen. With the advent of synthetic biology, it is now feasible to design, synthesize, and introduce biochemical pathways in vivo. We engineered a synthetic photorespiratory bypass based on the 3-hydroxypropionate bi-cycle into the model cyanobacterium, Synechococcus elongatus sp. PCC 7942. The heterologously expressed cycle is designed to function as both a photorespiratory bypass and an additional CO2-fixing pathway, supplementing the Calvin-Benson cycle. We demonstrate the function of all six introduced enzymes and identify bottlenecks to be targeted in subsequent bioengineering. These results have implications for efforts to improve photosynthesis and for the “green” production of high value products of biotechnological interest.

Oxygenic photosynthesis is the primary source of nearly all biological energy. In this process, light is converted into chemical energy, which is used to fix CO 2 in the CB 5 cycle through the enzyme RuBisCO. The carboxylase activity of RuBisCO results in the addition of one molecule of CO 2 to one molecule of ribulose-1,5-bisphosphate to create two molecules of 3-phosphoglycerate, thus fixing inorganic CO 2 into triose phosphates. However, the competing oxygenase activity of RuBisCO results in the loss of fixed carbon through a process termed photorespiration. One of the "holy grails" of photosynthesis research has been to engineer RuBisCO to improve CO 2 fixation and reduce photorespiration; however, these attempts have been met with limited success. It has been shown that biochemical constraints as well as abiotic factors are crucial considerations in addressing the protein engineering of RuBisCO (1,2). Given this complexity, a more promising approach may be to accept the inherent "flaws" of RuBisCO and improve net photosynthetic rates through engineered photorespiratory bypasses.
Photorespiration produces the toxic intermediate 2-phosphoglycolate, which is recycled through the photorespiratory C 2 cycle (see Fig. 1A). This pathway is costly, requiring ATP and reducing equivalents in an elaborate reaction sequence involving more than a dozen enzymes and transporters. Furthermore, the reaction catalyzed by glycine decarboxylase, converting two glycine molecules into one serine in the C 2 cycle, releases both NH 3 and CO 2 , resulting in a net loss of carbon and nitrogen. To date, only two studies have attempted to experimentally decrease the negative impacts of the photorespiratory C 2 cycle by expression of alternative glycolate metabolic pathways. Kebeish et al. (3) attempted to bypass most of the C 2 cycle by introducing the glycolate catabolic pathway from Escherichia coli (4) into Arabidopsis thaliana chloroplasts. This pathway circumvents the loss of nitrogen, but the glyoxylate carboligase reaction results in the release of one CO 2 per two glyoxylate molecules (see Fig. 1A). Although increased biomass was reported, interestingly, transformants expressing only the first enzyme of that pathway, glycolate dehydrogenase, showed similar results, rendering the approach controversial. In a second study, Maier et al. (5) introduced a glycolate oxidation cycle into Arabidopsis chloroplasts; however, this pathway results in the release of even more CO 2 than the heterologously expressed glycolate catabolism pathway. In both cases, CO 2 release occurs in the chloroplast, where it can potentially be refixed by RuBisCO. The challenges associated with designing experimental approaches to mitigate the losses associated with photorespiration are likewise underscored by results from sys-□ S This article contains supplemental Experimental Procedures and tems-level genome-scale metabolic modeling that suggests photorespiration is essential for optimal photosynthesis (6). Introduction of additional, synthetic CO 2 fixation pathways provides an approach to increasing photosynthesis that circumvents the complexities associated with manipulating the C 2 cycle (7). Of the six known CO 2 fixation cycles in nature, only the 3-hydroxypropionate (3OHP) bi-cycle is completely oxygen-insensitive (8,9), a key consideration when engineering pathways into oxygenic photoautotrophs. The 3OHP bi-cycle from the thermophilic anoxygenic phototroph Chloroflexus aurantiacus offers an attractive starting point for engineering efforts (10) because all of the necessary enzymes have been characterized (9). In this bi-cyclic pathway, bicarbonate is fixed by biotin-dependent acetyl-CoA carboxylase and propionyl-CoA carboxylase. The primary CO 2 fixation product resulting from the first cycle is glyoxylate, which is then fed into the second cycle, in which another bicarbonate is fixed and pyruvate is generated as the final product (9).
We designed a CO 2 -fixing synthetic photorespiratory bypass based on the 3OHP bi-cycle (see Fig. 1B). To experimentally validate the design, we introduced the requisite six genes, encoded in assembled DNA constructs spanning more than 16 kbp, to reassimilate the photorespiratory byproduct glyoxylate in the cyanobacterium S. elongatus PCC7942. We demonstrate activity for all of the gene products and identify metabolic bottlenecks to be addressed. In comparison with the conventional C 2 cycle, the synthetic bypass not only prevents the loss of NH 3 but also results in a net gain in carbon fixation rather than a net loss.

EXPERIMENTAL PROCEDURES
Cloning, Strains, and Growth Conditions-All constructs were cloned using the BglBrick assembly format (11) in E. coli and subsequently cloned into various neutral site destination vectors, which allow for genomic integration into the S. elongatus genome by previously described transformation protocols (12,13). Plasmids and strains that were generated and used are summarized in Table 1. All used primers are listed in supplemental Table S1. S. elongatus strains were maintained in BG-11 medium under appropriate selection with constant light at 30 or 37°C.
Cloning, Heterologous Expression of Recombinant Enzymes in E. coli, and Purification-The cloning, expression, and purification of the mesaconyl-C1-CoA hydratase (MCH) and mesaconyl-CoA C1:C4 CoA transferase (MCT) from C. aurantiacus were performed as described previously (9). Cloning, expression, and purification of the malyl-CoA lyase (MCL) from C. aurantiacus were described previously (14). For the expression and purification of the MCT from "Candidatus Accumulibacter phosphatis," see supplemental Experimental Procedures.
Enzyme Assays-One unit corresponds to an enzyme activity of 1 mol min Ϫ1 (mg protein) Ϫ1 . Unless stated otherwise, all assays were carried out at 30 or 37°C, depending on the growth conditions of the transformant cultures. Cells were harvested during exponential phase by centrifugation at 6000 ϫ g. Cell pellets were resuspended in a 2-fold volume of 200 mM MOPS/ KOH buffer (pH 7.5). The cell suspensions were sonicated, and the cell lysates were centrifuged at 20,000 ϫ g and 4°C for 30 min. Supernatants were either used directly for enzyme assays or stored at Ϫ80°C.
The malonyl-CoA reductase was measured by a slightly modified previously described assay (15) monitoring the malonyl-CoA-dependent oxidation of NADPH at a wavelength of 365 nm (⑀ 365 ϭ 3,400 M Ϫ1 cm Ϫ1 ). The assay mixture (400 l) contained 200 mM MOPS/KOH buffer (pH 7.5), 5 mM MgCl 2 , 0.4 mM NADPH, and cell extract. The reaction was started by the addition of 1 mM malonyl-CoA. Notably, two NADPH molecules are oxidized per one malonyl-CoA that is reduced to 3OHP.
Propionyl-CoA synthase activity was monitored either by a previously described and slightly modified spectrophotometric assay (16) or in an HPLC-based assay, as described in the two steps below. (i) The reaction mixture (400 l) for the photometric assay contained 200 mM MOPS/KOH buffer (pH 7.5), 0.4 mM NADPH, 100 mM KCl, 2 mM ATP, 0.5 mM CoA, and cell extract. The reaction was started by the addition of 5 mM 3OHP. (ii) The same reaction mixture was used for the HPLCbased assay only with 1 mM instead of 0.4 mM NADPH. Samples of 100 l were withdrawn after different time points and stopped by the addition of 10 l of 90% formic acid. The samples were kept on ice before precipitated protein was removed by centrifugation at 16,000 ϫ g. The supernatants were subjected to HPLC analysis to confirm propionyl-CoA formation.
The concerted function of the MCL, MCH, MCT, and mesaconyl-C4-CoA hydratase (MEH) was demonstrated in an HPLC-based assay. The reaction mixture (400 l) contained 200 mM MOPS/KOH buffer (pH 7.5), 5 mM MgCl 2 , 0.5 mM propionyl-CoA, and cell extract. The reaction was started by the addition of 5 mM glyoxylate and carried out at 37°C. Samples of 100 l were withdrawn after different time points, treated as described above, and subjected to HPLC analysis to confirm the formation of acetyl-CoA or other CoA-thioester intermediates. MCT activity was determined in an HPLC-based assay. The reaction mixture (0.4 ml) containing 200 mM MOPS/KOH (pH 7.5), 5 mM MgCl 2 , 2 mM propionyl-CoA, 10 mM glyoxylate, 3 units (formation of ␤-methylmalyl-CoA) of recombinant MCL, and 25 units of recombinant MCH was preincubated for 10 min at 37°C to form mesaconyl-C1-CoA as substrate for the CoA transferase. MCT or ApMCT was added to start the reaction. Samples of 100 l were withdrawn prior to and after MCT addition. Reactions were stopped by the addition of 5 l of 90% formic acid. Precipitated protein was removed by centrifugation, and the supernatants were analyzed for CoA thioesters by reversed-phase HPLC. To determine the K m value for mesaconyl-C1-CoA, a similar preincubation (0.2 ml) with 3 units of MCL and 25 units of MCH was performed for 15 min at 55°C with 10 mM propionyl-CoA and 30 mM glyoxylate. The reaction was stopped by the addition of 2 l of 90% formic acid. Precipitated protein was removed by centrifugation, and the enzymatically synthesized mesaconyl-C1-CoA was used in the MCT assay at varying final concentrations as determined by HPLC and absorption at 260 nm.
Other Methods-See supplemental Experimental Procedures.

RESULTS
To implement the proposed cycle shown in Fig. 1B, we first tested the constitutive expression and activity of the first four Chloroflexus enzymes required, beginning where glyoxylate enters the cycle (i.e. MCL, MCH, MCT, and MEH). The reactions catalyzed by this sequence of enzymes result in the formation of acetyl-CoA and pyruvate (Fig. 1B) from propionyl-CoA and glyoxylate. Dicistronic operons were assembled to express mcl with mch and mct together with meh ( Fig. 2A). Both dicistrons were driven by the previously characterized psbA1 promoter (17). The cassette expressing all four genes (referred to as PMS4032) was integrated into the S. elongatus genome at neutral site 1 (NS1) (18). The resulting transformants were assayed for activity of all four enzymes. Soluble cell extracts from the transformants were incubated with propionyl-CoA and glyoxylate, and the expected disproportionation into acetyl-CoA and pyruvate was confirmed, indicating activity of all four enzymes; the rate of catalysis, however, was low.
The intermediates involved in the last two steps needed to complete the pathway in S. elongatus, MCR and PCS, are toxic to cells. Accumulation of 3OHP, the product of MCR, can lead to organic acid toxicity (19); propionyl-CoA, the product of PCS, inhibits both pyruvate dehydrogenase and citrate synthase (20). The potential toxicity in conjunction with the difficulty of successfully reconstituting multistep metabolic pathways (21) presented major challenges. Moreover, both MCR and PCS are large multidomain enzymes, potentially presenting difficulty in proper folding and expression. For these reasons, mcr and pcs were driven by the IPTG-inducible promoter, pTrc ( Fig. 2A). The mcr gene was assembled upstream of the PMS4032 cassette to generate PMS4570 and integrated into NS1 without an additional terminator downstream of mcr, whereas pcs alone was inserted into neutral site 3 (NS3) (13). Double transformants (PCS/PMS4570) containing all six genes integrated into both NS1 and NS3 were generated and tested for expression and enzyme activity in response to varying IPTG concentra-tions. MCR activity was confirmed spectrophotometrically by measuring the malonyl-CoA-dependent oxidation of NADPH. PCS activity was measured by spectrophotometrically by monitoring the 3OHP-dependent oxidation of NADPH and by following the formation of propionyl-CoA by HPLC. Both MCR and PCS were copiously expressed (Fig. 2B) and found to be active in the cell extracts (Fig. 2, C and D, respectively). An IPTG concentration of 20 M yielded the highest enzyme activities; further increases in IPTG concentrations did not result in higher activities. Furthermore, the addition of the pTrc promoter upstream of mcr also increased the expression of the two downstream genes mcl and mch, as deduced from the results of enzyme activity assays. However, the conversion of propionyl-CoA and glyoxylate to acetyl-CoA and pyruvate was stalled at the mesaconyl-C1-CoA intermediate (Fig. 1). The addition of purified recombinant MCT to the assay resulted in immediate conversion of mesaconyl-C1-CoA to acetyl-CoA and pyruvate, indicating that mct expression was the bottleneck, whereas the meh gene downstream of mct was adequately expressed (Fig.  2E).
To relieve the bottleneck, a second copy of mct, driven by the IPTG-inducible pTrc promoter, was added upstream of mcr. We tested two strategies for introducing the additional mct gene: 1) adding a duplicate Chloroflexus mct to generate PMS4749 and 2) introducing a synthetic mct homolog (referred to as ApMCT) from the ␤-proteobacterium Candidatus Accumulibacter phosphatis (22,23) resulting in PMS4591 ( Fig. 2A). ApMCT is the most closely related mesophilic homolog to the Chloroflexus MCT (24). We confirmed its function by expressing in E. coli and purifying a recombinant His 10 -tagged version of the ApMCT, which catalyzed the expected intramolecular CoA transfer reaction within mesaconyl-CoA with a specific activity of 37 Ϯ 6 mol min Ϫ1 (mg protein) Ϫ1 at 37°C, corresponding to a turnover number (k cat ) of 58 s Ϫ1 per dimer. Its apparent K m value for mesaconyl-C1-CoA was determined to be 1.49 Ϯ 0.22 mM, which was surprisingly high. In comparison the K m value of the Chloroflexus MCT is only 0.24 mM (9). Moreover, the specific activity of the Chloroflexus MCT is much higher, 520 mol min Ϫ1 (mg protein) Ϫ1 at 55°C (9), even assuming the reaction would be halved each 10°C the temperature is decreased. Therefore, the overall efficiency of the ApMCT would be much lower.
Nevertheless, the double transformants encoding either a second mct gene from Chloroflexus or the Accumulibacter gene (PCS/PMS4749 or PCS/PMS4591, respectively) were generated ( Fig. 2A) and assayed for all enzyme activities. In both cases, the MCT activity was substantially increased, and the activity of all six enzymes engineered into S. elongatus was confirmed ( Fig. 2, C, D, E, and G). However, introduction of the additional mct gene upstream of mcr apparently led to a decrease in MCR, MCL, and MCH expression (Fig. 2B) and activity (Fig. 2F).
To estimate whether the resulting enzyme activities were high enough to allow the functioning of the synthetic photorespiratory bypass, we calculated the carbon assimilation rate of a S. elongatus wild-type culture using the equation dS/dt ϭ (/Y) ϫ X (25), which correlates the specific substrate consumption (dS) over time (dt) with the specific growth rate ().
The established growth yield (Y) corresponds to a bacterial cell dry mass of 1 g formed per 0.5 g of carbon fixed (ϳ50% of bacterial cell dry mass is carbon). Although X usually refers to the concentration of living cells, in this case it is used to account for the amount of total protein per 1 g of cell dry mass (in bacteria ϳ50% of cell dry mass is protein). We assumed a typical doubling time of 8 h for a wild-type culture under laboratory conditions with ambient CO 2 , which corresponds to a of 0.087 h Ϫ1 . This would require a net carbon assimilation rate of 121 nmol min Ϫ1 (mg protein) Ϫ1 . Taking into account an estimated loss of up to 25% of the fixed carbon due to photorespiration (26) results in 80 nmol min Ϫ1 (mg protein) Ϫ1 for the oxygenase activity of RuBisCO and the production of glycolate. To efficiently reassimilate glycolate in the synthetic bypass, the minimal specific activities of the involved enzymes need to be at least as high as the rate of glycolate generation. Based on that estimate, all but one of the introduced enzymes were well above the required threshold (Fig. 2G). Only the specific activity of PCS (ϳ25 nmol min Ϫ1 (mg protein) Ϫ1 ) in the transformant cell extracts was lower than the calculated threshold, despite very high expression (Fig. 2B).

DISCUSSION
This study is, to our knowledge, the first successful effort to express a synthetic CO 2 -fixing photorespiratory bypass in a photoautotrophic organism, the cyanobacterium S. elongatus PCC7942. Unlike previous studies, our pathway differs by directly avoiding the net loss of nitrogen and carbon in the photorespiratory C 2 cycle, which actually results in a net gain in carbon fixation through the enzyme acetyl-CoA carboxylase (ACCase).
The unique feature of our pathway is the additional carbon fixation, which must be accounted for in energy balance comparisons of other proposed photorespiratory bypasses. Therefore, we have assumed the stoichiometrically correct values for the formation of two glycolate molecules per CO 2 released in the C 2 cycle (Table 2). Thus, to reassimilate two glycolate molecules, our cyclic bypass requires 6 ATP equivalents and 4 NAD(P)H, while fixing two additional molecules of bicarbonate, the form of inorganic carbon concentrated in the cytoplasm of cyanobacteria, and circumventing the loss of NH 3 . Note that if pyruvate, which derives from our bypass, is to be used for replenishing the CB cycle, two more ATP equivalents are required per pyruvate molecule in gluconeogenesis by pyruvate phosphate dikinase because it is AMP-forming. Nevertheless, the synthetic bypass compares favorably over the canonical photorespiratory C 2 cycle of cyanobacteria in terms of energy demand; the combined function of the C 2 cycle and CB cycle requires 11 ATP equivalents, 4 NAD(P)H, and 2 reduced ferredoxins to first refix the lost CO 2 and NH 3 , as well as additionally fix two more CO 2 molecules to arrive at the same level of net carbon fixation as the synthetic bypass (see Fig. 1 and Table 2 for comparison of photorespiratory pathways).
Although the vast majority of metabolic engineering efforts focus on introducing linear pathways for the anabolic production of molecules of interest, our approach introduces a self-sustaining metabolic cycle that fixes CO 2 when glycolate/glyoxylate is available. We demonstrate that concomitant expression and activity of all six enzymes necessary to reconstitute the synthetic bypass can be achieved. This required heterologous expression of ϳ16 kbp of DNA and functional assembly of six multimeric enzymes ranging in molecular mass from 62 to 600 kDa.
However, an obvious physiological phenotype was not observed during growth experiments. The transformants exhibited only slight delay in growth when liquid cultures in air were inoculated from agar plates, but they reached the same doubling times and optical densities as the wild type (Fig. 2F). Our results immediately suggest next steps toward improvement. For example, our initial design used enzymes derived from the thermophile Chloroflexus, which are evolved to function at higher temperatures than the mesophilic growth conditions of plants and most cyanobacteria. This may underlie the low measured activity of heterologous PCS despite its strong overexpression in our transformants (Fig. 2B). Synthesis and assembly of such a large enzyme (ϳ600 kDa) might impose a considerable stress on the transformant strains. Substitution by a PCS homolog from a mesophile may improve assembly and function of this trimeric enzyme in S. elongatus. Mining genome databases for mesophilic homologs of the six enzymes that may exhibit faster enzyme kinetics at lower temperatures could greatly improve flux through the cycle. However, characterization of these mesophilic alternatives is necessary, as our results with the much less efficient ApMCT homolog demonstrate. Nevertheless, mesophilic enzymes may still be advantageous in terms of expression and correct folding at ambient temperatures.
Likewise, an increase in ACCase activity may be required. Our present design relies on the native enzyme to fix bicarbonate. ACCase is required for fatty acid biosynthesis, and endogenous levels of the enzyme may be insufficient to support optimal flux through the heterologously expressed cycle. However, overexpression of up to four separate subunits of the prokaryotic ACCases will significantly complicate DNA assembly and cloning strategies. Suitable alternatives may be eukaryotic ACCases, which have undergone gene fusion events creating one large single multifunctional gene (27).
pete with the synthetic bypass for substrate. In contrast, plants contain only the C 2 cycle, thus simplifying the fate of glyoxylate. With the localization of all six genes of our pathway to the chloroplast, only one additional enzyme, glycolate dehydrogenase, would be necessary to convert glyoxylate and bicarbonate to pyruvate. In fact, glycolate dehydrogenase has already been successfully targeted and expressed in chloroplasts of Arabidopsis (3).
Our results have implications beyond the optimization of photorespiration in plants and cyanobacteria. The successful introduction of half of the 3OHP bi-cycle into S. elongatus provides a platform in which to express the other half to attain the full bi-cyclic CO 2 fixation pathway. Given that CO 2 fixation limits the light-saturated rate of photosynthesis, the presence of two orthogonal CO 2 fixation pathways is expected to significantly enhance the conversion of solar energy into biomass. Although appealing, introducing the whole 3OHP bi-cycle will result in substantial carbon flux toward pyruvate, which could be detrimental to organisms that have evolved carbon metabolism based on sugar phosphates.
On the other hand, pyruvate or intermediates in the synthetic bypass could be redirected for biotechnological applications, such as biofuels or replacements for chemical feedstocks that are currently petroleum-derived (19). For example, we have shown that 3OHP, a precursor for bioplastics, can be derived from malonyl-CoA by the heterologous expression of MCR in cyanobacteria. Developing cyanobacteria as production strains requires increasing their tolerance to higher concentrations of 3OHP; this has been accomplished in E. coli (19). Likewise, the production of propionyl-CoA by the combined function of MCR and PCS in the synthetic bypass could be useful for the production of diverse polyhydroxyalkanoates such as polyhydroxyvalerate, polyhydroxymethylvalerate, or co-polymers.
It is generally assumed that an increase in carbon fixation or decrease in photorespiration will improve the efficiency of photosynthesis (7) and thereby increase growth rates or biomass production; however, the effects on such broad and complex physiological traits are determined by many additional factors (e.g. carbon allocation, translocation, and secretion). Although our growth phenotype is inconclusive (Fig. 2F), future efforts in improving the properties (e.g. finding mesophilic homologs) of the enzymes used in our pathway will be crucial to further investigate the physiological impacts of synthetic carbon fixation pathways and photorespiratory bypasses. However, although several strategies dealing with photorespiration are known to exist in cyanobacteria, its importance is not quite clear, especially in light of the cyanobacterial carbon-concentrating mechanisms that include the carboxysome. Moreover, we still do not fully understand all the effects of the introduced enzymes in vivo; thus, future efforts relying on global and systems approaches may shed light on potential avenues for pathway optimization, as utilized in E. coli metabolic engineering efforts (31,32). This study represents an advance toward understanding the effects of altering central carbon metabolism in photosynthetic autotrophs using synthetic biology and metabolic engineering approaches.
In summary, improving photosynthesis holds promise for increasing the sustainable production of food and biofuel crops to meet the challenges of global climate change and population growth, but introducing new pathways and cycles constitutes a daunting challenge. The synthetic photorespiratory bypass TABLE 2 Energy balance comparison of photorespiratory pathways to achieve the same level carbon gain as the 3OHP bypass * If pyruvate is used for the regeneration of 3-phosphoglycerate, 2 more ATP equivalents are required per pyruvate molecule by pyruvate phosphate dikinase (AMP-forming).
Only 6 ATP are required if pyruvate is channeled into other biosynthesis pathways than gluconeogenesis.
reported in this study provides both a precedent and a platform for future bioengineering efforts.