A synthetic pathway for the fixation of carbon dioxide in vitro

See allHide authors and affiliations

Science  18 Nov 2016:
Vol. 354, Issue 6314, pp. 900-904
DOI: 10.1126/science.aah5237

Optimizing designer metabolisms in vitro

Biological carbon fixation requires several enzymes to turn CO2 into biomass. Although this pathway evolved in plants, algae, and microorganisms over billions of years, many reactions and enzymes could aid in the production of desired chemical products instead of biomass. Schwander et al. constructed an optimized synthetic carbon fixation pathway in vitro by using 17 enzymes—including three engineered enzymes—from nine different organisms across all three domains of life (see the Perspective by Gong and Li). The pathway is up to five times more efficient than the in vivo rates of the most common natural carbon fixation pathway. Further optimization of this and other metabolic pathways by using similar approaches may lead to a host of biotechnological applications.

Science, this issue p. 900; see also p. 830


Carbon dioxide (CO2) is an important carbon feedstock for a future green economy. This requires the development of efficient strategies for its conversion into multicarbon compounds. We describe a synthetic cycle for the continuous fixation of CO2 in vitro. The crotonyl–coenzyme A (CoA)/ethylmalonyl-CoA/hydroxybutyryl-CoA (CETCH) cycle is a reaction network of 17 enzymes that converts CO2 into organic molecules at a rate of 5 nanomoles of CO2 per minute per milligram of protein. The CETCH cycle was drafted by metabolic retrosynthesis, established with enzymes originating from nine different organisms of all three domains of life, and optimized in several rounds by enzyme engineering and metabolic proofreading. The CETCH cycle adds a seventh, synthetic alternative to the six naturally evolved CO2 fixation pathways, thereby opening the way for in vitro and in vivo applications.

Autotrophic carbon fixation transforms more than 350 gigatons of CO2 annually. More than 90% of the carbon is fixed by the Calvin-Benson-Bassham (CBB) cycle in plants, algae, and microorganisms. The rest is converted through alternative autotrophic CO2 fixation pathways (1, 2). Despite this naturally existing diversity, the application of CO2-fixing enzymes and pathways for converting CO2 into value-added multi-carbon products has been limited in chemistry (3, 4) and biotechnology (5). Natural CO2 fixation delivers mainly biomass and not a dedicated product. Moreover, under optimal conditions, biological CO2 fixation is often directly affected by the inefficiency of the CO2-fixing enzymes and pathways. For instance, the CBB cycle’s carboxylase, RuBisCO (ribulose-1,5-bisphosphate carboxylase/oxygenase), is a slow catalyst that shows a strong side reaction with oxygen, which leads to the loss of fixed carbon and thus photosynthetic energy by up to 30% in a process called photorespiration (6).

Attempts to improve biological CO2 fixation (7) have included evolving RuBisCO toward higher reaction rate and specificity (8, 9), engineering more efficient photorespiration (10, 11), and transplanting natural CO2 fixation pathways into nonautotrophic organisms such as Escherichia coli (1214). In contrast to these efforts, which showed only limited success, the emerging field of synthetic biology provides an alternative approach to create designer CO2 fixation pathways. By freely combining different enzymatic reactions from various biological sources, completely artificial CO2 fixation routes may be constructed that are kinetically or thermodynamically favored relative to the naturally evolved CO2 fixation pathways. Several synthetic routes for CO2 fixation have been theoretically considered (15). However, the gap between theoretical design and experimental realization in synthetic biology has impeded the realization of such artificial pathways. For example, attempts to directly assemble synthetic pathways in living organisms are challenged by limited understanding of the complex interplay among the different enzymes used in these synthetic networks, as well as interference of the synthetic networks in the complex background of the host organism, which can lead to undesired effects such as side reactions and toxicity. Therefore, the realization of synthetic pathways requires novel strategies that first allow their testing and optimization in more defined conditions (1619). To overcome these limitations, we decided to take a radically different, reductionist approach by assembling a synthetic CO2 fixation cycle from its principal components in a bottom-up fashion (fig. S1).

A known bottleneck in natural CO2 fixation is the efficiency of a carboxylating enzyme in a given pathway (2022). To identify a suitable CO2 fixation reaction for our synthetic cycle, we first compared the different biochemical and kinetic properties of all known major carboxylase classes (23) (table S1). On the basis of this analysis, we decided to rely on coenzyme A (CoA)–dependent carboxylases, and enoyl-CoA carboxylases/reductases (ECRs) in particular, because of their favorable catalytic properties. ECRs are a recently discovered class of carboxylases that operate in secondary metabolism, as well as in central carbon metabolism of α-proteobacteria and Streptomycetes, but notably not in any autotrophic CO2 fixation pathway known so far (24). Relative to other carboxylases, including RuBisCO, ECRs span a broad substrate spectrum (25), are oxygen-insensitive, do not accept molecular oxygen as substrate, require only the ubiquitous redox cofactor NADPH (reduced nicotinamide adenine dinucleotide phosphate), and catalyze the fixation of CO2 with high catalytic efficiency (i.e., on average better than RuBisCO by a factor of 2 to 4) (table S1) (26).

We conceived several theoretical CO2 fixation routes that (i) start with a given ECR reaction, (ii) regenerate the carboxylation substrate to allow for continuous cycling, and (iii) feature a dedicated output reaction to channel the fixed carbon into a product (fig. S2). In contrast to earlier approaches (15, 27), we did not restrict our design to known enzymes; rather, we considered all reactions that seemed biochemically feasible (i.e., fall into one of the standard reaction classes defined by the Enzyme Commission). We obtained several prospective cycles, of which seven were considered further (fig. S2, A to G).

We first evaluated the thermodynamic feasibility of these theoretical cycles by calculating their Gibbs free energy profile (ΔrG′) under biologically relevant conditions and estimating the presumable consumption of adenosine triphosphate (ATP) and NADPH in each cycle per CO2 molecule converted (fig. S2 and supplementary text). Our calculations show that the synthetic cycles were generally more energy-efficient than a stand-alone CBB cycle and were also more energy-efficient than the core sequence of other naturally existing, oxygen-insensitive CO2 fixation pathways, such as the 3-hydroxypropionate (3HP) bicycle and the 3-hydroxypropionate/4-hydroxybutyrate (3HP/4HB) cycle (table S2). As an example, four of our synthetic cycles required on average one-third less ATP per CO2 converted into phosphoglycerate, relative to the core sequence of the CBB cycle. This would translate to at least seven fewer photons per phosphoglycerate formed from CO2 in a theoretical photosynthetically coupled process (Table 1 and figs. S3 and S4). In summary, these calculations suggested that our synthetic cycles could provide promising solutions to an efficient C-C bond formation under aerobic conditions that compete with naturally existing metabolic pathways.

Table 1 Comparison of selected theoretical and synthetic CO2 fixation cycles with the core sequence of naturally existing, aero-tolerant CO2 fixation pathways.

Shown are the number of reaction steps and the numbers of NAD(P)H, ATP, and FADH2 molecules required (negative numbers) or generated (positive numbers) during conversion of CO2 into one molecule of phosphoglycerate. These values were normalized to ATP equivalents required per CO2 molecule fixed into phosphoglycerate, or alternatively to the number of photons per CO2 fixed into phosphoglycerate in a theoretical photosynthetically coupled process (see supplementary text and figs. S3 and S4 for a more detailed analysis). Number of reaction steps does not take into account stereochemical conversions such as racemization or epimerization. For ATP per CO2 conversions, a P/O ratio of 2.5 for NAD(P)H and 1.5 for FADH2 was assumed (38). For photons per CO2 conversions, the number of biologically generated photons required for the generation of the NADPH and ATP in each cycle was calculated, assuming a stoichiometry of 1 NADPH and 1.28 ATP per 4 photons (39). The photon energy left is given in ATP equivalents. These calculations do not consider the additional generation of ATP through any FADH2 generated through a given pathway. HOPAC, hydroxypropionyl-CoA/acrylyl-CoA; FUMES, fumaryl-CoA/methylmalyl-CoA/succinyl-CoA; HITME, hydroxycrotonyl-CoA/itaconyl-CoA/methylmalonyl-CoA; n.c., not calculated.

View this table:

Next, we searched bioinformatic databases for enzymes that could sustain our drafts (fig. S2). For the crotonyl-CoA/ethylmalonyl-CoA/hydroxybutyryl-CoA (CETCH) cycle (fig. S2A), we were able to identify one or more potential enzyme candidates for each reaction of its core sequence. We purified, tested, and characterized these candidates and selected a set of 12 enzymes, each with a specific activity of >1 U mg−1 protein at 250 μM substrate concentration (table S3), to experimentally realize the CETCH cycle as a proof of principle. To demonstrate the feasibility of the CETCH cycle, we first reconstituted its central CO2 fixation reaction sequence stepwise. We provided cofactors and the chemically synthesized intermediate propionyl-CoA in a buffer containing NaHCO3. To this mixture, propionyl-CoA carboxylase (Pcc) was added first, followed by the subsequent addition of the other 11 enzymes. High-performance liquid chromatography–mass spectrometry (HPLC-MS) analysis demonstrated the stepwise transformation of propionyl-CoA into key intermediates of the pathway (fig. S5), indicating that this first version of the CETCH cycle (CETCH 1.0) was in principle functional.

To test for continuous operation of CETCH 1.0, we provided all enzymes, cofactors, and the artificial electron acceptor ferrocenium from the beginning and started the cycle by addition of propionyl-CoA (fig. S6A). To follow the dynamics of the cycle, we used NaH13CO3 as the CO2 source for this and all subsequent experiments. Because of the constant incorporation of CO2 during the course of the cycle, we could use enrichment of the 13C isotope in the intermediates to follow flux through the CETCH cycle in the continuous mode and under apparent steady-state conditions (fig. S7). However, the conversion of CoA esters stopped before a complete turn of the cycle at the level of methylsuccinyl-CoA, which indicated that the methylsuccinyl-CoA dehydrogenase (Mcd) (28) enzyme reaction used in the cycle was rate-limiting (fig. S6, B and C, and supplementary text).

Mcd belongs to the family of flavin adenine dinucleotide (FAD)–dependent acyl-CoA dehydrogenases, which are coupled via electron transfer flavoproteins to the membrane-bound ubiquinone pool and eventually to the respiratory chain. In the stepwise reconstitution of CETCH 1.0, we had added the artificial electron acceptor ferrocenium together with Mcd to substitute for the electron transfer cascade. Under continuous conditions, the use of the artificial electron acceptor from the beginning was limited to 0.1 mM, because higher concentrations caused other enzymes in the assay to precipitate. At 0.1 mM, however, the spontaneous reduction of ferrocenium to ferrocene by NADPH (29), as well as the slow reoxidation rate of ferrocene, most likely prevented effective cycling of CETCH 1.0.

To overcome the limitations caused by Mcd, we engineered the enzyme to directly use molecular oxygen as electron acceptor. On the basis of structural considerations (Fig. 1 and supplementary text), we introduced three point mutations—Thr317 → Gly (T317G), Glu377 → Asn (E377N), and Trp315 → Phe (W315F)—to convert the dehydrogenase into a functional methylsuccinyl-CoA oxidase (Mco) that catalyzes the oxygen-dependent oxidation of methylsuccinyl-CoA with a maximal catalytic rate vmax of 97 ± 6 mU mg−1 and an apparent Michaelis-Menten constant KM of 27 ± 5 μM. In version 2.0 of the CETCH cycle, we replaced Mcd with the engineered Mco. Fractional labeling of intermediates showed that CETCH 2.0 was turning at least twice within 45 min (fig. S8).

Fig. 1 Structure-guided engineering of Mcd into a Mco.

(A) Active-site comparison of the human short-chain acyl-CoA dehydrogenase (green backbone; PDB ID 2VIG) with a model of Mcd from Rhodobacter sphaeroides (blue backbone; modeled by the SWISS-MODEL server with 2VIG as template) and the short-chain acyl-CoA oxidase 4 from Arabidopsis thaliana (orange backbone; PDB ID 2IX5). The three residues that were targeted to introduce oxidase activity into Mcd are highlighted. (B) HPLC-MS–based analysis of wild-type and different single active-site mutants for oxidation of methylsuccinyl-CoA into mesaconyl-CoA with molecular oxygen as electron acceptor. The screen identified three substitutions (W315F, T317G, and E377N) that were combined into double and triple mutants. (C) Michaelis-Menten graphs of single, double, and triple mutants characterized in more detail. The triple mutant showed the best kinetic parameters.

After successful demonstration of the CETCH core cycle under continuous conditions, we next optimized its reaction sequence. We established a readout module to directly quantify CO2 fixation (fig. S9A). Starting from propionyl-CoA, the CETCH cycle converts two molecules of CO2 per turn into one molecule of glyoxylate. To measure glyoxylate formation, we used a malate synthase (Mas) from E. coli, which condenses glyoxylate with externally added acetyl-CoA to yield malate. From here on, we used the malate production rate as a readout—in combination with CoA ester analysis—to improve CO2 fixation efficiency of the CETCH cycle stepwise (CETCH 2.0–5.4; see supplementary text).

During the optimization phase, we added enzymes for ATP and NADPH regeneration, provided chemical and enzymatic protection against oxidative damage from H2O2-producing enzymes, and dealt with unwanted side reactions that prevented effective cycling. The CETCH cycle includes enzymes from very different biological sources. The combination of such diverse enzymes into a synthetic pathway poses the challenge that these enzymes face metabolites to which they were never exposed in their native metabolic context, causing undesired side reactions. Inspired by principles of natural metabolism, we pursued different strategies to overcome this problem (supplementary text and figs. S9 to S12). We replaced the original pathway design with alternative reaction sequences, used enzyme engineering to minimize side reactions of promiscuous enzymes, and introduced proofreading enzymes to correct for the formation of dead-end metabolites. Whereas the first two of these strategies are applied in biotechnology and synthetic biology to improve productivity (30, 31), the concept of metabolic proofreading has only rarely been considered for synthetic pathway design (16), although proofreading apparently exists in naturally evolved pathways (32). We believe that this design principle should be included more systematically in metabolic engineering in the future.

Over the course of the optimization, CO2 fixation efficiency in the CETCH cycle improved by almost a factor of 20 until version 5.4 (Fig. 2). CETCH 5.4 represents an in vitro enzyme network that is able to fix CO2 at a rate of 5 nmol min−1 mg−1 of core cycle proteins (Fig. 2C). This is comparable to the few reported attempts to measure the CBB cycle in cell extracts (1 to 3 nmol min−1 mg−1 CBB cycle protein, assuming that CBB cycle enzymes account for 30% of the proteome) (33). However, we note that cell extracts represent a more complex environment than our reductionist in vitro system.

Fig. 2 The CETCH cycle.

(A) Topology of the CETCH cycle (version 5.4), including proofreading and cofactor regenerating enzymes. See table S3 for numbering of reaction steps and enzyme abbreviations. (B) Dynamics of key intermediates of CETCH 5.4 over 90 min. Shown are the levels of six different intermediates, as well as their fractional labeling patterns from the incorporation of 13CO2 for each turn of the cycle (see fig. S7 for the expected labeling pattern). Intermediates are colored as in (A). (C) Left y axis: CO2 fixation efficiency of the CETCH cycle over the course of its optimization (see supplementary text). CO2 fixation efficiency is defined as CO2 equivalents fixed per acceptor molecule in the cycle (i.e., starting amount of propionyl-CoA). Right y axis: Absolute malate concentration formed over the course of 90 min in CETCH 5.4. The final assay (0.52 ml) contained 2.3 mg ml−1 protein of cycle core enzymes plus 0.8 mg ml−1 auxiliary enzymes and produced 540 μM malate over 90 min, which corresponds to 1080 μM fixed CO2.

In total, the 13 core reactions of CETCH 5.4, together with the auxiliary proofreading and cofactor regeneration processes, are catalyzed by 17 enzymes from nine different organisms and all three domains of life, including bacteria, archaea, plants, and humans. In addition, the cycle features three reactions that were created by rational active-site engineering of existing enzyme scaffolds to catalyze the desired activities, such as Mco and Pco. Notably, CETCH 5.4 relies solely on the reductive carboxylation of enoyl-CoA esters. Although ECRs belong to the most efficient CO2-fixing group of enzymes known to date, they were not selected for autotrophic CO2 fixation during evolution, as far as we know. The reallocation of reductive carboxylation as a key reaction for a synthetic autotrophic CO2 fixation cycle thus goes beyond simply improving or reshuffling naturally existing autotrophic CO2 fixation reactions and pathways. The CETCH cycle expands the diversity of the six naturally evolved CO2 fixation strategies by a seventh, synthetic alternative that did not require the serendipity of evolution to bring together all components in space and time (34).

The successful in vitro reconstitution of a synthetic enzymatic network for the conversion of CO2 into organic products that is superior to chemical processes and competes with naturally existing CO2-fixing solutions in vitro opens the door for several future applications. These could include the in vivo transplantation of synthetic CO2 fixation cycles into lithotrophic or photosynthetic organisms, paving the way to improved CO2 fixation (35); the use of synthetic CO2 fixation cycles in the development of artificial photosynthetic processes (e.g., in combination with photovoltaics or artificial leaves) (36); and the design of a self-sustained, completely synthetic carbon metabolism in artificial or minimal cells (37).

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S14

Tables S1 to S8

References (4163)

References and Notes

  1. Acknowledgments: We thank G. Fuchs for introducing us to the fundamental principles of microbial CO2 fixation. We thank U. Oppermann for the expression plasmid p2BP1, K. Castiglione for the expression plasmid of Fdh (D221A), J. Andexer for the expression plasmid for Pkk2 and the purified enzyme, and J. Zarzycki and B. Vögeli for contributions during pathway design. The work conducted by the U.S. Department of Energy Joint Genome Institute, a DOE Office of Science User Facility, is supported under contract DE-AC02-05CH11231. Also supported by European Research Council grant ERC 637675 “SYBORG,” Swiss National Science Foundation Ambizione grant PZ00P3_136828/1, ETH Zürich grant ETH-41 12-2, and the Max Planck Society. Supporting data are available in the supplementary materials. Author contributions: T.J.E. conceived the project; T.J.E., L.S.v.B., and T.S. designed the experiments, analyzed the data, and wrote the manuscript; T.J.E., T.S., and L.S.v.B. performed experiments; S.B. assisted in experiments; and N.S.C. performed mass spectrometry and analyzed the data.
View Abstract

Navigate This Article