Awakening the Sleeping Carboxylase Function of Enzymes: Engineering the Natural CO2-Binding Potential of Reductases.

Developing new carbon dioxide (CO2) fixing enzymes is a prerequisite to create new biocatalysts for diverse applications in chemistry, biotechnology and synthetic biology. Here we used bioinformatics to identify a "sleeping carboxylase function" in the superfamily of medium-chain dehydrogenases/reductases (MDR), i.e. enzymes that possess a low carboxylation side activity next to their original enzyme reaction. We show that propionyl-CoA synthase from Erythrobacter sp. NAP1, as well as an acrylyl-CoA reductase from Nitrosopumilus maritimus possess carboxylation yields of 3 ± 1 and 4.5 ± 0.9%. We use rational design to engineer these enzymes further into carboxylases by increasing interactions of the proteins with CO2 and suppressing diffusion of water to the active site. The engineered carboxylases show improved CO2-binding and kinetic parameters comparable to naturally existing CO2-fixing enzymes. Our results provide a strategy to develop novel CO2-fixing enzymes and shed light on the emergence of natural carboxylases during evolution.

To harvest atmospheric CO2 as a sustainable carbon source for (bio)catalytic and (bio)technological applications,[1−5] it is necessary to extend the repertoire of CO2-fixing reactions. One possibility is to engineer a carboxylation function into the scaffold of non-CO2-fixing enzymes. Generally, the interaction of CO2 with proteins is poorly understood.[6] However, for enoyl-CoA carboxylase/reductase from Kitasatospora setae (ECRKs), four conserved amino acids that form a CO2-binding pocket at the active site were described recently[7] (Figure a). These four amino acids anchor and position the CO2 molecule during catalysis, in which a reactive enolate is formed that attacks the CO2.[8]

Figure 1

CO2-binding pocket of ECR and its partial conservation in the MDR superfamily. (a) Active site of ECRKs.[7] The CO2-binding pocket is defined by four conserved residues (Asn81, Phe170, Glu171, His365). CO2 was modeled into the structure. (b) Maximum-likelihood tree of the MDR superfamily[9] with (potential) CO2-binding enzyme families highlighted in color.

To identify enzyme scaffolds capable of binding CO2 beyond the ECR enzyme family, we searched homologues of the MDR superfamily for the CO2-binding motif. Our search revealed two enzyme families that show the potential to bind CO2, the propionyl-CoA synthase (PCS) and an archaeal enoyl-CoA reductase (AER) family (Figure b). The PCS family clusters closely to ECRs and shows a fully conserved CO2-binding motif across individual family members (Figure S1). The AER family is more distantly related to the ECR family, and selected homologues only contain one or two of the four conserved residues of the CO2-binding pocket (Figure S2). We decided to test selected members of these enzyme families in their CO2-fixing capabilities.

PCS is a three-domain fusion enzyme that catalyzes the overall conversion of 3-hydroxypropionate to propionyl-CoA[10] (Figure a). The enzyme forms a central reaction chamber, in which three subsequent reactions take place in a synchronized fashion.[11] When we assayed PCS from Erythrobacter sp. NAP1, PCSEN, at 4.4 mM dissolved CO2, we detected minor amounts of methylmalonyl-CoA besides the main product propionyl-CoA. Incorporation of 13CO2-label confirmed the latent carboxylation activity of PCSEN (Figure b). Notably, the carboxylation function was not limited to the Erythrobacter enzyme, but was also detected with PCS from Chloroflexus aurantiacus (PCSCa, Table S1).

Figure 2

PCSEN possesses a “sleeping carboxylase function”. (a) Reaction sequence of PCS. PCS natively catalyzes the conversion of 3-hydroxypropionate into propionyl-CoA (solid lines) and possesses a low carboxylation activity yielding methylmalonyl-CoA (dashed line). (b) High-performance liquid chromatography-mass spectrometry traces of the PCSEN overall reaction showing 3-hydroxypropionyl-CoA, propionyl-CoA and (3-12/13C)-methylmalonyl-CoA at m/z 840.14, 824.15 and 868.13/869.13, respectively. Methylmalonyl-CoA is only detected in the presence of 12/13CO2 (provided as bicarbonate). Data represent an individual experiment with two replicates. (c) Active site of PCSEN reductase domain (cyan, PDB: 4EQO(11)) and ECRKs (blue[7]), both cocrystallized with NADP+. Acrylyl-CoA and CO2 are modeled into the active site. WebLogo-Illustration[12,13] of conserved active site residues using 129 PCS and 29 ECR sequences. Numbering according to PCSEN or ECRKs, respectively.

The last reaction in the three-reaction sequence of PCS is the reduction of acrylyl-CoA to propionyl-CoA, catalyzed by a reductase domain harboring the CO2-binding motif (Figure a,c). We directly tested the reductase domain for carboxylation activity with an E1027Q variant of PCSEN (PCSEN_ΔDH) that is unable to generate acrylyl-CoA. When PCSEN_ΔDH was provided with external acrylyl-CoA and 4.4 mM dissolved CO2, the enzyme showed a carboxylation yield (defined as percentage yield of carboxylated product compared with total product formed, including reduced side product) of 3 ± 1% (Table ). This showed that the reductase domain is able to carboxylate acrylyl-CoA directly.

Table 1

Reaction Parameters and Carboxylation Yield for the Reductase Domain of Different PCSEN Variantsa

PCS variant	app. kcat. (s–1) at 4.4 mM CO2	app. KM_acrylyl-CoA (mM)	% carboxylation at 4.4 mM CO2	app. KM_CO2 (mM)
PCSEN_ΔDH WT	7.4 ± 1.0	0.014 ± 0.002	3 ± 1	n.m.
PCSEN_ΔDH D1302S	1.77 ± 0.09	0.027 ± 0.003	20.9 ± 0.7	27 ± 5
PCSEN_ΔDH T1753M	6.5 ± 0.6	0.0197 ± 0.0012	10 ± 2	n.m.
PCSEN_ΔDH D1302S T1753M	0.46 ± 0.03	0.026 ± 0.003	69 ± 3	26 ± 5

kcat. shows combined reduction and carboxylation activity. KM values were determined from a Michaelis–Menten fit of at least 18 data points, with fixed acrylyl-CoA concentrations for KM_CO2 (Figures S3 and S4, Table S2 for kcat. values). Carboxylation yields are calculated from mean carboxylation yields over five time points in three replicates. Data are mean ± s.d. CO2 concentrations were calculated. n.m., not measurable.

To improve further the carboxylation efficiency of PCSEN, we compared the active site of PCSEN (PDB: 6EQO) with ECRKs. While the NADPH binding site, as well as the four CO2-binding pocket residues are structurally conserved (Figure c), we noticed differences in the second shell of the active site. ECRKs features a small hydrophilic residue (Thr82), which interacts with Asn81 that stabilizes CO2 through its carboxyamide NH2 group. The corresponding residue in PCSEN is occupied by an aspartate (Asp1302). Molecular dynamics (MD) simulations demonstrated that Asp1302 in PCSEN forms a strong anionic hydrogen bond to the carboxamide NH2 group of Asn1301 (Figures a and S5), locking Asn1301 in a position which prevents interactions with CO2. This finding is in line with the fact that we could not determine an apparent KM for CO2 with PCSEN_ΔDH and that replacing Asn1301 by an aspartate abolished carboxylation activity.

Figure 3

Directed mutagenesis to exploit the carboxylation activity of PCSEN. (a) Representative snapshot from the MD simulation of the active site in wild type PCSEN. (b) Active site model of PCSEN D1302S to unlock Asn1301. (c) Active site of wild type PCSEN. (d) Active site model of PCSEN T1753M to restrict water access to the active site. Acrylyl-CoA and CO2 were modeled into the active site.

We aimed at unlocking Asn1301 from its fixed position by replacing Asp1302 with different small hydrophilic residues. PCSEN_ΔDH variant D1302S (Figure b) showed an increased carboxylation yield of 20.9 ± 0.7% at 4.4 mM dissolved CO2, and notably also Michaelis–Menten-like behavior with CO2 at an apparent KM_CO2 of 27 ± 5 mM (Table ). Together with MD simulations that showed a more flexible asparagine residue (Figure S5) this demonstrated that unlocking Asn1301 improves CO2-binding and carboxylation efficiency in PCSEN.

Another, equally important catalytic principle in carboxylases is the exclusion of water from the active site to minimize protonation reactions which would prematurely quench C–C bond formation.[7,14−16] In ECRKs, a conserved methionine (Met356) restricts access of water to the CO2-binding pocket. In PCSEN, this residue is a threonine, which presumably allows water to enter the active site and displace the CO2 molecule (Figure c). When we introduced the methionine in PCSEN (PCSEN_ΔDH T1753M, Figure d), carboxylation yield increased to 10 ± 2% at 4.4 mM dissolved CO2. When combining the D1302S with the T1573M mutation, the carboxylation yield of PCSEN_ΔDH further increased up to 69 ± 3% at 4.4 mM CO2 (Table ). Under saturating CO2 concentrations (i.e., 44 mM CO2), PCSEN_ΔDH D1302S T1753M showed a carboxylation yield of 94.5 ± 0.7%, demonstrating that we successfully converted the reductase domain into a carboxylase. During engineering, the kcat. of reduction was strongly decreased, while the apparent kcat. for carboxylation was maintained (Table S2) and falls in the range of naturally existing ECRs.[4,17] The engineered carboxylase domain also improved carboxylation yield in the context of the overall reaction of PCSEN (Supporting Information I).

We next investigated the carboxylation potential in the AER enzyme family of unknown function. We chose Nmar_1565 (AERNm), a homologue from Nitrosopumilus maritimus, in which two of the four amino acids of the CO2-binding motif, namely Phe122 and Glu123, are conserved (Figure a,b). Although no function was assigned to AERNm so far, we speculated that the enzyme might catalyze the reduction of acrylyl-CoA in the 3-hydroxypropionate/4-hyroxybutyrate cycle of N. maritimus.[18,19] Indeed, the enzyme reduced acrylyl-CoA to propionyl-CoA at an apparent kcat. of 0.99 ± 0.11 s–1, confirming its reductase function.

Figure 4

Awakening the “sleeping carboxylase function” in AERNm. (a) Active sites of AERNm (salmon) and ECRKs (blue). Illustration of conserved active site residues, generated by WebLogo[12,13] using 21 AERNm and 29 ECR sequences. Residue numbering refers to AERNm or ECRKs sequence, respectively. (b) Model of the AERNm wild type active site carrying an Asp50 instead of a conserved Asn. (c) Model of the AERNm D50N active site. Homology models were created with an ECR from Streptomyces sp. NRRL 2288 (PDB: 4y0k(21)) using SWISS-MODEL.[22] Acrylyl-CoA and CO2 were modeled into the active site.

AERNm activity was very sensitive to salt and buffer composition (Supporting Information II). When we incubated the enzyme with NaHCO3, at concentrations corresponding to 1.31 mM free CO2, activity dropped 10-fold. However, under these conditions AERNm showed a latent carboxylation activity and converted acrylyl-CoA into methylmalonyl-CoA at a carboxylation yield of 4.5 ± 0.9% (Table ), despite the lack of two of the four amino acid residues of the CO2-binding motif. To increase the carboxylation efficiency of AERNm we decided to rebuild the CO2-binding pocket through introduction of asparagine and histidine. Reintroduction of histidine failed due to inactive protein, which might be a result of interrupted second-shell interactions to Thr307 or steric clashes. However, replacing Asp50 by asparagine increased carboxylation yield dramatically (to 82 ± 5%, Figure c, Table ). The increase in catalytic activity in AERNm D50N was accompanied by an improved KM_CO2 (0.18 ± 0.03 mM), indicating increased CO2-binding. AERNm D50N performed best in 100 mM phosphate buffer, where it showed kcat. and carboxylation yields comparable to those of naturally existing carboxylases, such as RubisCO[20] (Table ).

Table 2

Reaction Parameters and Carboxylation Yield for AERNm Variants in 100 mM KHPO4 (pH 7.5)a

AER variant	app. kcat. (s–1) at 1.31 mM CO2	% carboxylation at 1.31 mM CO2	KM_CO2 (mM)
WT	0.084 ± 0.011	4.5 ± 0.9	n.m.
D50N	1.6 ± 0.2	82 ± 5	0.18 ± 0.03

kcat. shows combined carboxylation and reduction activity. KM_CO2 values were determined from a Michaelis–Menten fit of at least 18 data points with a fixed concentration of acrylyl-CoA (Figure S6, Table S3 for kcat. values). Carboxylation yields were calculated from the mean carboxylation ratio over five time points in three replicates. Data are mean ± s.d. CO2 concentrations were calculated. n.m., not measurable.

In conclusion, we successfully reshaped the energy landscape of acrylyl-CoA reductases from the thermodynamically favored product propionyl-CoA (ΔrG′0 ≈ −63 kJ/mol) to the disfavored methylmalonyl-CoA (ΔrG′0 ≈ −43 kJ/mol).[23] Our engineering efforts show that improving CO2-binding (reduced energy barrier for carboxylation) and minimizing side reaction with water (increased energy barrier for reduction) are both required to establish a carboxylation activity in the scaffold of different reductases. This is in line with the idea that in catalysis stabilization of favorable transition states (“positive catalysis”) and destabilization of unwanted transition states (“negative catalysis”) are both important,[24−26] as further supported by the finding that suppression of competing protonation side reactions is essential for efficient CO2-fixation in ECRKs and 2-ketopropyl coenzyme M oxidoreductase/carboxylase.[14−16]

On a broader picture, our findings also raise questions about the emergence of natural carboxylases. How did carboxylation functions naturally evolve in the scaffold of proteins, such as RubisCO or ECR? It has been suggested that these enzymes originated from non-CO2-fixing ancestors.[27,28] Our data provides experimental evidence for this evolutionary scenario by demonstrating that the MDR superfamily, to which ECR belongs, naturally possesses the capacity to interact with the CO2-molecule. It apparently takes only a few mutations to transform latent carboxylases that convert CO2 at low efficiency and nonphysiological CO2 concentrations into decent CO2-fixing enzymes.

Another apparent question is why PCS and AER would possess a “sleeping carboxylase function”? One explanation might be that the latent carboxylation activity was selected for. PCS operates in the 3-hydroxypropionate bicycle in C. aurantiacus and a modified version thereof in Erythrobacter sp. NAP1 (Figure S7a),[29,30] while AERNm presumably works in the 3-hydroxypropionate/4-hyroxybutyrate cycle in N. maritimus (Figure S7b).[18] Bioenergetic considerations suggest that even a low carboxylation activity would increase biomass yield of these organisms, which thrive at a constantly low energy supply[18] (Supporting Information III).

In summary, our proof-of-principle study demonstrates that it is possible to exploit the active site of reductases to create novel carboxylases. This opens the possibility for the future engineering of novel CO2-fixing enzymes that could find application in biocatalysis and synthetic biology (e.g., in artificial pathways for the conversion of CO2[31,32]).