Substrate Channeling by a Rationally Designed Fusion Protein in a Biocatalytic Cascade

Abstract

Substrate channeling, where an intermediate in a multistep reaction is directed toward a reaction center rather than freely diffusing, offers several advantages when employed in catalytic cascades. Here we present a fusion enzyme comprised of an alcohol and aldehyde dehydrogenase, that is computationally designed to facilitate electrostatic substrate channeling using a cationic linker bridging the two structures. Rosetta protein folding software was utilized to determine an optimal linker placement, added to the truncated termini of the proteins, which is as close as possible to the active sites of the enzymes without disrupting critical catalytic residues. With improvements in stability, product selectivity (90%), and catalyst turnover frequency, representing 500-fold increased activity compared to the unbound enzymes and nearly 140-fold for a neutral-linked fusion enzyme, this design strategy holds promise for making other multistep catalytic processes more sustainable and efficient.

Introduction

Substrate channeling is a phenomenon by which an intermediate in a multistep catalytic cascade is prevented from reaching equilibrium with the bulk solution and is instead directed toward subsequent reaction centers.^1,2 Within biological systems, substrate channeling between enzymes in cascade reactions results in increased flux through preferred pathways and prevention of side reactions; this is a critical feature considering the plethora of catalysts and substrates all present in the same aqueous solution in typical organisms.^3,4 As research into and the utilization of multistep enzyme cascades continues to increase in recent years, methods by which substrate channeling can be applied to these cascade reactions are being actively pursued.⁵

Even in the absence of substrate channeling, there are many merits incentivizing the use of one-pot, multistep catalyst cascades to perform a desired chemical conversion. In such cases, processes are often made more environmentally conscientious, sustainable, simple in design, and efficient in terms of time and waste production.⁶⁻⁹ High value-added chemical conversions are often inaccessible or less optimally performed (e.g., by using more costly starting materials) with a single catalyst.¹⁰ Multicatalyst conversions allow for the removal of inefficient intermediate preconcentration steps and can eliminate the handling or storage of highly reactive or toxic intermediates, offering improved safety. However, Kieboom et al. identified that challenges remain for the use of such cascade reactions. In particular, byproduct accumulation can occur if reactions are not highly selective, and catalyst poisoning by even trace quantities of byproducts has been reported in some cases.⁵ This encourages the utilization of substrate channeling; selectivity is improved as the intermediate is directed toward the desired reaction center of interest and away from the bulk solution, where it may interact unfavorably with other catalysts or system elements. Enzymes perform highly stereo- and regioselective chemical conversions and frequently share similar condition optima, such as those which operate ideally under ambient pH and temperature or those from the same organism; this makes enzymes ideal candidates for one-pot catalyst cascade reactions.¹¹

Among the bioinspired mechanisms utilized by researchers to achieve substrate channeling between enzymes, which include chemical swing-arms,¹² spatial confinement,¹³ and electrostatic guidance,¹⁴ the synthesis of fusion proteins is of particular interest for their relatively simple design and fabrication;¹⁰ here, two or more enzymes are linked together at their carboxyl (C) or amino (N) termini, typically with a bridging amino acid sequence in between, and expressed in a single open reading frame. Alternatively, the polypeptide bridge can be functionalized with non-natural amino acids to facilitate alkyne/azide “click” reactions at specific sites on the constituent enzymes.¹⁵ The control over active site orientation made possible by this strategy provides further benefits in fusion protein optimization compared to successful methods with randomized orientations, such as typical fusion proteins or enzyme aggregates.¹⁶⁻¹⁸ The characteristics defining the linker, including its length, composition, and point of articulation with each constituent enzyme, are controlled at the genetic level and are thus relatively straightforward to modify using the expansive toolkit for DNA manipulation available to modern biochemists.

There is some disagreement about whether fusion proteins with simple linker sequences intended to provide only proximity and/or orientation of the enzymes constitute substrate channeling in a formal sense.¹⁹ In systems such as those which employ fusion proteins,^20,21 protein/DNA scaffolds,²²⁻²⁴ or surface coimmobilization of constituents,^3,17 significant catalytic rate enhancements have been observed which exceed expectations when proximity alone is considered; Sigman et al. illustrated that, for a typical enzyme (e.g., turnover frequency, k, of 10 s^–1) and a typical aqueous-phase small molecule (e.g., diffusion coefficient, D, of 10^–5 cm² s^–1), there is no appreciable enhancement of local concentrations of the intermediate species at subsequent reaction centers compared to the bulk solution.¹ That is, diffusion typically occurs at time scales that are orders of magnitude faster than enzyme turnover events, so proximity alone should not produce an observable overall rate increase for most cascade reactions. Therefore, evidence supports the notion that other mechanisms besides simply an increased local intermediate concentration must be present for these observed rate enhancements. Fu et al., in their work involving glucose oxidase and horseradish peroxidase held at controlled distances between one another using DNA coimmobilization to DNA origami tiles, observed significant activity enhancements when the two were 10 nm apart; the team was able to provide evidence that this rate enhancement beyond the expectations of pure Brownian diffusion may be due to constrained diffusion of hydrogen peroxide, the intermediate species, within the hydration layer of the proteins due to favorable intermolecular interactions.²⁵ While further experimentation is needed to support this hypothesis, this insightful investigation serves to demonstrate the distinction between claims of substrate channeling producing rate enhancement by reducing random diffusion times—typically untrue or impossible—and by constraining diffusion to occur along particular pathways by favorable intermolecular interactions.

Rather than the linker sequence only serving to hold the constituent enzymes in proximity and/or ideal orientation, a polypeptide linker in a fusion protein can perform an active role in intermediate transport through deliberate intermolecular interactions.²⁶ The use of fusion proteins to achieve electrostatic substrate channeling remains relatively unexplored compared to other methods, and both theoretical and experimental investigations confirm that this strategy can be complex and difficult to employ. Brownian dynamics simulations support the notion that, even for electrostatic channeling, the proximity of enzyme active sites remains an important factor and should remain below about 2.5 nm for any meaningful channeling effect to be observed.¹⁶ The limitations of fusion proteins in general, including increased risk of inclusion body formation or deactivation of enzyme activity through functionalization or sequence modification, also apply in these systems;^27,28 further, the initial design of fusion proteins is often extensive and time-consuming.^8,29 For fusion proteins combining two large constituents, heterogeneity in conformation with unpredictable effects on structure and catalytic capability can occur if the linker is not rigid.³⁰ Enzymes that derive catalytic activity from quaternary structure (e.g., dimerization) can be deactivated by fusion with another enzyme as well.⁸

Despite these challenges, it is clear from experimental evidence that electrostatic substrate channeling in fusion proteins can produce enormous improvements in terms of cascade conversion efficiency (i.e., the portion of intermediate species successfully converted to final product) and overall rates of reaction through the prevention of competing reactions. Earl et al. demonstrated that over 95% channeling efficiency was attainable for two enzymes utilizing a bridge of opposite charge to the intermediate species, even under suboptimal rates of intermediate consumption and diffusion.²⁶ Liu et al., in work supported by molecular dynamics simulations, linked glucose 6-phosphate dehydrogenase and hexokinase using a cation-rich poly(lysine) bridge to effectively channel glucose 6-phosphate, the anionic intermediate, attaining steady-state reaction rates faster than for the unbound or neutral-linked controls.³¹ Further simulations of this system supported a mechanism of transport and a theoretical limit of up to 92% channeling efficiency, decreasing with increasing bridge length and decreasing intermolecular forces.^32,33

On the basis of the findings of these works and others, a system that takes full advantage of the potential of substrate channeling must be designed with a strong emphasis on fusion protein structure, particularly with regard to linker features (length, composition, articulation sites) and retention of enzyme activity—that is, preservation of the native fold in residues comprising the active site—after modification. While the limitation of modern computation to accurately compute protein structures due to their complexity persists as a major obstacle, significant strides have been made in protein theoretical research;³⁴ in particular, the Rosetta protein folding software suite from the Baker lab has demonstrated widespread success in this area.³⁵ Rosetta employs a hybrid coarse- and fine-grained energy function rather than purely explicit atomic simulation to greatly reduce computational demand. By using tools for secondary structure prediction such as the YASPIN hidden neural network along with Rosetta to take this secondary structure information and determine the tertiary structure, certain protein structural information can be obtained with relative ease and accuracy.³⁶

In this work, we demonstrate the computational, rational design of a fusion protein of NAD(H)-dependent Escherichia coli alcohol dehydrogenase (ADH) (PDB ID: 4gkv) and NAD(H)-dependent Bacillus cereus aldehyde dehydrogenase (aldDH) (PDB ID: 5gtk), two proteins participating in the two-step, reversible interconversion of ethanol and acetate (Figure 1). These enzymes were chosen for their shared, mild condition optima (T_opt = 37 °C and pH_opt = 7.5), shared cofactor in NADH, and known crystal structures. The reaction cascade catalyzed by these enzymes also holds promise for the sustainable production of biofuel from waste materials.³⁷ For the first time, Rosetta is used to predict and optimize fusion protein structures with a short, cation-rich, α-helix-forming polypeptide linker positioned as close as possible to the protein active sites; this facilitates both rigid orientation between the two enzymes and electrostatic channeling of acetaldehyde by ion-dipole interactions. The fusion protein (f+), along with neutral-linked (f0) and unlinked controls, are synthesized and purified, then coimmobilized with an NADH regeneration system to drive the electrochemical reduction of acetate to ethanol as a demonstration of a useful biosynthesis scheme. Gas chromatography with a flame ionization detector (GC-FID) is used for the simultaneous quantitation of ethanol and acetaldehyde; by comparing the quantities produced by f+ and f0 to unlinked ADH and aldDH, we are able to observe the contributions to performance enhancement of the bioelectrosynthesis scheme and draw conclusions regarding the merits of this fusion protein strategy for improving multistep catalysis.

Figure 1.

Reversible interconversion of ethanol and acetate catalyzed by ADH and aldDH. The fusion protein, f+, is comprised of ADH and aldDH with a cationic linker to facilitate substrate channeling of acetaldehyde, preventing its diffusion to the bulk solution and enabling enhanced flux through the reaction cascade.

Experimental Section

Rosetta Calculations

Protein structure determinations were performed using the Loop Builder application of the Rosetta protein folding software package, which was compiled and available through the Center for High-Performance Computing (CHPC) at the University of Utah.³⁸ Calculations were performed using resources available through the same department; for a typical structure determination, which involved manipulations of 8–12 amino acids at a time, 5000 structures were generated at an approximate rate of 6 structures/hour/core with optimized parallel computing. Inputs were obtained as PDB files from crystal structures available on the Protein Data Bank and refined by removing lines which Rosetta does not parse (i.e., those representing water molecules in the structure), then using the Rosetta relax function. Outputs were assessed by “score,” a dimensionless metric provided by Rosetta which roughly represents stabilization energy, as well as root-mean-square difference (RMSD) compared to the top-scoring structure in each data set; adequate Monte Carlo sampling of the coordinate space was determined on the basis of a positive slope on a score-vs-RMSD plot of the top 400 structures and qualitatively by the clustering of low-scoring, low-RMSD structures. Further visual analysis of the top 5 scoring structures was performed to ensure these had similar conformations. The best structure as determined under these conditions was then analyzed using UCSF Chimera Molecular Modeling software, often by comparison to the unmodified native fold; residue positions, orientations, and relative distances were determined using tools available from Chimera. This entire process was repeated in triplicate for each structure determination, and the top structure determined in each case was further compared to one another to ensure convergence to similar final conformations.

Plasmid Construction

All chemicals were provided by Sigma-Aldrich unless otherwise noted. Water-containing samples use 18+ MΩ ultrapure water from a Milli-Q Reference Water Purification System. All experiments involving temperature-controlled gene operations were performed using the ProFlex 32-well PCR System. All primer synthesis and sequence verification (performed on all genes after ligation or modification) were performed by the University of Utah DNA Sequencing Core Facility. The genes coding for ADH, aldDH, and f+ were ordered as double-stranded DNA fragments from IDT DNA with 20-nucleotide Gibson assembly ligation regions at the 5′ and 3′ ends. Gibson assembly utilizing a Master Mix from New England Biolabs (NEB) was performed to integrate the genes into the pET-28a(+) plasmid from EMD Millipore. This plasmid features inducible expression by the lac operon and confers kanamycin resistance. The f+ plasmid was modified using the Q5 Site-Directed Mutagenesis Kit from NEB to produce the f0 plasmid.

All centrifugation operations were performed using the Eppendorf 5424R, Beckman-Coulter Allegra X-12R, or Beckman-Coulter Avanti J-E centrifuge, depending on scale and speed requirements. The Fisher Scientific Isotemp 637F Incubator was used for cell culture plate incubations. All experiments involving cell culture incubation with shaking were performed using a Thermo Scientific MaxQ 4000 or Stackable 6000 Incubated/Refrigerated Shaker. Transformant selection was performed using kanamycin from Streptomyces kanamyceticus at 50 μg/mL in all gel and liquid cultures. NEB DH5-α E. coli cells were transformed using the plasmids, and colony screening polymerase chain reaction (PCR) was performed to confirm the presence of the insert; briefly, portions of colony were combined with Q5 HotStart High-Fidelity 2× Master Mix from NEB and primers using the provided standard protocol to amplify the insert region, then analyzed for mass on a 0.8% agarose gel using Gel Loading Dye, Purple (6×) from NEB, GelRed Nucleic Acid Gel Stain from Biotium, and the 1 Kb Plus DNA Ladder from Invitrogen. Plasmids from colonies found to contain the insert were purified using the Thermo Scientific PureLink HiPure Plasmid DNA Purification Kit, then sequenced to confirm correct ligation of the insert (Table S1).

Protein Expression

Purified, sequence-verified plasmids were introduced into Overexpress C41(DE3) Chemically Competent Cells (E. coli) following established protocols. This strain was found to be most effective at reducing inclusion body formation among those tested during subsequent protein synthesis. Transformants were grown in a 1% preculture containing Luria–Bertani growth medium overnight at 30 °C, then added to a larger volume (typically 1.5 L in a large shaker flask), grown to an optical density at 600 nm (OD₆₀₀) of 0.6–0.8 at 37 °C, then induced using isopropyl-β-d-thiogalactoside (IPTG) from Gold Biotechnology at a final concentration of 0.4 mM. The temperature was reduced to 18 °C for subsequent protein expression for 5–8 h, which was found to be effective in the prevention of inclusion body formation. Cells were collected by centrifugation for 10 min at 8000g, then flash frozen using liquid nitrogen and stored at −80 °C until thawing for further operations, where necessary.

Cells were resuspended in 50 mM Tris-hydrochloride, 150 mM sodium chloride buffer (pH = 7.5), which was also used for further experimental steps, and lysed using a M-110P Laboratory Microfluidizer from Microfluidics, operated at 21 000 psi, in three passes of the lysate. Cell debris was removed by centrifugation at 8000g for 10 min, and the clarified lysate was purified using a HisTrap High Performance column (5 mL) with the ÄKTAprime Plus Fast Protein Liquid Chromatography (FPLC) System from Cytiva; briefly, the cell lysate was loaded onto the prewashed column, washed with buffer containing 20 mM imidazole, then gradient-eluted using a second buffer containing 300 mM imidazole. Each eluate fraction was assayed for protein content using a standard Bradford reagent method and for activity using an appropriate oxidative substrate (either ethanol or acetaldehyde) and NAD⁺ (from Research Products International) at saturating concentrations in buffered solution; this represented the reaction with the highest activity for ADH and aldDH (Figure S2 and S3). The increase in absorption at 340 nm, corresponding with the production of NADH, was monitored at 37 °C using a Synergy HTX Microplate Reader, as well as all further NAD(H) production/consumption assays. Active fractions were combined and concentrated using a 10 000 MWCO MilliporeSigma Amicon Ultra Centrifugal Filter Unit, fractionated, and flash frozen using liquid nitrogen. All further experimentation was conducted using freshly defrosted enzyme samples.

Electrochemical Analysis and Fusion Enzyme Bioelectrocatalysis

Ten microliters of 40 mg/mL diaphorase (Toyobo USA, 2.5 U/mg), 20 μL of 10 mg/mL cobaltocene modified poly(allylamine) (Cc-PAA), and 4 μL of 3% (v/v) aqueous ethylene glycol diglycidyl ether (EGDGE) (from Polysciences, Inc.) were combined with varying quantities of f+, f0, ADH, aldDH, or a combination of the two free enzymes in varying ratios. Three microliters of the mixture was drop cast onto Toray paper (0.25 cm²) and dried overnight at room temperature. The electrochemical experiments involving the cobaltocene poly(allylamine)-diaphorase NADH regeneration system described previously were performed with a CH Instruments, Inc. potentiostat (model 1230) with a carbon paper working electrode (0.5 × 0.5 cm), platinum mesh counter electrode, and saturated calomel reference electrode (SCE) in an H-cell setup.³⁹ All electrochemical analyses were performed anaerobically (<1 ppm of O₂) to avoid competitive O₂ reduction at the working electrode.

Staircase plots for Michaelis–Menten parametrization of the enzymes were performed at room temperature (retains >80% of max activity at 25 °C) in 2 mL of buffer solution by establishing a stable baseline current at −0.9 V vs SCE, then injecting 50 μL of 1 M NADH (from Research Products International) as the cosubstrate driving the reduction of the relevant substrate (acetaldehyde for ADH or acetate for all other enzymes); next, 50 μL of 1 M substrate was serially injected, with a new baseline established after each injection, until no further current increase is observed. A final injection of NAD⁺ and observation of a current increase after the experiment is concluded is used to confirm that the NADH regeneration system remains active.

Long-term bulk electrolysis was performed in 3 mL of buffer, with 50 μL of 1 M NADH again added to drive the electrochemical reduction. On the basis of specific activity determinations by staircase amperometric measurements, equal units of each enzyme were used; this represented 8 μg protein for f+, 1.1 mg for f0, 1.7 mg for ADH, and 7.1, 10.7, and 14.3 mg of aldDH for 1:1, 1.5:1, and 2:1 aldDH:ADH ratios, respectively. Once a stable current baseline was established, 3 mL of 1 M acetate are added to provide saturating substrate conditions throughout the 2-day electrolysis. A first sample of 500 μL of the mixture was taken immediately after the addition of acetate and is sealed in an Agilent 2 mL Screw Top GC vial, and a second sample is collected after 2 days of bulk electrolysis have passed.

Gas Chromatography for Acetaldehyde/Ethanol Quantitation

Gas chromatographic analysis of acetaldehyde and ethanol standards used for calibration as well as experimental samples was performed using a Thermo Scientific Trace 1310 GC Instrument; 200 μL headspace samples were manually injected in triplicate using a Hamilton Microliter Syringe onto an Agilent HP-PLOT/Q capillary column (0.320 mm internal diameter, 30 m length), with an inlet temperature of 230 °C, carrier pressure of 8.0 psi (argon gas), purge flow of 5.0 mL/min, and split flow of 10.0 mL/min. The oven temperature was held at 38 °C for 6 min, then ramped to 170 °C over the course of 4.25 min, and finally held at 170 °C for 3.75 min. The flame ionization detector (FID) was operated at 250 °C, with hydrogen, air, and makeup gas flow rates of 35.0 mL/min, 350.0 mL/min, and 40.0 mL/min, respectively. The provided software package was used for peak analysis, with quantities of ethanol (retention time = ∼5.3 min) or acetaldehyde (r.t. = ∼10.75 min) determined by peak area.

All uncertainties are reported as the 95% confidence interval (using Student’s t statistic) unless otherwise noted.

Results

The crystal structures of ADH from E. coli and aldDH from B. cereus from the Protein Data Bank served as initial structures for Rosetta calculations. It is important to note that the purpose of these calculations was to serve as a “best guess” for producing a fusion protein with the features necessary to produce substrate channeling effects, and any quantitation involving structural features is approximate. Through homology studies and detailed analysis of the crystal structures, the catalytic domain of ADH was determined to be comprised of the residues coordinating a Zn atom in a “classical type” interaction of two cysteines (C37 and C145), a histidine (H58), and a water molecule, near the labile proton of the NAD cofactor.⁴⁰ The line of residues along the 37–46 α-helix, aspartate (D47), histidine (H42), and threonine (T39), comprise a hypothetical proton relay similar to that reported in horse liver ADH;⁴¹ glycerol, the substrate used during crystallization studies, was observed near D47 in a complex with the protein, suggesting catalytic activity in the vicinity of this residue. For aldDH, the catalytic domain is comprised of asparagine (N168), cysteine (C300), and the nicotinamide group of the NAD cofactor.⁴² Both proteins were observed as crystals in a tetrameric structure; however, the catalytic domains do not appear to depend on interprotein surfaces for their structure or function, which would potentially be disrupted when the two proteins are fused.

To begin the fusion protein design, the rigid, cation-dense linker sequence bridging the two proteins was fabricated using YASPIN protein secondary structure prediction.³⁶ A motif forming an α-helix comprised of as many of the cationic amino acids lysine (K) and arginine (R) as possible was conceptualized; the most generalizable form meeting these criteria was found to be KKRQKKKRK, with the polar residue glutamine (Q) being necessary to encourage helix formation. Histidine (H), while also a cationic amino acid, was not an ideal choice as it has a low α-helix forming propensity compared to K or R.⁴³ Since secondary structure features primarily manifest according to local interactions, the sequence was continually checked for compatibility with local residues up- and downstream of the linker, and was found to be adequate for α-helix propensity.

With the goal of placing a hypothetical linker polypeptide in very close (<2 nm) proximity to the two active sites of these proteins without disrupting catalytic activity, two strategies were considered to reposition the C- or N-terminus where the linker would be placed: extending the terminus with a domain which would favorably interact with targeted regions of the protein surface, or truncating the terminus, removing small, nonessential protein domains. Both the C- and N-terminus of each protein resides on the surface of the protein, meaning issues with significant structure disruption upon addition of further residues to either terminus was unlikely. Therefore, a scaffold structure was initially conceptualized in which an α-helix domain would be added that would interact favorably with an existing surface-accessible α-helix similar to coiled-coil domains found in natural structures. In ADH, residues 310–322 comprising an α-helix were modified to facilitate isoleucine–leucine pairs with an extension region, KIAALKYK, drawing inspiration from Woolfson et al.;⁴⁴ however, preliminary Rosetta structure calculations indicated that this strategy would prove unsuccessful as the extension region seemed to form more favorable interactions with the solvent or random surface regions, preventing precise reorientation of the terminus. Therefore, further efforts focused on terminus truncation to achieve ideal linker positioning.

In the terminus truncation strategy, the structures of ADH and aldDH were visually assessed to find the shortest sequence of residues which could be removed to position the terminus very close to the active site of each; calculations would be performed on each enzyme separately with the goal of ensuring that the addition of the α-helical, cationic linker domain, KKRQKKKRK, to the repositioned terminus would have the smallest overall impact on the native fold of each protein. In ADH, removal of residues to the 10th through 13th positions from the N-terminus comprising a nonessential domain of the enzyme would move the terminus in the vicinity of the catalytic residue, D47 (Figure 2a). Four new structures were produced from the ADH crystal structure which were truncated to these sites, N-10, N-11, N-12, and N-13, and allowed to refold to their new most-stable conformations after removal of this domain using the Rosetta Relax algorithm. For each, the α-helical linker domain was then added to the repositioned terminus. The Rosetta Loop Builder algorithm was then used to seek out a stable conformation for this linker domain; as part of this process, if the linker was positioned in the vicinity of other regions of the ADH enzyme, those other regions would also be allowed to refold to a minor extent based on the presence of the linker. The retention of the relative orientation of critical residues D47, H42, T39, C37, C145, and H58, representing the proton relay and Zn binding domains of the active site, respectively, was used as a metric of whether the proximity of the linker would cause a reduction or deletion of catalytic activity. Of these truncation sites, N-12 was found to position the nearest cationic residue of the linker in closest proximity to D47 without significant refolding of the catalytic domain, with an interatomic distance from R10 (Nη1) of the linker to D47 (Cγ) of 6.4 ± 0.4 Å (Figure 2b), well within the preliminary goal of <2 nm.

Figure 2.

ADH crystal structure before (a) and Rosetta-calculated structure after (b) modification by truncation and subsequent addition of the cationic linker (cyan) at the N-terminus. Catalytic residue D47 (dark blue), truncation region (lime green), N-10 site (yellow), N-11 site (black), N-12 site (cyan), and N-13 site (magenta) are labeled. A superimposition of structure a is also visible in b.

Compared to ADH, the truncation design to correctly position the linker was much more complicated for aldDH. Removal of sections from the C-terminus could provide access to the putative active site or substrate access port of aldDH by any of five different truncation sites within relative proximity to the terminus: C-456, C-457, C-463, C-474, or C-475 (Figure 3a). It was reasoned that removal of greater portions beyond the 456th residue, which itself represents the removal of over 30 residues, would have too significant an effect on the critical regions of the protein structure (e.g., the NADH binding domain) to be an effective design strategy or to be simulated with any degree of accuracy. Since the Loop Building application of Rosetta is confined to simulations involving small regions of the protein (<12 amino acids) at a time, a series of calculations was necessary to attempt to observe the effect of placing the linker at these five locations. This involved removing a small section (∼5 residues) from the C-terminus, then sequentially simulating all proximal regions of the protein to observe whether significant structure disruption would occur. While greater accuracy could be expected if all affected regions could be simulated simultaneously or if the entire truncation region could be removed at once, we reasoned that this was adequate for a qualitative assessment within the limitations of protein folding simulations. Truncations to these five sites were completed, and the linker was added at the C-terminus for each structure; the best outcome was produced in C-463, where the nearest linker residue K464 (Nζ) was in close proximity (9.9 ± 0.3 Å) to active site residue C300 (Sγ), did not disrupt the relative orientation of the active site pocket (observed for C-456), and did not occupy or “plug” the substrate access port (observed for C-457 and C-475) (Figure 3b).

Figure 3.

AldDH crystal structure before (a) and Rosetta-calculated structure after (b) modification by truncation and subsequent addition of the cationic linker (red) at the C-terminus. Catalytic residues N168 and C300 (dark blue), truncation region from the C-terminus to position 475 and from positions 473–464 (lime green), C-475 site (black), C-474 site (orange), C-463 site (red), C-457 site (magenta), and C-456 site (yellow) are labeled. A superimposition of structure a is also visible in b.

To summarize the findings of the Rosetta calculations, truncation of the N-terminus of ADH up to the 12th position enabled ideal positioning of the cationic linker, KKRQKKKRK, in very close proximity to the ADH catalytic domain without disrupting it. For aldDH, this was accomplished by C-terminus truncation of residues to the 463rd position. A final fusion protein design was therefore fabricated as [aldDH C-463]-KKRQKKKRK-[N-12 ADH] with a C-terminal poly(histidine) tag for purification. A composite structure qualitatively depicting the overall structure of this fusion protein is depicted in Figure 4. This protein, along with a neutral α-helix linked version (LPALMALSV) and the C-terminus His-tagged ADH and aldDH, were expressed at milligram-scale in E. coli C41(DE3) (Figure S1). Preliminary activity assays revealed that the preferred reaction of ADH is ethanol oxidation (Figure S2), while acetaldehyde oxidation is preferred for aldDH (Figure S3). However, the enzymes are reversible and will perform reduction reactions in the presence of NADH. The conversion of C1 or C3 alcohols, aldehydes, and carboxylic acids was not investigated in this study.

Figure 4.

Rosetta-calculated (composite, approximate) structure of the ADH-aldDH fusion protein. ADH catalytic residue D47 (dark blue), aldDH catalytic residues N168 and C300 (dark blue), the linker (magenta), the aldDH region (cyan) and the ADH region (red) are labeled.

A system was devised to demonstrate the merits of this fusion enzyme, particularly the operational performance difference between f+, f0, and the unlinked enzymes. Since substrate channeling is contingent on increasing the propensity of an intermediate interacting with downstream reaction centers, effects will not be observed when the first reaction is fast compared to the second; in that case, the intermediate would accumulate in the bulk and be present in ample amounts at the second active site where the reaction takes place relatively slowly, regardless of channeling effects. Therefore, the reductive direction of operation for the fusion protein was chosen since ADH is catalytically fast compared to aldDH; thus, aldDH is rate-limiting, and the overall turnover frequency is enhanced when the intermediate is channeled from this reaction center to ADH. The maintenance of saturating NADH concentrations for long time periods is necessary to drive this reduction reaction rather than the activity-preferred oxidation; for this reason, enzymes were coimmobilized with diaphorase (DH) and the redox polymer, cobaltocene modified poly(allylamine) (Cc-PAA), which comprised a previously reported bioelectrochemical NADH regeneration system.³⁹

An applied potential of −0.9 V vs SCE supplies sufficient reducing power while minimizing overpotential for NAD⁺ reduction by the DH/Cc-PAA NADH regeneration system. Under these conditions, the apparent activities and substrate affinities for the enzymes were assessed using staircase amperometry to generate a Michaelis–Menten-like kinetics measurement. For ADH, acetaldehyde was serially injected, and acetate was injected for all other enzymes and enzyme combinations; addition of the substrate resulted in enzyme turnover and resultant consumption of NADH, which is rapidly regenerated electrochemically by diaphorase, producing a current response. Since the current response is the result of NADH consumption, we make the simplifying assumption that this occurs as a two-electron process, corresponding to a single molecule of NADH being converted to NAD⁺ per enzyme turnover event. It is important to note that, since both ADH and aldDH are NADH-dependent enzymes, the individual contributions of each catalyzed reaction to the overall activity cannot be parsed apart; that is, some turnover events correspond to acetate reduction to acetaldehyde and others to subsequent acetaldehyde reduction to ethanol, despite enzyme kinetics being assessed on the basis of overall acetate concentration only. A representative staircase amperometric measurement and nonlinear regression analysis are presented for f+ in Figure 5, and the apparent Michaelis–Menten constants (K_M,app), specific activities, and turnover numbers under saturating NADH and acetate conditions are presented in Table 1.

Figure 5.

Current–time staircase response curve for f+ with the serial addition of acetate (a), and the nonlinear regression analysis results for the determination of Michaelis–Menten parameters (b).

Table 1. Michaelis–Menten Parameters for ADH, aldDH, f0, and f+.

	ADH	AldDH	f0	f+
KM,app (mM)	65 ± 9	100 ± 10	80 ± 30	170 ± 20
Specific activity (U mg–1)	0.0173 ± 0.0007	0.0042 ± 0.0007	0.028 ± 0.004	3.8 ± 0.2
Turnover number (s–1)	0.0104 ± 0.0004	0.0039 ± 0.0006	0.040 ± 0.006	5.5 ± 0.3

The large K_M,app values (in the μM scale for typical enzymes) observed are likely a consequence of enzyme coimmobilization, which is known to affect substrate diffusion to the catalyst;⁴⁵ therefore, the values are not likely of physical significance except in comparison to one another. ADH seems to display a lower K_M,app than aldDH, and f0 does not demonstrate a statistically significant difference in substrate affinity compared to ADH or aldDH; this result may be expected if the fusion protein is merely comprised of these two components. In contrast, f+ has a much larger K_M,app, which may be a consequence of a more impactful structural change for this linker compared to the neutral control, such as a partial blockage of the active site port of aldDH by the linker which is not present in f0. However, other effects could account for the differences observed, such as conformational changes caused by substrate binding that affect the binding behavior at the other active site. The Michaelis–Menten treatment of this fusion protein to determine K_M,app, which includes assumptions that the protein is not self-inhibiting, would not be applicable in that case.

Assuming comparable purity, f0 displays about 4-fold activity compared to ADH as determined by their turnover numbers, or the number of molecules of substrate (NADH) converted to product (NAD⁺) per second per enzyme molecule. The addition of equimolar quantities of active aldDH to ADH would contribute a maximum of about a 40% increase under saturating intermediate (acetaldehyde) conditions based on their activity difference. This seems to indicate that the presence of the linker, either by virtue of the fixed proximity of the two active sites or the properties of the linker itself, causes an effective inhibition of the reverse reaction (acetaldehyde oxidation to acetate) and/or enhancement of the subsequent forward reaction (acetaldehyde reduction to ethanol). The results are also consistent with the f0 sample being 400% purer than isolated ADH with a totally inactive aldDH domain; though this is unlikely, it may contribute to the observed activity enhancement to some lesser extent. We postulate instead that the rate enhancement, at least in part, is a consequence of confined diffusion of acetaldehyde in the hydration shell of the fusion protein or due to a modification of protein conformation, in either the bound or unbound states, resulting in more favorable binding conditions and thus greater overall flux through the conversion pathway.²⁵

In the case of f+, a substantial rate enhancement is observed, representing over 500-fold increased activity compared to unbound ADH alone and nearly 140-fold increased activity compared to f0 when similar sample purity is assumed. The same reasoning as for f0 can be applied, with even greater enhancements observed as a consequence of the cationic linker in place of a neutral one: f+ provides a cationic surface bridging the two active sites which participates in ion-dipole interactions with the strongly polar, carbonyl-containing intermediate. This intermediate diffuses in a confined manner rather than reaching equilibrium conditions with the bulk solution, preventing loss to side reactions, which are particularly prevalent in electrochemical systems (e.g., reoxidation by the mediator species). With excess NADH conditions maintained and acetaldehyde accumulation confined to the vicinity of the ADH active site, equilibrium drives the reduction of acetate. For f0, the bridging linker is comprised of the sequence LPALMALSV, amino acids with predominantly hydrophobic side groups. We can therefore conclude that the more favorable interactions between the highly polar acetaldehyde and the charged side groups of the KKRQKKKRK linker in f+ are responsible for the higher observed activity; it is possible that the intermediate cannot efficiently diffuse in a confined manner between the hydration shells of the constituent proteins in f0 as there is no polar or charged surface uniting them. Alternatively, as the linker in f0 is neutral, the fusion protein may take on a very different conformation compared to f+ due to the relative difference in solvation energy.

The preceding discussion relies on the assumption of equal sample purity, which may be reasonable as all samples were purified by an identical method; however, it is very difficult to rule out the possibility that the observation of higher activity for f+ compared to other enzymes is simply a consequence of a greater portion of the f+ protein sample being present in an active form compared to other samples (e.g., by conferred stability due to the presence of the linker which results in slower degradation of the sample). Therefore, rather than utilizing equal quantities of protein for each sample, bulk electrolysis over the course of 2 days was performed on systems containing equal units of enzyme activity. Here, five different reaction vessels were constructed, containing f+, f0, and aldDH:ADH together in 1:1, 1.5:1, and 2:1 unit ratios; the results of this bulk electrolysis are presented in Figure 6.

Figure 6.

Normalized chronoamperometric curves for each enzyme or enzyme combination during bulk electrolysis. Enzyme ratios are written as aldDH:ADH for unlinked systems.

All current densities gradually decrease over time as expected except in the case of f+, behavior which is replicable; there appears to be some conferred stability of this fusion protein due to the cationic linker that is not conferred by the neutral linker in f0, perhaps due to structural or conformational differences related to the hydrophobicity of the linkers, though the definitive reason is unclear. Because of this, the total normalized charge passed by the end of the two-day bulk electrolysis was 315 C cm^–2 U^1– for f+ compared to just 170 C cm^–2 U^1– for 1:1 aldDH:ADH and 125 C cm^–2 U^1– for f0. When f+ is compared to unlinked ADH and aldDH in increasing ratios in these terms, the electrochemical system utilizing f+ provides an equivalent performance to about a 1.8:1 aldDH:ADH ratio (Figure 7). The added stability of the f+ enzyme during bulk electrolysis allows for greater efficiency in terms of protein loading and may allow for larger quantities of ethanol to be formed during longer operation of the electrosynthesis scheme.

Figure 7.

Comparison of normalized charge passed by each enzyme or enzyme combination during bulk electrolysis. Enzyme ratios are written as aldDH:ADH for unlinked systems.

To quantify the production of the acetaldehyde intermediate and ethanol product for the unbound ADH + aldDH, f0, and f+ enzymes, samples were taken immediately after injection of the acetate substrate as well as after the two-day bulk electrolysis experiment was concluded. These samples were then analyzed using GC-FID, and the results are summarized in Table 2. The f+ fusion protein provides the greatest intermediate conversion efficiency, calculated as the portion of total acetaldehyde quantity which was converted to ethanol, of 90 ± 20%; the total quantity of product is also larger for f+ than for f0 or unbound ADH + aldDH, which was expected as f+ also passed the greatest overall charge among these samples. Interestingly, f0 displays improved conversion efficiency compared to the unlinked form; this provides further evidence that a substrate channeling-like effect is present, even when the linker is neutral, and supports the hypothesis that the intermediate may be experiencing favorable intermolecular forces within the hydration shell of the protein surface, leading to an increased probability of diffusion to the subsequent reaction center. Since ADH is catalytically fast compared to aldDH, even the unbound form seems to display a nonzero conversion efficiency, though the result is not statistically significant.

Table 2. Ethanol and Acetaldehyde Quantities Produced during Bulk Electrolysis for f0, f+, and 1:1 ADH:aldDH.

	[acetaldehyde] (mM)	[ethanol] (mM)	conversion efficiency (%)
1:1 ADH:aldDH	0.24 ± 0.04	0.07 ± 0.05	20 ± 20
f0	0.14 ± 0.04	0.17 ± 0.05	50 ± 20
f+	0.16 ± 0.04	1.09 ± 0.05	90 ± 20

Discussion

The extensive rational design of a fusion protein linked in such a way as to facilitate substrate channeling by ion-dipole interactions with a polar intermediate proved worthwhile; assuming equivalent purity to other samples, the fusion protein is orders of magnitude more active than either the neutral-linked or unlinked ADH and aldDH enzymes. This activity of f+ far surpasses that which could be expected when equal stoichiometric amounts of ADH and aldDH are used (about 4-fold, compared to 500-fold); considering the fundamental limitation of the quantity of enzyme which can immobilized on an electrode surface per unit area, this efficiency improvement is significant for bioelectrocatalysis in particular. This fusion protein also was shown to have enhanced stability in long-term electrochemical experiments and to increase total flux through the enzyme cascade, resulting in 1.09 ± 0.05 mM of ethanol being produced by bulk electrolysis in 2 days compared to just 0.07 ± 0.05 mM ethanol for unlinked ADH and aldDH. With the ease of performing DNA manipulations using modern biochemical methods, the neutral-linked control was synthesized in parallel and displayed increased activity and total flux as well, though to a lesser extent than the cationic-linked fusion protein; these results are consistent with other findings where simply the proximity of two enzymes resulted in an increase in overall cascade activity. Further experiments are needed to test the mechanism which we and others propose in which a substrate channeling-like effect is present through intermolecular interactions between the intermediate and the polar surface of proteins, confining the intermediate and increasing the probability of subsequent reaction rather than loss to competing reactions in the bulk solution.

On the basis of the experiments involving Rosetta protein simulations, further work involving substrate channeling using a fusion protein design strategy can be performed with relative ease when a few key factors are considered; if two proteins are linked together, knowledge of the constituent protein structures is highly valuable and will give insight into which termini are most easily functionalized and whether a linker can be placed in sufficiently close proximity to the enzyme active site to facilitate favorable interactions with the intermediate species. Rosetta proved invaluable for providing evidence of activity retention upon addition of the linker in the local vicinity of the active site. It was difficult to predict even with knowledge of the distance of the linker to the active site whether critical residues would refold to accommodate the presence of the linker. While the original design emphasized the importance of a rigid α-helix-forming linker being used to avoid dramatic conformational changes, the relative orientation of constituent proteins to one another seems to be at least partially controlled by surface interactions to which the linker conforms regardless of this consideration.

The ADH-aldDH fusion protein is of particular interest for its potential application in direct carbon dioxide reduction to methanol, a reaction pathway with profound ramifications if it can be efficiently achieved.⁴⁶ While C1 reactants/products were not tested in this study, future work will involve the combination of this fusion protein with an appropriate formate dehydrogenase to attempt this reaction cascade. In addition, we are interested in exploring further linker feature modifications, such as length, amino acid composition, and secondary structure. We expect that an optimized linker could be determined through these investigations, and this may provide even further insight into the nature of substrate channeling and its application in synthetic systems.

Conclusion

Substrate channeling holds promise to improve multistep catalytic reactions in a number of ways. If a particularly toxic or reactive intermediate can be channeled between reaction centers, it does not accumulate in the bulk, where it may make a solution hazardous to workers or catalysts. Side reactions that may inhibit reactions or result in the accumulation of unwanted or low-value byproducts can be avoided. As evidenced here, there is the potential for stability, throughput, and activity advantages by utilizing substrate channeling in a fusion enzyme design, leading to simpler systems that require less protein loading and offer greater efficiency. While further investigations are necessary to fully understand how to apply substrate channeling in synthetic systems to the extent displayed in nature, the work presented here represents an attempt to gain further insight into this valuable phenomenon for the enhancement of multistep catalysis.

Glossary

Abbreviations

C-terminus

carboxyl terminus

N-terminus

amino terminus

DNA

deoxyribonucleic acid

ADH

Escherichia coli alcohol dehydrogenase

aldDH

Bacillus cereus aldehyde dehydrogenase

T_opt

optimal temperature

pH_opt

optimal pH

f+

cationic-linked ADH-aldDH fusion protein

f0

neutral-linked ADH-aldDH fusion protein

GC

gas chromatography

FID

flame ionization detector

CHPC

Center for High-Performance Computing

PDB

Protein Data Bank

RMSD

root-mean-square difference

PCR

polymerase chain reaction

IPTG

isopropyl-β-d-galactoside

NAD(H)

nicotinamide adenine dinucleotide

Cc-PAA

cobaltocene modified poly(allylamine)

EGDGE

ethylene glycol diglycidyl ether

SCE

saturated calomel electrode

N-n or C-n

truncation to the n-th residue from the protein terminus

DH

diaphorase

K_M,app

apparent Michaelis–Menten constant.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.1c00180.

• Sequencing information and primers for plasmid constructions, protein gel electrophoresis results for purified protein fractions, and preliminary activity determinations for ADH and aldDH (PDF)

Author Contributions

M.K. initiated the project, performed the majority of the experiments, analyzed data, and prepared the manuscript. Y.L. performed electrochemical experiments. M.Y. assisted in performing cell culturing experiments. B.A. assisted in gene experiments. J.Z. and E.B. assisted in experiments involving ADH and aldDH synthesis. S.M. conceived, initiated, and supervised the project. All authors participated in project direction and reviewed, edited, and approved the manuscript.

This work was supported by grant FA9550-16-1-0279 from the Air Force Office of Scientific Research (AFOSR).

The authors declare no competing financial interest.