All cloning was done in E. coli DH5α. Hydrogenase genes from Chlamydomonas reinhardtii and the ferredoxin I gene from Spinacia olearcea were commercially synthesized by Codon Devices (Cambridge, MA), codon optimized for expression in Saccharomyces cerevisiae and acceptable for use in E. coli for wide applicability (see Additional file 1, figure S1 for nucleotide sequences). Hydrogenases from Clostridium acetobutylicum and Clostridium saccharobutylicum were cloned from plasmids received from Matthew Posewitz (National Renewable Energy Laboratory, Golden, CO). Hydrogenase genes HydA and HydB were cloned from Shewanella oneidensis using colony PCR of bacterial cultures from Colleen Hansel (Harvard University, Cambridge, MA). Thermotoga maritima HydA was cloned from genomic DNA provided by Kenneth Noll (University of Connecticut, Storrs, CT). PFOR and ferredoxin 27 were cloned from Clostridium acetobutylicum genomic DNA (ATCC, Manassas, VA). Zea mays ferredoxin was cloned from genomic DNA isolated from corn using DNeasy Plant Mini Kit (Qiagen, Valencia, CA). PFOR from Desulfovibrio africanus was isolated from plasmid pLP1 28 provided by Laetitia Pieulle (Centre National de la Recherche Scientifique, Marseille, France) and ydbK was obtained through colony PCR of E. coli BL21. Plasmids for tetracycline-responsive expression of synthetic scaffold proteins were provided by John Dueber 26 (University of California, Berkeley), and ferredoxin/hydrogenase fusions to metazoan GBD, SH3, and PDZ ligand domains were constructed, bridged by flexible (Gly₄Ser)_x linkers. Oligonucleotides for metazoan ligand domains and linker sequences were purchased from Integrated DNA Technologies, with ligand sequences identical to those previously reported 26 .

Site directed mutagenesis of the Clostridium and Shewanella hydrogenase genes to remove restriction sites needed for cloning and for alteration of C. reinhardtii HydA ferredoxin binding surface was carried out using the Quikchange Multi-Site Directed Mutagenesis Kit according to manufacturer instructions (Stratagene, La Jolla, CA). Cloning and mutagenesis primers are listed in Additional file 1, Table S1 (Integrated DNA Technologies, Coralville, IA)

Cloning of hydrogenase-ferredoxin fusion proteins was done using BioBrick standard assembly 29 and subsequently cloned into Novagen Duet vectors (Novagen, Gibbstown, NJ) whose multiple cloning sites were mutated to accept BioBrick parts (Additional file 1, figure S2). Strep-II tag (WSHPQFEK), (Gly₄Ser) ₂, and (Gly₄Ser) ₄ oligonucleotides with flanking BioBrick sites were purchased from Integrated DNA Technologies. Longer (Gly₄Ser) linkers were made through BioBrick fusion of multiple linker sequences or through PCR amplification from other chimeric proteins 30 .

All protein expression and hydrogenase activity assays were performed in E. coli BL21 (DE3). Cells were transformed with modified pCDF-duet with C. reinhardtii HydEF in MCS1 and C. reinhardtii HydG in MCS2, and with modified pACYC-duet with C. acetobutylicum PFOR or E. coli ydbK in MCS1 or Desulfovibrio vafricanus PFOR cloned into the downstream NdeI and AvrII sites of MCS2. Hydrogenase/ferredoxin pairs were transformed either in each multiple cloning site of modified pET-duet, or for the S. oneidensis hydrogenase HydA in MCS1, HydB in MCS2, and ferredoxin in MCS1 of modified pCOLA-duet. To compare in vitro hydrogen production using maturation factors from Clostridium acetobutilicum, we used plasmids provided by Matthew Posewitz (pCDF-duet with CaHydE in MCS1, CaHydF in MCS2 and pET-duet with CaHydA in MCS1 and CaHydG in MCS2 18 ). Artificial scaffolds were expressed from pJD plasmids provided by John Dueber, previously described in Dueber et. al. 2009 26 .

Hydrogenase knockout (ΔhycE, ΔhyaB, ΔhybC) and Δfpr, ΔydbK, Δhcr, ΔyeaX, ΔhcaD, or ΔfrdB single deletion strains were made by sequential P1 transduction from the Keio collection 31 into BL21(DE3) ΔtonA, followed by removal of the Kan^R marker by standard procedures.

E. coli cells were lysed with Bacterial Protein Extraction Reagent (B-PER, Pierce, Rockford, IL), protein samples were normalized using the Bradford assay (Bio-Rad, Hercules, CA), diluted into SDS-PAGE loading buffer and loaded onto a 4-20% Tris/glycine/SDS acrylamide gel. α-Strep-tag II antibody (HRP-conjugated, Novagen, Gibbstown, NJ) or α-ferredoxin primary antibody (Agrisera, Vännäs, Sweden) and α-Rabbit IgG secondary antibody were used.

Bacterial cultures were grown aerobically for two hours until reaching an OD₆₀₀ of approximately 0.15 in LB media with appropriate antibiotic (50 μg/ml ampicillin, 25 μg/ml spectinomycin, 25 μg/ml kanamycin, and/or 12.5 μg/ml chloramphenicol) in 40 ml glass serum vials, induced with 1 mM IPTG (and 2 μg/ml anhydrous tetracycline when relevant for the induction of scaffold proteins) and sparged with argon. For the methyl viologen assay, adapted from King et. al. 18 , vials were sparged for 2 hours and then lysed with assay buffer containing 50 mM Tris pH 7.0, 50% B-PER, 10 mM methyl viologen (Sigma, St. Louis, MO) and 50 mM sodium dithionite (Fisher, Pittsburgh, PA), the vials were capped with rubber septa, the cells were vortexed and allowed to rock overnight at room temperature. Hydrogen concentration in the headspace gas was measured by gas chromatography (Shimadzu GC-14A). In vivo hydrogen production assays were performed in a similar fashion, except that cultures were supplemented with 0.5% glucose at the time of IPTG induction, sparged for 30 minutes and simply capped and shaken ovenight at 37°C before measuring headspace gas composition. Glucose curves were measured in cells pretreated overnight with 1 mM IPTG then immediately diluted into LB + variable glucose + 1 mM IPTG, then sparged and grown overnight. All hydrogen production values were normalized to an OD₆₀₀ of 0.15.

Homology model of C. reinhardtii HydA1 was made using the SWISS-MODEL 32 server with the Clostridium pasteurinium HydA X-ray structure (1FEH 24 ) as a template.

To create a synthetic electron metabolism circuit with hydrogenase as the terminal electron acceptor, we first investigated the activity of various hydrogenase genes heterologously expressed in the presence of appropriate maturation factors. We adapted a previously established in vitro hydrogenase activity assay 18 , and measured hydrogen production from crude lysates of bacteria expressing hydrogenases and maturation factors from several species in the presence of a chemical electron donor, methyl viologen. Previous reports have shown that the hydrogenase maturation factors from C. reinhardtii, HydEF and HydG, are unstable when heterologously expressed in E. coli 18 , likely due to the genes' high GC content, while the maturation factors from Clostridium acetobutylicum were able to mature [FeFe]-hydrogenases from a wide range of species. Using commercially synthesized, codon optimized maturation factors from C. reinhardtii we were able to alleviate the instability of the gene constructs. We found that in vitro hydrogen production from the Clostridium acetobutylicum hydrogenase was identical when coexpressed with the synthetic maturation factors or with HydE, HydF, and HydG from C. acetobutylicum (data not shown). All subsequent experiments were performed using the optimized C. reinhardtii maturation factors.

We compared the in vitro hydrogen production of [Fe-Fe] hydrogenases from Clostridium acetobutylicum, Clostridium saccharobutylicum, Chlamydomonas reinhardtii, Shewanella oneidensis, and Thermotoga maritima, all of which are homologous in their catalytic domain (Additional file 1, figure S3). All hydrogenases except HydA from Thermotoga maritima could be expressed at a high level in E. coli (figure 2A), and were functional in vitro (figure 2B). Hydrogen levels increased linearly for the first several hours of measurement (data not shown), and we found that levels of hydrogen gas in the headspace after overnight incubation correlated to the relative rate of hydrogenase activity during this linear phase. Our overnight in vitro results agree with previous reports of in vitro hydrogen production rates, with the hydrogenases from Clostridium species producing the highest levels of hydrogen 18 . The heterologously expressed hydrogenase from Shewanella oneidensis is functional at relatively low levels in vitro when both subunits are coexpressed in E. coli with maturation factors from C. reinhardtii.

The in vitro assay is useful to test and compare the activities of heterologously expressed hydrogenase genes, but as the assay uses an exogenous reducing agent, it does not provide information on the electron flux within normal metabolic pathways in vivo. To measure electron flux in vivo as a function of hydrogen production, hydrogenase activity must be integrated into a functional electron transfer pathway. One well established class of electron donors to hydrogenases are ferredoxins, small soluble proteins that contain iron-sulfur clusters. Construction of a system where hydrogenase activity depends on electron transfer from ferredoxin would allow for comparison to in vitro data to provide information on hydrogenase behavior and hydrogenase-ferredoxin interaction dynamics.

To produce hydrogen in vivo from glucose, the [FeFe]-hydrogenase was coexpressed with its required maturation factors, ferredoxin, and pyruvate-ferredoxin oxidoreductase (PFOR) from different species. In this heterologous circuit, PFOR oxidizes pyruvate to acetyl-CoA, reducing ferredoxin, which then transfers the electron to the hydrogenase. In normal E. coli metabolism, the oxidative breakdown of pyruvate to acetyl-CoA is performed either aerobically by the pyruvate dehydrogenase complex, reducing NAD⁺, or anaerobically by pyruvate formate lyase, generating formate (figure 1A). PFOR functions in certain anaerobic bacteria and in eukaryotic parasites that possess hydrogenosomes, organelles evolutionarily related to the mitochondrion that generate a proton gradient through the production of hydrogen gas 33 . PFOR is an attractive electron source for a synthetic hydrogen production circuit as overexpression of a putative E. coli PFOR homolog, YdbK, increases in vivo hydrogen production by heterologously expressed [FeFe]-hydrogenase and ferredoxin 20 , PFOR purified from Clostridium pasteurianum has been shown to reduce a number of ferredoxins in vitro 34 , and functional PFOR from Desulfovibrio africanus has been recombinantly expressed in E. coli 28 .

Consistent with the establishment of a synthetic electron transport circuit in vivo, we observed high levels of glucose-dependent hydrogen production upon coexpression of PFOR, hydrogenase and its maturation factors, and ferredoxin all from Clostridium acetobutylicum in an E. coli strain lacking endogenous hydrogenases (ΔhycE, ΔhyaB, ΔhybC, figure 2C). Hydrogen production was again measured after overnight incubation, as we found that hydrogen production in vivo from glucose was exhausted after 16 hours (data not shown). We were unable to detect hydrogen production in the parental strain of E. coli with the native hydrogenases deleted. Removal of any individual pathway component from the synthetic circuit drastically reduced in vivo hydrogen production. However, as has been previously reported, there was a small background level of hydrogen production from expression of hydrogenase and maturation factors alone 35 . Consistent with previous results 19 , we found this background hydrogen production was slightly increased upon overexpression of ferredoxin in addition to hydrogenase, indicating that there are E. coli proteins capable of reducing both hydrogenases and plant-type ferredoxins, several candidate proteins of which we deleted in the following section (figure 3).

The hydrogenase-ferredoxin-PFOR pathway constitutes a modular system, where each element can be exchanged with homologous genes from different organisms. By coexpressing pathway enzymes from diverse microorganisms, we were able to compare the relative interaction strengths of four hydrogenases, three ferredoxins (figure 2D), and three PFORs (figure 2E). All ferredoxins were able to transfer electrons between PFOR and hydrogenases from different species with varying levels of efficiency.

In vivo hydrogen production from circuits expressing each of the four hydrogenases (C. acetobutylicum, C. saccharobutylicum, C. reinhardtii, and S. oneidensis) followed the same trend as the in vitro experiments, with the highest hydrogen production observed with the clostridial hydrogenases (figure 2D). The relative interaction and electron transfer rates for hydrogenase and ferredoxin were explored by comparing the in vivo hydrogen production of circuits made up from all pairwise combinations of the four hydrogenases and ferredoxin from C. acetobutylicum, Spinacea olearcea, and Zea mays and the PFOR from C. acetobutylicum (figure 2D). All hydrogenases produced the highest output when co-expressed with bacterial type 2-[4Fe-4S] ferredoxin from Clostridium acetobutylicum, with a potential of -420mV 27 . Intermediate levels of hydrogen were produced using leaf-type [2Fe-2S]-ferredoxin I from spinach, S. olearcea (-420 mV 36 ) while the homologous root-type ferredoxin III from corn, Z. mays (-345 mV 36 ) led to significantly lower in vivo hydrogen levels in all cases. Interestingly, the difference in hydrogen production from circuits expressing bacterial versus plant-type ferredoxins was more significant for hydrogenases from bacterial species. Hydrogenase from C. reinhardtii, which naturally pairs with plant-type ferredoxins, produced similar levels of hydrogen when co-expressed with ferredoxin from C. acetobutylicum or S. olearcea (figure 2D).

The interaction of overexpressed PFOR from C. acetobutylicum, D. africanus, or the PFOR homolog YdbK from E. coli with the three ferredoxins was compared in a similar fashion in circuits containing the C. acetobutylicum hydrogenase (figure 2E). Overexpression of PFOR from C. acetobutylicum and YdbK from E. coli led to similar levels of hydrogen production, although surprisingly, the highest levels of hydrogen produced from YdbK occurred when it was coexpressed with plant-type ferredoxin from S. olearcea. Overall, the highest levels of hydrogen production were seen with the PFOR from D. africanus, coexpressed with the hydrogenase and ferredoxin from C. acetobutylicum.

Natural biological electron transfer circuits are insulated to prevent electron leaks that can cause damage by creating oxygen radicals and insulated from one another to prevent "short circuiting" 37 . We sought to insulate our hydrogen producing circuit from competing metabolism to improve levels of hydrogen production and to better understand natural biological pathway isolation, a priority for the design of synthetic metabolic pathways. Although our constructed pathway is made up of genes that are divergent from E. coli metabolic enzymes, given the non-specific electrostatic interactions that mediate many ferredoxin interactions 5 , native iron-sulfur proteins may interact with the proteins of the heterologous pathway. This is evidenced by the background hydrogen production in strains expressing only heterologous hydrogenases and ferredoxins (figure 3). Deletion of these potentially competing redox interaction partners should improve pathway function. To address these issues, we deleted six genes identified through their homology to plant-type ferredoxins or ferredoxin oxidoreductases that still allowed for viability (fpr, flavodoxin:NADP⁺ reductase 19 ; ydbK, the putative PFOR homolog 19 ; hcr, an NADH oxidoreductase; yeaX, a predicted oxidoreductase; hcaD, ferredoxin:NAD⁺ reductase; and frdB, fumarate reductase. Additional file 1, figure S4).

These six deletions were tested individually in a ΔhycE, ΔhyaB, ΔhybC background while expressing hydrogenase from C. acetobutylicum and maturation factors from C. reinhardtii, ferredoxin from S. olearcea, with or without co-expression of PFOR from D. africanus. Deletion of fpr and ydbK have been previously shown to slightly decrease the background level of hydrogenase activity in vivo 19 . We found that only the ydbK deletion had any significant effect on hydrogen production compared to the hydrogenase knockouts alone. The background level of hydrogen production from HydA and ferredoxin expressed alone was decreased by half in the ydbK deletion strain, whereas hydrogen production from the full pathway with the D. africanus PFOR was increased by 1.4 fold (figure 3). This is consistent with our finding that overexpression of ydbK led to high levels of electron transfer when co-expressed with ferredoxin from spinach, indicating that endogenous ydbK is able to disrupt the synthetic electron transfer pathway.

As an independent strategy, we attempted to insulate the pathway from competing electron metabolism through modification of the interaction surface of the hydrogenase and ferredoxin by rational protein design. Co-evolution of interacting protein pairs for preferential binding between natural partners likely plays a large role in the isolation of natural pathways 38 , and this principle has been used in designing synthetic signal transduction systems 22 . To test whether mutations in a component of the artificial pathway specifically enhanced activity by improving the ferredoxin-hydrogenase interaction, we compared in vivo activity, which is ferredoxin-dependent, with in vitro activity, in which a chemical reducing agent, methyl viologen, drives hydrogen production. This in vitro assay thus measures ferredoxin-independent hydrogen production, reflecting the activity and expression level of the hydrogenase itself.

The interaction between ferredoxins and clostridial hydrogenases is poorly characterized, with evidence that more than simple electrostatic reactions may play an role in mediating the transfer of electrons 34 . In contrast, the interaction surface between the hydrogenase from Chlamydomonas reinhardtii and its cognate ferredoxin has been extensively modeled in silico, with evidence that this interaction has a strong electrostatic component 39 40 . Long et. al. 25 proposed a structural model for the interaction between the hydrogenase HydA2 from Chlamydomonas reinhardtii and its cognate [2Fe-2S] ferredoxin, suggesting several mutations that might enhance this interaction due to improved charge complementarity (figure 4A). Ferredoxin is rich in negatively charged residues, and the mutations, E5K, P2K, M119K, or D126K are designed to increase the positive charge of the hydrogenase binding surface. We tested these mutations using HydA1 from C. reinhardtii, and ferredoxin from spinach, both of which are closely related to the proteins studied in the in silico model (Additional file 1, figure S5). We found that two mutations in HydA1, D126K and E5K, improved in vivo hydrogen production while in vitro these mutations showed less or no effect (figure 4B). As in vitro hydrogen production was closely correlated to hydrogenase expression level (data not shown), the improvement in in vitro hydrogen production that was seen for E5K may be the result of increased protein expression.

Many metabolic electron transfer pathways are insulated through the physical scaffolding of protein components in the membrane, for example in the electron transport chain of mitochondria or chloroplasts. Additionally, some electron transport proteins themselves are built from combinations of modular electron binding domains, including the hydrogenase from Clostridium 24 . We sought to isolate our synthetic pathway and improve hydrogen production through physically linking the hydrogenase and ferredoxin in order to increase the chance of binding and electron transfer between the desired partners. We were able to show improved function of the artificial pathway through genetic fusion of the hydrogenase and ferredoxin. Chimeras between ferredoxin and various ferredoxin reductases have been shown to be functional in vitro 41 , with improved electron transfer rates presumably due to the increased local concentration of reduced ferredoxin. We fused the hydrogenase from C. acetobutylicum (figure 5A) with ferredoxin from spinach (figure 5B & C) or from C. acetobutylicum (figure 5D) using flexible protein linkers of various lengths.

Our hydrogenase-spinach ferredoxin chimeric proteins were all expressed at a high level in E. coli (figure 5B) and active at identical levels in vitro (figure 5C), regardless of the orientation of the fusion or linker length, consistent with the lack of a requirement for ferredoxin in the in vitro assay. In vitro data for fusions with the C. acetobutylicum ferredoxin followed a similar trend (data not shown). However, in vivo activity of the fusion proteins when coexpressed with the PFOR from D. africanus depended on linker length as well as overall configuration. Fusions of the spinach ferredoxin to the hydrogenase C-terminus displayed decreased function when the linker is very short, a nearly five-fold improvement at intermediate length linkers, and activity falling gradually to the level of that of separate proteins at longer lengths. This length dependent behavior is consistent with models of changes in local concentration of reactants due to enzyme fusion 42 . Protein fusion to the C-terminus of the spinach ferredoxin, whether the hydrogenase or a short protein tag abrogated in vivo hydrogenase activity entirely (data not shown), likely due to the proximity of the ferredoxin C-terminus to the iron-sulfur cluster 41 .

The behavior of the fusion protein is similar with the ferredoxin from C. acetobutylicum, although the bacterial ferredoxin is equally active when fused at either terminus (figure 5C). When fused to the hydrogenase C-terminus, the linker-length dependent activity displays similar characteristics to the spinach ferredoxin. Interestingly, when fused to the hydrogenase N-terminus, in vivo activity with shorter linker lengths is improved. The putative ferredoxin-binding region is on the N-terminal domain of the hydrogenase (figure 5A), which includes all of the F-clusters that transfer electrons from the surface to the active-site H-cluster. Fusion with a short linker at the C-terminus makes it impossible to reach the N-terminal domain, resulting in the decreased activity compared to unfused proteins. At the N-terminus, however, a short linker still allows for interaction between the hydrogenase and its fused ferredoxin, resulting in increased activity.

In signal transduction, complex protein scaffolds isolate pathways by localizing pathway components into a complex, directing the flow of information. These scaffolds can be rewired in their natural contexts in eukaryotic cells in order to alter the output behavior of signaling cascades 43 44 . Synthetic, modular scaffold proteins have been implemented in E. coli in order to direct flux in synthetic metabolic pathways, improving pathway output by up to 77-fold 26 . These synthetic scaffolds are built from modular scaffold domains from eukaryotic signal transduction--PDZ, SH3, and GBD domains--which tightly bind cognate ligand peptides that can be fused to any protein of interest for localization to the scaffold. We imported these scaffold designs into our synthetic pathway and found that scaffolding of the hydrogenase and ferredoxin dramatically affected the function of the pathway.

Scaffolding of metabolic pathways with small-molecule intermediates can lead to a "pipeline" effect, where the increase in the local concentration of upstream intermediates can speed up the reaction. This can significantly affect the pathway output, particularly when the chemical intermediates of the reaction pathway are toxic to the cell and increased throughput can lead to cell death if the intermediate is not rapidly converted by the next enzymatic step in the pathway 26 . Instead of small molecule intermediates, our synthetic pathway relies on protein-protein interactions, as is the case in many signal transduction cascades. By channeling electron transfer through scaffolded interactions, the flux through the synthetic circuit can potentially be significantly increased in an insulated manner. Because the tertiary structures of the synthetic scaffolds have not been determined, however, it is also possible that the requirement of protein-protein interaction for pathway function may lead to a decrease in pathway flux due to non-optimal insulation of interacting partners. We sought to characterize the effects of different scaffold structures, ligands, and linker lengths on the function of the synthetic PFOR-ferredoxin-hydrogenase circuit with ferredoxin and hydrogenase localized to the synthetic scaffold. All experiments were performed using the PFOR from D. africanus, and ferredoxin and hydrogenase from C. acetobutylicum.

The scaffold is an artificial protein built up of several modular peptide binding domains. The GBD binding domain is at the N-terminus, followed by the SH3 domain and the PDZ domain at the C-terminus. The number of domains in each case is variable, and in Dueber et. al., the ratio of the domains to one another made significant differences in flux depending on the stoichiometry of the reactions in the synthetic pathway 26 . Although multiple diffusion-limited metabolic pathways could be enhanced using this design 26 , artificially scaffolded redox pathways have not yet been investigated. While we were primarily interested in exploring how different configurations of scaffold domains and linker lengths affected the interaction of the redox proteins and therefore hydrogen output, we measured a three-fold improvement of hydrogen production in the scaffolded vs. non-scaffolded conditions when all of the proteins were expressed off of the lower activity pTet promoter, leading to a decrease in the absolute value of hydrogen production when compared to the Duet vectors (data not shown).

The ratio of the different scaffold domains, the ligand bound to the pathway components, and the length of the linker between the ligand and the ferredoxin protein all had significant effects on the output of the synthetic circuit (figure 6). Because ferredoxin and hydrogenase need to physically interact for the circuit to function, suboptimal configurations for the protein scaffold could conceivably sequester these proteins from one another. Indeed, we found that hydrogen output was decreased when the hydrogenase and ferredoxin were bound farther from one another along the length of scaffold (figure 6A). Furthermore, the length of the linker connecting ferredoxin with the SH3 ligand also significantly affected the ability of the hydrogenase and ferredoxin to interact while bound to the scaffold. Increasing the linker length from five amino acids to twenty led to a 3-5 fold increase in hydrogen output from the scaffolded circuit (figure 6B). Linkers of intermediate length produced intermediate pathway output.

The ratio of GBD, SH3, and PDZ domains that made up the synthetic scaffold also significantly affect the function of the pathway in some cases. The stoichiometry of the hydrogen production reaction requires two ferredoxins for the reduction of a single hydrogen molecule, so it might be expected that scaffolds that localize more ferredoxin molecules will increase flux through the hydrogenase. Unfortunately, however, PFOR is substrate limited 45 , with increasing concentrations of ferredoxin leading to decreased enzymatic activity. Increasing the ratio of ferredoxin:hydrogenase bound to the scaffold decreases the output from the synthetic pathway (figure 6C). This effect was also seen when ferredoxin genes were fused in tandem without a scaffold, with output from the hydrogen production circuit steadily dropping with each added ferredoxin (figure 6D). This strategy may, however, increase pathway flux in other synthetic electron transfer pathways when the electron donor is not substrate limited.