SynFd, CpFd, Zea mays Fd (ZmFd), Anabaena variablis Fd (AnFd), Synechocystis sp. PCC 6803 FNR (SynFNR), and Rice root FNR (RrFNR) were expressed in vivo in Escherichia coli and purified as described previously (Lu et al., 2015). CpI was also expressed heterologously in E. coli and purified as reported in the previous work (Koo and Swartz, 2018).

Nicotinamide adenine dinucleotide phosphate (NADPH), dithionite, glucose-6-phosphate (G6P), G6P dehydrogenase (G6PD), superoxide dismutase (SOD) and catalase were purchased from Sigma-Aldrich.

The O₂ reduction experiments were conducted in 8.4 mL crimp vials where 840 μL reaction mixtures contained the following (unless stated otherwise): 50 mM Tris buffer pH 7.0, 10 mM G6P, 4 units of G6PD, 5.0 mM NADPH, 5.0 or 50 μM FNR, and 5.0 μM of Fd. The reaction mixtures were prepared inside a N₂-only glovebox (CHOA Engineering). Before sealing the vials with rubber septa, magnetic stir bars were added for mixing. After removal from the glovebox, the sealed vials were placed on a stir plate to initiate mixing at 300 rpm. O₂ was introduced to the headspace at t = 0 min (or 10 min when examining various pairs of FNR and Fd) by using a syringe with a 20-gage needle (Daehan Sciences); air or a gas-tight handheld tank filled with O₂ was used as a source. O₂ and H₂ concentrations were measured by sampling 200 μL of the headspace with a valved 23-gage needle (Daehan Sciences), and using gas chromatography (GC 6500, YL Instrument). The H₂ production experiments were done in the same manner with the addition of 10 nM CpI to the aforementioned reaction mixtures inside the anaerobic glovebox.

The amount of H₂¹⁸O formed by the reaction mixtures (50 mM Tris buffer pH 7.0, 10 mM G6P, 4 units of G6PD, 5.0 mM NADPH, 5.0 or 50 μM FNR, 5.0 μM of Fd, 4.2 U SOD and 4.2 U catalase) were measured as follows. The 8.4 mL crimp vials (Thermofisher Scientific) were first opened using a decapper, and the reaction buffer were immediately put in the 80 °C water bath (Daehan Sciences) for the following hour to terminate any ongoing enzymatic reactions. Next, the reaction buffer was distilled at 120 °C and only the water vapor was collected by condensation. The collected samples were mixed with distilled H₂O at two different ratios (2- and 10-fold dilution) and submitted together with a sample containing only distilled H₂O to the mass spectrometry (National Center for Inter-University Research Facilities, Seoul) for analysis of H₂¹⁸O content.

We have previously reported O₂ consumption by SynFd^red during fermentative H₂ production where electrons are extracted from NADPH and delivered to hydrogenases via RrFNR and SynFd (Koo et al., 2016). In this study, we replaced the source of O₂ from air to 10 vol% ¹⁸O₂. This allowed us to directly measure accumulation of H₂¹⁸O in the buffer using mass spectrometry. Since Fd^red was reported to be capable of only uni- and divalent reduction of O₂ (Allen, 1975), we modified the biochemical reaction as described in Figure 1. Instead of adding the hydrogenase enzyme for the NADPH-driven H₂ production (blue arrows from Fd^red), SOD and catalase were added in the absence of the hydrogenase (red arrows from Fd^red) to increase the likelihood that any reactive oxygen species (ROS) generated by Fd^red would be converted into H₂¹⁸O for isotopic detection.

Four reaction mixtures were prepared, incubated with 10 vol% ¹⁸O₂ for an hour, and analyzed to determine (1) consumption of the headspace ¹⁸O₂ and (2) formation of H₂¹⁸O. The concentrations of reagents were as follows: 10 mM G6P, 4 U G6PD, 5.0 mM NADPH, 5.0 μM RrFNR, 5.0 μM SynFd, 4 U SOD, and 4 U catalase in 50 mM Tris buffer, pH 7.0. Incremental addition of reagents in reaction sequence (Figure 1) starting at SynFd clearly confirmed our previous finding (Koo et al., 2016): SynFd^red is responsible for O₂ consumption (Figure 2). The results also suggested that SynFd^red is not capable of reducing O₂ fully into H₂O. This was evident from the observation that a significant amount of H₂¹⁸O was formed only in the presence of both SOD and catalase. The fact that Synechocystis sp. cells have native SOD and catalase is consistent with our interpretation in that these enzymes can rapidly remove harmful O₂ radicals generated by SynFd^red.

Assuming that SOD do not impact the O₂ reduction activity of SynFd^red, we expected to see the decrease in headspace O₂ reduce by half (Figure 1, following the red arrows only from Fd^red) in the mixture containing only up to SOD. This is indeed what we observed: The loss of headspace O₂ after an hour decreased from roughly 2,900 to 1,600 nmoles upon addition of only SOD, approximately 50% reduction considering experimental error. We also expected to see a further reduction in the amount by which headspace O₂ decreases upon addition of catalase to the mixture containing SOD (Figure 1) since O₂ is regenerated in the presence of both. Indeed, the loss of headspace O₂ after an hour further decreased to roughly 1,000 nmoles. The water molecules containing ¹⁸O were only formed when both SOD and catalase were added to the reaction mixture. Furthermore, the amount of H₂¹⁸O produced, 1,500 nmoles, is similar to the stoichiometrically expected value of 1,300 nmoles: Each peroxide molecule generated by SOD will be converted into a water molecule by catalase. Based on the headspace O₂ loss, the average turnover number (TON) of O₂ reduction by SynFd was 11 per minute per reduced ferredoxin. It is important to note that this TON is not physiologically relevant. Fd mostly delivers electrons for NADPH regeneration inside cells while, in our experiments, the SynFd molecules were reduced using the electrons extracted from NADPH to consume O₂.

We next examined how the rate of electron loss to O₂ varies with respect to O₂ partial pressure in the headspace. As discussed in the previous section, SOD and catalase affect the apparent O₂ consumption by regenerating about half of the putative O₂ radicals back to O₂. We thus conducted the following experiments in the absence of SOD and catalase. The concentrations of reagents inside the sealed glass vials were as usual except for RrFNR, which was increased by 10-fold to 50 μM RrFNR. The higher [RrFNR] was chosen in order to enhance electron flux to Fd (K_M of 18 or 39 μM for SynFd and CpFd, respectively) and thereby increase changes in the O₂ peaks for more accurate analysis.

The results summarized in Figure 3 and Supplementary Figure 1 indicated that the O₂ reduction rate, or equivalently the electron loss rate is directly proportional to the headspace [O₂]. A linear relationship between the two was observed for both types of Fd from 1 to 21 vol% O₂ in the headspace of the reactor vial. Within this range, the observed O₂ consumption rate varied from 21 to 255 or 46 to 479 nmoles/min for SynFd and CpFd, respectively. The corresponding specific rate of electron loss to O₂ ranged between 5 and 61 per minute for SynFd (assuming only univalent reduction of O₂) and approximately 1.9-fold higher for CpFd. As expected, increasing the [RrFNR] from 5 to 50 μM increased the specific rate of electron loss, but only by about three-fold.

Based on these observations, we analyzed the kinetic parameters for the reductive O₂ consumption. The following rate law (Eq. 1a) was proposed where the consumption is modeled as a non-enzymatic biochemical reaction between Fd^red and O₂:

where [O₂] refers to the concentration of O₂ in the aqueous buffer solution. For the cases where [Fd^red] is fixed, the rate law further simplifies into the equation only dependent on [O₂] (Eq. 1b). This will be true if the electron flux from FNR to Fd increases as [O₂] increases to maintain [Fd^red] approximately constant. We observe a linear fit for the plot of dO₂/dt versus [O₂] (Figure 3), which suggests that the O₂ reduction by Fd^red follows a pseudo first order rate law under the experimental conditions. The rate constant of this kinetic model (eqn. 2, b = 1), namely k1′, is calculated to be 0.0013 and 0.0025 per minute for SynFd and CpFd, respectively.

A multitude of factors may be responsible for the differences in the O₂ reduction activity of the two types of Fd. To begin with, SynFd harbors one [2Fe–2S] cluster while there are two [4Fe–4S] clusters in CpFd (Bertini et al., 1995; Van Den Heuvel et al., 2003). The former is known to be capable of transitioning between [2Fe–2S]²⁺ and [2Fe–2S]⁺ while the latter can undergo changes among three states—[4Fe–4S]³⁺, [4Fe–4S]²⁺, and [4Fe–4S]⁺ (Yao et al., 2012). The midpoint redox potentials (RE) are similar: −412 and −387 mV for SynFd and CpFd, respectively (Grzyb et al., 2018). The dissociation constants with RrFNR and CpI are also not significantly different. In this regard, the difference in the range and nature of oxidation states appear to be more important with respect to the Fd’s O₂ reduction activity. The results in the subsequent section suggests that the local environment surrounding the Fe–S clusters also matter.

We have previously reported that the rate of O₂ consumption is different when SynFd versus CpFd is used for the NADPH-driven H₂ production pathway (Koo et al., 2016). The rate was approximately twice as fast with CpFd, meaning that more electrons are lost to O₂ reduction instead of being used for H₂ production by the hydrogenase. The choice of CpFd over SynFd would therefore result in a lower NADPH-driven H₂ production efficiency. In this manner, the choice of Fd can impact the efficiency and overall yield of aerobic, biological H₂ production when Fd molecules deliver electrons to the hydrogenase enzyme.

We examined various combinations of Fd and FNR to study how the electron loss to O₂ differs during H₂ production (Figure 1, co-existence of both the blue and red arrow pathways) under 5.0 vol% O₂. Different FNRs were evaluated as well since the electron flux rate through the FNR could influence the steady state [Fd^red] (Bingham et al., 2012; Lu et al., 2015), which in turn would affect the rate of electron loss to O₂. The concentrations of the reagents were as follows: 10 mM G6P, 1 U G6PD, 5 mM NADPH, 5 μM FNR, 5 μM Fd, and 10 nM WT CpI in 50 mM Tris–HCl buffer pH 7.0. The reaction lasted for an hour, and we analyzed the data with respect to (1) anaerobic FNR TON during H₂ production, (2) relative aerobic H₂ yield, and (3) the overall change in the headspace [O₂].

The results (Table 1) indicated that electrons in Fd^red are lost to O₂ at a faster rate with a higher (anaerobic) FNR TON within the NADPH-driven H₂ production (r = 0.82). This was expected since a higher FNR TON means a greater electron flux to Fd. In contrast, there was insignificant correlation between the anaerobic FNR TON and the relative aerobic H₂ yield of the reaction (r = −0.25). This confirmed our previous finding from the CpI mutagenesis study where we reported insignificant relationship between the two variables (Koo and Swartz, 2018). This observation suggests that it may be possible to improve both the electron flux and aerobic H₂ yield by engineering FNR and/or Fd. Table 1 also revealed that AnFd is the most efficient in delivering electrons to the hydrogenase in the presence of O₂. In contrast, CpFd lost the most electrons to O₂ among the four types of Fd studied in this experiment. The difference in efficiency between the two ferredoxins was striking: The overall amount of O₂ consumed by CpFd^red over an hour was as much as four-fold of the amount by AnFd^red.

AnFd harbors one [2Fe–2S] cluster with the midpoint RE potential of −405 mV (Jacobson et al., 1993). This is not much different from ZmFd, for example, which also features one [2Fe–2S] cluster and the midpoint RE potential of −390 mV (Shinohara et al., 2017). The difference in terms of electron leakage to O₂ reduction during H₂ production, however, is nearly two-fold. The slightly higher midpoint RE cannot explain this difference since SynFd loses more electrons to O₂ in spite of the lower midpoint RE potential (−412 mV) than AnFd (Table 1). These results indicate that other factors affect the partitioning of the electron flux between H₂ production and O₂ reduction. Given the substantial difference for the same type of Fd when a different FNR is used, we suspect that the binding dynamics is one of the factors influencing the partitioning. The reported binding affinities between native Fds and FNRs used in this study range between 15 to 50 μM (Lu et al., 2015; Shiigi, 2015). No obvious relationships between these affinities and the FNR TON or electron flux to O₂ reduction were observed, likely owing to the complex nature of multi-component interactions in our assay (Figure 1). A fully exhaustive study on the binding interactions between each pair within the assay may help to elucidate the electron partitioning phenomena.