The flavodoxin FldA activates the class Ia ribonucleotide reductase of Campylobacter jejuni

Campylobacter jejuni is a microaerophilic zoonotic pathogen with an atypical respiratory Complex I that oxidizes a flavodoxin (FldA) instead of NADH. FldA is essential for viability and is reduced via pyruvate and 2‐oxoglutarate oxidoreductases (POR/OOR). Here, we show that FldA can also be reduced by FqrB (Cj0559), an NADPH:FldA reductase. An fqrB deletion mutant was viable but displayed a significant growth defect. FqrB is related to flavoprotein reductases from Gram‐positive bacteria that can reduce NrdI, a specialized flavodoxin that is needed for tyrosyl radical formation in NrdF, the beta subunit of class 1b‐type (Mn) ribonucleotide reductase (RNR). However, C. jejuni possesses a single class Ia‐type (Fe) RNR (NrdAB) that would be expected to be ferredoxin dependent. We show that CjFldA is an unusually high potential flavodoxin unrelated to NrdI, yet growth of the fqrB mutant, but not the wild‐type or a complemented strain, was stimulated by low deoxyribonucleoside (dRNS) concentrations, suggesting FldA links FqrB and RNR activity. Using purified proteins, we confirmed the NrdB tyrosyl radical could be regenerated in an NADPH, FqrB, and FldA dependent manner, as evidenced by both optical and electron paramagnetic resonance (EPR) spectroscopy. Thus, FldA activates RNR in C. jejuni, partly explaining its essentiality.

, and C. jejuni has been designated a World Health Organization (WHO) priority organism for developing novel antimicrobial agents. New insights into C. jejuni biology are needed if control measures are to be put in place that more effectively reduce chicken colonization and/or food-chain contamination.
C. jejuni is a microaerophile, adapted for growth at low oxygen niches in the host, with most strains unable to grow at atmospheric oxygen levels. Compared with conventional aerobes, C. jejuni is unique in utilizing oxidant-labile enzymes in central metabolic pathways, which are critical for growth (Kendall et al., 2014). In particular, it employs pyruvate and 2-oxoglutarate: acceptor oxidoreductases (POR and OOR), rather than oxygen-stable NAD-linked 2-oxoacid dehydrogenases, to transfer substrate-derived electrons to the respiratory chain (Taylor & Kelly, 2019;Weerakoon & Olson, 2008).
These enzymes contain Fe-S clusters vulnerable to oxidative damage; we have previously shown that exposure of C. jejuni cells to prolonged aeration causes their inactivation in vivo, which may be a major contributor to the microaerophilic phenotype (Kendall et al., 2014). Conversely, we could not demonstrate consistent growth under strictly anaerobic conditions, even in the presence of alternative electron acceptors that do allow growth under oxygenlimiting conditions (Sellars et al., 2002). This seems to be due to the presence of a single class Ia-type ribonucleotide reductase (RNR) that would require catalytic amounts of oxygen to sustain DNA synthesis.
The electron acceptor for OOR has been shown to be the flavodoxin FldA (Cj1382; CjFldA), which has a flavin mononucleotide (FMN) cofactor. FldA is the sole flavodoxin encoded in the genome of C. jejuni NCTC 11168 (Parkhill et al., 2000;Weerakoon & Olson, 2008). Reduced FldA is thought to be reoxidized by donation of electrons from FMN to the membrane bound Complex I of the respiratory chain. In the canonical Complex I found in well studied "model" bacteria, the nuoE and nuoF genes encode an NADHbinding module that delivers electrons to Fe-S centers in the NuoG subunit. However, in C. jejuni, these genes are replaced by two novel genes cj1575c (nuoX) and cj1574c (nuoY), the function of which are unknown (Calderon-Gomez et al., 2017;Smith et al., 2000).
Conceivably, they may form a docking site for reduced FldA to deliver electrons to the Fe-S clusters of NuoG (Taylor & Kelly, 2019;Weerakoon & Olson, 2008) but may also have roles independent of Complex I, as unlike all of the other nuo genes in C. jejuni, nuoY (and probably nuoX) cannot be deleted (Weerakoon & Olson, 2008). It has also not been possible to delete the fldA gene, suggesting that CjFldA must have essential functions (Weerakoon & Olson, 2008).
Apart from allowing acetyl-CoA synthesis via POR, one additional likely vital role is in isoprenoid biosynthesis via the methyl erythritol phosphate (MEP) pathway (Heuston et al., 2012). IspG and IspH are 4Fe-4S enzymes that catalyse the last two steps in the formation of the isoprenoid precursors isopentyl pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP); in other bacteria, both reactions require reduced flavodoxin as the electron donor (Puan et al., 2005;Rohdich et al., 2003).
A study in the closely related Helicobacter pylori (St Maurice et al., 2007) reported that a flavin adenine dinucleotide (FAD) containing and NADPH oxidizing enzyme named FqrB (HP1164) could also act as a flavodoxin reductase in addition to POR. However, based on in vitro studies, these authors proposed that the physiological function of FqrB was in the production of NADPH via the reversed reaction, that is, the transfer of electrons from pyruvate to NADPH via POR, flavodoxin and FrqB. In H. pylori, FqrB is seemingly an essential enzyme as attempts to produce an fqrB deletion mutant were unsuccessful (St Maurice et al., 2007). A homologue of FqrB (Cj0559) is also present in C. jejuni. Here, we show that CjFldA is an unusually high-potential flavodoxin, making NADP reduction unfavorable. Instead, we demonstrate that FqrB provides an additional route for flavodoxin reduction using NADPH as the ultimate electron donor and that both FqrB and FldA are involved in tyrosine radical formation in, and thus activation of, the C. jejuni RNR.

| Redox properties of the C. jejuni flavodoxin: CjFldA is an unusually high redox potential flavodoxin that can be reduced by both POR and OOR
The C. jejuni FldA was overproduced in Escherichia coli as a Cterminally his-tagged protein and purified by nickel-affinity and hydrophobic interaction chromatography (HIC) ( Figure S1a). The aspurified (oxidized) protein has a typical flavodoxin absorption spectrum ( Figure 1a; raw data for all figures in Table S1) with maxima at ~380 and ~460 nm that matched those of a previous study reporting the purification of this protein (Weerakoon & Olson, 2008). The semireduced protein displays a shift of the 380 peak to 360 nm and broad absorbance maxima at ~460 and ~600 nm, which are characteristics of neutral semiquinone ( Figure 1a). The latter was exploited in optically monitored potentiometric titrations (Figure 1b), to define the flavodoxin reduction potentials at pH 7 of −170 mV versus standard hydrogen electrode (SHE) for the oxidized-semiquinone transition (Q/SQ or E 2 ) and −190 mV versus SHE for the semiquinonehydroquinone transition (SQ/HQ or E 1 ). Independent assessment of the redox properties of CjFldA was afforded by cyclic voltammetry, a powerful electrochemical method to analyze the current response of a redox species with a linearly cycled potential (Figure 1c). For these experiments, the flavodoxin was deposited onto the surface of a pyrolytic graphite edge electrode coated with the surfactant didodecyldimethylammonium bromide (DDAB) (Rusling, 1998;Seagel et al., 2017). Cyclic voltammetry revealed two redox processes. The lower potential process had a mid-point potential of approximately − 250 mV, whereas the higher potential process had a mid-point potential of approximately −150 mV. Neither process was present in voltammetry of DDAB-coated electrodes that had not been exposed to flavodoxin ( Figure 1c). Thus, we assign the processes observed with the CjFldA sample to FMN either within FldA or that had dissociated from the protein. Direct electrochemistry of bacterial flavodoxins using solid electrodes has been reported by several groups (Barker et al., 1988;Heering & Hagen, 1996;Seagel et al., 2017). From those findings, there are two points of particular significance for interpretation of the voltammetry of CjFldA. Firstly, peaks for the SQ/HQ transformation are well-defined, whereas those of the Q/SQ transformation are typically absent. Secondly, free FMN typically contributes peaks with a mid-point potential of approximately −200 mV at pH 7. The voltammetry of CjFldA is consistent with Q/SQ and SQ/HQ midpoint potentials above −300 mV versus SHE. This conclusion is in agreement with the results from potentiometric titration. Based on voltammetry of CjFldA measured at several independently prepared electrodes and inspection of the relative peak heights and widths for the two processes, it is likely that the higher potential peaks describe the contribution of free FMN. However, we cannot rule out a contribution from the Q/SQ process at these potentials as noted in the voltammetry of Azotobacter vinelandii flavodoxin II studied at similarly DDAB-coated graphite electrodes (Seagel et al., 2017).
The E 1 and E 2 redox potential values determined from optical potentiometry ( Figure 1b) are remarkably similar, and the E 1 value is far more electropositive than many flavodoxins (Table 1), despite the FMN isoalloxazine binding regions of CjFldA (the W-loop and Y-loop regions; Sancho, 2006) containing several D and E residues that might be expected to destabilize the anionic hydroquinone and shift the E 1 potential to more negative values (Table 1). This is further considered in Section 3.
The ability of FldA to act as electron acceptor for the POR and OOR enzymes was tested by adding cell-free extract plus CoA to the protein under anaerobic conditions and initiating the reaction by addition of either pyruvate or 2-oxoglutarate. A rapid reduction in the FMN absorbance in the 460-nm region was observed with either F I G U R E 1 Redox properties of the C. jejuni flavodoxin FldA. (a) The electronic absorption spectra of the fully oxidized (orange), semireduced (purple) and fully reduced (blue) states of FldA are shown. The spectra were recorded for 50-µM protein in an anaerobic buffer (50-mM potassium phosphate buffer, pH 7.0, and path length = 1 cm). The semireduced state was produced by stoichiometric addition of sodium dithionite. Excess dithionite was added to obtain the fully-reduced state. (b) Optically monitored potentiometric titration of FldA at 610 nm. The data (blue circles) and fit (dashed line) to the equation describing sequential one electron additions to a center with E 2 = −170 mV and E 1 = −190 mV. Sample contained 50 µM protein in an anaerobic buffer (10-mM potassium phosphate buffer, pH 7.0.) Path length = 1 cm. (c) Cyclic voltammetry of CjFldA (continuous blue line) adsorbed on a DDAB-coated pyrolytic graphite edge electrode. Cyclic voltammetry of the DDAB-coated electrode alone is also shown (broken gray line). Measurements performed in 50 mM potassium phosphate, pH 7 at 20°C with a scan rate of 20 mV/s. (d) Reduction of CjFldA by electrons from pyruvate via POR (red trace) or 2-oxoglutarate via OOR (Blue trace) in cell-free extracts monitored at 460 nm, as described in Section 4. The traces are an average of two determinations (raw data in the Supporting Information) substrate ( Figure 1d). Although FldA has previously been identified as the electron acceptor for OOR (Weerakoon & Olson, 2008), these results show that POR also reduces FldA.  Figure 2a shows a phylogenetic analysis of C. jejuni Cj0559 and a range of PNDOR homologs identified by BLAST analysis. It is apparent that the Campylobacterota enzymes, including Cj0559, cluster with a specific group of reductases from Gram-positive bacteria found in Bacillus subtilis (YpdA) and Bacillus cereus/Banthracis anthracis (FNR3) but are distinct from thioredoxin reductases and CoA disulfide reductases. YpdA is a putative bacillithiol disulphide reductase (Gaballa et al., 2010;Mikheyeva et al., 2019), and FNR3 enzymes have been characterized as flavodoxin reductases; B. cereus FNR3 can reduce NrdI, a specific flavodoxin required for formation of the tyrosyl radical in the Mn-containing class Ib RNR's in many Gram-positive bacteria (Lofstad et al., 2016). The FNR1 and FNR2 enzymes also reduce NrdI, with FNR2 being most efficient (Lofstad et al., 2016), but these are more distantly related to FqrB (Figure 2a).

| Cj0559 (FqrB) is an NADPH: flavodoxin reductase related to those in Gram-positive bacteria
We tested the ability of the purified C-terminally his-tagged recombinant Cj0559 to act as an NAD(P)H-dependent reductase for the flavodoxin FldA. C. jejuni Cj0559 was heterologously overproduced and purified ( Figure S1b) as a yellow-colored protein with an absorption spectrum typical of a flavoprotein with maxima at ~375 and ~460 nm ( Figure 2b). In aerobic buffer at pH 7.5, it exhibited a low rate of NADPH oxidase activity (0.79 ± 0.07 µmol min −1 mg protein −1 , n = 3 determinations), but activity with NADH was over Using an estimated extinction coefficient for flavodoxin FMN of 12 mM −1 cm −1 at 460 nm (Mayhew & Tollin, 1992), we determined a rate of 14.5 ± 0.2 µmol FMN reduced min −1 mg FqrB protein −1 with NADPH as reductant (n = 3). These data show that C. jejuni FqrB is an efficient FldA reductase specific for NADPH.

| The growth defect of an fqrB mutant is partially responsive to dRNS
Given the phylogenetic relationship noted above between FqrB and some NrdI reducing PNDOR enzymes in Gram-positive bacteria, and the identification of FqrB as a redox partner for FldA, we sought in vivo evidence that FqrB might be involved in ribonucleotide reduction. Although hp1164 (fqrB) was reported to be essential in H. pylori, we were able to construct a deletion mutant in the cj0559 gene by replacing most of the coding region with a nonpolar kanamycin-resistance cassette via allelic exchange mutagenesis (see Section 4). The mutant was also complemented with a wildtype gene copy inserted at the cj0046 pseudogene locus, driven by the metK promoter. The mutant formed small colonies on plates, and in liquid culture on complex media, it exhibited a significant growth defect compared with the wild-type and complemented strain ( Figure 3a). When grown with a mixture of all four deoxyribonucleosides (dRNS) at 0.1 mM each (final concentration), the growth TA B L E 1 Comparison of redox potentials and FMN binding loop regions in selected flavodoxins and NrdI proteins Note: The FMN is sandwiched between two loop regions known as the 50s or W-loop and 90s or Y-loop in classical flavodoxins, or the 40s loop and 70s loop in NrdI proteins. These contain aromatic/hydrophobic residues (highlighted in orange) that stack over the isoalloxazine ring. A preponderance of anionic residues (blue) in the loops in flavodoxins is thought to contribute to lowering of the E 1 redox potential in particular, whereas in NrdI proteins in the environment is more neutral or charge compensated (cationic residues in red). of the wild-type and complemented strain were not significantly different from that of the controls, whereas the mutant showed an increased growth rate. Interestingly, in the presence of a higher concentration (1 mM each) of dRNS, the growth rate and final cell density of the wild-type and complemented strain were severely inhibited, an effect probably due to an imbalance in the intracellular concentrations of DNA precursors (Torrents, 2014). However, the growth of the fqrB mutant was much less affected ( Figure 3a). The mutant was also significantly less sensitive to the toxic nucleotide analogs 6-thioguanine and 6-mercaptopurine compared with the wild-type and complemented strains ( Figure 3b). Taken together, these results suggest that the lack of FqrB in the mutant affects dRNS uptake and metabolism; a slower conversion of ribonucleotides to deoxyribonucleotides via ribonucleotide reductase might explain the partial growth stimulation by exogenous dRNS. We further investigated the link between FqrB and RNR by in vitro studies with the beta subunit of the RNR.

| Characterization of NrdB, the RNR beta subunit of C. jejuni
Campylobacter jejuni NCTC 11168 encodes a single heterodimeric RNR (Parkhill et al., 2000). The alpha subunit (NrdA; Cj0024) is predicted to catalyse ribonucleotide reduction using a cysteine radical, generated by long-range transfer from a tyrosine radical formed in the beta subunit (Cj0231). In class Ia RNRs, the tyrosine radical in NrdB is generated by oxygen and electrons from two reduced iron centers in a μ-oxo-bridged configuration, whereas in class Ib enzymes that have oxygen unreactive manganese ions instead of iron, superoxide generated by a reduced flavodoxin (NrdI) is used to form the radical in the beta subunit, designated NrdF (Stubbe & Cotruvo, 2011). As noted above, NrdI is itself reduced by a specific reductase of the PNDOR family. The dRNS growth stimulation ob-

C. jejuni
NrdA subunit also has a typical N-terminal ATP-cone domain that is characteristic of a class Ia enzyme (Hofer et al., 2012). Therefore, we wished to test if radical formation in C. jejuni NrdB required flavodoxin, by producing purified radical free NrdB and reconstituting with FldA, FqrB, and NADPH to determine if the radical could be regenerated in vitro.
NrdB was overproduced in E. coli BL21 (DE3) as a C-terminally his-tagged protein and purified by nickel affinity chromatography.
Two bands were consistently seen on SDS-PAGE gels ( Figure S1c iron-or manganese-supplemented media ( Figure 4a). In each case, the molar ratio of Fe/protein (1.0, 1.4, and 0.85, respectively) was significantly higher than the Mn/protein ratio (0.14, 0.05, and 0.25, respectively). Overall, the data confirm that the C. jejuni RNR is indeed an Fe enzyme of the class Ia type and that the database annotation of Cj0231 as NrdF is incorrect. Although the expected ratio of Fe/monomer is 2, a suboptimal stoichiometry is commonly found in heterologously expressed RNRs (Roca et al., 2008). The optical absorption spectrum of the protein shows a typical sharp feature at 410-411 nm diagnostic of the presence of a stable tyrosyl radical as well as weaker charge transfer bands centered around 325 and 370 nm arising from the μ-oxo-bridged diferric center. We found that treatment of the protein with 1 mM of the radical scavenger  (Petersson et al., 1980) resulted in an estimation of the radical yield in the purified protein of 0.7 per monomer, comparable with data for other enzymes (Roca et al., 2008). In contrast to HA, treatment with the commonly used radical scavenger hydroxyurea (HU) did not result in quenching, even in the presence of the metal chelator EDTA (Figure 4d).

| Tyrosyl radical regeneration in NrdB can be catalyzed by FldA, FqrB, and NADPH
The ability to produce a radical free NrdB preparation allowed us to investigate the requirements for radical formation in this protein.
Incubation of HA-treated and dialyzed NrdB with an equimolar concentration of the flavodoxin FldA (50 μM) plus a catalytic amount of His-tagged NrdB was purified from E. coli BL21 (pETNrdB) as described in Section 4. (a) Iron (orange) and manganese (pink) content of NrdB, measured by ICP-MS. The data are the means and SD of three independent purifications from cells grown in LB alone, LB + Fe (50-µM ferrous ammonium sulfate) or LB + Mn (50-µM manganese chloride). The sulfur content of the samples was used to normalize the data to the protein content (Wang et al., 2010) and is expressed as a molar ratio. In each case, the Mn ratio is significantly lower than the Fe ratio (****p < .0001 by Student's t-test). (b) Optical absorption spectroscopy of NrdB. The blue trace shows the spectrum of as-purifed NrdB (100 µM) with the tyrosyl radical peak at 411 nm and charge transfer bands of the di-ferric center at 325 and 370 nm. The red trace shows the spectrum after 30-min treatment with 1-mM hydroxylamine (HA), which quenches the radical. The black trace is the difference spectrum, illustrating the tyrosyl radical contribution. The EPR spectrum in (c) shows that HA treatment does not significantly alter the g = 4.27 CW EPR signal of the high spin ferric iron in rhombic coordination in 50-µM NrdB as isolated (blue trace) or 50-µM NrdB + 1-mM hydroxylamine (red trace). Instrumental conditions used were: microwave frequency ν MW = 9.47 GHz; microwave power P MW = 3.165 mW; sweep rate v = 22.65 G s −1 (one scan); time constant τ = 81.92 ms; modulation amplitude A m = 5 G; modulation frequency ν m = 100 kHz; registration temperature T = 10 K. In (d), optical spectroscopy (50-µM NrdB) shows that a high concentration of hydroxyurea (HU), either with (blue trace) or without (red trace) EDTA does not appreciably quench the tyrosyl radical compared with HA in panel (b) resulted in no increase in absorbance at 411 nm (Figure 5b), showing that radical formation was dependent on reduced flavodoxin formed through the NADPH dependent reductase activity of FqrB.
In order to further confirm that the tyrosyl radical had indeed been regenerated in the reconstituted system with characteristics similar to the native NrdB, we compared the EPR spectrum of a sample of as-purified NrdB with HA-treated NrdB and that obtained after reconstitution with NADPH, FqrB, and FldA. The optical spectra of the same samples used for EPR are shown in Figure 5c, which again confirm the presence of the 411 nm band after reconstitution, and the corresponding 10 K EPR spectra are shown in Figure 5d. Essentially, identical EPR spectra were recorded from both the as-purified protein and after reconstitution with NADPH, FqrB, and FldA ( Figure 5d). The spectra also show that HA treatment effectively removed the radical from NrdB, with only very weak EPR signals remaining. The center of the free radical EPR signals is at ~g = 2.0052, and the lineshape is typical of that of Fe 2 -tyrosyl radical containing class Ia RNRs (Hoganson & Babcock, 1992;Sahlin et al., 1987).

| Is superoxide involved in FldA catalyzed tyrosyl radical formation?
The di-ferrous center in class Ia RNRs is readily oxygen reactive, and the electron input required for re-reduction has been proposed to be supplied in some bacteria by a small ferredoxin, YfaE (Wu F I G U R E 5 Reduced flavodoxin catalyzes tyrosyl radical regeneration in NrdB. In panel (a), a mixture of hydroxylamine-treated NrdB (50-µM final concentration), FldA (50 µM), FqrB (5 µM), and NADPH (1 mM) in 50-mM Tris-HCl buffer pH 7.5 was incubated at 37°C and spectra recorded at the times shown. The build-up of the tyrosyl radical in NrdB is evident at 411 nm. In (b), the spectra of a series of control incubations missing one of the components in (a), as indicated, was obtained after 80-min incubation at 37°C. In (c), the same mixture as in panel (a) was prepared, incubated for 80 min and the optical spectra compared with as-purifed NrdB and HA-treated NrdB. These same samples were then examined by EPR spectroscopy (d) as described in Section 4. The free radical area of the CW EPR spectra of NrdB is shown: 50-µM NrdB as isolated shows a typical EPR line shape of a RNR tyrosyl radical (black trace); 50-µM NrdB +1-mM hydroxylamine results in essentially complete reduction of the radicals (red trace); the free radical reconstituted by 50-µM flavodoxin (FldA) and 5-µM flavodoxin reductase (FqrB) in the presence of 1-mM NADPH (blue trace) has a line shape indistinguishable from that of NrdB as isolated. The instrumental conditions were microwave frequency ν MW = 9.47091 GHz (blue trace), 9.46769 GHz (black trace) and 9.47058 GHz (red trace; the three spectra were shifted along the magnetic field axis to a common g-value on the basis of these differences in the microwave frequencies); P MW = 5.016 × 10 −2 mW; v = 0.596 G s −1 (one scan); τ = 81.92 ms; A m = 3 G; ν m = 100 kHz; T = 10 K. Symbol "x" gives relative magnification of the spectra et al., 2007). yfaE is sometimes genetically linked to the nrdAB genes but is not present in Campylobacterota including C. jejuni. Given the C. jejuni RNR is a class Ia enzyme, it seems most likely that the role of FldA identified above is as a simple electron donor to NrdB in place of YfaE. However, in class 1b RNRs that use oxygen unreactive Mn as cofactor, the NrdI flavodoxin generates superoxide that acts as the oxidant to form the tyrosyl radical (Berggren et al., 2014). The possible involvement of superoxide in the regeneration of the tyrosyl radical in the C. jejuni NrdB by reduced FldA was investigated in two ways. First, we added excess superoxide dismutase to a mixture of FqrB, FldA, and HA-treated NrdB before adding NADPH to start the reaction. After prolonged incubation, tyrosyl radical formation was only slightly suppressed compared with incubations lacking SOD (Figure 6a,b), but was still readily detectable. The caveat here is that superoxide could be channeled from flavodoxin to NrdB by close interaction, so may not be sensitive to SOD (Berggren et al., 2014). Secondly, we incubated purifed HA-treated NrdB with the well-known xanthine oxidase/hypoxanthine superoxide generating system, as a way of generating superoxide independently of flavodoxin and FqrB. A time-dependent formation of the tyrosyl radical could be observed, as evidenced by the absorbance increase at 411 nm ( Figure 6c). This suggests that at least in vitro superoxide can result in tyrosyl radical formation in C. jejuni NrdB.

| D ISCUSS I ON
As the electron acceptor for the citric-acid cycle enzymes POR and OOR, FldA is a central redox shuttle in the physiology of C. jejuni, but other roles for this key protein have not been investigated, and genetic studies are hampered by its essentiality. In this work, we have identified the cj0559 encoded FqrB as an FldA reductase and phenotypic analysis of a deletion mutant lacking FqrB has revealed a hitherto unknown link between FldA and deoxyribonucleotide reduction.
FqrB from C. jejuni is an NADPH-specific flavodoxin reductase that has a phylogenetic relationship to a group of similar enzymes from Gram-positive bacteria, some of which are known to be redox partners for NrdI-like flavodoxins necessary for tyrosyl radical formation in class Ib RNRs. FqrB was initially identified in H. pylori as a quinone and flavodoxin reductase that was suggested to participate in the production of NADPH by operating in the reverse direction to that proposed here, that is, coupling the reduction of NADP to the oxidation of reduced flavodoxin that had been generated by the action of POR (St Maurice et al., 2007). This seemed to confirm earlier data from the study of Hughes et al., (1998) who detected pyruvate-, CoA-, and FldA-dependent NADPH production in H. pylori cell-free extracts (CFE). The same reaction could also be demonstrated in C. jejuni crude extracts (St Maurice et al., 2007). However, the rates of NADPH production in these studies were very low (of the order of a few nmol/min mg protein −1 ). We can now explain this in view of the relatively high midpoint redox potential of FldA (−180 mV) compared with NADPH (−320 mV), such that it is thermodynamically unfavorable for FldA to reduce NADP. It is therefore unlikely that this mechanism would be relevant in vivo, and it should be noted that other known enzymes in C. jejuni central metabolism operate to reduce NADP directly (e.g., isocitrate dehydrogenase). On the contrary, our data suggest that FqrB catalyzes NADPH-dependent reduction of FldA, in addition to POR and OOR, and this contributes to the maintenance of a reduced pool of FldA for diverse metabolic reactions F I G U R E 6 Superoxide can catalyze tyrosyl radical formation in NrdB in vitro. In panel (a), the complete reconstituted system of HAtreated NrdB (50 µM), FldA (50 µM), FqrB (5 µM), and NADPH (1 mM) was incubated at 37°C for 120 min as a control to show formation of the tyrosyl radical at 411 nm. In panel (b), the same components were incubated but with the addition of 20 units of SOD and the spectra recorded at 20-min intervals for the times shown. In panel (c), HA-treated NrdB alone (50 µM) was incubated at 37°C with 1 mM hypoxanthine and 0.5 units of xanthine oxidase for 3 hr and spectra recorded every 30 min as shown requiring this electron carrier (Figure 7). That NADPH can reduce FldA might also explain previous reports of (albeit low) NADPH dependent respiration in membranes and cell extracts of campylobacters (Lascelles & Calder, 1985;Weerakoon & Olson, 2008).
The phenotype of an fqrB deletion mutant implicates FqrB in ribonucleotide reduction. The fact that this mutant was viable, although with a growth defect, is presumably because POR and OOR are also still able to contribute to FldA reduction. Although POR is essential (Kendall et al., 2014), presumably because there is no other way to make acetyl-CoA, an oorB mutant has previously been reported (Weerakoon & Olson, 2008). The stimulation of fqrB mutant growth by dRNS was significant although not large in magnitude, but this would be consistent with the additional functions that FldA has in electron transport to Complex I and, for example, isoprenoid biosynthesis ( Figure 7) as well as other possible cellular redox processes yet to be identified. It should also be noted that although a potential transport system and phosphoribosyl phosphotransferase for nucleotide salvage has been identified in C. jejuni (Yahara et al., 2017), Biochemical evidence for a role of FqrB and FldA in RNR activation was obtained by an in vitro reconstitution approach with the relevant purified proteins. The tyrosyl radical in purified NrdB was remarkably resistant to quenching by HU, despite growth being sensitive to this compound and causing filamentation in treated cells consistent with RNR inhibition (Sellars et al., 2002). HA proved an effective radical scavenger however and we were able to show, with appropriate controls, that radical reformation could be promoted in the presence of NADPH, FqrB, and FldA. Importantly, the EPR data show that the characteristics and environment of the radical formed in vitro is indistinguishable from that in the as-purifed NrdB protein.
The fqrB mutant phenotype data taken together with the in vitro reconsitution experiments therefore support a physiological role for FqrB and FldA in NrdB tyrosyl radical formation, and thus ribonucleotide reduction, in C. jejuni, as illustrated in Figure 7. Although our data were obtained with strain NCTC 11168, FqrB, FldA, and NrdB are present in other strains of C. jejuni and, for example, are 97%-100% identical in two other commonly studied strains, 81-176 F I G U R E 7 Multiple pathways for FldA reduction and the role of FldA in electron transport, isoprenoid biosynthesis, and deoxyribonucleotide reduction. FldA can be reduced by electrons from at least three sources; from pyruvate oxidation via pyruvate:flavodoxin oxidoreductase (POR), from 2-oxoglutarate reduction via 2-oxoglutarate:flavodoxin oxidoreductase (OOR) or from NADPH oxidation via flavodoxin quinone reductase (FqrB). FldA is the only known electron donor for respiratory Complex I in C. jejuni, but given the relatively small redox potential difference (~100 mV) between FldA ox /FldA red couple and the menaquinone (MK)/menaquinol (MKH 2 ) couple, it is unclear if Complex I is proton translocating. Our results suggest that FldA is an electron donor to the NrdB subunit of ribonucleotide reductase (RNR), to generate the tyrosine radical (Y•). This is then transferred to the NrdA subunit to generate a radical at the active site cysteine (C•), necessary for reduction of ribonucleotides (RNTs) to deoxyribonucleotides (dRNTs). FldA is also thought to be required as a source of electrons for two consecutive steps in the final section of the MEP isoprenoid biosynthesis pathway, catalyzed by IspG which converts methylerythritol cyclic pyrophosphate (MEc-PP) into 4-hydroxy-3-methylbut-2-enyl pyrophosphate (HMB-PP) and IspH which converts HMB-PP into the products isopentyl pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP). MEc-PP is produced in several enzymatic steps from pyruvate and glyceraldehyde-3-phosphate (GA3P) and 81116. The model in Figure 7 should therefore be generally applicable.
We have characterized FldA in this work as a high redox potential flavodoxin, with the FldA ox /FldA sq and FldA sq /Fld hq couples having E m values of −170 and −190 mV, respectively. Before we consider the significance of this in relation to RNR, it is worth noting the implications of these redox properties for understanding the bioenergetics of Complex I in C. jejuni, which uses FldA as the electron donor instead of NADH. Aerobic bacteria that use NADH (E m − 320 mV) coupled with ubiquinone (E m + 100 mV) as the electron acceptor for Complex I will have an overall ΔE of ~ 420 mV to drive proton translocation electrogenically, against an existing Δp of ~ 180 mV. However, only menaquinone-8 (MK-8) and methylmenaquinone-8 (mMK-8) are present in C. jejuni membranes (Carlone & Anet, 1983). The E m for MK-8 is about −75 mV, and for mMK-8, it is −124 mV (Juhnke et al., 2009) or as low as −140 mV (Hein et al., 2018). The ΔE for FldA reduction of MK-8 is therefore only ~100 mV, and for mMK-8, it is ~50 mV; con-  (Table 1). This is also reflected in the pI values of such flavodoxins, which are usually acidic. NrdI proteins, however, have a bi-modal distribution with both low pI (mainly Firmicutes) and high pI (E. coli and most other proteobacteria) types (Johansson et al., 2010). Both types tend to have a more neutral (or charge compensated) environment in the loops surrounding the bound FMN, resulting in higher and often more similar E 1 and E 2 redox potentials than canonical flavodoxins (Table 1), but the picture is quite complex and other residues are also important.
The C. jejuni FldA is a long-chain flavodoxin with a predicted pI of 3.9, and it possesses conserved anionic residues in the W-and Y-loop regions involved in FMN binding ( Table 1) that might be expected to result in a much lower E 1 redox potential than the −190 mV determined here. A crystal structure will be needed to rationalize the reasons for this disparity. Nevertheless, the similarity in the smaller separation in the E 1 and E 2 potentials of FldA and most NrdI proteins is intriguing. In class Ia RNRs, a small ferredoxin, exemplified by YfaE in E. coli (Wu et al., 2007) has been implicated in supplying electrons to reduce the di-ferric centre in NrdB to allow oxygen to react with the di-ferrous state, necessary to form the tyrosyl radical. No homologue of YfaE exists in C. jejuni and indeed only 29% of 181 NrdAB containing genomes were found to encode a yfaE-like gene linked to nrdAB (Wu et al., 2007), implying that in most bacteria, other redox proteins satisfy this requirement. Our data suggest that C. jejuni uses the flavodoxin FldA instead of a ferredoxin as the electron donor to NrdB. YfaE is a 2Fe-2S ferredoxin that directly supplies electrons, one at a time, to reduce the NrdB diferric center (Wu et al., 2007).
FldA, with E 1 and E 2 separated by only 20 mV could act in a similar manner. Nevertheless, in vitro generated superoxide also formed the radical in NrdB, as found with the NrdI flavodoxins interacting with the class Ib NrdF (Berggren et al., 2014). However, the physiological significance of this is difficult to judge, given that the NrdB/class Ia type of Fe-containing RNR should be fully capable of generating the Tyr radical from the oxidation of ferrous to ferric iron at the di-iron center mediated by molecular oxygen alone. It would clearly be informative to examine the relative reactivity of the C. jejuni NrdB diiron center with superoxide versus oxygen in more detail.
In conclusion, our results explain why the sole flavodoxin of C. jejuni is essential and provide experimental evidence for at least one key role unrelated to respiratory electron transport. Recent work on the H. pylori FldA has resulted in the development of inhibitors that are novel drug candidates for therapeutic use (Salillas & Sancho, 2020), highlighting how the unique biochemistry of essential proteins like flavodoxin in pathogenic Campylobacterota can be exploited.

| Construction of an fqrB mutant and complemented strain
Isothermal assembly (ISA) cloning (Gibson et al., 2009) was used to generate a plasmid for transformation into C. jejuni, that deletes cj0559 (fqrB) by allelic exchange mutagenesis, replacing most of the coding region with a kanamycin resistance cassette derived from pJMK30 (van Vliet et al., 1998), which carries a constitutive promoter and no terminator. The ISA reaction assembled 4 PCR amplified fragments; HincII digested pGEM3Zf(-) vector, two regions of ~500 bp flanking cj0559, such that only the first few and last few codons of the gene were retained, and the kanamycin resistance cassette, to be inserted between the two flanking regions. The kanamycin resistance cassette was amplified from pJMK30 using primers Kan_F and Kan_R ( Table 2). The left flanking region of the gene was amplified using primers FQRB_ISA_F1F and F1KR (Table 2), and the right flanking region was amplified by FQRB_ISA_F2KF and F2R.
The F1F and F2R primers contain 30-bp adapters for the pGEM3Sf(-) vector cut with HincII and the F1KR and F2KF primers contain 30-bp adapters for the kanamycin resistance cassette. PCR amplifications were performed with phusion polymerase (NEB). The ISA reactions were performed as previously described (Liu & Kelly, 2015)  with the wild-type fqrB gene driven by its native promoter, the gene plus ~200 bp upstream sequence was amplified with primers FQRB_ COMP_F and FQRB_COMP_R (Table 2) and cloned into the pRRA vector (Cameron & Gaynor, 2014) at MfeI and XbaI sites. This recombines the gene at one of the 16S rRNA loci. This plasmid was electroporated into the fqrB mutant and colonies selected on Columbia blood agar plates plus kanamycin and apramycin.

| Overproduction and purification of proteins
The fqrB, fldA, and nrdB genes were cloned between the NdeI and XhoI sites of the pET21a vector (Novagen) such that a 6-his tag was attached in-frame to the C-terminus of the cognate proteins. The primers used are shown in Table 2 gradient of 50-mM Tris-HCl buffer pH 7.5. After purification, proteins were buffer exchanged into 50-mM Tris-HCl pH 7.5 using Vivaspin spin columns (Sartorius Stedim) with a molecular weight cut-off of at least 50% smaller than the protein size. Protein purity was assessed by SDS-PAGE.

| Mass spectrometry of NrdB
The two bands seen on SDS-PAGE gels of purified NrdB were cut out, divided into small pieces and de-stained in 500 μl acetonitrile/ ammonium bicarbonate buffer (  using an m/z window from 375 to 1,600, a resolution of 60,000, and an automatic gain control target setting of 1 × 106. Using collisioninduced dissociation (CID), the 20 most intense precursor ions were selected for the acquisition of tandem mass spectra in the dual cell linear ion trap at normal scan rate. Charge states 1 + were not included for precursor selection. Normalized collision energy was set to 35%, activation time to 10 ms, isolation width to 2 m/z, and automatic gain control value was set to 1 × 104. Identification of peptides/proteins was performed by MaxQuant (Cox & Mann, 2008), searching with default parameters.

| Inductively coupled plasma mass spectrometry
Purifed NrdB samples were added to 1 ml concentrated HNO 3 (65% v/v), left in acid overnight, and then analyzed at the University of Sheffield ICP-MS facility.  (Mayhew et al., 1969). The plot of semiquinone concentration versus sample potential was fitted to the equation describing two sequential one-electron transfers to a single site:

| Optical spectroscopy
, E is the sample potential, and E 1 and E 2 are the reduction potentials for addition of the first and second electron, respectively. Equivalent experiments without flavodoxin showed there was negligible spectral change from the mediators at 610 nm for potentials where the semiquinone was observed.

| Protein film electrochemistry
Cyclic voltammetry was performed using a three-electrode cell configuration inside a Faraday Cage within a N 2 -filled chamber (atmospheric oxygen < 5 ppm). The reference electrode was Ag/AgCl (saturated KCl) and the counter electrode a platinum wire. The working electrode was pyrolytic graphite with the edge plane exposed to sample. Preparation of the working electrode was essentially as described by Seagel et al. (2017). Firstly, the surface was polished with Semiquinone population = 1 1 + 1 + − 1 0.3μm Al 2 O 3 as an aqueous slurry, sonicated, rinsed with Milli-Q water and dried with a tissue. Secondly, 5 μl of 10-mM didodecyldimethylammonium bromide (DDAB) in Milli-Q water was placed on the electrode surface, the electrode covered with a beaker and left to dry overnight at ambient temperature. Finally, the DDAB-coated working electrode was taken into the N 2 -filled chamber and 5 μl of 200-μM flavodoxin (in 50-mM phosphate, 150-mM NaCl, and pH 7.5) deposited on the surface. After 20 min, the working electrode was introduced to the electrochemical cell, which contained an aqueous solution of 50-mM potassium phosphate, pH 7. Cyclic voltammetry was performed with a PGSTAT30 potentiostat (Metrohm Autolab) under the control of NOVA 1.11 software. Measured potentials were converted to values versus SHE by addition of 197 mV.

| EPR spectroscopy
All EPR spectra were measured on a Bruker EMX EPR spectrometer (X-band). A Bruker resonator ER 4122 (SP9703) and an Oxford Instruments liquid helium system were used to measure the lowtemperature (10 K) EPR spectra. Wilmad SQ EPR tubes (Wilmad Glass, Buena, NJ, USA) with OD = 4.05 mm and ID = 3.1 mm were used. Aliquots (250 μl) of the samples were placed in the EPR tubes, frozen in methanol and kept on dry ice (~195 K). The EPR tubes were then transferred to liquid nitrogen (77 K) and stored there until the measurements on the EPR spectrometer. Instrument settings are given in the relevant figure legends. The spectra were acquired by Dr Dima Svistunenko at the Biomedical EPR Facility, University of Essex, UK.

| Enzyme assays
All spectrophotometric assays were conducted at 37°C in a Shimadzu UV-2401 spectrophotometer. The ability of pyruvate and 2-oxoglutarate to act as electron donors to FldA via the action of the POR and OOR enzymes was assayed using cell-free extracts (CFE) of C. jejuni prepared by sonication under anaerobic conditions as described previously, in the presence an oxygen-scavenging system consisting of glucose oxidase, glucose, and catalase (Kendall et al., 2014). CFE was added to nitrogen-sparged 100-mM Tris-HCl buffer pH 8, 2-mM MgCl 2 , 0.2-mM Coenzyme A, 0.1-mM thiamine pyrophosphate, and 50-µM purifed FldA in stoppered anaerobic cuvettes. The reaction was started by injection of anaerobic pyruvate or 2-oxoglutarate to 5-mM final concentration. Reduction of the FMN of FldA was followed at 460 nm. The NAD(P)H oxidase activity of FqrB was assayed at 340 nm (ε= 6.22 mM −1 cm −1 ) in oxygen saturated 50-mM Tris-HCl buffer pH 7.5 after addition of FqrB (1 µM final) and either NADPH or NADH (300 µM final) to start the reaction. The reduction of FldA by FqrB and NAD(P)H was measured at 460 nm (estimated ε of 12 mM −1 cm −1 ; Mayhew & Tollin, 1992) in 50-mM Tris-HCl buffer pH 7.5, 50-µM FldA, 0.1-µM FqrB, and 150 µM either NADH or NADPH to start the reaction.

| Preparation of radical free NrdB and reconstitution with FqrB, FldA, and NrdB
Purified NrdB was treated with 1 mM HA for 30 min then repeatedly buffer exchanged into 50-mM Tris-HCl pH 7.5 until no HA could be detected. Residual HA was detected by reaction with 8-hydroxyquinoline to form green indooxime after heating, which absorbs at 710 nm (Lanvers et al., 2002). Reconstitution assays with radical free NrdB were done in 50 mM Tris-HCl buffer pH 7.5 with 50-µM NrdB, 50-µM FldA, 5-µM FqrB, and 1-mM NADPH. Where samples were to be monitored by EPR, the buffer was changed to 20-mM Tris-HCl pH 7.7 plus 50-mM l-arginine and 50-mM l-glutamate to enhance NrdB stability.

| Phylogenetic analysis
Protein sequences were obtained from Uniprot. Multiple sequence alignments were made with CLUSTAL Omega and phylogenetic rees were generated and analyzed in JALVIEW, using publicly available software at www.expasy.org.

ACK N OWLED G M ENTS
We thank Dr Dima Svistunenko (Director, Biomedical EPR Facility, University of Essex, UK) for obtaining the EPR spectra and Prof Nick

AUTH O R CO NTR I B UTI O N S
Conception and design of study; DJK. Acquisition, analysis and interpretation of data; AA, LA, JS, JB, and DJK. Writing of the manuscript; DJK, JB, and AA.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available in the supporting information of this article (Tables S1 and S2).