S. pneumoniae NOX was expressed in Escherichia coli and purified as previously described with minor modifications (Methods)4,6. The protein was solubilized using LMNG and showed a single band in SDS–PAGE and a homogeneous single peak on size-exclusion chromatography, corresponding to the monomeric protein (Supplementary Fig. 1a,b). Quick removal of imidazole after Ni-NTA purification was crucial to avoid protein aggregation. The ultraviolet–visible light (UV–vis) spectrum of the purified protein showed the characteristic Soret peak at 414 nm, indicating correct heme incorporation (Supplementary Fig. 1c). Oxidation of NADPH could be directly followed by monitoring absorbance at 340 nm under aerobic conditions. A similar experiment in the absence of O2 showed only initial activity because of residual O2 in the buffer and/or reduction of SpNOX cofactors before the oxidation activity ceased, confirming that O2 can act as an electron acceptor for SpNOX (Supplementary Fig. 1d). NADPH oxidation was also measured with a cytochrome c reduction assay4 (Fig. 1b) and Michaelis–Menten analysis gave an apparent KM value of 7.59 µM and an apparent kcat of 8.27 s−1. The protein also exhibited NADH-driven activity, with an apparent KM value of 3.41 µM and an apparent kcat of 3.33 s−1. Cytochrome c reduction assays under anaerobic conditions revealed direct electron transfer to cytochrome c (Supplementary Fig. 1e), likely from the outer heme, and to superoxide dismutase (SOD) (Supplementary Fig. 1f). Therefore, the rate of cytochrome c reduction by the anion superoxide could not be estimated with SOD inhibition. Our results, obtained under both aerobic and anaerobic conditions, indicate that previous efforts to measure specific superoxide production using SOD inhibition may have been subject to unforeseen artifacts. The initial rates of cytochrome c reduction under aerobic and anaerobic conditions were very similar (Supplementary Fig. 1g), suggesting cytochrome c acts as the major direct electron acceptor in the in vitro assay. Contrary to previous observations4, we could observe ferric reductase activity under anaerobic conditions, indicating direct electron transfer to ferrous iron (Supplementary Fig. 1h). However, the direct transfer rate of SpNOX to ferrous iron (3.6 min−1) was much lower than to cytochrome c (124.7 min−1) or molecular oxygen (48 min−1) at similar initial concentrations. Moreover, Michaelis–Menten analysis of ferric reductase activity showed apparent kcat (0.05 s−1) and KM (81.3 µM) values corresponding to a catalytic efficiency of 0.0006 s−1 µM−1 (Supplementary Fig. 1i). These experiments suggest that the ferric reductase activity of SpNOX might not be physiologically relevant but rather an effect of the promiscuity of SpNOX for the final electron acceptor in vitro.

Cryo-EM single-particle analysis of SpNOX was performed in LMNG micelles and showed that the protein preparation was homogeneous (Methods and Supplementary Figs. 2–5). In addition to a structure of SpNOX in the absence of electron donor, we also determined structures of the stably reduced protein bound to NADH under anaerobic conditions and of the wild-type (WT) protein and an F397A mutant under turnover with NADPH and O2 present. Despite the relatively small size of the protein (46 kDa) for single-particle analysis, the final maps reached nominal resolutions ranging from 2.2 to 2.6 Å (Table 1 and Supplementary Figs. 6 and 7), allowing us to resolve essentially all residues. The TM and DH domains reached similar overall local resolutions, with only small regions of the NADPH-binding lobe being less well resolved (Supplementary Fig. 6). In contrast to previously published small-angle neutron scattering data indicating potential interdomain flexibility6, our results indicate that the DH domain interacts rigidly with the TM domain, thus facilitating particle alignment. While this paper was under review for publication, a study describing high-resolution crystal structures of the substrate-free WT and F397W DH domains and a low-resolution F397W full-length inactive structure of SpNOX became available as a preprint12.

The structure of SpNOX corresponds to the canonical NOX structure, with an N-terminal TM domain that coordinates two hemes and a C-terminal DH domain bound to FAD (Fig. 1c–e). Despite sharing an overall sequence identity between 10% and 14% with human and cyanobacterial NOXs (Supplementary Fig. 8a), the SpNOX structure strongly resembles the structures of eukaryotic NOXs and csNOX5 TM and DH domains (Supplementary Fig. 9).

The SpNOX TM domain encompasses six transmembrane helices (TM1–TM6) with an overall pyramidal shape, triangular on the inner membrane side and narrower toward the extracellular space. This folding strongly resembles the ferric reductase domain of eukaryotic NOXs, with TMs 2–5 adopting an hourglass-shaped conformation that binds two B-type hemes orthogonal to the membrane plane, one located closer to the cytosolic side (inner heme) and the other closer to the outer side (outer heme). Both are hexacoordinated by two pairs of histidines (H69 and H129, inner heme; H83 and H142, outer heme) belonging to TM3 and TM5 (Fig. 2a,b and Supplementary Fig. 8b). The iron-to-iron distance is 21.47 Å, while the edge-to-edge distance is 9.84 Å. Interestingly, the outer heme of SpNOX is flipped ~180° with respect to eukaryotic NOXs (Supplementary Fig. 9) and this seems to be determined by the residues interacting with the propionate groups in the structure (Supplementary Fig. 8a).

a, Cartoon representation of the structure of NADPH-bound SpNOX. Areas of interest are highlighted with colored dashed boxes. b, The two B-type hemes are coordinated by two histidine pairs H83 + H142 (outer heme) and H69 + H129 (inner heme) of the transmembrane helices TM3 and TM5. The edge-to-edge distance is indicated with a dashed line. TM1 and TM2 are omitted for clarity. c, Detailed view of the FAD-binding site. Side chains and FAD are shown as sticks. d, Closer look into the interface interactions between the TM and the DH domains. In c,d, water molecules are shown as red spheres and atoms within H-bond distance are marked with cyan dashed lines. e, The C-terminal tail of SpNOX lies at the interface of the TM and DH domains, is oriented orthogonal to the membrane plane and interacts by H-bonds between K70 at TM3 and F399.

The C-terminal DH domain is connected to TM6 by a short linker and shows the canonical ferredoxin–NADP+ reductase (FNR) fold: an FAD-binding lobe consisting of six antiparallel β-strands forming a β-barrel flanked by an α-helix and an NADPH-binding lobe with a Rossman-like topology formed by a parallel β-sheet sandwiched between α-helices (Fig. 2a and Supplementary Fig. 10a). When compared to eukaryotic NOXs and csNOX5 (Supplementary Figs. 8c and 9), the DH domain of SpNOX shows up to 19% sequence identity with a simplified overall architecture. The main structural differences accumulate at the NADPH-binding lobe (Supplementary Fig. 9), where SpNOX shows an unstructured long linker connecting the parallel β-strands. In eukaryotic NOXs, insertions and rearrangements at this site serve important regulatory functions (for example, the calmodulin-binding region of NOX5).

SpNOX displays high affinity for FAD (~60 nM) and lower affinity for other smaller flavins such as flavin mononucleotide12. A clear density for FAD was well resolved in all maps (Supplementary Fig. 7) inside a positively charged pocket at the interface of the FAD-binding lobe and the TM domain (Supplementary Fig. 10b). The DH domain residues H233, S236, K250, S252, T256 and T297 interact with FAD by hydrogen bonds (H-bonds), whereas F397 interacts by π–π stacking with the isoalloxazine ring (Fig. 2c and Supplementary Fig. 10c). Residues H233 and S236 belong to the strictly conserved ‘HPF(S/T)’ motif (Supplementary Fig. 8c). The only direct interaction between FAD and the TM domain is through π–π stacking between the adenine and Y122 (Fig. 2c and Supplementary Fig. 10d), which is strictly conserved in bacterial NOX-like proteins (Supplementary Fig. 11). A Y122A mutant was characterized in the literature12 and showed a reduced apparent affinity for FAD and a reduced kcat, confirming the importance of this residue in FAD binding. The adopted geometry of FAD is the same as in NOX2, which achieves this interaction through the Y122-equivalent residue F202 (Supplementary Fig. 10e). Interestingly, this FAD conformation is also present in human DUOX1, in spite of the fact that different interactions (mainly salt bridges linking the phosphate from adenine monophosphate (AMP) with R1214 and R1131 of the TM domain) are responsible for forming the interaction with FAD (Supplementary Fig. 10f).

The DH domain is docked through direct polar interactions of residues at the FAD-binding lobe, mainly with residues located at the B-loop and TM6 of the TM domain (Fig. 2d). Moreover, K70 at TM3 interacts by H-bonds with F399, which, together with K398 and K400, is part of the C-terminal tail of SpNOX (Fig. 2e). A Δ398–400 mutant showed an increased KM for NADPH and a reduced kcat (Supplementary Fig. 14a), indicating that these residues might be relevant for the docking of the TM and DH domains and for NADPH binding.

NADPH-bound (2.2-Å resolution) and anaerobic NADH-bound (2.4-Å resolution) SpNOX maps showed well-resolved densities for the substrate (Fig. 3 and Supplementary Fig. 7). Aside from the bound substrate, all three structures appear largely the same with only minor side chain rearrangements (Supplementary Fig. 12a–c,e–g,i–k). The nicotinamide moiety of NADPH and NADH is accommodated by H-bonds and hydrophobic interactions inside a cavity generated by F397 at the C-terminal tail and the strictly conserved ‘XGXGX’ and ‘CG(S/P)’ motifs located at the loop connecting α2 and β8 and at the loop connecting β10 and α3, respectively (Fig. 3a and Supplementary Figs. 8c, 12d,h,l and 13a). A C370A mutant showed an increase in KM, confirming the relevance of the ‘CG(S/P)’ motif for substrate binding (Supplementary Fig. 14b).

a, Detailed view of the NADPH-binding site at the DH domain of SpNOX. Atoms within H-bond distance are marked with cyan dashed lines. FAD (gray) and amino acid side chains (coral) are shown as sticks. b, The lack of specificity toward NADPH in SpNOX can be explained by the absence of ionic interactions with the 2′-phosphate, unlike in eukaryotic NOXs including human DUOX1 (PDB 7D3F)9, in which R1424 and R1495 interact with the 2′-phosphate. c, The proposed electron-transfer path within SpNOX. Hemes, FAD, NADPH and the inter-heme hydrophobic residues are shown as sticks on the surface of SpNOX. Distances between the redox cofactors are represented as dashed black lines and were measured between the nicotinamide and the isoalloxazine ring (7.2 Å), between the isoalloxazine ring and the inner heme lower edge (9.9 Å) and between the edges of the inner and outer hemes (9.8 Å). d, F397 sits between the isoalloxazine ring of FAD and the nicotinamide ring of NADPH, impeding hydride transfer. e, The nicotinamide ring of NADPH moves closer to the isoalloxazine group of FAD in the F397A SpNOX mutant.

Overall, the AMP-binding site of both substrates shows the highest number of interactions with SpNOX. Previous structural studies of NADPH-specific FNR superfamily members, such as DUOX1 (ref. 9), Anabaena PCC 7119 FNR13 and human neuronal nitric oxide synthase (nNOS)14, showed that the AMP-binding region is critical for coenzyme specificity because of the presence of at least one arginine residue that establishes charge-to-charge interactions with the 2′-phosphate group of NADPH. In SpNOX, this group is coordinated by H-bonds with S348 and Y353 (Fig. 3a), which is remarkably different to the NADPH-binding mode of eukaryotic NOXs and other members of the NADPH-specific FNR superfamily. In human DUOX1, two arginine residues, one analogous to R320 of SpNOX (R1424) and another analogous to Y353 (R1495), both coordinate the 2′-phosphate by salt bridges (Fig. 3b). In human NOX2, R513 also occupies the position of Y353 (Supplementary Fig. 8c). In Anabaena FNR, a tyrosine residue analogous to Y353 interacts with the adenine moiety; however, as in eukaryotic NOXs, the 2′-phosphate is also coordinated by salt bridges with two arginine residues (Supplementary Fig. 13b). Therefore, NADPH specificity results, at least partially, from ionic interactions that are absent in SpNOX, which would explain the lack of substrate specificity. Supporting this, a Y353R mutant increased the specificity of SpNOX for NADPH fourfold (Supplementary Fig. 14c,d). Additionally, the 5′phosphate interacts by a salt bridge with R320, the adenine moiety is sandwiched between the M375 side chain and the aromatic ring of Y353 and the 3′-OH of the ribose performs H-bonding with S318 and S348. The nicotinamide-bound ribose is hydrated by several resolved water molecules but does not show any direct interaction with SpNOX (Supplementary Fig. 13c). The comparison of the low-resolution SpNOX crystal structure, high-calcium NADPH-bound DUOX1 and inactive NOX2 from previous studies9,12 suggests that eukaryotic NOXs may require a tighter interaction between the NADPH-binding and FAD-binding lobes instead of a large domain motion for efficient hydride transfer. Our comparison of NADPH-bound SpNOX and high-calcium human DUOX1 (Supplementary Fig. 13d,e) further supports this idea and suggests a potential role of an NOX-conserved positively charged residue (DUOX R1337 and SpNOX K250; Supplementary Fig. 8c) that allows the nicotinamide ring to approach the isoalloxazine ring by forming a salt bridge with the phosphate adjacent to the ribose (Supplementary Fig. 13f).

The apparent electron-transfer pathway of SpNOX corresponds to the previously described electron pathways of NOX proteins: NADPH → FAD → inner heme → outer heme → O2 (refs. 7,8,9,10,11). The distance between the isoalloxazine ring and the lower edge of the inner heme is 9.9 Å, whereas the edge-to-edge distance between hemes is 9.8 Å (Fig. 3c). The space between the two hemes is partially occupied by a cluster of hydrophobic amino acids that includes the aromatic residues F107 and Y136. F107 occupies a similar position to a conserved aromatic residue present in other NOXs (F215 in human NOX2, F1097 in human DUOX1 and W378 in csNOX5). Mutational analysis in mouse DUOX1 and NOX2 (ref. 8,11) showed that this amino acid could be the preferred route for electron transfer between the hemes. Here, we analyzed the activity of an F107L mutant and an F107L;Y136L double mutant (Supplementary Fig. 14e,f). Contrary to the previous observations in eukaryotic NOXs, we did not detect any notable reduction of SpNOX activity. These results indicate that, in SpNOX, neither F107 nor Y136 is required for efficient electron transfer and suggest that the loss of activity in other NOXs could be the consequence of a less optimal hydrophobic environment or structural rearrangements. This is in agreement with previous observations that electron tunneling between redox centers separated by distances below 14 Å occur at rates fast enough not to be a limiting factor for substrate turnover15.

Our substrate-bound structures represent an inactive state of the enzyme, as deduced by a 7.2-Å core-to-core distance between the nicotinamide C4 and the isoalloxazine N5, too long for efficient hydride transfer (Fig. 3c,d), which should require distances consistent with simultaneous bond breakage and formation in the transition state16,17,18. The fact that a productive structure could not be solved may indicate that the active state is transient and, thus, sparsely populated19,20. In our inactive conformations, F397 is stacked between the nicotinamide and the isoalloxazine rings (Fig. 3d), suggesting that during turnover a conformational change involving, at least, the displacement of the F397 lateral chain is required for hydride transfer. Remarkably, an analogous C-terminal aromatic residue is conserved within other members of the FNR superfamily including plant-type FNRs, NOSs and cytochrome P450 reductases14,21,22,23. Extensive structural and biochemical studies have suggested that the displacement of this residue is highly thermodynamically unfavored and may act as the rate-limiting step for flavin reduction21,24,25,26. This residue may also contribute to the regulation of NADPH-binding affinity and specificity and to the stabilization of the FAD semiquinone state24. Substitution of the C-terminal tyrosine of plant-like FNRs to a nonaromatic residue such as alanine or serine substantially increased the affinity for NADP+ and NADPH and induced an enzyme state with productive flavin–nicotinamide interaction18. However, in other FNR superfamily members such as human nNOS, such a substitution did not lead to a large change in the binding affinity for NADPH, although it increased the kcat for NADH22. Here, we analyzed the steady-state kinetics of an F397A SpNOX mutant (Supplementary Fig. 14g). Similarly to nNOS, we did not observe a large effect on the KM, which suggests that F397 does not have a major role in the control of NADPH-binding affinity in SpNOX. As for the F397S SpNOX mutant described in the previous study12, we observed a moderate increase in kcat, which could be explained by the elimination of the F397 displacement step. These data are supported by further analysis of a cryo-EM structure of the F397A mutant bound to NADPH under turnover conditions at 2.64-Å resolution (Table 1, Fig. 3e and Supplementary Fig. 5). Density can be observed for the bound substrate but, compared to the other substrate-bound structures described here, the nicotinamide is closer to the isoalloxazine ring (Fig. 3e and Supplementary Fig. 12i–k). The absence of the large side chain of F397 allows NADPH to take up a productive conformation in SpNOX, as seen previously for pea FNR Y308 mutants18. In this conformation, the nicotinamide ring does not lie parallel to the isoalloxazine ring but at a ~26° angle (Fig. 3e).

A search for potential oxygen-reduction centers close to the outer heme-binding pocket revealed a strikingly high degree of exposure of the outer heme to the extracellular space, which is more buried within the structure of other NOXs by long extracellular loops or domains (Fig. 4a)8,11. In SpNOX, the extracellular loops (A-loop, C-loop and E-loop) fold away from the heme cavity, directly exposing the outer heme to the solvent (Fig. 4b,c). In fact, although all putative oxygen reaction centers previously described for NOXs showed a similar conformation involving two histidine residues, an arginine and a propionate group of the outer heme8,9,10,11, we did not find any similar O2-binding site in SpNOX. However, a closer examination of the outer heme-binding pocket revealed a small cavity below the C-loop occupied by an ordered water molecule. This water molecule, which forms H-bonds with N84, N101 and Y105, could be occupying an O2-binding site within efficient electron-transfer distance to the outer heme (Fig. 4d). A potential path for O2 or O2•− entry and exit at this site was found using Hollow27, taking the modeled water molecule as the starting point (Supplementary Fig. 13g). However, N84A and Y105F mutants did not display a notable change in cytochrome c reduction or NADPH oxidation activity compared to WT (Supplementary Fig. 14h–j), suggesting that this site may not be an actual oxygen-reduction center. We also identified another putative O2-binding site composed of S86 and heme-coordinating H83. Both amino acids are highly exposed to the solvent and coordinate a water molecule that could be occupying the O2-binding position (Fig. 4e).

a, SpNOX (cartoon representation, blue) presents an outer heme more solvent exposed than eukaryotic NOXs such as NOX2 (cyan; PDB 8GZ3)10 or DUOX1 (purple; PDB 7D3F)9 because of a lack of extracellular domains or large loops. b, Side view of the extracellular loops of SpNOX. c, Top view of SpNOX surface showing the solvent-exposed outer heme. d,e, Two different putative oxygen-reduction centers were identified in SpNOX, one formed by N84, N101 and Y105 (d), and another formed by H83 and S86 (e). The distance between the coordinated water molecules (red sphere and density) and the outer heme is shown as a black dashed line. Atoms within H-bond distance are marked with cyan dashed lines.