The Crystal Structures of Bacillithiol Disulfide Reductase Bdr (YpdA) Provide Structural and Functional Insight into a New Type of FAD-Containing NADPH-Dependent Oxidoreductase

Low G+C Gram-positive Firmicutes, such as the clinically important pathogens Staphylococcus aureus and Bacillus cereus, use the low-molecular weight thiol bacillithiol (BSH) as a defense mechanism to buffer the intracellular redox environment and counteract oxidative stress encountered by human neutrophils during infections. The protein YpdA has recently been shown to function as an essential NADPH-dependent reductase of oxidized bacillithiol disulfide (BSSB) resulting from stress responses and is crucial for maintaining the reduced pool of BSH and cellular redox balance. In this work, we present the first crystallographic structures of YpdAs, namely, those from S. aureus and B. cereus. Our analyses reveal a uniquely organized biological tetramer; however, the structure of the monomeric subunit is highly similar to those of other flavoprotein disulfide reductases. The absence of a redox active cysteine in the vicinity of the FAD isoalloxazine ring implies a new direct disulfide reduction mechanism, which is backed by the presence of a potentially gated channel, serving as a putative binding site for BSSB in the proximity of the FAD cofactor. We also report enzymatic activities for both YpdAs, which along with the structures presented in this work provide important structural and functional insight into a new class of FAD-containing NADPH-dependent oxidoreductases, related to the emerging fight against pathogenic bacteria.


Section S1. Expression, Purification and Characterization
Expression and Purification of Bc YpdA, Sa YpdA, and Sa YpdA G10A mutant pET-22b(+) plasmids containing the genes for Bc YpdA (BC1495, Bc ATCC 14579, restriction enzymes NdeI and BamHI) 1 , Sa YpdA (SACOL1520, Sa COL, restriction enzymes NdeI and HindIII) or Sa YpdA G10A (SACOL1520, Sa COL, restriction enzymes NdeI and HindIII) (GenScript) were transformed into competent Escherichia coli One Shot TM BL21 (DE3) cells (Invitrogen, Thermo Fischer Scientific). Cells containing either of the three plasmids were grown in Terrific Broth medium containing 100 μg/mL ampicillin. Protein expression was induced by adding isopropyl β-D-thiogalactoside (IPTG) to a final concentration of 0.5 mM at OD 600nm = 0.7-0.9, and the cultures were incubated for 12-16 hours at 20°C with vigorous shaking before cells were harvested and frozen at -20°C. Cells were thawed and dissolved in 100 mM Tris-HCl, pH 7.5, 1 mM DTT, 5 μg/mL DNase, cOmplete Protease Inhibitor Cocktail (Roche) in a 1:4 cell wet weight to buffer ratio and lysed by sonication. Alternatively, Sa YpdA, as well as Sa YpdA G10A, were dissolved in 50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 5 μg/mL DNase, cOmplete Protease Inhibitor Cocktail (Roche), prior to cell lysis. The suspensions were centrifuged at 48,000g and the lysates were cleared from nucleic acids by streptomycin sulfate (2.5 %) precipitation, followed by centrifugation at 48,000g. From the lysates, these Tag-free proteins were precipitated with ammonium sulfate ((NH 4 ) 2 SO 4 ) to final concentrations of 0.25 g/mL and 0.22 g/mL for Bc and Sa YpdAs, respectively, and centrifuged at 39,000g. Proteins were dissolved in 50 mM Tris-HCl, pH 7.5, 1 mM DTT, and desalted using a HiTrap Desalting column (GE Healthcare). Desalted proteins were applied to a HiTrap HP Q column and eluted with linear or step-wise 0-0.5 M KCl or NaCl gradients. As a final polishing step, protein used for crystallization experiments were purified on a Superdex 200 or Superdex 200 Increase column (GE Healthcare) in 50 mM Hepes, pH 7.5, 100 mM KCl. All chromatographic steps were performed using an Äkta purifier FPLC system (GE Healthcare). Protein fractions were pooled, concentrated in Amicon Ultra-15 filter units (10 or 30 kDA MWCO, Merck-Millipore), flash-frozen in liq N2, and stored at -80°C.

Preparation of Se-Methionine Derivatives
The L-selenomethionine (Se-Met) derivatives of Bc and Sa YpdA were expressed and purified in a manner similar to that of the wild type proteins, with the following modifications. Cells were grown at 37°C in M9 minimal medium supplemented with L-methionine (50 mg/L) until the OD 600nm had reached 1, harvested, and resuspended in fresh M9 minimal medium without methionine. Cells were further incubated at 37°C until the addition of lysine, phenylalanine, and threonine (100 mg/L of each); isoleucine, leucine, and valine (50 mg/L of each); and Lselenomethionine (50 mg/L), prior to induction with 0.5 mM IPTG (at OD 600nm = 1) and incubation for 16 hours at 20°C with vigorous shaking before harvesting and freezing of the cell paste. Cell lysis and protein purification was performed as described for the wild type proteins.

Dynamic Light Scattering (DLS) and Native Polyacrylamide Gel Electrophoresis (PAGE) Analyses of Protein Oligomerization
In order to estimate the size and molecular weight of the purified Sa and Bc YpdA proteins, oligomeric states of the YpdA proteins were investigated by DLS (Zetasizer Nano), in 50 mM Hepes, pH 7.5, 100 mM KCl, at 25°C and with 50 μM protein concentrations, in three replicates each. In addition, protein samples of both Sa and Bc YpdA (2 μM and 3 μM, respectively) were analyzed on Native PAGE (NativePAGE™ Novex® 4-16% Bis-Tris Protein Gels, Thermo Fisher Scientific).

Preparation of BSSB Substrate
As bacillithiol disulfide (BSSB) is not commercially available, an oxidation of reduced BSH was performed, as described by Hamilton and coworkers 2 . In short, prior to enzymatic assays, a solution of NH 4 HCO 3 was added to reduced BSH (Jema Biosciences) dissolved in water at room temperature and stirred with exposure to air for 1 h, flash-frozen in liq N2, and stored at -80°C. Oxidation of BSH to BSSB was verified with 5,5'-dithiobis-(2-nitrobenzoic acid) (DTNB).

Enzymatic Assays of BSSB Reduction by Sa and Bc YpdAs
The reduction of BSSB by NADPH-dependent YpdAs was investigated through spectroscopic determination of the reductase activity. Activity was verified through following the decrease in absorption caused by the oxidation of NADPH at 340 nm. Enzymatic assays performed in the presence of dioxygen were executed at 25°C in 20 mM Tris, 1.25 mM EDTA, pH 8.0, 200 μM NADPH, and varying concentrations of BSSB (0.25-10 μM BSSB), followed by the addition of 1 μM purified Sa or Bc YpdA after 1 min, and spectra were collected every second for 1000-1200 seconds at 340 nm (Agilent Cary 60). Due to the oxygen sensitivity seen for both Sa and Bc YpdAs, enzymatic assays were also performed under anaerobic conditions in a glovebox (Plas-Labs Anaerobic Chamber 855-AC and an Agilent 8453 diode-array UV-vis spectrophotometer). The Sa YpdA G10A mutant was used as a control, proving the importance of Gly10 in binding of the FAD cofactor, crucial for enzymatic activity. All solutions were degassed on a Schlenk line before transfer to the glovebox. Buffers and stock solutions were sparged with argon for minimum 2 hours in vented vials, and protein samples were subjected to 5-6 cycles of evacuation and refilling with argon. Assays were performed at 25°C in 20 mM Tris, 1.25 mM EDTA, pH 8.0, 240 μM NADPH, with the addition of 12.5 μM purified Sa YpdA, Bc YpdA or Sa YpdA G10A mutant, followed by the addition of 50 μM BSSB, and spectra were collected every second for 250 seconds at 340 nm. Prior to addition of BSSB, reactions were run for 2 min to allow for the reduction of trace amounts of dioxygen by YpdA, in the presence of NADPH. All assays were performed in replicates.

Crystal Data Collection, Processing, and Refinement
Diffraction data were collected at MAX IV, Lund, Sweden, on beam line BioMAX (Sa YpdA and Se-Met Bc YpdA) and at ESRF, Grenoble, France, on beam line ID23-1 (Bc YpdA) through MXCuBE3 3 and ISPyB 4 at 100 K. The Se-Met Bc YpdA diffraction data set was collected at 0.976252 Å, a few eV above the theoretical Se absorption K-edge. Diffraction data were indexed and integrated with iMosflm 5 (Bc YpdA), auto-processed with EDNA 6 and XDS 7 (Sa YpdA), and autoPROC 8 and XDS (Se-Met Bc YpdA), and scaled and merged with Aimless in the CCP4 package 9 .
Although BLAST sequence alignment searches against the PDB databases indicated hits with up to 30% sequence identity, extended molecular replacement trials with different search models were not successful in solving the structure. This was most likely due to the number of subunits in the asymmetric unit, 4 and 8 combined for Bc and Sa, respectively, with the possibility of relative movements of the FAD and NADPH binding domains.
The Se-Met Bc YpdA structure was solved by single-wavelength anomalous dispersion (SAD) experiment. Matthews coefficient 10 analysis indicated four molecules in the asymmetric unit with a Matthews coefficient of 2.6 Å 3 /Da and solvent content of 53.2%, indicating that the anomalous signal was significant to about 3.8 Å, so the data set was scaled to 3.5 Å with an anomalous redundancy of 6.6. To solve the structure with SAD, CRANK2 11 was used through CCP4 online using SFtools, SHELX, SHELXD, REFMAC5, PEAKMAX, MAPRO, Solomon, Multicomb, Parrot, and Buccaneer 12-18 . The solved structure contained four molecules in the asymmetric unit, and initially refined to an R-factor of 36%. This initial model was used as the starting model for the higher resolution native Bc YpdA data set of 1.6 Å resolution. After an initial refinement with REFMAC5 further automatic model building was performed with ARP/wARP 19-21 through the CCP4i. This was followed by several cycles of refinement with phenix.refine 22 in the Phenix suite 23 and model building in Coot 24 . TLS refinements was performed with each of the four chains constituting a TLS group. During the refinement also a PDB_REDO run was performed 25 . Model validation was performed using MolProbity 26 . All structure figures were prepared with PyMOL (Schrödinger, LLC).
The Sa YpdA 1 and 2 structures were solved independently through molecular replacement (MR) using the Bc YpdA structure as a starting model. Eight molecules in the asymmetric unit would give a Matthews coefficient of 2.8 Å 3 /Da and solvent content of 55.6%. To solve the full structure with eight molecules in the asymmetric unit with Phaser 27 through CCP4i, one had to search for one copy of the Bc YpdA tetramer, and four copies of the Bc YpdA monomer. This was followed by several cycles of refinement initially with REMAC5 and subsequently phenix.refine in the Phenix suite, and model building in Coot. Model validation was performed using MolProbity.
For Bc YpdA, all residues have been modelled for chains A and B. For chain C, the C-terminal residue has been excluded, and for chain D, the N-terminal and C-terminal residues have been excluded due to poor electron density. It can additionally be noted that some of the modelled loops have limited electron density. For Sa YpdA, on average five C-terminal residues have not been modelled for any of the subunits (Table S1), and some of the modelled loops have limited electron density. Chains G and H have large areas with poor electron density, show high temperature factors, and are more distorted than the other chains (Table S1). Due to the very limited and unclear electron density in some areas of chain G, residues 159-215 (Sa YpdA 1) and 179-187 + 196-215 (Sa YpdA 2) were not built into the model for chain G. Although chains G and H are disordered, they were independently a solution in MR, and removing them from the structure increased both Rwork and Rfree, and was therefore retained in the final structures.
Clear electron density was observed for FAD and modelled in all subunits of both Bc YpdA and Sa YpdAs, although with less clear density in the two more distorted chains G and H in Sa YpdAs ( Figure S13A,B). No electron density was observed for NADPH in the Bc YpdA structure where both conformations were observed for the possible gating residue Tyr133 ( Figure 3F,G). Both Sa YpdAs showed clear electron density for NADPH in chains C, D and F, which was accompanied with the closed conformation of residues 295-301, and Tyr128 in open conformation with hydrogen bonding to NADPH ( Figure 3D,E,F S13B,D,E). NADPH was not observed in chains A, B and E, which was accompanied with the open conformation of residues 295-301, and Tyr128 in the closed conformation ( Figure 3D,E,G). In both chains C and F (Sa YpdA 1), one conformation of NADPH was modelled in, while in both chains D and F (Sa YpdA 2), two orientations of NADPH could be observed accompanied by a movement of Phe51 with two alternate conformations ( Figure 3H). The orientation of the nicotinamide parts of NADPH have not been fully resolved.
The standard Phenix restraints used for the FAD cofactor were modified for the 1.6 Å Bc YpdA resolution structure, to take into account potential X-ray radiation-induced reduction of the FAD cofactor making the isoalloxazine ring free to bend along the N5-N10 axis (butterfly bend) 28 . The angle of the butterfly bend of the isoalloxazine ring was calculated with the psico module in PyMOL, by calculating the angle between the two planes defined by atoms N5, C4X, C4, N3, C2, C10, N10, and N5, C5X, C3, C7, C8, C9, C9A, N10. A slight butterfly bend of average 4.2º was observed for the isoalloxazine rings in Bc YpdA, consistent with expected radiation-induced reduction ( Figure S13C).

Crystal Packing
The Bc YpdA P22121 crystals contained a homo tetramer in the asymmetric unit ( Figure S4A), with the monomers having an RMSD value of 1.2-1.3 Å relative to each other ( Figure S4B). The Sa YpdA P6122 crystals contained eight monomers (RMSD values 1.2-2.8 Å relative to each other, Figure S4E) in the asymmetric unit; a tetramer and two dimers ( Figure S4C). The Sa YpdA tetramer is similar to the Bc tetramer as seen from the overlay in Figure S4D (RMSD value of 1.7 Å). The closest symmetry equivalent subunits in the Sa YpdA crystal to the two dimers are shown in Figure S4F (yellow and pink), which shows that they also are part of tetramers. Two of the tetramers ( Figure S4F, palecyan and yellow) overlay well (RMSD value of 1.3 Å), while for the third tetramer (pink), each of the two dimers overlay well (RMSD value of 2.5 Å) with the two dimers of the other tetramers, however, overlaying one dimer results in the second dimer to be shifted 15 Å relative to the others ( Figure S4G). Therefore, the Sa YpdA P6122 crystal also contains tetramers, however, most likely due to crystal packing, one of the tetramers had to adopt by slightly sliding the two dimers relative to each other ( Figure S4G). PDBePISA was used to analyze the buried surface area of the YpdA multimers. The ABCD, E2F2 and G2H2 tetramers have a surface area of ~53000 Å 2 with a buried area of ~16000 Å 2 , which suggests a stable tetrameric structure in solution (Table S3). The tetramers have a higher percentage of buried area than the individual dimers, except the dimer CD with fully bound NADPH.

Comparison of the Active Site of Selected Flavoprotein Disulfide Reductases
The electron flow for most flavoprotein disulfide reductases (FDRs) are from NADPH/NADH positioned on the re-face of the isoalloxazine ring, through the FAD and to a cystine or a single Cys on the si-face ( Figure S14B,E,F,G,H). The active Cys residues on the si-face are also accessible to substrate or other Cys/SeCys-residues for further reaction ( Figure S14). Low Mr TrxRs are an exception, where the electron transfer only occurs on the re-face through a rotation of the NADPH domain relative to the FAD domain. The electrons go from NADPH to the FAD, then NADPH is rotated away so the cystine comes close to the FAD, and the cystine is reduced ( Figure S14C,D). YpdA does not have a conserved cysteine within 3-4 Å of the isoalloxazine ring nor that is accessible by a substrate. However, the channel above the isoalloxazine ring on the re-face is large enough to accommodate BSSB, and the disulfide in BSSB could be in a similar position to the isoalloxazine ring as the cystine in low-Mr TrxR FO (flavin oxidizing) for a potential direct reaction (Figures S14D and S12E).

BLAST Search for Homologous Sequences
The Sa YpdA sequence (Locus tag SACOL1520) was used for a BLAST (Basic Local Alignment Search Tool) search using the NCBI web interface (https://blast.ncbi.nlm.nih.gov/Blast.cgi) searching the reference proteins (refseq_protein) database searching for 20,000 sequences with a threshold E-value of 1e -6 using the BLOSUM62 matrix. The same search was also performed limiting the search to the different bacterial phyla.
The BLAST search resulted in 3977 bacterial organism hits. The dominating bacterial phylum for Sa YpdA homologous sequences was Firmicutes (1813 organism hits) followed by Bacteroidetes (1425), Proteobacteria (527), Deinococcus-Thermus (93), Actinobacteria (71) and Acidobacteria (49). Of these organisms, Proteobacteria use GSH as the main LMW thiol, and Actinobacteria use MSH, while low G+C Firmicutes and Deinococcus-Thermus contain BSH, and Bacteriodetes and Acidobacteria to a large extent contain Me-BSH 29 . Therefore, Proteobacteria and Actinobacteria were not further studied, while three of the top hits from Firmicutes, Deinococcus-Thermus, Bacteroidetes, and Acidobacteria were subjected to multiple sequence alignments and phylogenetic tree analysis through Jalview 30 . The alignments were performed with Clustal Omega 31 and tree analysis with average distances using the BLOSUM62 matrix. Figure S9 shows that the sequences fall into four clades corresponding to the four phylum classes the sequences belong to. The sequence identity compared to the Sa YpdA sequence was for the other Firmicutes in Figure S9 ~60%, for Bacteroidetes ~44%, Deinococcus-Termus ~37%, and Acidobacteria ~44%, and all these sequences have been annotated as putative YpdAs in the refseq_protein database. A cysteine in Sa YpdA has been suggested to function as an active site residue, involved in the reduction of BSSB 32 . This cysteine is found in the canonical FAD binding GXGXXG motif in Firmicutes, however, in Bacteroidetes, this residue is mainly replaced by isoleucine, in Deinococcus-Thermus valine, and in Acidobacteria threonine ( Figure S9). Further studies on the proposed YpdAs from e.g. Bacteroidetes, Deinococcus-Thermus, and Acidobacteria need to be performed to reveal if they are YpdAs or if they have other functions.

Structure Comparison -Structural Alignment Search with DALI (Distance-matrix ALIgnment)
A search for similar structures of the YpdA in the Protein Data Bank (PDB) was performed using the DALI protein structure comparison server using the Bc YpdA structure as a search template 33 . The most similar monomer structures were flavoprotein monooxygenases (FPMOs), thioredoxin reductases (TrxRs), thioredoxin-like ferredoxin NADP + oxidoreductases (FNRs), dihydrolipoyl dehydrogenases (DLDs), and gluthatione reductases (GRs) ( Table S2).
Although the NADPH and FAD domains of YpdA are similar to DLD and GR ( Figure S2, Table  S2), DLD and GR have 100-130 additional C-terminal residues. Therefore, these proteins were not included in the structural sequence alignments generated with DALI ( Figure S3). The secondary structure assignments were calculated with DSSP 34-35 and shown below the sequence. The figure was generated in JalView and colored by % identity.

Phylogenetic Analysis of YpdA with other Flavin Oxidoreductases in Selected Firmicutes
Through several BLAST searches the likely sequences of YpdAs, FNRs, TrxRs, and FPMOs in the selected Firmicutes Staphylococcus aureus subsp. aureus COL, Staphylococcus epidermidis RP62A, Bacillus cereus ATCC 14589, Bacillus anthracis str. sterne, and Bacillus subtilis subsp. subtilis str. 168 were found. Multiple sequence alignments were performed with Clustal Omega and phylogenetic tree analysis with average distances using the BLOSUM62 matrix ( Figure S6). The figures were generated in JalView and the sequence alignments were colored by % identity. The phylogenetic tree analysis shows that the four oxidoreductase types from the five phyla fall into four clades corresponding to the four oxidoreductases types, and that YpdA forms its own clade, indicating that it constitutes a separate type of flavin oxidoreductases ( Figure S6

Analysis of Conserved Residues of YpdA with ConSurf
To evaluate the degree of conservation of residues in YpdA, ConSurf [36][37][38][39] was run on the Bc YpdA structure. The run was based on the homologue search algorithm HMMER, searching sequences from UniRef90, and multiple sequence alignment with MAFFT. This gave 2266 unique HMMER hits, and ConSurf used a sample of 150 sequences that represented the list of homologues sequences to map the conservation on a 9-bin scale from turquoise (most variable) to maroon (most conserved). The conservation color coding was then mapped onto the Bc YpdA crystal structure and figures were generated with PyMOL.
Phylogenetic analysis on YpdA was performed in Jalview on the 150 sequences selected in the ConSurf runs (a few outlier sequences were not included). Clustal Omega was used for sequence alignment, and average distances in the phylogenetic tree were calculated with BLOSUM62.
The selected sequences fall into three clades in the phylogenetic trees showing that the most homologous sequences to Bc YpdA is found in other Firmicutes, Bacteroidetes and Acidobacteria ( Figure S10).
The surface representation of the ConSurf colored YpdA show that the most conserved part of the surface (white-to-maroon) is between the tetramers, which supports the conclusion of YpdA being a biological tetramer ( Figure S11).
The most conserved residues (maroon-to-lightmaroon) are found around the FAD cofactor, around the NADPH binding site, and the residues lining the solvent channel spanning the structure in connection with the FAD cofactor.

Analysis of Channels with HOLLOW
The potential channel for BSSB binding was generated with HOLLOW 40 using a 4 Å probe in cylinder mode between Phe51 and Tyr152(Bc)/Tyr147(Sa), or Lys170(Bc)/Lys165(Sa) and Tyr152 . Similarly, the sphere mode was used to look for any small channels around Cys14.

Modelling/Estimating the Position of BSSB in YpdA
To show potential BSSB binding sites around the FAD group, BSSB was manually positioned in Coot within the HOLLOW-generated channels and regularized with respect to stereochemical restraints in Coot. To span potential orientations within the whole channel, three molecules of BSSB were fitted. One was positioned with the disulfide bond close to the C4a (C4X, FAD numbering) atom of FAD (5 Å).

Ligand Interaction Generated with LigPlot + .
To make schematic diagrams of protein-ligand interactions from the protein structure (PDB files), the program LigPlot + was used with default parameters to show hydrogen bonds and hydrophobic contacts represented by dashed lines and arcs with spokes radiating toward the ligand atoms they contact, respectively 41-42 . Figure S1: NADPH consumption by YpdA under aerobic conditions. YpdA consumes NADPH at higher enzymatic rates with BSSB added to the reaction.