Structure–function analyses reveal key features in Staphylococcus aureus IsdB-associated unfolding of the heme-binding pocket of human hemoglobin

IsdB is a receptor on the surface of the bacterial pathogen Staphylococcus aureus that extracts heme from hemoglobin (Hb) to enable growth on Hb as a sole iron source. IsdB is critically important both for in vitro growth on Hb and in infection models and is also highly up-regulated in blood, serum, and tissue infection models, indicating a key role of this receptor in bacterial virulence. However, structural information for IsdB is limited. We present here a crystal structure of a complex between human Hb and IsdB. In this complex, the α subunits of Hb are refolded with the heme displaced to the interface with IsdB. We also observe that atypical residues of Hb, His58 and His89 of αHb, coordinate to the heme iron, which is poised for transfer into the heme-binding pocket of IsdB. Moreover, the porphyrin ring interacts with IsdB residues Tyr440 and Tyr444. Previously, Tyr440 was observed to coordinate heme iron in an IsdB·heme complex structure. A Y440F/Y444F IsdB variant we produced was defective in heme transfer yet formed a stable complex with Hb (Kd = 6 ± 2 μm) in solution with spectroscopic features of the bis-His species observed in the crystal structure. Haptoglobin binds to a distinct site on Hb to inhibit heme transfer to IsdB and growth of S. aureus, and a ternary complex of IsdB·Hb·Hp was observed. We propose a model for IsdB heme transfer from Hb that involves unfolding of Hb and heme iron ligand exchange.

Staphylococcus aureus is most commonly found as a member of the normal human flora, colonizing primarily on the hands and in the nostrils (1). However, S. aureus is also one of the main agents of nosocomial infections (2) and can cause a range of diseases in humans, from mild skin infections such as boils and folliculitis, to severe, life-threatening bloodstream infections (3). Establishing infection requires effective iron scavenging systems because iron trafficking is tightly controlled within the human body, both to mitigate the toxicity of free iron and to limit microbial growth. The latter has been referred to as a form of innate immunity called "nutritional immunity" (4). S. aureus overcomes this nutritional immunity by employing two mechanisms to access iron sources: the secretion and uptake of ironchelating siderophores to acquire iron from host proteins such transferrin and the expression of surface receptors for heme and hemoglobin. However, when presented with both heme and iron-bound transferrin as the only potential iron sources, S. aureus can acquire both forms of iron but preferentially utilizes heme (5).
S. aureus acquires heme using the Isd (iron-regulated surface determinant) system. This system consists of nine components: four surface proteins covalently anchored to the peptidoglycan that reversibly bind heme (IsdA, IsdB, IsdC, and IsdH); an ABC transporter (IsdF) with an associated lipoprotein (IsdE); and two intracellular heme-degrading enzymes (IsdG and IsdI) (6 -8). The function of a predicted membrane protein, IsdD, remains unknown. Lastly, sortase B (SrtB) is encoded in a gene cluster with isdCDEFG and functions to anchor IsdC to the peptidoglycan, whereas the remaining Isd surface proteins (Isd-ABH) are anchored by sortase A, the housekeeping sortase of the cell (7). The expression of all Isd components appears to be classically regulated by the Fur regulator and is repressed in the presence of iron (8). IsdB and IsdH stand apart in the Isd system in that they are the only components capable of binding hemoproteins. IsdH can bind hemoglobin (Hb), 2 haptoglobin (Hp), and the Hp⅐Hb complex. Hp is a high-abundance serum protein that binds free Hb with high affinity but does not, in itself, constitute an iron source. IsdB can bind Hb and the Hp⅐Hb complex but not Hp alone (9,10). Thus, heme is stripped from Hb at the bacterial cell surface by IsdB or IsdH and is transferred in a unidirectional relay to IsdE via IsdA and IsdC (11)(12)(13). The heme is then imported into the cell by the permease IsdF for degradation by the homologous enzymes IsdG and IsdI to liberate iron for use by the cell (14,15).
The four cell wall-anchored Isd proteins, IsdABCH, share three features: an N-terminal secretion signal, a C-terminal sortase signal for cell wall anchoring, and one to three copies of a NEAT (for near transporter) domain (16). IsdA and IsdC contain a single NEAT domain, which binds heme. IsdB and IsdH contain two and three NEAT domains, respectively, but only their C-terminal NEAT domains, IsdB N2 and IsdH N3 , respectively, bind heme. The N-terminal and central NEAT domains of IsdH, recombinantly expressed individually as IsdH N1 and IsdH N2 , are each able to bind Hb, Hp, and Hp⅐Hb (9,17); however, the N-terminal IsdB domain alone, IsdB N1 , does not bind Hb (18), despite Ͼ40% sequence identity between IsdB N1 , IsdH N1 , and IsdH N2 and the shared ability of the full-length proteins to bind Hb. Both IsdB N1 and IsdB N2 must be present and contiguous with the intervening "linker" region (represented as IsdB N1N2 ) for high affinity Hb binding. Moreover, IsdB N1N2 removes heme from oxidized Hb, known as methemoglobin (metHb), the form of Hb produced upon red blood cell lysis in the bloodstream. By contrast, IsdB N1N2 does not remove heme from oxyHb, the reduced, oxygen-bound form of Hb that is present in intact red blood cells, demonstrating specificity toward the probable biologically relevant form of Hb encountered by S. aureus during infection (18).
Structures of the Isd proteins have been elucidated by X-ray crystallography or NMR, revealing the mode of heme coordination within NEAT domains (17, 19 -25). Recently, crystal structures of various portions of IsdH in complex with Hb have also been reported: IsdH N1 ⅐metHb (17), IsdH N2 ⅐metHb (19), and a heme transfer-deficient variant of IsdH N2N3 (Y642A) complexed with metHb (19,26). These complex structures provide insight into the extensive interactions between IsdH and Hb. In these structures, the heme remains largely encapsulated by metHb with minor or no distortions of the ␣Hb heme binding pocket. Therefore, these structures only provide insight into the initial step of the heme transfer process. These insights likely extend to IsdB⅐Hb, given the similarities between the two surface receptors. Nevertheless, IsdB is more important than IsdH in both in vitro growth on Hb and a mouse abscess model of infection (10,27). IsdB is also a dominant antigen that is highly up-regulated in blood, serum, and a cage model of tissue infection (28 -31), corroborating the key role of this receptor.
Herein, we report the crystal structure of an IsdB N1N2 ⅐Hb complex in which the heme is positioned between the two proteins, consistent with an intermediate state of heme transfer. Two point mutations were then introduced into IsdB N1N2 to trap a similar species in solution, which was characterized spectroscopically. In concordance, kinetic analyses of heme transfer revealed a multistep transfer process. Furthermore, we demonstrate that although IsdB N1N2 can bind to the Hp⅐metHb, as was previously demonstrated for IsdH, it is unable to remove heme from the complex. This finding correlates with our observation that S. aureus did not utilize Hp⅐Hb as a sole iron source.

Crystal structure of an IsdB N1N2 ⅐Hb complex
Because IsdB N1N2 does not extract heme from oxyHb (18), this form of Hb was used to obtain an IsdB N1N2 ⅐Hb complex for crystallization. Crystals of an IsdB N1N2 ⅐Hb complex were grown from solutions of citric acid or malonic acid, pH 5.0 -5.5, and 2.1-2.4 M ammonium sulfate. Small crystals (ϳ0.05-0.1 m) appeared after 2-4 weeks and contained both proteins (Fig. 1, left panel). A 3.6 Å resolution data set from a single crystal was collected, and the structure was solved by molecular replacement of the individual components. The structure revealed a central Hb a 2 b 2 tetramer surrounded by four IsdB molecules in the asymmetric unit (Fig. 1, right panel). An intact copy of the recombinant IsdB N1N2 construct is bound to each ␣Hb subunit; however, only the first NEAT domain of IsdB (IsdB N1 ) is observed bound to each ␤Hb subunit. The 19-kDa band observed in the left panel of Fig. 1 is consistent with proteolysis of IsdB N1N2 during crystallization and is presumed to be IsdB N1 . Inspection of crystal packing within the unit cell revealed that modeling the missing components of IsdB would result in a large scale steric clash with ␣Hb-bound IsdB N1N2 , suggesting that the partial proteolysis allowed for crystallization of the complex.

S. aureus IsdB-associated unfolding of human hemoglobin
The structure of each IsdB N1N2 molecule resembles a dumbbell with the two NEAT domains joined by an ␣-helical linker ( Fig. 2A, left panel), as was observed for IsdH N2N3 (19,26). The loop between IsdB N1 and the linker region is flexible, as evidenced by poor electron density; conversely, well ordered density is observed between the linker and the second NEAT domain (IsdB N2 ), which includes a short 3 10 helix. The IsdB N1 and IsdB N2 domains of intact IsdB N1N2 each make numerous interactions with ␣Hb (average interface area of 780 and 736 Å 2 , respectively), and similar interactions are observed for IsdB N1 bound to ␤Hb (634 Å 2 ). No interactions are observed between the linker region and Hb.
The Hb chains of the complex described here reveal major structural rearrangements as compared with the structure of isolated oxyHb (PDB code 2DN1). These changes are situated primarily in the two ␣Hb chains of the IsdB N1N2 ⅐Hb structure, which superimpose poorly with the equivalent ␣ chains in the free oxyHb structure (r.m.s.d. of 3.3 and 3.4 Å over all C␣) as compared with alignments between the ␤Hb chains (r.m.s.d. of 1.4 and 1.3 Å over all C␣). In the ␣ subunit of free human oxyHb, heme is bound in a pocket between the E and F helices, with His 87 (F helix) coordinating directly to the heme iron and the distal His 58 (E helix) forming a hydrogen bond to the hemebound dioxygen molecule. To our knowledge, the heme iron axial residue is ␣His 87 in all previously reported ␣Hb crystal structures regardless of the gaseous heme ligand. Upon binding to IsdB N1N2 , a major deformation of the ␣Hb heme pocket occurs, with the F helix and part of the E helix of each ␣Hb chain (encompassing residues ␣Asp 74 to ␣Arg 92 ) unwinding entirely (Fig. 2B). Clear electron density in this region indicated that heme (modeled at full occupancy) remains bound to ␣Hb ( Fig.  3A) but has been displaced 5 Å toward the heme-binding pocket of IsdB N2 . This heme displacement was accompanied by the unexpected direct coordination of the heme iron by ␣His 58 and ␣His 89 (Fig. 3B). Through reorganization of the polypeptide chain, ␣His 87 has pivoted out of the heme pocket and . IsdB N2 directly interacts with the Hb heme and heme pocket. The linker is a three-helix bundle between the two NEAT domains. IsdB N1 binds onto the opposite face of ␣Hb. Right panel, the conformation of the ␣Hb C-D loop is significantly altered upon IsdB binding. B, stereo view of the interaction with IsdB, which results in the ␣Hb F helix becoming highly unwound. The ␣Hb chain of oxyHb (PDB code 2DN1; green) is overlaid to demonstrate the original state of the helix. The heme is illustrated in blue sticks with heme iron in red. C, stereo view of the interaction between ␤Hb (orange) and IsdB N1 (cyan). The ␤Hb chain of oxyHb with associated heme (PDB code 2DN1; gray) is overlaid. The orientation of the molecules is similar to that of A. Density for the ␤Hb F helix and heme is absent, as are the IsdB linker and NEAT2 domains.

S. aureus IsdB-associated unfolding of human hemoglobin
points away from the heme iron (N⑀2 translation of 13.4 Å). In its place, ␣His 89 (originally on the F helix) has moved from a solvent-exposed position to coordinate to the heme iron (13.2 Å translation in N⑀2). Binding to IsdB N1N2 also resulted in the extension of the C-terminal H helix by another half turn, such that ␣Tyr 140 occupies the void left by the unwound F helix, with an accompanying C␣ shift of 7.5 Å and the placement of the side chain phenol group within 3.2 Å from N⑀2 of ␣His 87 . Eleven residues between helices C and D (␣Tyr 42 to ␣Ser 52 ) have also rearranged to accommodate the interaction with IsdB N1N2 (Fig. 2A, right panel). For the ␤ chains of Hb, insufficient density was present to model most of the F helix (␤Leu 88 to ␤Val 98 in chain B and ␤Phe 85 to ␤Val 98 in chain D), and no bound heme was observed (Fig. 2C). Otherwise, minimal differences were observed between the ␤ chains of free and IsdBcomplexed Hb.
The IsdB N2 domain interacts with the heme-binding site of ␣Hb and superposed well with the structure of free IsdB N2 (20) (r.m.s.d. of 1.4 Å over 109 C␣). However, the ␤7-␤8 loop (Val 435 to Tyr 440 ), which coordinates heme iron via Tyr 440 , is curved inward in the complex structure, making contacts with ␣Hb, and is poised to receive the heme molecule (Fig. 3C). Tyr 440 closely abuts ␣His 89 of Hb, whereas Tyr 444 is positioned adjacent to the heme pyrrole ring, forming a -stacking network of interactions. Interestingly, Tyr 440 is positioned further back than Tyr 444 , suggesting that Tyr 444 plays a role beyond simply stabilizing the position of Tyr 440 in the heme pocket, as was previously proposed (20), and appears to be required for the heme transfer process. The second heme iron-coordinating residue of holo IsdB N2 , Met 362 , has poor electron density in the complex structure and is likely conformationally flexible. Conversely, the IsdB N2 propionate-binding residue Ser 361 remains engaged in hydrogen-bonding with the propionate, with Glu 354 also participating in this interaction. The second heme propionate is not observed to form H-bonds to either Hb or IsdB N1N2 .
Proteolyzed IsdB N1 (chains J and H) interact with ␤Hb and are structurally similar to the IsdB N1 domains in intact IsdB N1N2 (chains E and F), which interact with ␣Hb in the crystal structure (r.m.s.d. of 0.9 -1.2 Å over 133 C␣). Minor differences are observed mainly in the ␤2-␤3 and ␤7-␤8 loops (numbered ). An anomalous density map (pink) contoured at 3 is overlaid to unambiguously support correct placement of the heme. B, the complex structure is overlaid with the structure of the ␣Hb chain from the oxyHb structure (PDB code 2DN1). The 2DN1 ␣Hb structure is shown in green; ␣Hb and IsdB N1N2 from the IsdB N1N2 ⅐Hb structure are shown in beige and blue, respectively. Relevant amino acid side chains are shown as sticks. Only the heme moiety from the structure presented here is shown (dark green). C, the IsdB heme pocket is positioned to accept ␣Hb-heme. IsdB and ␣Hb residues situated at the binding interface near the heme are shown as blue and beige sticks, respectively. The remainder of the structure is shown as cartoons. The ␣Hb-heme is shown in green, with the heme iron (coordinated by ␣His 58 and ␣His 89 ) shown as an orange sphere.

S. aureus IsdB-associated unfolding of human hemoglobin
according to IsdA N1 (21)) and at the N and C termini. However, Hb-bound IsdB N1 differed substantially from the solution structure of IsdB N1 (25) (r.m.s.d. of ϳ2.3 Å over all C␣ of model 1). One significant structural difference was within the fourresidue aromatic motif that is important for Hb-binding by IsdB and IsdH (27,32). This motif, FYHY in IsdB (residues 164 -167), was disordered in the solution structure but formed a short ␣-helix in the Hb-IsdB N1N2 structure. This ␣-helix forms a close contact with Hb, with Phe 164 closely abutting ␣Trp 14 /␤Trp 15 in a T-shaped -stacking interaction.

An IsdB heme pocket variant traps a bis-His metHb
Although the IsdB N1N2 ⅐Hb structure clearly showed the Hbheme in a bis-His coordination state, a bis-His state involving the native proximal and distal heme ligands was previously observed in crystals of horse metHb grown at pH 5.4 as compared with pH 7.1 (33). Although the heme ligands differ, the low pH of our IsdB N1N2 ⅐Hb crystallization solution could have similarly induced bis-His coordination. However, incubation of metHb or oxyHb at pH 5.5 overnight, with or without IsdB N1N2 , did not produce spectral changes associated with formation of a bis-His state (data not shown). Nonetheless, the observation of conformational changes in horse metHb to give a bis-His coordination state is a precedent for the conformational rearrangements observed in the IsdB N1N2 ⅐Hb structure.
A double mutant of IsdB N1N2 (IsdB YFYF ) was created to prevent heme transfer by replacing Tyr 440 (heme iron coordinating) and Tyr 444 (H-bonds to Tyr 440 ) with Phe residues in the heme pocket. In the complex structure, these two tyrosine residues are also juxtaposed to ␣His 89 of Hb, the heme-coordinating residue that is part of the unwound F helix. Upon addition of free heme to apo-IsdB YFYF , the spectra displayed a Soret peak at 403 nm, slightly blue-shifted relative to holo-IsdB N1N2 , which peaks at 405 nm (Fig. 4A). The ␣/␤ region of the IsdB YFYF ϩheme spectrum was nearly featureless with a minor peak at ϳ600 nm, similar to the spectrum of free heme in that region (Fig. 4B).
As previously reported (18), the addition of IsdB N1N2 to metHb resulted in a large-scale shift in the intensity and shape of the Soret peak as heme was transferred, with minor changes in peak intensities in the visible region (Fig. 4A). Visible region peak wavelengths remained at ϳ500 and 535 nm, with a chargetransfer band at 630 nm, indicating that the heme was a highspin ferric species in both holo-IsdB N1N2 and metHb (34). Addition of IsdB YFYF to metHb also caused a large-scale shift in the Soret peak; however, the spectra was distinct from that formed upon heme addition to IsdB YFYF (Fig. 4A). The Soret peak became red-shifted, to 410 nm, and a distinct shoulder developed at ϳ360 nm. Moreover, dramatic changes occurred in the ␣/␤ region (Fig. 4B); the charge-transfer band at 630 nm disappeared, and the highest absorption peak shifted to 533 nm with a shoulder peak at 564 nm. This spectrum is nearly identical to that of a ferric, low-spin, bis-His hemichrome form of human Hb observed when incubating purified ␣or ␤-globin with heme (but not for heme addition to whole globin) (34). The change to a characteristic bis-His hemichrome spectrum upon addition of IsdB YFYF to 4 N metHb was titratable and saturable (Fig. 5, A and B). A plot of the absorption change at a single wavelength (410 nm) as a function of added IsdB YFYF plateaued at ϳ5 M, indicating the stoichiometry of the reaction was ϳ1:1 (Fig. 5C), as seen for interaction of IsdB N1N2 and Hb in the crystal structure.
The binding of IsdB YFYF to metHb was confirmed by ITC. Titration of metHb into IsdB YFYF resulted in an exothermic reaction as observed by the negative change in enthalpy (Fig.  S1). Analysis of the data with a one-site model gave a K d of 6 Ϯ 2 M and a stoichiometry (N) of ϳ0.5 (average of three runs), implying two IsdB YFYF molecules bound to one metHb monomer. Because Hb has two unequal subunits that may interact with IsdB N1N2 differently, the K d and stoichiometry measurements are assumed to be the average of binding to ␣Hb and ␤Hb. A previous ITC study between IsdB N1N2 and carboxyhemoglobin reported a K d of 0.42 Ϯ 0.05 nM (18). The weaker interaction of IsdB YFYF for metHb may be due to a conformational change in the structure of metHb analogous to that observed in the IsdB N1N2 ⅐Hb crystal structure.

Rapid heme transfer from metHb to IsdB N1N2
To identify potential intermediates in the heme transfer pathway, the kinetics of heme transfer from metHb to IsdB N1N2 was investigated by stopped-flow spectroscopy. Mixing of 2 N metHb with 20 M IsdB N1N2 resulted in spectral changes between 180 and 730 nm that were complete within 10 s, consistent with previous reports using a full-length IsdB construct (11). Coincident with heme transfer, the Soret peak underwent a large shift in intensity, particularly between ϳ350 and 420 nm, along with spectral changes in the ␣/␤ region (Fig. 6A).

S. aureus IsdB-associated unfolding of human hemoglobin
Although the intensity of the Soret absorption peak increased during the first ϳ60 ms (Fig. 6B), the overall effect was a reduction in intensity with broadening of the Soret band.
Two wavelengths (406 and 428 nm) were chosen for singlewavelength stopped-flow spectroscopy under pseudo-first order conditions, with metHb held constant at 1 N and IsdB N1N2 increasing from 5 to 40 M. The kinetics at 428 nm were simpler (Fig. 6C), because this wavelength was outside the range that increased in the first 60 ms (ϳ350 -420 nm). A single exponential fit yielded observed rate constants (k obs ) that varied hyperbolically with IsdB N1N2 concentration (Fig. 6D), consistent with a two-step heme transfer mechanism (20) with a rate constant for heme transfer from metHb to IsdB N1N2 of 0.35 Ϯ 0.02 s Ϫ1 . At 406 nm, the kinetics were more complex. Curve fitting yielded four phases with differing amplitudes, rates, and concentration dependences (Fig. S2). In the first phase only, the rate was linearly dependent on IsdB N1N2 concentration (Table  1; see also Fig. S2C) and thus reflected concentration dependent collision events. Phases 2-4 did not display strong concentration dependence in their rates (Fig. S2, D-F), suggesting that they are associated with steps in the heme transfer process after IsdB N1N2 ⅐metHb complex formation.

Hp prevents heme transfer from metHb to IsdB N1N2
Hp is a serum ␣2-sialoglycoprotein that binds metHb released from lysed erythrocytes to prevent oxidative damage (35,36). The fundamental unit of human haptoglobin is a polymorphic ␣␤ dimer composed of light chains (␣) and heavy chains (␤). The ␤ chain is encoded by a single allele, whereas the ␣ chains come in two forms: ␣ 1 and ␣ 2 (36). These alleles generate three possible phenotypes. Phenotype 1-1 (where both copies of the ␣ gene are ␣ 1 ) is the simplest, with Hp forming a homodimer of two ␣␤ heterodimers. Phenotypes 2-1 and 2-2 can form heterogeneous cyclical or linear multimers of increasing size (37). Within the human host, the normal plasma Hb concentration is less than 5 N, whereas Hp is generally present in the range of 0.3-2 mg/ml (35) or roughly 6 -50 M depending on the phenotype.
Binding of metHb by Hp has been proposed to make heme inaccessible to microbial pathogens (38). Therefore, to investigate the effect of Hp on heme uptake by IsdB from Hb, excess IsdB N1N2 was added to metHb preincubated with increasing concentrations of human Hp. A pooled mixed phenotype Hp was used to model normal human serum; however, the exact molar concentration was unknown. Therefore, Hp was used at the high end of the normal range for Hp serum concentration (2 mg/ml). Preincubation of 2 N metHb with 2 mg/ml Hp resulted in electronic spectra that remained unchanged for 5 min after the addition of IsdB N1N2 (Fig. 7A). Only by decreasing the Hp concentration to 0.02 mg/ml or lower was a spectral change in the Soret region observed, implying heme transfer similar to that observed in the absence of Hp (Fig. 7B). Thus, Hp effectively blocked heme transfer from metHb to IsdB N1N2 . Inhibition was not due to the inability of IsdB N1N2 to bind metHb when the latter is complexed to Hp because IsdB N1N2 was able to pull down metHb in both the presence and absence of Hp but did not interact with Hp alone (Fig. 7C). To elucidate the stoichiometry of the Hp inhibition, the heme transfer assay was repeated using Hp phenotype 1-1, which revealed a stoichiometric ratio of approximately one Hp 1-1 molecule (consisting of two ␣␤ subunits) to one a 2 b 2 metHb tetramer (Fig. S3).

S. aureus growth on Hb⅐Hp as sole iron source
To examine the effects of Hp on Hb-heme utilization in vivo, the growth of S. aureus strain Newman was evaluated in iron-depleted RPMI media. As expected, S. aureus was able to grow in medium supplemented with 2 N heme iron or 0.2 N oxyHb as a sole iron source ( Fig. 8; see also Fig. S4). A higher concentration of heme was

S. aureus IsdB-associated unfolding of human hemoglobin
required because 0.2 N heme did not support growth of S. aureus (data not shown). OxyHb is expected to rapidly oxidize to metHb in culture. Assuming one Hp ␣␤ dimer binds to one dimer of Hb, growth on Hb with the addition of 0.44 -10 g/ml Hp reduced the growth of S. aureus in a concentration-dependent manner. Conversely, addition of Hp to medium supplemented with heme did not reduce the growth of S. aureus, implying that Hp had a Hbspecific inhibitory effect.

Discussion
IsdB is the major Hb receptor functioning at the interface between the bacterial cell surface and the extracellular environment. The heme extraction function of IsdB is supported by the observed growth deficiency of S. aureus on nanomolar concentrations of Hb as a sole iron source upon deletion of IsdB (27) and more directly by heme transfer assays between metHb and IsdB (18). In contrast, IsdB N1N2 was unable to extract heme

S. aureus IsdB-associated unfolding of human hemoglobin
from oxyHb within hours of incubation, and this form of Hb was used in crystallization trials to obtain the structure of a complex between the two proteins.
Crystals of IsdB N1N2 ⅐Hb formed slowly over 2 weeks, sufficient time for oxyHb to be oxidized to metHb. Indeed, the heme groups from the ␤Hb subunits are no longer observed in the complex structure and presumably were extracted by IsdB N1N2 that was subsequently degraded, leaving only the IsdB-N1 domain. The conformations of the ␣Hb subunits in the complex with IsdB N1N2 are a large departure from all previously published structures of Hb. The large distortion of the polypeptide chain and the bis-His heme coordination state are accompanied by a shift of the heme from Hb to the interface with IsdB. The structure suggests that heme is removed from ␤Hb before ␣Hb, which may be a consequence of the kinetics or thermodynamics of crystal formation. The observed bis-His form of Hb observed in the crystal structure may be an artifact of crystallization and represent an off-path species. Attempts to identify this putative intermediate in solution by stopped-flow kinetics have not yet met with success, possibly because the putative intermediate is short-lived. However, spectroscopic analysis of the IsdB YFYF variant mixed with metHb supports the trapping of a bis-His heme complex in solution, thereby supporting the possible existence of a bis-His heme intermediate in the heme transfer pathway by the wild-type protein.
An attractive model of heme transfer from Hb to IsdB is suggested when the IsdB N1N2 ⅐Hb complex structure is overlaid with the structures of oxyHb (PDB code 2DN1; ␣ chain) and holo-IsdB N2 (PDB code 3RTL) (Fig. 9A). In the IsdB N1N2 ⅐Hb structure presented here, the heme moiety is midway between the Hb and IsdB heme pockets (heme iron is ϳ5 Å from either position). The motion of the heme is not solely translational, as one of the propionate groups in the complex is in nearly the same position as observed in the holo-IsdB N2 structure (Fig. 9). Instead, H-bond interactions between the propionate and both Ser 361 and Glu 354 appear to anchor the heme as it rotates ϳ90°f rom the Hb heme pocket to the IsdB heme pocket, providing the most parsimonious route of heme transfer between the start and end states (as represented by the uncomplexed structures). Interestingly, the structures of oxyHb and our complex may also provide insight into the ambiguity observed in the electron density for the heme methyl/ethyl groups in the previously solved structure of holo-IsdB N2 , which resulted in two possible heme conformations while maintaining the position of the propionate groups (20). Because it is unlikely that the heme face can flip over during transfer, one can argue that heme conformation A, as shown in Fig. 9B, is the biologically relevant heme binding mode when transferred from Hb.
The interaction between IsdB N1 and ␤Hb may represent an interaction after a successful heme transfer event, because the ␤Hb heme and the IsdB NEAT1 and linker domains are not present. Additionally, helix F of ␤Hb, which contains ␤His 92 (equivalent to axial heme-coordinating His 87 of ␣Hb) and ␤His 97 , are disordered in the absence of bound heme. ␤His 97 is a single turn helix away from the equivalent position of ␣His 89 and may participate in the formation of a similar bis-His conformation in ␤Hb. For the transfer to occur, ␤Hb heme must have oxidized because IsdB N1N2 cannot remove ferrous heme from Hb (18). This slow oxidation is expected because the crystals of the complex took weeks to form. Subsequent to the oxidation, the heme is rapidly transferred to IsdB N2 , and the flexible loop between IsdB N1 and the linker is cleaved to allow for growth of the crystal. Interestingly, under physiological conditions, ␣Hb oxidizes seven to ten times more quickly than ␤Hb, especially at acidic pH (our crystals were produced at pH 5.5) (39,40). The observation of heme removal from ␤Hb in the crystal structure may be explained by the differential positioning of ␣His 89 and ␤His 97 in ␣Hb versus ␤Hb, respectively, which may be a rate-determining factor in the heme uptake mechanism. The asynchronous heme extraction from ␣Hb and ␤Hb observed in the crystal structure is mirrored by the kinetic analysis of heme transfer monitored at 408 nm. The four distinct kinetic phases observed may correspond to steps in heme extraction from the each Hb domain; however, alternative models are possible because of the presence of different subunits and the complex allosteric nature of Hb. The complexity of the heme transfer kinetics is expected for a substrate (Hb) that is a tetramer with two non-equivalent sites where the rate of heme extraction and the spectral change would vary with the heme site and the number of heme molecules bound to the

S. aureus IsdB-associated unfolding of human hemoglobin
tetramer. This complexity may also be hindering our attempts to observe spectral features of the putative bis-His intermediate. Nonetheless, the rates observed are similar or faster than the overall rate of heme transfer observed under steady-state conditions with IsdA as the ultimate heme recipient (18).
Superposition of the ␣␤ tetramer of Hb in complex with IsdB N1N2 upon the tetramers of oxyHb (PDB code 2DN1), deoxyHb (PDB code 2DN2), or metHb (PDB code 3P5Q) revealed an unusual quaternary structure. The IsdB N1N2 -bound tetramer exhibited neither a T-like (deoxy) nor an R-like (oxy or met) state conformation. This quaternary structure was also observed in a crystal structure of a heme-transfer deficient mutant of IsdH N2N3 in complex with metHb (PDB code 4XS0) (26). However, the overall IsdH N2N3 ⅐metHb structure is more similar to that of oxyHb (␣Hb, r.m.s.d. of 1.2 Å; ␤Hb, r.m.s.d. of 0.5 Å) than our IsdB N1N2 ⅐Hb structure and oxyHb (␣Hb, r.m.s.d. of 3.3-3.4 Å; ␤Hb, r.m.s.d. of 1.3-1.4 Å). Dickson et al. (26) noted that the Hb dimer in IsdH N2N3 ⅐metHb structure was most similar to structures of unusual liganded hemoglobins in a T-like state, such as a human sickle-cell variant of embryonic Hb. Our structure of IsdB N1N2 ⅐Hb supports the hypothesis that distortion of the Hb quaternary structure into a liganded T-like state is a characteristic of the Hb-binding and heme-extraction process by Isd proteins.
Hp is present at 10 mg/ml in normal serum (37) and has a high affinity for dimeric Hb (K d ϭ ϳ10 Ϫ15 M (41)). Therefore, metHb present in serum is likely to exist as a complex with Hp. We have shown that heme transfer from metHb to IsdB N1N2 is inhibited by Hp, despite the ability of IsdB to bind Hp⅐Hb (9). In fact, under the conditions tested, heme transfer only occurred when the concentration of mixed-serotype Hp was decreased to 20 g/ml, or ϳ500-fold lower than the concentration of Hp in human serum. The data presented here are consistent with a recent study demonstrating Hp inhibition of heme transfer from metHb to a similar IsdB N1N2 construct but not to an analogous IsdH construct (42). Moreover, in contrast to IsdH, IsdB did not inhibit binding of Hb-Hp to CD163 (42), the macrophage receptor for recycling Hp⅐Hb complexes (43).
Binding of Hp to Hb has also been shown to stabilize the heme-Hb interaction (44). A structural alignment of our IsdB⅐Hb crystal structure with the porcine (PDB code 4F4O) and human (PDB code 4X0L) Hp⅐Hb structures revealed no steric clashes between IsdB N1N2 and Hp (Fig. 10). Thus, inhibition of heme uptake from the Hp⅐Hb complex may reflect modulated stability of heme in the Hb pocket rather than direct blocking of the IsdB binding site by Hp. More specifically, examination of the Hp⅐Hb structures reveals a Hp loop (Pro 327 -Gly 329 in human Hp; Pro 268 -Gly 270 in porcine Hp) that interacts with F helix at Pro 78 of ␣Hb. This interaction needs to be disrupted to unwind the F helix for subsequent heme transfer to IsdB.
The IsdB N1N2 construct could possibly lack regions required to remove heme from Hp⅐Hb complexes in vitro. However, S. aureus was not able to use Hb⅐Hp as a sole iron source, whereas nanomolar concentrations Hb alone supported S. aureus growth (Fig. 8). Previous studies have indicated that S. aureus can grow on Hb⅐Hp as sole iron source (45,46); however, the quality of Hb was lower, and the concentration used in the growth medium was much higher than the more physiologically relevant concentration used here. In the work of Francis et al. (46), cells were grown in iron-rich tryptone and yeast extract, conditions under which the expression of the Isd system is repressed, and iron uptake was measured based on radioactivity of the cell pellet, a method that cannot distinguish between surfacebound and internalized iron. Recent studies have highlighted the importance of using human Hb from donors over commercial lyophilized Hb because of contamination with free heme (27,47). During infection by S. aureus, hemolysin production and local hemolysis may be required to sufficiently increase the Hb concentration, allowing for excess Hb not bound to Hp.

Cloning
Plasmids were generated encoding residues 126 -459 of IsdB with an N-terminal His 6 tag and a thrombin cleavage site in pET28a(ϩ)as previously described (18). Briefly, the IsdB N1N2 construct was subcloned from a GST-tagged construct cloned from S. aureus N315 chromosomal DNA (20) using a modified whole plasmid polymerase chain reaction method (48). A dou- ble variant of IsdB N1N2 was created by subcloning from the IsdB N1N2 clone using the QuikChange site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA). All clones were verified by sequencing (Agencourt, Beverly, MA).

Protein expression and purification
Recombinant protein was overexpressed in Escherichia coli BL21 (DE3) cells. A 2-liter bacterial culture was grown from 2 ml of overnight culture in LB broth supplemented with 25 g/ml kanamycin at 30°C to an A 600 nm of 0.7-0.9 and then induced with 0.5 mM of isopropyl ␤-D-thiogalactopyranoside and grown for another 18 h at 25°C. The cells were pelleted by centrifugation; resuspended in 20 ml of 50 mM Tris, pH 8.0, 100 mM NaCl; and then lysed at 4°C using an EmulsiFlex-C5 homogenizer (Avestin, Ottawa, Canada). Insoluble material was removed by centrifugation; the soluble lysate contained a mixture of apo and holo His 6 protein, and apo protein could be separated at 4°C using a HisTrap nickel affinity column (GE Healthcare) by elution with an imidazole gradient. The apo protein was dialyzed against 50 mM Tris, pH 8.0, 100 mM NaCl and then cleaved with thrombin at a 1:500 ratio by weight of His 6 protein to remove the His 6 tag, leaving behind a two amino acid (Gly-Ser) N-terminal artifact. Recombinant protein was then dialyzed against 20 mM HEPES, pH 7.4, for cation exchange chromatography using a Source 15S column (GE Healthcare), and purified protein was obtained by elution with a NaCl gradient. The resulting pure (Ͼ95% by SDS-PAGE) apo protein was dialyzed against 20 mM HEPES, pH 7.4, 80 mM NaCl for all studies. Hemoglobin was prepared from fresh human blood as described previously.

Stopped-flow spectroscopic analysis of heme transfer from metHb to IsdB
An SX.18MV stopped-flow reaction analyzer (Applied Photophysics Ltd., Leatherhead, UK) was used to investigate the possibility of heme-coordination intermediates in the transfer process between metHb and IsdB N1N2 . The temperature of the optical cell and drive syringe chamber was maintained at 25°C using a circulating water bath. Multiple wavelength data from 180 to 730 nm were collected using the photodiode array detector of the system directly coupled to the xenon light source (the practical range was from 250 to 730 nm). 20 M apo-IsdB N1N2 and 2 N metHb samples were in a buffer composed of 20 mM NaH 2 PO 4 , pH 7.4, 50 mM NaCl. Each acquisition was 15 s long with spectra collected at logarithmic intervals; four acquisitions were averaged.
A monochromator was used to collect single-wavelength data at 406 and 428 nm for the reaction of metHb and IsdB N1N2 in 20 mM HEPES, pH 7.4, 80 mM NaCl. The reactions were carried out with 2 N metHb in one syringe (final concentration, 1 N) and concentrations of apo-IsdB N1N2 ranging from 10 to 80 M (final concentrations, 5-40 M) in the second syringe; a minimum 5-fold excess of the IsdB acceptor was used to attain pseudo-first order conditions. The temperature was again maintained at 25°C, and five 30-s acquisitions were performed at each wavelength for each concentration pair. The first 3 ms were in the dead time of the instrument and thus were excluded from analysis. At least four acquisitions were aver-aged, and curve fitting was performed using the ProDataSX software.

Crystallization and data collection
Solutions of apo-IsdB N1N2 and oxyHb in an equimolar (by monomer) ratio were mixed together at 10 mg/ml and immediately used to set up crystal trays. The IsdB N1N2 ⅐Hb complex was crystallized in space group P2 1 2 1 2 using reservoir solution containing 0.1 M malonic acid, pH 5.5, 2.2 M ammonium sulfate in 96-well sitting drop plates. Crystals appeared within a few days at room temperature but were allowed to grow to a sufficient size for ϳ4 weeks. Crystals were cryoprotected with 30% sodium malonate, pH 7.0, and flash frozen in liquid nitrogen. A 3.60 Å resolution data set was collected at the Stanford Synchotron Radiation Lightsource on Beamline 7-1 (1.000 Å wavelength) at 100 K. The data were processed using HKL2000 (49) ( Table 2). The data set was initially phased by molecular replacement using the oxyhemoglobin (PDB code 2DN1) and IsdB N2 (PDB code 3RTL) crystal structures and a SWISS-MODEL (50) generated IsdB N1 model (based on IsdH N1 PDB code 4IJ2) as search models with Phaser-MR (51) in the PHENIX suite of programs (52), which found the hemoglobin tetramer, three copies of IsdB N1 and three copies of IsdB N2 . Manual inspection of the placement of IsdB N1 and IsdB N2 molecules, particularly at non-conserved bulky residues such as Tyr and Phe, revealed that electron density for one Phaser-placed copy of IsdB N1 was better fit by IsdB N2 , and the model was thus corrected to include four copies of IsdB N2 and two copies of IsdB N1 . SWISS-MODEL was also used to generate a model for the IsdB linker region based on the Hb-IsdH complex crystal structure (PDB code 4IJ2), which was added in a subsequent round of molecular replacement. Manual rebuilding with Win-Coot (53) was used in all stages to complete the structure, and refinement was carried out using phenix.refine (54). Heme molecules were placed into omit difference maps, and the position of iron was confirmed by an anomalous difference map.

S. aureus IsdB-associated unfolding of human hemoglobin
The heme-iron ligand geometry was not restrained. The polypeptide chain has excellent stereochemistry, with 95.2% of residues in the favored region of the Ramachandran plots and 4.5% in allowed regions, and 0.3% were outliers.

Hb؉IsdB N1N2 crystal composition
Crystals from a well of oxyHbϩIsdB N1N2 were looped and soaked in well solution to remove adventitiously bound protein and then dissolved in 5 l of 20 mM HEPES, pH 7.4, 80 mM NaCl. Five small crystals in total were dissolved together and separated on a 15% SDS-PAGE gel to visualize the components of the crystal.

Haptoglobin inhibition assay
Human Hp of mixed serotype purified from plasma was purchased as a lyophilized solid (Athens Research and Technology, Athens, GA). 1 mg of Hp was dissolved in 100 l of 20 mM HEPES, pH 7.4, 80 mM NaCl to give a concentration of 10 mg/ml. Spectra of 2 N metHb alone or combined with 10 M IsdB N1N2 and 2 mg/ml Hp (precombined with metHb) were taken from 250 -750 nm in a conventional spectrophotometer (Cary50) with an optical path length of 1 cm in a quartz cell at room temperature, 22°C. Subsequently, the minimum inhibitory concentration of Hp was determined by adding decreasing amounts of Hp under the same conditions (beginning with 2 mg/ml as above) until the heme transfer reaction from metHb to IsdB N1N2 was observed to proceed. Spectra (250 -750 nm) were recorded immediately after mixing, as before. The reaction was monitored by recording additional spectra at 2 and 5 min after the initial spectrum.

IsdB N1N2 ⅐Hb⅐Hp pulldown assay
The ability of nickel bead-bound His 6 -IsdB N1N2 to pull down metHb⅐Hp complexes was tested. Six 25-l aliquots of nickel bead slurry (Chelating Sepharose Fast Flow, GE Healthcare; stored in 20% ethanol) were washed twice with 1 ml of dH 2 O, followed by a 1-ml wash of binding buffer, consisting of 20 mM HEPES, pH 7.4, 80 mM NaCl, 75 mM imidazole. The moderate amount of imidazole was a concentration at which His 6 -IsdB could bind the nickel beads but Hb could not. To each aliquot of washed beads, 50 l of 20 M His 6 -IsdB N1N2 , metHb, or Hp (alone for controls or together for pulldowns, as indicated) were added to the beads and kept at room temperature on the benchtop. Samples were incubated for 15 min and gently agitated occasionally. After incubation, the 50-l supernatant was removed, and beads with bound protein were washed twice with 500 l of binding buffer and then eluted with 1 M imidazole, pH 7.5. 4 l of each eluate was run on a 15% SDS-PAGE gel at 200 V for 1 h and 5 min and stained using Coomassie Blue.

Isothermal titration calorimetry (ITC)
ITC was conducted using a MicroCal ITC-200 instrument at 25°C. Syringe and cell samples were co-dialyzed overnight against 20 mM HEPES, pH 7.4, 80 mM NaCl. Dialysis buffer was collected and sterile-filtered for diluting samples and washing the ITC cell. ITC experiments were conducted with 800 N (200 M) of metHb as the titrant in the syringe and 80 M of IsdB YFYF in the cell. Binding isotherms were analyzed with MicroCal Origin 7.0 software using a one-site model. Three ITC experiments were conducted to obtain an average dissociation constant (K D ) and stoichiometry (N), and standard deviations were calculated and reported as error estimates.

IsdB YFYF titration of metHb
Spectra (250 -750 nm) of metHb, IsdB constructs, or mixtures thereof were taken in a conventional Cary50 spectrophotometer with an optical path length of 1 cm in a quartz cell at room temperature (22°C). 5 N metHb was mixed with 10 M of apo-IsdB N1N2 or apo-IsdB YFYF , and spectra were immediately recorded; spectra did not change within 5 min of first recording. Additionally, 4 N metHb in a total volume of 1000 l was titrated with aliquots (3.7 l) of apo-IsdB YFYF in 1 M increments from 1 to 9 M. Spectra (250 -750 nm) were taken in a Cary50 with an optical path length of 1 cm in a quartz cell at 22°C, with three replicates carried out. The initial absorbance for each replicate at 410 nm (i.e. 0 M IsdB YFYF ) was subtracted from each subsequent titration absorbance value to yield the ⌬410 nm, which was plotted against IsdB YFYF concentration; a straight line could be fit through the first five titration points to indicate stoichiometry.
Single colonies of S. aureus str. Newman on tryptic soy agar were inoculated into RPMI medium with 0.5 mM ethylenediamine-N,NЈ-bis(2-hydroxyphenylacetic acid) (EDDHA, LGC Standards GmbH, Teddington, UK). The cultures were incubated at 37°C for 16 -20 h on a shaker set to 200 rpm. Overnight cultures were centrifuged for 2 min at 11,000 ϫ g, and the pellet was washed three times in NRPMI with 0.5 mM EDDHA. The cells were then normalized to OD ϳ3 and subcultured 1:100 into 200 l of NRPMI with 0.5 mM EDDHA. In addition to a no-iron control, cultures were supplemented with 200 nN Hb, 200 nN human Hb, and 4 -400 g/ml Hp, mixed serotype (Athens Research and Technology), or 2 M of heme with and without 400 g/ml haptoglobin. The cultures were grown in triplicate with constant shaking at fast-speed and high-amplitude settings in a Bioscreen C instrument (Growth Curve USA, Piscataway, NJ) at 37°C. Growth experiments were repeated three times, and the average growth with standard deviations are plotted using GraphPad Prism 6 (GraphPad Software, La Jolla, CA).