Kinetics of α-Globin Binding to α-Hemoglobin Stabilizing Protein (AHSP) Indicate Preferential Stabilization of Hemichrome Folding Intermediate*

Background: α-Hemoglobin stabilizing protein (AHSP) facilitates hemoglobin production. Results: AHSP preferentially binds to ferric versus ferrous α subunits and induces reversible structural alterations within seconds of binding. Conclusion: AHSP exerts its effects by stabilizing a ferric α folding intermediate and inhibiting its participation in hemoglobin assembly. Significance: AHSP is a molecular chaperone for ferric α-globin. Human α-hemoglobin stabilizing protein (AHSP) is a conserved mammalian erythroid protein that facilitates the production of Hemoglobin A by stabilizing free α-globin. AHSP rapidly binds to ferrous α with association (k′AHSP) and dissociation (kAHSP) rate constants of ≈10 μm−1 s−1 and 0.2 s−1, respectively, at pH 7.4 at 22 °C. A small slow phase was observed when AHSP binds to excess ferrous αCO. This slow phase appears to be due to cis to trans prolyl isomerization of the Asp29-Pro30 peptide bond in wild-type AHSP because it was absent when αCO was mixed with P30A and P30W AHSP, which are fixed in the trans conformation. This slow phase was also absent when met(Fe3+)-α reacted with wild-type AHSP, suggesting that met-α is capable of rapidly binding to either Pro30 conformer. Both wild-type and Pro30-substituted AHSPs drive the formation of a met-α hemichrome conformation following binding to either met- or oxy(Fe2+)-α. The dissociation rate of the met-α·AHSP complex (kAHSP ≈ 0.002 s−1) is ∼100-fold slower than that for ferrous α·AHSP complexes, resulting in a much higher affinity of AHSP for met-α. Thus, in vivo, AHSP acts as a molecular chaperone by rapidly binding and stabilizing met-α hemichrome folding intermediates. The low rate of met-α dissociation also allows AHSP to have a quality control function by kinetically trapping ferric α and preventing its incorporation into less stable mixed valence Hemoglobin A tetramers. Reduction of AHSP-bound met-α allows more rapid release to β subunits to form stable fully, reduced hemoglobin dimers and tetramers.

importance of the loop separating ␣-helices 1 and 2 of AHSP. Collectively, this work suggests that AHSP stabilizes ␣ in vivo by preferentially binding an oxidized ␣ hemichrome folding intermediate and temporarily impairing ␣ assembly into HbA until reduction to the ferrous state has occurred.

EXPERIMENTAL PROCEDURES
Recombinant Human AHSP Expression and Purification-AHSP protein was obtained from pGEX-2T (GE Healthcare) with the full-length human AHSP gene inserted downstream of the Schistosoma japonicum glutathione S-transferase gene (pGEX-2T-AHSP) (GenBank TM accession number NM_016633.2) (11,12). AHSP mutants were generated using a QuikChange II site-directed mutagenesis kit (Agilent Technologies, Inc., Santa Clara, CA) in accordance with manufacturer's instructions and the following PCR primers: AHSP P30A  AHSP was expressed as a GST fusion protein using Escherichia coli BL21 cells (EMD Biosciences, Inc., San Diego, CA; Novagen brand) using methods developed previously (11,12). The soluble GST-AHSP present in the supernatant was captured using ϳ25 ml of glutathione-Sepharose FF medium and an Ä KTA FPLC system (GE Healthcare). During this process, PBS was used as a binding and wash buffer, and 50 mM Tris-HCl, 10 mM reduced glutathione, pH 8.0 at 25°C was used as an elution buffer. GST was cleaved from AHSP using 500 units of thrombin obtained from GE Healthcare. Reduced glutathione, free GST, thrombin, uncleaved AHSP, and other contaminants were then removed by size exclusion chromatography using a preparative grade Superdex 75 HiLOAD prepacked column (GE Healthcare). Yields of GST-free AHSP were ϳ10 mg/liter of bacterial culture. The AHSP produced using this method contains an extra N-terminal Gly-Ser dipeptide appendage due to the thrombin cleavage site (12).
Recombinant Hb Turriff Production-Hb Turriff was produced using an expression system developed previously (22) and a mutated version of the pHE2 plasmid provided by C. Ho and T-J. Shen (Carnegie Mellon University, Pittsburgh, PA). HbA bearing ␣ K99E was produced by site-directed mutagenesis and the following mutagenic primers: 5Ј-CCG GTT AAC TTC GAA CTG CTG TCT CAC TGC C-3Ј and 5Ј-GGC AGT GAG ACA GCA GTT CGA AGT TAA CCG G-3Ј. Recombinant Hb Turriff was expressed and purified using the methods described by Birukou et al. (23).
Native Human HbA Purification and Chain Isolation-HbA was purified from units of human blood obtained from the Gulf Coast Regional Blood Center (Houston, TX) using established methods (24). Separated ␣ and ␤ chains were isolated using established methods (25,26) that were modified as follows. Incubation of CO-liganded HbA with 4-(hydroxymercuri)benzoic acid was limited to ϳ4 h at 4°C instead of overnight. Following this incubation, samples were rapidly buffer-exchanged into 10 mM Tris-HCl, pH 8.0 at 4°C using a column containing 200 ml of Sephadex G-25 medium (Sigma-Aldrich). Samples were then applied to another column containing 100 ml of diethylaminoethyl cellulose medium that had been equilibrated with the same buffer (DE52 medium, Whatman). This column retains ␤ and tetrameric HbA while allowing ␣ to flow through. HbA was then eluted with 20 mM Tris-HCl, pH 7.4 at 4°C, and ␤ was eluted using 100 mM Tris-HCl, pH 7.0 at 4°C. Rather than regenerating sulfhydryl groups using the methods of Geraci et al. (25), a final concentration of 5 ml/liter ␤-mercaptoethanol was added to each sample on ice after which the samples were immediately exchanged into 10 mM Tris-HCl, pH 8.0 at 4°C using a column containing 200 ml of Sephadex G-25 medium. This entire process was done in less than 8 h, and all work was done in a room maintained at 4°C. Regeneration of sulfhydryl groups was assayed by Boyer titration (27). Chain isolations were done in 4-ml batches of 50 mg/ml HbA.
Protein Identity, Purity, and Stability Verification-Plasmid DNA was isolated from each AHSP expression and sent for sequencing to Lone Star Laboratories, Inc. (Houston, TX) using manufacturer-specified sequencing primers to verify the absence of any unwanted mutations. These primers were 5Ј-GGG CTG GCA AGC CAC GTT TGG TG-3Ј and 5Ј-CCG GGA GCT GCA TGT GTC AGA GG-3Ј. SDS-PAGE gels stained with Coomassie Blue were used to confirm protein expression; GST cleavage; and GST, thrombin, and contaminant removal. MALDI-TOF spectrometry performed at Rice University (Houston, TX) confirmed the identity and purity of AHSP. HbA, ␣, and ␤ purities and reassembly efficiency were verified by cellulose acetate electrophoresis (Helena Laboratories Corp., Beaumont, TX). Absorbance spectra and ligand binding kinetics were also verified to ensure that the samples retained normal function following purification and isolation (28). Heme protein concentrations were determined using extinction coefficients reported previously (29,30). With the exception of AHSP P30W , AHSP concentrations were determined by optical absorbance at 280 nm using the extinction coefficient 11,460 M Ϫ1 cm Ϫ1 , which was calculated using the ExPASy Proteomics Server ProtParam Tool. The extinction coefficient used for AHSP P30W was 16,960 M Ϫ1 cm Ϫ1 , which was calculated in the same manner.
Instrumentation and Materials-Manual mixing spectrophotometry was done in either a Cary 50Bio (Varian, Inc., Palo Alto, CA) or a UV2401PC spectrophotometer (Shimadzu, Inc., Columbia, MD) using cuvettes purchased from Starna Cells (Atascadero, CA). Stopped-flow spectrophotometry was done using either a modified Durrum Model D-110 (Palo Alto, CA) or an Applied Photophysics PiStar kinetic circular dichroism stopped-flow spectrophotometer (Leatherhead, Surrey, UK). Unless otherwise indicated, all stopped-flow fluorescence experiments were performed using exit and entrance slit widths of 5 nm for the excitation monochromator, an excitation wavelength of 280 nm, and a cutoff filter fitted to the sample housing that allowed the measurement of total sample fluorescence emission above a wavelength of 302 nm. The volume of the cell was 20 l, and the excitation pathlength of the incident light was 10 mm. The photomultiplier unit was positioned at a 90°a ngle from the incident light, and the fluorescence cell width was 1 mm (total cell dimensions were 10 ϫ 2 ϫ 1 mm). Shot volumes were between 100 and 200 l, and mixing was performed using equal volumes of reactant solutions. Unless otherwise noted, all experiments were performed using 50 -150 mM potassium phosphate buffer, pH 7.4 at 22°C, and all concentrations are given as postmixing. Glass syringes were used whenever possible in the stopped-flow experiments to prevent atmospheric gas contamination (Cadence Science, Lincoln, RI). All buffers, salts, and medium components used for these experiments were obtained from either Sigma-Aldrich or Fisher Scientific. CO and O 2 gases were obtained from Matheson Tri-gas, Inc. (Basking Ridge, NJ).
Data Analysis and Figure Production-Microsoft Excel was used for nonlinear least square data fitting (Microsoft Corp., Redmond, WA) (31). Theoretical data sets calculated from the expressions defined in the text were put into an Excel spreadsheet along with experimentally obtained data, and the sum of the squared residuals were minimized using the Excel Solver tool to obtain visual assessments of plots of the observed and theoretical time courses. Fitting routines in Origin were also used to verify the values obtained from Excel (OriginLab Corp., Northampton, MA). Structure images were created using the PyMOL Molecular Graphics System (Schrödinger, LLC, New York, NY).

RESULTS
Rates of AHSP Binding to Native ␣-Recombinant human wild-type AHSP (AHSP WT ) exhibits intrinsic fluorescence from several aromatic side chains (32). Although wild-type human ␣ and ␤ also contain these residues, neither of these subunits exhibit strong intrinsic fluorescence due to highly efficient fluorescence resonance energy transfer to their heme prosthetic groups (33,34). When ␣ binds to AHSP, this energy transfer quenches the intrinsic fluorescence of AHSP. The key residue in this process is thought to be AHSP Trp 44 (32), which is solvent-exposed in unbound AHSP and becomes buried when ␣ is bound (14).
Baudin-Creuza et al. (32) were the first to report that fluorescence quenching occurs upon formation of a complex between ␣ and AHSP. Their data are consistent with a simple one-step bimolecular association reaction (32). REACTION 1 We confirmed and extended their work using a stopped-flow fluorometer to measure the rates of association (kЈ AHSP ) and dissociation (k AHSP ) (19), and our initial kinetic measurements were subsequently confirmed by Brillet et al. (35).
As shown in Fig. 1A, mixing ␣ WT with AHSP WT resulted in rapid fluorescence quenching on time scales of less than 100 s. The observed time courses for wild-type AHSP are biphasic with a fast phase showing a rate that is linearly dependent on [␣ WT ] and a smaller (20 -30%) slow phase showing an observed rate that is concentration-independent and equal to 0.04 Ϯ 0.01 s Ϫ1 in 100 mM phosphate buffer at pH 7.4 at 22°C. This slow phase was not present when the concentration of ␣ WT was substantially less than that of AHSP WT .
The titration in Fig. 1A indicates that the fast phase of the binding reaction corresponds to a bimolecular association event, which is consistent with previous studies of the formation of reduced ␣⅐AHSP complexes (11,12,14,19). The associ-FIGURE 1. Time courses and association rate constants for ␣CO WT binding to AHSP WT . A, fluorescence signal changes after rapidly mixing ␣CO WT with AHSP WT . Numbers in the right margin represent the postmixing nanomolar (nM) concentrations of ␣CO WT . Control reactions designed to account for the potential effects of photobleaching, precipitation, denaturation, and aggregation were performed, and no changes occurred on the time scales shown (data not shown; see Ref. 28). B, determination of bimolecular association rate constant for AHSP WT binding to both reduced and oxidized ␣ WT (kЈ AHSP ) under pseudo-first-order conditions. The lines represent fits to an expression derived from Reaction 1 with the y-intercept fixed to values of k AHSP measured directly by ␤CO displacement (Table 1 and Fig. 5). Buffers were bubbled with the indicated gases prior to use. To generate ferric ␣ WT , a 5-fold molar excess of potassium ferricyanide was added to ␣O 2 WT prior to the experiment, and the samples were stored on ice to inhibit denaturation and precipitation of met-␣. All concentration values are postmixing.
ation rate constant, kЈ AHSP , was obtained by plotting the observed rates (k obs ) for the fast phase against [␣] under pseudo-first-order conditions when ␣ WT was in molar excess (Fig.  1B). In this plot, the slope of the line is equal to kЈ AHSP . Fitting data from three independent experiments gave a kЈ AHSP value of 10 Ϯ 1.9 M Ϫ1 s Ϫ1 for AHSP WT in 50 mM potassium phosphate buffer at pH 7.4 at 22°C. This bimolecular rate is not altered by oxidation of ␣ WT heme iron to the ferric (met-␣) state nor is the rate constant for ␣O 2 binding appreciably different from that of ␣CO binding (see Fig. 1B, Table 1 in this paper, and Fig. 2A in Khandos et al. (36)).
Santiveri et al. (13) reported that AHSP WT exists in solution as a spontaneously interconverting mixture of two structural conformers due to peptidylprolyl cis-trans isomerization at its Asp 29 -Pro 30 peptide bond. We hypothesized that the slow phase shown in Fig. 1A is related to this phenomenon. We mutated Pro 30 to an Ala, which cannot undergo cis-trans isomerization. This P30A mutation was reported previously to eliminate AHSP conformational heterogeneity in favor of a structure that is similar to the trans Asp 29 -Pro 30 conformation of AHSP WT (13,14). We also mutated Pro 30 to Trp, which had the same effect while conferring greater intrinsic fluorescence (not shown) and altered affinity for ␣.
When ␣ WT was rapidly mixed with either AHSP P30A or AHSP P30W , no slow phase was detected, and the entire reaction is a simple bimolecular process. Representative time courses for AHSP P30W and plots from these experiments are shown in Fig.  2, A and B. The amplitudes of the total fluorescence decreases for the P30A mutant are virtually identical to those for the sum of the fast and slow phases for wild-type AHSP (not shown; see Ref. 29). In contrast, the changes for AHSP P30W are roughly twice those for AHSP WT in Fig. 1A because of the extra Trp side chain. The observed values of kЈ AHSP for the AHSP Pro 30 mutants are similar to those for the fast phase in ␣ binding to AHSP WT (Table 1).
As depicted in Fig. 3, A and B, structural alignments from two previously published NMR studies indicate that AHSP P30A is more similar to the trans Asp 29 -Pro 30 peptide bond conformer  (28). B, determination of the bimolecular association rate constant for ␣CO WT binding to AHSP P30A and AHSP P30W under pseudo-first-order conditions. Lines represent fits to Reaction 1 with the y-intercept fixed to values of k AHSP measured directly by ␤CO displacement (Table 1 and Fig. 5).

TABLE 1 Rates of ␣ binding to and dissociation from AHSP WT , AHSP P30A , and AHSP P30W
The association (kЈ AHSP ) and dissociation (k AHSP ) rate constants were obtained from kinetic data similar to those in Figs. 1, 2, and 5. The equilibrium dissociation constant (K D(AHSP) ) was calculated from the ratio k AHSP /kЈ AHSP . The K D(AHSP) values reported by Gell et al. (12,16) are given in parentheses and were determined by isothermal titration calorimetry in 20 mM sodium phosphate buffer at pH 7.0 at 20°C using 450 mM sucrose in the case of ferric ␣Hb WT to mitigate precipitation. Gell et al. (16) did not report a K D(AHSP) for ␣CO binding to AHSP P30W ; the value for this reaction in the last row is for ␣O 2 . The association and dissociation rate constants (kЈ AHSP and k AHSP , respectively) reported by Brillet et al. (35) are shown in brackets and were determined in PBS at 37°C.
a A slow second phase (ϳ25% amplitude) is observed with a first-order rate equal to 0.04 s Ϫ1 that is independent of ͓␣͔ (Fig. 1A). b The rate parameters for met-␣ WT binding and release from AHSP P30A and AHSP P30W were determined from a more limited set of data than those for the other reactions.
Binding was estimated from one set of concentrations, and release was measured at one high ͓␤͔ with the value of kЈ ␣␤ fixed to the average value obtained by McGovern et al. (39). Time courses for these reactions are provided in Fig. 2 of Khandros et al. (36). c The time course for met-␣ dissociation from AHSP P30A indicated two phases, and fitting to a two-exponential expression gave a fast phase with an amplitude of ϳ33% and k AHSP of Ϸ0.04 s Ϫ1 and a slow phase with an amplitude of ϳ67% and k AHSP of Ϸ0.002 s Ϫ1 . The larger rate is still significantly slower that the rate of dissociation of reduced ␣ from this mutant. The value in the table was calculated from the half-time of the reaction. Similar analyses of the met-␣ AHSP WT and AHSP P30W dissociation reactions indicate that if any fast phases exist their amplitudes are Յ15% of the total fluorescence changes (see Fig. 2B in Khandros et al. (36)).
of AHSP WT than the cis conformer (13,14). Combined with our kinetic data, these structural data suggest that the fast phase shown in Fig. 1A is due to the association of ferrous ␣ WT with the trans Asp 29 -Pro 30 AHSP WT conformer. These data also suggest that the slow phase represents a rate-limiting cis-totrans isomerization of the Asp 29 -Pro 30 peptide bond followed by rapid binding. The reaction of AHSP WT with met-␣ WT also lacks a slow phase (Fig. 4). The total amplitude changes are similar to those for the AHSP reaction with either ␣O 2 WT and ␣CO WT , suggesting that met-␣ can bind to either the cis or trans AHSP WT conformers during bimolecular association perhaps because of greater conformational flexibility.
To investigate further our interpretation of the slow phase, peptidylprolyl isomerases were added to the solutions of AHSP WT prior to mixing with ␣CO. We tried recombinant cyclophilin-A and FK506-binding protein 4 (Prospec Protein Specialists, East Brunswick, NJ). However, no detectable change in the amplitude or rate of the slow phase was observed after preincubation of AHSP WT with either of these enzymes (data not shown). Either the active sites of these enzymes have specificities that preclude interaction with AHSP WT , or the origin of the slow phase results from some other type of conformational isomerization at the Pro 30 loop.
Rates of ␣ Dissociation from AHSP Complex-Gel filtration chromatography and electrophoretic mobility shift assays have shown that ␤ subunits are capable of competitively displacing ␣ from AHSP (11,12,15). HbA formation occurs during this process because (a) ␣ has a much higher affinity for ␤ than for AHSP (12,37) and (b) ␣ cannot simultaneously bind to AHSP and ␤ because both of these interactions involve the same set of interfacial ␣ helices (14,15). A scheme showing these interactions is given in Fig. 5A.
The only species in Fig. 5A that exhibits strong intrinsic fluorescence in solution is AHSP when it is not in complex with holo-␣ (19, 32). When ␣⅐AHSP complexes are mixed with ␤, displacement of ␣ from AHSP increases fluorescence emission  (13)). Blue, green, and yellow stick structures depict the trans peptide bond conformation of this residue based on other structural data (Protein Data Bank codes 1W0B (13), 1XZY (14), and 1W0A (13), respectively). Residue 30 in structure 1W0B (blue) is an Ala, in 1XZY (green) is a Pro, and in 1W0A (yellow) is a Pro. Corey-Pauling-Koltun coloring is otherwise used throughout. Notably, x-ray crystal structures for both ferric and ferrous ␣⅐AHSP complexes have been reported (14,15). These structures were both obtained using ␣ WT ⅐AHSP P30A complexes, suggesting that AHSP with a trans Asp 29 -Pro 30 peptide bond can accommodate ␣ of either oxidation state. A more recent x-ray structure of a ferric complex bearing AHSP Pro 30 and a cis Asp 29 -Pro 30 peptide bond conformation has also been reported (16). These structural data suggest that ferric ␣ can accommodate either AHSP peptide bond conformation. FIGURE 4. Effects of oxidation on ␣ WT binding to AHSP WT . A, fast phase for the association reaction. B, slow phase for the association reaction. The reactions were initiated by mixing 1 M ␣ WT O 2 or met-␣ WT with 1 M AHSP WT and monitoring changes in total fluorescence emission signal. The designation "met" is used to refer to the ferric iron or hemin oxidation state. The met-␣ sample was generated using the same ␣ WT O 2 solution by adding a 5-fold excess of ferricyanide to the sample a few minutes prior to the experiment. Note that the y axis in B is smaller, and the rate of the slow phase for ␣ WT O 2 to AHSP WT is approximately the same as that for ␣ WT CO binding shown in Fig. 1A.
as free AHSP is generated. The rate of this process allows measurement of k AHSP (19,32,35).
The observed time courses for the reaction of ␤ with ferrous ␣⅐AHSP complexes are monophasic at high concentrations, and the rate of displacement increases with increasing [␤]. Representative time courses for a set of displacement reactions are shown in Fig. 5B, and a more detailed comparison of time courses for ferric and ferrous ␣ dissociation from wild-type and the Pro 30 mutants of AHSP is given in Fig. 2B of Khandros et al. (36). The amplitudes of the fluorescence increases for ␣ displacement from the various AHSP complexes are slightly smaller in magnitude than the decreases observed in the association experiments (28). The smaller fluorescence increases are mostly likely due to the higher level of background absorbance and light scattering that occur with the addition of excess ␤. As expected, displacement of ␣ from ASHP P30W showed roughly twice the increase as from AHSP WT due to the presence of a second Trp.
At high AHSP and ␤ concentrations, the amount of free ␣ during the displacement reaction is small throughout the reaction. Under these conditions, the rate of ␣ displacement from AHSP by ␤ is given by the following equation, assuming a steady-state for free ␣, where k AHSP is the rate constant for dissociation of the ␣⅐AHSP complex, kЈ AHSP is the association rate constant for complex formation, and kЈ ␣␤ is the bimolecular rate constant for the association of free Hb subunits to form an ␣␤ dimer (38). By measuring the observed rate of replacement (r obs ) at increasing [␤], this equation can be used to obtain fitted values for k AHSP and the ratio of kЈ AHSP to kЈ ␣␤ (19). Representative fits to Equation 1 are shown in Fig. 5C.
In these studies, low concentrations of ␣⅐AHSP complexes were mixed with 5-fold or greater concentrations of ␤. These concentrations introduce a departure from pseudo-first-order conditions, making Equation 1 approximate because the free concentration of AHSP is increasing from zero during the reaction. To address this, we fixed the free [AHSP] in Equation 1 to 50% of the postmixing total AHSP concentration (0.125 M). Then r obs values were determined by either fitting the observed time courses to a single exponential expression or computed from the measured half-time (r obs ϭ ln2/t1 ⁄ 2 ). This simplified analysis provides fitted values for k AHSP that are identical to those obtained by numerical integration of the rate equations that allow the concentration of AHSP to increase with time (35), and good fits to plots of r obs versus [␤] were obtained (Fig. 5C).
For the fits to Equation 1 shown in Fig. 5C, the value of kЈ AHSP was fixed to the value determined from the bimolecular association reactions shown in Figs. 1B and 2B, and k AHSP and kЈ ␣␤ were allowed to vary. Table 1   stants for the corresponding ␣⅐AHSP complexes were calculated from the ratio k AHSP /kЈ AHSP .
Although kЈ AHSP appears to be invariant, the rate of ␣ dissociation from AHSP is strongly dependent on both the size of the amino acid at position 30 of AHSP and the oxidation state of ␣. For ␣CO WT , the P30A AHSP mutation had little effect on k AHSP , but the P30W mutation caused a marked 25-fold decrease in k AHSP ( Table 1). The value of k AHSP for the reduced forms of ␣ WT is Ϸ0.2 s Ϫ1 , but the value for met-␣ is 100-fold lower and gives rise to a subnanomolar value for K D(AHSP) . The time courses for met-␣ dissociation from AHSP P30A show heterogeneity (see Table 1, footnote c; see Fig. 2B in Ref. 36), suggesting multiple conformations for the mutant AHSP P30A ⅐ met-␣ complex. However, even the rate of the most rapid phase is 5-10-fold smaller than k AHSP for the corresponding reduced ␣ complex.
Effects of Other Mutations in AHSP Pro 30 Loop and ␣ K99E Subunit Variant-Previous studies of AHSP WT have revealed detailed structural information regarding the ␣⅐AHSP binding interface, including the identities of all the ␣ chain amino acids that are thought to interact directly with AHSP (14,15). In a previous study (40), we used this information as a basis for conducting literature-based searches for clinically significant ␣ variants possessing mutations at these positions. It was hypothesized that the phenotypes associated with these mutations might be a result of aberrant ␣ mutant ⅐AHSP WT interactions. Using a series of indirect binding studies, eight ␣ missense mutations were investigated, and one was found to affect the ␣-AHSP interface without detectably perturbing ␣␤ subunit interactions (40). This mutation, called Hb Turriff, replaces ␣ Lys 99 with Glu (40).
To extend this work, we designed a series of AHSP mutants, which were hypothesized to restore binding to ␣ K99E (40). As shown in Fig. 6A, the ␣ WT Lys 99 residue is positioned in close proximity to AHSP WT Pro 30 when the two proteins are bound together. The ␣ K99E mutation is predicted to perturb this region of the ␣-AHSP interface by introducing a negative charge and generating unfavorable electrostatic interactions with neighboring AHSP WT residues. To alter the charge on the complementary surface of AHSP WT and investigate the relative importance of electrostatic interactions in this region, Gln 25 and Asp 29 in AHSP were mutated to Lys and Arg, respectively. Hypothetical structures showing possible side chain rotamers in the mutant AHSP proteins are shown in Fig. 6B.
We expressed and purified these AHSP mutant proteins along with recombinant Hb Turriff. We then isolated the individual subunits of Hb Turriff to obtain ␣ K99E . All proteins were stable and found to behave similarly to their wild-type counterparts throughout the purifications. Although ␣ K99E exhibited altered mobility during diethylaminoethyl cellulose chromatography, this protein was still resolvable from both tetrameric Hb Turriff and ␤. The finding that Hb Turriff was expressed at the usual levels confirms our previous work, which suggested that the K99E mutation does not significantly perturb ␣␤ subunit interactions.
The following sets of binding and dissociation reactions were investigated using reduced ␣CO variants: 1) ␣ K99E binding to AHSP WT , 2) ␣ K99E binding to AHSP Q25K , 3) ␣ K99E binding to AHSP D29R , 4) ␣ WT binding to AHSP Q25K , and 5) ␣ WT binding to AHSP D29R . Representative time courses as well as fits to the kinetic expressions for association (k obs ϭ kЈ AHSP ϩ k AHSP ) and dissociation (Equation 1) for these experiments are shown in Fig. 7, A-D. Longer time scales were investigated, and small slow phases with highly variable amplitudes and rates were observed for all these AHSP variants, which still have Pro at position 30. The rate and equilibrium constants for the major fast phases of these reactions are given in Table 2.
These results agree with our previous work (40) and with the data of Feng et al. (14), which showed that the ␣ K99A mutation disrupts binding to AHSP WT . In the case of the ␣ K99E mutation, no binding to AHSP WT was detected at all even at high micromolar concentrations of both proteins. Weak binding could be restored with the Q25K AHSP mutation, and moderate binding occurred with the D29R AHSP variant. Thus, it seems likely that the ⑀-amino group of ␣ WT Lys 99 participates in favorable electrostatic interactions with one or more polar side  (15)). A, possible interactions at the AHSP-␣ interface. The parenthetical distance was determined using structural data obtained from Protein Data Bank structure code 1Y01 (14). B, hypothetical structures for AHSP revertant mutants (Q25K and D29R) that allow binding of ␣ K99E , Hb Turriff. In both panels, the color scheme is the same as in Fig. 5. Distances were measured and the theoretical model was generated using the PyMOL Molecular Graphics System distance measuring utility and site-directed mutagenesis wizard, respectively. chains on AHSP WT . However, because ␣ WT could still bind fairly strongly to AHSP Q25K and AHSP D29R , it also seems likely that the K99E ␣ mutation causes other conformational alterations in the globin structure.
Rapid and Reversible AHSP-induced ␣ Hemichrome Formation-HbA autooxidation is a spontaneous process in which either O 2 bound to a ferrous heme group spontaneously dissociates as a superoxide radical, or O 2 free in solution reacts with a transient aquo-deoxyheme, leaving the resulting iron in the ferric (Fe 3ϩ ) state (41)(42)(43). Following this reaction, the hemin iron is axially coordinated on one side by a histidine (the proximal or F8 histidine) and on the other by an H 2 O or OH Ϫ depending on pH (44,45). If the hemin pocket unfolds, the distal histidine, His 58 (E7) in ␣ (or other nearby basic amino acids on the distal side of the heme ring) can coordinate to the ferric iron atom (44 -48) The resulting bishistidyl adduct is called a hemichrome (44,45). Bishistidyl adducts can also form with ferrous heme groups, and the resulting products are termed hemochromes (44,45). These species can be identified by characteristic peaks in the visible spectra of the corresponding metHb and deoxy-Hb derivatives (44,45).
Previous work has shown that binding to AHSP accelerates the autooxidation of ␣O 2 to form a hemichrome with no readily identifiable intermediate aquo-met-␣ state (14,18). At neutral

Rates of ␣ WT and ␣ K99E binding to and dissociation from AHSP Q25K and AHSP D29R
The association (kЈ AHSP ) and dissociation (k AHSP ) rate constants were obtained from the kinetic data in Fig. 7, and the equilibrium dissociation constants (K D(AHSP) ) were calculated from the ratio k AHSP /kЈ AHSP . The abbreviation "ND" signifies that a given value could not be determined. No binding of ␣ K99E CO to AHSP WT could be detected presumably because of very large k AHSP and K D(AHSP) values. In the case of ␣ K99E binding to AHSP Q25K , we were unable to measure displacement reactions, and k AHSP was calculated based on the y-intercept of the association reaction measured under pseudo-first-order conditions (plot in Fig. 7B). pH, rapid oxidation of free ␣O 2 by ferricyanide initially produces met-␣ with water coordinated to the sixth position of the iron atom. The spectral differences between ␣O 2 and met-␣ are shown in Fig. 8A with the ferrous oxygenated form exhibiting strong absorption peaks at 541 and 576 nm and aquo-met-␣ exhibiting large charge transfer bands peaking at ϳ505 and 635 nm (29,30). Adding equimolar AHSP to met-␣ resulted in the immediate (Յ10 s) disappearance of the 505 and 635 nm bands and appearance of 535 and 565 nm bands, which are characteristic of hemichrome formation (Fig. 8A) (11,44). The kinetics of this spectral transition are shown in Fig. 8B. When 20 M met-␣ was mixed with 20 M AHSP, a large increase in absorbance was observed at 535 nm. The apparent bimolecular rate is 5-10 M Ϫ1 s Ϫ1 , which is identical to that observed when measuring binding by quenching of intrinsic AHSP fluorescence (Fig. 8B, gray time course). These results show that met-␣ hemichrome formation occurs simultaneously with binding to AHSP and that the conformational transition from aquo-met to the hemichrome form of ␣ occurs at rates Ͼ100 s Ϫ1 when bound to AHSP. If the hemichrome conformational change were slower, it would have been seen as a much slower 535-nm absorbance increase in Fig. 8B compared with the AHSP fluorescence decrease.
To investigate the reversibility of AHSP-induced hemichrome formation with respect to reduction, a solution of met-␣⅐AHSP complex was rapidly mixed with excess sodium dithionite. This reagent is a reducing agent, which both consumes O 2 in aqueous solutions and rapidly converts metHbA to ferrous deoxy-HbA (1,30). As is shown in Fig. 9, rereduction of met-␣⅐AHSP did not produce a spectrum indicative of a hemochrome, which would have had a moderate peak at ϳ530 nm and a very intense band at 558 nm (44,49). Instead, reduction with dithionite produced a deoxy-␣ spectrum with a single peak at ϳ560 nm, which is characteristic of pentacoordinate deoxy-␣ in its native and fully folded form even though the subunit is still bound to AHSP (29). The time course of the reduction reaction was measured in a rapid scanning, stopped-flow spectrophotometer. The resulting time courses indicate that rereduction occurred rapidly (Fig. 9, inset) with an observed rate that is on the same order as the rate of reduction of free met-␣ (data not shown), and no hemochrome intermediates were observed. Thus, the ␣ conformational changes after reduction are also very rapid even when the globin is bound to AHSP.

DISCUSSION
The two phases observed for the reaction of ␣ WT with AHSP WT (Figs. 1 and 4) seem to verify the structural heterogeneity first characterized by Santiveri et al. (13). The loss of the second, slow phase for the P30A and P30W AHSP mutations (Fig. 2) suggests strongly that the kinetic heterogeneity is due to peptidylprolyl cis-trans isomerization at the Asp 29 -Pro 30 pep-  . Chemical reduction of AHSP⅐met-␣ WT complexes. Spectra were recorded before and after the addition of excess sodium dithionite (DT) (ϳ15 mM) to 40 M ␣ WT ⅐AHSP WT complexes. Inset, emergence of the deoxy-␣ WT absorption peak at 558 nm after mixing AHSP⅐met-␣ WT with the same amount of reductant in a stopped-flow rapid mixing apparatus. Spectra were also recorded every 1 ms and analyzed (data not shown). There was no evidence for transient formation of a hemochrome intermediate. tide bond in AHSP. Based on the relative amplitude of the slow association phase, it appears that 20 -30% of free AHSP WT occupies a cis peptidylprolyl conformation that is relatively unreactive toward reduced ␣. Although this estimate differs from the 50% mixture reported by Santiveri et al. (13), it is consistent with previous reports that at equilibrium 60 -90% of model oligopeptides occupy the trans peptidylprolyl conformation and 10 -40% occupy the cis conformation (for a review, see Ref. 50). Although more recent work has revealed that the exact ratios are dependent on which amino acid precedes Pro, in cases of Asp-Pro, it has been shown that 11-19% of model oligopeptides occupy the cis peptidyl-prolyl conformation (51). Additionally, the observed rate of the slow phase, 0.04 Ϯ 0.01 s Ϫ1 in 100 mM potassium phosphate buffer at pH 7.4 at 22°C, is consistent both with the rate assignment of AHSP conformational interconversion by NMR (13) and with other studies that show that peptidylprolyl cis-trans interconversions occur spontaneously with half-times of 10 -100 s at 25°C (50). The observations that the rate of the slow phase was independent of ␣ concentration and that the slow phase was not observed when ␣ was limiting are also supportive of this model. This mechanism implies that reduced ␣ reacts rapidly with the trans Pro 30 conformer of AHSP in a simple bimolecular process (Reaction 1), whereas little or no binding occurs to the small remaining fraction of cis AHSP until it isomerizes to the trans conformation.
This interpretation is supported by the effects of the P30A and P30W AHSP mutations, which caused the formation of a single trans conformation at the 29-30 peptide bond (Fig. 3) (13). The time courses for ␣ binding to AHSP P30A and AHSP P30W are monophasic and bimolecular. The fitted values of kЈ AHSP for these mutants are almost identical to that obtained from the fast phase for ␣ binding to AHSP WT . This finding also suggests strongly that reduced ␣ binds rapidly to the trans peptidylprolyl Pro 30 conformer of AHSP WT . This idea is supported by the observation that the wild-type ␣⅐AHSP complex was very difficult to crystallize, and formation of crystals was greatly facilitated by the P30A mutation (14 -16).
However, none of these arguments for cis to trans peptidylprolyl isomerization are direct, and we did not observe an effect on the amplitude or rate of the slow phase when peptidylprolyl isomerases were added to the AHSP solution prior to mixing with ␣. Thus, it is possible that the slow phases observed for AHSP WT reflect some other type of conformational isomerization that causes more quenching of intrinsic AHSP fluorescence after reduced ␣ is bound. Some autooxidation is occurring when ␣O 2 is bound, but the AHSP⅐␣CO complex is stable with little or no UV-visible absorbance changes occurring on time scales of minutes to hours. The lack of a slow phase for met-␣ binding to AHSP WT suggests that the rapidly formed ␣ hemichrome binds to either prolyl conformer.
The importance of the loop separating helices 1 and 2 of AHSP WT is underscored by our studies of ␣ K99E (Hb Turriff). Our preliminary studies indicated that this variant is unstable in vivo due to a loss of binding to AHSP WT (40). Our new studies confirm that this mutant ␣ cannot bind AHSP even at micromolar concentrations and that the mild phenotype associated with Hb Turriff is caused by a loss of this interaction and not by disrupted assembly into ␣␤ dimers and tetramers, which appears normal.
The results in Table 2 show that introduction of positively charge amino acids in the Pro 30 loop of AHSP revertants can partially reverse the effects of the ␣ K99E mutation. However, the results are complex and indicate that, although favorable electrostatic interactions are important near the Pro 30 loop, other conformational factors also play an important role.
The association and dissociation rate constants reported for ␣CO and ␣O 2 binding to AHSP in Table 1 agree well with those in our initial report (19) and with the parameters reported subsequently by Brillet et al. (35). However, Brillet et al. (35) reported that the "… dissociation rates were not found to depend strongly on the Hb oxidation state (oxy versus met versus deoxy)." Although our data also show that kЈ AHSP is nearly the same for ferric and ferrous ␣ (Table 1), the rate of dissociation (k AHSP ) of the met-␣⅐AHSP WT complex is ϳ100-fold smaller than that for the reduced ␣⅐AHSP complexes. The net result is a 100-fold higher affinity and a very small subnanomolar K D for met-␣ binding to AHSP.
Gell et al. (16) observed a similar large relative decrease in K D for met-␣ versus ␣CO binding to AHSP using isothermal titration calorimetry (Table 1). However, the absolute K D values obtained by calorimetry are consistently higher than either we or Brillet et al. (35) determined from the ratio of the kinetic parameters. The reasons for this discrepancy are not obvious. The cause may be related to systematic errors during curve fitting that arise because of the high concentrations of protein (1-5 M) required by the isothermal titration calorimetry experiments and the low K D values (Յ0.1 M; Table 1).
Because AHSP binds to ␣ more rapidly than ␤, it is likely that AHSP out-competes ␤ kinetically for nascent free ␣ in vivo. If the ␣ heme groups are in the ferrous form, they can be more readily displaced by ␤ due to their dramatically higher affinity for ␤ chains (K D Ϸ 0.001 nM (37,(52)(53)(54)) than for AHSP (K D Ϸ 17 nM (Table 1)) and the relatively high rate of ferrous ␣⅐AHSP dissociation (k AHSP Ϸ 0.2 s Ϫ1 ). However, if bound ␣ is in the oxidized hemichrome form, displacement by ␤ chains is 100fold slower. Thus, participation of ferric ␣ WT in HbA assembly is inhibited kinetically by AHSP, but disassembly of the resultant met-␣⅐AHSP complex is facilitated by hemin reduction.
Hemoglobin assembly requires apoglobin synthesis, partial folding, heme uptake, reduction, and ␣␤ dimer formation. Although folding and assembly can occur without AHSP binding, we suggest that the highlighted pathway in Fig. 10 is favored in the presence of AHSP. AHSP first assists the folding of apo-␣ (17) and then uptake of hemin to generate the met-␣⅐AHSP hemichrome complex. It has been shown that hemichrome forms are highly populated intermediates during the folding of both holomyoglobin and holo-Hb (46 -48). The ␣ hemichrome folding intermediate is unstable when free in solution, loses hemin, and unfolds rapidly in the apoform, leading to precipitation (44 -48). However, this met-␣ folding intermediate is stabilized by AHSP. When the met-␣⅐AHSP complex is reduced, fully folded reduced deoxy-or ␣O 2 is formed, has a lower affinity for AHSP, and is rapidly displaced by ␤.
Free ␣O 2 monomers are reportedly more prone to autooxidation than ␤O 2 (56) and are produced in a slight excess during normal HbA production in vivo probably to compensate for this instability (for a review, see Ref. 19). In general, ferric subunits and metHbA are much less stable than their ferrous forms due to enhanced rates of hemin loss and subsequent rapid apoprotein denaturation (57)(58)(59)(60)(61)(62)(63)(64)(65). Our data and the scheme in Fig.  10 suggest that AHSP WT promotes the formation of more stable HbA end products by acting as a quality control checkpoint during the assembly process. The relative importance of this function is likely amplified during periods of oxidative stress or during the spontaneous accumulation of ferric ␣ WT . Instead of these subunits reacting directly with ␤ to form relatively unstable, partially oxidized HbA tetramers, ferric ␣ chains are sequestered kinetically in AHSP complexes until the iron can be reduced.
This redox proofreading function and stabilization of the hemichrome folding intermediate helps explain the findings of Khandros et al. (36) that demonstrate that Pro 30 substitutions (P30W and P30A) do not appear to significantly affect HbA production or AHSP function in vivo. The P30W mutation enhances the affinity of AHSP for ␣O 2 and slows its autooxidation rate, but these properties appear to be less important than high affinity binding to met-␣ chains. These in vivo data also suggest that cis-trans peptidylprolyl isomerization in AHSP WT is not critical physiologically because only the trans conformation occurs in the Pro 30 mutants, which have no obvious phe-notype. The key biochemical property of AHSP appears to be its high affinity for oxidized ␣.
The mechanism in Fig. 10 shows why the lack of K99E ␣ binding to AHSP indirectly destabilizes HbA. When this variant is oxidized, it cannot be sequestered by AHSP, which allows ferric ␣ K99E to either denature or react with ␤ and be incorporated into relatively unstable mixed valency tetramers. Both the stabilization and quality control functions help explain the observation that disruption of AHSP WT in mice worsens the phenotypes of both ␣and ␤-thalassemia syndromes (66,67).
Although HbA assembly almost certainly occurs through several parallel pathways, AHSP binding appears to act as a shunt for ferric ␣ until reduction can occur. Thus, even when ␤ subunits are in excess (␣-thalassemia), there will be more incorporation of unstable oxidized ␣ into HbA when AHSP is absent, leading to more globin precipitates. In the case of excess ␣ (␤-thalassemia), the situation is even worse in the absence of AHSP when no chaperone is available to stabilize the excess met-␣.

CONCLUSIONS
Cis-trans isomerization and electrostatic interactions in the Pro 30 loop between helices 1 and 2 of AHSP have marked effects on the rate and equilibrium constants for ␣ binding and can help explain the phenotypes of ␣ hemoglobinopathies that involve disruption of ␣ binding to AHSP but not to ␤. Compared with ferrous ␣, met-␣ binds 100-fold more tightly to AHSP, resulting in a very low rate of dissociation of the ferric ␣⅐AHSP complex. Consequently, AHSP can trap met-␣ kinetically and inhibit its incorporation into unstable mixed valence HbA tetramers. This kinetic property coupled with AHSP stabilization of the hemichrome folding intermediate supports a dual role for AHSP. It acts both as a molecular chaperone to facilitate ␣ folding and as a quality control protein to prevent assembly of unstable, partially oxidized Hb tetramers.