Probing the α1β2 Interface of Human Hemoglobin by Mutagenesis ROLE OF THE FG-C CONTACT REGIONS

The allosteric transition of hemoglobin involves an extensive reorganization of the α1β2 interface, in which two contact regions have been identified. This paper concerns the effect of two mutations located in the “switch” (αC3 Thr → Trp) and the “flexible joint” (βC3 Trp → Thr). We have expressed and characterized one double and two single mutants: Hb αT38W/βW37T, Hb βW37T, and Hb αT38W, whose structure has been determined by crystallography. We present data on: (i) the interface structure in the two contact regions, (ii) oxygen and CO binding kinetics and cooperativity, (iii) dissociation rates of deoxy tetramers and association rates of deoxy dimers, and (iv) the effect of NaI on deoxy tetramer dissociation rate constant. All the mutants are tetrameric and T-state in the deoxygenated derivative. Reassociation of deoxygenated dimers is not modified by interface mutations. DeoxyHb αT38W dimerizes 30% slower than HbA; Hb βW37T and Hb αT38W/βW37T dissociate much faster. We propose a binding site for I− at the switch region. The single mutants bind O2 cooperatively; the double one is almost non-cooperative, a feature confirmed by CO binding. The functional data, analyzed with the two-state model, indicate that these mutations reduce the value of the allosteric constant L.

The three-dimensional structure of liganded and deoxyhemoglobin (Baldwin and Chothia, 1979;Shaanan, 1983;Perutz et al., 1987) indicates that the allosteric transition can be described topographically as a change in the relative orientation of the two dimers identified as ␣ 1 ␤ 1 and ␣ 2 ␤ 2 , which rotate with respect to each other and slide along the ␣ 1 ␤ 2 and ␣ 2 ␤ 1 interfaces. Therefore, the amino acid residues that contribute to the ␣ 1 ␤ 2 (and ␣ 2 ␤ 1 ) interface (Fig. 1) play a major role in controlling the relative stability of the allosteric states and, as a consequence, cooperativity. This interface contains residues of helix C and the FG corner of both chains. In total 17 residues establish interactions across the ␣FG-␤C and the ␣C-␤FG contacts with a pseudo-symmetric arrangement (Fig. 1). Other interactions of minor significance are present across the ␣FG-␤FG and ␣C-␤C contacts (Baldwin and Chothia, 1979).
The ␣FG-␤C contact is extensive and forms a network of weak bonds, which is largely maintained in the allosteric transition, in spite of some changes in the orientation of specific amino acid side chains. ␤37 (C3) Trp in particular makes contacts with ␣92(FG4) Arg, ␣94(G1) Asp, and ␣95(G2) Pro in both oxy-and deoxyhemoglobin; nevertheless, it is a good probe of the allosteric transition because its optical spectrum is perturbed upon ligand binding (Briehl and Hobbs, 1970;Perutz et al., 1974). The amino acid residues at the ␣C-␤FG contact region of the ␣ 1 ␤ 2 interface undergo an important reorganization in the course of the T-R transition. In deoxyhemoglobin the side chain of ␤97(FG4) His lies between ␣44(CD2) Pro and ␣41(C6) Thr, while in oxyhemoglobin this same His settles between ␣41(C6) Thr and ␣38(C3) Thr. Thus, this region of the ␣ 1 ␤ 2 interface, which seems to be compatible with only two states, plays a key role in determining the allosteric transition (see Fermi and Perutz (1981)). Baldwin and Chothia (1979) summarize the function of the ␣ 1 ␤ 2 interface with the definition of the ␣FG-␤C contact region as a "flexible joint" and the ␣C-␤FG as a "switch." Hence the amino acid residue at position C3 has a different role in the ␣ and ␤ chains; ␤37(C3) Trp is involved in a series of contacts, which stabilize the tetramer in both the oxy and deoxy derivatives, while ␣38(C3) Thr participates in the reorganization of the ␣ 1 ␤ 2 interface associated with the allosteric transition.
Human Hb has been expressed in Escherichia coli and yeast, and several interesting mutants have been prepared to test specific hypotheses (Nagai et al., 1985;Martin de Llano et al., 1993;Komiyama et al., 1995). In view of the considerations reported above, we have investigated the role of the residue at C3 in both chains of human hemoglobin by synthesizing the site-directed mutants bearing in the ␤ chains the residue found in the ␣ chains and vice versa. The type of substitution to be inserted at C3 was chosen to probe the effect of changes in the number of atomic contacts (Schaad et al., 1993), avoiding the introduction of charged residues. Three hemoglobins were therefore expressed in E. coli: the single mutants ␣38 Thr 3 Trp (␣T38W) 1 and ␤37 Trp 3 Thr (␤W37T) and the double mutant ␣38 Thr 3 Trp/␤37 Trp 3 Thr (␣T38W/␤W37T, called herein the double mutant). It may be recalled that there are no known natural mutants of ␣38 Thr, while the two natural mutants of ␤37 Trp, Hb Hirose and Hb Rothschild, have Ser and Arg, respectively (Yamaoka, 1971;Gacon et al., 1977;Sasaki et al., 1978) (see also Huisman (1992)).
Our three mutants were compared to wild type HbA with respect to the reaction with oxygen and carbon monoxide, the stability of liganded and unliganded tetramers, the kinetics of association of deoxy dimers and the dissociation of deoxy tetramers, the effect of sodium iodide, and the changes in the molecular contacts at the interface by x-ray crystallography of deoxy-Hb ␣38 Thr 3 Trp or by modeling.
Structural analysis leads to simple predictions on the role of the ␣ 1 ␤ 2 interface, suggesting that (a) semi-conservative mutations at topological position C3 affect tetramer stability and cooperativity in a simple manner, without loss of allosteric behavior; (b) substitutions at equivalent positions of the ␣ and ␤ chains should yield different functional effects, in spite of the pseudo-symmetry of that interface. By-and-large these predictions are fulfilled by our results.

EXPERIMENTAL PROCEDURES
Materials-The site-directed mutant hemoglobins were produced in E. coli and purified as described previously (Nagai et al., 1985;Hoffman et al., 1990;Vallone et al., 1993). Reagents were of analytical grade.
Functional Properties-The oxygen binding isotherms were determined using the tonometric method (Rossi Fanelli and Antonini, 1958). The time course of CO binding to deoxyhemoglobin was followed using either the Applied Photophysics stopped flow apparatus (Leatherhead, United Kingdom) or the flash photolysis apparatus described by Brunori and Giacometti (1981).
The time course of dissociation of deoxyhemoglobin into dimers was monitored by recording the optical spectrum (in the Soret region) after mixing deoxyhemoglobin with a stoichiometric amount of haptoglobin (isoform 1.1 purchased from Sigma), as described by Ip et al. (1976). Since haptoglobin binds rapidly and tightly to the free dimers, the equilibrium is driven toward the dissociated state (Nagel and Gibson, 1972).
The time course of association of deoxyhemoglobin dimers was followed by mixing oxyhemoglobin (which at micromolar concentration contains a substantial amount of oxygenated dimers) with 50 mM sodium dithionite in the rapid scanning stopped flow spectrometer described by Bellelli et al. (1990). After rapid dissociation of oxygen, the dimers recombine in a slow, concentration-dependent process (Kellett and Gutfreund, 1970).
Transient spectra were collected into a matrix (A), each column being a difference spectrum and each row a time course, and analyzed with the singular value deconvolution algorithm (SVD; Golub and Reinsch, 1980;Henry and Hofrichter, 1992). This deconvolution yields the three matrices U, S, and V, whose product U ϫ S ϫ V T approximates A. The product U ϫ S may be envisaged as a matrix of extinction coefficients of the spectroscopic components detected by the algorithm, while V columns represent their time evolution. The SVD analysis allowed to magnify the small absorption changes coupled to the allosteric transition and to increase the signal to noise ratio.
Structural Properties-The deoxygenated derivative of the mutant ␣T38W was crystallized following Perutz (1968). X-ray data from a single crystal were collected and processed as described by Perutz et al. (1993). A total of 25,572 reflections were collected in the shell between 22.0 and 2.48 Å; these reduced with a merging R factor of 9.6% to 15,560 uniquely indexed reflections, or 76% of the unique reflections in the shell. Cell dimensions were the same as those of native deoxy-HbA, within experimental error. Mutant structure amplitudes were scaled to native with a scale factor and a temperature factor; the R factor between mutant and native was 13.5% for 15,442 reflections included in the map. A difference map was calculated from the differences of the scaled amplitudes with phases of the native model (Brookhaven Protein Data Bank code 2HHB; Fermi et al., 1984). The map was symmetryaveraged about the molecular dyad to improve accuracy.
Molecular modeling was carried out on a Silicon Graphics workstation (Silicon Graphics Inc., Mountain View, CA) by use of the Discover/ Insight package. Two structures were from the Brookhaven Protein Data Bank: deoxygenated recombinant hemoglobin (Kavanaugh et al., 1992a) and wild type oxyhemoglobin (Shaanan, 1983). For mutant ␣T38W we used the coordinates of the deoxy derivative. For the other mutants (i.e. the two proteins containing the substitution ␤W37T and the oxy form of Hb ␣W38T) simulation of the mutation in the corresponding wild type structure was performed by substituting the side chain in the starting position leading to no unfavorable contacts with neighboring side chains, followed by energy minimization allowing only amino acids in a sphere of 6 Å from the mutated amino acid to move.

RESULTS
Oxygen Equilibrium Experiments-The oxygen binding properties of the three site-directed mutants and HbA, depicted in Fig. 2, show that both single mutants are cooperative and display Hill coefficients larger than 2, while the double mutant is almost non-cooperative (n ϭ 1.3).
In spite of its limitations (see, for example, Ackers et al. (1992)), the analysis of the oxygen binding isotherms was carried out according to the two-state allosteric model (Monod et al., 1965). Fit of data is often beset by uncertainties in K R and L 0 . Nonetheless, since kinetic data on O 2 and CO have provided independent support that K R is identical or very close to that of the isolated ␣ and ␤ chains and of the ␣␤ dimer (Edelstein and Gibson, 1987;Szabo and Karplus, 1972), we fitted the data in Fig. 2 assuming K R to be the same for HbA and all the mutants, and thus similar to the oxygen affinity of the isolated chains. This assumption proved compatible with a good fit, with the parameters presented in Table I. By contrast, very poor fits were obtained if K T of the mutants was imposed to be the same as that of HbA, which is not surprising and agrees with the theory of Szabo and Karplus (1972). Within the limitations of the two-state model, the numerical values of the allosteric parameters indicate that mutations at C3 (i) affect the equilibrium constant between the two quaternary states, reducing the difference in stability between T 0 and R 0 in unliganded Hb (see values of L 0 ), and (ii) reduce to some extent the constraints imposed on the subunits by the quaternary assembly as deduced from the values of K T .
Kinetics of Combination with CO-As an additional test of the conclusions drawn from O 2 binding data, we have determined the CO combination rate constants by stopped-flow and compared them to the values obtained by partial flash photolysis. The time course of CO binding to HbA and Hb ␣T38W by stopped flow is autocatalytic (Fig. 3), which confirms cooperative ligand binding; for Hb ␤W37T and Hb ␣T38W/␤W37T, the same reaction appears non-autocatalytic, although the apparent rate constant is still much lower than that of the R-state.
The data reported in Fig. 3 were fitted to four consecutive and irreversible pseudo-first order processes, as described by Hopfield et al. (1971). The values of L 0 and c (K R /K T ) used to calculate the relative amount of T-and R-state Hb at each ligation state were taken from Table I; the values of R kЈ were obtained by partial flash photolysis, which is known to populate partially liganded species that recombine as fast as the isolated chains (Sawicki and Gibson, 1976;Vandegriff et al., 1991;Jones et al., 1992). The CO combination time courses in Fig. 3 are satisfactorily fitted with this model, regardless of the presence of evident kinetic cooperativity, and the rate constants are given in Table II. Regarding Hb ␤W37T, an autocatalytic time course may have been expected. However, it is known that speeding up of the CO combination is very sensitive to the value of c, which may not be identical for O 2 and CO, even though this difference is sometimes very difficult to demonstrate. Under these conditions the estimate of T kЈ is subject to some uncertainties, and the overall time course is essentially exponential.
Even complete photolysis of dilute solutions of HbCO usually follows a biphasic rebinding time course (Antonini and Brunori, 1971). This has been attributed to the presence of rapidly reacting dimers in equilibrium with the slowly reacting tetramers (Gibson and Antonini, 1967;Edelstein et al., 1970). Thus flash photolysis has also been used to probe the extent of dissociation of HbCO into dimers. Employing this technique, Vallone et al. (1993) have shown that Hb ␣T38W CO is a more stable tetramer than HbA CO, by approximately 0.6 kcal/mol. Similar experiments carried out as a function of Hb concentration (data not shown) have demonstrated that the CO derivative of the two ␤W37T mutants is almost completely dissociated into dimers at [Hb] ϭ 15 M, and thus a lower limit of 40 M can be set to the value of the tetramer-dimer dissociation constant of these mutants.
Since O 2 and CO exhibit nearly parallel equilibrium curves, the data in Tables I and II may be considered together even though most of the cooperativity is expressed in the dissociation rate constants in the case of O 2 and in the association rate constants in the case of CO (Antonini and Brunori, 1971;Szabo, 1978).
A significant conclusion is that all our mutants display a ligand affinity higher than HbA in the T-state. However, this increase is smaller for the two single mutants, and substantial for the double mutant, in keeping with the reduced number of contacts at the ␣ 1 ␤ 2 interface (see below). More interestingly, the value of L 0 decreases significantly in the mutants, being, in the double mutant, ϳ15-fold smaller than in HbA.

Recombination of Dimers following Oxygen Dissociation-
The time course of the recombination of deoxygenated ␣␤ dimers from HbA and the mutant Hbs was studied by recording the slow optical transitions that follow the rapid deoxygenation by dithionite of a dilute solution of HbO 2 (Antonini et al., 1968;Kellett and Gutfreund, 1970), given that, at micromolar concentration, HbO 2 is an equilibrium mixture of tetramers and dimers. This is possible because (a) the optical spectrum of deoxyhemoglobin in the T-state differs from that in the R-state (the transient R 0 Hb, as described by Sawicki and Gibson (1976); the deoxy dimers and the deoxy chains, as reviewed by Bellelli and Brunori (1994)), and (b) the T 0 -R 0 difference spectrum is, within errors, the same for HbA and our three mutants.

TABLE II CO combination rate constants
The value of T kЈ, determined by stopped flow, corresponds to the reaction of the unliganded T-state; R kЈ, determined by partial photolysis, is the highest estimate for the rapid phase and corresponds to the rate constant for the partially liganded R state (as well as dimers).
1.2 ϫ 10 5 6.2 ϫ 10 6 ␤W37T 4.2 ϫ 10 5 4.0 ϫ 10 6 ␣T38W/␤W37T 5.5 ϫ 10 5 6.2 ϫ 10 6 The experiment is described in the following scheme (Scheme 1). 2 An example of this type of experiment for HbA and the three mutants is reported in Fig. 4. The plot shows the second column of V (from the SVD analysis), which contains most of the slow optical transition. The faster phase seen in this figure is the time course of O 2 dissociation (upon mixing with dithionite), and is described by steps 1 and 3 of Scheme 1. The slower phase (step 2 in the same scheme) relates to the time course of association of deoxy dimers. The relative amplitude of this phase is a measure of the fraction of oxy dimers in equilibrium with oxy tetramers. Since addition of 0.5 M NaI promotes complete dissociation of the liganded tetramers but does not prevent the reassociation of unliganded dimers (Kellett and Gutfreund, 1970), experiments carried out under these experimental conditions yield the full amplitude of the slow phase. Estimates for the rate constant of deoxy dimers association for HbA and the three mutants (in the absence and in the presence of 0.5 M NaI) are reported in Table III.
Rate of Dissociation of Unliganded Tetramers into Dimers-The dissociation of deoxyhemoglobin was followed by the absorbance changes at 430 nm after reaction with haptoglobin, recorded either statically or kinetically (Ip et al., 1976). This method relies on the essentially irreversible binding of Hp to free ␣␤ dimers, which forces the tetramer-dimer equilibrium toward dissociation, the rate-limiting step being dimerization (see also Nagel and Gibson (1972)). In view of the results reported above, the effect of NaI has also been tested on unliganded Hb.
The static difference spectra obtained upon binding to Hp are very similar to those obtained by the kinetics of dimer recombination (data not shown). The total absorbance change for complete dissociation of deoxy tetramers into dimers was found to be 13% of the total absorbance of the sample at 430 nm. This is in good agreement with the value expected for the T 0 -R 0 spectral change (for a review, see Bellelli and Brunori (1994)). Fig. 5 shows the observed time courses of dissociation. This experimental approach suffers from the extremely slow time course, which, at lower temperatures, extends over several days; thus, for some proteins and under some conditions, the optical transition could not be followed to completion. Nonetheless, the time course was fitted in all cases to a single exponential following Ip et al. (1976). In addition, the asymptotic value was independently estimated from the total optical density change (see above). Table IV reports the first order rate constants determined from the data in Fig. 5. HbA and Hb ␣T38W behave similarly, while both Hb ␤W37T and the double mutant dissociate faster than HbA (in spite of being tetrameric in the deoxygenated derivative, as judged by the recovery of the expected optical transition). The effect of NaI on the dissociation of HbA and Hb ␤W37T is the same (see the ratio kЈ/k in Table IV). However, quite unexpectedly, this effect is considerably reduced in the two mutants in which ␣38 (C3) is mutated to Trp; this point will be discussed below, with reference to the structure of the interface.
2 Although Scheme 1 has been widely applied to the description of this reaction (Kellett and Gutfreund, 1970;Wiedermann and Olson, 1975), the data obtained by recording the complete optical transition over the range 380 to 520 nm, and analyzed by SVD (see "Experimental Procedures") suggest that the kinetics may be more complex than previously thought. This observation is being further explored and will be presented elsewhere, since Scheme 1 is sufficient for description of the experiments presented here.  A 11.0 ϫ 10 5 3.4 ϫ 10 4 ␣T38W 9.1 ϫ 10 5 5.5 ϫ 10 4 ␤W37T 6.2 ϫ 10 5 2.4 ϫ 10 4 ␣T38W/␤W37T 5.5 ϫ 10 Crystallographic Structure of Deoxy-Hb ␣38 Thr 3 Trp- Fig.  6 (panel A) shows the difference map F(Hb ␣T38W)-F(HbA) in the deoxy state; in panel B only the negative values of the difference map (red contours) are shown, and the HbA structure is displayed. The excellent fit of the Trp side chain to the main positive peak may be seen. The largest negative peak represents the absence in this mutant of Thr ␣38 O␥-1 and a bound water molecule, seen in the structure of HbA; a second less intense negative peak may correspond to another water molecule. In the mutant, Trp C␥ overlaps with the native Thr C␥-2 of HbA and thus no difference peak is observed. It may be clearly seen that there is no evidence of any change in position of neighboring residues of the ␣ 1 ␤ 2 interface.
Molecular Modeling-Modeling of the T and R structures of our mutants (over and above the crystallographic structure of deoxy-Hb ␣T38W) was limited to simple geometric computations, such as the number of atoms in contact with the side chain at position C3, and the number of side chains interactions lost or established after amino acid substitutions. Nevertheless even this simple analysis yielded information useful for the interpretation of functional experiments.
As shown in Table V, upon introduction of a Trp at ␣38 (C3) the increase in the number of interface contacts between ␣ 1 and ␤ 2 is negligible in deoxy, but considerable in oxy-Hb. Conversely the substitution Trp 3 Thr at ␤37 (C3) leads to a substantial decrease in the number of contacts in both oxy-and deoxy-Hb, with loss of a hydrogen bond between ␤37 Trp and ␣94 Asp in the deoxy-Hb (see Fig. 7, panel C). Hence one could deduce that compared to HbA, the ␣T38W mutant has a more compact and extensive ␣ 1 ␤ 2 interface especially in the R-state; on the other hand, Hb ␤W37T has a looser interface, especially in the T-state, with loss of a hydrogen bond.
Assuming that these effects are additive in the double mutant, one may expect that in this hemoglobin cooperativity will be severely impaired, since the R-state would be favored with a concomitant destabilization of the T-state; this prediction agrees well with the experimental data reported above.
In order to compare the data obtained on Hb ␤W37T with observations reported in the literature on Hb Hirose (which bears the substitution ␤W37S, Sasaki et al., 1978), we have measured the atoms in a sphere of 4 Å to a Ser in ␤C3. As may be seen in Table V, the number of contacts in Hb Hirose is reduced, in both oxy and deoxy state, compared to our mutant ␤W37T, in agreement with the experimental data (see "Discussion").
Inspection of the water accessible surface in the proximity of position C3 allows evaluation of the cavities at the edge of the ␣ 1 ␤ 2 interface, and how the shape of the surface is affected by the mutations. The crystallographic structure of deoxy-Hb ␣T38W shows that the Trp in ␣C3 lies flat on the edge of the interface, excluding ␤145 Tyr and ␤100 Pro from contact with the solvent, and thus seems to act as a "hydrophobic plug" (Fig.  7, panel F). Modeling the oxygenated state of the same mutant shows that the side chain of the Trp introduced in ␣C3 could also fit nicely in a cavity, in contact with ␤145 Tyr.
The substitution ␤W37T, on the other hand, seems to deepen a cavity at the edge of the opposite side of the ␣ 1 ␤ 2 interface, possibly allowing to ␣141 Arg greater accessibility to the external medium.  single mutants and one double mutant in the ␣ 1 ␤ 2 interface of human hemoglobin. The mutations are in the so called "flexible joint" and "switch" regions, which are pseudo-symmetric, ␣T38W being in the switch region, and ␤W37T in the flexible joint; the double mutant contains both mutations. The first mutation (Thr 3 Trp) might be expected to affect primarily the allosteric properties, while the second (Trp 3 Thr) might also affect the stability of liganded and unliganded tetramers. Most, but not all, of the results presented above bear out these predictions. The properties of these new mutants are discussed in the framework of the two-state MWC model (Monod et al., 1965), in spite of the limitations that have been discussed in the literature (see, for example, Ackers et al. (1992)). The model provides a convenient analytical description of the data and allows discussion of the more interesting findings with a nomenclature known to most. Hb ␣T38W, which bears two Trp residues in the ␣ 1 ␤ 2 interface, is a slightly more stable tetramer than HbA, especially in the liganded form where the tetramer dissociation constant is decreased 6-fold . The three-dimensional structure of this mutant in the deoxy state shows that the indole side chain of Trp ␣38 has been accommodated by expulsion of two water molecules hydrogen-bonded to ␣38 Thr in HbA, without significant changes of the adjacent residues (Fig.  6). Thus, the enhanced stability of the oxy tetramer is consistent with the increased hydrophobic character of the interface (which is not perturbed by the bulkier side chain), with the more extensive contacts of the indole (Table V), and with the hypothesis that the most significant contribution to the stability of the tetramer resides in the pseudo symmetric ␣FG-␤C contact, which is unmodified. The slightly increased affinity of the T-state observed for Hb ␣T38W as compared to HbA is not easily accounted for, given that the constraints in the T-state are not decreased. On the other hand, the lower value of L 0 is consistent with the increased stability of R and thus with a (slightly) reduced cooperativity.
Hb ␤W37T, which has no Trp residues in the ␣ 1 ␤ 2 interface, dissociates into dimers more readily than HbA, in both the unliganded and liganded derivatives, as shown by results from analytical ultracentrifuge and flash photolysis to be published elsewhere, indicating that Trp at ␤C3 contributes to the stability of the ␣ 1 ␤ 2 interface in both quaternary structures. In spite of the destabilization of this interface, cooperativity of ␤W37T is reduced but preserved, not only because the contacts in the ␣C-␤FG switch region (␤97 His, ␣41 Thr, ␣44 Pro, and ␣38 Thr) are preserved, but also because Thr at ␤37(C3) can to some extent fulfill the role of Trp in the flexible joint, as indicated by modeling (Table V). The remarkable but limited competence of Thr at ␤37 residue is reflected in a small increase of the O 2 affinity of the T-state (Table I) and a somewhat greater increase of the rate constant for CO binding to deoxy-Hb (Table II); thus, the constraints of the T-state are partially released by mutation of ␤C3 and essentially maintained by mutation of ␣C3.
The interesting behavior of the double mutant Hb ␣T38W/ ␤W37T may be understood qualitatively on the basis of the properties of the two single mutants. Although the double mutant, like HbA, has only one Trp residue at the ␣ 1 ␤ 2 interface, ligand binding and tetramer dissociation are both remarkably different from wild type. This is the best evidence that the effect of mutations at topological position C3 is asymmetric and that perturbing the flexible joint and the switch indeed has different consequences. This double mutant dissociates more readily into dimers in both the liganded and unliganded derivatives, which we attribute largely to the flexible joint mutation ␤W37T. However, cooperativity is also reduced, even when compared to that of the two single mutants, as shown by the increased O 2 affinity of the T-state (Table I). These properties may be understood on the basis of two synergistic effects, i.e. a significant release of the interface constraints due to ␤W37T, leading to a more relaxed T-state (hence the smaller K T ), and an increase in the stability of the R-state tetramer, related to mutation ␣T38W. As a result the energy difference between the two allosteric states is reduced, as indicated by the value of the allosteric constant L 0 , which is smaller than HbA by ϳ15-fold.
As already stated, there is no known natural mutant of position ␣38 (Huisman, 1992); however, two site-directed mutants of this residue, namely Hb ␣T38S and ␣T38V, have been expressed and characterized by Hashimoto et al. (1993). The functional properties of both these Hbs are by and large similar to those of HbA. However, the substitutions in these case are semiconservative, and it is easily conceivable that the residues effectively replace the wild type Thr.
As to position ␤37, our data may be compared with those of the natural mutants Hb Hirose and Hb Rothschild. Comparison of Hb ␤W37T with Hb Rothschild (␤37 Trp 3 Arg; Gacon et al. (1977)) is complex, because of arginine's positive charge. Nonetheless, consistent with the other ␤37 mutants, Hb Rothschild displays reduced cooperativity and an increased tendency to dissociate into dimers. However, the structure of the deoxygenated derivative (Kavanaugh et al., 1992b) shows a novel and strong chloride binding site, which affects the functional properties of this Hb.
Hb Hirose (␤37 Trp 3 Ser) displays extremely low (if any) cooperativity, very rapid CO binding by flow, and a high tendency to dissociate into dimers in both the presence and the absence of oxygen (Sasaki et al., 1978). Hb ␤W37T, on the other hand, although extensively dissociated into dimers when liganded, is fully associated as the deoxy derivative (Fig. 5) and maintains a higher cooperativity than Hb Hirose, confirming that Thr but not Ser is a partially competent substitute for ␤37 Trp at the flexible joint. The computer-simulated structures of oxy-and deoxy-Hb ␤W37T (Fig. 7) indicate that the methyl group of Thr ␤37 makes contacts with ␣95 Pro and ␣140 Tyr in the oxy derivative and with ␣140 Tyr in the deoxy derivative; these contacts are lost or less extensive with Ser (Table V).
A site-directed mutant of position ␤37, in which a Phe substitutes the Trp, has been obtained by Ishimori et al. (1992). In this case the Phe would be expected to provide a more effective replacement for Trp than either Hb Hirose or our mutant. Unfortunately the authors did not measure the tetramer-dimer equilibrium and kinetics on their mutant Hb. As with ␤W37T, Hb ␤W37F displays high oxygen affinity and low cooperativity, even though the 1 H NMR spectrum of this mutant is consistent with that typical of T-state HbA.

Rates of Dimer Reassociation and Tetramer Dissociation in
Deoxyhemoglobin-Fitting the time course of association of deoxy ␣␤ dimers to a second order reaction (as reported in Fig.  4) shows small but systematic deviations, which are more marked for the mutants than for HbA. Nevertheless, the time course of reassociation of deoxy ␣␤ dimers of HbA, when fitted to the simplest possible scheme (see Scheme 1, under "Results"), yields a rate constant at low salt concentration slightly larger than, but not inconsistent with, that reported in the literature (Gray, 1974;Wiedermann and Olson, 1975). Small differences may be due either to the experimental conditions or the analysis. Among the rate constants reported in Table III, those determined for Hb ␤W37T and for the double mutant are more reliable because these Hbs are completely dissociated into dimers as HbO 2 and thus the total signal is larger and the initial dimer concentration does not need to be fitted.
Interestingly, Table III shows that the rate constant of reassociation of unliganded dimers and the effect of NaI are the same (within a factor of two) for all four hemoglobins. Thus, the effect of interface mutation(s) on the stability of the deoxygenated Hb tetramer is only very slightly (if at all) reflected in its rate of association, in agreement with Ackers et al. (1992); the same conclusion holds for the dissociating effect of NaI. These observations imply that formation of the productive dimerdimer complex is marginally affected by one or two conservative changes at the contact interface. This is not wholly surprising in view of the large surface buried in the contact between two ␣␤ dimers (ϳ1500 Å 2 ).
On the other hand, the effect of mutations on the rate of dissociation of the deoxy tetramer is more marked and follows an interpretable trend (Table IV). In deoxy-Hb ␣T38W there are 8 intersubunit contacts compared to 7 in deoxy-HbA, consistent with just a 30% decrease in the rate of tetramer dissociation (Table IV). The ␤37 mutants, on the other hand, both dissociate faster than Hb A, which we attribute to the loss (i) of the contact between ␣141 Arg and ␤37 Trp with perturbation of the network of salt bridges present in the T-state, and (ii) of a hydrogen bond between ␤37 Trp and ␣94 Asp (Fermi and Perutz, 1981). Therefore, the reduced number of contacts (Table V) agrees well with the increased rate of dissociation of the ␤37 mutants (which can be as much as 25-fold).
However, it should be pointed out that the rate constants for dissociation of deoxy-Hb ␤W37T and Hb ␣T38W/␤W37T are quite different. This implies that the effect of the two mutations is not additive; otherwise, the double mutant would dissociate as fast as Hb ␤W37T. This deviation from simple additivity suggest that the interface may be also distorted by long range effects. This is not unreasonable, given that perturbation of a single residue even at the ␣ 1 ␤ 1 interface leads to a large destabilization of the T-state and loss of cooperativity (see, for example, Amiconi et al. (1989), Weber et al. (1993), and Zhang et al. (1996)). A thorough interpretation of non additive effects necessarily implies taking into account long range interactions, which indeed are well documented. However, we would rather to defer this analysis until the crystallographic structure of these mutants (now in progress) is available.
Effect of Sodium Iodide-The iodide ion strongly promotes dissociation of the hemoglobin tetramer (Kellett and Gutfreund, 1970), more than other anions such as Cl Ϫ (Antonini and Brunori, 1971). We found that I Ϫ affects both the combination and dissociation rate constants (Tables III and IV). The 25-50-fold slower reassociation of deoxy dimers may be due to binding of I Ϫ to several positively charged residues of the ␣ 1 ␤ 2 interface (␣92 Arg, ␤40 Arg, and ␣40 His). Nonetheless, the effect of I Ϫ on the bimolecular association rate constant of the mutants is approximately the same as for HbA (Table III).
On the other hand, the effect of I Ϫ on the rate of dissociation of the deoxy tetramer depends in a rational way on the amino acid residue at position C3. This is evident comparing the ratio of the rate constants kЈ and k reported in Table IV. With a Trp in the switch region as in Hb ␣T38W, I Ϫ accelerates dissociation only 3-fold compared to 13-fold in HbA. We interpret the reduced effect as resulting from the more hydrophobic character of the interface and from the position of Trp, which seems to act as a hydrophobic plug, shielding from contact with the solvent ␤145 Tyr and ␤100 Pro and forming a barrier to the penetration of this ion.
When Trp is removed from the flexible joint as in Hb ␤W37T, I Ϫ accelerates dissociation approximately 13-fold, just as in HbA. In deoxy-Hb, ␤37 Trp is on the edge of the ␣ 1 ␤ 2 interface ( Figs. 1 and 7), and its orientation is such that ␣95 Pro and ␣140 Tyr are also exposed to the external medium; mutation of Trp with Thr therefore does not result in exposure to the solvent of side chains previously buried in the interface. Thus, the similarity in the effect of I Ϫ on the dissociation rate constants of HbA and Hb ␤W37T mutant is fully consistent with the proposal that access of the anion to the ␣ 1 ␤ 2 interface is toward the switch region, near position ␣38, and therefore not affected by substitution in the flexible joint region. In support of this hypothesis, the effect of I Ϫ on the double mutant is again smaller and similar to that of Hb ␣T38W, because of the hydrophobic plug effect. Thus, we conclude that the effect of I Ϫ on the stability of the deoxy tetramer is linked to binding of this anion to the interface near position C3. This prediction may be checked by examination of the crystallographic structure of deoxy-HbA in the presence of NaI. FIG. 7. Residues at position C3 (shown in black) and their neighboring amino acids in a sphere of 4 Å for oxy-and deoxy-HbA (Table V). Residues belonging to the ␣ and ␤ chains are shown in light gray. The labels are indicated only for the monomer opposite the C3 residue. The dots represent the water accessible surface (probe radius 1.4 Å) within the HbA tetramer. Hydrogen bonds involving the C3 side chain (either with other amino acid residues or with water molecules) are represented by a gray dotted line. Oxy-Hb: a, ␣38Thr from the crystallographic structure (Shaanan, 1983); b, ␣38Trp modeled starting from a. Deoxy-Hb: c, ␤37Trp from the crystallographic structure (Kavanaugh et al., 1992b); d, ␤37Thr modeled from c; e, ␣38Thr from the crystallographic structure (Kavanaugh et al., 1992b); f, ␣38Trp form the crystallographic structure (this paper). It has to be noticed that the number of residues in a sphere of 4 Å from the amino acid side chains of interest are fewer when the side chain is smaller and larger when this is greater; therefore, the amino acids displayed in panels a and b, c and d, and e and f of this figure are not the same.