Oligomerization and ligand binding in a homotetrameric hemoglobin: Two high-resolution crystal structures of hemoglobin Bart's (γ₄), a marker for α-thalassemia

Abstract

Hemoglobin (Hb) Bart's is present in the red blood cells of millions of people worldwide who suffer from α-thalassemia. α-Thalassemia is a disease in which there is a deletion of one or more of the four α-chain genes, and excess γ and β chains spontaneously form homotetramers. The γ₄ homotetrameric protein known as Hb Bart's is a stable species that exhibits neither a Bohr effect nor heme–heme cooperativity. Although Hb Bart's has a higher O₂ affinity than either adult (α₂β₂) or fetal (α₂γ₂) Hbs, it has a lower affinity for O₂ than HbH (β₄). To better understand the association and ligand binding properties of the γ₄ tetramer, we have solved the structure of Hb Bart's in two different oxidation and ligation states. The crystal structure of ferrous carbonmonoxy (CO) Hb Bart's was determined by molecular replacement and refined at 1.7 Å resolution (R = 21.1%, R_free = 24.4%), and that of ferric azide (N₃⁻) Hb Bart's was similarly determined at 1.86 Å resolution (R = 18.4%, R_free = 22.0%). In the carbonmonoxy–Hb structure, the CO ligand is bound at an angle of 140°, and with an unusually long Fe-C bond of 2.25 Å. This geometry is attributed to repulsion from the distal His63 at the low pH of crystallization (4.5). In contrast, azide is bound to the oxidized heme iron in the methemoglobin crystals at an angle of 112°, in a perfect orientation to accept a hydrogen bond from His63. Compared to the three known quaternary structures of human Hb (T, R, and R2), both structures most closely resemble the R state. Comparisons with the structures of adult Hb and HbH explain the association and dissociation behaviour of Hb homotetramers relative to the heterotetrameric Hbs.

Mammalian hemoglobins (Hbs) function as heterotetramers comprising two alpha-type chains and two beta-type chains. In normal healthy humans, transcription and translation of the alpha-type and beta-type globin chains produces closely matched quantities of each type that spontaneously aggregate to form heterotetramers (Valdes and Ackers 1977). However, in the conditions known as the thalassemias mis-matched quantities of globin chains are produced. In the case of alpha (α) thalassemias, excess beta-type chains aggregate to form abnormal homotetrameric Hbs (Rigas et al. 1956; Ager and Lehmann 1958; Dance and Huehns 1962). Furthermore, as the α globin chain is expressed at all stages of human development, two main varieties of development-stage linked homotetramers are formed in α-thalassemias. During the adult stage the predominant abnormal homotetramer is HbH (β₄), while at the fetal stage Hb Bart's (γ₄) predominates. The beta (β) chains have a greater attraction than gamma (γ) chains for α chains in tetramers (Lehmann and Huntsman 1974); therefore, when both β and γ chains are present and there is an insuffient supply of α, HbA (α₂β₂) forms in preference to HbF (α₂γ₂). Excess γ chains then aggregate to form Hb Bart's. Only when the amount of β by itself is greater than α are significant quantities of HbH observed.

Alpha thalassemia is associated with four general clinical syndromes that correspond with the loss of increasing numbers of the four Hb α genes (Dickerson and Geis 1983; Bunn and Forget 1986): (1) The silent-carrier state (α-thalassemia 2). One α gene is missing. This condition is entirely symptomless with 1–2% Hb Bart's present at birth. (2) Classical α-thalassemia trait (α-thalassemia 1). Two α genes are missing. About 5% Hb Bart's is present at birth. The symptoms, minor red blood cell abnormalities but no anemia, are the same for either homozygous or heterozygous individuals. (3) Hemoglobin H disease. Three α genes are missing, resulting in the presence of Hb Bart's in infancy and HbH in aldulthood. There has also been a report (Dance and Huehns 1962) of rare HbH disease individuals with a tetramer of another β-like chain, delta (δ). HbH disease is characterized by a chronic hemolytic anemia of variable severity. (4) Hydrops fetalis. All four α genes are missing. This condition is fatal, leading to stillbirth in the last few weeks of pregnancy. Therefore, the presence of Hb Bart's, which was first recognized as such by Ager and Lehmann (1958), is an important indicator of a major health disorder. Hb Bart's occurs in tens of millions of humans worldwide, and reaches a frequency of over 40% in the Laotian population of South-East Asia (Wasi 1986).

The β₄ homotetramer has been studied in some detail, using both naturally isolated material and protein prepared from the aggregation of β chains isolated from normal adult hemoglobin (Huehns and Beaven 1962; Benesch and Benesch 1974^; Philo et al. 1988). In contrast to heterotetrameric hemoglobins, HbH has been found to have a high affinity for oxygen (Benesch et al. 1968) and almost no cooperativity in oxygen binding (Kurtz et al. 1981), and to be relatively unstable in solution (Dance et al. 1963). Crystal structures of the carbonmonoxy (Borgstahl et al. 1994a) and deoxygenated (Borgstahl et al. 1994b) forms of HbH indicate that the protein remains in the R-state quaternary structure both in the absence and presence of heme ligands.

In contrast to HbH, little is known of the structural and functional properties of Hb Bart's, partly due to the previous unavailablity of large amounts of protein. It is more stable in solution than HbH (Bunn and Forget 1986), and has a lower affinity for free alpha chains than does HbH (Huehns and Beavan 1962). It has also recently been shown (Adachi et al. 2000) that the γ₄ homotetramer self-assembles by a different pathway from that of β₄, involving stable γ₂ dimers. Hb Bart's has been crystallized previously (Czerwinski et al. 1981), but no structure is available. Indeed, although many studies on the functional properties of fetal hemoglobin (α₂γ₂, HbF) have appeared (e.g., Allen et al. 1953; Weatherall et al. 1974; Huisman 1981), the 2.5 Å structure of the deoxy form of this protein (Frier and Perutz 1977) remains the only structure available for a hemoglobin incorporating the γ-chain.

Here we present the three-dimensional structure of Hb Bart's (γ₄), in two liganded forms, determined at high resolution from crystals that were unexpectedly obtained during attempts to crystallize the human embryonic hemoglobin Portland (ζ₂γ₂) (Kidd et al. 2001). The crystal structures allow us to derive new information on the structural organization and ligand binding properties of this medically significant protein, and to obtain a second, high-resolution view of the hemoglobin γ-chain. Further, by comparing the subunit interfaces in α₂β₂, β₄, and now γ₄, we are able to probe the factors that stabilize the γ₂ dimer in solution, and offer an explanation as to why β-type globins aggregate to form homotetramers whereas α-type globins do not.

Results

Structure determination and model quality

The crystal structure of the carbonmonoxy form of ferrous Hb Bart's (CO–γ₄) was determined by molecular replacement using a search model derived from the 1.8 Å structure of CO–β₄ (Borgstahl et al. 1994a). The structure was refined at 1.7 Å resolution to a crystallographic R-factor of 0.211 (R_free = 0.244). Refinement results are summarized in Table 1. The structure of the azide form of ferric Hb Bart's (NNN-γ₄) was also determined by molecular replacement using a search model derived from the 1.7 Å model of CO–γ₄, and was refined at 1.86 Å resolution to an R-factor of 0.184 (R_free = 0.220). Both structures exhibit good stereochemistry with root-mean-square (rms) deviations from ideal geometry of 0.006 Å for bond lengths and 1.06° for bond angles for CO–γ₄ and 0.005 Å and 1.06° for NNN-γ₄ (Table 1).

Table 1.

Refinement and model statistics

	CO-γ4	NNN-γ4
Resolution limits (Å)	20.0–1.7	20.0–1.86
Number of reflections	26,914	22,315
Reryst (%)	21.1	18.4
Rfreea (%)	24.4	22.0
Bond length r.m.s. deviation (Å)	0.006	0.005
Bond angle r.m.s. deviation (deg.)	1.06	1.06
Average B-factor (Å2):
Main chain	15.4	15.0
Side chain	17.0	16.8
Solvent	30.4	28.2
Number of protein atoms (2 × 146 residues)	2260	2260
Number of heme atoms	86	86
Number of ligand atoms	4 (2 CO)	6 (2 N3−)
Number of water molecules	287	242

^a 10% randomly omitted reflections were used for R_free.

The final models for both structures comprise the complete γ-chain (146 residues) for each of the two monomers in the asymmetric unit, which have been independently refined. The carbonmonoxy-dimer contains 2260 nonhydrogen protein atoms, two CO–heme groups, and 287 water molecules, and the azide–dimer 2260 nonhydrogen protein atoms, two N₃⁻–heme groups, and 242 water molecules. With the exception of several side chains on the surface of the molecule that exhibit little or no electron density and have been assigned zero occupancies (two each in molecules A and B for CO–γ₄ and one in A, four in B for NNN–γ₄) the electron density for the entire polypeptide is in each case excellent (Fig. 1 ▶). There are indications of possible minor alternate conformations for other side chains (12 in molecule A and 12 in molecule B for both structures), but these have not been modeled. Root-mean-square (rms) differences when the two crystallographically independent molecules are superimposed are 0.28 Å for CO–γ₄ and 0.30 Å for NNN–γ₄, for all main-chain atoms. Ramachandran plots of main-chain torsion angles (φ,ψ) calculated with PROCHECK (Laskowski et al. 1993) show that 91.5% of residues lie in the most favored region for CO–γ₄ and 93.8% for NNN–γ₄, with no residues in disallowed areas. The heme groups are well defined, and there is clear density visible for the ligands bound to each heme.

Fig. 1.

Electron density for CO–γ₄ and identification of the HBG1 gamma gene product. Normally, two gamma genes (HBG1 and HBG2) are expressed in humans. Their gene products differ at residue 136 (out of 146) where glycine is found in the G-gamma product (HBG2) and alanine is found in the A-gamma product (HBG1). (A) Density around residue 136 (middle of figure) modeled as glycine and refined. The 2F_o–F_c maps (blue) are contoured at 1.2σ. The F_o–F_c maps are contoured at ±3σ with positive density in green and negative density in red. (B) Density around residue 136 after modeling as alanine and refinement. The density corresponds best with an alanine, identifying the cloned gene as HBG1. Figures drawn using BOBSCRIPT (Esnouf 1997) and RASTER3D (Merritt and Murphy 1994).

Humans possess two γ genes adjacent to each other on the same chromosome (Dickerson and Geis 1983). These two genes differ only by a single residue (Ala versus Gly) at position 136, and are designated ^Aγ and ^Gγ. During the cloning of the embryonic hemoglobin genes into a yeast expression system (Mould et al. 1994; Hofmann et al. 1995) it was initially not known which γ gene had been inserted; however, DNA sequencing of the cloned γ gene indicated that ^Aγ had been cloned (R. Mould and T. Brittain (unpubl.) Based on our 1.7 Å and 1.86 Å electron density maps (Fig. 1 ▶), it is clear that the residue at position 136 is an alanine, and this result confirms the cloning of the ^Aγ gene.

Tertiary structure

The overall molecular structure of Hb Bart's is shown in Figure 2 ▶. The two monomers of the asymmetric unit, labeled A and B (Fig. 2A ▶), are analogous to subunits α1 and β1 of adult HbA, respectively. The complete homotetramer (Fig. 2B ▶) is generated by crystallographic symmetry; the two symmetry-related monomers are labeled A′ and B′, and are analogous to HbA α2 and β2, respectively. The CO–γ₄ and NNN–γ₄ structures are very similar, with rms differences of 0.34 Å for main-chain atoms and 0.96 Å for side-chain atoms, after superposition of the entire structures.

Fig. 2.

Structure of the carbonmonoxy form of hemoglobin Bart's (γ₄). (A) Ribbon diagram of the two subunits (labeled A and B) of γ in the asymmetric unit. The hemes are depicted as ball-and-stick models. (B) Ribbon diagram of the CO–γ₄ tetramer. The tetramer was generated through crystallographic symmetry (symmetry-related globins are labeled A′ and B′). Figures were drawn using the program MOLSCRIPT (Kraulis 1991), and rendered with RASTER3D (Merritt and Murphy 1994).

Superposition of the γ subunits on to other β-type globin subunits shows a generally close correspondence. Taking the A subunit of CO–γ₄ as a reference, the entire chain, with the exception of the N-terminal residue, Gly1, can be superimposed on subunit A of CO–β₄ (Borgstahl et al. 1994a) with an rms difference of 0.73 Å for 145 Cα atoms. The most significant deviation involves the N-terminal A helix (residues 7–18), which is tilted slightly inwards towards helix E so that its N-terminus is displaced by about 2 Å; the movement is further propagated through the extended residues 1–6 such that the N-terminal residue of the γ-chain, Gly1, is displaced 4 Å from that of the β-chain (see Fig. 4 ▶). This displacement of the N-terminal residue seems to be a characteristic feature of the γ-chain (compared with the β-chain) as it is also seen in fetal Hb (Frier and Perutz 1977), and is implicated in the greater tetramer stability of HbF relative to HbA (Manning et al. 1999). Interestingly, superposition of the γ subunits on to either of the β subunits of heterotetrameric oxy-HbA (Shaanan 1983) shows somewhat less agreement than for the β subunits of homotetrameric β₄, with an rms difference of 0.90 Å (145 Cα atoms). This implies that quaternary structure differences related to homotetramer formation do have some slight impact on the tertiary structure.

Fig. 4.

Superposition of CO–γ₄ (sky-blue) upon R-state adult oxy-hemoglobin, O₂–α₂β₂ (α in yellow, β in orange). The alignment was based on superposition of the two γ-chains on to the two β-chains, and shows clearly the displacement of the A helix and N-terminus (indicated) of the γ-chain relative to the β-chain. On the other hand, the N-terminus of the α-chain in α₂β₂ corresponds closely with that of the γ-chain in γ₄). As observed for β₄ (Borgstahl et al., 1994a), γ₄ most closely resembles the R-state of liganded adult hemoglobin (Shaanan, 1983). Figures were drawn using MOLSCRIPT (Kraulis 1991) and RASTER3D (Merritt and Murphy 1994).

Comparing the location and orientation of the heme group within the globin folds, if the A subunit of CO–γ₄ is superimposed on to the A subunit of CO–β₄, using the globin Cα atoms alone, the heme groups match with a mean positional difference (for all atoms except those of the vinyl and propionate groups) of 0.35 Å. Most of this difference comes from a slight translational displacement of 0.25 Å, in which the movement of the heme accompanies a similar displacement of helix F, to which it is attached.

Heme stereochemistry

The heme groups in the CO and azide complexes match very closely (rms difference 0.42 Å for 32 heme atoms when the structures are superimposed on the basis of their main chain atoms). In both complexes the hemes are remarkably planar, with little buckling or ruffling. This must be a genuine property of the heme stereochemistry for Hb Bart's, at the pH of crystallization, because the hemes in both subunits of both structures were independently refined using minimal bond length, angle, dihedral, and improper constraints, and using two different refinement protocols. In each case the iron atom lies in the plane of the porphyrin, and its bond with Nɛ2 of the proximal histidine ligand, His93, is about 2.1 Å (Table 2), as in other liganded Hbs. The only differences in heme stereochemistry are for two of the heme substituents. First, one of the vinyl groups is rotated ∼180° about its C-C bond in the NNN–γ₄ structure relative to the CO–γ₄ structure, a difference that is found in both subunits. Second, one of the propionic acid groups of CO–γ₄ is in two conformations, again in both subunits, whereas only a single conformation is found in NNN–γ₄.

Table 2.

Metal–ligand geometry and interactions

	CO-γ4			NNN-γ4
	A	B		A	B
A. Angles (deg.)
Fe-C-O	140	135	Fe-N1-N2	115	115
Fe-C/heme normal	9	10	Fe-N1/heme normal	−1	1
			N1-N2-N3	5	11
B. Distances (Å)
Fe-C	2.25	2.26	Fe-N1	2.16	2.19
Fe-Nɛ2(His92)	2.12	2.09	Fe-Nɛ2(His92)	2.14	2.10
C...Nɛ2(His63)	3.05	2.97	N1...Nɛ2(His63)	2.72	2.84
O...Nɛ2(His63)	3.78	3.85	N2...Nɛ2(His63)	3.55	3.76
C...Cδ2(His63)	3.66	3.55	N1...Cδ2(His63)	3.40	3.38
C...Cγ2(Val67)	3.35	3.34	N1...Cγ2(Val67)	3.19	3.24
O...Cγ2(Val67)	3.51	3.48	N2...Cγ2(Val67)	3.22	3.28
			N3...Cγ2(Val67)	3.66	3.79
O...Cδ2(Leu106)	3.72	3.86	N3...Cδ2(Leu106)	3.40	3.28

Ligand binding

Both the carbon monoxide and azide complexes show full occupancy of their bound ligands. The crystals of CO–γ₄ were grown in carbon monoxide-saturated conditions, in the dark, and in the presence of a reducing agent, to avoid oxidation and to promote full CO occupancy. Initial omit maps (both 2F_o–F_c and F_o–F_c) showed strong electron density for bound CO in both independent subunits. After model building and refinement, strong electron density in the 2F_o–F_c map (Fig. 3A ▶) and low temperature factors for the CO atoms (average B values of 16.7 Ų and 19.2 Ų for the C and O atoms, respectively) are consistent with full occupancy. Similarly, the azide ligands in NNN–γ₄ were also represented by strong electron density in 2F_o–F_c and F_o–F_c omit maps (Fig. 3B ▶), and have full occupancy as judged by their low temperature factors (average B value of 15.8 Ų).

Fig. 3.

Heme groups with bound ligands. The maps for both structures were calculated with Fourier coefficients 2F_o–F_c, and were contoured at 1.2σ. (A) Two views of one of the heme groups in the 1.7 Å structure of CO–γ₄. The carbon monoxide ligand is depicted as a ball-and-stick model with carbon in yellow and oxygen in red. The Fe-C-O angle is 140° and the Fe-C distance is 2.25 Å. (B) Two views of one of the heme groups in the 1.86 Å structure of NNN–γ₄. The azide ligand is depicted as a ball-and-stick model with nitrogen in blue. The Fe-N1-N2 angle is 115° and the Fe-N1 distance is 2.16 Å. Figures were drawn using BOBSCRIPT (Esnouf 1997) and RASTER3D (Merritt and Murphy 1994).

Carbon monoxide binding

The CO ligand in Hb Bart's is bound in a single well-defined conformation, with a stereochemistry that is essentially the same in both independently refined subunits of CO–γ₄; details are given in Table 2. Two features of the CO binding mode in Hb Bart's appear to differ significantly from that in other Hbs and from the Fe-CO geometry found in model CO–porphyrin complexes (Peng and Ibers 1976). First, the Fe-C bond length (2.25 Å) is markedly greater than in other CO–Hbs where it is consistently around 1.8 Å (Derewenda et al. 1990; Borgstahl et al. 1994a). Second, the CO ligand is bound in an angular mode, with an Fe-C-O angle of ∼135–140°. This compares with Fe-C-O angles of ∼170° in R-state CO–HbA (Derewenda et al. 1990), R2-state CO-HbA (Silva et al. 1992) and CO–β₄ (Borgstahl et al. 1994a) and angles close to 180° in CO–porphyrin complexes (Peng and Ibers 1976). Accompanying this, the Fe-C bond deviates by 10° from colinearity with the heme normal through the Fe atom, markedly more than in the other CO–Hb structures where the deviation is ∼6° (Derewenda et al. 1990; Borgstahl et al. 1994a), and in CO–porphyrin complexes where it is colinear (Peng and Ibers 1976). Both subunits were refined independently and without restraints on the Fe-C bond lengths or Fe-C-O bond angles, strongly suggesting that these differences are a real feature of the Hb Bart's crystal structure.

The geometry of ligand binding is likely to be influenced by the crystallization conditions, especially pH, as well as by the structural constraints of the ligand binding pocket (Ray et al. 1994). The unusually planar heme group should allow greater Fe–porphyrin back-bonding and thereby reduce Fe-CO back-bonding (Li and Spiro 1988). Considering that the extent of heme buckling varies in different CO–Hbs, however, yet the Fe-C distance is consistently around 1.8 Å, heme planarity is unlikely to be the main factor in the substantially lengthened and more angular Fe-CO bond in Hb Bart's.

The closest approach to the CO ligand is from the distal histidine, His63, whose side chain Nɛ2 atom is only 3 Å from the CO carbon atom. His63 is in contact with solvent (its Nδ1 atom forms a 2.8 Å hydrogen bond with a well-ordered water molecule), and at the pH of crystallization (4.5–4.9) protonation of His63 is likely. The pKa of His63 in Hb Bart's is not known but in Mb the equivalent distal histidine, His64, has a pKa of 4.4 (Yang and Phillips 1996, and references therein). Because the Nɛ1. . .C(CO) interaction cannot be a hydrogen bond, the increased displacement of the CO ligand from the heme normal is most probably due to this (repulsive) contact. In Mb, the distal histidine can relieve this repulsion at low pH by swinging out of the ligand-binding pocket at low pH (Yang and Phillips 1996), but this does not occur in the present structure. Positive charge on His63 should also change the polarization of the CO ligand (Kushkuley and Stavrov 1997), and hence, the Fe-C bond order. The CO oxygen atom is positioned between the side chains of His63 and Val67 (Fig. 3A ▶) such that it is in van der Waals contact with His63 Cɛ2, Val67 Cγ1 and Leu106 Cδ2. This packing is similar to that in other CO–Hb structures, however, and we suggest that the longer Fe-C bond and more angular bonding in Hb Bart's results from reduced Fe-CO back-bonding, due to the positive charge on the nearby His63.

Azide binding

In contrast to the CO binding in CO–γ₄, the azide binding in NNN–γ₄ appears close to optimal. The average Fe-N1-N2 angle is 115° (Table 2), close to the ideal sp² bond angle of 120°, and very similar to that found in other azido–met globin structures; 118° in the hemoglobin-like molecule from Vitreoscilla stercoraria (VtHb, PDB entry 2VHB; Tarricone et al. 1997), 119° in horse myoglobin (HMb, PDB entry 1AZI; Maurus et al. 1998), and 117° in sperm whale myoglobin (SWMb, PDB entry 1SWM; Rizzi et al. 1993). This orientation allows a short, linear hydrogen bond with the distal histidine in which the lone pair on the azide N1 atom and a hydrogen on the protonated Nɛ2 of His63 would be directed precisely towards each other. This interaction presumably also explains why the azide ligand is bound with the Fe-N1 bond colinear with the heme normal. The Fe-N1 bond length (average 2.17 Å) is also similar to that in Mb (Maurus et al. 1998). Back-bonding between Fe(III) and azide should be less significant than between Fe(II) and CO, and protonation of His63 should in this case be favorable to ligand binding. The overall orientation of the azide ligand in each subunit of NNN–γ₄ is also very similar to that in the other met–azido globin structures; it is oriented towards the hydrophobic back of the heme pocket, in contact with His63, Val67, and Leu106, with the N3 atom positioned approximately over the C1C heme atom.

Quaternary structure

Three quaternary states have been characterized crystallographically for heterotetrameric Hbs; one unliganded T state (Fermi et al. 1984; Kavanaugh et al. 1992) and two liganded states; the R (Shaanan 1983; Derewenda et al. 1990) and the R2 (Silva et al. 1992) or Y state (Smith et al. 1991; Smith and Simmons 1994). The R and R2 states are similar, and there is debate as to whether one or other is an intermediate in the transition between liganded and unliganded quaternary structures, or whether both are variants of a single R state (Tame 1999).

Multiple superpositions show that the liganded Hb Bart's structures most closely resemble the standard R state of the heterotetrameric Hbs. In the latter, the αβ dimer interface (α1β1 or α2β2) is the most extensive in the tetramer, and changes least during the R-to-T transition, that is, the αβ dimer remains largely intact. Superposition of the γA subunit of Hb Bart's on to the individual subunits of R-state HbA (human oxy-Hb; Shaanan 1983) gives an rms difference of 1.03 Å (132 Cα) for the HbA α subunit and 0.87 Å (145 Cα) for the β subunit. Superposition of the entire γ₂ (AB) dimer of Hb Bart's on to the αβ dimer of HbA shows that they, too, are very similar; the rms difference of 1.11 Å (277 Cα) for the whole dimer is only slightly more than for the individual subunits. If one subunit pair is first superimposed (e.g., γA on α) only a small rotation (1.8°) is required to superimpose the other (γB on β). If the R2-state HbA structure (Silva et al. 1992) is used, a larger rotation (4.0°) is needed. Considering now the tetramer; if one dimer pair is superimposed (γ₂(AB) on α1β1) a rotation of 8.3° is required to superimpose the other dimer pair (γ₂(A′B′) on α2β2) if the R state is assumed, and 19.9° if the R2 state is assumed. Thus, the γ₄ homotetramer much more closely resembles the standard R state (Fig. 4 ▶).

Essentially the same quaternary structure is shared by the liganded β₄ homotetramer (Borgstahl et al. 1994a); the γ₂ dimer can be superimposed on to the β₂ dimer with an rms difference of only 0.82 Å (individual subunits 0.75 Å) and a rotation of only 1.0° is then required to superimpose the other dimer pair; the whole γ₄ tetramer can be superimposed on to β₄ with an rms difference of only 0.90 Å. We conclude that this quaternary structure is characteristic of a liganded β-type homotetramer, and is not affected by sequence differences in the individual subunits.

To compare the subunit interfaces of CO–γ₄ with those of CO–β₄ and oxy–α₂β₂ we calculated the surface area buried in each interface, using a 1.4-Å probe, with the program AREAIMOL from the CCP4 program suite (Collaborative Computational Project Number 4 1994); these results are summarized in Table 3.

Table 3.

Comparison of subunit interfaces ^a

	HbA (α2β2)	Bart's (γ4)	HbH (β4)
α1β1 interface	1805	1580	1735
α1α2 interface	800	700	500
α1β2 interface	1010	1200	1325
β1β2 interface	350	700	500

^a Buried surface areas (Ų) given for each interface.

The most extensive interface in HbA is the α1β1 (or α2β2) interface, formed primarily by residues from the BC and GH corners of each subunit; residues 30–36 and 103–126 of the α subunit and 30–35 and 111–131 of the β subunit. In CO–γ₄ and CO–β₄, this interface is slightly reduced in size; to 1580 Ų in γ₄ and 1730 Ų in β₄, compared with 1805 Ų in HbA. The greater reduction in γ₄ is due largely to the substitution of His116 in β by Leu116 in γ, and of Cys112 in β by Thr112 in γ, and a 2–3° rotation of helix G, to which these residues belong, in γ₄ relative to β₄. In HbA and in β₄, His116 Nɛ2 is hydrogen bonded to the carbonyl oxygen of Pro114 across the interface, but this interaction cannot occur in γ₄ because of the His → Leu substitution. In β₄, the Cys112 side chains from both subunits adopt dual conformations to avoid mutual steric conflicts; the closest approach of the two sulfur atoms is 2.99 Å. In γ₄, however, the rotation of helix G prevents a steric clash between the two Thr112 side chains despite conformations remarkably similar to the dual conformations of the cysteine side chains of β₄.

The greatest difference between γ₄ and HbA is in the α1β2 interface, which is significantly enhanced in the homotetramer, to be almost as extensive as the α1β1 interface. This enhancement is also a feature of β₄. This interface involves primarily the 3₁₀-helix C (residues 35–42), and residues 96–101 of the FG corner, and the increased interfaces in γ₄ and β₄ arise from interactions between the Arg40 and Asp99 side chains from both subunits. In the β₄ structure (Borgstahl et al 1994a), equally populated dual conformations are found for both Arg40 and Asp99, in both subunits, allowing two populations of Arg40. . .Asp99 salt bridges at this interface. The equivalent γ₄ interface also has Arg40 and Asp99 in close proximity to each other and Arg40 and Asp99 of the adjacent symmetry-related subunit. However, the electron density for these residues is not as clearly defined in γ₄, suggesting that these four residues at this interface are mobile.

Homotetramer symmetry requires that the α1α2 and β1β2 interfaces are equivalent in γ₄. These interfaces involve mainly the chain termini, residues 1–2 and 143–146 and are significantly larger in γ₄ than in β₄ (Table 3) because of the displacement of the γ-chain N-terminus relative to that of the β-chain (Fig. 4 ▶). This brings it closer to the C-terminus of the neighboring subunit. The substitution of Ser143 in the γ-chain for His143 in the β-chain also increases the interface in γ₄. In heterotetrameric HbA, in contrast, the α1α2 interface is slightly larger than that in γ₄ (and includes a salt bridge between the chain termini that is not present in either homotetramer), whereas the β1β2 interface is much smaller.

Discussion

An important feature of hemoglobin is that the individual globin chains are synthesized separately and must then assemble into functional heterotetrameric Hb molecules. Because the β-type (but not α-type) globin chains also have the ability to self-assemble into stable homotetramers, these homotetrameric species must dissociate prior to formation of the functional heterotetramer. Structural differences between the γ₄, β₄ and α₂β₂ Hb tetramers thus offer clues as to what factors influence their association/dissociation behavior, and ultimately correct Hb self-assembly.

Differences in association/dissociation behavior of β and γ chains

Both the β and γ globins can form stable homotetramers, and our structural comparisons show that their quaternary structures are extremely similar. Why, then, does the self-assembly mechanism for the γ chain go through a stable γ₂ dimer intermediate (Adachi et al. 2000; Kidd et al. 2001), whereas the β chain shows very little dimer formation, and the β₄ homotetramer assembles directly from monomers (Bunn and Forget 1986)? The nature of the γ₂ dimer is unknown, but the largest interface in the γ₄ homotetramer is that equivalent to the α1β1 (or α2β2) interface in HbA, and it is reasonable to assume that this might be the γ₂ dimer interface. Differences in this interface may therefore explain the differences in homodimer formation. Moreover, we note that the other major interface in the homotetramers, equivalent to α1β2, is essentially the same in γ₄ and β₄, dominated by the Arg40/Asp99 pairs contributed by both subunits.

Four residues at the α1β1 interface in β₄ differ from those in γ₄. These are β51Pro, β112Cys, β116His, and β125Pro versus γ51Ala, γ112Thr, γ116Ile, and γ125Glu. Mutagenesis studies suggest that only the residue at position 116 contributes to differences in heterotetramer formation with α chains (Adachi et al. 2000). This residue also accounts for the ability of γ chains to form stable homodimers. When γ116Ile is mutated to His, γ chains then form monomers in addition to dimers just like β. Size-exclusion chromatography results suggest that the substitution of His for Ile at 116 promotes the dissociation of γ₄ and causes it to follow an assembly pathway analogous to the 4β ↔ β₄ pathway instead of the wild-type 4γ ↔ 2γ₂ ↔ γ₄ pathway (analogous to HbA). Curiously, the interface equivalent to α1β1 is actually larger in β₄ than in γ₄ (Table 3), yet it is less stable. The His64Ile mutation suggests that it is the burial of the hydrophobic Ile116 in the γ oligomer, compared with the protic His116 in the β oligomer, that is critical in giving enhanced stability, both because of the hydrophobic effect and because His116 is expected to be protonated in the pH range in which oligomer dissociation occurs. In addition, repulsive interactions between the two sulfurs of βCys112 in this interface, even if in two conformations, should be stronger in β₄ compared to the two oxygens of γThr112 in γ₄, thus weakening this interface even further in β₄.

Why are homotetramers formed by β-type chains but not by α-type chains?

Comparison of the subunit interfaces in the γ₄ and β₄ heterotetramers with those in the α₂β₂ heterotetramer of HbA highlights several notable features. First, in both γ₄ and β₄ homotetramers the interface equivalent to α1β2 is more extensive (Table 3) and more tightly packed (Borgstahl et al. 1994a) than the same interface in HbA, and includes the cluster of Arg40/Asp99 pairs. Much of this enhancement comes from the presence of β-chain residues Trp37 and Arg40 (both Thr in the α-chain). Second, the β1β2 interface in heterotetramers is very small and contains no specific interactions, whereas the equivalent interface in the homotetramers is significantly larger (Table 3) and contains specific interactions between the N- and C-terminal residues. These latter interactions are more extensive in γ₄ than in β₄, which may perhaps imply greater homotetramer stability for γ₄. In combination, however, these two effects, which stabilize homotetramer formation by the β- and γ-globin chains, arise from features specific to these chains, explaining why α-chains do not form stable homotetramers.

Materials and methods

Crystallization and data collection

The Hb Bart's crystals used in this work were obtained unexpectedly during attempts to crystallize the human embryonic hemoglobin Portland ζ₂γ₂; crystals grew over a wide range of pH (4.5–8.5) but the poorly diffracting needle-shaped crystals obtained at higher pH proved to be Hb Portland, whereas the block-shaped crystals obtained at lower pH were of Hb Bart's. Evidently, dissociation of the ζ and γ subunits occurred at pH below 5.0, leading to the reassociation of the γ subunits as the homotetramer γ₄ (Kidd et al. 2001).

The recombinant hemoglobin Portland used was expressed in Saccharomyces cerevisiae and purified as previously described (Mould et al. 1994; Hofmann et al. 1995). The preparation of the NNN–γ₄ crystals is described in Kidd et al. (2001). For preparation of the carbonmonoxy form of the protein, carbon monoxide gas was gently bubbled through solutions of hemoglobin Portland. To prevent oxidation, 2 mM sodium dithionite was added to the protein prior to crystallization. All crystallization trials were set up in a glove box filled with carbon monoxide gas, and all solutions were degassed and purged with carbon monoxide. Crystallization conditions were found using a search procedure based on orthogonal arrays (Kingston et al. 1994). Crystals were grown in the dark, by hanging-drop vapor diffusion at 18°C. Equal volumes of protein solution (40 mg/mL in 0.1 M HEPES buffer (pH 7.0) with 2 mM sodium dithionite) and reservoir solution were mixed, with the best-diffracting crystals (0.2 × 0.2 × 0.2 mm) obtained with a reservoir solution of 0.2 M acetic acid/KOH (pH 4.5–4.9) and 21% MePEG 5000. The best met–azido crystals were similarly obtained at low pH (pH 4.9).

Both the carbonmonoxy and met–azido crystals were orthorhombic, P2₁2₁2, with nearly identical unit-cell parameters a = 61, b = 82, and c = 53 Å (Table 4), and differed from those originally obtained for the met form of Hb Bart's (Czerwinski et al. 1981). Assuming two globin chains in the asymmetric unit, the crystal density (V_m) is calculated to be 2.1 ų Da⁻¹ with a solvent content of 41% (v/v) for both structures (Matthews 1968). A data set to 1.86 Å resolution was collected for a flash-frozen crystal of the met–azido form as described in Kidd et al. (2001). For the carbonmonoxy form, 30% glycerol was added to the reservoir solution as a cryoprotectant. X-ray data to 1.7 Å were collected at 110 K using a MAR Research 345 image plate detector with synchrotron radiation from beamline 7–1 at the Stanford Synchrotron Radiation Laboratory. Data were collected as series of 0.5° oscillation frames, each of 10-s exposure, at a crystal to detector distance of 200 mm. Image data were processed and scaled using the DENZO and SCALEPACK (Otwinowski and Minor 1997) software. Data collection statistics for both crystal forms are summarized in Table 4.

Table 4.

Crystal and data collection statistics

	CO-γ4	NNN-γ4
Space group	P21212	P21212
Cell dimensions:
a (Å)	60.62	60.54
b (Å)	82.84	81.63
c (Å)	53.58	53.04
Reflections measured	105,161	156,828
Unique reflections	27,848	22,691
Maximum resolution (Å)	1.7	1.86
Completenessa (%)	91.5 (98.5)	99.7 (98.3)
Multiplicitya	3.8 (3.7)	6.9 (5.9)
Rmergea,b (%)	7.4 (28.3)	6.4 (26.1)
MeanaI/σI	8.7 (3.0)	14.1 (6.9)

^a Figures in parentheses are for the outermost resolution shell: 1.76–1.70 Å for CO-γ₄, and 1.93–1.86 Å for NNN-γ₄.

^bR_merge = ∑_h∑_i|I_h,i − <I_h>|/∑_h∑_iI_h,i.

Structure determination and refinement

Both structures were solved by molecular replacement. Because the crystals grew from solutions of either CO– or NNN–ζ₂γ₂, we assumed that a ζγ dimer was present in the crystal asymmetric unit. An αγ dimer from the 2.5 Å structure of deoxy fetal Hb (PDB code 1FDH; Frier and Perutz 1977), with side chains that differed between α and ζ truncated to Ala or Gly, was therefore taken as an initial search model. Molecular replacement using AMoRe (Navaza 1994), with data in the resolution range 10.0 to 3.5 Å failed to give a clear rotation solution, however. Assuming the failure was due to use of a T-state, unliganded, Hb as search model, we next used a liganded αβ dimer from the 1.7 Å structure of carbonmonoxy adult Hb (PDB code 1BBB; Silva et al. 1992) with side chains that differed between α and ζ chains and between β and γ chains truncated to their common atoms. Using CNS (Brünger et al. 1999), rotation and translation function calculations based on data between 15 and 4 Å gave a single unambiguous solution. After rigid-body refinement in which each subunit was treated as a separate rigid body the R-factor was 45.4% and R_free was 47.2%. For the R_free calculation (Brünger 1992) in this and subsequent refinement steps, 10% of the reflections were randomly selected. The model was then built into 2F_o–F_c and F_o–F_c electron density maps using O (Jones et al. 1991). The γ-chain side chains were straightforward to build; however, the side chains for a number of the ζ chain residues did not fit the density. Model building from a "generic" globin poly-alanine dimer model yielded the same result: 100% of the γ chain residues fit their density but only about 60% of the ζ chain residues fit well. Interestingly, the ζ chain had extra density around residue 50, at the site of a six-residue insertion that is found in β-type but not α-type globins. At this point, we suspected that we had a gamma dimer in the asymmetric unit and had actually crystallized γ₄ rather than ζ₂γ₂. Sequencing via density revealed that over 95% of the structure could be built as gamma. This was confirmed by the automatic model-building program wARP (Lamzin and Wilson 1997). Biochemical analysis (Kidd et al. 2001) then confirmed that the crystals contained only γ chains, implying that we had crystallized γ₄, Hb Bart's.

Molecular replacement calculations using CNS were repeated using a β₂ dimer from the 1.8 Å structure of CO–β₄ (PDB code 1CBM; Borgstahl et al. 1994a), with CO removed and side chains that differed between β and γ truncated to their common atoms. The R-factor after rigid-body refinement was 27.6% and the R_free 27.6%. The structure was refined by alternating cycles of CNS refinement (using data between 25 and 1.7 Å) with model rebuilding using O. Typically, CNS refinement included the following steps (in order): simulated annealing, individual B-factor refinement, water picking, and conjugate gradient minimization refinement. To check that the planar heme and the long Fe-C bond were not artefacts of the refinement protocols, the final stages of refinement were repeated using REFMAC (Murshudov et al. 1997), with no restraints on the Fe-C bond length or Fe-C-O bond angle. This resulted in minimal change. The final R (R_free) values were 21.1% (24.4%).

NNN–γ₄ was built directly from CO–γ₄ (CO atoms removed) and after refinement has a final R (R_free) value of 18.4% (22.0%). The refinement protocol followed that for CO–γ₄. Full refinement statistics for both structures are given in Table 1. Structural superpositions were carried out with the program LSQKAB from the CCP4 program suite (Collaborative Computing Project Number 4 1994), and the quality of the final model was analyzed with PROCHECK (Laskowski et al. 1993). RCSB Protein Data Bank accession numbers for CO–γ₄ and NNN–γ₄ are 1I3D and 1I3E, respectively.

Abbreviations

• Hb, hemoglobin

• HbH, β₄ homotetramer

• Hb Bart's, γ₄ homotetramer

• CO-γ₄, carbonmonoxy γ4 Hb

• NNN-γ₄, met-azide γ₄ Hb

• CO-β₄, carbonmonoxy β₄ Hb

• HbA, adult Hb (α₂β₂)

• HbF, fetal Hb (α₂γ₂)

• Hb Portland, ζ₂γ₂

• MePEG, methoxypolyethylene glycol

• rms, root-mean-square