Sequencing a protein by x-ray crystallography. II. Refinement of yeast hexokinase B co-ordinates and sequence at 2.1 A resolution.

Although the amino acid sequence of yeast hexokinase B has not been determined by chemical means, crystallographic refinement of the hexokinase monomer was carried out at 2.1 Å resolution to improve both the atomic co-ordinates and the amino acid sequence, which had been obtained from a 2.5 Å electron density map. The atomic co-ordinates were adjusted by real-space refinement into a multiple isomorphous replacement map, followed by automated difference Fourier refinement, and restrained parameter structure factor least-squares refinement. The amino acid sequence was altered periodically after visual inspection of (F_o ? F_c) difference electron density maps. Evidence of the improvement in the amino acid sequence was provided by the better agreement between the X-ray and chemically derived amino acid compositions, and most importantly by the ability to locate two short peptides which had been chemically sequenced. While only 6 out of the 18 residues in these two peptides agree with the sequence of the original model, 12 residues agree with the sequence of the refined model and the others differ by only an atom or two. The refined model contains 3293 of of the 3596 non-hydrogen atoms expected from the amino acid composition and 152 bound water molecules. The crystallographic R factor at 2.1 Å is 0.25.We show that there are several advantages to refining the structure of even a protein of unknown sequence. (1) Improved phases can be obtained to the resolution limit of the diffraction pattern starting with a model derived from a 2.5 Å map. (2) The accuracy of the amino acid sequence derived by X-ray methods alone can be substantially improved. (3) Functionally important residues can be identified before chemical sequence information is available. (4) The improved X-ray sequence should greatly reduce the effort required to obtain a chemical sequence; since peptides as short as eight or nine residues can be located in the refined X-ray sequence, peptides do not need to be overlapped by chemical means.     

Atomic Structure of the Degraded Procapsid Particle of the Bacteriophage G4: Induced Structural Changes in the Presence of Calcium Ions and Functional Implications

Bacteriophage G4 and φX174 are members of the Microviridae family. The degree of similarity of the structural proteins ranges from 66% identity of the F protein to 40% identity of the G protein. The atomic structure of the φX174 virion had previously been determined by X-ray crystallography. Bacteriophage G4 procapsids, consisting of the structural proteins F, G, D, B, H, and small traces of J but no DNA, were set up for crystallization. However, the resultant crystals were of degraded procapsid particles, which had lost the assembly scaffolding proteins D and B, resulting in particles that resembled empty virions.The structure of the degraded G4 procapsid has been determined to 3.0 Å resolution. The particles crystallized in the hexagonal space groupP6₃22 with unit cell dimensionsa=b=414.2(5) Å andc=263.0(3) Å. The diffraction data were collected at the Cornell High Energy Synchrotron Source (CHESS) on film and image plates using oscillation photography. Packing considerations indicated there were two particles per unit cell. A self-rotation function confirmed that the particles were positioned on 32 point group special positions in the unit cell. Initial phases were calculated to 6 Å resolution, based on the known φX174 virion model. Phase information was then extended in steps to 3.0 Å resolution by molecular replacement electron density modification and particle envelope generation.The resulting electron density map was readily interpretable in terms of the F and G polypeptides, as occur in the mature capsid of φX174. In a few regions of the electron density map there were inconsistencies between the density and the published amino acid sequence. Redetermining the amino acid sequence confirmed that the density was correct. The r.m.s. deviation between the C^αbackbone of the mature capsid of φX174 and the degraded G4 procapsid was 0.36 Å for the F protein and 1.38 Å for the G protein. This is consistent with the greater conservation of the F protein compared to the G protein sequences among members of the Microviridae family. Functionally important features between φX174 and G4 had greater conservation.Calcium ions (Ca^{2 +}) were shown to bind to G4 at a general site located near the icosahedral 3-fold axis on the F protein capsid, equivalent to sites found previously in φX174. Binding of Ca^{2 +}also caused the ordering of the conserved region of the DNA binding protein J, which was present in the degraded procapsid particle in the absence of DNA.     

Structure determination of crystalline lobster D-glyceraldehyde-3-phosphate dehydrogenase

Single crystal X-ray data were collected on film for the holoenzyme of lobster d-glyceraldehyde-3-phosphate dehydrogenase to 3·0 Å resolution. Films of potassium tetraiodomercurate, K₂HgI₄, comprising a complete low resolution set, with some additional high resolution terms, were given to us by Drs H. C. Watson and L. J. Banaszak. A 3·0 Å high resolution data set was collected of a p-chloromercuri-phenylsulfonate derivative. All these films were processed on a computer controlled Optronics film scanner. The K₂HgI₄ derivative difference Patterson was initially interpreted in terms of four single sites, one for each polypeptide chain, consistent with the previously determined molecular 222 symmetry. Single isomorphous replacement phases were then sufficient to identify other heavy atom sites. Least-squares refined parameters were used to give multiple isomorphous replacement phases at low resolution, and single isomorphous replacement phases at high resolution. The resultant electron density map was oriented along the molecular 2-fold axes and then averaged over all four equivalent subunits. This process produced a much improved electron density map, which could easily be interpreted in terms of a single polypeptide chain per subunit consistent with the known amino acid sequence. The use of non-crystallographic symmetry to improve the electron density map is equivalent to the molecular replacement method. A comparison is also made with other dehydrogenases.     

A crystallographic model for azurin a 3 A resolution.

The structure of the blue copper protein azurin (M_r 14,000) from Pseudomonas aeruginosa has been determined from a 3.0 Å resolution electron density map computed with phases based on a uranyl derivative to 3 Å resolution and a platinum derivative to 3.7 Å. Interpretation of the somewhat noisy map was based on comparison of the density of the four molecules in the asymmetric unit with their averaged density. The polypeptide chain folds into an eight-strand β barrel with an additional flap containing a short helix. The copper atom is bound at one end and on the inside of the barrel, probably to a cysteine, a methionine, and two histidine residues.     

Structure of arthropod hemocyanin

Hemocyanins are large multi-subunit copper proteins that transport oxygen in many arthropods and molluscs. The amino acid sequence of subunit a of Panulirus interruptus hemocyanin (657 residues) has been completed and fitted to the electron-density map (3.2 Å resolution). Comparison of amino acid sequence data for several different hemocyanin subunits of arthropod species indicated that the general features of the polypeptide architecture as found in spiny lobster hemocyanin occur in all arthropods. This structure must therefore be at least as old as the estimated time of divergence of crustaceans and chelicerates, 540–600 million years ago.     

Structure of 2-keto-3-deoxy-6-phosphogluconate aldolase at 2.8 Å resolution

The structure of 2-keto-3-deoxy-6-phosphogluconate aldolase has been extended to 2.8 Å resolution from 3.5 Å resolution by multiple isomorphous replacement methods using three heavy-atom derivatives and anomalous Bijvoet differences to 6 Å resolution (〈m〉 = 0.72). The replacement phases were improved and refined by electron density modification procedures coupled with inverse transform phase angle calculations. A Kendrew model of the molecule was built, which contained all 225 residues of a recently determined amino acid sequence, whereas only 173 were accounted for at 3.5 Å resolution. The missing residues were found to be part of the interior of the molecule and not simply an appendage. The molecule folds to form an eight-strand α/β-barrel structure strikingly similar to triosephosphate isomerase, the A-domain of pyruvate kinase and Taka amylase. With a knowledge of the sequence, the nature of the interfaces of the two kinds of crystallographic trimers have been examined, from which it was concluded that the choice of trimers selected in the 3.5 Å resolution work was probably correct for trimers in solution. The active site region has been established from the position of the Schiff base forming Lys144 but it has not been possible to confirm it conclusively in independent derivative experiments. An apparent anomaly exists in the location of Glu56 (about 25 Å from Lys144). The latter has been reported to assist in catalysis.     

Predominant end-products of prophage Mu DNA transposition during the lytic cycle are replicon fusions

The structure of l-arabinose-binding protein (M_r 33, 100), an essential component of the osmotic shock-sensitive, high-affinity l-arabinose transport system in Escherichia coli, has been determined at 2.4 Å resolution. The phases were solved by the method of multiple isomorphous replacement, using four derivatives, p-chloromercuribenzenesulfonate and CdI₂ (data to 2.4 Å resolution), and p-chloromercurinitrophenol and (NH₄)₂PtCl₄ to 3.5 Å resolution. A final mean figure of merit of 0.65 was obtained for 9628 reflections.With the aid of the amino acid sequence determined by Hogg &; Hermodson (1977), a complete model of the protein molecule has been determined using initially an optical comparator. The entire model was subsequently examined in detail using a computer graphic system.The protein molecule is ellipsoidal (axial ratio of 2:1), and consists of two globular domains (designated P and Q). Each domain is made from two separate polypeptide chain segments. Despite the discontinuity in the folding, the arrangements of the secondary structure in the two domains are very similar. Both domains contain a six-stranded parallel β-sheet (with the exception of the sixth anti-parallel strand in the Q domain) flanked by two α-helices on either side. The packing topology is α/β. A C-terminal helix is shared by both domains.The two domains show significant conformational similarity but lack sequence homology. A comparison of the two domains revealed that of the 139 α-carbons in the P domain and 152 in the Q domain, 92 were found to be equivalent with a root-mean-square distance of 2.6 Å.The cleft formed by the packing of the two domains is predominantly lined with hydrophilic residues. The sugar-binding site is located in this cleft.     

Molybdenum active centre of DMSO reductase from Rhodobacter capsulatus: crystal structure of the oxidised enzyme at 1.82-Å resolution and the dithionite-reduced enzyme at 2.8-Å resolution

The 1.82-Å X-ray crystal structure of the oxidised (Mo^(VI)) form of the enzyme dimethylsulfoxide reductase (DMSOR) isolated from Rhodobacter capsulatus is presented. The structure has been determined by building a partial model into a multiple isomorphous replacement map and fitting the crystal structure of DMSOR from Rhodobacter sphaeroides to the partial model. The enzyme structure has been refined, at 1.82-Å resolution, to an R factor of 14.8% (R _free?=?18.4%). The molybdenum is coordinated by seven ligands: four dithiolene sulfurs, Oγ of Ser147 and two oxo groups. The four sulfur ligands, at a metal-sulfur distance of 2.4?Å or 2.5?Å, are contributed by the two molybdopterin guanine dinucleotide (MGD) cofactors. The coordination sphere of the molybdenum is different from that in previously reported structures of DMSOR from R. sphaeroides and R. capsulatus. The 2.8-Å structure of DMSOR, reduced by addition of sodium dithionite, is also described and differs from the structure of the oxidised enzyme by the removal of a single oxo ligand from the molybdenum coordination sphere. A structure, at 2.5-Å resolution, has also been obtained from crystals soaked in mother liquor buffered at pH?7.0. No differences are observed in the structure at pH?7 when compared with the native crystal structure at pH?5.5.     

Tandem genetic duplications in Salmonella typhimurium: amplification of the histidine operon.

Bovine pancreatic phospholipase A₂ (M_r = 14,000) has been crystallized and its three-dimensional structure determined by X-ray diffraction analysis to a resolution of 2.4 Å. Three heavy-atom derivatives were used in the phase calculations with inclusion of the anomalous dispersion differences. The resulting electron density map allowed an easy and unambiguous tracing of the peptide chain. Two of the seven disulfide connections appeared to be different from what was suggested by the earlier chemical and structural work. The bovine phospholipase A₂ structure contains about 50% α-helix and 10% β-structure. The bovine enzyme structure was found to deviate substantially from the previously published porcine prophospholipase structure.     

The structure of glycogen phosphorylase alpha at 2.5 A resolution

The structure of the glucose-inhibited form of glycogen phosphorylase a has been determined at a resolution of 2.5 Å. With the aid of the primary sequence derived by Titani et al. (1977) for this enzyme, we have constructed an atomic model of the 97,400 molecular weight monomer. A substantial improvement in the electron density map over that reported previously (Fletterick et al., 1976b) was achieved by extension of the data set to 2.5 Å and the inclusion of three additional “heavy-atom” derivatives in the phasing procedure. Main-chain and side-chain electron density are clearly resolved in the map, allowing an unambiguous correlation with the published primary structure. The course of the polypeptide backbone in the C-terminal half of the molecule has been modified at two positions from that reported in the 3.0 Å resolution interpretation.The enzyme is clearly organized into two domains, both with $\text{α}\text{β}$ packing topology. The catalytic site lies in a crevice at the interface between the two domains. α-d-Glucose, which stabilizes the inactive (T) conformation in the parent crystal, is bound at this site in the C_(6′) chair equatorial conformation within 6 Å of the pyridoxal phosphate coenzyme which is covalently bound through the ?-amino group of lysine 679.The larger, N-terminal domain is differentiated by folding architecture and tertiary contacts into two lobes or subdomains which share the same β-sheet backbone: the predominantly helical glycogen storage (maltoheptaose binding) lobe and the N-terminal subdomain. The latter is involved in a variety of protein-protein interactions with the monomer related by the 2-fold axis of the physiological dimer, and contains the serine 14-phosphate moiety and the AMP (positive effector) binding site. The core of the second domain is the complex (βαβ)′ folding unit previously characterized as the nucleotide binding fold (Rao &; Rossmann, 1973).     

Structure determination of feline panleukopenia virus empty particles

Various crystal forms of the single-stranded DNA, feline panleukopenia virus (FPV), a parvovirus, have been grown of both full virions and empty particles. The structure of empty particles crystallized in an orthorhombic space group P2₁2₁2₁, with unit cell dimensions a = 380.1 Å, b = 379.3 Å, and c = 350.9 Å, has been determined to 3.3 Å resolution. The data were collected using oscillation photography with synchrotron radiation. The orientations of the empty capsids in the unit cell were determined using a self-rotation function and their positions were obtained with an R-factor search using canine parvovirus (CPV) as a model. Phases were then calculated, based on the CPV model, to 6.0 Å resolution and gradually extended to 3.3 Å resolution by molecular replacement electron density averaging. The resultant electron density was readily interpreted in terms of the known amino acid sequence. The structure is contrasted to that of CPV in terms of host range, neutralization by antibodies, hemagglutination properties, and binding of genomic DNA. © Wiley-Liss, Inc.     

Crystal structure of Bence Jones protein rhe (3 A) and its unique domain-domain association.

The crystal structure of protein Rhe, a lambda type V_L dimer, has been determined at a resolution of 3 Å by the method of multiple isomorphous replacement supplemented with anomalous scattering data. A crystallographic sequence was assigned from an interpretation of the electron density map in an optical comparator and is compared with a chemically determined partial amino acid sequence. The monomeric unit of Rhe, as determined crystallographically, contains 113 amino acids, 110 belonging to the variable region and three belonging to the constant segment of a light chain. The single polypeptide chain constituting the monomer forms a nine-stranded β-barrel characteristic of V domains. The β-pleated sheet surrounds an ellipsoidally shaped interior core of approximately 10 Å × 15 Å × 25 Å in size. The monomers that are related by the crystallographic dyad are held together as dimers by interdomain hydrogen bonds and hydrophobic interactions. At one end of the dimer is an opening which is lined exclusively by residues from the hypervariable regions.A comparison of Rhe with Rei, a kappa type V_L dimer (Epp et al., 1975), revealed that monomers of Rhe and Rei dimerized differently. Their respective dyad and pseudodyad of dimerization are not the same, and this causes a variation in the overall steric arrangement of the hypervariable regions in the two cavities. In adition a dissimilarity was observed in the non-hypervariable segment linking the first and second hypervariable regions. This segment is in the form of a loop and it includes most of the residues participating in the interdomain interactions stabilizing dimer formation in both proteins and these loop positions differ by as much as 7 Å. Our results also show that there is a good correlation between the dissimilarity of the loop position and the difference in the domain-association. Our preliminary analysis indicates that the positions of the corresponding non-hypervariable loops in V domains may be determined in part by the residues in the hypervariable regions.Accordingly, the three-dimensional structure of Rhe suggests that this nonhypervariable loop in V_L and its counterpart in V_H may have an important biological function in antibody specificity and variability by virtue of their influence over the architecture of the complementarity site.     

X-ray crystallographic studies of seal myoglobin: The molecule at 2.5 Å resolution

The three-dimensional structure of metmyoglobin from the common seal has been determined at 2.5 Å resolution. The isomorphous replacement technique has been employed using two derivatives, the mercuri-iodide and the aurichloride. Four-circle diffractometer data to a Bragg angle θ = 18.05 ° were measured for one complete set of Friedel pairs of reflexions from each type of protein crystal. Atomic positions for the individual atoms in the (HgI^?₃₎-group at the two sites of attachment were obtained from three-dimensional difference electron density maps and were further refined. A ‘best’ electron density map of the native protein based on refined heavy-atom parameters was interpreted with the help of the known amino acid sequence, and co-ordinates for all the non-hydrogen atoms were measured from the model. Those of the globin were further constrained according to the ‘modelfit’ procedure of Dodson et al. (1976). The molecule is described in detail; the conformations of the side-chains relative to the positions of the heavy atoms and to the interface between neighbouring molecules are discussed. A preliminary residue by residue comparison of the seal and sperm whale myoglobin molecules is presented in the accompanying paper.     

Crystal structure of Turkey egg-white lysozyme: Results of the molecular replacement method at 5 Å resolution

Turkey egg-white lysozyme differs from hen egg-white lysozyme in its primary structure in 7 of the 129 residues. We have determined the rotational and translational parameters relating the known co-ordinates of hen egg-white lysozyme molecule to the turkey lysozyme. The rotational parameters were determined using the rotation function, the translational parameters were determined by placing the properly rotated molecule systematically at all positions within the unit cell and searching for those positions producing few intermolecular contacts between the α-carbon atoms of one molecule and all its neighbors. These parameters were refined by minimizing the conventional R factor between observed and calculated structure amplitudes. The final rotational and translational parameters give an R value of 46.7% for reflections with d spacings between 6 Å and 12 Å and have 7 intermolecular contacts closer than 5 Å between the a carbon atoms of one molecule and all its neighbors. An electron density map has been calculated at 5 Å resolution; the packing of the molecules in this form appears to present the entire length of the active cleft in the vicinity of the crystallographic 6-fold axis and does not appear to be blocked by neighboring molecules.     

X-ray structure and characterization of a thermostable lipase from Geobacillus thermoleovorans

Thermo-alkalophilic bacterium, Geobacillus thermoleovorans secrets many enzymes including a 43?kDa extracellular lipase. Significant thermostability, organic solvent stability and wide substrate preferences for hydrolysis drew our attention to solve its structure by crystallography. The structure was solved by molecular replacement method and refined up to 2.14?Å resolution. Structure of the lipase showed an alpha-beta fold with 19 α-helices and 10 β-sheets. The active site remains covered by a lid. One calcium and one zinc atom was found in the crystal. The structure showed a major difference (rmsd 5.6?Å) from its closest homolog in the amino acid region 191 to 203. Thermal unfolding of the lipase showed that the lipase is highly stable with T_m of 76?°C. ¹³C NMR spectra of products upon triglyceride hydrolysate revealed that the lipase hydrolyses at both sn-1 and sn-2 positions with equal efficiency.     

Crystal structure of cat muscle pyruvate kinase at a resolution of 2.6 A

The structure of pyruvate kinase (EC 2.7.1.40) has been determined from a 2.6 Å resolution electron density map. This map shows more detail than the previous 3.1 Å map (Stammers &; Muirhead, 1977) and has enabled a detailed chain folding to be established for two out of the three domains which make up each of the four identical subunits. A provisional chain folding has been established for the third domain. The results have been briefly reported in a previous paper (Levine et al., 1978). Details of the structure determination and a further discussion of the results are presented in this paper.Domain A (the three domains of pyruvate kinase are referred to as A, B and C) can be described in terms of a cylindrical eight-stranded parallel β sheet and an outer coaxial cylinder of eight α helices. The α helices connect adjacent strands of the β sheet. Domain B is made up of a closed anti-parallel β sheet structure. Domain C is a five-stranded β sheet of which the fourth strand is anti-parallel and the rest parallel. These strands are also interconnected by α helices.Domain A can be dissected into eight consecutive β strand—α helix units starting from the N-terminus. The arrangement of these relative to each other can be most simply described by relating them to eight planes, each at 40 ° to the cylinder axis and symmetrically placed around the cylinder. When unit 2 is aligned with one of these planes then units 1, 3, 4, 5 and 8 are also closely aligned with a plane. This analysis is also applied to triosephosphate isomerase and a strikingly similar arrangement is found. A detailed comparison of the two structures is presented. Although the lack of a chemical sequence makes it difficult to identify the amino acid residues of pyruvate kinase, side-chains are clearly visible in the map and this information is correlated with the results of previous 6 Å substrate soaking experiments and with the structure of triosephosphate isomerase. The similarities and differences are discussed in terms of similarities and differences in the reactions catalysed and also of different subunit packing.     

Crystallographic study of turkey egg-white lysozyme and its complex with a disaccharide

The crystal structure of turkey egg-white lysozyme, determined by the molecular replacement method at 5 Å resolution (Bott & Sarma, 1976) has now been refined to 2.8 Å resolution and a model has been built to fit the electron density. A comparison of the co-ordinates with those of hen lysozyme indicate a rootmean-square deviation of 1.6 Å for all the main-chain and side-chain atoms. A significant difference is observed in the region of residues 98 to 115 of the structure. The molecules are packed in this crystal form with the entire length of the active cleft positioned in the vicinity of the crystallographic 6-fold axis and is not blocked by neighboring molecules. A difference electron density map calculated between crystals of turkey lysozyme soaked in a disaccharide of N-acetyl glucosamine—N-acetyl muramic acid and the native crystals showed a strong positive peak at subsite C, a weak positive peak at subsite D and two strong peaks that correspond to the subsite E and a new subsite F′. This new site F′ is different from the subsite F predicted for the sixth saccharide from model building in hen lysozyme. The interactions between the saccharides bound at subsites E and F′ and the enzyme molecules are discussed.     

Structural characterization of the putative ABC-type 2 transporter from Thermotoga maritima MSB8

This study describes the structure of the putative ABC-type 2 transporter TM0543 from Thermotoga maritima MSB8 determined at a resolution of 2.3 Å. In comparative sequence-clustering analysis, TM0543 displays similarity to NatAB-like proteins, which are components of the ABC-type Na⁺ efflux pump permease. However, the overall structure fold of the predicted nucleotide-binding domain reveals that it is different from any known structure of ABC-type efflux transporters solved to date. The structure of the putative TM0543 domain also exhibits different dimer architecture and topology of its presumed ATP binding pocket, which may indicate that it does not bind nucleotide at all. Structural analysis of calcium ion binding sites found at the interface between TM0543 dimer subunits suggests that protein may be involved in ion-transporting activity. A detailed analysis of the protein sequence and structure is presented and discussed.     

Structure of cycloheptaamylose inclusion-complexes: crystal structure of substituted benzoic acid and phenol derivatives

Cycloheptaamylose has been crystallized with 2,5-diiodobenzoic acid as guest. The X-ray crystal structure at 1.2-Å resolution with space group C2 and cell dimensions a  19.192 (13), b  24.759 (20), c  15.739 (13) Å, and β  109.6 (3)° was solved by using rotation-translation functions. Complexes of other meta-substituted guests were found to be isomorphous, and were solved by using the phases of the cycloamylose of the 2,5-diiodobenzoic acid complex. The complex with 2-bromo-5-tert-butylphenol having a  19.235 (11), b  24.662 (17), c  16.018 (11) Å, and β  108.9 (2)° was determined at 1.0 Å resolution, and the complexes with m-bromobenzoic acid, m-iodobenzoic acid, m-iodophenol, m-toluic acid, and 2-bromo-4-tert-butylphenol were determined at 2.0-Å resolution. In all cases, the guest molecule was disordered. However, by using information from all the structures, it may be concluded that the functionally important carboxylic acid group lies in the primary-hydroxyl end of the cycloheptaamylose molecule. As studies in solution have shown that the hydrogen-bonding groups of guest molecules interact with the secondary-hydroxyl end of the cycloheptaamylose molecule, it is concluded that the structure seen in the crystals here does not correspond to a catalytically active species. Cyclo-heptaamylose exists as a dimer in the crystal by means of extensive hydrogen bonding across the secondary-hydroxyl ends of two cycloheptaamylose molecules. A continuous channel throughout the crystal is achieved by the stacking of these dimer units.     

The structure of horse methaemoglobin at 2-0 A resolution

The structure of horse methaemoglobin has been redetermined by phase extension and refinement. This has improved our knowledge of the haem geometry and the stereochemistry of the interfaces between the subunits, and confirmed the disorder of the C-terminal residues. Using new four-circle diffractometer data between the limiting spheres of radius 10 and 2.0 Å^?1, the co-ordinates determined by Perutz et al. (1968a,b) were subjected to successive cycles of real-space refinement into electron density maps calculated with observed ¦F¦ values and phases derived from the latest refined model, until the reliability index had dropped from an initial value of 0.45 to 0.23. The positions of the iron atoms relative to the planes of the porphyrin rings were refined separately, and checked by Fourier syntheses based on anomalous scattering and by difference Fourier syntheses calculated with coefficients from which the iron contributions had been removed. The general root-mean-squared error in atomic positions is 0.32 Å; the probable error in the displacement of the iron atoms from the porphyrin planes is 0.06 Å. The difference Fourier synthesis, obtained after refinement of the protein was complete, showed 41 bound water molecules per asymmetric unit and also revealed five errors in amino acid sequence, one of which was confirmed chemically.The secondary structures of the subunits are stabilized by hydrogen bonds formed by main-chain NH and CO groups either with each other or with nearby polar side-chains. There are few internal hydrogen bonds linking the various chain segments; many of the external polar side-chains help to stabilize the tertiary structure by forming hydrogen bonds with each other or through bound water molecules. Several of the helical segments are irregular and the terminal residues are disordered. The contacts between the subunits are more polar than the earlier 2.8 Å map had led us to believe, because it had failed to show up the 15 bound water molecules at the α₁β₁ and the four at the α₁β₂ contact. Their inclusion has raised the number of hydrogen bonds between neighbouring subunits at α₁β₁ from five to 17 or possibly 19, and at α₁β₂ from two to six or possibly seven. The remaining 22 water molecules are distributed over the internal cavity and the molecular surface; most of them make hydrogen bonds with at least two polar groups of the protein. Despite several amino acid differences, the structure of the α₁β₁ contact, including the bound water, is the same as in human deoxyhaemoglobin (Fermi, 1975).