Solution and crystal structures of a sperm whale myoglobin triple mutant that mimics the sulfide-binding hemoglobin from Lucina pectinata.

The bivalve mollusc Lucina pectinata harbors sulfide-oxidizing chemoautotrophic bacteria and expresses a monomeric hemoglobin I, HbI, with normal O2, but extraordinarily high sulfide affinity. The crystal structure of aquomet Lucina HbI has revealed an active site with three residues not commonly found in vertebrate globins: Phe(B10), Gln(E7), and Phe(E11) (Rizzi, M., Wittenberg, J. B., Coda, A., Fasano, M., Ascenzi, P., and Bolognesi, M. (1994) J. Mol. Biol. 244, 86-89). Engineering these three residues into sperm whale myoglobin results in a triple mutant with approximately 700-fold higher sulfide affinity than for wild-type. The single crystal x-ray structure of the aquomet derivative of the myoglobin triple mutant and the solution 1H NMR active site structures of the cyanomet derivatives of both the myoglobin mutant and Lucina HbI have been determined to examine further the structural origin of their unusually high sulfide affinities. The major differences in the distal pocket is that in the aquomet form the carbonyl of Gln64(E7) serves as a H-bond acceptor, whereas in the cyanomet form the amido group acts as H-bond donor to the bound ligand. Phe68(E11) is rotated approximately 90 degrees about chi2 and located approximately 1-2 A closer to the iron atom in the myoglobin triple mutant relative to its conformation in Lucina HbI. The change in orientation potentially eliminates the stabilizing interaction with sulfide and, together with the decrease in size of the distal pocket, accounts for the 7-fold lower sulfide affinity of the myoglobin mutant compared with that of Lucina HbI.

Despite substantial amino acid sequence variability, the reversible binding of molecular oxygen in myoglobin (Mb) 1 and hemoglobin (Hb) is achieved by a surprisingly invariant protein folding topology: a heme group imbedded within 7-8 packed ␣-helices (1,2). Examination of the more than 300 known sequences of vertebrate Hbs and Mbs and the Ͼ130 nonvertebrate globin-like sequences demonstrates that only two residues appear to be conserved, the proximal His at helical position F-8 and the Phe parallel to the heme surface at an interhelical loop, position CD1 (1,3). The residue considered to be pivotal in the stabilization of the bound O 2 is the one whose side chain can provide a neutral donor for a hydrogen bond with the ligand, such as His or Gln at the distal position E-7. Among vertebrate Mbs and Hbs, His predominates, with Gln occurring only in elephant Mb (4), shark Mb (5), hagfish Hb (6), and the ␣ chain of opossum Hb (7). Although His is also found in the majority of nonvertebrate Hbs and Mbs, Gln is not uncommon and, in addition, other residues may occupy the E-7 position, e.g. Val in Aplysia (8) and several other mollusc Mbs (9), Tyr in the Mb from Paramphistomun epiclitum (10,11), and Leu in the monomeric components of Glycera Hb (12). Although capable of O 2 binding, a number of the variant globins exhibit some anomalous functional properties. Parallel functional and structural studies on genetically engineered mutants of mammalian Mb have shown that changes in only a very limited number of residues in the distal pocket often transfers the essence of the unusual behavior of the natural genetic variant to the appropriate reference protein mutant (13)(14)(15)(16)(17).
Cytoplasmic Hbs occur in many symbiotic associations between bacteria and invertebrates or plants and may have a physiological role in the symbiosis (18 -20). The bivalve mollusc Lucina pectinata, found in sulfide-rich coastal sediments, harbors sulfide-oxidizing chemoautotrophic bacteria (21,22). Its abundant cytoplasmic hemoglobins consist of three singlechain components, each having a moderately high affinity for oxygen (P 50 ϭ 0.1-0.2 torr), which is achieved, however, by a very different balance of combination and dissociation rates. Reactions of the monomeric HbI with oxygen are rapid, whereas those of HbII and HbIII are extraordinarily slow (23).
Recent crystal structures of Lucina aquometHbI and the sulfide bound form (24,25) have revealed a distal Gln 64 (E7) and Phe 29 (B10), neither of which is rare among invertebrate Mbs, Hbs. However, the E-11 position is uniquely occupied by Phe rather than the aliphatic amino acids Val, Ile, or Leu. To examine the structural origin of strong sulfide binding in Lucina HbI, we have systematically replaced His 64 (E7), Leu 29 (B10), and Val 64 (E11) with Gln, Phe, and Phe, respectively, in sperm whale myoglobin and then determined the structures of the resulting multiple mutants by NMR and x-ray crystallography. This approach is similar to that used to study the unusual functional properties of Aplysia Mb, Ascaris Hb, and elephant Mb, and provides comparison between the native protein and the synthetic mimic in two ligation/spin states and between protein structures in crystal and solution (13-17, 26, 27). The NMR studies were pursued to define more carefully the positions of the labile protons of Gln 64 (E7) and their role as hydrogen bond donors to the bound ligand. Both the paramagnetic relaxation and induced dipolar shifts in metMbCN and metHbICN can be used to model the position of distal residues (14 -17, 26, 29, 30). The combined NMR and crystallographic results reveal systematic differences between the orientation of Gln 64 (E7) in the aquomet and cyanomet complexes of Lucina HbI and L29F/H64Q/V68F-Mb, which demonstrate that Gln 64 (E7) may serve as either a H-bond acceptor or donor to bound ligands and that a difference in the orientation of the Phe 68 (E11) ring may account for the difference in sulfide affinity between Lucina HbI and the sperm whale triple mutant Mb mimic.

EXPERIMENTAL PROCEDURES
Protein Preparation-The pUC19 plasmid containing V68F sperm whale myoglobin was made by Egeberg et al. (31). The plasmid containing L29F/H64Q/V68F sperm whale Mb was constructed from pEMBL19 containing L29F Mb using cassette mutagenesis as described in Springer et al. (32). Vectors harboring the mutated gene were transformed into Escherichia coli strain TB1 and expressed constitutively using a 100-liter fermentor. The resulting soluble holomyoglobin protein was purified as described previously (33,34).
Protein Crystallization-Crystals of the recombinant L29F/H64Q/ V68F-Mb were grown by the batch method (35,36) in a temperaturecontrolled environment (17°C). Large single crystals developed out of solutions ranging from 2.4 to 2.7 M ammonium sulfate buffered at pH 9 (20 mM Tris-HCl, 1 mM EDTA).
X-ray Data Collection and Structure Determination-Crystals were hexagonal, space group P6, with one molecule per asymmetric unit as originally solved by Phillips et al. (35). They were mounted in sealed quartz capillary tubes prior to data collection. A single data set was collected from one crystal at room temperature using a Rigaku R-axis IIc imaging plate system and copper K␣ radiation from a Siemens rotating anode operated at 50 mV and 90 mA. The unit cell dimensions of the L29F/H64Q/V68F-metMbH 2 O were a ϭ 91.39, c ϭ 45.75. A total of 74,100 measured reflections from 60 images (⌬ ϭ 1.5°) with an R merge of 5.7% for all data were reduced to 15,526 unique intensities (5.0 to 1.85 Å, 87.3% complete) using the program XDS.
Crystal Structure Refinement-Starting coordinates were generated using the L29F/H64Q-MbCO (Protein Data Bank entry 1MCY, Brookhaven National Laboratory) (17) to calculate initial phases. The coordinates for the additional mutated residue (V68F) were built from an electron density omit map. Solvent atoms were introduced in positive density peaks over four . Cycles of conventional positional refinement were carried out with XPLOR (37-39) alternated with manual fitting using the CHAIN software package (40). The crystal structures were all refined with Engh and Huber (41) topology and parameters with no restraints on the iron atom to remove bias in metal position. The crystallographic refinements converged to an R-factor of 16.8% for the triple mutant metmyoglobin structure with root mean square bond deviations of 0.017 Å. Coordinates have been deposited in the Protein Data Bank (entry 1OBM).
1 H NMR Measurements-All the 1 H NMR spectra were collected on the GE Omega 500 MHz spectrometer. The strongly relaxed signals were optimally detected in water-eliminated Fourier transform spectra (42). Nonselective T 1 s for the resolved strongly relaxed protons were measured via inversion-recovery experiment. Steady state NOEs were recorded as described previously (43). The phase-sensitive TOCSY (44,45), NOESY (46), and conventional magnitude COSY (MCOSY) (47) employed the method described by States et al. (48) to provide quadrature detection in the t 1 dimension. Solvent suppression, when required, was achieved by direct saturation in the relaxation delay period. 512 blocks were collected with 25.0 kHz spectral widths to include all resonances, and 10 kHz to improve resolution for the diamagnetic envelope. 128 to 256 scans were accumulated with repetition rate of 0.7 s Ϫ1 or 1.2 s Ϫ1 for each block with free induction decays of 2048 complex points. The data were processed as described previously (49); details are given in the figure captions. All two-dimensional data were processed on Silicon Graphics workstation using the software package Felix from Biosym/MSI (San Diego).
Magnetic Axes Determination-The magnetic axes were determined as described previously (15,16,29,30,50). Experimental dipolar shifts for the structurally conserved proximal side of the heme were used as input to search for the Euler rotation angles, ⌫(␣,␤,␥), that transforms the molecular pseudo-symmetry coordinates (xЈ, yЈ, zЈ, or R, Ј, ⍀Ј (Fig.  1)) readily obtained from crystal coordinates (17,24,25,51) into magnetic axes, x, y, and z, by minimizing the global error function, ⌬ ax and ⌬ rh are axial and rhombic anisotropies, and ␦ DSS (obs) is the observed chemical shift referenced to DSS. ␦ dia is the shift in the isostructural diamagnetic MbCO complex (52,53), or calculated for protons whose ␦ dia are not available as described in detail by Qin et al. (15,16). Minimizing the error function F/n in Equation 1 was performed over three parameters, ␣, ␤, and ␥, using available ⌬ ax and ⌬ rh , or extended to all five parameters to yield both the Euler angles and anisotropies as described in detail previously (29). Dipolar Shift Simulations-The position of a substituted or perturbed residue can be determined by minimizing a local error function. This local error function, designated F*(residue)/nЈ to distinguish it from that global error function in Equation 1, is given by Ref. 30, where ␦ dip (, ⍀, r) represents the ␦ dip (calc) as a function of a bond rotation or a translation of a residue by a distance r in a direction defined by the angle ⍀, using the magnetic axes derived from conserved structural elements; nЈ is the number of residue dipolar shifts. The bond angle, , or movement defined by ⍀ angle and distance r, that mimize the residual error function F * (residue)/nЈ, defines structural changes as described in detail previously (15)(16)(17)29). When available, the influence of paramagnetic relaxation, T 1 Ϫ1 ϰ R Fe Ϫ6 , allowed estimate of R Fe using the relation, where the His(F8) ring NH (R Fe ϭ 5.0 Å) and heme methyl (R Fe ϭ 6.1 Å) provide upper and lower limits to a proton of interest (54). The molecular modeling was carried out on a Silicon Graphics INDIGO from available crystal coordinates (17,24,25,51) using the INSIGHT II (Biosym/MSI, San Diego) and MIDAS (UCSF) programs.

Hydrogen Sulfide and O 2 Binding to Lucina HbI and Sperm Whale Mb
A brief summary of the affinities of the myoglobin mutants and Lucina HbI for O 2 and hydrogen sulfide 2 is shown in Table  I (17,23,28). Lucina HbI shows a 5,000-fold higher affinity for sulfide than recombinant wild-type sperm whale myoglobin, whereas the two proteins have similar affinities for oxygen. There are progressive increases in K H 2 S as the active site of the whale myoglobin is altered to resemble that of the mollusc hemoglobin. All three single mutations, L29F, H64Q, and V68F produce large, 6 -20-fold increases in sulfide affinity, and the effects on K H 2 S are roughly additive in the double and triple mutants. In contrast, these mutations produce markedly different effects on O 2 affinity. For example, the single H64Q and L29F mutations produces a 5-fold decrease and 15-fold increase, respectively, in O 2 affinity. As a result, K O 2 for the triple mutant is only 3-4-fold higher than wild-type myoglobin and very similar to that of Lucina HbI (Table I).

High Resolution Crystal Structure of Sperm
The three mutated side chains in the recombinant myoglobin do not alter the overall tertiary structure of myoglobin nor do they appear to sterically hinder the bound ligand. Two of these mutated residues, Gln 64 (E7) and Phe 68 (E11), occupy positions equivalent to those observed in the structures of corresponding H64Q-Mb and V68F-Mb single mutants (36,55). In contrast, the aromatic ring at position 29 in the triple mutant is perpendicular to the orientation found in sperm whale L29F-and L29F/H64Q-MbCO ( Fig. 2) (17,56). A detailed comparison of the distal pockets of the L29F/H64Q/V68F-metMbH 2 O and Lucina metHbIH 2 O (24) is presented in Table II and Fig. 2. The overall positions of the three mutated residues in the recombinant sperm whale metmyoglobin are close to those of the corresponding residues in the clam protein, but the exact orientations of the phenyl side chains are significantly different. The distal glutamines in both proteins form hydrogen bonds with the coordinated water molecule.

Heme Pocket Structure of L29F/H64Q/V68F-metMbCN and Lucina metHbICN
Resonance Assignments-The 500 MHz 1 H NMR spectra for L29F/H64Q/V28/F-metMbCN, V68F-metMbCN, and Lucina-metHbICN in 1 H 2 O at 25°C are shown in Fig. 3, A-C, respectively. The signals for the heme for each protein were located and assigned by the characteristic pattern of dipolar contacts about the heme periphery among the TOCSY identified vinyl, propionate groups, and the pyrrole methyl, meso-Hs in a fashion standard for both diamagnetic and paramagnetic heme proteins. The pattern of heme chemical shifts is similar to that observed in wild-type sperm whale metMbCN (not shown). Resonances for residues near the heme are assigned to the degree possible, as limited by spectral congestion and paramagnetic relaxation, and by standard backbone dipolar connectivities together with characteristic TOCSY and/or COSY connectivities for several side chains (15,16,49). Other residues not addressable by these approaches are assigned on the basis of dipolar contacts to the heme and/or to other assigned residues. Inasmuch as this assignment strategy has been reported in detail for both wild-type and mutant metMbCN (17, 29, 30, 43), FIG. 1. Schematic structure of the heme cavity in sperm whale mutant L29F/H64Q/V68F-Mb and Lucina HbI with the proximal (squares) and distal (circles) residues in contact with heme, axial His, and ligand identified by the position on helices. In cases where the residues differ for the two proteins, the residue is labeled first for L29F/ H64Q/V68F-Mb and second for Lucina HbI. The solid lines reflect dipolar (NOESY) contacts observed for both proteins; dotted and dashed lines represent such dipolar contacts observed only in the triple mutant and Lucina HbI, respectively. For all but Gln 64 (E7) in Lucina HbI, the observed dipolar contacts are completely consistent with the crystal structure of their respective aquo-met forms. The xЈ,yЈ,zЈ coordinate system represents an iron-centered coordinate system derived from crystal coordinates of MbCO. The magnetic axes, x, y, z, in which the paramagnetic susceptibility tensor is diagonal, are related by the Euler rotation ⌫(␣, ␤, ␥) via (x, y, z) ϭ (xЈ, yЈ, zЈ)⌫(␣, ␤, ␥), where ␤ represents the tilt of major magnetic axes from the heme normal, ␣ is the direction of this tilt, defined as the angle between the projection of -z on the xЈ,yЈ plane and xЈ axis, and ϳ ␣ ϩ ␥ locates the projection of the rhombic axes onto the heme plane.  (23); wild-type and mutant myoglobins. 2 two-dimensional data are shown only as relevant to the placement of perturbed distal pocket residues in both protein complexes.
The fingerprint portion of the NOESY and COSY maps (not shown) for each protein locates two extended helical fragments each, the shorter one corresponding to F helix residues F4-F9 that includes the strongly hyperfine shifted axial His(F8). The strongly relaxed and hyperfine shifted proximal FG corner residues (His 97 (FG3) and Ile 99 (FG5) in Mb mutants, Arg 99 (FG1) and Ile 101 (FG4) in Lucina HbI) are identified by their characteristic NOESY cross-peaks to the 5-CH 3 . For each protein, the pattern of heme residue and inter residue dipolar contacts and paramagnetic relaxation reveal a proximal side heme pocket structure that is indistinguishable from that in the respective reference crystal structures (Fig. 1). The chemical shifts for relevant assigned resonances for the three protein complexes are listed in Table III.
The longer helical fragment represents the E helix residues E-7-E-14 in each protein, with residues E-10, E-11 and E-14 exhibiting the expected NOESY cross-peaks in the heme, including that of the Phe 68 (E11) ring to the 2-vinyl group. The TOCSY/MCOSY detected fragment NHC ␣ H-C ␤ H for Gln 64 (E7) in each protein shows strong NOESY cross-peaks to a moderately relaxed and hyperfine shifted CH-CH 2 fragment which is assigned to the remainder of Gln 64 (E7); the C ␤ H-C ␤ ЈH TOCSY/ COSY connectivities are not observed because the cross-peak is too close to the diagonal in each protein (Figs. 4 and 5). A strongly relaxed (T 1 ϳ 12 ms) and strongly low-field hyperfine shifted labile proton, when saturated, exhibits an intense NOE to another upfield shifted and moderately relaxed (T 1 ϳ 50 ms) labile proton in Lucina metHbICN (Fig. 3E), but not detected because of overlap with the solvent signals in the triple mutant, as well as to the Gln 64 (E7) C ␥ Hs (and for the triple mutant, to Thr 67 (E10); Fig. 3D) and confirms the two labile protons as arising from the N ⑀ Hs of Gln 64 (E7). The T 1 s via Equation 5 reveal, R Fe (N ⑀ H) ϭ 4.2 Ϯ 0.2 Å for the low-field NH for both proteins, and for Lucina metHbICN, the upfield resolved R Fe (N ⑀ HЈ) ϭ 4.7 Ϯ 0.2 Å. The resulting R Fe are listed in Table  II.
One strongly relaxed (for C H) and hyperfine shifted aromatic ring with NOESY contact to 5-CH 3 and Gln 64 (E7) (not shown) which uniquely identifies Phe 43 (CD1). Another strongly hyperfine shifted and relaxed (C H 19 ppm, T 1 ϳ 33 ms) aromatic ring exhibits NOESY cross-peaks to Phe 68 (E11), Gln 64 (E7), Gly/Ala 65 (E8), and Phe 43 (CD1) and definitively assigns Phe 29 (B10) for each protein (Figs. 4 and 5). The chemical shifts of relevant assigned distal residues for the three proteins are included in Table III. The pattern of dipolar contacts between distal residues and the heme, and among distal residues, are predicted by the crystal structure, as shown schematically in Fig. 1. The difference that are noted, as described below, involve primarily the residues Gln 64 (E7).
Orientation of Distal Residues in L29F/H64Q/V68F-metM-bCN-Using either the wild-type MbCO crystal coordinates with the mutated Phe 29 (B10) inserted with an orientation as found in the crystal structure of L29F/H64Q-MbCO (half-open squares in Fig. 6A), or directly the crystal coordinates of the presently characterized L29F/H64Q/V68F-metHbH 2 O (open squares in Fig. 6A), a qualitative, but not quantitative, correlation between ␦ dip (calc) and ␦ dip (obs) are observed, and the R Fe (C H ϭ 4.7 Å) is shorter than indicated by T 1 ϭ 33 ms (R Fe ϭ 5.1 Ϯ 0.3 Å). However, altering 1 by 5°from that in the L29F/H64Q-metMbCO and a 50°rotation of 2 leads to a Phe 29 (B10) orientation whose ␦ dip (calc) correlate very well with ␦ dip (obs) (closed squares in Fig. 6A) and for which the R Fe (C H) ϭ 5.1 Å is consistent with the T 1 value.
Introduction of Gln 64 (E7) into the wild-type MbCO crystal structure based on its orientation in the H64Q-MbCO (or L29F/ H64Q-MbCO) crystal structure results in reasonable agreement with both T 1 data and ␦ dip (calc) (half-closed circles in Fig.  6A). However, when directly using the crystal coordinates for L29F/H64Q/V68F-metMbH 2

Structure of Sulfide-binding Globins
Gln 64 (E7) in the L29F/H64Q/V68F-metMbH 2 O crystal structure leads to an excellent fit for ␦ dip (obs) versus ␦ dip (calc) (closed circles in Fig. 6A) and yields R Fe (N ⑀ H) ϭ 4.2 Å, in good agreement with the 4.2-Å value obtained from the T 1 ϭ 14 ms via Equation 5. The other N ⑀ H is correctly predicted to resonate close to the solvent signal by the optimized ␦ dip (calc). Similarly small variations in 1 , 2 starting with the Gln 64 (E7) orientation in the single mutant lead to a Gln orientation essentially the same as that obtained starting with the coordinate of the triple mutant (not shown). The anomalous variable temperature slopes (see above), small ␦ dip (calc) which are relatively insensitive to ring orientation, and the presence of comparable magnitude but opposed sign ␦ dip (calc) and ring current shift rendered the optimization of the orientation of Phe 68 (E11) ring on the basis of ␦ dip (obs) completely impractical (50). The pattern of small ␦ dip for Phe 68 (E11), nevertheless, are consistent with observed dipolar shifts and NOESY contacts. The orientation for these three mutated residues in L29F/H64Q/V68F-metMbCN are shown in Fig. 2A, and the relevant side chain bond angles and distances to the iron are listed in Table II Fig. 6B); the correlation is particularly poor for the labile N ⑀ Hs where both models predict even the wrong sign for ␦ dip (calc). Moreover, the low-field N ⑀ H closest to the iron has R Fe ϳ3.5 Å and 5.2 Å in the two crystal structures while the observed T 1 ϭ 12 ms predicts R Fe ϳ 4.2 Ϯ 0.2 Å via Equation 5. Since the Gln 64 (E7) orientation in Lucina metHbIH 2 O had the N ⑀ H oriented toward the ligated water, which is expected to be a H-bond donor, the Gln orientation with 3 rotated by 180°to interchange the carbonyl and amide groups was considered. This altered orientation leads to different N ⑀ H 2 dipolar shifts (not shown) that fit ␦ dip (obs) better, but still unacceptably. Attempts to obtain a fit for ␦ dip (obs) by sequential 1 , 2 , 3 rotation failed since ␦ dip (calc) for C ␤ Hs is quite insensitive to the 1 . Instead, the combination of the large low-field ␦ dip (obs) and R Fe ϳ 4.2 Å (from the T 1 ϭ 12 ms) was used to uniquely locate the low-field N ⑀ H in the crystal coordinates, and a search made for the position in space of the other N ⑀ H (with intra N ⑀ H distance 1.88 Å) to satisfy both the upfield ␦ dip (obs) and R Fe ϳ 4.6 Ϯ 0.2 Å obtained from its T 1 ϭ 22 ms. Upon obtaining a reasonable fit for the N ⑀ Hs that satisfy both relaxation and ␦ dip (obs) constraints, a search was pursued for the range of 1 , 2 , and 3 allowed by the fixed C ␣ H and NH 2 positions. The good correlation between ␦ dip (calc) and ␦ dip (obs) for this optimized Gln orientation is shown in Fig. 6B (closed  circles). The resulting tortional angles for Gln 64 (E7) are listed in Table II where they are compared with values in the crystal structure of Lucina metHbIH 2 O.
The orientation of Phe 29 (B10) in the crystal structures of either Lucina metHbIH 2 O (24) or metHbIH 2 S (25) yield very similar ring proton ␦ dip (calc) that correlate equally well with ␦ dip (obs) (open and closed squares in Fig. 6B, respectively). The crystallographic R Fe (C H) (5.6 Å), however, is somewhat larger than the value reflected in its T 1 ϳ 40 ms (R Fe ϭ 5.2 Ϯ 0.2 Å via Equation 5. A ϳ5°change in 1 alters the ␦ dip (calc) inconsequentially but results in a reasonable R Fe (C H) ϳ5.3 Å. Phe 68 (E11) exhibits only minor ␦ dip (obs) and, like in the triple mutant Mb, yields anomalous Curie plots that reflect variable population of alternate orientations (50). The crystallographic orientation, however, is consistent both with ␦ dip (calc) (not shown) and NOESY cross-peak pattern (Fig. 1). It is noted that for Lucina metHbICN, in contrast to L29F/H64Q/V68F-metM-bCN, the crystallographic distal Phe 43 (CD1) placement in either crystal structure yields similarly poor fits for ␦ dip (obs) versus ␦ dip (calc) (open triangles in Fig. 6B). Moreover, the Gln 64 (E7) orientation deduced above results in a ϳ0.3 Å van der Waals overlap with the crystallographically defined Phe 43 (CD1) ring. Moving the Phe 43 (CD1) ring 0.3 Å parallel to the heme so as to abolish this steric interaction leads to a much better fit between ␦ dip (obs) and ␦ dip (calc) for Phe 43 (CD1) (closed triangles, Fig. 6B). The positions of the distal residues for metHbIH 2 O in the crystal and metHbICN in solution are compared in Fig. 2B, and reveal significant differences primarily for the Gln 64 (E7) orientation.

DISCUSSION
Solution NMR Cavity Structure-The conventional two-dimensional NMR approach allowed unambiguous assignment for all relevant residues in the heme cavity. The intrinsic paramagnetism of cyano-metMb, instead of being an impediment to assignment, was actually an advantage since the dipolar shift with minimal relaxation provide significantly enhanced resolution for the E and F helical segments. For the proximal side of the heme, the NOESY cross-peak patterns and paramagnetic relaxation for the two Mb mutants were indistinguishable from those observed for wild-type metMbCN and predicted by the wild-type MbCO structure. A similarly strongly conserved proximal structure is confirmed for Lucina metHbICN. These conserved proximal structures allow the determination of meaningful magnetic axes for each protein. The robust nature of the magnetic axes for L29F/H64Q/V68F-metMbCN is confirmed by the indistinguishable magnetic axes determined using either the wild-type MbCO or triple mutant crystal structures.
For the distal residues, whose relaxation properties, dipolar shifts and/or NOESY cross-peak pattern differed from that expected for the reference crystal structure, the combined restraints for dipolar shifts and paramagnetic relaxation allow the determination of the their orientations in a more definitive manner than allowed by the NOESY cross-peak pattern themselves. The accuracy of the distal orientations is limited only by the basic assumption that the helical backbone positions are inconsequentially perturbed in the mutant and/or ligation variant relative to the available crystal structure. It is noted that the labile protons of the distal Gln 64 (E7) are readily assigned. The combined effect of the dipolar shift and paramagnetic relaxation allow the placement of the labile protons in the distal pocket so as to more accurately describe hydrogen bonding between Gln 64 (E7) and the bound ligand than is possible in some of the crystal structures. The aromatic ring for Phe 68 (E11) is oriented away from the expected binding site with the ring inserting into a largely pre-existing cavity in the distal pocket (55). The absence of a methyl group at the ␥ 2 position results in less direct hindrance of the bound ligand than is observed for the naturally occurring V68F metMbCN and Lucina metHbICN. This orientation is probably similar to that in the oxy complexes.
Comparison of Distal Pockets in Solution and Crystal Structures-For all but select distal residues, the heme pocket structure for each protein is essentially the same in cyanomet and aquo-met derivatives. In the case of the sperm whale triple mutant, the major difference in the Gln 64 (E7) orientation between the cyanomet and aquomet complexes is ϳ180°rotation about 3 that interchanges the carboxy and amide termini (Table II, Fig. 2A). The 180°rotation of the Gln side chain terminus is completely consistent with the stabilizing roles of the side chain for the different ligands. For the cyano complex, the N ⑀ H 2 is oriented to serve as a H-bond donor to bound cyanide, whereas in the aquomet complex, the carbonyl oxygen is oriented to accept a H-bond from coordinated water. The small changes in 1 , 2 for Gln 64 (E7) (Table II) help to accommodate the larger cyanide ligand.
As shown in Fig. 2B, the solution orientation of Gln 64 (E7) in Lucina metHbCN I differs substantially from that reported for the corresponding aquomet crystal. The altered orientation is firmly and independently established by the NOESY pattern from C ␥ Hs to Phe 29 (B10), the dipolar shift simulation, and the paramagnetic relaxation. However, both structures place the amino group in a position to interact with the bound ligand. The electron density in the crystal structure does not allow differentiation between the terminal carbonyl O atom and the N ⑀ atom. 4 However, the carbonyl O atom rather than the reported NH 2 must be pointing toward the coordinated water to serve as H-bond acceptor. Donation of proton to the bound water would give the ligand a partial positive charge (i.e. H 3 O ϩ character) causing substantial electrostatic repulsion with the net ϩ1 charge on the hemin iron atom. Thus, the solution NMR data, through the dipolar shift and paramagnetic relaxation, provide the most definitive location of the NH 2 group. The small movements of Phe 29 (B10) and Phe 43 (CD1) in Lucina HbI appear to reflect minor accommodation of the larger cyanide ligand. However, the orientations of Gln 64 (E7) do differ in solution and crystal, as shown in Fig. 2B.
The orientation of the Gln 64 (E7) side chain in the sulfide complex of Lucina metHbI (not shown) is intermediate between that for the aquomet crystal and that for metHbICN in solution (Fig. 2B). The CO versus NH 2 terminus orientation in the sulfide complex of Lucina metHbI also differs from that in metHbICN in solution by a 180°rotation at 3 . Again, this difference is consistent hydrogen bond donation to bound cyanide and acceptance from bound H 2 S. Implications for Sulfide Binding-All three single distal pocket substitutions in sperm whale Mb cause an enhancement of sulfide binding but have variable effects on oxygen affinity. The key mutation is H64Q, which causes a 25-fold increase in sulfide affinity by itself. The triple mutant, L29F/H64Q/V68F-metMb, has a sulfide affinity which is ϳ700 times greater than that of wild-type metMb and only ϳ7-fold less than that of Lucina HbI. Thus the triple mutant exhibits an increased sulfide binding free energy of ϳ3.8 kcal/mol or about ϳ75% of the stabilization of the bound sulfide (ϳ5.1 kcal/mol) by native Lucina HbI relative to that by wild-type sperm whale Mb. Hence, a major portion, but not all, of the remarkable sulfide affinity of Lucina HbI can be transferred to a mammalian Mb by replacing the three residues in contact with bound sulfide with those found in Lucina HbI. The high affinity of Lucina HbI for sulfide was attributed (24,25) to a combination of the hydrogen-bond acceptance by the Gln 64 (E7) side chain carbonyl, the hydrophobic pocket provided by the "cage" generated by the three Phe side chains, B10, CD1, and E11, and highly favorable electrostatic interactions between the sulfur and the edges of the three aromatic rings. While the distal pockets for the triple mutant Mb and Lucina HbI are similar (Fig. 2), there are some significant differences that relate to their differences in affinity for sulfide.
The side chains of Gln 64 (E7) and Phe 43  protein (Fig. 2C). However, both Phe 29 (B10) and Phe 68 (E11) in the sperm whale mutant are rotated ϳ90°about 2 relative to their positions in Lucina HbI and their edges are closer to the iron atom in the mollusc Hb (Fig. 2C). As a result of the differences, Lucina hemoglobin has a significantly larger ligand-binding site than that found in the myoglobin mutants, which should facilitate the binding of the large H 2 S ligand. The change in orientation of Phe 29 (B10) should have little affect on stabilization of bound sulfide since the positive edge of the ring multipole is pointing toward the bound ligand in both proteins. In Lucina HbI, the positive edge of the Phe 68 (E11) ring also points toward the bound ligand (Fig. 2C). This orientation of the Phe(E11) side chain is caused by direct steric interactions Phe 28 (B9) in Lucina HbI (25). Stabilization of bound sulfide by Phe 68 (E11) ring does not appear to occur in the myoglobin triple mutant since the E-11 side chain is allowed to take a position more perpendicular to the plane of the heme since the B9 position is occupied by a smaller Ile residue. A more complete and quantitative interpretation of the sulfide binding data will require detailed volume and electrostatic calculations for the entire set of mutants listed in Table I; such studies are now in progress.