Structures of an unliganded neurophysin and its vasopressin complex: Implications for binding and allosteric mechanisms

Abstract

The structures of des 1–6 bovine neurophysin-II in the unliganded state and as its complex with lysine vasopressin were determined crystallographically at resolutions of 2.4 Å and 2.3 Å, respectively. The structure of the protein component of the vasopressin complex was, with some local differences, similar to that determined earlier of the full-length protein complexed with oxytocin, but relatively large differences, probably intrinsic to the hormones, were observed between the structures of bound oxytocin and bound vasopressin at Gln 4. The structure of the unliganded protein is the first structure of an unliganded neurophysin. Comparison with the liganded state indicated significant binding-induced conformational changes that were the largest in the loop region comprising residues 50–58 and in the 7–10 region. A subtle binding-induced tightening of the subunit interface of the dimer also was shown, consistent with a role for interface changes in neurophysin allosteric mechanism, but one that is probably not predominant. Interface changes are suggested to be communicated from the binding site through the strands of β-sheet that connect these two regions, in part with mediation by Gly 23. Comparison of unliganded and liganded states additionally reveals that the binding site for the hormone α-amino group is largely preformed and accessible in the unliganded state, suggesting that it represents the initial site of hormone protein recognition. The potential molecular basis for its thermodynamic contribution to binding is discussed.

The protein neurophysin (NP) plays a central role in the targeting of the hormones oxytocin and vasopressin to regulated neurosecretory vesicles and to the storage of the hormones within these vesicles (for review, see Breslow and Burman 1990). Biologically, each hormone is compartmentalized with a separate neurophysin, but the oxytocin- and vasopressin-related neurophysins are very closely related structurally and each reacts similarly with either hormone in vitro. We previously have reported the structure of bovine vasopressin-related neurophysin (BNP-II) bound to oxytocin (Rose et al. 1996) and to the dipeptide p-iodo-l-phenylalanyl-l-tyrosine amide, which also binds to the hormone-binding site (Chen et al. 1991). However, questions about the relationship between structure and function in this system, particularly as to mechanisms underlying allosteric behavior (e.g., Nicolas et al. 1980; Breslow et al. 1991) and the thermodynamics of ligand binding (e.g., Breslow and Burman 1990) are not answered by the structure of the bound state alone. Neurophysin was probably the first protein for which binding was found to be thermodynamically linked to increased dimerization (Nicolas et al. 1978), and the mechanism responsible for this linkage has yet to be elucidated; the hormone-binding site and subunit interface are not proximal (Chen et al. 1991; Rose et al. 1996). Given what is now known to be the importance of binding-induced dimerization to many other biologic processes, the mechanism of this phenomenon in NP has the potential for more general applicability. Similarly, mechanisms underlying the thermodynamics of hormone–NP interaction (Breslow and Burman 1990; Breslow et al. 1999) cannot be addressed completely without knowledge of the unbound state.

In the present study, we report the structure of des 1–6 BNP-II in the unliganded state at 2.4 Å resolution. This derivative is the first NP that has yielded crystals in the unliganded state suitable for X-ray diffraction. Its general crystallization properties (Wu et al. 1996) and aspects of its solution properties (Zheng et al. 1997) have been reported previously. It differs from the full-length protein principally in having a moderately increased dimerization constant in the unliganded state (Zheng et al. 1997). We also report the structure of the vasopressin complex of this derivative. No structure of a vasopressin–NP complex has been reported previously. The structure is of interest because it allows rigorous comparison both with the corresponding unliganded state and with the known structures of the full-length protein complexed with oxytocin and with dipeptide, and because it represents the first crystallographic view of vasopressin conformation.

Results

Comparison of the general folding of liganded and unliganded states

Figure 1 ▶ compares the backbone ribbon structures of the dimer of unliganded des 1–6 BNP-II with that of its lysine vasopressin (VK) complex and with that of the oxytocin complex of the full-length protein. Bound hormone also is shown. Note that the tripeptide tail of the hormones is resolvable only in the oxytocin complex, and that NP residues 1–4 and 1–6 of subunits A and B, respectively, are not resolved in the oxytocin complex. The overall similarity of folding in the three structures is evident in Figure 1 ▶, with the distinct exception of the external loop at the carboxyl end of the 3,10 helix, which differs significantly between the unbound and ligated protein structures as discussed further below. The structures have an axial ratio of approximately 3 in the unliganded state, which is reduced only slightly (by ∼10%) on ligation. The asymmetry of the unliganded dimer is consistent with estimates based on hydrodynamic studies (Rholam et al. 1982), although the retention of this asymmetry on ligation differs significantly from the assignment of the liganded state as a sphere based on the same hydrodynamic studies. Like the oxytocin complex, the crystalline VK complex contains only one bound hormone per subunit, with no evidence of the controversial second site reported for lysine vasopressin in solution (cf. Nicolas et al. 1980; Breslow and Burman 1990).

Fig. 1.

Ribbon drawings of the backbone structures of neurophysin in the oxytocin complex of full-length BNP-II (left); the VK complex of des 1–6 BNP-II (center); unliganded des 1–6 BNP-II (right). Bound hormones are shown in black.

Disulfide pairing patterns (Burman et al. 1989) and crystallographic data (Chen et al. 1991; Rose et al. 1996) have shown that each NP chain consists of two domains, a larger amino domain and a smaller carboxyl domain that has homology with the major segment of the amino domain. The crystallographic data also have shown that the homologous segments of the two domains consist predominantly of β-sheet and turns, with the nonhomologous segment of the amino domain additionally containing an 11-residue stretch of 3,10 helix that extends from residues 39 to 49 and that forms part of the hormone or peptide-binding site. A long unstructured loop of approximately nine residues (residues 50–58), which also contains part of the binding site, extends from the last five residues of the amino domain through the first four residues of the carboxyl domain. Figure 2 ▶ is a superposition of the backbone structures of the unliganded des 1–6 protein and its VK complex. There is no difference between the two in β-structure content, or in the length of the single 3,10 helix. However, the long unstructured loop between residues 50–58 and the unstructured region between residues 7 and 10 each contain binding-induced displacements in the 2–3-Å range. Smaller backbone changes, such as those at the end of the 3,10 helix, are seen elsewhere in the structure, some of which are accompanied by significant side-chain positional differences. These include changes in the subunit interface as discussed below.

Fig. 2.

Superimposition of the backbones of residues 8–85 of unliganded des 1–6 BNP-II (black) and its VK complex (white).

Comparison of the structures of bound oxytocin and lysine vasopressin and of their interactions with NP

Figure 3 ▶ compares the structures of bound oxytocin and VK. The tripeptide tails of the hormones are not shown, because the VK tail is not seen, presumably because of greater disorder than the oxytocin tail. Note that the tails of the two hormones participate at best very weakly in binding, as evidenced by both the crystal structure of the oxytocin complex (Rose et al. 1996) and solution data on bound hormones (Blumenstein and Hruby 1977). Table 1 lists the NP residues in close contact with the hormone ring (hormone residues 1–6) in the two structures. The data show that Cys 1 and Tyr 2 of the hormones are superimposable in the two structures and interact with identical residues on NP. Cys 1 and Tyr 2 contain the groups most critical to binding—the α-amino group and the phenyl ring in position 2 (Breslow and Burman 1990). As in the dipeptide and oxytocin complexes, the vasopressin α-amino group is hydrogen-bonded to the carboxyl group of Glu 47 (with which it also interacts electrostatically) and the backbone carbonyl groups of Leu 50, Ser 52, and Glu 47 (described in greater detail below). The Tyr 2 phenyl ring is bound in a tight pocket with most of its interactions to the backbone and/or side-chain regions of residues 21–24, the 10–54 disulfide bridge, and the side chains of residues 47 and 48. In addition, and not previously appreciated, are two weak (∼4.4 Å and 4.9 Å) van der Waals contacts between the phenyl ring of Tyr 2 of vasopressin and CB of Asp 76, and 4–5-Å contacts of CB of Tyr 2 with the Asp 76 side chain; similar contacts also are seen in the oxytocin and dipeptide complexes. It is relevant that the VK complex, which represents the first NP complex in which bound water is visualized, also shows three water molecules in contact with the Tyr 2 ring at <4.5 Å, one in contact with CD2 and two in contact with CD1, one of which also contacts CE1 at a distance of 4.1 Å. This is surprising, because water was predicted by modeling to contact only one CD position and not to contact a CE position (Breslow et al. 1999). One notable difference between the two hormones in Tyr 2 interactions lies in the hydrogen-bonding pattern of the Tyr 2 hydroxyl group, which is hydrogen-bonded to the carbonyl oxygen of both Cys 44 and Gly 23 in the VK complex, but only to the Cys 44 oxygen in the oxytocin complex.

Fig. 3.

Structures of the ring (residues 1–6) of bound oxytocin (left) and bound lysine vasopressin (center). (Right) Crystal structure of (unliganded) pressinoic acid (Langs et al. 1986). Colors are as follows: (yellow) residue 1 (Cys/2); (light blue) residue 2 (Tyr); (purple) residue 3 (Ile in oxytocin, Phe in VK and pressinoic acid); (red) residue 4 (Gln); (green) residue 5 (Asn); (yellow) residue 6 (Cys/2).

Table 1.

Comparison of NP contacts for ring residues of oxytocin and vasopressin in their NP complexes with BNP-II and des 1–6 BNP-II, respectively

	Protein contacts at <4.5 Å (residues)
Hormone residue	Oxytocin	Vasopressin
Cys 1	8, 47, 48, 50–54	8, 47, 48, 50–54
Tyr 2	10, 21–24, 44, 47, 48, 54, 76	10, 21–24, 44, 47, 48, 54, 76
Ile 3	5, 7, 53–55
Phe 3		53, 54
Gln 4	55, 75, 76	76
Asn 5	48	48, 76a
Cys 6	53	51, 53

^a The apparent difference between the two complexes in interactions with Asp 76 is exaggerated by the 4.5 Å distance cutoff. The closest contact between Asn 5 and Asp 76 is 4.43 Å in the vasopressin complex and 4.65 Å in the oxytocin complex.

Supporting interactions involving residues 1 and 2 of the hormones are van der Waals interaction of the residue 1 side chain with the protein and a hydrogen bond between the carbonyl of residue 1 and the −NH of Cys 54 of the protein. These are identical in the VK and oxytocin complexes. Not identical are the interactions of residue 3, which is Phe in vasopressin and Ile in oxytocin. In both cases, the −NH of residue 3 is hydrogen-bonded to the oxygen of NP Cys 54 as in the oxytocin and dipeptide complexes. However, the Ile and Phe side chains establish different contacts with NP. The Ile side chain is in van der Waals contact with Leu 5, Leu 7, Pro 53, and Gln 55 of NP, with contacts to residues 7 and 55 at <4 Å. In contrast, the Phe 3 side chain interacts more weakly with residue 55 (closest contact is 4.83 Å, compared with 3.28 Å for Ile), and there are no contacts to residue 7. The differences between Phe and Ile in position 3 in their interactions with NP reflect several factors. First, as shown in Figure 3 ▶, these residues occupy overlapping, but nonidentical space, with a difference of 1.17 Å in the CB of each in the bound state. Second, residue 5 of NP is absent in the des 1–6 protein, from which the VK complex was prepared, and residue 7 of NP is markedly displaced from the position it occupies in the oxytocin complex of the intact protein. This possibly reflects cooperative interactions between residues 5 and 7 of native NP in interactions with residue 3 of the hormone; these are not possible in the des 1–6 protein, weakening potential interactions of the Phe 3 side chain with residue 7. A difference in the orientation of Gln 55 in the two complexes (CD positions differ by 1.3 Å) is a probable contributor to the weaker interactions of Phe 3 than of Ile 3 with Gln 55 in the two complexes and potentially derives from the above difference in the position of residue 7, which is in close van der Waals contact to Gln 55 in the oxytocin complex, but not in the VK complex.

The most significant difference between the two complexes, however, lies in the orientation of Gln 4 (Fig. 3 ▶). The CD position of Gln 4 in the two complexes differs by 5.03 ± 0.13 Å, the average reflecting the fact that the two subunits of the oxytocin complex are not strictly identical. Depending on the subunit of the oxytocin complex used as reference, backbone differences between the two bound hormones in the vicinity of Gln 4 range from 0.49 Å to 0.70 Å, which can be compared with values in the range 0.2–0.4 Å elsewhere in the ring. Additionally (Table 1), Gln 4 of vasopressin interacts with only a subset of the residues that contact Gln 4 of oxytocin and more weakly; for example, while distances between Gln 4 and residues 55 and 75 in the VK complex are minimally 6.9 Å, interactions of Gln 4 with these residues in the oxytocin complex occurs with closest contacts of 3.75 ± 0.15 and 4.5 ± 0.3 Å, respectively. The differences are primarily the consequence of the different orientation of Gln 4 in the two complexes, because the differences in protein structure in the vicinity of residues 55 and 75, although real, are small relative to the difference in contact distance. The results suggest intrinsic differences in the structures of oxytocin and vasopressin in the vicinity of Gln 4 that are not annulled by the general similarity in their more critical interactions with NP.

Because of differences between the two hormones in conformation, we compared their conformations with that of the crystal structure (Langs et al. 1986) of (unliganded) pressinoic acid, which represents the ring of vasopressin without the tripeptide tail. As seen in Figure 3 ▶, there is little resemblance of the pressinoic acid structure to that of either of the bound hormone rings. This indicates either the influence of binding on the hormone structures, or of crystal packing forces or the lack of a tripeptide tail on pressinoic acid.

Conformational changes at the Tyr 2 binding site induced by binding

Establishment of the binding pocket for the Tyr 2 ring requires significant movement of several residues. Figure 4A ▶ superimposes the positions of key residues involved in this binding (<5 Å from the ring) in the VK complex and unliganded states. Large movements are seen in Asn 48 and Glu 47, changes in the Glu 47 side chain reaching 3 Å and potentially being attributable in part to its additional role in salt bridge formation (see below). Large movements also are seen in the 10–54 disulfide pair, the sulfurs of which contact the Tyr 2 ring and move ∼3 Å (Cys 10) and ∼2 Å (Cys 54) on binding. Associated backbone displacements occur in these residues, the largest of which involve the Cys 54 carbonyl oxygen. This moves 3.2 Å on binding, which also may result from its hydrogen bonding to the −NH of hormone residue 3 (vide supra); similarly, the −NH of Cys 54 moves 1.6 Å on binding, possibly in part because of its hydrogen bonding to the carbonyl of Cys 1 of the hormone. Smaller changes, of potential mechanistic significance (see Discussion) occur in Gly 23 and Pro 24, both of which establish multiple contacts to the Tyr 2 ring.

Fig. 4.

Comparison of key binding site residues in the unliganded state (black) and in the VK complex (white). Residues are numbered. (A) Superimposition of binding site residues for the Tyr 2 ring; (B) superimposition of binding site residues for the α-amino group, also showing the water molecules of the unliganded state; (C) comparison of hydrogen-bonding patterns in the VK complex (left) and in the unliganded state (right) at the binding site for the α-amino group. Residue 1' is Cys 1 of the hormone. Note that a hydrogen bond between the backbone O of Glu 47 and the N of Leu 50, which is present in both states, is omitted for clarity (see Table 2).

Although substantial movements of some residues accompany binding to the Tyr 2 ring, changes in intrasubunit van der Waals contacts of binding site residues are relatively limited, in part because most of the intramolecular contacts of binding site residues arise from atoms that are adjacent in the sequence or are brought into proximity by disulfide bridges. These changes—or lack thereof—are central to an ultimate understanding of NP binding thermodynamics. Several of the NP atoms that contact the Tyr 2 ring are largely buried in both the bound and unbound states, and binding to these involves the acquisition of contacts to ligand, accompanied in some cases by the net establishment of additional intramolecular contacts. Consider contacts to the buried backbone N and O of Gly 23 at <5 Å. Establishment of seven contacts to the bound Tyr 2 ring by the N of Gly 23 is accompanied by the addition of four new relatively weak intrachain contacts and both increases and decreases in distances to other residues, but only one contact characteristic of the unbound state is increased to >5 Å. The picture with the O of Gly 23 is similar—in this case binding also accompanied by a significant decrease in backbone hydrogen bond length (Discussion). The net increase in packing interactions on binding in these cases is clearly stabilizing. For some atoms, however, binding involves significant compensatory adjustments in pre-existing contacts. For example, the Cys 44 O is hydrogen-bonded to the −NH of Glu 47 in the unbound state at a distance of 2.85 Å. In the bound state, it is hydrogen-bonded to the −OH of hormone Tyr 2 at a distance of 2.69 Å (in addition to its Tyr 2 ring contact), while the hydrogen bond to Glu 47 is stretched to 3.23 Å, with accompanying large increases in van der Waals contact distances to several other ligand atoms. These compensatory changes on hydrogen bonding of the Tyr-OH potentially account for the negligible contribution of the hydroxyl group (e.g., Breslow and Burman 1990) to the free energy of binding.

For ligand atoms that are largely or completely exposed to water in the unbound state, the establishment of contacts to the Tyr 2 ring is associated with diminished exposure to bulk solvent and the acquisition of contacts to bound water. The side-chain N and O atoms of Asn 48 are reasonably accessible to bulk solvent in the unbound state with no evidence of bound water, while they are each hydrogen-bonded to a water in the bound state (bond lengths of 2.64 Å and 3.02 Å, respectively) and more buried. A similar situation prevails with the ring carbons of Pro 24.

Conformational changes at the α-amino binding site

Residues involved in binding the essential hormone amino group also move significantly on binding. Figure 4B ▶ is a superimposition of these residues in the bound and unbound states. Changes of 2–3 Å occur in essentially all of the residues (Glu 47, Arg 8, Leu 50, and Ser 52), although the positions of these residues relative to each other remains similar—not identical— in both states. This similarity partially reflects the fact that the salt bridge between Arg 8 and the Glu 47 carboxyl group, which is part of the more complex salt bridge involving the hormone α-amino group in the bound state, persists to a significant extent in the unbound state. Figure 4C ▶ and Table 2 compare interactions at the α-amino binding site in the unliganded and liganded des 1–6 structures. In subunit A in the unliganded state, the Glu 47 carboxyl oxygens are multiply hydrogen-bonded to the terminal nitrogens of Arg 8, and to bound waters, one of which (water 11) also is hydrogen-bonded to the backbone oxygen atoms of Glu 47 and Leu 50, and a second of which (water 4) is hydrogen-bonded to the N of Leu 11 and the O of Cys 21 (not shown). The carbonyl oxygen of Glu 47 is additionally hydrogen-bonded to the backbone −NH of Leu 50 (Table 2). One of the terminal Arg 8 nitrogens is hydrogen-bonded to the carbonyl oxygen of Ser 52, while the other is hydrogen-bonded to a third bound water (water 36). Subunit B (data not shown) is virtually identical to subunit A in this region, but only one bound water (corresponding to water 4 of subunit A) is present, probably reflecting the observed greater solvent exposure of Glu 47 in subunit B so that bound water is replaced by bulk water. Interestingly, solution studies suggest that the Glu 47 salt bridge in unliganded bovine NP-I is less well formed than in NP-II (Zheng et al. 1996), an observation that might be explained by a greater average solvent exposure of the NP-I site than of the NP-II site.

Table 2.

Comparison of key hydrogen-bonding interactions at the α-amino binding site of subunit A in the unliganded and liganded states

Group	Hydrogen-bonding partners and distances
Unliganded state (subunit A)
Glu 47 OE1	Arg 8 NH2 (2.82), water 11 (2.71)
Glu 47 OE2	Arg 8 NH2 (2.93), Arg 8 NH1 (2.88), water 4 (2.40)
Glu 47 O	Leu 50 N (3.01), water 11 (2.64)
Arg 8 NH1	Glu 47 OE2 (2.88), water 36 (2.71)
Arg 8 NH2	Glu 47 OE1 (2.82), Glu 47 OE2 (2.93), Ser 52 O (3.06)
Leu 50 O	Water 11 (2.71)
Ser 52 O	Arg 8 NH2 (3.06)
Water 4	Glu 47 OE2 (2.40), Leu 11 N (2.55), Cys 21 O (3.02)
Water 11	Glu 47 OE1 (2.17), Glu 47 O (2.64), Leu 50 O (2.71)
Water 36	Arg 8 NH1 (2.71)
Vasopressin complex
Cys 1 α-N (VK)	Glu 47 OE1 (2.77), Leu 50 O (2.68), Ser 52 O (2.77), Glu 47 O (2.76)
Glu 47 OE1	Cys 1 α-N (2.77)
Glu 47 OE2	Arg 8 NH1 (2.52), Leu 11 N (3.08)
Glu 47 O	Leu 50 N (3.04), Cys 1 α-N (2.76)
Arg 8 NH1	Glu 47 OE2 (2.52)
Arg 8 NH2	Ser 52 O (2.76)
Leu 50 O	Cys 1 α-N (2.68)
Ser 52 O	Cys 1 α-N (2.77), Arg 8 NH2 (2.76)

Numbers in parentheses are distances in Angstroms. Waters are assumed hydrogen bonded when distances are <3.4 Å.

In the liganded state (both subunits identical), all bound waters are displaced by the incoming protonated hormone α-amino group, which interacts with the Glu 47 carboxyl group both electrostatically and by a hydrogen bond and is additionally hydrogen-bonded to the carbonyl oxygens of Glu 47, Leu 50, and Ser 52. The carbonyl oxygen of Glu 47 also remains hydrogen-bonded to the −NH of Leu 50 in the complex at a distance almost unchanged from its unliganded state (Table 2). Thus, both the Glu 47 carbonyl oxygen and one amino proton are each involved in two hydrogen bonds. Finally, the specific salt bridge interactions between Glu 47 and Arg 8 are slightly weakened by binding in the VK complex. Only one hydrogen bond remains (Table 2), although both terminal −NH groups are in van der Waals contact with the carboxyl group, and the Arg 8 terminus appears more solvent-exposed. The same bonding pattern is seen in subunit B of the oxytocin complex but is slightly altered in subunit A. The bound amino group of the liganded state therefore interacts almost identically to water 11 of the unbound state, with the exception of its direct formation of a hydrogen bond to the carbonyl oxygen of Ser 52. Despite this similarity, the salt bridge site is translocated 2–3 Å on ligand binding. The significance of the structure of this site in the unliganded state is discussed below.

Conformational changes at the subunit interface induced by binding

Interactions between NP subunits to form the dimer involve both the amino and carboxyl domains (Chen et al. 1991; Rose et al. 1996) of the protein. In the amino domain, interactions across the interface consist of reciprocal antiparallel β-sheet hydrogen bonding between residues 34 and 36 of each chain, supported by other contacts that principally involve the side chains of these residues and of residues 32, 35, 38, and 40. In the carboxyl domain, similar intersubunit β-sheet hydrogen-bonding interactions involve residues 77, 79, and 81 of the two chains and are supported principally by side-chain interactions of these residues and of residues 78 and 80. Additionally, there are weak side-chain interactions between residue 36 of the amino domain and residue 72 of the carboxyl domain across the interface. Evidence of hydrogen bonds involving α-CH groups and main chain carbonyl groups (e.g., Vargas et al. 2000) and of other backbone interactions also is seen.

Structural rearrangements of the subunit interface associated with binding peptide are relatively small. Figure 5A ▶ compares the interface backbone of the unliganded state with that of the VK complex; residues shown are those establishing intersubunit contacts—32–40, 72, and 77–81. The identity of the residues establishing contacts between the two strands, the relative orientation of the two strands, and the relationship of the amino domain segment of the interface to that of the carboxyl segment are unaltered by binding. However, small backbone displacements (typically 0.2–0.3 Å) and still larger side-chain displacements (up to 1 Å) are seen in a subset of residues, some of the largest side chain changes occurring in residues 36, 40, 77, and 81 (Fig. 5B ▶). Collectively, these changes lead to small alterations in the van der Waals contacts and hydrogen-bonding distances across the subunit interface, the nature of the changes on average indicating a slight tightening of the interface in the bound state relative to the unbound state. Of the ∼300 interatomic contacts across the interface, 58 differ in the two states by >0.2 Å. Of these, 46 are closer in the VK complex than in the unliganded state with an average difference between the two states of 0.34 Å, whereas only 12 are closer in the unliganded state with an average difference between the two states of 0.38 Å. Approximately two-thirds of the contacts that are closer in the VK complex are also at least 0.2 Å closer when the oxytocin complex is compared with the unliganded state, suggesting that they do not arise from crystallization conditions and do not represent random differences. Of the contacts that appear at least 0.2 Å looser in the bound state than in the unbound state, only one-third are preserved in the oxytocin complex. A particularly striking binding-induced change in the interface that is common to both the VK and oxytocin complexes, and also seen in the dipeptide complex, is the formation of a hydrogen bond between the −OH of Thr 81 of one subunit and the carbonyl oxygen of Glu 77 of its partner subunit, markedly reducing the solvent accessibility of the −OH (Fig. 5B ▶).

Fig. 5.

Superimposition of the subunit interface in the unliganded state (black) and in the VK complex (white). Bond radii in this figure have been decreased from 0.2 Å to 0.1 Å for clarity. (A) Backbones of the amino domain segment of the interface (residues 32–40) and of the carboxyl domain segment of the interface (residues 72, 77–81); note that the two domains have been rotated relative to each other for clarity; (B) binding-induced changes in side-chain orientation of selected residues, including demonstration of the formation of a hydrogen bond between the −OH of Thr 81 and the backbone O of Glu 77. The distance across the interface between the OG1 of Thr 38 and the backbone O of Glu 77 is decreased by 0.4 Å in the complex, and the distance between the CG2 atoms of Val 36 is decreased by 0.7 Å in the complex. In contrast, distances between the terminal oxygens of Glu 40 and the ring of Phe 35 increase by ∼0.6 Å and 1 Å in the complex.

Discussion

Significance of structural differences between the oxytocin and vasopressin complexes

Although the structures of the oxytocin and vasopressin complexes are very similar, we have pointed to differences between them in the positions of binding site residues on NP, and in the conformations and protein contacts of the two bound hormones. The large difference in bound hormone conformation at residue 4 is difficult to rationalize unless it is intrinsic to the relative conformational preferences of the two hormones and potentially plays a role in the observed NP conformational differences between the two complexes near regions of the binding site. However, the possibility (Results) that some of the differences reflect the absence of NP residues 1–6 in the VK complex is supported by the significantly fewer interactions with NP of Phe 3 (of VK) than of Ile 3 (of oxytocin) in their crystalline complexes. Solution data suggest similar interactions between an Ile and a Phe in position 3 of small peptides that bind to the hormone-binding site of the intact protein (e.g., Breslow et al. 1973; Breslow and Burman 1990) and indicate similar binding affinities of oxytocin and vasopressin to the intact protein (e.g., Breslow and Walter 1972; Nicolas et al. 1980). Large differences in Gln 58 in the two complexes (CD positions differ by 3.1 ± 0.7 Å) also are not obviously explained by differences in hormone structure, because Gln 58 does not contact the bound hormone. The crystal structure of the oxytocin complex of the des 1–6 protein is needed to resolve this issue.

Significance of the comparative structures and environments of binding site residues in bound and unbound states

A significant finding is the extent to which the binding site for the hormone α-amino group pre-exists in the unliganded state. The hydrogen bonds formed by water 11 in the unliganded state indicates that it has almost identical hydrogen-bonding partners to those of the bound amino group (Table 2). The sole difference between the amino group and water 11 is the absence of a hydrogen bond from water 11 to the Ser 52 carbonyl oxygen in the unbound state and the presence of a hydrogen bond from the amino group to the Ser 52 carbonyl oxygen in the bound state, which occurs without loss of the hydrogen-bonding interaction of the carbonyl to the Arg 8 terminus. The similarity between water 11 and the amino group in hydrogen-bonding interactions and the number of hydrogen bonds to oxygen of water 11 in the unbound state suggests that, like the amino group, it might be bound in the protonated state. However, this is unlikely, both because of the huge binding-induced change in water pK_a it would represent and because such protonation would be difficult to reconcile with known binding-induced changes in proton equilibria (e.g., Breslow et al. 1971). Both the third hydrogen bond from water and the fourth hydrogen bond to the amino group therefore most are likely to represent bifurcated hydrogen bonds.

Considerable ambiguity surrounds the potential contributions of salt bridges to protein folding and interactions (e.g., Hendsch and Tidor 1994; Xu et al. 1997). In the case of NP, the negative free energy of binding the hormone α-amino group has been estimated as >5 Kcal/mole based on the effects on binding of deleting the amino group and the assumption of free energy additivity (Breslow and Burman 1990). A recently observed cooperative interaction energy of ∼1 Kcal/mole between binding of the hormone tyrosine ring and salt bridge stability (Breslow et al. 1999) now suggests a minimum stabilizing contribution of 4 Kcal/mole for the bound amino group. Several factors potentially contribute to this stability. These include the negative dipole at the end of the 3,10 helix, a role for which is suggested by the requirement for the protonated α-amino group in binding even when the Glu 47 carboxyl is protonated (e.g., Breslow and Burman 1990). Another potential contributor is the additional hydrogen bond to the Ser 52 carbonyl oxygen in the bound state formed with the amino group (Table 2). Assuming ∼5 Kcal/mole for the gas phase stability of a hydrogen bond, the net addition of a hydrogen bond by the bound amino group would be strongly stabilizing. Moreover, whereas the hydrogen-bonding interactions to waters 4 and 36 are lost in the bound state (Table 2; Fig. 4 ▶), these appear to be compensated for by a new hydrogen-bonding interaction of Leu 11 (Table 2) and increased solvent exposure of Arg 8. Finally, the protonated amino group should form stronger hydrogen bonds than those formed by the water molecule it displaces. Recent data also support the concept that salt bridges are more stable when part of a hydrogen-bonding or ion-pair network (e.g., Albeck et al. 2000), as is the case here.

The fact that the amino group binding site pre-exists to a significant extent in the unbound state and is relatively superficial in location suggests that it might be the initial site of hormone–NP recognition, anchoring the hormone while the Tyr 2 ring inserts into its less accessible site. Nonetheless, the amino binding site moves 2–3 Å to attain the bound state. Movement of Ser 52 can be explained by its hydrogen bonding to the amino group, but a plausible reason for the remainder of the change lies with conformational changes necessary to optimize binding of the tyrosine ring. Ligand atoms involved in the binding of the Tyr 2 ring include CB and CG of Glu 47 (potentially accounting for the observed interaction energy between binding of the Tyr ring and salt bridge strength; Breslow et al. 1999), the side chain of Asn 48, which is adjacent to Glu 47, and the Cys 10 sulfur, which is sequentially near Arg 8 of the amino group salt bridge. The final bound state therefore can be viewed as one in which concerted movements of the region around Cys 10 and Glu 47 optimize contacts to the Tyr 2 ring while preserving salt bridge contacts. Such movements also are suggested to account for a demonstrably increased internalization of Glu 47 in the bound state. Thus their energetic cost would be paid for both by the stabilizing contacts to the Tyr 2 ring and by a decreased dielectric constant in the vicinity of the salt bridge, additionally increasing salt bridge stability.

Implications for allosteric mechanism of differences between unliganded and liganded forms in the structure of the subunit interface

Potential mechanisms underlying the increase in dimerization associated with peptide binding include conformational differences between unliganded monomers and dimers at or near the binding site that preferentially facilitate binding to dimer, and binding-induced conformational change at the interface that strengthens intersubunit contacts. Support for the former mechanism is found in nuclear magnetic resonance data that indicate conformational differences between unliganded monomers and dimers of BNP-I in the vicinity of the hormone-binding site (e.g., Breslow et al. 1995), and in kinetic studies indicating a significantly slower on-rate for binding to monomer than to dimer (Pearlmutter and McMains 1977). In the present study, the subtle tightening of interactions across the subunit interface in the liganded state alternatively suggests a role for binding-induced changes at the interface and is in qualitative agreement with the different effects of pressure on dimer dissociation in liganded and unliganded states, which were interpreted to suggest a lower solvent penetrability of the monomer–monomer interface in the liganded state (Breslow et al. 1991). The question remains as to the extent to which this tightening contributes to the ∼2.8 Kcal/mole change in the free energy of dimerization (estimated from the 100-fold increase in dimerization constant [e.g., Kanmera and Chaiken 1985]) that is induced by peptide binding.

Available data indicate that the linkage between binding and dimerization is independent of the identity of the bound peptide, provided that the peptide occupies the normal hormone-binding site, and indicate that the linkage is similar in BNP-I and BNP-II (e.g., Nicolas et al. 1978; Breslow et al. 1991). Assuming this to be the case, binding-induced changes at the interface that contribute significantly to allosteric mechanism should be the same in the crystalline complexes of BNP-II with vasopressin, with oxytocin, and with dipeptide and also should be the same in BNP-I and -II. However, examination of the crystal structure of the dipeptide complex of BNP-II reveals that the bond lengths of most of the altered contacts are slightly longer than in the other complexes, and that not all the intersubunit van der Waals contact changes common to both the VK complex and oxytocin complex are manifest in the dipeptide complex. This may reflect the less sophisticated methodology used in the refinement of the dipeptide data than used for the later structures, or the fact that the dipeptide complex was unique in crystallizing as a pseudotetramer with five bound dipeptides (Chen et al. 1991). Nonetheless, the primary structure of BNP-I (Chauvet et al. 1983) also indicates that the binding-induced change in hydrogen bonding of the Thr 81 −OH group present in BNP-II complexes is less likely in BNP-I; residue 81 is Glu in NP-I and normally would not be expected to serve as a proton donor at neutral pH. Residue 81 is also an aliphatic residue in several neurophysins (Chauvet et al. 1983). Accordingly, the energetic consequences of some of the contact alterations seen in the hormone complexes are likely to be small. A preliminary estimate of the maximum likely energetic contribution of the interface tightening in the VK complex, calculated solely from van der Waals energies as described earlier (Breslow et al. 1999), and with no allowance for negative conformational entropy changes, places it at less than one-fourth of the total change in dimerization energy; reasonable allowances for a net contribution of changes in Thr 81 hydrogen bonding do not increase this fraction to greater than one-half. Other factors, such as differences between monomer and dimer in the unliganded state, therefore are likely to contribute the remainder.

By what mechanism does hormone binding induce interface changes; that is, which residues serve as the structural link between the interface and the binding site? No residue is common to both the hormone-binding site and the interface. The otherwise attractive possibility that the mechanism involves communication between Glu 47 and the amino domain segment of the interface via the connecting 3,10 helix is diminished by evidence that large changes in helix structure do not have significant effects on dimerization (Zheng et al. 1997). One likely connection between the interface and binding site involves the linked series of β-sheet strands (e.g., Rose et al. 1996) that connect these two regions. We particularly note that interactions between strands receive additional stability in the bound state from Gly 23, which interacts strongly with the Tyr 2 ring, at the same time markedly strengthening its hydrogen bonds to the N atoms of Ser 25 and Ile 26, these bonds stabilizing both a turn and the β-sheet in this region. Other binding-induced changes that might alter the interface are those in Asp 76, immediately adjacent to the interface, and Phe 22, the oxygen of which directly contacts the Tyr 2 ring. In the bound state, the intrachain contact distance between the Phe 22 O and Asp 76 α-carbon is reduced from 3.83 Å to 2.95 Å, consistent with formation of an α-proton to oxygen hydrogen bond that would connect the amino and carboxyl segments of the subunit interface. These structural features of the VK complex are also manifest in the complexes of NP with oxytocin and with dipeptide, thereby passing the minimum test for any proposed allosteric mechanism.

Higher resolution data and the application of rigorous computational methodology are needed to further define the contribution of binding-induced changes at the interface to NP properties. The present data provide evidence of a potential contribution of such changes to allosteric mechanism and allow initial hypotheses as to the route of communication between the interface and the binding site. However, at the same time, they strongly suggest that other factors, such as the conformational differences between unliganded monomers and dimers, ultimately may prove to be of greater mechanistic significance.

Materials and methods

Crystallization and data collection

Des 1–6 BNP II was prepared as described previously (Wu et al. 1996). Lys 8 vasopressin (VK) was purchased from Sigma (V6879) and used without further purification. Diffraction quality crystals for both des 1–6 BNP-II and its VK complex were grown by vapor diffusion using conditions previously reported (Wu et al. 1996; Hu et al. 1999).

The des 1–6 BNP-II data were collected at room temperature on a crystal measuring 0.3 × 0.3 × 0.3 mm mounted in a sealed glass capillary with a small amount of mother liquor to prevent dehydration. Data were recorded on a Rigaku R-Axis IV image plate detector system using 5 kW mirror (YALE/MSC) focused Cu Kα X-rays. The crystal to detector distance was 20 cm, which gives a maximum resolution of 2.43 Å at the detector edge. The data set consisted of 180, 1° oscillation images each exposed for 20 min. Data were indexed, integrated, and scaled using HKL (Otwinoski and Minnor 1997). Details of the data collection and processing are collected in Table 3.

Table 3.

Data collection and processing statistics

Data collection and processing statistics	des 1–6 BNP-II	VK complex
Unit cell and space group
a (Å)	48.92	50.9
c (Å)	78.52	105.8
Space group	P32	P6122
Molecules per asymmetric unit	2	1
Data collection statistics
Temperature (K)	298	110
Resolution (Å)	2.4	2.3
Completeness (%)	100	96.2
Rmergea (%)	4.2	5.8
Refinement statistics
Resolution range (Å)	23–2.4	15–2.3
R value all data	0.218	0.234
Free R value	0.249	0.299
Free R value test set size (%)	5.2	5.0
RMSD from ideality
Bond lengths (Å)	0.005	0.010
Bond angles (°)	1.4	1.5
Dihedral angles (°)	24.7	26.6
Improper angles (°)	1.01	1.33
Mean B value (Å2)	37.8	26.6
Coordinate errorb (Å)	0.30	0.3
Number of protein atoms	1118	551
Number of ligand atoms	0	52
Number of solvent atoms	43	36
Number of hetero atoms	0	3

^a

^b Estimated coordinate error from the Luzzati (Luzzati 1952) plot.(RMSD) Root mean square deviation.

The data for the VK complex of des 1–6 BNP-II were recorded on a crystal measuring 0.15 × 0.15 × 0.2 mm mounted in 0.3-mm fiber loop (Teng 1990) and flash-cooled (Hope 1988) to 110K in a nitrogen gas cold stream. Data were recorded on a MarResearch 30-cm image plate system using 5-kW mirror (YALE/MSC) focused Cu Kα X-rays. The crystal to detector distance was 18 cm, which gives a maximum resolution of 2.26 Å at the detector edge. The data set consisted of 360 quarter-degree oscillation images each exposed for 6 min. Data were indexed, integrated, and scaled using X-GEN (Howard et al. 1987).

Structure determination

The structures of both des 1–6 BNP-II and its VK complex were determined by molecular replacement (Rossmann and Blow 1962) using AMoRe (Navaza 1994). In both cases, residues 7–86 of the dimer of the BNP-II dipeptide complex (NP-FY) containing the A and C chains (Protein Data Bank entry 2BN2) were used as the search model.

The des 1–6 BNP-II analysis was conducted in space groups P3₁ and P3₂ using data in the range from 12 Å to 3.5 Å. The top four solutions from the rotation function analysis then were used in the translation search. The solution, a = 60.2, b = 77.0, c = 142.7, Tx = 0.87, Ty = 0.48, and Tz = 0.23 in space group P3₂, gave the lowest residual (45.9 with the next lowest value being 49.7).

The analysis of the VK complex of des 1–6 BNP-II was performed in space groups P6₁22 and P6₅22 by using data in the range from 8 Å to 4 Å. The top five solutions from the rotation function analysis then were used in the translation search. The solution, a = 38.8, b = 56.6, c = 15.2, Tx = 0.47, Ty = 0.94, and Tz = 0.25 in space group P6₁22, gave the lowest residual (43.0 with the next lowest value being 54.0). Interestingly, although the asymmetric unit of the complex contains only a single liganded NP chain, no clear molecular replacement solution was obtained using only the NP-FY monomer as the search model.

Refinement

The structures of both des 1–6 BNP-II and its VK complex were refined by simulated annealing with X-PLOR (Jones et al. 1991) using the Engh and Huber stereochemistry library (Brunger et al. 1989). The free R-factor (Engh and Huber 1991) based on a random set of reflections was monitored throughout the refinement. After the initial round of positional refinement, omit maps (omitting 5% of the total number of residues) were used to manually adjust (Brunger 1992) the models as necessary. The models then were subjected to further rounds of positional refinement followed by a round of simulated annealing refinement. At this point, individual atoms were assigned isotropic B-factors, followed by several cycles of B-factor refinement. Solvent atoms were identified (peak heights above 4σ and good hydrogen-bonding geometry) using difference Fourier maps and included in the refinement.

The final model for des 1–6 BNP-II includes residues 7–86 of each chain of the NP dimer and 43 solvent molecules modeled as water. Residues 87–95 of each NP monomer were not observed in the electron density maps and are presumed to be disordered. The R-factor is 21.8% for all data between 23.0 Å and 2.4 Å resolution. The refined model has good stereochemistry with root mean square deviations in bond length and angles of 0.005 Å and 1.4°, respectively. Analysis of the Ramachandran plot (PROCHECK; Laskowski et al. 1993) showed no residues in disallowed regions.

The final model for the VK complex of des 1–6 BNP-II includes NP residues 7–85, residues 1–6 of the bound hormone, three Cd⁺² ions, and 36 solvent molecules modeled as water. Residues 86–95 of the NP monomer and residues 7–9 of the hormone were not observed in the electron density maps and are presumed to be disordered. The refined model has an R-factor of 23.4% for all data between 15.0-Å and 2.3-Å resolution and good stereochemistry with root mean square deviations in bond length and angles of 0.010 Å and 1.5°, respectively. Analysis of the Ramachandran plot (Laskowski et al. 1993) showed no residues in disallowed regions.

Details of the refinement for both structures are collected in Table 3. Atomic coordinates have been deposited with the Protein Data Bank (Bernstein et al. 1977). The file for unliganded des 1–6 BNP-II is 1 JK6 des 1–6 NP. The file for the VK complex is 1 JK4 NP-VP complex.

Distance calculations

Interatomic distances and binding-induced changes in atomic positions were measured using the Swiss Protein Data Bank modeling system and the MIDAS display system (Ferrin et al. 1988). In comparing the different subunits of the same system, small differences in interatomic distances occasionally were seen between the two subunits of the unliganded des 1–6 protein, but not in its symmetrical VK complex. Differences also were seen between the two subunits of the oxytocin complex and among the four subunits of the dipeptide complex. To handle these, we have made the simplifying assumption that such differences are imposed by the crystalline state and, unless otherwise specified, calculate each distance as the average of that in each of its subunits; distances involved in the central conclusions of this work are not affected significantly by this simplification. Binding-induced changes in absolute atomic positions were calculated by superimposing the structure of the unliganded state on that of each liganded state to obtain the best backbone fit and the distance between any atom in the two states directly measured. Figures utilized the MIDAS display system (Ferrin et al. 1988) and the Swiss Protein Data Bank Modeling system.

Abbreviations

• BNP, bovine neurophysin

• BNP-I, bovine neurophysin-I

• des 1-6 BNP-II, des 1-6 bovine neurophysin-II

• NP, neurophysin

• NP-FY, the complex of BNP-II with p-iodo-l-phenylalanyl-l-tyrosine amide

• VK, lysine vasopressin