Contributions of the interdomain loop, amino terminus, and subunit interface to the ligand-facilitated dimerization of neurophysin: Crystal structures and mutation studies of bovine neurophysin-I

Abstract

Current evidence indicates that the ligand-facilitated dimerization of neurophysin is mediated in part by dimerization-induced changes at the hormone binding site of the unliganded state that increase ligand affinity. To elucidate other contributory factors, we investigated the potential role of neurophysin's short interdomain loop (residues 55–59), particularly the effects of loop residue mutation and of deleting amino-terminal residues 1–6, which interact with the loop and adjacent residues 53–54. The neurophysin studied was bovine neurophysin-I, necessitating determination of the crystal structures of des 1–6 bovine neurophysin-I in unliganded and liganded dimeric states, as well as the structure of its liganded Q58V mutant, in which peptide was bound with unexpectedly increased affinity. Increases in dimerization constant associated with selected loop residue mutations and with deletion of residues 1–6, together with structural data, provided evidence that dimerization of unliganded neurophysin-I is constrained by hydrogen bonding of the side chains of Gln58, Ser56, and Gln55 and by amino terminus interactions, loss or alteration of these hydrogen bonds, and probable loss of amino terminus interactions, contributing to the increased dimerization of the liganded state. An additional intersubunit hydrogen bond from residue 81, present only in the liganded state, was demonstrated as the largest single effect of ligand binding directly on the subunit interface. Comparison of bovine neurophysins I and II indicates broadly similar mechanisms for both, with the exception in neurophysin II of the absence of Gln55 side chain hydrogen bonds in the unliganded state and a more firmly established loss of amino terminus interactions in the liganded state. Evidence is presented that loop status modulates dimerization via long-range effects on neurophysin conformation involving neighboring Phe22 as a key intermediary.

Elucidation of the mechanisms by which information is communicated over long distances within proteins is essential to a complete understanding of protein structure–function relationships. The facilitation or induction of protein dimerization by bound ligand, a process in part involving such communication and integral to many cell signaling pathways is, with some exceptions, poorly understood. As seen in the ligand-induced dimerization of epidermal growth factor (for example, see Dawson et al. 2005), the interactions involved can be very complex. The present study probes factors involved in the ligand-facilitated dimerization of a smaller system, the hormone-binding protein neurophysin (NP)¹ (for example, see Breslow and Burman 1990).

Neurophysins are a family of closely related proteins derived from either the oxytocin or vasopressin precursors (or analogous precursors in other species), and were among the first proteins for which ligand-facilitated dimerization was demonstrated (Nicolas et al. 1976, 1980). With one ambiguous exception, all ligands capable of binding to the neurophysin hormone-binding site—oxytocin, vasopressin, and related smaller peptides—similarly increase dimerization constants (Breslow et al. 1973, 1991; Fassina and Chaiken 1988). Given the importance of both hormone–NP interaction and NP dimerization to precursor stability (Barat et al. 2004), the linkage between binding and dimerization is likely to play a particularly significant role in folding in vivo.

The effect of ligand on NP dimerization is allosteric, as evidenced by the crystallographic demonstration of a clear separation between neurophysin's hormone-binding site and the dimer subunit interface (for example, see Rose et al. 1996; Wu et al. 2001). In a recent study of the weakly dimerizing H80E mutant of BNPI (bovine oxytocin-related NP), we presented evidence that dimerization of the unliganded protein is associated with changes at the hormone-binding site, supporting the view that differences in binding site conformation between unliganded monomeric and dimeric states, favoring binding to dimer, are significant contributors to the allosteric mechanism (Naik et al. 2005). The present study is aimed principally at exploring the contributions to ligand-facilitated dimerization of the neurophysin interdomain loop, a segment consisting of four or five residues that are distant from the dimer subunit interface, but that serve as a covalent bridge and potential hinge between the protein's amino and carboxyl domains. Because NP dimerization involves both domains, their relative orientation is important; changes in this orientation, controlled by an interdomain hinge, would have the ability to modulate dimerization (Eubanks et al. 2001; Nguyen and Breslow 2002).

We explored the role of the loop by examining the effects of loop residue mutation and of deletion of the six amino-terminal NP residues. Undefined pH-dependent intramolecular interactions of the first two residues of BNPII (bovine vasopressin-related NP) were first observed by 1D NMR (Lord and Breslow 1979) and subsequently shown to be released upon occupancy of the hormone-binding site (Zheng et al. 1997). Deletion of the first six residues of BNPII led to increases in dimerization and binding constants by factors of four and two, respectively, and, based on consideration of the crystal structure of WT liganded BNPII, it was suggested that the amino terminus might interact with the loop in the unliganded state (Zheng et al. 1997). In contrast, 1D NMR (Lord and Breslow 1979) did not indicate analogous interactions of the amino terminus of BNPI, the allosteric properties of which are similar, but possibly not identical, to those of BNPII (compare Kanmera and Chaiken 1985^; Fassina and Chaiken 1988^; Breslow et al. 1991). Thus, the potential importance of the amino terminus to NP allosteric effects was unclear.

More recently, however, NMR investigation of the structure of H80E (the H80E mutant of BNPI) demonstrated interactions of residues 3 and 4 with loop residue 55 in the unliganded monomeric state (Nguyen and Breslow 2002), an interaction now also seen by NMR in the unliganded WT BNPI dimer (H. Lee, M.T. Naik, C. Bracken, and E. Breslow, in prep.), suggesting that such interactions might modulate dimerization in both BNPI and BNPII, as we explore here. Interactions in the monomer NMR structure are also seen between the 1–6 sequence and both loop residue 56 and binding site residues 53 and 54, adjacent to the loop. Note that crystallography has provided little information about residues 1–6. These residues are largely undetectable (and assumed disordered) in complexes of full-length WT BNPII (Chen et al. 1991; Rose et al. 1996) and were deleted in other studies to improve data resolution (Wu et al. 2001).

Loop residues are strongly, perhaps completely, conserved among mammalian neurophysins (for example, see Chauvet et al. 1983). BNPI is chosen for these studies because of the body of preexisting NMR data on this system (for example, see Breslow et al. 1992; Nguyen and Breslow 2002; Naik et al. 2005) and our reliance on NMR to assess dimerization (Zheng et al. 1997). However, with the exception of the NMR structure of the unliganded H80E monomer (Nguyen and Breslow 2002), no BNPI structures have previously been solved. Instead, due to their easier crystallizability, BNPII crystal structures have provided the basis to date of structure-based insights into the effects of NP mutation. Crystal structures have respectively been solved for dimeric WT BNPII bound to the dipeptide FY and to oxytocin, and of des 1–6 BNPII both in the unliganded dimeric state and in the liganded dimeric state bound to vasopressin (Chen et al. 1991; Rose et al. 1996; Wu et al. 2001). Nonetheless, because BNPI and BNPII differ in ∼20% of their sequences (▶) and no structure of dimeric BNPI has been available, and because of the effects of loop mutations reported below, the present investigation also includes elucidation of the structures of des 1–6 BNPI and its Q58V mutant in their liganded dimeric states, as well as the structure of desBNPI (as its F91STOP mutant) in its unliganded dimeric state. Discussion of these structures here will principally focus on features relevant to allosteric mechanism.

Figure 1.

Amino acid sequences of BNPI and BNPII. With the exception of the F91 STOP mutants, our BNPI constructs terminate at Ser92 (Eubanks et al. 1999). Interdomain loop residues 55–59 are shown in bold. Residues at the subunit interface in the unliganded state are underlined. Binding site residues are italicized. Note that residue 82 is not an interface residue, although inadvertently listed as such in an earlier reference (Naik et al. 2005).

Results

Effects of mutation on folding

Under standard folding conditions, all BNPI mutants in this study exhibited normal folding efficiencies with the exception of S56E and K59G, for which efficiencies of both were reduced by ≥50% (see Materials and Methods). In the case of the S56E mutant, the principal cause of the problem is a major reduction in peptide-binding affinity (▶), so that the folded state is not stabilized by bound ligand under folding conditions as normally (with only one known exception) required for efficient NP folding (for example, see Eubanks et al. 2000). However, because both the binding and dimerization properties of the folded K59G mutant were not significantly weaker than those of many mutants that did fold normally (▶)—particularly those of the closely related K59A mutant—and because neither CD nor NMR indicated any significant conformational differences from normally folded protein, a folding pathway defect was suspected. Consistent with this, changing the folding procedure (see Materials and Methods) from the β-mercaptoethanol procedure in which the thiol catalyzing disulfide rearrangement is air-oxidized during folding, to the glutathione buffer procedure, which does not involve thiol oxidation and thereby allows a longer time for rearrangement of protein disulfides, resulted in normal folding efficiency. A potential explanation of the pathway defect is provided in the Discussion.

Table 1.

Effects of mutations on peptide binding and on dimerization^a

Effects of mutations on binding and dimerization constants and on allosteric mechanism

▶ summarizes the effects of the mutations studied on peptide-binding and dimerization constants. For illustrative purposes, examples of binding and dimerization data contributing to several key values in ▶ are shown in ▶ and ▶, respectively. Data previously reported for G57 mutants (Eubanks et al. 2001) are included for the sake of completeness. Although significant mutation-induced effects on binding and/or dimerization constants are present for most mutants, all of the mutants in ▶, with the exception of S56E, for which no binding could be demonstrated, retained the property of ligand-facilitated dimerization; i.e., addition of peptide markedly increased the fraction of dimer, as shown for the strongly dimerizing des 1–6 Q58V double mutant in ▶. Nonetheless, the high dimerization constant of the liganded state prevented quantitative determination by NMR of the effects of mutation on this value; protein concentrations allowing partial dissociation of the liganded dimers were too low for accurate measurements of monomer/dimer ratios.

Figure 2.

Scatchard plots of binding the peptide phenylalanyl-tyrosine amide to WT BNPI (open circles), the single mutants Q58V (filled triangles) and Q55A (filled squares), and the double mutant Q58V/Q55A (filled circles). Ordinate units are inverse molarities (M⁻¹). Conditions: ∼0.1 mM protein, pH 6.2, 25°C. Binding constants in units of inverse molarity for these specific experiments calculated from the data as described in Materials and Methods are: Q55A (2480), WT (14,860), Q58V/Q55A (17,780), and Q58V (64,400). Note the curvature of the data obtained at values of <0.5 for the double mutant as described in Materials and Methods, which, in this case, would not have significantly influenced the calculated binding constant.

Figure 3.

NMR spectra of WT BNPI and its des 1–6 and desQ58V mutants at ∼0.2 mM concentration illustrating the determination of dimerization constants. Signals listed as M and D are the monomer and dimer signals respectively of the α-proton signal of Cys28. The signal marked with an x is assigned as noise or a trace contaminant based on its absence in other spectra of the same proteins and is not included in calculations. (Top) WT BNPI, 0.23 mM, M/D = 1.01 ± 0.08, calculated dimerization K = 4350 M⁻¹. (Middle) des 1–6 BNPI, 0.16 mM, M/D = 0.84 ± 0.01, calculated dimerization K = 8200 M⁻¹. (Bottom) desQ58V, 0.23 mM, M/D = 0.425, calculated dimerization K = 17,950 M⁻¹. Note that calculated dimerization constants apply only to the individual spectra shown and variably deviate to some degree from the composite results reported in ▶.

Figure 4.

Effect of binding the dipeptide FF to the hormone-binding site of the des 1–6 Q58V mutant of BNPI as monitored by 1D proton NMR. The chemical shifts of Cys28 α-proton signals in monomeric (M) and dimeric (D) states are indicated; note that these are essentially independent of the state of protein ligation (for example, see Barat et al. 2004). FF is chosen for NMR binding studies because its signals in free and bound states do not obscure the Cys28 signals; the binding constant of FF is similar to that of FY. (Lower spectrum) 0.1 mM protein, pH 7. (Upper spectrum) Same sample after addition of 0.5 mM FF. The protein is saturated with peptide under these conditions. Note the complete loss of both the Cys28 monomer signal and the 6.6 ppm signal in the presence of peptide.

Some of the reductions in binding affinity and/or dimerization reported in ▶ can be explained in the context of the previously reported BNPII crystal structures (see Discussion). Additionally, the observed mutation-induced increases in both parameters associated with deletion of residues 1–6 parallel those previously seen in BNPII (Zheng et al. 1997), assisting their interpretation. However, origins of the increases in one or both parameters seen in the Q55A, S56A, and Q58V, A and G mutants required more detailed investigation (see Discussion). Of the Q55 and Q58 mutations, the only large increases seen simultaneously in both parameters accompany the Q58V mutation, which by itself increased the binding constant by a factor of four to five when introduced into either the WT protein or the des 1–6 WT protein, and increased the dimerization constant by factors of approximately two and three in the WT and des 1–6 WT proteins, respectively.

Note in ▶ that the thermodynamic effects of the Q58V, Q55A, and des 1–6 mutations exhibit variable additivity with each other and that the degree of additivity is not necessarily the same in a given double mutant for both binding and dimerization. For example, the effects on binding of the Q58V and Q55A mutations are almost completely additive, but their combined effects on dimerization exhibit extreme nonadditivity. The des 1–6 mutation is completely additive with the Q58V mutation when binding is measured, but is markedly nonadditive with the Q55A mutation, while slightly potentiating the effects of both these mutations on dimerization.

Table 2.

Degree of additivity of the contributions of key mutants to the free energies of binding and dimerizations^a

Properties of the Q58V mutant were particularly difficult to explain by any of the BNPII crystal structures (see above) or by the structure of the H80E monomer. To explore the possibility that its increased binding constant reflected altered interactions between the protein and peptide in the liganded state and to elucidate factors contributing to dimerization in the unliganded state, crystal structures were needed of the FY complexes of WT BNPI and its Q58V mutant and of unliganded dimeric WT BNPI.

Crystal structures of the FY complex of desBNPI and of unliganded desBNPIF91STOP

The BNPI crystals most suitable for X-ray diffraction analysis were those of its des 1–6 derivative in the liganded state and its des 1–6 F91STOP mutant in the unliganded state. Phe91 is close to the carboxyl terminus (▶); the F91 STOP mutation increased the dimerization constant by 50%–90% when either the full-length or des 1–6 derivatives were compared, but was otherwise benign (▶). (The dimerization increase is of undetermined origin since, as with BNPII, residues beyond 86 or 87 are not seen in the crystal structures.) ▶ shows the crystal structure of the asymmetric unit of desBNPIfy. Like the crystal structure of BNPIIfy (and unlike complexes of BNPII with hormone), the asymmetric unit of the unit cell contains more than one dimer. In the case of BNPIIfy, the two dimers interact directly via weak interactions (Chen et al. 1991). In desBNPIfy, the two dimers (AB and CD) are linked together indirectly by a single orthogonally bridging dimer subunit (E) from an adjacent asymmetric unit, with contacts to E involving subunits A and D. The E subunit appears to be a crystal phenomenon and is not considered in our analyses. However, despite the crystal packing and sequence differences between BNPI and BNPII (▶) and despite the fact that residues 1–6 are absent in the BNPI structure, the conformations of the individual chains of desBNPIfy and BNPIIfy are basically similar, as shown by a comparison of their backbones (▶).

Figure 5.

Asymmetric units of liganded and unliganded derivatives of BNPI. (Left) des BNPIfy. The bound dipeptides are shown as stick structures. (Right) desBNPIF91STOP in the absence of peptide. Lower and upper dimers for both structures are AB (yellow and green chains) and CD (blue and orange chains), respectively. E subunits are shown in light purple. Structures were visualized by PyMOL (http://www.pymol.org).

Figure 6.

Comparison of the backbone structure of desBNPIfy (black) with that of WT BNPIIfy (white) and desBNPIF91STOP (white). Positions of the interdomain loop and amino (N) and carboxyl (C) termini are indicated. Residues 1–6 of BNPIIfy were deleted from the comparison to allow the superposition. (Left) Superposition of desBNPIfy and BNPIIfy. (Right) Superposition of desBNPIfy and desBNPIF91STOP.

More detailed comparisons of the structure of desBNPIfy with that of either the BNPII dipeptide complex or the des 1–6 BNPII vasopressin complex do, however, indicate a number of differences in side chain positions (data not shown). At least some of these represent external loop residues probably influenced by crystallization conditions; most are also otherwise of uncertain significance given the similar binding, dimerization, and allosteric properties of the two proteins in the neutral pH region (for example, see Nicolas et al. 1980; Breslow et al. 1991). These functional similarities are clearly reflected by shared structural features. For example, residues involved in binding dipeptide are identical in the two proteins with the exception of residue 76, which is Asp in BNPII and Pro in BNPI (▶). Nonetheless, despite the residue 76 substitution, all binding site residues interact similarly with ligand in the two proteins, including the Asp and Pro side chains of residues 76, which both interact weakly with a common region of the dipeptide tyrosine.

Residues directly participating in dimerization in both complexes are also, with one significant exception, the same in sequence position, albeit not necessarily in identity (▶)—principally 32–38, 40, 72, and 77–81—and show similar modes of interaction in the two neurophysins. The significant exception, which remains true on comparison of desBNPIfy with all BNPII complexes of known structure, lies in the finding that Ser25 forms a hydrogen bond across the interface only in the BNPI complex, an apparent consequence of the difference between BNPI and BNPII in interface residues 77 and 81 (▶). In both liganded and unliganded BNPI and BNPII dimers, the backbone –NH of 81 is hydrogen bonded across the interface to the carbonyl oxygen of 77 of the partner subunit. However, an additional intersubunit hydrogen bond is present in the liganded state that in both cases involves the residue 81 side chain.

In BNPII complexes, the –OH of Thr81 hydrogen bonds to the carbonyl oxygen of 77 of its partner subunit, representing a second hydrogen bond to this oxygen (Wu et al. 2001). The same bond cannot form at neutral pH to the side chain of Glu81 of BNPI. Instead, in three of the four chains of the two complete desBNPIfy dimers, the carboxyl group of 81 hydrogen bonds across the interface to the –OH of Ser25 (▶). The absence of this hydrogen bond in unliganded desBNPI is evidenced by a lack of visible electron density of the residue 81 side chain in any of the four dimer subunits, indicating disorder, as well as by structure refinement. RMSD comparison of the subunit interfaces of liganded and unliganded states indicates that formation of this hydrogen bond represents the largest ligand-induced interface change. For example, after structure refinement, the average RMSD between liganded and unliganded states of the side chains of all interface residues in the AB dimer, exclusive of residue 81, is 0.8 ± 0.4 Å, while that for residue 81 is 2.7 ± 0.3 Å. Amino domain interface changes are particularly small.

Figure 7.

Hydrogen bonding of residues 25, 77, and 81 across the subunit interface of the AB dimer in the liganded state. Dashed lines indicate hydrogen bonds. Subunit identity is given in parentheses.

▶ also shows the asymmetric unit of the crystal structure of the F91STOP mutant of unliganded desBNPI. Like the liganded structure, it contains five subunits, representing two dimers (AB and CD) and one dimer subunit (E). In this case, the AB and CD dimers are in direct contact, with interactions principally involving subunits A, B, and C. The E subunit (which is again not considered in our analyses) contacts only the CD dimer, largely through van der Waals interactions with both subunits. The backbone structure of unliganded desBNPIF91STOP, as determined by crystallography, is shown in ▶ and differs significantly from that of the liganded state in the position of the 55–59 loop, as also previously observed on comparison of liganded and unliganded structures of des 1–6 BNPII (Wu et al. 2001). The nature and significance of the 55–59 loop differences will be discussed below in the context of the effects of mutation. The other major differences between the two structures lie in the region encompassing residues 44–54 and reflect ligand-induced change in binding site residues (e.g., ▶) that in turn contribute to the ligand-induced changes in the adjacent interdomain loop. The connection of the loop to the binding site is an integral component of the allosteric mechanism.

Crystal structure of the dipeptide complex of the Q58V mutant of des 1–6 BNPI

Despite the effects on binding and dimerization of the Q58V mutation, the crystal structure of the dipeptide complex of desQ58V (including the arrangement of subunits within the asymmetric unit) is very similar to that of the corresponding des1–6 WT protein, as particularly exemplified by RMSD comparison of the structures of their respective A subunits (▶). Although additional residues show high RMSD values when other subunits are compared, almost all such residues (including those in subunit A), with the exception of the 77–81 region, represent solvent-exposed residues of no known function in the folded protein. In the rare instances where binding site residues are involved, deviations from the WT protein appear unlikely to contribute significantly to the higher binding constant of the mutant. However, differences in side chain orientation for residues 77 and 81, which interact with each other across the interface via main chain hydrogen bonding (see above), are seen in all subunits (e.g., ▶) and in part represent differences between the two protein complexes in side chain hydrogen bonding in this region of the subunit interface. Specifically, the intersubunit hydrogen bond between Glu81 and Ser25 side chains (see above) is present in only two of the four subunits of the two complete des Q58Vdimers—replaced in one chain by a hydrogen bond between a carboxyl oxygen of Asp77 and a ring –NH of His80 on its partner chain (see below).

Figure 8.

Superficial similarity between the crystal structures of the FY complexes of desBNPI and its Q58V mutant. (Top) Region (residues 50–61) surrounding residue 58 in the WT (black) and mutant (gray) proteins. Chain runs from left to right. Note similar side chain positions in the two proteins for residue 58 and apparent similarity in backbone structure. (Below) RMSD comparison of the A subunit structures of the two proteins. Ordinate is Ångstrom values for backbone residues (diamonds) and side chains (squares). RMSD values of 0 for some side chains indicates Gly residues.

The ultimate origin of any differences is inevitably the site of the mutation, which shows relatively trivial differences between the two complexes on superficial inspection (▶); i.e., despite their differences in hydrophobicity, both the Val and Gln side chains at position 58 are similarly externalized in their complexes and the neighboring backbone conformations are largely similar. However, analysis of hydrogen-bonding patterns indicates that the Gln to Val mutation leads to subtle differences in hydrogen-bonding interactions between the loop and carboxyl domain β-sheet segments, as evidenced by the differences in hydrogen bonding to the terminus of Arg66 (▶).

Table 3.

Comparison of hydrogen bonding interactions between loop residue atoms and the terminus of Arg66 in the dipeptide complexes of des1–6 BNP1 and its Q58V mutant

Conformational effects in the unliganded state of mutation of Gln58 and Gln55 and of deletion of residues 1–6

1D NMR spectra of the dipeptide complexes of the WT and Q58V mutants or of their des 1–6 derivatives provided no obvious evidence of differences between the WT and mutant proteins in the liganded state (data not shown). However, at pH 6.2 in the unliganded state, the Q58V mutant of the full-length protein showed a discrete signal at ∼6.6 ppm not present in the WT protein under similar conditions (▶). The intensity of the signal and its response to concentration was almost identical to that of the 6.4 ppm dimer signal from Cys28, indicating its association with dimer; its chemical shift was independent of pH in the range 6–8. A similar signal was present in the unliganded dimers of Q58A, G and D mutants (data not shown) and was retained in the unliganded double mutant des 1–6Q58V (▶, ▶), but was absent in the unliganded double mutant Q55A/Q58V (▶), as well as in the liganded state (▶) and in mutants (liganded or unliganded) representing single mutations at other positions. The signal was abolished by nitration of the sole protein tyrosine, Tyr49 (▶), while 2D COSY spectra of the unnitrated protein indicated its connectivity to the Tyr 2,6 ring proton region to which the 3,5 ring protons of Tyr49 connect (▶). The data assign the 6.6-ppm signal to Tyr49 ring protons originating from a new dimer conformer (see Discussion). Changes near 0.4 ppm also accompanied the Q58 mutations, but were not unambiguously assigned.

Figure 9.

Evidence that the 6.6 ppm signal of Q58 mutants is from Tyr49. (Top) DQF-COSY spectrum of the Q58V mutant. (Bottom panel, spectra from top to bottom), 1D proton spectra of Q58V mutant; WT protein; double mutant des 1–6,Q58V; and Q58V nitrated at Tyr49. (Nitration moves Tyr49 ring protons downfield from 7 ppm to the Phe ring proton region, which is not shown in the 1D spectra.) In the DQF-COSY spectrum, the principal signal from the 3,5 ring protons of Tyr49 is located at ∼6.75 ppm and shows connectivity solely to the 2,6 ring protons located at 7.13 ppm. The 6.6 ppm signal shows connectivity only to 7.09 ppm, the latter representing the 2,6 protons of the same ring also upfield shifted relative to the WT protein. Protons downfield from the Tyr49 protons in the COSY spectrum are from Phe residues and show no connectivity to the 6.6 ppm signal. Signals at ∼6.4 and 6.15 ppm are the dimer and monomer signals, respectively, of the Cys28 α-proton.

Figure 10.

Representative NMR spectra showing the effect of the des 1–6 mutation on the WT protein and the effect of the Q55A mutation on the Q58V mutant. (Top to bottom) WT BNPI; des 1–6 BNPI; Q58V mutant of BNPI; and the double mutant Q55A/Q58V. Arrow points to 6.62 ppm.

A different additional Tyr49 signal near 6.6 ppm was exhibited in the unliganded dimeric state by the des 1–6 derivatives of both WT protein and mutant proteins (▶). At pH 6.2, the signal was manifest as a very weak concentration-dependent peak at ∼6.65 ppm on or just upfield from the main Tyr49 3,5 proton signal (located at ∼6.75 ppm). However, at slightly higher pH values, it moved further upfield so that it appeared as a more discrete peak that was similar to, but somewhat less resolved than, the 6.6 ppm signal seen in Q58 mutants (see Discussion). Evidence suggestive of an effect of the des 1–6 mutation on conformation was also seen by CD, the mutation increasing the conformationally sensitive 245/280 nm disulfide band ratio (see Materials and Methods) by ∼20% (data not shown).

Evidence of yet a different mutation-induced conformational change was seen in the case of the Q55A mutant, the NMR spectra of which showed two new peaks in the unliganded state, one at ∼0.65 ppm and another at ∼7.0 ppm (data not shown), the 0.65-ppm peak also seen in the liganded state. Both signals were lost upon nitration of Tyr49, consistent with a tentative assignment of the downfield peak to Tyr49, but the upfield peak remains ambiguous. Most clearly indicative of a conformational effect of the Q55A mutation is that, as noted above, the 6.6 ppm peak of the Q58V mutant is absent in the Q55A/Q58V double mutant (▶).

Discussion

The ligand-induced increase in BNPI dimerization involves an approximate 100-fold increase in dimerization constant (for example, see Kanmera and Chaiken 1985), the other thermodynamic face of which is an approximate 10-fold tighter binding of hormone (or related ligand) to a dimer subunit than to an isolated monomer (for example, see Breslow et al. 1991). Values for the effect of ligand on BNPII dimerization constant range from the same as that for BNPI (Breslow et al. 1991) to ∼40% of that value (compare Kanmera and Chaiken 1985^; Fassina and Chaiken 1988). Considerable evidence, both from kinetics—which indicate faster binding to dimer than to monomer (Pearlmutter and Dalton 1980)—and from NMR structural studies (see Introduction), argues that differences between unliganded monomeric and dimeric states in NP conformation contribute to the increased binding affinity of the dimer. While such differences could, in principle, be sufficient in themselves to also account for the increased dimerization constant of the liganded state, the present results will be shown to represent evidence that ligand-induced changes involving the interdomain loop—particularly those leading to changes in the hydrogen bonding of residues 58, 56, and 55 and in interactions of the loop (and possibly also residues 53–54) with the amino terminus—potentially contribute to the ligand-induced increase in dimerization. Results will be interpreted in the context of the BNPI crystal structures, which additionally provide new evidence for a role for ligand-induced changes at the subunit interface as part of the allosteric mechanism. Effects of loop mutations not obviously related to allosteric mechanism will be discussed solely to the extent that they provide new insights into neurophysin properties.

Crystal structures of liganded desBNPI and its Q58V mutant and of unliganded desBNPIF91STOP: Ligand-induced changes at the subunit interface

Because interpretation of much of the data obtained rests on the crystal structures of liganded and unliganded BNPI, we point out here that significant differences are often seen among the individual chains of the crystals’ multichain asymmetric units, which we discuss as relevant. These differences are likely to reflect both crystal packing forces and the heterogeneity of monomer and dimer conformers in solution, the latter demonstrated by NMR in both unliganded monomeric and dimeric states (Breslow et al. 1992; Nguyen and Breslow 2002). Note that available data on the bovine neurophysins (for example, see Nicolas et al. 1980) argue against the presence in solution of any unliganded states higher than dimer under the conditions used here and that we have recently confirmed this by analytical ultracentrifugation for the strongly dimerizing des 1–6 Q58V mutant. This is also true for studies of the liganded state by ultracentrifugation (for example, see Nicolas et al. 1980), and fluorescence anisotropy studies have suggested that only trace quantities of liganded oligomers higher than dimer are present at 0.1 mM (Breslow et al. 1991), the upper concentration limit for most binding studies here (▶). Also, because we assume the lack of such higher oligomers in solution, the structures of E subunits, which are likely present only in such ensembles, are omitted from the present analyses, as also noted above.

The elucidation of the crystal structure of desBNPIfy provides the first structure of this protein in the liganded state, where, in regions important to its physiological role, it strongly resembles structures of liganded BNPII, with which it shares extensive sequence and functional similarity (see above). The most notable difference between the two structures so far in fact reflects an underlying similarity. That is, despite the differences between the two proteins in interface residues 77 and 81 (▶), an additional hydrogen bond involving the side chain of residue 81 forms across the interface in the liganded state of both proteins, albeit with different hydrogen bond partners. The binding-induced formation of a new interface hydrogen bond from Thr81 in BNPII was previously noted as a potential contributor to the increased dimerization constant of the liganded state, but, given that the same bond was not possible in BNPI, its importance was uncertain (Wu et al. 2001). The formation of a different ligand-induced interface hydrogen bond by residue 81 in BNPI increases the likelihood that this mechanism contributes significantly to the increased dimerization constant of the liganded state in both neurophysins.

In further support of the significance of this hydrogen bond, the finding that it is present in only three quarters of the chains of the two desBNPIfy dimers is paralleled by a similar situation in the dipeptide complex of BNPII, the asymmetric unit of which also contains two dimers (see above). Even in the Q58V mutant of desBNPI—which required different crystallization conditions from the WT protein, and which contains only two of the three ligand-induced hydrogen bonds between Ser25 and Glu81 found in the WT protein—a different ligand-induced interface hydrogen bond is present in a third subunit, so that the total number of ligand-induced new interface hydrogen bonds is the same. Nonetheless, the universality of this mechanism in other neurophysins remains to be demonstrated, since several do not contain polar residues in position 81 (for example, see Chauvet et al. 1983). Moreover, although formation of this hydrogen bond is the largest ligand-induced interface change, it is not the only one. Multiple small interface adjustments of unmeasured significance accompany ligand binding, as evidenced by RMSD comparison of liganded and unliganded BNPI (see above) and as also reported for BNPII (Wu et al. 2001).

Effects of mutation of Lys59: Role of residue 59 in folding

Lys59 joins the interdomain loop to the carboxyl domain, its carbonyl oxygen involved in the first β-sheet hydrogen bond of that domain. The present studies indicate a significant role for Lys59 in the folding pathway and represent the first mutation-induced folding pathway defect seen in NP. The fact that the K59G mutant, but not the K59A or other mutants, required a longer equilibration time with thiol than WT protein to fold normally (see Results) strongly suggests a role for position 59 in guiding the folding path of the carboxyl domain—the smaller range of phi, psi angles preferentially experienced by Ala or Lys residues than by Gly residues most likely reducing the frequency or stability of “incorrect” disulfide partners during folding, decreasing the time needed for disulfide equilibration to the correct structure.

Mutation of Gly57: Effects of steric hindrance and phi, psi angles on binding and dimerization

G57S and G57R mutations of human vasopressin-related NP are a cause of familial neurogenic diabetes insipidus (Ito et al. 1991; Rittig et al. 1996). Effects on peptide binding of mutating residue 57 to Ser or Arg in BNPI (▶) have previously been discussed in the context of BNPII crystal structures and assigned to steric hindrance by the mutated residues in the liganded state (Eubanks et al. 2001). The BNPI crystal structures support this explanation of the binding effects but, as with BNPII structures, predict only minor steric hindrance effects of the G57S mutation on the unliganded dimer. The similar effects on dimerization of the Ser and Arg substitutions (▶) also argue against a strictly steric hindrance effect. These considerations and the sensitivity of dimerization to loop conformation as demonstrated by the effects of other loop mutations (see below) suggest that changes in loop phi, psi angles or flexibility, arising solely from the substitution of Gly57 by amino acids with more restricted conformations, might account for the effects on dimerization of mutation at this position.

Mutation of Ser56: Effects of side chain hydrogen bonding

Residue 56 mutants were chosen to explore a potential functional role of Ser56 side chain hydrogen bonding as well as the earlier observation that succinylation of Ser56 removed the concentration dependence of binding, implying a loss of ligand-facilitated dimerization (Huang et al. 1993). The 50% decrease in binding constant accompanying mutation of Ser56 to either Gly or Ala (▶) is consistent with the fact that, in the liganded state, the –OH of Ser56 is inaccessible to solvent and is hydrogen bonded in all subunits of both BNPI and BNPII to the carbonyl oxygen of Cys21 (▶). In contrast, Ser56 is solvent accessible in the unliganded dimer and hydrogen bonds between Ser56 and Cys21 are absent, replaced in 50% of subunits by hydrogen bonding of the Ser –OH group to Gln58 (▶). The absence of the –OH in the Gly and Ala mutants therefore leaves the Cys21 oxygen without a hydrogen bond in the nonpolar environment of the liganded state, but has a less destabilizing impact on the unliganded state because of the smaller number and greater solvent accessibility of Ser –OH hydrogen bonds to other residues. In fact, the unliganded dimer gains stability relative to monomer by deletion of the Ser56 –OH group. The dimerization constant of the S56A mutant is increased relative to WT in the unliganded state, suggesting that the few Ser56 –OH hydrogen bonds in the unliganded WT dimer destabilize the unliganded dimer relative to monomer. This is confirmed and its significance discussed below in the context of the role of Gln58.

Table 4.

Hydrogen bonding and significant van der Waals interactions of loop residue side chains^a

The S56E mutant was selected for investigation in an attempt to duplicate and further explore the effects of hydroxyl group succinylation (see above). The mutation (▶) decreased the dimerization constant in the absence of peptide by ∼50%, an effect similar to that of succinylation (Eubanks et al. 2001), but decreased binding affinity to an extent even greater than the 95% produced by succinylation, precluding measurement of the extent of allostery in the mutant. The effect on binding is possibly the consequence of a deeper burial of the negatively charged Glu carboxyl than of the hydroxyl-esterified succinyl group in the nonpolar environment of the bound state, but a 30% lower than normal 245/280 nm disulfide CD ratio for this mutant (see Materials and Methods) suggests an altered conformation as well.

Effects of mutating Gln58: Evidence for a role in allosteric mechanism

Comparison of BNPI and BNPII crystal structures indicates that the environment of Gln58 similarly identifies the state of dimer ligation in both proteins. (Investigation of the liganded monomeric state is so far precluded by its high dimerization constant.) In the unliganded dimeric states of both proteins, the Gln58 carboxamide faces the protein interior, in relatively close proximity (∼5 Å) to the ring edge of Phe22, as shown specifically for BNPI in ▶. This is even more the case for the H80E mutant in its unliganded monomeric state (Nguyen and Breslow 2002; H. Lee, M.T. Naik, C. Bracken, and E. Breslow, in prep.), where the Gln58 terminus is only ∼3.5 Å from the Phe22 ring edge. However, in unliganded dimeric BNPI and BNPII structures, but not thus far seen in the unliganded H80E monomer, the Gln58 carboxamide nitrogen is hydrogen bonded to the backbone oxygens of Ser56 and Gly57, while its carboxamide oxygen hydrogen bonds in one des BNPI subunit to the Ser56 hydroxyl (▶; ▶). In contrast, the Gln58 side chain projects into the solvent in the liganded dimeric state of both proteins, its carboxamide now ∼8–9 Å from the Phe22 ring and not hydrogen bonded to another residue (▶; ▶).

Figure 11.

Effect of binding ligand on the environment and interactions of the Gln58 carboxamide in des 1–6 BNPI. (Left) The unbound state showing proximity of the Gln58 carboxamide to the ring of Phe 22 in the protein interior, and hydrogen bonding of the carboxamide to the O of Gly57, and the –OH of Ser56. (Right) The bound state showing movement of the Gln58 carboxamide away from Phe22 and the loss of its hydrogen-bonding interactions. In the bound state, the distance between the Gln58 NE2 and Gly57O has increased to 5.46Å and the distance between the 58OE1 and 56OG has increased to 11.58Å.

It had initially occurred to us that, if this change in solvent exposure of Gln58 was a simple transition from the protein interior to the protein exterior, it should be reflected in binding constant differences between hydrophilic and hydrophobic position 58 substitutions, the former having higher binding constants because of a more favorable free energy of transfer to water. ▶ indicates that this is not the case. The most hydrophobic residue, Val, yielded a mutant with the highest binding constant. Of particular interest is that all Q58 mutants, with the exception of the Asp mutant, have dimerization constants approximately double that of the WT protein.

To explain the effect of these mutations on dimerization, we note that all of the Q58 mutations have a similar effect on the conformation of the unliganded dimer (see Results). Moreover, what they also have in common is that they either lack side chains capable of hydrogen bonding (Gly, Ala, Val mutants) or (in the case of the Asp mutant) of forming the same hydrogen bonds that the Gln58 terminus forms in the WT unliganded dimer. These findings suggest that the Gln58 hydrogen bonds in the unliganded dimer constrain its conformation and reduce its stability advantage over monomer, loss of these hydrogen bonds on average doubling the dimerization constant of the Gly, Ala, and Val mutants and changing the conformation of the unbound dimer. This interpretation is strongly supported by the increased dimerization constant of the S56A mutant (see above), which has similarly lost the hydrogen bond between the Ser hydroxyl and Gln58 terminus and additionally has lost a hydrogen bond between the Ser56 hydroxyl and the backbone N of Gln58 that might also destabilize the dimer (▶). (The fact that the dimerization constants of the S56G and Q58D mutants are not obviously increased relative to WT [▶] suggests negative effects on dimerization of the increased flexibility associated with the S56G mutation and of electrostatic repulsion arising from the internalized carboxyl of the Q58D mutant.) Most importantly, since the Gln58 hydrogen bonds are also lost when Gln58 is externalized by ligand binding, the results suggest that the ligand-induced conformational change in Gln58 contributes a factor of approximately two to the ligand-induced 100-fold increase in dimerization constant.

The origins of the high binding constant of the Q58V mutant are not definitively established by the present studies, but the data suggest that the explanation lies principally in the properties of the liganded state, as opposed to a mutation-induced destabilization of the unliganded state. The binding difference cannot be explained solely by the NMR difference between the WT protein and the Q58V mutant in their conformations in the unbound state, since this difference and the accompanying higher dimerization constant is shared by other Q58 mutants (Q58A and Q58G) that do not have high binding constants. It also seems unlikely that the single difference between the two proteins in dimer interface hydrogen bonding in the bound state accounts for their different binding properties. The Q58V mutation similarly increases the binding affinities of WT and des 1–6 BNPI, which were determined at pH 6.2 and 7.4, respectively (▶). However, the His80 ring should be a significantly stronger hydrogen bond donor in the protonated state than in the unprotonated state—all the more so in this case because the hydrogen bond involves the carboxyl oxygen of Asp77—and, at least in the WT protein, has a pK value of 6.87 in the unliganded state (Griffin et al. 1975). A significantly lower effect of the mutation on binding by the des 1–6 protein than on WT would therefore have been expected.

We suggest instead that the stronger binding by the mutant might result directly from the effects of the mutation on loop backbone hydrogen bonding (▶), the increased number of hydrogen bonds in the mutant—if not compensated for by differences elsewhere in the protein—having the potential to directly increase the stability of the liganded mutant relative to that of the liganded WT protein. This suggestion does not in itself explain why the binding constants of the Q58A and G mutants are not similarly elevated. However, preliminary modeling suggests that the answer might lie in a less restricted local loop conformation in the bound state for these mutants (as well as for WT) relative to that in the branched chain Val mutant, providing greater entropy but diminishing the enthalpic stability of adjacent hydrogen bonds.

Effects of deletion of residues 1–6: Potential allosteric role of the amino-terminal tail

▶ demonstrates that, as with BNPII (Zheng et al. 1997), removal of the amino-terminal six residues in BNPI increases binding and dimerization, both by a factor of approximately two in the WT protein. Thus, interactions between the amino terminus and the 53–59 sequence reduce the binding and dimerization constants of the unliganded state. Because such interactions are known to be lost in the liganded state of BNPII (see above), where excision of residues 1–6 leads to a fourfold increase in dimerization constant, the results argue that, in the case of BNPII, loss of amino terminus interactions contributes a factor of approximately four to the ligand-induced increase in dimerization constant. A caveat is that, in contrast to the effects of excision, residues 1–6 remain covalently attached to liganded full-length NP. However, in crystal structures of complexes of WT BNPII (see Introduction), residues 1–6 are unresolved in most chains, suggesting that they are disordered. In chains where they are resolved, no interactions with other regions of the protein are seen. A significant effect on dimerization of the covalent attachment of residues 1–6 in the liganded state therefore seems unlikely.

In the case of BNPI, no firm evidence was obtained for loss of interactions involving the amino-terminal tail in the liganded state, in part because the crystal data represent only the des 1–6 protein. However, the crystal data indicate an average ligand-induced increase of ∼2 Å in the distance between residues 55 and 7, consistent with such a loss. Moreover, the fact that, at least in the monomer structure (see above), these interactions involve binding site residues 53 and 54 in addition to loop residues increases the likelihood of their loss in the liganded state. The lack of more definitive evidence in part reflects the lack of influence of these interactions on 1D NMR spectra of unliganded BNPI (see above), a probable result of the difference in amino-terminal sequences of BNPI and BNPII (▶) and the consequences of this for the nature and spectral effects of the interaction. The data therefore allow, but do not confirm, a contribution of a factor of approximately two from loss of amino terminus interactions to the ligand-induced increase in BNPI dimerization.

Long range nature of the effects of Gln58 mutation and of deleting residues 1–6

Both Gln58 mutations and deletion of residues 1–6 lead to changes in Tyr49 NMR spectra in the unliganded dimeric state. Although the proximity of Tyr49 to the 1–6 sequence is unknown in this state, the distance of closest approach between Tyr49 and Gln58 is ∼20 Å. The chemical shifts of the altered Tyr49 signals and the pH dependence of that in the des 1–6 derivatives suggest that they represent a subset of previously observed conformers of Tyr49 in unliganded dimeric BNPI, found to differ in WT and des 1–8 proteins and thought to be influenced by the titration of His80 (Breslow et al. 1992), the latter ∼15 Å from Tyr49. In all, these data indicate that effects on the unliganded dimeric state of Gln58 mutation and of excision of the amino terminus represent long-range effects on protein conformation. They additionally suggest that effects on Tyr49 NMR signals upon excision of residues 1–8, originally attributed to the loss of Arg8 (Peyton et al. 1986; Breslow et al. 1992), arise at least in part from the loss of residues 1–6.

The mechanism by which mutation or other change in status of residue 58 affects dimerization is likely to involve Phe22 (e.g., ▶), which contacts residues in both the amino and carboxyl domains that form paths of noncovalent linkages to the subunit interface. This suggestion is supported by the previously demonstrated dimerization-induced change in the contact of Phe22 to carboxyl domain residue Ala68 (for example, see Nguyen and Breslow 2002), which has van der Waals contacts to interface residues in both monomeric and dimeric states. Consistent also with such a role for Phe22, the distance between Gln58 and Phe 22 differs in each of the three relevant protein states—unbound monomer, unbound dimer, and bound dimer (see above). On the other hand, a specific mechanism by which both Gln58 mutation and excision of residues 1–6 might affect Tyr 49 conformers—other than altering conformational constraints at the end of the 3, 10 helix of which Tyr49 is the terminus—is less apparent.

Effects of Gln55 mutation: Evidence of differences between BNPI and BNPII

Interactions of the residue 55 side chain are nonidentical in the different bovine NP complexes, some of these differences reflecting the presence or absence of a third residue in the ligand and its identity if present (Wu et al. 2001). However, these interactions differ most significantly on comparison of the unliganded states of the two bovine neurophysins (▶). In the crystal structure of unliganded des 1–6 BNPII, no side chain atoms of Gln55 are visible beyond the beta position, suggesting that they are disordered. In crystals of unliganded desBNPIF91STOP, the carboxamide terminus of Gln55 is completely identified and shown to be hydrogen bonded in at least half of dimer chains (▶). These differences suggest that residue 55 either plays no role in allosteric mechanism or that its role is different in different neurophysins.

The Q55A mutation is of particular interest in this context. The increase in dimerization constant resulting from mutation to Ala indicates that the γ-CH₂ and/or the carboxamide of Gln55 negatively affect dimerization. The NMR structure of the H80E monomer indicates no hydrogen bonding of the Gln55 side chain. The data accordingly indicate that, as with Gln58, the weaker dimerization of the WT protein than of the Q55A mutant reflects constraints on dimerization placed by the specific bonding interactions of the Gln terminus in the unliganded dimer. Also, analogous to the effects of Gln58 mutations, release of these constraints by mutation appears manifest by conformational change in the unliganded state (see Results). Therefore, since these particular hydrogen bonding interactions are lost in the liganded state (▶), the results tend also to suggest that this loss contributes to the ligand-induced increase in BNPI dimerization constant and that the contributions of Gln55 to ligand-facilitated dimerization differ in BNPI and BNPII.

There are two caveats, however. One is that, in contrast to the Gln58 side chain, the Gln55 carboxamide has significant bound-state interactions in both BNPI and BNPII (▶), which might have an effect of their own on dimerization in both proteins. The importance of these contacts to the stability of the bound state is one probable cause of the reduced binding constant of the Q55A mutant (▶). To the extent that they might also alter dimerization, their impact on the net contribution of ligand-induced changes in Gln55 to the ligand-induced increase in dimerization needs to be considered for both BNPI and BNPII. Second, although Gln55 side chain hydrogen bonds in the unliganded state serve to constrain dimerization only in BNPI, their quantitative contributions even in BNPI are ambiguous given the extreme nonadditivity of Gln55 and Gln58 mutation effects on dimerization (▶)—an effect likely to reflect the nonadditivity of their effects on conformation in the unliganded state (see Results) or (less likely given their different NMR effects) the possibility that the two mutations act via a common process. Accordingly, any quantitative difference between the two neurophysins in the contribution of Gln55 to ligand-facilitated dimerization remains to be demonstrated.

With respect to nonadditivity, it is also relevant to point to the nonadditivity of effects of the Q55A and des1–6 mutations, particularly but not exclusively on binding (▶). Given the NMR-demonstrated contacts between residue 55 and the 1–6 region (see Introduction), this nonadditivity is not surprising and supports the view that the effects of the 1–6 region on dimerization are mediated at least in part via its interactions with residue 55. A potential explanation of the nonadditivity would invoke a tightening of these interactions in the Q55A mutant. Since these interactions impede both binding and dimerization, this effect would contribute to the low binding constant of the Q55A mutant and reduce the dimerization constant of the Q55A mutant relative to what it would be in its absence. Loss of this effect in the double mutant would increase both its binding and dimerization constants relative to those predicted by additivity, as observed.

Significance and conclusions: Importance of loop interactions to allosteric mechanism

The 100-fold increase in NP dimerization upon occupancy of the hormone-binding site represents an increase in the standard negative free energy of NP dimerization of ∼2.8 kcal/mol, a value that includes the as yet unquantitated contribution of dimerization-induced changes in the unliganded state that increase binding affinity. The present studies reveal two contributions to the 2.8 kcal/mol additional to the effects of dimerization on binding affinity—a ligand-induced change in the nature of the interface and the presence of loop and amino terminus interactions that constrain dimerization in the unliganded state and that are absent in the liganded state. First, the new crystal structures demonstrate the ligand-induced formation of an additional intersubunit hydrogen bond by residue 81. Together with earlier studies of des 1–6 BNPII (Wu et al. 2001), these results provide the clearest evidence to date of a ligand-induced change in the interface of potential energetic significance and an explanation of earlier analysis of pressure-induced NP dimer dissociation, which suggested an extension of the subunit interface or a decrease in its water accessibility upon ligand binding (Breslow et al. 1991). Second, the data demonstrate that, in BNPI, the side chain hydrogen bonding of loop residues 55, 56, and 58 and the interactions of the amino terminus with the loop (particularly with residue 55), and possibly also with residues 53 and 54, reduce the dimerization constant of the unliganded state and that most, but not necessarily all, of these interactions are lost or altered in the liganded state. Similarities and potential differences between BNPI and BNPII in the details of loop involvement are observed.

The separate contribution of ligand-induced changes of the subunit interface to ligand-facilitated dimerization cannot yet be calculated, in part because its independence from loop contributions is unknown; e.g., effects on dimerization of changes in the loop might include interface changes. However, the potential role of ligand-induced changes in loop and amino terminus interactions, regardless of other effects they might generate, is significant. In BNPI, the combined constraints on dimerization of the unliganded state generated by hydrogen bonding interactions of Gln55, Gln58, and Ser56 side chains and by interactions of the amino terminus are ∼1 kcal/mol as judged by the increased dimerization associated with the double mutant Q55A/Q58V and the des 1–6 mutation (e.g., ▶). In BNPII, where the Gln55 side chain does not appear to be involved in the unliganded state, but where the consequences for dimerization of excision of the amino terminus are twice as high as in BNPI (see above), estimates of loop constraints on dimerization in the unliganded state range from 0.8 to ∼1.3 kcal/mol depending on assumptions made about Gln58; hydrogen-bonding interactions of the Gln58 terminus in the two proteins are similar but not identical (▶).

The realized contribution of these constraints to the ∼2.8 kcal/mol ligand-induced increase in the negative free energy of dimerization depends on the extent to which these constraints are actually lost in the liganded state. In BNPII, in which ligand-induced release of both amino terminus interactions and Gln58 carboxamide hydrogen bonding is demonstrable, these contributions should approximately equal loss of the constraints they impose on dimerization in the unliganded state—roughly 40% of the total change in dimerization free energy. In BNPI, these contributions range from ∼19% to ∼36% of the total free energy change depending on whether amino terminus interactions are indeed lost in the liganded state. Potential bound state contributions of the Gln55 terminus in both proteins remain to be evaluated.

The present results also have implications for the mechanism of dimerization in the unliganded state, specifically providing evidence that such dimerization involves formation of intraloop hydrogen bonds that constrain both the dimerization constant and conformation of the unliganded state. The inability to form these hydrogen bonds increases dimerization, albeit to a conformationally altered state. The results underscore the importance of loop interactions to neurophysin conformation in general and to dimerization in particular and are consistent with their potential role in domain orientation (see Introduction). A specific path involving Phe22 is suggested by which the status of the loop at residue 58 is communicated to the rest of the protein.

The mechanism by which specific interactions constrain neurophysin dimerization in the unliganded state contrasts with that operative in epidermal growth factor receptor (for example, see Dawson et al. 2005). In the receptor case, the unliganded state is monomeric, and such interactions directly restrain residues that ultimately form part of the dimer subunit interface in the liganded state. With neurophysin, the constraints operate at a distance on the subunit interface to maintain a low but significant dimerization constant in the unliganded state. Release of these constraints upon ligand-binding is mediated by proximity of the constraining interactions to the binding site and leads principally to optimization and enhancement of preexisting interchain contacts.

Materials and methods

Preparation of BNPI mutants

Single site mutation of WT BNPI DNA for expression in Escherichia coli was accomplished by PCR as previously described (Eubanks et al. 2000, 2001). Shortening of the protein at the C terminus was accomplished by mutation of individual C-terminal codons to STOP codons. Deletion of the first six residues of the WT protein utilized a pair of 46 bp primers for PCR that complement the WT plasmid but skip the codons for the N-terminal six residues of the protein. DNA sequences were confirmed by sequencing before mutants were expressed.

The WT and mutant proteins were expressed in E. coli and isolated from inclusion bodies in their misfolded state also as described earlier (Eubanks et al. 1999, 2000). The misfolded state is a disulfide-scrambled state and is normally folded at pH 8 by stirring in air for 48 h, in the presence of β-mercaptoethanol to facilitate disulfide exchange and ligand peptide to drive the folding reaction (Eubanks et al. 2000). This procedure was used successfully for all mutants, but gave low folding yields (≤50% of normal) for the K59G and S56E mutants. The low yield of the S56E mutant was shown to reflect a low affinity for peptide (see Results). However, yields for the K59G mutant were restored to normal by folding in a glutathione buffer (3 mM GSSG, 2 mM GSH, 10 mM FY in 0.1 M sodium acetate at pH 7.4) in the absence of air (see Discussion). After folding, protein was separated from the other components of the folding mixture by gel filtration and lyophilized. With the exception of the S56E mutant, correctly folded protein was separated from misfolded protein by chromatography on a ligand-linked affinity column and dialyzed and lyophilized as previously described (Eubanks et al. 2000). Affinity chromatography could be not used for the S56E mutant because of the low peptide affinity of its folded state, and the folded state of this protein was separated from misfolded states by HPLC as previously described for other mutants that retain some folding ability but that cannot bind peptide (Eubanks et al. 2001). Masses of all purified folded proteins were confirmed by mass spectrometry. Folding was confirmed by circular dichroism; all folded mutants exhibited the normal two conformationally sensitive disulfide signals at ∼245 and 280 nm with—except as indicated for the S56E and des1–6 mutants (see Results and Discussion)—chain molar ellipticities of ∼+20,000 and −20,000 deg cm²/dmol, respectively (for example, see Eubanks et al. 2001).

Determination of dimerization constants

Dimerization constants were measured by NMR at 25°C as previously described (for example, see Zheng et al. 1997; Eubanks et al. 2000), using the relative intensities of Cys28 α-proton signals, located at ∼6.15 and ∼6.4 ppm in monomer and dimer, respectively. Data were obtained on a Varian Inova 600 MHz spectrometer. For most runs, samples were dissolved in D₂O containing 10 mM pH 6.2 phosphate buffer readjusted with NaOD or DCl to a final pH of 6.2 (uncorrected electrode reading in D₂O) after addition of protein. Several reported studies were carried out without buffer, but at measured pH. Protein concentrations were determined from the CD intensity at 280 nm, assuming a molar (chain) ellipicity of −20,000 deg cm²/dmol (see above). Comparison of concentrations so calculated with values obtained from 280 nm absorbance measurements indicated that this method gave reliable values even for proteins in which the 245/280 ratios were atypical, and was preferred because of its insensitivity to random contaminants from solvent. To obtain the weight ratios (M/D) of monomer (M) to dimer (D), spectra were processed by different methods to reduce effects of noise and baseline uncertainties and then magnified. Intensities of monomer and dimer signals in the magnified spectra were determined by weight. Dimerization constants calculated for individual experiments are averages of the different processing methods. Spectra were obtained at different concentrations as needed to reduce ambiguities.

Other NMR methods

DQF-COSY experiments (Rance et al. 1983) of Q58V and nitrated Q58V consisted of 256 t ₁ increments. Spectral widths in both dimensions were 6999.7 Hz. The concentrations of Q58V and nitrated Q58V were 1.5 mM and 0.7 mM, respectively. Processing of the NMR data was carried out by NMRpipe (Delaglio et al. 1995) and analyzed by the Sparky program (SPARKY 3, T.D. Goddard and D.G. Kneller, University of California, San Francisco).

Determination of binding affinities

Peptide binding was measured by CD as previously described (Breslow et al. 1973) using ∼0.1 mM mononitrated protein that had been prepared as also previously described and further purified by affinity chromatography (Rabbani et al.1982). For most proteins, measurements were made at pH 6.2 in 0.1 M ammonium acetate containing 2 mM MES buffer. However, the peptide complexes of the des 1–6 mutants became insoluble under these conditions, so studies of these proteins were conducted at pH 7.4 in 50 mM Tris acetate, 50 mM ammonium acetate. In all cases, results with the mutant are compared with those of the WT protein under identical conditions.

Binding constant measurement involves determination from the nitrotyrosine CD signal of the fraction of protein in the liganded state at different concentrations of total peptide. Protein concentrations are determined by nitrotyrosine absorbance. Free peptide concentrations are calculated as the difference between bound peptide and total peptide. Because of the linkage between binding and dimerization, Scatchard plots of binding data exhibit a slight curvature that becomes more marked at lower degrees of protein saturation. Also, estimates of free peptide concentration become less reliable at low degrees of protein saturation for strongly binding proteins. Accordingly, for comparison of different peptides or different proteins, as is the case here, we have typically not utilized data obtained at levels of fractional protein saturation () <0.4 in binding constant calculations. Constants are instead calculated from Scatchard plot slopes at and above values of = 0.5 (for example, see Breslow et al. 1991) and represent hybrids of monomer and dimer affinities.

Crystal preparation and data collection

Crystals of the FY complex of des 1–6 BNPI were obtained by hanging-drop vapor diffusion by mixing equal volumes protein and dipeptide solutions (15 mg/mL des 1–6 BNPI and 10 mg/mL FY.HCl in water) and then mixing half of the resultant solution with an equal volume of crystallization buffer: 0.2 M ammonium acetate, 0.1 M tri-sodium citrate dihydrate (pH 5.6), 60% (w/v) polyethylene glycol 4000. Crystals appeared after 2 d at 20°C.

Crystals of the FY complex of the des 1–6 Q58V mutant of BNPI were similarly prepared but with a different crystallization buffer—0.2 M calcium chloride dihydrate, 0.1 M sodium acetate trihydrate (pH 4.0), 22% (v/v) iso-propanol. Crystals appeared after 3 d at 20°C.

The most useful crystals of unliganded BNPI were obtained using the des 1–6 derivative of the F91STOP mutant. These crystals were also obtained by hanging-drop vapor diffusion, in this case mixing equal volumes of protein solution (20 mg/mL in water) and crystallization buffer—1.5 M sodium chloride, 11% (v/v) ethanol. Crystals appeared after 6 d at 20°C.

The BNP1FY and Q58VFY data were collected using a Rigaku RU-H3R generator and R-AXIS IV detector system with optical focused Cu Kα X-rays. The BNP1 data were collected at the Brookhaven National Laboratory X-4A beamline with λ = 0.9793 Å. Data were processed using HKL2000. Details are listed in ▶.

Table 5.

Data collection and processing

Structure determination

Structures were determined by molecular replacement using the program MOLREP. Des1–6 BNPII (PDB entry 1JK6) was used as the starting search model for desBNPIfy. Subsequently, desBNPIfy was used as the search model for desQ58Vfy and desBNPIF91STOP. Ligand was traceable from desBNP1fy and desQ58Vfy electron density maps. Initial structures were refined by simulated annealing with CNS followed by B-factor refinement. The final structures of each protein include residues 7–86 of five chains, as well as bound dipeptides for the complexes desBNPIfy and desQ58Vfy. A number of water molecules are also assigned in the structures of the two complexes, the larger number in the WT complex reflecting the higher degree of resolution of this structure. Water molecules were also seen in the structure of the unliganded protein, but the lower resolution of this structure made their assignment less certain, so these are not reported. Details are shown in ▶. Structures were visualized by the VMD program (Humphrey et al. 1996) unless otherwise indicated.

Table 6.

Structure solution and refinement statistics

Atomic coordinates for the three crystal structures have been deposited with the Protein Data Bank and given the following identification codes: desBNPIfy = 2HNU; desQ58Vfy = 2HNV; desBNPIF91STOP = 2HNW.