A Core Mutation Affecting the Folding Properties of a Soluble Domain ...

4 downloads 7 Views 948KB Size Report
Soluble Domain of the ATPase Protein CopA from. Bacillus .... cates identical residues, the dot indicates ..... produces a decrease in the unfolding free energy,22.


J. Mol. Biol. (2003) 331, 473–484

A Core Mutation Affecting the Folding Properties of a Soluble Domain of the ATPase Protein CopA from Bacillus subtilis Lucia Banci, Ivano Bertini*, Simone Ciofi-Baffoni, Leonardo Gonnelli and Xun-Cheng Su Department of Chemistry, The Magnetic Resonance Center CERM, University of Florence Via Luigi Sacconi 6, 50019 Sesto Fiorentino, Florence, Italy

The two N-terminal domains of the P-type copper ATPase, CopAa and CopAb, from Bacillus subtilis differ in their folding capabilities in vitro. Whereas CopAb has the typical babbab structure and is a rigid protein, CopAa is found to be largely unfolded. A sequence analysis of the two and of orthologue homologous proteins indicates that Ser46 in CopAa may destabilise the hydrophobic core, as also confirmed through a bioinformatic energy study. CopAb has a Val in the corresponding position. The S46V and S46A mutants are found to be folded, although the latter displays multiple conformations. S46VCopAa, in both apo and copper(I) loaded forms, has very similar structural and dynamic properties with respect to CopAb, besides a different length of strand b2 and b4. It is intriguing that the oxygen of Thr16 is found close, though at longer than bonding distance, to copper in both domains, as it also occurs in a human orthologue domain. This study contributes to understanding the behaviour of proteins that do not properly fold in vitro. A possible biological significance of the peculiar folding behaviour of this domain is discussed. q 2003 Elsevier Ltd. All rights reserved.

*Corresponding author

Keywords: folding; S46V mutant; NMR structure; copper transporting protein; CopA

Introduction In recent years it has started to become clear that metal ion uptake, trafficking, excretion and regulation are processes performed and tightly controlled by several groups of proteins which guide the metal ions from outside the cell through the membrane to their final destination.1,2 Recently, we have been developing a structural genomic study on a series of proteins involved in copper homeostasis.3,4 Among these, are P-type ATPases, membrane proteins involved in metal Supplementary data associated with this article can be found at doi:10.1016/S0022-2836(03)00769-1 Abbreviations used: HSQC, heteronuclear single quantum coherence; RMSD, root-mean-square deviation; NOESY, nuclear Overhauser effect spectroscopy; TOCSY, total correlation spectroscopy; NOE, nuclear Overhauser effect; TPPI, time-proportional phase incrementation; REM, restrained energy minimization; DTT, dithiothreitol; CD, circular dichroism; SC, side-chain. E-mail address of the corresponding author: [email protected]

homeostasis.5 The N-terminal region of these metal transporting P-type ATPases contains a variable number of soluble domains having a predicted babbab fold.3 Specifically, they are one or two in bacteria while the number ranges between two and six in eukaryotes. An ATPase, CopA, in Bacillus subtilis contains two soluble domains. We have already determined the NMR solution structure of the second soluble domain and shown that it adopts the predicted babbab fold.6 On the contrary, the 1H – 15N heteronuclear single quantum coherence (HSQC) spectrum of the apo form of CopA(1 – 72) (apo-CopAa hereafter) indicates that it is not in a well-structured state. The two protein segments are highly similar in sequence (40% identity, 67% similarity). However, a detailed sequence analysis indicated that at position 46, instead of the highly conserved Val or Ala, a Ser is present. Energy, secondary structure and hydrophobic core analysis of the modelled first domain confirmed that this is a key position of the amino acid sequence with respect to the protein stability. A mutation from Ser to Val in this position determines the stabilisation of the folding. Indeed, the

0022-2836/$ - see front matter q 2003 Elsevier Ltd. All rights reserved.


solution structure determination of the apo and Cu(I) forms of the S46VCopAa construct, which are reported here, shows that the babbab fold is completely recovered. A structural comparison with the second domain of CopA from B. subtilis shows that the overall folding is essentially identical, with small differences in the vicinity of the copper-binding site. Our results represent an interesting example of how a single core mutation can lead towards a well-folded state in a ferrodoxinlike fold, providing a significant contribution on how non-local interactions can determine an increased stability of a fold through the formation of a more compact hydrophobic core. A possible biological significance of the peculiar folding behaviour of this metal-binding domain is finally discussed.

Results The two soluble domains located in the N terminus of the copper ATPase protein CopA are roughly constituted by amino acid residues 1 –71 and 73– 147, respectively.3 It has been previously observed through NMR spectroscopy that the polypeptide 73– 147 forms a folded protein in a

Folding Properties of an N-terminal Domain of CopA

monomeric state adopting the predicted babbab fold,6 while the 1H – 15N HSQC spectrum (Figure 1(A)) of the apo-CopAa polypeptide chain shows a limited dispersion of the signals. All the backbone 15N and 1H resonances are grouped in the narrow 110 – 130 and 7.7– 8.5 ppm range, respectively. Also the NHs of the side-chains of Asn and Gln residues are essentially all degenerate in the typical region of side-chain NHs. Furthermore, the number of cross-peaks observed in the 1H – 15N HSQC spectrum of the apo-CopAa is lower than expected considering the number of amino acid residues. Since the integrity of the polypeptide chain is conserved as checked through electrophoresis, we might assume that some cross-peaks are broadened beyond detection due to (i) conformational fluctuations on the millisecond timescale or (ii) NH exchange with bulk water which is faster than the frequency difference in proton resonances. Such line broadening appears to be a characteristic of molten globules and of some disordered regions of proteins.7 – 10 Molten globules are compact intermediates in protein folding that have native-secondary structure but lack fixed packing interactions.11,12 Normally, few resonances can be observed in the 1H – 15N HSQC spectrum for these folding intermediates.13 The lack of peaks in the

Figure 1. 2D 1H – 15N HSQC spectra (600 MHz, 298 K) of (A) apoCopAa and (B) apo-S46VCopAa. For both samples the protein concentration was about 1.5 mM, in 20 mM phosphate buffer (pH 7).

Folding Properties of an N-terminal Domain of CopA

H – 15N HSQC spectrum of apo-CopAa might suggest that the protein remains in a collapsed state, with residual secondary structure elements restricting the conformation of the polypeptide chain. Indeed, the far-UV circular dichroism (CD) spectrum of apo-CopAa shows the presence of a-helical structures without any strong negative band at 198 nm, characteristic of a random coil protein. Changes in buffer, pH and dithiothreitol (DTT) concentrations did not produce any relevant improvement in the spectra. Sequence alignment (Figure 2) between the two soluble domains was performed in order to locate amino acid changes that might determine the lack of structurally important interactions for the stabilisation of the folding. The two protein segments are highly similar in sequence (40% identity, 67% similarity), both having the copper binding motif MXCXXC and a few hydrophobic residues conserved between all bacterial ATPases.3 Therefore, the different folding properties of the two domains observed in vitro could be ascribed to one or more key residues in the sequence which are of different nature in the two domains and which are essential to induce its folding in vitro. Attention was directed to those positions that show differences in the length of the side-chain and/or in charge and/or hydrophobicity of the residue. Subsequently, the selected positions were restricted to those present in the secondary structure elements predicted for the first soluble domain. The PROSA program was used as a tool to evaluate the energetic properties of this protein fold as a function of amino acid sequence position.14 In the energy graphs positive values indicate strained residues of the chain whereas negative values correspond to stable parts of the molecule. In the case of poorly defined regions such as loops, usually variable and unreliable values can be found depending on the structural model quality.14 For this analysis a structural model for the sequence of the first CopA domain was calculated, modelling it using the solution structure of apo-CopAb6 as template. The whole analysis identified two positions, 27 and 46, where both the sequence alignment (Figure 2) and the energetic parameters of PROSA (Figure 3) show meaningful differences, suggesting that they



Figure 3. Energy graph of CopAb structure (B), CopAa model (X) and S46VCopAa model (O), based on the pair energy Ep and surface energy Es as obtained from PROSA. Positions 27 and 46 are shown.

could represent “weak” positions in the sequence determining a less stable fold of the first domain. A variable energy value is found for residue Thr17 depending on the input structural model as a consequence of its position in a disordered loop region. A comparative structural genomic analysis of ATPase proteins, starting from the sequence of the first soluble domain of the copper transporting yeast ATPase, Ccc2, is available.3 In a subset of 53 soluble domains of bacterial membrane-bound ATPases, position 27 is found not to have recurrent amino acids, and is therefore discarded as a possible target of “weak” positions in the babbab fold, while in position 46 an hydrophobic residue is always present with only two exceptions, where a Ser is present as in the present case. On the basis of this analysis, two mutants, S46A and S46V, were designed and two structural models were obtained using the same template as before. In the model of S46V mutant, Val46 is in contact with several hydrophobic residues (Met10, Val12, Ile24, Leu28, Val39, Val48, and Ile64), which are all highly conserved residues in the soluble domains of metal transporting ATPases and which constitute the hydrophobic core of the babbab fold.3 In addition, the mutants S46V and S46A have a lower energy in position 46, comparable with that found for the

Figure 2. Sequence alignment of the two soluble domains of CopA from B. subtilis. Positive (Arg and Lys) and negative (Glu and Asp) residues are indicated in blue and in red, respectively. The stars indicates identical residues, the dot indicates similar residues. Residues on which the attention was pointed out in this research are shaded in green. The predicted secondary structure elements, reported above the alignment, refer to the CopAa domain.


Folding Properties of an N-terminal Domain of CopA

second, well folded domain of CopA (Figure 3). These observations support again the idea that position 46 could play a relevant role in stabilizing the folding of the protein. Therefore, we expressed the two variants S46V and S46A. Both mutants were stable and monomeric as determined from gel filtration analysis. Also the 1H – 15N HSQC map of the S46V variant in the apo form shows a good signal dispersion and linewidths consistent with a monomeric folded protein (Figure 1(B)), while the 1 H – 15N HSQC spectrum of the S46A variant indicates a folded state but with the presence of multiple conformations in solution, as evidenced by a number of 1H – 15N resonances larger than that expected. The binding of copper has also been characterized for the S46V variant. The atomic absorption data of Cu(I)-S46VCopAa shows that one copper ion is bound to the protein (metal/protein ratio 0.9). Therefore, we selected the S46VCopAa variant for further characterization and for determining the solution structure of both apo and copper(I)-bound forms. NMR structure of apo- and Cu(I)-S46VCopAa Assignments of the resonances of apo- and Cu(I)-S46VCopAa started from the analysis of the 1 H – 15N HSQC maps, which allowed the identification of the 15N and 1HN resonances. Analysis of 15 N-edited 3D NOESY-HSQC and of 2D NOESY and TOCSY maps allowed sequence-specific assignment. Signals of 73 out of 76 residues were assigned both in the apo and Cu(I) forms, with the first two residues and Ala18 being not identified in both forms. In the apo and in the Cu(I) loaded proteins about 96% of the proton resonances could be located in the maps and 71 out of 74 15N backbone amide resonances were assigned. A few signals exhibit sizable shift variations between the apo and the copper-bound forms. Comparison of the weighted average chemical shift differences Davg(HN) (Figure 4)15 reveals that the shift differences are located in the stretch 15 – 24, with Thr16, Ala19 and Cys20 experiencing the largest chemical shift changes. All the other residues of the two forms have small or negligible chemical shift differences. The pattern of assigned NOEs indicated the presence of a few secondary structure elements, which involve two helices, characterized by a high number of sequential and medium range connectivities, and four antiparallel b-strands indicated by the presence of long range NOEs, thus confirming that the S46V mutation produces a folded protein. It is also found that the secondary structure is not significantly affected by the presence or absence of the copper ion. In the apo form, 3132 NOE cross-peaks were assigned and integrated, providing 1960 unique upper distance limits, of which 1440 are meaningful. A total of 41 proton pairs were stereospecifically assigned. Eighty-nine angle constraints, 46 f and 43 c were experimentally determined and used in

Figure 4. 1H and 15N amide chemical shift differences between Cu(I)-S46VCopAa and its apo form. The weighted average chemical shift differences Davg(HN) (i.e. ð½ðDHÞ2 þ ðDN=5Þ2 =2Þ1=2 ; where DH and DN are chemical shift differences for 1H and 15N, respectively) are shown in the bottom plot. The secondary structure elements of S46VCopAa are reported at the top.

the calculations. After restrained energy minimization (REM) on each of the 30 conformers of the family, the RMSD to the mean structure (for resi˚ , for the backbone, and dues 4 –73) is 0.37(^ 0.13) A ˚ for all heavy atoms; the penalty is 0.78(^ 0.12) A ˚ 2 for distance constraints and 0.39(^ 0.05) A 0.11(^ 0.02) rad2 for angle constraints. The RMSD values per residue to the mean structure of the REM family are given in Figure 5(A). In the Cu(I) form, 2913 NOE cross-peaks were assigned and integrated, providing 1818 unique upper distance limits, of which 1357 are meaningful. A total of 35 proton pairs were


Folding Properties of an N-terminal Domain of CopA

Table 1. Statistical analysis of the REM family and the mean structure of apo-S46VCopAa from B. subtilis REM (30 structures) ˚ )a RSM violations per experimental distance constraint (A Intraresidue (271) 0.0155 ^ 0.0018 Sequential (358) 0.0082 ^ 0.0025 0.0021 ^ 0.0016 Medium rangeb (358) Long range (453) 0.0130 ^ 0.0017 Total (1440) 0.0150 ^ 0.0011 Phi (46) (deg) 0.664 ^ 0.153 Psi (43) (deg) 0.279 ^ 0.328 Average number of violations per structure Intraresidue Sequential Medium rangeb Long range Total Phi Psi Average no. of NOE violations ˚ larger than 0.3 A ˚ 2) Total NOE square deviations (A Average torsion deviations (rad2) RMSD to the mean structure (4-73) ˚) (A

Figure 5. RMSD per residue to the mean structure of apo-S46VCopAa (A) and of Cu(I)-S46VCopAa (B) for the backbone (filled circles) and all heavy atoms (open squares) of the conformer structure.

stereospecifically assigned and 89 angle constraints, for the same angles as in the apo structure, were used in the calculations. After restrained energy minimization on each of the 30 conformers of the family, the RMSD to the mean structure (for ˚ for the backbone residues 4– 73) is 0.38(^ 0.10) A ˚ and 0.79(^ 0.08) A for all heavy atoms; the penalty ˚ 2 for distance constraints and is 0.27(^ 0.03) A 2 0.12(^ 0.02) rad for angle constraints. The RMSD values per residue to the mean structure of the REM family are given in Figure 5(B). The statistical analyses of the REM family of apo- and Cu(I)-S46VCopAa structures are reported in Tables 1 and 2, respectively. Both structures show the typical folding pattern of copper chaperones, the “open-faced b-sandwich” fold (b1-a1b2-b3-a2-b4),16 – 18 displaying the following secondary structure elements: 6– 13 (b1), 19– 30 (a1), 33 – 40 (b2), 44 – 51 (b3), 56 –68 (a2), 70 –72 (b4), which are consistent with the analysis of the NOE patterns. In addition, in both structures a 310-helix occurs between strands b2 and b3 involving residues 41– 43. In Figure 6, 30 conformer structures of apo- and Cu(I)-S46VCopAa are represented as a

Structural analysisc % Of residues in most favourable regions % Of residues in allowed regions % Of residues in generously allowed regions % Of residues in disallowed regions H-bond energy (kJ mol21) Overall G-factor Experimental restraints analysisd % Completeness of experimentally ˚ cut-off observed NOE up to 4 A distance % Completeness of experimentally ˚ cut-off observed NOE up to 5 A distance

kREMl 0.0171 0.0084 0.0192 0.0123 0.0145 0.0 0.0

6.57 ^ 1.82 3.73 ^ 1.24 12.10 ^ 1.64 9.73 ^ 2.36 32.13 ^ 3.93 1.83 ^ 0.78 0.57 ^ 0.67

6 4 10 6 26 0 0



0.39 ^ 0.05 0.11 ^ 0.02 0.37 ^ 0.13 (BB) 0.78 ^ 0.12 (HA)

0.37 0.17



16.1 1.3

15.2 1.5

0.5 3.05 ^ 0.05 20.30 ^ 0.02

0.0 3.34 -0.33





REM indicates the energy minimized family of 30 structures, kREMl is the energy minimized average structure obtained from the coordinates of the individual REM structures. a The number of experimental constraints for each class is reported in parenthesis. b Medium range distance constraints are those between residues ði; i þ 2Þ; ði; i þ 3Þ; ði; i þ 4Þ and ði; i þ 5Þ: c As it results from the Ramachandran plot analysis. For the PROCHECK statistics, an average hydrogen-bond energy in the range of 2.5–4.0 kJ mol21, and an overall G-factor larger than 20.5 are expected for a good quality structure. d As it results from AQUA analysis.

tube, whose radius is proportional to the backbone RMSD of each residue. Mobility of apo-and Cu(I)-S46VCopAa The experimental R1 ; R2 and 1H – 15N NOE values of the backbone amide 15N nuclei of apoand Cu(I)-S46VCopAa were measured at 600 MHz (Figure 7). Reliable relaxation values have been obtained for all 71 assigned backbone NH


Folding Properties of an N-terminal Domain of CopA

Table 2. Statistical analysis of the REM family and the mean structure of Cu(I)-S46VCopAa from B. subtilis REM (30 structures) ˚ )a RSM violations per experimental distance constraint(A Intraresidue (280) 0.0201 ^ 0.019 Sequential (366) 0.0102 ^ 0.0015 0.0161 ^ 0.0015 Medium rangeb (303) Long range (408) 0.0046 ^ 0.0017 Total (1357) 0.0133 ^ 0.0080 Phi (46) (deg) 2.605 ^ 1.110 Psi (43) (deg) 0.659 ^ 0.491 Average number of violations per structure Intraresidue 16.73 ^ 2.03 Sequential 8.87 ^ 2.29 7.77 ^ 1.36 Medium rangeb Long range 2.60 ^ 1.52 Total 35.97 ^ 3.45 Phi 2.60 ^ 1.14 Psi 0.90 ^ 0.75 Average no. of NOE violations ˚ larger than 0.3 A ˚ 2) Total NOE square deviations (A Average torsion deviations (rad2) RMSD to the mean structure (4–73) ˚) (A

Structural analysisc % Of residues in most favourable regions % Of residues in allowed regions % Of residues in generously allowed regions % Of residues in disallowed regions H-bond energy (kJ mol21) Overall G-factor Experimental restraints analysisd % Completeness of experimentally ˚ cut-off observed NOE up to 4 A distance % Completeness of experimentally ˚ cut-off observed NOE up to 5 A distance

kREMl 0.0218 0.0138 0.0159 0.0050 0.0146 1.269 0.881 18 8 7 3 36 1 1



0.27 ^ 0.03 0.12 ^ 0.02 0.38 ^ 0.10 (BB) 0.79 ^ 0.08 (HA)

0.32 0.11



19.7 1.6

22.7 1.5

0.4 3.09 ^ 0.05 20.35 ^ 0.02

0.0 2.90 -0.38





Figure 6. Backbone atoms for the solution structures apo-S46VCopAa (A) and Cu(I)-S46VCopAa (B) as a tube with variable radius, proportional to the backbone RMSD value of each residue. a-Helices are coloured in red, b-strands in cyan and 310-helices in orange. The side-chains of Cys17, Cys20 and the Cu(I) ion are also shown. The secondary structure elements are indicated. The Figure was generated with the program MOLMOL.30

REM indicates the energy minimized family of 30 structures, kREMl is the energy minimized average structure obtained from the coordinates of the individual REM structures. a The number of experimental constraints for each class is reported in parenthesis. b Medium range distance constraints are those between residues ði; i þ 2Þ; ði; i þ 3Þ; ði; i þ 4Þ and ði; i þ 5Þ: c As it results from the Ramachandran plot analysis. For the PROCHECK statistics, an average hydrogen-bond energy in the range of 2.5 –4.0 kJ mol21, and an overall G-factor larger than 20.5 are expected for a good quality structure. d As it results from AQUA analysis.

resonances for both forms. Average values of R1 ; R2 and 1H – 15N NOE are 1.95(^ 0.16) s21, 7.57(^ 1.32) s21 and 0.72(^ 0.29) for the apo form and 1.97(^ 0.33) s21, 8.10(^ 1.50) s21 and 0.70(^ 0.30) for the Cu(I) form, respectively. These values are homogeneous along the entire polypeptide sequence. Only residues in the loop 1 region, which contains the Cys ligands, display R2 values higher than the average in the apo form, while the C and N terminus of both forms show a significant

Figure 7. 15N relaxation parameters R1, R2 (using a refocusing delay of 450 ms) and heteronuclear NOE versus residue number for apo-S46VCopAa collected at 600 MHz.

Folding Properties of an N-terminal Domain of CopA

decrease of the 1H – 15N NOE and R2 values. The latter behaviour suggests a flexibility in the ps –ns timescale of these N and C-terminal protein segments, consistent with their low number of NOEs and higher RMSD values. A few residues experience R2 values larger that the average, thus suggesting the presence of conformational exchange processes that contribute to R2 : They are Thr16 and Cys17 (loop1), Ala19 and Cys20 (a1), Asn57 and Ala58 (a2). Of these, Thr16, Cys17, Ala19, Ala58 and, in addition, Ile49 (b3), show a dependence of the R2 values on neff, as determined in CPMG R2 measurements as a function of tCPMG, according to equation (1). The dependence of R2 with neff indicates an exchange contribution to the transverse relaxation rate occurring with correlation times tex in the range of 100– 400 ms (see Materials and Methods). Cys20 (a1) and Asn57 (a2) have R2 higher than average, but do not experience a dependence on R2 with the tCPMG, suggesting the presence of exchange processes occurring at rates faster than those accessible with the present experimental conditions. Other residues, such as Gln11 (b1), Val12 (b1), Met31 (loop2), Val48 (b3), Gly56 (a2) and Tyr50 (b3), display a significant increase of R2 only at the weakest neff value, indicating exchange processes at rates slower than those accessible (see Materials and Methods). Interestingly, some of the residues involved in the conformational exchange processes belong to strand b3, which contains Val46. The others are mostly located close to the copper-binding region, as already found in the apo form of other babbab chaperones.19,20 From the R2 =R1 ratio an estimate of the overall tumbling correlation time (tm) of the apo-and Cu(I)-S46VCopAa was calculated to be 5.3(^ 0.3) ns and 5.1(^ 0.9) ns, respectively. This value is consistent with that expected from the Stokes – Einstein isotropic model21 for a molecule of about 8 kDa and indicates absence of aggregation. This value is very similar to that determined for other metal chaperones with analogous fold using the same approach.20

Discussion Structural effects of the S46V mutation In the present work it is shown that the change of a small hydrophilic residue at position 46 into hydrophobic residues determines dramatic changes in the folding properties of the babbab fold. Indeed, from a substantially collapsed state of the wild-type protein, the formation of a folded state occurs. The S46V mutation produces a single, well-folded form in solution, while the S46A mutation produces a folded state characterized by multiple conformations, indicating that a small side-chain does not produce the optimal packing contacts. This result indicates that the hydrophobic interactions formed by the residue in this position


constitute a major contribution to the stability of the native structure in vitro. A similar behaviour has been observed in the cytochrome c folding properties, where the side-chain of Leu94 is critical in determining packing of the N and C-terminal helices.22 In that case, replacement of Leu with Ala produces a decrease in the unfolding free energy,22 indicating that the smaller Ala causes severe packing defects at the helix –helix interface. A further indication of the role of Val46 in determining strong hydrophobic packing in S46VCopAa comes from its strong and numerous contacts with surrounding residues as documented by the several NOEs between its hydrophobic side-chain and those of the conserved hydrophobic residues Met10, Val12 and Val39. The presence of a strong hydrophobic patch surrounding Val46 is in agreement with the solvent accessibility, which is remarkably low (1% and 3% accessible surface of Val46 in apo and copper-bound forms, respectively). All these data represent clear evidence that position 46 participates in a network of hydrophobic interactions when a hydrophobic residue is present in that position, while a hydrophilic residue, such as a serine, produces a not well-folded state in vitro conditions. Our results therefore support the idea that one of the key driving forces for folding is the contact among neighbouring hydrophobic residues that facilitate the interactions of more distant hydrophobic residues, thus favouring the formation of the complete structure. A single residue can determine a large destabilization of the hydrophobic core. From these results a question arises on whether there are functional reasons that justify the instability of wild-type domain, or the unfolding is due to subtle differences between the in vitro and in vivo conditions. The presence of a Ser at position 46 in another organism out of 53 disfavours the possibility that the present construct is an allele or a sequence mistake. On the other hand, the possibility of a specific role of the partially unfolded protein is disfavoured by comparison with the more numerous well behaving constructs. It is possible, however, that the in vivo conditions provide a folded and functionally active protein. In the analogous Wilson protein, the N-terminal domains are shown to interact with the ATP-binding domain23 and therefore such interaction, or the interaction with the membrane, may stabilize the folding in vivo. Comparison between the solution structures of S46VCopAa and wild-type CopAb for both apo and Cu(I) states The apo and copper(I) forms of S46VCopAa and CopAb structures, which have 40% of residue ˚ identity, have an overall backbone RMSD of 1.90 A ˚ , respectively. The overall folding of the and 2.00 A two domains is essentially identical. Secondary structure elements are well superimposed with the exception of strand b2 in the apo forms (Figure 8).


Folding Properties of an N-terminal Domain of CopA

Figure 8. (A) Comparison between the backbone of apo-S46VCopAa (red) and apo-CopAb (green) from B. subtilis. (B) Comparison between the backbone of Cu(I)-S46VCopAa (red) and Cu(I)-CopAb (green) from B. subtilis. The secondary structure elements are indicated. The cysteine ligands involved in the copper binding, the Cu(I) ion, Thr16 and Val46 are also indicated. The Og of Thr16 is depicted as a sphere.

Strand b2 in both forms of S46VCopAa is longer than in the second domain.24 This might be due to the lack of a proline at position 106 in CopAb, which breaks strand b2. The last strand b4 is significantly shorter in the S46VCopAa structure. Unlike the CopAb structure, a 310-helix is present between strands b2 and b3 involving residues 41 –43 of the S46VCopAa domain. The mutated residue Val46 takes the same conformation as the corresponding Val115 in the CopAb structure for both the apo and the copper(I) forms (Figure 8). Also several NOEs involving Val46, as described before for S46VCopAa, are conserved for Val115 in the CopAb domain, thus suggesting the key role of this residue in making hydrophobic contacts and therefore in determining the global correct fold. In the apo forms, the largest differences are observed for loop 1, which contains the copperbinding ligands, and loop 3, where large conformational changes are present, while loops 2 and 5 are well superimposed (Figure 8(A)). In both apo forms, the N-terminal copper binding cysteine is ˚ ; Cys85, quite disordered (Cys17, on SC 2.75 A ˚ RMSD on SC 2.91 A) spanning variable conformations, while the second copper-binding cysteine, located in helix a1, has a well-defined confor˚ ; Cys85, RMSD on SC mation (Cys17, on SC 1.00 A ˚ 0.87 A) and similar in both structures (Figure 8(A)). In the copper-bound forms, at variance with the copper-free ones, loops 1 and 3 have a very similar conformation in the two structures (Figure 8(B)). Both cysteine residues are well defined and very similar in both copper(I)-bound structures, even if the position of the C-terminal cysteine is a bit translated from one structure to the other (Figure 8(B)). In addition to one of the copper(I)-binding cysteine residues, loop 1 also contains a threonine residue, Thr16, whose oxygen is pointing toward the metal ion in both copper(I) structures (Og – ˚ in CopAb6 and Cu(I) average distance 3.45 A ˚ 3.50 A in S46VCopAa) (Figure 8(B)). As in the cop-

per bound form of the first domain of the human ATPase the corresponding Thr is found, from NMR data, close to the copper ion,25 Thr16 oxygen might be considered a long distance metal ligand. This residue is, indeed, highly conserved as either Thr or Ser in all the bacterial and eukaryotic copper transporter sequences known up to now.3

Concluding Remarks In conclusion, we have shown that, through a detailed analysis of the primary sequence and of the structural and energetic properties, mutants can be designed that would then produce well folded proteins. The mutation on only one selected position can indeed produce the necessary interactions, which energetically stabilize the global fold of the protein. We have solved the structures with and without copper(I) of a stable construct. Finally, the unanswered question of why a destabilizing amino acid is sometimes used by nature has arisen.

Materials and Methods Structural prediction and homology modelling Secondary structure prediction was performed with the PHD Method†,26,27 which uses evolutionary information in the form of multiple sequence alignments that are used as input in place of single sequences. Structures were modelled with the program MODELLER v. 4.0,28 using as reference the available solution structure of apo-CopAb, i.e. the second domain of the bacterial CopA protein from B. subtilis (PDB ID 1JWW). The program PROSA II (version 3.0, 1994)14 for protein structure analysis was used to estimate the energy distribution over the protein model. It provides an adimensional energy parameter per residue, which is a combination of pair residue interaction energies and † http://pbil.ibcp.fr


Folding Properties of an N-terminal Domain of CopA

surface energies. The surface term is used to model the energetic features of solvent – protein interactions as it takes into account solvent exposure.29 In energy graphs positive values points to strained sections of the chain whereas negative values correspond to stable parts of the molecule. Solvent accessibility for individual residues, residue – residue contacts and surface potentials were evaluated with MOLMOL v. 2.6.30 Two residues were assumed to be in contact if at least five pairs of their atoms were clo˚ . Buried residues were defined as those havser than 4 A ˚ 2. For most of ing a solvent accessibility lower than 25 A the residues, this value corresponds to 90% or more of the surface being buried.31

time)36 were obtained at 800 MHz with an INEPT delay of 5.3 ms, a recycle time of one second and spectral windows of 15 ppm and 33 ppm for the 1H and 15N dimensions, respectively. HNHA and HNHB experiments37,38 were also performed at 600 MHz. Quadrature detection in the indirect dimensions was performed in the TPPI mode,39 and water suppression was achieved through WATERGATE sequence40 in all NMR experiments. All 3D and 2D spectra were collected at 298 K, processed using the standard Bruker software (XWINNMR) and analysed through the XEASY program.41

Apo-S46VCopAa cloning and purification

Distance constraints for structure determination were obtained from a 15N-edited 3D NOESY-HSQC and from 2D NOESY spectra by converting NOE cross-peaks intensities into interproton upper distance limits. Stereospecific assignments of diastereotopic protons have been obtained by the analysis of the HNHB experiment and through the program GLOMSA.42 3JHNHA coupling constants were determined through the HNHA experiment. Secondary structure elements were determined on the basis of the 3JHNHA coupling constants and of the backbone NOEs. Backbone dihedral f angles were derived from 3JHNHA coupling constants through the appropriate Karplus equation.38 Backbone dihedral c angles for residue i 2 1 were also determined from the ratio of the intensities of the daN ði 2 1; iÞ and dN aði; iÞ NOEs, present on the 15N(i) plane of residue i in the 15N-edited 3D NOESY-HSQC.43 The copper ion was included in the calculations of the copper-loaded form by adding a new residue in the amino acid sequence as already described.17 Structure calculations were performed using DYANA.44 A total of 300 random conformers were annealed in 10,000 steps using NOE and dihedral angle constraints. The 30 conformers with the lowest target function constitute the final family. Restrained energy minimization was then applied through the SANDER module of the AMBER 5.0 program package.45 The force field parameters for the copper(I) ion were adapted from similar systems.17 The NOE and dihedral angle constraints were applied ˚ 22 and with force constants of 50 kcal mol21 A 32 kcal mol21 rad22, respectively. The program CORMA,46 which is based on relaxation matrix calculations, was used to check the agreement between the experimental and the back-calculated NOESY cross-peaks, evaluated in the final structure. The quality of the structures were evaluated through Ramachandran plots and energetic parameters using the programs PROCHECK,47 PROCHECK-NMR48 and AQUA.48 Structure calculations were run on a cluster of Linux PCs.

The plasmid for the protein expression of CopAa from B. subtilis was prepared as described.6 The single amino acid substitutions were created using the QuickChangee Site-Directed Mutagenesis Kit from Stratagene. Sequencing of the engineered DNA fragments was achieved using an automatic sequencer ABI 377. The expression and purification of the mutants of CopAa domain were performed as described.6 Sample preparation As a precaution against disulphide formation in those proteins such as the present ones containing CXXC motifs, the samples were prepared in a Vac Atmospheres nitrogen chamber at room temperature. Protein concentrations were determined using a calculated extinction coefficient of 2680 M21 cm21.32 The copper(I) derivative was prepared following the procedure described for similar proteins.6,17 Copper content was checked through atomic absorption measurements with a Perkin Elmer 2380 instrument. CD spectra were collected on a JASCO J-810 spectropolarimeter with a fused quartz cuvettes with 0.1 cm path length (Merck). The NMR samples of apo and copper-loaded apoS46VCopAa were in 20 mM sodium phosphate buffer (pH 7), 90% H2O/10% 2H2O. The final protein concentration ranges between 1 mM and 1.5 mM. 2 mM DTT was added in the apo and Cu(I) samples. Approximately 0.6 ml of sample was loaded into high quality NMR tubes, which were capped with latex serum caps in the Vac Atmospheres chamber. In order to check the invariance of the sample during each experiment, the volumes of the peaks as well as chemical shifts in HSQC spectra collected at the beginning and at the end of each set of experiments were checked. NMR experiments NMR spectra were acquired at 298 K on Avance 800, 700 and 600 Bruker spectrometers operating at a proton nominal frequency of 800.13 MHz, 700.13 MHz and 600.13 MHz, respectively. All the triple resonance (TXI 5-mm) probes used were equipped with pulsed field gradients along the z-axis. 2D TOCSY33 spectra were recorded on the 600 MHz spectrometer with a spin-lock time of 100 ms, a recycle time of one second and a spectral window of 15 ppm. Two dimensional NOESY maps34,35 were acquired on the 800 MHz or 700 MHz spectrometers with a mixing time of 100 ms, a recycle time of 1 ms and a spectral window of 15 ppm. The 15Nedited 3D NOESY-HSQC experiments (100 ms mixing

Structure calculations

Relaxation measurements and analysis Relaxation experiments were collected at 298 K on a 1.5 mM apo- and Cu(I)-S46VCopAa sample on a Bruker Avance 600 spectrometer, operating at proton nominal frequencies of 600.13 MHz. 15N R1 ; R2 ; and steady-state heteronuclear NOEs were measured with the gradientenhanced, sensitivity-enhanced, pulse sequences as described by Farrow et al.49 R2 were measured using different refocusing times (tCPMG) in the Carr-PurcellMeiboom-Gill (CPMG) detection scheme to determine the occurrence of exchange processes.50 Sets of


Folding Properties of an N-terminal Domain of CopA

experiments were collected at six CPMG refocusing delays: 450, 550, 700, 850, 1000 and 1150 ms, which allow the observation of exchange between two conformational sub-states with average lifetimes ranging from 100 ms to 400 ms. All experiments use the “water flipback” scheme to suppress the water signal without its saturation.51 A recycle delay of three seconds was used for R1 and R2 : The steady-state heteronuclear 1H – 15N NOE was obtained by recording spectra with and without proton saturation. In the case of reference spectra without proton saturation, a relaxation delay of six seconds was employed, whereas a delay of three seconds prior to the three seconds of proton saturation was employed for spectra with proton saturation. The latter was achieved with a train of 1208 1H pulses at 20 ms intervals. 1024 £ 256 data points were collected for each map, using eight scans for each experiment. Spectral window of 40 ppm in the F1 (15N frequency) dimension and of 16 ppm in the F2 (1H frequency) dimension were used. Quadrature detection in F1 was obtained by using the TPPI method.39 Integration of cross-peaks for all spectra was performed by using the standard routine of the XWINNMR program. Relaxation rates R1 and R2 were determined by fitting the cross-peak intensities measured as a function of the delay within the pulse sequence, to a single exponential decay.52 Errors on the rates were estimated through a Monte Carlo approach.53 Heteronuclear 1H– 15N NOE values were calculated as the ratio of peak volumes in spectra recorded with and without 1H saturation. The heteronuclear 1H – 15N NOE values and their errors were estimated by calculating the mean ratio and the standard error from the available data sets. An estimate of the overall tumbling correlation time and the local correlation times for the NH vector of each residue were derived from the measured R2 =R1 ratios (R2 values with a refocusing delay of 450 ms are used in this estimate). In this analysis, care was taken to remove from the input the relaxation data of those NHs that have exchange contributions to the R2 value or that are exhibiting internal motions producing low 1H– 15N NOE values. R2 measurements in the presence of a CPMG spinecho pulse train The effects of spin-echo pulses on transverse relaxation rates due to exchange processes depend on the average field strength produced by the CPMG train. The latter is determined by the repetition time (tCPMG) with which the 3608 pulses are applied: neff ðs21 Þ ¼

1 2ðTp þ tCPMG Þ


where Tp is the duration of a 1808 15N CPMG pulse (160 ms for the experiments collected at 600 MHz). When the tex of the exchange process is longer than the tCPMG delay between refocusing pulses, no effect on the relaxation rates is observed. Experimental limits on the spin-echo delay length are determined by the duty cycle of the transmitter for short delays and the evolution of 1H– 15N coupling during long delays. Atomic coordinates Resonance assignments and the derived atomic coordinates for a family of acceptable structures are available at the BioMagResBank (accession numbers BMRB5769

(apo form) and BMRB5768 (copper(I) form)) and at the Protein Data Bank (PDB ID 1OQ3 (apo form) and PDB ID 1OQ6 (copper(I) form), respectively.

Acknowledgements We thank Professor R. Udisti for atomic absorption spectroscopy. This work was supported by the European Community (SPINE n8QLG2-CT2002-00988, “Structural proteomics in Europe”) and by Ente Cassa Risparmio di Firenze.

References 1. Finney, L. A. & O’Halloran, T. V. (2003). Transition metal speciation in the cell: insights from the chemistry of metal ion receptors. Science, 300, 931– 936. 2. Puig, S. & Thiele, D. J. (2002). Molecular mechanisms of copper uptake and distribution. Curr. Opin. Chem. Biol. 6, 171– 180. 3. Arnesano, F., Banci, L., Bertini, I., Ciofi-Baffoni, S., Molteni, E., Huffman, D. L. & O’Halloran, T. V. (2002). Metallochaperones and metal transporting ATPases: a comparative analysis of sequences and structures. Genome Res. 12, 255– 271. 4. Banci, L. & Rosato, A. (2003). Structural genomics of proteins involved in copper homeostasis. Accts Chem. Res. 36, 215– 221. 5. Solioz, M. & Vulpe, C. D. (1996). CPx-type ATPases: a class of P-type ATPases that pump heavy metals. Trends Biochem. Sci. 21, 237– 241. 6. Banci, L., Bertini, I., Ciofi-Baffoni, S., D’Onofrio, M., Gonnelli, L., Marhuenda-Egea, F. C. & Ruiz-Duen˜as, F. J. (2002). NMR characterization of the N-terminal domain of a potential copper translocating P-type ATPase from Bacillus subtilis. J. Mol. Biol. 317, 415– 429. 7. Baum, J., Dobson, C. M., Evans, P. A. & Hanley, C. (1989). Characterization of a partly folded protein by NMR methods: studies on the molten globule state of guinea pig a-lactalbumin. Biochemistry, 28, 7 –13. 8. Wang, Y. & Shortle, D. (1995). The equilibrium folding pathway of Staphylococcal nuclease: identification of the most stable chain– chain interactions by NMR and CD spectroscopy. Biochemistry, 34, 15895– 15905. 9. Wang, Y., Alexandrescu, A. T. & Shortle, D. (1995). Initial studies of the equilibrium folding pathway of staphylococcal nuclease. Philos. Trans. Roy. Soc. ser. B, 348, 27 – 34. 10. Eliezer, D. & Wright, P. E. (1996). Is apomyoglobin a molten globule? Structural characterization by NMR. J. Mol. Biol. 263, 531– 538. 11. Ptitsyn, O. B. (1995). Molten globule and protein folding. Advan. Protein Chem. 47, 83 – 229. 12. Dobson, C. M. (1994). Protein folding. Solid evidence for molten globules. Curr. Biol. 4, 636– 640. 13. Schulman, B. A., Kim, P. S., Dobson, C. M. & Redfield, C. (1997). A residue-specific NMR view of the non-cooperative unfolding of a molten globule. Nature Struct. Biol. 4, 630– 634. 14. Sippl, M. J. (1993). Recognition of errors in the

Folding Properties of an N-terminal Domain of CopA







21. 22.



25. 26.

27. 28. 29.


three-dimensional structures. Proteins: Struct. Funct. Genet. 17, 355– 362. Garrett, D. S., Seok, Y. J., Liao, D. I., Peterkofsky, A., Gronenborn, A. M. & Clore, G. M. (1997). N-terminal domain of enzyme I of the Escherichia coli phosphoenolpyruvate:sugar phosphotransferase system by multidimensional NMR. Biochemistry, 36, 2517– 2530. Rosenzweig, A. C., Huffman, D. L., Hou, M. Y., Wernimont, A. K., Pufahl, R. A. & O’Halloran, T. V. (1999). Crystal structure of the Atx1 metallo˚ resolution. Struct. Fold chaperone protein at 1.02 A Des. 7, 605– 617. Banci, L., Bertini, I., Ciofi-Baffoni, S., Huffman, D. L. & O’Halloran, T. V. (2001). Solution structure of the yeast copper transporter domain Ccc2a in the apo and Cu(I)-loaded states. J. Biol. Chem. 276, 8415–8426. Gitschier, J., Moffat, B., Reilly, D., Wood, W. I. & Fairbrother, W. J. (1998). Solution structure of the fourth metal-binding domain from the Menkes copper-transporting ATPase. Nature Struct. Biol. 5, 47– 54. Steele, R. A. & Opella, S. J. (1997). Structures of the reduced and mercury-bound forms of MerP, the periplasmic protein from the bacterial mercury detoxification system. Biochemistry, 36, 6885– 6895. Arnesano, F., Banci, L., Bertini, I., Cantini, F., CiofiBaffoni, S., Huffman, D. L. & O’Halloran, T. V. (2001). Characterization of the binding interface between the copper chaperone Atx1 and the first cytosolic domain of Ccc2 ATPase. J. Biol. Chem. 276, 41365– 41376. Einstein, A. (1956). Investigations on the Theory of the Brownian Movement, pp. 19 – 34, Dover, New York. Colon, W., Elove, G. A., Wakem, L. P., Sherman, F. & Roder, H. (1996). Side chain packing of the Nand C-terminal helices plays a critical role in the kinetics of cytochrome c folding. Biochemistry, 35, 5538–5549. Tsivkovskii, R., MacArthur, B. C. & Lutsenko, S. (2001). The Lys1010 –Lys1325 fragment of the Wilson’s disease protein binds nucleotides and interacts with the N-terminal domain of this protein in a copper-dependent manner. J. Biol. Chem. 276, 2234–2242. Arnesano, F., Banci, L., Bertini, I., Huffman, D. L. & O’Halloran, T. V. (2001). Solution structure of the Cu(I) and apo forms of the yeast metallochaperone, Atx1. Biochemistry, 40, 1528– 1539. Opella, S. J., DeSilva, T. & Veglia, G. (2002). Structural biology of metal-binding sequences. Curr. Opin. Chem. Biol. 6, 217– 223. Rost, B. & Sander, C. (1994). Combining evolutionary information and neural networks to predict protein secondary structure. Proteins: Struct. Funct. Genet. 19, 55– 72. Rost, B. & Sander, C. (1993). Prediction of protein secondary structure at better than 70% accuracy. J. Mol. Biol. 232, 584– 599. Sali, A. & Blundell, T. L. (1993). Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. 234, 779– 815. Babajide, A., Hofacker, I. L., Sippl, M. J. & Stadler, P. F. (1997). Neutral networks in protein space: a computational study based on knowledge-based potentials of mean force. Fold. Des. 2, 261– 269. Koradi, R., Billeter, M. & Wu¨thrich, K. (1996). MOLMOL: a program for display and analysis of macromolecular structure. J. Mol. Graph. 14, 51 – 55.


31. Rodionov, M. A. & Blundell, T. L. (1998). Sequence and structure conservation in a protein core. Proteins: Struct. Funct. Genet. 33, 358– 366. 32. Gill, C. S. & Von Hippel, P. H. (1989). Calculation of protein extinction coefficients from amino acid sequence data. Anal. Biochem. 182, 319– 326. 33. Griesinger, C., Otting, G., Wu¨thrich, K. & Ernst, R. R. (1988). Clean TOCSY for 1H spin system identification in macromolecules. J. Am. Chem. Soc. 110, 7870 –7872. 34. Macura, S., Wu¨thrich, K. & Ernst, R. R. (1982). The relevance of J cross-peaks in two-dimensional NOE experiments of macromolecules. J. Magn. Reson. 47, 351 –357. 35. Wider, G., Macura, S., Kumar, A., Ernst, R. R. & Wu¨thrich, K. (1984). Homonuclear two-dimensional 1H NMR of proteins. Experimental procedures. J. Magn. Reson. 56, 207– 234. 36. Wider, G., Neri, D., Otting, G. & Wu¨thrich, K. (1989). A heteronuclear three-dimensional NMR experiment for measurements of small heteronuclear coupling constants in biological macromolecules. J. Magn. Reson. 85, 426– 431. 37. Archer, S. J., Ikura, M., Torchia, D. A. & Bax, A. (1991). An alternative 3D NMR technique for correlation backbone 15N with side chain Hb resonances in larger proteins. J. Magn. Reson. 95, 636– 641. 38. Vuister, G. W. & Bax, A. (1993). Quantitative J correlation: a new approach for measuring homonuclear three-bond J(HNHa) coupling constants in 15N enriched proteins. J. Am. Chem. Soc. 115, 7772– 7777. 39. Marion, D. & Wu¨thrich, K. (1983). Application of phase sensitive correlated spectroscopy (COSY) for measurements of proton– proton spin – spin coupling constants in proteins. Biochem. Biophys. Res. Commun. 113, 967– 974. 40. Piotto, M., Saudek, V. & Sklenar, V. (1992). Gradienttailored excitation for single quantum NMR spectroscopy of aqueous solutions. J. Biomol. NMR, 2, 661 –666. 41. Eccles, C., Gu¨ntert, P., Billeter, M. & Wu¨thrich, K. (1991). Efficient analysis of protein 2D NMR spectra using the software package EASY. J. Biomol. NMR, 1, 111 – 130. 42. Gu¨ntert, P., Braun, W. & Wu¨thrich, K. (1991). Efficient computation of three-dimensional protein structures in solution from Nuclear Magnetic Resonance data using the program DIANA and the supporting programs CALIBA, HABAS and GLOMSA. J. Mol. Biol. 217, 517– 530. 43. Gagne’, R. R., Tsuda, S., Li, M. X., Chandra, M., Smillie, L. B. & Sykes, B. D. (1994). Quantification of the calcium-induced secondary structural changes in the regulatory domain of troponin-C. Protein Sci. 3, 1961 –1974. 44. Gu¨ntert, P., Mumenthaler, C. & Wu¨thrich, K. (1997). Torsion angle dynamics for NMR structure calculation with the new program DYANA. J. Mol. Biol. 273, 283– 298. 45. Pearlman, D. A., Case, D. A., Caldwell, J. W., Ross, W. S., Cheatham, T. E., Ferguson, D. M., et al. (1997). AMBER 5.0, University of California, San Francisco. 46. Borgias, B., Thomas, P. D. & James, T. L. (1989). COmplete Relaxation Matrix Analysis (CORMA).(5.0), University of California, San Francisco, CA. 47. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & Thornton, J. M. (1993). PROCHECK: a program to check the stereochemical quality of protein structures. J. Appl. Crystallog. 26, 283– 291.


Folding Properties of an N-terminal Domain of CopA

48. Laskowski, R. A., Rullmann, J. A. C., MacArthur, M. W., Kaptein, R. & Thornton, J. M. (1996). AQUA and PROCHECK-NMR: programs for checking the quality of protein structures solved by NMR. J. Biomol. NMR, 8, 477– 486. 49. Farrow, N. A., Muhandiram, R., Singer, A. U., Pascal, S. M., Kay, C. M., Gish, G. et al. (1994). Backbone dynamics of a free and phosphopeptide-complexed Src homology 2 domain studied by 15N NMR relaxation. Biochemistry, 33, 5984– 6003. 50. Mulder, F. A., Van Tilborg, P. J., Kaptein, R. & Boelens, R. (1999). Microsecond time scale dynamics in the RXR DNA-binding domain from a combination

of spin-echo and off-resonance rotating frame relaxation measurements. J. Biomol. NMR, 13, 275–288. 51. Grzesiek, S. & Bax, A. (1993). The importance of not saturating H2O in protein NMR. Application to sensitivity enhancement and NOE measurements. J. Am. Chem. Soc. 115, 12593– 12594. 52. Marquardt, D. W. (1963). An algorithm for leastsquares estimation of nonlinear parameters. J. Soc. Ind. Appl. Math. 11, 431– 441. 53. Peng, J. W. & Wagner, G. (1992). Mapping of the spectral densities of N– H bond motions in eglin c using heteronuclear relaxation experiments. Biochemistry, 31, 8571 –8586.

Edited by M. F. Summers (Received 19 March 2003; received in revised form 10 June 2003; accepted 12 June 2003)

Suggest Documents