J Biol Inorg Chem (2002) 7: 617–622 DOI 10.1007/s00775-002-0340-8
O R I GI N A L A R T IC L E
Silvio Aime Æ Nicola D’Amelio Æ Marco Fragai Yong-Min Lee Æ Claudio Luchinat Æ Enzo Terreno Gianni Valensin
A paramagnetic probe to localize residues next to carboxylates on protein surfaces Received: 30 August 2001 / Accepted: 21 December 2001 / Published online: 14 February 2002 SBIC 2002
Abstract It is shown that the paramagnetic properties of lanthanides can be exploited to obtain information on specific parts of a protein surface. Owing to the high affinity of coordinatively unsaturated lanthanide complexes for oxygen donors, carboxylate groups can be used as preferential targets for the interaction. The DO3A ligand is particularly useful in these studies, as it coordinates lanthanides in a heptadentate fashion, leaving two sites available for exogenous donors. A solution of a 15N-labeled sample protein, calbindin D9k (75 residues), was titrated with up to 200% of Gd(III)DO3A complex, and an inversion recovery 15N-1H HSQC experiment was used to measure the paramagnetic contributions to the longitudinal relaxation rates of the amide protons. Relaxation data were used as distance constraints to estimate the number of interacting complexes and the occupancies of their binding sites. Four preferential interaction sites on the protein surface
M. Fragai Æ Y.-M. Lee Æ C. Luchinat (&) Magnetic Resonance Center (CERM), University of Florence, Via L. Sacconi 6, 50019 Sesto Fiorentino, Italy E-mail:
[email protected]fi.it Tel.: +39-055-4574262 Fax: +39-055-4574253 S. Aime Æ E. Terreno Department of Chemistry IFM, University of Turin, Via P. Giuria 7, 10125 Turin, Italy N. D’Amelio Æ G. Valensin Department of Chemistry, University of Siena, Pian dei Mantellini 44, 53100 Siena, Italy M. Fragai Department of Chemistry, University of Florence, Via L. Sacconi, 6, 50019 Sesto Fiorentino, Italy C. Luchinat Department of Agricultural Biotechnology, University of Florence, Via L. Sacconi 6, 50019 Sesto Fiorentino, Italy
are found. Inspection of the various carboxylate side chains on the surface of the protein indicates that Gd(III)-DO3A interacts preferentially with carboxylaterich regions, rather than with isolated carboxylates, suggesting the possibility of chelation of one Gd(III)DO3A molecule by two carboxylate groups. Gd(III)DO3A is thus a valuable semi-selective probe for clusters of negative charges on the protein surface. Keywords Paramagnetic probe Æ DO3A Æ Protein surface
Introduction The genome sequencing project provides new challenges in structural biology and molecular biophysics in general. In order to understand the molecular basis of life, ideally the main features of all biomolecules involved in the complex system of the cell must be determined. The determination of the three-dimensional structure plays a central role. X-ray and NMR techniques are progressing rapidly to this end. Another key issue is that of monitoring protein-protein (or protein-nucleic acid) interactions and/or predicting them on the basis of protein surface features. In this scenario, attention has been dedicated to target protein surface groups. In several cases, this was done through the use of paramagnetic effects. Three main kinds of paramagnetic probes have been exploited. Non-specific probes, like organic radicals, mainly affect the relaxation behavior of all exposed atoms on the macromolecular surface, leaving internal parts less affected [1, 2, 3]. Ideally, these probes do not interact with any specific surface residue. Specific probes, on the other hand, recognize specific targets on the macromolecular structure [4, 5, 6, 7, 8, 9]. They are often based on the formation of selective covalent bonds. One example is the site-directed spin labeling on cysteine residues [10]. A third kind of probe lies in between the first two. They can be broadly defined as probes that interact with
618
surface residues through either electrostatic interactions or coordination bonds. A successful example of the former is the use of lanthanide complexes with strongly negative chelating ligands to selectively interact with positively charged surface residues [11]. The simplest probes that interact through coordination bonds are the aqua ions of paramagnetic metals, such as Co2+ [12, 13], Mn2+ [14], Gd3+ [15], and other lanthanides [16], which have been used for more than 30 years. They form coordination bonds with several different surface residues, so they effectively act as non-specific probes although their effect may not be homogeneously sensed over the whole protein surface. More recently, paramagnetic complexes of Gd3+ and Cr3+with different chelating ligands such as EDTA and DTPA have been used [17, 18, 19]. These complexes may have a limited specificity for carboxylate groups on the protein surface, which is however lowered by the negative overall charge of the complex. A neutral bis-amide derivative of DTPA has been shown to perform well, but again as a non-specific surface probe [20]. That the formation of selective coordination bonds on protein residues is possible is beautifully demonstrated by, for example, the histidinespecific ruthenium(III) complexes developed by Gray and co-workers [21]. It would be desirable to devise a complex of a high relaxing metal such as Gd3+ which could have a specificity for carboxylate groups comparable to that of ruthenium for histidines. In this work we describe the use of a gadolinium complex targeted at solvent-exposed carboxylate groups to yield information on residues nearby. The chosen complex is the Gd(III)-DO3A ligand (Fig. 1), which is known to bind anions such as carbonate, acetate, malonate, lactate, and citrate [22, 23], and has been suggested to interact with donor atoms on the surface of proteins [22]. Indeed, Gd(III)-DO3A, functionalized with hydrophobic substituents in order to increase its affinity for hydrophobic protein pockets, loses its contrast enhancing features owing to the replacement of two coordinated water molecules by carboxylate groups from the protein surface [23, 24]. Gadolinium(III) as a probe is an obvious choice because it has the ability of largely affecting not only the T2 values of nearby nuclei (as all lanthanides do when interacting with macromolecules) but also their T1 [25]. This enables us to measure longitudinal relaxation at a relatively low concentration of the paramagnetic species, thus preventing dramatic
Fig. 1. Interaction of Gd-DO3A with carboxylic moieties
loss in sensitivity. Here we demonstrate that the Gd(III)DO3A complex interacts specifically with exposed carboxylate groups of a sample protein, thus providing an easy and quick tool for obtaining insight on nuclei in the proximity of surface carboxylates. The nuclei chosen are the peptide NH protons, monitored through HSQC spectra on the 15N-labeled protein. The sample protein selected for this study is calbindin D9k, a 9 kDa protein containing two Ca2+ sites. The choice of calbindin D9k was driven by the fact that both X-ray [26, 27, 28] and solution [29] structures are known and that paramagnetic constraints (originating from lanthanide substitution in one of the two Ca2+ binding sites) have been already considered for structure refinement in solution [30]. Calbindin D9k is also a good probe for assessing the different selectivity of Gd(III)-DO3A for different carboxylate groups, owing to the occurrence of several surface Asp and Glu residues.
Materials and methods Sample preparation Protein expression [31] and purification [32] of both the Ca2+ and the apo form of bovine calbindin D9k were performed as reported. The P43M mutant [33, 34] was used to avoid any conformational heterogeneity due to cis-trans isomerization as found for the wildtype protein. The expression system was a generous gift of Prof. S. Forse`n. Uniformly 15N-labeled P43M calbindin D9k was obtained from M9 minimal medium containing (15NH4)2SO4 as the sole nitrogen source. NMR experiments were performed on a sample of 1 mM calbindin D9k in 90% H2O/10% D2O at 300 K, pH 6.0. The paramagnetic Gd(III)-DO3A complex was prepared by titration of the ligand with a GdCl3 solution, following the disappearance of the NMR signals of the diamagnetic species. The paramagnetic complex was then added to calbindin D9k in five steps, corresponding to 3%, 12%, 48%, 98%, and 198% of the protein concentration. NMR spectroscopy NMR spectra were recorded on a Bruker 700 spectrometer equipped with a TXI probe. Amide HN resonances were detected through the 1H-15N HSQC pulse sequence [35, 36] implemented with the sensitivity enhancement scheme [37, 38]. Proton longitudinal relaxation times were measured through a combination of the inversion recovery [39] and the HSQC sequence. This was obtained by introducing a 1H 180 pulse followed by a variable delay in front of the HSQC sequence. T1 values were obtained from a series of spectra with the following delays (in ms): 80, 160, 320, 480, 640, 960, 1280, 1920, 2560, and 3480. Shorter delays (20, 40 ms) were introduced in place of the last two longest delays as the concentration of the paramagnetic complex was increased. The measurement was repeated six times: on the free protein and after each addition of the paramagnetic complex (3%, 12%, 48%, 98%, and 198% of the protein concentration). Peak volumes as a function of the variable delay were fitted by utilizing the Levenberg-Marquardt algorithm for non-linear least-square minimization of the v2 error function [40], using as a model function a three-parameter monoexponential recovery. The plotting and fitting program jplot (Dr. J. Craven, University of Sheffield) was used. This program permits the simultaneous fit and display of multiple magnetization recovery curves. Paramagnetic contributions R1p were calculated as the difference in R1 before and after the complex addition.
619 Structure calculation T1 values were converted into distance constraints as described in the data analysis section. Simulated annealing in the torsion angle space was performed with 10,000 steps using 300 random starting positions of Gd(III)-DO3A complexes around the protein structure, generated with the in-house developed PSEUDYANA [41] module of the program DYANA [42]. During the calculation, only the Gd(III)-DO3A complexes were allowed to move while the structure of calbindin D9k was left unperturbed, fixing the torsion angle values in the input file. No structural refinement of calbindin D9k has been performed.
Results and discussion T1 measurements Titration of calbindin D9k with Gd(III)-DO3A up to a 2:1 molar ratio shows drastic alterations of 15N-1H HSQC cross-peak intensities and line widths, but virtually no shifts of the signals. This demonstrates that interactions with the paramagnetic complex leave the backbone structure unperturbed. Paramagnetic effects are readily quantitated by measuring paramagnetic contributions to R1 (R1p) of each amide HN at increasing concentrations of Gd(III)DO3A in solution. Figure 2 shows the R1p values as a function of residue number for a 0.98:1 Gd(III)DO3A:protein ratio at a protein concentration of 1.0 mM. A different relaxation behavior is observed, depending on the position of the NH protons along the primary structure. Figure 2 shows the presence of several ‘‘hot’’ spots (N-terminus, residues 4, 5, 18, 26, regions 44–48, 64–65, and the C-terminus) along the sequence. The location of Asp and Glu residues is also shown in Fig. 2; some correlation with the R1p effects is immediately apparent. Of course, effects in different
parts of the sequence may be also due to the interaction with the same complex if these regions are close in the three-dimensional structure of the protein. Data analysis The above data suggest that Gd(III)-DO3A has some potential as a reporter for carboxylate groups. However, its future suitability as a carboxylate-specific probe depends on a more detailed analysis of the present Gd(III)DO3A-calbindin D9k system. R1p data were thus used as distance constraints to assess the number and occupancies of interacting sites. The Gd(III)-DO3A complex can be localized on the protein surface using the in-house developed PSEUDYANA [41] module of the program DYANA [42]. Since the DYANA program is able to perform minimization only on one molecular frame, the Gd(III)-DO3A complexes were linked through a large number of pseudoatoms with zero van der Waals radius to one end of the protein and to one another. These linkers made of pseudoatoms do not interfere in the structure calculation (they can penetrate the calbindin D9k structure without causing any steric repulsion) and they allow us to randomly locate more than one Gd(III)DO3A molecule with respect to the protein. In a typical run, 300 random starting positions were generated in this way without varying the protein structure. R1p values were converted into distance constraints in the following way. First, the overall paramagnetic contribution of all Gd(III)-DO3A molecules to the longitudinal relaxation rate of proton m, R1pm, was extracted by subtracting the free protein R1mdia values from those obtained in presence of complex, R1mmeas: dia R1pm ¼ Rmeas 1m R1m
ð1Þ
The experimental quantity R1pm was theoretically interpreted as: R1pm ¼
N X n
fn R1pnm ¼ Kdip
N X fn 6 r nm n
ð2Þ
where N is the number (to be determined) of protein sites capable of interacting with a Gd(III)-DO3A molecule, R1pnm is the contribution to R1pm from the nth site, fn is the occupancy of the nth site, rnm is the distance of proton m from the gadolinium(III) ion in the nth site, and Kdip is the sum of the Solomon and Curie constants [25, 43, 44]:
Fig. 2. Value of the paramagnetic contribution to the relaxation rate for amide protons for a 0.98:1 Gd(III)-DO3A:protein ratio. Black bars refer to glutamate amide protons, white bars refer to aspartate amide protons, gray bars refer to amide protons for different amino acids
2 l 0 2 2 2 2 Kdip ¼ KSolomon þ KCurie ¼ lB ge cI ðSðS þ 1ÞÞ 15 4p ( ) sc2 3sc1 6sc2 þ þ ð1 þ ðxI xS Þ2 s2c2 Þ 1 þ x2I s2c1 ð1 þ ðxI þ xS Þ2 s2c2 Þ 2 l0 2 l4B g4e x2I ðS 2 ðS þ 1Þ2 Þ 3sc þ ð3Þ 5 4p ð1 þ x2I s2c Þ ð3kT Þ2 ðsc1 Þ1 ¼ ðs1e Þ1 þ ðsr Þ1 þ ðsM Þ1
ð4Þ
620
ðsc2 Þ1 ¼ ðs2e Þ1 þ ðsr Þ1 þ ðsM Þ1
ð5Þ
ðsc Þ1 ¼ ðsr Þ1 þ ðsM Þ1
ð6Þ
where l0 is the permeability of vacuum, S is the spin quantum number of the paramagnetic species, lB is the electron Bohr magneton, ge is the free-electron g factor, cI is the proton gyromagnetic ratio, xI and xS are the proton and electron precession frequencies, sc1, sc2, and sc are correlation times, sr is the rotational correlation time of the protein, sM the lifetime of the protein Gd(III)-DO3A adduct, and se1 and se2 are the longitudinal and transverse electronic relaxation times of the metal ion. Contact contributions on protein nuclei were considered negligible, the van der Waals radius of the complex being around 6 A˚. The values of longitudinal and transverse electronic relaxation times at a magnetic field of 16.4 T (700 MHz) can be estimated as se1=5.3·10–7 s and se2=4.3·10–10 s by using the value of electronic relaxation time at zero field of the Gd(III)-DO3A complex (ss0=1.3·10–10 s), and by the mean squared fluctuation D2 of the dynamic ZFS (4.6·1019 s–2) and the associated correlation time for its modulation (sv=1.4·10–11 s) [24, 45, 46] using the equation: ðse1 Þ
1
1 2 sv 4sv ¼ D ð4SðS þ 1Þ 3Þ þ 25 ð1 þ x2S s2v Þ ð1 þ 4x2S s2v Þ ð7Þ
ðse2 Þ1 ¼
1 2 5sv 2sv þ 3 þ D ð4SðS þ 1Þ 3Þ 50 ð1 þ x2S s2v ð1 þ 4x2S s2v Þ
R1pm Kdip
1=6 ¼
N X fn 6 r nm n
!1=6 ð9Þ
The target function to be minimized in the simulated annealing procedures is thus defined as the sum of the differences squared between the experimental R1pm and the R1pm which is calculated by the program using Eqs. 2 and 3. It is clear from Eqs. 2 and 9 that an assumption is needed on the number of metals N and the occupancies fn associated with each of them. Several calculations were thus performed by increasing each time the number of metals N from 1 to 10. For each N, different combinations of fn values were tested. Determination of interaction sites The results of the above PSEUDYANA calculations are summarized in Fig. 3. The minimum effective number of metals N can be inferred by the values of the target function becoming small. However, care must be taken, because, as the number of metals increases, the value of the target function tends to decrease owing to the larger number of degrees of freedom of the system. The expected behavior is the following: 1. The target function will decrease dramatically on passing from 1 to the minimum effective number of metals. 2. The decrease will slow down by overestimating the number.
ð8Þ
Possible changes of these parameters upon binding of the complex to the protein should not affect Eq. 3 much, as terms containing sc2 are dispersed, while terms containing sc1 are dominated by either sr or sM, because sc1 is so long. The rotational correlation time of calbindin, sr, is around 4·10–9 s [47]. A contribution of sM to the correlation time is possible. However, the value of sM observed when the two available sites of Gd(III)-DO3A are occupied by water (sM=7.7·10–8 s–1) [22], and which can be considered a lower limit for sM of a carboxylate ligand, is already much longer than sr, suggesting that sM is unlikely to modulate the dipolar interaction. Therefore, sr is likely to dominate both sc1 and sc in Eq. 3. Exchange contributions to R1p were not considered for similar reasons, the values of the paramagnetic relaxation times being of the order of seconds. This assumption is further justified by the fact that the observed relaxation times change dramatically along the sequence. After insertion of sr as a correlation time in Eq. 3, the R1p values allowed the estimate of the average distance of each amide HN from the paramagnetic centers, weighted by the fractional occupancy of each interaction site:
Fig. 3. Target function (Tf) values as a function of the number of complexes supposed to interact with calbindin. Error bars reflect the variability in target function values observed with different occupancy combinations
621
3. The system will tend to reject additional metals by placing them far away from the protein, thus quenching their paramagnetic effects. In the case of one metal, the high value of the target function (Fig. 3), together with the position of the complex in the center of the protein (large van der Waals violations), clearly shows that one metal cannot justify the experimental data. Two metals, either with the same or different occupancies, are in all cases much better than one, giving sizably lower values of the target function. The two complexes are located on the surface and the van der Waals steric hindrance is removed. The calculation was repeated with an increasing number of complexes and, for each N value, by increasing the number of different fn combinations1, until the system started to reject the paramagnetic complexes. As shown in Fig. 3, the target function decreases with an increasing number of metals up to four metals, after which its value remains nearly constant, the exceeding complexes being rejected. In all calculations, the assumption of full occupancy is justified by (1) the relatively low concentration of the complex with respect to the available interaction sites and (2) by sample calculations assuming half overall occupancy, for which the complexes ended up too close to the surface and yielded serious van der Waals violations. Once the number of interacting complexes was determined to be four, some modifications were made in order to check the reliability of the localization of the metal binding sites. To do so: 1. The complex was substituted by a single atom with a van der Waals radius of 2.5 A˚ to avoid any steric clash that could originate from a wrong approach with the protein (the complex may approach from the side where no coordination site is available for the protein). 2. In the PSEUDYANA calculation the van der Waals terms of the atoms of the carboxylates in the terminal part of the glutamate and aspartate side chains were removed in order to avoid possible wrong steric clashes due to the fact that the side chains have a fixed conformation in the model structure and are not allowed to move to interact with the complex.
the third, the third twice the fourth). The resulting position of the four sites is shown in Fig. 4. The target function value indicates site 1 as the main interaction site. Interestingly, this site is the first to be occupied when assuming only two bound metals (the other being located in a average position among sites 2, 3, and 4). A closer inspection of the shape of the protein surface reveals that the four sites are located close to several of the carboxylate groups of the protein. Site 1 may be constituted by Glu4-Glu5 plus the C-terminal end, which are close in space, site 2 by Glu17-Asp19-Glu26, which again are close in space, site 3 by Asp47-Glu48Glu51, and site 4 by Glu64. Apparently, with the exception of the isolated Glu64, carboxylate-rich regions seem to be preferred for the interaction. The possibility of simultaneous chelation by two carboxylates is demonstrated by the existence of an adduct of an amide analogue of Gd(III)-DO3A with lactate (K=4000 M–1 at 298 K) and malonate (K=20,000 M–1 at 298 K) [23, 48]. Interestingly, the Eu(III) complex of the same ligand interacting with the carbonate ion had a lower binding constant value (K=1000 M–1 at 295 K, pH 7.4), suggesting the likelihood to achieve higher affinity for close carboxylate pairs [23] than for isolated carboxylates. The Glu64 (Glu65 is involved in the binding of calcium) is isolated but very much projected towards the solvent and largely accessible even for large molecules. The presence of only four main regions further suggests that synergism between different carboxylate residues is one of the factors determining the selectivity of the interaction. This also explains the larger effect observed at both termini. In conclusion, the Gd(III)-DO3A complex interacts with exposed carboxylate groups. The interaction with the test protein calbindin D9k indicates a good selectivity for carboxylate residues, preferentially when close to one another. In some cases a chelate interaction through two carboxylates belonging to residues well separated in the primary structure can be hypothesized, provided they are close enough for a proper binding. Moreover, the interaction has been shown to leave the backbone structure unperturbed, as evident from the absence of significant shifts in the 15N-1H HSQC spectrum after the addition of the complex.
For the reasons explained above, in all cases full association was assumed. The lowest target function value was obtained in the case where each site has a different occupancy (the first twice the second, the second twice 1 For instance, in the case of N=4, the following cases were considered: 1. n equal occupancies: fn1=fn2=fn3=fn4=1/4. 2. Three metals having twice the f value of the remaining one: fn1=fn2=fn3=2/7, fn4=1/7. 3. One metal having twice the value of the remaining three: fn1=2/5, fn2=fn3=fn4=1/5. 4. Two metals having twice the value of the remaining two: fn1=fn2=1/3, fn3=fn4=1/6. 5. The first metal having twice the value of the second, the second twice of the third, and the third twice the fourth: fn1=8/15, fn2=4/15, fn3=2/15, fn4=1/15
Fig. 4. Stereo representation of the interaction sites in calbindin. Gd(III)-DO3A complexes are shown as spheres. Positions of the metals refer to the best fit assuming full occupancy
622 Acknowledgements This work has been supported by Murst ex 40%, Italy, the European Union, LSF contract HPRI-CT-199900009, CNR, Progetto Finalizzato Biotecnologie, Italy, contract 990039349, and CNR, Comitato nazionale per le scienze chimiche, Italy, contract 970113349.
References 1. Esposito G, Lesk AM, Molinari H, Motta A, Niccolai N, Pastore A (1992) J Mol Biol 224:659–670 2. Scarselli M, Bernini A, Segoni C, Molinari H, Esposito G, Lesk AM, Laschi F, Temussi P, Niccolai N (1999) J Biomol NMR 15:125–133 3. Yuan T, Ouyang H, Vogel HJ (1999) J Biol Chem 274:8411–8420 4. Kosen PA, Scheek RM, Naderi H, Basus VJ, Manogaran S, Schmidt PG, Oppenheimer NJ, Kuntz ID (1986) Biochemistry 25:2356–2364 5. De Jong EAM, Claesen CAA, Daemen CJM, Harmsen BJM, Konings RNH, Tesser GI, Hilbers CW (1988) J Magn Reson 80:197–213 6. Rehmann JP, Barton JK (1990) Biochemistry 29:1710–1717 7. Ramos A, Varani G (1998) J Am Chem Soc 120:10992–10993 8. Dunham SU, Turner CJ, Lippard SJ (1998) J Am Chem Soc 120:5395–5406 9. Kechuan T, Gochin M (1999) J Am Chem Soc 121:9276–9285 10. Battiste JL, Wagner G (2000) Biochemistry 39:5355–5365 11. Dick LR, Geraldes CFGC, Sherry AD, Gray CW, Gray DM (1989) Biochemistry 18:7896–7904 12. McDonald CC, Phillips WD (1969) Biochem Biophys Res Commun 35:43–51 13. Sherry AD, Teherani J (1983) J Biol Chem 258:8663–8669 14. Lane AN, Jardetzky O (1985) Eur J Biochem 152:411–418 15. Reuben J (1971) Biochemistry 10:2834–2838 16. Agresti DG, Lenkinski RE, Glickson JD (1977) Biochem Biophys Res Commun 76:711–719 17. Kopple KD, Zhu XH (1983) J Am Chem Soc 105:7742–7746 18. Kemple MD, Ray BD, Lipkowitz KB, Prendergast FG, Rao BDN (1988) J Am Chem Soc 110:7863–7888 19. Petros AM, Mueller L, Kopple KD (1990) Biochemistry 29:10041–10048 20. Liepinsh E, Baryshev M, Sharipo A, Ingelman-Sundberg M, Otting G, Mkrtchian S (2001) Structure 9:457–471 21. Yocom KM, Shelton JR, Schroeder WA, Worosila G, Isied SS, Bordignon E, Gray HB (1982) Proc Natl Acad Sci USA 79:7052–7055 22. Aime S, Botta M, Geninatti Crich S, Giovenzana G, Pagliarin R, Sisti M, Terreno E (1998) Magn Reson Chem 36:S200–S208 23. Bruce JI, Dickins RS, Govenlock LJ, Gunnlaugsson T, Lopinski S, Lowe M, Parker D, Peacock RD, Perry JJ, Aime S, Botta M (2000) J Am Chem Soc 122:9674–9684
24. Caravan P, Ellison JJ, McMurry TJ, Lauffer RB (1999) Chem Rev 99:2293–2351 25. Bertini I, Luchinat C (1996) NMR of paramagnetic substances. Coord Chem Rev 150:1–296 26. Svensson LA, Thulin E, Forse´n S (1992) J Mol Biol 223:601– 606 27. Szebenyi DME, Moffat KJ (1986) J Biol Chem 261:8761–8777 28. Anderson M, Malmendal A, Linse S, Ivarson I, Forse´n S, Svensson LA (1997) Protein Sci 6:1139–1147 29. Kordel J, Skelton NJ, Chazin WJ (1993) J Mol Biol 231:711– 734 30. Allegrozzi M, Bertini I, Janik MBL, Lee Y-M, Liu G, Luchinat C (2000) J Am Chem Soc 122:4154–4161 31. Brodin P, Grundstrom T, Hofmann T, Drakenberg T, Thulin E, Forse´n S (1986) Biochemistry 25:5371–5377 32. Johansson C, Brodin P, Grundstrom T, Thulin E, Forse´n S, Drakenberg T (1990) Eur J Biochem 187:455–460 33. Malmendal A, Carlstro¨m G, Hambraeus C, Drakenberg T, Forse´n S, Akke M (1998) Biochemistry 37:2586–2595 34. Chazin WJ, Ko¨rdel J, Drakenberg T, Thulin E, Brodin P, Grundstrom T, Forse´n S (1989) Proc Natl Acad Sci USA 86:2195–2198 35. Bodenhausen G, Ruben DJ (1980) Chem Phys Lett 69:185–188 36. Kay LE, Keifer P, Saarinen T (1992) J Am Chem Soc 114:10663–10665 37. Palmer AG III, Cavanagh J, Wright PE, Rance M (1991) J Magn Reson 93:151–170 38. Cavanagh J, Palmer AG III, Wright PE, Rance M (1991) J Magn Reson 91:429–436 39. Vold RL, Waugh JS, Klein MP, Phelps DE (1968) J Chem Phys 48:3831–3832 40. Press WH, Flannery BP, Teukolsky SA, Vetterling WT (1988) Numerical recipes in C: the art of scientific computing. Cambridge University Press, New York 41. Banci L, Bertini I, Cremonini MA, Gori Savellini G, Luchinat C, Wu¨thrich K, Gu¨ntert P (1998) J Biomol NMR 12:553–557 42. Gu¨ntert P, Mumenthaler C, Wu¨thrich K (1997) J Mol Biol 273:283–298 43. Eaton DR, Fischer RD, Hawkins CJ, Horrocks WD Jr, Jesson JP, Kreilick RW, Kurland RJ, La Mar GN, Langford CH, Phillips WD, Pignolet LH, Rettig MF, Swift TJ (1973) NMR of paramagnetic molecules. Academic Press, New York 44. Bertini I, Luchinat C (1986) NMR of paramagnetic molecules in biological systems. Benjamin/Cummings, Menlo Park, Calif 45. Bloembergen N, Morgan LO (1961) J Chem Phys 34:842–850 46. Rubinstein M, Baram A, Luz Z (1971) Mol Phys 20:67–80 47. Bertini I, Lee Y-M, Luchinat C, Piccioli M, Poggi L (2001) ChemBioChem 2:550–558 48. Aime S, Botta M, Bruce JI, Maniero V, Parker D, Terreno E (2001) Chem Commun 115–116
