Eur. J. Biochem. 210,73-77 (1992)
0FEBS 1992
The effect of metal binding on the structure of annexin V and implications for membrane binding ', Robert HUBER' and Gerhard BOD03 Laboratoire pour I'Utilisation du Rayonnement Electromagnetique, Centre Universitaire Paris Sud, Orsay, France Max-Planck-Institutfur Biochemie, Martinsried, Federal Republic of Germany
Anita LEWIT-BENTLEY', Solange MORERA
Bender & Co., Vienna, Austria
(Received May 20, 1992) - EJB 92 0704 The structure of annexin V, crystallised in the presence of two calcium or barium ions for each protein molecule, was solved by molecular replacement to 0.24 nm resolution. The two metal ions are found in domains I and IV, i.e. on the same side of the channel that lies in the centre of the molecule. The structures of the barium and calcium form are extremely close, the only differences localised in the metal-binding sites that lie on the surface of the molecule. The occupancies of the metal ions, however, are lower for barium than for calcium, expressing the lower affinity of the protein for the former. The packing of the annexin molecules in the crystal asymmetric unit may represent a model for the calcium driven association of membrane-bound annexins that leads to membrane fusion. Annexins are a family of proteins that bind phospholipids in a calcium-dependent manner. They are found in a wide range of tissues, with not well understood, but apparently very diverse, physiological effects, a fact that has lead to some confusion in their nomenclature [l]. Their amino acid sequences show a high degree of similarity, with a segment of about 70 amino acids being repeated four times (eight times in annexin VI). Within each of these segments, a highly conserved sequence of 17 amino acids was suggested as the site of calcium and perhaps phospholipid binding [2]. The differences between members of the annexin family reside in the N-terminal region which has 4- 167 amino acids (annexins VI and VII, respectively). Interest in these proteins is mainly stimulated by their anticoagulant and anti-inflammatory properties, as well as by the capacity of some of them to promote membrane fusion, to generate voltage-dependent ion-conducting channels [3,4] or to associate with cytoskeletal proteins (for review see [5, 61). The crystal structure of annexin V has been solved recently [7 -91. The structure is highly a-helical, with each sequence domain corresponding to a structural domain. Each domain consists of five a-helices (A-E), four of which are roughly antiparallel, while helix C is approximately perpendicular to them. The overall dimensions of the molecule are 6.4 x 4.0 x 3.0 nm, with the N- and C-termini on one, somewhat concave, side (Fig. 1). The molecule is formed by two halves, clearly separated from each other, but related to each other by a local twofold axis of symmetry. One half of the molecule consists of domains I and IV, the other of domains I1 and 111; within each half, a local twofold axis, parallel to the pricipal one, relates the domains to each other. While the interactions between domains within each half of the molecule (I and IV on the one hand, and I1 and 111 on the other hand) are fairly tight and hydrophobic, there is a clear separation between the two halves, with only two short stretches of polyCorrespondence to A. Lewit-Bentley, LURE, Blt. 209D, Universite Paris Sud, F-91405 Orsay, France
peptide chain (residues 87 - 89 and 245 - 247) connecting them together at the concave side of the molecule. The two halves are separated by several tenths of a nanometer at the convex side (top of Fig. l), creating a space lined with polar side chains and which was proposed as the site for the calcium channel 141. Calcium ions were seen only in a crystal form which had been soaked in a high concentration of calcium salts 181. They were indeed found within the 17-amino-acid sequence as predicted, except in domain 111. The calcium site is in each case formed by a tight loop between helices A and B, with three carbonyl oxygens within the loop, a carboxyl group from an amino acid some 39-40 residues further on in the sequence, and one water molecule as ligands. The sequence of the calcium site is very specific: hKGf;GT, where the glycines allow the tight-loop geometry. The sequence of the third repeat in annexin V, as well as of the first repeat in annexins I and 11, does not correspond to this pattern, and thus cannot bind calcium in the same manner. All the calcium-binding loops lie on the side opposite to the N- and C-termini, on the convex side of the molecule and protrude above the surface of the molecule. The binding of annexins to phospholipids is dependent on the presence of calcium. Among other divalent cations studied, zinc can increase the binding in the presence of low calcium concentrations, but does not promote phospholipid binding by itself. Magnesium and barium inhibit binding of annexin V to phospholipids [lo]. The affinity of annexin V for divalent and trivalent cations seems closely related to their ionic radius (ionic radii of Ca2+, La3+ and BaZ+ are 0.1 nm, 0.103 nm and 0.135 nm, respectively) Ill], with BaZf having the lowest (Kd = 1 mM) and La3+ the highest ( K d = 10 pM) affinity constants (H. Haigler, personal communication). We have crystallised annexin V at low ionic strength and low calcium concentration, as well as in the presence of barium. An analysis of differences between these structures and those reported earlier show significant movements within the molecule in response to crystal packing and/or calcium binding.
74 Table 1. Crystal and refinement data. Rsym= C ( I - ( I ) ) / C ( r ) for equivalent measurements. RMS, root mean square.
Ion
Parameter resolution
Rsym
nm Ca
Ba
1.20-0.24 1.20-0.24
Z > 2a(4
cornplete-
no. of reflections
ness
total
R factor
83.6 74.3
unique
bonds
nm 98 82
no. of solvent
cutoff
%
0.067 0.086
resolution amplitude RMS deviation
98492 31600 50558 23533
0.181 0.208
MATERIALS AND METHODS Recombinant human annexin V was prepared as before [12]. Crystals were grown at low ionic strength using poly(ethylene glycol) 20000 as a precipitant (50 mM Tris/maleate, pH = 6.6) with 2-3 mol Ca2+/mol protein [13]. The conditions for the barium-containing crystals were the same, except that a higher excess of the metal ion was used (up to 4 mol Ba2+/molprotein). The crystals are monoclinic, space group P21, with two molecules in the asymmetric unit and with the following cell dimensions: a = 8.39 nm, b = 8.09 nm, c = 7.14nm, j l = 108.7' (Ca crystals); a = 8.41 nm, b = 8.06 nm, c = 7.17 nm, /I = 107.8" (Ba crystals). Data were collected on a Xentronics area detector using the data-reduction programme XENGEN [14] for the Ca data and XDS [15] for the Ba data. The calcium and barium data are not isomorphous beyond 0.3 nm resolution, and we therefore decided to refine the two structures separately. The starting point for refinement of the calcium structure was obtained by molecular replacement, using the hexagonal annexin-V crystal-structure coordinates [7] as a model. The Patterson space rotation function [16] (programme package PROTEIN) gave two principal peaks with maximum correlation of 13.5 and 12.8 (the next maximum, very broad, was at 12.5), the two related by a rotation obtained from a selfrotation function search. The subsequent translation function searches gave peaks of 13.7 root mean square and 11.9 root mean square with PROTEIN. We confirmed the solution using the CCP4 programme package [ 171: the self-rotation function confirmed a non-crystallographic twofold axis (a peak at 0.8 of the maximum) lying perpendicular to the crystal h-axis and bisecting the monoclinic angle p. Refinement of the structures was carried out using XPLOR [18], followed by manual corrections, addition of metal ions and solvent molecules on the graphics display using FRODO [19]. In the final stages, the programme PROLSQ [20] was used. Table 1 summarises the measurement and refinement data for both crystal forms. The poorer refinement result for the barium structure is certainly due to poorer crystal quality. The monoclinic crystal form is indecd difficult to obtain reproducibly and in the form of single crystals, with multiple crystals often impossible to diagnose.
RESULTS AND DISCUSSION Calcium sites
In the previous crystallographic study of annexin V [8], three calcium sites and two extra lanthanum sites were described. The exchange of lanthanum for calcium went in the
0.7-0.24 0.7-0.24
24F) 248')
angles
planes
nm
nm
0.0017 2.4" 0.0017 2.4"
0.0004 0.0004
molecules
266 265
order of sites 1 > 2 > 3, where sites 1, 2 and 3 correspond to domains 11, IV and I, respectively, suggesting that the calcium is most strongly bound to domain 1. In the P21 crystal form, we find only two calcium sites, which are in domains 1 and IV (corresponding to sites 3 and 2 respectively of the previous paper) [8]. The calcium binding is therefore asymmetric with respect to the internal symmetry of the molecule, with both calcium ions binding to the same side of the central pore of the molecule. The calcium-binding loop of domain I1 has adopted a conformation different from either the low calcium (P6,) [7] or the high calcium (R3) [8] form (Fig. 2). The annexin V structure presented here, with two ions of calcium bound to domains 1 and IV, shows significant differences from either of the structures published so far. In the first crystal form (P63, low Ca) prepared in low calcium concentration, no calcium could be found. In the second one (R3, high Ca), obtained by soaking crystals at high calcium concentration, three calcium ions were bound to the protein. When the low-Ca and high-Ca structures were superimposed, the high-Ca form was shown to be more open by about 4', via a motion of the two halves of the molecule about a hinge situated between them at the level of the connecting peptides, thus increasing the channel between the two halves [8]. The structure of the intermediate calcium form described here opens even further by roughly the same amplitude (Fig. 3). We suggest effects of calcium binding and/or different packing in the different crystal forms compared to be at the origin of the conformational differences observed, which certainly document a remarkable flexibility of the molecule. Calcium affinity of annexins in solution is relatively low, becoming much higher in the presence of phospholipids (the affinity constants change from millimolar to micromolar, the exact value being different for each annexin). An early study of calcium binding by annexin V [21] indicated two conformational changes occuring at two different relative calcium concentrations, suggesting the presence of Ca sites with two different affinities. Binding studies of annexin V to phospholipids indicated the presence of at least two Ca sites, and the process was shown to be cooperative [22,23]. It would certainly be interesting to know if the calcium substitution seen in the three crystal forms studied so far corresponds to the observed behaviour in solution, as well as the implications for membrane binding. Barium and possible zinc binding
The structure of the annexin-V crystals prepared in the presence of barium shows that the barium ions have replaced calcium ions in the same two sites, in domains I and IV. The geometry of the sites is similar, but affected by the difference
75 in ionic radius of the two ion types: the carbonyl-oxygenCa distances of 0.235 - 0.245 nm increase to 0.295 - 0.305 nm for Ba. The loops containing the carbonyl-oxygen ligands move away from the barium ion (the maximum displacement is 0.185 nm at the Ca position of the central ligand), while the carboxylic-acid group is displaced only slightly compared to the calcium form (Fig. 4). The perturbation of the structure due to the presence of barium is limited to the metal ion site, the calcium and barium structures being superimposable everywhere else (root-mean-square difference at Ca positions for the two molecules in the asymmetric unit is 26 pm). This is certainly due to the special position of the metal-binding sites on the molecule surface
The occupancy of the ion sites is, however, much lower for barium (0.15 -0.45) than for calcium (1 .0), and the protein loop coordinating the metal ion is slightly less well ordered for barium than for calcium. Furthermore, the occupancy of site 3 , (domain I) is higher than that of site 2 (domain IV) in both molecules of the asymmetric unit (0.45 and 0.35 for site 3, and 0.15 and 0.3 for site 2, for the two molecules, respectively), even though both sites are equally accessible to solvent in the crystal packing. In fact, the barium ions of sites 3 and 2 from symmetrically related molecules come to within 1.2 nm of each other in the crystal lattice. In the second molecule, crystallographically distinct, the difference in occupancy is smaller but still significant. The difference in occu-
Fig. 1. Ribbon drawing [27]of the schematic of the annexin-V structure. Domains I, 11, 111 and IV are cyan, red, yellow and blue respectively. Fig. 3. Superimposition of the Ca backbones of the three crystal forms. Domains IT and 111 were aligned to give the best superimposition. The colours are the same as in Fig. 2.
Fig. 2. The conformation of the calcium site in domain I1 in the three form with no calcium is in red, the P21 form crystal forms. The with two calcium ions in yellow, the R3 form with three calcium ions in blue.
Fig. 4. Superimposition of the metal-binding loop in domain I for the calcium (pale green) and barium (red) structures.
16
/
I
,
Fig. 5. The hypothetical Zn-binding site. The dotted lines represent distances of 0.22-0.35 nm from the zinc ion, which is represented by a small pyramid.
Fig. 6. The interactions of the N- and C-termini of the two molecules within the asymmetric unit. Hydrogen bonds are represented by dotted lines.
pancies is a further confirmation of a difference in affinity for metal ions of the three binding sites within the annexin-V molecule (the affinity decreasing in the order site 3 > 2 9 1). In the case of the more bulky B a z f , for which the overallbinding affinity is lower, the relative difference in the affinity between the different sites becomes more pronounced. The decreased membrane-binding affinity of annexin V in the presence of barium [lo] may therefore most likely be related to the lower affinity of the protein for the metal ion, rather than to any perturbation of the protein conformation. The sequence of annexin V contains three histidine residues, of which two (His98 and His267) lie only about 0.3 nm below the molecular surface, on either side of the central channel, extending across it. When considering the synergistic effect of zinc at low calcium concentration upon membranebinding affinity [lo], these two histidines seem good candidates for zinc coordination. In the P21 crystal form, a zinc ion could lie at a distance of 0.23nm from the N n position of both histidines, with other charged side chains within a reasonable distance (Fig. 5). The synergistic effect of zinc ions at low calcium concentrations could be explained by its bridging of the protein molecule across the central cleft via the histidine side chains, and thus help to bring about the conformational change (movement across the channel) seen in the high-Ca form.
phenomenon has been observed for annexins VII, VI and IV in solution using light scattering and fluorescence [24]. The Ca-dependent self-association of synexin (annexin VII) was seen to start at about 2 mol Ca2+/mol protein. It became much more pronounced in the presence of membranes and, furthemore, the self-association of annexin molecules accompanied the aggregation of chromaffin granule membranes. In a comparative study of the Ca-dependent interactions of annexins I, 11, 111, V and VI with phospholipid vesicles, annexins I and I1 were found to promote their aggregation and the fusion of large unilamelar phospholipid vesicles [25]. Calpactin I (annexin 11) has been shown by freeze-etch electron microscopy to form a bridge between the surfaces of liposomes and chromaffin vesiclesjust before their fusion [26]. In the P21 crystal form the asymmetric unit contains two molecules that are related by a non-crystallographic twofold axis. These two molecules interact through their concave surfaces that contain the N- and C-termini. The two termini are in fact stabilised by the dimer interaction and have become well ordered compared to the crystal forms studied earlier. There is a network of hydrogen bonds and salt bridges connecting the N-terminus of one molecule (residues 2 - 6) with the C-terminus of the second molecule (residues 319 - 320, the annexin-V amino acid sequence is numbered starting with Ala2) and vice versa (Fig. 6). Assuming that membrane binding occurs through the calcium sites situated on the opposite sides of the molecules, it is tempting to suggest that the annexin dimer as it condensed in our crystals represents a model for a role of annexin in the aggregation of membranes: the molecules bind to the surface of separate membranes via their calcium sites, possibly in contact with each other under the effect of calcium concentration. In annexins I and 11, the length of the N-terminal sequence is much larger than in annexin V (by 27 and 18 residues, respectively), while annexin I1 interacts via its N-terminus with a small protein, p l l , to
Implications for membrane interactions
The preparation of the crystals, grown in a low-ionicstrength medium, allowed a fairly good control of the stoichiometry of protein vs calcium-ion concentration. In fact, as soon as the Ca2+ concentration rose above about 3 mol/mol protein, the protein started aggregating to form a precipitate and microcrystals, and above 5mol CaZf/mol protein the precipitation became very fast and pronounced. A related
77 form a heterodimer. The proposed arrangement could promote the association through the N- and C-termini of these annexins even more. We thank Dr. C. Cambillaud of the LCCMB, Marseille, and Dr. G. A. Bentley of the Pasteur Institute, Paris, for giving us the possibility to collect data in their laboratories.
REFERENCES 1. Crumpton, M. J. & Dedman, J. R. (1990) Nature 345, 212. 2. Glenney, J. R. (1986) J . Biol. Chem. 261, 7247-7252. 3. Rojas, E., Pollard, H. B., Haigler, H. T., Parra, C. & Burns, A. I . . (1990) J . Biol. Chem. 265,21207-21215. 4. Karshikov, A,, Berendes, R., Burger, A,, CavaliC, A,, Lux, H. D. & Huber, R. (1992) Eur. Biophys. J . 20, 337-344. 5. Klee, C. B. (1988) Biochemistry 27, 6645-6653. 6. Crompton, M. R., Moss, S. E. & Crumpton, M. J. (1988) Cell 55, 1-3. 7. Huber, R., Romisch, J. & Paques, E.-P. (1990) EMBO J. 9,3867 3874. 8. Huber, R., Schneider, M., Mayr, I., Romisch, J. & Paques, E.-P. (1990) FEBS Lett. 275, 15-21. 9. Huber, R., Berendes, R., Burger, A,, Schneider, M., Karshikov, A., Luecke, H., Romisch, J. & Paques, E.-P. (1992) J. Mol. Biol. 223, 683 - 704. 10. Andree, H. A. M., Reutelingsperger, C. P. M., Hauptman, R., Hemker, H. C., Hermens, W. T. & Willems G. (1990) J. B i d . Chem. 265,4923-4928.
11. Pauling, L. (1960) Nature of the chemical bond, 3rd edn, Cornell University Press, Ithaca, New York. 12. Maurer-Fogy, I., Reutelingsperger, C. P. M., Peiters, J., Bodo, G., Stratowa, C. & Hauptman, R. (1988) Eur. J . Biochem. 174, 585 - 592. 13, Lewit-Bentley, A., DoubliC, S., Fourme, R. & Bodo, G. (1989) J . Mol. Biol. 210, 875 - 876. 14. Howard, A. J., Gilliland, G. L., Finzel, B. C., Poulos, T. L., Ohlendorf, D. H. & Salemme, F. R. (1987) J . Appl. Crystallogr. 20, 383 - 387. 15. Kabsch, W. (1988) J . Appl. Crystallogr. 21,916-924. 16. Huber, R. (1965) Acta Crystallogr. 19, 353. 17. CCP4 (1979). The SERC ( U K ) collaborative computing project no. 4 , Daresbury Laboratory, Warrington. 18. Brunger, A. T., Kuriyan, M. & Karplus, M. (1987) Science 235, 458. 19. Jones, T. A. (1978) J. Appl. Crystallogr. 11, 268-272. 20. Konnert, J. H. & Hendrickson, W. A. (1980) Acta Crystallogr. A36, 344 - 350. 21. Shadle, P. J., Gerke, V. & Weber, K. (1985) J . Biol. Chem. 260, 16354-16360. 22. Glenney, J. R. Jr, Tack, B. & Powell, M. A. (1987) J . CellBiol. 104, 503-511. 23. Schlaepfer, D. D., Mehlman, T., Burgess, W. H. & Haigler, H. T. (1987) Proc. Natl Acad. Sci. USA 84, 6078 -6082. 24. Zaks, W. J. & Creutz, C. E. (1991) Biochemistry 30,9607-9615. 25. Blackwood, R. A. & Ernst, J. D. (1990) Biochem. J . 266, 195200. 26. Nakata, T., Sobue, K. & Hirokawa, N. (1990) J . Cell Biol. 110, 13-25. 27. Priestley, J. P. (1988) J . Appl. Crystallogr. 21, 572-576.
0FEBS 1992
The effect of metal binding on the structure of annexin V and implications for membrane binding ', Robert HUBER' and Gerhard BOD03 Laboratoire pour I'Utilisation du Rayonnement Electromagnetique, Centre Universitaire Paris Sud, Orsay, France Max-Planck-Institutfur Biochemie, Martinsried, Federal Republic of Germany
Anita LEWIT-BENTLEY', Solange MORERA
Bender & Co., Vienna, Austria
(Received May 20, 1992) - EJB 92 0704 The structure of annexin V, crystallised in the presence of two calcium or barium ions for each protein molecule, was solved by molecular replacement to 0.24 nm resolution. The two metal ions are found in domains I and IV, i.e. on the same side of the channel that lies in the centre of the molecule. The structures of the barium and calcium form are extremely close, the only differences localised in the metal-binding sites that lie on the surface of the molecule. The occupancies of the metal ions, however, are lower for barium than for calcium, expressing the lower affinity of the protein for the former. The packing of the annexin molecules in the crystal asymmetric unit may represent a model for the calcium driven association of membrane-bound annexins that leads to membrane fusion. Annexins are a family of proteins that bind phospholipids in a calcium-dependent manner. They are found in a wide range of tissues, with not well understood, but apparently very diverse, physiological effects, a fact that has lead to some confusion in their nomenclature [l]. Their amino acid sequences show a high degree of similarity, with a segment of about 70 amino acids being repeated four times (eight times in annexin VI). Within each of these segments, a highly conserved sequence of 17 amino acids was suggested as the site of calcium and perhaps phospholipid binding [2]. The differences between members of the annexin family reside in the N-terminal region which has 4- 167 amino acids (annexins VI and VII, respectively). Interest in these proteins is mainly stimulated by their anticoagulant and anti-inflammatory properties, as well as by the capacity of some of them to promote membrane fusion, to generate voltage-dependent ion-conducting channels [3,4] or to associate with cytoskeletal proteins (for review see [5, 61). The crystal structure of annexin V has been solved recently [7 -91. The structure is highly a-helical, with each sequence domain corresponding to a structural domain. Each domain consists of five a-helices (A-E), four of which are roughly antiparallel, while helix C is approximately perpendicular to them. The overall dimensions of the molecule are 6.4 x 4.0 x 3.0 nm, with the N- and C-termini on one, somewhat concave, side (Fig. 1). The molecule is formed by two halves, clearly separated from each other, but related to each other by a local twofold axis of symmetry. One half of the molecule consists of domains I and IV, the other of domains I1 and 111; within each half, a local twofold axis, parallel to the pricipal one, relates the domains to each other. While the interactions between domains within each half of the molecule (I and IV on the one hand, and I1 and 111 on the other hand) are fairly tight and hydrophobic, there is a clear separation between the two halves, with only two short stretches of polyCorrespondence to A. Lewit-Bentley, LURE, Blt. 209D, Universite Paris Sud, F-91405 Orsay, France
peptide chain (residues 87 - 89 and 245 - 247) connecting them together at the concave side of the molecule. The two halves are separated by several tenths of a nanometer at the convex side (top of Fig. l), creating a space lined with polar side chains and which was proposed as the site for the calcium channel 141. Calcium ions were seen only in a crystal form which had been soaked in a high concentration of calcium salts 181. They were indeed found within the 17-amino-acid sequence as predicted, except in domain 111. The calcium site is in each case formed by a tight loop between helices A and B, with three carbonyl oxygens within the loop, a carboxyl group from an amino acid some 39-40 residues further on in the sequence, and one water molecule as ligands. The sequence of the calcium site is very specific: hKGf;GT, where the glycines allow the tight-loop geometry. The sequence of the third repeat in annexin V, as well as of the first repeat in annexins I and 11, does not correspond to this pattern, and thus cannot bind calcium in the same manner. All the calcium-binding loops lie on the side opposite to the N- and C-termini, on the convex side of the molecule and protrude above the surface of the molecule. The binding of annexins to phospholipids is dependent on the presence of calcium. Among other divalent cations studied, zinc can increase the binding in the presence of low calcium concentrations, but does not promote phospholipid binding by itself. Magnesium and barium inhibit binding of annexin V to phospholipids [lo]. The affinity of annexin V for divalent and trivalent cations seems closely related to their ionic radius (ionic radii of Ca2+, La3+ and BaZ+ are 0.1 nm, 0.103 nm and 0.135 nm, respectively) Ill], with BaZf having the lowest (Kd = 1 mM) and La3+ the highest ( K d = 10 pM) affinity constants (H. Haigler, personal communication). We have crystallised annexin V at low ionic strength and low calcium concentration, as well as in the presence of barium. An analysis of differences between these structures and those reported earlier show significant movements within the molecule in response to crystal packing and/or calcium binding.
74 Table 1. Crystal and refinement data. Rsym= C ( I - ( I ) ) / C ( r ) for equivalent measurements. RMS, root mean square.
Ion
Parameter resolution
Rsym
nm Ca
Ba
1.20-0.24 1.20-0.24
Z > 2a(4
cornplete-
no. of reflections
ness
total
R factor
83.6 74.3
unique
bonds
nm 98 82
no. of solvent
cutoff
%
0.067 0.086
resolution amplitude RMS deviation
98492 31600 50558 23533
0.181 0.208
MATERIALS AND METHODS Recombinant human annexin V was prepared as before [12]. Crystals were grown at low ionic strength using poly(ethylene glycol) 20000 as a precipitant (50 mM Tris/maleate, pH = 6.6) with 2-3 mol Ca2+/mol protein [13]. The conditions for the barium-containing crystals were the same, except that a higher excess of the metal ion was used (up to 4 mol Ba2+/molprotein). The crystals are monoclinic, space group P21, with two molecules in the asymmetric unit and with the following cell dimensions: a = 8.39 nm, b = 8.09 nm, c = 7.14nm, j l = 108.7' (Ca crystals); a = 8.41 nm, b = 8.06 nm, c = 7.17 nm, /I = 107.8" (Ba crystals). Data were collected on a Xentronics area detector using the data-reduction programme XENGEN [14] for the Ca data and XDS [15] for the Ba data. The calcium and barium data are not isomorphous beyond 0.3 nm resolution, and we therefore decided to refine the two structures separately. The starting point for refinement of the calcium structure was obtained by molecular replacement, using the hexagonal annexin-V crystal-structure coordinates [7] as a model. The Patterson space rotation function [16] (programme package PROTEIN) gave two principal peaks with maximum correlation of 13.5 and 12.8 (the next maximum, very broad, was at 12.5), the two related by a rotation obtained from a selfrotation function search. The subsequent translation function searches gave peaks of 13.7 root mean square and 11.9 root mean square with PROTEIN. We confirmed the solution using the CCP4 programme package [ 171: the self-rotation function confirmed a non-crystallographic twofold axis (a peak at 0.8 of the maximum) lying perpendicular to the crystal h-axis and bisecting the monoclinic angle p. Refinement of the structures was carried out using XPLOR [18], followed by manual corrections, addition of metal ions and solvent molecules on the graphics display using FRODO [19]. In the final stages, the programme PROLSQ [20] was used. Table 1 summarises the measurement and refinement data for both crystal forms. The poorer refinement result for the barium structure is certainly due to poorer crystal quality. The monoclinic crystal form is indecd difficult to obtain reproducibly and in the form of single crystals, with multiple crystals often impossible to diagnose.
RESULTS AND DISCUSSION Calcium sites
In the previous crystallographic study of annexin V [8], three calcium sites and two extra lanthanum sites were described. The exchange of lanthanum for calcium went in the
0.7-0.24 0.7-0.24
24F) 248')
angles
planes
nm
nm
0.0017 2.4" 0.0017 2.4"
0.0004 0.0004
molecules
266 265
order of sites 1 > 2 > 3, where sites 1, 2 and 3 correspond to domains 11, IV and I, respectively, suggesting that the calcium is most strongly bound to domain 1. In the P21 crystal form, we find only two calcium sites, which are in domains 1 and IV (corresponding to sites 3 and 2 respectively of the previous paper) [8]. The calcium binding is therefore asymmetric with respect to the internal symmetry of the molecule, with both calcium ions binding to the same side of the central pore of the molecule. The calcium-binding loop of domain I1 has adopted a conformation different from either the low calcium (P6,) [7] or the high calcium (R3) [8] form (Fig. 2). The annexin V structure presented here, with two ions of calcium bound to domains 1 and IV, shows significant differences from either of the structures published so far. In the first crystal form (P63, low Ca) prepared in low calcium concentration, no calcium could be found. In the second one (R3, high Ca), obtained by soaking crystals at high calcium concentration, three calcium ions were bound to the protein. When the low-Ca and high-Ca structures were superimposed, the high-Ca form was shown to be more open by about 4', via a motion of the two halves of the molecule about a hinge situated between them at the level of the connecting peptides, thus increasing the channel between the two halves [8]. The structure of the intermediate calcium form described here opens even further by roughly the same amplitude (Fig. 3). We suggest effects of calcium binding and/or different packing in the different crystal forms compared to be at the origin of the conformational differences observed, which certainly document a remarkable flexibility of the molecule. Calcium affinity of annexins in solution is relatively low, becoming much higher in the presence of phospholipids (the affinity constants change from millimolar to micromolar, the exact value being different for each annexin). An early study of calcium binding by annexin V [21] indicated two conformational changes occuring at two different relative calcium concentrations, suggesting the presence of Ca sites with two different affinities. Binding studies of annexin V to phospholipids indicated the presence of at least two Ca sites, and the process was shown to be cooperative [22,23]. It would certainly be interesting to know if the calcium substitution seen in the three crystal forms studied so far corresponds to the observed behaviour in solution, as well as the implications for membrane binding. Barium and possible zinc binding
The structure of the annexin-V crystals prepared in the presence of barium shows that the barium ions have replaced calcium ions in the same two sites, in domains I and IV. The geometry of the sites is similar, but affected by the difference
75 in ionic radius of the two ion types: the carbonyl-oxygenCa distances of 0.235 - 0.245 nm increase to 0.295 - 0.305 nm for Ba. The loops containing the carbonyl-oxygen ligands move away from the barium ion (the maximum displacement is 0.185 nm at the Ca position of the central ligand), while the carboxylic-acid group is displaced only slightly compared to the calcium form (Fig. 4). The perturbation of the structure due to the presence of barium is limited to the metal ion site, the calcium and barium structures being superimposable everywhere else (root-mean-square difference at Ca positions for the two molecules in the asymmetric unit is 26 pm). This is certainly due to the special position of the metal-binding sites on the molecule surface
The occupancy of the ion sites is, however, much lower for barium (0.15 -0.45) than for calcium (1 .0), and the protein loop coordinating the metal ion is slightly less well ordered for barium than for calcium. Furthermore, the occupancy of site 3 , (domain I) is higher than that of site 2 (domain IV) in both molecules of the asymmetric unit (0.45 and 0.35 for site 3, and 0.15 and 0.3 for site 2, for the two molecules, respectively), even though both sites are equally accessible to solvent in the crystal packing. In fact, the barium ions of sites 3 and 2 from symmetrically related molecules come to within 1.2 nm of each other in the crystal lattice. In the second molecule, crystallographically distinct, the difference in occupancy is smaller but still significant. The difference in occu-
Fig. 1. Ribbon drawing [27]of the schematic of the annexin-V structure. Domains I, 11, 111 and IV are cyan, red, yellow and blue respectively. Fig. 3. Superimposition of the Ca backbones of the three crystal forms. Domains IT and 111 were aligned to give the best superimposition. The colours are the same as in Fig. 2.
Fig. 2. The conformation of the calcium site in domain I1 in the three form with no calcium is in red, the P21 form crystal forms. The with two calcium ions in yellow, the R3 form with three calcium ions in blue.
Fig. 4. Superimposition of the metal-binding loop in domain I for the calcium (pale green) and barium (red) structures.
16
/
I
,
Fig. 5. The hypothetical Zn-binding site. The dotted lines represent distances of 0.22-0.35 nm from the zinc ion, which is represented by a small pyramid.
Fig. 6. The interactions of the N- and C-termini of the two molecules within the asymmetric unit. Hydrogen bonds are represented by dotted lines.
pancies is a further confirmation of a difference in affinity for metal ions of the three binding sites within the annexin-V molecule (the affinity decreasing in the order site 3 > 2 9 1). In the case of the more bulky B a z f , for which the overallbinding affinity is lower, the relative difference in the affinity between the different sites becomes more pronounced. The decreased membrane-binding affinity of annexin V in the presence of barium [lo] may therefore most likely be related to the lower affinity of the protein for the metal ion, rather than to any perturbation of the protein conformation. The sequence of annexin V contains three histidine residues, of which two (His98 and His267) lie only about 0.3 nm below the molecular surface, on either side of the central channel, extending across it. When considering the synergistic effect of zinc at low calcium concentration upon membranebinding affinity [lo], these two histidines seem good candidates for zinc coordination. In the P21 crystal form, a zinc ion could lie at a distance of 0.23nm from the N n position of both histidines, with other charged side chains within a reasonable distance (Fig. 5). The synergistic effect of zinc ions at low calcium concentrations could be explained by its bridging of the protein molecule across the central cleft via the histidine side chains, and thus help to bring about the conformational change (movement across the channel) seen in the high-Ca form.
phenomenon has been observed for annexins VII, VI and IV in solution using light scattering and fluorescence [24]. The Ca-dependent self-association of synexin (annexin VII) was seen to start at about 2 mol Ca2+/mol protein. It became much more pronounced in the presence of membranes and, furthemore, the self-association of annexin molecules accompanied the aggregation of chromaffin granule membranes. In a comparative study of the Ca-dependent interactions of annexins I, 11, 111, V and VI with phospholipid vesicles, annexins I and I1 were found to promote their aggregation and the fusion of large unilamelar phospholipid vesicles [25]. Calpactin I (annexin 11) has been shown by freeze-etch electron microscopy to form a bridge between the surfaces of liposomes and chromaffin vesiclesjust before their fusion [26]. In the P21 crystal form the asymmetric unit contains two molecules that are related by a non-crystallographic twofold axis. These two molecules interact through their concave surfaces that contain the N- and C-termini. The two termini are in fact stabilised by the dimer interaction and have become well ordered compared to the crystal forms studied earlier. There is a network of hydrogen bonds and salt bridges connecting the N-terminus of one molecule (residues 2 - 6) with the C-terminus of the second molecule (residues 319 - 320, the annexin-V amino acid sequence is numbered starting with Ala2) and vice versa (Fig. 6). Assuming that membrane binding occurs through the calcium sites situated on the opposite sides of the molecules, it is tempting to suggest that the annexin dimer as it condensed in our crystals represents a model for a role of annexin in the aggregation of membranes: the molecules bind to the surface of separate membranes via their calcium sites, possibly in contact with each other under the effect of calcium concentration. In annexins I and 11, the length of the N-terminal sequence is much larger than in annexin V (by 27 and 18 residues, respectively), while annexin I1 interacts via its N-terminus with a small protein, p l l , to
Implications for membrane interactions
The preparation of the crystals, grown in a low-ionicstrength medium, allowed a fairly good control of the stoichiometry of protein vs calcium-ion concentration. In fact, as soon as the Ca2+ concentration rose above about 3 mol/mol protein, the protein started aggregating to form a precipitate and microcrystals, and above 5mol CaZf/mol protein the precipitation became very fast and pronounced. A related
77 form a heterodimer. The proposed arrangement could promote the association through the N- and C-termini of these annexins even more. We thank Dr. C. Cambillaud of the LCCMB, Marseille, and Dr. G. A. Bentley of the Pasteur Institute, Paris, for giving us the possibility to collect data in their laboratories.
REFERENCES 1. Crumpton, M. J. & Dedman, J. R. (1990) Nature 345, 212. 2. Glenney, J. R. (1986) J . Biol. Chem. 261, 7247-7252. 3. Rojas, E., Pollard, H. B., Haigler, H. T., Parra, C. & Burns, A. I . . (1990) J . Biol. Chem. 265,21207-21215. 4. Karshikov, A,, Berendes, R., Burger, A,, CavaliC, A,, Lux, H. D. & Huber, R. (1992) Eur. Biophys. J . 20, 337-344. 5. Klee, C. B. (1988) Biochemistry 27, 6645-6653. 6. Crompton, M. R., Moss, S. E. & Crumpton, M. J. (1988) Cell 55, 1-3. 7. Huber, R., Romisch, J. & Paques, E.-P. (1990) EMBO J. 9,3867 3874. 8. Huber, R., Schneider, M., Mayr, I., Romisch, J. & Paques, E.-P. (1990) FEBS Lett. 275, 15-21. 9. Huber, R., Berendes, R., Burger, A,, Schneider, M., Karshikov, A., Luecke, H., Romisch, J. & Paques, E.-P. (1992) J. Mol. Biol. 223, 683 - 704. 10. Andree, H. A. M., Reutelingsperger, C. P. M., Hauptman, R., Hemker, H. C., Hermens, W. T. & Willems G. (1990) J. B i d . Chem. 265,4923-4928.
11. Pauling, L. (1960) Nature of the chemical bond, 3rd edn, Cornell University Press, Ithaca, New York. 12. Maurer-Fogy, I., Reutelingsperger, C. P. M., Peiters, J., Bodo, G., Stratowa, C. & Hauptman, R. (1988) Eur. J . Biochem. 174, 585 - 592. 13, Lewit-Bentley, A., DoubliC, S., Fourme, R. & Bodo, G. (1989) J . Mol. Biol. 210, 875 - 876. 14. Howard, A. J., Gilliland, G. L., Finzel, B. C., Poulos, T. L., Ohlendorf, D. H. & Salemme, F. R. (1987) J . Appl. Crystallogr. 20, 383 - 387. 15. Kabsch, W. (1988) J . Appl. Crystallogr. 21,916-924. 16. Huber, R. (1965) Acta Crystallogr. 19, 353. 17. CCP4 (1979). The SERC ( U K ) collaborative computing project no. 4 , Daresbury Laboratory, Warrington. 18. Brunger, A. T., Kuriyan, M. & Karplus, M. (1987) Science 235, 458. 19. Jones, T. A. (1978) J. Appl. Crystallogr. 11, 268-272. 20. Konnert, J. H. & Hendrickson, W. A. (1980) Acta Crystallogr. A36, 344 - 350. 21. Shadle, P. J., Gerke, V. & Weber, K. (1985) J . Biol. Chem. 260, 16354-16360. 22. Glenney, J. R. Jr, Tack, B. & Powell, M. A. (1987) J . CellBiol. 104, 503-511. 23. Schlaepfer, D. D., Mehlman, T., Burgess, W. H. & Haigler, H. T. (1987) Proc. Natl Acad. Sci. USA 84, 6078 -6082. 24. Zaks, W. J. & Creutz, C. E. (1991) Biochemistry 30,9607-9615. 25. Blackwood, R. A. & Ernst, J. D. (1990) Biochem. J . 266, 195200. 26. Nakata, T., Sobue, K. & Hirokawa, N. (1990) J . Cell Biol. 110, 13-25. 27. Priestley, J. P. (1988) J . Appl. Crystallogr. 21, 572-576.
