Science in China Ser. C Life Sciences 2004 Vol.47 No.4 359—367
359
Homology modeling three-dimensional structure of AnxB1 and reducing its immunogenicity by sequence-deleted mutagenesis YAN Hongli1, SONG Yunlong2, LIU Fan1, HE Yan1 & SUN Shuhan1 1. Department of Medical Genetics, Second Military Medical University, Shanghai 200433, China; 2. Department of Medicinal Chemistry, Second Military Medical University, Shanghai 200433, China Correspondence should be addressed to Sun Shuhan (email:
[email protected])
Received April 28, 2003
Abstract AnxB1, a novel annexin previously isolated from Cysticercus cellulose, shows high thrombi affinity and anticoagulant activity in vivo. In order to investigate the relationship between structure and biological function, a predicted three-dimensional (3D) model of AnxB1 was generated by homology modeling. This model contains four homologous internal-domains and the Cα trace of domain I, II and IV shows high similarity. Based on the structure characterization, four sequence-deleted mutants were constructed and expressed as GST fusion proteins in E. coli. Two of the mutants, GST-M3 and GST-M4 reserved high anticoagulant activity (p<0.01 vs. GST). Furthermore, compared with the wild type GST-AnxB1, the immunogenicity of GST-M3 and GST-M4 was reduced significantly (p<0.01) and the molecular weight was lowered to 27 kD and 34 kD, respectively. These observations laid a solid foundation for further study on developing new thrombolytic agents with higher efficiency and lower side effect. Keywords: annexin, homology modeling, sequence-deleted mutant, anticoagulant, immunogenicity. DOI: 10.1360/03yc0085
We previously isolated and functionally identified a new annexin from Cysticercus cellulosae, which belongs to a novel annexin subfamily and is designated as AnxB1[1]. In the presence of Ca2+, AnxB1 can bind to negatively charged phospholipids like phosphatidylserine (PS) with high affinity, whereas hardly associates with phosphatidylcholine (PC) even at high Ca2+ concentration. PS is the major component of the platelet membrane and plays a key role in the process of coagulation cascade. Under viable and unperturbed conditions, PS localizes predominantly in the membrane leaflets facing the cytosol. However, when platelets are activated by agonists like thrombin and collagen, PS becomes surface-exposed and thereby
provides a catalytic surface for the procoagulant reactions, such as the activation of coagulant factor Xa. AnxB1 possesses the Ca2+-dependent PS binding property; therefore, it can inhibit the activation of coagulant factor Xa by shielding PS on the surface of the activated platelets and subsequently inhibit the thrombin generation[2]. In addition, AnxB1 can also act as a potential biopharmaceutical agent for detection of thrombi. We have previously constructed a chimeric protein AnxB1ScuPA, in which AnxB1 is attached to a plasminogen activator (low molecular weight prourokinase), the thrombolytic activity was about 3-fold that of urokinase[3]. However, two points should be noted: firstly, as a non-human origin protein, Copyright by Science in China Press 2004
360
Science in China Ser. C Life Sciences
AnxB1 has a strong immunogenicity; secondly, although AnxB1 is a relatively small protein (Mr. 38 kD), when attached to the plasminogen activator, the fusion protein was larger than 70 kD, protein of this size is difficult to express and purify.
galactopyranoside (IPTG), Sarcosyl, deoxycholate were purchased from Sigma, and the standard reagents used in KPTT test were provided by Changhai Hospital. All other chemicals used in this study were of analytical grade or better.
Homology modeling is one of the most effective methods to predict unknown protein’s three-dimensional structure by using the solved crystal structures as templates. The predicted model is quite reliable if the amino acid sequence similarity between the target protein and the template is over 30%[4]. Annexins are structure-conserved proteins widely distributed in eukaryotes and amino acid sequence identities between different members are usually over 30%. Nervertheless, up to date, more than 30 crystal structures of different annexins have been established and deposited to Brookhaven Protein Data Bank (PDB). These foundations lead to feasibility to build a reliable model for AnxB1. In this study, a predicted 3D model for AnxB1 was firstly built by homology modeling. In order to obtain some proteins with lower molecular weight and immunogenicity, four sequence-deleted mutants were constructed based on the structure characterizations and expressed as GST fusion proteins in E. coli. Two of the mutants, GST-M3 and GST-M4, showed similar anticoagulant activity to the wild type GST-AnxB1. Furthermore, the molecular weights of the two mutants are reduced to 27.3 kD and 34 kD, and their immunogenicity also greatly decreased.
1.2
1
Materials and methods
1.1 Materials The recombinant plasmid pUC18-AnxB1, which contains the full cDNA sequence of AnxB1, was previously constructed and preserved in our laboratory. E. coli strain k802 and plasmid pGEX-5T[5] was a kindly gift from Dr. Wu S. M. The following reagents were obtained from commercial sources: restriction enzymes, T4 DNA ligase, pfu DNA polymerse were purchased from TaKaRa or Huamei Corp.; middle molecular weight protein marker was from Huashun Corp.; glutathione-sepharose 4B was from Pharmacia; Dithiothreitol (DTT), phenylmethylsulfonylfuoride (PMSF) were from AMRESCO; Isopropyl-β-D-thio-
Homology modeling
Molecular modeling was carried out on an SGI workstation using Insight II program (98.0 Biosym/MSI, San Diego, CA). To summarize, the process consists of the following steps: (i) Template selection. By searching the Brookhaven Protein Data Bank (PDB) with the FASTA program, four crystal structures were selected as templates by overall sequence identity: human AnxV (PDB: 1ANW), human AnxIV (PDB: 1ANN), human AnxI (PDB: 1AIN), Hydra vulgaris AnxXII (PDB: 1AEI). (ii) Determination of the structurally conserved regions (SCRs). By aligning the amino acid sequence of AnxB1 with those of templates, the structurally conserved regions and an average Cα framework structure of the templates were determined. (iii) Rough model building. The backbone of each SCR in AnxB1 was then built by fitting the corresponding SCRs from one of the known homologs to the appropriate region in the framework. Loops and the structurally variable regions (SVRs) are difficult to build because these regions are highly diverse in different templates. These regions were constructed by a knowledge-based approach. For each SVR, appropriate fragments with the same length were retrieved from rotamer libraries. Side chains of the model were added both taking in to account the backbone secondary structure and the side chain conformation at the corresponding residues of the templates. (iv) Model refinement. After adding all the hydrogen atoms, the rough model was refined using energy minimization and molecular dynamics (MD) so that the steric strain introduced during the model-building process can be relieved. All the process was performed using Insight II/Discover module and calculated under CFF force field. Firstly, the model was subjected to a rough energy minimization: first optimized by 2000 cycles on constrained backbone using steepest descent method, then using conjugate gradient method to constrain backbone until the final root mean square (RMS) gra-
Homology modeling three-dimensional structure of AnxB1 and reducing its immunogenicity by sequence-deleted mutagenesis
dient was less than 0.5 kcal/(mol·Å). Subsequently, the model was subjected to constrained molecular dynamics simulations. The system was simulated for 500 ps at room temperature (300 K). On every 10 ps, one conformation was collected and energy minimization was started according to the following procedure: first 300 cycles by steepest descent method and then 200 cycles by conjugate gradient method to constrain backbone. The conformation having the lowest energy was chosen for further evaluation. The refined structure was examined using the Profile-3D program to evaluate the model’s internal consistency and reliability. 1.4
Construction of GST fusion mutants
Four sequence-deleted mutants were constructed as GST fusion proteins. Five oligonucleotide primers were used. 1: 5′-gGAATTCcGGGTTCTCACCGACCGC-3′; 2: 5′-gGAATTCcCTATCCAATCACCGTGG3′; 3: 5′-gGTCGACttaACCGGCGAGCGCTAGC-3′; 4: 5′-gGTCGACttaGGCGTAGCGAACTCTAG-3′; 5: 5′-gGTCGACttaTGCAGGGCCGATGAG-3′. 1, 2 were forward primers, with an EcoR I restriction site (underlined) at 5′ ends; 3, 4, 5 were reverse primers, introducing a stop codon TAA (boxed) and a Sal I restriction site (underlined) at 5′ ends. DNA fragments of AnxB1 encoding amino acids 97—173 (M1), 97— 272 (M2), 97—345 (M3) and 30—345 (M4) were PCR amplified using primers 2—3, 2—4, 2—5 and 1—5, respectively. The PCR products were digested with EcoR I/Sal I and then cloned into an EcoR I/Sal I digested pGEX-5T vector, yielding four expression plasmids: pGEX5T-M1, pGEX5T-M2, pGEX5T-M3 and pGEX5T-M4. The plasmids were verified by DNA sequencing and transformed into E. coli strain k802. Protocol of transformation, digestion was described previously[6]. 1.5 Expression and purification of GST fusion mutants The recombinant E. coli strains containing the above expression plasmids were grown in 1 liter of 2 YT broth containing 100 μg/mL ampicillin at 37℃. 2—3 h later, the culture reached an optical density of
361
A600 = 0.6. Expression of GST fusion mutants was induced by 0.5 mmol/L IPTG at 30℃ for 6 h (the lower temperature was used to increase the solubility of the fusion protein). Recombinants of GST-M1, GST-M2, GST-M3 and GST-M4 were purified as described in ref. [7] with a slight modification: Bacterial cultures were centrifuged at 6000 g for 10 min at 4℃ and the pellet resuspended in 50 mmol/L phosphate buffered saline (PBS), pH 7.3. Cells were sonicated on ice, followed by a centrifugation at 12000 g for 10 min. The supernatant was mixed with Glutathione-Sepharose 4B and incubated at room temperature for 30 min. After a 500 g centrifugation for 5 min, the pellet was extensively washed with PBS. The bound GST fusion protein was harvested by adding elution buffer (50 mmol/L TrisHCl, 10 mmol/L reduced glutathione, pH 8.0). The eluted fractions were analyzed by 10% SDS-PAGE and stored at −20℃. As a control, the 26-kD glutathione S-transferase (GST) of Schistosoma japonicum was expressed from the pGEX-5T vector (without a fusion insert) in the bacterial strain k802 and purified according to the same procedure described previously. 1.6
Kaolin partial thromboplastin time (KPTT) test
Anticoagulant activity of GST-M3 and GST-M4 were assayed using kaolin partial thromboplastin time (KPTT) test. All the samples for test were diluted with PBS to the final volume of 100 μL and incubated with 100 μL kaolin active thrombofax at 37℃ for 5 min. Then 100 μL standard human serum was added and incubated for 3 min. Finally, 100 μL 25 mmol/L CaCl2 was added and the fibrin formation time was detected. 1.7
Measurement of IgG titers
4—6 weeks KM male mice were purchased from the laboratory animal center of our university. 15 mice were randomly divided into 3 groups (n =5 each): GST-AnxB1 control, GST-M3, GST-M4. All the solutions for immunization were diluted with phosphatebuffered saline (PBS) so that the final volume was 1 mL/kg. Groups of five mice were immunized (i.p.)
362
Science in China Ser. C Life Sciences
with purified GST-AnxB1, GST-M3, GST-M4 at a dose of 50 μg per mouse. Complete Freund’s adjuvant (CFA) was used as adjuvant. 14 d later, the mice were boosted with a same dose. Blood samples were collected from each mouse every 5 d after the last injection. Sera were obtained by centrifugation and stored at −20℃ for later assay. Specific IgG titers were determined by enzyme-linked immunosorbent assay (ELISA). 1.8 SDS-polyacrylamide gel electrophoresis and determination of protein concentration 10% SDS-polyacrylamide gel electrophoresis was performed according to Lammeli[8]. Protein concentrations was determined by the method of Bradford[9] using bovine serum albumin as a standard. 1.9
Statistical analysis Data were expressed as the mean ± standard de-
viation ( X ±SD). Statistical analysis was done using Student’s t-test. P < 0.05 was taken to indicate a significant difference. 2
Results
2.1 Homology modeling of AnxB1 and structure analysis Four homologous sequences were retrieved by searching the protein data base: AnxV (1ANW), AnxIV (1ANN), AnxI (1AIN), AnxXII (1AEI). Those proteins have 45.2%, 41.4%, 39.3% and 36.3% amino acid identities to AnxB1, respectively. Pair-wise sequence alignments revealed that AnxV (1ANW) is the highest homologue to AnxB1. Therefore, the crystal structure of AnxV was selected as the major template. Fig. 1(a) shows the final model of AnxB1. The model was evaluated using three independent methods. Firstly, eighteen environmental parameters of the model were calculated using profile 3D program. The overall compatibility score is 141.85, nearing the expected compatibility score of 158.08; secondly, each residue also has a validity score, S > 0 means the residue is allowable. As shown in fig. 2(b), most of the residues have a score > 0, with an exception of several residues in domain II and III. Thirdly, fig. 2(a) shows
the Ramachandran plot of the φ /ψ distribution, with 87% residues in most favored regions, 9.5% in allowed or generously allowed regions, only 3.5% in disallowed regions. All the above data indicated that the quality of the model is quite reasonable. The predicted model of AnxB1 has a similar protein topology to the crystal structure of AnxV: The polypeptide chain was folded into four homologous internal-domains and arranged in a cyclic array, giving the molecule an overall flat, slightly curved shape. Each domain consists of five α-helices, the connector between domain I and II is short and extended, whereas the connector between domain III and IV is a little longer and internal-facing folded. The two loops are tightly contact and form a calcium-dependent phospholipid binding site[10]. Two distinct differences have been observed. First, the N-terminus of AnxB1 is composed of 27 amino acid residues, much longer than that of AnxV (12 aa). Second, the connector between domain II and III in AnxB1 is more extended than the corresponding region in AnxV, which is consistent with the result of pair-wise sequence alignment. 2.2
Domain I, II and IV have a similar structure
Fig. 3 shows the superimposition of the Cα trace between four domains in AnxB1. Intriguingly, root mean square (RMS) deviation among the residues in domain I, II and III is lower than 1.2 Å. Furthermore, RMS deviation between the key residues involved in calcium binding sites, Gly43, Gly45, Glu85 residues in domain I, Gly115, Gly117, Asn157 residues in domain II and Gly289, Gly291, Asn331in domain IV, is lower than 0.5 Å, which revealed that the structure of the three domains is in high similarity (fig. 3(a)). However, domain III is not the case. The RMS scores between domain III and other domains are nearly 3 Å (fig. 3(b), table 1). 2.3 Construction, expression and purification of GST fusion mutants Paul et al.[11] have proposed that the internal homology domains in Annexin family were generated by gene duplication. In this study, superimposition of the Cα traces of domain I, II and IV indicated that the
Homology modeling three-dimensional structure of AnxB1 and reducing its immunogenicity by sequence-deleted mutagenesis
363
Fig. 1. Comparison of the structure model of AnxB1 and its major template. (a) Structure model of AnxB1 generated by homology modeling. (b) Crystal structure of AnxV (1ANW). N terminal region is shown in purple. Domain I, II, III and IV are labeled in aqua, yellow, green and red colors, respectively. The blue region in (a) indicates the long connector between domain II and III.
Fig. 2. Theoretical validation of the model. (a) Ramachandran plot. (b) Validation by Profile-3D. S>0 means the residue is allowable.
364
Science in China Ser. C Life Sciences
Fig. 3. Superimposition of the Cα trace between the internal domains. (a) Domain I, II and IV. The purple ball indicates calcium atom, the labeled residues are the key residues involved in calcium binding site. (b) Domain I, II, III and IV. Cα traces of domain I, II and IV are shown in green lines. The red line indicates domain III. Table 1 Root mean square (RMS) deviation of the Cα trace between domain I, II, III and IV Domain I II III IV
I 0.0 0.91565 3.31369 1.16842
II 0.91565 0.0 3.3588 1.15225
III 3.31369 3.3588 0.0 3.34861
IV 1.16842 1.15225 3.34861 0.0
three domains have a similar topology and each domain contains a typical Ca2+-dependent phospholipid binding site. Based on these findings, we proposed that if constructing some truncated mutants only reserving one or two homologous domains, theoretically, they might reserve the anticoagulant property for possessing entire calcium binding sites. Nevertheless, reduced molecular weights will lower their immuno-
genicity. To verify our hypothesis, four sequence-deleted mutants were constructed (fig. 4). The amino acid boundaries of each mutant were defined as consistent with the structure model of AnxB1. Domains reserved in each mutant were: M1, domain II; M2, domain II—III; M3, domain II—IV; M4, domain I—IV. DNA fragments encoding these mutants were PCR amplified and cloned in pGEX-5T expression vector. The recombinant expression vectors were transformed into E. coli K802. Expression of the GST fusion proteins was induced by 0.5 mmol/L IPTG under tac promoter. SDS-PAGE analysis revealed that four predominant bands appeared after induction: a 37 kD band in E. coli containing pGEX-5T-M1, a 45 kD
Fig. 4. Schematic diagram of the amino acid boundaries for the sequence-deleted mutants. internal domain.
: N terminal region.
: GST moiety.
: Homology
Homology modeling three-dimensional structure of AnxB1 and reducing its immunogenicity by sequence-deleted mutagenesis
band in E. coli containing pGEX-5T-M2, a 53 kD band in E. coli containing pGEX-5T-M3 and a 60 kD E. coli containing pGEX-5T-M4 (fig. 5(a)). Each band amounted to more than 40% of the total bacterial proteins. Given the GST moiety is 26 kD, molecular weight of the mutant part is 11 kD, 19 kD, 27 kD, 34 kD, respectively, which is consistent with their theoretical molecular weights: M1 (11 kD), M2 (19.4 kD), M3 (27.3 kD) and M4 (34 kD). The products were purified by affinity chromatography and finally presented as a single band (fig. 5(b)). 2.4
KPTT test and immunogenicity analysis
Anticoagulant activity of GST fusion mutants was assayed by kaolin partial thromboplastin time test (fig. 6(a)). As compared with GST and PBS control, GST-AnxB1, GST-M3, GST-M4 prolonged KPTT time significantly (P<0.01), whereas GST-M1, GST-M2 showed no statistically significant difference with control. Three groups of mice were immunized with some dose of GST-M3, GST-M4 and GST-AnxB1. IgG levels in the pooled sera from five mice of the same group were determined by ELISA test. The results (as showed in fig. 6(b)) that 20 d after the final immuniza-
365
tion, IgG level in the group of GST-M3, GST-M4 is only half that of the GST-AnxB1 group, meaning the immunogenicity of GST-M3, GST-M4 reduced significantly. 3
Discussion
One of the tasks of the post-genomic era is to rationalize and exploit fruitfully the enormous amount of information generated by sequencing thousands of genes. However, although the sequences are available, it is difficult to provide detailed functional role of the proteins. In order to make a deep investigation on there functions, three-dimensional structures are usually required to be determined. But solving a highresolution structure has been proved to be difficult because of some technological limitations. Therefore, knowledge-based protein structure prediction is an effective substitute method to solve this problem. We choose homology-modeling method to build the 3D model of AnxB1 based on the following facts: First, previous studies have proved that structure of annexins is mainly composed of α helixes and is conserved in all members. We have also superimposed the Ca framework of the four templates before building the model of AnxB1, and found the SCRs in templates
Fig. 5. Expression and purification of the GST fusion proteins. 10% SDS-PAGE analysis of the bacterial protein is shown. (a) Total bacterial proteins. Lane 1, Total SDS lysate from uninduced E. coli containing the pGEX-2T vector; lane 2, total SDS lysate of the E. coli containing the pGEX-5T-AnxB1 after induction by IPTG for 5 h; lanes 3—7, total SDS lysate of the E. coli containing the pGEX-5T-M4, pGEX-5T-M1, pGEX-5T-M2, pGEX-5T-M3, pGEX-5T, respectively, after induction by IPTG for 5 h. (b) Purified GST fusion protein after elution from affinity medium. Lanes 1—6, GST-AnxB1, GST-M4, GST-M1, GST-M2, GST-M3, GST, respectively; M indicates Mr markers.
366
Fig. 6. Biochemical activity analysis of GST fusion mutants. (a) KPTT test. Data are expressed as means ± SD of three individual tests. *, P<0.05 compared to GST control; **, P < 0.01 compared to GST control. (b) IgG titers in mice immunized with GST-AnxB1 and its mutants. Data are expressed as means ± SD of five mice. *, P < 0.05 compared to GST-AnxB1; **, P < 0.05 compared to GST-AnxB1.
are highly conserved. Intriguingly, the calcium-dependent phospholipid binding sites all locate at the SCRs. Second, since the first crystal structure of human annexinV was determined by Huber in 1990, more than 31crystal structures of different annexins have been solved. These findings make us easily find the proper templates. Analysis of the predicted model of AnxB1 revealed that the structures of domain I, II and IV are highly consistent and each contains a calcium-dependent phospholipid binding site. These observations have provided us some theoretic fundamentals for further constructing sequence-deleted mutants. However, although all the four mutants possess at least one calcium-dependent phospholipid binding site, only
Science in China Ser. C Life Sciences
GST-M3 and GST-M4 kept the anticoagulant activity. These results indicated that GST-M1 and GST-M2 might be incorrectly folded, or other residues in domain II and IV might be involved in the anticoagulant process, which needs further investigation. As compared with GST-AnxB1, the immunogenicity of GSTM3 and GST-M4 was significantly reduced. Two reasons might explain these results: first, it is generally believed that immunogenicity of a protein is partly dependent on its molecular weight. M3 and M4 are truncated AnxB1 mutants and their molecular weights have been reduced to 27.3 and 34 kD. Therefore, the lower molecular weight might contribute to their immunogenicity reduction. The second reason might be more important: Annexins are a structurally conserved gene family. All annexins can be divided into two modules: C-terminal “core” and N-terminal “tail”. The core region contains four internal-homology domains and has a similar topology in all annexins. However, the tail is unique and only has a low sequence similarity between different subfamilies. Therefore, although the core region is much larger than the tail, immunogenicity of AnxB1 mainly depends on the short tail. Lack of N-terminal tail in M3 and M4 might result in a greatly reduced immunogenicity. Based on its ability to displace phospholipidsdependent coagulant factors, AnxB1 has been found to be a potent anticoagulant. Intriguingly, unlike anticoagulants such as heparin, AnxB1 does not directly act on thrombin and thereby leads to less side effects such as bleeding. In addition, the ability to bind activated platelets with high affinity revealed that AnxB1 can be used as a promising vector to develop some thrombitargeted thromblytic agents. In this work two mutants reserving the anticoagulant activity of AnxB1 were obtained by sequence-deleted mutagenesis. Compared with wild type AnxB1, both of them have a lower molecular weight and reduced immunogenicity, which laid a solid foundation for further study on developing thrombolytic agents with higher efficiency and less side effects. Acknowledgements This work was supported by the National High Technology Research and Development Program of China (863 Program) (Grant No. 101—060504)and the National Natural Science Foundation of China (Grant No. 30271167).
Homology modeling three-dimensional structure of AnxB1 and reducing its immunogenicity by sequence-deleted mutagenesis
References 1. Yan Hongli, Sun Shuhan, Chen Ruiwen, Cloning and functional identification of a novel annexin subfamily in Cysticercus cellulosae, Molecular & Biochemical Parasitology, 2002, 119: 1—5. 2. Zhang Yi, Chen Ruiwen, Sun Shuhan, The influence of Annexin32, a new Ca2+-dependent phospholipids-binding protein, on coagulation time and thrombosis, Science in China, Ser. C, 2002, 45(2): 186—190. 3. Yan Hongli, Sun Shuhan, Chen Ruiwen et al., Construction and expression of a low relative molecular mass single-chain urokinase-type plasminogen activator (ScuPA32k) and annexin32, Acad. J. Sec. Mil. Med. Univ., 2002, 23(4): 378—380. 4. Orengo, C. A., Jones, D. T., Thornton, J. M., Protein superfamilies and domain superfolds, Nature, 1994, 372: 631. 5. Berthold, H., Frorath, B., Scanarini, M. et al., Plasmid pGEX-5T: An alternative system for expression and purification of recombinant proteins, Biotech. Lett., 1992, 14(4): 245—250.
367
6. Sambrook, J., Fretsch, E. F., Maniatis, T., Molecular Cloning: A Laboratory Manual, 3rd ed., New York: Cold Spring Harbor Laboratory Press, 2001, 66—117. 7. Marshak, D. R., Kadonaga, J. T., Burgess, R. R. et al., Strategies for Protein Purification and Characterization: A Laboratory Course Manual Cold Spring Harbor Laboratory Press, 1996, 6—38. 8. Laemmli, U. K., Cleavage of structural proteins during the assembly of the head of bacteriophage T4, Nature, 1970, 227: 680—685. 9. Bradford, M. M., A rapid and sensitive method for the quantization of microgram quantities of protein utilizing the principle of protein-dye binding, Anal. Biochem., 1976, 112: 195—203. 10. Swairjo, M. A., Concha, N. O., Kaetzel, M. A. et al., Ca2+-bridging mechanism and phospholipid head group recognition in the membrane-binding protein annexin V, Nature Structural Biology, 1995, 2: 968—974. 11. Smith, P. D., Moss, S. E., Structural evolution of the annexin supergene family, Trend in Genetics, 1994, 10(7): 241—245.