Variability among the sites by which curaremimetic toxins bind to torpedo acetylcholine receptor, as revealed by identification of the functional residues of alpha-cobratoxin.

alpha-Cobratoxin, a long chain curaremimetic toxin from Naja kaouthia venom, was produced recombinantly (ralpha-Cbtx) from Escherichia coli. It was indistinguishable from the snake toxin. Mutations at 8 of the 29 explored toxin positions resulted in affinity decreases for Torpedo receptor with DeltaDeltaG higher than 1.1 kcal/mol. These are R33E > K49E > D27R > K23E > F29A >/= W25A > R36A >/= F65A. These positions cover a homogeneous surface of approximately 880 A(2) and mostly belong to the second toxin loop, except Lys-49 and Phe-65 which are, respectively, on the third loop and C-terminal tail. The mutations K23E and K49E, and perhaps R33E, induced discriminative interactions at the two toxin-binding sites. When compared with the short toxin erabutoxin a (Ea), a number of structurally equivalent residues are commonly implicated in binding to muscular-type nicotinic acetylcholine receptor. These are Lys-23/Lys-27, Asp-27/Asp-31, Arg-33/Arg-33, Lys-49/Lys-47, and to a lesser and variable extent Trp-25/Trp-29 and Phe-29/Phe-32. In addition, however, the short and long toxins display three major differences. First, Asp-38 is important in Ea in contrast to the homologous Glu-38 in alpha-Cbtx. Second, all of the first loop is insensitive to mutation in alpha-Cbtx, whereas its tip is functionally critical in Ea. Third, the C-terminal tail may be specifically critical in alpha-Cbtx. Therefore, the functional sites of long and short curaremimetic toxins are not identical, but they share common features and marked differences that might reflect an evolutionary pressure associated with a great diversity of prey receptors.

␣-Cobratoxin, a long chain curaremimetic toxin from Naja kaouthia venom, was produced recombinantly (r␣-Cbtx) from Escherichia coli. It was indistinguishable from the snake toxin. Mutations at 8 of the 29 explored toxin positions resulted in affinity decreases for Torpedo receptor with ⌬⌬G higher than 1.1 kcal/mol. These are R33E > K49E > D27R > K23E > F29A > W25A > R36A > F65A. These positions cover a homogeneous surface of approximately 880 Å 2 and mostly belong to the second toxin loop, except Lys-49 and Phe-65 which are, respectively, on the third loop and C-terminal tail. The mutations K23E and K49E, and perhaps R33E, induced discriminative interactions at the two toxin-binding sites. When compared with the short toxin erabutoxin a (Ea), a number of structurally equivalent residues are commonly implicated in binding to muscular-type nicotinic acetylcholine receptor. These are Lys-23/Lys-27, Asp-27/Asp-31, Arg-33/Arg-33, Lys-49/Lys-47, and to a lesser and variable extent Trp-25/Trp-29 and Phe-29/ Phe-32. In addition, however, the short and long toxins display three major differences. First, Asp-38 is important in Ea in contrast to the homologous Glu-38 in ␣-Cbtx. Second, all of the first loop is insensitive to mutation in ␣-Cbtx, whereas its tip is functionally critical in Ea. Third, the C-terminal tail may be specifically critical in ␣-Cbtx. Therefore, the functional sites of long and short curaremimetic toxins are not identical, but they share common features and marked differences that might reflect an evolutionary pressure associated with a great diversity of prey receptors.
In the course of the past 2 decades, a large number of curaremimetic toxins have been isolated from venoms of elapid and hydrophiid snakes (1). They are small proteins that all adopt a leaf-like shape, with three adjacent loops rich in ␤-pleated sheet which emerge from a small globular core where four disulfide bonds are invariably located (2)(3)(4)(5)(6). Despite this overall structural homogeneity, curaremimetic toxins can be divided into at least two major subfamilies on the basis of both their amino acid sequences (1) and functional properties (7). One subfamily involves the long chain toxins with 66 -74 residues and 5 disulfide bonds, which bind with high affinities to both muscular-type and ␣7 neuronal receptors (1,7). The other subfamily involves short chain toxins with 60 -62 residues and 4 disulfide bonds, which bind with high affinity to musculartype nicotinic acetylcholine receptors only (AChRs). 1 Several specific structural deviations differentiate toxins from both categories. Thus, the long chain toxins have a longer C-terminal tail, an additional small loop cyclized by a fifth disulfide bond located at the tip of the central loop and a shorter first loop. One of these deviations, i.e. the fifth disulfide bond, has been associated with the unique ability of long chain toxins to bind to ␣7 receptors with high affinities (7). In addition, curaremimetic toxins isolated from the Laticauda colubrina venom do not fall into either of these two major subfamilies. Such toxins have the size of long chain toxins (69 residues) but lack the fifth disulfide bond and bind with high affinity to musculartype AChRs only (7). Therefore, in contrast to what has been believed during the past 2 decades, curaremimetic toxins do not form a homogeneous family of proteins. This conclusion raises, therefore, the possibility that short and long chain toxins may not achieve their comparable binding to muscular-type AChR by identical means. To approach this question, the sites where both types of toxins bind to the same type of AChR need to be elucidated.
Previously, mutational studies have been carried out with a view to understanding how short chain toxins recognize muscular-type AChRs. Initially, 10 residues were found to be important for erabutoxin a (Ea), a toxin from the sea snake Laticauda semifasciata, to bind to AChR from Torpedo (8,9). Among these, Lys-27, Trp-29, Asp-31, Arg-33, Glu-38, and Lys-47 are highly conserved in both short and long chain curaremimetic toxins, whereas Ser-8, Gln-7, Gln-10, and Ile-36 are more variable. In these studies, no attempt was made to identify the contribution of these functional residues to either one of the two toxin-binding sites. More recently, other studies also showed that the invariant residues Lys-27 and Arg-33, and to a lesser extent Lys-47, are important for a short chain toxin from Naja mossambica mossambica to bind to mouse muscle AChR and contribute differently to the two toxin-binding sites (10,11). Therefore, the greater understanding of the functional sites of short chain toxins is emerging. In sharp contrast, and despite a number of preliminary studies based on numerous chemical modifications (12)(13)(14)(15)(16)(17) and one genetic approach (18), the functional sites of long chain toxins remain unclear.
The aim of this paper was to identify residues that are important for ␣-cobratoxin (␣-Cbtx), a long chain toxin from Naja kaouthia, in its binding to muscular-type AChR. To approach this question, we (i) constructed a cDNA encoding ␣-Cbtx; (ii) expressed it as a fusion protein in Escherichia coli; (iii) showed that the unfused recombinant toxin (r␣-Cbtx) is indistinguishable from the venom toxin; (iv) performed 35 mutations at 29 different positions covering the whole loop I of ␣-Cbtx, part of loops II and III, and its C-terminal tail. The data demonstrate that ␣-Cbtx and the short chain Ea share both similarities and marked differences in binding to the same muscular-type receptor.

EXPERIMENTAL PROCEDURES
Construction of the Synthetic Gene for ␣-Cobratoxin-A synthetic gene for ␣-Cbtx was designed by back-translating the primary amino acid sequence of native ␣-Cbtx and then optimizing the codon usage for expression in E. coli. To obtain optimal expression of the recombinant ␣-Cbtx, we used a ZZ fusion protein strategy with the pCP vector, a pET3a derivative plasmid previously described (19). Expression of ZZ fusion proteins is inducible in the E. coli host strain BL21(DE3) which contains the structural gene for the T7 bacteriophage RNA polymerase under lac repressor control.
To create the desired synthetic gene, 10 oligomers ranging in length from 26 to 59 nucleotides were synthesized and purified by polyacrylamide/urea gel electrophoresis followed by elution in 0.5 M ammonium acetate. Equimolar mixtures of sense and complementary oligonucleotides were annealed at 95°C for 5 min and cooled to room temperature overnight, and the five double-stranded oligonucleotides were then ligated by T4 DNA ligase (Appligene Oncor) in three steps. A small quantity of the complete synthetic gene (213 base pairs) flanked by KpnI and BamHI restriction sites and containing the following sequence (sense strand) was obtained: 5Ј-ATCCGTTGCTTCATTACTCCTG-ATATCACCTCCAAGGACTGTCCGAACGGTCACGTTTGCTACACCA-AGACTTGGTGCGATGCTTTCTGCTCCATCCGTGGAAAACGTGTTG-ACCTGGGATGTGCTGCTACCTGCCCGACCGTTAAAACCGGTGTT-GACATCCAGTGTTGCTCCACTGATAACTGTAACCCTTTCCCTACT-CGTAAGCGCCCT-3Ј.
The final construct was amplified using standard polymerase chain reaction procedures with two oligonucleotides A (5Ј-GGCGCCGGCCG-CGGTACCCATGATCCGTTGCTTCATTACT-3Ј) and B (5Ј-GGCGCGG-GCCGCGGATCCTTATTAAGGGCGCTTACGAGTAGGG-3Ј). The polymerase chain reaction product was isolated by low melting agarose electrophoresis and then digested with KpnI and BamHI (Roche Molecular Biochemicals). The doubly digested fragment was ligated into a KpnI/BamHI-digested pCP plasmid at 16°C for 3 h. DH5␣-competent cells were transformed with the ligation mixture, and single colonies were screened for insert by polymerase chain reaction using two oligonucleotides specific for pCP. Five clones were sequenced by doublestranded DNA sequencing (Amersham Pharmacia Biotech T7 Sequencing TM Kit). Isolated plasmid (Promega, Wizard Plus Minipreps Kit) was used to transform the BL21(DE3)-inducible strain.
Probes for Site-directed Mutagenesis-The nucleotide sequences of the probes used for mutagenesis are available upon request. They were synthesized by Eurobio and purified by polyacrylamide/urea gel electrophoresis followed by elution in 0.5 M ammonium acetate. Site-directed mutagenesis was performed using the QuikChange TM Site-directed Mutagenesis Kit from Stratagene. The mutated genes were determined using the ABI.PRISM TM A310 Genetic Analyses of Perkin-Elmer applied biosystems.
Fusion Protein Expression-The bacterial host used for cytoplasmic expression of ZZ-Cbtx hybrid wild-type and mutants was E. coli BL21(DE3). Bacteria were grown at 37°C to late log phase (A 600 ϭ 0.6 to 0.8) in Luria broth (Difco) supplemented with 200 g/ml ampicillin and 30 g/ml chloramphenicol. Bacteria were then induced with 1 mM isopropyl-1-␤-D-thio-1-galactopyranoside (Eurobio) for 3 h. Cells were centrifuged and resuspended in lysis buffer (30 mM Tris, pH 8, 5 mM EDTA, 20% sucrose, 0.1 mg/ml lysozyme (Sigma)) and 0.5 mM phenylmethylsulfonyl fluoride (Sigma). After three freezing-unfreezing steps, 2 mg/ml protamine sulfate (Sigma) was added, and the solution was centrifuged at 12,000 rpm for 10 min. The supernatant was then applied to an IgG-Sepharose-6FF column (Amersham Pharmacia Biotech), and the hybrid was eluted with 0.5 M acetic acid, pH 3.4, and lyophilized. Protein concentration was estimated by UV absorption at 280 nm.
Cleavage by CNBr-The lyophilisate was resuspended with 0.1 N HCl in the presence of a 100-fold excess of CNBr and incubated at room temperature for 6 h. The solution was then diluted 10-fold with water and lyophilized to stop the reaction.
Protein Refolding and Purification-Treated sample at the concentration of 0.1 mg/ml was refolded in 100 mM phosphate buffer, pH 7.8, containing 4 mM GSH and 2 mM GSSG (Sigma). The refolded ␣-Cbtx was purified by chromatography on a reverse phase HPLC column (Vydac, C 4 , 10 m, 1 ϫ 25 cm). The column was equilibrated in 0.1% trifluoroacetic acid, and elution was performed using an H 2 O/trifluoroacetic acid 0.1%, CH 3 CN gradient of 21-40% acetonitrile.
Characterization of Recombinant ␣-Cbtx-The purity of wild-type and ␣-Cbtx mutants was assessed by (i) SDS-PAGE electrophoresis with silver staining, (ii) chromatography on an analytical reverse phase HPLC column (Vydac, C 4 , 5 m, 0.46 ϫ 25 cm), and (iii) electrospray mass spectroscopy. Dichroic spectra were recorded at 22°C using a CD 6 Jobin-Yvon dichrograph to show that the toxins were correctly folded. The concentration of purified ␣-Cbtx was determined (i) by measuring the absorbance at 278 nm of a given solution of toxin and (ii) by using amino acid analysis.
Binding Assays-Competition experiments with AChR from Torpedo marmorata, prepared as described previously (8,9), were performed at equilibrium, using 3 H-toxin ␣ (18 Ci/mmol) as radioactive tracer. Varying concentrations of wild-type or mutated ␣-Cbtx were incubated in 0.25 ml of Ringer buffer containing 4 nM toxin-binding sites and 6.4 nM 3 H-toxin ␣, for at least 5 h at 20°C. The same results were obtained when the solutions were incubated overnight, indicating that our experiments were performed at equilibrium. The mixtures were filtered through Millipore filters (HAWP) that had been soaked in Ringer buffer. The filters were washed with 6 ml of Ringer buffer, dried, and after addition of 10 ml of scintillation solution (Lipoluma) counted on a Rackbeta counter (Amersham Pharmacia Biotech). Binding measurements were analyzed using the empirical Hill equation in order to define the Hill coefficient, which could be correlated to the presence of one (n H ϭ 1) or two binding sites (n H less than unity). The functions describing competitive binding of r␣-Cbtx mutants to either a single site or two independent sites were then fitted by nonlinear regression. Statistical improvement in the computer fit for a two-site model versus a one-site model was determined by the extra sum of squares principle (20) utilizing the F test. The two-site model was accepted when the statistical comparison between the two models gave a value of p Ͻ 0.05. The IC 50 values determined by nonlinear regression analysis were converted to apparent dissociation constants values (K d ) as described by Cheng and Prusoff (21). All these fits and the statistical analyses were performed using the Prism software (GraphPad). The experimental conditions were such that the minimal K i value that can be determined from competition data was not lower than about 6 pM. This value can be readily deduced from the equation by Cheng and Prusoff (21), in which the dissociation constant of the radioactive ligand is equal to 20 pM (8,9), the radioactive ligand concentration is 6.4 nM, and the concentration of binding sites 4 nM. As the K i value of the wild-type ␣-Cbtx was 58 pM (see below), our experimental conditions allowed us to show that a mutation could raise binding affinity by no more than about 9-fold (58/6 pM). However, as discussed below, such a situation was never encountered.

Production and Characterization of the Recombinant ␣-Cbtx
(r␣-Cbtx)-By using a strategy that was previously successfully applied to the production of short chain curaremimetic toxins (19), the long chain ␣-Cbtx was produced in the cytoplasm of E. coli fused to ZZ, a two IgG binding domain from protein A. The fusion protein was isolated by chromatography on an IgG-Sepharose 6FF column. The eluted fusion protein was produced with a yield of 30 mg/liter culture. Critical conditions were required for efficient cyanogen bromide cleavage of the fusion protein. The medium should be highly acidic (0.1 N HCl) to reduce the cleavage time to 6 h, at room temperature. Also, the fusion protein should not be re-oxidized before the cleavage treatment. These conditions constituted an appropriate compromise for the fusion protein to be cleaved without significant degradation of the cleaved moieties. The resulting material was refolded and purified. Fig. 1 shows SDS-PAGE of the recombinant toxin at different stages of production and purification, from the fusion protein to the purified refolded toxin. The final material migrated as a single band at the expected molecular weight, like ␣-Cbtx purified from venom. The recombinant material was chemically pure and comparable to the venom toxin as inferred from various lines of evidence. First, the recombinant toxin co-eluted with the native toxin in analytical reverse phase chromatography. Second, the recombinant and venom toxins have identical amino acid compositions and Nterminal amino acid sequences. Third, the far UV CD spectra of the venom and recombinant toxins were virtually superimposable (Fig. 2). The two toxins have similar contents of ␤-sheet, the major structural component of ␣-Cbtx (5,22). Fourth, electrospray mass analyses indicate the same mass value for both toxins: 7820.3, close to the theoretical value of 7821. Finally, the molar absorbance of the recombinant and folded toxin was 9670 M Ϫ1 ⅐cm Ϫ1 , a value identical to that determined for the venom toxin. The recombinant toxin therefore displays all the expected physicochemical characteristics of ␣-Cbtx. The final yield in toxin production was 1.2 mg/liter of culture.
Binding affinities of venom and r␣-Cbtx to Torpedo AChR were determined on the basis of competition experiments, using 3 H-toxin ␣ as a radioactive tracer. Typical competition binding curves obtained with the native and recombinant toxins are shown in Fig. 3. These data tally well with competition for a single class of binding sites. The K d value deduced from these competition experiments indicates an equilibrium binding constant for r␣-Cbtx equal to 58 Ϯ 7 pM, a value that agrees with that obtained for the venom toxin, 39 Ϯ 5 pM ( Fig. 3 and Table I).
Selection of r␣-Cbtx Mutants-At first, we identified residues that are conserved at structurally homologous positions in both long and short chain toxins. As illustrated in Fig. 4, the threedimensional structures of these toxins are readily superimposable, especially at the level of their second and third loops (3)(4)(5)(6). Inspection of a chart of toxin amino acid sequences (1) revealed that 14 structurally homologous positions, other than those occupied by the four half cystines, are occupied by the same (or chemically similar) residue in more than 90% of 84 short and long chain toxins. These are Gly-20, Tyr-25, Lys/Arg-27, Trp-29, Asp-31, Arg-33, Gly-34, Asp/Glu-38, Gly-40, Pro-44, Val/Ala-46, Lys-47, Thr/Ser-57, and Asn/Asp-61, using the numbering of short chain toxin amino acid sequences. Some of these positions are implicated in the capacity of short toxins to bind to muscular-type AChR (8 -10). These are Lys-27, Trp-29, Asp-31, Arg-33, and Glu-38 which belong to the second loop of the toxin, and Lys-47 which is present in the third loop. Since long and short toxins exert comparable binding functions on muscular-type AChRs (1, 7), we first probed the role of the structurally homologous Lys-23, Trp-25, Asp-27, Arg-33, Asp-38, and Lys-49 in r␣-Cbtx. We performed mutations K23E, W25H, W25F, W25A, D27N, D27R, R33E, D38L, and K49E. We also mutated Tyr-21 because its structural role has been often proposed (1). The introduced mutations were Y21A and Y21F.
Although the tertiary structures of both short and long toxins nicely superimposed on their second and third loops, no structural fit was seen between their first loops, the tip of their loop II, and in their C-terminal tails (Fig. 4). To study the role of the first loop in r␣-Cbtx, we mutated all residues from Phe-4 to Asp-13 into Ala; only Lys-12 was changed into Glu. The role of the tip of the central loop was first explored by mutating all residues located within the disulfide bond Cys-26 -Cys-30. Thus, in addition to mutations made at positions 27, we introduced the mutations A28G, A28R and F29W, F29L, F29A. Then, the two half-cystines of the fifth disulfide bond were mutated into serine, and the vicinal Ser-31, Ile-32, Lys-35, and Arg-36 were mutated into alanine. The role of the C-terminal tail was probed by mutating the highly conserved Phe-65 and Pro-66 into alanine. Finally, we mutated Thr-47 and Asp-53 as a result of their spatial proximity to the invariant Lys-49.
Properties of the Mutants-35 mutants were prepared in satisfactory yields, not much different from that observed for the wild-type toxin. These mutants were purified by HPLC and displayed the expected mass, as determined by electrospray mass analysis. Only one difficulty was encountered with the mutant Y21A; it could be produced as an unfolded protein but failed to refold, suggesting a possible structural role for Tyr-21, as anticipated previously (1).
Competition binding experiments were performed with all mutants, and typical curves are shown in Fig. 3. Table I displays all the affinity constants and the corresponding difference in energy of binding (⌬⌬G) deduced from such curves.
As deduced from binding curves, the vast majority of the mutants was characterized by Hill coefficients of 1.1 Ϯ 0.2, indicating similar affinities for the two AChR-binding sites. In contrast, the Hill coefficients calculated with the mutants K23E and K49E were equal to 0.6 Ϯ 0.1, suggesting that these residues interact differentially with the two toxin-binding sites on Torpedo AChR. We then tentatively fitted these two binding curves using non-linear regression, assuming the presence of either a single site or two distinct sites. A comparison of these two models through statistical analysis indicated that the binding of the two mutants K23E and K49E fit better with the presence of two distinct classes of receptor sites (p Ͻ 0.01). These results agree with recent data reporting that mutations K27E and K47E in a short chain toxin (NmmI) differentially influence the binding at the two receptor sites with, respectively, higher and lower affinities at the ␣␦ and ␣␥ sites (10,11). Furthermore, the mutants K23E and K49E in ␣-Cbtx differ 14-and 17-fold, respectively, at the two sites, and these values are comparable to those found for the corresponding NmmI mutants. Differential binding may also occur with the ␣-Cbtx mutant R33E, as seen with the mutant R33E in NmmI (10), although the decrease in affinity was great, and material was insufficient to complete the binding curve.
Over 6-fold affinity decreases occurred with the following mutations K23E, W25A, D27R, F29L, R33E, R36E, K49E, and F65A (Table I). To assess the structural consequences of the introduced mutations, we monitored the CD spectra of all the mutants. They were all similar, with three signals of compara-TABLE I Dissociation constants of toxin wild types and mutants on the Torpedo receptor The left part shows dissociation constants Ϯ S.E. of wild-type and mutated ␣-Cbtx, as deduced from binding competition data. The right part shows dissociation constants of wild-type and mutated erabutoxin a (Ea), as previously reported (8,9). Only mutations that caused affinity changes in Ea are shown. These mutation-sensitive positions are compared with those that occupy a structurally comparable position in ␣-Cbtx. Some mutations indicated the presence of two toxin-binding sites, whose K d values are indicated by two values. ⌬⌬G values are the difference in free energy of binding between wild-type and mutant toxins. ⌬⌬G ϭ ⌬G MUT Ϫ ⌬G WT ϭ RT ln(KdЈ/Kd), with R ϭ 1.99 cal/mol/K and T ϭ 293K. The residues whose mutations caused a ⌬⌬G higher than 1.1 kcal/mol are in bold. ble intensities as follows: one positive around 197 nm, one negative around 212 nm, and one positive around 230 nm. All these spectra were virtually superimposable (not shown) with that of the wild-type toxin (Fig. 2). Therefore, the wild-type toxin and its mutants have the same secondary structure content and presumably also the same tertiary structure. From these observations we suggest that mutations did not affect the architecture of the toxin and therefore that the affinity decreases observed upon mutation most probably reflect the direct contribution of the mutated residue to the binding of ␣-Cbtx to the Torpedo receptor. Four of the eight mutation-sensitive residues are positively charged, and three of them, Lys-23, Arg-33, and Lys-49, are highly conserved in curaremimetic toxins. By far the largest affinity decrease was observed upon reversion of the positive charge of Arg-33 which caused a difference in binding free energy of approximately 4 kcal/mol. The functional role of this highly conserved residue has often been predicted for long chain toxins (1), but this is the first time that this view receives experimental confirmation. It is striking that Arg-36, which is close to Arg-33 (8.6 Å), is also sensitive to mutation. Thus, the mutation R36A resulted in a 7.5-fold affinity decrease (Table I), which clearly indicates that a strong, positively charged character in this region is important in toxin binding.
The role of Lys-23 and Lys-49 in long toxins, as judged from previous chemical modifications, is somewhat controversial (12,13,17,23). Our data show that mutations K23E and K49E differentially affected the binding of ␣-Cbtx at the two receptor sites, causing a weak effect at one site and 28-and 53-fold affinity decreases, respectively, for the other site (Table I). Therefore, positive charges at positions 23 and 49 are critical for the long chain toxin to bind at one of the two sites. It is known that ␣-conotoxins are appropriate ligands for identification of the subunit composition of the high and low affinity sites of muscle AChRs (10, 24 -28). Nevertheless, the affinity ratio for the two sites is 46,000 and 19 for the muscle-derived cells and Torpedo AChRs, respectively (25). This low ratio observed with the Torpedo receptor did not allow us to identify which of the ␣␥ or ␣␦ interfaces was preferentially affected by mutations K23E and K49E.
Although highly conserved, the negative charge of Asp-27 is not in itself an important element for binding since the mutation D27N caused virtually no effect on toxin affinity. This is in agreement with the observation that mutation into alanine of the homologous Asp-30 in the long chain ␣-Bgtx had no effect on toxin affinity (18). However, we made a greater change by reversing its charge, so as to probe a possible role of Asp-27 in ␣-Cbtx. We found that mutation D27R caused a 31-fold affinity decrease ( Fig. 3 and Table I). This strongly suggests that the residue is involved in the binding, although at the moment we are unable to explain how.
Three mutation-sensitive residues have an aromatic side chain (Table I; Fig. 3), and two of them, Trp-25 and Phe-29, are highly conserved among curaremimetic toxins. At present, some ambiguity exists in the literature concerning the role of Trp-25, its chemical modification being associated with controversial results (15,29,30). In the light of previous work (8,9), we anticipated that mutations at Trp-25 would lower the affinity, and introduction in ␣-Cbtx of an alanine at position 25 did indeed reduce affinity 11-fold. However, neither mutation W25F nor W25H caused any detectable effect. The probable functional role of Trp-25 seems predominantly associated with its aromatic character. The role of an aromatic residue at this position in long chain toxins can be further illustrated with neuronal -bungarotoxin, a toxin that has a low affinity for muscular-type AChR. Substitution of its structurally homologous Glu-26 by a tryptophan caused a 5-fold affinity increase for the Torpedo receptor (31).
The role of Phe-29 in long chain toxins has not been investigated previously. Its aromatic character seems to be important since its substitution by leucine caused a 12-fold affinity decrease (Table I). In contrast, its mutation to the bulkier tryptophan (F29W) caused no effect on toxin binding activity. Surprisingly, the mutation F29A had no effect. We cannot explain this observation, although reducing the size of a functional residue may create a space that is compensated by vicinal side chains (32), and position 29 is sometimes naturally occupied by a threonine residue (1).
Mutation F65A caused a 7-fold affinity decrease, suggesting that the C-terminal tail of ␣-Cbtx may play some binding role, in agreement with previous data (33).
All other mutations in the second loop had little if any functional consequences. In particular, although Tyr-21, Ala-28, Cys-26 -Cys-30, Ser-31, Lys-35, and Asp-38 are highly conserved among the sequences of long chain toxins, their mutation revealed that they contribute little if at all to the interaction with Torpedo AChR (Table I). None of the mutations introduced in the first loop at the 10 positions between Phe-4 and Asp-13 had any effect on the stability of the toxinreceptor complex. Clearly, this loop is not involved in the binding of ␣-Cbtx to Torpedo AChR. The long C-terminal tail is so specific to long chain toxins that it is not entirely involved in the interaction with the muscular receptor type, since the mutation P66A caused no affinity decrease. Finally, although being located on loop III close to the critical Lys-49, Thr-47 and Asp-53 are unlikely to be functionally important since their respective mutations into alanine and lysine caused no significant affinity decreases (Table I). DISCUSSION We have identified residues of a long chain curaremimetic toxin whose mutations affect toxin binding to a muscular-type nicotinic acetylcholine receptor (AChR). The selected toxin was ␣-cobratoxin (␣-Cbtx) from venom of the cobra N. kaouthia. It is a typical long chain curaremimetic toxin whose structural properties have been largely elucidated by x-ray diffraction (5,22) and NMR spectroscopy (34) and whose functional properties have also been analyzed (35,36).
The recombinant toxin (r␣-Cbtx) was chemically, functionally and structurally indistinguishable from the venom toxin and was produced with a yield of approximately 1.2 mg per liter of culture. Other long chain toxins have been previously produced recombinantly. A few years ago, ␣-bungarotoxin (␣-Bgtx) from venom of the krait Bungarus multicinctus (37) was produced in E. coli, using another expression system (18). However, the recombinant toxin possessed 10 extra residues in its N-terminal sequence, and the yield was substantially lower. More comparable to our results were the data on the production of the recombinant neuronal -bungarotoxin (-Bgtx) (31). By using a similar approach based on the production of a fusion protein, an active -Bgtx was produced in E. coli. We suggest that the recombinant system described in the present paper may be of general applicability for small proteins rich in disulfide bonds.
On the Residues by Which ␣-Cbtx Binds to Torpedo AChR-If we consider as functionally important the residues for which at least one mutation reduced by more than 1.1 kcal/mol the affinity toward at least one of the two toxin-binding sites, the binding surfaces of ␣-Cbtx to Torpedo receptor then include Arg-33 Ͼ Lys-49 Ͼ Asp-27 Ͼ Lys-23 Ͼ Phe-29 Ն Trp-25 Ͼ Arg-36 Ն Phe-65. These residues cover a surface of approximately 880 Å 2 , with their side chains located on the toxin concave face and pointing in a similar direction (Fig. 4). Thus, if one looks at the toxin structure with the concave side facing the viewer and the three loops hanging down (see Fig. 5), they form a stretched surface that essentially crosses the second and third loops, with a large contribution of positively charged and aromatic residues.
Examination of residues that are insensitive to mutations reinforced this delineation. Thus, Fig. 5 shows that the left upper region of the molecule is uncritical for binding, since none of the residues of the first loop were sensitive to mutation. The lack of functionality of this loop was unexpected for three reasons. First, various residues, such as Phe/Tyr-4, Thr-6, Pro-7, and Ser-11, are found in more than 80% of the long chain toxins, and such a conservation is usually considered to reflect some functional and/or structural involvement, although this is clearly not the case here. Second, an NMR study of the complex between ␣-Bgtx and a receptor fragment (T␣185-196) indicated an interacting role of the toxin residues Thr-5, Thr-6, Ala-7, and Ile-11 (2). However, ␣-Bgtx is an atypical toxin, especially regarding the constitution of its first loop. Furthermore, the toxin might establish differential interactions with a receptor fragment or the whole receptor. Third, as will discussed below, the first loop of erabutoxin a, another curaremimetic toxin, is critically involved in binding to the same receptor (9).
Other regions of the toxins were also excluded from the receptor binding region. Thus, the base and part of the tip of the central loop seem to be excluded as judged from the unimportance of Tyr-21, Asp-38, Ala-28, Ser-31, and Ile-32. That the convex face of the toxin is excluded is suggested by the observation that Lys-35, whose side chain points toward this face, is insensitive to mutation. Although the role of the C-terminal tail is not yet completely understood, part of it is not implicated as shown by the insensitivity of the mutation P66A. Finally, the region on the right of the molecule is also poorly functional since of the three residues probed in the third loop, only Lys-49 is sensitive to mutation (Fig. 5B).
Therefore, the data showing both the positions that are sensitive to mutations and those that are insensitive delineate the FIG. 5. Comparison of the residues though which Ea (A) and ␣-Cbtx (B) interact with the Torpedo receptor. The concave face of the toxins is shown in both cases. Residues whose mutations caused a decrease in affinity for Torpedo AChR of less than 3-fold are in green, between 3-and 9-fold in yellow, 10-and 100-fold in orange, and higher than 100-fold in red. The only residue for which mutation induced an affinity increase is in blue. Data for Ea are from Ducancel et al. (40). site by which ␣-Cbtx recognizes the Torpedo AChR. This site includes the residues shown in Fig. 5B, some of them being involved in one or both toxin-binding sites.
Comparison of Functional Sites of Long and Short Chain Toxins-It is well documented and rather puzzling that the family of snake curaremimetic toxins include long and short members to achieve the same type of action, high affinity blockage of AChR (1). The present data obtained with the long ␣-Cbtx and those previously reported for the short Ea (8,9) allow us to examine how the two types of toxins exert this common function. Such a comparison is all the more appropriate as follows: (i) the studies of ␣-Cbtx and Ea were based on similar competition binding assays, using similar preparations of AChR from T. marmorata, and (ii) the equilibrium dissociation constants are quite similar for the two toxins, 70 and 60 pM for Ea and ␣-Cbtx, respectively.
The residues by which Ea and ␣-Cbtx interact with Torpedo receptor are shown in Fig. 5. A number of these functionally important residues are identical in nature and located at homologous positions in the two toxins (Fig. 5, Table I). These are, in Ea and ␣-Cbtx, respectively, Lys-27/Lys-23, Trp-29/Trp-25, Asp-31/Asp-27, Phe-32/Phe-29, Arg-33/Arg-33, and Lys-47/Lys-49. These residues are also conserved in most other short and long chain toxins and so may constitute a common functional core for most curaremimetic toxins. Since Arg-33, Lys-27, and Lys-47 of the short toxin NmmI bind close to the ␣-subunit region 188 -200 of the muscle receptor (11,38), we suggest that the functionally homologous residues of the long chain toxin recognize a similar region of the receptor.
Beside their remarkable functional similarities, the two toxins display a number of differences that can be appreciated at different levels. Thus, mutations at homologous residues of the functional core may not be followed by fully identical functional consequences. This is probably the case of the positively charged residues. Recent analysis performed with recombinant mouse muscle AChR subunits expressed in HEK cells indicated that the two sites can be differentiated by some toxin mutants (10). Although discrimination of the two binding sites is less straightforward with the Torpedo receptor than with mammalian muscle-type AChRs (25), some site selectivity could be identified from binding competition curves for the mutants K23E and K49E of ␣-Cbtx ( Fig. 3 and Table I) and perhaps with its R33E mutant. Our previous competition data obtained with Ea mutants were initially treated by assuming that the two binding sites on Torpedo AChR were equivalent (8,9). On reviewing these data, we noted that the two binding sites could also be discriminated with the Ea mutants S8T, Q10A, K27E, D31H, R33E, and K47E. Thus, the discriminating factors of EaK27E and ␣-CbtxK23E are, respectively, 11-and 13.5-fold, like those for EaK47E and ␣-CbtxK49E, which were 18-and 17-fold, respectively. However, the binding energy differences associated with mutations K27E (3.41 and 2 kcal/mol) and K47E (3.35 and 1.64 kcal/mol) in the short chain toxin are substantially higher than those of K23E (1.94 and 0.4 kcal/mol) and K49E (2.31 and 0.66 kcal/mol) introduced in the long chain toxin (Table I). These observations suggest that the two critical lysines and perhaps the critical arginine of the two toxins do not interact in strictly identical ways with the two binding sites on the Torpedo AChR. The homologous Trp-25/29 also may not play identical roles in the two toxins since the reduction in size of its aromatic group through mutation into His or a Phe affected the binding affinity in the short chain toxin only. Therefore, although the two toxins exploit a core of identical homologous residues for binding to the same target, these residues may perceive differential local environments and hence might not have strictly identical binding functions.
A number of other homologous residues were clearly sensitive to mutations in one toxin only. This is the case for Asp-38/ Glu-38 whose mutation to leucine affects the binding affinity of Ea but not ␣-Cbtx. A conserved residue, therefore, even if it is functionally important in one toxin member, may not necessarily be critical in all members of the toxin family, in contrast to what was previously anticipated (1,39).
Obviously, the most striking difference that can be seen between the two toxins concerned their first loop. Three residues at the tip of this loop are highly mutation-sensitive in Ea (40), whereas all residues in the corresponding loop in ␣-Cbtx are insensitive to mutation. However, alignments of toxin amino acid sequences could suggest some equivalence between a number of residues in the two toxins. Thus, Ser-8 in short chain toxins is frequently aligned with the highly conserved (80%) Thr-6 of long chain toxins (1). Nevertheless, although mutation at Ser-8 caused large affinity decreases in Ea, especially for one of the two binding sites (Table I), mutation at Thr-6 had no effect on the affinity of ␣-Cbtx. Ser-8 and Thr-6 are therefore not only functionally distinct but more generally a threonine and a serine cannot be always be considered theoretically to play comparable roles, even when they occupy a comparable location in two structurally and functionally similar proteins.
With hindsight, that the tip of the first loop does not play equivalent roles in short and long toxins is unsurprising. The first loop is substantially longer in the short toxin, precluding a superimposition with that of the long toxin (Fig. 4). Thus, considering the common and highly functional residue Arg-33 as a common binding reference for the two toxins, the closest residues at the tip of loop I in Ea and ␣-Cbtx are located at quite different distances. Thus, C␣R33 is at 12.4 Å from C␣S8 in Ea and at 17.8 Å from C␣T6 in ␣-Cbtx. We suggest that the shortness of the first loop in ␣-Cbtx makes its residues somewhat inaccessible to the interactive receptor surface. In contrast, the mutation-sensitive Phe-65 in ␣-Cbtx is not only closer to Arg-33 (C␣R33-C␣F65 ϭ 13.5Å) but is located at a distance that is comparable to that between Arg-33 in Ea and the tip of the first loop (see above). In other words, the C-terminal tail in ␣-Cbtx might be functionally equivalent to the critical residues located on the first loop of Ea.
At first sight, the short toxin seems to have more functional residues than the long toxin ( Fig. 5 and Table I), but 74% of the positions have been probed in Ea (40) and only 41% in ␣-Cbtx. Possibly, therefore, a number of functional residues in ␣-Cbtx remain to be identified, in particular within the C-terminal tail and the third loop, which have not been fully explored. It is also possible that the backbone contributes differentially in the two toxins and of course the mutational approach is uninformative about this. Further studies are needed to clarify these points. It is not inconceivable that the two toxins use a different number of residues to interact with the same receptor, even with a nearly identical high affinity. This would only imply that the individual binding contributions of the residues in the two toxins are different. We are unable to assess this point in the present analysis, where the introduced mutations did not necessarily reflect the genuine binding contribution of the original side chain, as evidenced by the differential effects of the introduction of different residues at the same position. Some mutations like those in which the original charge was reversed might have considerably amplified the effective binding contribution.
Concluding Remarks-In conclusion, the present study shows that long and short curaremimetic toxins exploit a common core composed of Lys-27/23, Trp-29/25, Asp-31/27, Phe-32/ 29, Arg-33/33, and Lys-47/Lys49, for binding to one or both toxin recognition sites of Torpedo AChR. However, these toxins also differ functionally in two ways. First, the binding energy contribution of some residues of the common core may be different in the two toxin families. Second, additional residues may be functionally important in one toxin only. Therefore, two curaremimetic toxins recognize the same receptor with comparably high affinities, through substantial variations around a common binding core. We are now investigating whether the same strategy occurs when the same toxin binds with high affinities to two different receptors, as observed with long toxins that bind to both muscular-type and ␣7 neuronal receptors (7).