Delineation of the Functional Site of α-Dendrotoxin

We identified the residues that are important for the binding of α-dendrotoxin (αDTX) to Kv1 potassium channels on rat brain synaptosomal membranes, using a mutational approach based on site-directed mutagenesis and chemical synthesis. Twenty-six of its 59 residues were individually substituted by alanine. Substitutions of Lys5 and Leu9 decreased affinity more than 1000-fold, and substitutions of Arg3, Arg4, Leu6, and Ile8 by 5–30-fold. Substitution of Lys5 by norleucine or ornithine also greatly altered the binding properties of αDTX. All of these analogs displayed similar circular dichroism spectra as compared with the wild-type αDTX, indicating that none of these substitutions affect the overall conformation of the toxin. Substitutions of Ser38 and Arg46 also reduced the affinity of the toxin but, in addition, modified its dichroic properties, suggesting that these two residues play a structural role. The other residues were excluded from the recognition site because their substitutions caused no significant affinity change. Thus, the functional site of αDTX includes six major binding residues, all located in its N-terminal region, with Lys5 and Leu9 being the most important. Comparison of the functional site of αDTX with that of DTX-K, another dendrotoxin (Smith, L. A., Reid, P. F., Wang, F. C., Parcej, D. N., Schmidt, J. J., Olson, M. A., and Dolly, J. O. (1997)Biochemistry 36, 7690–7696), reveals that they only share the predominant lysine and probably a leucine residue; the additional functional residues differ from one toxin to the other. Comparison of the functional site of αDTX with those of structurally unrelated potassium channel-blocking toxins from venomous invertebrates revealed the common presence of a protruding key lysine with a close important hydrophobic residue (Leu, Tyr, or Phe) and few additional residues. Therefore, irrespective of their phylogenetic origin, all of these toxins may have undergone a functional convergence. The functional site of αDTX is topographically unrelated to the “antiprotease site” of the structurally analogous bovine pancreatic trypsin inhibitor.

Therefore, all of these toxins may possess a functional surface that is complementary to this loop, an observation that raises the question as to how similar these surfaces are from one toxin to another. The answer to such a question may not only shed light on the evolution of these toxins but should also help characterize the surface by which Kv1 channels interact with these toxins. Mutational analyses have finely delineated the functional sites of scorpion toxins (26) and sea anemone toxins (11,25). Although the sea anemone and scorpion toxins are not structurally related, their functional sites share some similarities. They are all flat surfaces of comparable size (ϳ700 Å 2 ) with five functionally important residues, including a similar critical functional diad (11). This diad in both toxins comprises a lysine, which always plays a predominant part in binding, and an aromatic residue separated from the lysine by 6.6 Å. It was proposed that the lysine of scorpion toxins projects into the conduction pore of the channel (26 -29), and the most critical lysine of sea anemone toxins was predicted to play a similar role (11,25). Invertebrate toxins that have unrelated architectures but commonly block voltage-gated potassium channels therefore seem to display quite similar functional sites, suggesting that they underwent a convergent functional evolution. To investigate whether this proposal could be extended to toxins from vertebrates, we decided to delineate the functional site of ␣-dendrotoxin (␣DTX), a well known voltage-gated potassium channel-blocking toxin, isolated from venom of the green mamba Dendroaspis angusticeps (30).
This toxin is an extensively studied prototype of the family of dendrotoxins isolated from mamba venoms, which all block voltage-gated potassium channels (6 -8, 31). Its three-dimensional structure, as revealed by crystallographic studies (15), is highly similar to that of BPTI (14), the well known protease inhibitor. Comparison of amino acid sequences of members of the dendrotoxin family suggested that one or more of the lysines present in the triplet 28 -30 of ␣DTX may be involved in the interaction with Kv1 channels (6). However, preliminary site-directed mutagenesis experiments with ␣DTX did not confirm this proposal (32). In the present study, 26 residues of ␣DTX were individually substituted by an alanine, and the corresponding analogs were characterized according to their secondary structure and their capacity to inhibit the binding of radioactive ␣DTX to rat brain synaptosomal membranes. We thus identified a functional topography composed of six residues. Comparison of this site with functionally important residues recently identified in DTX-K (33), another dendrotoxin, shows that these toxins possess different functional sites that share only a common critical lysine. Furthermore, comparison of the ␣DTX functional site with those of structurally unrelated Kv1 channel-blocking toxins shows that all of these toxins share a minimal functional anatomy, composed of a key lysine and a hydrophobic residue, suggesting that they have undergone functional convergent evolution. It is also shown that the functional site of ␣DTX is topographically unrelated to that of BPTI, a well known trypsin inhibitor, although the two proteins share the same fold.

EXPERIMENTAL PROCEDURES
Materials-Oligonucleotides were synthesized on an Applied Biosystems 381A synthesizer. Amino acid composition analyses were performed using an Applied Biosystems device (420A Derivatizer and 130A Separation system), and N-terminal sequencing was done using the Applied Biosystems sequencer (477A protein sequencer) on-line with the phenylthiohydantoin analyzer (120A analyzer). Dichroic spectra were recorded on a Jobin-Yvon CD6 dichrograph. Expression vector pEZZ18 was generously provided by Prof. Mathias Uhlèn (Royal Institute of Technology, Stockholm, Sweden). Mass determination was performed on an API IIIϩ (Perkin-Elmer Sciex) mass spectrometer equipped with a nebulizer-assisted electrospray source.
Production of ␣DTX Analogs in Escherichia coli-Recombinant ␣DTX analogs were produced using the periplasmic expression system previously described (32). As for previous recombinant ␣DTXs (32), each analog was recovered as two isoforms: one with an N-terminal glutamine residue and the other, as found in wild-type ␣DTX, with an N-terminal pyroglutamic residue. The latter was used for structural and functional characterization.
Chemical Synthesis of ␣DTX and ␣DTX Analogs-The synthesis of native ␣DTX and of analogs P2A, K5A, K5Nle, K5Orn, L6A, I8A, L9A, E33A, R34A, and I58A was performed using the Fmoc/tert-butyl and maximal temporary protection strategy on an Applied Biosystems 431A synthesizer. The following standard chemical procedure was used for each monosubstituted analog: 0.05 mM of Fmoc-Gly-4-hydroxymethylphenoxy resin, 20-fold excess of each amino acid, dicyclohexylcarbodiimide/1-hydroxy-7-azabenzotriazole activation. Deprotection (1.5 h) and cleavage (300 mg of peptide ϩ resin) were achieved using the mixture trifluoroacetic acid/triisopropylsilane/water (9/0.5/0.5, v/v/v). The acidic mixture was then precipitated in 100 ml of cold diethylether, redissolved in 50 ml of 10% acetic acid in water, and lyophilized. Oxidation of the reduced peptide was achieved at 0.1 mg/ml in degassed potassium phosphate buffer (100 mM, pH 7.8) using the redox couple reduced glutathione (5 mM)/oxidized glutathione (0.5 mM). The disappearance of the reduced peptide was monitored by reverse phase high-pressure liquid chromatography on a C18 analytical column (4.6 ϫ 250 mm, Vydac) using a linear gradient of 40 min starting from 0.1% trifluoroacetic acid in water to 90% acetonitrile/ 0.1% trifluoroacetic acid in water. The crude oxidized peptide was then purified in two steps: reverse phase high-pressure liquid chromatography on a C18 semipreparative column (10 ϫ 250 mm, Vydac) in the same conditions as above followed by a final purification on a MonoS column (Amersham Pharmacia Biotech) in the conditions used for recombinant analogs (32).
Characterization of ␣DTX Analogs-For all ␣DTX analogs, the protein concentration was estimated using molar extinction coefficients determined by amino acid composition analyses. The presence of the desired substitution in the ␣DTX analogs produced in E. coli was checked by Edman degradation of the N-terminal glutamine isoform, or, for analogs S44A, K48A, E51A, R54A, and R55A, by sequencing peptides resulting from endoproteinase Lys-C (Boehringer Mannheim) cleavage of reduced and alkylated protein (32). For the chemically synthesized ␣DTX analogs, the substitution was assessed by mass spectrometry.
Disulfide Bridge Assignment of Chemically Synthesized Analogs-70 g of each analog and wild-type ␣DTX purified from venom were cleaved by trypsin. The resulting peptides were separated by reverse phase high-pressure liquid chromatography at a flow rate of 0.8 ml/min on a C18 column (4.6 ϫ 250 mm, Vydac) using a 65-min gradient of 10 -60% eluent B (0.085% trifluoroacetic acid, 50% acetonitrile in water) in eluent A (0.1% trifluoroacetic acid in water). Peptides were identified by means of amino acid composition analyses. For the analog R34A, the peptide containing half-cysteines 16, 32, 40, and 53 was further cleaved by chymotrypsin. Preparation of [ 125 I]␣DTX-Synthetic ␣DTX (10 g) was incubated at room temperature in a micro test tube (Eppendorf) coated with 1 g of iodogen (Pierce) with 2 mCi of 125 I (Amersham Pharmacia Biotech) in 200 l of 0.1 M sodium phosphate. After 15 min, 20 l of 0.1 M sodium thiosulfate were added, and the reaction mixture was injected onto a C18 column (Vydac). After washing the column with 25% eluent B (0.085% trifluoroacetic acid, 50% acetonitrile) in eluent A (0.1% trifluoroacetic acid), separation was achieved using a 40-min gradient of 25-60% eluent B in eluent A (1 ml/min). The fraction containing pure monoiodinated ␣DTX (2000 Ci/mmol) was kept at 4°C after the addition of bovine serum albumin (1 mg/ml).
The binding capacity of monoiodinated ␣DTX was 84%, as determined by saturating 4 pM of labeled toxin with increasing concentrations of rat brain synaptosomal membranes (0.35-24 g of protein/ml). This value remained constant for at least 85 days.
Preparation of Rat Brain Synaptosomal Membranes-Ten adult male rats (Sprague-Dawley, 150 g) were killed by decapitation. Decerebellate brains were homogenized in 100 ml of 10 mM Tris-HCl, pH 7.4, 10% sucrose, 0.1 mM phenylmethylsulfonyl fluoride using an UltraTurrax T25 (IKA-Labortechnik) and then a glass/Teflon homogenizer. This whole homogenate was centrifuged at 1500 ϫ g for 5 min, and the supernatant was recentrifuged for 10 min at 17,000 ϫ g. The pellet was resuspended in 80 ml of 5 mM Tris-HCl, pH 8, briefly homogenized with a glass/Teflon homogenizer, and left for 2 h at 4°C. Sucrose was then added to the lysed membrane suspension to a final concentration of 36%. The suspension was centrifuged for 90 min at 28,000 rpm using a SW28 rotor (Beckman) (66,100 -141,000 ϫ g) in tubes containing three layers of sucrose (36%: membrane suspension, 28.5 and 10%). The synaptic membrane fraction (lower interface) was diluted to less than 10% sucrose in 10 mM Tris-HCl, pH 7.4, pelleted by centrifugation at 17,000 ϫ g for 10 min, washed twice with 10 mM Tris-HCl, pH 7.4, aliquoted, flash-frozen in liquid nitrogen, and stored at Ϫ80°C. Protein concentration was determined by the method of Lowry, using bovine serum albumin as a standard.
Binding Assays-All binding experiments were done at room temperature, in 20 mM Tris-HCl, pH 7.4, 0.1 M NaCl, 0.1% bovine serum albumin. After 1.5-2 h, the incubation medium (1-6 ml) was rapidly filtered through Whatman GF/C glass fiber filters that had been presoaked in 0.5% (w/v) polyethylenimine, followed by three washes with ice-cold filtration buffer (20 mM Tris-HCl, 0.15 M NaCl; 3 ml per wash). Radioactivity in the filters was then counted in a gamma counter.

RESULTS
Choice of Substitutions-In the early stages of this study, there was no clear indication to suggest which analogs should be prepared. It was only established (32) that substitutions D12N and K28A/K29A/K30G had little effect on the biological property of the toxin, suggesting that these positions are unlikely to be important for toxin binding to Kv1 channels, whereas the substitution Z1Q caused affinity to increase 5-fold.
In the present study, therefore, selection of the positions of ␣DTX to be substituted was based on a comparison of the primary structures of dendrotoxins and functionally different proteins possessing the same fold ( Fig. 1). First, we expected that functionally important features of dendrotoxins would emerge from comparison of their primary structures with those of structurally related proteins. Five residues (Lys 5 , Lys 19 , Lys 28 , Gln 31 , and Ser 38 ) are more specific to dendrotoxins. Four (Lys 5 , Lys 19 , Gln 31 , and Ser 38 ) were substituted in this study, and one (Lys 28 ) in our previous work (32). Asp 36 , which is present in three of the four DTXs, was also substituted. Second, according to their primary structures, two groups of dendrotoxins can be distinguished: DTX-K and ␦DTX, and ␣DTX and DTX-I, which possess two additional residues in their N-terminal. Because this structural partition corresponds to distinct functional properties of DTXs (see under "Discussion"), we anticipated that residues uniquely conserved in both ␣DTX and DTX-I but not in DTX-K and ␦DTX may be associated with specific behavior of the first two toxins. Thirteen residues (Pro 2 , Arg 4 , Leu 6 , Ile 8 , His 10 , Arg 11 , Tyr 17 , Gln 27 , Lys 29 , Glu 33 , Trp 37 , Ser 44 , and Ile 58 ) were therefore substituted either in this study or in the previous one (Lys 29 ) (32). Third, because dendrotoxins are more basic than any other Kunitz-type proteins, we substituted the positively charged residues not yet included by the above selection procedure (Arg 3 , Arg 15 , Arg 34 , Arg 46 , Lys 48 , Arg 54 , and Arg 55 ). In addition, Asp 18 , which is specific to ␣DTX, was also substituted. We also mutated Glu 51 and, finally, Leu 9 and Asn 43 because they were found to be close to functionally important residues identified during this study.
Production of ␣DTX Analogs-Using the procedure previ-ously described (32), the production of ␣DTX analogs in E. coli yielded only 10 -90 g of each isoform per liter of culture. Additionally, attempts to produce the mutant ␣DTX-W37A failed, probably because Trp 37 belongs to a small hydrophobic core (15). To increase analog yields, we decided to produce them by a chemical approach (43). Peptide synthesis led to milligram amounts of synthetic ␣DTX, with a 6% yield from the starting resin. We identified the disulfide bonds of the resulting synthetic ␣DTX by submitting it to trypsin digestion. The peptide map obtained with the resulting fragments was identical to that of wild-type ␣DTX (not shown), indicating that the disulfide bridges are identical in the wild-type and synthetic toxins. In mice, intracerebroventricular injections of synthetic ␣DTX indicated an LD 50 of 5 ng/g, a value that is similar to that of the toxin isolated from venom (32). Furthermore, the synthetic and native toxins displayed a similar ability to inhibit the binding of radiolabeled ␣DTX to rat brain synaptosomal membranes (not shown). Using the chemical approach, we therefore synthesized chemically the analogs possessing the substitutions P2A, K5A, L6A, I8A, L9A, E33A, R34A, and I58A; all other alanine analogs were produced by the recombinant approach. We introduced nonnatural residues at position 5 (K5Nle and K5Orn) to probe in more detail the role of lysine 5, which was  39, amino acid 12 is an asparagine, but we found an aspartic acid at this position (32). However, all of the analogs produced in this study possess an asparagine in position 12.) Secondary structure elements are indicated above the primary structures (hatched bars, ␣-helix and 3 10 helix; checked bars, ␤-strands).
found to be critical for toxin binding (see below). The analog N43A was also chemically synthesized, but we could not oxidize it, suggesting that the introduced substitution prevented the correct folding of the protein.
Binding of [ 125 I]␣DTX to Rat Brain Synaptosomal Membranes-Rat brain synaptosomal membranes are frequently used to study the binding properties of various ligands toward Kv1 channels, using radiolabeled ␣DTX as a tracer (11,25,32,44). High proportions of synaptosomal membranes are usually used in such experiments. Because our mutational analysis required numerous competitive binding experiments, we established new binding conditions using the lowest possible proportions of membranes. A typical saturation binding experiment is shown in Fig (45), we assessed the validity of the apparent K d value, by determining the kinetics of binding of [ 125 I]␣DTX binding to rat brain synaptosomal membranes. Dissociation was initiated by adding a 1000-fold molar excess of ␣DTX (Fig. 3). The mean value (Ϯ S.D.) from all individual experiments (n ϭ 3) for the dissociation rate constant (k off ) was 3.8 Ϯ 1.2 ϫ 10 Ϫ4 s Ϫ1 . The association rate constant of [ 125 I]␣DTX binding was determined using a constant concentration of synaptic membranes and three different [ 125 I]␣DTX concentrations (Fig. 4A). The k on was deduced from the plot of k obs versus [ 125 I]␣DTX concentrations (Fig. 4B). The mean value of k on (Ϯ S.D.) from two individual experiments was 2.0 Ϯ 1.2 ϫ 10 8 M Ϫ1 s Ϫ1 . The dissociation constant K d calculated from the ratio of these values is 1.9 pM, a value that agrees with that determined by saturation binding to equilibrium. The values of the apparent K d and k on deduced from our experiments differed from those reported in Ref. 45, probably as a result of experimental differences in protein concentrations and/or binding sites (46 -48).
Binding Properties of ␣DTX Analogs- Fig. 5A shows typical plots of the competitive binding of [ 125 I]␣DTX and wild-type ␣DTX and three analogs with lower relative affinities to rat brain synaptosomal membranes. Using similar experiments, we investigated the inhibitory potency of 28 analogs of ␣DTX (26 alanine analogs plus K5Nle and K5Orn). The resulting K i values are listed in Table I, and the relative affinities are shown in Fig. 5B. Clearly, some mutations caused marked decreases in affinity, whereas several others did not. Similar changes in affinity were found when binding experiments were performed with higher concentrations of binding sites (not shown). The results of these mutations will now be presented by dividing the toxin structure into four regions: (i) 1-18, which contains a 3 10 helical structure between residues 5 and 9; (ii) 19 -38, with a double-stranded antiparallel ␤-sheet and a ␤-turn between residues 27 and 31; (iii) 39 -49, with no canonical secondary structure; and (iv) 50 -58, which adopts a helical structure (15).
Twelve of the 18 positions of the N-terminal region have been explored (Fig. 5B), revealing that substitutions of six nearly consecutive residues (Arg 3 , Arg 4 , Lys 5 , Leu 6 , Ile 8 , and Leu 9 ) caused affinity decreases. K5A and L9A seem to be critical, and they lowered affinity by more than three orders of magnitude. This is in agreement with recent data that indicated that the selective modification of Lys 5 also greatly affected the toxin affinity of the highly analogous dendrotoxin I (49). In addition, we observed that substitution of Lys 5 by ornithine ((CH 2 ) 3 -NH 3 ϩ : 5 Å) and norleucine (which lacks the ⑀-amine) lowered affinity by approximately two and three orders of magnitude, respectively (Fig. 5C). The substitution R3A only induced a less than 10-fold decrease, whereas the three substitutions R4A, L6A, and I8A resulted in 10 -30-fold decreases in affinity. The other substitutions (P2A, H10A, R11A, R15A, Y17A, and D18A) and the previously described mutation D12N (32) did not significantly reduce the affinity. In the region 19 -38, the substitutions K19A, Q27A, Q31A, R34A, and D36A had no significant effect on the affinity of ␣DTX, in agreement with previous results that showed that substitutions in this region had little effect on the biological properties of ␣DTX (32). In contrast, the substitution S38A was associated with a 7-fold decrease in affinity. Of the three substitutions that were introduced in the 39 -50 region, S44A and K48A caused no change in binding properties, whereas R46A induced a nearly 200-fold affinity decrease. Finally, although the C-terminal helix 50 -58 is spatially close to the critical N-terminal region, none of the substitutions E51A, R54A, R55A, and I58A caused any significant change in binding affinity.
Structural Properties of ␣DTX Analogs-Before concluding that an affinity decrease reflects the functional contribution of a substituted residue, it is essential to establish that the substitution has not altered the toxin structure. To probe the conformational state of the toxin analogs displaying weaker binding affinity, we compared their content in secondary structure with that of the wild-type protein, by monitoring their circular dichroic spectra (Fig. 6). None of the substitutions introduced at positions 5 and 9 were associated with any detectable change of the dichroic content of the toxin, indicating that no major conformational change occurred upon either of these substitutions. A similar conclusion was reached with the analogs R3A, R4A, L6A, and I8A (data not shown). Therefore, the decreases in affinity observed upon substitutions at posi-tions 3, 4, 5, 6, 8, and 9 ( Fig. 5B and Table I) probably do not reflect a structural role of these side chains.
The dichroic spectrum of the analog S38A (Fig. 6) was markedly different from that of the wild-type toxin; the band at 210 nm observed for the native toxin was shifted to 205 nm, and a shoulder became observable around 225 nm. Such a change is difficult to interpret, but it may suggest that the helical structure is reinforced and/or that the ␤-sheet content of the toxin is decreased. Such a finding is not too surprising because Ser 38 establishes critical interactions with other parts of the toxin, its side chain oxygen atom being involved in hydrogen bonds with main-chain nitrogen atoms of Cys 16 , Cys 40 and in a bifurcated hydrogen bond with carbonyl oxygen atoms of Pro 13 and Cys 40 (15). Thus, its substitution by alanine may prevent formation of these bonds and hence alter the toxin structure. Similarly, the circular dichroic spectrum of ␣DTXR46A was different from that of wild-type ␣DTX (Fig. 6), suggesting that the substitution also induced structural perturbations. Exam- decreased the stability of an analog of BPTI in which only the disulfide 5-55 was present (52). The consequences may be similar in ␣DTX upon substitution of Arg 46 by an alanine. We therefore suggest that the affinity decreases noted with substitutions at positions 38 and 46 may result from perturbation of ␣DTX conformation. DISCUSSION The Functional Site of ␣DTX-Although ␣DTX has been widely studied for more than 15 years (reviewed in Ref. 8), little is known regarding the molecular features that are associated with its capacity to recognize Kv1 channels. We used extensive mutational analysis to delineate its functional site. Twenty-six positions were individually substituted by an alanine. This suppresses the side chain beyond the ␤-carbon and does not introduce a new chemical function. Such an approach proved to be successful in identifying functionally important residues in various proteins (53), but it provides little information on how the residue contributes to the binding of the protein to its target. If one includes the substitutions described in our previous report (32) (K28A/K29A/K30G, D12N, and Z1Q), we have now probed 31 positions of ␣DTX.
All of the analogs were tested for their capacity to inhibit the binding of iodinated ␣DTX to rat brain synaptosomal membranes. ␣DTX is known to recognize various subtypes of Kv1 channels with different affinities. Using cloned Kv1 channels, electrophysiological data revealed that ␣DTX blocks Kv1.1, Kv1.2, and Kv1.6 with comparable affinities, but has much lower affinities for Kv1.3, Kv1.4, and Kv1.5 (reviewed in Refs. 7-8 and 31). Furthermore, the ␣DTX-specific Kv1 channels from rat brain can be heteromeric and may form a heterogeneous population of molecules (54,55). Almost all ␣DTX-acceptors contain Kv1.2 ␣-subunits, and 50% of them also possess Kv1.1 ␣-subunits. In contrast, the ␣DTX-sensitive potassium channels containing Kv1.6 or Kv1.4 ␣-subunits appear to be much less abundant (54). Despite this heterogeneity, our present and previous binding experiments suggest the presence of a predominant single class of binding sites.
Six residues (Arg 3 , Arg 4 , Lys 5 , Leu 6 , Ile 8 , and Leu 9 ) were identified as being functionally important because their substitution caused a substantial decrease in the binding affinity of ␣DTX to rat brain synaptosomal membranes without affecting the toxin conformation. These residues are represented in Fig. 7, where they have been colored red, orange, and yellow for substitutions causing affinity decreases of above 100-fold, be-  tween 10-and 100-fold, and between 5-and 10-fold, respectively. Glp 1 , the glutamine substitution of which was previously shown to increase the affinity by 5-fold (32), is shown in yellow. All of these residues are concentrated in one region. However, before concluding that the functional site is thus delineated, the borders of the site also had to be identified. We found that 18 other positions, i.e. Pro 2 , His 10 , Arg 11 55 , and Ile 58 , can be substituted without causing affinity changes, suggesting that they are not major binding participants. These functionally excluded residues are colored green in Fig. 7. Arg 46 and Ser 38 , the alanine substitu-tion of which induced affinity decreases and structural perturbations, are also colored green. All of these residues are not only widely spread on the toxin surface but a number of them, such as Pro 2 , Gln 27 , Gln 31 , Ser 44 , and Arg 55 , also closely surround the six functionally important residues. We therefore suggest that most of the side chains by which ␣DTX binds to potassium channels from rat brain synaptosomes have been identified.
The six functional residues identified in this study are nearly consecutively located along the peptide stretch 2-8 (Fig. 7A). They form a surface that covers approximately 700 Å 2 , a value that is compatible with those usually observed at the interface The backbone of all residues and the side chains of residues that have not been tested are colored white, and the side chains of substitutioninsensitive residues are green. The side chains of Ser 38 and Arg 46 , the mutation of which causes both structural and functional perturbations (see text), are also colored green, as are those of Asp 12 , Lys 28 , Lys 29 , and Lys 30 , because it was previously shown that substitutions D12N and K28A/K29A/K30G did not alter the binding affinity of the toxin (32). Side chains of functionally important residues are colored red, orange, and yellow according to the decrease in affinity observed upon their alanine substitution: red, Ͼ100-fold; orange, 10 -100-fold; yellow, 5-10-fold. Substitution of Glp 1 into glutamine was previously shown to increase the affinity 5-fold (32). This residue for which only the ␣-carbon coordinates are available is also colored yellow.

FIG. 8. Comparison of the functional topographies of DTX-K and
␣DTX. The side chains of the functional site of DTX-K (33) (left) and ␣DTX (this work) (right) are shown in cyan and magenta (common critical lysine) on a tube representation of their three-dimensional structures (16,15). Only the side chains of the residues of DTX-K, the substitution of which lowered affinity at least 5-fold (33), are shown. of protein-protein complexes (56). The center of this surface is occupied by Lys 5 , one of the two most important residues, which protrudes from the functional surface (Fig. 7B). To investigate the role of the different parts of its side chain, we substituted Lys 5 into ornithine and norleucine. The results (Fig. 5C) indicated a critical role of both its ⑀-amine group and, to a lesser extent, the distance between its ␣-carbon and this amine. The second most important residue is Leu 9 , the ␦-carbon of which is 6.9 Å from the ␣-carbon of Lys 5 . The three moderately important residues Arg 4 , Leu 6 , and Ile 8 surround Lys 5 in a rather compact manner, whereas the less important residue (Arg 3 ) and Glp 1 are more remote from the two important residues Lys 5 and Leu 9 , and are presumably at the border of the functional site.
Dendrotoxins Do Not All Possess the Same Functional Site-As mentioned above, dendrotoxins can be divided into two structural subclasses. One of them includes ␣DTX and DTX-I, and the other comprises ␦DTX and DTX-K (see Fig. 1). Dendrotoxins from the two groups are characterized by different biological properties. Thus, ␣DTX is relatively selective in blocking slow-inactivating K ϩ currents in rat dorsal root ganglion cells, whereas ␦DTX, which belongs to the DTX-K group, is more selective against non-inactivating currents (57). DTX-I, of the ␣DTX group, blocks channels formed by expression in Xenopus oocytes of ␣-subunits Kv1.1, Kv1.2, and Kv1.6, whereas DTX-K is highly selective for Kv1.1 (58). However, despite 95% sequence identity between DTX-K and ␦DTX, DTX-K is much more potent than ␦DTX in blocking Kv1.1 channels (59). Therefore, although all members of the dendrotoxin family bind to voltage-gated potassium channels, they display slightly distinct biological activities, suggesting differences in their functional sites.
Recently, nine variants of DTX-K were produced and tested for their capacity to inhibit the binding of [ 125 I]DTX-K to rat brain synaptic membranes (33). Four residues were identified as playing a major role in the toxin binding function: Lys 3 and Lys 6 in the N-terminal part of the toxin, and Trp 25 and Lys 26 , which are located in the ␤-turn that joins the two antiparallel ␤-strands. Comparison of the functional sites of ␣DTX and DTX-K (Fig. 8) revealed that they share only a common lysine (Lys 3 in DTX-K and Lys 5 in ␣DTX). Interestingly, the affinity decrease observed upon the substitution K3A in DTX-K (1250fold) is similar to that induced by the substitution K5A in ␣DTX, suggesting that these residues play a similar important role in the two toxins. Because the second most important residue of ␣DTX (Leu 9 ) is found in all dendrotoxins, we suspect that this residue, which has not yet been probed in DTX-K, may also be functionally important in this toxin. All of the other functional residues of DTX-K differ in ␣DTX and most are specific to one dendrotoxin subclass. Thus, among the four additional functional residues in ␣DTX (Arg 3 , Arg 4 , Leu 6 , and Ile 8 ), Arg 3 is only found in ␣DTX, whereas the three others (Arg 4 , Leu 6 , and Ile 8 ) are also found in DTX-I but not in the other dendrotoxins. Similarly, among the additional functional residues identified in DTX-K (Lys 6 , Trp 25 , and Lys 26 ), Trp 25 and Lys 6 are also found in ␦DTX, but not in ␣DTX or DTX-I. Therefore, dendrotoxins from the same subclass may share similar functional sites, which, however, may differ from one subclass to the other. Only a common core, composed of Lys 5 (or Lys 3 in DTX-K) and, probably, Leu 9 (or Leu 7 in DTX-K), seems to be shared by all dendrotoxins. This could explain the heterogeneity in biological properties observed between the two subclasses. A similar situation was recently observed with curaremimetic toxins. On the basis of their primary structures, these toxins are currently classified as long-chain and shortchain toxins, the longer toxins possessing, in particular, an extra small cyclic loop (60). Both long and short chain toxins bind to muscular nicotinic acetylcholine receptors with high affinities (61), whereas the presence of the extra loop uniquely provides long-chain toxins with a high affinity for ␣7 neuronal receptors.
␣DTX Shares a Functional Anatomy with Other Potassium Channel Inhibitors-The binding sites of three types of structurally unrelated Kv1 channel blocking toxins are known. These are those of ␣DTX (this work) and DTX-K (33), scorpion toxins (26 -28), and sea anemone toxins (11,25). Although these toxins have distinct architectures, they are all likely to bind to the same region of Kv1 channels (11, 18 -24) and display functional surfaces that share several molecular features. Their functional surfaces have a comparable size of approximately 700 Å 2 . They all have a protruding lysine, which appears to be a key center associated with the binding to Kv1 channels (11,(25)(26)(27)(28)33). In scorpion toxins, the positive charge of the key lysine (Lys 27 ) is thought to mimic the positive charge of potassium ions (26 -29). Furthermore, in all models aiming at docking the position of a scorpion toxin molecule bound to a potassium channel, the side chain of Lys 27 projects into the conduction pore of the channel, whereas the other important residues interact with residues of the outer wide vestibule (26 -29, 62-64). That Lys 5 of ␣DTX may play a role similar to that of Lys 27 of scorpion toxins was suggested by mutational data of Lys 5 into norleucine and ornithine. Clearly, its ⑀-amine group, presumably positively charged, as well as the distance between the ␣-carbon and this amine group, are important factors. Comparable results were obtained with the most critical lysine of BgK for binding to rat brain Kv1 channels. 2 Thus, we propose that Lys 5 of ␣DTX, the side chain of which protrudes from the functional site (Fig. 7B), is a key residue that may fit into the pore of the vestibule of voltage-gated potassium FIG. 9. Comparison of the functional topographies of BgK, ␣DTX, and charybdotoxin. The side chains of the functional residues of BgK (11), ␣DTX (this work), and charybdotoxin (CTX) (26) are shown on tube representations of their three-dimensional structures (11,15,9). The critical lysine and the hydrophobic residue of the diad (see text) are colored magenta and yellow, respectively. The other functional residues are shown in cyan. channels, as does Lys 27 of scorpion toxins and, presumably, the key lysine of sea anemone toxins (11,25).
The key lysine of both sea anemone and scorpion toxins is always associated with a 6.6 Ϯ 1 Å distant key aromatic residue (␣-carbon of lysine-center of aromatic ring), either a tyrosine or a phenylalanine. Such a diad of residues was proposed to form a conserved functional core in potassium channelblocking toxins from invertebrates (11). The functional site of ␣DTX does not possess such an aromatic residue, but it does have the highly critical and hydrophobic Leu 9 , the distance from Lys 5 of which (␣-carbon lysine-␦-carbon leucine: 6.9 Å) is comparable to that observed in the diads of invertebrate toxins. Therefore, all toxins that block Kv1 channels seem to possess a comparable functional diad, which, however, should be more broadly defined as being composed of a hydrophobic residue separated from the key lysine by 6.7 Ϯ 0.9 Å (␣-carbon of lysine-center of aromatic ring or ␦-carbon of the aliphatic side chain) (Fig. 9). Enzymes with comparable functions possess critical catalytic diads or triads, although they adopt different overall architectures. We find it striking that structurally distinct toxins may possess binding diads with comparable recognition functions.
It is remarkable that the functional site of ␣DTX, which is produced by a venomous vertebrate (a mamba), shares some features with the functional sites of structurally unrelated potassium channel-blocking toxins produced by venomous invertebrates. This observation suggests that all of these toxins, irrespective of the phylogenetic origin of the venomous animals that produce them, underwent a comparable convergent evolution, the Kv1 channels of the prey possibly acting as evolutionary sieves.
␣DTX and BPTI Have Topographically Unrelated Functional Sites-As already mentioned, the fold adopted by ␣DTX is also observed in proteins with unrelated functions. This is the case, in particular, of BPTI (14), which blocks the catalytic action of proteases. That a conserved protein fold is capable of exerting unrelated functions is not uncommon. Thus, it is well known that antibodies (65) or enzymes (66) with conserved overall scaffolds can exert distinct functions. In these cases, however, the respective functional sites, i.e. paratopes or catalytic cavities, are confined in similar areas. Here, we are facing a completely different situation because the functional sites of BPTI and ␣DTX are located in topographically unrelated regions of the protein. The "antiprotease site" region of BPTI (67) comprises residues Pro 13 , Lys 15 , Arg 17 , Ile 19 , and Arg 39 , which are located on "top" of the BPTI structure (Fig. 10), whereas the ␣DTX functional site is at the opposite end of the fold. Clearly, therefore, the surface of the BPTI/DTX fold can accommodate distinct functional sites in unrelated regions, a scenario that agrees with previous proposals made for other toxin folds (68 -70). It will now be interesting to examine whether additional regions of the DTX/BPTI fold are associated with other functions, such as the calcium channel-blocking activity of calcicludine (34). In any case, the available data suggest that the DTX/BPTI fold undergoes a natural "engineering" resulting in divergent functional topographies. As a result, the DTX/BPTI fold appears to be a promising additional template for engineering of novel functions, as initiated with the scorpion toxin fold (71)(72).